Chemical Reviews: Modern Topics in Chemical Sensing issue 2

Volume 108, Number 2

Introduction: Modern Topics in Chemical Sensing Chemical Sensors is a large branch of analytical chemistry that has unique characteristics. It is very broad, ranging from solid-state physics to molecular biology. Because of this breadth, choice has to be made between the all-encompassing, but rather shallow treatment of the subject on one hand or selection of a few topics covered in depth by experts in the field. The latter approach has been chosen for this issue. It does not pretend to cover all new aspects of this active and growing field. Several rather important modern topics of chemical sensing that could have been included have not been, mainly because of the lack of available and willing authors. There is some confusion in the terminology. The label “chemical sensor” is often used to describe an analytical procedure that should be more appropriately called an “analytical assay” or “sensing system”. The main difference between the two lies in the mode of information acquisition. While a chemical sensor acquires information continuously, a sensing system obtains information in discrete steps. It does not matter that most modern analytical, specifically bioanalytical, assays are automated and can run unattended for long periods of time. The discontinuous nature of their operation still distinguishes them from true chemical sensors. The two groups of procedures are fully complementary and valuable tools of analytical chemistry. This thematic issue is limited to true chemical sensors only. A typical modern chemical sensor consists of a physical, “transducer” and a chemically selective material. Different strategies can be employed to extract maximum information about the sample. Among these are multivariate analysis of data obtained from sensing arrays, use of spatially and temporally distributed sensors, and integration of sensors with solid-state processing technology. Two examples of important sensor applications have been also included in this issue. Ion selective electrodes are one of the oldest chemical sensors. In the past decade, some significantsalthough not always positivesdevelopments have taken place, such as the lowering of detection limits by nonequilibrium operation. The important issue of internal contact in solid-state potentiometric sensors has been one of the most significant contributions of Johan Bobacka, Ari Ivaska, and Andrzej Lewenstam. Joseph R. Stetter and Jing Li focused their review of amperometric sensors on the sensing of gases. After a rather brief overview of the general amperometric sensing principles, the authors concentrate on new nanomaterials that have been recently utilized in these devices. Somewhat neglected but very interesting semiconductor

Jirˇ´ı Janata is Georgia Research Alliance Eminent Scholar in the School of Chemistry and Biochemistry, Georgia Institute of Science and Technology. Between 1991 and 1997 he was an Associate Director of Environmental Molecular Sciences Laboratory, Pacific Northwest National Laboratory, in Richland, Washington. Prior to that appointment, he was Professor of Materials Science and Professor of Bioengineering at the University of Utah for 17 years. He was born in Czechoslovakia, where he received his Ph.D. degree in analytical chemistry from the Charles University (Prague) in 1965. His current interests include interfacial chemistry, chemical sensors, and electroanalytical chemistry with particular emphasis on development of chemical sensors for environmental and security applications.

junction gas sensors are thoroughly reviewed by Karin PotjeKamloth. With the advent of organic semiconductors, the importance of this type of chemical sensor is expected to grow, as is indicated by the very large number of references included in her review. This review should also be of interest to researchers dealing with organic electronics, such as organic field-effect transistors. Four reviews are devoted to advances in optical sensing. Colette McDonagh, Conor S. Burke, and Brian D. MacCraith emphasize new hardware that has become available thanks to rapid advances in communication and signal processing technologies. Utilization of optical sensing principles in combination with biologically derived selectivity is covered in the review by Sergey M. Borisov and Otto S. Wolfbeis. It includes selectivity derived from enzymatic, immunochemical, oligonucleotide, and organismal principles. Sensors based on plasmon resonance are discussed in the review by Jirˇ´ı Homola and in the closely related review by Matthew E. Stewart, Christopher R. Anderton, Lucas B. Thompson, Joana Maria, Stephen K. Gray, John A. Rogers, and Ralph G. Nuzzo. The latter introduces the new subject of nanostructuring.

10.1021/cr0680991 CCC: $71.00 © 2008 American Chemical Society Published on Web 01/26/2008

328 Chemical Reviews, 2008, Vol. 108, No. 2

Cantilevers are a rapidly growing subgroup of mass sensors. That growth is fueled by the wide use of microfabricated tips for various types of scanning microscopy. The subject is reviewed by Karen M. Goeders, Jonathan S. Colton, and Lawrence A. Bottomley. A review of sensors for radionuclides has been prepared by Jay W. Grate, Oleg B. Egorov, Matthew J. O’Hara, and Timothy A. DeVol and is believed to be the first of its kind. For that reason, the criterion of “continuous data acquisition”, used to define true chemical sensors, has been somewhat relaxed, allowing some crossover to sensing systems. There is much to be gained by increasing the “order” of chemical sensors, in direct analogy with hyphenated analytical techniques such as GC-MS, MS-MS, etc. Many deficiencies in the performance of single sensors can be avoided by grouping diverse sensors into higher order arrays, as shown by Andreas Hierlemann and Ricardo Gutierrez-Osuna. The benefits of increasing the sheer number of sensors to “very large sensing arrays” are demonstrated in the review by Christopher N. LaFratta and David R. Walt. This article also highlights an elegant approach to the tricky problem of addressing individual elements in such an array. Very large scale integration of electronics is an inevitable requirement in the realization of sensing arrays. That aspect is addressed in the review by Hierlemann and is also exhaustively treated by Segyeong Joo and Richard B. Brown Rather interesting coupling of the Internet with chemical sensing is discussed in a review by Dermot Diamond, Shirley Coyle, Silvia Scarmagnani, and Jer Hayes. It clearly shows that sensors are taking over the world as “wireless sensing networks”. The extraction of distributed spatial and temporal information through multiple sensing channels and the acquisition of information from the dynamic behavior of sensing arrays are demonstrated in a review by Takamichi Nakamoto and Hiroshi Ishida. Electronic noses have been in existence for

Editorial

at least two decades. The current state of this mode of multivariate sensing is discussed by Frank Ro¨ck, Nicolae Barsan, and Udo Weimar. Development of new sensing materials is essential for the advancement of modern chemical sensors. It has been discussed to some extent in all of the above reviews. However, several contributions have been devoted exclusively to this important aspect of research. A review by Jay W. Grate focuses on the development of materials for gas and vapor sensing that are based on acidic hydrogen-bonding polymers. David W. Hatchett and Mira Josowicz treat composites of intrinsically conducting polymers with various nanomaterials. These heterogeneous sensing materials include metals, carbon-based nanomaterials, and semiconductors. A broad overview of combinatorial strategies for the development of new sensing materials is presented by Radislav A. Potyrailo and Vladimir M. Mirsky. The issue concludes with two topics of perennial interest: glucose biosensors, by Joseph Wang, and application of modern sensors to critical care medicine, by Bruce A. McKinley. Chemical sensors have achieved a rather dubious distinction. They often serve as the default mode for a material or procedure that “did not quite work out” for its original purpose or to justify results for which there is no obvious “other use”. The phrase “...and it can be also used for chemical sensing” is typical and unfortunately much too common in the chemical literature. Hopefully, the reviews compiled in this issue will change this paradigm at least to some extent. Jirˇ´ı Janata Georgia Institute of Technology CR0680991

Chem. Rev. 2008, 108, 329−351

329

Potentiometric Ion Sensors Johan Bobacka,*,† Ari Ivaska,† and Andrzej Lewenstam‡,§ Åbo Akademi University, Process Chemistry Centre, c/o Laboratory of Analytical Chemistry, Biskopsgatan 8, FI-20500 Turku-Åbo, Finland; Faculty of Material Science and Ceramics, AGH-University of Science and Technology, Al. Mickiewicza 30, PL-30059 Cracow, Poland; and Åbo Akademi University, Process Chemistry Centre, c/o Center for Process Analytical Chemistry and Sensor Technology (ProSens), Biskopsgatan 8, FI-20500 Turku-Åbo, Finland Received April 25, 2007

Contents

1. Introduction

1. Introduction 2. Theory of Potentiometric Response 2.1. Total-Equilibrium Models (Classical Models) 2.1.1. Phase-Boundary-Potential Approach 2.1.2. Total-Membrane-Potential Approach 2.1.3. Semiempirical Equations 2.1.4. Comments on Model Benefits and Drawbacks of Total-Equilibrium Models 2.2. Local-Equilibrium Models (Diffusion-Layer Models) 2.2.1. Comments on Model Benefits and Drawbacks of DLM Models 2.3. Advanced Nonequilibrium Models 2.3.1. Comments on the Benefits and Drawbacks of the Advanced Model 3. Solid-Contact ISEs 3.1. Conducting Polymers 3.2. Conducting Polymers as Solid Contact 3.2.1. Polypyrroles 3.2.2. Polythiophenes 3.2.3. Polyanilines 3.3. Conducting Polymers Dissolved in the Ion-Selective Membrane 3.4. Conducting Polymers as Sensing Membranes 3.5. Comments on the Ion-to-Electron Transduction Process 4. Miniaturized ISEs 4.1. Conventional Microelectrodes 4.2. Solid-Contact Microelectrodes 4.3. Microelectrodes in Flow Analysis 4.4. Life Science and Biology Applications 5. New Wave 6. Conclusions 7. Acknowledgments 8. References

329 329 331 331 332 332 333 333 335 335 336 338 338 340 340 340 342 342 343 344 345 345 346 347 347 347 348 348 348

* Corresponding author. E-mail: [email protected]. Phone: +358 2 215 3246. Fax: +358 2 215 4479. † Åbo Akademi University, Process Chemistry Centre, c/o Laboratory of Analytical Chemistry. ‡ AGH-University of Science and Technology. § Åbo Akademi University, Process Chemistry Centre, c/o Center for Process Analytical Chemistry and Sensor Technology (ProSens).

Potentiometric ion sensors or ion-selective electrodes (ISEs) are an important subgroup of electrochemical sensors.1-3 ISEs are characterized by small size, portability, low-energy consumption, and low cost, which are attractive features concerning practical applications. ISEs based on polymeric membranes containing neutral or charged carriers (ionophores) are available for the determination of a large number of inorganic and organic ions, as described in detail, about a decade ago, in extensive reviews.4,5 However, during the past decade, the chemical sensing abilities of ISEs have been improved to such an extent that it has resulted in a “new wave of ion-selective electrodes”.6,7 This can be attributed to several important findings, such as the considerable improvement in the lower detection limit of ISEs, new membrane materials, new sensing concepts, and deeper theoretical understanding and modeling of the potentiometric response of ISEs. The aim of this review is to highlight some of these modern topics in the field of potentiometric ion sensors. This review is focused on recent achievements since the beginning of this millenium and emphasizes the results from the last 5 years (2002-2006). Section 2 gives a critical overview, placed in a historical perspective, on the theory of the potentiometric response, including classical equilibrium models as well as advanced nonequilibrium models. Section 3 deals with recent advances in the field of solidcontact ISEs, emphasizing the application of conducting polymers as ion-to-electron transducers. Recent developments in the area of miniaturized ISEs, including aplications in flow analysis, life science, and biology, are discussed in section 4. Finally, the new wave of ISEs is commented on in section 5. We hope that the issues discussed will illustrate the great possibilities offered by modern ISEs and encourage further innovations in the rapidly expanding field of chemical sensors in the years to come.

2. Theory of Potentiometric Response The response of potentiometric ion sensors, i.e., ionselective electrodes (ISEs) and/or ion-sensitive sensors (ISSs) (i.e., sensors with solid-state contact made of, e.g., conducting polymer film), is a complex time-dependent phenomenon that depends on the electroactive material (membrane/film) and the bathing solution as well as the membrane|solution interface and their composition, thermodynamic, and kinetic properties. All these features are the subjects of the theoretical modeling of the response. Modeling in the ion-sensors area serves two roles.8 One classical role is in supporting the practitioner (the user of sensors) with very basic principles

10.1021/cr068100w CCC: $71.00 © 2008 American Chemical Society Published on Web 01/12/2008

330 Chemical Reviews, 2008, Vol. 108, No. 2

Johan Bobacka was born in 1962 in Larsmo, a small village on the west coast of Finland. He received his M.Sc. degree in chemical engineering in 1988 and his Ph.D. degree in analytical chemistry in 1993 from Åbo Akademi University. The Ph.D. work on “Electrochemical characteristics of poly(3-octylthiophene) film electrodes” was performed under the supervision of Professor Ari Ivaska and Professor Andrzej Lewenstam. During the years 1994−1997, Johan Bobacka worked as a research scientist at Kone Instruments Corporation (now Thermo Fisher Scientific), and since 1997, he is a Docent in electroanalytical chemistry at Åbo Akademi University. He was a senior researcher (1997−2001) and Academy of Finland Research Fellow (2001−2006) at the Laboratory of Analytical Chemistry, which is part of the Åbo Akademi Process Chemistry Centre (Finnish Centre of Excellence 2000−2011). At present, Johan Bobacka is Professor of Analytical Chemistry at Åbo Akademi University (2006−2009). His current fields of interest include electroanalytical chemistry, process analytical chemistry, chemical sensors, conducting polymers, and electrochemical impedance spectroscopy, with special emphasis on the development of potentiometric ion sensors based on conducting polymers.

Bobacka et al.

Andrzej Lewenstam obtained his Ph.D. (1977, supervisor prof. A. Hulanicki) and his DSc (1987) from Warsaw University and since 1990 he has been a Professor. Currently he is Professor of Sensor Technology at Åbo Akademi University and Director of the Center for Process Analytical Chemistry and Sensor Technology ‘ProSens’ at this University, Professor in Chemistry, AGH University of Science and Technology, Cracow, and Adjunct Professor in Chemical Engineering, Laval University, Quebec, Canada. Since 2006, he is a chairman of the Working Group on Selective Electrodes and Biosensors of the International Federation of Clinical Chemistry (IFCC). His current fields of interest are chemical sensors and biosensors, clinical chemistry, membrane potential and biomimetics, conducting polymers, hydrometallurgy, and mathematical modelling, as well as methodology of chemistry and philosophy of Science. He is the author of approximately 200 papers in chemistry and philosophy of Science and 20 patents and is a member of the editorial board of 4 international journals.

Figure 1. Methodology used in models of potentiometric ionsensors response.

Ari Ivaska was born in 1946 in Kuopio, Finland. He received his M.Sc. degree in chemical engineering in 1971 and his Ph.D. in analytical chemistry in 1975 at Åbo Akademi University, Finland. The Ph.D. thesis “Potentiometric titration of weak acids and their binary mixtures” was supervised by Professor Erkki Wa¨nninen. Ari Ivaska did his post doc in 1978/79 at Chelsea College, University of London, England, and 1982/ 83 at Northwestern University Evanston, Illinois, U.S.A. He also worked as UNESCO expert at UNICAMP in Brazil, 1980/81. In 1985/86, he worked at the research centre of Neste Company in Finland and was nominated to the Chair of Analytical Chemistry at Åbo Akademi University in 1987. He has been as visiting Professor at University of Washington, Seattle, U.S.A., in 1991/92 and 1996/97 and the spring term 2003 at University of Wollongon, Australia. He is the head of the Process Analytical Group at the Åbo Akademi Process Chemistry Centre nominated to the Centre of Excellence in research by the Academy of Finland for 2000−2011. Ari Ivaska is currently the director of the Research Institute of the Åbo Akademi University Foundation. His fields of interest are electroanalytical chemistry, process analytical chemistry, electroactive materials in general, and metal ions in the paper and pulp chemistry.

of sensor response to help the application and to support quantitative measurement by simple equations. Another adVanced role is to provide a fundamental understanding of the sensor response for those interested in electrochemical

theory as the forefront of sensor technology, by helping to map electric potential and concentration changes in space and time. Related to these two roles are a few levels of modeling generality and/or idealization, as schematically shown in Figure 1. Classical models are more idealized to (intentionally) avoid mathematical, numerical, and computational difficulties stemming from solving nonlinear equations, inherent to advanced models. Classical models are easier to comprehend and to be presented and are the subject of many papers.1-7 However, the use of advanced models is the only way to achieve a fundamental understanding (often nonintuitive) of a sensor response. The main reason is that the classical models disregard migration and, therefore, do not provide adequate space and time-dependent characteristics of sensor response, whereas the advanced models do, as shown in Figure 2. Summarized below are the recent advances in the present quantitative theory of potentiometric ion sensors in contrast to more classical approaches. In all the models considered here, the potentiometric ion sensor is represented by the following scheme: sample|ion-sensitive membrane/ film|internal contact (e.g., solution, gel, solid contact). In all the models discussed, it is assumed that the potentiometric response is modeled under open-circuit conditions (under the zero-current condition). Furthermore, the sensor is made of separate, homogeneous, ionically conducting phases that form well-defined, flat interfaces; the

Potentiometric Ion Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 331

glass, and plastic), they also apply to the membranes where complexation/association processes should be taken into account, e.g., the neutral carrier-based membranes.

2.1. Total-Equilibrium Models (Classical Models) In all the classical models, the electric potential (EM) of an ion sensor is represented by the sum of a boundary potential (EPB) at the sample|ion-sensitive membrane (film) boundary (PB) and by the diffusion potential inside the membrane or film (ED). The constant added includes the potential at the internal interfaces (internal contact).

EM ) EPB + ED + constant

(1)

(For simplicity in the equations below, it is arbitrarily assumed that constant ) 0.) In the classical modeling of the response of ion sensors, two possibilities are consideredsone recognizes diffusion potential, while the other disregards it. These two approaches are presented below.

2.1.1. Phase-Boundary-Potential Approach

Figure 2. Schematic presentation of differences between totalequilibrium and advanced models: (a) concentration profiles and (b) electrical potential profiles.

interfaces are unblocked for ionic charge-transfer processes (faradaic currents), which applies to both ISEs and ISSs (see section 3.5); also, the sensor’s phases are characterized by standard chemical potentials of the components and ionic mobilities invariant in space and time. It is also assumed that the only driving forces for ion-fluxes are gradients in ion concentrations and in electric potentials (the gradients perpendicular to the sensor surfaces (1D formulation)) and that the pressure and temperature in the modeled system are constant and solvent flow (osmotic effects) is ignored. It should be mentioned that, although the models presented here refer to pure ion-exchanger membranes (solid state,

The phase-boundary-potential model is based on two idealizing assumptions: (1) The phase-boundary potential at the sample|membrane interface (phase boundary) governs the membrane response, i.e., EM ) EPB. Migration effects in the membrane are ignored, which means that the kinetic parameters of all the charged species involved, i.e., all ionic mobilities, are equal. Consequently, the diffusion potential is ignored, which formally means that ED ) 0, i.e., the electroneutrality in the membrane, except of the boundary, is assumed (the electroneutrality assumption). [On some occasions, the diffusion potential is assumed to be ED * 0 ) const (the pseudoelectroneutrality assumption) and, in this way, is disregarded.] (2) Electrochemical equilibrium is assumed at the sample|membrane interface; difference in chemical potential for any ion able to transfer the interface is balanced by a difference of the inner electrical potentials EM (EM is called the equilibrium potential). Additionally, it is assumed that the electric potentials and the concentrations of ions in the phases in contact are independent of the distance (except of the phase boundaries) and of time; there are no ion concentration drops in the respective phases over distance (the total-equilibrium assumption). The two assumptions specified above provide grounds for implementing Guggenheim’s concept of the electrochemical potential, µ˜ i,9

µ˜ i ) µi + ziFφ ) µ0i + RT ln(ai) + ziFφ

(2)

where µi is the chemical potential in the phase (µ0i under standard conditions), zi is the ion valency, ai is the single free ion activity, φ is the electric (inner) potential in the phase, and R, T, and F are the universal gas constant, the absolute temperature, and the Faraday constant, respectively. Using further idealizing assumptions, namely, that only an ion “i” can transfer through the interface (the ideal permand ion-selectiVity assumptions); the ion transfer is fast and reversible (the infinite kinetics assumption); the phases in

332 Chemical Reviews, 2008, Vol. 108, No. 2

Bobacka et al.

contact are not only of distinctly different chemical properties but are immiscible (the ideal immiscibility assumption); the convention for a single ion activity instead of a mean ionic activity (e.g., by using Debye-Hu¨ckel theory and convention) is adopted and it is assumed that the single ion activity (ai) in each phase is equal to its concentration (ci) (the ideal phase assumption); and there is no flux of solvent through the membrane/film (the solVent impermeability assumption), it is possible to employ Guggenheim’s concept to derive EPB as a function of ci. By using eq 2 for each phase, for the condition of electrochemical equilibrium between the phases (µ˜ i ) µ˜ i) and by recognizing that they are chemically distinctively different (µ0i * µj 0i ), after separation of electrical and chemical terms, one immediately arrives at a Nernstian-like equation for the electrical potential difference at the ionsensor interface,

RT RT ci EM ) EPB ) ln ki + ln ziF ziF cji

(3)

where ci denotes the concentration of an ion i in the solutions (in mol/dm3) and the barred symbol denotes the membrane/ film phase, and ki is an ion-partition constant defined as ki ) exp{(µ0i - µ j 0i )/RT}, where µ0i and µj 0i are standard chemical potentials of ion i in the respective phases. The theoretical models of the ion-sensor responses exploiting Guggenheim’s PB concept started with the paper by Nikolskii.10 The author considered the contact of a pHsensitive glass membrane with the bathing solution containing the main ions (i ) H+) and the monovalent interfering ions (j ) Na+) and obtained the equation for ion-sensor response in the form

EM ) EPB )

RT RT ci RT ln ki + ln ) ln ki + ziF ziF cji ziF

( )

cicjj cji RT ci(cji + cjj) RT RT ln ln ki + ln ) ) ziF cji(cji + cjj) ziF ziF cji + cjj RT ci + Ki,jcj RT RT ln ki + ln ln(ci + Ki,jcj) ) const + ziF ziF cji + cjj ziF (4)

(

(

)

ci +

)

where Ki,j used in the derivation above is the equilibrium constant

Ki,j )

kj cicjj ) ki cjcji

(5)

for the ion-exchange reaction,

i+ + j + a i + + j +

(6)

and cji + cjj ) const (the bar means the ions are in the membrane phase). Nikolskii was aware that including potentials arising from the different mobilities of ions in the membrane is a complex mathematical task, which was completed three decades later by Eisenman.11

He considered different membrane types, i.e., solid-state, glass, and membranes containing ion-exchangers and neutral ligands. He considered cases with and without association between ions in the membrane. Eisenman’s modeling made under classical assumptions of a total equilibrium yielded, for most cases considered, the equations in the form first obtained by Nikolskii. For instance, the equation for the fully dissociated ion-exchanger case was derived in the form

EM ) EPB + ED )

( )(

) )]

ujj RT RT ci RT ln ki + ln + ln jci + cjj ) ziF ziF cji ziF uji ujj ci + Ki,jcj RT RT ln ki + ln cji + cjj ) ziF ziF cji + cjj uji ujj RT ln ci + Ki,jcj (7) const′ + ziF uji

[(

(

)

where uji and ujj are the ion mobilities for i+ and j+ ions, respectively. If uji ) ujj, then eq 7 is identical to eq 4. As is shown above, both the Nikolskii and the Eisenman models provided strict analytical derivations of equations, possible for equal charges of the main and interfering ions (e.g., 1:1, 2:2). These models were later applied on many occasions to describe the response all kinds of ion sensors (for review, see refs 4 and 5). Analytical derivation of the above equations for unequal charges (zi * zj) in the frame of the total-equilibrium approach is impossible.12-15 To cover such cases, a caseby-case approach for each nonequal charges is used and relatively complex formalisms with implicit methods of equation solving are employed. They are based on additional ad hoc assumptions, such as ignoring the changes in concentration of the main ion in the membrane. Some ad hoc formal stratagems are employed as well, e.g., using the same mathematical equation to bind two independent variables depicting the primary ion, one representing the primary ion activity in the sample without interference from the other sample ions and the other representing the primary ion activity in the mixed sample.4,5,16 In this situation, to cover the cases of unequal charges while keeping the totalequilibrium assumption valid, the semiempirical equations, in a similar form to eqs 4 and 7, were postulated, as shown below.

2.1.3. Semiempirical Equations The semiempirical equations reflect the emphasis on the practical applications of ion-selective membrane electrodes. As a result of the arbitral decision of IUPAC, it was postulated to merge and extend the previously mentioned eqs 4 or 7 in the form of a general equation for all ionselective electrodes and ion sensors, covering all charges of ions. This equation is known today as the NikolskiiEisenman (NE) equation (for simplicity given below for one interfering ion).17

EM ) const′ +

RT pot zi/zj log(ci + Ki,j cj ) ziF

(8)

2.1.2. Total-Membrane-Potential Approach Eisenman abrogated the idealizing assumption that ED ) 0 and extended Nikolskii’s eq 4 for ED * 0 in the membrane.

This equation can be further extended to include the low detection limit (L) of the ion sensor:18

Potentiometric Ion Sensors

EM ) const′ +

Chemical Reviews, 2008, Vol. 108, No. 2 333

RT pot zi/zj log(ci + Ki,j cj + L) ziF

(9)

pot Both parameters Ki,j and L are constitutive for analytical practice and theoretical modeling. Theoretical modeling attempts to help predict their values, and, if it occurs, to show the evolution of the response for different concentrations of main vs interfering ions, over a period of time.

2.1.4. Comments on Model Benefits and Drawbacks of Total-Equilibrium Models The predominant way of modeling under total-equilibrium PB models is to ignore the diffusion potential without any formal (mathematical) justification, despite the reports on its significance even under total equilibrium.15,19,20 A typical way of ignoring ED in the PB models is formulated verbally, not mathematically, e.g., by saying that “by the end of the 1980s, it started to become clear that subtle effects of ionic mobilities in ISE membranes may be typically ignored without a significant loss in accuracy”.21 What is overlooked in the argument just cited is the fact that ignoring diffusion potential means arbitrarily placing and affixing the membrane potential at the phase boundary only andsat the same times justifying an uncritical use of the boundary-potential concept. Interestingly, this practice remains in contrast with wellbalanced criticism of the phase-boundary approach by the inventors of this concept. Namely, according to Guggenheim, “the conception of splitting the electrochemical potential µ˜ i of an ion i into the sum of a chemical term µi and an electrical term ziFφ has no physical significance; for one can assign an arbitrary value to φ for some point in each medium and this will for the ions of each type i determine µi, so as to give µ˜ i, the value which determines all the physical processes involving ions of type i”.9 Nikolskii added to Guggenheim by saying that his model “involves some difficulties, for in this case one deals with thermodynamically undefined variables, interface potential differences, diffusion potentials, and the activities of single ions”.22 The advantage of the total-equilibrium PB models lies in their simplicity. Characteristic of the phase-boundary modeling is that it considers ion complexion/association equilibria in a membrane in a similar way to solution chemistry and to coupling the membrane chemistry with that of the bathing solution via a pivotal phase boundary, eq 3. This methodology provides similar flexibility, as known from the complex formation theory, in considering many ion-equilibria and complex formation constants, the mass conservation, and the charge balance (electroneutrality) equations. Therefore, the methodology used in the total-equilibrium modeling has value: delivering the basic theoretical support and formal instrumentation needed for the practical applications of sensors, while offering an intuitive but essential understanding of the principles underlying sensor response. For these reasons, this modeling has been used extensively to provide a simple semiquantitative description of most ionselective electrode related experiments and has assisted every period of development in the ion-sensor technology. The modeling was, and still is, applied in ion-selective electrodes4,5 and conducting polymer ion sensors, including the sensors with a solid contact.23-25 The major drawbacks of the total-equilibrium models relate directly to the idealizations used; i.e., to the counterfactual assumption that all individual mobilities in the membrane are equal. This allows the migration effects (ED ) 0) to be

ignored and the assumption that (i) the electroneutrality condition is valid even in the proximity of the phase boundary, (ii) the concentrations of primary and interfering (and their complexes) as well as oppositely charged sites in the layers adjacent to the membrane/film surface are equal to those in the respective bulks, and (iii) the ion-transfer rates are infinite (see Figure 2). Primarily, assuming that the systems modeled are at total equilibrium leaves us with the sensor response interpretation that the electric potential, sensitivity, and, in particular, selectivity coefficients and detection limit are time-independent. The consideration in the total-equilibrium models that the sensor response is time-independent contradicts numerous time-dependent empirical reports in the field of potentiometric sensors, especially those delivered by the nonequilibrium potentiometry, such as the lowering of the detection limit by using transmembrane fluxes, which are timedependent.

2.2. Local-Equilibrium Models (Diffusion-Layer Models) In these models, called the diffusion-layer models (DLMs), the local equilibrium at the sensor interfaces is assumed (the local equilibrium assumption). (It means that eqs 3 and 7 as well as eqs 8 and 9 apply by substitution of the bulk concentration of ions by respective surface concentrations.) Additionally, it is assumed that the concentration of ions in the membrane phase and contacting phases are dependent on the distance but are independent of time, i.e., the sensor system is at steady state, or are dependent on time by diffusion of ions to/from membrane|solution interface(s) controlling the equilibration rate. The need to extend the time-independent thermodynamic modeling as described above was already noticed some decades ago, owing to the observations of variable and timedependent selectivity in the case of ion sensors with solidstate and plastic membranes.26-28 The problem of selectivity changes with time is still an issue of significant interest in the area of bio-ion-sensitive membranes/films.29 Sokalski et al.’s recent discovery of lowering the detection limit due to transmembrane fluxes gave an impetus to the consideration of ion fluxes30 and resulting concentration gradients in the theoretical modeling using the DLM frame. The diffusion-layer model (DLM) was first introduced by Lewenstam,31 and was continued in a number of papers,28,31-37 to model changes of selectivity coefficients with concentrations and time. This model is based on the assumption of local equilibrium at the solution|membrane interface, and consequently, a starting point in DLM are eqs 3 and 4 with surface (local) concentrations instead of bulk concentrations. The model assumes steady-state ion fluxes given by linear concentration gradients between the interface(s) and respective bulks, and it assumes constant and time-independent diffusion layers in the “local” areas. In DLM, the time-dependent response (for ions of equal charges and an ignored detection limit) is obtained by using a pivotal parameter “s(t)” called the surface coverage or site filling factor, characterizing the distance of the system under local equilibrium from total equilibrium over time (t), which is defined as29,33,35-37

s(t) )

cj0(t) ci0(t) + cj0(t)

)

Ki,jcj0(t) ci0(t) + Ki,jcj0(t)

(10)

334 Chemical Reviews, 2008, Vol. 108, No. 2

Bobacka et al.

Here, the barred concentrations refer to ion concentrations in the membrane or film surface at time t, and the sum ci0(t) + cj0(t) is equal to cib, where cib is the concentration of the main ion in the bulk of the membrane. The concentrations without bars ci0(t) and cj0(t) are the solution ion concentrations on the membrane surface at time t. Assuming a constant diffusion-layer thickness with linear concentration changes in the diffusion layers (the Nernst approach) for any response time, and by coupling the ion fluxes (Ji,Jj) from/to the interface by virtue of mass conservation, one obtains

Jj )

ntot ds(t) cj - cj0(t) ci0(t) - ci ) Dj ) Di ) -Ji (11) 3 A dt 10 δ 103δ

where A is the electrode surface area (m2), ntot is the number of active exchangeable sites occupied by the i and j ions (mol) on the membrane surface, and δ is the diffusion-layer thickness (m). By combining eqs 5, 7, and 10 rewritten in local concentrations and eq 11, one further obtains

EM(t) ) const +

(

)

uj [1 - s(t)]Ki,j + s(t) Ki,j c + Dj c i ui Di j RT ln (12) Dj F c ib Ki,j[1 - s(t)] + s(t) Di where the const is the term including contributions independent of the concentration of the ions i and j; Di and Dj are the diffusion coefficients of the primary ion and interfering ion in the aqueous diffusion layer (m2 s-l), respectively; and uji and ujj represent the ionic mobilities of ions in the membrane phase (m2 s-1 V-1). After separation of the variables s(t) and t and integrating within time limits, t to t )+∞, and corresponding s limits, s(t) and seq, it is possible to obtain a function binding s(t) and t,

Dj ‚c Dj Di j s(t) Ki,j s(t) - Ki,j ln 1 ) Di ci + Ki,jcj seq (ci + Ki,jcj)Ct (13)

[

]

ci +

(

)

with C ) DjA/ntotδ × 10-3 and where t is a time (s), 10-3 is a unit conversion factor (dm3 to m3), and seq is given by

seq )

cjj0(t f ∞) cji0(t f ∞) + cjj0(t f ∞)

)

Ki,jcj ci + Ki,jcj

(14)

where cji0(t f ∞) and cjj0(t f ∞) represent the ion concentration in the membrane phase at the interface at steady state and/or at total equilibrium (for both cases, ds(t)/dt ) 0). In this way, in the DLM, potential is connected to time. The DLM, in contrast to total-equilibrium models, contains time as a model parameter. The time is introduced ad hoc to describe attaining of total equilibrium (i.e., equilibration) via diffusion-controlled ion transport. The DLM predicts that pot changes during the equilithe selectivity coefficient Ki,j bration process (i.e., in the measurement) as a function of s(t) and, thus, with time (t).33-37 According to DLM, the value

pot of the selectivity coefficient Ki,j can vary between two limiting valuessone for a short response time dictated by the ion-transport properties in solution,

pot Ki,j (t f 0) )

Dj ≈1 Di

(15)

and the other at steady state and/or total equilibrium, where pot Ki,j is characterized exclusively by the membrane-related parameters and expressions known from total-equilibrium models, eqs 4 and 7. In the DLM, eqs 4 and 7 are limiting cases for t f ∞: pot (t f ∞) ) Ki,j (for ED ) 0) Ki,j

ujj pot or Ki,j (t f ∞) ) Ki,j (for ED * 0) (16) uji

The above prediction given by eq 15 is of great practical importance in applications of ion sensors because it predicts that the sensor response for short readout times is characterized not by the equilibrium selectivity as given by eq 16 but by ion transport and the selectivity given by eq 15. In consequence, if it occurs, the electrode senses “equally” the main and interfering ions. This prediction resulted in important practical benefits that are also of interest today. It was, for instance, used to kinetically discriminate strong interferences by short readout times (t f 0)38,39 or, alternatively, to benefit from the measurements of strong interferents, such as heparin on chloride ISE. The latter is realized pot ≈ 1 and using the response by taking advantage of KCl,heparin to the interfering (heparin) translated “1:1” into the signal of the main (chloride) ions.40,41 DLM was also used to interpret nonmonotonic transients in ion-sensor response36 and for interpretation of long-term drifts in sensors in which a thin aqueous layer is formed between the membrane and the substrate electrode.42 DLM was used successfully to demonstrate that the poor apparent selectivities and detection limits have a common origin in the increased (vs bulk) surface concentrations of the main ion.43-46 In consequence, it was shown for the first time that, using the ion sensors in the regime of nonequilibrium response, induced as a result of the ion-complexation processes, both true (unbiased) selectivity coefficients44,46 and much lower detection limits for solid-state membranes can be achieved.45 In 1999, Sokalski et al.47 used the DLM frame to interpret the effect of lowering of the detection limit for plastic membranes by analyzing transmembrane ion fluxes in a symmetric solution/membrane/(internal) solution system. In this formulation of DLM, the diffusion potential is ignored, the fluxes of ions are treated under steady state, and the concentrations of all species are assumed to change linearly within the diffusion layers, i.e., in the adjacent solution layer and over the membrane. Ion-exchange as well as coextraction processes are considered to analyze the low detection limits. The model in this form provides a possibility to analytically find an equation for the detection limit (L) vs different model parameters, especially the concentrations of ions in the internal solution and membrane, but not vs time. An important outcome of this interpretation is reflected in the equation that allows the calculation of the steady-state surface concentration of the main ion, which dictates the low detection limit for (plastic) membranes:

Potentiometric Ion Sensors

ci0(t f ∞) ) ci +

D h iδ [cj (t f ∞) - cjid(t f ∞)] Did i0

Chemical Reviews, 2008, Vol. 108, No. 2 335

(17)

The above equation indicates that adjusting the composition of the internal solution (i.e., by keeping low free main ion concentration due to ion complexation and optimizing vs the membrane selectivity, the interfering ion concentration in the internal solution or the solid-contact film), the membrane thickness (d), the diffusion coefficient of in the membrane (D h i), the thickness of the diffusion layer in the sample (δ), and the diffusion coefficient in the sample (Di) plays an essential role in forming an ISS response in the concentration range close to the low detection limit, as shown schematically in Figure 3. Predictions from this version of DLM are better visualized by the graphical representations of the ISS responses instead of using rather complicated algebraic equations (see Figure 4). In fact, the application of a simple commercial equation solver (e.g., Mathcad) and its graphical tools proved to be quite feasible.47 The observation of Sokalski et al. on the lowering of the ISS detection limit,30 and its mathematical modeling by DLM, heralded a new chapter in exploring the power of potentiometric sensors under local equilibrium, as proved by a number of papers (for recent review, see ref 48).

Figure 3. Schematic presentation of the concentration profile of the main ion influencing the lower detection limit.

2.2.1. Comments on Model Benefits and Drawbacks of DLM Models The DLM allows the sensor response to be theoretically modeled in a way that was not possible for total-equilibrium models; it allows modeling of the variation in the selectivity vs time and the effect of lowering the detection limit under steady state to be interpreted. From the theoretical point of view, the advantages of DLM-type modeling do not compensate for the drawbacks already known in the total-equilibrium models, as explained above. However, by extending the power of the totalequilibrium concepts, DLM, in a relatively simple way, can truly support practical advances in nonequilibrium potentiometry, since it exploits close-to-the membrane and transmembrane fluxes for comprehensive analytical benefits. One possible application of DLM is by assuming the invalidity of steady state for concentration (∂ci(x,t)/∂t * 0) and employment of the second Fick’s law for ions treated as neutral species, as shown very recently.49 The main problem of the nonequilibrium potentiometry is in finding a “niche of stability” for the reproducible analytical readouts in a variety of matrixes and in time. The models presented so far are overidealized and not able to answer the basic questions, especially important in the case of lowering the detection limit: what is the role of the membrane, its thickness and dielectric constant, in shaping the sensor response? Similarly, how are other important questions to be answered, such as the following: How do the different diffusibilities of ions and ion-exchange rates influence the signal? When can the subtle effects of ionic mobilities in ISE membranes be ignored without a significant loss in accuracy? What are the bulk-to-bulk concentration and potential changes over time? Finally, how can the signal be stabilized if the sensor works in a nonequilibrium regime in different matrixes? All these questions call for a new “upper floor” approach in theoretical modeling, namely, for the advanced nonequilibrium models, which are described below.

Figure 4. Calculated EMF functions for a series of ISEs having the same membrane but different primary ion concentrations, ci in the internal solution, from 1 to 10-15 M in the inner solution. Curves are labeled with the corresponding negative logarithm.47

2.3. Advanced Nonequilibrium Models A fundamental difference between the advanced nonequilibrium models and the (total or local) equilibrium models is in abandoning of two constitutive conditions, the electroneutrality and (total or local) equilibrium condition. It is possible to do so, owing to the admission of charge separation from the Poisson equation and the finite ion-transfer rates described by the appropriate heterogeneous ion-transfer rate constants. Modeling of the nonequilibrium potentiometric response of the ion-sensitive sensors requires employment of explicit space and time domains, which provide the platform for a set of relevant thermodynamic, extrathermodynamic, and kinetic data. This platform may be constructed by use of the Nernst-Planck and Poisson equation system (NPP)50 or by a similar system that is an appropriate and rich enough tool to encompass the ion-sensor response. The first implementation of the NPP to model nonequilibrium (non-steadystate) response of ion sensors was recently reported by Sokalski and Lewenstam51 and in the contributions that followed.52-54 The NPP equation system allows calculations of the electric potential difference and concentration profiles as a function of space and time for the conventional sensors with internal solution or for the sensors with solid internal contacts. In contrast to the total-equilibrium and diffusionlayer models, the NPP model does not require any arbitral splits of the membrane potential into phase boundaries and diffusion-potential terms (compare Figure 2). Additionally, it does not use idealizing assumptions of the total and/or local equilibrium and the electroneutrality conditions. The

336 Chemical Reviews, 2008, Vol. 108, No. 2

Bobacka et al.

NPP model applies for different membrane and film-based ISSs and for ions of every charge, without the need for a case-by-case approach as known from the previous models, described earlier.4,13 In the NPP model, multilayers and adjacent (diffusion) layers are admitted. The membrane and the bathing solution(s), or films in contact, may have a thickness and dielectric permittivity of choice. The membrane may contain unequally mobile and/or immobile charged and uncharged sites/ligands. Ions of any charge can be considered in the charge-transfer and transport processes. Different degrees of association between extracted ions and sites and ligands in the membrane are permissible. Furthermore, the gradient of the chemical potential of the solvent across the membrane and convective flows are allowed. In the present formulations of the NPP model, all activity coefficients in the membrane are assumed to be one, ion extraction is described by the first-order rate constants, and the 1D geometry is used. The core of the NPP model is that ion fluxes in space (x) and time (t) are described by the Nernst-Planck equation,

[

fi(x,t) ) -Di

∂ci(x,t) F - zici(x,t) E(x,t) ∂x RT

( )

]

(18)

where fi(x,t) is the flux of the ith ion, ci(x,t) is the concentration of the ith ion in space point x and time t, E(x,t) is the electric field, Di is the diffusion coefficient of the ith ion, and zi is the charge of the ith ion, as in the models discussed above. In order to solve eq 18, two additional equations are used to relate fi(x,t), ci(x,t), and E(x,t). The first choice is the law of mass conservation:

∂fi(x,t) ∂ci(x,t) )∂t ∂x

(19)

The second is the Poisson equation, rewritten for the total current density (I) as

∂E(x,t)

I(t) ) F‚∑ zi‚fi(x,t) + ‚ i

(20) ∂t

where  is the dielectric permittivity. In the calculations, the Chang-Jaffe boundary conditions are used,

fi0(t) ) B k i‚ci,bL - A k i‚ci0(t) fid(t) ) - B k i‚ci,bR + A k i‚cid(t)

(21)

where fi0, fid, ci0, and cid are the fluxes and concentrations at x ) 0 and x ) d (where d is membrane thickness), respectively; B ki and A ki are the forward and backward rate constants, respectively, and their ratio is a partition coefficient; and ci,bL and ci,bR are the concentrations in the bathing solutions on the left (L) and right side (R) of the membrane, respectively. The system of nonlinear partial differential equations 18, 19, and 20 is solved to obtain the resolution in space and time by the finite-difference51-54 or finite-element method.55 Some authors stress the complexity of these procedures.21 We do not share this opinion; to the contrary, the imple-

Figure 5. Time-dependent concentration profiles for site R- ([i+] ) 10-4, [j+] ) 10-3, D h i/D h j ) 0.5, Ki,j ) 0.1, and R h TOTAL ) 10-3). Curves a-g show profiles after the following: (a) 4 × 10-4, (b) 1.64, (c) 13.1, (d) 26.2, (e) 104.8, (f) 420, and (g) 13 440 s (steady state).50

mentation and access are simple when using appropriate commercial platforms.

2.3.1. Comments on the Benefits and Drawbacks of the Advanced Model The total-equilibrium and diffusion-layer models are merely special cases (concretizations) of the NPP model. In a strict formal (mathematical) sense, all that is predicted by the PBMs and DLMs can be obtained from the NPP model by deduction (as shown in Figure 11 and discussed below). Of course, in simple, theoretical cases and routine laboratory practice, the full use of NPP is not necessary. (While admitting this, it should be emphasized that the argument repeatedly given by some authors in favor of phase-boundary models (PBMs),21 namely, that, in the NPP, the knowledge of individual mobilities of relevant ions and their transfer rates at the phase boundary is required, is not in contrast to the phase-boundary models. Actually, in the PB models, as was shown above, even stronger assumptions are used, i.e., that the rate constants are infinite and all individual mobilities are equal.) However, for modeling of time-dependent processes, in cases of suspected nonlinearities, and for someone working with sensors under nonequilibrium, as happens in the case of fast readout times or in lowering the detection limit, the NPP approach or a similar theoretical tool is a must. The reason is simple: as the model from the “upper floor”, the NPP model provides hard numerical facts instead of verbal declarations and unproven intuitions, so often made by advocates of simple modeling. The NPP approach has already provided novel insights and numerical answers to problems in ISE practice that have been intriguing for decades and which the DLM and PB models were unable to describe: the concentration and potential profiles over equilibration time and even under steady state (or equilibrium) revealed striking nonlinearities (see Figure 5), shown for R-, which in the PB and DLM models is assumed to be not dependent on distance (x) and time (t). It was presented that the contribution of the socalled diffusion potential prior to equilibrium is significant and varies with time (Figure 6), and in consequence, prediction of NPP and PB models for the same conditions can be strikingly different (Figure 7): In fact, the overall linearity of the calibration curves depends on the distance from the steady state or equilibrium in the sensor system, and even under steady state may be significantly influenced by migrational effects (Figure 8). (For simplicity, these

Potentiometric Ion Sensors

Figure 6. Time-dependent and distance-dependent potential profiles, calculated as φ(x,t) ) ∫ E(x,t) dx. Total membrane potential b,R E(x,t) dx, where xb,L and xb,R are the points in the is EM(x,t) ) ∫xxb,L bulk of bathing sample solution (on the left, L) and internal solution/ film (on the right, R). Curves a-g show profiles after the following: (a) 4 × 10-4, (b) 1.64, (c) 13.1, (d) 26.2, (e) 104.8, (f) 420, and (g) 13 440 s (steady state).52

Figure 7. Comparison of the (a) steady-state NPP potential profile from Figure 6 with that presumed by (b) the phase-boundary model.52

effects are shown here schematically; for numerical details, see refs 52 and 53.) Furthermore, the NPP model shows unequivocally that the concentration profile of the species confined in the membrane phase (the oppositely charged sites) can dramatically change over the membrane distance along with highly nonlinear (and thus “nonintuitive”) profiles. For this reason, a strict electroneutrality condition, represented by the distance-independent charge-balance equation, used in PB models16,21 is expedient to allow binding of opposite charges in the membrane boundary but is clearly inadequate.53,55 Recent results from use of the NPP model allowed the analysis of concentration profiles with time for membranes bathed by strongly interfering ions (a chloride-sensitive membrane bathed with perchlorates),53 as illustrated in Figure 9, where the theoretical prediction and experimental results are compared. In the same work, selectivity changes as a function of time were examined, which had been inaccessible for the totalequilibrium model (PB model) and only partially accessible to the diffusion-layer model (DLM). Moreover, the access to spatial distributions of ions and the electric potential vs distance during equilibration time allows inspection of the underlying reasons for selectivity coefficient changes and their magnitudes.53,54 A similar methodology was used in the numerical analysis of the detection limit due to the transmembrane ion fluxes,

Chemical Reviews, 2008, Vol. 108, No. 2 337

Figure 8. Calibration curves in the presence of primary ([i+] ) (10-10 to 10-1 M)) and interfering ([j+] ) 10-3 M) ions calculated according to the NPP model (other data as in Figure 5). Pie charts express contributions of phase-boundary potential (EPB) (white) and inner membrane (diffusion) potential (Ed) (black) for the outermost calibration curves (Di/Dj ) 0.1 or 10) at [i+] ) 10-4 and [i+] ) 10-10 M, Ki,j ) 0.1 The curves correspond to the following diffusion coefficient rations (D h i/D h j): (a) 0.1, (b) 0.5, (c) 1, (d) 2, and (e) 10.52

to find the parameters important for its evolution and stabilization over a period of time. The NPP model allows numerical analysis of the important role of transmembrane ion fluxessas well as the appearance of the electric potential drops at one interface (sample|membrane) or at two interfaces (sample|membrane and membrane|internal contact interfaces)s in shaping the effect of the lowering of the detection limit.55,56 Consequently, for the first time, the significant influence of ion transport (diffusion), distribution and rate parameters, ion charges, dielectric constant, and thickness of the membrane are predicted. The NPP approach provides realistic concentration profiles as a function of time, which previously could only be addressed by the DLM, assuming linear concentration drops (compare Figure 3 and Figure 10a). In addition, the NPP model provides the profile of the electric potential in space and time (Figure 10b), which, according to the DLM (being a steady-state model), cannot be modeled. There is no doubt that enhanced sensitivity of measurements under nonequilibrium may be paid for by a decreased reliability of the results. Thus, an important and so far untouched question concerning the influence of the diffusion potential and its variation from sample-to-sample and over readout time, which may unfavorably and in an uncontrolled manner influence the precision of measurements with ISSs, can now be considered. Demonstrably stable and reproducible measurements have not yet been convincingly achieved. However, via the nonequilibrium model (NPP), a comprehensive analysis of this problem will soon be presented.56 The inevitable conclusion is that the NPP offers a novel tool for solving a number of, up until now, inaccessible problems in nonequilibrium potentiometry. The NPP is, by far, more general than the DLM and PB models, with the latter being special cases of the NPP, as shown in Figure 11. The relationship between these potentiometric models is characteristic of empirical sciences, for instance, relativistic mechanics and its special case, classical mechanics (even if we have ample confirmation from everyday routines in contrast with relativistic theory). For

338 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 9. Theoretical and experimental calibration curves for ionexchanger electrode chloride ISS.53 (a) Calibration curves calculated for membrane conditioned in i- and j- for solutions of (1) j- (5 min), (2) i- (5 min), (3) i- (15 min), (4) i- (30 min), and (5) i(60 min), and (6) calibration curve obtained with a membrane conditioned in i- for a solution of i- (60 min). i is the preferred ion, and j is the discriminated ion. The inset shows the EMF time dependence for an ISE conditioned in discriminated ion for 10-4, 10-3, 10-2, and 10-1 M, in order from top to bottom. (b) Calibration curves obtained with a membrane conditioned in Cl- for solutions of ()) Cl- (5 min/conc.), (0) ClO4- (5 min/conc.), (∆) ClO4- (15 min/conc.), (3) ClO4- (30 min/conc.). Calibration curve obtained with a membrane conditioned in ClO4- for a solution of (O) ClO4(60 min/conc). ClO4- is the preferred ion, and Cl- is the discriminated ion. The inset shows the EMF time dependence for an ISE conditioned in Cl- for 10-4, 10-3, 10-2, and 10-1 M in order from top to bottom.

this reason, the theoretical value of the NPP is unquestionable. To summarize, there is currently no doubt that, according to a hierarchy of the generality of the models discussed, the NPP approach provides clear added value, with regard to both nonequilibrium response and characteristics of the steady state of potentiometric ion sensors. The NPP also allows assessment of the migrational effects, finite kinetics, and permittivity of the membrane, none of which can be accessed via more idealized models because of their limited dictionaries and formalisms, as shown in Table 1. Additionally, Table 1 shows that, although the NPP itself has its own intrinsic idealizations and limitations and shares common idealizing assumptions and parameters with the DLM and NE, it is simply more powerful. This again reflects the process of the cumulation of knowledge, which is wellknown for empirical sciences.8 A new chapter in modeling on the level of the generality dictated by the NPP has now been breached, and questions insurmountable for the simpler DLM and PB models can now be answered. There is a list of problems concerning ISE/ISSs to be addressed, as well as many technical possibilities to be investigated. In 2D and 3D, modeling of

Bobacka et al.

fluxes through nanotubes and biological channels is still under development. Assumptions such as those listed above, i.e., equating activities and concentrations, uncoupled fluxes, complex-forming processes in the membrane, higher-order rate equations, influence of the electric field on ion-transfer rates, e.g., by extending the Chang-Jaffe boundary conditions in a form of Butler-Volmer conditions, must all be considered. (The latter condition admits the influence of the electric field on rate constants; although this is formally and programwise a simple extension as will be shown shortly,54 the problem of the ion-transfer mechanisms is still controversial and open.57-59) Furthermore, the variability of the dielectric constant along the membrane distance and at the interfaces needs to be considered, and finally, increasing the library of transport parameters in the membranes and films relevant to sensor technology needs to be recognized.60 These are just some examples of the interesting problems requiring further theoretical work on the assumed “level” of approximation and experimental verification.55,56 Present applications of ion sensors in the nonequilibrium regime, and the need for proper theoretical support, show that understanding of the response mechanism of potentiometric sensors is a new and open challenge. Without ignoring the achievements and power of the earlier classical or diffusion-layer models, a call for advanced modeling is nonetheless unquestionable.

3. Solid-Contact ISEs Elimination of the internal filling solution from conventional ISEs results in solid-contact ISEs (ISSs), which are more durable and easier to miniaturize than their conventional counterparts. However, in order to obtain solid-contact ISEs with stable electrode potential, it is necessary to have sufficiently fast and reversible ion-to-electron transduction in the solid state without any contribution from parasitic side reactions.61 Research and development of solid-contact ISEs had already started in the beginning of the 1970s with the invention of the coated-wire electrode (CWE), which indeed represents a simple and robust design.62 The main drawback of the CWE is obviously the poor potential stability resulting from the blocked interface that forms between the purely electronic conductor (metal) and the purely ionic conductor (ion-selective membrane). Solid-contact ISEs with improved potential stability have, therefore, been produced by utilizing electroactive materials showing mixed electronic and ionic conductivity that serve as ion-to-electron transducers between the electronic conductor and the ion-selective membrane.61 Among the electroactive materials available today, electroactive conjugated polymers (conducting polymers) have emerged as one of the most promising ion-to-electron transducers for solid-contact ISEs.63-66 Other approaches to solid-contact ISEs involve the use of Ag/AgCl,67 Ag/AgCl/hydrogel,68 redox-active self-assembled monolayers,69,70 Prussian Blue,71 carbon-based composites,72,73 Ag-based composites,74 and Ag/AgCl/porous carbon loaded with ionophore and plasticizer that resulted in a solidcontact Pb2+-ISE with an impressive lower limit of detection below 10 pM.75

3.1. Conducting Polymers The discovery and development of conjugated polymers that can be made electronically conducting by partial oxidation (p-doping) or reduction (n-doping) has had a great

Potentiometric Ion Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 339

Figure 10. Calculated by NPP, the main ion concentration (a) and the electrical potential changes in space and time (b).54 ciL ) 10-9, ciR ) 10-7, cjL ) 10-7, and cjR ) 10-1; cji ) cjR- ) 10-3 (mol/dm3); Di ) 10-9 and Dj ) 10-8 and D h i ) 10-11 and D h j ) 10-10 (m2/s); khi,l ) -7 kCi,l ) 1; khi,l ) 10 and kCi,l ) 1 (m/s), L and R denote the left and right sides of the membrane. Diffusion potential is calculated at 1 µm distance from the interfaces in the interior of the membrane.

impact on several fields of science and technology, which naturally goes far beyond the scope of this review.76-84 However, it should be mentioned here that conducting polymers have been applied in a large variety of chemical sensors and several reviews have been published on this topic.63-66,85-114 A few of these reviews focus entirely on the application of conducting polymers in potentiometric sensors.63-66,92,97,114 Although conducting polymers have been known in the field of potentiometric ion sensors since the 1980s, development is still continuing. Conducting polymers have some key features that are useful when applied as ion-to-electron transducers in solidcontact ISEs. Conducting polymers are electronically conducting materials that can form an ohmic contact to materials with a high work function, such as carbon, gold, and platinum, which ensures a proper electronic (ohmic) contact. Conducting polymers can be deposited on the electronic conductor by electropolymerization of the monomer or by solution-casting of the soluble conducting polymer, which gives some flexibility in the manufacturing process. Conducting polymers are electroactive materials with mixed

electronic and ionic conductivity, which means that they can transduce an ionic signal into an electronic one in the solid state. Furthermore, the properties of conducting polymers can be tailored via functionalization, e.g., by covalent bonding of side groups to the conjugated polymer backbone and by immobilization of functional doping ions. These are important features that make conducting polymers suitable as solid contacts in combination with conventional ionselective membranes.63 In this type of solid-contact ISE, where the conducting polymer is coated with a conventional ion-selective membrane, the ion-selectivity is determined mainly by the ionselective membrane, which allows the utilization of various ionophore-based polymeric membrane formulations that are available.4,5 Recent progress in the area of ISEs such as the lowering of the detection limit toward the picomolar level,30 enabling potentiometric trace-level analysis,6 gives an additional impetus for the development of solid-contact ISEs with improved analytical performance. Conducting polymerbased solid-contact materials are boosting “the new wave of ion-selective electrodes”.7

340 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 11. Interrelation between models in potentiometry. (*) The condition reads as follows: there is no change in electric field strength and concentration of ions over time in the system sample/ membrane/internal contact (solution or film). (**) The condition reads as follows: there is total equilibrium over distance in the system sample/membrane/internal contact (solution) or film, i.e., electrochemical potential of each special along x is constant; additionally, electrical potential over distance in each phase, and in consequent concentrations of each species, are contact except the phase boundaries.

3.2. Conducting Polymers as Solid Contact Over the past few years, the analytical performance of ISEs where conducting polymers are used as ion-to-electron transducers (solid-contact ISEs) has been dramatically improved. Solid-contact ISEs for determination of both inorganic and organic ions were constructed by using conducting polymers as the ion-to-electron transducers. Of particular interest is the lowering of the detection limit down to the nanomolar level. The conducting polymer materials were based mainly on electropolymerized or chemically polymerized derivatives of pyrrole (Figure 12), thiophene (Figure 13), and aniline (Figure 14).

3.2.1. Polypyrroles Polypyrrole was already used as an ion-to-electron transducer in solid-contact ISEs in the beginning of the 1990s115 and is frequently still used for the same purpose today. Recent developments were focused on improving the fabrication techniques and the analytical performance of such solid-contact ISEs. Polypyrrole was found to be useful as a solid contact in planar ISEs.116,117 A critical comparison of conducting polymer- and hydrogel-based solid contacts in K+-ISEs showed that polypyrrole doped with potassium hexacyanoferrate results in solid-contact K+-ISEs with better long-term potential stability than those based on the hydrogel contact.118 Polypyrrole doped with Tiron was used as a solid contact in Ca2+-ISEs.119-121 Since Tiron complexes Ca2+, the detection limit of such ISEs was found to be as low as 10-9 M. Similar detection limits were obtained for a Pb2+-ISE using polypyrrole doped with hexacyanoferrate as a solid contact

Bobacka et al.

when the electrode was used in a flow-through cell.122 The detection limit of Cl--ISEs based on polypyrrole as the solid contact could be lowered by more than 3 orders of magnitude when applying an anodic current that compensated the Clleakage from the ion-selective membrane (due to selfdischarge of polypyrrole).123 The influence of spontaneous charge-transfer processes of polypyrrole on the linear response range and selectivity of Ca2+-ISEs using polypyrrole as a solid contact was studied.124 Galvanostatic polarization of solid-contact ISEs was found to offer some control of the ion flux through the electrode and, consequently, could be used to lower the detection limit.125,126 Interestingly, K+-ISEs based on methacrylic/acrylic membranes in combination with polypyrrole as the solid contact resulted in K+-ISEs with both improved detection limit (below 10-6 M) and excellent stability.127 These results show that polypyrrole-based solidcontact ISEs with polymeric ion-selective membranes are promising also for low-concentration measurements. Polypyrrole doped with tetraphenylborate was used as a solid contact in K+-ISEs in order to have well-defined ion transfer between the conducting polymer and the ionselective plasticized poly(vinyl chloride) (PVC)-based membrane.128 Polypyrrole doped with different anions was also employed as a solid contact in pH electrodes based on polymer membranes containing tertiary amine ionophores.129 The physicochemical properties of polypyrrole can be significantly influenced by the doping ion, which offers possibilities of improving, e.g., adhesion of polypyrrole to the ion-selective membrane. This is well-illustrated by the use of polypyrrole doped with cobalt bis(dicarbollide) ions as the solid contact in pH-, Cu2+-, and K+-selective microelectrodes.130-132 Furthermore, a composite of polypyrrole and Nafion was applied as the solid contact in pH electrodes based on glass membranes.133,134 Polypyrrole was used as an ion-to-electron transducer also in solid-contact ISEs for determination of oxytetracycline hydrochloride and methacycline hydrochloride, which are antibiotics belonging to the tetracycline family.135,136 Poly(1-hexyl-3,4-dimethylpyrrole), which is soluble in tetrahydrofuran (THF), was used as a solid contact in carbonate-selective ISEs based on a silicone rubber membrane.137,138 Solution-processable conducting polymers offer some additional flexibility in the electrode manufacturing process, when compared to electropolymerization.

3.2.2. Polythiophenes Poly(3-octylthiophene) (POT) was the first one of the polythiophenes to be used as a solid contact in ISEs.139 More recently, solution-cast films of POT on screen-printed gold substrates and on platinum (silicon-based substrates) were evaluated as solid contacts in miniature Cl--ISEs.140,141 Improved analytical performance was obtained by using an additional adhesive layer (3-aminopropyltriethoxysilane) between the screen-printed gold electrode and the POT film, while there was no significant difference between PVC and polyurethane (PUR, Tecoflex) used as the ion-selective membrane matrices.140,141 Pb2+-ISEs were constructed by using solution-cast POT as the solid contact to a Pb2+-selective membrane based on poly(methylmethacrylate)/poly(decylmethacrylate) (MMA/ DMA).142 POT is highly lipophilic, which helped to prevent the formation of an internal water layer between the solid contact and the ion-selective membrane. Interestingly, the solid-contact Pb2+-ISE showed a much faster response at low

Potentiometric Ion Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 341

Table 1. Principles of the Models in Potentiometry model advanced models here, Nernst-Planck-Poisson (NPP)

parameters used (model dictionary) EM(x,t) ci(x,t) cji(x,t) ki B ki and A (including Ki,j) x (including δ, d) Di, uji  t

main assumptions

main benefits of model application in potentiometry (current state)

(1) general assumptions (see p 5); zero-current, no pressure and temperature gradients, no solvent flow, no convection (2) membrane homogeneous and flat, ionically conducting (3) one-dimension (1D) (4) infinite dilution/ideal phase (activities ) concentrations) (5) sharp border between phases (6) permittivity ) const over distance in each phasea (7) no flux couplinga

time- and space-dependent electrode response, selectivity, and low detection limit, access to membrane physicochemical properties (e.g., migrational effects, permittivity), finite charge-transfer rates

local-equilibrium models here, EM(t f ∞) ci0, cji0, diffusion-layer model (DLM) Ki,j δ, d Di, uji t (?)

(1) as in NPP (1-5) (2) electroneutrality (3) infinite rate constants (4) local equilibrium (5) steady state (limited access to time and space domains) (6) linear concentration changes over distance; electrical potential changes only at the phase boundaries

electrode response, selectivity, and low detection limit under steady state

total-equilibrium models here, Nikolskii-Eisenman (NE)

(1) as in DLM (1-3) time-independent electrode response, (2) total equilibrium selectivity, and low detection limit; (3) constant concentrations and electrical potentials access to chemical binding in each phase except of phase boundaries processes in the membrane

a

EM ci, cji Ki,j ui (?)

6 and 7 specific for NPP.

Figure 12. Pyrrole-based monomer units of polymers applied as ion-to-electron transducers in solid-state ISEs over the past few years: (1) pyrrole, (2) N-methylpyrrole, and (3) 1-hexyl-3,4dimethylpyrrole.

Figure 14. Aniline-based monomer units of polymers applied as ion-to-electron transducers and/or in solid-state ion-selective electrodes over the past few years: (8) aniline, (9) R-naphthylamine, (10) o-aminophenol, (11) o-phenylenediamine, (12) N-phenylglycine, and (13) o-anisidine.

Figure 13. Thiophene-based monomer units of polymers applied as ion-to-electron transducers in solid-state ISEs over the past few years: (4) 3-methylthiophene, (5) 3-octylthiophene, (6) 3,4ethylenedioxythiophene, and (7) 3,4-dioctyloxythiophene.

concentrations and even a slightly better detection limit (10-9.3 M) compared to the corresponding liquid-contact ISE. This has resulted in renewed interest in POT as a solidcontact material in ISEs. The use of solution-cast films of undoped POT as a solid contact together with plasticizerfree acrylate-based ion-selective membranes resulted in a number of solid-contact ISEs (Ag+, Pb2+, Ca2+, K+, I-) with detection limits close to the nanomolar (10-9 M) range.143,144 Solution-cast POT was found to be a very suitable solid-

contact material also for Ca2+-ISEs utilizing conventional PVC-based ion-selective membranes, showing, in addition, a low detection limit.145 However, the long-term potential stability of solid-contact NO3--ISEs based on POT as a solid contact was found to be somewhat inferior to conventional NO3--ISEs with internal filling solution.146 The potential stability of solid-contact K+-ISEs was found to correlate well with the bulk redox capacitance of the conducting polymer, as shown by using poly(3,4-ethylenedioxythiophene) (PEDOT) as a solid-contact material.147 This can be understood when considering that the transduction of an ionic signal into an electronic one via the redox reaction of a conducting polymer results in charging/discharging (doping/undoping) of the conducting polymer layer. From this point of view, chronopotentiometry was found to be a convenient and fast experimental method to evaluate the potential stability of solid-contact ISEs.147

342 Chemical Reviews, 2008, Vol. 108, No. 2

PEDOT is highly electroactive and shows good environmental stability in its oxidized (p-doped) form. Consequently, PEDOT has been studied rather intensively as a solid-contact material in recent years. PEDOT doped with poly(styrene sulfonate) (PSS), i.e., PEDOT(PSS), was applied as an ionto-electron transducer in solid-contact ISEs for various ions, including K+,147,148 Ag+,149,150 Na+,151 Cs+,151 Ca2+,152 and some aromatic cations (N-methylpyridinium, bupivacaine).153,154 The potential of solid-contact K+-ISEs based on PEDOT as a solid contact was found to be less sensitive to O2 and CO2 (pH) compared to those based on polypyrrole.147,148 A smallvolume radial flow cell for solid-contact ISEs using PEDOT(PSS) as a solid contact was also described.155 The solid contacts were prepared by electropolymerization of the 3,4ethylenedioxythiophene monomer and by solution casting of the commercially available aqueous dispersion of PEDOT(PSS) (Baytron P). Solution-cast films of PEDOT(PSS) (Baytron P) were applied as the solid contact also to screenprinted gold substrates.106 The water-solubility of the solution-cast PEDOT(PSS) film was decreased by ionic crosslinking with multicharged cations (Mg2+, Ca2+, Fe2+/3+, or Ru(NH3)62+/3+) before application of the plasticized PVCbased K+-selective membrane.156 Recently, PEDOT was applied as the solid contact in newly designed microcavitybased solid-contact ion-selective microelectrodes.157 PEDOT doped with hexacyanoferrate was used as the solid contact in Cu2+-ISEs.158 The improvement in the detection limit was attributed to spontaneous accumulation of Cu2+ in the solid contact, causing an influx of Cu2+ ions at the ionselective membrane/solution interface.158 The detection limit and stability of Pb2+-ISEs using PEDOT(PSS) as a solid contact was found to be influenced by the presence of interfering ions in the conducting polymer layer.159 Furthermore, by using soluble poly(3,4-dioctyloxythiophene) (PDOT), which is more lipophilic than PEDOT, it was possible to show the importance of the ion content of PDOT when used as a solid-contact layer.160 All-plastic disposable Ca2+-ISEs and K+-ISEs were prepared via solution-casting of PEDOT/PSS (Baytron P) and plasticized PVC-based membranes on plastic substrates.161 Here, the PEDOT/PSS worked both as the ion-to-electron transducer and as the electronic contact. The same approach was used also for Cu2+-ISEs where the leakage of primary ions from the membrane was eliminated as indicated by a super-Nernstian response for Cu2+ activities below 10-4 M.162 Poly(3-methylthiophene) (PMT) doped with BF4- and modified with EDTA as a complexing agent was used as the solid contact in Ca2+-ISEs for low-level concentration measurements.163 The Ca2+-ISEs showed a super-Nernstian response for Ca2+ activities below 10-5 M, indicating an influx of Ca2+ ions to the ion-selective membrane as a result of complexation of Ca2+ with EDTA present in the solid contact.163

3.2.3. Polyanilines Solution-cast polyaniline (PANI) was used as a solid contact in miniature Cl--ISEs.141 The analytical performance of these solid-contact Cl--ISEs was found to be similar to those based on POT as a solid contact.141 Electrosynthesized PANI was applied as a solid contact in ISEs based on plasticized PVC-based ion-selective membranes. Good overall analytical performance was obtained for such solid-contact ISEs selective to pH164 and Tl3+.165 The stability of the PANI solid contact in K+-ISEs

Bobacka et al.

based on plasticized PVC was studied by using UV-visible spectroscopy. Partial conversion of PANI from its conducting emeraldine salt form to its nonconducting emeraldine base form was found to take place during long-term measurements (1-3 months) but not during short-term measurements (4 days).166 Different derivatives of polyaniline, including poly(Rnaphtylamine), poly(o-aminophenol), poly(o-phenylenediamine), and poly(N-phenylglycine), were used as solid contacts in plasticized PVC membrane-based ISEs for determination of various organic compounds of pharmaceutical importance, including dimedrol,167 chlordiazepoxide,168 propranolol,169 papaverine,170 amiodarone,171 and dopamine.172 The potential stability of some of the solid-contact ISEs was evaluated by current reversal chronopotentiometry,168,171,172 a method used earlier for K+-ISEs based on PEDOT as the solid contact.147

3.3. Conducting Polymers Dissolved in the Ion-Selective Membrane Solid-contact ISEs, where the conducting polymer is dissolved in the ion-selective membrane, were initially called single-piece electrodes.173 Since the conducting polymer is dissolved in the ion-selective membrane, it can influence the selectivity of the electrode. Solid-contact Li+-ISEs based on plasticized PVC containing 1% (w/w) PANI protonated (doped) with bis(2-ethylhexyl) phosphoric acid were studied.174 The solid-contact Li+-ISEs showed the same dynamic response range as the conventional ISE with internal solution. However, the presence of PANI in the ion-selective membrane increased the H+ interference, due to the pH sensitivity of PANI.174 PANI protonated (doped) with phosphoric acid dihexadecyl ester was used both as an ion-to-electron transducer and as a pH-sensitive component in plasticized PVC-based membranes.175 Membranes containing 50% (w/w) PANI and 50% plasticized PVC showed the best pH sensitivity among those studied. Interestingly, the emeraldine salt-emeraldine base transition of PANI was facilitated by lipophilic cations and hindered by lipophilic anions added to the membrane. However, the analytical performance of these pH electrodes was slightly inferior to that of electrochemically synthesized PANI.175 Solid-state ISEs for determination of linear alkylbenzenesulfonates were constructed by using plasticized PVC membranes containing 5% (w/w) electrochemically synthesized polypyrrole doped with dodecylbenzenesulfonate.176 Here, polypyrrole worked both as an ion-to-electron transducer and as an anion-exchanger for alkylbenzenesulfonate anions. Solid-state K+-ISEs were prepared by using plasticized PVC membranes containing 2% (w/w) polypyrrole, PANI, or poly(o-anisidine) that were doped (protonated) with diesters of sulfosuccinic acid.177 The membranes were solution-cast on planar silver electrodes. The resulting solidcontact K+-ISEs showed comparable selectivity with the corresponding hydrogel-contact K+-ISEs. However, the presence of the conducting polymer, especially polypyrrole, in the ion-selective membrane improved the reproducibility and repeatability of the response. The signal stability of this type of solid-contact K+-ISE containing 2% (w/w) polypyrrole doped with di(2-ethylhexyl)sulfosuccinate was improved when the membrane was cast on a gold substrate instead of silver,178 which can be related to the higher work function

Potentiometric Ion Sensors

of gold compared to silver. The presence of polypyrrole in the membrane did not influence the ion-selectivity in this case.

3.4. Conducting Polymers as Sensing Membranes Solid-state ion-selective electrodes based on immobilization of ion-recognition sites in the conducting polymer membrane represent a research area of great potential. Over the past few years, the main focus has been on conducting polymers that contain ion-recognition sites in the form of immobilized doping ions.24 Covalent binding of ion-recognition sites to conducting polymers was already suggested in the 1980s.179 Covalent binding of ion-recognition sites to the conducting polymer backbone allows integration of the ionrecognition sites and the ion-to-electron transducer even within the same (macro)molecule, which may be of great importance for the construction of durable micro- and nanosized ion sensors in the future. However, the synthesis of such functionalized monomers and their polymerization is relatively demanding. A solid-state Zn2+-ISE based on electrochemically synthesized polypyrrole doped with tetraphenylborate was developed.128,180 The selectivity coefficients (log KZn,j) determined by the fixed interference method were as follows: j ) Ca2+ (-2.7), Mg2+ (-2.1), Pb2+ (-1.6), Ni2+ (-0.6), and Co2+ (-0.6).128,180 Solid-state ISEs for different cations (Ca2+, Mg2+, Cu2+, and Zn2+) based on electrochemically synthesized polypyrrole doped with metal-complexing ligands were further developed.181,182 In addition to metal-complexing groups, the ligands contain sulfonate groups that compensate for the positive charge of the oxidized (p-doped) polypyrrole backbone. The effects of chemical (soaking) and electrochemical (oxidation/reduction) conditioning on the potentiometric sensitivity and selectivity of polypyrrole doped with metal-complexing ligands were studied in detail.181,182 Polypyrrole doped with adenosine triphosphate (ATP) was found to give a near-Nernstian response to Ca2+ and Mg2+, and the response was not influenced by Na+.183 It was suggested that the polypyrrole doped with ATP can be used as artificial membranes in order to model ATP-mediated processes of real biological membranes.183 Furthermore, there was a correlation between the film topography and the potentiometric response of PEDOT doped with ATP, which was sensitive to Ca2+ and Mg2+.184 Smoother films generally showed a more stable and faster potentiometric response than the rougher ones.184 Polypyrrole and PEDOT doped with heparin were also found to give a near-Nernstian response to Ca2+ and Mg2+, and the response was not influenced by Na+ or K+.185 The response remained unchanged even after 1 year of soaking, indicating the high stability of this type of electrode.185 A solid-state pH electrode were developed by using electrochemically synthesized polypyrrole doped with cobalt bis(dicarbollide).186 The electrode showed a quasi-Nernstian response (-50 mV/pH unit) and a linear range from pH 3 to 12. Electrodeposited polymers based on various monomers containing amino groups (1,3-diaminopropane, diethylenetriamine, pyrrole, p-phenylenediamine, and aniline) were studied as pH sensors, showing a linear response in the range from pH 2 to 11.187 Polypyrrole-based pH sensors were miniaturized.188 Electrochemically synthesized PPy doped with bicarbonate was applied as a pH electrode in a Severinghaus CO2 sensor.189 Electrochemically synthesized

Chemical Reviews, 2008, Vol. 108, No. 2 343

polypyrrole doped with dodecylsulfate (DS-) was used for the development of a tubular solid-state ISE for determination of DS- in a flow-injection analysis system.190 A correlation between the spontaneous charging/discharging processes and the potentiometric detection limit of conducting polymers, such as polypyrole, poly(N-methylpyrrole), and PEDOT, was found.191 Furthermore, the potentiometric detection limit for electrosynthesized PEDOT(PSS) could be lowered to 10-6-10-7 M by polarization with a small cathodic current that prevented cation leakage from the polymer film into the solution. However, similar to most nonfunctionalized conducting polymers, PEDOT itself showed low selectivity.192 Solid-state Ag+-ISEs based on polypyrrole and PEDOT doped with sulfonated calixarenes (and resorcarenes) were developed.193,194 The response mechanism was studied for Ag+-ISEs based on PEDOT doped with hexabromocarborane.195 Pretreatment of these electrodes by cyclic voltammetry in KNO3 solution was found to improve the potentiometric response to Ag+.195 Electrochemically mediated doping/templating by repeated oxidation/reduction in AgNO3 solution was employed to enhance the potentiometric Ag+ response of polypyrrole that was synthesized by electropolymerization of pyrrole in the presence of Eriochrome BlueBlack B.196 Even undoped poly(3-octylthiophene) (POT) was found to give a selective potentiometric response to Ag+, indicating that Ag+ interacts with the polythiophene backbone (sulfur atoms, π-electrons).197 Electrochemically overoxidized polypyrrole (OPPy) gave a potentiometric response to alkali and alkaline earth metal cations, albeit with low selectivity.198 Overoxidation was suggested to result in oxygen-containing groups acting as “hard” Lewis bases that form complexes with hard cations, while the redox interference was simultaneously decreased because of the lower electronic conductivity of overoxidized polypyrrole.198 Novel electropolymerized films based on some functionalized polyanilines, polypyrrole,s and amino heterocyclic compounds were studied as solid-state ISEs for determination of anions, such as amino acids and ascorbic acid.199 In fact, oxidized (p-doped) conducting polymers are inherently suitable for anion sensors because of the polycationic backbone.199 A very simple procedure for manufacturing of a solid-state NO3--ISE by electrosynthesis of polypyrrole doped with NO3- on a pencil lead was described.200 Similarly, PEDOT doped with ClO4- worked very well as ClO4- sensors with similar selectivity as commercial ClO4-ISEs.201 Following the same approach, polypyrrole doped with valproate was found to give a well-functioning valproate sensor.202 Solid-state Cl--ISEs based on chemically synthesized undoped POT containing trihexadecylmethylammonium chloride (THMACl) ions were studied.203 In contrast to tridodecylmethylammonium, the more lipophilic trihexadecylmethylammonium cation required the addition of a plasticizer (2-nitrophenyloctyl ether, o-NPOE) to the POT film in order to give a functioning Cl--ISE with the following composition: 35% (w/w) POT, 23% (w/w) THMACl, and 42% (w/w) o-NPOE.203 Solid-state Cu2+-ISEs were developed by using electrosynthesized undoped polycarbazole and polyindole as sensing membranes.204 However, these electrodes showed a severely super-Nernstian response to Cu2+ at concentrations higher than 10-4 M.

344 Chemical Reviews, 2008, Vol. 108, No. 2

Bobacka et al.

A solid-state pH nanoelectrode was constructed by electrodeposition of PANI onto an ion-beam conically etched carbon fiber with a tip diameter of ca. 100-500 nm.205 The pH electrode gave a slope of ca. -60 mV/pH unit in the linear pH range of 2.0-12.5. The selectivity coefficients (log KH,j) were around -12 with respect to K+, Na+, Ca2+, and Li+, which is comparable to conventional glass pH electrodes. A solid-state pH microelectrode based on PANI electrodeposited on a microband electrode was used for in situ pH measurement of the self-oscillating Cu(II)-lactate system.206 The pH sensitivity of PANI and its derivatives was found to depend on the substituent and the doping anion incorporated in PANI during electropolymerization.207,208 The best pH response was obtained for PANI doped with chloride, while N-substituted PANI showed no pH sensitivity, because the N-substituent hindered the emeraldine salt-emeraldine base transition.207,208 A solid-state ISE for determination of dodecylsulfate (DS-) was prepared by using electrochemically synthesized PANI doped with DS-. The DS--ISE showed a Nernstian response to DS- in the linear range from 10-9 to 10-5.5 M. The electrode showed relatively high selectivity to DS-, with the strongest interference being observed from dodecylbenzenesulfonate (DBS-) (log KDS,DBS ≈ -2.2).209 Alternatively, when PANI was electrosynthesized in the presence of DBSas the doping anion, the electrode became selective to DBS(e.g., log KDBS,DS ≈ -2.2) with a Nernstian response to DBSin the linear range from 10-5.3 to 10-2.4.210 The potentiometric response of PANI toward aniline based on the formation of charge-transfer complexes between aniline (donor) and the oxidized form of PANI (acceptor) was explained theoretically.211 The theory could also explain the potentiometric response of other similar sensor materials to aromatic amines, pyrrole, and phenols.211 A novel approach for the potentiometric detection of saccharides using poly(aniline boronic acid) was presented.212 The detection was based on pKa changes of the poly(aniline boronic acid) resulting from boronic acid-diol complexation.

3.5. Comments on the Ion-to-Electron Transduction Process Not only solid-state ISEs/ISSs but also conventional ISEs with internal filling solution (and reference electrodes as well) are asymmetric systems, because ions do not enter electronic equipment used to measure the potential. At some point, there is a transduction of the ionic signal into an electronic signal via a reversible redox reaction. In the case of a conventional Ag/AgCl electrode in contact with chloride ions, the reversible redox reaction involves the Ag/Ag+ redox couple, as follows:

Ag + Cl- h AgCl + e-

(22)

The ion-to-electron transduction is completely analogous for other electroactive materials, such as conducting polymers.63-65 Since the electrode contains a finite amount of redox-active material, the electrode has a finite redox capacitance (C).147 For a given constant current (i), there will be a change in the electrode potential (E) with time (t):

∆EC i ) ∆t C

(23)

Additionally, the electrode resistance (R), including all timeindependent resistances of the electrode, will give a constant potential shift:

∆ER ) iR

(24)

Equations 23 and 24 represent the simplest possible case that neglects diffusion/migration processes in the ion-selective membrane and the ion-to-electron transducer. However, this simple RC model was verified experimentally and is used mainly to conceptualize the ion-to-electron transduction process in ISEs.147,213 In practical potentiometric measurements, the current (i) of the high-impedance voltmeter is small enough so that (∆EC/∆t) and ∆ER can be neglected. In contrast, currents induced by electrical noise may be significant enough to influence the potential. Furthermore, miniaturization of electrodes tends to decrease C and increase R, resulting in lower potential stability. This is valid for conventional ISEs with internal filling solution as well as for solid-contact ISEs. The same reasoning is also true for ion-selective electrodes with a completely blocked interface (coated-wire electrodes) where the redox capacitance is replaced by a double-layer capacitance. Consisely, the potential stability of coated-wire electrodes should improve by increasing the contact area between the ion-selective membrane and the electronic contact, because of the increase in the double-layer capacitance. The conjugated polymers so far used as ion-to-electron transducers can be divided into two main groups of materials based on their redox capacitance, which is related to the oxidation level (degree of p-doping) of the conjugated polymers. POT represents a conjugated polymer with a relatively high oxidation potential, and therefore, it has a very low degree of p-doping under ambient conditions. This means that POT has a low electronic conductivity and a low redox capacitance under ambient conditions. On the contrary, PPy, PEDOT, and PANI represent conjugated polymers that are relatively stable in their highly oxidized (p-doped) state, showing high electronic conductivity and high redox capacitance. On the basis of the magnitude of the redox capacitance, highly p-doped conducting polymers, such as PPy, PEDOT, and PANI, should, therefore, be more suitable as ion-toelectron transducers compared to POT. However, this is not the complete account, partly because there may be electrochemical side reactions taking place in parallel with the main ion-to-electron transduction process. Such parallel reactions may be due to the presence of, e.g., O2, CO2, and H2O that can reach the solid contact, even if it is coated with a polymeric ion-selective membrane. Side reactions due to redox couples in the solution are, of course, more likely to play a role when conducting polymers are dissolved in the ion-selective membrane and especially when conducting polymers are used as sensing membranes. Electrochemical side reactions that change the redox state of the conducting polymer will not only influence the electrode potential but also cause a flux of ions into or out of the conducting polymer, which may, for example, influence the detection limit of the ISE. This is likely to happen in the case of highly p-doped conducting polymers that are electroactive in a broad potential range. PEDOT is known to be less sensitive to CO2 and pH compared to PPy and PANI. However, similar to other highly p-doped conducting polymers, PEDOT is also sensitive to redox interference, which results in long-term potential drift. Regardless, the

Potentiometric Ion Sensors

higher the redox capacitance, the smaller is the potential change for a given current and ion flux, i.e., for a given rate of an electrochemical side reaction or a given applied current that compensates for such a side reaction (chronopotentiometry). Simply expressed, this means that a high capacitance stabilizes the potential of the solid contact itself but does not eliminate the ion flux associated with side reactions. The above discussion suggests that a high capacitance is only a partial solution to the stability problem. It would be of utmost importance to have a conducting polymer (or any other electroactive solid material) with a high redox capacitance (in a narrow potential range) that would not participate in any side reactions at all. However, even in the absence of electrochemical side reactions, there may be ion-exchange processes between counterions of a highly p-doped conducting polymer and ions in the ion-selective membrane (in the case of solid-contact ISEs) or solution (in the case where the conducting polymer is used as a sensing membrane). Such ion-exchange processes depend on the selectivity of the ion-selective membrane and the solid-contact materials. Even salt formation at the interface between the conducting polymer and the ionselective membrane is possible, depending on the solubilitity of the ions present in the two phases. The presence of water at the interface leads to an unwanted situation resembling the conventional liquid-contact ISE where the conducting polymer plays the role of an internal reference electrode in contact with a very small volume of internal filling solution, the ion composition of which can easily change with time. As mentioned previously, POT has a low redox capacitance and a low electronic (and ionic) conductivity. This means that the potential of POT is more sensitive to current flow than PPy, PEDOT, and PANI. However, because of its low conductivity, POT is less electroactive and may not participate in side reactions to the same extent as the highly p-doped conducting polymers. Additionally, POT has a low content of ions and is relatively lipophilic, which prevents the accumulation of water and salt inside the solid contact. In the case of solid-contact ISEs, the compatibility between the material used as a solid contact and the plasticizer (and polymer) used in the ion-selective membrane is a key issue that has not been studied thus far. Furthermore, leakage of plasticizer, ionophore, and additives to the solution may become a serious problem in the case of miniaturized ISEs based on plasticized polymer membranes. It is interesting to note that low detection limits for several ions were achieved by using POT as a solid contact together with plasticizer-free acryl-based ion-selective membranes. This indicates that POT may be a good alternative as a solidcontact material, despite its low redox capacitance, if the main target is to reach a low detection limit. In contrast, POT is expected to be more sensitive to external electrical noise compared with the highly doped conducting polymers. The above discussion focused on ion-to-electron transducers based on conducting polymers, because these materials have been extensively studied for many years and they appear to be very promising. Conducting polymers make it possible to fabricate, e.g., all-plastic ISEs161 and pH nanoelectrodes.205 However, recent results indicate that excellent potential stability (potential drift ca. 11.7 µV/h) can be achieved by using three-dimensionally ordered macroporous carbon as the solid contact.214 Despite the relatively complicated manufacturing procedure, this latter approach is indeed very promising. In this case, the large contact area between the

Chemical Reviews, 2008, Vol. 108, No. 2 345

ion-selective membrane and the electronically conducting carbon material should result in a large double-layer capacitance that would stabilize the potential. Further characterization of this type of solid-contact ISE by, e.g., electrochemical impedance spectroscopy may provide important fundamental information concerning the ion-to-electron transduction process of solid-contact ISEs. Consequently, which material is the best ion-to-electron transducer? The above discussion indicates that the choice of conducting polymer or any other material used as an ionto-electron transducer must be done on a case-to-case basis. Requirements, such as analytical performance, mechanical durability, lifetime, fabrication methods, and cost of the ion sensor, have to be considered. Finding a solid material that fulfills all possible requirements of a good ion-to-electron transducer for all possible applications is really a great challenge.

4. Miniaturized ISEs 4.1. Conventional Microelectrodes Ion-selective microelectrodes are used in many different applications from life sciences to environmental control. There are several reviews, books, and sections of books dedicated to this topic, e.g., by Purves,215 Ammann,216 and Takahashi et al.217 Microelectrodes based on the concept of micropipets have already been used for a longer time, especially in life-science applications.218 Two different pipet electrodes, one as the indicator electrode and the second as the reference electrode, have been used, as well as the construction of a doublebarreled micropipet electrode. The pipets have to be silanized before applying the ion-selective cocktail into the pipet. An improved method for this procedure was described recently.219 There is also a patent for the apparatus and method to selectively induce hydrophobicity in a single barrel of a multibarreled ion-selective microelectrode.220 Scanning electrochemical microscopy (SECM) is a technique where ion-selective microelectrodes have found wide areas of application and where the advantages are the micrometer dimensions, selectivity, and low detection limit.221 pH probes based on an antimony microdisk electrode222 or a H+-sensitive liquid membrane have been used in pHmicroscopy where pH profiles have been measured in corrosion studies223,224 or in measuring biological activities.225,226 The technique has also been applied to study many other kinds of surfaces and processes.227,228 A novel type of solid-contact ammonium-selective microelectrode was recently constructed for SECM measurements.229 Doublebarreled chloride-selective microelectrode were used to map in situ Cl- ion distribution in localized corrosion systems.230 There are also microelectrodes for some “not-so-common” ions. A new Schiff’s base was synthesized and tested as the ionophore for yttrium ions in an ion-selective microelectrode.231 That sensor was tested in complexometric titration and direct determination of yttrium in dissolved yttriumaluminum alloy samples. Environmental measurement of ionic species with ionselective microelectrodes is also an interesting application area. Several ions have been measured in pore water samples,232 and a phosphate-selective microelectrode based on cobalt as the sensing material was used in studying a biological phosphorus removal process.233

346 Chemical Reviews, 2008, Vol. 108, No. 2

By selecting an optimum composition for the inner reference solution, the detection limit of a calcium micropipet electrode could be lowered.234 The calibration curve was found to be linear down to 10-8 M of Ca2+. Ion-selective microelectrodes were recently used in determination of extremely low levels of calcium, lead, and silver ions, 10-10 M in the presence of 10-5 or 10-6 M sodium ions.235 It was possible to detect 300 attomol of those ions at a constant ionic background.

4.2. Solid-Contact Microelectrodes During the past decade, microsensors have gained a widespread and increasing interest both by scientists, engineers, and people working with practical applications of these sensors. The collaboration between scientists and microtechnology engineers has been fruitful, leading to different manufacturing procedures for making both thin- and thickfilm microelectrodes. New technologies enabling mass production of miniaturized ion-selective electrodes were reviewed.236 The discussion concerned mainly applications in biology and medicine, but procedures to make thin- and thick-film microelectrodes were also reviewed.236 Most of the currently used microelectrodes are manufactured on solid substrates. Gold, silver, platinum, and different kinds of carbon materials are used. A recent review on advanced chemical microsensors deals with the design and performance of miniaturized chemical sensors based on Si transducers: ion-selective field effect transistors and solidstate electrodes.67 Si nitride substrate with a polymeric solid contact was used in a miniaturized sodium-selective electrode.237 Silicon substrate was also used in constructing a pH-sensitive microelectrode where the sensitivity is due to ion-selective PVC membrane on top of polypyrrole layer doped with cobalt bis(dicarbollide).238 Pulsed-laser deposition is also a practical method for making solid-state micro-ionselective electrodes. Cd2+ selective microelectrode was constructed by pulsed laser deposition (PLD) technique and using CdSAgIAs2S3 as the sensitive material.239 Screen-printing technology is a useful and practical way to produce solid-contact electrodes. A procedure to make an array of all-solid-state potassium selective electrodes on screen-printed gold substrates has been described.156 Screenprinting technology was also used to produce thick-film Ag sulfide electrodes.240 The performance of these strip sensors was comparable to those of conventional electrodes. Another way to make silver- and sulfide-selective electrodes is to etch the end of a silver wire to a sharp tip and then to insulate the wire except for the very end that then functions as the sensing part of the probe.241 Even an all-solid-state potentiometric sensor for ascorbic acid was constructed based on this screen-printing technology.73 Laser-ablation, screenprinting, and molecular-imprinting techniques were used in making nanoliter-volume vials with carbon and Ag/AgCl ring electrodes embedded in the sidewalls.242 Polypyrrole was electropolymerized on the carbon rings with nitrate as the doping ion, and the vials were used in determination of nitrate in nanoliter samples. On the basis of earlier reports on anionselectivity of polypyrrole, it is somewhat suprising how selective are the nitrate sensors that the authors have been able to produce by using nonfunctionalized polypyrrole. The main advantage of that system lies in the small sample size. Screen-printing technique is also used in making planar-form solid electrolyte modified Ag/AgCl reference electrodes.243,244 Construction and design of this type of reference electrodes

Bobacka et al.

make them suitable for measurements in physiological samples and in flow analysis. A novel method to make miniaturized solid-contact ionselective electrodes has been presented.157 The gold wire in a glass capillary was etched to form a microcavity in the electrode body. The gold disk electrode was then coated by electropolymerization with the conducting polymer poly(3,4ethylenedioxythiophene) (PEDOT) doped with poly(4-styrenesulfonate) (PSS). Potassium- and calcium-selective microelectrodes were then made by filling the cavity with the corresponding ion-selective membrane. These microelectrodes were found to have similar potentiometric characteristics as conventional electrodes. Solid-contact microelectrodes are suitable for use in harsh and remote environments. An exotic example of such an application is the use of a microelectrode array for measurement of ion concentrations in the Martian soil.245 A prototype miniature array of polymer-membrane and solid-state ionselective electrodes was developed to perform in situ analysis of soil samples for a number of ions. The array consisted of 27 microelectrodes for nine ions. Each electrode was in three replicates.68 Screen-printing technology has also been used in constructing miniaturized solid-contact electrodes for chloride ions. Either carbon or silver was used as the screen-printed substrate. The conducting polymers polypyrrole or polyaniline with an adhesive admixture of 3-aminopropyltriethoxysilane was placed on the substrate. The ion-selective PVC membrane containing the ion-exchanger for chloride (methyl-tri-n-tetradecylammonium chloride) was then placed on the conducting polymer film. This construction showed the same performance and stability as the conventional electrodes with inner reference electrode and solution.116,246 The potential stability of microfabricated potassium ionselective electrodes with two types of solid contact was studied and compared.118 Hydrogel and polypyrrole doped with potassium hexacyanoferrate(II)/(III) were used as the solid contact between the PVC-based potassium-selective membrane and the screen-printed silver or platinum substrates. The electrodes were incorporated into a flow-through cell where the measurements were conducted. The electrodes with the polypyrrole solid contact showed much higher potential stability than the electrodes with hydrogel contact.118 A microfabricated ion-selective microelectrode-array platform has been constructed and characterized.247,248 The platform contained 24 micropipets individually filled with a Ca2+ selective membrane and was developed for in vitro intracellular measurements. Different chalcogenide glasssensitive materials have also been used to make an array of miniaturized ion-selective electrodes.249 It was demonstrated in that work that the sensor array allows the problem of an insufficient selectivity of single sensors to be overcome. A microsensor array of miniaturized solid-state ion-selective electrodes to analyze sweat samples for sodium, potassium, and chloride was developed for point-of-care diagnosis of cystic fibrosis.250 Miniaturized, planar ion-selective electrodes fabricated by thick-film technology were recently reviewed.251 Different manufacturing procedures were discussed and screen printing was found to be a suitable technique because of its simplicity, low cost, high reproducibility, and efficiency in large-scale production. Thick-film technology was also demonstrated to be a proper method to produce ionophore-based ion-selective

Potentiometric Ion Sensors

electrodes on inexpensive substrates. The analytical parameters of these electrodes were comparable with those of conventional electrodes. Miniaturized Ag, Ag/AgCl, or iridium oxide solid-state potentiometric sensors were used to measure the capacitance and the response time of the electrodes by an electrochemical time-of-flight method.252,253

4.3. Microelectrodes in Flow Analysis Solid-contact ion-selective microelectrodes have many advantages when used in flow channels. They are robust and rather easy to fabricate and install. A miniaturized ionselective Cu2+ electrode was developed for fluidic microsystems.254 Cu was deposited on a silicon wafer and then converted to CuS by hydrogen sulfide. The electrode performed well in the system. The authors found differences in the adhesion of the CuS layer for Si, SiO2, and Si3N4. An all-solid-state potassium microelectrode for flow analysis was developed for measurements in flowing solutions177 as well as a nitrate sensor for determination of nitrates in wastewaters, fertilizers, and pharmaceuticals by the flow-injection analysis technique.255 A small-volume flow cell was developed by incorporating all-solid-state electrodes.155 Ca2+ and pH measurements were carried out in the Lab-on-Valve instrument by using all-solid-state microelectrodes.152 Lowtemperature cofired ceramic technology was used to construct continuous-flow analytical microsystems.256 Ammonium- and nitrate-selective microelectrodes were incorporated in that microfluidic sysytem, allowing a complete on-chip integration for potentiometric detection. A microfluidic device was developed for measurement of pH as well as of Ca2+ and K+ concentrations with in-channel-incorporated all-solid-state ion-selective electrodes.257 Construction of different planar solid-state ion-selective microelectrodes as well as reference electrodes for flow analysis was recently discussed in an overview.117

4.4. Life Science and Biology Applications The use of microelectrodes to investigate transportation of inorganic ions in plants has been reviewed.258 The construction of multibarreled ion-selective microelectrodes for measurements in biological tissues was also reviewed.259 It was also demonstrated how cell volume, membrane potential, and intracellular ion concentrations can simultaneously be determined by using potentiometric measurement with multibarreled ion-selective electrodes. A special study on the cell volume regulation mechanism of nerve cells by using multibarreled ion-selective microelectrodes was made.260 Different requirements for the charge-exchange processes between a dielectric layer and the ions in the sample in ion-selective biosensors made on silicon were discussed in another study.261 It was found that standard dielectric materials normally employed in microelectronics technology can be used in making ion-sensitive field-effect transistors and ion-selective microelectrodes for biosensing applications. Use of ionselective microelectrodes and fluorescent dyes in measurement of intracellular pH has been compared and discussed.262 Ion-selective microelectrodes were used in plant physiology in studying the NH4+, H+, and NO3- fluxes around the roots of plants263 and their concentration profiles in nitrifying biofilms.264 By using these electrodes, the authors were able to determine the source of nitrogen for a particular plant. The mechanism of pH homeostasis in Listeria monocytogens subjected to, e.g., acid stress was studied by using a pH

Chemical Reviews, 2008, Vol. 108, No. 2 347

microelectrode,265 and ammonium transport across the plant cell membrane was measured with an ammonium-selective microelectrode.266 Ionized magnesium has been measured with an ionselective microelectrode in physiological solutions,267,268 and the release of ionized Ca from bones has been measured by a Ca-selective microelectrode.269 Ion-selective microelectrodes were also used in studying ion fluxes across cardiac membrane patches.270 Potassium, calcium, sodium, and proton fluxes were measured, and the authors were, additionally, able to quantify those fluxes. The effect of interfering ions in using ion-selective microelectrodes in biological applications has been discussed.271 The kinetics of ion fluxes in some plant leaf and root tissues was determined by using ion-selective microelectrodes,272,273 as was ion fluxes in bacteria.274 Fluxes of H+ and Ca2+ ions during the growth of a single-celled fungus were also studied by ion-selective microelectrodes.275 Salicylate selective microelectrodes were even constructed for biological applications to measure fluxes of salicylate ions.276 Ion-selective electrodes are used in many biosensors to detect, e.g., ammonia or H+ released in the enzymatic reactions involved in the detection procedure. Ion-selective microelectrodes are needed for miniaturization of such biosensors.277 Some of the main problems in using ion-selective microelectrodes are the slow response time and the susceptibility to noise because of the high electrical resistance of the convential micropipet electrodes (see section 3.5). These problems are especially crucial in biological in vivo measurements but can partially be overcome by using a concentric inner micropipet.278 Response times in the order of a few ms were obtained for pH and Ca2+ electrodes used in extracellular measurements.

5. New Wave During the past decade, the developments in the field of potentiometric ion sensors (ISEs) have been exceptionally fruitful. The discovery of the low detection limit in 1997 represents a major breakthrough in the field of ISEs.30 Much research has been devoted to the design and optimization of liquid-contact ISEs to measure low ion concentrations.234,235,279-285 The “new wave” of ISEs has unquestionably arrived.6,7 The question is: where will it lead us in the future? Solid-contact ISEs for trace-level analysis have already been demonstrated by several groups.75,120-123,125,142-144,163 When such powerful ion sensors are combined with solidstate reference electrodes,243,286-292 mass-producible miniaturized ion-sensor systems with unforeseen analytical capabilities are within reach. Deep theoretical understanding based on advanced modeling of the potentiometric response will definitely boost this development. Parallel developments of plasticizer-free membranes,293-304 covalently bound ionophores,293,296,298,302,305-309 and charged sites304 will enhance the durability of ion sensors. Other alternative membrane materials and construction principles add new tools for the future as well. Here, we would like to mention anion-sensitive epoxy resins,310,311 membranes prepared by the sol-gel method,312 polymer-supported liquidcrystal membranes,313 polymer membranes located inside an electrochemically inert porous matrix,314 monolithic capillarybased membranes,284 and fluorous membranes with exceptionally low polarity.315

348 Chemical Reviews, 2008, Vol. 108, No. 2

There are several other interesting developments that may have an impact on the future developments of ISEs. The socalled sandwich membrane method, which was actually introduced already in 1979, is a useful tool for studying complexation in real membranes.316,317 Polycation-sensitive electrodes can be used to detect even polycationic dendrimers.318 Highly lipophilic closo-dodecacarborane anions showing very weak ion-pair formation are promising alternatives to the commonly used tetraphenylborate derivatives as anionic additives in polymeric ion-selective membranes.319 Under certain conditions, conventional ISEs with internal filling solution can be calibrated by varying the composition of the internal solution, which may be particularly useful in applications where the electrode cannot easily be removed from the sample (e.g., process control and in vivo measurements) during calibration.320 A solvent-free method for making hotpressed ion-selective membranes is an environmentally friendly approach that avoids the use of organic solvents.321 Application of a controlled current can be used to influence transmembrane ion fluxes and is an elegant approach toward lowering the detection limit of ISEs.322,323 This method has also been applied to solid-contact ISEs using polypyrrole as the ion-to-electron transducer.126 As discussed in section 3.5, highly p-doped conducting polymers with a high redox capacitance can simultaneously work as “source” and “sink” of electrons and ions without any significant change in potential if the magnitude of the applied current is sufficiently low. Therefore, chronopotentiometry may become a useful method to control transmembrane ion fluxes and detection limits of solid-contact ISEs. Under some conditions, currentreversal chronopotentiometry allows the equilibrium potential to be determined although the electrode is significantly polarized by the applied current.213 Furthermore, a significant increase in sensitivity was achieved by imposing current pulses to the ISE, leading to the concept of so-called pulstrodes.324 Being a fascinating research area in itself, the new wave of ISEs will most probably also result in new applications of potentiometric ion sensors in the future.

6. Conclusions The response of potentiometric ion sensors, i.e., ionselective electrodes (ISEs), is a complex time-dependent phenomenon that calls for advanced theoretical modeling in addition to classical equilibrium models. This is particularly important when ISEs are used in their nonequilibrium regimes, e.g., in order to reach low detection limits, which is intensively being explored all over the world. The prerequisites for obtaining solid-contact ISEs with stable potential are well-documented, and during the past decade, conducting polymers have gained widespread acceptance as ion-to-electron transducers in solid-contact ISEs. However, in spite of extensive research in this area, obtaining solidcontact ISEs with reproducible standard potentials is still a great challenge. Miniaturized versions of both conventional and solid-contact ISEs have been used in various applications. Cost-effective miniaturized ion-sensor systems with unforeseen analytical capabilities are within reach. Advances made by numerous research groups all over the world allow us to conclude that the future of potentiometric ion sensors looks very prospective indeed.

7. Acknowledgments A.L. thanks Dr. Tomasz Sokalski and Dr. Witold Kucza for very valuable discussions and assistance during the

Bobacka et al.

preparation of this text. M.Sc. Krzysztof Szyszkiewicz, M.Sc. Peter Lingenfelter, Prof. Robert Filipek, and Prof. Marek Danielewski are acknowledged for their interest and invaluable support. The financial support (AL) by the Polish Committee for Scientific Research (KBN) project 3T09A 175 27 is acknowledged. This work is part of the activities at the Åbo Akademi Process Chemistry Centre within the Finnish Centre of Excellence Programme (2000-2011) by the Academy of Finland.

8. References (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) (29) (30) (31) (32) (33) (34) (35) (36) (37) (38) (39) (40) (41) (42) (43) (44) (45) (46)

Bakker, E.; Telting-Diaz, M. Anal. Chem. 2002, 74, 2781. Bakker, E. Anal. Chem. 2004, 76, 3285. Bakker, E.; Qin, Y. Anal. Chem. 2006, 78, 3965. Bakker, E.; Bu¨hlmann, P.; Pretsch, E. Chem. ReV. 1997, 97, 3083. Bu¨hlmann, P.; Pretsch, E.; Bakker, E. Chem. ReV. 1998, 98, 1593. Bakker, E.; Pretsch, E. TrAC, Trends Anal. Chem. 2005, 24, 199. Pretsch, E. TrAC, Trends Anal. Chem. 2007, 26, 46. Zytkow, J. M.; Lewenstam, A. Fresenius J. Anal. Chem. 1990, 338, 225. Guggenheim, E. A. J. Phys. Chem. 1929, 33, 842. Nikolskii, B. P. J. Phys. Chem. (U.S.S.R.) 1937, 10, 495. Eisenman, G. In Ion-selectiVe electrodes; Durst, R. A., Ed.; NBS Special Publication 314; NBS: Washington, DC, 1969. Kakiuchi, T. Anal. Chem. 1996, 68, 3658. Morf, W. E.; Badertscher, M.; Zwickl, T.; de Rooij, N. F.; Pretsch, E. J. Phys. Chem. B 1999, 103, 11346. Sokalski, T.; Lingenfelter, P.; Lewenstam, A. J. Phys. Chem. B 2003, 107, 2443. Mikhelson, K. N.; Lewenstam, A. Anal. Chem. 2000, 72, 4965. Bakker, E.; Meruva, R. K.; Pretsch, E.; Meyerhoff, M. E. Anal. Chem. 1994, 66, 3013. IUPAC. Pure Appl. Chem. 1994, 66, 2527. Lewenstam, A. Scand. J. Clin. Lab. InVest. 1994, 54, 11-19. Mikhelson, K. N.; Lewenstam, A. Sens. Actuators, B 1998, 48, 344. Mikhelson, K. N.; Lewenstam, A.; Didina, S. E. Electroanalysis 1999, 11, 793. Bakker, E.; Buhlmann, P.; Pretsch, E. Talanta 2004, 63, 3. Nikolskii, B. P.; Shultz, M. M.; Belijustin, A. A. Glass Electrodes for Hydrogen and Other Cations; Eisenman, G., Ed.; Marcel Dekker: New York, 1967. Lewenstam, A.; Bobacka, J.; Ivaska, A. J. Electroanal. Chem. 1994, 368, 23. Migdalski, J.; Blaz, T.; Lewenstam, A. Anal. Chim. Acta 1996, 322, 151. Michalska, A.; Lewenstam, A. Anal. Chim. Acta 2000, 406, 159. Moody, G. J.; Thomas, J. D. R. Talanta 1971, 18, 1251. Hulanicki, A.; Lewandowski, R. Chem. Anal. (Warsaw) 1974, 19, 53. Hulanicki, A.; Lewenstam, A. Talanta 1977, 24, 171. Paczosa, B.; Migdalski, J.; Lewenstam, A. Electrochim. Acta 2006, 51, 2173. Sokalski, T.; Ceresa, A.; Zwickl, T.; Pretsch, E. J. Am. Chem. Soc. 1997, 119, 11347. Lewenstam, A. Ph.D. Thesis, Warsaw University, Poland, 1977. Hulanicki, A.; Lewenstam, A. Talanta 1977, 24, 171. Hulanicki, A.; Lewenstam, A. Anal. Chem. 1981, 53, 1401. Hulanicki, A.; Lewenstam, A. Talanta 1982, 29, 661. Morf, W. E. Anal. Chem. 1983, 55, 1165. Lewenstam, A.; Hulanicki, A.; Sokalski, T. Anal. Chem. 1987, 59, 1539. Lewenstam, A.; Hulanicki, A. SelectiVe Electrode ReV. 1990, 12, 161; and 1991, 13, 129 (erratum). Ilcheva, L; Cammann, K. Fresenius J. Anal. Chem. 1985, 320, 664. Paczosa-Bator, B.; Blaz, T.; Migdalski; J. Lewenstam, A. Bioelectrochem. 2007, published on Web: doi:10.1016/j.bioelechem.2007. 01.002. Fu, B.; Bakker, E.; Yun, J. H.; Yang, V. C.; Meyerhoff, M. Anal. Chem. 1994, 66, 2250. Langmaier, J.; Samcova, E.; Samec, Z. Anal. Chem. 2007, 79, 2892. Morf, W. E.; Badertscher, M.; Zwickl, T.; Reichmuth, Pretsch, E. J. Phys. Chem. B 2000, 104, 8201. Hulanicki, A.; Lewenstam, A. Talanta 1976, 23, 661. Lewenstam, A.; Sokalski, T.; Hulanicki, A. Talanta 1985, 32, 531. Hulanicki, A.; Sokalski, T.; Lewenstam, A. Mikrochim. Acta 1988, III, 119. Sokalski, T.; Maj-Zurawska, M.; Hulanicki, A. Mikrochim. Acta 1991, I, 285.

Potentiometric Ion Sensors (47) Sokalski, T.; Zwickl, T.; Bakker, E.; Pretsch, E. Anal. Chem. 1999, 71, 1204. (48) Szigeti, Z.; Vigassy, T.; Bakker, E.; Pretsch, E. Electroanalysis 2006, 13-14, 1254. (49) Morf, W. E.; Pretsch, E.; De Rooij, N. F. J. Electroanal. Chem. 2007, 602, 43. (50) Brumleve, T. R.; Buck, R. P. J. Electroanal. Chem. 1978, 90, 1. (51) Sokalski, T.; Lewenstam, A. Electrochem. Commun. 2001, 3, 107. (52) Sokalski, T.; Lingenfelter, P.; Lewenstam, A. J. Phys. Chem. B 2003, 107, 2443. (53) Lingenfelter, P.; Bedlechowicz-Sliwakowska, I.; Sokalski, T.; MajZurawska, M.; Lewenstam, A. Anal. Chem. 2006, 78, 6783. (54) Kucza, W.; Danielewski, M.; Lewenstam, A. Electrochem. Commun. 2006, 8, 416. (55) Filipek, R.; Szyszkiewicz, K.; Danielewski, M.; Sokalski, T.; Lewenstam, A. In preparation. (56) Sokalski, T.; Kucza, W.; Lewenstam, A. Anal. Chem., in preparation. (57) Cai, C.; Tong, Y.; Mirkin, M. V. J. Phys. Chem. B 2004, 108, 17872. (58) Frank, S.; Schmickler, W. Electroanal. Chem. 2006, 590, 138. (59) Kakiuchi, T.; Tsujioka, N. J. Electroanal. Chem. 2007, 599, 209. (60) Heng, L. Y.; Toth, K.; Hall, E. A. H. Talanta 2004, 63, 73. (61) Nikolskii, B. P.; Materova, E. A. Ion-Sel. Electrode ReV. 1985, 7, 3. (62) Cattrall, R. W.; Freiser, H. Anal. Chem. 1971, 43, 1905. (63) Bobacka, J.; Ivaska, A.; Lewenstam, A. Electroanalysis 2003, 15, 366. (64) Bobacka, J.; Lindfors, T.; Lewenstam, A.; Ivaska, A. Am. Lab. 2004, 36, 13. (65) Bobacka, J. Electroanalysis 2006, 18, 7. (66) Michalska, A. Anal. Bioanal. Chem. 2006, 384, 391. (67) Wro´blewski, W.; Dybko, A.; Malinowska, E.; Brzo´zka, Z. Talanta 2004, 63, 33. (68) Lukow, S. R.; Kounaves, S. P. Electroanalysis 2005, 17, 1441. (69) Fibbioli, M.; Bandyopadhyay, K.; Liu, S.-G.; Echegoyen, L.; Enger, O.; Diederich, F.; Gingery, D.; Bu¨hlmann, P.; Persson, H.; Suter, U. W.; Pretsch, E. Chem. Mater. 2002, 14, 1721. (70) Grygołowicz-Pawlak, E.; Wygladacz, K.; Sek, S.; Bilewicz, R.; Brzo´zka, Z.; Malinowska, E. Sens. Actuators, B 2005, 111-112, 310. (71) Gabrielli, C.; He´mery, P.; Liatsi, P.; Masure, M.; Perrot, H. J. Electrochem. Soc. 2005, 152, H219. (72) Gallardo, J.; Alegret, S.; Mun˜oz, R.; De-Roma´n, M.; Leija, L.; Herna´ndez, P. R.; del Valle, M. Anal. Bioanal. Chem. 2003, 377, 248. (73) Veltsistas, P. G.; Prodromidis, M. I.; Efstathiou, C. E. Anal. Chim. Acta 2004, 502, 15. (74) Legin, A. V.; Makarychev-Mikhailov, S. M.; Kirsanov, D. O.; Vlasov, Yu. G. Russ. J. Appl. Chem. 2002, 75, 926. (75) Fouskaki, M.; Chaniotakis, N. A. Anal. Chem. 2005, 77, 1780. (76) Shirakawa, H.; Louis, E. J.; MacDiarmid, A. G.; Chiang, C. K.; Heeger, A. J. J. Chem. Soc., Chem. Commun. 1977, 578. (77) Shirakawa, H. Synth. Met. 2002, 125, 3. (78) MacDiarmid, A. G. Synth. Met. 2002, 125, 11. (79) Heeger, A. J. Synth. Met. 2002, 125, 23. (80) Skotheim, T. A., Ed. Handbook of Conducting Polymers; Marcel Dekker: New York, 1986. (81) Skotheim, T. A., Elsenbaumer, R. L., Reynolds, J. R., Eds. Handbook of Conducting Polymers, second ed.; Marcel Dekker: New York, 1998. (82) Skotheim, T. A., Reynolds, J. R., Eds. Handbook of Conducting Polymers, third ed.; CRC Press: Boca Raton, FL, 2007. (83) Nalwa, H. S., Ed. Handbook of Organic ConductiVe Molecules and Polymers; Wiley: Chichester, U.K., 1997. (84) Lyons, M. E. G., Ed. ElectroactiVe Polymer Electrochemistry; Plenum Press: New York, 1994. (85) Ivaska, A. Electroanalysis 1991, 3, 247. (86) Imisides, M. D.; John, R.; Riley, P. J.; Wallace, G. G. Electroanalysis 1991, 3, 879. (87) Bidan, G. Sens. Actuators, B 1992, 6, 45. (88) Zotti, G. Synth. Met. 1992, 51, 373. (89) Bartlett, P. N.; Birkin, P. R. Synth. Met. 1993, 61, 15. (90) Teasdale, P. R.; Wallace, G. G. Analyst 1993, 118, 329. (91) Bartlett, P. N.; Cooper, J. M. J. Electroanal. Chem. 1993, 362, 1. (92) Josowicz, M. Analyst 1995, 120, 1019. (93) Emr, S. A.; Yacynych, A. M. Electroanalysis 1995, 7, 913. (94) Sadik, O. A. Anal. Methods Instrum. 1996, 2, 293. (95) Barisci, J. N.; Conn, C.; Wallace, G. G. TRIP 1996, 4, 307. (96) Adeloju, S. B.; Wallace, G. G. Analyst 1996, 121, 699. (97) Rover, L. L., Jr.; De Oliveira Neto, G.; Kubota, L. T. Quı´m. NoVa 1997, 20, 519 (in Portuguese). (98) Go¨pel, W.; Schierbaum, K.-D. In Handbook of Organic ConductiVe Molecules and Polymers, Vol. 4. ConductiVe Polymers: Transport, Photophysics and Applications; Nalwa, H. S., Ed. Wiley: Chichester, U.K., 1997; p 621. (99) Fabre, B.; Simonet, J. Coord. Chem. ReV. 1998, 178-180, 1211.

Chemical Reviews, 2008, Vol. 108, No. 2 349 (100) Giuseppi-Elie, A.; Wallace, G. G.; Matsue, T. In Handbook of Conducting Polymers, second ed.; Skotheim, T. A., Elsenbaumer, R. L., Reynolds, J. R., Eds.; Marcel Dekker: New York, 1998; p 963. (101) Lewis, T. W.; Wallace, G. G.; Smyth, M. R. Analyst 1999, 124, 213. (102) Wallace, G. G.; Smyth, M.; Zhao, H. Trends Anal. Chem. 1999, 18, 245. (103) McQuade, D. T.; Pullen, A. E.; Swager, T. M. Chem. ReV. 2000, 100, 2537. (104) Palmisano, E.; Zambonin, P. G.; Centonze, D. Fresenius J. Anal. Chem. 2000, 366, 586. (105) Kane-Maguire, L. A. P.; Wallace, G. G. Synth. Met. 2001, 119, 39. (106) Fabre, B. In Handbook of AdVanced Electronic and Photonic Materials and DeVices, Vol. 8, Conducting Polymers; Nalwa, H. S., Ed.; Academic Press: San Diego, CA, 2001; p 103. (107) Leclerc, M. In Sensors Update, Vol. 8; Baltes, H., Go¨pel, W., Hesse, J., Eds.; Wiley-VCH: Weinheim, Germany, 2001; p 21. (108) Ramanaviciene, A.; Ramanavicius, A. Crit. ReV. Anal. Chem. 2002, 32, 245. (109) Janata, J.; Josowicz, M. Nat. Mater. 2003, 2, 19. (110) Trojanowicz, M. Mikrochim. Acta 2003, 143, 75. (111) Adhikari, B.; Majumdar, S. Prog. Polym. Sci. 2004, 29, 699. (112) Persaud, K. C. Mater.Today 2005, April, 38. (113) Bobacka, J. In Encyclopedia of Sensors; Grimes, C. A., Dickey, E., Pishko, M. V., Eds.; American Scientific Publishers: California, 2006; Vol. 2, p 279. (114) Maksymiuk, K. Electroanalysis 2006, 18, 1537. (115) Cadogan, A.; Gao, Z.; Lewenstam, A.; Ivaska, A.; Diamond, D. Anal. Chem. 1992, 64, 2496. (116) Zielin´ska, R.; Mulik, E.; Michalska, A.; Achmatowicz, S.; MajZ˙ urawska, M. Anal. Chim. Acta 2002, 451, 243. (117) Mamin´ska, R.; Wro´blewski, W. Electroanalysis 2006, 18, 1347. (118) Gyurcsa´nyi, R. E.; Rangisetty, N.; Clifton, S.; Pendley, B. D.; Lindner, E. Talanta 2004, 63, 89. (119) Konopka, A.; Sokalski, T.; Michalska, A.; Lewenstam, A.; MajZ˙ urawska, M. Chem. Sensors, Suppl. B 2004, 20, 472. (120) Konopka, A.; Sokalski, T.; Michalska, A.; Lewenstam, A.; MajZ˙ urawska, M. Anal. Chem. 2004, 76, 6410. (121) Konopka, A.; Sokalski, T.; Lewenstam, A.; Maj-Z˙ urawska, M. Electroanalysis 2006, 18, 2232. (122) Sutter, J.; Lindner, E.; Gyurcsa´nyi, R. E.; Pretsch, E. Anal. Bioanal. Chem. 2004, 380, 7. (123) Michalska, A.; Duman´ska, J.; Maksymiuk, K. Anal. Chem. 2003, 75, 4964. (124) Michalska, A.; Maksymiuk, K. J. Electroanal. Chem. 2005, 576, 339. (125) Michalska, A. Electroanalysis 2005, 17, 400. (126) Pawłowski, P.; Michalska, A.; Maksymiuk, K. Electroanalysis 2006, 18, 1339. (127) Michalska, A. J.; Appaih-Kusi, C.; Heng, L. Y.; Walkiewicz, S.; Hall, E. A. H. Anal. Chem. 2004, 76, 2031. (128) Pandey, P. C.; Singh, G.; Srivastava, P. K. Electroanalysis 2002, 14, 427. (129) Han, W.-S.; Yoo, S.-J.; Kim, S.-H.; Hong, T.-K.; Chung, K.-C. Anal. Sci. 2003, 19, 357. (130) Zine, N.; Bausells, J.; Teixidor, F.; Vin˜as, C.; Masalles, C.; Samitier, J.; Errachid, A. Mater. Sci. Eng., C 2006, 26, 399. (131) Marques de Oliveira, I. A.; Pla-Roca, M.; Escriche, L l.: Casabo´, J.; Zine, N.; Bausells, J.; Teixidor, F.; Crespo, E.; Errachid, A.; Samitier, J. Electrochim. Acta 2006, 51, 5070. (132) Zine, N.; Bausells, J.; Vocanson, F.; Lamartine, R.; Asfari, Z.; Teixidor, F.; Crespo, E.; Marques de Oliveira, I. A.; Samitier, J.; Errachid, A. Electrochim. Acta 2006, 51, 5075. (133) Kaden, H.; Jahn, H.; Berthold, M. Solid State Ionics 2004, 169, 129. (134) Vonau, W.; Gabel, J.; Jahn, H. Electrochim. Acta 2005, 50, 4981. (135) Sun, X. X.; Zhang, X.; Aboul-Enein, H. Y. Il Farmaco 2004, 59, 307. (136) Sun, X. X.; Aboul-Enein, H. Y. Anal. Sci. 2007, 23, 231. (137) Kim, I. T.; Lee, S. W.; Elsenbaumer, R. L. Synth. Met. 2004, 141, 301. (138) Song, F.; Ha, J.; Park, B.; Kwak, T. H.; Kim, I. T.; Nam, H.; Cha, G. S. Talanta 2002, 57, 263. (139) Bobacka, J.; McCarrick, M.; Lewenstam, A.; Ivaska, A. Analyst 1994, 119, 1985. (140) Paciorek, R.; van der Wal, P. D.; de Rooij, N. F.; Maj-Z˙ urawska, M. Electroanalysis 2003, 15, 1314. (141) Paciorek, R.; Maj-Zurawska, M. Chem. Sensors, Suppl. B 2004, 20, 556. (142) Sutter, J.; Radu, A.; Peper, S.; Bakker, E.; Pretsch, E. Anal. Chim. Acta 2004, 523, 53. (143) Chumbimuni-Torres, K. Y.; Rubinova, N.; Radu, A.; Kubota, L. T.; Bakker, E. Anal. Chem. 2006, 78, 1318.

350 Chemical Reviews, 2008, Vol. 108, No. 2 (144) Rubinova, N.; Chumbimuni-Torres, K.; Bakker, E. Sens. Actuators, B 2007, 121, 135. (145) Sutter, J.; Pretsch, E. Electroanalysis 2006, 18, 19. (146) Khripoun, G. A.; Volkova, E. A.; Liseenkov, A. V.; Mikhelson, K. N. Electroanalysis 2006, 18, 1322. (147) Bobacka, J. Anal. Chem. 1999, 71, 4932. (148) Va´zquez, M.; Bobacka, J.; Ivaska, A.; Lewenstam, A. Sens. Actuators, B 2002, 82, 7. (149) Bobacka, J.; Lahtinen, T.; Koskinen, H.; Rissanen, K.; Lewenstam, A.; Ivaska, A. Electroanalysis 2002, 14, 1353. (150) Bobacka, J.; Va¨a¨na¨nen, V.; Lewenstam, A.; Ivaska, A. Talanta 2004, 63, 135. (151) Lisowska-Oleksiak, A.; Lesin´ska, U.; Nowak, A. P.; Bochen´ska, M. Electrochim. Acta 2006, 51, 2120. (152) Kikas, T.; Ivaska, A. Talanta 2007, 71, 160. (153) Bobacka, J.; Alaviuhkola, T.; Hietapelto, V.; Koskinen, H.; Lewenstam, A.; La¨msa¨, M.; Pursiainen, J.; Ivaska, A. Talanta 2002, 58, 341. (154) Alaviuhkola, T.; Bobacka, J.; Nissinen, M.; Rissanen, K.; Ivaska, A.; Pursiainen, J. Chem.sEur. J. 2005, 11, 2071. (155) Va´zquez, M.; Bobacka, J.; Ivaska, A.; Lewenstam, A. Talanta 2004, 62, 57. (156) Va´zquez, M.; Danielsson, P.; Bobacka, J.; Lewenstam, A.; Ivaska, A. Sens. Actuators, B 2004, 97, 182. (157) Sundfors, F.; Bereczki, R.; Bobacka, J.; To´th, K.; Ivaska, A.; Gyurcsa´nyi, R. E. Electroanalysis 2006, 18, 1372. (158) Ocypa, M.; Michalska, A.; Maksymiuk, K. Electrochim. Acta 2006, 51, 2298. (159) Michalska, A.; Ocypa, M.; Maksymiuk, K. Anal. Bioanal. Chem. 2006, 385, 203. (160) Michalska, A.; Skompska, M.; Mieczkowski, J.; Zago´rska, M.; Maksymiuk, K. Electroanalysis 2006, 18, 763. (161) Michalska, A.; Maksymiuk, K. Anal. Chim. Acta 2004, 523, 97. (162) Michalska, A.; Ocypa, M.; Maksymiuk, K. Electroanalysis 2005, 17, 327. (163) Michalska, A.; Konopka, A.; Maj-Zurawska, M. Anal. Chem. 2003, 75, 141. (164) Han, W.-S.; Chung, K.-C.; Kim, M.-H.; Ko, H.-B.; Lee, Y.-H.; Hong, T.-K. Anal. Sci. 2004, 20, 1419. (165) Kharitonov, S. V.; Zarembo, Y. V.; Zarembo, V. I. Electroanalysis 2006, 18, 1354. (166) Lindfors, T.; Ivaska, A. Anal. Chem. 2004, 76, 4387. (167) Kuznetsova, M. V.; Ryasenskii, S. S.; Gorelov, I. P. Pharm. Chem. J. 2003, 37, 599. (168) Gorelov, I. P.; Ryasenskii, S. S.; Kartamyshev, S. V.; Fedorova, M. V. J. Anal. Chem. 2005, 60, 65. (169) Kartamyshev, S. V.; Kuznetsova, M. V.; Ryasenskii, S. S.; Gorelov, I. P. Pharm. Chem. J. 2005, 39, 43. (170) Kuznetsova, M. V.; Kartamyshev, S. V.; Ryasenskii, S. S.; Gorelov, I. P. Pharm. Chem. J. 2005, 39, 97. (171) Kholoshenko, N. M.; Ryasenskii, S. S.; Gorelov, I. P. Pharm. Chem. J. 2006, 40, 289. (172) Kholoshenko, N. M.; Ryasenskii, S. S.; Gorelov, I. P. Pharm. Chem. J. 2006, 40, 334. (173) Bobacka, J.; Lindfors, T.; McCarrick, M.; Ivaska, A.; Lewenstam, A. Anal. Chem. 1995, 67, 3819. (174) Grekovich, A. L.; Markuzina, N. N.; Mikhelson, K. N.; Bochenska, M.; Lewenstam, A. Electroanalysis 2002, 14, 551. (175) Lindfors, T.; Ervela¨, S.; Ivaska, A. J. Electroanal. Chem. 2003, 560, 69. (176) Shafiee-Dastjerdi, L.; Alizadeh, N. Anal. Chim. Acta 2004, 505, 195. (177) Zachara, J. E.; Toczyłowska, R.; Pokrop, R.; Zago´rska, M.; Dybko, A.; Wro´blewski, W. Sens. Actuators, B 2004, 101, 207. (178) Toczyłowska, R.; Pokrop, R.; Dybko, A.; Wro´blewski, W. Anal. Chim. Acta 2005, 540, 167. (179) Garnier, F. Angew. Chem. 1989, 101, 529. (180) Pandey, P. C.; Singh, G. Sens. Actuators, B 2002, 85, 256. (181) Migdalski, J.; Blaz˘ , T.; Lewenstam, A. Chem. Anal. (Warsaw) 2002, 47, 371. (182) Migdalski, J. Chem. Anal. (Warsaw) 2002, 47, 595. (183) Migdalski, J.; Blaz˘ , T.; Paczosa, B.; Lewenstam, A. Mikrochim. Acta 2003, 143, 177. (184) Paczosa-Bator, B.; Peltonen, J.; Bobacka, J.; Lewenstam, A. Anal. Chim. Acta 2006, 555, 118. (185) Paczosa, B.; Blaz˘ , T.; Migdalski, J.; Lewenstam, A. Polish J. Chem. 2004, 78, 1543. (186) Masalles, C.; Borro´s, S.; Vin˜as, C.; Teixidor, F. Anal. Bioanal. Chem. 2002, 372, 513. (187) Lakard, B.; Herlem, G.; Lakard, S.; Guyetant, R.; Fahys, B. Polymer 2005, 46, 12233. (188) Lakard, B.; Segut, O.; Lakard, S.; Herlem, G.; Gharbi, T. Sens. Actuators, B 2007, 122, 101.

Bobacka et al. (189) Tongol, B. J. V.; Binag, C. A.; Sevilla, F. B., III Sens. Actuators, B 2003, 93, 187. (190) A Ä lvarez-Romero, G. A.; Morales-Pe´rez, A.; Rojas-Herna´ndez, A.; Palomar-Pardave´, M.; Ramı´rez-Silva, M. T. Electroanalysis 2004, 16, 1236. (191) Michalska, A.; Maksymiuk, K. Talanta 2004, 63, 109. (192) Michalska, A.; Gałuszkiewicz, A.; Ogonowska, M.; Ocypa, M.; Maksymiuk, K. J. Solid State Electrochem. 2004, 8, 381. (193) Mousavi, Z.; Bobacka, J.; Ivaska, A. Electroanalysis 2005, 17, 1609. (194) Va´zquez, M.; Bobacka, J.; Luostarinen, M.; Rissanen, K.; Lewenstam, A.; Ivaska, A. J. Solid State Electrochem. 2005, 9, 312. (195) Mousavi, Z.; Bobacka, J.; Lewenstam, A.; Ivaska, A. J. Electroanal. Chem. 2006, 593, 219. (196) Zanganeh, A. R.; Amini, M. K. Electrochim. Acta 2007, 52, 3822. (197) Va´zquez, M.; Bobacka, J.; Ivaska, A. J. Solid State Electrochem. 2005, 9, 865, and 2006, 10, 1012 (erratum). (198) Erso¨z, A.; Gavalas, V. G.; Bachas, L. G. Anal. Bioanal. Chem. 2002, 372, 786. (199) Volf, R.; Kra´l, V.; Hrdlicˇka, J.; Shishkanova, T. V.; Broncova´, G.; Krondak, M.; Gro¨tschelova´, M.; Sˇt’astny´, M.; Kroulı´k, J.; Valı´k, M.; Mateˇjka, P.; Volka, K. Solid State Ionics 2002, 154-155, 57. (200) Bendikov, T. A.; Harmon, T. C. J. Chem. Educ. 2005, 82, 439. (201) Bendikov, T. A.; Harmon, T. C. Anal. Chim. Acta 2005, 551, 30. (202) Sabah, S.; Aghamohammadi, M.; Alizadeh, N. Sens. Actuators, B 2006, 114, 489. (203) Sjo¨berg-Eerola, P.; Bobacka, J.; Sokalski, T.; Mieczkowski, J.; Ivaska, A.; Lewenstam, A. Electroanalysis 2004, 16, 379. (204) Prakash, R.; Srivastava, R. C.; Pandey, P. C. J. Solid State Electrochem. 2002, 6, 203. (205) Zhang, X.; Ogorevc, B.; Wang, J. Anal. Chim. Acta 2002, 452, 1. (206) Leopold, S.; Herranen, M.; Carlsson, J.-O.; Nyholm, L. J. Electroanal. Chem. 2003, 547, 45. (207) Lindfors, T.; Ivaska, A. J. Electroanal. Chem. 2002, 531, 43. (208) Lindfors, T.; Ivaska, A. J. Electroanal. Chem. 2002, 535, 65. (209) Mousavi, M. F.; Shamsipur, M.; Riahi, S.; Rahmanifar, M. S. Anal. Sci. 2002, 18, 137. (210) Karami, H.; Mousavi, M. F. Talanta 2004, 63, 743. (211) Vinokurov, I. A.; Kankare, J. J. Electroanal. Chem. 2003, 543, 101. (212) Shoji, E.; Freund, M. S. J. Am. Chem. Soc. 2002, 124, 12486. (213) Bobacka, J.; Lewenstam, A.; Ivaska, A. J. Electroanal. Chem. 2001, 509, 27. (214) Lai, C.-Z.; Fierke, M. A.; Stein, A.; Bu¨hlmann, P. Anal. Chem. 2007, 79, 4621. (215) Purves, R. D. Microelectrode Methods for Intracellular Recording and Ionophoresis; Academic Press: London, 1981. (216) Amman, D. Ion-selectiVe microelectrodes: Principles, Design and Applications; Springer-Verlag: Berlin, 1986. (217) Takahashi, A.; Camacho, P.; Lechleiter, J.; Herman, B. Physiol. ReV. 1999, 79, 1089. (218) Voipio, J. In pH and Brain Function; Kaila, K., Ransom, B. R., Eds.; Wiley-Liss: New York, 1998; p 95. (219) Deveau, J. S. T.; Lindinger, M. I.; Grodzinski, B. Biol. Procedures Online 2005, 7, 31. (220) Deveau, J. S. T.; Lindiger, M. I. U.S. Pat. Appl. 2003-345409 20030116, 2004. (221) Bard, A. J., Mirkin, M. V., Eds.; Scanning Electrochemical Microscopy; Marcel Dekker: New York, 2001. (222) Horrocks, B. R.; Mirkin, M. V.; Pierce, D. T.; Bard, A. J.; Nagy, G.; Toth, K. Anal. Chem. 1993, 65, 1213. (223) Toth, K.; Nagy, G.; Horrocks, B. R.; Bard, A. J. Anal. Chim. Acta 1993, 282, 239. (224) Klusmann, E.; Schultze, J. W. Electrochim. Acta 2003, 48, 3325. (225) Liu, B.; Cheng, W.; Rotenberg, M.; Mirkin, M. V. J. Electroanal. Chem. 2001, 500, 590. (226) Csoka, B.; Kovacs, B.; Nagy, G. Biosens. Bioelectron. 2003, 18, 141. (227) Gyurcsa´nyi, R. E.; Nyba¨ck, A.-S.; To´th, K.; Nagy, G.; Ivaska, A. Analyst 1998, 123, 1339. (228) Mathieu, E.; Schulte, A.; Mann, S.; Guntram, J.; Dietzel, I. D.; Schumann, W. Anal. Chem. 2004, 76, 3682. (229) Gyetvai, G.; Sundblom, S.; Nagy, L.; Ivaska, A.; Nagy, G. Electroanalysis 2007, 19, 1116. (230) Lin, C.-J.; Du, R.-G.; Nguyen, T. Corrosion 2000, 56, 41. (231) Ganjali, M. R.; Daftari, A.; Nourozi, P.; Salavati-Niasari, M. Anal. Lett. 2003, 36, 1511. (232) Timms, W. A.; Hendry, M. J. Ground Water Monit. Rem. 2004, 24, 67. (233) Wang, J. J.; Bishop, P. L. EnViron. Technol. 2005, 26, 381. (234) So¨dergård, M.; Csoka, B.; Nagy, G.; Ivaska, A. Anal. Lett. 2003, 36, 2909. (235) Malon, A.; Vigassy, T.; Bakker, E.; Pretsch, E. J. Am. Chem. Soc. 2006, 128, 8154. (236) Lindner, E.; Buck, R. P. Anal. Chem. 2000, 72, 336A.

Potentiometric Ion Sensors (237) Wygladacz, K.; Durnas, M.; Parzuchowski, P.; Brzozka, Z.; Malinowska, E. Sens. Actuators, B 2003, 95, 366. (238) Zine, N.; Bausells, J.; Ivorra, A.; Augilo´, J.; Zabala, M.; Teixidor, F.; Masalles, C.; Vin˜as, C.; Errachid, A. Sens. Actuators, B 2003, 91, 76. (239) Kloock, J. P.; Moreno, L.; Bratov, A.; Huachupoma, S.; Xu, J.; Wagner, T.; Yoshinobu, T.; Ermolenko, Y.; Vlasov, Y. G.; Scho¨ning, M. J. Sens. Actuators, B 2006, 118, 149. (240) Tymecki, L.; Zwierkowska, E.; Glab, S.; Koncki, R. Sens. Actuators, B 2003, 96, 482. (241) Eftekhari, A. Anal. Lett. 2001, 34, 1087. (242) Lenihan, J. S.; Ball, J. C.; Gavalas, V. G.; Lumpp, J. K.; Hines, J.; Daunert, S.; Bachas, L. G. Anal. Bioanal. Chem. 2007, 387, 259. (243) Tymecki, Ł.; Zwierkowska, E.; Koncki, R. Anal. Chim. Acta 2004, 526, 3. (244) Liao, W.-Y.; Chou, T.-C. Anal. Chem. 2006, 78, 4219. (245) Kounaves, S. P.; Buehler, M. G.; Hech, M. H.; West, S. ACS Symp. Ser. 2002, 811 (EnVironm. Electrochem.), 306. (246) Paciorek, R.; Bieganowski, P.; Maj-Zurawska, M. Sens. Actuators, B 2005, 108, 840. (247) Guenat, O.; Dufour, J.-F.; Berdondin, L.; Morf, W.; de Rooij, N. F.; van der Wal, P. D.; Koudelka-Hep, M. Chem. Sens. 2004, 20, 728. (248) Guenat, O. T.; Dufour, J.-F.; van der Wal, D, P.; Morf, W. E.; de Rooij, N. F.; Koudelka-Hep, M. Sens. Actuators, B 2005, 105, 65. (249) Mourzina, Y. G.; Schubert, J.; Zander, W.; Legin, A.; Vlasov, Y. G.; Schoning, M. J. Electrochim. Acta 2001, 47, 251. (250) Lynch, A.; Diamond, D.; Leader, M. Analyst 2000, 125, 2264. (251) Lukasz, T.; Glab, S.; Koncki, R. Sensors 2006, 6, 390. (252) Slowinska, K.; Majda, M. J. Solid State Electrochem. 2004, 8, 763. (253) Elsen, H. A.; Slowinska, K.; Hull, E.; Majda, M. Anal. Chem. 2006, 78, 6356. (254) Huller, J.; Pham, M. T.; Howitz, S. Sens. Actuators, B 2003, 91, 17. (255) Hassan, S. S. M.; Sayour, H. E. M.; Al-Meherezi, S. S. Anal. Chim. Acta 2007, 581, 13. (256) Ibanez-Garcia, N.; Mercader, M. B.; da Rocha, Z. M.; Seabra, C. A.; Gongora-Rubio, M. R.; Chamarro, J. A. Anal. Chem. 2006, 78, 2985. (257) Liao, W.-Y.; Weng, C.-H.; Lee, G.-B.; Chou, T.-C. Lab Chip 2006, 6, 1362. (258) Miller, A. J.; Cookson, S. J.; Smith, S. J.; Wells, D. M. J. Exp. Botany 2001, 52, 541. (259) Dierkes, P. W.; Neumann, S.; Muller, A.; Gunzel, D.; Schlue, W.R. New Trends Electrochem. Technol. 2003, 2, 526. (260) Dierkes, P. W.; Neumann, S.; Klees, G.; Schlue, W.-R. Electrochim. Acta 2003, 48, 3373. (261) Bauselis, J.; Errachid, A.; Zine, N. Proc.sElectrochem. Soc. 2003, 2003-1, 48. (262) Schwiening, C. J. Soc. Exp. Biol. Semin. Ser. 1999, 68, 1. (263) Garnett, T. P.; Shabala, S. N.; Smethurst, P. J.; Newman, I. A. Funct. Plant Biol. 2003, 30, 1165. (264) Verschurren, P. G.; van der Baan, J. L.; Blaauw, R.; de Beer, D.; van den Heuvel, J. C. Fresenius J. Anal. Chem. 1999, 364, 595. (265) Shabala, L.; Budde, B.; Ross, T.; Siegumfeldt, H.; Jakobsen, M.; McMeekin, T. Appl. EnViron. Microbiol. 2002, 68, 1794. (266) Wells, D. M.; Miller, A. J. Plant Soil 2000, 221, 103. (267) Zhang, X.; Fakler, A.; Spichiger, U. E. Electroanalysis 1998, 10, 1174. (268) Guenzel, D.; Schlue, W.-R. BioMetals 2002, 15, 237. (269) Berger, C. E. M.; Horrocks, B. R.; Datta, H. K. Electrochim. Acta 1999, 44, 2677. (270) Kang, T. M.; Markin, V. S.; Hilgemann, D. W. J. Gen. Physiol. 2003, 121, 325. (271) Knowles, A.; Shabala, S. J. Membr. Biol. 2004, 202, 51. (272) Garnett, T. P.; Shabala, S. N.; Smethurst, P. J.; Newman, I. A. Plant Soil 2001, 236, 55. (273) Shabala, S.; Shabala, L. Physiol. Plant. 2002, 114, 47. (274) Shabala, L.; Ross, T.; Newman, I.; McMeekin, T.; Shabala, S. J. Microbiol. Methods 2001, 46, 119. (275) Branka, Z. Ann. N. Y. Acad. Sci. 2005, 1048, 487. (276) O’Donnell, M. J.; Rheault, M. R. J. Exp. Biol. 2005, 208, 93. (277) Tinkilin, N.; Cubuk, O.; Isidak, I. Anal. Chim. Acta 2002, 452, 29. (278) Fedirko, N.; Svichar, N.; Chesler, M. J. Neurophysiol. 2006, 96, 919. (279) Ceresa, A.; Radu, A.; Peper, S.; Bakker, E.; Pretsch, E. Anal. Chem. 2002, 74, 4027. (280) Bedlechowicz, I.; Maj-Z˙ urawska, M.; Sokalski, T.; Hulanicki, A. J. Electroanal. Chem. 2002, 537, 111. (281) Vigassy, T.; Gyurcsa´nyi, R. E.; Pretsch, E. Electroanalysis 2003, 15, 375.

Chemical Reviews, 2008, Vol. 108, No. 2 351 (282) Radu, A.; Telting-Diaz, M.; Bakker, E. Anal. Chem. 2003, 75, 6922. (283) Malon, A.; Radu, A.; Qin, W.; Qin, Y.; Ceresa, A.; Maj-Zurawska, M.; Bakker, E.; Pretsch, E. Anal. Chem. 2003, 75, 3865. (284) Vigassy, T.; Huber, C. G.; Wintringer, R.; Pretsch, E. Anal. Chem. 2005, 77, 3966. (285) Bereczki, R.; Taka´cs, B.; Gyurcsa´nyi, R. E.; To´th, K.; Nagy, G.; Langmaier, J.; Lindner, E. Electroanalysis 2006, 18, 1245. (286) Simonis, A.; Lu¨th, H.; Wang, J.; Scho¨ning, M. J. Sens. Actuators, B 2004, 103, 429. (287) Wilburn, J. P.; Ciobanu, M.; Buss, N. I.; Franceschetti, D. R.; Lowy, D. A. Anal. Chim. Acta 2004, 511, 83. (288) Vonau, W.; Enseleit, U.; Gerlach, F.; Herrmann, S. Electrochim. Acta 2004, 49, 3745. (289) Blaz˘ , T.; Migdalski, J.; Lewenstam, A. Analyst 2005, 130, 637. (290) Ha, J.; Martin, S. M.; Jeon, Y.; Yoon, I. J.; Brown, R. B.; Nam, H.; Cha, G. S. Anal. Chim. Acta 2005, 549, 59. (291) Kisiel, A.; Marcisz, H.; Michalska, A.; Maksymiuk, K. Analyst 2005, 130, 1655. (292) Mamin´ska, R.; Dybko, A.; Wro´blewski, W. Sens. Actuators, B 2006, 115, 552. (293) Lugtenberg, R. J. W.; Egberink, R. J. M.; van den Berg, A.; Engbersen, J. F. J.; Reinhoudt, D. N. J. Electroanal. Chem. 1998, 452, 69. (294) Heng, L. Y.; Hall, E. A. H. Anal. Chem. 2000, 72, 42. (295) Heng, L. Y.; Hall, E. A. H. Anal. Chim. Acta 2000, 403, 77. (296) Heng, L. Y.; Hall, E. A. H. Electroanalysis 2000, 12, 178. (297) Heng, L. Y.; Hall, E. A. H. Electroanalysis 2000, 12, 187. (298) Malinowska, E.; Gawart, L.; Parzuchowski, P.; Rokicki, G.; Brzo´zka, Z. Anal. Chim. Acta 2000, 421, 93. (299) Heng, L. Y.; Chern, L. H.; Ahmad, M. Sensors 2002, 2, 339. (300) Qin, Y.; Peper, S.; Bakker, E. Electroanalysis 2002, 14, 1375. (301) Peper, S.; Ceresa, A.; Qin, Y.; Bakker, E. Anal. Chim. Acta 2003, 500, 127. (302) Qin, Y.; Peper, S.; Radu, A.; Ceresa, A.; Bakker, E. Anal. Chem. 2003, 75, 3038. (303) Edelmann, K.; Reiche, A.; Sandner, B.; Marstalerz, J.; Mu¨ller, H. Polymer 2005, 46, 407. (304) Qin, Y.; Bakker, E. Anal. Chem. 2003, 75, 6002. (305) Pu¨ntener, M.; Fibbioli, M.; Bakker, E.; Pretsch, E. Electroanalysis 2002, 14, 1329. (306) Qin, Y.; Bakker, E. Anal. Chem. 2004, 76, 4379. (307) Pu¨ntener, M.; Vigassy, T.; Baier, E.; Ceresa, A.; Pretsch, E. Anal. Chim. Acta 2004, 503, 187. (308) Le Goff, T.; Braven, J.; Ebdon, L.; Scholefield, D. Anal. Chim. Acta 2004, 510, 175. (309) Bereczki, R.; Gyurcsa´nyi, R. E.; A Ä gai, B.; To´th, K. Analyst 2005, 130, 63. (310) Trojanowicz, M.; Opara, W. Talanta 2002, 56, 213. (311) Shin, J. H.; Lee, H. L.; Cho, S. H.; Ha, J.; Nam, H.; Cha, G. S. Anal. Chem. 2004, 76, 4217. (312) Jime´nez-Morales, A.; Galva´n, J. C.; Aranda, P. Electrochim. Acta 2002, 47, 2281. (313) Oosaki, S.; Kawai, Y.; Yajima, S.; Kimura, K. Anal. Sci. 2004, 20, 1165. (314) Wendzinski, F.; Borchardt, Romberg, B.; Knoll, M. Anal. Lett. 2003, 36, 21. (315) Boswell, P. G.; Bu¨hlmann, P. J. Am. Chem. Soc. 2005, 127, 8958. (316) Stefanova, O. K. Elektrokhimia 1979, 15, 1707. (317) Shultz, M. M.; Stefanova, O. K.; Mokrov, S. B.; Mikhelson, K. N. Anal. Chem. 2002, 74, 510. (318) Nevins, Buchanan, S. A.; Balogh, L. P.; Meyerhoff, M. E. Anal. Chem. 2004, 76, 1474. (319) Peper, S.; Telting-Diaz, M.; Almond, P.; Albrecht-Schmitt, T.; Bakker, E. Anal. Chem. 2002, 74, 1327. (320) Malon, A.; Bakker, E.; Pretsch, E. Anal. Chem. 2007, 79, 632. (321) Le Goff, T.; Marsh, J.; Braven, J.; Ebdon, L.; Scholefield, D. Green Chem. 2002, 4, 486. (322) Lindner, E.; Gyurcsa´nyi, R. E.; Buck, R. P. Electroanalysis 1999, 11, 695. (323) Morf, W. E.; Badertscher, M.; Zwickl, T.; De Rooij, N. F.; Pretsch, E. J. Electroanal. Chem. 2002, 526, 19. (324) Makarychev-Mikhailov, S.; Shvarev, A.; Bakker, E. J. Am. Chem. Soc. 2004, 126, 10548.

CR068100W

Chem. Rev. 2008, 108, 352−366

352

Amperometric Gas SensorssA Review Joseph R. Stetter* SRI International 333 Ravenswood Avenue, Menlo Park, California 94025-3493

Jing Li NASA Ames Research Center, Moffett Field, California 94035-1000 Received September 4, 2007

Contents 1. Introduction 2. Characteristics of Amperometric Gas Sensors 2.1. Electrochemical Gas Sensors 2.2. Amperometry 2.3. Amperometric Gas Sensors 2.4. Theory of the Limiting Current 2.5. Structure and Geometry of an AGS 2.5.1. Gas Membrane 2.5.2. Electrolyte 2.5.3. Electrodes 2.6. Electrode Reactions 2.7. Analytes 3. Survey of Old and New Amperometric Gas Sensors 3.1. Working Electrode and Sensor Design 3.1.1. Clark Sensor 3.1.2. Diffusion Electrodes 3.2. Electrolytes 3.2.1. Nonaqueous Electrolyte 3.2.2. Solid Polymer Electrolytes 3.2.3. Ceramic Solid Electrolytes at Elevated Temperature 3.3. MEMS and Nanotechnology 3.4. Sensor Array 4. AGSs For Practical Applications 4.1. Industry 4.2. Environment 4.3. Defense and Homeland Security 4.4. Medical/Biomedical 5. Summary 6. Acknowledgment 7. References

352 352 352 354 354 355 355 355 356 356 357 357 357 357 357 359 360 360 360 361 361 362 363 364 364 364 365 365 365 365

1. Introduction The amperometric gas sensor, or AGS, belongs to a large and important class of electrochemical gas sensors, and they play an ever-increasing role in environmental monitoring, medical and health applications, industrial safety, security, surveillance, and the automotive industry. The fundamental * To whom correspondence should be addressed. E-mail: joseph.stetter@ sri.com.

principles and constituents of the AGS sensor device as well as some important events in the design and development of the AGS technology are reviewed herein. The modern AGS is based on a series of scientific discoveries, and it is has contributed much to society in its rich history. Amperometric gas sensors (AGSs) are best divided by the temperature at which they operate and can be represented by two well-known commercial gas sensors that include the ambient-temperature liquid electrolyte gas sensors used for medical and industrial hygiene applications and the hightemperature solid electrolyte O2 sensor used in automotive, process, and stack gas applications. The AGS can operate at low or high temperatures, ranging from below freezing to over 1000 °C, and the materials of construction, including the electrolyte, change significantly. The choice of materials can result in totally different designs. The room-temperature AGS uses liquid or polymer electrolytes and is the system of choice for industrial hygiene, health, safety, and medical applications, while the high-temperature AGS using solid electrolytes has been primarily developed for use in automotive exhaust and stack gas process monitoring. This review focuses more on the low-temperature AGS, but some discussions are included on the high-temperature sensors. The fact that the AGS serves many important practical applications speaks to the superb simultaneous combination of features that can be realized with this approach including: low power, low cost, small size, selectivity, stability, sensitivity, and sometimes fast response time. Given these combined characteristics, the AGS is often described as providing relatively high analytical performance at modest cost.

2. Characteristics of Amperometric Gas Sensors 2.1. Electrochemical Gas Sensors Electrochemical gas sensors can be divided into three main classes according to the operating principle: amperometric, potentiometric, and conductometric sensors. Many sensor types exist within each class. We will not cover potentiometric techniques or conductometric techniques for gas sensors but restrict discussions to amperometric sensors for the gas-phase analytes. In the AGS, the current generated by reaction of an analyte at an electrode is measured as the sensor signal and can be measured at a fixed or variable electrode potential, although it is very common at a fixed applied potential. The reaction rate, reflected by the current at the sensing electrode, occurs at a thermodynamically

10.1021/cr0681039 CCC: $71.00 © 2008 American Chemical Society Published on Web 01/18/2008

Amperometric Gas SensorssA Review

Dr. Joseph R. Stetter obtained a Ph.D. in Physical Chemistry from the University at Buffalo (SUNY) in 1975. In the 1970s, Dr. Stetter was Director of Chemical Research at the Energetics Sciences Division of Becton Dickinson and Company where he developed the first diffusion-type electrochemical CO sensors; the earliest diffusion CO dosimeters; solidstate gas sensors for CO, NOx, SO2, and other toxic gases; and an electrochemical hydrazine sensor still in use by NASA. While at the Argonne National Laboratory in Chicago, IL, in the 1980s, he led the development of the first integrated and operational “sensor-array-based” instrument with pattern recognition (now called electronic nose). In the 1980s, Dr Stetter founded TRI (Transducer Research, Inc.), where he developed portable instruments and sensors for CO and CO2, end-ofservice filter indicators, chlorinated hydrocarbon sensors, NOx sensors, personal protection instruments, and low-cost effective protection equipment for human health and the environment. In the 1990s, he sold TRI and became Professor of Chemistry at the Illinois Institute of Technology, started a sensor research group, and founded the International Center for Sensor Research and Engineering at IIT, mentoring both M.S. and Ph.D. students. He founded Transducer Technology, Inc. (TTI), a startup company focusing on nanotechnology enabled sensors and instruments in 1999. In 2007, TTI merged with KWJ Engineering, Inc., of Newark, CA, making nanosensors for health, safety, and process control applications. Recently, Dr. Stetter is Director of the Microsystems Innovation Center for SRI International (Menlo Park, CA). R&D focuses on new sensors, unique structures/materials, artificial senses, chem/biosensors, novel MEMS for drug and vaccine delivery, vacuum microelectronics including microelectron and ion sources, and micro/nanostructures and bio-MEMS. Dr. Stetter has published more than 100 refereed scientific and technical articles and has more than 25 domestic and foreign patents. He has served as Chairman of the Electrochemical Society Sensor Division and served on the boards of national and international technical societies. He has organized national and international scientific meetings and symposia in his field and serves as editor and reviewer for scientific and engineering journals. His awards include three IR-100 Awards for new product development; the Federal Laboratory Consortium Special Award for Excellence in Technology Transfer; the Argonne National Laboratory Inventor’s award; the Technology Management Association of Chicago’s 2002 “Entrepreneur-of-the-Year” award; and two NASA New Technology Awards. He is on the board of directors for several start-up companies.

determined potential for any given reaction and, when operated under appropriate diffusion-limited conditions, is simply proportional to the concentration of the analyte. The relationship between current and concentration is linear, typically over 3 orders of magnitude, and measurements with high sensitivity (ppm and ppb) are possible with excellent measurement accuracy under constant potential conditions. Faraday’s Law is often applied to the observed currents, and the design of the sensor is chosen to control many kinetic factors including mass transfer of the analyte to the electrode as well as the electrocatalytic activity of the electrode material. The geometry and dimensions of the sensor device have a profound effect on the AGS analytical performance, including the observed sensitivity, selectivity, response time, and signal stability.

Chemical Reviews, 2008, Vol. 108, No. 2 353

Dr. Jing Li received her Ph.D. in Materials Science in 1996 from University of Utah in Salt Lake City. She is currently a Principal Investigator in NASA Ames Nanotechnology Branch. She leads the effort and has developed a space-qualified nanochemical sensor unit for flight demonstration in a satellite that has been launched and operated successfully since March 2007. The nanotechnology-based chemical sensor platform that she developed has drawn interest from other NASA centers, many industry partners, and government Labs. She is a principle investigator (PI) on the Nano ChemSensor Unit project, which won 2007 NASA Ames Center Honor Award. She worked with a team at NASA on nanoelectronic devices that resulted in a NASA TGIR Award (2002). She is an internationally known expert in the field of chemical sensors. Her research focus is on the chemical sensor development utilizing the nanostructured materials and a sensor array with signal processing techniques for intelligent sensor systems. Prior to joining NASA Ames, she was a lead scientist at Cyrano Science, Inc., where she developed polymer-based chemical sensors for an electronic nose instrument. Dr. Jing Li worked at Pacific Northwest National Lab as a DOE postdoc fellow on the chemical sensor development for waste tank monitoring. She has extensive experience of both research and commercialization in developing chemical sensors and intelligent sensing systems, which includes the following: (1) investigating the sensing materials such as nanostructured carbon nanotube and metal, metal oxides nanowires, and their intercalation in polymers as composites; (2) studying the sensing mechanisms in scope of the charge transfer (first and secondary) interaction and diffusion/partition process between sensing materials and analytes characterized by electrical, electrochemical, and spectroscopic techniques such as UV−vis, FTIR, Raman, work function measurements with Kelvin probe, and adsorption/absorption measurement with quartz crystal microbalance and Micromeritics; and (3) developing the signal-processing techniques (i.e., pattern recognition) for a chemical sensor array to be used as an intelligent system for digitizing the chemical information. She is currently a vice chair for Sensor Division in the Electrochemical Society. Her research interests are chemical sensors and intelligent sensing systems for space applications and terrestrial applications. Table 1. Comparison of Amperometric and Potentiometric Sensors electrochemical class

sensor signal vs [analyte]

amperometric, AGS potentiometric

i ) kP E ) Ec + k ln P

principle kinetic, Faraday law thermodynamic, Nernst law

In a potentiometric electrochemical class of sensor, the open-circuit potential between two electrodes is monitored, and this potential is typically proportional to the logarithm of the concentration of analyte. The relationship is often expressed by the Nernst equation, as shown in Table 1. A potentiometric sensor can measure analyte concentrations over a very wide range, often more than 10 decades, and the sensing principle is thermodynamic, i.e., it assumes that all reactions relating to the sensing are at equilibrium. In potentiometry, the chemical and diffusion processes must be at equilibrium conditions in the sensor for a thermody-

354 Chemical Reviews, 2008, Vol. 108, No. 2

Stetter and Li

electrode and a counterelectrode, and the electrolyte solution in which the two electrodes are immersed. Missing from this for the moment is an interface to enable a gaseous analyte to contact the working electrode, but this is discussed later. When applying a proper potential between the two electrodes, electroactive species in the electrolyte solution will participate in electrochemical reactions on the surfaces of each of the electrodes. A reduction reaction occurs at the cathode and can be expressed as shown in eq 1, and at the same time, an oxidation reaction takes place at the anode and can be written generally as shown in eq 2:

Figure 1. Illustration of amperometric sensor with two-electrode configuration for liquids.

namically accurate signal to be observed, while the amperometric sensors rely on Faraday’s law and a dynamic reaction achieving a stready-state condition in the sensor. As may be implied, the amperometric sensor signal will get smaller with the size of the electrode and the rate of analyte reacting at the electrode surface, while the potentiometric sensor assumes a thermodynamic potential independent of size of the electrode. This situation becomes interesting in light of nanostructures wherein the thermodynamic potential is characteristic of aggregates of atoms, while the amperometric sensor reation rate is typically enhanced by the high surface areas afforded by nanostructured electrode materials.

2.2. Amperometry Amperometry is a conventional electroanalytical technique that encompasses coulometry, voltammetry, and constantpotential techniques and is widely used to identify and quantify electroactive species in the liquid or gas phase. For liquid-phase analytes, the electrodes and analytes are immersed in a common electrolyte solution. In contrast, application of amperometry to gas-phase analytes involves a unique gas/liquid/solid boundary (analyte-electrolyteelectrode) and an interfacial transport process that frequently controls the response characteristics and analytical performance of the AGS. A major consideration in the specific design of the sensor device is to choose a geometry and porosity/composition of materials to enable the gaseous analyte to effectively and efficiently transport to the interface of the electrode/electrolyte where a fast, reversible, redox reaction can occur. This interfacial redox reaction provides the charge-transfer reaction of the analyte and the analyte’s characteristic sensor signal.

2.3. Amperometric Gas Sensors Amperometric gas sensors are sometimes called voltammetric, microfuel cell, polarographic, amperostatic, potentiostat, limiting current, or other sensor names.1,2 The most descriptive name is “amperometric”, and this refers to either constant-potential amperometry or variable-potential amperometry. A simple amperometric cell with two-electrode configuration is illustrated in Figure 1 wherein the electrochemical system consists of two electrodes, i.e., a working

O1 + ne ) R1

(1)

R2 ) O2 + me

(2)

The cathode reaction takes electrons from the cathode and combines them with an oxidized species, O1, to produce a reduced species, R1. At the anode, a reduced species, R2, reacts to produce an oxidized species, O2, and electrons are given to the anode. The electrons on the anode or lack of electrons on the cathode would build up a charge on the electrode surface, except that the electrodes are conductors and are connected in an external circuit in which the electrons flow. This is an illustration of Faraday’s Law in that an exact number of electrons are generated or used per each analyte molecule and so the charge (the number of electrons or Coulombs of charge) is exactly related to the number of analyte molecules reacting in the system and the current (electron flow) at the electrode is exactly related to the rate of the electrochemical reactions (current is Coulombs/s). In the AGS, typically the current under diffusion-limited reaction conditions is directly related to the rate of electrode reaction taking place on the electrode surface and, hence, as discussed in more detail below, the rate of the electrode reaction is proportional to the concentration of reactant, i.e., the analyte. Consequently, the electrical charge or current can be used as a sensor signal and will be related to the concentration of the analyte in an AGS. The common characteristic of all AGSs is that measurements are made by recording the current in the electrochemical cell between the working electrodes and counterelectrodes over time as a function of the analyte concentration. When this experiment is performed at a fixed potential controlled by a potentiostatic circuit, the technique should be properly called “constant-potential amperometry.” The AGS produces a current when exposed to a gas/vapor containing an electroactive analyte because the analyte diffuses into the electrochemical cell, to the working electrode surface, and thereon participates in an electrochemical reaction that either produces or consumes electrons (i.e., a redox reaction). Selectivity to certain gaseous analytes is an important consideration for sensing applications. For the AGS, selectivity can be realized by optimizing the electrode material to facilitate or catalyze only selected reactions or by control of the sensing electrode potential to select an applied bias that thermodynamically favors either the oxidation reaction or the reduction reaction for a particular analyte. Noble metals such as gold and platinum, mostly in porous or high surface area designs, have been used for the working or sensing electrode in the AGS because of their excellent chemical stability in electrolyte solutions and high electrocatalytic activity toward analytes like CO, H2S, O2, Cl2, and NO. The applied potential between working and reference electrodes set by the potentiostat fixes the thermodynamic operating

Amperometric Gas SensorssA Review

Chemical Reviews, 2008, Vol. 108, No. 2 355

potential for the working electrode reaction and is critical for both the observed sensitivity and the selectivity of the sensor. The selection of operating potential provides an effective way to change or optimize the performance of the sensor.

2.4. Theory of the Limiting Current The observed current in the AGS can depend on many factors, and the resultant mathematical expression relating the current, potential, and concentration can be complex. There are two special operating conditions, called limiting conditions, that can be used to simplify the resulting expression for relating the rate of an electrode reaction (rr) and the rate of mass transport, e.g., by diffusion (rd), of the reactant to the electrode surface. These conditions include the situation where rr , rd (eq 1) and the situation where rr . rd (eq 2). Both cases are important for sensor applications, and in both these cases, the electrode process can be easily understood and the resulting expression can be simplified. The current obtained at the two rate-limiting conditions is called the limiting current. In the first case, where rr , rd, i.e., the rate of reaction at the electrode surface is the rate-limiting step, the limiting current is controlled by the rate of the electrode kinetics. In this case, the analyte reaches the surface much faster than it is reacted, and so the concentration at the electrode surface is the same as in the bulk of the solution and the gas surrounding the electrode and the expression for the current will take the form of eq 3,

ilim ) nFkAC exp(RnFE/RT)

(3)

where k represents the standard rate constant, F is the faraday constant, R is the gas constant, T is the Kelvin temperature, A is the area of the electrode, C is the concentration, n represent the number of electrons per molecule reacting, and R and E are the transfer coefficient and overvoltage of the electrode reaction, respectively. If all conditions are held constant, the limiting current is proportional to the concentration (C) of analyte with an exponential temperature coefficient, and the derivation of this expression can be found in many textbooks.3,4 The limiting current that is obtained in this kinetically controlled electrochemical region, however, is not always stable because of the degradation of the electrode’s catalytic activity and can be problematic for a practical AGS. This drawback can be overcome by operating the AGS in the region of diffusion control rather than the electrode’s kinetic control. Under this condition, where rr . rd, the rate of diffusion of the reactant toward the electrode surface is much slower than the rate of reaction and now the limiting current is controlled by the diffusion of analyte. In this case, the concentration of the analyte at the electrode surface is virtually zero and every molecule of analyte that reaches the surface is immediately reacted. Under these conditions, the rate of electrode reaction may be limited by the rate of diffusion through a gas membrane or a capillary that is placed somewhere between the bulk analyte sample (e.g., the air containing the anayte) and the catalyst layer of the electrode. On the basis of Fick’s diffusion law, the limiting current, ilim, should now be governed by the rate at which the analyte transports to the surface, and this can usually be written as shown by eq 4,

ilim ) k[CO]gas

(4)

where the current is directly proportional to the gaseous concentration in some convenient units like ppmv (parts per million by volume). The current can also be limited by the rate of diffusion across the diffusion layer in solution, and an expression similar to eq 4 is also obtained. Figure 2 illustrates the two limiting conditions described above wherein the concentration of the analyte is plotted vs the distance from the electrode surface. When the current is limited by the rate of the electrode reaction, the analyte supply is much faster than its consumption and the analyte concentration is almost the same in the bulk and at the electrode surface. For a mass transport (diffusion) limited electrode reaction, the analyte supply is much slower than its consumption by reaction at electrode surface, the analyte concentration falls to zero at electrode surface, and the current is now limited by the supply of analyte. This condition is often preferred by the designer of the AGS since the physical control of the analyte supply by diffusion allows one to build a more stable sensor with the square root temperature dependence of the signal. When operating the AGS using a potentiostat, the sensor can be operated at an electrochemical potential where the reaction is facilitated, and in this case, the magnitude of the sensor signal is practically independent of the electrode potential. In theory, the limiting current region can be achieved in any case when the rate-limiting step is a step prior to electron transfer. Typically, an electrode with relatively high electrocatalytic activity is sought, and the benefits of a stable long-lasting sensor are realized.

2.5. Structure and Geometry of an AGS Figure 3 illustrates the design of a typical three-electrode AGS for CO sensing having a working electrode (WE) reaction [CO + H2O ) CO2 + 2H+ + 2e-] and counterelectrode (CE) reaction [1/2O2 + 2H+ + 2e- ) H2O] and reference electrode (RE) embedded within the cell. The WE, CE, and RE are in electrolytic contact (all touch the electrolyte), and the overall cell reaction is CO + 1/2O2 ) CO2. The common feature of the typical gas sensor is an interface at the working electrode that facilitates transport of the gas to the WE/electrolyte interface. This inlet can be simple diffusion or aided by a small air pump that transports the sample to the backside of a gas porous membrane through which the analyte diffuses/permeates. The gas membrane can be used to control the gas flow into the sensor and it can aid selectivity, allowing only the analyte gas to pass as well as providing a barrier to prevent the leakage of the electrolyte from the interior of the sensor. The design and materials chosen for each part of the sensor are critical and determined, to a large part, by the application and the desired analytical performance. In the following paragraphs, we discuss alternative materials and designs that are used to provide different performance capabilities.

2.5.1. Gas Membrane Several types of porous and gas-permeable membranes exist and can be made of polymeric or inorganic materials. Most of them are commercially available very thin solid Teflon films, microporous Teflon films, or silicone membranes. Issues concerning the choice of membrane include permeability to the analyte of interest, ability to prevent electrolyte leakage, manufacturability, and the thickness and

356 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 2. Concentration distribution of analyte (A) for different current limit conditions.

Stetter and Li

β-alumina, and NASICON. For AGS applications, the electrolyte solution needs to support both counterelectrode and sensing electrode reactions, form a stable reference potential with the reference electrode, be compatible with the materials of construction, and be stable over long periods of time under various operating conditions. In some cases, hydrogels or an electrolyte inside a porous matrix are used to replace free liquid electrolytes to raise viscosity, lower evaporation rates, and resist leakage of the electrolyte from sensor devices. The polymers or hydrogels can prevent the evaporation of electrolyte during sensor fabrication, especially for microsensor devices where very small amounts of electrolyte are used. Solid electrolytes (polymers and hightemperature ionic materials) allow fabrication of a solid-state sensor, which avoids the danger of leakage of liquid electrolytes entirely. In order to get sufficient ionic conductivity, elevated operating temperatures may be required. High-temperature operation can be an advantage for operation in harsh environments, in an internal combustion engine control system, or for exhaust gas monitoring.

2.5.3. Electrodes

Figure 3. Illustration of amprometric gas sensor for CO with threeelectrode configuration.

durability of the membrane. For example, a semipermeable membrane composed of an acrylonitrile butadiene copolymer can be used to selectively detect the partial pressure of oxygen in blood samples by allowing oxygen transport and effectively preventing the transport of the other species present in the sample. In a number of cases, the rate of mass transport through the membrane controls the limiting current and, hence, the sensitivity of the AGS. An ideal gas membrane will have a constant permeability to the target gaseous analyte during sensor operation over a wide temperature range and possess mechanical, chemical, and environmental stability.

2.5.2. Electrolyte The role of the electrolyte is to transport charge within the sensor, contact all electrodes effectively, solubilize the reactants and products for efficient transport, and be stable chemically and physically under all conditions of sensor operation. The electrolyte is an ionically conducting medium that ionically transports charge within the cell. Examples of commonly used electrolytes include the following: aqueous electrolytes, such as sulfuric acid, sodium hydroxide, and potassium chloride; nonaqueous electrolytes, such as propylene carbonate with lithium perchlorate; and solid electrolytes, such as the polymer Nafion when wet and, for hightemperature sensors, yttria-stabilized zirconia or YSZ,

The electrode reaction, i.e., electron-transfer reaction at the WE, involves several steps,1,2 including adsorption of the analyte onto the surface, electroreaction, and desorption of products from the electrode surface. Typically, the working electrode is made from a noble metal, such as platinum or gold, which is capable of making a defined interface with the electrolyte in the cell and is porous to allow efficient diffusion of the gas phase to a large and reactive electrode/ electrolyte interface. The noble metals generally exhibit excellent stability under polarized potentials that may be corrosive to other metals. The noble metals are also excellent catalysts for many analyte reactions. Carbons, including graphite and glassy carbon, are also popular materials for working electrodes, especially for sensing bioanalytes since many forms of carbon are biocompatible. The selection of the WE material is, therefore, based on the electrochemical and electrocatalytic properties as well as its stability and manufacturability. The counterelectrode must also be stable in the electrolyte and efficiently perform the complementary half-cell reaction that is opposite of the analyte reaction.2 A Pt electrode is very often used as the CE in an AGS. In addition to the working electrodes and counterelectrodes, a reference electrode is also present when a potentiostat is used.1 The reference electrode must form a stable potential with the electrolyte, be compatible with the sensor manufacture, and, generally, not be sensitive to temperature (T), pressure (P), relative humidity (RH), or other contaminants or reactants in the sensor system. The RE must be able to maintain the WE at a constant electrochemical potential during the sensing application. An Ag/AgCl reference electrode, which shows good reversibility, is commonly used for this purpose. The other popular reference electrode is a Pt/air electrode, which is not a classical refernce electrode but is sometimes called a pseudo-reference electrode because, while it forms a stable potential, the potential is not that of a well-defined thermodynamic reaction and the electrode must be calibrated with an electrode of known potential. In a two-electrode system, a single electrode, called the auxiliary electrode, can function as both the RE and the CE for the purposes of a given analytical experiment.

Amperometric Gas SensorssA Review

Chemical Reviews, 2008, Vol. 108, No. 2 357

Table 2. Example Electrode Reactions for Aqueous Electrolyte Amperometric Gas Sensors target gas

electrode reaction

H2 CO O2 NO2 NO H2S SO2

H2 + 2OH- ) 2H2O + 2eCO + H2O ) CO2 + 2H+ + 2eO2 + 4H+ + 4e- ) 2H2O NO2 + 2H+ + 2e- ) NO + H2O NO + 2H2O ) NO32- + 4H+ + 3eH2S + 4H2O ) SO42- + 10H+ + 8eSO2 + 2H2O ) SO42- + 4H+ + 2e-

2.6. Electrode Reactions For an example of electrode electrochemical reactions, the typical room-temperature liquid electrolyte CO sensor is considered. The CO sensor has an H2SO4 electrolyte and three Pt electrodes, typically made with high-surface-area Pt powder. The fundamental process for the electrochemical reaction is the transfer of electrons from the reacting CO and H2O to the WE surface and the formation of some H+ to transport the charge in the cell and the formation of one additional oxygen-carbon chemical bond changing the valence of the carbon atom:

CO + H2O f CO2 + 2H+ + 2e-

(5)

In order to maintain charge neutrality in the electrochemical cell, a second reaction, the reduction of oxygen to water, will occur at the CE that uses the acid and electrons to form water:

/2O2 + 2H+ + 2e- f H2O

1

(6)

The WE current generated in the electrochemical cell is related to the rate of CO oxidation at the working electrode and is directly proportional to the rate of CO arriving at the working electrode. Upon completion of the electrochemical reaction, the products must desorb from the electrode surface. Following desorption, reaction products (CO2) diffuse away from the electrode area. If the products of the reaction are sensor poisons, the sensor lifetime or response characteristics may be severely limited. Thus, the choice of materials and methods for the electroanalytical processes are critical in the sensor design. Table 2 lists electrochemical reactions for gas analysis. However, the actual electrode reaction at the electrode surface is dependent on the nature of the electrode material, the electrolyte solution, the thermodynamic potential, the interface of the electrode-electrolyte, and, of course, the analyte. To be complete, each of these analyte reactions is balanced with a second half-cell reaction that occurs at the CE, and this was discussed in the electrolyte section. The counterreaction is important to consider because it can limit the overall performance of the resulting sensor. It is typically chosen to be a fast, reversible, and noninterfering reaction. This is not always easily achieved and has limited some successful commercialization of amperometric sensors.

2.7. Analytes Many gases can be measured with an appropriately designed AGS, such as H2, O2, CO, NO2, NO, O3, SO2, H2S, and organic vapors with electroactive functional groups likes alcohols or aldehydes. Each sensor can have a unique design and a different set of materials and geometries for mem-

branes, electrolytes, and electrodes in order to take advantage of chemical properties of a specific target analyte and survive under various operating conditions. This use of a variety of materials, unique geometries or structures, and different methods is what gives the sensors their analytical properties and allows them to serve diverse applications. It is because of the discovery of unique combinations of materials, geometries, and methods that the AGS keeps improving and the widespread use of the AGS for industrial safety, biochemistry, clinical chemistry, health and medicine, agriculture, food safety, environmental protection, automotive technology, space exploration, military threat detection, and process control continues to grow. The next section presents a historical perspective on the development of AGS technology and its path to becoming one of the most widely used gas-sensing devices.

3. Survey of Old and New Amperometric Gas Sensors In general, the AGS technology results from the basic understanding of electrochemistry arising from the works of Nernst, Faraday, and Heyrovsky as well as the understanding of mass transport and electrochemical reactions provided by Fick, Cottrell, Buttler-Volmer, and Levich. Their fundamental scientific advances allow us to make sensors and understand the many dynamic aspects of the observed AGS signals.3,4

3.1. Working Electrode and Sensor Design 3.1.1. Clark Sensor The earliest successful AGS was developed by Clark et al. in 1953 for oxygen determination.5 The original Clark sensor was first introduced to measure the oxygen content in blood samples, and the major innovation was to cover the shiny platinum button WE cathode with cellophane. Unlike previously designed O2 sensors, which employed a bare platinum electrode, the Clark sensor showed enhanced selectivity for oxygen due to the presence of the cellophane membrane, which allowed O2 to penetrate but prevented interferences from red blood cells and some other gases. The success of the Clark sensor eventually resulted in the use of cellophane-covered platinum electrodes to regulate the oxygen tension in the arterial line of the dispersion oxygenator during total bypass of the heart and lungs.5 As demonstrated by Clark, a membrane diffusion barrier for the working electrode offers several advantages over a bare electrode. For instance, the membrane eliminates interferences from ions and other substances to which it is impermeable, thus improving the selectivity for the analyte of interest. Also, the membrane serves as a diffusion layer for gaseous analytes. By appropriate selection of the membrane material and size, one can control the analytical characteristics of a sensor, permitting the analysis of several analytes over a wide range of concentrations. The earliest Teflon-bonded diffusion electrodes were prepared by Niedrach and Alford6 for use in fuel cells. Later, we will also show a method to use bare electrodes with a polymer electrolyte and thereby eliminate the membrane to improve response times.7 In 1958, Sawyer et al. investigated sensing of several gases such as oxygen, nitrogen, chlorine, bromine, sulfur dioxide, nitrogen dioxide, nitrous oxide, hydrogen, carbon monoxide,

358 Chemical Reviews, 2008, Vol. 108, No. 2

Stetter and Li

Figure 4. SEM micrographs of KrF laser patterned PTFE membrane electrodes: (a) PTFE-bonded platinum black layer irradiated with a fluence of 500 mJ/cm2; (b) gold film irradiated with a fluence of 60 mJ/cm2; (c) the edge of the electrode as shown in (a); and (d) timedependent limiting currents of the SO2 sensors with different pattern sizes compared to the unpatterned sensor. The concentration of the SO2 gas is 998 ppm: (a) unpatterned sensor, (b) patterned sensor with the pattern size of 100 µm, and (c) patterned sensor with the pattern size of 60 µm. Reprinted with permission from Qin, Z.; Wang, P. N.; Wang, J. Sens. Actuators, B 2005, 107, 805. Copyright 2005 Elsevier Science.

hydrogen sulfide, and hydrogen cyanide with a membranecovered platinum electrode.8 The best performance characteristics were achieved using polyethylene, as opposed to saran, Mylar, and natural rubber, as the membrane material. As shown by their results, the incorporation of the membrane into the sensor system greatly enhanced the overall practicality of the sensor, leading to many new applications. However, it remained difficult to construct practical gas sensors at that time because there was no convenient way to package the electrodes for field use. Modern Clark electrodes are fitted with a porous poly(tetrafluoroethylene) (PTFE) membrane. Because of the hydrophobicity of the material, the pores (typical diameter 10 µm) are not wetted by the aqueous solution but allow the transport of dissolved gases to the electrode-electrolyte interface.9 As the mass-transfer rate of the analyte is slow compared to electron transfer, the Faradaic current is controlled by diffusion rather than the kinetics of the electrode reaction, and this assures a linear dependence of the current on concentration of the dissolved oxygen. The arrangement can be used for the measurement of other dissolved electroactive gases, most notably for the determination of chlorine in swimming pools. The layer of electrolyte solution between membrane and electrode is kept thin in order not to compromise sensitivity and response time. This design is, in principle, suitable for measurements in the gas phase, but in practice it is found that the cell is not stable because of variations in the thickness of the thin liquid film used in the Clark design, most likely because of evaporation of the water used as a solvent for the electrolyte. For the measurement of electroactive species in the gas phase, designs derived more from fuel cell technology have been

adopted. PTFE membranes can be coated with vacuumevaporated metals for improved sensitivity.10 The potential drop across any given sensor cell includes the potentials of the working electrodes and counterelectrodes and the potential over the galvanic cell’s internal resistance. For a given material of working electrode, the sensitivity of the galvanic cell depends on the surface area on which the electrochemical reaction takes place.11-15 The assembly must provide as large a surface area as possible for the threephase boundary (TPB ) interface between the gas, the electrode, and the electrolyte), because the electrode reaction is assumed to take place primarily in this region. The particular “hair morphology” of a catalyst surface, e.g., nanostructures, can dramatically increase the three-phase boundary area, followed by an enhancement of the sensor’s sensitivity. A further improvement of the sensitivity from ppm to ppb gas concentration levels can be obtained using the new membrane-electrode assembly composed of a PTFE membrane and a nanocomposite material of carbon nanotubes and PTFE. Carbon nanotubes can greatly increase the surface area of the TPB because the carbon nanotubes have very high surface area, e.g., single-walled carbon nanotubes can achieve 1600 m2/g Brunauer-Emmett-Teller (BET) value.16,17 The catalyst is deposited electrochemically on the membrane in order to get a nanostructured surface and to increase the TPB area.18-20 Wang’s group developed a method to pattern the PTFE membrane electrodes by KrF laser ablation21 (see Figure 4). Patterning of the sensing electrode can be wellcontrolled and is expected to increase the sensitivity of the sensor because the sidewall of the patterned catalyst layer can be considered as the effective TPB. Compared with the unpatterned sensor, the sensitivity of the sensor with a pattern

Amperometric Gas SensorssA Review

size of 60 µm was improved by 14 times, and the response time is reduced half. The dramatic increase of the limiting current and the shortening of the response time were obtained from the increase of the TPB area and the rapid gas diffusion through the membrane electrode. Other nanostructures are used for this purpose as well, such as composites of Gd0.7Ca0.3CoO3-δ and Ce0.8Gd0.2O1.9 that have potential applications as oxygenpermeable membranes in an amperometric sensor for NOx detection in exhaust gases.22 In addition, a series of manganites was investigated for membranes with the composition A0.7E0.3MnO3 (A ) Gd, Y, and Pr and E ) Ca and Sr).23 The presence of slow diffusion-coupled redox processes could be detected in all manganites at low and high oxygen activities vs an air reference electrode. The oxygen ion conductivity was considerably lower than 10-5 S cm-1 in all investigated manganites. Nevertheless, the comparison of results for glass-encapsulated and nonencapsulated microcontacts showed that Pr0.7Ca0.3MnO3 and, in particular, Pr0.7Sr0.3MnO3 are quite active as catalysts for the oxygen electrode reaction at the three-phase boundary air/zirconia/ manganite, whereas Y0.7Sr0.3MnO3 does not catalyze the oxygen electrode reaction significantly at the surface. Studies have been carried out on the selectivity, mass transfer rates, and effects of temperature on PTFE films as diffusion barriers.24 The advent of using nanostructures has made a significant impact on the design of the modern AGS as it has on other sensors.19

3.1.2. Diffusion Electrodes A major achievement in the evolution of the modern AGS occurred in 1965 with the development of the gas-diffusion electrode by Niedrach.6 The original Niedrach electrode consisted of a sintered Teflon-Pt catalyst covered with a porous Teflon film. Its performance was evaluated in hydrogen-oxygen and hydrogen-air fuel cells at ambient temperature in 6N KOH and 5N H2SO4. In both cases, it was shown that water was efficiently retained in the electrolytic solutions and did not leak into the gas fuel feed stream. The Niedrach electrode provided a model for the contemporary gas-diffusion electrode, being “semihydrophobic” and allowing an open porous electrode of the finely dispersed metal catalyst yet not allowing complete penetration by the electrolyte. These electrodes are prepared by mixing Teflon emulsion with a high-surface-area noble metal catalyst powder and then depositing the slurry on a metal wire screen or on the surface of a totally hydrophobic porous Teflon film. The electrochemical cell that is based on a gasdiffusion electrode composed of a porous PTFE membrance and a porous electrode is shown in Figure 5. The resulting gas-diffusing electrode consists of highly interlocked matrices of gas pores, electrolyte channels, electronically conducting paths, and electrocatalytic surfaces. It is porous enough to affect efficient gas permeation, has sufficient metal catalyst to be a good electronic conductor, and is hydrophilic enough to make intimate contact with the electrolyte for ionic conduction and facilitation of electrochemical reactions involving gases. In contrast to the Clark electrode, gas-diffusion electrodes based on back-side metallized porous membranes are not affected by evaporation of water because the porous electrode is directly in contact with the bulk of the electrolyte solution. The mass transfer of analyte from the sample to the working electrode can be faster, resulting in shorter response times

Chemical Reviews, 2008, Vol. 108, No. 2 359

Figure 5. Schematic of a cell based on a gas-diffusion electrode (GDE) composed of a porous PTFE-membrane and a porous electrode. Reprinted with permission from Knake et al., Analytica Chimica Acta, 2005, 549, 1. Copyright 2005 Elsevier Science.

and higher currents that lead to higher sensitivity. Also, the real surface area of the electrode can be several orders of magnitude higher, allowing species with relatively poor electroactivity to produce measurable currents. Therefore, they are especially valuable in those cases when the kinetics of the electrochemical reaction are very slow, such as the oxidation of hydrocarbons and CO. Examples of the earliest of these sensors reveal excellent practical sensor characteristics.25-28 A kinetic study of NO gas at the Pt/Nafion electrode with application to the amperometric NO and NO2 sensing has been carried out by Ho’s group.29,30 Sensors based on metallized porous membranes and porous catalytic electrodes have been produced from several manufacturers that are widely used in industrial applications.1,2 On the other hand, relatively few publications on this specific structure of electrode have appeared in the scientific literature in the past decade,31-33 indicating the maturity of this type device. The fuel-cell type electrode was never utilized in an AGS until, in 1969, a breakthrough in design combined it with a Pt/air reference electrode to achieve the practical ethanol and CO sensor design introduced by Oswin, Blurton, and coworkers.26 This new design was used to detect various toxic gases, such as ethanol, CO, H2S, NO2, and NO. The sample (an electrochemically oxidizable or reducable species) enters the cell through porous electrodes similar to those used in a fuel cell (see Figure 6);34 hence, the device has been incorrectly called a fuel-cell sensor, although its purpose is not the conversion of chemical to electrical energy. The relationship between the cell current and the concentration of chemical species of interest present in a more or less constant matrix can be established empirically and optimized to be linear in most cases over a fairly broad range of concentrations. In the electrochemical cell that becomes the sensor of Oswin and Blurton,26 the gas was detected from the current generated by the electrochemical reaction at a fixed potential of 0.9-1.5 V vs SHE. This avoided the generation of undesired current from reactions involving the oxygen-water redox couple within the cell. The original Oswin/Blurton cell significantly simplified sensor design and provided a model for several two- and three-electrode sensors that were eventually introduced in the 1970s including ethanol,26 CO, and other gases.28 Recently, a high-temperature solid electrolyte amperometric CO2 sensor was fabricated using a Pt/NASICON (Na+ conductor, Na3Zr2Si2PO12)/Pt cell and a porous Na2CO3-

360 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 6. Fuel-cell-type amperometric sensor.

BaCO3 auxiliary layer.35 A platinum-loaded zeolite electrode was developed for high-temperature measurement of total NOx in combustion environment.36 A Pt-loaded Y (PtY) is a catalyst electrode for NOx equilibration with electrochemical oxidation of NO on a yttria-stablized zirconia (YSZ) electrolyte. By applying a low anodic potential of 80 mV, the NO in the NOx equilibrated mixture can be oxidized at a Pt working electrode on the YSZ electrolyte at 500 °C. A novel reference electrode for a NOx sensor was developed that did not require air, and this makes the design simpler and more robust.37 With the advent of nanotechnology, the use of nanostructured materials in fuel-cell sensors to enhance the sensor performance becomes ever more popular. The details will be discussed in the following section.

3.2. Electrolytes In addition to the progress made in the development of electrodes and sensor geometries, the 1970s witnessed improvements in the use of different electrolytes. Prior to 1970, electrolyte solutions for amperometric sensors were mainly aqueous in nature. Even nowadays the aqueous solutions are still used as the electrolytes for gas sensing, e.g., acid or halide-halate electrolytic solutions are used for acidic gas38 and for other gas-detection applications.39 As illustrated in the preceding examples, aqueous electrolytes are effective in many amperometric gas sensors. However, they are also limited in their specificity, service life, operating temperature range, and electrical potential range. As a result of the problems and limitations associated with aqueous electrolytes, sensor research began to focus on the design and development of alternative electrolytes. By the middle 1970s, nonaqueous and solid polymer electrolytes emerged for the AGS.6,40-48 As nanotechnology becomes more prevalent and many nanostructures are developed for use as working electrodes, parallel studies of nanosturctures for use as the electrolyte are also being conducted.

3.2.1. Nonaqueous Electrolyte One of the earliest amperometric sensors employing a nonaqueous electrolyte was described by Schneider in 1978.47 The novel electrochemical cell, used for the detection of chlorine, contained a nonaqueous electrolyte composed of

Stetter and Li

lithium or sodium perchlorate as supporting electrolyte dissolved in an organic solvent selected from the group consisting of γ-butyrolactone, sulfolan, or propylene carbonate. All these solvents have a high boiling point (>200 °C) and possess a high anodic limit with the Pt electrode. This sensor was superior to previously designed chlorine sensors, which frequently failed because the electrode materials selective to chlorine often dissolved in the aqueous electrolyte medium. The new sensor was effective in detecting chlorine down to the ppm range, had a fast response time of <1 min, and had an extended useful life of >1 year. A word of caution must be added that the use of perchlorates in organic solvents always represents an explosion hazard, particularly when the assembly is, inadvertently, allowed to dry up or be exposed to high temperatures. Similarly, Venkatasetty described the invention of a carbon monoxide sensor incorporating a nonaqueous electrolyte.44 This sensor, based on an aprotic electrolyte system, was also used to amperometrically detect toxic gases such as nitrogen oxides, SO2, and H2S. Other electrochemical cells employing nonaqueous electrolytes were also introduced during the same time period. These systems were used to fabricate amperometric sensors for several gases such as methane,45 hydrazine, and carbon dioxide.47

3.2.2. Solid Polymer Electrolytes Solid polymer electrolytes became important during the mid-1970s because of the inefficiencies and maintenance requirements of liquid electrolytes. Originally, a solid polymer electrolyte (SPE) was described as a solid plastic sheet of perfluorinated sulfuric acid polymer that, when saturated with water, became an excellent ionic conductor. Nafion, a typical solid polymer electrolyte, is a hydrated copolymer of poly(tetrafluoroethylene) (PTFE) and polysulfonyl fluoride vinyl ether containing pendant sulfonic acid groups (Dupont). Nafion has good proton conductivity, high gas permeability, and chemical stability, and it has been widely used in chloralkalai, fuel-cell, and sensor applications. In the Nafion-based AGS design, Nafion film is used both as the electrolyte and as a support for the electrode structure. The ionic conductivity results from the mobility of the hydrated hydrogen ions that move through the polymer sheet by passing from one sulfuric acid group to another.19 An early amperometric gas sensor incorporating a solid polymer electrolyte was described by LaConti and later improved by Sedlak et al. The sensor, originally used to detect carbon monoxide,43 employed Nafion. Being a cation-exchange membrane, Nafion permitted the passage of cations and blocked the passage of anions. Thus, the hydrogen ions produced at the working electrode through oxidation of carbon monoxide were transported through the ion-exchange membrane to the counterelectrode, where they reduced oxygen with the addition of electrons to produce molecular water. In addition to its excellent ion-exchange capacity, the Nafion electrolyte exhibited other positive features, such as high structural stability, resistance to acids and strong oxidants, and good thermal stability. As a result of its useful characteristics, Nafion has been used to detect gases other than carbon monoxide,49,50 such as CO2 and NH3,51 NOx,52,53 hydrogen,54-56 H2S,57 SO2,58 O2,59,60 O3,61,62 and alcohol vapors.63 The only limitation for field use was the issue of the water in the Nafion freezing at low temperatures. In addition, Kuwata et al. illustrated two different amperometric oxygen sensors using a Nafion membrane opera-

Amperometric Gas SensorssA Review

tive at room temperature.64,65 Also, Alberti and Casicola presented an amperometric three-electrode sensor for CO operating at room temperature and using thin layers of R-zirconium phosphate as a solid electrolyte.41

3.2.3. Ceramic Solid Electrolytes at Elevated Temperature The use of ceramic solid electrolytes in amperometric gas sensors at elevated temperatures allows for the detection of different gases under harsh conditions. There is an increasing need for rugged and reliable gas sensors capable of monitoring carbon monoxide, nitric oxides, sulfur oxides, and hydrocarbons from automotive exhausts. Because of such harsh environments, e.g., high temperature and high pressure, aqueous electrolytes, liquids, or polymeric materials cannot be used in sensor designs. Ceramic solid electrolyte sensors are typically designed for high-temperature applications such as stack gas, combustion process, and automotive engine exhaust. The solid electrolyte provides ionic conduction and also supports the electrodes. The interface between the solid electrolyte and the electrode plays an important role in the resulting gas sensor’s analytical characteristics. Liu and Weppner described amperometric gas sensors using solid electrolytes such as NASICON, cubic stabilized zirconium oxide, and polycrystals of tetragonal zirconium oxide to selectively detect the partial pressures of several gases.66 An AGS incorporating solid electrolytes made from rare earth elements was also designed for high temperature, and Stetter and Cao67 described an amperometric sensor for the detection of halogenated compounds such as chloropentafluorobenzene, chlorobenzene, bromobenzene, and iodobenzene. This sensor, which used a bead of sodium lanthanum fluoride silicate as the solid electrolyte, was shown to operate around 550 °C.67 Similarly, Coillard et al. discussed a sensor employing commercial zirconia lambda gauge for the amperometric detection of NO in automobile exhausts at high temperature.68 As a result of their versatility, solid electrolytes have become important components of amperometric gas sensors. An amperometric NO sensor using a LaGaO3 solid electrolyte was investigated, and the response was based on the difference of catalytic activity of the electrodes.69 It was found that the oxygen pumping current increased upon exposure to NO when Sr0.6La0.4Mn0.8Ni0.2O3 and La0.5Sr0.5MnO3 (LSM-55) were used as active and inactive electrodes for NO oxidation, respectively. Compared with various common electrolytes such as YSZ or Sm-doped CeO2, it was clearly demonstrated that much higher sensitivity in the amperometric mode is achieved by using LaGaO3-based fast oxide-ion conductors. The fuel-cell configuration was used for the investigation of three types of metallic electrodes for NO detection in the exhaust gases.70 Their results show that the cell defined as RhPtAu/YSZ/ RhPtAu gives a relatively high NO sensitivity of ∼14 nA/ppm and a low O2 sensitivity of 0.02 nA/ppm. The availability of electrode materials with selectivity opens up possibilities for the development of highly sensitive amperometric NO gas sensors with a single working electrode, i.e., the NO concentration in O2 and NO containing gas mixtures can be directly measured without oxygen pumping.

3.3. MEMS and Nanotechnology Microelectrodes (MEMS), especially micro-working electrodes, with a very small electrode surface area, have been used for amperometric gas sensors. In recent years, MEMS

Chemical Reviews, 2008, Vol. 108, No. 2 361

technologies have increasingly been employed to fabricate microelectrode structures. The microelectrodes have shown several advantages of high sensitivities, fast response speeds, small sample amounts, and well-defined and reproducible geometries. In contrast to the development of sensor parts, such as membranes, electrodes, and electrolytes, which occurred during the 1960s and 1970s, amperometric gas sensor research of the 1980s focused on the production of novel sensor designs. One of the highlights of the 1980s in relation to analytical chemistry was the introduction of microsensors. Microfabrication of electrochemical devices has numerous advantages over standard fabrication procedures. These advantages include precise reduced sensor size, reduced cost, smaller sample size, faster response time, higher concentration sensitivity, well-defined geometrical features, and potential for mass production. These advantages, however, must be obtained without degradation of the signal-to-noise ratio as the sensor size is reduced. Progress in the development of microamperometric sensors was slow when compared to the production of micropotentiometric sensors. Microamperometric sensors of the early 1980s consisted only of microfabricated electrodes on a suitable substrate. These electrodes were then coated with the liquid electrolyte solution that also carried the sample to the electrode area for analysis. An example of an early microamperometric sensor was presented by Sleszynski and Osteryoung in 1984.71 The given sensor, which used very small electrodes constructed from nonconductive epoxy and reticulated vitreous carbon (RVC), was shown to yield desirable results as compared to standard fabricated sensors. However, the goal of constructing complete, practical, and commercially successful amperometric gas sensors using the microfabrication approach has still not been achieved. During the late 1980s, microfabrication technology for the construction of amperometric sensors was investigated with the introduction of novel sensor electrode designs. In 1988, Maclay and co-workers72,73 introduced a series of Nafionbased microfabricated amperometric gas sensors using gold sensing electrodes in the shape of a square grid. The newly designed sensors were evaluated by comparing their analytical characteristics with those of conventional sensors. The response time of the miniaturized sensors was >1 order of magnitude faster than conventional sensors, although they had a lower sensitivity. This work also elucidated the fact that the sensitivity of the device depended not only on the chemical nature of the electrode surface but also on the specific structure of the electrocatalytic surface and the interface created by gas/electrode/electrolyte. In an effort to improve the sensitivity of microamperometric sensors, Buttner et al.73 constructed devices with an integrated design in which the working electrodes, counterelectrodes, and reference electrodes were photolithographically etched onto a gold-coated silicon oxide surface of a silicon wafer and spin-coated with a thin film of a solution of Nafion. The working electrode of the sensor was an ultrafine square grid with evenly spaced regions of gold and holes every 50 µm. The signal observed for the sensor following exposure to gases such as H2S and NO greatly exceeded the response that would be expected on the basis of simple geometric considerations of microelectrodes.73 The analytical characteristics of microamperometric sensors can also be improved by varying the sensor components, such as electrolyte composition. For example, Hauser et al.31

362 Chemical Reviews, 2008, Vol. 108, No. 2

recently described the preparation of a carbon dioxide sensor using mixed organic solvents as the electrolyte. As shown by the results of this study, a mixed acetonitrile/dimethyl sulfoxide solvent enhanced the overall performance of the sensor by producing a large potential window over which the CO2 reduction currents can be measured.31 As illustrated by these examples, advances in microtechnology during the 1980s have resulted in the production of newly designed amperometric gas sensors with characteristics superior to those of conventional sensors. More recently, gold microelectrodes are used with and without membranes in nonaqueous media for oxygen, carbon dioxide, and nitrous oxide detection simultaneously over a wide concentration range.74,75 It is believed that, by using sufficiently small microelectrodes, it should be possible to effectively “outrun” the reaction kinetics by the much faster loss of species from the electrode/electrolyte interface by diffusion, thereby giving rise to independent oxygen and carbon dioxide signals. Hahn and co-workers also conducted the microelectrode studies of isoflourane vapor, an inhalation anesthetic agent, and oxygen mixtures in dimethyl sulfoxide.76 However, such sensors never demonstrated the long lifetimes required by field applications nor the low cost and, thus, have not been commercialized for popular use. Very recently, nanosize working electrodes using carbon nanotubes have been reported. The advantages of such microsensors are small in size, low cost, low power, high signal-to-noise ratio, and low detecting limit. For the discussion of using the nanostructures for making the microsensors, a sensor array consisted of various highly sensitive carbon nanotube-based sensors developed by Li’s group are described here, such as sensors using pristine carbon nanotubes,77 carbon nanotubes loaded with metal clusters,78,79 and carbon nanotubes coated with polymers.80 An interdigitated electrode (IDE) was utilized as a transducer, with many of them on a chip that was fabricated on a silicon wafer.81 This electrochemical sensor depends on the transfer of charge from one electrode to another electrode. The gases and vapors introduced to this sensor array are NO2, HCN, HCl, Cl2, acetone, and benzene in parts per million (ppm) concentration levels in dry air. These are toxic gases and vapors that are of interest to both military and civilian personnel in defense applications, as well as in industry process and environmental monitoring. The detection limit of carbon nanotube sensors can achieve the ppb concentration level.82 Carbon nanotubes can be coated with metals to prepare hydrogen sensors83 or with Nafion to make RH sensors.20 For many years, it was generally reported that potentiometric sensors were more easily microfabricated because the signal size did not depend at all on the electrode size while amperometric sensor currents were directly proportional to electrode size. An ideal use of nanostructures to create large surface area in small-sized devices is found here. With the use of nanostructures, the electrode area of the amperometric sensor can be reduced by orders of magnitude while preserving the sensor signal magnitude, and this is a perfect example of the importance of nanostructures in the design and construction of the AGS. Several other designs for amperometric gas sensors were also introduced during the 1990s. For example, Wallgren and Sotiropoulos reported84 the construction of an oxygen sensor based on the planar sensor design. The design consisted of both the working electrodes and the counter-

Stetter and Li

electrodes being vacuum-deposited as nonporous Au layers on the same face of a Nafion electrolyte and in contact with the gas sample. Potential advantages of this design include faster response times and higher sensitivities per useful electrode area due to the high mass transport rates of the gas to the sensing electrode since no membrane or significant liquid film barrier is present. Also, smaller quantities of precious metal catalyst could be used for the working electrode, and the labor-intensive stacking steps required for the fabrication of conventional-type designs are avoided. As shown by the results of this research, oxygen reduction leads to an exponential current rise over a wide potential range, indicating very high mass transport rates and implying that the electroactive gas reacts at the line formed by the gas/ solid electrolyte/metal layer interface.84 However, the practicality of the design may be questioned because, in practice, if the water content of the membrane changes, so do the dimensions, and this can lead to stresses that will fracture the electrodes and alter the response. Berger and Edelman also described the production of a planar sensor used to detect the partial pressure of oxygen in blood samples.85 In this sensor, a semipermeable membrane, comprising an acrylonitrile butadiene copolymer or an acrylate-based copolymer, was provided to act as an impermeable barrier for ions and blood constituents other than oxygen.85 In 2004, Meyerhoff and co-workers reported a planar amperometric nitric oxide sensor based on platinized platinum anode.86 A microporous poly(tetrafluoroethylene) gaspermeable membrane was used. Platinization of the working platinum electrode surface dramatically improves the analytical performance of the sensor by providing ∼10-fold higher sensitivity (0.8-1.3 pA/nM), ∼10-fold lower detection limit (e1 nM), and extended (at least 3-fold) stability (>3 d) compared to sensors prepared with bare Pt electrodes. In this article, both experimental results and theory of NO measurements as a function of sensor diameter and distance from a point source were discussed in the context of an investigation of monitoring NO release from diazeniumdiolate-doped polymeric films. By modifying the above-discussed NO amperometric gas sensor with thin hydrophilic polyurethane films containing catalytic Cu(II)/(I) sites, the direct amperometric detection of S-nitrosothiol species (RSNOs) is realized.87 Catalytic Cu(II)/(I)-mediated decomposition of S-nitrosothiols generates NO(g) in the thin polymeric film at the distal tip of the NO sensor. Another type of planar thick-film sensor with centrosymmetric diffusion geometry for detection of hydrocarbons in oxygen, nitrogen, and hydrocarbon gas mixtures has been developed for monitoring in exhaust gases.88 In this work, propene (C3H6) is chosen for sensor tests as a model hydrocarbon. The solid-state electrochemical cell based on yttria-stabilized zirconia (YSZ) operates in the amperometric mode. Oxygen is pumped out at the Au/YSZ electrode, and C3H6 is oxidized at the Pt/YSZ electrode (see Figure 7). At a gas temperature of 600 °C, the sensitivity is 2 nA/ppm C3H6 for this high-temperature amperometric gas sensor design.

3.4. Sensor Array The 1980s also witnessed the microcomputer revolution, which, coupled with the advent of chemometrics and available AGSs, encouraged the introduction of arrays of gas sensor systems.89-93 While the earlier research occurred in

Amperometric Gas SensorssA Review

Chemical Reviews, 2008, Vol. 108, No. 2 363

Figure 7. Schematic of the sensor design: (A) one side of the sensor showing the counterelectrodes upon the YSZ substrate, (1) Au/YSZ ring electrode (RE) and (2) Pt/YSZ point electrode (PE); (B) back side of the sensor, showing the diffusion barrier covering the working electrodes; and (C) propene characteristics of the Pt/YSZ point electrode at various oxygen concentrations. Reprinted with permission from Schmidt-Zhang, P.; Guth, U. Sens. Actuators, B 2004, 99, 258. Copyright 2004 Elsevier Science.

the early 1980s93,94 in conjunction with a K-nearest neighbor or KNN algorithm for pattern recognition, a very recent development features an amperometric sensor array for a gas mixture of NOx, CO, O2, and SO2 detection using Nafion/Pt (or Au)/ceramic plate microelectrodes prepared by microfabrication technologies as the sensing electrodes and a backpropagation neural network (BPN) as the patternrecognition algorithm.95,96 The limiting current potentials for anodic oxidation of NO on Au, anodic oxidation of SO2 and CO on Pt, and cathodic reduction of NO2 on Au and O2 on Pt were applied to each of the sensors in the array, respectively. The system was optimized for identifying these gases with their concentrations in the mixture. Sensor array technology has been utilized in chemical detection of gases and is sometimes known as an electronic nose97,98 and, when used in liquids, is sometimes know as the electronic tongue.99 However, the instrument using a sensor array as a detector has not been fast to get into the commercial analytical market and remains in the stage of a new technology in various application areas. One of the main reasons for the slow introduction of arrays into the market is the difficulty in calibration and in the isolation of the analyte response from the matrix response. Recent conferences, reviews, and books have chronicled progress in sensor arrays incorporating all types of sensors.90 A recent renaissance has occurred and provided new interest in sensor-array-based analysis. Much of this interest has to do with important applications like the detection of lung cancer, recently shown to be possible with a gas sensor array.100 While the relationship of the sensor array signals to lung cancer is undisputed, there is something unsettling about an analytical technique for which the molecular basis for the response is clearly not well-understood if understood at all. Statistically, while we expect the analysis to be somewhat robust and to apply to unknowns, the statistical basis for this can be calculated and the molecular basis is only implied. Thus, the sensor array work often blurs the line between functional dependence and association of a signal and complex endpoint. It will take progress in multivariate calibration, the construction of multivariate models like the net analytical signal recently used for NIR determination of glucose, or the development of another robust

method in order to gain widespread analytical acceptance for routine analytical determinations. What the sensor array has always been good at is the comparison of two complex mixtures, one a standard and one sample or a mixture of standard and a sample to the standard. In analysis, the sensor array can be a highly valuable tool for determination or comparison of complex nonmolecular endpoints like taste, odor, or flavor, and the AGS was one of the earliest contributors.93,94 More comparisons of different analytical myths and operational aspects of the electronic nose can be found89 for practical applications.

4. AGSs For Practical Applications The modern AGS can be considered one of the primary methods for field detection of CO in air for industrial hygiene applications since the 1970s, when it was introduced to replace the hopcalite catalyst method.101 The 1970s through the 1990s also brought about many additional applications of amperometric sensors in field analytical use.102 The AGS provides ppm level sensitivity, stability, low cost, low power, good selectivity, and fast response all at the same time, and this makes for a powerful and versatile sensor. Also, because the analysis is “on the spot”, samples do not have to be preserved or returned to a lab for a slow and costly analysis. By manipulating the standard three-electrode potentiostat arrangement of amperometric gas sensors, one can design an amperometric detector for various target analytes and conditions in field analysis. Hauser et al. recently described a simplified amperometric detector for capillary electrophoresis. This system employed only two electrodes in total to analyze catecholamines, ascorbic acid, carbohydrates, and heavy metals. Detection was carried out with a single working electrode by using an electrophoretic counterelectrode as a polarized pseudo-reference electrode with an appropriate electronic circuitry. As shown by the results of this study, the amperometric detector achieved similar performance characteristics as compared to the traditional amperometric gas sensor.31 The amperometric gas sensor has a long history of development and some of them have produced mature for field applications. With the advent of microfabrication

364 Chemical Reviews, 2008, Vol. 108, No. 2

Stetter and Li

Figure 8. (A) Schematic cross section of a typical amperometric oxygen sensor with a channel-type diffusion barrier and (B) principle of an amperometric two-electrode cell for the simultaneous detection of oxygen and NO. Reprinted with permission from Reinhardt, G.; Mayer, R.; Rosch, M. Solid State Ionics 2002, 150, 79. Copyright 2002 Elsevier.

technology, computational power in a PC, and nanotechnology, improved sensors with better working electrodes, different types of electrolytes, and miniature design have evolved, and this has expanded the spectrum of real-world applications.103 The challenges are common to all applications including the following: inexpensive fabrication steps, sensor robustness, reliability, reproducibility, low power consumption, and easy system integration.

4.1. Industry AGSs possess high sensitivity, small size, and low power consumption, which can be used to quickly verify incoming raw materials at the delivery point. The AGS technology can significantly reduce the amount of time and money spent analyzing materials in a lab, as well as reduce the amount of material handling. Most changes in chemical processes can be reflected in the changing composition of the vapor phase surrounding or contained within the process. Thus, vapor-phase sensors can enable the quick assessment of the chemical status of most industrial processes. Examples are found across many sectors including food processing (coffee roasting and fermentation), petrochemical (plastics manufacture and gasoline blending), and consumer products (detergents and deodorants). Much like vision inspection is used to assess the visual integrity (color, shape, and size) of products, olfactory inspection assesses the chemical integrity (consistency and presence of contaminants) of products.1,2,89,91,109,110,111

4.2. Environment Increasing awareness of new regulations for safety and emission control makes environmental monitoring one of the most important applications for reliable gas sensors. Current methods for air quality analysis typically require the use of very costly and bulky analytical equipment. For applications in environmental monitoring or indoor air quality measurement, sensors that are able to selectively detect various gases at concentration levels ranging from a few ppb to hundreds of ppm are valuable. New developments in MEMS sensors are trying to address these difficult measurement tasks.103 With unique combined advantages of high sensitivity, small size, and low power consumption, the amperometric gas sensor is uniquely suited for environmental applications, but it is often challenged with sensitivity or selectivity demands not encountered in other applications. For example, the amperometric sensors based on stabilized zirconia have been developed for the detection of oxygen-containing gases like NOx or combustible gases such as CO, H2, and hydrocarbons;104 for NOx detection;105 for lean combustion gas control;106 and for selective detection of propane.107 Multielectrode amperometric sensors are able to detect O2 and NO or O2 and combustibles at the same time, and such a

multielectrode cell is illustrated in Figure 8. The basic idea is to adjust geometric and working parameters in such a way that, at each electrode, essentially one-electrode reaction takes place. Selectively reacting electrodes are desired to improve sensor performance. Recently, SrTi0.65Fe0.35O3-δ (STF-35) has received considerable attention for use as a resistive oxygen sensor because of its unusual temperature-independent conductivity above 700 °C and pO2 > 1 Pa.108 This material has the potential for automotive oxygen sensor applications. Combined with physical sensors for temperature, anemometry, pressure, and humidity monitoring, an AGS system can be built to get comprehensive information for an environment simultaneously. The Indoor air quality application is just now beginning to emerge as exposure to environmental pollutants in the air is being suspected for many human health issues. The Exposure Biology program at the National Institutes of Health addresses many of these concerns. The application to measure trace levels of pollutant exposure of children and the population at large will require new sensors for measurements as well as new modeling tools (http://www.gei.nih.gov/ exposurebiology/index.asp). This need for personal measurement sensors at ppb levels for specific pollutants represents a new challenge that is ideally met with the AGS since it has all of the important logistical features of low power, light weight, small size, and low cost. However, the challenge is obtaining increases in sensitivity and selectivity and often response time and stability in the AGS for analytes such as ozone, CO, NOx, and others.

4.3. Defense and Homeland Security Chemical sensors like the AGS are suitable for security and defense applications because of their portability and low power consumption. AGS potentially can offer higher sensitivity, specificity, and lower power consumption for toxic gas detection. Some examples include monitoring filter breakthrough (e.g., toluene concentration inside the mask), personnel badge detectors (e.g., toxins accumulation), embedded suit leak-detection sensors (e.g., soldiers’ suits for upcoming warfare agents), and other applications. Additionally, a wireless capability with the sensor can be used for networked mobile and fixed-site detection and early warning systems for military bases, work facilities, and battlefield areas. Explosives detection at the airport is recently a priority application for Department of Homeland Security (DHS), and sensors or detectors will require quick detection and a low false-alarm rate. One type of the liquid explosive hazard that is easily detected by amperometry is hydrogen peroxide. The challenges will be to obtain the stability, reproducibility, and low maintenance of the fielded sensors to achieve quick screening and early warning at the airport. The detection of mustard gas and HCN are two specific agents that can be

Amperometric Gas SensorssA Review

addressed with the AGS, as well as TNT, when the electrochemical sensors are used in arrays.109-116

4.4. Medical/Biomedical Chemical sensors will provide physicians with a quicker and more accurate diagnostic tool. Applications will include breath analysis that can provide objective information on the identity of certain chemical compounds such as in exhaled air and in the excreted urine or other body fluids that are directly related to specific metabolic conditions, certain skin diseases, or bacterial infections, such as those common to burn wounds. Additionally, the chemical sensors may also provide more accurate, real-time patient monitoring during anesthesia administration. An amperometric enzyme/gas-diffusion electrode is developed that combines a porous air-breathing gas-diffusion electrode with an enzyme immobilized on its catalytic layer. The oxygen needed for the enzymatic reaction is supplied to the reaction zone of the electrode in the gas phase through the gas-supplying layer of a gas-diffusion electrode. The performance of a gas-diffusion/enzyme electrode operating in oxygen-containing solutions is practically independent of the concentration of the dissolved oxygen.

5. Summary Amperometric gas sensors have a long and rich history. The present review shows that AGSs are an important member of the electrochemical class of chemical sensors. Many research and development efforts have resulted in practical life-saving sensors, such as the AGS sfor CO, H2S, NO2, O2, or other electrochemically active gases that are now routinely used in industrial hygiene and safety monitoring of people and workplaces. There is an ever-expanding list of analyte targets for the AGS, and its use for electroactive materials like TNT will challenge the sensitivity of the amperometric approach. With the advent of arrays, the AGS is contributing to the detection of a diverse set of complex endpoints including toxicity, cancer detection, and off-odor of materials. New applications such as homeland security and defense against threats have arisen and will challenge the sensitivity and selectivity of the electrochemical technique in general and the amperometric approach specifically. As we demand ever-lower detection levels and since the sensor signal in amperometry is directly proportional to the electrode area, the challenge is to maintain and improve the signal/ noise ratio for the analyte. However, the challenge of amperometry is to utilize the ability provided by electronics to detect smaller and smaller amounts of charge/current and to use nanostructures to continue to increase the electrode reactive surface area even with less material used in the electrode. The advancement of MEMS technology and the fast pace of nanotechnology will no doubt enable new AGS designs and materials such that new applications are found for large gas molecules and weakly electroactive gas molecules in applications that will analyze for chemical threats, toxins, food flavors, and fragrances. Much of the modern AGS work focuses on new designs that incorporate microfabrication and nanofabrication to achieve smaller size, low power, lower cost, and portable sensors and sensor arrays with intelligence. New nanomaterials development changes the building blocks that will provide well-organized nanostructures with high surface area, high chemical reactivity at lower temperature,

Chemical Reviews, 2008, Vol. 108, No. 2 365

good mechanical strength, and better thermal stability, leading to new catalysts for selectivity, new electrolytes for higher-temperature operation, multiple working electrodes for self-amplifying sensors, and combinations with bioanalytical approaches for biosensors and enzyme-based sensors. MEMS technology and nanotechnology combined with new computational power brighten the future of the AGS and its use within analytical chemistry and especially in field analytical measurements. We now begin a new era of AGS development using new materials, new time-resolved and spatially resolved sensor array approaches, and smaller sensor devices. It is clear that the utility of amperometry will continue and will preserve its place in gas-sensor applications now and in the future37 and will continue to serve for the betterment of human health, safety, and the environment.

6. Acknowledgment The authors would like to thank Sheng Yao for help with some of the illustrations used herein and to thank our colleagues because, without their work, this review could not have been written. Also, we express regret if some work has not been included as it is only possible to include a fraction of the work in this field.

7. References (1) Stetter, J. R.; Butner, W. J.; Cao, Z. Electroanalysis 1992, 4. (2) Stetter, J. R.; Chang, S. C.; Cha, C. S. Talanta 1993, 40. (3) Bard, A. J.; Faulkner, L. R. Electrochemical Methods: Fundamentals and Applications, 2nd ed.; John Wiley & Sons: New York, 2001. (4) Kissinger, P. T.; Heinemann, W. Laboratory Techniques in Electroanalytical Chemistry, 2nd ed.; Marcel Dekker Inc.: New York, 1996. (5) Clark, L.; Wolf, R.; Granger, D.; Taylor, Z. J. Appl. Physiol. 1953, 6. (6) Niedrach, L. W.; Alford, H. R. Chem. Abstr. 1965, 62, 11416c. (7) LaConti, A. B. U.S. Patent 4,227,984, 1979. (8) Sawyer, D.; George, R. S.; Rhodes, R. C. Anal. Chem. 1959, 31. (9) Knake, R.; Jacquinot, P.; Hodgson, A. W. E.; Houser, P. C. Anal. Chim. Acta 2005, 549, 1. (10) Wienecke, M.; Bunescu, M. C.; Pietrzak, M.; Deistung, K.; Fedtke, P. Synth. Met. 2003, 138, 165. (11) Parsonage, E. E.; Debe, M. K. U.S. Patent 5,333,8430, 1994. (12) Thomas, J. H.; Ziegler, R. J.; Debe, M. K.; Haugen, G. M.; Steinbach, A. J. U.S. Patent 5,879,827, 1999. (13) Lauer, J. M.; Akmal, N. Patent EP0777121, 1997. (14) Ziegler, P.; Demisch, U.; Pahlke, J. U.S. Patent 5,733,436, 1998. (15) Lee, S. B.; Maclay, J. G.; Keyvani, D. U.S. Patent 5,716,506, 1998. (16) Cinke, M.; Li, J.; Chen, B.; Cassell, A.; Delzeit, L.; Han, J.; Meyyappan, M. Chem. Phys. Lett. 2002, 365, 69. (17) Stetter, J. R.; Hesketh, P. J.; Hunter, G. W. Interface 2006, 15 (1), 66. (18) Hong, Y. J.; Oh, S. M. Sens. Actuators, B 1996, 32, 7. (19) Stetter, J. R.; Maclay, G. J. AdVanced Micro and Nano Systems; Baltes, H.; Brand, O.; Fedder, G. K.; Hierold, C.; Korvink, J. G.; Tabata, O., Eds.; Wiley-VCH Verlag-GmbH and Co.: Weinheim, Germany, 2004; Vol. 1, p 357. (20) Star, A.; Han, T. R.; Joshi, V.; Stetter, J. R. Electroanalysis 2004, 16 (1-2), 108. (21) Qin, Z.; Wang, P. N.; Wang, J. Sens. Actuators, B 2005, 107, 805. (22) Nigge, U.; Wiemhofer, H. D.; Romer, E. W. J.; Bouwmeester, H. J. M.; Schulte, R. R. Solid State Ionics 2002, 146, 163. (23) Wiemhofer, H. D.; Bremes, H. G.; Nigge, U.; Zipprich, W. Solid State Ionics 2002, 150, 63. (24) Do, J. S.; Wu, K. J.; Tsai, M. L. Sens. Actuators, B 2002, 86, 98. (25) Oswin, H. Chem. Abstr. 1970, 72, 106582t. (26) Bay, H. W.; Blurton, K. F.; Lieb, H. C.; Oswin, H. G. Nature 1972, 240. (27) Sedlak, J. M.; Bay, H. W.; Valentine, A. M. Anal. Chem. 1974, 46. (28) Stetter, J. R.; Blurton, K. F. J. Chromatogr. 1978, 155. (29) Ho, K. C.; Liao, J. Y.; Yang, C. C. Sens. Actuators, B 2005, 108, 820. (30) Ho, K. C.; Hung, W. T. Sens. Actuators, B 2001, 79, 11. (31) Hauser, P.; Jordan, L.; Dawson, G. A. Electroanalysis 1997, 9. (32) Hauser, P.; Kappes, T. Analyst 1999, 124.

366 Chemical Reviews, 2008, Vol. 108, No. 2 (33) Hauser, P.; Dawson, G. Electroanalysis 2000, 12. (34) Janata, J. Principles of Chemical Sensors; Plenum Press: New York, 1989. (35) Lee, J. S.; Lee, J. H.; Hong, S. H. Sens. Acuators, B 2003, 96, 663. (36) Yang, J. C.; Dutta, P. K. Sens. Actuators, B 2007, 123 (2), 929. (37) Yao, S.; Stetter, J. R. J. Electrochem. Soc. 2004, 151 (4), H75. (38) Imaya, H.; Ishiji, T.; Takahashi, K. Sens. Actuators, B 2005, 108, 803. (39) Ho, K. D.; Hung, W. T. Sens. Actuators, B 2001, 79, 11. (40) Lee, S. B.; Maclay, J. G.; Keyvani, D. U.S. Patent 5,716,506, 1998. (41) Alberti, G.; Casicola, M. Solid State Ionics 1993, 61. (42) Campbell, D. N.; Davis, R. C.; Schmidt, J. C. U.S. Patent 4,744,954, 1986. (43) LaConti, A. B. Chem. Abstr. 1979, 90, 60474a. (44) Venkatasetty, H. V. U.S. Patent 4,522,690, 1985. (45) Otagawa, T.; Zaromb, S.; Stetter, J. R. J. Electrochem. Soc. 1985, 132. (46) Schmidt, J. C.; Campbell, D. N.; Clay, S. B. U.S. Patent 4,595,486, 1985. (47) Scheider, A. A. U.S. Patent 4,184,937, 1978. (48) Albery, W. J. U.S. Patent 4,377,446, 1982. (49) Xing, X. K.; Liu, C. C. Electroanalysis 1991, 3, 111. (50) Yan, H.; Liu, C. C. Sens. Actuators, B 1994, 17, 165. (51) Mayo, N.; Harth, R.; Mor, U.; Marouani, D.; Hayon, J.; Bettelheim, A. Anal. Chim. Acta 1995, 310, 139. (52) Opekar, F. Electroanalysis 1992, 1, 133. (53) Do, J. S.; Shieh, R. Y. Sens. Actuators, B 1997, 37, 19. (54) Sakthivel, M.; Weppner, W. Sens. Acutators, B 2006, 113, 998. (55) Samec, Z.; Opekar, F.; Crijns, J. E. F. Electroanalysis 1995, 7, 1054. (56) Chao, Y.; Buttner, W. J.; Yao, S.; Stetter, J. R. Sens. Actuators, B 2005, 106 (2), 784. (57) Schiavon, G.; Zotti, G.; Toniolo, R.; Bontempelli, G. Anal. Chem. 1995, 67, 318. (58) Schiavon, G.; Zotti, G.; Toniolo, R.; Bontempelli, G. Analyst 1991, 116, 797. (59) Hwang, B. J.; Liu, Y. C.; Hsu, W. C. J. New Mater. Electrochem. Syst. 1999, 2, 33. (60) Hwang, B. J.; Liu, Y. C.; Hsu, W. C. J. Solid-State Electrochem. 1998, 2, 378. (61) Schiavon, G.; Zotti, G.; Bontempelli, G.; Farnia, G.; Sandona, G. Anal. Chem. 1990, 62, 293. (62) Knake, R.; Hauser, P. C. Anal. Chim. Acta 2002, 459, 199. (63) Schiavon, G.; Comisso, N.; Toniolo, R.; Bontempelli, G. Electroanalysis 1996, 8, 544. (64) Kuwata, S.; Miura, N.; Yamazoe, N. Chem. Lett. 1988, 7. (65) Kuwata, S.; Miura, N.; Yamazoe, N. Nippon Kagaku Kaishi 1989, 10. (66) Liu, J.; Weppner, W. U.S. Patent 5,322,601, 1994. (67) Stetter, J. R.; Cao, Z. U.S. Patent 5,226,09, 1993. (68) Coillard, V.; Juste, L.; Lucat, C. Meas. Sci. Technol. 2000, 11. (69) Dutta, A.; Ishihara, T. Sens. Actuators, B 2005, 108, 309. (70) Dchmidt-Zhang, P.; Zhang, W.; Gerlach, F.; Ahlborn, K.; Guth, U. Sens. Actuators, B 2005, 108, 797. (71) Sleszynski, N.; Osteryoung, J. Anal. Chem. 1984, 56. (72) Maclay, G. J.; Buttner, W. J.; Stetter, J. R. IEEE Trans. Electron DeVices 1998, 35. (73) Buttner, W. J.; Maclay, G. J.; Stetter, J. R. Sens. Mater. 1990, 2. (74) Floate, S.; Hahn, C. E. W. J. Electroanal. Chem. 2005, 583, 203. (75) Wadhawan, J. D.; Welford, P. J.; McPeak, H. B.; Hahn, C. E. W.; Compton, R. G. Sens. Actuators, B 2003, 88, 40. (76) Floate, S.; Farmery, A. D.; Hahn, C. E. W. Sens. Actuators, B 2005, 109, 200. (77) Li, J.; Lu, Y.; Ye, Q.; Cinke, M.; Han, J.; Meyyappan, M. Nanoletters 2003, 3 (7), 929. (78) Lu, Y.; Li, J.; Han, J.; Ng, H.-T., Binder, C.; Partridge, C.; Meyyappan, M. Chem. Phys. Lett. 2004, 391, 344. (79) Young, P.; Lu, Y.; Terrill, R.; Li, J. J. Nanosci. Nanotechnol. 2005, 5 (9), 1509. (80) Li, J.; Lu, Y.; Meyyappan, M. IEEE Sens. J. 2006, 6 (5), 1047. (81) Li, J.; Lu, Y.; Ye, Q.; Delzeit, L.; Meyyappan, M. Electrochem. Solid State Lett. 2005, 8 (11), H100. (82) Qi, P.; Vermesh, O.; Grecu, M.; Javey, A.; Wang, Q.; Dai, H.; Peng, S.; Cho, K. J. Nano Lett. 2003, 3 (3), 347. (83) Stetter, J. R.; Bradley, K.; Cumings, J.; Gabriel, J.-C. P.; Gru¨ner, G.; Star, A. Nanotech 2003, 3, 313. (84) Wallgren, K.; Sotiropoulos, S. Sens. Actuators, B 1999, 60. (85) Berger, J.; Edelman, P. U.S. Patent 6,068,748, 2000. (86) Lee, Y.; Oh, B. K.; Meyerhoff, M. E. Anal. Chem. 2004, 76, 536.

Stetter and Li (87) Cha, W.; Lee, Y.; Oh, B. K.; Meyerhoff, M. E. Anal. Chem. 2005, 77, 3516. (88) Schmidt-Zhang, P.; Guth, U. Sens. Actuators, B 2004, 99, 258. (89) Stetter, J. R.; Penrose, W. R. Understanding Chemical Sensors and Chemical Sensor Arrays (Electronic Noses): Past, Present, and Future. In Sensors Update; Wiley-VCH: Weinheim, Germany, 2002; Vol. 10, Chapter 2.3, p 189. (90) Stetter, J. R., Penrose, W. R., Eds. Artificial Chemical Sensing: Proc. Eighth International Symposium on Olfaction and the Electronic Nose; ISOEN8-ISOEN 2001; The Electrochemical Society: Pennington, NJ, 2001; ISBN 1-56677-321-0. (91) Stetter, J. R.; Penrose, W. R. The Electrochemistry Encyclopedia; http://electrochem.cwru.edu/ed/encycl/art-n01-nose.htm (accessed 2001). (92) Zaromb, S.; Stetter, J. R. Sens. Actuators 1985, 6, 225. (93) Stetter, J. R.; Zaromb, S.; Penrose, W. R.; Findlay, M. W.; Otagawa, T.; Sincali, A. J. Hazard. Mater. Spills Conf. Proc., PreV., BehaV., Control Cleanup Spills Waste Sites (Ludwigson, John, Ed.) 1984, 183. (94) Stetter, J. R.; Jurs, P. C.; Rose, S. L. Anal. Chem. 1986, 58, 860. (95) Do, J. S.; Chen, P. J. Sens. Actuators, B 2007, 122, 165. (96) Pardo, M.; Dwong, L. G.; Sberveglieri, G.; Brubaker, K.; Schneider, J. F.; Penrose, W. R.; Stetter, J. R. Sens. Actuators, B 2005, 106, 136. (97) Li, J. Sensors 2000, 17 (8), 56. (98) Gardner, J.; Bartlett, P. N. Electronic Nose: Principles and Applications; Oxford University Press: New York, 1999. (99) Nanto, H.; Stetter, J. R. Introduction to Chemosensors. In Handbook of Machine Olfaction, Electronic Nose Technology; Pearce, T. C., Ed.; Wiley-VCH: Weinheim, Germany, 2003; Chapter 4. (100) Mazzone, P.; Mekhail, T.; Czich, C. Am. J. Respir. Crit. Care Med. 2006, A777. (101) Stetter, J. R. Ann. Am. Conf. GoV. Ind. Hyg. 1984, 11, 225. (102) Miller, M.; David, F.; Wilson, E.; Kling, R. J. Air Pollut. Control Assoc. 1971, 21. (103) Bryzek, J. S.; Roundy, S.; Bircumshaw, B.; Chung, C.; Castellano, K.; Vestel, M.; Stetter, J. R. IEEE Circuits DeVices, 2006, 22 (2), 8 (ISSN 8755-3996). (104) Reinhardt, G.; Mayer, R.; Rosch, M. Solid State Ionics 2002, 150, 79. (105) Nakamura, T.; Sakamoto, Y.; Saji, K.; Sakata, J. Sens. Actuators, B 2003, 93, 214. (106) Ivers-Tiffee, E.; Hardtl, K. H.; Menesklou, W.; Riegel, J. Electrochim. Acta 2001, 47, 807. (107) Ueda, T.; Plashnitsa, V. V.; Nakatou, M.; Miura, N. Electrochem. Commun. 2007, 9, 197. (108) Litzelman, S. J.; Rothschild, A.; Tuller, H. L. Sens. Actuators, B 2005, 108, 231. (109) Stetter, J. R. Amperometric Electrochemical Gas Sensors: Description and Applications. In NIST Workshop on Gas Sensors: Strategies for Future Technologies, Sept 8-9, 1993; NIST Pub. #865; National Institute of Standards and Technology: Gaithersburg, MD, 1993; pp 61-64. (110) Findlay, M. W.; Penrose, W. R.; Stetter, J. R. Quality Classification of Grain Using a Sensor Array and Pattern Recognition. Anal. Chim. Acta 1993, 284, 1. (111) Stetter, J. R. Instrumentation to Monitor Chemical Exposure in the Synfuel Industry. Ann. Am. Conf. GoV. Ind. Hyg. 1984, 11, 225. (112) Strathmann, S.; Penrose, W. R.; Stetter, J. R.; Gopel, W. Detection of TNT with chemical sensors. Proc. International Symposium on Olfaction and the Electronic Nose, (ISOEN 99), Tuebingen, Germany, Sept 20-22, 1999. (113) Wang, J. Electrochemical Sensing of Explosives. In Counterterrorist Detection Techniques of ExplosiVes; Yinon, J., Ed.; Elsevier: Amsterdam, The Netherlands, 2007; Chapter 4. (114) Wang, J. Electrochemical Sensing of Explosives.Electroanalysis 2007, 4, 415. (115) Wang, J.; Pumera, M.; Collins, G. E.; Mulchandani, A. Measurements of Chemical Warfare Agent Degradation Products Using an Electrophoresis Microchip with Contactless Conductivity Detector. Anal. Chem. 2002, 74, 6121. (116) Wang, J.; Pumera, M.; Chatrathi, M. P.; Escarpa, A.; Musameh, M.; Collins, G.; Mulchandani, A.; Lin, Y.; Olsen, K. Single-Channel Microchip for Fast Screening and Detailed Identification of Nitroaromatic Explosives or Organophosphate Nerve Agents. Anal. Chem. 2002, 74, 1187.

CR0681039

Chem. Rev. 2008, 108, 367−399

367

Semiconductor Junction Gas Sensors Karin Potje-Kamloth* Institut fuer Mikrotechnik Mainz, Carl-Zeiss-Strasse 18-20, 55129 Mainz, Germany Received July 11, 2007

Contents 1. Introduction 2. Properties of Organic Semiconductors 2.1. Electronic Properties of Organic Semiconductors 2.1.1. Molecular Semiconductors 2.1.2. Conducting Polymers 2.2. Doping in Organic Semiconductors and Self-localized States 2.2.1. Molecular Semiconductors 2.2.2. Conducting Polymers 2.3. Transport Mechanism in Organic Semiconductors 2.3.1. Molecular Semiconductors 2.3.2. Conducting Polymers 3. Metal/Semiconductor Junctions and Their Use in Gas Sensing 3.1. Introduction 3.2. Interfacial Electronic StructuresIdeal Junction Characteristics 3.2.1. Formation of a Schottky Barrier 3.2.2. Charge Carrier Transport Mechanism across the Junction Potential Barrier 3.2.3. Ohmic Contact and Space Charge Limited Current 3.2.4. Tunneling through the Barrier 3.3. Interfacial Electronic StructuresDeviations from the Ideal Case 3.3.1. Introduction of the Ideality Factor 3.3.2. Schottky Barrier Lowering 3.3.3. Effect of Interface States 3.3.4. Metal/Organic Semiconductor Junctions with an Interfacial Layer 3.4. Gas/Solid Interactions in Schottky Barrier Junction Devices 3.4.1. Bulk Effect upon Gas DopingsModulation of the Electron Work Function 3.4.2. Formation of a Dipole and an Interfacial Layer 3.5. Extraction of Schottky Diode Parameters for Sensor Applications 3.5.1. Introduction 3.5.2. Current−Voltage Characteristic 3.5.3. Capacitance−Voltage Characteristic 3.5.4. Impedance Measurement 4. Schottky Diodes and Their Sensor Applications 4.1. Schottky Diodes Based on Inorganic Semiconductors * E-mail: [email protected].

367 368 368 368 369 369 370 370 371 372 372 372 372 373 373 374 374 376 376 376 377 377 377 379 379 380 380 380 381 382 382 383 383

4.1.1. Silicon Based Schottky Diodes 4.1.2. GaAs Based Schottky Diodes 4.1.3. InP Based Schottky Diodes 4.1.4. GaN and AlGaN Based Schottky Diodes 4.1.5. SiC Based Schottky Diodes 4.1.6. CdSxSe1.x Based Schottky Diodes 4.2. Schottky Diodes Based on Organic Semiconductors 4.2.1. Phthalocyanines 4.2.2. Polythiophene/Poly(3-alkylthiophenes) 4.2.3. Polyanilines 4.2.4. Polypyrrole 5. Summary and Perspectives 6. List of Symbols and Abbreviations 7. References

383 385 385 385 386 386 386 388 388 389 390 395 395 396

1. Introduction Over the past three decades, a number of studies have been undertaken to characterize the interactions of gases and chemical species with semiconductors as well as to develop useful sensing devices based on these effects.1 Examples include gas detectors using metal-oxide semiconductor fieldeffect transistors,2-4 metal-oxide semiconductor capacitors,5 as well as Schottky barrier diodes.6 Among those, Schottky barrier diode sensors are devices which are extremely simple to fabricate, obviating the need for photolithography or hightemperature diffusion/oxidation steps. These sensors may be broadly classified into two categories: (1) interfacecontrolled deVices, in which the species of interest are adsorbed at the metal surface and thus affect interfacial polarization, and (2) bulk-controlled deVices, which are dependent on the change in bulk properties of the semiconductor caused by absorption of and interaction with the diffusing species. Schottky barrier based gas sensors have been fabricated with a number of inorganic semiconductors using catalytic metals as the metal contact,6-11 mostly for detection of hydrogen. Due to various hydrogen-induced changes in catalytic metals, the change of the interface properties of the Schottky barrier diode readily lends itself to applications in interface-controlled sensors devices.6 The metal/inorganic semiconductor Schottky diodes can exhibit a very high sensitivity and low detection limit to hydrogen. The success of these devices translates into a huge number of papers published on the subject of hydrogen sensing. Unfortunately, the range of applicability is limited to the detection of hydrogen and hydrogen-producing gases or vapors, which comes from the exclusive solubility and permeability of atomic hydrogen in such materials. This limitation does not apply to the class of organic semiconductors, which includes organic molecules as well

10.1021/cr0681086 CCC: $71.00 © 2008 American Chemical Society Published on Web 01/30/2008

368 Chemical Reviews, 2008, Vol. 108, No. 2

Karin Potje-Kamloth studied chemistry at Ludwigs Maximilians Universita¨t Mu¨nchen and received her Ph.D. and Habilitation in physical chemistry. Currently, she is a senior scientist at Institut fuer Mikrotechnik Mainz GmbH. Her main research areas of interest are low cost chemical sensors, thin film technology of advanced polymer materials, energy harvesting, and chemical sensing platforms for microsystems

as conjugated polymers. They show almost unlimited permeability for gases and vapors, and the concomitant ability to form charge-transfer complexes with the matrix. Moreover, they offer the possibility to tune the sensitivity and selectivity toward gases and vapors by chemical alternation of the polymer backbone, addition of side groups, alternation of the lengths of the polymer chain, and alternation of the nature of the dopant. Variations in the chemical and physical properties of the organic semiconductors are manifold and represent a further advantage for organic semiconductors in this application area. For this reason, the focus of this review is on the progress in metal/organic semiconductor based Schottky barrier junction sensors. The use of organic semiconducting materials has received considerable interest as the active component in electronic device structures,12-18 such as organic light emitting diodes (OLEDs),19,20 solar cells,21,22 and organic field-effect transistors,23,24 Schottky diodes,25 and gas sensors.26-30 In the case of field-effect transistors (FETs), the organic semiconductor can either be the active component, i.e., the current carrying material, of the organic FET (OFET) or the selective layer used as the gate electrode in a chemical sensitive FET (CHEMFET). The interaction of gases or vapors with the bulk of organic semiconductors affects the work function of the organic semiconductor, which modifies the electrical properties of the electronic device. This leads to the application of organic semiconductors in bulk-controlled devices. The metal/semiconductor interface is the obvious component in any semiconductor device and controls the Schottky barrier device characteristic. The theoretical aspects of interfaces in devices based on organic semiconductors closely resemble those of inorganic ones. This allows in general the use of the extensive knowledge accumulated during many years of studies of the electronic properties of inorganic semiconductors. The purpose of this review is to introduce the reader to the field of metal/semiconductor junctions, in particular metal/organic semiconductor junctions, by providing insights into some of the theoretical and experimental approaches that have been employed thus far, although it is not meant to be an exhaustive study of all of the research work and techniques employed in the study of interface formation in semiconductor devices. This topic is too broad and is beyond the scope of this review.

Potje-Kamloth

The second section gives an introduction to the physics of organic semiconductors, which are relevant to the understanding of the peculiar properties of interfaces in organic semiconductor devices. The third section deals with some theoretical aspects concerning metal/semiconductor junctions and their use in chemical sensing. The fourth section is a review of this area of semiconductor junction gas sensors and deals with the electrical and gas sensing properties of Schottky barrier diodes. Both sections cover inorganic and organic semiconductor Schottky barrier diodes. Furthermore, section 4 shows how experimental and structural parameters of semiconductors can influence the Schottky barrier junction and hence the diode characteristics. The reader who is familiar with organic semiconductor polymers and the electrical properties of semiconductor contacts can skip directly to section 4, and those interested in an overview of this class of materials and of these types of devices may find sections 2 and 3 to be useful.

2. Properties of Organic Semiconductors The word “organic semiconductors” includes polymers and low-molecular-weight organic materials. The former are usually characterized by ill-defined chain lengths while the latter show both a well-defined composition and length. The use of the name “organic semiconductor” is based on the extrinsic semiconducting properties of organic systems, i.e., the capacity to transport charge generated by light, injected by electrodes, or provided by chemical dopants. Due to the weak electronic coupling between organic molecules, disorder plays a central role in the understanding of these materials. Disorder manifests itself in varying the energy levels of the electronic states of different molecules and the structure of the solid state. Both strongly influence the transport mechanism of charge carriers and charge injection at metal/semiconductor interfaces.

2.1. Electronic Properties of Organic Semiconductors 2.1.1. Molecular Semiconductors In inorganic semiconductor crystals such as silicon or germanium, the strong coupling between the atoms and the long-range order lead to the delocalization of the electronic states and the formation of allowed valence and conduction bands, separated by a forbidden energy gap. By thermal activation or photoexcitation, free electrons are generated in the conduction band, leaving behind positively charged holes in the valence band. The transport of these free charge carriers is described in the quantum mechanical language of Bloch functions, k-space, and dispersion relations familiar to solid-state physicists.31 In organic solids, intramolecular interactions are mainly covalent, but intermolecular interactions are due to much weaker van der Waals and London forces. As a result, the transport bands in organic crystals are much narrower than those of their inorganic counterparts and the band structure is easily disrupted by introducing disorder in the system. Thus, even in molecular crystals, the concept of allowed energy bands is of limited validity and excitations as well as interactions localized on individual molecules play a predominant role. The common electronic feature of molecular semiconductors is the π-conjugated system, which is formed by the overlap of carbon pz-orbitals. In Figure 1 the

Semiconductor Junction Gas Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 369

Figure 1. Molecular structure of metallophthalocyanine tetrasulfonate (MPcTS) salt.

molecular structure of a typical representative of this class of semiconductors, metallophthalocyanine tetrasulfonate (MPcTS), is shown. The orbital system of the ring comprises 42 π-electrons. Due to the orbital overlap, the π-electrons are delocalized within the molecule and the energy gap between the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO) is small, i.e., exhibiting transition frequencies within the visible range. Activation of free electrons by photoexcitation or chemical doping can lead to a tremendous increase of the electric conductivity from about 10-12 S cm-1 to values in the range of 102 S cm-1, which indicates the semiconducting character of the material.

2.1.2. Conducting Polymers Polymers are typically associated with flexible, processible materials having electrically insulating properties. Although this is true of most polymers, a special class of these materials called conjugated polymers has the electrical and optical properties traditionally associated with metals and semiconductors, yet they retain the mechanical properties and the processability of plastics. Conjugated polymers are characterized by repeated units in which atomic valence is not satisfied by covalent bonds. The leftover valence electrons of adjacent carbon atoms overlap and form double bonds that give rise to π-bonds, in which the π-electrons are delocalized over large segments of the polymer chain. They are forming the so-called π-conjugated system, which is responsible for the electronic properties of the conjugated polymers. The chemical structures of the most thoroughly studied conducting polymers are shown in Figure 2. Polymers can consist of on the order of 105 to 106 (or even a greater number) monomers strung together to form a macromolecular chain. Typical conjugated polymers include straight chain units, five or six-membered rings, and all combinations of these. The addition of heteroatoms (atoms other than carbon and hydrogen) and side chains allows for an even larger variety in this class. Their electronic properties can be tailored by the synthesis, and their electrical conductivity can be varied from about 10-12 S cm-1 to values in excess of 105 S cm-1. In order to understand the operation of polymer devices better, it is useful to first examine some basic properties of organic semiconductors based on conjugated polymers, because they differ qualitatively from those of crystalline inorganic semiconductors in several important respects. In crystalline materials, the properties are determined by the three-dimensional arrangement of the atoms and their resulting interactions. Conjugated polymers are treated as a quasi-one-dimensional system, wherein the polymer chains

Figure 2. Schematic representation of the chemical structures of the most commonly studied conjugated polymers, as well as their corresponding nomenclature and energy band gap Eg.

are assumed to behave independently of one another, and their physical and chemical properties depend on interactions within the single chains. The π-bonding scheme of conjugated polymers decreases the gap between occupied (HOMO) and unoccupied (LUMO) states. The valence band and conduction band of conducting polymers are generally derived from such π-bonds. The band gap of these polymers tends to lie between 1.5 and 3 eV (see Figure 2), in the same range as that of inorganic semiconductors. Since the bonds formed by σ-orbitals are stronger than the π-bonds, the polymers do not break apart when excited states are created in the π-electron levels. Also, in most polymers, there is only a weak overlap between the π-orbitals between neighboring polymer chains, so electrons and holes tend to be delocalized on individual polymer chains, although they can hop between chains. The π-electron delocalization along the chain and the weak interchain bonding give conjugated polymers a quasi-one-dimensional nature and give rise to strong anisotropies when the macromolecules are chain extended and chain aligned. Due to the electron-phonon-coupling of the π-electrons, which leads to a lattice distortion (Peierls distortion), a gap is generated at the Fermi level. If an electron is added or removed to/from the polymer chain, a self-localized electronic state within the previously forbidden semiconductor band gap is formed as a result of the easy deformability of the polymer lattice without chain scissoring. The creation of electronic defects allows charge transport within a single chain. This property of conjugated polymers demonstrates that many ideas are used to explain why inorganic semiconductors do not directly carry over to conjugated polymers. Early on, most of the research effort was concentrated on the conducting properties of the doped conjugated polymers,32 which were often referred to as ‘‘conducting polymers’’. In this review, this notation is used throughout the text to describe this class of polymers when they are in both pristine and doped states.

2.2. Doping in Organic Semiconductors and Self-localized States Doping of a disordered organic semiconductor by charged moieties has two counteracting effects: (1) the increase of concentrations of charge carriers and, thus, the change of the Fermi level;33,34 (2) the increase of energetic disorder by formation of additional deep Coulomb traps of the opposite polarity. Therefore, the average hopping rate is controlled

370 Chemical Reviews, 2008, Vol. 108, No. 2

by the charge carrier release from the Coulomb traps.35 The former effect facilitates conductivity; the latter strongly suppresses the charge carrier hopping rate. This process determines, together with the intrinsic order, the temperature dependence of the mobility. Arkhipov et al. showed that the doping efficiency on charge carrier hopping strongly depends on the energy disorder and the external field.35 Upon doping by ionized species, charge neutrality must be maintained. There are two ways to accomplish this. One is electrochemical doping. If the ionization (or reduction) potential of the electrode more or less matches the HOMO of the organic semiconductor, a negatively charged majority carrier can be injected, provided that the electrolyte supplies appropriate counterions that can diffuse into the semiconductor. For p-conducting organic semiconductors, positively charged majority carriers are injected, as the oxidation potential of the electrode matches the LUMO. An example is the oxidation of polyhexylthiophene by injection of holes from a solution containing tetraethylammonium perchlorate. Charge injection is compensated for by concomitant “doping” with perchlorate anions. The alternative method is doping by a neutral entity, whose electron affinity is different enough to allow for charge transfer from the semiconductor to the dopant. In both cases, excess mobile majority carriers and immobile countercharges (i.e., dopants) are generated that roughen the energy landscape, in which the charge carriers migrate, but in the “neutral” doping case, in addition, charge redistribution can and does occur. The majority of charge carriers will actually form metastable geminate pairs, whose dissociation is facilitated by the ambient phonon bath and the external electric field. A high charge carrier concentration can be reached without introducing counterions due to either field effect or high level of monopolar charge injection across a contact. Under these circumstances, the Coulomb interaction between charge carriers can strongly change the effective potential landscape. The interaction between charges of the same sign is repulsive and, therefore, cannot create Coulomb traps. It gives rise to transiently fluctuating potential barriers that affect the mobility.

2.2.1. Molecular Semiconductors Doping of organic molecular films has been investigated relatively little compared to the doping of inorganic semiconductors or of conjugated polymers. The main reason is that, unlike inorganic semiconductors, traditional n- and p-doping has not been a requirement for achieving bipolar transport in the most common molecular devices, i.e., OLEDs and OFETs. The ability to stack electron and hole transport layers, to build organic/organic or metal/organic junctions, alleviates the need to “dope” the organic materials in order to inject electrons or holes into the active layer(s) of the organic device. However, the performance of organic devices is now reaching a level at which electrical doping seems to be attractive as a means to further improve efficiency. In this respect, considerable work remains to be done, however, to understand and control doping in materials which exhibit fundamental differences with standard inorganic semiconductors. The weak intermolecular bonds, large energy gaps, and small dielectric constants of these materials are not particularly conducive to low dopant ionization energies. Several groups have started to investigate doping mechanisms and their effect on the electronic structure of the host organic molecular semiconductors.36-39 Particularly important are the relative energies of the dopant and the host molecular levels,

Potje-Kamloth

which determine the ‘‘ionization energy’’ and doping efficiency. Doping of metal/organic contacts with inorganic donors, such as lithium in aluminum tris(8-hydroxychinoline) (Alq3),40 and with inorganic acceptors, such as antimony pentachloride in N,N′-diphenyl-N,N′-bis(3-methylphenyl)1,1′-biphenyl-4,4′-diamine (TPD),41 has shown the potential for significant improvements in current injection. Similar results could be obtained by doping with molecular acceptors, e.g., tetrafluorotetracyanoquinodimethane (F4-TCNQ),37,42 and molecular donors, e.g., bis(ethylenedithio)tetrathiafulvalene (BEDT-TTF).43

2.2.2. Conducting Polymers When an electron is added to (or withdrawn from) a rigid band semiconductor, it must go into the conduction band (or come from the valence band) because the lower energy valence band is completely filled. However, when an electron is added to (or withdrawn from) a conducting polymer, a chain deformation takes place around the charge, which costs elastic energy and puts the charge in a lower electronic energy state. The competition between elastic deformation energy and electronic energy determines the size of the lattice deformation, which can be on the order of 20 polymer units long.44 This localized charged particle together with the simultaneous chain deformation is known as a polaron and is defined in the semiconductor literature as an electron that is dressed by a phonon cloud. The lattice distortion by the charges is also called self-trapping. In Figure 3A, the doping process of p-type polypyrrole (PPy) is depicted, which represents a typical conducting polymer. The corresponding band structures and the allowed electronic transitions are shown in Figure 3B. Bipolarons are similar to polarons, but they are double charged. Instead of being a single charge that distorts the chain, there are two charges, which are bound together in and by the same chain deformation. Although these charges repel each other via the Coulomb interaction, they remain bound together by their common chain deformation; the energy increase to form two distinct chain deformations rather than a single one is greater than the Coulomb energy gained by their separation. Associated with each one of these excitations are energy states located in the energy gap of the p-doped polymer (Figure 3B). The polaron has two subgap states, of which the lower one is singly occupied and the higher one is empty for a positive polaron (i.e., an electron and a hole). The doubly charged bipolaron has its subgap states completely empty for a positive bipolaron. A significant difference between polarons and bipolarons is their spin signatures: polarons have charge (1 and spin 1/2, the same spin and charge as a free electron. Bipolarons have a zero spin and are doubly charged, mainly because of the pairing of the electrons. The charge injection into the macromolecular chain, i.e., the creation of any of the above-mentioned defects (polaron, bipolaron) in the conjugated backbone of the polymer, is called doping. Besides this charge injection, doping in polymers also implies the insertion or repulsion of a counterion (referred to as “dopants” or “doping ions”) to maintain charge neutrality. In order to match conventions between physicists and chemists, a polymeric cation-rich material is called p-doped, and a polymeric anion-rich material is called n-doped. Since every monomer is a potential redox site, conducting polymers can be doped to a high density of charge carriers. The doping level is up to 5

Semiconductor Junction Gas Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 371

Figure 3. (A) Neutral (undoped) state of polypyrrole and doped (polaron and bipolaron) structures of lightly and heavily doped polypyrrole, respectively. (B) Corresponding schematic energy band structures and the allowed electronic transitions.

orders of magnitude greater than that in common inorganic semiconductors. The standard value of the doping levels lies in the range of 0.27-0.5 positive charge per monomer (2750% by mol), depending on the anion and the type of conducting polymer. A slight decrease in the doping level is observed with an increase of the molecular weight and the negative charge of the anion. For all conjugated polymers, except for polyacetylene, polarons and bipolarons are the entities through which charge transport is accomplished in conducting polymers.45 They can travel along the chain as an entity, with the atoms in the path changing their positions so that the deformation travels with the electron or hole.46 Doping in polymers can be accomplished chemically, electrochemically, and photochemically, as well as by charge injection at the metal/insulator/semiconductor (MIS) interface.47 In the case of electrochemical and/or chemical doping, the induced electrical conductivity is permanent, until the charge carriers are purposely removed by dedoping, i.e., by reversing the electrochemical reactions, or until the charge carriers are chemically compensated. Initially, conducting polymers have been regarded as onedimensional semiconductors, because the π-conjugated systems extend over the whole polymer chains, thus allowing for delocalized states in one direction. In their pioneering work, Su, Schrieffer, and Heeger48 studied the interaction of an excitation with an ideal one-dimensional lattice, neglecting Coulomb interactions and disorder effects. Today, conducting polymers are discussed either as one-dimensional semiconductors with strong electron/phonon interactions or as disordered molecular solids with strong Coulomb interactions. The discussion of the precise nature and dynamics of excited states in conducting polymers is still very much alive. The one-dimensional semiconductor nature of conducting polymers might play a role in experiments on highly ordered, stretch-oriented materials or on isolated single chains. Furthermore, in conducting polymers, the influence of chemical dopants on the morphology of the polymer layer and on the charge-transporting states is still a matter of

controversy. The (quasi) one-dimensional nature of the conducting states may explain the exceptionally high conductivities in doped conducting polymer systems.49

2.3. Transport Mechanism in Organic Semiconductors A feature common to all amorphous and disordered systems, among them organic semiconductors, is the frequency dependent conductivity that increases approximately linearly with frequency, at least in the range 101-107 Hz, i.e., σ(ω) ∝ σωs, where the frequency exponent s e 1.50 The origin of this obviously universal behavior of the dispersive component σac of the conductivity has been ascribed either to the relaxation process or to a transfer process. The former is caused by the motion of charge carriers by thermally assisted quantum mechanical tunneling between localized states lying deep within the band gap. The latter is called hopping and involves classical thermal activation over the barrier height, separating two sites with an energy difference ∆E. The frequency exponent s is predicted to have a temperature dependence, and the magnitude of s at any temperature is determined by the binding energy of the charge carrier in its localized site.51 The temperature dependence of the dc conductivity term provides first-hand information on the possible processes involved in the conduction in amorphous materials. One of the theoretical models most successfully used to predict the probable transport mechanism in disordered materials is the Mott’s variable-range hopping (VRH).52 It describes a phonon-assisted quantum mechanical transport process, in which a balance is obtained between the thermodynamics constraint on a charge carrier moving to a nearby localized state of different energy and the quantum mechanical constraint53 on a charge carrier moving to a localized state of similar energy, but spatially far away. The VRH model is equally applicable to charge carriers such as electrons, polarons, or bipolarons, provided that the appropriate wave

372 Chemical Reviews, 2008, Vol. 108, No. 2

function is incorporated. This model and its field of applications have been extensively reviewed by various authors.52,54 According to the phonon-assisted quantum mechanical transport process, the temperature dependence of the dc conductivity provides information about the conduction mechanism and the dimensionality of the charge transport mechanism, which ranges from a three-dimensional to a onedimensional transport process. Considering a one-dimensional system, the most commonly invoked descriptions are charging-energy-limited tunneling (CELT),55 fluctuation-induced tunneling (FIT),53 and quasi-one-dimensional variable-range hopping.52,56

2.3.1. Molecular Semiconductors Charge carrier hopping within a positionally random and energetically disordered system of localized states was shown to be an adequate model for the description of both the equilibrium and the nonequilibrium conductivity in noncrystalline organic semiconductors.57-60 Ba¨ssler et al. discussed the effect of doping on the charge carrier hopping in disordered organic semiconductors.58,61 Doping such a system by charged moieties will create a random distribution of dopant ions that will coulombically interact with charge carriers localized in randomly located intrinsic hopping sites62 and, thus, broaden the effective density of states distribution. This effect is especially important for molecular semiconductors, because the dielectric permittivity is low, and the range of the Coulomb potential is large in organic solids.

2.3.2. Conducting Polymers Much work has focused on the nature of charges in doped conducting polymers. Even though there is a high density of conduction electrons at the Fermi level for the highly doped polymers, the charge carriers may be spatially delocalized, so that they cannot participate in the electronic transport except through hopping. The prime source of localization in conducting polymers is structural disorder. Most conducting polymers are noncrystalline, i.e., amorphous materials, although some of these systems possess a molecular low crystallinity, in which regions of the material are more ordered and other regions are more disordered. The percentage of crystallinity may vary from near zero to 11% for polypyrroles,63 20-30% for polyanilines,64 and approximately 80% for highly doped polyacetylenes.65 The charge carrier mobilities in these low-conducting amorphous solids are frequently estimated to be very low, µ , 1 cm2 V-1 s-1.66 These values bear witness against a band transport mechanism. However, detailed knowledge of their charge transport is usually quite limited. Nevertheless, the basic principles underlying electronic states and electronic transport in this class of solid materials are those of amorphous solids. It should be emphasized that the great variety of polymer constituents and structures yields a diversity of interesting and perhaps specific electronic properties. The dimensionality of the system can play a major role in determining the nature of both electronic states and electronic transport. The intrinsic contributions to charge transport are strongly intermixed, and it is often not trivial to quantify the individual contributions from various parameters (e.g., interchain interactions, anisotropic diffusion of charge carriers, the role of dopant ions, the extent of disorder, etc.). The theoretical modeling of transport properties in conducting polymers is a challenging problem because of the extreme complexity of these systems.

Potje-Kamloth

For a transport mechanism to be described in a certain material, charge carriers must be present. In conducting polymers, upon injection of an electron or a hole, a defect in the polymer chain is created, i.e., a polaron or a bipolaron, as already mentioned. These charged defects are the real charge carriers in nondegenerated conducting polymers. Charge transport in conducting polymers consists generally of two components: intrachain and interchain transport.67,68 Intrachain charge transport occurs along the polymer backbone and requires less energy than interchain charge transport, which involves the hopping of the charge to neighboring chains. The dopant compounds play a very important role in the hopping process during interchain charge transport. The charge carrier mobility may be thought of as a measure of the ease with which the charge carriers move through the material. It is sensitive to the level of structural order present in the polymer. Thus, structural defects decrease the conductivity by lowering the mobility. It is interesting that the theoretical conductivity of perfectly aligned defectfree polyacetylene is believed to be greater than 106 S cm-1.67 As a general consideration, the conjugation length appears to be the most significant structural parameter controlling the charge mobility. It has been recognized that the effective conjugation length of most conducting polymers is finite, because of the imperfect planarity of the conjugated system. It is supposed that the chain architecture of a conducting polymer consists of rigid blocks of a conjugation length of several polymer units separated by flexible nonconjugated polymer segments. This kind of macromolecular chain architecture should be kept in mind when one deals with the structure-property relationship of conducting polymers. Currently, the description of the conductivity in these kinds of disordered materials remains open, and the understanding of the influence of the disorder, which is present at every scale of the structure (defects on the chains, heterogeneities in the doping distribution, etc.) is far from being complete. Moreover, several studies have been devoted to the theoretical search for microscopic pictures leading to different new descriptions, such as Fermi glass,69 Coulomb glass,70 superlocalization,71 multifractal localization,72 etc. Such a great variety of theoretical approaches reflects the wide range of phenomena involved in electronically conducting polymers. Furthermore, the transport process may differ from one polymer to another, as well as from one sample to another, principally according to the preparation method.

3. Metal/Semiconductor Junctions and Their Use in Gas Sensing 3.1. Introduction In the following sections, the behavior of basic metal contacts with semiconducting materials and their application in chemical gas sensing are described. The elucidation of the interfacial electronic structure forms the basis for understanding and improving the performance of electronic devices. A key theoretical prediction is that the electric field at the semiconductor/metal interface should, in principle, respond to changes in the Fermi level of the contacting metal.73,74 The behavior of such contacts has been discussed thoroughly in several books and reviews articles,75-77 although the interfacial chemistry of these systems is still controversial.

Semiconductor Junction Gas Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 373

Figure 4. Formation of a barrier between a metal and a semiconductor: (A) neutral and isolated; (B) electrically connected; (C) in perfect n-type rectifying contact (φm >φs); (D) n-type ohmic contact (φm < φs); (E) p-type ohmic contact (φm > φs); (F) p-type rectifying contact (φm < φs). o denotes electron in conduction band; + denotes donor ion.

Metal/semiconductor contacts are components of any semiconductor device. At the same time, such contacts cannot be assumed to have a resistance as low as that of two connected metals. In particular, a large mismatch between the Fermi energy of the metal and the semiconductor can result in a high-resistance rectifying contact. A proper choice of materials can provide a low-resistance ohmic contact. However, for a lot of semiconductors, there is no appropriate metal available. Instead, one then creates a tunnel contact. Such a contact consists of a thin barrier, obtained by heavily doping the semiconductor, through which charge carriers can readily tunnel. Thin interfacial layers also affect the formation, which is discussed for metal/organic semiconductors contacts.

3.2. Interfacial Electronic StructuresIdeal Junction Characteristics 3.2.1. Formation of a Schottky Barrier Experiments on rectifying contacts started in 1874 with the pioneering work of Braun, who observed asymmetries in the transport of electrical current across metal/semiconductor interfaces.78 The following decades brought a variety of technical applications, but it took more than 60 years until Schottky,79 and independently Mott,80 used microscopic concepts to describe these so-called Schottky contacts. Since then, metal/inorganic semiconductor contacts have been extensively studied and have become rather well understood. The subject of interfacial electronic structure can be roughly divided into two aspects: (1) the energy level alignment at the interface and (2) the band bending in the space-charge region.81 The former is important for charge carrier injection; the latter is essential for charge carrier separation. A complete

overview of the works on Schottky contacts is beyond the scope of this review. Several excellent textbooks report on their electronic properties and the development of the field.81-83 Figure 4 illustrates the band diagram of Schottky’s Gedankenexperiment for the formation of a Schottky barrier. The metal and the semiconductor are supposed to be electrically neutral and to be separated from each other for an n-type semiconductor with a work function φs smaller than that of the metal φm (Figure 4A). When the metal and the semiconductor come in electrical contact, the two Fermi levels are forced to coincide and electrons pass from the semiconductor into the metal. The result is an excess of negative charge on the metal surface and the formation of a positive charge depletion zone with a thickness WS in the semiconductor near its surface (Figure 4B). These excess charges form an interface dipole and produce an electric field, directed from the semiconductor to the metal. By bringing the metal and the semiconductor closer together, the gap between the two materials vanishes and the electric field corresponds now to a gradient of the electron potential in the depletion layer, resulting in the well-known band-bending regime (Figure 4C). The Mott-Schottky model leads to the Schottky barrier height φb. For an ideal contact between a metal and an n-type semiconductor, the height of the barrier φb measured relative to the Fermi level is given by

φb ) φm - χS

(1)

φb ) eVd + ξ ) φbi + ξ

(2)

or

374 Chemical Reviews, 2008, Vol. 108, No. 2

Potje-Kamloth

where χs is the electron affinity of the semiconductor, defined as the difference in energy between an electron at rest outside the surface and an electron at the bottom of the conduction band just inside the surface, e is the electronic charge, ξ is the difference between the bottom of the conduction band EC and the Fermi level of the semiconductor, eVd is the band bending at zero bias voltage, and eVd, the diffusion potential at zero bias voltage, usually called the built-in potential φbi or the contact potential formed between the metal and the semiconductor, is given by82

eNDWS2 φbi ) eVd ) 2S

(3) J ) A**T2 exp

where ND is the donor concentration, s is the permittivity of the semiconductor, and WS is the depletion layer width. Equation 2 has been referred to as the Mott-Schottky limit. In obtaining it, a number of important assumptions have to be made, namely that (1) the surface dipole contributions to φm and χS do not change when the metal and semiconductor are brought into contact, (2) there are no localized states on the surface of the semiconductor, and (3) there is a perfect contact between the semiconductor and the metal (i.e., there is no interfacial layer). In the case that the work function of the n-type semiconductor φs is larger than that of the metal φm, the contact is biased so that electrons flow from the semiconductor to the metal. They encounter no barrier (Figure 4D). If a bias voltage is applied such that electrons flow in the reverse direction, the comparatively high concentration of electrons in the region where the semiconductor bands are bent downward (usually referred to as the accumulation region) behaves like a cathode, which is easily capable of providing a copious supply of electrons. The current is then determined by the bulk resistance of the semiconductor and the applied voltage. Such a contact is termed ohmic contact. A p-type semiconductor/metal junction, for which φm exceeds φS, represents an ohmic contact (Figure 4E). The case of a p-type semiconductor, for which φS exceeds φm, is shown in Figure 4F. Bearing in mind that holes have difficulty in going underneath a barrier, one sees that Figure 4F is the p-type analogue of Figure 4C and gives rise to rectification, which is the case of Schottky contacts based on p-type organic semiconductors. The barrier height φb for an ideal contact between a metal and a p-type semiconductor is given by

φb ) Eg - (φm - χs)

metal/semiconductor interface, contribute to the current flow. Quantum mechanical tunneling through the barrier takes into account the wave-nature of the electrons, allowing them to penetrate through thin barriers. In a given junction, combinations of all three mechanisms exist. However, typically one finds that only one charge carrier mechanism dominates. Thermionic Emission Theory. Thermionic emission theory assumes that the barrier height φb is much larger than kT and that electrons, with an energy larger than the top of the barrier, will cross the barrier, provided they move toward the barrier. The actual shape of the barrier is hereby ignored. The current density, J, can be expressed as

(4)

3.2.2. Charge Carrier Transport Mechanism across the Junction Potential Barrier The models initially developed for charge carrier transport across potential barriers in crystalline materials have also been used for noncrystalline systems. The current across a metal/semiconductor junction is mainly due to majority carriers. Three distinctly different mechanisms exist: diffusion of charge carriers from the semiconductor into the metal, thermionic emission of charge carriers across the Schottky barrier, and quantum mechanical tunneling through the barrier. The diffusion theory assumes that the driving force is distributed over the length of the depletion layer. The thermionic emission theory, on the other hand, postulates that only energetic charge carriers, which have an energy equal to or larger than the conduction band energy at the

( )[ ( ) ] φb eV exp -1 kT kT

(5)

or

[ (eVkT) - 1]

J ) J0 exp

(6)

where A** is the effective Richardson constant for thermionic emission, neglecting the effects of optical phonon scattering and quantum mechanical reflection (equal to 120 A cm-2 K-2 for free electrons), T is the absolute temperature, k is the Boltzmann constant, V is the bias voltage, and J0 is the saturation current density, given as

( )

J0 ) A**T2 exp -

φb kT

(7)

Diffusion Theory. For semiconductors with low charge carrier mobility (µ < 10-4 cm2 V-1 s-1), it has been shown that the dominant barrier transport mechanism is diffusion.81 The diffusion theory by Schottky81 is derived from the assumption that the depletion layer is large compared to the mean free path, so that the concepts of drift and diffusion are valid. The resulting current density equals

( )[ ( ) ]

J ) eNcµeE¨ max exp -

φb eV exp -1 kT kT

(8)

eV -1 kT

(9)

or

[

J ) JSD exp

]

where NC is the effective density of states in the conduction band, µe is the electron mobility, and E¨ max is the maximum field strength at the metal/semiconductor interface given by E¨ max ) eNDW/s. It can be seen that eq 9 is almost, but not quite, of the form of the ideal rectifier characteristic for thermionic emission. The difference arises because E¨ max is not independent of bias voltage but is proportional to (V0 V)1/2. For large values of reverse bias, the current does not saturate but increases roughly with |V|1/2.

3.2.3. Ohmic Contact and Space Charge Limited Current A metal/semiconductor junction results in an ohmic contact (i.e., a contact with voltage independent resistance), if the Schottky barrier height φb is zero or negative. In such cases, the charge carriers are free to flow in and out of the semiconductor so that there is a minimal resistance across the contact (see Figure 4D and E). For an n-type semiconductor, the work function of the metal must be close to or smaller

Semiconductor Junction Gas Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 375

than the electron affinity of the semiconductor. For a p-type semiconductor, it requires that the work function of the metal must be close to or larger than the sum of the electron affinity and the band gap energy. Since the work function of most metals is less than 5 V and a typical electron affinity is about 4 V, it can be problematic to find a metal that provides an ohmic contact to p-type semiconductors with a large band gap such as GaN, SiC, or organic semiconductors. The values of the work function of organic semiconductors strongly depends on the preparation method and the doping level. The values range between 4.0 and 5.3 and can even differ for the same type of polymer and doping because of differences in the treatment of the material. Therefore, the reader is referred to the literature to obtain information on the work function values of the organic semiconductors of interest. Effect of Background Doping-Bulk Limited Transport. In J-V characteristics, a slope of the ln J versus ln V plot of approximately unity indicates an ohmic conduction. Assuming conduction is via holes, the current flow may be expressed in the form84

(VL)

J ) eh0µh

(10)

where h0 is the concentration of thermally generated holes in the valence band, µh is the hole mobility, and L is the thickness of the semiconductor. The concentration of holes at thermal equilibrium is given by

[

h0 ) NV exp -

]

(EF - EV) kT

(11)

where NV is the effective density of states in the valence band and (EF - EV) is the separation of the Fermi level from the valence band edge. The current density in the ohmic region becomes

() [

J ) eµhNV

]

(EF - EV) V exp L kT

(12)

By plotting ln(J/V) against 1000/T, the values eµhNV and (EF - EV) can be calculated from the slope and the intercept at 1/T ) 0. When holes are injected into the p-type bulk of a metal/ p-type semiconductor Schottky diode, the total hole density NV(x) at the point x measured from the ohmic (metal) contact consists of two parts: (1) the density of the existing holes h0 induced by the background acceptors and (2) the density [NV(x) - h0] of the injected holes. The total density of the holes determines the current. The space charge is determined by the distribution of injected holes, provided that the charge of the existing holes h0 is compensated by the ionized fixed dopant ions. As long as the total hole density NV(x) is equal to the doping density over most of the sample thickness, there is no space charge present, and it is expected that Ohm’s law will be obeyed. The electric field in the doped samples becomes constant over practically the whole length of the sample. This assumption would be equivalent to performing an electrochemical experiment in which the resistance of the solution dominates and no electrochemical activity at the surface electrode (i.e., contact) can be detected. In effect, such a case corresponds to a chemiresistor in which the useful information is obtained from the changes of bulk resistivity of the sensing layer. This is not the case for Schottky barrier junctions.

If current is passing through a forward biased junction with a low doped semiconductor or a semiconductor exhibiting low mobility, a depletion layer will be formed in front of the ohmic contact of the Schottky barrier diode, as discussed by Chen et al.85 It can be described in analogy to the depletion polarization effect in an electrochemical experiment by a limited charge carrier transport process inside the semiconductor toward the ohmic contact. The width of the mass-transport limited contact resistance can be modulated by the external field. The time dependence of the polarization resistance is controlled by the charge carrier mobility and by the geometry of the contact. In most Schottky barrier devices, the ohmic contact resembles an electrode with a rectangular geometry, for which the polarization resistance increases with 1/xt at constant applied voltage. The current increases roughly with |V|1/2 (see also section 3.2.2). Moreover, a second bulk effect has to be taken into account in the presence of an intrinsic or low doped semiconductor forming the Schottky barrier junction. If the semiconductor has no free or low charge carrier density due to doping or due to intrinsic thermal ionization, the theory of space charge limited current (SCLC) applies and the current density J is given by the V2 (Mott-Guerney) Law81,86

9 V2 J ) s µ 3 8 L

(13)

where s is the permittivity of the semiconductor and L is the thickness of the sample. With increasing applied voltage, the charge carrier density of the undoped sample increases continuously with x. The injected charge carrier density becomes considerably larger than the doping induced density in most of the volume of the sample and the electric field increases with xx. The behavior of the current density at high currents, and therefore at high voltages, is quite different. The curves for high and low doped semiconductors become identical. One would therefore expect that the current would obey the SCLC V2 law at high voltages in the doped material as well. The transition from Ohm’s law to SCLC takes place at the point at which the Ohm’s law straight line and the SCLC line (corresponding to the V2 law) meet. Therefore, an expression for the voltage Vtr, at which the transition occurs, can be derived by equating the ohmic current and the SCLC (eqs 10 and 13):

8eh0L2 Vtr ) 9s

(14)

This equation provides a method for determining the background doping concentration. The transition voltage Vtr can be experimentally determined. Effect of Trapping and Field Dependent Mobility. A modification of the V2 law occurs when trapping and the dependence of the mobility on the electric field are taken into account. The effects of trapping and field dependent mobility on the J-V curves have been discussed by several authors.86-89 Most workers have assumed an exponential distribution of traps Ntr given by

( )

Ntr(E) ) NV exp -

E kTtr

(15)

where NV is the density of states for a p-type semiconductor

376 Chemical Reviews, 2008, Vol. 108, No. 2

Potje-Kamloth

semiconductor that there is only a thin barrier separating the metal from the semiconductor. For heavily doped semiconductors (doping level: 1019 cm-3 or higher) exhibiting a depletion region width at the metal/semiconductor interface on the order of 3 nm or less, or for operation at low temperature, it may be possible for electrons with energies below the top of the barrier to readily tunnel across such barriers by quantum mechanical tunneling.82 If the tunneling current dominates the current flow, the current-voltage relationship is of the form

( )[

Figure 5. Dark J-V characteristic of an Al/doped PPy Schottky diode. The dashed line indicates the fit of ln J vs V according to thermionic emission (see section 3.2.2).

and Ttr is the characteristic temperature of the distribution. The ratio of free to trapped charge carrier density θ for a p-type semiconductor is given by

θ)

[

]

NV -(Et - EV) exp Ntr kT

(16)

where (Et - EV) is the activation energy of hole traps. For the distribution of traps, the current density J is given by the J ) ktrVm power law, where ktr is a constant and the value of the exponent m depends on the characteristic temperature Ttr. The effect of field dependent mobility has also been discussed by several authors.89,90 Most authors have used the following equation for the mobility:

µ ) µ0 exp(βxE)

(17)

Variations of this equation are discussed by Campbell et al.89 It is shown that both SCLC with traps and field dependent mobility can give rise to the equation J ) ktrVm. However, experimental currents measured at different temperatures could not be described with one value of m. At a given temperature and in a small voltage range, the Vm law with constant m has been experimentally verified.89 Figure 5 shows a typical current density vs applied voltage curve of an Al/PPy Schottky diode. The diode is forward biased at positive voltages, implying that the Al contact is negative. The current density in the low bias region between 0.1 and 1.5 V varies exponentially with the bias voltage and fits closely with the thermionic emission theory (eq 6). At a bias voltage > 1.5 V, the current starts to become linear, indicating a dominant contribution of the bulk resistance, which is in series with the resistance of the depletion layer of the diode. As the voltage increases to more than 5 V, the current increases again. This indicates the transition from the contribution of the ohmic current to the contribution of the SCLC, according to eq 14. The current density J varies with Vm. The determination of the exponent m by plotting J as a function of Vm gives information if the SCLC regime is trap free and the mobility is field independent (m ) 2) or controlled by trapping or by the effect of the electric field on the mobility (m > 2).

3.2.4. Tunneling through the Barrier An alternate and more practical contact is a tunnel contact, if no appropriate metal is available to form an ohmic contact. It can be particularly problematic to find a metal that provides an ohmic contact to doped organic semiconductors that usually show work functions of about 5 V or higher. Tunnel contacts do have a positive barrier at the metal/semiconductor interface. They also have a high enough doping level in the

J ) J0 exp -

φb eV exp -1 E00 kT

( )

]

(18)

where E00 ) (h/4π)(ND/m*s)1/2/eV, where h is Planck’s constant and m* is the effective mass of the electron. The equation indicates that the tunneling will increase exponentially with xND. A possible mechanism for the charge transfer from the metal to a conductive polymer based on tunneling through the barrier is given by Gustafsson et al.91 Mobile charges in the conducting polymer exist in the form of polarons and bipolarons which occupy states in the band gap (see section 2.2.2). They are different from the valence band holes normally encountered in inorganic semiconductors. Injection of a hole by tunneling or by thermionic emission into the polymer should be similar to the photogeneration of charge in conducting polymers, where, after its formation, a rapid relaxation to polaronic/bipolaron states occurs.92 Thus, the majority carrier transfer from the metal to the polymer is based on (1) tunneling through the barrier to the valence band, followed by (2) an immediate relaxation to a polaron or bipolaron defect state, whereas, in the opposite direction, the charge transfer occurs by direct tunneling from the defect state to the metal. This gives rise to different tunneling paths at different energy levels in the forward and reverse bias voltage directions, which probably can affect the rectification ratio of the device.

3.3. Interfacial Electronic StructuresDeviations from the Ideal Case 3.3.1. Introduction of the Ideality Factor Real Schottky diodes do not always follow the expressions derived for the ideal case of a charge carrier transport mechanism across a potential barrier. Deviations from ideal behavior arise from imperfections in fabrication, or factors, which are not included in the relatively simple theories taken for derivation of eqs 5-9 and eq 18. A few of the major limitations are given below. In practice, diodes never satisfy eq 5 exactly but can be more closely described by the modified eq 19.82

J ) J0 exp

eV (nkT )[1 - exp eVkT]

(19)

or

J ) J0 exp

eV (nkT )

for V .

3kT e

(20)

where n is the ideality factor, which can be estimated from the slope of the straight part of the ln J vs V plot and is usually greater than unity. The ideality factor n is introduced in eq 19 to account for the failure of the simple, single-

Semiconductor Junction Gas Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 377

Figure 6. Schematic band diagram of the band bending interface in the presence of surface states for a metal/semiconductor: (A) before contact; (B) after contact formation. The Fermi level is pinned by a high density of surface states of the semiconductor.

carrier theories of barrier behavior, e.g., when image lowering of the potential barrier occurs. There are many other possible reasons why n should exceed unity, e.g., series resistance of the bulk semiconductor, generation-recombination within the depletion region, tunneling through the barrier, interface states, and the presence of an insulating interfacial layer.82,93 In the case of polymeric diodes, bulk resistance and interfacial layer effects are likely to be the most important.

3.3.2. Schottky Barrier Lowering It has been assumed that the Schottky barrier height remains constant under all conditions of applied voltage. However, the barrier height varies with applied voltage because conduction electrons experience a force from their image charges in the metal. This force attracts the electrons toward the metal surface, effectively lowering the barrier and allowing voltage-dependent deviations from ideal behavior. In theory, this “image force” should give the reverse current a fourth-power dependence of bias upon voltage, rather than the constant value implied by eq 7. This effect is usually not observed, because charge carrier generation in the depletion region at high reverse bias and tunneling effects dominate the reverse leakage. At forward biases above approximately 0.1 V, the effect causes a slight deviation of the ideality factor n from unity. For a diode that is assumed to be ideal except for barrier lowering, the ideality factor is

(

n) 1-

)

dφb dV

-1

(21)

where dφb/dV is the derivative of the barrier height with respect to the applied voltage. The image force lowering of the barrier ∆φbi is given as

∆φbi )

( ) eE¨ max 4π0

1/2

(22)

E¨ max is the maximum field at the interface, and 0 ) 8.85 × 10-14 C cm-1 V-1 is the permittivity of free space.

3.3.3. Effect of Interface States Very often, a deviation from the ideal Mott-Schottky behavior is observed. It can be measured in terms of the slope parameter

S)

dφb dφm

(23)

describing the dependence of the Schottky barrier height on the metal work function. According to the Mott-Schottky

model, S should be equal to unity, while S is smaller than unity for any deviation. An important limitation of the Mott-Schottky model is the neglect of intrinsic surface states with energy levels located in the semiconductor band gap which are pinning the metal Fermi level as shown in Figure 6. In this extreme case, the Schottky barrier height does not depend at all on the metal work function, i.e., S ) 0. Considering a continuum of interface states, a phenomenological formula for the Schottky barrier height can be formulated76

φb ) S(φm - χs) + (1 - S)φ0

(24)

where S is the slope factor and φ0 is the barrier height when the charge neutrality level of the continuum of interface states coincides with the Fermi level at the interface. The barrier height becomes independent of the metal work function and is determined entirely by the doping and surface properties of the semiconductor. In real situations, the constant S lies between 0 and 1. “Fermi level pinning” indicates that the Fermi level position at the surface of the semiconductor measured relative to the vacuum level does not vary when either the work function or the Fermi level or the contacting phase is varied. Under strong Fermi level pinning, a constant electric potential is dropped across the semiconductor regardless of the nature of the metal. The remaining potential is dropped across a thin dielectric layer near the metal/ semiconductor junction. Besides the interface states, other reasons have been proposed for inorganic semiconductor/metal junctions as explanations for the Fermi level pinning behavior, i.e., chemical reactions such as alloy formation,94 or other stoichiometry changes,95 and/or quantum mechanically induced interface states.96 Although these theories were primarily developed to explain the behavior of inorganic semiconductor/metal contacts, many of them also make testable predictions regarding the behavior of contacts comprising organic semiconductors in Schottky barrier junction sensors.

3.3.4. Metal/Organic Semiconductor Junctions with an Interfacial Layer The elucidation of the interfacial electronic structure, particularly the energy-level alignment at organic/substrate interfaces, forms the basis for understanding and improving the performance of organic electronic devices.97,98 Despite the advances in device application, there is still only a limited understanding of the interface between organic materials and metals. A number of recent studies have helped to understand the basic properties of the interface; see, e.g., the review by

378 Chemical Reviews, 2008, Vol. 108, No. 2

Potje-Kamloth

Figure 7. Energy diagram of the organic/metal interface with (A) band bending leading to Fermi level alignment and (B) without band bending. Evacuum denotes the vacuum level, EF is the Fermi level, eVi is the shift of the vacuum level at the interface due to the dipole layer formation, and ∆ is the interfacial layer thickness.

Seki at al.99 and references therein. In particular, ultraviolet and X-ray photoemission spectroscopy (UPS, XPS) have been used to investigate metal/organic100,101 and organic/ organic102-104 interfaces. The organic layers employed in such studies were nominally undoped, although the purity of the organic layers is usually insufficient to achieve intrinsic behavior. The simplest model used to discuss the energy level offset at these interfaces is the “vacuum-level alignment” model (i.e., Mott-Schottky limit, eq 2). The band of an organic semiconductor in contact with a metal is bent to achieve the alignment of the Fermi level of the metal with the Fermi level of the organic layer fixed somewhere in the band gap, simply by analogy to an interface between a doped inorganic semiconductor and a metal. As shown for several interfaces between an undoped organic semiconductor and a metal layer prepared in UHV, this is not always true.105 If the Fermi level fixing and its alignment with the metal Fermi level occur by atmospheric doping for an organic semiconductor, this assumption may become valid for practical devices prepared and used under atmospheric conditions such as ambient air. The doping process is expected to cause a characteristic band bending at organic semiconductor/metal interfaces. Several problems have remained in understanding whether or not this simple model is applicable to typical organic/inorganic and organic/ organic interfaces. Among these, the two main problems are (1) little is confirmed experimentally about the effects of the molecular orientation and/or packing structure on electronic structures near organic/organic and organic/inorganic interfaces and (2) it is not easy to clearly distinguish the electronic structure in the “contact interface” from that in the bulk region.106 Changes in the molecular orientation during film growth may cause unintentional dipole-layer formation in the film107 and unintentional changes in the ionization potential of the film.108 In contrast, several groups have reported the electronic structure of a well-defined organic/substrate contact interface.109-111 The organic/substrate interface can be divided into two adsorption systems:99 (1) a chemisorbed system, in which the interface dipole is formed by chemical interactions and electron transfer, which gives rise to pinning of the energy levels of the organic semiconductor to the substrate Fermi level, independent of the initial semiconductor work function, and (2) a physisorbed system, in which the interface dipole is formed by the push-back effect and electron transfer. The important aspects can be discussed using Figure 7. When a metal and an organic solid come into contact, a dipole layer may be formed right at the interface, due to

rearrangement of electronic charge. Several mechanisms have been identified to explain this type of interface dipole.38,112 Possible origins of the rearrangements are, e.g., charge transfer across the interface, redistribution of the electron cloud spilled out of the metal surface, mirror force, formation of an interfacial electronic state, interfacial chemical reaction, and aligned permanent dipoles of adsorbed molecules.99 Such interfacial dipole layer formation results in an abrupt shift of the potential across the dipole layer, leading to a shift of hypothetical vacuum levels (eVi) at the interface, as shown in Figure 7. If the organic semiconductor is doped, for example by an acceptor, as shown in Figure 7A, the dopant can be ionized to form a fixed space charge, and the energies of the occupied and unoccupied levels are shifted with increasing distance from the interface, indicating the space-charge region at the metal/semiconductor interface. However, most organic semiconductors are used, at least nominally, undoped and the electronic structure of the semiconductor is represented without surface “band bending”, as shown in Figure 7B. The presence of a space-charge region at the interface between a doped semiconductor and a metal, with a significant molecular level bending in the doped layer, has been amply demonstrated.39,113 The magnitude of the spacecharge region has also been shown to be larger in wide band gap semiconductors with low intrinsic charge carrier concentrations and low conductivity,114 which are common characteristics of organic semiconducting materials. Nishi et al. discussed the effect of bulk oxygen doping on the interfacial electronic structure of titanyl phthalocyanine films (TiOPc) deposited on various metal substrates.105 Films prepared under UHV did not show the alignment of the metal Fermi level with the Fermi level of the organic semiconductor at fixed energy within the band gap, as shown in Figure 7B, possibly because of insufficient charge density in the TiOPc films. On the other hand, there is a conversion from n- to p-type semiconductor seen, when the film is exposed to O2, as well as a clear alignment of the Fermi level together with the formation of a space-charge region and an upward molecular level bending (Figure 7A). Heeger et al. presented a model to understand the interactions at doped polymer/metal interface48,115 which was refined by Brazovskii et al.116 The model considers specifically the charge carriers in degenerated conjugated polymers. The nondegenerate continuum model of Davids et al.117 extended the work toward conjugated polymers with a nondegenerate ground state. Based on the relative energies between the Fermi level of the metal (EF) and the formation energy of polarons and bipolarons of the polymer, Eform,

Semiconductor Junction Gas Sensors

conclusions about the extent of charge transfer during interface formation can be made. If EF of the contact is higher in energy than the Eform of the negative bipolaron/polaron, the transfer of a large negative charge density into the polymer occurs, whereas a large positive charge density is transferred if EF of the contact is lower than Eform of the positive bipolaron/polaron. In addition, Davids et al. suggested that the presence of traps in the polymer will accommodate charges as well.117 After values are assigned to the polaron and bipolaron formation energies, as well as to the trap energies and densities, it is possible to model the potential and charge density at the metal/polymer interface for a given Fermi energy with respect to the center of the energy gap by solving the Poisson equation. As a result, the transferred charge remains close to the metal/polymer interface. By considering the metal/polymer interactions from the microscopic point of view of quantum interactions, a molecular modeling was carried out, which took into account the quasi-one-dimensionality of the polymer system.118-120 As a result of the difference in the Fermi levels of the metal and the conducting polymer, it is expected that charge transfer may occur across the interface. Furthermore, the quantum calculation approach considers specific chemical reactions or even charge donation between the metal and the polymer as a result of bringing metal atoms in the vicinity of a conjugated system and observing the changes in the electronic structure of the resulting system. Based on such calculations, it has been postulated that the metal plays an important role in determining the extent of charge transfer that occurs in metal/conducting polymer interfaces: Ca and Na appear to transfer charge at the interface without significantly altering its chemistry.118,119 Al, on the other hand, disrupts the chemical structure of the surface of the conducting polymer.119 The results obtained in the case of Ca and Na agree with the general behavior postulated by Davids et al.,117 whereas Al deposition introduces an additional effect of surface chemical reactions, which are not considered by the continuum model.

3.4. Gas/Solid Interactions in Schottky Barrier Junction Devices The key difference between organic and inorganic Schottky barrier junction devices is the ability of many gases and vapors to penetrate through the organic semiconductor either to change the Schottky junction resistance or to interact with the semiconductor itself, which causes a work function change of the material, as shown in section 3.4.1. In the case of an inorganic Schottky barrier junction device, gas permeation toward the metal/semiconductor interface works only for hydrogen or hydrogen-producing compounds and then gives rise to formation of a dipole layer, as discussed in section 3.4.2. Adsorption at the catalytic metal surface has only a limited secondary effect, i.e., the dissociation of molecular hydrogen to atomic hydrogen.

3.4.1. Bulk Effect upon Gas DopingsModulation of the Electron Work Function The rectifying behavior of the Schottky diodes relates directly to the Fermi level of the layer of the organic semiconductor and can be influenced not only by the structure, the dopant type, and the doping level,121 but also by the interaction of the semiconductor with gases or vapors. By exposure of the organic semiconductors to certain gaseous

Chemical Reviews, 2008, Vol. 108, No. 2 379

species exhibiting electron donor or acceptor behavior, the Fermi level position is changed27,30,122 by either a decrease or an increase in the doping level. In this case, a gas-induced change of the Schottky barrier characteristics can be observed, which can be exploited for gas sensing measurements. In 1980, Van Ewyk et al. proposed the formation of a charge-transfer complex due to an electron donator/acceptor interaction during the absorption of strong electron donating or accepting gases on organic p-type semiconductors.123 The absorption causes a change in the doping level of the semiconductor and thus a change in its conductivity. The transfer of a partial amount of electron density without complete ionization of the reacting species is generally known for adsorption processes of molecules on conducting surfaces and has been treated comprehensively, from both the catalysis124 and chemical125 points of view. It was assumed that this type of reaction is also responsible for the chemical modulation of the electron work function of the polymer layer due to gas absorption.126 This type of interaction process is often described as a secondary doping of the sensitive semiconducting layer. The primary doping process is carried out during formation of the semiconducting sensing layer by incorporating doping ions. In the case of an organic semiconductor, primary doping is carried out during the chemical or electrochemical preparation process.67,127 Janata presented a model that describes the potential concentration relationship based on the formation of a chargetransfer complex between the secondary dopant (gas or vapor) and the matrix (sensitive phase), which is combined with a fractional charge transfer δ.128 The dependence of the Fermi level position of the sensitive material on the partial pressure, pgas, of the dopant gas or vapor is given by

EF ) E/0 +

kT ln(pgas + const) 2δ

(25)

where in E/0 the donor/acceptor level of the organic semiconductor and the equilibrium coefficients of all relevant reactions are combined. For pgas ) 1, EF equals E/0, denoted as the value of the Fermi level under standard conditions. A change of the Fermi level is reflected in the change of the work function EWF of the semiconductor

∆EF ) ∆EWF

(26)

Equation 25 has the familiar form of the Nernst equation for ion and electron transfer across the interface of two condensed phases, except that they account for the fractional charge transfer δ. The magnitude and the polarity of the response depend on the value of the charge-transfer coefficient δ, and with 2|δ| < 1, a fractional value of the slope is possible. It depends on the ability of the entering gas or vapor molecules to exchange charge density with the semiconductor matrix either by oxidizing (2δ < 0) or reducing (2δ > 0) the active sites of the matrix. The gas/vapor behaves like an electron acceptor and donor, respectively. In the absence of interfacial layer and Fermi level pinning, the gas-induced change of the Fermi level causes a change in the Schottky barrier characteristics in the low bias region. The diode current is then exponentially related to the change of the Fermi level, i.e., the work function of the semiconductor and the applied voltage, according to eq 5.

380 Chemical Reviews, 2008, Vol. 108, No. 2

Potje-Kamloth

adsorbed species. In the steady state, the coverage is

()

c1 θ ) d1 (1 - θ)

Figure 8. Band diagrams of a catalytic metal/semiconductor Schottky diode without and with hydrogen adsorption, respectively.

Figure 9. Cross-sectional view of a typical geometrical arrangement of a Schottky barrier diode used for chemical sensing applications.

It is known that a number of metals, e.g., palladium or platinum, adsorb and dissolve hydrogen and that the adsorbed atoms change the work function of the metal surfaces. By using a thin metal film as the catalytic metal electrode of the Schottky diode, hydrogen and other gases that react with the catalytic metal can be detected. Hydrogen molecules are adsorbed and dissociated on the outer metal surface. Then, the atomic hydrogen permeates through the bulk lattice toward the metal/semiconductor, causing a perturbation at this interface that gives rise to a change in the sensor output signal. Two hypothetical steps were proposed to explain the detection mechanism:129 (1) the hydrogen atom is polarized at the metal/semiconductor interface, which gives rise to a dipole layer, or (2) an excess of charge states at the metal/semiconductor interface is created in the presence of hydrogen and reduces the Schottky barrier height. Figure 8 illustrates the band diagrams without and with hydrogen adsorption, respectively. The hydrogen-polarized layer, arising from the intrinsic electric field of the diode, redistributes the charges in the depletion region and abates the degree of band bending. Hence, the corresponding current is correlated with the number of hydrogen atoms adsorbed at the metal/semiconductor interface. When atomic hydrogen has been formed on the outer metal surface, which is exposed to the ambient (see Figure 9), an equilibrium between the hydrogen concentration at this metal surface and that at the metal/semiconductor interface is reached. Based on this model, the change in the J-V characteristics and the decrease in the Schottky barrier height are strongly related to the hydrogen concentration. Assuming that the adsorption is not affected by conditions outside the metal surface, the metal/semiconductor interface has similar adsorption properties as the free metal surface. The fractional coverage θ is equal at the outer surface and the inner interface in the steady state. At the outer surface, the following reaction occurs in pure hydrogen: c1

H2(g) 79 8 2Ha d

(27)

1

where (g) refers to gas-phase species and (a) refers to

pH2(g)1/2

(28)

where θ is the fractional coverage of hydrogen and pH2(g) is the hydrogen gas pressure. Equation 28 is valid, for both the metal/semiconductor interface and the outer metal surface. The change in the build-in voltage can be written as130

∆Vbi ) -µNadsθ/S

(29)

where µ is the dipole moment of an adsorbed hydrogen atom, Nads is the density of adsorption sites, and S is the permittivity of the semiconductor. Moreover, the increase of the interfacial charge density causes the change in the ideality factor. The analytical relation between the ideality factor and the interfacial charge density is given by131

n)1+

3.4.2. Formation of a Dipole and an Interfacial Layer

1/2

(

)

∆ s + eNssb i W

(30)

where ∆ is the interfacial layer thickness (see Figure 7), i and s are the permittivities of the interfacial layer and the semiconductor substrate, respectively, W is the depletion layer width, e is the electronic charge, and Nssb (cm-2 eV-1)131 is the density of the interface states that are in equilibrium with the semiconductor.

3.5. Extraction of Schottky Diode Parameters for Sensor Applications 3.5.1. Introduction In Figure 9, a cross section of a typical chemical sensor based on a Schottky junction is shown. Each Schottky barrier diode comprises two contacts, or junction areas, by definition: (1) between metal I and the semiconductor forming the Schottky barrier junction, which is the origin of the sensor signal, and (2) between metal II and the semiconductor forming the ohmic contact, which has to be inactive and, hence, should not contribute to the sensor response. The type of contact is defined by the work function of the materials used for fabrication (see section 3.2.1 and Figure 4). It should be noted that the area of the ohmic contact of the sandwich structure shown in Figure 9 is much larger than that of the Schottky junction, in analogy to auxiliary and working electrodes in electrochemical experiments. In Figure 10, the schematic band diagram and the equivalent circuit of a chemical sensor based on an ideal Schottky junction device, i.e., without an appreciable interfacial layer or interface states, is shown. The equivalent circuit of the Schottky barrier diode in Figure 10 shows those parts which can be influenced by the chemical interaction of gases with the chemical sensing material: (1) the barrier region (of thickness WJ) formed by metal I and the semiconductor having a capacitance CJ in parallel with the junction resistance RJ, and (2) the neutral bulk (of thickness WB) of the semiconductor with a bulk capacitance CB in parallel with the bulk resistance RB. Both metal I and the semiconductor, forming the Schottky barrier junction, can be used as the chemical sensing components interacting with gases and vapors. Their choice and combination defines the working principle as well as

Semiconductor Junction Gas Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 381

Figure 11. Typical change in the J-V characteristics of a Schottky barrier diode due to interaction with a gas or vapor of interest. ∆V indicates the gas or vapor induced voltage shift at constant current density.

Figure 10. Schematic band diagram and equivalent circuit of an ideal Schottky barrier diode.

the sensitivity and the selectivity of the sensor device. The generation of the “sensor response”, i.e., sensor signal, of the Schottky barrier diode is originated in the change of the junction characteristics (i.e., J-V or C-V characteristics) by alternating the Schottky barrier height or the built-in voltage. The response can be due either to adsorption of the species of interest at the metal surface affecting interfacial polarization by formation of a dipole layer or to absorption of gases or vapors of interest by the semiconductor and their interaction with the semiconductor, which changes its work function and, hence, the contact potential or built-in voltage of the diode. The former is mostly related to Schottky barrier diodes based on inorganic semiconductors and is discussed in section 4.1, whereas the latter is related to Schottky barrier diodes based on organic semiconductors discussed in section 4.2. It should be noted that the interaction of gases or vapors with the semiconductor, which causes a change of the doping level, leads not only to a change of the work function of the semiconductor, i.e., a change of the junction resistance RJ and the junction capacitance CJ of the Schottky barrier junction, but also to a change of the bulk resistance RB of the semiconductor (see Figure 10). In this case, the sensor device works as both a junction controlled device, with a diode current in the low bias voltage region exponentially related to the change of the built-in voltage, and a bulk controlled device in the high bias region, with a diode current linearly related to the change in conductivity. The latter represents a simple chemiresistor device. These compensatory effects are discussed in section 4.2.4.3. In general, the gas sensitivity of diodes is represented by the so-called gas-induced voltage shift, ∆V, extracted from current voltage characteristics and is defined as29,132

∆V ) Vgas - Vair at J ) constant

(31)

where Vgas and Vair are the applied voltages at a constant current density in gas and in air, respectively. In Figure 11 a typical change in the J-V characteristic of a Schottky barrier diode due to interaction with a gas or vapor of interest is shown. ∆V indicates the gas or vapor induced voltage shift

at constant current density. It can be either junction (i.e., ∆VJ) or bulk (i.e., ∆VB) controlled, as discussed above. The extraction of the sensing information from Schottky barrier diodes can be done by operation in constant current mode. Hence, the sensor response is the output voltage. Figure 20 shows a typical example of an experimentally obtained transient response of the output voltage of a Schottky barrier diode operated in the constant current mode.

3.5.2. Current−Voltage Characteristic Taking into account the image force between an electron and the surface of the metal, eq 7 can be rewritten as82

[

J0 ) A**T2 exp -

]

φb - ∆φbi kT

(32)

For metal/semiconductor contacts, the value of J0 can be readily correlated with a thermodynamic property of the interface, the barrier height of the junction, φb. Provided that the current-voltage characteristics are sufficiently close to ideal characteristics to allow a reliable value of J0 to be determined, the effective barrier height can be deduced in two ways: (1) If A** is known, the value of J0, found by extrapolating the straight part of the ln{J/[1 - exp(-eV/kT)]} vs V plot to V ) 0 (eq 19), immediately gives the effective barrier height φe ) φb - ∆φbi. A** is often not known with any great precision, but because an error of a factor of 3 in A** gives rise to an error of about kT/e in φe, an imprecise value can usually be used unless a very accurate measurement of φe is required. The determination of the zero-bias barrier height from current-voltage characteristics is only reliable if the semilog plot of J vs V is linear with a low value of n (1 < n < 1.1). For large values of n or a nonlinear plot of ln J as a function of V, the Schottky barrier is far from ideal. This is probably due to the presence of a thick interfacial layer and/or to recombination in the depletion region and/or to tunneling conduction. (2) If A** is not known, a plot of ln(J0/ T2) against T-1 should give a straight line of slope -φe/k and intercept on the vertical axis equal to ln A**. This is the most common case for the Schottky diodes based on organic semiconductors, because of the lack of the effective mass of charge carriers. The barrier height is generally a decreasing function of temperature, because the expansion of the lattice causes a change in the band gap. To a first approximation, one can write φe(T) ) φe(0) - bT, in which case the method gives the barrier height at absolute zero.

382 Chemical Reviews, 2008, Vol. 108, No. 2

Potje-Kamloth

A problem arises if the diode has a large series, i.e., bulk resistance (see Figure 10), or the semiconductor is low doped. This is typical for most polymeric Schottky diodes.133 The resistance is mainly due to the resistivity of the polymer film. In this case, the region over which the plot of ln J vs V is linear may be small, and accurate extrapolation to zero voltage may be difficult. Moreover, the polymer surface cannot be exposed to any chemical etching process; therefore, it can be expected that irregularities will be brought into contact with the metal and the effective area of the junction will be somewhat uncertain. To circumvent these difficulties, the use of a Chot plot,134 which is a modified Norde plot, has been suggested by Ingana¨s135 to determine the effective barrier height.

3.5.3. Capacitance−Voltage Characteristic An independent way to determine the barrier height involves the measurement of the differential capacitance of the Schottky barrier contact. The procedure is known as the Mott-Schottky or capacitance-voltage method. According to the Schottky theory, the depletion layer capacitance (/F‚ cm-2) of the metal/semiconductor contact

C)

δQ δVr

(33)

can be expressed as81

C

-2

)

(

2 Vbi - V -

kT e

)

eSN

(34)

where Q is the charge associated with the band bending, Vr is the reverse bias voltage, V is the applied voltage, and N is the free charge carrier concentration. If φb is independent of V (i.e., if there is no appreciable interfacial layer), a plot of C-2 against V (Mott-Schottky plot) should give a straight line with an intercept -VI on the horizontal axis equal to -(Vbi - kT/e). The barrier height is then given by

φb ) eVbi + ξ ) eVI + kT + ξ

(35)

The free charge carrier concentration, N, can be determined from the slope of the C -2 vs V plot. The depletion layer width, W, is given by

W)

( ) 2sφbi eN

1/2

( )[ ] -2 dV 2 C es d

tion, which disappears at low frequency. A compromise must be found to minimize these two contributions. This method of barrier height determination is generally straightforward for metal/semiconductor contacts, yielding the flat-band barrier height, which does not include the image force lowering present at the junction. However, when some of the applied voltage is dropped across an interfacial dielectric layer, or when traps are present in an interfacial layer, the capacitance-voltage data will not generally adhere to the simple relationships of eq 33 and eq 34, respectively. Hence, the capacitance method gives results which can differ significantly from those obtained by the current-voltage methods.

3.5.4. Impedance Measurement To gain more insight into the nature of junctions and to obtain the equivalent circuit models for diodes, the complex impedance is usually measured as a function of frequency at different bias voltages.137 This methodology is analogous to impedance analysis used in electrochemical experiments. It has been proposed that an ideal Schottky diode (i.e., without an appreciable interfacial layer) can be divided into two regions: (1) the barrier region (of thickness WJ), with a capacitance CJ in parallel with the junction resistance RJ, and (2) the neutral bulk (of thickness WB), with a bulk capacitance CB in parallel with the bulk resistance RB (see Figure 12A).133 In most practical metal/semiconductor contacts, the conditions RJ . RB, CJ . CB, and ωmaxτB , 1 are satisfied.133,138 The plane plot of the complex impedance is shown in Figure 12B. From Figure 12B, the bulk resistance RB and the junction resistance RJ can be estimated by the intercepts of the high- and low-frequency extrapolations of the semicircle with the horizontal axis, respectively. Using these values of RJ and RB, the junction capacitance (for each frequency) and thus the depletion width can be fitted according to

(36)

If N is not constant, the plot is no longer linear. In this case, the differential capacitance method can be used to determine the doping profile. The derivative of eq 35 gives81

N)

Figure 12. (A) Common equivalent circuit of a Schottky diode junction. (B) Impedance plane plot. Arrows indicate the direction of increasing frequency. The dotted lines imply that the bulk capacitance CB is often not appreciably distributed.

Zr )

RJ 1 + ω Rj Cj

Ri ) (37)

In practice, the junction is reverse polarized, and small oscillations δV are superposed. If no trap exists, the variation of the charge associated with the band bending δQ corresponds to the variation of the free charge carrier concentration at the edge of the depletion zone. If traps exist, they contain part of the charge δQ. To get rid of them, one has to perform measurements at high frequency (>100 kHz).136 For such a frequency, only the semiconductor depletion charge changes. Another problem then arises: the bulk resistance contribu-

2

2

2

+

ωRJ2CJ

RB 1 + ω2RB2CB2

1 + ω2RJ2CJ

2

ωRB2CB 1 + ω2RB2CB2

(38)

(39)

In most practical metal/semiconductor contacts, the ideal situation shown in Figure 12A is never reached because there is usually a thin insulating layer with a capacitance CI in parallel with the resistance RI (see Figure 12B). The insulating layer has three effects:82 (1) because of the potential drops in the layer, the zero-bias barrier height is lower than it would be in an ideal diode, (2) the electrons have to tunnel through the barrier presented by the insulator so that the current for a given bias is reduced in a manner equivalent to a reduction in the Richardson constant A**,

140

1.7 × 1.33/-0.02 (500 ppm) 0.59/-0.005 (800 ppm) -/-0.01428 (500 °C) -/-0.00547 (500 °C) 1.73/-0.05 13.5

RT

synth air

1.7 × 10-4/2.1 × 10-4

h h ) homogeneous. a

sat. current/gas induced sat. current/A

Pt-Pd/(porous)Si

From the rapid progress in semiconductor materials and devices, solid-state hydrogen sensors based on the Schottky diode structure with a catalytic palladium metal have attracted intensive interest.10,140 Silicon based Schottky diodes are inexpensive, and they are compatible with monolithic silicon integrated circuit technology. The measured diode current

h h h h h

4.1.1. Silicon Based Schottky Diodes

Table 1. Electrical Parameters of Schottky Barrier Diodes Based on SiC and Si

In the field of solid-state chemical sensors, Schottky diodes and related devices (MOSFET, MIS structures, etc.) incorporating catalytic metals have been widely studied for their application in the detection of hydrogen and hydrogenproducing compounds.6 In hydrogen sensors based on Schottky diodes, the key role is played by the Schottky metallization, mostly belonging to the platinum group metals, which ensures the catalytic dissociation of molecular hydrogen into hydrogen atoms (eq 27). It should be noted that the semiconducting substrate is only used to provide sufficiently high Schottky barrier heights, necessary for good sensing performance, but is not the gas-sensitive part of the sensor. According to the dipole model, originally developed by Lundstro¨m,139 the hydrogen sensitivity of catalytic gate devices is based on the dissociation of hydrogen molecules on the catalytic metal surface and the diffusion through the metal film to form a polarized layer at the metal/insulator interface (see section 3.4.2). The polarized layer gives rise to a shift of the capacitance-voltage (C-V) curve of a MOS capacitor or the current-voltage (J-V) curve of a Schottky diode. Catalytic gate devices also respond to hydrogencontaining molecules such as hydrocarbons, provided that the molecules are also dissociated on the catalytic metal surface. Oxygen atoms are also dissociated on the catalytic metal surface. Water formation with oxygen atoms from oxygen-containing molecules consumes hydrogen and, therefore, decreases the sensor response. In other words, catalytic metal gates have a direct response to hydrogen and hydrocarbons as well as an indirect response to oxygen molecules, whose effect is to decrease the direct response. Besides Si, compound semiconductors such as GaAs, InP, SiC, and GaN have been alternatively employed as substrate materials for Schottky diode type hydrogen sensors. Tables 1-3 give an overview of the electrical and gas sensing parameters of the Schottky barrier diodes mentioned above.

ideality factor/gas induced shift, n/Dn

4.1. Schottky Diodes Based on Inorganic Semiconductors

synth air synth air synth air synth air synth air

4. Schottky Diodes and Their Sensor Applications

202-400 500 400-600 400-600 RT

(41)

Pt/n-SiC Pt/n-SiC Pd/-SiC Pt/-SiC Pt-Pd/(polished)Si

e δV kT δ(ln J)

background atmos

n)

barrier height/gas induced shift, φb/∆φb (eV)

in a manner which can be described in terms of an ideality factor n (greater than unity) defined by eq 41

NO/10-500 ppm H2/800 ppm H2/800 ppm CH4/4800 ppm NO2/5-20 ppm NO/50-250 ppm NO2/5-20 ppm

charge carrier density, NA, ND (cm-3)

(40)

operating temp (°C)

kT e

gas/conc

for V .

metal layera

eV (nkT )

structure metal/semiconductor

J ) J0 exp

ref

and (3) when a bias is applied, part of the bias voltage is dropped across the insulating layer, so that the barrier height φb is a function of the bias voltage. This effect changes the shape of the current voltage characteristics of the tunneling current given by81

170 11 165 165 140

Chemical Reviews, 2008, Vol. 108, No. 2 383

1016

Semiconductor Junction Gas Sensors

structure metal/semiconductor Pt/n-GaAs Pt/n-GaAs Pt/n-GaAs Pd/n-GaAs Pd/n-GaAs Pt/n-GaAs Ir/n-GaAs Pt/Al0.3Ga0.9As Pd/Al0.3Ga0.9As Pd/n-GaAs Au/n-GaAs Au/n-GaAs Pd/n-GaAs Pd/n-GaAs (MOS)

metal layera h h h h h p p h h h p p h h

operating temp (°C)

gas/conc H2 /1000 ppm H2 /23-1350 ppm H2 /23-1350 ppm H2 /23-1350 ppm H2 /23-1350 ppm NH3 /6-95 ppm NH3 /23-1350 ppm H2 /49-9090 ppm H2 /49-9090 ppm H2 /1-5% CO /1-5% NO2 /1-5% H2 /500-1500 ppm H2 /5-5000 ppm

50-200 °C 150 °C 50 °C 150 °C 50 °C 150 °C 150 °C 30-120 °C 30-120 °C RT RT RT-250 °C 80 °C/110 °C

background atmos

sat. current/gas induced sat. current/A

synth air synth air/N2 synth air/N2 synth air/N2 synth air/N2 synth air/N2 synth air/N2 synth air synth air synth air N2 N2 synth air N2

6 × 10-10/9 × 10-9 A

ideality factor/gas induced shift, n/∆n

barrier height/gas induced shift, φb/∆φb (eV)

1.57/+0.16 (500 ppm) 1.083/-0.009 (1010 ppm)

charge carrier density, NA, ND (cm-3)

-/-0.19, -0.15 -/-0.02 (100 ppm, N2) -/-0.001 (100 ppm, N2) -/-0.075 (100 ppm, N2) -/-0.025 (100 ppm, N2)

4 × 10

-/-0.074 (9090 ppm) -/-0.07 (9090 ppm) -/-0.28 (5%)

2 × 1017 2 × 1017 2 × 1017 1 × 1015 1 × 1015 1 × 1015 1.5 × 1017

0.95(RT)/-0.08 (500 ppm) 0.786/-0.046 (1010 ppm)

ref

14

7 8 8 8 8 8 8 148 148 147 151 151 150 150

384 Chemical Reviews, 2008, Vol. 108, No. 2

Table 2. Electrical Parameters of Schottky Barrier Diodes Based on GaAs and AlGaAs

Table 3. Electrical Parameters of Schottky Barrier Diodes Based on InP and GaN structure metal metal/semiconductor layerb

gas/conc

Pd/InP (MOS) Pd/n-InP Pd/n-InP (MOS) Pd/n-InP Pd-Ge-Pd/p-InP (pseudojunction) Pd*/n-InPa Pt/AlGaN/GaN

h h h h h

H2 /5-5000 ppm H2 /100 ppm H2 /15-1010 ppm H2 /15-1010 ppm NO2/10 ppm

h h

Pt/AlGaN/GaN IrPt/AlGaN/GaN PdAg/AlGaN/GaN Pt/GaN

h h h h

H2/5 vol % H2/5 vol % H2/5 vol % H2/500-2000 ppm

80 °C/110 °C RT RT-110 °C RT-110 °C RT-100 °C

background atmos

sat. current/gas induced sat. current/A

ideality factor/gas induced shift, n/∆n

N2 synth air synth air synth air synth air

1.072/-0.047 (1010 ppm) 7.5 × 10-8/4.3 × 10-6 1.8/-0.88

H2 /15 ppm-1 vol % RT-118 °C CO/8-160 ppm 250-300 °C

synth air synth air

/3.2 × 10-6

1.12/+2.55

200-800 °C 200-800 °C 200-800 °C 400 °C

N2 N2 N2 synth air

3.5 × 10-12

1.5 1.4 1.6 1.37

Pd* indicates electroless plated. b h ) homogeneous.

1.06/+0.14

barrier height/gas induced shift, φb/∆φb (eV)

charge carrier density, NA, ND (cm-3)

0.677/-0.21 (1010 ppm) 0.72/-0.19 (100 ppm) 0.63/-0.15 (1010 ppm) 0.42/-0.02 (1010 ppm) 0.906/-0.039 (10 ppm)

1.5 × 1016 1.5 × 1017 1.5 × 1017 5 × 1016

150 10 9 9 155

1017

152/153 163

0.588/-0.177 (5000 ppm) 1.01/-0.06 (250 °C, 8 ppm) 1.12/-0.03 (300 °C, 8 ppm) -/-0.0114 -/-0.0246 -/-0.0941 1.15/-

1017

ref

162 162 162 161

Potje-Kamloth

a

operating temp (°C)

Semiconductor Junction Gas Sensors

varies exponentially with the lowering of the hydrogeninduced Schottky barrier height.141 Therefore, these Schottky diodes often exhibit excellent hydrogen detection sensitivities. The formation of palladium silicide at the Pd/Si Schottky interface leads to the degradation of detection capacity.142 An ultrathin (∼30 Å) oxide layer introduced into the Pd/Si interface prevents the formation of palladium silicide,142,143 forming a conducting MIS (metal-insulator-semiconductor) diode. However, the sensitivity toward hydrogen is significantly decreased, if the oxide layer is not properly defined. Therefore, the high quality of the oxide layer is a key factor for the device performance. This certainly increases the fabrication complexity and the cost. In the MIS configuration, the Schottky barrier height of the Pd/Si system is found to depend quite strongly on the Pd effective work function and hence on the hydrogen ambient. Zhang at al. investigated Pt-Pd/Si-Al Schottky diode structures for detection of NO2 from 6 to 22 ppm gas concentrations and of NO from 50 to 250 ppm gas concentrations, respectively. The sensor operates at room temperature. The response is roughly proportional to the logarithm of the gas concentration.140 Mesoscopic polysilicon wires coated with a thin film of palladium have been used as hydrogen sensors. The wires show an increase in their electrical resistance, when hydrogen reacts with the sensitive layer.144 Porous silicon obtained from p+-type silicon wafers were impregnated with Pd nanoparticles using an electroless process. It has been shown that the distribution of nanoparticles over the porous media creates a higher gas sensitivity toward hydrogen.145

4.1.2. GaAs Based Schottky Diodes In a metal/semiconductor junction, the height of the Schottky barrier depends on the work function difference between the metal and the semiconductor according to eq 4. However, as for many III-V semiconductors, the observed value of the barrier height of a metal/GaAs contact does not follow the simple theory in terms of the metal work function and the electron affinity of GaAs due to the high density of surface states, which causes a pinning of the Fermi level. Many models have been developed to explain this discrepancy. Among them, the effective work function model by Freeouf et al.146 emphasizes the role of the intermediate layer formed by the reaction between the metal and GaAs to determine the barrier height. However, some authors have found evidence of unpinning of the Fermi level in metal/ GaAs interfaces fabricated on clean (100) GaAs surfaces. This offered a way to fabricate metal/GaAs Schottky diodes with high sensing capability.147,148 Lechuga et al. reported a Pt/GaAs Schottky diode reaching a low detection limit for hydrogen of 6 ppm in nitrogen and 200 ppm in air.7,8 Liu et al. proposed the fabrication of a planar-type Pd/GaAs Schottky diode as a hydrogen sensor exhibiting an interfacial oxide, which weakens the Fermi level pinning and, hence, improves the barrier height.149 The use of a porous GaAs semiconductor has been investigated to improve the performance of Schottky diode hydrogen sensors, both in terms of sensitivity and response. Pd/porousGa/As and Au/porous-GaAs Schottky junctions have been fabricated, which show sensitivity toward hydrogen and polar gases, e.g., CO, NO, respectively.150,151 Pd/porous-GaAs Schottky diode gas sensors exhibited large sensitivity of more

Chemical Reviews, 2008, Vol. 108, No. 2 385

than three times higher than that of Pd/GaAs contacts with a response time of 1 s toward 500 ppm hydrogen gas. It has been found that this promoting effect could be related to the formation of pore structures (effective adsorption centers) in the Pd contact induced by the porosity of the GaAs wafer. A urea sensor has been presented using an ammonia gas sensor based on a Pt/GaAs Schottky diode and a ureaseimmobilized membrane. The authors used discontinuous platinum films which proved to be effective and reliable in solid-state chemical sensors for the detection of NH3 gas.8 The Schottky diodes have been coated with an organic layer in order to protect the devices during measurements in solution, as well as to use it as a support for biological material.

4.1.3. InP Based Schottky Diodes Pd/InP Schottky diodes have been regarded as promising devices for hydrogen sensing.149,152 At room temperature, it has been found that the Pd/InP Schottky diode exhibits superior sensing performances for the detection of hydrogen. Yousuf et al. demonstrated essentially high hydrogen response in the J-V characteristics of a Pd/InP Schottky diode.10 Although enormous current variations are observed, the low barrier height associated with the high defect state density at the Pd/InP interface severely restricts the allowable variation in the barrier height. Recently, electroplating techniques have also been proposed to fabricate the Pd/InP Schottky diodes.153 Although uniformity and adhesion related problems must still be solved compared with traditional vacuum deposition techniques, electrochemical techniques do exhibit the advantages of easy operation, simple equipment, and low cost.154 Particularly, because of the specificity the low-energy process, the electrochemical techniques can eliminate the Fermi-level pinning effect and lead to well behaved Schottky contact properties, which can remarkably improve the sensing performances of Pd/InP Schottky diodes. Talazac et al. have presented a different type of Schottky based gas sensor.155 The detection mechanism is based on the electron exchange of the gaseous molecules with a thin semiconducting layer between the Schottky contact and the semiconducting substrate forming a pseudo-Schottky barrier diode. High sensitivities in the ppb range toward NO2 and O3 could be found. However, the sensors suffer from longterm aging effects.

4.1.4. GaN and AlGaN Based Schottky Diodes GaN and AlGaN are attractive as materials for the fabrication of various chemical and biochemical sensors:156 the bulk and surface properties are chemically stable due to large band gap energies, the materials allow sensing operation at high temperatures, and the materials are environment friendly. The Schottky barrier heights are much more dependent on the metal work function157 than those for other III-V materials, indicating a weaker Fermi level pinning. However, Schottky diodes formed on GaN and AlGaN materials exhibit excess reverse leakage currents, which are many orders of magnitude larger than the prediction of the standard thermionic emission model (see section 3.2.2). Another issue related to the metal/semiconductor interface is a poor ohmic contact. Large leakage currents in Schottky diodes are particularly problematic for constructing high-performance sensors.158 As a model which explains large leakage currents in nitride-based Schottky barriers, Hasegawa et al. recently

386 Chemical Reviews, 2008, Vol. 108, No. 2

proposed a thin surface barrier model159 involving donors near the surface. Due to the presence of ionized high-density donors, the width of the Schottky barrier is greatly reduced, and the electrons tunnel through this barrier by the thermionic field emission or the field emission mechanism. On free surfaces of GaN and AlGaN, high-density surface states exist, which cause charge-discharge transients leading to performance instability, such as current collapse, and poor long-term reliability. Recently, experimental and theoretical investigations on the role of oxygen impurities in GaN and AlGaN have indicated that oxygen acts as a shallow donor close to the surface with an activation energy of about 30 meV.160 Therefore, the elimination of oxygen from AlGaN and GaN layers is highly desirable prior to the sensor fabrication in order to realize high-performance sensors, e.g., by using an ultrathin Al layer as a getter material for oxygen.157 Schottky diodes on AlGaN/GaN heterostructures with Pt, IrPt, and PdAg catalytic metals have been fabricated and characterized from 200 to 800 °C for hydrogen sensing operation.161 The hydrogen sensitivity of Pt and IrPt diodes improves with the increase in temperature due to a more effective hydrogen dissociation. The sensitivity of PdAg diodes degrades with increase in temperature due to thermal instability of PdAg.162 Schottky gas sensors for CO were fabricated using a PtAlGaN/GaN Schottky diode. The CO sensors show good sensitivity in the temperature range 250-300 °C.163

4.1.5. SiC Based Schottky Diodes The operating temperature of gas sensors using silicon substrates is limited below 250 °C due to the small band gap of silicon. This restricts their use in specific environments such as automotive, aeronautical, and environmental areas. Semiconductor substrates with a large energy band gap such as silicon carbide and diamond can be used for sensors operating at high temperatures. The use of catalytic metal gate SiC devices as gas sensors for hydrogen and hydrocarbons has been extensively investigated by several groups. Janson et al. reviewed the effect of hydrogen on the wide band gap semiconductor SiC and its influence on catalytic metal/SiC Schottky diodes.164 Kim et al. investigated hydrogen and methane gas sensors using Pt-SiC and Pd/SiC Schottky diodes operating at temperatures in the range 300-500 °C.11,165 Annealing effects on Pd/6H-SiC Schottky diodes have also been investigated.166 The studies indicated that the response of the diode to hydrogen degraded after annealing at 425 °C due to interfacial diffusion of silicon into the Pd film region after dissociation from SiC. Palladium silicides are formed as the major interfacial products. The use of a Pd0.9Cr0.1 alloy film deposited on a 6H-SiC epilayer results in Schottky diodes exhibiting a stable catalytic surface and a significant improvement of the device sensitivity.167 Some reports deal with the response of catalytic metal gate SiC devices to nitric oxides (NOx). Zubkans et al.168 reported the direct response to NOx, which could be increased after a treatment with ammonia. Katsube et al. showed that thin catalytic metal gate Schottky diodes and p/n heterojunctions on Si substrates can be used for direct NOx detection at room temperature.169 They extended their investigations using SiC substrates for high-temperature operation170 for direct detection of NOx at elevated temperatures up to 400 °C at

Potje-Kamloth

minimum concentrations down to 10 ppm. The Langmuir adsorption model can describe the response of these devices.

4.1.6. CdSxSe1.x Based Schottky Diodes It has been shown that the bulk photoluminescence of semiconductors can be used to probe the Schottky barrier characteristics of both semiconductor/metal diodes171 and photoelectrochemical cells.172 In particular, the electric field in the semiconductor can be estimated from the PL intensity using a dead-layer model: electron-hole (e--h+) pairs, formed within a distance from the interface in the order of the depletion width, do not contribute to the bulk photoluminescence. Schottky diodes have been fabricated with Pd because of the known sensitivity of the current-voltage characteristic to hydrogen. Carpenter et al.173 showed that, besides the decrease of the Pd work function on exposure to hydrogen, the variation in Schottky barrier height with hydrogen strongly influences the bulk photoluminescence intensity of the Pd/CdS Schottky diode. The bulk photoluminescence intensity is enhanced upon exposure to hydrogen. In air the bulk photoluminescence intensity of the Schottky diode returns to its original value. The spectral changes are consistent with a reduction in Schottky barrier height resulting from the dissolution of hydrogen in Pd. Qualitatively, the bulk photoluminescence is expected to rise because the smaller electric field in the semiconductor allows a larger fraction of e--h+ pairs to radiatively recombine. By regarding the region supporting the electric field as being completely nonemissive, a quantitative expression for the relative intensity of the bulk photoluminescence can be obtained:

φair/φH2 ) exp(-R∆D)

(42)

where φair and φH2 are the radiative quantum yields in air and in hydrogen, respectively, ∆D is the difference in deadlayer thickness between the two media, and R is the sum of the solid’s absorptivities for the exciting and emitted light, respectively. ∆D can be used to calculate the reduction of the Schottky barrier height. By substitution of the depletion width W by ∆D, eq 36 gives the change in barrier height upon exposure to hydrogen.

4.2. Schottky Diodes Based on Organic Semiconductors Electrical conductivities of organic semiconductors can be varied over the full range from insulator to semiconductor through p-doping or n-doping (see section 2.2). They offer a viable alternative to conventional inorganic semiconductors in many applications because of their unusual electrical properties, diversity, ease of fabrication, large area, and potentially low cost. Much research has been carried out on the use of organic semiconductors as active materials in electronic devices. Fabrication of electronic devices, such as OFETs, OLEDs, and devices based on Schottky barriers is the most important application of these materials. The use of different semiconducting organic films for the detection of gas and vapors has been reported by a number of groups.27-30,174 Whereas the chemical stability in a given temperature range and in ambient environment is a basic requirement of organic semiconductors to be used as active material in electronic devices, the modulation of physical properties of the organic semiconductors as a function of gas or vapor concentration is a prerequisite to be used as

Semiconductor Junction Gas Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 387

Table 4. Parameters of Molecular Semiconductor Schottky Barrier Diodes Based on Phthalocyanines (Pc) doping

barrier height φb /eV

charge carrier density Ns /cm-3

oxygen as deposited oxygen

1.03 1.09 0.65

oxygen as deposited oxygen

1.2

1.7 × 1019 7.5 × 1018 2.2 × 1017 5.1 × 1018 1.6 × 1018 3.4 × 1017 1.1 × 1016

device config Au/AlPcCl/Al ITO/AlPcCl/Al Au/FePcCl/Au Au/TiOPc/HOPG Au/NiPc/Al

p-type p-type p-type p-type p-type n-type

depletion width W/ nm

169

charge carrier mobility µ /m2 V-1 s-1

ref

1.73 × 10-8 1.58 × 10-8 1.2 × 10-6

194

5.8 × 10-6

176

196 193 105

Table 5. Electrical Parameters of Polymer Schottky Barrier Diodes Based on Polythiophene (PT), Poly(3-alkylthiophenes) (P3AT)a, Polyanilines (PANI), and Polypyrroles (PPy)a structure Al/PT/Au Al/P3DT/Au-Sn Al/P3HT/Au Al/P3HT/ITO Al/P3MeT/Pt Al/P3MeT/Au Al/P3OT/ITO Al/P3OT/Au

doping agent ClO4 undoped undoped PF6 reduced BF4 reduced PF6 undoped FeCl3 undoped

ideality factor n 13.0

contact or built-in potential φbi/eV PolythiophenessPT 0.38

barrier height φb/eV 0.66 0.4

3.9 3.0 4.5 2.5

0.35 1.2 0.6

6.1 4.9 1.2

0.52 0.7

0.34 0.75 1.5 0.8 0.45 1.01 0.72

charge carrier density NA/cm-3 3.0 × 1014 1.0 × 1017 3.5 × 1015 6.4 × 1019 4.6 × 1016 2.9 × 1017 7.0 × 1016

ref 211 217 91 205 216 239 239 205 205 203 204

PolyanilinessPani Al/Pani/Au

Al/Panib/Au Al/Pani/ITO In/Pani/Pt Al/PPy/Au Al/PPy-NMPy/Au In/PPy-NMPy/Au Au/PPy/Pt

HCl TOS PAA undoped HCl HCl DBS/HCl

1.24 1.3 1.7 1.95 1.46-1.64 4.2 3.12

BF4 BF4 ClO4 CuPcTS PbPcTS NiPcTS CoPcTS

4-7 1.4-2 1.2 3.6 3.8 4.0 4.2

0.72 0.82 0.9 0.65 0.69 PolypyrrolessPPy ∼0.7 0.78-0.82 0.23 0.26 0.24 0.23 0.2

0.38 0.78 0.87 1.03 0.33-0.41 0.76 0.803 0.7-0.76 0.7 0.81 0.63

2.8 × 1018 1.4 × 1018 1.8 × 1017 3.1 × 1017 3.06 × 1017 1.0 × 1022 1016-1017 7.0 × 1017 1.2 × 1020 4.6 × 1019 3.7 × 1019 1.9 × 1019

222 220 222 228 221 185 185 182 186

a ClO ) perchlorate; CuPcTS ) copper phthalocyanine tetrasulfonate; PbPcTS ) lead phthalocyanine tetrasulfonate; NiPcTS ) nickel 4 phthalocyanine tetrasulfonate; CoPcTS ) cobalt phthalocyanine tetrasulfonate; DBS ) dodecylbenzenesulfonic acid; D ) dodecyl; O ) octyl; M ) methyl; H ) hexyl; NMPy ) N-methylpyrrole. b Modified with metal-2.5-diaminobenzenesulfonic acid.

the sensing element in chemical sensors. However, the details of the gas response mechanism have not been completely elucidated. From a practical point of view, the gas sensing characteristic is strongly influenced by a number of processing parameters: for example, the growth rate, the thermal annealing,29 the polymerization temperature,175 and the nature of the dopants.132 By exposure of the polymers to certain gaseous species exhibiting electron donor or acceptor behavior,176,177 the Fermi level position is changed by either increasing or decreasing the doping level according to section 3.4.1. In this case, a gas-induced change of the Schottky diode barrier characteristic can be observed. Furthermore, the increase or decrease in the ideality factor during gas/ polymer interactions, which is due to a generation or reduction of the interface state density, can influence the gas sensitivity of the junction as well. Despite these advances in the use of chemical sensing applications, the detailed properties and roles of interfaces in organic semiconductor devices remain elusive. Parker suggests that charge injection into organic materials proceeds via tunneling across interfacial barriers.178 Improvement of device performance through interface modification has also

been reported.179 Up to now, most Schottky diodes based on organic semiconductors were fabricated and characterized using metals with low work function, such as Al, In, or Ca.27,180-182,217 These diodes show unstable electrical and chemical sensing properties, which have been attributed to the high chemical reactivity of the metal used.183 The nature of the interfacial layer between these metals and the semiconducting polymers is commonly difficult to define.181 In Tables 4-7, diode parameters of organic Schottky diodes are given. The data show that, unlike the case of conventional metal/semiconductor contacts, the electrical properties of contacts based on organic semiconductors can be manipulated not only through the choice of materials, methods, and conditions of preparation but also through the type of dopants incorporated into the organic semiconductor (see section 2.2). This can be exploited to tune the diode characteristics in a definite way. On the other hand, a careful control of the preparation parameters is necessary to obtain reproducible results. The choice of the metal forming the ohmic contact depends on the work function of the organic material used. For the commonly used polymer-based diodes, consisting

388 Chemical Reviews, 2008, Vol. 108, No. 2

Potje-Kamloth

Table 6. Electrical Parameters of PPy-MPcTS Schottky Barrier Diodes Extracted from J-V and C-V Characteristics, Measured in Air and Exposed to 11 pm NO2 saturation current density J0/A cm-2

ideality factor n

rectification ratio (at |V| ) 1 V)

built-in voltage Vbi/V

dopant

air

NOx

air

NOx

air

NOx

air

NOx

CuPcTS PbPcTS NiPcTS CoPcTS TOS

8.1 × 1.5 × 10-3 6.5 × 10-3 1.0 × 10-2 2.5 × 10-2

7.0 × 7.9 × 10-4 2.4 × 10-3 4.8 × 10-3 1.1 × 10-2

3.6 3.8 4.0 4.2 4.9

3.2 3.5 3.7 3.8 4.3

7.2 5.2 3.7 3.2 2.1

13.5 8.7 5.1 4.7 2.9

0.26 0.24 0.23 0.20 0.15

0.29 0.27 0.26 0.22 0.17

10-4

10-5

Table 7. Electrical Parameters of PPy-MPcTS Schottky Barrier Diodes Extracted from J-V and C-V Characteristics: Change of NO2 Sensitivity for Different Polymerization Temperatures of PPy, Measured in Air and in 11 pm NO2 polymerization temp/K

rectification ratio at |V| ) 0.9 V

saturation current density J0/A cm-2

ideality factor n

charge carrier conc NA/1019 cm-3

278 283 288 293 303 313 323 333

2.1 3 5 6.3 27.3 156 307 68

3.1 × 10-2 6.3 × 10-3 1.9 × 10-3 7.9 × 10-4 1.2 × 10-4 7.6 × 10-5 5 × 10-5 1 × 10-4

4 3.7 3.4 3.6 3.3 3.1 2.5 2.9

97.5 52.6 37.1 12 2.1 0.82 0.37 1.4

of poly(3-octylthiophene),27 poly(3-methylthiophene),184 and PPy,29,185,186 gold was used to form an ohmic contact. The work function of gold has been assumed to be about 5.1 eV. Several studies show that the work function of π-conjugated polymer films is about 5 eV, depending upon the dopant type and the doping level.64,126 In this case, the barrier between the p-type polymer and gold is relatively small but is predicted to be negative, representing a quasi-ohmic contact, with a symmetric current-voltage dependence. Nevertheless, the experimental contacts are sometimes rectifying,133 indicating a higher work function of the polymer than that of gold. Hence, the junction characteristics of the ohmic gold/polymer barrier are strongly influenced by the gas sensing behavior of the polymer layer, exhibiting either symmetric or a non-ohmic, i.e., asymmetric, current-voltage characteristics. To overcome this uncertainty, platinum is recommended as metal to form an ohmic contact with the doped p-type conducting polymers.

4.2.1. Phthalocyanines Phthalocyanines (Pc) are a class of organic compounds of high thermal and chemical stability, which are classified as p-type semiconductors characterized by a low mobility and low charge carrier concentration.187 They have potential advantages for use as active layers in electroluminescent,188 photovoltaic,189 and gas sensor devices,190 because they are easily processable in low cost and large area device fabrication.191 However, it is known that oxygen profoundly influences the electrical properties of phthalocyanines. It has been demonstrated that the dark rectification ratio of Au/Pc/M (M ) Au, Cu, Cr, Al) is related to the presence of O2.192,193 When the devices are made and studied under vacuum, no rectifying effect is observed. However, by exposure to air, a strong rectifying effect has been found. Acceptor levels are generated within the band gap of these materials in the presence of O2, and hence, their thermal activation energy is lowered and the conductivity is enhanced.192,194

Besides O2, any other electron acceptors, such as NOx and Cl2, induce acceptor levels, while electron donors, such as H2 or NH3, remove them. Both change the Schottky diode characteristics. With most of the phthalocyanines, gold is found to form the ohmic contact,14,15 and aluminum is found to form the blocking contact.192,195 Under forward bias, oxygen-doped phthalocyanine thin films show ohmic conduction at low voltages and space charge limited conduction controlled by an exponential distribution of traps above the valence band edge at higher voltages.192,194,196 As already discussed above, almost all phthalocyanines show excellent gas-sensing properties for both electron acceptors, such as NOx,197-199 HCl,197 Cl2200 and electron donors, such as H2 or NH3.105 The change in the doping level induces a Fermi level shift and a change in the Schottky barrier height. It is supposed that a charge-transfer complex128 between the organic semiconductor and the gas or vapor is formed during the interaction. Spectroscopic investigations show the accumulation of the gas species in the bulk of the sensitive film and the formation of organic radical cations.197,201 Studies of phthalocyanines indicated that the sensitivity and selectivity of the sensor device are strongly influenced by the nature of the central metal ion and also by the crystal structure and/or crystallinity. Nieuwenhiuzen et al.202 reported that the sensitivities for detection of NOx decrease in the order Co > Cu > Pb > H > Fe > Mg > Ni, whereas the sensitivities toward NH3 decrease in the order Pb . Fe > Cu > Ni > Mg. Pc films either prepared by spin-coating or by vacuum deposition show a detection range of 0.3-200 ppm for NO2197,202 and one of 1-200 ppm for NH3202 without further annealing after film formation. Independent of the central metal ion, phthalocyanines show no sensitivity toward CO, CO2, CH4, C6H14, C7H8, SO2, H2O, and H2S.202 High stability and reproducibility are observed in the sensing behavior of all phthalocyanines, with the exception of PbPc and FePc. Both show a deformation during exposure to NH3 and NO2, respectively. Additionally, the crystal structure and size have been shown to vary with operating time.198,202 The presence of disordered phases between the PbPc particles reduces the response and recovery times during the gas interaction. It was concluded by Sadaoka et al. that the gassensing properties of PbPc can be improved by annealing the Pc films at 300 °C in order to form a homogeneous layer of triclinic crystals with a mean diameter less than 0.2 µm.198

4.2.2. Polythiophene/Poly(3-alkylthiophenes) For more than one decade, polythiophene and its derivates have been the subject of intensive experimental and theoretical work.203-205 Polythiophene and its 3-methyl derivative exhibit a poor processability due to their insolubility. The available materials have not been of the high electronic

Semiconductor Junction Gas Sensors

quality necessary to use them in electronic devices. In 1986, poly(3-alkylthiophenes) with an alkyl group of more than four carbon atoms became available.206 They are soluble in several organic solvents and even fusible, which makes them attractive for use in electronic devices ranging from fieldeffect transistors207 and light emitting diodes208 to nonlinear optoelectronic components.209 Sundberg et al. investigated the manufacture of metal/ polymer contacts.210 The comparison of the current-voltage and capacitance-voltage characteristics of melt-processed and sputtering-processed devices shows no significant difference between these structures. Both processes influence the polymer surface. The sputtering involves some relatively high-energy particles impinging on the polymer surface, whereas melt-processing can give rise to surface modifications and, hence, to a stronger enhancement of the doping concentration close to the interface than obtained for sputtering. For both devices, C-V characteristics indicate an additional constant capacitance in series with the variable space-charge capacitance of the polymer, which is assigned to an insulating oxide layer at the metal/polymer junction. In the melt-processed device, this layer is probably thicker because the aluminum is exposed to air before making the contact. Similar results were obtained when the rectifying contact was formed by pressing aluminum on the polymer surface.211 In this context, it should be taken into account that reactive metals such as Al and In might form a thin insulating interfacial oxide layer and probably react with the dopants.212 This would lead to a lower accumulation of the dopant species at the interface and a wider barrier. Lous et al. investigated Schottky contacts on highly doped organic semiconductors based on thiophene oligomer by current-voltage and capacitance-voltage measurements.213 The Schottky diodes obtain their rectifying behavior from a partly undoped p- layer at the metal/oligomer interface with respect to the higher doped p+ bulk layer. This low conductive p- layer implies a large electric field at the interface which can cause a substantial lowering of the Schottky barrier. Schottky diodes with poly(3-octylthiophene) as the semiconductor and aluminum- and indium-doped tin oxide as the rectifying and ohmic contact, respectively, show a rectification ratio up to 3-4 orders of magnitude. The shape of the J-V characteristics, the high value of the ideality factor, and the observation of series resistance indicate an interfacial layer at the aluminum/polymer interface and the effects of trapping in a region adjacent to the metal.214 The effect of oxygen is reversible on structures based on poly(3-methylthiophene).215,216 The effect of various atmospheres on the electrical characteristics of Mo/poly(3methylthiophene) Schottky barrier diodes was investigated by Tagmouti et al.216 The electrical characteristics were mainly affected by the presence of water vapor and oxygen. The capacitance shows a frequency dispersion, which is characteristic for the presence of deep traps induced by oxygen forming acceptor states in the band gap. The authors concluded that oxygen fills the traps and dopes the polymer. The effect of water vapor is manifest by reduction in dopant density and a filling of deep traps, which is irreversible. Gas-sensitive characteristics toward water vapor, chloroform, and ethanol were observed in poly(3-dodecylthiophene) Schottky diodes.217 The two types of gases have opposite effects on the diode characteristics; one type (water vapor) increases and the other type (chloroform and ethanol)

Chemical Reviews, 2008, Vol. 108, No. 2 389

Figure 13. Chemical structure of polyaniline, showing three different oxidation states.

decreases the reverse bias current. The results were explained by the acceptor and donor behavior, respectively, which results in a change in the Schottky barrier height by doping or undoping of impurity levels in the band gap of the p-type semiconductor.

4.2.3. Polyanilines The advantages of polyaniline compared with other conducting polymers are the easy and cheap oxidative synthesis, the thermal and environmental stability, and the simple nonredox doping in protic acids.218 Polyaniline is a p-type organic semiconductor, and it can exist in several oxidation states, ranging from fully reduced leucoemeraldine over intermediate and moderately oxidized (doped) emeraldine to the fully oxidized pernigraniline, as named by Green and Woodhead219 (Figure 13). Acid-doping of the soluble emeraldine base, using organic and inorganic acids, leads to the insoluble highly conducting emeraldine salt. Schottky diodes have been prepared using either casting220,221 or electrochemical techniques.222,223,228 The polymers investigated involve undoped and acid-doped polyanilines, which behave as p-type semiconductors. The diodes, in which aluminum was used as low work function metal to form a Schottky junction, exhibit a moderate rectifying behavior and low leakage current. The values of ideality factor n lie between 1 and 2, indicating either a recombination process in the barrier or a thin insulating layer at the interface. The acid-doping of polyaniline, e.g., with toluene-4-sulfonic acid, can cause a higher rectifying effect and photovoltaic conversion efficiency. Calderone et al. investigated the metal/polymer interface theoretically, focusing on the interaction between aluminum and the fully reduced form of polyaniline, i.e., leucoemeraldine.224 The study obtained for the model system Al/pphenylenediamine indicated that an electron charge is transferred from the metal to the organic system, and the π-electronic levels are strongly perturbed. These results are considered as the basis of the Al/leucoemeraldine interaction. Chemical sensitivities toward gases and vapors, such as ammonia,225 hydrogen cyanide,226 hydrogen,227 and methane,228 have been shown. Campos used a sandwich type device based on Al/polyaniline as a sensor for the detection of methane gas. HCl-doped polyaniline was electrochemi-

390 Chemical Reviews, 2008, Vol. 108, No. 2

cally deposited on an indium tin oxide coated glass electrode, which serves as the ohmic contact metal in the Schottky diode. Methane decreased the forward and reverse bias current, indicating an increase in the barrier height.

4.2.4. Polypyrrole 4.2.4.1. General Aspects. Among various conducting polymers, polypyrrole (PPy) has been extensively investigated because of properties that are attractive from the practical point of view: that is, relatively good environmental stability, high conductivity, and ease of preparation either by chemical or electrochemical methods, as well as the possibility of forming homopolymers or composites with optimal properties. A general difficulty of the reproducible PPy preparation arises from its complexity. The structure and hence the properties of the resulting PPy are strongly influenced by a number of variables, e.g., the polymerization potential, the monomer concentration, and the preparation temperature,229 which are not perfectly controllable. Therefore, the results on PPy vary widely. Two basic methods are used for the preparation of PPy: chemical and electrochemical synthesis. Anions, present in the polymerization solution, are simultaneously incorporated into the polymer as dopants. The final polymer chain bears a charge of unity for every three or four pyrrole rings, which is counterbalanced by the dopant anion.229 A lot of work has been carried out in order to improve both the chemical and electrochemical synthesis of pyrrole, as reviewed by Rodrı´guez et al.229 There exists a huge number of publications dealing with parameters influencing the final properties of PPy samples during the synthesis. Only some important parameters are briefly discussed below. The counterion is indispensable for the charge compensation of PPy; at the same time the size, shape and structure of the counterion are expected to influence the electrical conductivity. There are a large number of reports comparing the conductivity values of PPy containing aromatic sulfonate anion, such as p-toluene sulfonates, aromatic carboxylates, etc., and inorganic anions, such as BF4-, Cl-, ClO4-, SO42-, etc., and it was found that conductivity was always higher in the case of aromatic dopants.230 Few studies have dealt with PPy containing anionic transition-metal complexes, e.g., tetrasulfonated metallophthalocyanines (MPcTS).231,232 Saunders et al.231 have studied the physical and spectroscopic properties of PPy-MPcTS (M ) Co, Cu, Fe, and Ni) prepared using both aqueous and nonaqueous solution. The conductivities of these materials were found to rapidly decrease upon exposure to air. As a reason for this, it was suggested67 that the large size of the MPcTS counteranions produced packing inconsistencies within PPy-MPcTS, which facilitates the chemical attack of the polymer backbone by atmospheric oxygen. Walton et al.232 have studied a competitive doping of PPy during polymerization in an aqueous and acetonitrile solution of ClO4- and copper phthalocyanine tetrasulfonate. The effect of temperature on the polymerization process has been studied. It has been shown that the temperature has a significant effect on the conductivity of PPy depending on the combination of solvent and dopant. Toluensulfonateand perchlorate-doped PPy films, deposited at a lower temperature (280 K) in water, exhibit higher conductivities than films prepared at 293 or 313 K.229 The authors found that an increase in the temperature favors the polymer growth, which has been related to an overpotential decrease.

Potje-Kamloth

4.2.4.2. Metal/Polymer Interface. The polypyrrole family of polymers is a promising candidate for device fabrication because of its relatively high stability in ambient air. Oxidized (p-type) PPy is reported to have a work function close to 5 eV.122 From theoretical considerations, therefore, the junctions between PPy and metals with low work functions are expected to form a Schottky barrier. Several reports are available on metal/PPy junctions.182,233,235 Almost all of the results were explained on the basis of the Richardson equation, eq 6. The high charge carrier concentration found in doped PPy (NA ∼ 1022 cm-3 assuming that all dopant charges give rise to one hole polaron or half a bipolaron) results in the formation of a very thin Schottky barrier with a narrow depletion width (∼ 10 Å), which may be the reason for the bad rectifying behavior of the polymer Schottky junction due to the dominant tunneling process of the charge carrier transport mechanism (see section 3.2.4). It has been suggested that the introduction of N-substituted pyrroles to PPy lowers the conductivity of the homopolymer.234 The relationship between the density of dopants and the conductivity has led to investigations of the rectifying characteristics of various copolymers having pyrrole and N-methylpyrrole units. The introduction of N-methylpyrrole would less influence the delocalization of π-electrons along the polymer chain than that of other N-substituted pyrroles. Schottky junction devices from a pyrrole/N-methylpyrrole copolymer have been fabricated exhibiting a relatively low charge carrier concentration of NA ∼ 1017 cm-3 and a conductivity of 10-210-4 S cm-1.182,185 The values of the ideality factor (n ) 1.20-1.38) are closer to that of an ideal diode, and the rectification ratio is relatively high. In or Al have mostly been used as low work function metals to from a non-ohmic Schottky barrier with PPy.137,181-183,233 Gupta et al. reported leaky Schottky barriers of reactive metals, such as Al and In, with the polymer.233 Some authors concluded that the formation of metal salts at the metal/polymer junction as a result of reaction between the dopant anion and the low work function metal leads to insulating interphases.137 Other reports show that the ambient (i.e., oxygen) influences the current-voltage characteristics of contacts between these metals and PPy and that the observed hindrance of charge transport was due to the formation of insulating metal oxides.181,183,235,236 The thin insulating layer at the interface between the metal and the polymer was assumed to be the reason for the anomalous J-V characteristics with a small plateau at low forward bias.217,237,238 The forward characteristics of this type are comparable with nonequilibrium effects in MIS diodes.237 Taylor et al. noted that this effect could be further produced when the polymer Schottky diode comprising an aluminum rectifying contact was given a thermal post-metal annealing in air.239 On the other hand, it has been shown by Nguyen et al. that a metal with a high work function (e.g., gold) can be used to form a non-ohmic contact with PPy when metal phthalocyanine tetrasulfonates (MPcTS) are used as dopants.132 These contacts show an exponential and asymmetrical behavior of the J-V curves (Figure 14). As can be seen in Table 4 and Figure 14, the junction parameters of Au/PPy-MPcTs diodes are strongly influenced by the type of dopant. The junction between PPy-TOS and Au shows a low rectifying ratio and a relatively high saturation current density. This junction seems to be practi-

Semiconductor Junction Gas Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 391

The capacitance-bias voltage curve shown in Figure 17 indicates the formation of a Schottky junction135 and allows the use of the simple depletion layer theory as already discussed in section 3.5.3. Nguyen et al.132 determined a value of 0.26 V for the built-in voltage Vbi and a charge carrier concentration NA (concentration of polarons and/or bipolarons) of 1.2 × 1020 cm-3 for the CuPcTS-doped PPy layer of 120 nm in thickness. This is in good agreement with NA ) 1.5 × 1020 cm-3, reported by Gupta233 for electrochemically prepared PPy of 3 µm in thickness, doped with tetrafluoroborate anion. Figure 14. Dark current density vs applied voltage (J-V) characteristics of Au/PPy Schottky barrier diodes doped with different dopants.

Figure 17. Mott-Schottky plot (C-2 vs bias voltage) for a Au/ PPy-CuPcTS Schottky barrier diode. Figure 15. Dark current density vs applied voltage (J-V) characteristics of a Au/PPy-CuPcTS Schottky barrier diode.

cally quasi-ohmic. The most optimal diode parameters, such as the rectifying ratio and the ideality factor, were found for the Au/PPy-CuPcTS junction (Figure 15). The rectification ratios (forward bias current/reverse bias current at 0.6 V) were observed to be between 7.7 and 3.2. The values are slightly larger than that of the Schottky diode based on trans-polyacetylene and Pb (n ) 1.5) reported by Kanicki,133 and are in good agreement with that for Ti/PPy (n ) 10) reported by Gupta et al.240 and for Al/PPy Schottky diodes (n ) 6) reported by Bantikassegn et al.137 Impedance and capacitance measurements give insight into the nature of the diode junctions and the physical properties of PPy as an organic semiconductor.132 Figure 16 shows the plane plot of the complex impedance of a Au/PPyCuPcTS Schottky barrier diode, measured at zero bias. The device shows only a single semicircular arc. The experimental curve can be fitted well by applying the equivalent circuit of the diode shown in Figure 16. This indicates the formation of a negligible small insulating interfacial layer (see Figure 7A).

The charge carrier concentration can also be estimated from the doping level, determined by electrochemical quartz microbalance (EQCM) measurements.241 A doping level of 30.2% is deduced, and a charge carrier concentration of about 3.3 × 1021 cm-3 is expected. This value is in accordance with that reported by Buhks et al.242 (about 3 × 1021 cm-3). It can be seen that the acceptor concentration calculated from capacitance measurements is far different from that estimated from the doping level obtained from the EQCM measurements. This may be due to the large concentration of defects present in PPy that strongly influence the conjugation lengths of the polymer chain231 (see section 4.2.6.1). 4.2.4.3. Chemical Sensing Properties. Influence of Dopants. Several authors,64,121,243 have reported that the sensitivity of PPy to certain vapors or gases changes with the doping anion. Sensitivities toward organic vapors, such as alcohols, chlorinated hydrocarbons,121,244 aromatic hydrocarbons,245 nitrogen oxide gases, and organophosphorous compounds,121 have been reported. The influence of different metal phthalocyanines incorporated in PPy on the NOx sensing properties of the Au/PPy diodes has been studied.121 NOx exhibits a strong electron acceptor effect on PPy when doped with metal phthalocyanines.27,29,64 Nguyen at al.

Figure 16. Complex impedance plane plots of a Au/PPy-CuPcTS Schottky barrier diode at zero bias, and a complex plane model.

392 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 18. Mott-Schottky plot (C-2 vs bias voltage) for a Au/ PPy-CuPcTS Schottky barrier diode measured in air and in 11 ppm NOx.

Figure 19. Change in J-V characteristics of Au/PPy-CuPcTS measured in air and in 11 ppm NOx. ∆V indicates the NOx induced voltage shift at constant current density.

Figure 20. Response curve of a Au/PPy Schottky barrier sensor toward exposure to NOx by applying a constant current density (J ) 2 × 10-3 A cm-2).

showed that the exposure of a Schottky diode with PPy as the semiconducting organic layer to NOx causes an increase in the doping level of PPy. This leads to a positive shift of its Fermi level and, therefore, increases the built-in potential at the Au/PPy junction.132 This effect could be confirmed by capacitance voltage measurements (Figure 18).

Potje-Kamloth

The change in the J-V characteristics is shown in Figure 19 for the same measurement. The gas sensitivity of the diodes is represented by the so-called gas-induced voltage shift, ∆V, at a constant current density in gas and in air, respectively, which was extracted from current voltage characteristics as discussed in section 3.5.1. Figure 20 shows the transient response of the output voltage of a Au/PPy-CuPcTS Schottky barrier diode to exposure to NOx, measured at constant current density. As already discussed in section 3.5.1, an ideal Schottky barrier diode (i.e., without an appreciable interfacial layer) can be divided into two regions: (1) the barrier region (of thickness WJ), having a capacitance CJ in parallel with the junction resistance RJ, and (2) the neutral bulk (of thickness WB), with a bulk capacitance CB in parallel with the bulk resistance RB. The NOx exposure can lead to an increase in the built-in voltage Vbi and the depletion width WJ; hence, the resistance RJ of the barrier region can increase as well. In contrast, NOx exposure may cause a decrease in the bulk resistance RB of the PPy layer because of the strong acceptor behavior of NOx. These effects have been confirmed by impedance measurements.132 Figure 21 shows the plane plot of the complex impedance of the Au/PPy-CuPcTS Schottky barrier diode, measured in air and in NOx after different time intervals. It can be clearly seen that when the diode is exposed to NOx, the bulk resistance decreases and the junction resistance simultaneously increases. A combination of those compensatory effect can decrease the gas sensitivity of the Schottky barrier diodes. In the low bias region, the junction voltage increases slightly due to NOx-exposure, while in the high bias region it increases largely. This suggests a large change in the interface layer resistance upon NOx-exposure, which has been assumed to play an important role in the charge carrier transport across the Schottky barrier.82 The change of the diode parameters extracted from current voltage measurements of PPy-MPcTS Schottky diodes exposed to NOx is listed in Table 6. For comparison, the sensitivity of a PPy-TOS Schottky diode is included. The data listed in Table 6 show that the NOx sensitivity is strongly influenced by the nature of the dopants. A lower saturation current density is accompanied by a higher rectification ratio and by a lower value of the ideality factor, which are related to a larger value of the NOx-induced voltage shift measured in the low bias range (Vbias < 0.3 V). The ideality factor shows a tendency to decrease when the diode is exposed to NOx. It has been reported by Bardeen246 that the discrepancy between the real Schottky diode and the ideal may be due

Figure 21. Plane plots of complex impedance of a Au/PPy-CuPcTS Schottky barrier diode (1) in air, (2) and (3) in 11 ppm NOx.

Semiconductor Junction Gas Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 393

Figure 23. Doping level (O) and built-in voltage (0) as a function of PPy film thickness of Au/PPy-CuPcTS Schottky barrier diodes.

Figure 22. Dependence of the diode parameter and sensor response of Au/PPy-MPcTS Schottky barrier diodes on the nature of dopants.

to the effect of surface states and that the increase or decrease in the ideality factor may be due to a generation or reduction of the interface state density. Hence, the tendency of the ideality factor to decrease when the diode is exposed to NOx may be attributed to the compensation of the interface states by NOx molecules.132 In the low bias range, the diode current is primarily controlled by the depletion region and the gas response is usually explained in terms of the change of the work function (i.e., Fermi level) in the polymer layer. Capacitance-voltage (C-V) measurements show that PPy-CuPcTS exhibits the highest charge carrier density of all layers studied and hence the most positive work function. Hence, PPy-CuPcTS leads to the highest built-in voltage of all polymer diodes studied (Figure 22A). The interaction of PPy-CuPcTS with NO2 causes the highest change in work function, which is related to the highest change in the charge carrier concentration, as C-V measurements have shown (Figure 22B). Both effects are enhancing the sensor signal of the Schottky barrier diode when compared with work function measurements using Kelvin Probe or POSFET.247 The charge carrier concentration of the PPy layer decreases from CuPcTS to TOS as dopant. The junction between PPy-TOS and Au shows a very low rectifying ratio due to the low built-in voltage and is practically quasi-ohmic. In this case, the bulk resistance cannot be neglected any more even in the low bias range. Hence, the tendency of the NO2-induced junction voltage shift to decrease when the built-in voltage of the diode decreases can be explained in terms of the enhancement of the compensatory effect in the PPy bulk. This can be seen in Figure 22C along the series CuPcTS, PbPcTS, NiPcTS, CoPcTS, and TOS. This effect is consistent with results obtained by Cabala et al.121 They observed (1) an increase of the work function

of doped PPy during exposure to NOx, which was assumed to be due to an increase of the doping level, and (2) a decrease of the sensitivity toward NOx from PPy-CuPcTS to PPy-PbPcTS and PPy-TOS. According to the model given by Janata,128 describing the chemical modulation of the work function of a sensitive polymer layer upon exposure to an interacting gas or vapor (see section 3.4.1), the magnitude and the polarity of the response depend on the value of the charge-transfer coefficient δ, with |2δ| < 1 (eq 25). The charge-transfer coefficient δ can be estimated from quartz crystal microbalance (QCM) measurements during NOx exposure. From the obtained data, an absorbed NOx gas concentration of about 1.14 × 1020 cm-3 is deduced when the PPy layer is exposed to 11 ppm NOx. Assuming that most absorbed NOx molecules participate in the formation of a charge-transfer complex, a value of 0.36 can be deduced for the charge-transfer coefficient δ, taking into account the change of the charge carrier concentration in the PPy layer extracted from C-V measurements (NA ) 4.2 × 1019 cm- 3). Evaluating the change in built-in voltage with NOx gas concentration, a value of 0.41 is deduced for the charge-transfer coefficient δ. This value is slightly larger than that estimated from QCM measurements, and it indicates that the change in the Fermi level of PPy does not follow exactly the change in the charge carrier concentration according to eq 25 and eq 26, respectively. This fact may be attributed to the pinning of the Fermi level due to the presence of interface states, as discussed in section 3.3.3 Influence of Doping LeVelsFilm Thickness. It is wellknown that the doping level of MPcTS-doped PPy layers prepared by electropolymerization is strongly influenced by the PPy layer thickness. Rosenthal et al. have shown that the doping level of a MPcTS-doped PPy layer electrochemically formed at a constant potential decreases sharply when the PPy layer thickness increases.248 The parasitic oxidation of the MPcTS dopant gives rise to a depletion layer near the electrode surface and, hence, to a decrease in the concentration of the dopant anions incorporated into the polymer backbone. The influence of the film thickness on the doping level of PPy has been monitored in situ using EQCM measurements.241,249 J-V and C-V measurements of Au/PPyCuPcTS Schottky barrier diodes comprising PPy layers of different thicknesses showed that a decrease in the doping level causes a decrease in the Fermi level of PPy. As shown in Figure 23, the built-in voltage formed between the PPy layer and Au and, in turn, the depletion width of the junctions decrease according to eq 3.

394 Chemical Reviews, 2008, Vol. 108, No. 2

Potje-Kamloth

Figure 24. Change in junction voltage ∆VJ as a function of PPy film thickness of Au/PPy-CuPcTS Schottky barrier diodes.

The two compensatory effects of NOx on the gas sensitivity of the Schottky barrier diodes in the bulk of PPy and at the interface region can be studied more extensively with the variation of the polymer film thickness, which strongly influences the doping level of the electropolymerized MPcTsdoped PPy layers.248 They can be expressed as the changes in the junction and bulk voltages, which can be determined from the J-V characteristics according to the following equations:

( )

(43)

VB ) Vbias - VJ

(44)

Figure 25. Dependence of the electrical parameter and sensor signal of a Au/PPy-CuPcTS Schottky barrier diode on the preparation temperature.

The influence of the polymer film thickness of PPyCuPcTS/Au diodes on chemical sensing properties, expressed as the change in the junction voltage ∆VJ upon exposure to NOx, is shown in Figure 24. The NOx-induced voltage shift ∆VJ, which was measured in the low bias range, decreases when the PPy film thickness increases. The increase in the bulk resistance due to the decrease of the doping level causes a stronger influence of the neutral bulk effect. A decrease in the diode current exposed to NOx is observed in the low bias voltage region, because the junction resistance RJ is dominant relative to the bulk resistance RB. In the higher forward bias region, where the flat-band condition is reached, the depletion width WJ decreases to zero, the junction resistance reaches a minimum value, and the bulk resistance of the PPy layer may become dominant. Therefore, the gas response of the current-bias behavior is dominated by the gas-induced change of the PPy layer resistance. Influence of Doping LeVelsFilm Preparation Temperature. The effects of substrate temperature during film formation, of the annealing process on the film morphology and the crystalline structure, and the influence on the gassensing properties for NO2 have been studied. Several authors have reported that, by manipulation of the film morphology, it is possible to control the sensitivity and response characteristics of the gas sensors.198,250,251 Masui et al. found that the gas sensitivity of the amorphous copper phthalocyanine (CuPc) film is higher than that of the polycrystalline CuPc film because the amorphous film has a looser stacking of CuPc molecules and, hence, more adsorption sites for NOx.251 Schottky barrier diodes based on PPy layers prepared at room temperature (295 K) do not show very large rectification ratios (Table 4). This may be due to a very thin barrier formed at Au/PPy junctions and may be an effect of the tunneling current.137 With an acceptor concentration on the

order of 1019 cm-3 and assuming the depletion width to scale as the square root of the acceptor density (eq 37), a depletion width of about 1-10 nm is expected for these devices. Hence, the tunneling current component cannot be neglected at room temperature (section 3.2.4). This is associated with the relatively large values of the ideality factor (see Table 4). In general, the rectification ratio of diodes can be improved by the enhancement of the depletion width. According to eq 3, this may be reached either by increasing the built-in potential (i.e., contact potential) or by decreasing the active charge carrier concentration. It was found that the charge carrier concentration calculated from 1/C2 versus V curves is far different from that estimated from the doping level obtained from EQCM measurements.241,249 (∼1021 cm-3), assuming that all dopant charges give rise to one hole polaron (or half a bipolaron). This may be due to the existence of a large concentration of defects in amorphous semiconductor materials. Due to these structural defects, the majority (about 80 to 90%) of the self-localized electronic defects (polarons or bipolarons) are trapped, such that they cannot contribute any more to the charge carrier transport through the polymer layer. The concentration of structural defects increases with increasing preparation temperature, because the polymerization rate rises simultaneously.252 In Figure 25 and Table 7 the electrical parameters of Au/ PPy-CuPcTS Schottky barrier diodes can be found as a function of the polymerization temperature of PPy.249 The rectification ratio increases rapidly with the increase of the preparation temperature, and simultaneously, the saturation current density decreases largely. This may be due to the enhancement of the defect concentration in PPy with the increase in the preparation temperature, which decreases the so-called active acceptor concentration and, hence, enhances the depletion width of the Schottky barrier according to eq 37.

VJ =

nkT J ln e J0

Semiconductor Junction Gas Sensors

Figure 26. Transient response of a Au/PPy-CuPcTS Schottky barrier diode electrochemically prepared at different temperatures, exposed to 11 ppm NOx in ambient air.

Therefore, the concentration of the active charge carriers is dependent on the preparation temperature and decreases with increasing preparation temperature (Figure 25A). Simultaneously, the size of the space charge region increases and therefore the junction resistance is enhanced with respect to the bulk resistance. Hence, the NO2-induced shift of junction voltage increases for the same concentration (Figure 25D), enhancing the sensitivity of the Schottky barrier diode. Furthermore, with increasing temperature, the crystallinity decreases and is zero when reaching the glass transition temperature (which is between 315 and 328 K for PPyCuPcTS64). The gas or vapor can only be absorbed by the amorphous part of the polymer, which increases with the preparation temperature. Hence, NO2 interacts more strongly with layers prepared at higher temperatures, leading to a higher relative change in the active charge carrier concentration (Figure 25B) and a higher change in the built-in voltage for the same NO2 concentration (Figure 25C). Both effects cause a further increase of the gas-induced junction voltage shift with the preparation temperature. The diffusion of gas or vapor in conducting polymers and their adsorption on polymer surfaces are the fundamental processes that control the chemical sensing response. It is supposed that diffusion processes in the crystalline polymer phase are much slower compared to those in the amorphous phase.253 Hence, the polymerization temperature is an important parameter influencing the diffusion processes of polymeric sensor transducers. Figure 26 shows the response curves of diodes with PPy films exposed to 11 ppm NO2. The diode based on the PPy film prepared at 323 K exhibits the response time and the recovery time, which are much faster than those of diodes based on PPy films prepared at 283 and 303 K. The decrease in response and recovery time can be explained by an increase in the amount of the amorphous phase compared to that of the crystalline phase in PPy when the preparation temperature increases.

5. Summary and Perspectives Schottky barrier devices are promising candidates for sensor applications. The devices are simple to fabricate, obviating the need for expensive processes used in microtechnology. Both the metal and the semiconductor forming the Schottky barrier junction can be used as the chemical sensing components interacting with the gases or vapors. The material choices and combinations define the working principle, the sensitivity, and the selectivity of the sensor devices.

Chemical Reviews, 2008, Vol. 108, No. 2 395

In Schottky barrier devices, particular attention must be paid to the formation of the junction barrier between the metal and the semiconductor. In the case of organic semiconductors, there is still only limited understanding of the interface between the organic material and the metal, which makes data interpretation complicated. Experimental data show that the junction barrier turns out to be not ideal because of the presence of interface states as well as the undesired formation of an interfacial layer caused either by instability of the components in air or by chemical reaction between the metals and semiconductors. The deviation from ideality can influence the sensor characteristics up to a complete breakdown of the ability to detect gases or vapors. Therefore, a high quality of the interface is the key factor affecting the sensing performances of the Schottky barrier devices. This certainly can increase the fabrication complexity and cost. The key difference between organic and inorganic Schottky barrier junction devices is the ability of many gases and vapors to penetrate through the organic semiconductor to the interface either to change the Schottky junction resistance or to interact with the bulk of the semiconductor, which causes a work function change of the material. In the case of inorganic Schottky barrier junction devices, gas permeation toward the metal/semiconductor interface works only for hydrogen or hydrogen-producing compounds and causes formation of a dipole layer. Organic semiconductors offer a viable alternative to conventional inorganic semiconductors for application in chemical sensors. They show sensitivities toward many gases or vapors, ranging from organic solvents to inorganic gases. Moreover, their porosity enables an ease of penetration of gases or vapors. The mechanical flexibility, environmental stability, and solution processability offer an enormous potential for applications within the field of microsensors. The macromolecular character and the high degree of flexibility in preparation make various physical and chemical properties realizable. The tunability of the sensing properties by the nature of the dopants as well as by the preparation procedures is an important benefit. Individual modification of each sensor is possible in only one step. This allows for the inexpensive fabrication of multisensing arrays, an aspect that makes sensors based on organic semiconductors suitable for commercialization. They are inherently compatible with solid-state integrated chemical sensors, because they can be readily incorporated into microfabricated structures. Additional benefits are the low device prices, the small dimensions, and low power consumption.

6. List of Symbols and Abbreviations ∆ δ 0 s i φb φe ∆φbi φbi µp,n µ θ τ τtr

interfacial layer thickness, cm2 fractional charge-transfer coefficient upon gas interaction permittivity of free space, A s V-1 cm-1 permittivity of the semiconductor, A s V-1 cm-1 permittivity of the interfacial layer, A s V-1 cm-1 Schottky barrier height, eV effective barrier height, eV image force lowering of the barrier height, eV contact potential or built-in potential, eV charge carrier mobility, cm2 V-1 s-1 dipole moment, A s cm-1 ratio of free to trapped charge carriers dielectric relaxation time, s electron lifetime within the depletion layer, s

396 Chemical Reviews, 2008, Vol. 108, No. 2 χ ω ξ A A** C E EC EF EWF Eg Emax

electron affinity, eV angular frequency, Hz energy difference between EC and EF, eV area or junction area, cm2 effective Richardson constant, A cm-2 K-2 capacitance or capacity per area, F cm-2 energy, eV conduction band edge, eV Fermi level, eV electron work function, eV energy band gap, eV maximum field strength at the metal/semiconductor interface, V cm-1 EV valence band edge, eV eV0 band bending at zero applied voltage ()φbi), eV e electronic charge, A s h Planck constant, J s h0 concentration of thermally generated holes in the valence band, cm-3 J0 saturation current density, A cm-2 J current density, A cm-2 k Boltzmann constant, J K-1 L layer thickness, cm m* reduced mass of electron, kg Nads density of adsorption sites, cm-3 Nssb density of interface states, cm-2 eV-1 NC effective density of states in the conduction band, cm-3 ND donor concentration, cm-3 Ntr effective density of traps, cm-3 NV effective density of state in the valence band, cm-3 n ideality factor ni intrinsic electron concentration, cm-3 pgas partial gas pressure, N m-2 S slope parameter s frequency exponent T absolute temperature, K V voltage or bias voltage at Schottky contact, V Vbi built-in voltage, V Vd diffusion voltage at zero bias voltage, V Vr reverse bias voltage, V W charge depletion zone or depletion layer width, cm index “s” semiconductor index “m” metal index “h” hole index “e” electron index “b” bulk index “j” junction Alq3 aluminum tris(8-hydroxychinoline) BEDT-TTF bis(ethylenedithio)tetrathiafulvalene BN benzonitrile CELT charging-energy-limited tunneling CHEMFET chemical sensitive field-effect transistor DCM dichloromethane EQCM electrochemical quartz crystal microbalance F4-TCNQ tetrafluorotetracyanoquinodimethane FIT fluctuation-induced tunneling HOMO highest occupied molecular orbital LUMO lowest unoccupied molecular orbital MIS metal-insulator-semiconductor MOSFET metal-oxide-semiconductor field-effect transistor MPcTS metallophthalocyanine tetrasulfonate OLED organic light emitting diode OFET organic field-effect transistor Pc phthalocyanine QCM quartz crystal microbalance THF tetrahydrofuran TPD N,N′-diphenyl-N,N′-bis(3-methylphenyl)-1,1′-biphenyl-4,4′-diamine VRH variable-range hopping

Potje-Kamloth

7. References (1) Go¨pel, W.; Hesse, J.; Zemel, J. N. Sensors, A ComprehenisiVe SurVey; Vol. 2, Chemical and Biochemical Sensors; VCH Weinhein: New York, 1991. (2) Janata, J. In Solid-State Chemical Sensors; Janata, J., Huber, R. J., Eds.; Academic Press: New York, 1985; and references therein. (3) Lundstro¨m, I.; Shivaraman, M. S.; Svensson, C. M. J. Appl. Phys. Lett. 1975, 46, 3876. (4) Janata, J.; Josowicz, M. Nat. Mater. 2003, 2, 19. (5) Steele, M. C.; Hile, J. W.; Maclver, B. A. J. Appl. Phys. 1976, 47, 2537. (6) Lundstro¨m, I.; Svensson, C. In Solid State Chemical Sensors; Janata, J., Huber, R. J., Eds.; Academic Press: New York, 1985; and references therein. (7) Lechuga, L. M.; Calle, A.; Golmayo, D.; Tejedor, P.; Briones, F. J. Electrochem. Soc. 1991, 138, 159. (8) Lechuga, L. M.; Calle, A.; Golmayo, D.; Briones, F. Sens. Actuators, B 1992, 7, 614. (9) Liu, W. C.; Pan, H. J.; Chen, H. I.; Lin, K. W.; Cheng, S. Y.; Yu, K. H. IEEE Trans. Electron DeVices 2001, 48, 1938. (10) Yousuf, M.; Kuliyev, B.; Lalevic, B.; Poteat, T. L. Solid-State Electron. 1982, 25, 753. (11) Kim, C. K.; Lee, J. H.; Lee, Y. H.; Cho, N. I.; Kim, D. J. Sens. Actuators, B 2000, 66, 116. (12) Heeger, A. J. Angew. Chem., Int. Ed. 2001, 40, 2591. (13) Burroughes, J. H.; Bradley, D. D. C.; Brown, A. R.; Marks, R. N.; MacKay, K.; Friend, R. H.; Burn, P. L.; Holmes, A. B. Nature 1990, 347, 539. (14) Burn, P. L.; Holmes, A. B.; Kraft, A.; Bradley, D. D. C.; Brown, A. R.; Friend. R. H.; Gymer, R. W. Nature 1992, 356, 47. (15) Alidiss, M., Ed. Intrinsically Conducting Polymers: An Emerging Technology; NATO ASI Series E; Kluwer: Dordrecht, 1993; Vol. 246. (16) Sirringhaus, H.; Tessler, N.; Friend, R. H. Science 1998, 280, 1741. (17) Wienk, M. M.; Kroon, J. M.; Verhees, J. H.; Knol, J.; Hummelen, J. C.; van Hal, P. A.; Janssen, R. A. J. Angew. Chem., Int. Ed. 2003, 42, 3371. (18) Teh, K. S.; Lin, L. J. Micromech. Microeng. 2005, 15, 2019. (19) Ho, P. K. H.; Thomas, D. S.; Friend, R. H.; Tessler, N. Science 1999, 285, 233. (20) Pei, Q. B.; Yu, G.; Zhang, C.; Yang, Y.; Heeger, A. J. Science 1995, 269, 1086. (21) Brabec, C. J.; Zerza, G.; Cerullo, G.; De Silvestri, S.; Luzzati, S.; Hummelen, J. C.; Sariciftci, N. S. Chem. Phys. Lett. 2001, 340, 232. (22) Padinger, F.; Rittberger, R. S.; Sariciftci, N. S. AdV. Funct. Mater. 2001, 11, 1. (23) Scheinert, S.; Paasch, G. Phys. Status Solidi A 2004, 201, 1263. (24) Nishide, T.; Oyamada, T.; Akiyana, S.; Sasabe, H.; Adachi, C. AdV. Mater. 2006, 18, 3120. (25) Bandyopadhyay, S.; Bhattacharyya, A.; Sen, S. K. J. Appl. Phys. 1999, 85, 3671. (26) Janata, J. Phys. Chem. Chem. Phys. 2003, 5, 5155. (27) Assadi, A.; Spetz, A.; Willander, M.; Svensson, C.; Lundstro¨m, I.; Ingana¨s, O. Sens. Actuators, B 1994, 20, 71. (28) Miasik, J. J.; Hooper, A.; Tofield, B. C. J. Chem. Soc., Faraday Trans. 1986, 82, 1117. (29) Tuyen, Le. T. T.; Potje-Kamloth, K.; Liess, H.-D. Thin Solid Films 1997, 292, 293. (30) Hanawa, T.; Kuwabata, S.; Yoneyama, H. J. Chem. Soc., Faraday Trans. 1, 1988, 84, 1587. (31) Ashcroft, N, W.; Mermin, N. D. Solid State Physics; Saunders: Philadelphia, 1976. (32) Skotheim, T. A., Ed. Handbook of Conducting Polymers, Vol. 1 and 2; Marcel Dekker: New York, 1986. (33) Arkhipov, V. I.; Heremans, P.; Emelianova, E. V.; Adriaenssens, G. J.; Ba¨ssler, H. Appl. Phys. Lett. 2003, 82, 3245. (34) Tanase, C.; Blom, W. M.; de Leeuw, D. M.; Meijer, E. J. Phys. Status Solidi 2004, 201, 1236. (35) Arkhipov, V. I.; Heremans, P.; Emeliannova, E. V.; Ba¨ssler, H. Phys. ReV. B 2005, 71, 055214. (36) Koch, N.; Rajagopal, A.; Ghijsen, J.; Johnson, R. L.; Leising, G.; Pireaux, J.-J. J. Phys. Chem. B 2000, 104, 1434. (37) Gao, W.; Kahn, A. Org. Electron. 2002, 3, 53. (38) Shen, C.; Kahn, A.; Hill, I. G. In Conjugated Polymer and Molecular Interfaces; Salaneck, W. R., Seki, K., Kahn, A., Pireaux, J.-J., Eds.; Marcel Dekker: NewYork, 2001. (39) Blochwitz, J.; Pfeiffer, M.; Fritz, T.; Leo, K.; Alloway, D. M.; Lee, P. A.; Armstrong, N. R. Org. Electron. 2001, 2, 97. (40) Ganzorig, C.; Sugo, K.; Fujihira, M. Mater. Sci. Eng., B 2001, 85, 140. (41) Ganzorig, C.; Fujihira, K. Appl. Phys. Lett. 2000, 77, 4211.

Semiconductor Junction Gas Sensors (42) Zhou, C.; Pfeiffer, M.; Blochwitz, J.; Werner, A.; Nollau, A.; Fritz, T.; Leo, K. Appl. Phys. Lett. 2001, 78, 410. (43) Nollau, A.; Pfeiffer, M.; Fritz, T.; Leo, K. J. Appl. Phys. 2000, 87, 4340. (44) Ritsko, J. J.; Fink, J.; Crecelius, G. Solid State Commun. 1983, 46, 477. (45) Chance, R. R.; Boudreaux, D. S.; Bredas, J. L.; Silbey, R. In Handbook of Conducting Polymers; Skotheim, T. A., Ed.; Marcel Dekker: New York, 1986; Vol. 2, Chapter 24. (46) Conwell, E. M. In Organic ConductiVe Molecules and Polymers; Nalwa, H. S., Ed.; John Wiley & Sons: Chichester, 1997; Vol. 4, Chapter 1. (47) Heeger, A. J. In Conjugated Polymers and Related Materials; Salaneck, W. R., Lundstro¨m, I., Ranby, B., Eds.; Oxford Science Publications: Oxford, 1993; Chapter 4. (48) Heeger, A. J.; Kivelson, S.; Schrieffer, J. R.; Su, W. P. ReV. Mod. Phys. 1988, 60, 781. (49) Joo, J.; Long, S. M.; Pouget, J. P.; Oh, E. J.; MacDiarmid, A. G.; Epstein, A. J. Phys. ReV. B 1998, 57, 9567. (50) Austin, I. A.; Mott, N. F. AdV. Phys. 1969, 18, 41. (51) Elliot, S. R. Philos. Mag. 1971, 36, 1291. (52) Mott N. F.; Davis, E. A. Electronic Processes in Non-Crystalline Materials; Clarendon-Press: Oxford, 1979. (53) Sheng, P.; Abeles, B.; Arie, Y. Phys. ReV. Lett. 1973, 31, 44. (54) Mott, N. F. Metal-Insulator Transition; Taylor-Francis: London, 1990; and references therein. (55) Abeles, B.; Sheng, P.; Coutts, M. D.; Arie, Y. AdV. Phys. 1975, 407. (56) Joo, J.; Oblakowski, Z.; Du, G.; Pouget, J. P.; Oh, E. J.; Wiesinger, J. M.; Min, Y.; MacDiarmid, A. G. Phys. ReV. B 1994, 49, 2977. (57) Pope, E:, Swenberg, C. E. Electronic Processes in Organic Crystalls and Polymers, 2nd ed.; Oxford University Press: Oxford, U.K., 1999. (58) Ba¨ssler, H. Phys. Status Solidi B 1993, 175, 15. (59) Dieckmann, A.; Ba¨ssler, H.; Borsenburger, P. M. J. Chem. Phys. 1993, 99, 8136. (60) Novikov, S. V.; Vannikov, A. V. J. Phys. Chem. 1995, 99, 14573. (61) Ba¨ssler, H. Phys. Status Solidi B 1993, 175, 15. (62) Arkhipov, V. I.; Emelianova, E. V.; Adriaenssens, G. J. Phys. ReV. B 2001, 63, 081202(R). (63) Mitchell, G.; Davis, F.; Legge, C. Synth. Met. 1988, 26, 247. (64) Cabala, R. Ph.D. Thesis, Charles University, Prague, 1998. (65) Epstein, A.; Rommelmann, H.; Bigelow, R.; Gibson, H.; Hoffmann, D.; Tanner, D. Phys. ReV. Lett. 1983, 50, 1866 and references therein. (66) Pochtennyi, A. E.; Misevich, A. V. Phys. Solid State 1996, 38, 1422. (67) Saunders, B. R.; Murray, K. S.; Fleming R. J.; Korbatich, Y. Chem. Mater. 1993, 5, 809 and references therein. (68) Pfluger, P.; Weiser, G.; Scott, J. C.; Street, G. B. In Handbook of Conducting Polymers; Skotheim, T. A., Ed.; Marcel Dekker Inc.: New York, 1986; Vol. 2, Chapter 38. (69) Anderson, W. Commun. Solid State Phys. 1970, 2, 193. (70) Tenelsen, K.; Schreiber, M. Phys. ReV. B 1994, 49, 12662. (71) Levy, Y. E.; Souillard, B. Europhys. Lett. 1987, 4, 233. (72) Schreiber, M.; Grussbach, H. Phys. ReV. Lett. 1995, 51, 663. (73) Sze, S. M.; Rhoderick, S. M.; Sharma, B. L.; Schottky, W. Z. Phys. 1939, 113, 367. (74) Mott, N. F. Proc. Cambridge Philos. Soc. 1938, 34, 568. (75) Sze, S. M.; Viturro, R.; Masilhiot, C.; Shaw, J. L.; Brillson, L. J.; LaGaffe, D.; Margaritondo, G.; Pettit, C. D.; Woodall, J. M. J. Vac. Sci. Technol. 1989, A7, 855. (76) Cowley, A. M.; Sze, S. M. J. Appl. Phys. 1965, 36, 3212. (77) Rhoderick, E. H. J. Appl. Phys. 1975, 46, 2809. (78) Braun, F. Poggendorff’s Ann. Phys. 1874, 153, 556. (79) Schottky, W. Naturwissenschaften 1938, 26, 843. (80) Mott, N. F. Proc. Cambridge Philos. Soc. 1938, 34, 568. (81) Sze, S. M. Physics of Semiconductor DeVices; John Wiley & Sons: New York, 1981. (82) Rhoderick, E. H.; Williams, R. H. Metal-Semiconductor Contacts; Oxford Science Publications: Oxford, U.K., 1988. (83) Milnes, A. G.; Feucht, D. L. Heterojunctions and Metal-Semiconductor Junctions; Academic Press, Inc.: New York, 1972. (84) Sadaoka, Y.; Jones, T. A.; Go¨pel, W. J. Mater. Sci. Lett. 1989, 8, 1095. (85) Chen, H.; Rambhatla, A.; Janata, J.; Potje-Kamloth, K. J. Electrochem. Soc. 2007, 154, H354. (86) Lampert, M. A.; Mark, P. Current Injection in Solids; Academic Press: New York, 1970. (87) Crone, B. K.; Davids, P. S.; Campbell, I. H.; Smith, D. J. Appl. Phys. 1998, 84, 833. (88) Karg, S.; Meier, M.; Riess. W. J. Appl. Phys. 1997, 82, 1951. (89) Campbell, A. J.; Weaver, M. S.; Lidzey, D. G.; Bradley, D. D. J. Appl. Phys. 1998, 84, 6737. (90) Crone, B. K.; Campbell, I. H.; Davids, P. S.; Smith, D. L. J. Appl. Phys. 1999, 86, 5767.

Chemical Reviews, 2008, Vol. 108, No. 2 397 (91) Gustafsson, G.; Sundberg, M.; Ingana¨s, O.; Svensson, C. J. Mol. Electron. 1990, 6, 105. (92) Townsend, P. D.; Friend, R. H. Phys. ReV. 1989, B40, 3112. (93) Henisch, H. K. Semiconductor Contacts; Clarendon: Oxford, U.K., 1984. (94) Rhoderick, H. E.; Freeouf, J. L.; Woodall, J. M. Appl. Phys. Lett. 1981, 39, 72. (95) Spicer, W. E.; Kendelewicz, T. J.; Newman, N.; Chin, K. K.; Lindau, I. Surf. Sci. 1986, 168, 240. (96) Tersoff, J. Phys. ReV. Lett. 1984, 52, 476. (97) Salaneck, W. R., Seki, K., Kahn, A., Pireaux. J. J., Eds. Conjugated Polymer and Molecular Interfaces: Science and Technology for Photonic and Optoelectronic Applications; Marcel Dekker: New York, 2002. (98) Cahen, D.; Kahn, A.; Umbach, E. Mater. Today 2005, 8, 32. (99) Ishii, H.; Sugiyama, K.; Ito, E.; Seki, K. AdV. Mater. 1999, 11, 605. (100) Hill, I. G.; Rajagopal, A.; Kahn, A. J. Appl. Phys. 1998, 84, 3226. (101) Mori, T.; Fujikawa, H.; Tokito, S.; Taga, Y. Appl. Phys. Lett. 1998, 73, 2763. (102) Schlaf, R.; Parkinson, B. A.; Lee, P. A.; Nebesny, K. W.; Armstrong, N. R. J. Phys. Chem. B 1999, 103, 2984. (103) Lee, S. T.; Wang, Y. M.; Hou, C. W.; Tang, C. W. Appl. Phys. Lett. 1999, 74, 670. (104) Seki, K.; Nish, T.; Tanaka, S.; Komatsu, K.; Ikame, T.; Ishii, H.; Kanai, K. Proc. Int. Symp. Super-Functionality Organic DeVices; IPAP Conf. Ser. 6; Institute of Pure and Applied Physics: Tokyo, 2005, p 178. (105) Nishi, T.; Kanai, K.; Ouchi, Y.; Willis, M. R.; Seki, K. Chem. Phys. 2006, 325, 121. (106) Salaneck, W. R., Seki, K., Kahn, A., Pireaux. J. J., Eds. Conjugated Polymer and Molecular Interfaces: Science and Technology for Photonic and Optoelectronic Applications; Marcel Dekker: New York, 2002. (107) Yamane, H.; Yabuuchi, Y.; Fukagawa, H.; Kera, S.; Okudaira, K. K.; Ueno, U. J. Appl. Phys. 2006, 99, 093 705. (108) Fukagawa, H.; Yamane, H.; Kataoka, T.; Kera, S.; Nakamura, M.; Kudo, K.; Ueno, N. Phys. ReV. B: Condens. Matter Mater. Phys. 2006, 73, 24531. (109) Schroeder, P. G.; France, C. B.; Park, J. B.; Parkinson, B. A. J. Appl. Phys. 2002, 91, 3010. (110) Morikawa, Y.; Ishii, H.; Seki, K. Phys. ReV. B: Condens. Matter Mater. Phys. 2004, 69, 041403. (111) Witte, G.; Lukas, S.; Bagus, P. S.; Wo¨ll, C. Appl. Phys. Lett. 2005, 87, 263502. (112) Kahn, A.; Koch, N.; Gao, W. J. Polym. Sci., Part. B: Polym. Phys. 2003, 41, 2529. (113) Gao, W.; Kahn, A. J. Appl. Phys. 2003, 94, 359. (114) Kronik, L.; Shapira, Y. Surf. Sci. Rep. 1999, 37, 1. (115) Heeger, A. J.; Parker, I. D.; Yang, Y. Synth. Met. 1994, 67, 23. (116) Brazovskii, S.; Kirova, N. N. Synth. Met. 1993, 55-57, 1421. (117) Davids, P. S.; Saxena, A.; Smith, D. L. J. Appl. Phys. 1995, 78, 4234. (118) Dannetum, P.; Fahlman, M.; Fauquet, C.; Kaeijama, K.; Sononda, Y.; Lazzaroni, R. Synth. Met. 1994, 6, 133. (119) Dannetum, P.; Lo¨gland, M.; Salaneck, W. R.; Fredriksson, C.; Staftsto¨m, S.; Holmes, A. B.; Brown, A. R.; Graham, S.; Friend, R. H.; Lhost, O. Mol. Cryst. Liq. Cryst. 1993, 43, 228. (120) Choong, V.; Park, Y.; Hsiejh, B. R.; Gao, Y. J. Phys. 1999, D30, 1421. (121) Cabala, R.; Meister, V.; Potje-Kamloth, K. J. Chem. Soc., Faraday Trans. 1997, 93, 131. (122) Bott, B.; Thorpe, S. C. In Techniques and Mechanism in Gas Sensing; Moseley, P. T., Norris, J., Williams, D. E., Eds.; Adam Hilger Series on Sensors; IOP Publishing: Philadelphia, PA, 1991; Chapter 5. (123) Van Ewyk, R. L.; Chadwick, A. V.; Wright, J. D. J. Chem. Soc., Faraday Trans. 1 1980, 76, 2194. (124) Morrison, S. R. The Chemical Physics of Surfaces; Plenum Press: New York, 1977. (125) Go¨pel, W. Prog. Surf. Sci. 1985, 20, 9. (126) Blackwood, D.; Josowicz, M. J. Phys. Chem. 1991, 95, 493. (127) Thieblemont, J. C.; Planche, V.; Petrescu, C.; Bouvier, J. M.; Bidan, G. Polym. Degrad. Stab. 1994, 43, 29. (128) Janata, J. Anal. Chem. 1991, 63, 2546. (129) Ruths, P.; Ashok, S.; Fonash, S.; Ruths, J. IEEE Trans. Electron DeVices 1981, ED-28, 1003. (130) Johansson, M.; Lundstro¨m, I.; Ekedahl, L.-G. J. Appl. Phys. 1998, 84, 44. (131) Card, H. C.; Rhoderick, E. H. J. Phys. D: Appl. Phys. 1971, 4, 1586. (132) Nguyen, V. C.; Potje-Kamloth, K. Thin Solid Films 1999, 338, 142. (133) Kanicki, J. In Handbook of Conducting Polymers; Skotheim, T. A., Ed.; Marcel Dekker Inc.: New York, 1986; Chapter 17. (134) Chot, T. Phys. Status Solidi A 1981, 66, K43. (135) Ingana¨s, O.; Skotheim, T.; Lundstro¨m, I. J. Appl. Phys. 1983, 54, 3636.

398 Chemical Reviews, 2008, Vol. 108, No. 2 (136) Magaud, L.; Cyrot-Lackmann, F. Encyclopedia of Applied Physics; VCH: New York, 1996; Vol. 16. (137) Bantikassegn, W.; Ingana¨s, O. J. Phys. D: Appl. Phys. 1996, 29, 2971. (138) Macdonald, J. R.; Johnson, W. R. In Impedance Spectroscopy; Macdonald, J. R., Ed.; John Wiley & Sons, Inc.: New York, 1987. (139) Lundstro¨m, I.; Shivaraman, M. S.; Svensson, S.; Lundkvist, I. J. Appl. Phys. Lett. 1975, 26, 55. (140) Zhang, W.; de Vasconcelos, E. A.; Uchida, H.; Katsube, T.; Nakatsubo, T.; Nishioka, Y. Sens. Actuators, B 2000, 65, 154. (141) Fang, Y. K.; Hwang, S. B.; Lin, C. Y.; Lee, C. C. Appl. Phys. Lett. 1990, 57, 2686. (142) Tongson, L. L.; Knox, B. E.; Sullivan, T. E.; Fonash, S. J. J. Appl. Phys. 1979, 50 1535. (143) Lundstro¨m, I. Sens. Actuators, B 1981/82, 2, 105. (144) Tibuzzi, A.; Di Natale, C.; D’Amico, A.; Margesin, B.; Brida, S.; Zen, M.; Soncini, G. Sens. Actuators, B 2002, 83, 175. (145) Rahimi, F.; Iraji Zad, A. Sens. Actuators, B 2006, 115, 164. (146) Freeouf, J. L.; Woodhall, J. M. Appl. Phys. Lett. 1981, 39, 727. (147) Cheng, S. Y. Mater. Chem. Phys. 2002, 78, 525. (148) Cheng, C.-C.; Tsai, Y.-Y.; Kin, K.-W.; Chen, H.-I.; Hsu, W.-H.; Chuang, H.-M.; Chen, C.-Y.; Liu, W.-C. Semicond. Sci. Technol. 2004, 19, 778. (149) Liu, W. C.; Pan, H. J.; Chen, H. I.; Lin, K. W.; Wang, C. K. Jpn. J. Appl. Phys. 2001, 40, 6254. (150) Salehi, A.; Nikfarjam, A.; Jamshidi, D.; Kalantari, D. J. Sens. Actuators, B 2006, 113, 419. (151) Salehi, A.; Kalantari, D. J. Sens. Actuators, B 2007, 122, 69. (152) Chen, H.-I.; Chu, Y.-I. Semicond. Sci. Technol. 2004, 19, 39. (153) Chen, H.-I.; Chou, Y.-I.; Hsiung, C.-K. Sens. Actuators, B 2003, 92, 6. (154) Chen, H.-I.; Chou, Y.-I. Semicond. Sci. Technol. 2003, 18, 104. (155) Talazac, L.; Barbain, F.; Varenne, C.; Mazet, L.; Pellier, S.; Soulier, C. Sens. Actuators, B 2002, 83, 149. (156) Stutzmann, M.; Steinhoff, M.; Eickhoff, M.; Ambacher, O.; Nebel, C. E.; Schalwig, J.; Neuberger, R.; Mu¨ller, G. Diamond Relat. Mater. 2002, 11, 886. (157) Hasegawa, H. Curr. Appl. Phys. 2007, 7, 318. (158) Matsuo, K.; Hashizume, T.; Hasegawa, H. Appl. Surf. Sci. 2005, 244, 273. (159) Hasegawa, H.; Oyama, S. J. Vac. Sci. Technol., B 2002, 20, 1647. (160) Wang, H.; Chen, A.-B. J. Appl. Phys. 2000, 87, 7859. (161) Ali, M.; Cimalla, V.; Lebedev, V.; Romanus, H.; Tilak, V.; Merfeld, D.; Sandvik, P.; Ambacher, O. Sens. Actuators, B 2006, 113, 797. (162) Song, J.; Lu, W.; Flynn, J. S.; Brandes, G. R. Solid-State Electron. 2005, 49, 1330. (163) Zhao, G.; Sutton, W.; Pavlidis, D.; Piner, E. L.; Schwank, J.; Hubbard, S. IECE Trans. Electron. 2003, E36-C, 2027. (164) Janson, M. S.; Linnarsson, M. K.; Halle´n, A.; Svensson, B. G.; Achtziger, N.; Une´us, L.; Spetz, A. L.; Forsberg, U. Phys. Scr. 2004, T108, 99. (165) Kim, C K.; Lee, J. H.; Choi, S. M.; Noh, I. H.; Kim, H. R.; Cho, N. I.; Hong, C.; Jang, G. E. Sens. Actuators, B 2001, 77, 455. (166) Hunter, G. W.; Neudeck, P. G.; Gray, M.; Androjna, D.; Chen, L.Y.; Hoffman, R. W.; Liu, C. C.; Wu, Q. H. Silicon Carbide, IIINitride and Related Materials; Transtech Publications: Switzerland, 2000; Vols. 338-342, p 1439. (167) Chen, L.-Y.; Hunter, G. W.; Neudeck, P. G.; Knight, D. Solid-State Electron. 1998, 42, 2209. (168) Zubkans, J.; Spetz, A. L.; Sundgren, H.; Winquist, F.; Kleperis, J.; Lusis, A.; Lundstro¨m, I. Proc. 8th Int. Conf. Solid-State Sens. Actuators, Technical Digest, Vol. 1, Stockholm, Sweden, IEEE: New York, 1995; p 726. (169) Wenyi, Z.; Vasconcelos, E. A.; Uchida, H.; Katsube, T.; Nakatsubo, T.; Nishioka, Y. Sens. Actuators, B 2000, 65, 154. (170) Khan, S. A.; Vasconcelos, E. A.; Uchida, E. A.; Katsube, T. Sens. Actuators, B 2003, 92, 181. (171) Hollingsworth, R. E.; Sites, J. R. J. Appl. Phys. 1982, 53, 6367 and references therein. (172) Hobeon, W. S.; Ellis, A. B. J. Appl. Phys. 1983, 54, 6956 and references therein. (173) Carpenter, M. K.; Van Ryswyk, H. V.; Ellis, A. B. Langmuir 1985, 1, 605. (174) Bailey, R. A.; Persaud, K. C. In Polymer Sensors and Actuators; Osada, Y., De Rossi, D. E., Eds.; Springer-Verlag: Berlin, 2000. (175) Liang, W. L.; Lei, J.; Martin, Ch. R. Synth. Met. 1992, 52, 227. (176) Riad, A. S. Physica B 1999, 270, 148. (177) Hiramoto, M. Jpn. J. Appl. Phys. 2003, 42, 672. (178) Parker, I. D. J. Appl. Phys. 1994, 75, 1656. (179) Wang, H. L.; MacDiarmid, A. G.; Wang, Y. Z.; Gebler, D.; Epstein, A. J. Synth. Met. 1996, 78, 33. (180) Singh, R.; Narula, A. K.; Tandon, R. P.; Mansingh, A.; Chandra, S. J. Appl. Phys. 1996, 79, 1476.

Potje-Kamloth (181) Bantikassegn, W.; Dannetun, P.; Ingana¨s, O.; Salaneck, W. R. Thin Solid Films 1993, 224, 232. (182) Koezuka, H.; Etoh, S. J. Appl. Phys. 1983, 54, 2511. (183) Bantikassegn, W.; Dannetun, P.; Ingana¨s O.; Salaneck, W. R. Synth. Met. 1993, 55-57, 36. (184) Glenis, S.; Tourillon, G.; Garnier, F. Thin Solid Films 1986, 139, 221. (185) Singh, R.; Narula, A. K. Appl. Phys. Lett. 1997, 71, 2845. (186) Nguyen, V. C. Ph.D. Thesis, Universita¨t der Bundeswehr Mu¨nchen, 2000. (187) Gould, R. D. Coord. Chem. ReV. 1996, 156, 237. (188) Fujii, A.; Ohmori, Y.; Yoshino, K. IEEE Trans. Electron. DeVices 1997, 44, 1204. (189) Law, K. Y. Chem. ReV. 1993, 93, 439. (190) Guillaud, G.; Simon, J.; Gewain, J. P. Coord. Chem. ReV. 1998, 17880, 1433. (191) Gould, R. D.; Shafai, T. S. Superficies Vacio 1999, 9, 226. (192) Riad, A. S. Physica B 1999, 270, 148. (193) Rajesh, K. R.; Menon, C. S. Can. J. Phys. 2005, 83, 1151. (194) Samuel, M.; Menon, C. S.; Umnikrishnan, N. V. Semicond. Sci. Technol. 2006, 21, 677. (195) Brown, A. R.; Bradley, D. D. C.; Burroughes, J. H.; Friend, R. H.; Greenham, N. C.; Burn, P. L.; Holmes, A. B.; Kraft, A. Appl. Phys. Lett. 1992, 61, 2793. (196) Sharma, G. D. Synth. Met. 1995, 74, 227. (197) Honeybourne, C. L.; Ewen, R. J. J. Phys. Chem. Solids 1983, 44, 833. (198) Sadaoka, A.; Jones, T. A.; Go¨pel, W. Sens. Actuators, B 1990, 1, 148. (199) Gould, R. D.; Ibrahim, N. A. Thin Solid Films 2001, 398-399, 432. (200) Wang, H. Y.; Ko, W. H.; Batzel, D. A.; Kenney, M. E.; Lando, J. B. Sens. Actuators, B 1990, 1, 1138. (201) Schoch, K. F.; Temofonte, T. A. Thin Solid Fillms 1988, 165, 83. (202) Nieuwenhiuzen, M. S.; Nederlof, A. J.; Barendsz, A. W. Anal. Chem. 1988, 60, 230. (203) Musa, I.; Eccleston, W. Jpn. J. Appl. Phys. 1998, 37, 4288. (204) Assadi, A.; Fu, Y.; Willander, M.; Svensson, C. Jpn. J. Appl. Phys. 1993, 32, 1696. (205) Fang, Y.; Chen, S.-A. Mater. Chem. Phys. 1992, 32, 380. (206) Jen, K. Y.; Miller, G. G.; Elsenbaumer, R. L. J. Chem. Soc., Chem. Commun. 1986, 1346. (207) Garnier, F.; Yassar, R.; Hajlaoui, R.; Horowitz, G.; Deloffre, F.; Servet, B.; Ries, S.; Alnot, P. J. Am. Chem. Soc. 1993, 115, 8716. (208) Greenham, N. C.; Moratt, S. C.; Bradley, D. D.; Friend, R. H.; Homes, A. B. Nature 1993, 365, 628. (209) Chittibabu, K. G.; Kamath, M.; Kumar, J.; Tripathy, S. K. Chem. Mater. 1994, 6, 475. (210) Sundberg, M.; Gustafsson, G.; Ingana¨s, O. Appl. Phys. Lett. 1990, 57, 733. (211) Turut, A.; Ko¨leli, F. J. Appl. Phys. 1992, 72, 818. (212) Kuo, C. S.; Wakim, F. G.; Sengupta, S. K.; Tripathy, S. K. Jpn. J. Appl. Phys. 1994, 33, 2629. (213) Lous, E. J.; Blom, P. W. M.; Molenkamp, L. W.; de Leeuw, D. M. Phys. ReV. B 1995, 51, 17251. (214) Musa, I.; Eccleston, W. Jpn. J. Appl. Phys. 1998, 37, 4288. (215) Abdou, M. S. A.; Orfino, P.; Xie, Z. W.; Deen, M.; Holderoft, S. AdV. Mater. 1994, 6, 838. (216) Tagmouti, S.; Oueriagli, A.; Outzourhit, A.; Khaidar, M.; Ameziane, El. L.; Yassar, A.; Youssioufi, H. K.; Garnier, F. Synth. Met. 1997, 88, 109. (217) Ohmori, Y.; Takahashi, H.; Kawai, T.; Yoshino, K. J. Appl. Phys. 1990, 29, L1849. (218) Triverdi, D. C. In Organic ConductiVe Molecules and Polymers; Nalwa, H. S., Ed.; John Wiley & Sons: Chichester, U.K., 1997; Vol. 2. (219) Green, A. G.; Woodhead, A. E. J. Chem. Soc. Trans. 1910, 97, 2388. (220) Chen, S.-A.; Fang, Y. Synth. Met. 1993, 60, 215. (221) Gupta, R. K.; Singh, R. A. Comput. Sci. Technol. 2005, 65, 677. (222) Yang, C.-H. J. Electrochem. Soc. 1999, 146, 1939. (223) Abthagir, P. S.; Saraswathi, R. J. Mater. Sci.: Mater. Electron. 2004, 15, 81. (224) Calderone, A.; Lazzaroni, R.; Bredas, J. L. Synth. Met. 1993, 5557, 4620. (225) Wu, S.; Zeng, F.; Li, F.; Zhu, Y. Eur. Polym. J. 2000, 4, 679. (226) Li, J.; Petelenz, D.; Janata, J. Electroanalysis 1993, 5, 791. (227) Conn, C.; Sestak, S.; Baker, A. T.; Unsworth, J. Electroanalysis 1998, 10, 1137. (228) Campos, M.; Bulhoes, L. O. S.; Lindino, C. A. Sens. Actuators, A 2000, 87, 67. (229) Rodrı´guez, J.; Grande, H.; Otero, T. In Organic ConductiVe Molecules and Polymers; Nalwa, H. S., Ed.; John Wiley & Sons: Chichester, England, 1997; Vol. 2 and references therein. (230) Warren, F. L.; Anderson, P. J. Electrochem. Soc. 1987, 134, 101.

Semiconductor Junction Gas Sensors (231) Saunders, B. R.; Fleming, R. J.; Murray, K. S. Chem. Mater. 1995, 7, 1082. (232) Walton, D. J.; Hall, C. E.; Chyla, A.; Viney I. V. F.; Mure, J.-M. Synth. Met. 1993, 55-57, 1465. (233) Gupta, R.; Misra, K. C.; Malhotra, B. D.; Beladakere, N. N.; Chandra, S. Appl. Phys. Lett. 1991, 58, 51. (234) Kanazawa, K. K.; Diaz, A. F.; Kronubi, M. T.; Street, G. B. Synth. Met. 1981, 4, 119. (235) Ingana¨s, O.; Lundstro¨m, I. Synth. Met. 1984/5, 10, 5. (236) Dhawan, S. K.; Triverdi, D. C. Synth. Met. 1993, 60, 67. (237) Green, M. A.; King, F. D.; Shewchun, J. Solid-State Electron. 1974, 17, 55. (238) Ohmori, Y.; Takahashi, H.; Muro, K.; Uchida, M.; Kawai, T.; Yoshino, K. Jpn. J. Appl. Phys., Part 2 1991, 30, L1247. (239) Taylor, D. M.; Gomes, H. L. J. Phys. D: Appl. Phys. 1995, 28, 2554. (240) Gupta, R. Appl. Phys. Lett. 1991, 58, 51. (241) Topart, P.; Josowicz, M. J. Phys. Chem. 1992, 96, 7824. (242) Buhks, E.; Hodge, I. M. J. Chem. Phys. 1983, 11, 5976.

Chemical Reviews, 2008, Vol. 108, No. 2 399 (243) Nagase, H.; Wakabayashi, K.; Imanka, T. Sens. Actuators, B 1993, 13-14, 596. (244) Topart, P.; Josowicz, M. J. Phys. Chem. 1992, 96, 7824. (245) Barisci, J. N.; Wallace, G. G.; Andrews, M. K.; Partridge, A. C.; Harris, P. D. Sens. Actuators, B 2002, 84, 252. (246) Bardeen, J. Phys. ReV. 1947, 71, 717. (247) Potje-Kamloth, K. Crit. ReV. Anal. Chem. 2002, 32, 121. (248) Rosenthal, M. V.; Skotheim, T. A.; Linkous, C. A. Synth. Met. 1986, 15, 219. (249) Nguyen, V. C.; Potje-Kamloth, K. J. Phys. D: Appl. Phys. 2000, 33, 1. (250) Lee, Y.-L.; Tsai, W.-C.; Maa, J.-R. Appl. Surf. Sci. 2001, 173, 352. (251) Masui, M.; Sasahara, M.; Wada, T.; Takeuchi, M. Appl. Surf. Sci. 1996, 92, 643. (252) Novak, P. Electrochim. Acta 1992, 37, 1227. (253) Hsieh, J. C.; Liu, C. J.; Ju, Y. H. Thin Solid Films 1998, 322, 98.

CR0681086

Chem. Rev. 2008, 108, 400−422

400

Optical Chemical Sensors Colette McDonagh,† Conor S. Burke,‡ and Brian D. MacCraith*,† Biomedical Diagnostics Institute, Dublin City University, Glasnevin, Dublin 9, Ireland and Optical Sensors Laboratory, National Centre for Sensor Research, Dublin City University, Glasnevin, Dublin 9, Ireland Received July 12, 2007

Contents

1. Introduction

1. Introduction 2. Sensing Platforms 2.1. Introduction 2.2. Fiber Optic Sensor Platforms 2.2.1. Passive FOCS 2.2.2. Active FOCS 2.3. Planar Waveguide-Based Sensor Platforms 2.3.1. Fluorescence-Based PWCS 2.3.2. Absorption-Based PWCS 2.3.3. Refractometric PWCS 2.3.4. Light Coupling Strategies for PWCS 2.4. Summary 3. Direct Sensors 3.1. Introduction 3.2. Direct Spectroscopic Sensing 3.2.1. Absorption-Based Sensors 3.2.2. Direct Fluorescence Sensing 3.2.3. Raman and SERS Sensing 3.3. Summary 4. Reagent-Mediated Sensors 4.1. Introduction 4.2. Reagents 4.2.1. Reagents for Colorimetric Sensing 4.2.2. Reagents for Luminescence Sensing 4.2.3. Summary 4.3. Immobilization Matrices 4.3.1. Introduction 4.3.2. Sol−Gel Matrices 4.3.3. Polymer Matrices 4.3.4. Interaction of Reagent and Support Matrix 4.4. Recent Developments in Absorption-Based Sensors 4.5. Recent Advances in Luminescence-Based Sensors 4.5.1. Introduction 4.5.2. Intensity-Based Sensing 4.5.3. Lifetime-Based Sensing 5. Key Trends and Future Perspectives 6. References

400 402 402 402 402 404 405 406 406 407 408 409 409 409 409 409 411 411 412 412 412 412 412 413 413 414 414 414 414 415 416 416 416 416 417 418 419

* To whom correspondence should be addressed. Phone: +353 (0)1 7005299. E-mail: [email protected]. † Biomedical Diagnostics Institute, Dublin City University. ‡ National Centre for Sensor Research, Dublin City University.

The field of optical chemical sensors has been a growing research area over the last three decades. A wide range of books and review articles has been published by experts in the field who have highlighted the advantages of optical sensing over other transduction methods.1-5 An appropriate definition of a chemical sensor is the so-called “Cambridge definition”:1,6 Chemical sensors are miniaturised deVices that can deliVer real time and on-line information on the presence of specific compounds or ions in eVen complex samples. Figure 1 shows a schematic of a sensor system, illustrating the three main elements, the sample (or analyte), transduction platform, and signal-processing step. Optical chemical sensors employ optical transduction techniques to yield analyte information. The most widely used techniques employed in optical chemical sensors are optical absorption and luminescence, but sensors based on other spectroscopies as well as on optical parameters, such as refractive index and reflectivity, have also been developed. This review will deal mainly with the transduction stage in Figure 1. There will be only limited discussion of sample preparation/enrichment and signal-processing techniques. Recent developments in the field have been driven by such factors as the availability of low-cost, miniature optoelectronic light sources and detectors, the need for multianalyte array-based sensors particularly in the area of biosensing, advances in microfluidics and imaging technology, and the trend toward sensor networks. In the case of luminescencebased sensors, direct intensity detection has been replaced in many applications by lifetime-based sensing, often using sophisticated phase-based techniques. This review will focus on developments in optical chemical sensing over the last 10 years with major emphasis placed on the literature from 2000 to the present day. Recent novel developments will be highlighted, and future trends will be discussed. While the optical principles used in chemical sensing have not changed substantially over the years, in many cases the transduction platforms have changed considerably, yielding sensors with vastly improved performance, the most relevant performance parameters being sensitivity, stability, selectivity, and robustness. The structure adopted for this review is shown in Figure 2. Because of the important role played by the sensor platform in current sensor developments, this review begins with a section on different platforms and platform technologies. This section includes a description of waveguide sensor technology, including fiber-based and planar waveguide systems. Refractometric platforms are also dealt with here as are interferometric sensors.

10.1021/cr068102g CCC: $71.00 © 2008 American Chemical Society Published on Web 01/30/2008

Optical Chemical Sensors

Colette McDonagh studied undergraduate physics at the National University of Ireland in Galway and was awarded her Ph.D. degree in Physics from Trinity College, Dublin, in 1980. After postdoctoral work at Trinity College and the Department of Applied Science at the University of California, Davis, she took up an academic position in the School of Physical Sciences at Dublin City University. She currently holds the position of Associate Professor, and her research interests include development of sol−gelbased optical sensors for environmental monitoring, luminescence-based optical biosensors, and development of strategies for luminescence enhancement in biochips including metal-enhanced luminescence and highbrightness nanoparticles.

Chemical Reviews, 2008, Vol. 108, No. 2 401

Brian MacCraith is Director of the Biomedical Diagnostics Institute (BDI) at Dublin City University. The BDI is a Science Foundation Ireland (SFI) Centre for Science, Engineering & Technology (CSET) focused on developing the underpinning science leading to next-generation biomedical diagnostics. Established in October 2005, the BDI is an academic−industry partnership involving six industrial and four academic partners and has been funded for 5 years in the first instance. The funding awarded to the BDI includes over 6 million euros from its industry partners and 16.5 million euros from SFI. He was founding Director of the National Centre for Sensor Research at Dublin City University and held this position from its establishment in October 1999 until the establishment of the BDI. Currently, the NCSR comprises over 200 full-time researchers working on the fundamental science and applications of chemical sensors and biosensors. With a strong track record and international reputation in the field of optical chemical sensors and biosensors, he has published widely (over 150 publications) on these topics as well as developing significant Intellectual Property.

Conor S. Burke received his B.Sc. (Hons.) degree in Physics with French from Dublin City University in 1999. In the same year, he started his Ph.D. studies with the Optical Sensors Laboratory in the School of Physical Sciences at Dublin City University under the supervision of Professor Brian MacCraith. He was awarded his Ph.D. degree in 2004 with a thesis entitled “Development of microfabricated optical chemical sensor platforms using polymer processing technology”. His work involved development of both colorimetric and fluorometric chemical sensors based on novel, enhanced optical platforms using polymer microprocessing technology. Since then, he has worked as a postdoctoral researcher at the National Centre for Sensor Research, Dublin City University, where his primary research interests include development of novel optical sensors for breath monitoring and environmental applications.

In general, optical chemical sensors may be categorized under the headings of direct sensors and reagent-mediated sensors. In a direct optical sensor, the analyte is detected directly via some intrinsic optical property such as, for example, absorption or luminescence. In reagent-mediated sensing systems, a change in the optical response of an intermediate agent, usually an analyte-sensitive dye molecule, is used to monitor analyte concentration. This latter technique is useful particularly in the case where the analyte has no convenient intrinsic optical property, which is the case for many analytes. Section 3 of this review deals with direct optical sensing using a range of optical parameters. This section deals exclusively with direct spectroscopic sensing,

Figure 1. Principal stages in the operation of a sensor.

including infrared (IR) and ultraviolet (UV) absorption techniques, direct fluorescence measurements, as well as Raman and surface-enhanced Raman spectroscopy (SERS). The section on infrared absorption includes a brief discussion of sensing based on the relatively newly developed quantum cascade lasers (QCL). Direct refractometric sensing is

402 Chemical Reviews, 2008, Vol. 108, No. 2

McDonagh et al.

Figure 2. Structure of review.

discussed in section 2. Section 4 on reagent-mediated sensors includes a discussion of analyte-sensitive reagents and associated immobilization approaches, especially those based on polymer and sol-gel matrices, and is followed by a comprehensive overview of recent developments in absorption- and luminescence-based sensors. This review does not contain formal sections on optical biosensors or surfaceplasmon-resonance (SPR) based sensors as these topics are dealt with in separate review articles in this issue. However, throughout this review, examples of biosensor applications are given where appropriate, which is a reflection of the current predominance of optical biosensors in sensor research and the overlap between chemical sensing and biosensing.

2. Sensing Platforms 2.1. Introduction The successful development of an optical chemical sensor is intrinsically linked to the nature of the physical platform on which it is based. When careful consideration of the transduction mechanism dictates the design of the sensor platform, this leads to development of highly efficient integrated optical sensors that combine light delivery/ collection with intrinsic sensing functionality. This section will present an overview of optical chemical sensor platforms with an emphasis on innovations in platform design during the past decade. In order to lend clarity to the discussion, the various types of sensor platform are grouped according to the underlying waveguide geometry on which the sensor is based, e.g., optical fiber based or planar, with further differentiation achieved according to the transduction mechanism employed by the sensor.

2.2. Fiber Optic Sensor Platforms The optical fiber is arguably the most exploited platform in the development of optical chemical sensors, a fact that is reinforced by the sheer volume of publications reporting such systems (over 2500 in the past 10 years). Indeed, this area of sensor research alone merits the publication of regular reviews of the field.5 Given the scope of fiber optic chemical sensor (FOCS) development, it is difficult for any review of the field to be completely exhaustive. This section will cover the principal fiber optic configurations and sensing schemes that are currently driving sensor research in this area. The scope of this discussion will be limited to sensing strategies

that are not the subject of other reviews in this issue. With this in mind, fiber optic sensors that utilize surface plasmon resonance (SPR) techniques and quantum-cascade laser (QCL) spectroscopy or those used to produce high-density sensor arrays will not be described in any great detail, except where such sensors are illustrative of a more general sensing configuration or strategy. While there exists a multitude of FOCS, the physical configurations employed are relatively few in number, with most sensors falling under one of the fiber categories illustrated in Figure 3, e.g., standard (unmodified) fiber, declad, active cladding, fiber bundle, bifurcated fiber bundle, U-bend, or tip based. In some cases, these configurations are combined in a single sensor, i.e., a standard fiber with a modified tip or a U-bend fiber with an active cladding. A significant fraction of the sensors reviewed here utilize one or more of these configurations/modifications. In this section, FOCS are classified according to the role played by the fiber in the operation of the sensor, i.e., passive or active. The fiber’s role is considered to be passive if the sensor response is not linked in any way to an intrinsic change in the optical properties of the fiber, which acts merely to transport the optical signal to and from the sensing environment. An active FOCS utilizes a fiber that has been modified so as to impart intrinsic analyte sensitivity to the fiber, for example, by doping the fiber cladding with an analyte-sensitive indicator. In this manner, the optical properties of the fiber are in some way modulated by the presence of the analyte. Passive and active systems are treated separately in the following subsections.

2.2.1. Passive FOCS A wide variety of passive FOCS have been developed over the past two decades, with perhaps the most prevalent example being that of the fiber-coupled spectrometer. However, only those systems developed within the past decade will be reviewed here. A passive FOCS commonly takes the form of a separate sensor element that is interrogated in some fashion by the fiber optic assembly. Several configurations are typically used to achieve this. The reflectance-based configuration is common in the development of colorimetric or absorption-based FOCS. A typical embodiment of such a sensor comprises an analytesensitive material (typically deposited on a planar support) that changes color upon interaction with the analyte of

Optical Chemical Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 403

Figure 3. Commonly used configurations for FOCS: (a) unmodified; (b) declad; (c) active or doped cladding; (d) fiber bundle; (e) bifurcated fiber bundle; (f) U-bend (shown declad); (g-i) tip based ((g) tip with active cladding, (h) etched tip, (i) modified end-face).

interest and a fiber bundle (as shown in Figure 3(d)) that is used to deliver light to this material and collect the reflected light. In normal operation, the light is delivered to the analyte-sensitive component via the central fiber in the bundle and the reflected light is collected by the outer fibers, which are arranged around the central fiber. Changes in the color (i.e., the absorption coefficient at a particular wavelength) of the sensor material are detected as changes in the intensity of the reflected light. Such sensor configurations benefit from the commercial availability of fiber bundles (as standard reflectance probes) in addition to the relatively simple optical setup. A reflectance-based system has been used in the development of a sensor for amine vapors based on silica microspheres coated with the pH indicator Bromocresol green.7 A bifurcated fiber bundle (see Figure 3(e)) has been used in a similar fashion for detection of sulfur dioxide by an organopalladium complex encapsulated in a PVC membrane.8 However, the use of a reflectance-based configuration is not limited to detection of analytes by colorimetric means. The same fiber bundles can equally be applied to the excitation and detection of fluorescence. In the case of the fiber bundle depicted in Figure 3(d), the central fiber delivers the excitation light to the external sensor element and the outer fibers are used to collect and transport the emitted fluorescence to a suitable detector. This configuration has been applied to detection of dissolved oxygen.9 The bifurcated fiber bundle can be used to deliver excitation light with one fiber and collect the excited fluorescence with the second. This configuration has been used extensively for the fluorescence-based detection of a

variety of analytes including oxygen,10 iodine,11 2,6-dinitrophenol,12 pH,13 and chloride.14 Transmission-based configurations are also commonly employed in the development of FOCS, but in recent years, research has moved away from the conventional configuration employed in many spectrometers, which sees two fibers aligned on opposing sides of a solid sensor element or optical cuvette. While current work still involves the use of fibers that are aligned in such a manner, the sensor element that is interrogated has seen some changes. A subnanoliter spectroscopic gas sensor has been developed. This consists of input and output fibers that are inserted into a fused silica capillary, which acts as a transmission cell for gaseous samples (in this case acetylene) that can be detected using an infrared source.15 A similar configuration was employed by Eom et al. in the development of a dual-wavelength measurement system for refractive index and turbidity compensation in liquid flow systems.16 In that case, a length of Teflon tubing played the role of transmission cell and the input/output fibers were used to transport visible light signals from two LED sources, one blue (λpeak ) 468 nm) the other red (λpeak ) 621 nm), to a photodiode detector. In this case, the red LED was used to provide a reference signal. Such configurations are not limited to cylindrical sample cell geometries. A variety of planar (passive) fiber-coupled sensor platforms have arisen in recent years due to the emergence of microfluidic systems as technology platforms and, in particular, the drive to develop opto-fluidic sensor chips that provide integrated, on-chip optical coupling and detection. One example of this recent trend is a PDMS-based microflow cytometer that incorporates grooves for several

404 Chemical Reviews, 2008, Vol. 108, No. 2

optical fibers positioned at different angles on either side of a flow channel that can be used to excite and detect fluorescence from particles within the channel in addition to scattered light detection.17 A second PDMS-based device has been developed that utilizes fiber alignment grooves with integrated 2D lenses on either side of a microfluidic channel for fluorescence detection.18 While these two examples exploit fluorescence detection, the optical configurations used are also inherently suited to transmission-based sensing strategies. Many opto-fluidic microsystems make use of fiber coupling to fulfill either the role of light delivery or light collection but not necessarily both. Yang et al. reported a LED-based capillary electrophoresis system that employs fiber optic detection of the LED-induced fluorescence from within a fused silica capillary.19 A PDMS microchip for the fluorescence-based detection of human serum albumin has also been described.20 This system employs an organic LED (OLED) as an excitation source and uses an integrated optical fiber to transport the fluorescence signal to a spectrometer. Chabinyc et al. described development of an integrated, PDMS-based microfluidic system for fluorescence detection that uses a fiber-coupled LED to excite fluorescence from within the microchannels of the system while employing a microavalanche photodiode, encapsulated within the chip, as a photodetector.21 These examples serve to highlight the importance of optical fibers in rendering these increasingly important platforms optically addressable at a localized level.

2.2.2. Active FOCS In this section, active FOCS are further categorized under one of three headings, which reflect the transduction mechanism employed: (1) fluorescence based, (2) absorption based (colorimetric and spectroscopic), and (3) refractometric. (1) Fluorescence-Based Active FOCS. In order to transform a standard optical fiber into an intrinsically fluorescence-based optical chemical sensor, it is necessary to impart analyte-sensitive fluorescence to the fiber in some manner. This is done through addition of fluorescent indicator molecules to the fiber platform. This may be achieved by replacing the cladding of the fiber over a portion of its length with a solid matrix that contains the fluorescent compound, a process that involves removing a portion of the original cladding of the fiber and coating the declad region with a liquid sensor material, which is subsequently cured to form a solid, fluorescent cladding (Figure 3c). A slight variation of this involves coating the declad, distal tip of the fiber with the sensor material (Figure 3g). In some instances, the distal tip may be chemically etched and the fluorescent material deposited within the etched cavity (Figure 3h), while yet another variation involves doping the fiber with the fluorescent material during the fiber fabrication process. However, the latter approach is rarely employed due to its adverse effect on the guiding properties of the fiber. A variation of the fluorescent cladding configuration was employed by Ahmad et al., who described a system based on two “positive” fibers attached to either end of a “negative” fiber.22 The system actually consisted of a single fiber, which was declad for a portion of its length and then coated along that portion with a solution of rhodamine 6G in glycerol. The higher refractive index of the glycerol relative to the silica core of the fiber caused any light guided within the fiber to leak out, making this the negative fiber component of the system, while the clad portions on either side of this

McDonagh et al.

section constituted the positive (i.e., guiding) fibers. The excitation light that leaked out from the negative fiber induced fluorescence from the rhodamine 6G in the glycerol film, which was coupled into the output positive fiber for detection. Park et al. described development of fiber optic sensors for detection of inter- and intracellular dissolved oxygen (DO) that utilize “pulled tip” fibers.23 These are multimode fibers, the tips of which have been pulled down to submicrometer dimensions and then dip coated with a liquid PVC matrix incorporating two luminescent dyes, one that is oxygen sensitive and the other acting as a reference. A similar configuration was employed by Preejith et al. in the realization of a tapered fluorescent fiber optic evanescent wave-based sensor for serum protein.24 In that work, a fiber was declad over a length of 5 cm from its end and the declad region was then contoured to produce a tapered fiber tip. The tip was then coated with a sol-gel layer in which the fluorescent complex NanoOrange was encapsulated. Evanescent wave excitation of fluorescence was employed for detection of serum albumin. Both this system and the previous one described by Park et al. differ from that produced by Ahmad et al. in that the latter system does not employ evanescent wave interrogation of the fluorescent cladding, but all three systems demonstrate the use of such a cladding or sensing layer in the production of an active FOCS. Another example of an active, fluorescence-based FOCS is the platform of Walt et al., which centers on development of high-density fiber bundles for fluorescence-based sensing.25,26 While this is a high-density array-based platform, it also demonstrates the use of chemical etching to transform the distal tip of a fiber into a well, which can be filled with a fluorescent material. In this case, the fluorescent probe takes the form of suitably functionalized microspheres that are distributed into individual microwells (which correspond to the etched distal tips of individual fibers in a high-density fiber bundle) and can bind specific, fluorescently labeled biomolecules for fluorescence detection in a highly parallel format. (2) Absorption-Based Active FOCS. Absorption-based optical sensors can be colorimetric or spectroscopic in nature. Colorimetric sensors, as the name would suggest, are based upon detection of an analyte-induced color change in the sensor material, while spectroscopic absorption-based sensors rely on detection of the analyte by probing its intrinsic molecular absorption. FOCS that utilize these techniques have been modified in order to facilitate the interaction of the light guided within the fiber with either the colorchanging indicator or the analyte itself, depending on the nature of the sensor. Both colorimetric and spectroscopic platforms will be discussed in this section, but it will become clear that some of the fiber configurations described for fluorescence-based sensors are equally useful for development of both colorimetric and spectroscopic sensors. Indeed, the modified cladding configuration described above has been used extensively in a number of modes for development of absorption-based sensors, and it should be stressed that only some of the more recent representative examples of such work are reported here. Recent examples include development of sensors for ammonia (concentration range 0.03-1%),27 pH,28,29 and ethanol.30 In some cases the sensor consisted of a dye-doped

Optical Chemical Sensors

section of fiber cladding that was interrogated by the evanescent field of the guided light,27 while in other cases a U-bend configuration was also used in conjunction with the dye-doped cladding in order to increase the interaction of the evanescent field with the analyte-sensitive reagent, thereby enhancing device sensitivity.28 The U-bend configuration was also used in conjunction with a declad section of fiber for spectroscopic monitoring of ethanol.30 In this case, a dye-doped cladding was not required as the sensor was being used to probe the intrinsic absorption of ethanol molecules in solution and it was necessary merely to remove the cladding from the section of the fiber forming the U-bend in order to yield a sensor element. This work actually described a distributed sensor network and utilized optical time domain reflectometry (OTDR) to interrogate a series of U-bend probes along a 1 km length of fiber. In a recent review of fiber optic sensors, James and Tatam described the development of a Langmuir-Blodgett (LB) film-based sensor for pH.29 This was fabricated by deposition of a LB film onto a side-polished optical fiber and is another example of a modified cladding-type sensor. With this platform, it was possible to monitor pH-induced changes in the absorption of the LB film at 725 nm, thereby providing the basis for the pH sensor. Refractometric FOCS. A range of refractometric FOCS has been developed in the past decade that involves addition of refractive index-sensitive optical structures to the optical fiber. Such structures include fiber Bragg gratings (FBG’s), Fabry-Perot cavities, and metal films for SPR measurements. FBG’s are written into photosensitive optical fibers (typically germanosilicate fibers) using a suitable laser light source. An interference pattern is produced by passing the laser output through a phase mask or splitting and recombining the laser beam. The fiber is exposed to this pattern, which causes a local modulation of the refractive index of the fiber that matches the pitch of the interference pattern. This refractive index modulation is known as a FBG, and its sensing capabilities derive from the dependence of the Bragg wavelength (and, therefore, the transmission of the grating) on the period of the grating and the effective refractive index, Neff, of the medium. FOCS based on FBG’s detect changes in Neff and require that the fiber section bearing the FBG be declad in order to impart sensitivity toward the refractive index of the environment in which the fiber is placed. A twin FBG-based sensor, developed by Sang et al., made use of the second grating in order to provide temperature compensation for the sensor.31 The device was used to detect different concentrations of sugar and propylene glycol solutions. Long period fiber gratings (LPFG’s) are similar structures that are fabricated in the same manner as FBG’s but have periods in the range 100-1000 µm. These structures are also refractive index sensitive and have been used for detection of analytes such as sodium chloride,32 ethylene glycol,32 and antibodies.33 Analyte specificity can be imparted to such sensors through deposition of an overlay29 that possesses its own refractive index response to particular analytes. A distal tip-based sensor for dissolved ammonia was recently reported by Pisco et al.34 The sensing element consisted of a tin dioxide (SnO2) thin film that was deposited onto the distal tip of a fiber by electrostatic spray pyrolysis (ESP). This is an example of a Fabry-Perot interferometric (FPI) FOCS, where changes in the thickness and refractive

Chemical Reviews, 2008, Vol. 108, No. 2 405

index of the SnO2 film at the tip of the fiber and/or changes in the refractive index of the sensing environment cause changes in the reflectivity of the fiber film interface. A zeolite thin-film-based FPI sensor was similarly employed by Liu et al. for detection of organic solvents in water.35 The optical configuration employed was similar to that of the ammonia sensor mentioned above with the zeolite thin film replacing the SnO2 layer. An optical humidity sensor for breathmonitoring applications that was also based on a fiber tip FPI configuration was described by Kang et al.36 In that case, the Fabry-Perot cavity was fabricated through layer-by-layer self-assembly of polyelectrolytes. Other examples of refractometric FOCS include a sensor for methane developed by Benounis et al., which exploited a cryptophane-doped section of fiber cladding coupled with evanescent wave interrogation.37 The specific absorption of methane by the cryptophane molecules caused a change in the cladding refractive index, which was indicated by a change in the transmitted optical power of the system. Jung et al. reported a refractive index sensor based on a threesegment optical fiber platform that consisted of a coreless silica fiber (CSF), sandwiched between two standard multimode fibers.38 The core modes from the first multimode fiber couple to cladding modes in the CSF, which then couple to core modes in the second multimode fiber. The coupling conditions are dependent on the refractive index difference between the CSF and the surrounding medium, thereby providing the system with refractive index detection capabilities.

2.3. Planar Waveguide-Based Sensor Platforms Compared with FOCS, planar waveguide chemical sensors (PWCS) are a relatively recent innovation in the field of optical chemical sensors. While most FOCS developed to date have adopted one of a small number of relatively conventional configurations (as demonstrated in the previous section), PWCS platforms research is exhibiting rapid growth in terms of innovative design and integration of multiple functionalities onto a single sensor chip. This is largely due to the compatibility of the planar geometry with a range of advanced microfabrication technologies and the ease with which such a sensor geometry can be integrated with microfluidic, lab-on-a-chip systems. Yet another advantage of such a configuration is its robust nature, which, when compared with optical fiber-based systems, is an attractive characteristic when contemplating development of practical sensor devices intended for deployment outside a laboratory environment. These attributes have made planar waveguides an ideal platform for development of integrated optical sensors. As for the previous section on FOCS, PWCS are categorized here according to the three principal transduction mechanisms employed in this field, i.e., fluorescence, absorption, and refractometry. Before examining each of these areas, it is useful to define the general configuration of a PWCS, the concept of which is illustrated in Figure 4. At its core, a PWCS comprises a planar substrate (e.g., glass, plastic, or silicon) that forms the basis of the sensor chip. In some cases, this substrate acts as the waveguide, while in others an additional waveguide layer is deposited onto the substrate. With regard to the waveguide layer, several configurations have been employed that impart various optical functionalities to the sensor platform, and some of these are reported in the following sections. In many cases,

406 Chemical Reviews, 2008, Vol. 108, No. 2

McDonagh et al.

Figure 4. Generalized configuration of a PWCS. In many cases, evanescent wave interactions with the analyte form the basis for the sensor. In some cases, a sensing layer may be employed to facilitate transduction by imparting colorimetric or fluorometric properties to the sensor.

the light that propagates within the waveguide facilitates the operation of the platform as a sensor through the interaction of its evanescent field with the sensing environment above the waveguide. However, this is not always the case, and examples of sensors employing interrogation principles other than evanescent field-based techniques are also given in the sections to follow.

2.3.1. Fluorescence-Based PWCS PWCS based on detection of fluorescence have been developed for a broad range of analytes including oxygen,39-42 pH,41 carbon dioxide,43 and a variety of biological species.44-46 Many of these systems employ the technique of evanescent wave excitation of fluorescence (also referred to as total internal reflection fluorescence). The sensor configuration is relatively straightforward, consisting of a planar waveguide to which light is coupled using one of a variety of techniques, such as prism coupling, grating coupling, or end-fire coupling. The evanescent field of the guided light extends into the superstrate (i.e., the sensing environment) and induces fluorescence in susceptible molecules located within a distance of typically 100-200 nm from the waveguide surface. Such a technique is particularly well suited to optical biosensing applications due to the high degree of surface selectivity afforded through the use of the evanescent field for excitation purposes. Chronis and Lee reported development of a total internal reflection-based biochip platform that excites fluorescence from within a microfluidic network using a single reflection,47 while the platforms described by Rowe-Taitt et al.45 and Duveneck et al.46 (the Zeptosens platform) are examples of multiple reflection (i.e., waveguiding based) evanescent wave excitation systems, which are well suited to development of array-type optical sensor chips. In some cases, the excited fluorescence is detected either above or below the sensor platform, but it is also possible to use the waveguide itself to capture fluorescence for detection at the output end face of the waveguide.39,42,48 With such a configuration, it is possible to utilize direct excitation of the fluorescence as opposed to evanescent wave excitation, which results in higher signal levels and improved sensor performance, while improving the inherent ability of the platform to discriminate between the excitation light and the fluorescence.39,48 Zourob et al. reported an alternative configuration, based on a metal-clad leaky waveguide platform, that also affords enhanced interrogation of the sensing medium, and this system can equally be applied to absorption-based or refractometric sensing schemes49(Figure 5a,b).

2.3.2. Absorption-Based PWCS One of the most prevalent examples of absorption-based PWCS platforms is that based on evanescent-wave absorp-

Figure 5. (A) Mode profile of a metal-clad leaky waveguide covered by water at a wavelength of 620 nm using a 1 mm glass substrate. (Reprinted with permission from ref 49. Copyright 2003 Elsevier Science B.V.) (B) Instrumental set up for optical sensor based on a metal-clad leaky waveguide using a (a) 473 nm solidstate laser or 610 nm LED, (b) collimating lens, (c) filter, (d) cylindrical lens, (e) polarizer, (f) grating, (g) sensor chip, (h) linear CCD detector or photomultiplier detector. (Reprinted with permission from ref 49. Copyright 2003 Elsevier Science B.V.)

tion. The configuration is similar to that described for evanescent-wave excitation-based fluorescence sensors. In this case, the sensing functionality comes about due to changes in the absorption coefficient of the sensing environment. Such changes result in more or less absorption of the evanescent field intensity by the environment, which is reflected as a change in the detected output intensity of the sensor. Sensing layers that are doped with colorimetric, analyte-sensitive indicators can be deposited onto the upper surface of the waveguide, and any analyte-induced color changes can be probed by the evanescent field of a suitable light source (i.e., one that is spectrally matched to the indicator used). In cases where the analyte is detected using direct spectroscopic or refractometric techniques, such a sensing layer may not be necessary, although transparent enrichment layers with high permeability coefficients for the analyte(s) of interest are sometimes employed. PWCS that utilize thin-film-based sensing layers have been employed for detection of analytes such as gaseous ammonia,50,51 pH,52 iodine,53 and water vapor.54 However, in recent years, a number of platforms have been developed that demonstrate a shift away from what might be seen as conventional evanescent wave absorption-based platforms. These platforms typically incorporate design features intended to enhance the interaction of the interrogating light with the sensing environment, thereby improving platform sensitivity. The integrated waveguide absorbance optode (IWAO) developed by Puyol et al. incorporates a PVC-based sensor membrane that is located between two

Optical Chemical Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 407

Figure 7. Schematic of a Mach-Zehnder interferometer.

Figure 6. Integrated waveguide absorbance optode (IWAO) configuration developed by Puyol et al. (Reprinted with permission from ref 55. Copyright 1999 American Chemical Society.)

antiresonant reflecting optical waveguides (ARROWs)55 (see Figure 6). The sensor membrane also serves as a waveguide and transports the interrogating radiation from the input ARROW to the output waveguide. The increased optical path length within the sensing layer provides enhanced sensitivity, and the device was applied to detection of potassium chloride solutions of varying strength. This configuration was later improved upon to incorporate curved ARROW waveguides and achieved improved sensitivity and response time.56 Hisamoto et al. reported development of ion-selective optodes57 that employ a similar strategy in that the sensing layer also serves as the waveguide (referred to as an “active waveguide”). This system is based primarily on a prism-coupled glass platform, which incorporates the PVC-based sensor membrane and has been applied to detection of a variety of analytes including potassium, sodium, and calcium. Qi et al. reported a composite optical waveguide (COWG) platform that consists of a potassium-ion-exchanged (PIE) waveguide onto which a tapered film of bromothymol blue (BTB)sa colorimetric pH indicatorsis deposited.58 The BTB film acts as a singlemode waveguide, and the sensor achieved a limit of detection (L.O.D.) of 1 ppb ammonia. Other systems that exploit enhanced interrogation of the sensing medium include the previously mentioned metal-clad leaky waveguide platform developed by Zourob et al.,49 while Burke et al.59 reported an injection-moulded waveguide platform that facilitated enhanced interrogation of a thin absorbing sensing layer through correct choice of the incident angle of interrogation, which was dictated by the design of integrated refractive optical elements.

2.3.3. Refractometric PWCS PWCS designed to measure refractive index changes represent one of the most extensively developed types of planar waveguide platforms. These systems exploit techniques such as interferometry, surface plasmon resonance, and light-coupling strategies to transduce refractive index changes. Comprehensive reviews of integrated optical chemical sensors based on such devices have recently been published by Lambeck60 and Gauglitz.61 A representative cross-section of cutting-edge sensor technology in this field will be delivered here, but readers should refer to these reviews for further reading on this sector of sensor technology. Sensors based on surface plasmon resonance (SPR) are perhaps the most widely known examples of refractometric

optical sensor platforms. However, this family of sensors will not be discussed here as it is the subject of a separate review in this issue. The first branch of refractometric PWCS to be discussed here is that of interferometric sensor platforms. A significant number of refractometric PWCS are based on exploitation of interferometric techniques, and one of the most commonly employed optical configurations is the Mach-Zehnder interferometer (MZI), shown in Figure 7. The MZI consists of a single-input waveguide whose transported optical power is split equally between two parallel waveguide branches using a Y-splitter. A second Y-splitter is used to recombine the optical signals from both branches into a single-output waveguide, at the end of which the output power is measured using a suitable photodetector. For the purposes of chemical sensing applications, one of the waveguide branches is either coated with a sensor membrane or exposed to the sensing environment while the other unmodified branch serves as a reference waveguide. Changes in the refractive index of the sensing layer/environment influence the effective refractive index, Neff, in the sensing channel, which induces a phase shift in the optical signal that propagates through this channel. Upon recombination, interference of the two optical signals (i.e., sensor and reference) occurs and the measured output power changes depending on the phase shift between these two signals. In the recent review by Lambeck, a comprehensive treatment of MZI interferometric platforms has been given, including a discussion of a range of strategies intended to enhance sensitivity and general sensor performance. This involves optimizing parameters such as operating wavelength, interaction length, and, most importantly, the waveguide composition. The MZI configuration has recently been exploited by a number of groups for detection of a variety of chemical and biochemical species.62-66 Improved performance can be achieved by implementing serrodyne modulation, which involves inserting electro-optical modulators into each branch.67 Other interferometry-based PWCS include the Young interferometer,68-70 the Michelson interferometer,71 and the difference interferometer (also referred to as a polarimeter).72-74 These platforms can be described as derivatives of the MZI and operate in an analogous fashion (see ref 53 for further details). Some noteworthy examples are the Young-type interferometer that has been commercialized by Farfield Sensors [www.farfield-scientific.com] and the Zeeman interferometer,75 which is a variant of the difference interferometer that is capable of resolving refractive index changes on the order of 10-8. In addition, Kribich et al. describe development of a multimode interference (MMI) coupler with tunable sensitivity that has been used to detect changes in relative humidity.76 This platform consists of single-mode input and output waveguides that are coupled to a central

408 Chemical Reviews, 2008, Vol. 108, No. 2

McDonagh et al.

Figure 8. RifS configurations and measurement principle. (Reprinted with permission from ref 61. Copyright 2005 Springer-Verlag.)

multimode waveguide. The field profile within the multimode region is dependent on the refractive index of the superstrate, and changes in this are detected as changes in the output light intensity. Lavers et al. described a related coupled-waveguide-type refractometer, which consists of a buried, single-mode potassium ion-exchanged (PIE) waveguide in a BK7 glass substrate, onto which a multimode polymer waveguide is deposited by photolithographic means.51 While this device also exploits the effects of coupling between dissimilar waveguides, it is more properly an example of a resonanttype refractometric PWCS, other examples of which include the resonant mirror platform,77 metal-clad leaky waveguides,49,78 and some grating-based devices.79-81 Other grating-based devices have been developed that rely on the refractive-index- and thickness-dependent nature of diffraction to yield a sensor response that manifests itself as a change in the intensity or spatial distribution of the diffraction pattern.82-86 Platforms based on this phenomenon are under development by Axela Biosensors [www.axelabiosensors. com], although in that particular case the sensor response is primarily due to changes in the thickness of the receptor layer that defines the diffraction grating as opposed to alterations in refractive index, and the platform is included here in order to illustrate the perceived utility and commercial potential of such a sensor configuration. Several other noteworthy PWCS platforms have been developed that exploit thickness-sensitive measurement principles, as opposed to refractometry. Reflectometric interference spectroscopy (RIfS),87 developed by Gauglitz, is one such technique. The principle of RIfS is illustrated in Figure 8 and, in brief, centers on detection of changes in the interference pattern generated by light that is reflected from the interfaces of a thin film deposited on a substrate. Changes in the thickness of the thin film (e.g., due to swelling upon interaction with the analyte of interest) result in a shift in the observed interference pattern, and the technique is capable of resolving a thickness change of 1 pm. Nikitin et al.88 exploited the interference pattern generated by light reflected from the upper and lower surfaces of a

glass cover slip in the development of the Picoscope. This system measures the correlation signal of two interferometerss the cover slip and a scanned Fabry-Perot interferometers to monitor changes in the thickness of the cover slip (due, for example, to biomolecular binding events) with a reported resolution of 3 pm. Lambeck et al. proposed three theoretical strip waveguide configurations implemented in SiON technologyssegmented, absorptive, and directional coupler basedsthat are designed to transduce thickness changes through changes in the field profile of the waveguide modes with a calculated resolution of 3 × 10-5 nm.89

2.3.4. Light Coupling Strategies for PWCS An important design consideration in the development of any PWCS is the method by which light is coupled to the sensor platform. The most commonly employed techniques are prism coupling,49,57,58,77 grating coupling,45,46,50,69,70,78 and end-fire coupling,42,55,62,63,65,66,68,72,74-76 with the choice of technique typically involving a tradeoff between fabrication costs, practicality, and sensor performance. The use of prism couplers provides relatively high coupling efficiency but detracts from the cost effectiveness, planarity, and overall robustness of the sensor platform, while end-fire coupling requires extremely precise alignment optics in order to be effective, which is not an attractive feature in a field deployable sensor but is an obstacle that can be overcome by employing fiber-pigtailing strategies (although this makes easy exchange of the sensor element problematic). The use of grating couplers can be seen as a viable alternative as this is a low-cost option that preserves the robustness of the sensor, but the reduced coupling efficiency may adversely affect sensor performance. Some efforts have been made to integrate the coupling functionality into a single chip using polymer microprocessing techniques such as injection molding59 or hot embossing.41 These techniques are compatible with the production of integrated grating couplers and refractive/prismatic optical elements that provide adequate coupling efficiency, do not reduce the robustness of the sensor, and also emphasize the trend toward integration of multiple functionalities on a single sensor platform.

Optical Chemical Sensors

2.4. Summary A range of recently developed optical chemical sensor platforms have been categorized and discussed according to the underpinning waveguide geometry employed. Further categorization was achieved according to the transduction mechanism upon which the sensors were based, and development of a number of highly innovative sensor platforms has been highlighted. This is particularly true of research in the field of planar waveguide-based devices where this robust, microfabrication-compatible geometry has been shown to lend itself well to development of optical chemical sensors having a range of integrated functionalities. However, integration of multiple functionalities (e.g., light delivery and analyte detection capability) is not limited to a planar geometry, as evidenced by the volume of work on active fiber-based sensors. The broad range of exciting developments in the field of optical chemical sensor platform development is indicative of the increasing interest in these devices and an acknowledgment of their effectiveness in providing sensitive, practical sensor solutions for a variety of application areas.

3. Direct Sensors 3.1. Introduction Direct optical detection involves the measurement of some intrinsic optical property of the analyte. Many direct sensors, for example, gas sensors, measure the intrinsic optical absorption, usually in the infrared (IR), using a variety of techniques including Fourier transform infrared (FTIR) spectroscopy or correlation spectroscopy. A range of optical configurations has been used, from free-space sensors to fiber-based configurations to optical waveguide sensors. Fiber optic and planar waveguide sensing platforms have been discussed in section 2 of this review as well as refractometric platforms such as the Mach Zehnder interferometer. Developments in IR sensing have been driven largely by the emergence of new diode laser sources, for example, QCL lasers. This section includes a comprehensive discussion of direct IR absorption sensing including FTIR-based systems. While a review of QCL-based sensing appears elsewhere in this issue, a short section on this topic is included here under the heading of diode laser-based sensing. A brief section on UV absorption is included, mainly concerned with environmental sensing of organic pollutants. Similarly, direct fluorescence sensing is dealt with mainly in the context of biomedical applications. Recent developments in SERSbased sensing are reported, which reflect the huge growth in this area in the last 10 years.

3.2. Direct Spectroscopic Sensing 3.2.1. Absorption-Based Sensors IR Absorption. For many applications, spectroscopic detection has been a reliable method of detecting chemical species. IR spectroscopy has been widely used for detection of gases, for example, CO2, CO, NO2, NH3, and CH4. In its simplest form, the technique involves confining a sample of the gas in an optical absorption cell and measuring the absorption at specific IR wavelengths, which are characteristic of the vibrational modes of the molecule. The system components usually include an IR source, optical filters to select specific absorption wavelengths, and a detector, which

Chemical Reviews, 2008, Vol. 108, No. 2 409

is sensitive at the wavelength of interest. Referencing can also be employed using a wavelength that is not absorbed by the analyte molecule. The fundamental vibrational transitions of many molecules of interest occur in the near- and mid-IR, spanning wavelengths from 1 to >10 µm. Gas analysis is becoming increasingly important in control of various industrial processes, for example, in combustion processes and monitoring industrial emissions. The need for environmental control of emissions has also led to development of analyzers to measure the concentration of a variety of gases. Much of this instrumentation is based on near- and mid-infrared absorption spectroscopy. While conventional IR sensing uses broadband IR sources, recent developments in diode laser technology, including the relatively recent availability of novel QCL sources, have led to development of highly sensitive and selective sensor systems. Fourier transform infrared (FTIR) spectroscopy is also a very powerful analytical tool. Nondispersive infrared (NDIR) sensors are widely used for gas analysis, and many commercially available gas analyzers are based on this principle. NDIR Sensors. NDIR sensors use band-pass filtering in order to select the specific analyte absorption wavelength as opposed to dispersion by a prism or a grating as employed in a spectrometer. NDIR-based sensors have been developed to monitor CO2 concentrations in the ocean in order to monitor the effects of increased atmospheric CO2 on the global carbon cycle. Total inorganic carbon measurement involves acidification of the seawater sample and quantitative measurement of the evolved CO2. NDIR detection has been found to yield a precision of 0.11-0.25% and an accuracy of ∼0.1% for total inorganic carbon detection.90,91 Kaltin et al.92 employed NDIR to measure the extracted CO2, and the system continually compares the signal to a certified reference material, which constitutes the baseline signal. A precision of ∼0.05% was achieved with an accuracy of ∼0.2%, which was comparable to the performance of the standard colorimetric technique, and the sampling time of 5 min was 3 times faster than the colorimetric method. In order to produce the speed of measurement desired to enable highspeed continuous analysis, Bandstra et al.93 employed a gaspermeable hydrophobic membrane contactor, which continuously strips the CO2 out from a flowing stream of seawater. The system continuously quantifies the CO2 concentration using an NDIR analyzer, which uses a lead selenide detector and broadband blackbody source, yielding an accuracy and precision better than (0.1% and a response time of 6s. FTIR-Based Analyzers. FTIR-based sensing is an alternative to NDIR-based analyzers and has advantages for many applications. Current FTIR spectrometers offer precise quantification of a wide range of analytes for concentrations down to single-digit ppb levels and are now available as integrated and relatively compact units compared to the complex and bulky units offered in the past. An FTIR spectrometer consists of an interferometer, usually a Michelson type, which generates an interferogram from the IR emission of the sample and then performs a Fourier transform to obtain the spectrum. The ability of FTIR to measure multiple analytes simultaneously differentiates the technique from NDIR sensors. Castro et al.94 recently reported the use of FTIR to characterize low-temperature combustion gases in biomass fuels. The spectrometer, operating in absorption mode, was placed in front of an IR source and mounted such

410 Chemical Reviews, 2008, Vol. 108, No. 2

that the optical line of sight was just above the sample, allowing the absorption of the emitted gases to be measured. A high acquisition speed was necessary in order to evaluate temporal evolution of gas concentrations and improve the signal-to-noise ratio. A spectral resolution of 1 cm-1 was selected as a compromise between required spectral resolution and acquisition time. The concentration of three gases, CO2, CO, and CH4, which are the most important carbonrelated products of biomass combustion (shrub species in this case), was measured in situ. These results can contribute to improved modeling of pyrolysis for predicting forest fire behavior. FTIR has also been used for analysis of vehicular emissions; for example, Reyes et al.95 monitored a range of pollutants including CO, NO, SO2, and NH3 under different driving conditions using a hybrid car. However, no detailed performance parameters were given. FTIR imaging has been used for high-throughput analysis of processes such as epoxy curing,96 gaseous or solid-phase reactions,97 and monitoring of pharmaceutical formulations under different conditions.98 Chan et al.98 applied FTIR imaging to the study of the behavior of drug formulations such as ibuprofen as a function of different environmental parameters. They combined a microdrop device with an FTIR imaging system, which consisted of a step scan spectrometer coupled to a macrochamber extension and a 64 × 64 focal plane array (FPA) detector. Spectra were measured with a 16 cm-1 resolution and a spectral range of 3950-900 cm-1. They demonstrated for the first time the simultaneous analysis of 100 samples. The system allows fast screening of many formulations under controlled environments. FTIR spectroscopy has been used for on-line monitoring of formaldehyde-based resin synthesis.99 The system used a FTNIR spectrometer fiber optic data acquisition, and spectra were collected at 8 cm-1 optical resolution. Chemometrics was employed to provide fast and reliable process information on an industrial scale. No sample manipulation was required, and data was acquired in less than a minute. Diode Laser Sensing Systems. While FTIR is well suited to multicomponent analysis of gases and other chemical species for a range of processes, laser spectroscopy is the method of choice for trace gas analysis because of its higher sensitivity and specificity, which arises when a spectrally narrow laser source probes a narrow absorption feature of the analyte. Werle100 and Allen101 produced comprehensive reviews of diode laser-based gas sensor systems for applications such as automated control of industrial processes, environmental monitoring, and combustion process monitoring. The lasers used for gas analysis span the near- and midIR spectral regions with wavelengths ranging from 1 to >2µm for conventional diode lasers and from 2 to >4µm for recently developed QCL lasers. Fiber-coupled multiwavelength diode laser sensor systems have been designed which use modulation spectroscopy and multipass absorption cells to increase sensitivity. Wavelength modulation spectroscopy has been used to monitor gas emission in industrial processes.102 Tunable diode laser absorption spectroscopy (TDLAS) systems, using lead salt lasers which are based on IV-VI semiconductors, have been used to achieve parts per billion detection levels of atmospheric gases, for example, methane.103 However, these systems require laser cooling and are not suited to unattended routine industrial applications. While lead salt lasers operated at cryogenic temperatures cover the fundamental gas absorp-

McDonagh et al.

tion bands required for ultrasensitive gas analysis, nearinfrared diode lasers, operated at room temperature, probe mainly the weaker overtone and combination bands of the analyte molecule. Therefore, there is a trade off between sensitivity and convenience of operation. Sensitivity problems can be overcome using appropriate signal processing and double-modulation techniques.104 QCL lasers are semiconductor lasers based on transitions in a multiple quantum well heterostructure. The emission wavelength depends mainly on the quantum well thickness rather than on the size of the band gap, as is the case with conventional diode lasers. QCLs operate at wavelengths in the MIR starting at about 3 µm, which match well with the fundamental vibrational absorption bands of many gases and other chemical species in comparison with conventional diode sources where the laser emission generally matches the weaker overtone bands. QCLs operate at near room temperature, produce milliwatts of radiation, and offer the possibility of tailoring the emission wavelength within a broad range of frequencies.105,106 The most technologically developed QCLs are based on InGaAs-InAlAs and GaAs-AlGaAs heterostructures.107 They are usually fabricated as single-mode lasers and often in a distributed feedback (DFB) configuration. QC-DFB lasers are of particular interest for gas sensing as they produce highly monochromatic radiation that is suitable for highresolution spectroscopy.108 The first reports of gas sensing using a CW QC-DFB laser was by Sharpe et al.,109 where they presented absorption data on NO and NH3 at low pressure. The first QCL-based system for determination of CO2 in aqueous solution was reported by Schaden et al.110 A QCL gas sensor system for measuring trace gases in air was reported by Kosterev.111 Gases such as CH4 and N2O were detected down to ppb levels. A QCDFB laser system was used to measure ppb levels of NO in vehicle emissions,112 while high-precision measurements of atmospheric N2O and CH4 were reported by Nelson et al.113 Sensing in liquids has also been reported. A Fabry-Perot QC laser system, consisting of two lasers, one tuned to a water vibration and the second to a water window, was used in a flow injection system equipped with a fiber optic flow cell with an adjustable path length. Adenine and xanthosine were used as test analytes, and the results demonstrated the advantages of QCL lasers in overcoming solvent interference for liquid sensing.114 UV Absorption. Direct UV absorption sensing has been used mainly in environmental applications to monitor pollutants such as heavy metals, hydrocarbons, and volatile organic compounds (VOC) in air and water. A sensor to detect Cr(VI) in water that is based on a flexible fused silica capillary and intrinsic evanescent-wave UV absorption of Cr ions in a water sample in the capillary has been reported.115 A Cr(VI) absorption band peaking at 374 nm was monitored, and a limit of detection of 31 parts in 109 was achieved using an absorption path length of 30 m of capillary. The concentrations of ozone and NO2 in the atmosphere were measured by UV-differential optical absorption spectroscopy (UV-DOAS), and the results were used to evaluate the contribution of NO2 in limiting ozone formation.116 The UV-DOAS system used a broad-band xenon source that was placed at a level of 12 m on top of a building, and the detector was placed 150 m away. A premeasured clean-gas reference spectrum was used to generate the differential absorption spectrum. UV-DOAS was

Optical Chemical Sensors

also used to monitor VOCs and aromatic hydrocarbons in ambient air in the vicinity of a refinery.117 The system detected high levels of benzene and toluene and also monitored seasonal variations in VOC concentrations. UVDOAS was one of four VOC monitoring techniques used in a field study in Mexico to identify VOCs that were primarily due to vehicle emissions.118 The data were analyzed to understand the spatial distributions, diurnal patterns, and reactivity of various VOCs. One of the key results of this study was a better understanding of the interaction between vehicular activity and meteorological processes.

3.2.2. Direct Fluorescence Sensing Direct fluorescence sensing is widely used in biomedical applications. Autofluorescence spectroscopy is a useful tool for noninvasive detection of precancerous development of the epithelium, where most human cancers originate. This fluorescence arises mainly from two intrinsic fluorophores, reduced nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FAD). The fluorescence of these species can be used to monitor cell activity. For example, the ratio of free and bound NADH, differentiated by their different lifetimes, is a good indicator of the cell metabolic state. Wu et al.119 used a time-resolved confocal fluorescence spectroscopy system to study cell metabolism and were able to differentiate between cancerous and normal cells by analyzing the dual-exponential decay of NADH. Depthresolved fluorescence spectroscopy, using a confocal microscope and multiple excitation wavelengths, was used to study different layers of the epithelium.120 By exciting at 355 and 405 nm, the system was able to isolate the fluorescence from different sublayers. In particular, the ratio of NADH to FAD fluorescence can be used to indicate the presence of precancerous cells. Autofluorescence patterns have also been used to follow changes in cervical tissue121,122 and oral tissue.123 The intrinsic fluorescence of tryptophan has been used to study protein folding and unfolding. Tryptophan fluorescence is very sensitive to the environment, whereby the peak emission wavelength shifts from 308 to 350 nm for the range of protein conformational changes.124 This spectral behavior has been used to study protein stability125 and protein aggregation.126 Fluorescence can be used to characterize different oils as the fluorescence is dependent on chemical composition. For example, crude petroleum oils have been analyzed using synchronous fluorescence, where both the excitation and emission monochromators are scanned simultaneously, and lifetime measurements were carried out.127,128 Recently, biocrude oils, derived from seeds, have been analyzed.129 Synchronous fluorescence has also been used to quantitatively discriminate between virgin olive oil and sunflower oil,130 and fluorescence spectroscopy combined with artificial neural networks has been used to classify a range of edible oils.131

3.2.3. Raman and SERS Sensing Like IR-based absorption spectroscopy, Raman spectroscopy probes the vibrational energy levels of molecules and therefore can be used as a highly selective technique which will distinguish between similar molecules. The Raman effect occurs when a photon interacts with the vibrational energy levels of a molecule and is scattered.132 Raman spectroscopy normally requires laser excitation and relies on good optical filtering to separate the scattered photon from the intense

Chemical Reviews, 2008, Vol. 108, No. 2 411

incident beam. Unfortunately, Raman scattering crosssections are very small, many orders of magnitude smaller than fluorescence cross-sections. However, it is still a useful technique for sensing of gaseous and liquid analytes. It has a particular advantage over IR absorption for liquid sensing in that there is no interference from the vibrational spectrum of water. For this reason, Raman is a useful technique for life science applications, including biomedical diagnostics, for example, the study of living cells. Lasers with large spot sizes (>10 µm) can be used to study whole cells and tissues. Raman imaging using diffraction-limited spot-size lasers can be used to map the chemical distribution in a cell.133 Uzunbajakava et al.134 used a laser with high lateral resolution to map the DNA and RNA distribution in GeLa cells. Principal components analysis of Raman spectra has been used to distinguish between different cancer cells.135,136 A Raman imaging system, based on fiber optic probes, has been used to image breast tissue.137 Different tissue features were identified from differences in the Raman spectra. Toxins such as sulfur mustard and ricin have been identified in cells with an identification accuracy of 88.6% and 71.4%, respectively, using Raman spectroscopy.138 The disadvantages of conventional Raman, such as the low scattering cross-sections, can be overcome to a large extent using the SERS technique. The strength of Raman scattering from a molecule in close proximity to a nanostructured metal surface can be enhanced by as much as a factor of 108, compared to conventional Raman. This surface-enhanced effect is strongest for silver and due to the interaction of the localized surface plasmon, generated at the metal surface, with the vibrational levels of the molecule. The large electromagnetic field generated induces a dipole in the molecule that is adsorbed on the metal surface, hence producing the hugely enhanced Raman signal.139 The two principal conditions required for SERS to be observed are the presence of a suitable SERS-active nanostructured metal surface and the requirement that the sample under investigation be immobilized on or in close proximity to the surface. In the past, routine application of the technique has been hampered by a poor understanding of the theoretical background and lack of reproducibility of SERS substrate materials. Recently, with advances in materials fabrication and better understanding of the details of the plasmonic interaction, SERS is being increasingly used in diverse fields such as biomedicine and environmental analysis. A number of review articles have been published recently with particular emphasis on biomedical applications of SERS.140-142 The use of SERS to obtain quantitative in-vivo glucose measurements has been reported,143 where a suitable SERS surface was achieved using a silver-coated self-assembled monolayer (SAM) of polystyrene nanospheres. The SERS probe was implanted in a rat, and the data obtained agreed well with simultaneous data monitored using a conventional electrochemical glucose sensor. The same substrate has also been used to detect the presence of the chemical warfare agent half-mustard.139 A combination of near-field scanning optical microscopy and SERS has been used to detect dyelabeled DNA with 100 nm resolution.144 SERS has been used to detect specific dyes in works of art. Anthraquinone145 and alizarin dyes have been identified.146 For the alizarin investigation, silver nanoparticles were deposited directly onto a sample of the painting. The advantage of SERS over conventional Raman was highlighted in this work as strong fluorescence of the dye completely masked the Raman signal,

412 Chemical Reviews, 2008, Vol. 108, No. 2

while in the SERS measurement, the fluorescence was quenched due to the presence of the metal substrate, enabling detection of the enhanced Raman signal. In environmental sensing, a SERS optode has been used to detect polycyclic aromatic hydrocarbons (PAHs) in seawater.147

3.3. Summary This section on direct sensing has focused exclusively on spectroscopic techniques. A short review of the range of IR absorption techniques and configurations used in gas analysis and sensing of organic compounds was given. NDIR- and FTIR-based sensing are both used extensively, FTIR being more appropriate for high-throughput, multianalyte applications. Laser-based systems are more complex and expensive but offer very high sensitivity, allowing detection down to ppb levels. The availability of QCL sources has been a novel development in this area and enables relatively compact, robust systems for use in the field. The recent growth of SERS-based sensing was discussed, and examples of diverse applications, including some biosensing applications, were given. In summary, it has to be said that, currently, the most widely used direct optical sensing technique is that of IR absorption-based sensing, mainly for gas analysis. For example, of the total commercial gas analyzer market in the United States, NDIR sensors had about a 50% share in 2005. It will be interesting to observe whether this technique still dominates the optical sensing market over the next decade. It will also be interesting to see if the current growth in SERS sensing will be sustained and whether it will lead to commercial sensor systems.

4. Reagent-Mediated Sensors 4.1. Introduction When an analyte does not exhibit a convenient spectroscopic optical response such as absorption or luminescence, sensing can be achieved by monitoring the optical response of an intermediate species or reagent, whose response is modulated in some way by the presence of the analyte. Reagent-mediated sensing is illustrated very well by the principle of optical oxygen sensing, whereby the luminescence intensity or decay time of an oxygen-sensitive luminescent complex, for example, a ruthenium polypyridyl or a porphyrin complex, is quenched in the presence of oxygen. This enables the oxygen partial pressure to be measured as a function of the intensity or luminescence decay time of the complex. Indirect colorimetric sensors also require an intermediate reagent, for example, many optical pH sensors have been based on monitoring the change in optical absorption of pH indicators such as bromocresol purple. This indirect sensing technique requires the reagent to be immobilized, either in the liquid or in the solid phase, to facilitate interaction with the analyte. In recent years, reagentbased optical sensing has been based mainly on solid-phase immobilization matrices, where the reagent dye can be adsorbed, covalently or ionically attached, or simply encapsulated in a solid matrix that is permeable to the analyte. If the immobilization matrix has the capability of being coated on a substrate in liquid form, as is the case for sol-gel glasses or polymer coatings, a wide range of sensor configurations is enabled including, for example, fiber optic, planar waveguide and array-based sensors. These configura-

McDonagh et al.

tions have already been described in section 2. Section 4.2 of this review will report on the wide range of reagents used in recent years in optical absorption- and luminescence-based chemical sensors, while section 4.3 describes two widely used sensor immobilization matrices, namely, sol-gel and polymer materials. Subsequent sections review the state-ofthe-art in absorption- and luminescence-based systems, dealing mainly with work that has been published in this decade.

4.2. Reagents 4.2.1. Reagents for Colorimetric Sensing Indirect colorimetric pH sensing uses organic pH indicators, the absorbance of which is modified by the pH of the environment. The pKa of these indicators indicates the center of the measurable pH range, for example, cresol red, bromophenol blue, and bromocresol purple respond to acidic pH (pH < 7), while cresol red, naphtholbenzene, and phenolphthalein respond at basic pH (pH > 7).148 The relatively narrow pH response range of most of the above dyes has been addressed whereby several indicators with different pKa values have been combined in one sensor in order to produce a linear pH response over a wide range.149 Makedonski et al.150 synthesized new reactive azo dyes for use in pH sensing where the dyes were covalently bound to a polymer matrix for increased sensor stability. Recently, phenolphthalein has been used to monitor the behavior of corrosion-resistant polymers in alkaline solutions in chemical plants.151 Since colorimetric CO2 sensing is normally achieved by measuring the pH change of an indicator in response to carbonic acid generated by the acidic CO2 gas, many of the same reagents that are used for pH sensing are also used in absorption-based CO2 sensing. Thymol blue immobilized in a sol-gel matrix has been used for gaseous CO2 sensing,152 while bromothymol blue in an ionic liquid matrix has been used for both gas-phase and dissolved CO2 sensing.153 Cresol red has been used in an optical fiber configuration in order to measure in vivo gastric CO2.154 Recent reviews by Mohr155,156 reported the development of new indicator dyes for neutral and ionic analytes. Many of these use reversible covalent bond formation to detect analytes such as amines, cyanide, formaldehyde, nitrites, and peptides. Selective colorimetric optical probes based on pyrylium dyes have been developed recently for detection of aliphatic amines.157,158 Novel, highly selective probes for formaldehyde detection have been developed by Suzuki et al.,159 while a formaldehyde sensor, based on the reagent 4-amino dydrazine-5mercapto-1,2,4-triazole, has been reported recently.160 A highly selective colorimetric Hg2+ sensor, which uses a squaraine reporter dye, has been reported.161 This highly toxic heavy-metal ion has also been detected optically by immobilizing dithizone on a triacetylcellulose membrane.162 Squaraine derivatives have also been used for colorimetric detection of cyanide ions.163 A fluoride-selective sensor based on the complexation of fluoride ion with an aluminum octaethylporphyrin ionophore dye has been reported.164 The sensor exploits the fact that aluminum(III) forms very stable complexes with fluoride in aqueous solutions. A breath sensor for acetone measurement, which detects the colored product that is formed when acetone reacts with alkaline salicylaldehyde, has been reported by Teshima et al.165

Optical Chemical Sensors

4.2.2. Reagents for Luminescence Sensing Analytes such as pH, CO2, ammonia, O2, and many cations and anions can be measured using luminescent probes. Luminescence is intrinsically more sensitive than absorption as a sensing technique, so, for many applications, luminescence-based sensing offers higher sensitivity than absorption sensors. As in the case of absorption-based sensors, the principle of luminescence measurement of pH, CO2, and ammonia relies on the change in luminescence of a pH indicator. For these analytes, the literature is dominated by two pHdependent luminescent probes, namely, fluorescein and 8-hydroxypyrene-1,3,6-trisulfonic acid (HPTS). Both dyes absorb at convenient LED-emission wavelengths in the blue, while emission occurs just above 500 nm. Fluorescein has been used recently in an optical fiber imaging sensor based on drop-on-demand inkjet printing.166 Fluorescein-related dyes have been used in order to facilitate covalent immobilization to the sensor matrix to eliminate dye leaching. For example, pH sensing has been realized using a fluoresceinamine isomer II complex that was covalently bound to a sol-gel matrix.167 Ratiometric sensing is a self-referencing technique where either an analyte-insensitive excitation or a luminescence band of the dye is ratioed with the analytedependent band. Excitation ratiometric sensing has been achieved using methacryloyl-modified HPTS that facilitates covalent bonding to a polymer matrix and has a sensing range of pH 6-9.168 Ratiometric pH sensing has also been demonstrated using mercurochrome, another fluorescein-related dye, in a solgel matrix.169 A newly synthesized boron-dipyrromethene derivative has been used in a single-excitation dual-emission ratiometric scheme to measure pH in human gastric juices.170 Naphthalimides have also been investigated as luminescent probes,171,172 while a pH sensor using covalently immobilized piperazinyl-1,8-napthalimide has been published recently.173 Luminescent transition-metal complexes have larger Stokes shifts and longer lifetimes than, for example, fluorescein or HPTS. This allows more flexibility in excitation/detection optoelectronics as well as enabling lifetime-based sensing schemes. pH sensing has been demonstrated recently using a range of ruthenium and rhenium complexes based on the pH-sensitive ligand 5-carboxyl-1,10-phenanthroline.174 In this work, the complexes have been tailored for maximum dynamic range. Aminofluorescein has been used for optical ammonia sensing where an enhancement has been achieved compared to fluorescein-based sensing due to the reaction of the ammonia with the amine group on the fluorescent dye.175 Ruthenium(II) tris(4,7-dephenyl-1,10-phenthroline) (Ru(dpp)3) has been ion paired with the pH indicator bromophenol blue.176 On exposure to ammonia, the increase in the deprotonated band of the pH indicator gives rise to energy transfer whereby the luminescence and lifetime of the Ru(dpp)3 complex decreases. In a novel, multianalyte approach, Nivens et al.177 used HPTS to detect pH, CO2, and NH3 by designing multilayer sol-gel structures whereby, for CO2 and NH3 sensing, a hydrophobic outer layer is used to eliminate cross-reactivity to pH. HPTS has been widely used for CO2 sensing, where, as above, the matrix has been tailored to prevent ingress of protons. HPTS is stable and water soluble, and the pKa is ideal for CO2 detection.178 Lifetime-based sensing based on HPTS in polymer matrices has been reported.179,180 A HPTS-

Chemical Reviews, 2008, Vol. 108, No. 2 413

based ratiometric sensor has been published recently for monitoring CO2 in marine sediments.181 The luminophores used for optical oxygen sensing have been based mainly on organometallic complexes of ruthenium, in particular ruthenium polypyridyl complexes, and also metalloporphyrin complexes. However, a number of papers have reported on a range of polycyclic hydrocarbons (PAHs) that are efficiently quenched by oxygen.182-184 Ruthenium polypyridyl complexes are characterized by high quantum efficiency and convenient absorption and emission maxima located in the visible spectrum with a large Stokes shift and relatively long lifetime (∼1 µs). These complexes are efficiently quenched in the presence of oxygen, giving rise to a decrease in both luminescence intensity and lifetime. The complex Ru(dpp)3, referred to above in the context of pH sensing, has one of the longest unquenched lifetimes (∼5 µs) and been widely reported for oxygen-sensing applications, both in polymer and sol-gel matrices.185-187 Porphyrin complexes have longer lifetimes (∼100 µs), emit at the red end of the visible spectrum, and are generally considered to have inferior photostability properties compared to ruthenium complexes. However, fluorinated porphyrins have been shown to be highly stable against photo-oxidation and photoreduction.188 The most widely reported complexes for oxygen-sensing applications are the platinum and palladium octaethylporphyrin (PtOEP and PdOEP).189,190 Platinum and palladium tetrakis(pentafluorophenyl)porphyrin (PtTFPP and PdTFPP) have also been reported.191 Ruthenium complexes have also been used for luminescence-based relative humidity sensing based on the quenching of a phenazine ligand by protons.192-194 There are extensive literature reports on the use of luminescent probes to measure anionic and cationic species as well as organic chemicals. Some of this work has been referenced in the recent review by Wolfbeis.5 Probes used for chloride sensing include the chloride-quenchable lucigenin dye14 and the halide-selective ionophore9 mercuracarborand-3.195 Aluminum(III) concentration has been measured using the dye 1,4-dihydroxyanthraquinone (quinizarin). Sensing is based on the highly fluorescent complex which is formed when Al3+ complexes with quinizarin.196 Molecularly printed polymers (MIPs) have also been used where Al3+ is used as the template and 8-hydroxy-quinoline sulfonic acid is the luminescent tag which complexes to the Al3+ ion.197 Badagu et al.198 developed a range of cyanide-sensitive probes that are based on the binding of the cyanide ion to a boronic acid functional group. Both intensity and lifetimebased sensing have been demonstrated. Luminescence-based nitrate sensing has been achieved using the cationic potentialsensitive dye 4-(4-(dihexadecylamino)styryl-N-methylpyridinium iodide), which acts as an anion-exchange catalyst that extracts nitrate from aqueous solution to form a luminescent complex.199

4.2.3. Summary Clearly, a comprehensive account of the large number and variety of absorption and luminescence probes reported in recent years is beyond the scope of this general review of optical chemical sensors. The most widely used reagents in areas such as pH and oxygen sensing have been highlighted. In the case of ion sensing, a selection of recently published work has been cited, which deals with some environmentally topical analytes such as chloride and aluminum ions.

414 Chemical Reviews, 2008, Vol. 108, No. 2

4.3. Immobilization Matrices 4.3.1. Introduction In most reagent-based optical sensors, the reagent is immobilized in a solid matrix usually in the form of a monolith or a thin film. The matrix serves to encapsulate the reagent such that it is accessible to the analyte while being impervious to leaching effects. In this section, two commonly used immobilization matrices will be described, namely, sol-gel and polymer materials, and a selection of work published in recent years will be reported for sol-gel and polymer-based absorption and luminescence sensors. Since the immobilization matrix very often influences the sensor response, a brief discussion of reagent-matrix interactions will be given.

4.3.2. Sol−Gel Matrices The sol-gel process provides a relatively benign support matrix for the immobilization of analyte-sensitive reagents and dyes. The basic process involves the hydrolysis and polycondensation of the appropriate metal alkoxide solution to produce a porous glass matrix in which the reagent is encapsulated in a nanometer-scale cage-like structure and into which the analyte molecules can diffuse. The versatility of the process facilitates tailoring of the physicochemical properties of the material in order to optimize sensor performance. For example, key sol-gel process parameters such as sol pH, precursor type and concentration, water content, and curing temperature can be adjusted to produce materials of desired porosity and polarity.200 This tailorability was demonstrated for oxygen-sensitive films doped with the Ru(dpp)3 complex and were fabricated using different ratios of tetraethoxyorthosilane (TEOS) and methyltriethoxysilane (MTEOS) precursors.201 In a later study, the effect of the alkyl (R) chain length on the oxygen sensitivity was measured.202 Monoalkoxysilanes of the form (CnH2n+1)-Si-(OR)3, where n ranges from 1 to 12, have been investigated and exhibit a linear oxygen response which increases up to C8 and then begins to decrease. The decrease in response at longer alkyl chain lengths was correlated with a decrease in oxygen diffusion coefficient in the matrix for these films. Very high oxygen sensitivities have been achieved using fluorinated sol-gel precursors. For example, oxygen sensitivities have been achieved using precursors such as 3,3,3-trifluoropropyl-trimethoxysilane (TFP-TMOS), which are ∼10 times greater than that achieved for MTEOSbased sensors.203,204 This high sensitivity is attributed to the highly hydrophobic and nonpolar nature of the fluoro films. These materials also exhibit very high stability, with stabilities of up to 2 years being achieved for some formulations. Novel phenyl-substituted ORMOSILs were used as matrices for oxygen sensors based on both the Ru(dpp)3 complex and a platinum(II)-octaethylporphyrin complex. A highly stable and sensitive oxygen response was obtained, and the films were sterilizable.205 As an alternative to physical entrapment, the analytesensitive dye can also be covalently bound to the sol-gel matrix. This can increase stability and eliminate dye-leaching effects in aqueous environments. Some examples have already been referred to in section 4.2.167,168 A triethoxysilane-modified ruthenium tris(bipyridyl) (Ru(bpy)32+) complex has been anchored to a range of polysiloxane matrices,206 and an oxygen sensor based on the same complex has been recently reported where the complex was covalently grafted

McDonagh et al.

to 3-aminopropyltriethoxysilane.207 The sensor exhibited no leaching and was used for dissolved oxygen measurement. Sol-gel matrices have also been used for other analytes such as in pH and CO2 sensing. As discussed in section 4.2, for pH sensing a hydrophilic matrix which allows ingress of protons is required, whereas for CO2 sensing a more complex matrix is required which retains the water required for conversion of the gas to carbonic acid in the films but has sufficient hydrophobicity to be pH insensitive. These conditions have been achieved for both sol-gel and polymer matrices using an ion-pairing approach. This will be described in more detail in the next section as the majority of the literature on optical CO2 sensors is based on polymer matrices. However, sol-gel-based optical CO2 sensors for both gas and dissolved phase have been reported.208,209,43 Standard sol-gel materials have a nonordered, amorphous structure where diffusion of analytes can be limited by the random microporosity of the structure. Mesostructured porous films, on the other hand, have large open porosity, which can offer enhanced diffusion and accessibility for analytes. Mesoporous sol-gel sensor films have been realized via an evaporation-induced self-assembly (EISA) approach.210,211 These films have a highly structured mesoporosity and excellent optical quality. The films are based on the surfactant cetyltrimethylammonium bromide (CTAB). The CTAB/Si and probe/Si ratio is optimized to give the required structure. A sensor to detect metallic cations was reported which was based on the chelating properties of dibenzoylmethane.211 An oxygen sensor based on a mesostructured sol-gel silica matrix was also reported.212 A ruthenium complex was grafted to the mesostructured network that was based on the precursor 3-(triethoxysilyl)propyl isocyanate and CTAB as surfactant. It was established that the stability, homogeneity, and sensitivity of the matrix was superior to a nonmesoporous matrix where the dye was just physically entrapped. A novel sol-gel-based sensor application has been reported whereby an optical fiber core is fabricated from reagent-doped porous sol-gel. Hence, the core acts as an active fiber core optical sensor (AFCOS). The fabrication and optical properties of the fiber were investigated, and a humidity sensor was constructed as proof of principle.213 Oxygen sensing has been reported using calcined mesoporous silica spheres which have been postdoped with the Ru(bpy)32+ complex.214 The oxygen sensitivity was dependent on the pore morphology in the sphere, and the response times (2-4 min) were longer than for thin films and most likely related to the relatively large sphere diameter of 0.1-0.2 mm.

4.3.3. Polymer Matrices Polymers have been widely used as support materials for a broad range of optical sensors. They have many desirable features and compare well with sol-gel matrices for most applications. While polymers may not be as photochemically stable or printable as sol-gels, some polymers are more suitable than sol-gels for high-temperature applications such as autoclavation. The most widely used materials include polystyrene (PS), polyvinyl chloride (PVC), polymethyl methacrylate (PMMA), polydimethyl siloxanes (PDMS), and polytetrafluoroethylenes (PTFE) and cellulose derivatives such as ethyl cellulose. As discussed in the previous section, hydrophobic matrices, for example, PMMA and PDMS, are selected as the optimum matrix for optical oxygen sensing,

Optical Chemical Sensors

while more hydrophilic matrices such as ethyl cellulose have been widely used for pH sensing. Much has been published on polymer-based oxygen sensors. Mills215 carried out a comprehensive investigation of the effect of various polymers and plasticizers on the oxygen sensitivity for mainly ruthenium-complex-doped materials. It was clear from this work that the value of the oxygen diffusion coefficient in the material was the main contributing factor in the oxygen sensitivity. Of all materials studied, PDMS yielded the highest diffusion coefficient, which is associated with the opening and closing of void volumes in the polymer. It is to be noted that similarly high oxygen sensitivities were achieved using fluoro-sol-gel materials by Bukowski.204 This has been attributed to the high O2 diffusivity in fluorinated materials.216 Polymer-based oxygen optrodes have been reported which withstand autoclavation. Polysulfone and polyetherimide materials were found to withstand steam sterilization at 135 °C.217 It is thought that these polymers have a different water uptake compared with other materials such as polyvinylnaphthalene, which are destroyed under these conditions. The dissolved oxygen sensors based on these polymers, once calibrated, do not need recalibration in between sterilization cycles. Polysulfone was also used successfully to monitor cell viability by measuring the oxygen consumption.218 A novel oxygen sensor membrane has been reported by Holmes-Smith et al.219 where an electropolymerized diphenyl-di(4-aminophenyl)porphyrin (Pt(II)DAPP) has been characterized and tested as an optical oxygen sensor. The sensitivity obtained was comparable to other Pt-porphyrin systems reported in the literature, but the main advantages of this sensor were the versatility of deposition via the electropolymerization technique as well as the lack of leaching as the porphyrin is part of the polymer. Optical CO2 sensors are generally based on the Severinghaus principle, which relies on conversion of the gas to carbonic acid. In the presence of bicarbonate, it has been shown that the partial pressure of CO2 is directly related to the pH change of a colorimetric or luminescent pH indicator.178 In solid optical CO2 sensors, a phase-transfer agent such as tetraoctylammonium hydroxide (TOAOH), a quaternary ammonium hydroxide, is used to solubilize the pH indicator into a hydrophobic polymer which acts as a proton barrier and reduces cross-sensitivity to pH. This gives rise to formation of an ion pair which facilitates encapsulation of the dye in the gas-permeable polymer. The quaternary ammonium hydroxide is normally associated with sufficient water molecules to facilitate formation of carbonic acid. A fiber optic microsensor based on ion-paired HPTS in an ethyl cellulose matrix has been reported for highresolution pCO2 monitoring in the marine environment.220 A gas-permeable but ion-impermeable coating was used as a protective layer to eliminate interferences such as chloride and pH, and the limit of detection achieved was 60 ppb. A highly stable, autoclavable sensor, based on HPTS ion paired to cetyltrimethylamonium hydroxide (CTMAOH) in a twocomponent silicone film, has been investigated for use in biofermentation.221 PDMS and vinyl-terminated dimethylsiloxane copolymers were used to make the sensor film using a three-step process. The sensor used ratiometric detection, was stable for several months and sterilizable, and had an LOD of 0.03% CO2. It has been observed that CO2 sensors based on a quaternary ammonium hydroxide base can have a limited shelf life in ambient air due to the ingress of other acidic

Chemical Reviews, 2008, Vol. 108, No. 2 415

gases such as NOx and SO2, which cause irreversible protonation of the indicator. It has been suggested that this is due to a process called Hofmann degradation.222 The feasibility of replacing the quaternary base with a neutral phosphazene base which would not be susceptible to degradation has been investigated by Schroder et al.223 Ethyl cellulose was the polymer used, and p-xylenol blue was the colorimetric pH indicator. The CO2 response of the new phosphazene base system was generally comparable to that of the quaternary base system. However, the new base could only be used in dissolved phase due to its volatility, and the sensor film exhibited cross-sensitivity to relative humidity. Molecularly imprinted polymers (MIPs) are being increasingly used as sensor supports. MIPs are synthesized by copolymerizing functional and cross-linking monomers in the presence of a target analyte molecule which acts as a template or imprint. When the template molecule is removed, the resulting cavity has a “memory” of the target analyte which imparts a high degree of selectivity to the MIP material.224 A polyurethane-based MIP sensor imprinted with anthracene was reported where a theoretical model of the sensor response was developed in order to provide an optimization strategy for sensor design.225 MIP sensors for detection of PAHs have been synthesized where the selectivity was tuned using different polystyrenes. Selectivity was tuned down to differences in size as little as one methyl group.226 Because of their selectivity, MIPs are now being developed for biosensor applications where antibodies and enzymes are used as templates. Reviews dealing with MIPs for biosensor and other applications include those by AlKindy et al.227 and Kindschy et al.228 Apart from their high degree of selectivity, MIP sensor materials are generally very robust, stable, and resistant to interferences such as pH and humidity.225

4.3.4. Interaction of Reagent and Support Matrix In order to be efficiently encapsulated in the support matrix, the reagent must be soluble in the support material. To enhance solubility, an ion-pairing approach can be used as discussed above for CO2-based sensing. In the case of sol-gel matrices, for example, in order to encapsulate oxygen-sensitive ruthenium complexes, the counterion and sol-gel solvent can be chosen to facilitate homogeneous distribution of the dye in the matrix. In the case of optical oxygen sensors, the linearity of the Stern-Volmer (SV) sensitivity plot is strongly dependent on the matrix. For most sol-gel and polymer materials, a downward curved SV plot is obtained due to the inhomogeneous nature of the amorphous matrix where each dye molecule experiences a slightly different microenvironment. However, linear SV plots have been obtained, for example, where the indicator dye is soluble in the matrix and homogeneously entrapped in a defined polymer microdomain.229 Linear SV plots have also been obtained for sol-gel materials which use precursors with long alkyl chain groups.202,204 The optical properties of some reagents are very sensitive to the environment; thus, often these reagents will experience spectroscopic shifts when encapsulated in a solid matrix. Of all the oxygen-sensitive ruthenium complexes, the spectroscopic properties of Ru(dpp)3 are relatively insensitive to the sensor matrix.230 In fact, these complexes exhibit enhanced fluorescence and longer decay times in solid matrices compared to in solution due to the reduction in nonradiative decay pathways. A substantial improvement in

416 Chemical Reviews, 2008, Vol. 108, No. 2

selectivity and sensitivity has been observed in the case of the metal-ion indicator pyrocatechol violet (PV) when encapsulated in a plasticized PVC membrane. A shift in the working wavelength was accompanied by enhanced sensitivity to Cu(II) ions. This is possibly due to a reduction in the complexation ability of the dye and the different permeability selectivity imparted by the PVC matrix compared to solution behavior.231 Many optical sensor reagents photodegrade under conditions of high-intensity illumination, and in many cases the matrix has on influence on the degree of photodegradation. In general, reagents entrapped in rigid supports exhibit increased photostability due to reduced ligand photodegradation compared to a solution environment.232 It has been shown that reagents encapsulated in sol-gel matrices have higher photostability than in polymer matrices. It has been suggested that the more organic polymer environment provides less stability than the more inorganic sol-gel matrix. In the case of oxygen-sensitive dyes, as mentioned in section 4.1, transition-metal complexes, for example, the ruthenium complexes, have higher stability than porphyrin complexes in general, but fluorinated porphyrin complexes exhibit high photostability.182 It has been established that, for oxygen sensors, singlet oxygen, which is a byproduct of the quenching process, causes ligand dissociation in oxygensensitive dyes.233 This degradation process is minimized in fluorinated matrices, either polymer or sol-gel based, which provide highly stable environments due to the high electronegativity of the C-F bond.182,203 In general, the behavior of the reagent can be dependent on the matrix as well as the analyte concentration.

4.4. Recent Developments in Absorption-Based Sensors Reagent-mediated absorption or colorimetric sensing has already been dealt with in section 4.2 on reagents, where numerous examples were given of colorimetric sensing for a wide range of analytes including pH, ammonia, and some ion-sensing applications. This brief section will highlight some recent examples of colorimetric sensors including some novel sensing techniques. A generic sensing platform has been reported which is based on a flexible fused silica capillary.234 The inner surface of a flexible light-guiding capillary was coated with a reagent-doped film, either polymer or sol-gel, to produce a long-path evanescent-wave-based sensor. The authors suggest that the flexibility and robustness of the fused silica capillary and the increased sensitivity provided by the increased length offer advantages over conventional optical fiber-based sensors as well as over previously reported capillary-based sensors. Sensing has been demonstrated for ammonia, toluene, and Cu(II) using this system. Cu(II) was detected down to 2.5 ppb, while ammonia was detected down to 15 ppm in the as-yet unoptimized system. A dynamic technique for measuring pH over a very wide range has been reported, where the limited measurable pH range of the indicator in normal steady state measurements has been overcome using a flow system and optimizing the diffusion of the analyte in the sensor membrane.235 In this work, the pH indicator was immobilized in a triacetyl cellulose membrane. The optical response to pH was measured before equilibrium was reached. At a fixed time before equilibrium, a linear plot is obtained for absorbance versus pH at any flow rate. The time and flow rate are

McDonagh et al.

optimized for maximum linear dynamic range. pH was determined over a range of up to 11 pH units. A combination of a colorimetric and fluorometric reagent has been used to measure CO2 where the luminescence change of a tetraphenylporphyrin dye due to the CO2dependent change in absorbance of a co-immoblized R-naphtholphthalein dye was measured.236 The technique relied on the overlapping absorption and luminescence of the colorimetric and fluorometric reagents, respectively, and the fluorophore which was insensitive to CO2 acted as an internal reference. The sensor operated in both gas and dissolved phase, was immune to photobleaching, and exhibited short response times of <6 s in the gas phase. The same authors used a thymol blue-europium complex combination for gasphase CO2 sensing.237 Mesoporous sensor matrices have already been discussed in section 4.3. A fiber-based colorimetric pH sensor based on bromothymol blue in a mesostructured silica film has been reported.238 The sensor responds over the pH range 2-5. Dye leaching is overcome by tailoring the mesopore dimensions to entrap the dye molecule, while the hydrophilic nature of the pore wall allows adsorption of the dye via hydrogen bonds. A CO2 sensor based on ion-paired bromothymol blue, which has been dissolved in an ionic liquid matrix, has been reported.239 The solubility of CO2 in water- miscible ionic liquids, such as 1-methyl-3-butylimidazolium tetrafluoroborate used in this work, is 10-20 times that observed in conventional solvent and polymer matrices. Sensing of CO2 was demonstrated both in gas and dissolved phase. However, regeneration of the reagent phase using nitrogen was necessary, and pH interference studies have not been reported in this preliminary report. The same authors reported a luminescence-based CO2 sensor based on HPTS dissolved in similar ionic liquid matrices.240

4.5. Recent Advances in Luminescence-Based Sensors 4.5.1. Introduction Reagent-based luminescence sensing can be based on monitoring the intensity of the luminescence, often using a ratiometric approach as mentioned in section 4.2.2, or the reagent lifetime can be monitored via a lifetime-dependent phase measurement for example. Simply monitoring the intensity, in the absence of an analyte-independent reference, can give rise to drift and instability due to light source and detector drift, changes in optical path, and drift due to degradation or leaching of the dye. These problems can be overcome to a large extent by employing ratiometric sensing which, as explained in section 4.2.2, involves taking the ratio between an analyte-dependent luminescence or excitation band and a second band which is analyte independent. Referenced sensing can also be achieved using techniques such as phase fluorometry and dual-luminophore referencing (DLR), which will be explored in more detail in section 4.5.3. This section will be divided into a review of recently reported intensity-based sensing and a discussion of lifetime-based sensing, including some recent examples.

4.5.2. Intensity-Based Sensing Fiber optic oxygen sensors based on two different Pt(II) complexes encapsulated in a fluorinated sol-gel matrix have been reported.241 A composite matrix of n-propyltrimethoxy-

Optical Chemical Sensors

silane and 3,3,3-trifluoropropyltrimethoxysilane was doped with either platinum tetrakis(pentafluorophenyl)porphine (PtTFPP) or platinum octaethylporphine (PtOEP). The higher oxygen sensitivity of metalloporphyrin complexes compared to ruthenium complexes has already been discussed in section 4.3, as also has the stability and enhanced oxygen permeability offered by fluoro sol-gel matrices. Both sensor materials demonstrated very high oxygen sensitivity, linear SV response, and short response times of ∼5 s. The disadvantages of intensity sensing have been addressed by the work of Tang et al.,242 where tailored solgel sensor arrays were employed to yield improved accuracy and precision in O2 detection with the use of artificial neural networks (ANN). Pin-printed arrays, where each element has a different O2 response profile, are produced. The tailored oxygen response is achieved by co-doping the sol-gel with different ratios of two ruthenium complexes which have very different oxygen sensitivities. The ANN was “trained” to identify the CCD images from the array. Overall, a 5-10fold improvement in accuracy and precision compared to a single-element sensor was achieved for quantifying O2 in unknown samples. Transparent nanostructured metal oxide oxygen sensor matrices were fabricated by embedding oxides such as aluminum oxide, silicon oxide, and zirconium oxide in PVA.243 Organometallic complexes of ruthenium or iridium were used as oxygen-sensitive dyes, which were incorporated in the nanostructured matrix. SV constants for these matrices were ∼100 times larger than for the same dyes incorporated in conventional polymer matrices. Furthermore, the materials were sterilizable by autoclavation and γ radiation, exhibited response times of <1 s, and were stable for up to 9 months. The superior oxygen sensor performance was attributed mainly to the nanoporosity and high total pore volume of the nanostructured matrices. The majority of both colorimetric and fluorometric pH indicators reported in the literature and in this review operate in the acidic or neutral pH range. Recently, luminescent pH indicators based on Schiff bases have been developed which respond in the pH range 7-12.244 The indicators chlorophenyliminopropenylaniline (CPIPA) and nitrophenyliminopropenylaniline (NPIPA), with pKa values of 10.30 and 8.80, respectively, have been immobilized in PVC. Photostability was superior to fluorescein for both dyes, and CPIPA in PVC matrix exhibited an enhanced quantum yield and reversible pH response over the basic pH range. Both dyes have convenient absorption and emission bands in the visible spectrum. A Li+ sensor for clinical applications, which is based on a polymer derivative of a novel lithium fluoroionophore, has been reported.245 The fluoroionophore is based on a tetramethyl blocking subunit of a 14-crown-4 as a Li+-selective binding site and 4-methylcoumarin as a luminophore. Ratiometric detection is enabled by the quenching of the main emission band by Li+ and the simultaneous appearance of a blue-shifted shoulder. The sensor exhibits good reversibility and immunity from interference for pH and the major biological interfering cations, K+, Ca2+, and Mg2+. Highly selective K+ sensing has been demonstrated using a fluoroionophore based on aza-18-crown-6 as K+ chelator and the fluorophore BODIPY (from Molecular Probes Inc.). The cavity size of the crown ether is chosen to be selective for K+ and discriminate against Na+ ions. By using novel substitution reactions of the BODIPY complex, the system

Chemical Reviews, 2008, Vol. 108, No. 2 417

was tailored such that the emission band shifts from ∼529 to 504 nm in the presence of K+, which facilitates ratiometric optical detection.246 Advances in array-based biosensing have been summarized in a recent review by Ligler.247 The main focus of this review is development of a portable, multianalyte luminescencebased array biosensor by the Naval Research Laboratories (NRL). The device uses planar waveguide optical coupling and PDMS microfluidics to allow simultaneous detection and quantification of multiple target analytes and multiple samples.248 The array-based biosensor has been used for antibody detection, for example, to detect food borne contaminants such as staphylococcal enterotoxin B and other toxins249 and the toxin deoxynivalenol, which is a common food contaminant. LODs of <1 ng/mL were achieved in each case. The system has also been used to detect bioterrorism agents such as anthrax.250

4.5.3. Lifetime-Based Sensing A primary disadvantage associated with the use of fluorescence intensity detection techniques is the susceptibility of sensors based on such techniques to signal fluctuations caused by changes in excitation source intensity, photobleaching of the fluorescent indicator complex, or changes in the concentration of the indicator within the sensing environment (due, for example, to leaching). A more robust fluorescence detection strategy involves measuring analyteinduced changes in the lifetime of the indicator compound, as this parameter is unaffected by the factors mentioned above. Sensors based on detection of fluorescence lifetime may be broadly divided into those that directly measure the lifetime and those that employ phase measurement techniques to indirectly monitor the lifetime. Systems that employ direct lifetime measurement techniques typically include a pulsed laser source and gated, or delayed, detection with a suitable photodetector, e.g., a photon-counting photomultiplier tube or intensified CCD camera. Such systems have been applied to detection of water in organic solvents,192 pH,251-253 and oxygen204,254 in recent years. However, the cost and complexity of systems capable of direct fluorescence lifetime measurements render them less attractive for development of low-cost, mass-producible systems. This has been an important driver in optical chemical sensor research in recent years, and while some efforts have been made to develop low-cost lifetime measurement systems,255 more effort has been applied to development of phase measurement instrumentation, which is well suited to development of low-cost sensor platforms. Perhaps the most common example of a luminescence-based sensor that utilizes phase measurement techniques is that of a phase fluorometric oxygen sensor.10,39,40,256-260 The concept of phase fluorometry is illustrated in Figure 9. The excitation source is modulated at a particular frequency, and the emitted fluorescence is similarly modulated but phase shifted relative to the excitation signal. Phase measurement electronics are used to measure this phase shift, which is related to the luminescence lifetime by eq 1

tan φ ) 2πfτ

(1)

where φ is the measured phase angle, f is the modulation frequency, and τ is the lifetime of the luminescent complex. This measurement concept can be implemented using lowcost phase detection electronics in conjunction with LED

418 Chemical Reviews, 2008, Vol. 108, No. 2

McDonagh et al.

Figure 9. Principle of phase fluorometry.

sources and photodiode detectors, making it an extremely attractive candidate for development of commercial oxygen sensors by companies such as PreSens Precision Sensing GmbH [www.presens.de], YSI environmental [www.ysi.com], Gas Sensor Solutions [www.gss.ie], Ocean Optics [www.oceanoptics.com], and Interlab [www.interlab.es]. The technique has also recently been applied to development of optical humidity sensors.193,194 An important element in development of a phase fluorometric sensor is the use of an indicator complex that is sufficiently long lived so as to be accessible to the modulation frequency range provided by the low-cost sensor electronics, which typically requires that the lifetime be at least in the single microsecond regime. In the case of phase fluorometric oxygen sensors, indicators based on ruthenium or porphyrin dye complexes adequately fulfill this role, but for analytes such as carbon dioxide, pH, or chloride, the fluorescent indicators employed are not sufficiently long lived to be directly accessible to phase fluorometric detection strategies. In response to this, a number of strategies have been developed with the goal of rendering the intrinsically intensity-based measurements required for detection of such analytes compatible with phase fluorometry instrumentation. This would then allow for development of multianalyte sensor technology requiring only a single instrumentation platform. Fluorescence resonance energy transfer (FRET) has been exploited in the past to bring intensity measurements into the phase domain.208,261 However, the most predominant example of a technique that fulfils this goal is dual-lifetime referencing (DLR), also referred to as dual-luminophore referencing, which was first reported by Huber et al. in 2000262 and has been applied to detection of analytes such as CO2,195, 20943, 263 pH,264,265 nitrate,266 copper(II) ions,267 and chloride.199 As the name suggests, the DLR technique makes use of two luminescent indicators that are typically coimmobilized within a solid matrix. One of the indicators is analyte sensitive and possesses a short luminescence lifetime; the other is analyte insensitive and long lived. Changes in the fluorescence intensity ratio of the two compounds are reflected as changes in the measured phase angle (see Figure 10), thereby facilitating the use of phase detection instrumentation in the realization of any sensor that exploits this technique. Additionally, Borisov et al. recently reported the use of a modified DLR (m-DLR) technique that facilitates the simultaneous detection of two analytes.268 A related technique is that of gated phase-modulation (GPM) fluorometry,269 which is similar to DLR in that it makes use of two

Figure 10. Principle of dual-lifetime referencing (DLR). Changes in the amplitude of the indicator signal that are caused by changes in analyte concentration result in a modification of the measured phase angle, φm. This is shown above for two different indicator signal amplitudes.

luminophores with long and short lifetimes, but unlike DLR, square-wave modulation is employed in conjunction with gated detection in order to develop sensors for analytes such as pH.

5. Key Trends and Future Perspectives While the basic sensor transduction principles employed in optical chemical sensors are now well established and remain largely unchanged over the years, there have been significant recent advances in both the design of sensor platforms and strategies for performance enhancement. These developments, which are likely to drive future trends in the research and development of optical chemical sensors, include advances in nanomaterials, microfluidics, wireless networks, as well as optical materials and components. As in many other areas of scientific research, many developments in optical chemical sensors (OCS) are to be found at the micro-nano interface. The massive investment in nanotechnology across the world is leading to new materials and structures, the unique properties of which can have significant implications for the design and implementation of OCS. For example, the area of plasmonics, which is based primarily on the novel optical properties associated with localized surface plasmon resonance on metal nanoparticles and nanostructured surfaces, is attracting considerable research attention with conferences and a new journal dedicated to the topic. The key implications of plasmonics for optical chemical sensing are in nanoscale sensors (e.g., refractometric) and enhancements of spectroscopic sensors. For example, surface-enhanced Raman scattering (SERS) is an area that has been the subject of intensive research over a relatively short time frame for development of ultrasensitive diagnostic devices. A number of novel geometries/platforms based on metal nanoparticles/nanostructures have been developed recently,134,135 and these have resulted in improved sensitivity and sensor stability. Research in this area is likely to accelerate further in coming years. Similarly, the area of metal-enhanced fluorescence270-273 (MEF) is attracting considerable attention and likely to play a key role in sensitivity enhancement strategies over the coming years as the underlying principles of this phenomenon become better understood. Another area where the impact of nanotechnology is likely to be seen is in the use of nanowires; the use

Optical Chemical Sensors

of subwavelength nanowires as evanescent wave-based optical sensors was recently reported.274 Nanomaterials must be supported on a suitable platform which must often provide additional functionalities such as sample delivery and signal collection. Such platforms typically employ features in micrometer dimensions, and this highlights the increasingly important role that microsystems and microfluidics will play in the future development of OCS. More generally, OCS employing microfluidics (labon-a-chip; micro-total analysis systems (µTAS); Bio-MEMS) will see significant growth especially in the context of (bio)sensor arrays and the need for miniaturized systems. It is important to note here that such microfluidic systems may also enable the use of chemical sensors based on renewable reagents (using irreversible indicators which would otherwise not be suitable for sensors) by virtue of the low volumes involved. Such miniaturized systems are also critical in the field of wireless sensor networks, which is the subject of a separate review in this issue. This area is likely to have a profound impact on our society as a whole, potentially providing real-time information on environmental and healthrelated parameters in a spatially distributed context. Another area where developments in microfluidics are likely to see major application is in the fabrication of low-cost diagnostic platforms for widespread use, e.g., in the home but more significantly for application in low resource environments (global diagnostics). Advances in optical materials and platforms are likely to play an important role in future developments of OCS. For example, refractometric sensor platforms based on geometries such as microsphere275,276 or microring277 resonators and 2D photonic crystals278 show promise for development of compact, array-based chemical sensors, while a number of groups have emphasized and demonstrated the importance of integration of all aspects of the sensor platform (i.e., source, sample, detector) onto a single chip using, for example, CMOS technology.279-281 The use of quantum dots for optical sensing is a rapidly developing area. In particular, quantum dots are increasingly being used as high-brightness fluorescent labels in optical biosensing. These nanoscale colloidal semiconductors have many advantages over conventional dye labels, for example, very high quantum efficiencies, narrow, tunable emission bands, and enhanced stability. It is likely that ongoing improvements in synthesis and functionalization of quantum dots will lead to a new generation of luminescence-based optical sensors.282-285 Another recent innovation in fluorescence-based sensor research is the exploitation of fluorescence emission anisotropy in the design of highly efficient, ultrasensitive sensor platforms based on a range of optical elements and platforms.285-288 Such platforms can enhance sensitivity by a factor of up to 100 and will have significant implications for development of efficient, low-cost diagnostic platforms. In conclusion, the field of OCS, which is based on solid fundamental principles, has an exciting future as the need for rapid supply of measurement information increases and the convergence of disparate technologies provides significant opportunities for performance enhancement.

6. References (1) Wolfbeis, O. S. Fibre Optic Chemical Sensors and Biosensors; CRC Press: Boca Raton, 1991; Vols. 1 and 2.

Chemical Reviews, 2008, Vol. 108, No. 2 419 (2) Orellana, G.; Moreno-Bondi, M. C. Frontiers in Chemical Sensors: NoVel Principles and Techniques; Springer: New York, 2005. (3) Baldini, F.; Chester, A. N.; Homola, J.; Martellucci, S. Optical Chemical Sensors; NATO Science Series 224; Springer: New York, 2006. (4) Optical Sensors - Industrial, EnVironmental and Diagnostic Applications; Edited by R. Narayanaswamy and O. S. Wolfbeis, Springer Series on Chemical Sensors and Biosensors (ed. O. S. Wolfbeis), Vol. 1, Springer, 2004. (5) Wolfbeis, O. S. Anal. Chem. 2006, 78, 3859. (6) Cammann, G. G.; Guilbault, E. A.; Hal, H.; Kellner, R.; Wolfbeis, O. S. The Cambridge Definition of Chemical Sensors, Cambridge Workshop on Chemical Sensors and Biosensors; Cambridge University Press: New York, 1996. (7) Oberg, K. I.; Hodyss, R.; Beauchamp, J. L. Sens. Actuators, B: Chem. 2006, 115, 79. (8) Alves, F. L.; Raimundo, I. M.; Gimenez, I. F.; Alves, O. L. Sens. Actuators, B: Chem. 2005, 107, 47. (9) O’Neal, D. P.; Meledeo, M. A.; Davis, J. R.; Ibey, B. L.; Gant, V. A.; Pishko, M. V.; Cote, G. L. IEEE Sens. J. 2004, 4, 728. (10) Valledor, M.; Carlos Campo, J.; Sanchez-Barragan, I.; Carlos Viera, J.; Costa-Fernandez, J. M.; Sanz-Medel, A. Sens. Actuators, B: Chem. 2006, 117, 266. (11) Chen, L. X.; Niu, C. G.; Xie, Z. M.; Long, Y. Q.; Song, X. R. Anal. Sci. 2006, 22, 977. (12) Wang, X.; Zeng, H. L.; Zhao, L. X.; Lin, J. M. Talanta 2006, 70, 160. (13) Vasylevska, G. S.; Borisov, S. M.; Krause, C.; Wolfbeis, O. S. Chem. Mater. 2006, 18, 4609. (14) Huber, C.; Klimant, I.; Krause, C.; Wolfbeis, O. S. Anal. Chem. 2001, 73, 2097. (15) Alfeeli, B.; Pickrell, G.; Wang, A. B. Sensors 2006, 6, 1308. (16) Eom, I. Y.; Dasgupta, P. K. Talanta 2006, 69, 906. (17) Tung, Y. C.; Zhang, M.; Lin, C. T.; Kurabayashi, K.; Skerlos, S. J. Sens. Actuators, B: Chem. 2004, 98, 356. (18) Camou, S.; Fujita, H.; Fujii, T. Lab Chip 2003, 3, 40. (19) Yang, B. C.; Tian, H. Z.; Xu, J.; Guan, Y. F. Talanta 2006, 69, 996. (20) Hofmann, O.; Wang, X. H.; deMello, J. C.; Bradley, D. D. C.; deMello, A. J. Lab Chip 2005, 5, 863. (21) Chabinyc, M. L.; Chiu, D. T.; McDonald, J. C.; Stroock, A. D.; Christian, J. F.; Karger, A. M.; Whitesides, G. M. Anal. Chem. 2001, 73, 4491. (22) Ahmad, M.; Chang, K. P.; King, T. A.; Hench, L. L. Sens. Actuators, A: Phys. 2005, 119, 84. (23) Park, E. J.; Reid, K. R.; Tang, W.; Kennedy, R. T.; Kopelman, R. J. Mater. Chem. 2005, 15, 2913. (24) Preejith, P. V.; Lim, C. S.; Chia, T. F. Meas. Sci. Technol. 2006, 17, 3255. (25) Bowden, M.; Song, L. N.; Walt, D. R. Anal. Chem. 2005, 77, 5583. (26) Ahn, S.; Kulis, D. M.; Erdner, D. L.; Anderson, D. M.; Walt, D. R. Appl. EnViron. Microbiol. 2006, 72, 5742. (27) Moreno, J.; Arregui, F. J.; Matias, I. R. Sens. Actuators, B: Chem. 2005, 105, 419. (28) Gupta, B. D.; Sharma, N. K. Sens. Actuators, B: Chem. 2002, 82, 89. (29) James, S. W.; Tatam, R. P. J. Opt. A: Pure Appl. Opt. 2006, 8, S430. (30) King, D.; Lyons, W. B.; Flanagan, C.; Lewis, E. Meas. Sci. Technol. 2004, 15, 1560. (31) Sang, X. Z.; Yu, C. X.; Yan, B. B.; Ma, J. X.; Meng, Z. F.; Mayteevarunyoo, T.; Lu, N. G. Chin. Phys. Lett. 2006, 23, 3202. (32) Falciai, R.; Mignani, A. G.; Vannini, A. Sens. Actuators, B: Chem. 2001, 74, 74. (33) DeLisa, M. P.; Zhang, Z.; Shiloach, M.; Pilevar, S.; Davis, C. C.; Sirkis, J. S.; Bentley, W. E. Anal. Chem. 2000, 72, 2895. (34) Pisco, M.; Consales, M.; Campopiano, S.; Viter, R.; Smyntyna, V.; Giordano, M.; Cusano, A. J. LightwaVe Technol. 2006, 24, 5000. (35) Liu, N.; Hui, J.; Sun, C. Q.; Dong, J. H.; Zhang, L. Z.; Xiao, H. Sensors 2006, 6, 835. (36) Kang, Y.; Ruan, H.; Wang, Y.; Arregui, F. J.; Matias, I. R.; Claus, R. O. Meas. Sci. Technol. 2006, 17, 1207. (37) Benounis, M.; Jaffrezic-Renault, N.; Dutasta, J. P.; Cherif, K.; Abdelghani, A. Sens. Actuators, B: Chem. 2005, 107, 32. (38) Jung, Y.; Kim, S.; Lee, D.; Oh, K. Meas. Sci. Technol. 2006, 17, 1129. (39) Burke, C. S.; McGaughey, O.; Sabattie, J. M.; Barry, H.; McEvoy, A. K.; McDonagh, C.; MacCraith, B. D. Analyst 2005, 130, 41. (40) McDonagh, C.; Kolle, C.; McEvoy, A. K.; Dowling, D. L.; Cafolla, A. A.; Cullen, S. J.; MacCraith, B. D. Sens. Actuators, B: Chem. 2001, 74, 124. (41) Malins, C.; Niggemann, M.; MacCraith, B. D. Meas. Sci. Technol. 2000, 11, 1105. (42) Chang-Yen, D. A.; Gale, B. K. Lab Chip 2003, 3, 297.

420 Chemical Reviews, 2008, Vol. 108, No. 2 (43) Burke, C. S.; Markey, A.; Nooney, R. I.; Byrne, P.; McDonagh, C. Sens. Actuators, B: Chem. 2006, 119, 288. (44) Huang, S. H.; Tseng, F. G. J. Micromech. Microeng. 2005, 15, 2235. (45) Rowe-Taitt, C. A.; Golden, J. P.; Feldstein, M. J.; Cras, J. J.; Hoffman, K. E.; Ligler, F. S. Biosens. Bioelectron. 2000, 14, 785. (46) Duveneck, G. L.; Abel, A. P.; Bopp, M. A.; Kresbach, G. M.; Ehrat, M. Anal. Chim. Acta 2002, 469, 49. (47) Chronis, N.; Lee, L. P. Lab Chip 2004, 4, 125. (48) Gouin, J. F.; Doyle, A.; MacCraith, B. D. Electron. Lett. 1998, 34, 1685. (49) Zourob, M.; Mohr, S.; Fielden, P. R.; Goddard, N. J. Sensors Actuators, B: Chem. 2003, 94, 304. (50) Malins, C.; Doyle, A.; MacCraith, B. D.; Kvasnik, F.; Landl, M.; Simon, P.; Kalvoda, L.; Lukas, R.; Pufler, K.; Babusik, I. J. EnViron. Monit. 1999, 1, 417. (51) Lavers, C. R.; Itoh, K.; Wu, S. C.; Murabayashi, M.; Mauchline, I.; Stewart, G.; Stout, T. Sens. Actuators, B: Chem. 2000, 69, 85. (52) Yang, L.; Saavedra, S. S. Anal. Chem. 1995, 67, 1307. (53) Yang, L.; Saavedra, S. S.; Armstrong, N. R. Anal. Chem. 1996, 68, 1834. (54) Skrdla, P. J.; Armstrong, N. R.; Saavedra, S. S. Anal. Chim. Acta 2002, 455, 49. (55) Puyol, M.; del Valle, M.; Garces, I.; Villuendas, F.; Dominguez, C.; Alonso, J. Anal. Chem. 1999, 71, 5037. (56) Puyol, M.; Salinas, I.; Garces, I.; Villuendas, F.; Llobera, A.; Dominguez, C.; Alonso, J. Anal. Chem. 2002, 74, 3354. (57) Hisamoto, H.; Kim, K.; Manabe, Y.; Sasaki, K.; Minamitani, H.; Suzuki, K. Anal. Chim. Acta 1997, 342, 31. (58) Qi, Z. M.; Yimit, A.; Itoh, K.; Murabayashi, M.; Matsuda, N.; Takatsu, A.; Kato, K. Opt. Lett. 2001, 26, 629. (59) Burke, C. S.; Polerecky, L.; MacCraith, B. D. Meas. Sci. Technol. 2004, 15, 1140. (60) Lambeck, P. V. Meas. Sci. Technol. 2006, 17, R93. (61) Gauglitz, G. Anal. Bioanal. Chem. 2005, 381, 141. (62) Hua, P.; Luff, B. J.; Quigley, G. R.; Wilkinson, J. S.; Kawaguchi, K. Sens. Actuators, B: Chem. 2002, 87, 250. (63) Prieto, F.; Sepulveda, B.; Calle, A.; Llobera, A.; Dominguez, C.; Abad, A.; Montoya, A.; Lechuga, L. M. Nanotechnology 2003, 14, 907. (64) Qi, Z. M.; Matsuda, N.; Itoh, K.; Murabayashi, M.; Lavers, C. R. Sens. Actuators, B: Chem. 2002, 81, 254. (65) Sepu´lveda, B.; Rı´o, J. S. d.; Moreno, M.; Blanco, F. J.; Mayora, K.; Domı´nguez, C.; Lechuga, L. M. J. Opt. A: Pure Appl. Opt. 2006, 8, S561. (66) Esinenco, D.; Psoma, S. D.; Kusko, M.; Schneider, A.; Muller, R. ReV. AdV. Mater. Sci. 2005, 10, 295. (67) Heideman, R. G.; Lambeck, P. V. Sens. Actuators, B: Chem. 1999, 61, 100. (68) Ymeti, A.; Greve, J.; Lambeck, P. V.; Wink, T.; van Hovell, S. W. F. M.; Beumer, T. A. M.; Wijn, R. R.; Heideman, R. G.; Subramaniam, V.; Kanger, J. S. Nano Lett. 2007, 7, 394. (69) Hradetzky, D.; Mueller, C.; Reinecke, H. J. Opt. A: Pure Appl. Opt. 2006, 8, S360. (70) Hoffmann, C.; Schmitt, K.; Brandenburg, A.; Hartmann, S. Anal. Bioanal. Chem. 2007, 387, 1921. (71) Yuan, W.; Ho, H. P.; Wong, C. L.; Kong, S. K.; Lin, C. L. IEEE Sens. J. 2007, 7, 70. (72) Koster, T.; Lambeck, P. Sens. Actuators, B: Chem. 2002, 82, 213. (73) Gut, K. J. Phys. IV 2005, 129, 109. (74) Stamm, C.; Dangel, R.; Lukosz, W. Opt. Commun. 1998, 153, 347. (75) Ayras, P.; Honkanen, S.; Grace, K. M.; Shrouf, K.; Katila, P.; Leppihalme, M.; Tervonen, A.; Yang, X.; Swanson, B.; Peyghambarian, N. J. Opt. A: Pure Appl. Opt. 1998, 7, 1261. (76) Kribich, K. R.; Copperwhite, R.; Barry, H.; Kolodziejczyk, B.; Sabattie´, J.-M.; O’Dwyer, K.; MacCraith, B. D. Sens. Actuators, B: Chem. 2005, 107, 188. (77) Goddard, N. J.; Singh, K.; Hulme, J. P.; Malins, C.; Holmes, R. J. Sens. Actuators, A: Phys. 2002, 100, 1. (78) Malins, C.; Fielden, P. R.; Goddard, N. J. Meas. Sci. Technol. 2004, 15, 948. (79) Cunningham, B.; Li, P.; Lin, B.; Pepper, J. Sens. Actuators, B: Chem. 2002, 81, 316. (80) Challener, W. A.; Edwards, J. D.; McGowan, R. W.; Skorjanec, J.; Yang, Z. Sens. Actuators, B: Chem. 2000, 71, 42. (81) Yuen, P. K.; Fontaine, N. H.; Quesada, M. A.; Mazumder, P.; Bergman, R.; Mozdy, E. J. Lab Chip 2005, 5, 959. (82) Schueller, O. J. A.; Duffy, D. C.; Rogers, J. A.; Brittain, S. T.; Whitesides, G. M. Sens. Actuators, A: Phys. 1999, 78, 149. (83) Morhard, F.; Pipper, J.; Dahint, R.; Grunze, M. Sens. Actuators, B: Chem. 2000, 70, 232. (84) St. John, P. M.; Davis, R.; Cady, N.; Czajka, J.; Batt, C. A.; Craighead, H. G. Anal. Chem. 1998, 70, 1108.

McDonagh et al. (85) Dubendorfer, J.; Kunz, R. E.; Jobst, G.; Moser, I.; Urban, G. Sens. Actuators, B: Chem. 1998, 50, 210. (86) Wiki, M.; Gao, H.; Juvet, M.; Kunz, R. E. Biosens. Bioelectron. 2001, 16, 37. (87) Hanel, C.; Gauglitz, G. Anal. Bioanal. Chem. 2002, 372, 91. (88) Nikitin, P. I.; Gorshkov, B. G.; Nikitin, E. P.; Ksenevich, T. I. Sens. Actuators, B: Chem. 2005, 111, 500. (89) Lambeck, P. V.; van Lith, J.; Hoekstra, H. J. W. M. Sens. Actuators, B: Chem. 2006, 113, 718. (90) O’Sullivan, D. W.; Millero, F. J. Mar. Chem. 1998, 60, 75. (91) Kimoto, H.; Nozaki, K.; Kudo, S.; Kato, K.; Negishi, A.; Kayanne, H. Anal. Sci. 2002, 18, 247. (92) Kaltin, S.; Haraldsson, C.; Anderson, L. G. Mar. Chem. 2005, 96, 53. (93) Bandstra, L.; Hales, B.; Takahashi, T. Mar. Chem. 2006, 100, 24. (94) De Castro, A. J.; Lerma, A. M.; Lopez, F.; M. Guijarro, M.; Diez, C.; Harnando, C.; Madrigal, J. Infrared Phys. Technol. 2007, 51 (1), 21. (95) Reyes, F.; Grutter, M.; Jazcilevich, A.; Gonzalez-Oropeza, R. Atmos. Chem. Phys. 2006, 6, 5339. (96) Eidelman, N.; Raghavan, D.; Forster, A. M.; Amis, E. J.; Karim, A. Macromol. Rapid Commun. 2004, 285, 259. (97) Snively, C. M.; Oskarsdottir, G.; Lauterback, J. Catal. Today 2001, 67, 357. (98) Chan, K. L. A.; Kazarian, S. G. J. Comb. Chem. 2005, 7, 185. (99) Dessipri, E.; Minopoulou, E.; Chryssikos, G. D.; Gionis, V.; Paipetis, A.; Panayiotou, C. Eur. Polym. J. 2003, 39, 1533. (100) Werle, P. Spectrosc. Acta, Pt. A: Mol. Biomol. Spectrosc. 1998, 54, 197. (101) Allen, M. G. Meas. Sci. Technol. 1998, 9, 545. (102) Linnerud, L.; Kaspersen, P.; Jaeger, T. Appl. Phys. B 1998, 57, 297. (103) Werle, P.; Popov, A. Appl. Opt. 1999, 38, 1494. (104) Werle, P.; Slemr, F.; Maurer, K.; Kormann, R.; Mucke, R.; Janker, B. Opt. Lasers Eng. 2002, 37, 101. (105) Kosterev, A. A.; Tittel, F. K. IEEE J. Quantum Electron. 2002, 38, 582. (106) Hvozdara, L.; Pennington, N.; Kraft, M.; Karlowatz, M.; Miziakoff, B. Vib. Spectrosc. 2002, 30, 53. (107) Capasso, F.; Gmachl, C.; Paiella, R.; Tredicucci, A.; Hutchinson, A. L.; Sivco, D.; Baillargeon, J. N.; Cho, A. Y. IEEE J. Sel. Top. Quantum Electron. 2000, 6, 931. (108) Hvozdara, L.; Gianordoli, S.; Strasser, G.; Schrenk, W.; Unterrainer, K.; Gornik, E.; Murthy, C. S.; Kraft, M.; Pustogow, V.; Mizaikoff, B.; Inberg, A.; Croitoru, N. Appl. Opt. 2000, 39, 6926. (109) Sharpe, S. W.; Kelly, J. F.; Hartman, J. S.; Gmachl, C.; Capasso, F.; Sivco, D. L.; Baillargeon, J. N.; Cho, A. Y. Opt. Lett. 1998, 23, 1396. (110) Schaden, S.; Haberdorn, M.; Frank, J.; Baena, J. R.; Lendl, B. Appl. Spectrosc. 2004, 58 (6), 667. (111) Kosterev, A. A.; Curl, R. F.; Tittel, F. K.; Gmachl, C.; Capasso, F.; Sivco, D. L.; Baillargeon, J. N.; Hutchinson, A. L.; Cho, A. Y. Appl. Opt. 2000, 39, 4425. (112) Weber, W. H.; Remillard, J. T.; Chase, R. E.; Richert, J. F.; Capasso, R.; Gmachl, C.; Hutchinson, A. L.; Sivco, D. L.; Baillargeon, J. N.; Cho, A. Y. Appl. Spectrosc. 2002, 56, 706. (113) Nelson, D. D.; McManus, B.; Urbanski, S.; Herndon, S.; Zahniser, M. S. Spectrosc. Acta, Pt. A: Mol. Biomol. Spectrosc. 2004, 60, 3325. (114) Kolhed, M.; Haberkorn, M.; Pustogov, V.; Mizaikoff, B.; Frank J.; Karlberg, B. Lendl, B. Vib. Spectrosc. 2002, 29, 282. (115) Tau, S.; Sarma, T. V. S. Opt. Lett. 2006, 31 (10), 1423. (116) Wu, B., Chang, C.; Sree, U.; Chiu, K.; Lo, J. Anal. Chim. Acta 2006, 576, 91. (117) Lin, T.; Sree, R.; Tseng, S.; Chiu, K.; Wu, C.; Lo, J. Atmos. EnViron. 2004, 38, 4111. (118) Velasco, E.; Lamb, B.; Westberg, H.; Allwine, E.; Sosa, G.; ArriagaColina, J. L; Jobson, B. T.; Alexander, M. L.; Prazeller, P.; Knighton, W. B.; Rogers, T. M.; Grutter, M.; Herndon, S. C.; Kolb, C. E.; Zavala, M.; de Foy, B.; Volkamer, R.; Molina, L. T.; Molina, M. J. Atmos. Chem. Phys. 2007, 7, 329. (119) Wu, Y.; Zheng, W.; Qu, J. Y. Opt. Lett. 2006, 31 (21), 3122. (120) Wu, Y.; Qu, J. Y. J. Biomed. Opt. 2006, 11 (5), 054023 1. (121) Drezek, R.; Brookner, C.; Pavlova, I.; Boiko, I.; Malpica, A.; Lotan, R.; Follen, M.; Richards-Kortum, R. Photochem. Photobiol. 2001, 73, 636. (122) Brookner, C. K.; Follen, M. Boiko, I.; Galvan, J.; Thomsen, S.; Malpica, A.; Suzuki, S.; Lotan, R.; Richards-Kortum, R. Photochem. Photobiol. 2000, 71 (6), 730. (123) Onizawa, K.; Okamura, N.; Saginoya, H.; Yoshida, H. Oral Oncol. 2003, 39, 150. (124) Vivian, J. T.; Callis, P. R. Biophys. J. 2001, 80, 2093. (125) Monsellier, E.; Bedouelle, H. Protein Eng. Des. Sel. 2005, 18, 445. (126) Duy, C.; Fitter, J. Biophys. J. 2006, 90, 3704. (127) Patra, D.; Mishra, A. K. Anal. Bioanal. Chem. 2002, 373, 304.

Optical Chemical Sensors (128) Ryder, A. G. J. Fluoresc. 2004, 14, 99. (129) Sarma, A. K.; Ryder, A. G. Energy Fuels 2006, 20, 783. (130) Pouilli, K. I.; Mousdis, G. A.; Georgiou, C. A. Anal. Bioanal. Chem. 2006, 386, 1571. (131) Scott, S. M.; James, D.; Ali, Z.; O’Hare, W. T.; Rowell, F. J. Analyst 2003, 128, 966. (132) Laserna, J. An Introduction to Raman Spectroscopy: Introduction and Basic Principles; Wiley: New York, 2001. (133) Owen, C. A.; Notingher, I.; Hill, R. J. Mater. Sci.: Mater. Med. 2006, 17, 1019. (134) Uzunbajakava, N.; Lenferink, A.; Kraan, Y.; Volokhina, E.; Vrensen, G. Greve, J.; Otto, C. Biophys. J. 2003, 6, 3698. (135) Shafer-Peltier, K. E.; Haka, A. S.; Motz, J. T.; Fitzmaurice, M.; Dasari, R. R.; Field, S. J. Cell Biochem. 2002, 39, 125. (136) Puppels, G. J. J. Raman Spectrosc. 2002, 7, 496. (137) Kneipp, J.; Schut, T. B.; Kliffen, M.; Menke-Pluijmers, M.; Puppels, G. Vib. Spectrosc. 2002, 1, 67. (138) Notingher, I.; Green, C.; Dyer, C.; Perkins, E.; Hopkins, N.; Lindsay, C.; Hench, L. L. J. R. Soc. Interface 2004, 1, 79. (139) Stuart, D. A.; Biggs, K. B.; Van Duyne, R. P. Analyst 2006, 131, 568. (140) Vo-Dinh, T.; Yan, F.; Wabuyele, M. B. Surface-enhanced Raman Scattering: Physics and Application. Top Appl. Phys. 2006, 103, 409. (141) Zhang, X.; Shah, N.; Van Duyne, R. P. Vib. Spectrosc. 2006, 42, 2. (142) Yonzon, C. R.; Stuart, D. A.; Zhang, X.; McFarland, A. D.; Haynes, C. L.; Van Duyne, R. P. Talanta 2005, 67, 438. (143) Stuart, D. A.; Yuen, J. M.; Shah, N.; Lyandres, O.; Yonzon, C.; Glucksberg, M. R.; Walsh, J. T.; Van Duyne, R. P. Anal. Chem. 2006, 78, 7211. (144) Deckert, V.; Zeisel, D.; Zenobi, R.; Vo-Dinh, T. Anal. Chem. 1998, 70 (13), 2646. (145) Chen, K.; Leona, M.; Vo-Dinh, K.; Yan, F.; Wabuyele, M. B.; VoDinh, T. J. Raman Spectrosc. 2006, 37, 520. (146) Chen, K.; Vo-Dinh, K.; Yan, F.; Wabuyele, M. B.; Vo-Dinh, T. Anal. Chim. Acta 2006, 569, 234. (147) Schmidt, H.; Nguyen, B. H.; Pfannkuche, F.; Amann, H.; Kronfeldt, H. D.; Kowalewska, G. Mar. Pollut. Bull. 2004, 49, 229. (148) Lobnik, A. In Optical Chemical Sensors; Baldini, F., Chester, A. N., Homola, J., Martellucci, S., Eds.; NATO Science Series 224; Springer: New York, 2006. (149) Lin, J.; Liu, D. Anal. Chim. Acta 2000, 408, 49. (150) Makedonski, P.; Brandes, M.; Grahn, W.; Kowalsky, W.; Wichern, J.; Weise, S.; Johannes, H. Dyes Pigm. 2004, 61, 109. (151) Gotou, T.; Noda, M.; Tomiyama, T.; Sembokuya, H.; Kubouchi, M.; Tsuda, K. Sens. Actuators, B: Chem. 2006, 119, 27. (152) Segawa, H.; Ohnishi, E.; Arai, Y.; Yoshida, K. Sens. Actuators, B: Chem. 2003, 94, 276. (153) Oter, O.; Ertekin, K.; Tapkaya, D.; Alp, S. Sens. Actuators, B: Chem. 2006, 117, 295. (154) Baldini, F.; Falai, A.; De Gaudio, A. R.; Landi, D.; Lueger A.; Mencaglia, A.; Scherr, D.; Trettnak, W. Sens. Actuators, B: Chem. 2003, 90, 132. (155) Mohr, G. J. Anal. Bioanal Chem. 2006, 386, 1201. (156) Mohr, G. J. Sens. Actuators, B: Chem. 2005, 107. 2. (157) Wetzl, B. K.; Yarmoluk, S. M.; Craig, D.; Wolfbeis, O. S. Angew. Chem. 2004, 116, 5515. (158) Hofelschweiger, B.; Durkop, A.; Wolfbeis, O. S. Anal. Biochem. 2005, 344, 122. (159) Suzuki, Y.; Nakano, N.; Suzuki, K. EnViron. Sci. Technol. 2003, 37, 5695. (160) Kawamura, K.; Kerman, K.; Fujihara, M.; Nagatani, N.; Hashiba, T.; Tamiya, E. Sens. Actuators, B: Chem. 2005, 105, 495. (161) Ros-Lis, J.; Martinez-Manez, R.; Rurak, K.; Sancenon, F.; Soto, J.; Spieles, M. Inorg. Chem. 2004, 43, 5183. (162) Safavi, A.; Bagheri, M. Sens. Actuators, B: Chem. 2004, 99, 608. (163) Ros-Lis, J. V.; Martinez-Manez, R.; Soto, J. Chem. Commun. 2002, 19, 2248. (164) Badr, I. H. A.; Meyerhoff, M. E. Anal. Chem. 2005, 77, 6719. (165) Teshima, N.; Li, J.; Toda, K.; Dasgupta, P. K. Anal. Chim. Acta 2005, 535, 189. (166) Carter, J. C.; Alvis, R. M.; Brown, S. B.; Langry, K. C.; Wilson, T. S.; McBride, M. T.; Murick, M. L.; Cox, W. R.; Grove, M. S.; Colston, B. W. Biosens. Bioelectron. 2006, 21, 1359. (167) Duong, H. D.; Sohn, O.; Lam, H. T.; Thee, J. I. Microchem. J. 2006, 84, 50. (168) Kermis, H. R.; Kostov, Y.; Rao, G. Analyst 2003, 128, 1181. (169) Sanchez-Barragan, I.; Costa-Fernandez, J. M.; Sanz-Medel, A.; Valledor, M.; Ferrero, F. J.; Campo, J. C. Anal. Chim. Acta 2006, 562, 197. (170) Ando, Y.; Iino, S.; Yamada, K.; Umezawa, K.; Iwasawa, N.; Citterio, D.; Suzuki, K. Sens. Actuators, B: Chem. 2007, 121, 74. (171) Le, T. P.; Rogers, J. E.; Kelly, L. A. J. Phys. Chem A 2000, 104, 6778.

Chemical Reviews, 2008, Vol. 108, No. 2 421 (172) De Silva, A. P.; Rice, T. E. J. Chem. Soc., Chem. Commun. 1999, 2, 163. (173) Li, Z.; Niu, C.; Zeng, G.; Liu, Y.; Gao, P.; Huang, G.; Mao, Y. Sens. Actuators, B: Chem. 2006, 114, 308. (174) Higgins, B.; DeGraff, B. A.; Demas, J. N. Inorg. Chem. 2005, 44 (19), 6662. (175) Chen, X.; Lin, L.; Dai, Y.; Wang, X. Anal. Chim. Acta 2004, 506, 9. (176) Mohr, G.; Draxler, S.; Trznadel, K.; Lehmann, F.; Lippitsch, M. E. Anal. Chim. Acta 1998, 360, 119. (177) Nivens, D. A.; Schiza, M. V.; Angel, S. M. Talanta 2002, 58, 543. (178) Mills, A.; Eaton, K. Quim Anal. 2000, 19, 75. (179) Wolfbeis, O. S.; Klimant, I.; Werner, T.; Huber, C.; Kosch, U.; Krause, C.; Neurauter, G.; Durkop, A. Sens. Actuators, B: Chem. 1998, 51, 17. (180) Cajlakovic, M.; Bizzarri, A.; Ribitsch, V. Anal. Chim. Acta 2006, 573-574, 57. (181) Zhu, Q.; Aller, R. C.; Fan, Y. Mar. Chem. 2006, 101, 40. (182) Amao, Y. Microchim. Acta 2003, 143, 1. (183) Amao, Y.; Okura, I.; Miyashita, T. Bull. Chem. Soc. Jpn. 2001, 74, 1156. (184) Fujiwara, Y.; Okura, I.; Miyashita, T.; Amao, Y. Anal. Chim. Acta 2002, 471, 25. (185) Choi, S. J.; Choi, B. G.; Park, S. M. Anal. Chem. 2002, 74 (9), 1998. (186) Fuller, Z. J.; Bare, W. D.; Kneas, K. A.; Xu, W. Y.; Demas, J. N.; DeGraaf, B. A. Anal. Chem. 2003, 75 (11), 2670. (187) McDonagh, C.; Bowe, P.; Mongey, K.; MacCraith, B. D. J. NonCryst. Solids 2002, 306, 138. (188) Lee, S. K.; Okura, I. Anal. Commun. 1997, 34, 185. (189) Kerry, J. P.; Papkovsky, D. P. Research AdVances in Food Science; Global Research Network: Kerala, India, 2002; Vol. 3, p 121. (190) O’Mahony, F. C.; O’Riordan, T. C.; Papkovskaia, N.; Kerry, J. P.; Papkovsky, D. B. Food Control 2006, 17, 286. (191) O’Riordan, T. C.; Voraberger, H.; Kerry, J. P.; Papkovsky, D. B. Anal. Chim. Acta 2005, 530, 135. (192) Glenn, S. J.; Cullum, B. M.; Nair, R. B.; Nivens, D. A.; Murphy, C. J.; Angel, S. M. Anal. Chim. Acta 2001, 448, 1. (193) Bedoya, M.; Diez, M. T.; Moreno-Bondi, M. C.; Orellana, G. Sens. Actuators, B: Chem. 2006, 113, 573. (194) McGaughey, O.; Ros-Lis, J. V.; Guckian, A.; McEvoy, A. K.; McDonagh, C.; MacCraith, B. D. Anal. Chim. Acta 2006, 570, 15. (195) Xu, C.; Qin, Y.; Bakker, E. Talanta 2004, 63, 180. (196) Quinti, L.; Allen, N. S.; Edge, M.; Murphy, B. P.; Perotti, A. J. Photochem. Photobiol. A: Chem. 2003, 155, 79. (197) Ng, S. M.; Narayanaswamy, R. Anal. Bioanal. Chem. 2006, 386, 1235. (198) Baduaga, R.; Lakowicz, J. R.; Geddes, C. D. Anal. Chim. Acta 2004, 522, 9. (199) Huber, C.; Klimant, I.; Krause, C.; Werner, T.; Wolfbeis, O. S. Anal. Chim. Acta 2001, 449, 81. (200) Wright, J. D.; Sommerdijk, N. A. Sol-gel Materials: Chemistry and Application; CRC Press: Boca Raton, 2000. (201) MacCraith, B. D.; McDonagh, C. J. Fluoresc. 2002, 12 (3,4), 333. (202) Tao, Z.; Tehan, E. G.; Tang, Y.; Bright, F. V. Anal. Chem. 2006, 78, 1939. (203) Bukowski, R. M.; Ciriminna, R.; Pagliaro, M.; Bright, F. V. Anal. Chem. 2005, 77, 2670. (204) Bukowski, R. M.; Davenport, M. D.; Titus, A. H.; Bright, R.V. Appl. Spectrosc. 2006, 60 (9), 951. (205) Klimant, I.; Ruckruh, F.; Liebsch, G.; Stangelmayer, A.; Wolfbeis, O. S. Mikrochim. Acta 1999, 131, 35. (206) Holder, E.; Oelkrug, D.; Egelhaaf, H.; Mayer, H. A.; Lindner, E. J. Fluoresc. 2002, 12 (3,4), 383. (207) Zhang, H.; Li, B.; Lei, B.; Li, W.; Lu, S. Sens. Actuators, B: Chem. 2007, 12 (1), 208. (208) Von Bultzingslowen, C.; McEvoy, A.; McDonagh, C.; MacCraith, B. D. Anal. Chim. Acta 2003, 480, 275. (209) Von Bultzingslowen, C.; McEvoy, A.; McDonagh, C. M.; MacCraith, B. D.; Klimant, I.; Krause, C.; Wolfbeis, O. S. Analyst 2002, 127, 1478. (210) Grosso, D.; Cagnol, F.; Soler-Illia, G. J.; Crepaldi, E. L.; Amenitsch, H.; Brunet-Bruneau, A.; Bourgeois, A.; Sanchez, C. AdV. Funct. Mater. 2004, 14, 309. (211) Nicole, L.; Boissiere, C.; Grosso, D.; Hesemann, P.; Moreau, J.; Sanchez, C. Chem. Commun. 2004, 20, 2312. (212) Lei, B.; Li, B.; Zhang, H.; Lu, S.; Zheng, Z.; Li, W.; Wang, Y. AdV. Funct. Mater. 2006, 16, 1883. (213) Tao, S.; Winstead, C. B.; Jindal, R.; Singh, J. P. IEEE Sens. J. 2004, 4 (3), 322. (214) Zhang, P.; Guo, J.; Wang, Y.; Pang, W. Mater. Lett. 2002, 53, 400. (215) Mills, A. Sens. Actuators, B: Chem. 1998, 51, 60. (216) Reiss, J. G.; Krafft, M. P. Biomaterials 1998, 19, 1529. (217) Voraberger, H. S.; Kreimaier, H.; Biebernik, K.; Kern, W. Sens. Actuators, B: Chem. 2001, 74, 179.

422 Chemical Reviews, 2008, Vol. 108, No. 2 (218) Florescu, M.; Katerkamp, A. Sens. Actuators, B: Chem. 2004, 97, 39. (219) Holmes-Smith, A. S.; Zheng, X.; Uttamial, M. Meas. Sci. Technol. 2006, 17, 3328. (220) Neurauter, G.; Klimant, I.; Wolfbeis, O. S. Fresnius J. Anal. Chem. 2000, 366, 481. (221) Ge, X.; Kostov, Y.; Rao, G. Biosens. Bioelectron. 2003, 18, 857. (222) Waldner, A.; Barnard, S. M. U.S. Patent 6,338,822, 2002. (223) Schroder, C. R.; Klimant, I. Sens. Actuators, B: Chem. 2005, 107, 572. (224) Haupt, K.; Mosback, K. Chem. ReV. 2000, 100, 2495. (225) Chen, Y.; Brazier, J. J.; Yan, M.; Bargo, P. R.; Prahl, S. A. Sens. Actuators, B: Chem. 2004, 102, 107. (226) Lieberzeit, P. A.; Gazda-Miarecka, S.; Halikias, K.; Schirk, C.; Kauling, J.; Dickert, F. L. Sens. Actuators, B: Chem. 2005, 1112, 259. (227) Al-Kindy, S.; Badia, R.; Suarez-Rodriguez, J. L.; Diaz-Garcia, M. E. Crit. ReV. Anal. Chem. 2000, 30 (4), 291. (228) Kindschy, L. M.; Alocilya, E. C. Trans. ASAE 2004, 47 (4), 1375. (229) Orellana, G.; Garcia-Fresnadillo, D. In Optical sensors: industrial, enVironmental, and diagnostic applications; Narayanaswamy, R., Wolfbeis, O. S., Eds.; Springer: New York, 2004; p 338. (230) Mongey, K. F.; Vos, J. G.; MacCraith, B. D.; McDonagh, C.; Coates, C.; McGarvey, J. J. J. Mater. Chem. 1997, 7, 1473. (231) Steinberg, I. M.; Lobnik, A.; Wolfbeis, O. S. Sens. Actuators, B: Chem. 2003, 90, 230. (232) Orellana, G.; Moreno-Bondi, M. C.; Garcia-Fresnadillo, D.; Marazuela, M. D. In Frontiers in Chemical Sensors: NoVel Principles and Techniques; Orellano, G., Wolfbeis, O.O., Eds.; Springer: New York, 2005; p 189. (233) Roth, T. P. Ph.D. Thesis, Swiss Federal Institute of Technology, 2001. (234) Tao, S.; Gong, S.; Fanguy, J. C.; Hu, X. Sens. Actuators, B: Chem. 2007, 120, 724. (235) Safafi, A.; Banazadeh, A. R. Anal. Chim. Acta 2007, 583, 326. (236) Amao, Y.; Komori, T. Talanta 2005, 66, 976. (237) Nakamura, N.; Amao, U. Sens. Actuators, B: Chem. 2003, 92, 98. (238) Miled, O. B.; Grosso, D.; Sanchez, C.; Livage, J. J. Phys. Chem. Solids 2004, 65, 1751. (239) Oter, O.; Ertekin, K.; Topkaya, D.; Alp, S. Sens. Actuators, B: Chem. 2006, 117, 295. (240) Oter, O.; Ertekin, K.; Topkaya, D.; Alp, S. Anal. BioAnal. Chem. 2006, 386, 1225. (241) Chu, C.; Lo, Y. Sens. Actuators, B: Chem. 2007, 124 (2), 376. (242) Tang, Y.; Tao, Z.; Bukowski, R. M.; Tehan, E. C.; Karri, S.; Titus, A. H.; Bright, F. V. Analyst 2006, 131, 1129. (243) Fernandez-Sanchez, J. F.; Cannas, R.; Spichiger, S.; Steiger, R.; Spicher-Keller, U. E. Anal. Chim. Acta 2006, 566, 271. (244) Derinkuyu, A.; Ertekin, K.; Oter, O.; Denizalti, S.; Cetinkaya, E. Anal. Chim. Acta 2007, 588, 42. (245) Citterio, D.; Takeda, J.; Kosugi, M.; Hisamoto, H.; Sasaki, S.; Komatsu, H.; Suzuki, K. Anal. Chem. 2007, 79, 1237. (246) Barauh, M.; Qin, W.; Vallee, R. A. L.; Beljonne, D.; Rohand, R.; Dehaen, W.; Boens, N. Org. Lett. 2005, 7 (20), 4377. (247) Ligler, F. S.; Sapsford, K. W.; Golden, J. P.; Shriver-Lake, L. C.; Taitt, C. R.; Dyer, M. A.; Baraone, S.; Myatt, C. J. Anal. Sci. 2007, 23, 5. (248) Taitt, C. R.; Anderson, G. P.; Lingerfelt, B. M.; Feldstein, M. J.; Ligler, F.S. Anal. Chem. 2002, 74, 6114. (249) Sapsford, K. E.; Taitt, C. R.; Loo, N.; Ligler, F. S. Appl. EnViron. Microbiol. 2005, 71, 5590. (250) Shriver-Lake, L. C.; Ligler, F. S. IEEE Sens. J. 2005, 5 (4), 751. (251) Qin, W. W.; Baruah, M.; De Borggraeve, W. M.; Boens, N. J. Photochem. Photobiol. A: Chem. 2006, 183, 190. (252) Kuwana, E.; Liang, F.; Sevick-Muraca, E. M. Biotechnol. Prog. 2004, 20, 1561. (253) Ryder, A. G.; Power, S.; Glynn, T. J. Appl. Spectrosc. 2003, 57, 73. (254) Hradil, J.; Davis, C.; Mongey, K.; McDonagh, C.; MacCraith, B. D. Meas. Sci. Technol. 2002, 13, 1552.

McDonagh et al. (255) Kostov, Y.; Rao, G. ReV. Sci. Instrum. 2003, 74, 4129. (256) Chodavarapu, V. P.; Shubin, D. O.; Bukowski, R. M.; Titus, A. H.; Cartwright, A. N.; Bright, F. V. IEEE Trans. Circuits Syst. I: Regul. Pap. 2007, 54, 111. (257) Kostov, Y.; Rao, G. Sens. Actuators, B: Chem. 2003, 90, 139. (258) Ogurtsov, V. I.; Papkovsky, D. B. Sens. Actuators, B: Chem. 2006, 113, 917. (259) O’Mahony, F. C.; O’Riordan, T. C.; Papkovskaia, N.; Ogurtsov, V. I.; Kerry, J. P.; Papkovsky, D. B. Packag. Technol. Sci. 2004, 17, 225. (260) Shinar, R.; Zhou, Z. Q.; Choudhury, B.; Shinar, J. Anal. Chim. Acta 2006, 568, 190. (261) Neurauter, G.; Klimant, I.; Wolfbeis, O. S. Anal. Chim. Acta 1999, 382, 67. (262) Huber, C.; Klimant, I.; Krause, C.; Werner, T.; Mayr, T.; Wolfbeis, O. S. Fresenius J. Anal. Chem. 2000, 368, 196. (263) Borisov, S. M.; Krause, C.; Arain, S.; Wolfbeis, O. S. AdV. Mater. 2006, 18, 1511. (264) Stahl, H.; Glud, A.; Schroder, C. R.; Klimant, I.; Tengberg, A.; Glud, R. N. Limnol. Oceanogr. Meth. 2006, 4, 336. (265) Liebsch, G.; Klimant, I.; Krause, C.; Wolfbeis, O. S. Anal. Chem. 2001, 73, 4354. (266) Huber, C.; Klimant, I.; Krause, C.; Werner, T.; Wolfbeis, O. S. Anal. Chim. Acta 2001, 449, 81. (267) Mayr, T.; Klimant, I.; Wolfbeis, O. S.; Werner, T. Anal. Chim. Acta 2002, 462, 1. (268) Borisov, S. M.; Neurauter, G.; Schroeder, C.; Klimant, I.; Wolfbeis, O. S. Appl. Spectrosc. 2006, 60, 1167. (269) Rowe, H. M.; Chan, S. P.; Demas, J. N.; DeGraff, B. A. Appl. Spectrosc. 2003, 57, 532. (270) Aslan, K.; Gryczynski, I.; Malicka, J.; Matveeva, E.; Lakowicz, J. R.; Geddes, C. D. Curr. Opin. Biotechnol. 2005, 16, 55. (271) Lakowicz, J. R. Anal. Biochem. 2005, 337, 171. (272) Stranik, O.; McEvoy, H. M.; McDonagh, C.; MacCraith, B. D. Sens. Actuators, B: Chem. 2005, 107, 148. (273) Zhang, J.; Lakowicz, J. R. Opt. Express 2007, 15, 2598. (274) Sirbuly, D. J.; Tao, A.; Law, M.; Fan, R.; Yang, P. D. AdV. Mater. 2007, 19, 61. (275) Arnold, S.; Khoshsima, M.; Teraoka, I.; Holler, S.; Vollmer, F. Opt. Lett. 2003, 28, 272. (276) Hanumegowda, N. M.; White, I. M.; Oveys, H.; Fan, X. D. Sens. Lett. 2005, 3, 315. (277) White, I. M.; Zhu, H. Y.; Suter, J. D.; Hanumegowda, N. M.; Oveys, H.; Zourob, M.; Fan, X. D. IEEE Sens. J. 2007, 7, 28. (278) Camargo, E. A.; Chong, H. M. H.; De La Rue, R. M. Appl. Opt. 2006, 45, 6507. (279) Thrush, E.; Levi, O.; Cook, L. J.; Deich, J.; Kurtz, A.; Smith, S. J.; Moerner, W. E.; Harris, J. S. Sens. Actuators, B: Chem. 2005, 105, 393. (280) Cristea, D.; Craciunoiu, F.; Modreanu, M.; Caldararu, M.; Cernica, I. Opt. Mater. 2001, 17, 201. (281) Adams, M. L.; Enzelberger, M.; Quake, S.; Scherer, A. Sens. Actuators, A: Phys. 2003, 104, 25. (282) Medintz, I. L.; Clapp, A. R.; Mattoussi, H.; Goldman, E. R.; Fisher, B.; Mauro, J. M. Nat. Mater. 2003, 2, 630. (283) Wang, X. J.; Ruedas-Rama, M. J.; Hall, E. A. H. Anal. Lett. 2007, 40, 1497. (284) Costa-Fernandez, J. M.; Pereiro, R.; Sanz-Medel, A. Trac-Trends Anal. Chem. 2006, 25, 207. (285) Blue, R.; Kent, N.; Polerecky, L.; McEvoy, H.; Gray, D.; MacCraith, B. D. Electron. Lett. 2005, 41, 682. (286) Holthoff, W. G.; Tehan, E. C.; Bukowski, R. M.; Kent, N.; MacCraith, B. D.; Bright, F. V. Anal. Chem. 2005, 77, 718. (287) Ruckstuhl, T.; Seeger, S. Appl. Opt. 2003, 42, 3277. (288) Sauer, U.; Preininger, C.; Krumpel, G.; Stelzer, N.; Kern, W. Sens. Actuators, B: Chem. 2005, 107, 178.

CR068102G

Chem. Rev. 2008, 108, 423−461

423

Optical Biosensors Sergey M. Borisov† and Otto S. Wolfbeis* Institute of Analytical Chemistry, Chemo- and Biosensors, University of Regensburg, D-93040 Regensburg, Germany Received April 2, 2007

Contents 1. Introduction and Scope 2. General Remarks 2.1. Definition of Biosensors 2.2. Classification of Biosensors 2.2.1. Catalytic Biosensors 2.2.2. Affinity Biosensors 2.3. General Aspects of Signal Generation, Immobilization of Biomolecules, and Sample Handling 2.4. Frequently Used Spectroscopies and Internal Referencing 3. Enzymatic Biosensors 3.1. General Considerations 3.2. Enzymatic Glucose Biosensors 3.3. Other Enzyme-Based Biosensors 4. Immunosensors 4.1. General Remarks 4.2. Immunosensor Formats 4.2.1. Direct Immunosensors 4.2.2. Competitive Immunosensors 4.2.3. Sandwich Immunosensors 4.2.4. Displacement Immunosensors 4.2.5. Immunosensors Based on Binding Inhibition 4.2.6. Comparative Study on Immunosensor Formats 4.3. Preferred Optical Readout Formats in Immunosensing 4.3.1. Conventional Readout Formats 4.3.2. Evanescent Wave, Capillary, and Other Readouts 4.4. Immobilization of Antibodies on Sensor Surfaces and Nonspecific Protein Binding 4.5. Specific Examples of Immunosensors 4.5.1. Biosensors for Proteins and Antibodies 4.5.2. Biosensors for Toxins 4.5.3. Biosensors for Drugs 4.5.4. Biosensors for Bacteria Cells 4.5.5. Biosensors for Pesticides 4.5.6. Multianalyte Biosensors 5. Biosensors Based on Ligand−Receptor Interactions 5.1. Receptor-Based Biosensors for Saccharides and Glycoproteins

423 424 424 424 424 424 425 425 426 426 427 430 436 436 437 437 437 437 439 439 439 440 440 440 440 441 441 442 443 443 444 444 445 445

* Corresponding author. E-mail: [email protected]. Tel.: (+49) (941) 943-4066. Fax: (+49) (941) 943-4064. † New address: Institute of Analytical Chemistry and Radiochemistry, University of Technology of Graz, Stremayrgasse 16, A-8010 Graz, Austria.

5.2. Receptor-Based Biosensors for Inorganic Ions 5.3. Receptor-Based Biosensors for Gaseous Species 5.4. Receptor-Based Biosensors for Toxins 6. Nucleic Acid Biosensors 6.1. Single DNA Sensors on Solid Supports and on Fiber-Optics 6.2. DNA Arrays 6.3. Molecular Beacons in DNA Sensors 6.4. Liposome-Based DNA Assays 6.5. Aptamer-Based DNA Sensing 7. Whole-Cell Biosensors 7.1. Catalytic Whole-Cell Biosensors 7.2. External Stimuli-Based Cellular Biosensors 7.3. Genetically Engineered Whole-Cell Biosensors 8. Solid Supports for Use in Optical Biosensors, and Other Methods of Immobilization 9. Outlook 10. List of Abbreviations and Acronyms 11. References

447 448 448 449 449 450 451 451 452 452 452 453 453 454 456 456 457

1. Introduction and Scope Numerous kinds of biosensors do exist, but this chapter is confined to sensors and systems where the information is gathered by the measurement of photons (rather than electrons as in the case of electrodes). More specifically, it relates to sensors based on the measurement of absorbance, reflectance, or fluorescence emissions that occur in the ultraviolet (UV), visible, or near-infrared (NIR). It does not cover sensors based on infrared or Raman spectroscopy, nor those based on surface plasmon resonance. Molecular imprintsswhile very promisingsare not covered either since they do not match the definition of a biosensor (which asks for a biological recognition element to be at work). Fluorescence is by far the method most often applied and comes in a variety of schemes. Parameters that are being measured in such sensors include intensity, decay time, anisotropy, quenching efficiency, luminescence energy transfer, and the like. Optical layouts include plain sensor foils (stripes) and also waveguide optical systems, capillary sensors, and arrays. Chemical sensors and biosensors do not have separation capabilities unless coupled to respective additional devices that, however, make the system more complex, require larger instrumental effort (and power consumption!), and prevent sensing to be combined with imaging. Hence, specificity can only be based on selective (bio)molecular recognition. To achieve this goal, use is made of more or less specific biorecognition elements such as enzymes, antibodies, oligonucleotides, or even whole cells and tissues. The variety

10.1021/cr068105t CCC: $71.00 © 2008 American Chemical Society Published on Web 01/30/2008

424 Chemical Reviews, 2008, Vol. 108, No. 2

Sergey M. Borisov, born in 1978, at present is a postdoctoral fellow at the Institute of Analytical Chemistry and Radiochemistry at the University of Technology in Graz, Austria. He recived a Ph.D. degree in chemistry from the Herzen State Pedagogical University (St. Petersburg, Russia) in 2003 and worked as a postdoctoral fellow at the University of Regensburg before he assumed his present position in Nov. 2006. His research interest is in the chemistry of porphyrins, in the application of fluorescent probes to biosensing schemes, and in the use of (fluorescent) micro- and nanomaterials in bioanalytical methods. He has authored several articles in recent years, mainly in analytical and material science journals.

Borisov and Wolfbeis

solid-phase-based diagnostic devices such as those for glucose, pregnancy markers, or cardiac markers. They are better referred to as test strips. Others (mainly bioorganic chemists) often refer to molecular bioprobes as biosensors. However, true biosensors are solid state, not certain molecules, give a reading after having been contacted with the sample to be analyzed, and do not require the addition of reagent(s). It is noted at this early stage of the review that the world still does not have a fully reversible glucose sensor for in vivo use over >1 months, which would be a great relief to the 4-5% of the population suffering from various forms of diabetes and that would enable the construction of an artificial pancreas. Definitions of biosensors have been given but are diverse.28-30 However, all include the use of a biological component such as an enzyme, an antibody, a polynucleic acid, or even whole cells or tissue slices. In other words, a pH electrode capable of sensing the pH of blood is not a biosensor by all current definitions, as is xenon gas that can be used to probe the structure and dynamics of a protein.31 Certain authors confuse the terms sensor and probe; we are referring such authors to the homepage of the world’s largest manufacturer of bioprobes (www.probes.com), which never would refer to its many bioprobes as “sensors”. Over time, this has led to the undesirable situation that electronic searches for literature on biosensors result in two sets of data. The first (still larger one) is on true sensors of all kind (electrochemical, fluorescence, piezo, thermal, surface plasmon resonance (SPR), reflectometric, chemo/ bio-luminescent, IR, and the like). The second set of data is on (mainly optical) molecular probes whichsin a wrong fashionsare referred to as biosensors.

2.2. Classification of Biosensors

Otto S. Wolfbeis, born in 1947, is a Full Professor of Analytical and Interface Chemistry at the University of Regensburg, Germany. His research interests are in optical chemical sensing and biosensing, in the design of novel schemes in analytical fluorescence spectroscopy, in fluorescent probes, beads, and labels, in biosensors based on thin gold films and molecular imprints, and in the design of advanced materials for use in (bio)chemical sensing. He has authored numerous papers on optical (fiber) chemical sensors and fluorescent probes, has edited a book on Fiber Optic Chemical Sensors and Biosensors, acts as the editor of the Springer Series on Methods and Applications of Fluorescence, is the Editor-inChief of Microchimica Acta, is a member of the board of various journals including Angewandte Chemie, and acts as the chairman of the Permanent Steering Committee of the biannual conference on Methods and Applications of Fluorescence (MAF) since 1989. For details, see www.wolfbeis.de.

of approaches that have been made in the past makes it rather difficult to sort biosensors systematically. Excellent books and reviews cover the first few decades of research on biosensors,1-27 but none of those describes the state of the art as comprehensively as the one presented here, since it covers the work from the early stages to the state as of early 2007.

2. General Remarks 2.1. Definition of Biosensors Unfortunately, the terminology on biosensors is not systematic. Medical doctors tend to refer to “biosensors” as

The biosensors described in this review can be divided into two kinds of groups, viz. biosensors of the catalytic type and biosensors of the affinity type. Their features are briefly discussed in the following.

2.2.1. Catalytic Biosensors These make use of biocomponents capable of recognizing (bio)chemical species and transforming them into a product through a chemical reaction. This type of biosensor is represented mostly by enzymatic biosensors, which make use of specific enzymes or their combinations. Many wholecell biosensors also rely on biocatalytic reactions. More recently, catalytically active polynucleotides (DNAzymes) have been used as well. This type of biosensor also includes biosensors based on measurement of the rate of inhibition of a catalytic reaction by an inhibitor such as a heavy metal ion or a pesticide. Catalytic whole-cell sensors often are employed to sense sum parameters such as toxicity, antibiotic activity, or cell viability.

2.2.2. Affinity Biosensors These make use of the specific capabilities of an analyte to bind to a biorecognition element. This group can be further divided into immunosensors (which rely on specific interactions between an antibody and an antigen), nucleic acid biosensors (which make use of the affinity between complementary oligonucleotides), and biosensors based on interactions between an analyte (ligand) and a biological receptor.

Optical Biosensors

Some whole-cell biosensors act as recognition elements responding to (trigger) substances by expressing a specific gene.

2.3. General Aspects of Signal Generation, Immobilization of Biomolecules, and Sample Handling The most usual format of a biosensor is that of a biological compound immobilized on the surface of a transducer. The function of the latter is to gather the analytical information when in contact with the sample and to convert it into an electrical signal. Optical transducers respond to an analyte by undergoing a change in their optical properties, such as absorption, reflectance, emission, or a change in an interferometric pattern. Signal changes are recorded by a photodetector and, thus, transformed into an electrical signal. The development of appropriate (and stable) materials probably is more of a challenge in biosensor development than the development of appropriate spectroscopies. In all kinds of biosensors, recognition is accomplished by a biomolecule. In the overwhelming majority of biosensors, this biomolecule is immobilized on the surface of the sensor. Immobilization serves one or more of the following purposes: (a) It enables continuous sensing of analytes in flowing systems such as blood, bioreactor fluids, or water samples. (b) The biomolecule is “added” to the sample in welldefined quantity. (c) The biosensor becomes reusable or regenerable. The stability of immobilized biomolecules is a serious issue. It is noted that many articles on biosensors do not consider aspects of long-term stability in a proper way. Much more often than chemical sensors, biosensors have been combined with (micro)fluidic devices such as (micro)flow injection analyzers or lab-on-a-chip devices. Optical biosensors are particularly useful in the case of the latter, where voltages of up to several kV are applied that may disturb (or make impossible) electrochemical detection. Valcarcel and Luque de Castro32 have reviewed the state of flow-through “biosensors”, which, however, often are based on flow-injection and corresponding detectors.

2.4. Frequently Used Spectroscopies and Internal Referencing There are two main types of optical biosensors: The first exploits any changes that can occur in the intrinsic optical property of the biomolecule as a result of its interaction with the target analyte. Such changes can occur in absorbance, emission, polarization, or luminescence decay time of a receptor. Such sensors are not numerous because their sensitivity is usually low, and because many effects occur in the deep UV where spectroscopy has academic merits but is difficult to implement when it comes to analyzing complex (such as environmental or clinical) samples. An additional challenge when using intrinsic biosensors consists of the separation of the shortwave signal from background fluorescence (or absorbance). Enzymes using FAD as a coenzyme are examples of more longwave absorbing receptors that undergo intrinsic spectral changes on binding a ligand (during catalytic conversion), as are some cytochromes and hemoglobin. The second type of biosensor is making use of optical labels and probes of various kinds. This requires the

Chemical Reviews, 2008, Vol. 108, No. 2 425

biomolecule to be covalently labeled (an extra step) but enables the analytical wavelength(s) to be shifted to almost any desired value. Moreover, luminescence decay times and anisotropy can be adjusted to desired values, and effects such as dynamic or static quenching can be exploited in a more systematic manner. Not surprisingly, luminescent labels are widely used for this purpose. In recent years, the use of luminescent nanoparticles has strongly increased. Absorptiometry and reflectometry are still the most widely used methods, both in solution assays and in test strips. Absorptiometry is well-established (hardly any lab does not have a photometer) and has the unique merit of being selfreferenced (i.e., the intensity of the signal measured is always referenced to the intensity of the incoming light beam in the case of two-path photometers as used in cuvette assays or microtiter plates). Surprisingly enough, fluorescence intensity is by far the most often used analytical parameter when it comes to biosensing. According to Parker,33 luminescence intensity (I) is proportional to the concentration of an analyte present,

I ) I0clφfk where I0 is the intensity of the exciting (laser) beam,  is the molar absorption coefficient (molar absorbance) of the fluorescent probe or label, c is its concentration, l is the length of penetration, φf is its quantum yield, and k is a geometrical factor that accounts for the geometry of the optical system. This linear relationship between measured fluorescence (phosphorescence) intensity and I0 is valid only for solutions of low molar absorbance. Fluorescence intensity at a single wavelength is not referenced and obviously depends on numerous variables (and can be compromised by drifts in the photodetection system). Ratiometric (two-wavelength) measurements have, therefore, become quite widespread.34-40 This either requires addition of an inert reference fluorophore or the application of a FRET system (i.e., a donor dye and an acceptor dye). FRET systems have often been employed in immunosensors,41-44 nucleic acid sensors,40,45-47 and those based on ligand-receptor interactions.48-56 Another method for self-referencing consists in the measurement of luminescence decay time.52,57 Since the measurement of decay times in the order of a few nanoseconds (or even picoseconds) so far has required complex and expensive instrumentation (this has changed in recent years, though), labels and indicators were employed with decay times in the order of µs and ms. Decay time-based sensing is widely employed in optical oxygen sensing and in enzyme sensing based on oxygen transduction.58-60 Measurement of decay time also was reported for a fluorescent hydrogen peroxide transducer (a europium(III) complex) for use in a glucose biosensor.61 A final self-referenced method is based on measurement of fluorescence polarization,62-64 which also is independent of various variables (such as the degree of labeling, photobleaching, quenching, and solvent effects) that make other methods prone to errors. Table 1 summarizes the more important (self-referenced) methods of read-out in luminescence and their respective merits. Measurement of intensity is most common because they are easily performed, routine instrumentation is available, and one label is required only. Ratiometric (2-λ) methods are more reliable but require the availability of appropriate probes ands labels. Dual-lifetime referencing (DLR) is quite powerful, too, but requires the presence of a reference dye

426 Chemical Reviews, 2008, Vol. 108, No. 2

Borisov and Wolfbeis

Table 1. Fluorescent Schemes and Their Ability to Compensate for Interferences of Various Origin. (++): Well Compensated for; (+) Partially Compensated for; (-) Not Compensated for. These Are General Statements Only That May Be Different in Particular Situations parameter interference resulting from

intensity

2-λa

ref dye

FRET

lifetime

optical components (e.g., filters) instrumental drift (light source, detector) optical misalignment background fluorescence of sample light scatter by sensor material/ sample intrinsic color of sample dye leaching/bleaching temperature inhomogeneous dye loading

-

+ ++ + + +e ++

+b ++ + + -

+ -c + +e 0

++ ++ ++ +d ++ ++ + ++

anisotropy + ++ + ++ ++ ++

DLR + ++ ++ ++ + +e +

a Using a dual-wavelength label or probe. b Only compensated for if detected with the same set of optical components. c Not used in fiber-optic sensors. d Well-compensated only in case of time domain for probes having decay times in the µs or ms range. e Assuming both dyes to display the same temperature-dependence of their spectral properties.

with long decay time, so it is more complicated. DLR has not been applied to biosensing so far. The measurement of decay time (“lifetime”) is superior to measurement of intensity in many respects. In the case of affinity sensors, probes are needed whose decay time (that usually is not strongly affected by binding) changes upon biomolecular interactions. It has been demonstrated, for example, that the fluorescence lifetime of certain fluorescent labels (supposed to be inert in terms of changes of decay time) is a useful parameter to detect affinity binding between biotin and streptavidin and between biotinylated bovine serum albumin and streptavidin.65 Lifetime also can be determined in FRET, preferably if a long-lived donor dye is used. Refractometry also is self-referenced and has been used in immunosensors.66-68 Less common spectroscopic techniques such as reflectometric interference spectroscopy,69 optical waveguide lightmode spectroscopy, 70 supercritical angle fluorescence,71 and light scattering72 also shall only be mentioned here.

3. Enzymatic Biosensors 3.1. General Considerations Determination of such analytes as glucose, lactate, urea, ethanol, phenols, pesticides, and many others is of high significance in clinical medicine, food and environmental analysis, and bioprocess monitoring. The lack of indicators that give changes in color or luminescence at room temperature without addition of (aggressive) reagents and at nearneutral pH, in reasonably short time and in a fully reversible way, has made researchers look for alternatives. Enzymes catalyze reactions with a high degree of specificity, and the products of these reactions (or of reactants consumed) are detected directly if colored or luminescent, or by using optical transducers. The steady-state concentration of detectable species is, thus, related to the concentration of the analyte. Some enzymatic reactions require the presence of other specific reactants called coenzymes, e.g., nicotinamide adenine dinucleotide or flavine mononucleotide, which change their optical properties during the reaction. A cross section of the typical enzymatic biosensor is shown in Figure 1. An indicator layer (often sensitive to oxygen or pH) is spread over a transparent inert support, usually a polyester film. An indicator dye is either directly dissolved in a polymer matrix or, alternatively, covalently immobilized or physically adsorbed on a surface of microbeads, which

Figure 1. Cross section of a fiber-optic enzymatic biosensor. The analyte (substrate) enters the enzyme layer where it is converted into products. The indicator (sensing) layer consists of an indicator dye in a polymer layer and registers the formation of reaction products or the consumption of coreactants such as oxygen. The transparent support is inert and used only to facilitate manufacturing. It may as well be omitted. Exc and Em symbolize the paths of exciting and emitted light, respectively.

then are dispersed in the matrix polymer. The indicator layer is responsible for sensing of either cosubstrates consumed or of products produced during the enzymatic reaction. Enzyme(s) can be chemically immobilized onto the surface of a polymer membrane (e.g., cellulose, nylon, or inorganic porous glass) or physically entrapped into a polymer network, e.g., sol-gels, hydrogels, or Langmuir-Blodgett films. To avoid leaching of the enzyme, it is often cross-linked to bovine serum albumin via glutaraldehyde linkers. Alternatively, preactivated membranes may be used. When the analyte (the substrate) diffuses into the enzyme layer, it is converted into products. The indicator layer registers the formation of reaction products or the consumption of coreactants such as oxygen. In Figure 1, the sensor “sandwich” is mounted on the tip of an optical fiber that transmits excitation light from a light source to the sensor foil and emitted (reflected) light back to a photodetector. However, the majority of biosensors are not based on fiberoptics. Optical sensors that exploit chemi- and bioluminescent reactions are usually simpler because no indicator layer is required. The chemical species generated during an enzymatic process are involved in subsequent reactions that result in the production of light. In this case, other substrates (“reagents”) are needed along with the sample solution. Most chemi- and bioluminescent reactions are catalyzed by enzymes that have to be co-immobilized in the enzyme layer. Biosensors that make use of the intrinsic optical properties of the enzyme do not require optical transducers and, thus, usually include an enzyme layer placed

Optical Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 427

on a planar support or at the tip of an optical fiber, preferably in a hydrogel.

3.2. Enzymatic Glucose Biosensors Not surprisingly, this is by far the most often investigated type of biosensor. Those based on the use of glucose oxidase (GOx) function on the basis of the following reactions: GOx

D-glucose + O2 98 D-gluconolactone + H2O2 (1) D-gluconolactone + H2O 98 D-gluconate + H+ (2) The concentration of glucose thus can be related to (a) the amount of oxygen consumed,35,58,59,73-88 (b) the amount of hydrogen peroxide produced,39,61,89-102 or (c) the decrease in pH due to the conversion of D-gluconolactone to Dgluconic acid.103-105 The above equations indicate that the response of such a sensor depends on a number of variables (notwithstanding the effects of temperature and diffusion). The first is pH. If pH transduction is used, the initial pH and the buffer capacity of the sample will govern the shape and the relative signal change. pH also affects enzyme activity. The second is oxygen. Depending on how its concentration is related to that of glucose, different shapes of the response curve and different signal changes will be observed, as can be seen in Figure 2. If the sample is anaerobic (i.e., does not contain any oxygen), no signal change will be detectable. If oxygen is present in large excess, the concentration of oxygen is low, and diffusional processes are fast, hardly any signal changes will be detectable once the steady-state equilibrium is reached. It also needs to be reminded that the shapes are quite different for standing samples, stirred samples, and flowing samples. Finally, the quantity (more precisely, the activity) of immobilized GOx will strongly affect the signal change and the response time. This is shown in Figure 2 for flowing samples. The sensor is first equilibrated with a buffer solution saturated with air. The flow of buffer is then replaced by a flowing sample at time t1. At time t2, the sample is replaced by a flow of buffer again. Various curves (A-E) are obtained depending on the levels of oxygen and glucose in the sample: (A) No oxygen and no glucose in the sample; the shape of the response is mainly determined by the rate of the diffusion of oxygen out of the sensor membrane into the sample flow. The same signal level is reached (even faster) if the sample contains no oxygen but a relatively large concentration of glucose. (B) Response to a sample where [O2] , [glucose]; all oxygen in the sensor is quickly consumed as a result of enzymatic oxidation and of diffusion. (C) Air-saturated sample where [O2] > [glucose]; the shape is mainly determined by the rate of the enzyme-catalyzed oxidation of glucose. (D) Sample where [O2] . [glucose]; the steady-state signal is smaller than that in (C). (E) Sample without glucose where the pO2 is lower than at air saturation. These are exemplary plots; the shapes and steady-state intensities also depend on the activity of the enzyme, the flow rates, and the thicknesses of the various layers (and, thus, on the oxygen storage capacity). Note that the shapes for flowing samples are quite different from those obtained

Figure 2. Typical signal shapes that can be obtained if a glucose sensor based on immobilized glucose oxidase and using an oxygen sensor as the transducer (in contact with air-saturated buffer) is exposed to flowing samples containing various levels of oxygen and glucose, respectively, and then again to air-saturated buffer. See the text for an explanation of shapes and signal changes.

with standing samples and that even these can differ depending on whether they are stirred or not. Detection of glucose via the quantity of hydrogen peroxide formed appears to be the most attractive approach since it works at virtually zero background, even though it also is affected by the initial pO2 in the sample. Optical continuous sensors for H2O2 are scarce, however. The group of Luebbers106 probably were the first to describe a glucose sensor based on transduction via oxygen, which acts as a dynamic quencher of the luminescence of certain indicator dyes. The sensors consisted of an oxygen sensor (using pyrenebutyric acid as the oxygen probe) onto which GOx was deposited as a thin layer. The sensor reported glucose in physiological concentrations. The temperature dependence of the biosensor was studied in some detail. Temperature is known to exert an effect on various parameters including the rate of diffusion, the activity of the enzyme, the efficiency of quenching of the indicator by oxygen, and the quantum yield of the fluorophore used. In essence, a reduced analytical range and a steeper slope of the response curve toward glucose is observed. Diphenylanthracene (DPA) was used as a probe for oxygen in a sol-gel based glucose biosensor. The sensing material was obtained107 by controlled hydrolytic polycondensation of tetraethoxysilane (TEOS) to give a fairly inert inorganic glassy matrix whose porosity and size of pore network can be varied by polymerization conditions. Both DPA and GOx were entrapped into the sol-gel. Because the material has no absorption in the near UV and visible, it is well-suited for fabrication of optical sensor membranes. Enzymes in sol-gels can be substantially stabilized by addition of polycations.108 Other biosensors based on oxygen transduction made use of polyaromatic hydrocarbons such as pyrene, decacyclene, and their derivatives, which were dissolved in silicone.73-75,77 Following their discovery as probes for oxygen in 1986,109 ruthenium(II) complexes with ligands such as bipyridyl (Rubipy), 1,10-phenanthroline (Ru-phen), and 4,7-diphenyl1,10-phenanthroline (Ru-dpp) rapidly replaced the polycyclic aromatic hydrocarbons. They possess visible absorption, relatively long decay times (0.6-6 µs), and good photostability and, therefore, are widely used oxygen probes.35,76,78,80-88 The probe Ru-dpp is a preferred indicator because of its good brightness (Bs; defined as the product of quantum yield and the molar absorption coefficient at the excitation wavelength), which is 10 500 M-1 cm-1 at 465 nm excitation.110 In being cationic, ruthenium probes can be adsorbed

428 Chemical Reviews, 2008, Vol. 108, No. 2

onto silica gel beads (which are negatively charged at pH 7) and then be dispersed in silicone, which results in an good sensitivity to oxygen. In addition, this material is highly scattering, which increases the efficiency of collection of fluorescence. Platinum(II) and palladium(II) porphyrins represent another group of viable luminescent oxygen indicators because of their high chemical and photochemical stability, large Stokes’ shifts, good brightness, and long luminescence lifetimes. They also are often used in pressure-sensitive paints. Papkovsky58 used a phosphorescent platinum(II) complex with octaethylporphyrin (PtOEP) dissolved in polystyrene as the oxygen transducer for the glucose sensor. Luminescence intensity and decay times were measured. The sensor was applied to the determination of 0.05-1.2 mM of glucose with a limit of detection of 0.05 mM. Recently, glucose was dually sensed by immobilizing (a) a europium probe acting as a reporter for hydrogen peroxide; (b) the iridium-trisbipyridine complex as a reporter for oxygen; and (c) glucose oxidase in a hydrogel membrane.111 This sensor measures the hydrogen peroxide formed (without any background) and can compensate for variations in oxygen partial pressure in the sample, which has a strong effect on the shape of the response function (see Figure 2). Miniaturized glucose sensors are particularly attractive for a number of clinical applications, including measurements of glucose in extremely small volumes or monitoring of localized events where high spatial resolution is desired. Microsensors also are attractive because they produce less injury to patients. Rosenzweig and Kopelman82,83 designed a fiber-optic glucose microsensor in which a ruthenium oxygen probe and GOx were incorporated into an acrylamide polymer covalently attached to the surface of an optical fiber (of an outer diameter between 2 and 100 µm). The analytical range of the sensor was rather high (0.7-10 mM), but the detectable quantity of glucose was very small because of the small sample volumes needed. The group of Klimant reported on a fiber-optic flow-through biosensor for online monitoring of glucose.112 A microdialysis membrane in a Tygon tubing contained a fiber-optic sensor composed of immobilized glucose oxidase and an oxygen transducer layer, and a reference oxygen sensor was used to compensate for interfering effects. The authors have also demonstrated outstanding selectivity of the sensor, which makes use of an oxygen optode as a transducer.113 No interference was observed from ascorbic acid, acetylsalicylate, uric acid, mannitol, and dopamine in concentrations exceeding physiological levels by several folds. Measurement of glucose in humans via a sensing catheter was demonstrated. Xu et al.35 prepared luminescent probes that were encapsulated into nanoparticles to give so-called PEBBLE sensors designed for intracellular glucose imaging. The polyacrylamide nanoparticles of 45 nm diameter incorporate GOx, the oxygen indicator (a sulfonated Ru-dpp derivative), and an oxygen-insensitive fluorescent dye, Oregon Green 488dextrane, that is used as a reference for the purpose of ratiometric intensity measurements. The small size and inert matrix of these sensors allows them to be inserted into living cells with minimal physical and chemical perturbations of their biological functions. Because glucose biosensors based on oxygen transducers measure the consumption of oxygen during the enzymatic reaction, the response of such sensors to glucose is actually dependent on concentration of oxygen in the analyzed

Borisov and Wolfbeis

medium. This is shown in Figure 2 for the two extremes of ratios of concentrations of glucose and oxygen. To overcome problems associated with variable oxygen supply, oxygen should be present in large excess (compared to the amount consumed) or its concentration should be kept constant (which is not always easy to achieve) or known. Dual biosensors have, therefore, been developed and represent one possible solution to the above problem.80,85 Such sensors do contain both an oxygen-sensitive and a glucose-sensitive element located in the proximity, e.g., on a distal end of an imaging fiber.80 The glucose-sensitive element is prepared by covering the oxygen sensor with an enzyme layer. Wolfbeis et al.85 tested three different combinations of oxygen transducer and sol-gel immobilized GOx. In the first, GOx was sandwiched between a sol-gel layer doped with Ru-dpp and a second sol-gel layer composed of pure sol-gel. Such configuration provided the highest enzyme activity and the largest dynamic range (0.1-15 mM) but suffered from a distinct decrease in sensitivity upon prolonged use. In the second, which provided the fastest response time (t90 ) 50 s), a sol-gel layer doped with Ru-dpp was covered with sol-gel-entrapped GOx. In the third sensor type, both the oxygen-sensitive sol-gel powder and the sol-gel powder containing GOx were incorporated into a single sol-gel phase. Such a sensor type provided the best operational lifetime. The authors also have derived equations that describe how the effect of varying oxygen supply can be compensated for by making use of two sensors, one sensitive to oxygen only and the other sensitive to both oxygen and glucose. Since both the decay time of the luminescence of oxygen indicators (especially for ruthenium(II) complexes)114,115 and the quenching by oxygen itself116 are highly temperaturedependent, the performance of all biosensors based on oxygen transducers also is influenced by temperature. The temperature dependence of such sensors may be compensated for by making use of dual sensors for oxygen and temperature,116,117 but so far this problem has not been addressed for glucose sensors, which, therefore, need to be thermostatted. This option has not been applied to glucose sensors. The concentration of glucose may also be related to the amount of protons produced in reaction 2; however, only few optical glucose biosensors made use of pH transducers. The fluorescence of the pH probe 8-hydroxypyrene-1,3,6trisulfonate (having a pKa ≈ 7.3) contained in a protonpermeable hydrogel served as a signal to monitor pH changes during enzymatic reaction.103 The limit of detection (LOD) for glucose was 0.1 mM. Polyaniline was found to exhibit pH-sensitive spectra and, thus, was used itself as a pH transducer.105 The enzymatic reaction can be monitored at 550-650 nm (where the absorbance decreases) or at 700900 nm (where the absorbance increases). The LOD for glucose was 1 mM. An interesting approach was made by McCurley.104 A pHinsensitive fluorophore linked to cadaverine was incorporated, along with GOx, into a cross-linked acrylamide-based hydrogel placed at the end of an optical fiber. The amine moiety of cadaverine is responsible for the pH-dependent swelling of the hydrogel. When the volume of the hydrogel increases, a decrease in fluorescent intensity is observed because the total quantity of fluorophore remains constant. The sensor was operative in the range from 0 to 1.6 mM of glucose.

Optical Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 429

Finally, the amount of hydrogen peroxide produced during the enzymatic reaction can be related to the concentration of glucose. Most “sensors” rely on irreversible chromogenic reactions of hydrogen peroxide. The H2O2 transducer can be considered a biosensor itself because it makes use of the oxidation of luminol (5-amino-2,3-dihydrophthalazine-1,4dione) catalyzed by horseradish peroxidase (POx) as shown in eq 3: POx

luminol + 2H2O2 + OH- 98 3-aminophthalate + N2 + 3H2O + hν (3) The intensity of chemiluminescence (peaking at 430 nm) is proportional to the concentration of hydrogen peroxide and, in the case of the glucose biosensor, to the concentration of glucose. The method was pioneered by Freeman and Seitz,118 who immobilized POx in a polyacrylamide gel to monitor H2O2 in concentrations as low as 1 µM. The chemiluminescent reaction was widely used for determination of hydrogen peroxide93,119-121 and glucose.39,90,91,93 The sensors typically operate in the flow injection mode where GOx and POx are immobilized in a polymer membrane immersed into a solution containing luminol and sample. The problem to overcome is a mismatch between the optimal pH needed for enzymatic oxidation of glucose (neutral pH) and for enzymatic oxidation of luminol (pH ≈ 9). While an intermediate pH may be used, other possible solutions include the use of cetyltrimethylammonium bromide micelles to incorporate luminol and HPOx90 or the use of an internal solution of POx and luminol, located close to a membrane containing immobilized GOx.93 Similar to the oxidation of luminol, POx also catalyzes the reaction of H2O2 with other substrates such as homovanillic acid122 and Amplex Red.123 The products of oxidation are highly fluorescent species whose intensity can be monitored. Unlike in chemiluminescent “sensors”, the signal is not transient in these cases. Heo and Crooks102 used the POx-Amplex Red system for simultaneous determination of glucose and galactose in a microfluidic array biosensor. The enzymes (GOx and POx or galactose oxidase and POx) were entrapped in hydrogel micropatches where they show good storage stability. Amplex Red was added to the analyte solution, which was pumped over the surface of the sensor. The fluorescence of resorufin, the product of the oxidation of Amplex Red, was imaged via a conventional charge-coupled device camera. Glucose was determined in the range of 1-5 mM. By using specific enzymes located in different micropatches, several analytes can be determined simultaneously, as was demonstrated for the sensing of glucose and galactose mixtures. Production of resorufin also was monitored in a biosensor for superoxide ion, which makes use of superoxide dismutase and POx.124 Luminol also may be electrochemically oxidized by hydrogen peroxide, a reaction that does not require the enzyme POx and gives strong electrochemiluminescence (ECL). In this case, the polymer membrane containing immobilized GOx is placed on a carbon electrode,92,94,96 and the intensity of ECL is monitored from the other side. A sol-gel containing the enzyme also was coated on the surface of an electrode,99 and ECL was measured. Alternatively, the enzyme may be immobilized in a ceramic-carbon composite material.100 This graphite-containing sol-gel material was placed in a glass tube and served as an electrode to generate the ECL of luminol.

Several other kinds of H2O2 transducers were reported for use in glucose biosensors. Thus, a mixture of titanium(IV) ion and a pyridylazophenol dye was found to produce a reddish-purple product.89 Formation of a colored adduct with a dinuclear iron(III) complex was used to quantify H2O2 and glucose.95 The colored form of Prussian Blue was formed from the colorless one (Prussian White) upon oxidation by hydrogen peroxide.97,98 Wolfbeis et al.61,101 introduced a novel hydrogen peroxide transducer, which is based on the luminescent europium(III) tetracycline complex (EuTc).125 The probe is excitable by visible light and responds to H2O2 by an ∼15-fold increase in luminescence intensity. Unlike in previous methods, the determination of H2O2 does not require the addition of POx, and the reaction is fully reversible although rather slow in both directions (∼10 min). Moreover, the transducer operates at neutral pH. The large Stokes’ shift of ∼200 nm and the long-lived emission (with decay times in the microsecond time domain) enable the time-resolved suppression of fluorescent species. The probe immobilized into a hydrogel was successfully used for sensing101 and time-resolved imaging61 of glucose. Trettnak and Wolfbeis126 exploited the intrinsic fluorescence of GOx, which contains the fluorophore flavine adenine dinucleotide (FAD). FAD displays visible absorption (∼380-450 nm) and weak emission at ∼530 nm. The optical properties of the enzyme are slightly different for the reduced form (FADH2) that is produced during reaction with glucose and, therefore, could be used for analytical purposes. The sensor shows full reversibility (FADH2 is back-oxidized by molecular oxygen), but the analytical range is narrow (0.50.8 mM). The same authors used the intrinsic fluorescence of lactate mono-oxygenase for determination of lactate.127 Similarly, Chudobova et al.128 used the intrinsic absorption of GOx to obtain the sensor with a wider analytical range (1-10 mM) and limit of detection (LOD) of 2 mM. As expected, the LOD was significantly lower (0.5 mM) when the sample was deoxygenated. Sierra et al. reported on the use of the intrinisc fluorescence of GOx for the determination of glucose in serum.129 The green fluorescence of FAD is strongly quenched by serum proteins so that high concentrations of enzyme are required. Therefore, the authors suggested to exploit the UV fluorescence of the protein part of GOx, which peaks at 334 nm at λexc ) 224 nm. Even though such a biosensor is suitable, in principle, for the determination of glucose in the range from 0.5 to 20 mM, there are substantial drawbacks that include the lack of affordable (semiconductor-based) excitation light sources for 224 nm and interferences by other luminescent species. De Marcos et al.130 have investigated sensors based on intrinsic fluorescence of GOx immobilized on different polymer supports and in polymer matrixes, with respect to sensitivity, leaching of the enzyme, and sensor shelf life. The best results were achieved when GOx was immobilized on photopolymerized polyacrylamide. The sensor polymer films had a lifetime of over 2 months and adequate analytical characteristics. The linear range was between 1.67 and 11 mM of glucose. In an attempt to shift analytical wavelengths into the visible, De Marcos et al.131 have labeled GOx with fluorescein and found an increase in fluorescence intensity in the presence of glucose, probably a result of an inner filter effect. In fact, the absorption spectrum of GOx-bound FAD (but not of FADH2) overlaps that of fluorescein. Consequently,

430 Chemical Reviews, 2008, Vol. 108, No. 2

Borisov and Wolfbeis

when FAD is reduced to FADH2 during enzymatic action, fluorescence is enhanced (λexc ) 492 nm). No such enhancement is observed if GOx is labeled with Cy-5 or Texas Red. For fluorescein-labeled GOx, the linear response is from 0.55 to 5.5 mM of glucose. The labeled GOx was entrapped into polyacrylamide gel, and the sensor was used in the flow injection mode.132 Another optical biosensor for glucose makes use of the intrinsic absorption properties of horseradish peroxidase (POx), which undergo spectral changes upon binding of H2O2.133 Both GOx and POx were entrapped in a polyacrylamide gel matrix. When glucose is present, H2O2 is produced and reversibly bound by POx. The intermediate species produced during enzymatic activity display different absorption spectra between 400 and 450 nm. The sensor has a linear response between 1.5 and 300 µM of glucose. The intrinsic optical properties of enzymes also were used for determination of nitrate (absorbance of nitrate reductase)134 and nitrite ions (absorbance of cytochrome cd1 nitrite reductase)135 as well as for sensing ethanol,136 pyruvate,137 and lactate,138 using intrinsic fluorescence of alcohol dehydrogenase, lactate dehydrogenase, and lactate oxidase, respectively. The characteristics of various glucose biosensors based on the use of GOx are summarized in Table 2 along with those for other enzyme-based sensors.

3.3. Other Enzyme-Based Biosensors A number of biosensors were designed by analogy to GOxbased glucose sensors by making use of other oxidases. Enzymatic oxidation of glutamate, for example, can be described by eq 4, L-glutamate

GlOx

+ H2O + O2 + H+ 98 2-oxoglutarate + NH4+ + H2O2 (4)

where GlOx stands for glutamate oxidase. Thus, biosensors for glutamate can be based on an oxygen transducer (e.g., decacyclene in silicone)77 or on detection of H2O2 by chemiluminescence (via the GlOx/POx system).139,140 It was found that peroxidase from Arthromyces ramosus produced a 100 times stronger luminescence signal than the commonly used POx from horseradish. A glutamate biosensor based on an ammonia transducer also was reported.141 The enzyme glutaminase (GLase) was used in a glutamine biosensor in which glutamine was converted into glutamate according to eq 5, GLase

glutamin + H2O 98 glutamate + NH3

oxidized by molecular oxygen. The enzymes were immobilized on an eggshell membrane and showed a remarkable long-term stability there. When stored at room temperature over a period of 6 months, the sensor retained >95% of its initial activity. When the enzymes were immobilized in plasticized poly(vinyl chloride), the sensor lost ∼45% of its activity in 5 days. The group of Lu¨bbers also reported on biosensors for xanthine, lactate, and cholesterol using oxygen transduction. Pyrene butyric acid acted as a fluorescent probe for oxygen and was covered with a layer containing the appropriate oxidase.155 Similar to the aspartame biosensor, a dual enzymatic system was used for determination of choline-containing phospholipids (e.g., lecithin).156 Phospholipids were hydrolyzed by phospholipase D to produce choline, which was subsequently oxidized by oxygen in the presence of choline oxidase. A similar bienzymatic system was used by Kotsira and Clonis,157 however, in combination with a pH optical transducer. Production of betaine during oxidation of choline results in a change of pH and protonation of the indicator bromothymol blue. Lactate monooxygenase is more stable than lactate oxidase and was used in a lactate biosensor.158 Lactate was monitored by measurement of the oxygen consumption via the fluorescence of decacyclene and can be determined with an LOD of 0.3 mM. Its stability in sol-gel matrix can be improved by addition of polycations.108 Detection of hydrogen peroxide via chemi- and electrochemiluminescence of luminol was used in biosensors for lactate,94,96,159 ethanol,160 choline,161 lysine,119,162 sulfite,163 xanthine and hypoxanthine,140,164 and choline.96,165,166 Uric acid and D-amino acids were detected by bienzymatic systems including uricase and POx,167 and D-amino acid oxidase and POx,168 respectively. Here, thiamine was oxidized by hydrogen peroxide to give fluorescent thiochrome, whose fluorescence was detected.

(5)

Glutamate produced in the first reaction was subsequently oxidized to 2-oxoglutarate (reaction 4). Hydrogen peroxide was detected by the chemiluminescence resulting from the oxidation of luminol that was catalyzed by either hexacyanoferrate(III) ion142,143 or POx.139 To avoid interference by glutamate, which can be present in samples along with glutamine, an ion-exchange resin was used142,143 to remove glutamate, which is an anion at pH 7. Sensors that make use of an oxidase-type enzyme and an oxygen transducer also were designed for lactate,142 ethanol and methanol,145-147 cholesterol,148-150 sulfite,151 bilirubin,152 and phenol.153 An aspartame biosensor incorporated both R-chymotrypsin and alcohol oxidase.154 Aspartame is hydrolyzed by R-chymotrypsin to produce methanol, which is

Figure 3. Schematic of an electrochemiluminescent multifunctional biosensing chip: GCE, glassy carbon electrode; Pt, platinum pseudo-reference electrode; S, silicone spacer; W, plexiglass window. A solution containing different analytes is injected in each channel, a working potential of +850 mV is then applied, and the emitted light (integrated over 3 min) is detected by a CCD camera. Reprinted with permission from Marquette, C. A.; Degiuli, A.; Blum, L. J. Biosens. Bioelectron. 2003, 19, 433. Copyright 2003 Elsevier.

Optical Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 431

Table 2. Overview of Optical Enzymatic Biosensors: Abs, Absorbance; BL, Bioluminescence Intensity; CL, Chemiluminescence Intensity; EL, Electroluminescence Intensity; LI, Luminescence Intensity; LL, Luminescence Lifetime analyte acetylcholine acetylcholine acetylcholine acetylcholine ATP ATP ATP ATP ADP AMP L-alanine alcohols ethanol ethanol ethanol ethanol ethanol ethanol ethanol ethanol methanol D-amino acid aspartame captan Captan + ORP (paraoxon) bilirubin cholesterol cholesterol cholesterol choline choline choline with phospholipids fructose H2O2 H2O2 H2O2 H2O2 H2O2 H2O2 H2O2 hypoxanthine hypoxanthin glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose

enzyme AChE AChE AChE AChE f. luciferase f. luciferase f. luciferase f. luciferase hexokinase + pyruvate kinase + glucose 6-P dehydrogenase adenylate kinase + creatine kinase + luciferase L-alanineDH ADH ADH + aldehyde-DH ADH AOx AOx AOx ADH + OR + b. luciferase ADH ADH AOx + HPOx D-amino acid oxidase + HPOx R-chymotrypsin + AOx GST GST + AChE bilirubin oxidase cholesterol oxidase cholesterol oxidase cholesterol oxidase ChOx + HRP ChOx phospholipase-D + ChOx GFOR HPOx HPOx HPOx HPOx HPOx HPOx HPOx XOx + HPOx XOx + POx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx + HPOx GOx + HPOx GOx + HPOx

analytical range

transducer pH pH pH pH

oxygen H2O2 oxygen

oxygen H2O2 oxygen

oxygen oxygen oxygen oxygen H 2O 2 H 2O 2 oxygen

H 2O 2 H 2O 2

oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen oxygen pH pH pH H 2O 2 H 2O 2 H 2O 2

LOD

spectroscopy

indicator or substrate

ref

0.5-20 mM 0-20 µM 2-13 mM ? 0.1 nM-1 µM 0.1 nmol-1 µM ? ? 0.1-20 µM

0.5 mM ? 1 mM 50 µM 0.05 nM 0.1 nmol 0.1 pmol 10 pmol µm

LI ratio of LIs LI LI BL BL BL BL LI

?

25 pmol

BL

0.45-4.5 mM ? 1-100 mM 10-1000 mM 50-500 mM 3-750 µM 0.5-9 mM 0.4-70 µM 0.043-1 mM 0-1.1 mM 80 µM-60 mM 0-10 mM 0.056-3.07 mM 0-2.0 ppm 0-2.0 mM

? 0.9 mM ? ? 10 mM 3 µM 0.5 mM 0.4 µM ? ? 80 µM 0.3 µg/mL 32 µM ? ?

LI. LI. LI. LI. LI. CL LI. BL LI. LI. LI. LI. LI. Abs. Abs.

223 219 183 220 145 160 146 233 PEG-NAD+ 221 PEG-NAD 222 Ru-dpp 147 thiamine 168 Ru-dpp 154 CDNB + GSH 202 synthestic substrate 203

0.1-300 µM 0.2-3 mM 0.15-3.0 mM 0.07-18 mM 3-150 µM 10 pmol-30 nmol 0.08-3.00 g/L

0.1 µM 0.2 mM 0.15 mM 0.05 mM 3.0 µM 10 pM 0.08 g/L

LI. LI. LI. LI. CL EL. Lum. I.

Ru-dpp decacyclene Ru-dpp Ru-dpp luminol luminol Ru-dpp

0.278-331 mM ? 0.01-1 mM 0.05-1.2 mM 0.1-3 mM 17-117 µM 1-130 µM 0.5-250 µM 1-320 µM 0.5 µM-1 mM 0.5-0.8 mM 1.7-11 mM 1-10 mM 0.1-20 mM 0.1-500 mM 0.01- 2 mM 0.06-1 mM 0.05-1 mM 0.03-1.2 mM 0.1-8.3 mM 0-2.5 mM 0-20 mM 0.5-15 mM 0.7-10 mM 0.06-30 mM 0.1-15 mM 0.30-2.0 mM 0.3-5 mM 9.0-200 µM 0.1-0.8 mM 0.1-2 mM 0-1.7 mM 1-30 mM 0.25-250 nmol 0.3-300 µM 1-5 mM

0.278 mM 1 µM 1 µM 0.025 mM 0.67 mM 16.7 µM 1µM ? 0.55 µM ? 0.5 mM ? 0.5-2 mM 0.05 mM 0.1 mM 0.01 mM 0.06 mM ? 0.05 mM 0.1mM 80 µM 0.6 mM ? 0.75 mM 6 µM 0.1 mM 0.3 mM 0.3 mM 9.0 µM

intrinsic LI. CL CL CL CL CL LI. LI. CL CL intrinsic LI. intrinsic LI. intrincic Abs. LI. LI. LI. LI. LI. LI.,LL LI. LI. LI. LI. LI. LI. LI. LI. ratio of LIs LI. LI. LI. LI. Abs. CL CL CL

0.1 mM ? 1 mM 0.25 nmol 0.1 µM 0.43 mM

FITC SNARF HPTS FITC

206 34 208 207 228 229 234 235 NAD+ (coenzyme) 236 235 PEG-NAD+ NAD+ (coenzyme) NAD+ (coenzyme) NAD+ (coenzyme) Ru-bipy luminol a Ru(II) complex

152 148 149 150 165 96, 166 156

224 118 119 93 120 121 122 123 164 140 126 130 128 decacyclene 73 decacyclene 74 decacyclene 75 Ru-phen 76 decacyclene 77 PtOEP 58, 59 Ru-dpp 78 Al-ferron complex 79 Ru-ligand complex 80 Rudpp 81 Ru-phen 82 Ru-dpp 84 Ru-dpp 85 Ru-dpp 86 sulfonated Ru-dpp 35 a Ru complex 87 Ru-dpp 88 HPTS 103 Rhodamine (inert) 104 polyaniline 105 luminol 91 luminol 90 luminol 93

luminol luminol luminol luminol luminol homovanilic acid amplex red luminol luminol

432 Chemical Reviews, 2008, Vol. 108, No. 2

Borisov and Wolfbeis

Table 2 (Continued) analyte

enzyme

analytical range

LOD

spectroscopy

0.01-0.8 mM 1-5 mM 0-10 mM 60 pmol-5 µmol 60 pmol-0.6 µmol 50 µM-10 mM 0.01-10 mM 17 µM-15 mM 30-200 µM 0.05-2.0 mM 0.1-5 mM 0.1-2 mM 1.1-11 mM 0-0.55 mM 0.055-55.5 mM 1-12 µM 0.1-60 µM 1 µM-1 mM ? 0-18 µM 1-100 µM 1 µM-2.5 mM 0.5-1 mM 0.3-6.0 mM 0.02-0.5 mM 3-200 µM 0.1-1 µmol 30-300 pmol 2-50 µM 0.2-1.0 mM 11-150 mg/L 10-250 IU/L 20-120 µM 10 µM-1 mM 5 µM-10 mM 0-0.1 g/L 0.04-0.12 µM 1 nM-10 µM 1 nM-3 µM 0.3 nM-3 µM 2 pmol-1 nmol 5.5 nM-1 µM 5-500 nM 1 nM-1µM 10-500 pmol 0-1.5 µM 0.07-1.25 µM

80 µM 0.8 mM 0.3 mM 60 pmol 60 pmol 26 µM 8.16 µM 17 µM 10 µM ? 0.2 mM 0.05 mM 0.6 mM ? 55 µM 0.1 µM 0.1 µM 1 µM 0.13 µM 0.2 µM 1 µM 1 µM ? 0.3 mM ? 3 µM ? 30 pmol 2 µM ? ? 10 IU/l 20 µM 10 µM 5 µM ? 0.04 µM 1 nM 0.3 nM ? 2 pmol 1 nM 5 nM 1 nM ? 0.125 µM 0.075 µM

LI. LI. EL. EL. EL. EL. EL. Abs. Abs. Abs. LI. LL. LI. LI. intrinsic LI. LI. CL CL LI. LI. CL CL intrinsic, LI. LI. LI. LI. CL EL. LI. Abs. & LI. LI. BL Abs. CL CL LI. Abs. Refl. BL BL BL BL BL BL BL intrinsic abs. intrinsic abs.

10-380 µM

?

Abs.

p-nitrophenyl phosphate alkaline phospatase

0-40 µM

?

LI.

ORP (paraoxon) ORP carbamate pesticides ORP (paraoxon) ORP ORP (Carbaryl) ORP (paraoxon) ORP (paraoxon) ORP (paraoxon) ORP (DFP) oxaloacetate

0.5-16 µM ? 0.8-3.0 mg/L ? ? 0.1-8.0 mg/L 0.01-0.48 mM 0.8-15 µM 20-140 µM 2-400 µM 3 nM-2 µM

0.2 µM 0.1-58 mg/L 25 ng 27 ppb 2 ppm 108 µg/l 2 µM 0.8 µM 20 µM 0.05 µM 1 nM

Abs. LI. Refl. LI. Abs. Abs. Abs. ratio of LI ratio of LI ratio of LI BL

0-10 mM 0.25-10 mM 0.1-10 mM 0.3-10 mM 0.1-25mM 1-10 mM 0.5-8 mM 0.25-10 mM 0.08-40 mM

? 75 µM 0.1 mM ? 0.1 mM 1 mM ? 0.25 mM 0.08 mM

Abs. LI. LI. Abs. LI. LI. Abs. & Refl. LI. LI.

GOx + HPOx GOx + HPOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GOx GDH GDH GFOR GlOx GlOx + POx GlOx + POx GlDH + GPT GlDH GAH + GlOx GAH + GlOx + POx lactate monooxygenase lactate monooxygenase LOx LOx + HPOx LOx + HPOx LOx LDH LaDH LDH + GPT OR + b. luciferase phospholipase-D + ChOx LyOx + HPOx LyOx + POx mannitol-DH urease urease OR + b. luciferase OR + b. luciferase OR + b. luciferase OR + b. luciferase OR + b. luciferase OR + b. luciferase OR + b. luciferase nitrate reductase cytochrome cd1 nitrite reductase p-nitrophenyl phosphate alkaline phospatase

glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glucose glutamate glutamate glutamate glutamate glutamate glutamine glutamine lactate lactate lactate lactate lactate lactate lactate/pyruvate lactate lactate LDH lecitin lysine lysine mannitol Hg2+ Hg2+ NADH NADH NADH NADH NADH NADH NADH nitrate ion nitrite ion

penicillin penicillin penicillin penicillin penicillin penicillin penicillin penicillin phenol

AChE AChE AChE AChE AChE AChE OPH OPH OPH OPH malate DH + OR + b. luciferase penicillinase penicillinase penicillinase penicillinase penicillinase penicillinase penicillinase penicilinase tyrosinase

transducer H2O2 H2O2 H2O2 H2O2 H2O2 H2O2 H2O2 H2O2 H2O2 H2O2 H2O2 H2O2

NH3 H2O2 H2O2 H2O2 H2O2 oxygen oxygen H2O2 H2O2 H2O2

pH H2O2 H2O2 NH3 pH

pH pH pH pH pH pH pH pH pH pH pH pH pH oxygen

indicator or substrate Amplex Red Amplex Red luminol luminol luminol luminol luminol Ti(IV) reagent Fe(III) complex Prussian White EuTc EuTc NAD+ (coenzyme) PEG-NAD+ carboxyfluorescein luminol luminol NAD+ (coenzyme) NAD+ (coenzyme) luminol luminol decacyclene decacyclene homovanillic acid luminol luminol NAD+ (coenzyme) NAD+ (coenzyme) PEG-NAD+ BTB luminol luminol PEG-NAD+ Nile Blue pH ind. strip

p-nitrophenyl phosphate umbelliferyl phosphate AMPT indoxyl acetate chlorophenol red FITC o-nitrophenol bromcresol purple paraoxon carboxy SNARF-1 DDAO phosphate carboxy SNARF-1 bromocresol green acrylofluorescein FITC azo dye FITC FITC phenol red acryloylfluorescein Ru-dpp

ref 39 102 92 94 96 99 100 89 95 97, 98 101 61 213 222 224 141 139 140 217 218 142, 143 139 127 158 77, 144 122 159 96 214 216 221 232 157 162 119 222 210 211 225 226 227 228 230, 231 229 159 134 135 197 198 199 200 204 206 201 205 209 36 37 38 233 172 175 176 173 177 178 174 179 153

Optical Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 433

Table 2 (Continued) analyte

enzyme

L-phenylalanineDH phenylpyruvateDH LOx + LDH sorbitolDH + OR + b. luciferase sulfite sulfite oxidase sulfite sulfite oxidase + HPOx superoxide radical superoxide dismutase + HPOx urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease urea urease uric acid uricase + HPOx xanthine XOx + HPOx

transducer

analytical range

LOD

spectroscopy

indicator or substrate +

ref

0.6-6 mM 0-0.7 mM 0-0.1 mM 20 nM-10 µM

? ? 8.4 µM 20 nM

LI. LI. LI. BL

PEG-NAD PEG-NAD+ NAD+ (coenzyme)

223 222 215 233

oxygen H2O2 H2O2

0-100 ppm 1-100 µM ?

? 0.5 µM 20 nM

LI. CL LI.

perylene luminol Amplex Red

151 163 124

pH pH pH pH pH pH pH pH pH pH pH pH NH3 NH3 NH3 NH3 NH3 NH3 NH3 NH3 H2O2 H2O2

0-40 mM 0-2 mM 40-250 µM ? 0-1 mM 0-4 mM 0-100 mM 0.2-100 mM 0.06-1 M 2.0-12.0 mM 0.001-10 mM 0-400 µM 0.05-2.5 mM 0.25-8 mM 0.1-5 mM 0.01-1 mM 0.1-100 mM 10 µM-100 mM 0.1-10 mM 0.1 mM-0.1 M 3-30 mM 3.1-320 µM

? ? ? ? 20 µM ? ? 0.1 mM 0.06 M ? 2.5 µM ? ? 0.25 mM ? 0.03 mM 0.1 mM ? ? 0.1 mM 0.9 mM 2.2 µM

Abs. Refl. Refl. LI. LI. Refl. Abs. Abs. Abs. Abs. ratio of LI ratio of LI LI. Abs. LI. LI. Abs. Abs. Abs. LI. LI. CL

BTB BTB phenol red FITC NBD-PE FITC PVP Prussian Blue polypyrrole Prussian Blue FITC SNARF CF BTB HPTS Nile Blue acridine orange ETH 5350 brilliant yellow octadecyl dichlorofluorescein thiamine luminol

172 180 181 182 186 183 187 185 184 97 39 34 188 189 190 191 192 193 194, 195 196 167 164

L-phenylalanine phenylpyruvate pyruvate sorbitol

A microarray biochip for simultaneous electrochemiluminescent detection of several analytes was reported169,170 that contains the following six enzymes: GOx, glutamate oxidase, choline oxidase, lactate oxidase, lysine oxidase, and uricase. They were noncovalently immobilized (along with luminol) on anion-exchanger beads consisting of diethylaminoethyl sepharose, and the resulting beads were dispersed along with the luminol beads into poly(vinyl alcohol) bearing styrylpyridinium groups. This “cocktail” was spotted on the surface of a glassy carbon electrode, giving spots of 0.8 mm in diameter (Figure 3). The spots were allowed to polymerize under UV light. The electrochemiluminescence from the sixchannel, six-parameter biosensor was read by a CCD camera. Simultaneous measurements of glucose, glutamate, choline, lactate, lysine, and uric acid could be performed in the ranges 20 µM-2 mM, 1 µM-0.5 mM, 2 µM-0.2 mM, 2 µM-0.2 mM, 1 µM-0.5 mM, and 1 µM-25 µM, respectively. A biosensor for acetate also was developed.170 Acetate kinase (pre-immobilized on sepharose beads) and pyruvate kinase were immobilized in a layer brought into contact with the sample solution, while pyruvate oxidase was entrapped in a layer placed between the kinase layer and the glassy carbon electrode. Acetate could be measured in the range from 10 µM to 100 mM. A related microarray biochip of nine screen-printed graphite electrodes was prepared for determination for glucose and lactate.171 A reproducibility of within 4.4% was found at an optimum luminol oxidation potential of +650 mV. The LODs for simultaneous determination of glucose and lactate were 10 and 3 µM, respectively. Like in glucose biosensors, pH transducers were also used in biosensors for penicillin and urea. Penicilloate and protons are produced from penicillin in the enzymatic hydrolysis catalyzed by penicillinase. The decrease in pH is monitored by changes of absorption (reflectance)172-174 or emission

intensity175-179 of pH indicators. Fluorescein-derived indicators were used almost exclusively in order to monitor pH changes in neutral media. The pH indicator is usually contained in a hydrogel (most often a polyacrylamide gel), which is permeable for protons formed during the enzymatic reaction. A photopolymerization process was used to obtain a pH/penicillin array biosensor with spot diameters of ∼27 µm located on the surface of a 350 µm thick optical fiber.179 Penicillin and pH could be measured simultaneously (via imaging of fluorescence intensity with a CCD camera), and effects of changing pH, which often are large in complex fermentation media, could thus be compensated for. Biosensors based on pH transduction suffer from the fact that pH changes depend on the buffer capacity of the sample medium, which often is unknown and can hardly be compensated for. Urease-catalyzed hydrolysis of urea leads to formation of ammonium ions (eq 6) but also results in an increase in pH: urease

urea 98 2NH4+ + HCO3- + OH-

(6)

Consequently, two types of urea biosensors can be developed. Those based on pH transducers are designed analogously to penicillin optical sensors and make use of absorptionbased172,180,181 or fluorescent34,39,182,183 pH indicators. In contrast to most optical biosensors that rely on measurements of fluorescence intensity only, the one designed by Tsai and Doong39 employs a ratiometric scheme of self-referencing. Here, the intensity of the indicator (a fluorescein isothiocyanate-dextrane conjugate) is referenced to the pH-independent intensity of tetramethylrhodamine isothiocyanatedextrane. Such referencing makes it possible to overcome drawbacks of intensity-based measurements (for example, drifts in the intensity of the light source) but also seems to

434 Chemical Reviews, 2008, Vol. 108, No. 2

Borisov and Wolfbeis

increase the sensitivity of measurements (LOD ) 2.5 µM, compared to 20-100 µM for other urea biosensors). The ratiometric approach also was used by Yadavalli et al.,34 who prepared sensor arrays composed of poly(ethylene glycol) hydrogel microspots with a diameter of ∼200 µm containing urease labeled with a seminaphthofluorescein indicator (SNAFL-1). The ratio of the intensities for the acidic (green emission) and basic (red emission) forms of the indicator was determined using a microscope with two different sets of excitation and emission filters. Polypyrrole was found to exhibit a pH-dependent intrinsic absorption with a peak at 650 nm. It was used in an urea biosensor where it acts as a support and a pH indicator simultaneously.184 In other work, Prussian Blue (also having a pH-dependent near-IR absorption) along with the enzyme was chemically incorporated into polypyrrole films.97,185 Brennan et al.186 co-immobilized the fluorescent probe nitrobenzoxadiazole dipalmitoylphosphatidylethanolamine (NBD-PE) and urease on the surface of an optical fiber whose surface was modified with alkylamine monolayers. Alterations of pH during the enzymatic reaction result in a change of physical and electrostatic structure of the membranes, which in turn alters the emission intensity of the NBD-PE. Fluorescence intensity decreases with the degree of ionization of head groups, consistent with an increase in self-quenching. By analogy to the approach made by McCurley,104 a pHdependent swelling of a polymer also was used for sensing urea.187 A layer of poly(vinyl pyrrolidone) cross-linked with sulfonated bisazidostilbenes was coated onto a metal mirror. Protons and certain other ions cause swelling of the material, and the changes in reflectance can be monitored. The analytical range of this fully reversible sensor is from 0 to 100 mM of urea. A number of urea biosensors are based on the determination of ammonia gas produced during hydrolysis of urea.188-196 Two types of ammonia transducers were used. In the first, a pH indicator is contained in a buffer solution positioned behind a gas-permeable membrane, in which the enzyme is immobilized. Gaseous ammonia diffuses through the membrane and dissolves in the buffer. This results in an increase of pH of the internal solution and in deprotonation of the indicator. Changes in absorbance or fluorescence intensity of the indicator are related to the ammonia concentration present in the external solution and, thus, to the level of urea. This sensor type was introduced by Rhines and Arnold188 and often used in later work.189,190,194,195 A completely different scheme is utilized in the second type of ammonia transducers. Such transducers contain the NH4+-selective neutral ionophore nonactin, a proton-selective neutral chromoionophore (a pH indicator), and a lipophilic anionic counterion dissolved in a plasticized poly(vinyl chloride) (PVC) matrix. The sensor layer is covered with a gas-permeable membrane to warrant selectivity for ammonia gas by inhibiting a direct ion-exchange reaction between the sample solution and the sensor membrane. A membrane with immobilized urease is mounted on top of the sensor. Ammonia gas diffuses through the gas-permeable membrane and reaches the PVC layer, where the reaction outlined in eq 7 occurs:

IndH+ + NH3 + IP f Ind + IP-NH4+

(7)

Here, Ind and IP are the neutral chromoionophore and the

neutral ionophore, respectively. Again, both absorptionbased192,193 and fluorescent191,196 pH indicators came to use. Kawabata et al.192 manufactured a 140 µM thick urea microsensor based on this principle. Certain enzymatic reactions do not require optical transduction via a chemical sensor because optically detectable species are generated or consumed during the reaction. Such sensors usually consist of a membrane that contains the immobilized enzyme. Chromogenic or fluorogenic substrates and any cosubstrates are added to the sample into which the sensor is submerged. Because of the absence of a transducer, such biosensors are often referred to as direct optical biosensors. One type of a direct optical biosensor is based on hydrolysis of a substrate catalyzed by a hydrolase-type enzyme. The principle was first demonstrated by Arnold,197 who used immobilized alkaline phosphatase to catalyze hydrolysis of p-nitrophenyl phosphate, which results in the formation of yellow p-nitrophenolate. A linear dependence of the change in absorbance on the concentration of the substrate was observed. The method was further developed by Freeman and Bachas,198 who introduced a sensor that makes use of a competition between two substrates (4-methylumbelliferyl phosphate and p-nitrophenyl phosphate) for the active state of the model enzyme alkaline phosphatase. If the sensor was placed in a solution containing 4-methylumbelliferyl phosphate, the highly fluorescent anion of 4-methylumbelliferone was produced upon hydrolysis. In the presence of the analyte (p-nitrophenyl phosphate), the rate of fluorescence change caused by production of 4-methylumbelliferone was decreased. The method also was demonstrated to work for the determination of adenosine monophosphate, another substrate of alkaline phosphatase. A sensing scheme for determination of organophosphorous pesticides such as paraoxon199 is based on the inhibitor action of acetylcholine esterase (AChE). The biosensor makes use of a synthetic yellow substrate that is converted into a blue product by AChE. Inhibition of the reaction by pesticides is monitored spectroscopically. The LOD for paraoxon was 200 nM. Although the immobilized enzyme showed a very good long-term stability, that of the synthetic substrate was rather low in that the half-lifetime was ∼2 weeks only at room temperature. Others200 have used indoxyl acetate as a substrate for AChE. The fluorescence intensity of indoxyl (λmax ) 470 nm) was related to the concentration of the inhibiting pesticide. Similar to the work of Freeman and Bachas,198 enzymatic hydrolysis of substrate o-nitrophenyl acetate was used also for determination of organophosphates.201 Choi et al.202 developed a biosensor for captans, a group of systemic organophosphorus fungicides and pesticides. The enzyme glutathione-S-transferase (GST) converts the substrates, 1-chloro-2,4-dinitrobenzene and glutathione, into yellow S-(2,4-dinitrophenyl) glutathione. In the presence of captans, GST is inhibited and the amount of the product is decreased. A dual enzymatic system consisting of GST and AChE was shown to be suitable for simultaneous determination of both paraoxon and captan.203 The absorbance of S-(2,4-dinitrophenyl)glutathione (the product of the reaction catalyzed by GST) and R-naphthol (the product of the reaction catalyzed by AChE) was detected at 400 and 500 nm, respectively. It was observed that AChE was inhibited by both captan and organophosphorus compounds, while GST was inhibited by captan only. Thus, simultaneous detection of both analytes becomes possible.

Optical Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 435

Inhibitors of AChE can also be detected using pH optical transducers. The hydrolysis of acetylcholine chloride according to eq 8 results in the formation of acetic acid and, therefore, in a decrease in pH: AChE

acetylcholine + H2O 98 choline + acetate + H+ (8) Different absorption-based204,205 and fluorescent34,206,207 indicators were used for determination of organophosphorous and carbamate pesticides as well as of acetylcholine itself.208 All biosensors that make use of the inhibition of AChE show good sensitivity but are severely limited in specificity because AChE is the target of a wide variety of toxic inhibitors. These range from heavy metal ions to chemical warfare agents. Both organophosphate- and carbamate-based pesticides inhibit AChE. Additionally, most sensors using AChE inhibition have lengthy response times because of long incubation periods, inhibition is often irreversible, and subsequent reactivation of AChE sometimes is impossible. More recently, interest has been directed to organophosphorus hydrolase (OPH), which is not susceptible to nonspecific inhibition and offers much better specificity than AChE. OPH hydrolyzes a range of organophosphate esters, including pesticides such as paraoxon and chemical warfare agents such as soman or sarin. Catalytic hydrolysis of these compounds is accompanied by a release of protons, which makes possible determination of organophosphorus pesticides (ORP) using pH transducers.36,38 Hydrolysis of some ORP also produces detectable chromophoric products.209 An interesting approach was introduced by Simonian et al.37 The sensing scheme is far from a conventional enzymatic biosensor but resembles a competitive immunoassay. The enzyme OPH is covalently attached to the surface of a gold nanoparticle. The fluorophore 7-hydroxy-9H-(1,3dichloro-9,9-dimethylacridin-2-one)phosphate binds weakly to the active site of the enzyme. The fluorescence of the bound fluorophore is enhanced because of the proximity to the gold surface. An inhibitory substrate is added that has a much higher affinity for the active site of OPH and, thus, replaces the fluorophore. When far away from the gold surface, no enhancement of fluorescence is observed any longer. Biosensors for heavy metal ions have been reported that exploit the inhibition of urease by heavy metal ions.210,211 Urease is chosen as the enzyme since it is particularly sensitive to ions such as Pb(II), Cd(II), Ag(I), and Hg(II). In contrast to the catalytic biosensors that monitor inhibition of enzymatic activity, certain biosensors use a different mechanism of signal transduction. Walker and Asher212 designed an ultasensitive biosensor for the pesticide parathion. The sensor utilizes an array of colloidal polymer particles (which diffract light in the visible spectral region) emulgated into a polyacrylamide-based hydrogel. AChE is covalently attached to the hydrogel backbone where it irreversibly binds parathion, which in turn results in the formation of a charged product. This induces swelling of the hydrogel network and results in a shift of the wavelength of the diffracted light that is proportional to the concentration of the analyte. The sensor is capable of sensing parathion in the fM to pM concentration range. The LOD is 4.3 fM of parathion, which is several orders of magnitude lower than those for other sensors for organophosphates. As expected, ionic strength severely influences swelling properties and, thus, the performance of the sensor; therefore, 30 min of

washing with deionized water is necessary after incubation of the sensor with parathion (30 min). A number of enzymatic redox reactions require the presence of coenzymes. Most systems involve nicotinamide adenine dinucleotide (NAD+), to which a hydrogen atom and an electron can be transferred, while the substrate is oxidized according to eq 9: dehydrogenase

substrate + NAD+ 98 product + NADH (9) The formation of the reduced form of nicotinamide adenine dinucleotide (NADH) can be monitored via its characteristic absorption at λmax 350 nm and emission peaking at λmax 450 nm, and this enables the optical quantification of substrate concentrations. In early work, the dehydrogenases were immobilized on (or in) a hydrophobic polymer membrane, and the “sensors” were operated in the flow-injection mode, where the analyzed substrate and NAD+ are passed over the sensor layer. Such sensors were developed for glucose,213 lactate and pyruvate,214-216 glutamate,217,218 and alcohols183,219 using glucose dehydrogenase, lactate dehydrogenase, glutamate dehydrogenase, and alcohol dehydrogenase, respectively. Enzymatic oxidation of lactate by NAD+ results in the formation of pyruvate and NADH. The reaction thus was used not only for determination of lactate but also in the reverse direction for pyruvate,215 with NADH being supplied instead of NAD+. Evidently, the main disadvantage of such sensors relies on the fact that the coenzyme needs to be added to the sample solution. Therefore, some effort was focused on designing a self-contained biosensor, i.e., a sensor that does not require the supply of any additional components. For example, NAD+ was immobilized together with alcohol dehydrogenase in a sol-gel monolith.220 However, leaching of the coenzyme into the solution was not completely excluded, and the response of the sensor was rather slow (∼30 min). A novel approach was proposed by Scheper and Buckmann,221-223 who used a poly(ethylene glycol) molecular weight-enlarged NAD+ (PEG-NAD+) instead of NAD+. A pair of dehydrogenase-type enzymes (for substrate detection and for regeneration of the coenzyme) and PEG-NAD+ were enclosed in the sensing compartment between the ultrafiltration membrane and the fiber-optic tip. The analytes and products were allowed to diffuse freely through the ultrafiltration membrane. In contrast to previous sensors (where NAD+ was supplied in solution), this sensor type allows for the regeneration of PEG-NAD+ in a subsequent reaction such as the one in eq 10: LDH

pyruvate + NADH + H+ 98 lactate + NAD+ (10) The scheme was used for determination of glucose, lactate, ethanol, pyruvate, mannitol, formate, L-alanine, and L-phenylalanine. The unique enzyme glucose-fructose oxidoreductase (GFOR) is capable of dehydrogenating glucose to gluconolactone and of simultaneously reducing fructose to sorbitol in a ping-pong mechanism according to GFOR

glucose + fructose 98 gluconolactone + sorbitol (11) Both the intrinsic absorbance and fluorescence of NADH can be measured and enable optical detection of both substrates. In the GFOR-based biosensor,224 the enzyme was

436 Chemical Reviews, 2008, Vol. 108, No. 2

cross-linked with glutaraldehyde and placed between an optical fiber and a dialysis membrane. Glucose was sensed via the increase in fluorescence of the enzyme due to formation of NADH, and fructose was sensed via the decrease in fluorescence due to consumption of NADH. The system can be regenerated by passing fructose or glucose solutions, respectively, over it. Apart from measurements of its intrinsic absorbance or fluorescence, NADH can be detected with much higher sensitivity via reactions 12 and 13, which are catalyzed by bacterial enzymes and result in blue-green bioluminescence:

Borisov and Wolfbeis

entrapped in a collagen membrane, as well as that of adenosine monophosphate (AMP) and adenosine diphosphate (ADP) by entrapping two additional enzymes, adenylate kinase and creatine kinase, responsible for conversion of AMP and ADP into ATP. The sensitivity for ATP was significantly lower and the limit of detection was significantly higher (10 pmol) than for sensors using luciferin in solution. ADP was also determined via the fluorescence of NADH that is formed in the following sequence of reactions236 (eqs 15-17) that are catalyzed by the enzymes pyruvate kinase (PyKin), hexokinase (HexKin), and glucose-6-phosphate dehydrogenase (Glu-6P-DH):

oxidoreductase

NADH + H+ + FMN 98 NAD+ + FMNH2 (12) bacterial luciferase

FMNH2 + R-CHO + O2 98 FMN + R-COOH + H2O + hν (13) Here, FMN is flavine mononucleotide and R-CHO is a longchained aldehyde, e.g., decanal. The emission of bioluminescence peaks at 490 nm. Initially, biosensors for NADH only made use of enzymes immobilized onto a polymer support (a preactivated polyamide membrane), while the cofactor FMN and the longchained aldehyde were added to the solution to be analyzed.225-229 Attempts were made to design a self-contained biosensor that would not require the addition of coreactants and, therefore, would operate in a reagentless mode.230,231 Hence, the flavine cofactor was noncovalently entrapped in a matrix of poly(vinyl alcohol), which allowed its controlled release in the vicinity of the immobilized enzymes. The method works well but the aldehyde needs to be continuously supplied to the reaction medium. Oxidation of a substrate by a dehydrogenase-type of enzyme coupled to bioluminescent detection of NADH also was used for determination of the activity of lactate dehydrogenase232 (lactate was supplied together with FMN and the aldehyde) and for analysis of sorbitol, ethanol, and oxaloacetate.233 Compared to the ethanol biosensors based on direct detection of NADH and those using alcohol oxidase, the biosensor with coupled bioluminescent detection of NADH proved to be 1-2 orders of magnitude more sensitive, with a typical LOD being 0.4 µM. At the same time, such a system is more complicated because it makes use of three enzymes and requires cosubstrates such as FMN and an aldehyde to be added. The luciferin/luciferase bioluminescent system with its λmax of 560 nm was adapted to the determination of adenosine triphosphate (ATP). Oxidation of luciferin is catalyzed by firefly luciferase according to eq 14, firefly luciferase

ATP + luciferin + O2 98 AMP + oxyluciferin + P2O74- + CO2 + hν (14) and results in green luminescence. As in the case of bioluminescent determination of NADH, biosensors for ATP are extremely sensitive (LODs are <1 pmol). In earlier systems, luciferin again had to be added to the sample solution.228,229,234 To overcome this inconvenience, a reagentless biosensor was designed.235 Here, luciferin was incorporated into acrylic (Eudragit) microspheres entrapped in a film of poly(vinyl alcohol). Such a controlled-release system allowed the determination of ATP via firefly luciferase

PyKin

ADP + phosphoenol pyruvate 98 ATP + pyruvate (15) HexKin

ATP + glucose 98 ADP + glucose-6-phosphate (16) Glu-6P-DH

glucose-6-phosphate + NAD+ 98 6-phosphogluconate + NADH + H+ (17) ADP could be determined in concentrations as low as 0.1 µM. It was discovered recently that the europium tetracycline (EuTC) 1:1 complex can act as a luminescent probe for nucleoside phosphates including AMP, ADP, and ATP. The probe can be excited with the 405 nm laser diode and is nonspecific, but the response to the various phosphates is different. It has been applied to the determination of the activity of soluble kinases (which are important in highthroughput screening for new drugs).237 The same group has used EuTC to monitor the activity of alkaline phosphatase or the efficiency of an inhibitor by determining the amount of phosphate released by the enzyme from phenyl phosphate.238 An overview on enzyme-based biosensors is given in Table 2 along with typical data of merit. In conclusion, it can be stated that most enzymatic biosensors (a) are fairly easy to fabricate; (b) do not require labeling but a transducer capable of detecting reaction products or coreactants; (c) are suitable for continuous analyte monitoring; (d) have moderate sensitivity and limits of detection; (e) are prone to poisoning; and (f) are thermally labile (with few exceptions) and frostsensitive in aqueous solutions.

4. Immunosensors 4.1. General Remarks Affinity biosensors make use of specific interactions between an antibody (Ab) and an antigen (Ag) or a hapten. Antibodies are large Y-shaped proteins (∼150 kD) used by the immunosystem to identify and neutralize alien objects like bacteria and viruses. The affinity of Ag’s to Ab’s is very strong (Ka 1012-1014) but of strictly noncovalent nature. Binding of an antibody to its specific antigen can cause precipitation of the Ab-Ag complex, result in blocking of viral receptors, or mark the Ag for digestion by phagocytes. Smaller molecules such as pesticides or hormones often cause immune response only when attached to a large carrier such as a protein and usually are referred to as haptens. Antibodies to the hapten-carrier adduct produced by the body are able, however, to bind the hapten. It should be stressed that most

Optical Biosensors

components of physiological pathways are not immunogenic, e.g., glucose, citrate, fatty acids, amino acids, and the like. Immunosensors are mainly used for determination of concentration of antigens or haptens or, alternatively, for sensing antibodies because their presence can indicate an infection. We differentiate between immunoassays (performed in solution and not treated here except for certain examples) and immunosensors (on solid supports). The latter are treated here but actually are not sensors in their strictest definition because they are not capable of continuously and reversibly recording a parameter. Solid-phase immunoassays make use of a recognition element (Ab or Ag) immobilized on the surface of an inert support which, however, also may act as an optical fiber or a planar waveguide. Despite the fact that a binding event between an antibody and an antigen is reversible and noncovalent, most immunoreactions are irreversible in practice because of very large association constants and very slow dissociation rates. As a result, practically all immuno“sensors” are suitable for a single measurement only. This makes calibration difficult and requires an enormous reproducibility in manufacturing. Given this, attention has been paid to the regeneration of sensors (e.g., by washing with solutions of high osmolarity, high ionic strength, or low pH), which allows for multiple measurements with a single sensor. However, regeneration procedures do not always result in full recovery of the activity. In recent years, a number of devices were developed that made possible simultaneous detection of several analytes performed automatically.

4.2. Immunosensor Formats The most widely used formats are illustrated in Figure 4, where the upper panel represents the situation before immunobinding has occurred and the lower panel represents the situation after it. A selection of immunosensors for various analytes is presented in Table 3.

4.2.1. Direct Immunosensors These sensors are fairly straightforward but have been reported for a limited number of analytes only.43,44,66-68,72,239-248 The sensing format is schematically shown in Figure 4. An unlabeled antigen binds to an unlabeled antibody. Interferometric readout is common since it has the advantage of not requiring a label.66-68 The change of refractive index, however, is much smaller than in a fluorescent or radiolabel sandwich format (Figure 3c) because antigens and particularly haptens possess relatively low molecular mass. The intrinsic fluorescence of benzo[a]pyrene tetraol (BTP) was used as analytical information; the anti-BTP antibodies were immobilized onto silica microbeads.239,240 The optical signal of such single-shot probes is directly proportional to the amount of BTB captured. The LOD is 0.5 nM. Another example is represented by a biosensor for human serum albumin (HSA).241 When HSA binds to dansyl-labeled antibody attached to the surface of an optical fiber, an increase in fluorescence is observed because the antigen shielded the label from quenching water molecules. Engstro¨m et al.249 observed an enhancement of the intrinsic UV fluorescence of tryptophan of monoclonal mouse antibodies immobilized on the surface of a quartz slide upon binding maltose and panose (a rather rare triglucoside). The low affinity of the antibodies for the saccharides enabled a virtually reversible sensing, with no need for sensor regeneration. The analytical range was from 0 to 8 mM of the

Chemical Reviews, 2008, Vol. 108, No. 2 437

Figure 4. Typical formats of heterogeneous optical immunoassay; situations before (upper part) and after equilibration (lower part). The enzyme-linked immunosorbent assay (ELISA) in practically all cases is a modification of the sandwich method as it makes use of an enzyme as a label. Thus, it requires a subsequent enzymatic reaction to produce a colored or fluorescent product whose concentration can be determined, usually in solution and not on the surface of the sensor.

respective sacharides, and the LODs were 5 and 15 µM, respectively, for panose and mannose. Reck et al.250 reported a homogeneous immunoassay for thyroxine hormone. Quenching of the intrinsic fluorescence of the thyroxine-binding globulin was observed upon binding the thyroxine. Although the initial response was achieved after 5 min of incubation, almost 2 h were needed until the system reached saturation. The main drawback of this approach is its low sensitivity, since the LOD was found to be ∼100-fold higher than the concencentration of free thyroxine in serum. Other fluorescent immunosensors used in the direct format242-245 can only serve as model systems, since labeling of the analyte is necessary in the case of the assay, which is difficult (or even impossible) for real samples.

4.2.2. Competitive Immunosensors In this format (see Figure 4b),41,42,49,246,251-261 an unlabeled antigen (the analyte) and its labeled form compete for a limited number of binding sites of the immobilized antibody. Fluorescence intensity is inversely proportional to the amount of the analyte concentration. The application of the methods requires a labeled antigen to be available. The method can be inversed to enable the detection and assay of antibodies via the competitive binding of labeled and unlabeled antibodies, respectively, to an immobilized antigen.262

4.2.3. Sandwich Immunosensors Such assays (Figure 4c)60,243,261,263-272 are widely used and require relatively large antigens that contain at least two epitopes (the site of a macromolecule that is recognized by an antibody) for the antigen to be bound to the immobilized capture antibody and to the labeled second antibody. Fluorescence intensity is proportional to the concentration of the fluorescently labeled antibody, which, in turn, is related to the concentration of the antigen. Two different protocols are usual. In the “stepwise” protocol, the antigen and fluorescently labeled second antibody are added sequentially to the biosensor. In the “premixed” protocol, the antigen and antibody are premixed before injection into the biosensor. The stepwise protocol is said to give a significantly higher

analyte

assay format

analytical range

LOD

scheme

label

direct

0-9 mg/L

?

FI

dansyl/anti-HSA attached to fiber

direct direct direct

0.011-0.11 g/L 1-100 nM 10-250 nM

0.011 g/L 0.5 nM 20 nM

FI intrinsic FI FI

FITC/anti-rabbit IgG

anti-goat IgG IgG trinitrobenzene PSA LDH hemoproteins

direct direct direct direct direct direct

0.3-10 mg/L 0-5 mg/L 1-8 µg/L 1-100 µg/L 0.125-5.0 mg/L 10 nM-10µM

0.3 mg/L ? 1 µg/L 0.5 mg/L 0.03 mg/L ?

FI FI FI FI FI FI & FRET

Salmonella typhimurium

direct

?

1.03 × 105 cfu/L

ratio of FI

anti-rabbit IgG phenytoin

competitive competitive homogeneous competitive homogeneous competitive competitive competitive competitive competitive competitive competitive competitive competitive competitive

0-150 nM 1-20 µM

8 nM 1 µM

FI FI & FRET

FITC/anti-rat IgG Q-dot/protein A on optical fiber Cy-5/trinitrobenzol allophycocyanin/anti-PSA antibody FITC/LDH FITC labeled anti-hemoprotein IgG on LB monolayer complex of Alexa Fluor 546/ anti-Salmonella IgG and Alexa Fluor 594/protein G on optical fiber; FRET FITC/anti-rabbit IgG Texas red/anti-phenytoin IgG

0-300 µM

?

FI & FRET

Texas red/anti-theophylline IgG

1 nM-1 mM 0.5-200 nM 10-4-10-1 g/L 0.01-1 µM 0.1 nM-1 µM 10-1000 µg/L 1-1000 µg/L 20-200 µg/L 1-100 µg/L 50 µg/L-10 mg/L

1 nM 0.5 nM ? 5 µg/L ? 10 µg/L 5 µg/L 20 µg/L 2.5 µg/l 10 ppb

FI FI FI FI FI FI FI FI FI FI

aminofluorescein /inazethapyr fluorescein/atrazine FITC/rabbit IgG fluorescein and benzoylecgonine fluorescein/CCA Cy-5-labeled TNT sulfonate Cy-5/EDTA-TNB Cy-5/trinitrophenyl Cy-5/EDTA-RDX fluorescein/TCPB

competitive sandwich sandwich

1-50 mg/L 0-50 nM 0.03-1.2 nM

? ? 30 pM

FI FI FI

sandwich sandwich

1.5-200 mg/L 10-100 µg/L

2 mg/L ?

FI FI

Cy-5/theophylline FITC/anti-hCG IgG TRITC/ anti-botulinium toxin A IgG FITC/anti-AFS IgG NIR dye1/goat anti-human IgG

sandwich sandwich sandwich

40-300 ng/L 0.1-10 mg/L 3 × 107-3 × 1011 cfu/L 0.1-250 µg/L 0-1 mg/L 30-400 ng/L 1-1000 µg/L 5-30 µg/L 0.5-25 µg/l 2.5-10 µg/L 10-50 µg/L

40 ng/L 0.03 mg/L 107 cfu/L

FI FI FI

100 ng/L 10 µg/L 30 ng/L 1 µg/L 5 µg/L 0.5 µg/L 2.5 µg/L 10 µg/L

FI FI FI FI FI LL LL FI

theophylline inazethapyr atrazine human IgG cocaine CCA TNT TNT TNT RDX TCPB theophylline hCG Clostr. botulinium toxin A ASF protein human IgG mouse IgG LDH Salmonellax typhimurium ricin SEB SEB hCG TNT mouse IgG LDH TNT

sandwich sandwich sandwich sandwich sandwich sandwich sandwich displacement

material

assay time

ref

30 min

241

? 45 min ?

244 239, 240 245

30 min 10 min 5 min ? 4 min 30 min

242 44 246 247 243 248

5 min

43

20 min 15 min

262 41

15 min

42

2 min 10 min 15 min 15 min 20 min 4 min 5 min 5 min 5 min 20 min

252, 253 254 255 256 258 257 259 246 259 260

5 min 2 min 2 min

261 263 264, 265

40 min 30 min

266 267

Cy-5/anti-mouse IgG FITC/anti-LDH IgG Cy-5/anti-Salmonella IgG

anti-AFS IgG on Immobilon membrane goat anti-human IgG on PMMA droplet on fiber anti-mouse IgG on capillary anti-LDH IgG on fiber anti-Salmonella IgG on fiber

20 min 4 min 60 min

271 243 268

Cy-5/anti-ricin IgG Cy-5/anti-SEB IgG Cy-5/sheep anti-SEB IgG Cy-5/anti-hCG IgG Cy-5/anti-TNT IgG GOx/anti-mouse IgG GOx/anti-LDH Cy-5/trinitrophenyl

anti-ricin IgG on fiber anti-SEB IgG on PS waveguide sheep anti-SEB IgG on a capillary surface anti-hCG IgG on waveguide anti-TNT IgG on waveguide mouse IgG on PtOEPK/PS layer LDH on PtOEPK/PS layer anti-TNT IgG on waveguide

15 min 10 min 20 min 5 min 15 min 1h 1h 5 min

269 270 271 261 246 60 60 246

FITC/protein A

rabbit IgG on silica beads anti-BTP IgG on silica beds (7 µm) human IgG on an ion exchange waveguide goat IgG on a patterned waveguide anti-TNT IgG on a waveguide anti-LDH IgG on fiber

rabbit IgG on fiber phycoerythrin-phenytoin; semipermeable membrane phycoerythrin-theophylline in a well with semipermeable membrane sheep anti-inazethapyr IgG on fiber anti-atrazine IgG on fiber anti-human IgG on waveguide anti- benzoylecgonine IgG on fiber anti-CCA IgG on fiber anti-TNB IgG on fiber anti-TNT IgG on fiber anti TNT IgG on waveguide anti-RDX IgG on fiber anti- polychlorinated biphenyls IgG on fiber anti-theophylline IgG on waveguide anti-hCG IgG on waveguide anti-botulinium toxin A IgG on fiber

Borisov and Wolfbeis

human serum albumin (HSA) anti-rabbit IgG BTP protein A

438 Chemical Reviews, 2008, Vol. 108, No. 2

Table 3. Overview of Immunosensors (Acronyms Used for Schemes: FI, Fluorescence Intensity; FRET, Fluorescence Resonance Energy Transfer; LL, Luminescence Lifetime; RI, Refractive Index)

2 min 1h 30 min 10 min 20 min 15 min 15 min 15 min 15 min 15 min 15 min 12 min

ref assay time material

anti-paclitaxel on a capillary aminohexylatrazine on fiber aminohexylatrazine on fiber atrazine on the interference layer atrazine derivative on waveguide atrazine derivate on waveguide isoproturon derivate on waveguide estrone derivate on waveguide 2,4-D on fiber dichloroanilineglutaric acid on waveguide paraquat derivative on waveguide testosterone derivative on waveguide

label

Rhodamine/paclitaxel FITC/anti-triazine IgG anti-triazine IgG mouse anti-atrazine IgG Cy-5.5/anti-atrazine IgG Cy-5.5/anti-atrazine IgG Cy-5.5/anti-isoproturon IgG Cy 5.5/anti-estrone IgG Cy-5/anti-2,4-D Cy-5/anti-2,4-D Cy-5.5/anti-paraquat IgG Cy-5.5/anti-testosterone IgG

scheme

FI FI RI RI FI FI FI FI FI FI FI FI 1 µg/L 0.1 mg/L 15 mg/L 0.25 ppb ? 0.16 µg/L 0.05µg/L 0.08µg/L ? 0.03 µg/L 10 ng/L 0.2 ng/L

LOD analytical range

1-100 µg/L 0-10 mg/L 20-200 mg/L 0.1-1 mg/L 0.1-1000 µg/L 0.35-1.47 µg/L 0.11-2.83µg/L 0.17-10.7µg/L 0.3-10 µM 0.07-1.8 µg/L 0.01-100 µg/L 0.2 ng/L-100 µg/L

assay format

displacement binding inhibition binding inhibition binding inhibition binding inhibition binding inhibition binding inhibition binding inhibition binding inhibition binding inhibition binding inhibition binding inhibition

analyte

paclitaxel terbutryn terbutryn atrazine atrazine atrazine isoproturon esterone 2,4-D 2,4-D paraquat testosteron

Table 3 (Continued)

273 274 275 69 276 277 277 277 279 280 281 282

Optical Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 439

response than the premix protocol. Typical examples of fluorescent sandwich assays are listed in Table 3. Gold nanoparticles may be used to enhance the fluorescence signal of a labeled antibody. Hong et al.272 have designed a biosensor operating in a sandwich format by immobilizing protein C antibody, protein C, and secondary antibodies that were labeled with Cy-5 and Alexa Fluor 647. Their signal was typically inhanced by 10-200% when gold nanobeads were added. Self-assembled nanolayers (SAMs) can be used to control the distance between the nanobeads and the labeled antibodies. Maximal enhancement of the signal was achieved with SAM thicknesses of 2 nm. It was also found that using ethanol instead of water resulted in an up to 10-fold enhancement of fluorescence intensity.

4.2.4. Displacement Immunosensors This less common format (Figure 4d)246,273 requires an initial saturation of all the antibody binding sites with a fluorescently labeled antigen. Upon introduction of the unlabeled antigen, displacement of the labeled antigen occurs and is measured in this sensor as a decrease in the fluorescence intensity. It needs to be kept in mind that the “sensor” only works if the reaction rates of displacement are adequately fast. In fact, such sensors usually are quite slow.

4.2.5. Immunosensors Based on Binding Inhibition In contrast to other formats, such sensors (Figure 4e)69,70,274-282 require immobilization of an unlabeled analyte derivative on the surface of a waveguide. In the absence of the antigen, the labeled antibody can bind to the surface. Binding is inhibited, however, in the presence of the analyte because it blocks the binding sites (paratopes) of the antibody. For the binding inhibition assay to be quantitative, the number of high-affinity binding sites on the surface has to be much higher than the number of labeled antibodies in the bulk solution. If an antigen or an antibody is immobilized on a solid surface, the flexibility of the system becomes limited. More flexible sensors can be prepared by immobilization of an arbitrarily chosen ligand (antigen) on the surface.276 Different analyte derivatives that are conjugated to an antibody can be then attached to the ligand for assay operation in the binding-inhibition format.

4.2.6. Comparative Study on Immunosensor Formats Snapsford et al.246 compared the performance of four immunoassay formats (direct, competitive, sandwich, and displacement) that can be used for determination of 2,4,6trinitrotoluene (TNT). Anti-TNT was attached to waveguide surface via avidin-biotin linkage. Cy-5-labeled trinitrobenzene was used in direct, competitive, and displacement assay formats, and Cy-5-labeled antibody was used in a sandwich format. While TNT itself is unsuitable for detection using the sandwich assay format, it was conjugated to ovalbumin (OVA), so that the OVA-TNT complex can be bound not only to the antibody attached to the waveguide but also to the tracer antibody. Each assay format resulted in different LODs and dynamic ranges. The LOD was the lowest in the direct assay (1 µg/L); however, it also had the narrowest dynamic range (1-8 µg/L). On the other hand, the dynamic range was the widest in the competitive assay (20-200 µg/ L), but it had the highest LOD (20 µg/L). The LODs in the displacement and sandwich assay formats were 10 and 5 µg/ L, respectively. While 15 min was required to perform the

440 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 5. Principle of the total internal reflection fluorescence in an optical fiber waveguide. On reflection at dielectric interface, light penetrates into the second phase that has a lower refractive index than that of the core. Intensity decreases exponentially over the penetration depth dp (which typically is about as long as the wavelength of the light employed). Any labeled antibodies located in the declad zone within dp are excited to produce fluorescence, while those located outside this distance will not.

two-step sandwich assay, the other assays required only 5 min. It was also shown that complete regeneration of the sensor was possible within 2 min by passing a regeneration buffer containing 50% ethanol over the sensor layer. No loss of activity was observed after 10 regeneration cycles.

4.3. Preferred Optical Readout Formats in Immunosensing 4.3.1. Conventional Readout Formats In the most simple version, immunosensor spots on glass, metal, or plastic supports are read out by either absorption, fluorescence, interferometry, various methods of polarization spectroscopy, or surface plasmon resonance (treated elsewhere).283 They can be combined with methods of optical spectroscopy. Fluorescence intensity serves as the analytical parameter in most immunosensors (see Table 3) and is mostly read at a single wavelength, but this may cause difficulties in measuring reproducible data. According to Parker’s law (see section 2.4), luminescence intensity depends not only on the concentration of the fluorophore but also on other variables such as the intensity of the exciting light and the geometry of the experimental arrangment. Self-referenced methods, where the latter parameters are being referenced out, are, therefore, preferred. Measurement of intensity at two wavelengths (e.g., after addition of a reference dye or by making use of fluorescence resonance energy transfer from a donor fluorophore to an acceptor fluorophore) is one common self-referenced method. Solution immunoassays often are performed by measuring polarization, but less often in solid-state devices for obvious reasons. The measurement of fluorescence decay time represents another, albeit less common, self-referenced method.

4.3.2. Evanescent Wave, Capillary, and Other Readouts It was recognized rather early that the solid supports required in biosensors also may act as optical components. Evanescent wave spectroscopy (EWS) has become particularly useful and is often applied in immunosensors (which contrasts the situation in the case of enzymatic biosensors). EWS can be performed in various ways, but total internal reflection fluorescence (TIRF) is, by far, the most often applied. A schematic of how TIRF works is shown in Figure 5. Light transported by a waveguide (here, an optical fiber) excites the fluorescence of a label on its surface only within the evanescent field. This has several advantages: (a) the unbound labeled species in solution remain unexcited and, thus, do not form a background signal; (b) measurements can be performed in absorbing or turbid media such as most biological solutions; and (c) background fluorescence of the serum can also be largely reduced.

Borisov and Wolfbeis

Figure 6. Capillary flow sensor. The capillary acts (a) as a sample compartment (or flow-through cell); (b) as an optical waveguide; and (c) as the solid support for immobilized antibodies. Light is coupled into and out of the capillary through grating couplers. Antibodies are deposited on the inner surface of the capillary. The fluorescence of labeled antibodies or antigens is interrogated by the evanescent wave mode. Reprinted with permission from Weigl, B. H.; Wolfbeis, O. S. Anal. Chem. 1994, 66, 332. Copyright 1994 American Chemical Society.

In contrast to surface plasmon resonance and interferometric sensors, the response of TIRF immunosensors does not depend on the mass of an analyte, which makes possible detection of even small haptens. If planar waveguides are used, the fluorescence is typically collected perpendicularly to the surface of the waveguide. In most fiber-optic biosensors, the emission is, however, collected at the distal end of the optical waveguide, as shown in Figure 5. The use of capillaries284,285 for optical immunosensing is very attractive because a capillary can not only guide fluid but also light in its wall (see Figure 6). Usually, the excitation light is introduced at the end of the capillary and propagates on the inner surface as an evanescent wave. When the evanescent excitation generates a signal from an antigenantibody-fluorophore complex, the emitted light is coupled into and propagates along the capillary. At the distal end, a grating can be used to couple the light out of the waveguide into a photodetector. When the active surface area is increased inside the capillaries, the fluorescence signal integrates over their length, but the electronic background noise remains constant. Sensitivity of the sensor, thus, is significantly improved. The group of Seeger71 has designed an immunosensor that exploits changes in the supercritical angle of the fluorescence (SAF) of molecules bound to a glass surface. Because the detection volume in the aqueous sample is significantly reduced in this technique, bulk (i.e., background) fluorescence from solution is strongly suppressed. The SAF signal can be captured by a parabolic glass lens, thus leading to high spatial collection efficiency and detection sensitivity. As little as 2 pM concentrations of labeled rabbit IgG could be detected in a direct immunoassay format. The sensitivity could be further improved by using tight focusing and smaller excitation spots. In this case, however, rapid photobleaching is a serious limitation. On the other side, spot diameters of 60 µm allow for up to 200 measurements with photodegradation not exceeding 1%. This is more than adequate to obtain a smooth response plot. The group also reported on a confocal reader for biochip screening and fluorescence microscopy.286

4.4. Immobilization of Antibodies on Sensor Surfaces, and Nonspecific Protein Binding There are several widely used methods for immobilization of large proteins such as antibodies on solid support. One is

Optical Biosensors

based on the creation of a covalent linkage between the support and the protein, often via a spacer group. The surface of a support is rendered reactive with groups such as -COOH, maleinimide, -NH2, or, less often, iodoacetamide groups, isothiocyanate groups, or boronic acid. These can be reacted with amino groups, thiol groups, carboxy groups, or saccharide groups of antibodies to form covalent chemical bonds. Glass surfaces and metal oxide nanoparticles are often derivatized using silyl reagents (aminopropyltrimethoxysilane being a typical example), while gold surfaces are derivatized with thiolated reagents of the type HS-(CH2)n-X, where X is one of the reactive groups given before. Such thiols readily bind to gold to form self-assembled monolayers on its surface. Quantum dots based on metal (Zn, Cd) sulfides and selenides also are surface-modified first by making use of an appropriate thiol chemistry. Plastic materials are more difficult to derivatize unless they contain intrinsic chemical groups such as in the poly(acrylonitrile-co-acrylic acid) copolymers.287 Another important method is based on the strong affinity of biotin to (strept)avidin.67, 68, 246, 270,288 Neutravidin also has been used recently.289-291 They all have four binding sites for biotin. Typically, the surface of a biosensor is modified (as described in the previous paragraph) by introducing biotin groups, in the overwhelming majority by using di- or tri(ethylene glycol)-modified biotin of the chemical structure

biotin-CONH-(CH2-CH2-O)n-CH2-CH2-NH2 (where n ) 2 or 3). Any protein that has been modified with (strept)avidin will strongly bind to such a surface. The affinity (binding) constants of the resulting noncovalently linked conjugates are in the order of 1012 to 1014 depending on the protein and the surface (or particle) used. One may wonder why the rather affordable avidin (a glycoprotein) is used so much less often than the rather expensive streptavidin (not a glycoprotein). On the other side, streptavidin has an isoelectric point (pI) of 5 and is less prone to nonspecific binding as compared to avidin with its pI of 10.5. Proteins also can be immobilized via a polyhistidine tag. The method is based on an amino acid motif in proteins that consists of at least six histidine (His) residues. It also is known as hexahistidine tagging, or 6xHis-tag, or by the tradename His-tag. The polyhistidine tag can be used for the immobilization of proteins on a nickel- or cobalt-coated microtiter plate, on glass, or on another protein. The most simple way to add a poly-His unit to a protein is to insert a protein DNA in a vector encoding a His-tag so that it will be automatically attached to one of its ends. The other technique is to perform a PCR with primers that have repetitive histidine codons next to the start or stop codon in addition to several (16 or more) bases encoding specifically to the protein to be tagged. Another widely used method (with particular applicability to the immobilization of antibodies) is based on the use of protein A. This is a 40-60 kD surface protein that strongly binds to immunoglobulins from many mammalian species. Specifically, it binds to the Fc region through interaction with the heavy chain and, thus, does not strongly compromise its affinity to the respective antibody. Shriver-Lake et al.292 investigated different heterobifunctional cross-linkers for covalent attachment of antibodies through thiol-terminated silanes onto glass, silica, silicone, and other surfaces. A variety of cross-linkers were found to be suitable for effective immobilization of antibodies. The

Chemical Reviews, 2008, Vol. 108, No. 2 441

use of carbohydrate-reactive cross-linkers resulted in immobilized antibodies having higher activities than when using reactive succinimide residue but required a more complex procedure, which implied the risk of denaturation of some antibodies. Preininger et al.293 have investigated three different types of polymer supports used for immobilization of antigens and antibodies with respect to specific binding and nonspecific binding (NSB, better referred to as nonspecific adsorption) and regeneration of the sensor. Interaction of rhodaminelabeled anti-human IgG with immobilized human IgG was used in a direct assay format for investigation of specific binding. The degree of NSB was determined using antihuman IgG and was found to be quite high (∼80% of the level of specific binding) for human IgG immobilized onto a sol-gel support. Carboxylated poly(vinyl chloride) (PVC) support showed ∼50% of nonspecific absorption, while the NSB to polystyrene was negligible. However, only 35% of initial signal was regenerated when polystyrene support was used, while the regeneration level of 50% was possible for carboxy-modified PVC. Liu et al.294 have shown that, when using polystyrene fibers, the extent of NSB can be significantly reduced by introducing a spacer between the polystyrene surface and the photoimmobilized antibody. Poly(ethylene glycol) crosslinkers with five ethylene glycol units decrease the degree of NSB dramatically, and additional treatment of the surface with BSA eliminates it completely. The authors also showed that the “stepwise” protocol of sandwich assay formats resulted in a much higher sensitivity than when using the more convenient “premix” protocol. NSB also can be significantly reduced by using a dextrane linker.295 Carboxymethylated dextrane was attached to the surface of a fiber-optic waveguide whose surface was treated with aminopropyltriethoxysilane; this was followed by carboxamide formation using activation with EDC and formation of a reactive NHS ester. The same method was applied to covalently immobilize an antibody to dextrane. The amount of NSB was shown to be only 2% of the level obtained for the untreated glass chip. Different immobilization techniques for glass fibers were investigated by Tedeschi et al.296 Immobilization of antibodies via glycidyloxypropyl-trimethoxysilane-dextrane resulted in the highest density of active sites. The Langmuir-Blodgett (LB) technique represents an alternative for immobilization of antibodies.255 Protein A has a specific affinity for a specific section of IgG and can be prepared as a stable monolayer by the LB film technique. Such a monolayer was immobilized on an alkylsilanized hydrophobic synthetic quartz plate. Anti-human IgG antibody was self-assembled on the protein A film. Rabbit IgG labeled with fluorescein isothiocyanate was used in a competitive assay for determination of human IgG over the analytical range from 10-4 to 10-1 g/L. Anderson et al.297 showed that the sensitivity of fluorescent immunoassays for determination of antigens was similar when the antibody was covalently attached to the support or via protein A.

4.5. Specific Examples of Immunosensors 4.5.1. Biosensors for Proteins and Antibodies Barnard and Walt251 developed a kind of reversible immunosensor for continuous measurements of IgG over a prolonged period of time using a controlled-release system. The fluorescein-labeled antibody and the Texas Red-labeled

442 Chemical Reviews, 2008, Vol. 108, No. 2

antigen were separately incorporated into an poly(ethyleneco-vinyl acetate) matrix, allowing for controlled release of the components. The analyzed IgG from the media and the two released biocomponents diffuse into the reaction chamber, where the competitive immunoreaction occurs. FRET is observed in the absence of IgG but is suppressed once the complex between fluorescein-labeled antibody and unlabeled IgG has been formed. Fluorescence intensity is monitored with the help of an optical fiber located above the reaction chamber. The steady-state rate of release is achieved after 2 days, allowing continuous monitoring of 0-500 mg/L of IgG over a period of 1 month. The approach, thus, can be applied to situations where continuous monitoring of analyte is required over prolonged periods of time and where rapid response (limited here by the diffusion of the analyte from the bulk solution or the release rate) is not an issue. Lepesheva et al.248 developed a FRET assay for hemoproteins in the analytical range from 10 nM to 10 µM. In the direct biosensor, fluorescein-labeled anti-hemoprotein IgG was absorbed onto a LB monolayer contained on a quartz support. Fluorescence intensity decreased in the presence of hemoprotein as the result of the quenching of the fluorescence of the label by heme. Luminescent quantum dots are viable optical markers44 and have been used in a direct assay for IgG. Protein A was labeled with CdSe/Zn Q-dots with a fluorescence λmax of 655 nm and then was immobilized at the tip of an optical fiber. Once the immunoreaction with IgG occurs, a decrease in fluorescence intensity is observed as a result of FRET from the Q-dot to the bound protein. Enzyme-linked immunosorbent assays employ enzymes as labels. Several attempts were made to design biosensors based on the use of enzyme labels. Papkovsky et al.60 demonstrated the feasibility of an enzyme-linked biosensor with the IgG-anti-IgG model system. Mouse IgG antigen was absorbed onto a surface of a glass fiber membrane combined with an optical oxygen transducer (a phosphorescent metalloporphyrin contained in polystyrene). Mouse IgG was detected in a sandwich ELISA using anti-mouse IgG antibodies labeled with glucose oxidase (GOx) as the secondary antibody. The amount of the antigen was quantified by measurement of the consumption of oxygen that results from the enzymatic reaction in the presence of glucose. The (rather long) luminescence decay time of the oxygen probe was monitored. This is in contrast to immunosensors based on the measurement of fluorescence intensity. A glass cover was used to limit oxygen access, and this significantly improved sensitivity, the LOD being 0.5 µg/L. Lactate dehydrogenase (LDH) was detected analogously using anti-LDH antibodies labeled with GOx. As little as 2.5 µg/L ()10 pM) of LDH could be sensed. The assay time was 1 h. Kim et al.298 made use of microbeads (made from a modified acrylamide) of ∼1 µm diameter that were assembled onto an amino-functionalized glass surface. A network of biotin and anti-biotin couples was attached to the beads (that also act as microlenses) via photopolymerization with aminobenzophenone. In the absence of the analyte, the immobilized antigens and antibodies interact with each other, which results in microspheres that are in the “on”state (Figure 7). The interactions between the attached antigens and antibodies are disrupted when a sample containing antigen (biocytin) is introduced. The microlenses are transformed from the “on”-state to the “off”-state as a result

Borisov and Wolfbeis

Figure 7. Microscope images of a microlens-based optical immunosensor array. Its sensitivity can be tuned by changing the concentration of the antibody used in the cross-linking stage. The concentrations of antibiotin were (a) 6.7 µM, (b) 2 µM, (c) 1 µM, and (d) 0.6 µM. With decreasing concentrations of antibiotin, the microlenses are inversed at lower concentration of biocytin. Reprinted with permission from Kim, J.; Singh, N.; Lyon, L. A. Angew. Chem., Int. Ed. 2006, 45, 1446. Copyright 2006 Wiley.

of gel swelling once the concentration of the analyte exceeds a certain critical value. The changes can be monitored microscopically. The sensitivity can be tuned by changing the concentration of the antibody used in the photochemical cross-linking stage. When the hydrogel microlenses are prepared with an excess of binding pairs above the critical point, they swell only after a suitably large number of displacement events have occurred. However, if the number of cross-linked units is just slightly above this critical point, only a few displacement events will result in gel swelling. The system was demonstrated to be fully reversible as the microlenses return to the initial “on” state when the antigen is removed by washing the sensor with phosphate-buffered saline.

4.5.2. Biosensors for Toxins In the integrating capillary biosensor described by the group of Ligler et al.,271 antibodies (anti-mouse IgG or sheep anti-staphylococcal enterotoxin B (SEB) were coated onto the entire inner surface of the capillary. Immunosensing of mouse IgG and SEB was accomplished in a sandwich format using antibodies labeled with Cy-5. Compared to conventional fiber-optic biosensors and planar waveguide-based about ∼2 orders of magnitude (40 and 30 ng/L for IgG and SEB, respectively). The analytical range of the sensor is from 40 to 300 and from 30 to 400 ng/L of IgG and SEB, respectively. Moreover, multianalyte detection can be attained by passing the sample through multiple capillaries, each coated with a different antibody, either sequentially or in parallel, depending on the amount of sample available. An ELISA type of biosensor for antibodies against cholera toxin B (CTB) was developed by Konry et al.299 An electroconductive surface was created on a fiber-optic waveguide by coating it with indium tin oxide to allow surface electropolymerization of biotin-pyrrole monomers. Biotin-conjugated CTB was attached to the surface using avidin. Anti-CTB was quantified via a competitive assay format in which the sensor was incubated first with the analyzed antibodies and then with goat anti-rabbit IgG labeled with horseradish peroxidase. Fibers were then immersed into a solution of luminol and oxidizing agent (a standard kit), and chemiluminescence was monitored. The

Optical Biosensors

LOD of the sensor is said to be 1 × 106 “titers”. The total assay time was 65 min. In another biosensor for anti-CTB, the antigen was attached to the surface of the optical fiber via a photoreactive benzophenone derivative,300 and the same LOD was achieved. An analogous system was developed for detection of antibodies against anti-West Nile virus IgG,301 and the LOD was similar. A 2-fold increase in sensitivity was observed compared to a chemiluminescent ELISA, presumably because light emission occurs near the optical fiber, which enhances the efficiency of light collection.

4.5.3. Biosensors for Drugs A capillary displacement-type immunosensor for the drug paclitaxel was designed by Sheikh and Mulchandani.273 The anti-paclitaxel antibodies were attached via protein A to the silanized inner walls of a glass capillary, and the binding sites were saturated with rhodamine-labeled paclitaxel. Although fluorescence intensity of the displaced labeled antibodies was monitored in another chamber, thus taking no advantage of the light-guiding properties of the capillary itself, the sensor still proved to be very sensitive. In fact, the detection limits were 1 and 4 ng/mL at flow rates 0.1 and 1 mL/min, respectively. The assay time ranged from 2 min at 1 mL/min to 8 min at 0.1 mL/min. Moreover, the regeneration of the capillary column was possible without affecting the performance of the biosensor. Anderson and Miller41 developed a homogeneous immunoassay for the anticonvulsant drug phenytoin where Bphycoerythrin-labeled phenytoin and Texas Red-labeled antibody were contained in a 200 µm cellulose dialysis tube connected to the distal end of an optical fiber (see Figure 8). The two species form a complex in which fluorescence resonance energy transfer (FRET) occurs from phycoerythrin to Texas Red. The interaction thus results in the quenching of the fluorescence of phycoerythrin. Phenytoin is small enough to diffuse through the dialysis membrane, where it displaces some of the phycoerythrin-phenytoin conjugates in the complex. The increase in the fluorescence of phycoerythrin is, thus, proportional to the concentration of phenytoin. This biosensor, notably operating in a fully reversible way(!), was suitable for determination of phenytoin with an LOD of 5 µM and a measurement time of 15 min. Later, the system was optimized to detect phenytoin in concentrations as low as 1 µM.49 A similar system was used for determination of theophylline.42 This biosensor design is of wider interest because full reversibility is achieved. However, the system is applicable only to antibodies having high reverse rate constants, which is not usually the case (see Figure 8).

Figure 8. Schematic representation of the reversible competitive immunosensor for phenytoin. Phenytoin and the phycoerythrinphenytoin conjugates competitively bind to the antiphenytoin-TR complex. Redrawn with permission from Anderson, F. P.; Miller, W. G. Clin. Chem. 1988, 34, 1417. Copyright 1988 American Association for Clinical Chemistry.

Chemical Reviews, 2008, Vol. 108, No. 2 443

4.5.4. Biosensors for Bacteria Cells Ko and Grant developed a FRET-based immunosensor for determination of Salmonella typhimurium.43 S. typhimurium antibody was labeled with a donor dye, while protein G was labeled with an acceptor. Both were immobilized on the surface of an optical fiber. In the absence of the antigen, very little fluorescence is observed from the acceptor. FRET occurs as a result of conformational changes when the antigen binds to the antibody. The ratio of the fluorescences of donor and acceptor serves as the analytical information. Concentrations of the bacterium as low as 1 × 105 cfu/L can be detected in 5 min. Antimicrobial peptides (AMPs) were recently shown to be suitable as recognition elements for microbial cells. AMPs are produced by many organisms for protection against invasion of harmful microbes. AMPs recognize microbes by interacting with their membranes, which then are destroyed. Several immunosensors reported recently302,303 rely on this recognition ability of AMPs. A direct immunosensor302 for Escherichia coli and Salmonella typhimurium was demonstrated to work with the AMP magainin I, which was either covalently attached to the patterned microscope slide surface or immobilized via biotin-avidin chemistry. Cy-5-labeled cells could be detected with detection limits comparable to analogous antibody-based assays. The LODs for E. coli and S. typhimurium were 1.6 × 105 and 6.5 × 104 cfu/mL, respectively, in the case of covalently immobilized magainin, and 6.8 × 105 and 5.6 × 105 cfu/mL, respectively, for the AMP immobilized via biotin-avidin. The assay time was 70 min. AMPs also were applied in the more-practical sandwich types of assays.303 Here, the immobilized AMPs were used to capture the unlabeled targets, while detection of bound cells was accomplished using fluorescently labeled antibodies. A significant degree of nonspecific binding was found in the case of tracer antibodies labeled with Cy-5 and Alexa Fluor 647 dyes. Replacement of the marker to Cy-3 was found to significantly reduce the amount of nonspecific binding. It was also found that high peptide density was necessary for optimal results. Limits of detection for E. coli and S. typhimurium were 5 × 104 and 1 × 105 cfu/mL, respectively, when magainin I was used, and 2-10-fold higher with other peptides. Martinez et al.304 reported on a biosensor for the protective antigen (PA) and for cellular components of Bacillus anthracis using SiONx waveguides. The sensor can detect 83 µg/L (i.e., 1 picomolar concentrations) of PA in a complex fluid within 10 min when operated in a sandwich assay format, but it possibly can become even more sensitive if interferences by background fluorescence and nonspecific binding can be further reduced. When whole cells are monitored immunologically by optical methods, the use of ultrasonic standing waves significantly improves the sensitivity of a biosensor. Zourob et al.72 showed that ultrasonication for 3 min enhanced sensitivity of detection of Bacillus subtilis cells by 2 orders of magnitude because the diffusion-limited capture rate is replaced by much faster cell movement. Rabbit anti-B. subtilis antibodies were immobilized on the surface of a metal-clad leaky waveguide. Evanescent light-induced scattering was detected by a CCD camera. Obviously, fluorescent labeling was not required, and the analytical range of the sensor operated in the direct assay format was from 1 × 106 to 1 × 1012 cfu/L.

444 Chemical Reviews, 2008, Vol. 108, No. 2

Borisov and Wolfbeis

4.5.5. Biosensors for Pesticides A biosensor for the pesticide atrazine made use of polystyrene nanobeads dyed with a luminescent europium(III) chelate.278 Beads containing carboxy groups were covalently coupled to atrazine antibodies, which, in turn, were used in a binding inhibition assay format. An indium tin oxide (ITO) waveguide was used for immobilization of the capture analyte derivative to minimize nonspecific binding of the beads. Particles with diameters of 107, 304, and 396 nm were investigated in some detail. A decrease in their size resulted in faster binding but did not increase the assay sensitivity, which was comparable with the sensitivity of a known ELISA for atrazine. The use of such labels (having decay times of 0.1-1 ms) enables an almost complete elimination of background fluorescence by applying time-resolved measurements on the ITO waveguide.

4.5.6. Multianalyte Biosensors Lately, significant effort was devoted to the development of automatted biosensor devices capable of simultaneous immunosensing of several analytes of importance in environmental monitoring but, in particular, for the determination of (bio)chemical warfare agents and explosives. A compact, portable, multichannel fiber-optic instrument named MANTIS (of 5.5 kg weight)270 was reported to be capable of automatically transporting sample, buffer, and labeled antibodies to fibers and to perform fluorescence measurements. It enables four fluorescence immunoassays to be performed simultaneously on the surface of miniaturized polystyrene fiber-optic probes. The device was demonstrated to work for determination of staphylococcal enterotoxin B (SEB). Antibodies to SEB were immobilized on the surface of a polystyrene waveguide through avidin-biotin bridging. Cy5-labeled SEB antibodies were used in the sandwich assay format. The analytical range of the biosensor was from 10 µg/L to 1 mg/L, and the time required was 10 min. The improved version of the MANTIS device (termed RAPTOR) was shown to be able to determine 105 cfu/mL of Bacillus globigii, 107 cfu/mL of Erwinia herbicola, and 109 pfu/mL of MS2 coli phages simultaneously.305 Analogously, the RAPTOR system can determine staphylococcal enterotoxin B, ricin, Francisella tularensis, and Bacillus globigii simultaneously.305 In other work,306 plastic capillaries with immobilized antibodies were used for simultaneous determination of hormones prolactin, follitropin, and human chorionic gonadotropin (hCG). Fluorescein-labeled antibodies were used in a sandwich assay format. The detection limits were 1.3 µg/L, 2.3 IU/L, and 3.6 IU/L for prolactin, follitropin, and hCG, respectively. Feldstein and others at the Naval Research Laboratory307 developed an automatted array biosensor for determination of biological warfare agents. The central element of the sensor is a planar optical waveguide (Figure 9) used for direct excitation of antibodies that are bound to the waveguide surface within the penetration depth of the evanescent field. A physically isolated patterning method has been developed to manufacture an array of recognition elements (each ∼1 mm2 in size). In this technique, the multichannel patterning cell is placed on the prefunctionalized surface of the waveguide. Then, the recognition species (e.g., antibodies) are introduced into appropriate channels and are patterned on the waveguide surface during incubation, after which the cells are removed. The sensor is then used in a sandwich format.

Figure 9. Optical system of an array biosensor. Light from a diode laser is coupled into the waveguide. Fluorescence from the waveguide surface is focused by a graded index lens array (GRIN) through optical filters onto a Peltier-cooled CCD imaging array. Reprinted with permission from Feldstein, M. J.; Golden, J. P.; Rowe, C. A.; MacCraith, B. D.; Ligler, F. S. J. Biomed. MicrodeV. 1999, 1, 139. Copyright 1999 Springer Science and Business Media.

Simultaneous determination of the explosives 2,4,6trinitrotoluene (TNT) and hexahydro-1,3,5-trinitro-1,3,5triazine (RDX) was performed by a system named “Analyte 2000” developed at the Naval Research Laboratory.259 Two probes for determination of TNT were coupled with two probes for RDX. Cy-5-labeled analyte derivatives were used in a competitive immunoassay format to determine as low as 5 µg/L of TNT and 2.5 µg/L of RDX. Only a minimal cross-reactivity for the two haptens was observed in the multianalyte immunosensor, which was, therefore, capable of analyzing samples containing mixtures of the two compounds. The same array biosensor307 was used (a) for simultaneous screening of human serum for antibodies against bacterial and viral antigens including Staphylococcus aureus enterotoxin B, tetanus toxin, diphtheria toxin, and hepatitis B, with detection limits from 0.2 to 3 µg/mL and an LOD as low as ∼100 fg;308 (b) for the mycotoxin deoxynivalenol (LOD ) 0.2 ng/mL in buffer);309 (c) for simultaneous determination of large food pathogens such as Campylobacter jejuni (LOD ) 500 cfu/mL) and of small toxins such as aflatoxin B1 (LOD ) 0.3 ng/mL);310 (d) for staphylococcal enterotoxin B and botulinum toxin A (with LODs of 0.1 and 20 ng/mL, respectively);311 (e) for ochratoxin A in cereals and beverages (LODs ranging from 4 to 100 ng/g);312 (f) for Salmonella typhimurium (LOD ) 8 × 104 cfu/mL within 15 min and 8 × 104 cfu/mL within 1 h);313 (g) for the aggressive Escherichia coli species O157/H7 in food samples (with LOD of 5 × 103 cfu/mL in buffer and 1-5 × 104 cfu/mL in spiked food matrixes, and an assay time of 30 min only);314 and (h) for Campylobacter and Shigella species in food (LOD ) 9.7 × 102 and 4.9 × 104 cfu/mL, respectively).315 These articles reveal that the standard 6 × 6 array sensor can be used to analyze six samples for up to six different analytes. Taitt et al. showed316 that the same format is suitable for analyzing a single sample for 36 different analytes by using complementary mixtures of capture and tracer antibodies. Mixtures were optimized to allow detection of closely related species without significant cross-sensitivity. The approach was demonstrated to work when analyzing a sample for 9 targets with a simple 3 × 3 array. The only limitation of the approach is that the quantity of reagents needed increases significantly. Several other array biosensors suitable for simultaneous immunosensing were developed. The RIANA (river analyzer)280 array biosensor is based on a TIRF arrangement that was combined with a flow injection technique for

Optical Biosensors

automated and reproducible fluid handling. Herbicides can be detected by a heterogeneous binding inhibition immunoassay. Analyte derivatives were attached via an aminodextrane linkage to defined detection spots (L 3 mm) located on a glass waveguide, with a 2.5 mm distance between the spots. Analyte-specific antibodies labeled with a fluorescent marker were preincubated with analyte present in the sample, and the remaining (nonblocked) antibodies were then selectively attached to the spots. Each test cycle includes regeneration with hydrochloric acid of pH 1.9 and washing with an acetonitrile/water/propionic acid (49/50/1) mixture and is finished within 15 min. Because the glass surface included three sensitive spots, simultaneous detection of three analytes was possible. The following pesticides were detected: atrazine, simazine, isoproturon, alachior, 2,4-dinitrophenoxyacetic acid, and pentachlorophenol. The respective LODs were 0.03, 0.03, 0.11, 0.07, 0.07, and 4.23 mg/L. In related work,277 the pollutants atrazine and isoproturon and the hormone and endocrine disruptor estrone were determined simultaneously with LODs of 0.155, 0.046, and 0.084 µg/L using the binding inhibition assay format. No interference in the analysis of target compounds was observed upon simultaneous quantification. It was shown later317 that much lower concentrations of the pollutants can be detected by the RIANA biosensor. The highest standard deviation observed at very low analyte concentration results from the inaccuracy of the dilution procedures that require up to 11 steps. Much lower standard deviations were observed when stock solutions were used for each concentration. The errors resulting from dilutions can be compensated for by using a statistical method. The LODs for atrazine, estrone, and isoproturon were 0.002, 0.019, and 0.016 µg/L, which is ∼1 order of magnitude lower than if the statistical method was not applied. The RIANA array biosensor also was employed for monitoring testosterone in water samples (LOD 0.2 ng/L),282 progesterone in water (LOD 0.37 ng/L318 and 0.2 ng/L319), progesterone in milk (LOD 45 ng/L),319 the pesticide propanil in aqueous samples (LOD 0.6 ng/L),320 and other antibiotics, hormones, and endocrine-disrupting chemicals with similar LODs.321 A completely different approach toward array immunosensors was developed by Rissin and Walt.322 The underlying scheme of this type of microsensor is similar to the one used in DNA array sensors that were developed by the same group. In a typical example, antibodies to lactoferrin and IgA were covalently immobilized on the surface of 3 µm sized poly(methylstyrene) microspheres. These, in turn, were positioned onto the array of ∼50 000 individual optical fibers. A luminescent europium(III) chelate in two different concentrations was applied for encoding purpose, i.e., to establish the position of the two types of microbeads. Lactoferrin and IgA were determined in a sandwich assay format, with the secondary antibodies being labeled with Alexa Fluor 546. IgA can be determined in concentrations between 700 pM and 100 nM, while for lactoferrin the range is between 385 pM and 10 nM. While simultaneous determination of only two analytes was demonstrated, the approach is likely to be suitable for multianalyte sensing. In conclusion, it can be stated that immunosensors (a) are versatile because they enable the determination of highly different species that range from haptens to viruses and cells; (b) display excellent sensitivity and have very low limits of detection; (c) are highly specific; (d) act irreversibly and,

Chemical Reviews, 2008, Vol. 108, No. 2 445

therefore, are not suitable for continuous sensing in most cases; (e) are less prone to poisoning than enzyme-based biosensors; (f) are thermally labile and frost-sensitive in aqueous solutions; and (g) can be multiplexed.

5. Biosensors Based on Ligand−Receptor Interactions This type of affinity biosensor makes use of specific interactions between a biological receptor and an analyte. Reports on receptor-based biosensors are less numerous than those on immunosensors for the following reasons. This is mainly due to the fact that working with receptors is limited by the facts that (a) their isolation often is tedious ands unlike the production of antibodiessrequires individual protocols for each receptor; (b) they need the natural (cellular) environment to function best; (c) they are not stable over time; and (d) their molecular diversity requires an individual labeling protocol for each single species.

5.1. Receptor-Based Biosensors for Saccharides and Glycoproteins The first optical receptor-based biosensors were developed for sensing glucose and made use of its specific interaction with the glucose-binding protein concanavalin A (ConA).323-325 The approach (see Figure 8 for a closely related scheme) is similar to competitive immunoassays, with the notable exception that it acts reversibly. ConA is immobilized on the inner wall of a mm-sized hollow dialysis chamber via a 1,6-hexanediimine glutaraldehyde spacer. The chamber is placed at the distal end of an optical fiber. Fluoresceinlabeled dextrane is contained in the solution filling the chamber. In the absence of glucose, it binds to ConA. Unlike the large ConA, glucose can freely diffuse through the membrane and bind to ConA, and this results in the displacement of dextrane. The released fluorescein-labeled dextrane is distributed within the chamber and “seen” by the fiber if located in the cone determined by the numerical aperture. Because the walls of the chamber are located out of the aperture of the fiber, no fluorescence is registered in the absence of glucose. The sensor operates in the 0-50 mM glucose range and has a response time of ∼7 min. A FRET sensor based on the same principle was designed later.48 Here, the interaction of fluorescein-labeled dextrane with rhodamine B-labeled ConA resulted in a decrease in fluorescence intensity because of the more efficient FRET. Aggregation of ConA is prevented by chemical modification of the protein with succinic anhydride followed by labeling with the rhodamine and results in more stable calibration plots.326 Concentrations of glucose as high as 0.08 M could be analyzed. Russell et al.327 provided another solution for the FRET system. Concanavalin A was labeled with the FRET acceptor dye tetramethylrhodamine isothiocyanate (to give TRITCConA) and covalently immobilized in photopolymerized poly(ethylene glycol) hydrogel spheres with an average diameter of ∼2 µm. The FRET donor dye fluorescein isothiocyanate dextrane (FITC dextrane) was physically entrapped in the hydrogel. It can bind TRITC-Con A in the absence of glucose, while in its presence, FITC-dextrane is liberated. An increase in its fluorescence is observed as a result. The dynamic range of the sensor was from 0 to 44 mM of glucose, and the response time is 10 min for a glucose concentration step from 0 to 11 mM.

446 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 10. Schematic of an affinity glucose sensor (redrawn with permission from Ballerstadt, R.; Schultz, J. S. Anal. Chem. 2000, 72, 4185; Copyright 2000 American Chemical Society). (Left) The Alexa-488-ConA conjugate is bound to dextrane and, therefore, confined in the porous microbeads, which do not allow the excitation light to penetrate. (Right) As glucose diffuses through the dialysis membrane, it liberates Alexa-488-ConA conjugate so that fluorescence is enhanced a result of better exposure to the exciting light beam and because fluorescence emission is no longer screened off.

Ballerstadt and Schultz328 have further developed the ConA-based system, thereby providing a most elegant solution. The system is based on an inner filter effect. Highly porous Sephadex beads were colored with two red dyes (safranine and para-rosaniline), which were selected to block the excitation and emission spectra of the fluorescent Alexa 488-ConA conjugate (see Figure 10). The latter is bound to dextrane inside the beads but is liberated when glucose is present. Once located outside the beads, the conjugate is fully exposed to the excitation light and fluorescence intensity, therefore, increases. The sensor has a dynamic range from 0.2 to 30 mM of glucose, and the total signal change was much higher than that for the FRET-based system.327 Moreover, faster response times were accomplished (4-5 min). In other work,329 the IR-dye Alexa-647 was conjugated to ConA in order to make the sensor work in the IR, which is less prone to spectral interferences by the intrinsic fluorescence of serum samples. The authors also carefully investigated the long-term stability of the biosensors by monitoring their performance over a 4 month period. The sensors displayed an initial increase in fluorescence over the first 3-4 weeks, which later on gradually decreased with an approximately linear drop of 25% per month. The decrease in fluorescence was not due to denaturation of the ConA but rather due to leakage of the fluorescently labeled ConA through the interface between the outer sealant and the membrane. If this problem can be coped with, the sensors potentially are suitable for continuous usage for up to 1 year. The main drawback of all sensors based on the use of ConA is their poor selectivity because they also respond to many other carbohydrates. For example, binding of fructose is ∼3 times stronger than that of glucose. This is in distinct contrast to the high specificity of enzymatic glucose sensors. On the other side, they have lower limits of detection. Other schemes for affinity sensing of glucose make use of the E. coli glucose-binding protein (GBP) that can be gene engineered. In contrast to ConA, GBP binds glucose with very high selectivity. In fact, the affinity to other saccharides is 100-1000-fold weaker than that for glucose or galactose. These proteins are rather stable and can be stored several months without degradation of activity.330 Since GBP

Borisov and Wolfbeis

undergoes significant conformational changes upon binding of glucose, a polarity-sensitive fluorescent indicator introduced at specific positions can act as an optical transducer. The proof of principle for the case of glucose was demonstrated by Marvin and Hellinga,331 who used GBP labeled with acrylodan. A homogeneous assay for glucose was developed that operates in the micromolar concentration range (0-2 µM). However, it was much earlier shown that a similar scheme could be applicable to sensing of maltose using a fluorescently labeled maltose-binding protein.332 A 60-180% increase of fluorescence intensity was observed in the presence of maltose. Concentrations of maltose from 0 to 200 µM could be determined. Ge et al.330 showed that the sensitivity of such assays strongly depends on the brightness (Bs) of the label used (Bs ) ‚QY). In fact, GBP labeled with the thiol label 2-(4′iodoacetamidoanilino)naphthalene-6-sulfonate resulted in a working range from 0.3 to 10 µM of glucose. For comparison, the assay based on the glutamine binding protein (GlnBP) labeled with a much brighter acrylodan was performed in the analytical range from 0.03 to 3.2 µM. The response of the proteins is very fast in solution (<1 min) but, however, much slower in dialysis cassettes (5-12 min). The recovery in the dialysis cassettes (which have a volume of 1 mL) was unacceptably slow in taking several hours, but this can be possibly accelerated by using smaller volumes. A ratiometric assay also was reported333 where GBP was labeled with both an environmentally sensitive fluorophore acrylodan and a reference luminophore, a ruthenium(II) polypyridyl complex. Sensing of glutamine using GlnBP was also demonstrated by Dattelbaum and Lakowicz.63 The E. coli GlnBP was covalently modified with acrylodan and other environmentally sensitive probes and used in a homogeneous assay format to detect glutamine from 0.05 to 6.4 mM. Time-resolved and polarization-based sensing of glutamine also was demonstrated. No interference by glutamate was demonstrated, which is a common drawback of the enzymatic glutamine biosensors based on the enzyme glutaminase. Ye and Schultz53 engineered a novel glucose indicator protein (GIP) that makes use of FRET from the donor (green fluorescent protein, GFP) to the acceptor (yellow fluorescent protein, YFP). In the absence of glucose, the two fluorescent proteins are in close proximity so that FRET occurs. When excited at 395 nm (corresponding to the absorbance maximum of GFP), the emission from the YFP (peaking at 527 nm) is observed. The spatial separation between the two moieties increases when GIP binds glucose, and FRET is reduced. In the biosensor, a solution of GIP was brought into a hollow cellulose dialysis fiber (L 190 µm), which was placed into a microcuvette, and fluorescence intensity was monitored at two wavelengths and the ratio was determined. The sensor responded reversibly to glucose with response and recovery times of ∼100 s, although some drift in the baseline occurred. Glucose could be determined in concentrations between 0 and 10 µM. Chinnayelka and McShane54 have used an inactive form of glucose oxidase as a selective glucose-binding protein. The apoenzyme labeled with tetramethylrhodamine isothiocyanate was placed inside nanoengineered polymeric microcapsules together with fluorescein-labeled dextrane. The FRET that was observed in the absence of glucose was reduced when glucose diffused into the microcapsules and replaced labeled dextrane in a competitive way. The ratio of fluorescence intensities of the two labels was used as the

Optical Biosensors

analytical information. The sensor reversibly responded to glucose concentrations in the range from 0 to 40 mM, was fast (1-2 min), and was rather selective (a 5-10-fold higher sensitivity for glucose compared to other saccharides). The sensitivity of the system can be easily tuned by varying the concentration of the GOx-dextrane complex. Ogasawa et al.334 designed an affinity sensor for the riboflavin-binding protein (RBP). It exploits the intrinsic green fluorescence of riboflavin. Hydrophobic 3-octylriboflavin is adsorbed on an optical fiber whose surface was made lipophilic by modification with octadecyl groups. On exposure to a solution of RBP, it binds to the surface-immobilized octylriboflavine. The sensor can be renewed with good reproducibility by removing the RBP-riboflavine complex and loading the surface again with octylriboflavine. In fact, 15 individual sensors for RBP were prepared on a single optical fiber, which varied in performance by <5%. When RBP binds to the immobilized riboflavin, quenching of fluorescence is observed. As little as 0.73 µM of the RBP can be detected in 5 min. Medintz et al.55 prepared two kinds of biosensors for maltose, which make use of quantum dots (QDs) and a maltose-binding protein (MBP) from E. coli. The first type of the sugar-sensing nanoassemblies consisted of 560-nm donor QDs conjugated to 10 molecules of MBP (in average). The displaceable dextrane was labeled with a fluorescent acceptor dye. No FRET was observed when maltose replaced dextrane in the sugar-binding site. The second maltose biosensor consists of 530-nm donor QDs loaded with 10 cyanine-labeled MBP molecules. The protein-bound label served as a bridging acceptor/donor for ultimate energy transfer to maltose-displaceable labeled dextrane. Both sensors respond to maltose concentrations from 10 nM to 1 mM. Another nanobiosensor for maltose56 was reported that is based on the use of thiolhexanoate-capped CdSe nanoparticles (L 3.0-3.5 nm) conjugated to the MBP, which was labeled with a luminescent ruthenium(II) complex. Little fluorescence is detectable in the absence of maltose because of the electron transfer from the ruthenium complex to the nanobead. Fluorescence is enhanced as a result of conformational changes that occur upon binding of maltose. Glucose and lactose were shown not to interfere, while maltose could be sensed from 10 nM to 10 µM. An unusual approach toward receptor-based sensing of glucose in blood was proposed by Sanz et al.335 It is based on the fact that oxidation of hemoglobin (Hb, contained in the sample) by H2O2 (generated after addition of glucose oxidase GOx according to eq 1) results in distinct changes in the absorption spectrum of Hb. To make the assay operative, blood samples prepared without pretreatment and reactions of H2O2 with other blood components such as catalase need to be blocked (which is achieved by addition of azide). The activity of GOx should be high enough for the “chemistry” to work, given the fact that the activity of GOx is inhibited by azide by ∼30%, and low enough to avoid its interaction with hemoglobin. The linear response range of the system was from 0.1 to 30 mM of glucose.

5.2. Receptor-Based Biosensors for Inorganic Ions Biosensors for cations exploit the need of enzymes for certain ions in order to function. This results in two kinds of sensing schemes. The first is the effect of the catalytic ion on the actiVity of the enzyme (which is treated in section

Chemical Reviews, 2008, Vol. 108, No. 2 447

3). The second is the effect of the catalytic ion on the conformation of the enzyme (regardless of its activity). Respective sensors are treated in this section. In such sensors, an enzyme serves as a kind of chelator for certain metal ions, but with much higher selectivity than classical chelators such as EDTA. The ions usually are required as enzyme cofactors. For example, Thompson and Jones336 reported a highly sensitive zinc(II) biosensor, which makes use of the enzyme carbonic anhydrase (CA). The Zn2+ ion is its natural cofactor and is bound by CA with excellent selectivity. The apoenzyme prepared by dialysis was contained in a chamber together with the fluorescent probe dansylamide. The ionpermeable chamber was attached to an optical fiber. The probe does not bind to the apoenzyme and, thus, remains weakly fluorescent in water. However, when (practically irreversibly) bound to CA in the presence of Zn2+, the fluorescence intensity increases significantly. The analytical range of the sensor was from 40 to 1000 nM. The main drawback of the system is its moderate brightness and the limited applicability to real samples because UV excitation is required at ∼330 nm where background fluorescence of most samples is very strong. A more flexible system for detection of Zn2+ is based on fluorescence quenching.52 In the presence of the metal cation, the active site of CA labeled with the fluorescent donor fluorescein permits binding of the colored quencher azosulfamide. The fluorescence decay time (measured by phase fluorometry) was shown to decrease with increasing concentrations of zinc cation, which could be determined in the concentration range from 1 to 100 µM. A cobalt biosensor was reported51 that is based on the finding that the d-d absorption of Co2+ coordinated to CA labeled with a cyanine dye promoted radiationless FRET since the absorption of the Co2+ ion overlaps the emission of the label. The labeled apoenzyme was entrapped in a polyacrylamide gel positioned at the distal end of an optical fiber. The sensor was capable of sensing Co2+ in the concentration range from 0 to 20 µM with response times of a few minutes. The fact that certain variants of CA exhibit different selectivity to metal ions was used by Zeng et al.57 to design a sensor for Cu2+. Two variants of human CA II were labeled (with Oregon Green and Alexa Fluor 660, respectively) and immobilized at the distal end of an optical fiber. An ∼85% drop in fluorescence intensity and decay time (as measured by phase fluorometry) was observed upon binding of Cu2+. The analytical range of the system was from 0.1 to 100 pM. Ions such as Co2+ and Ni2+, and even Zn2+, were shown to interfere only if present in much higher concentration, because the affinity of the CAs to these ions is much lower than that for Cu2+. It was demonstrated that the sensor was suitable for real-time analysis of Cu2+ in seawater. The phosphate-binding protein isolated from Escherichia coli and labeled with the fluorophore acrylodan is potentially suitable for biosensing purposes337 by showing a 50% increase of fluorescence intensity in the presence of micromolar concentrations of phosphate. In another detection scheme for phosphate,338 two cysteine mutations were introduced into the phosphate-binding protein, allowing it to be labeled with two rhodamine fluorophores. When close to each other (in the absence of phosphate), the rhodamine molecules form a noncovalent and nonfluorescent dimer. A linear correlation between fluorescence intensity and the concentration of phosphate was observed in the range from 0 to 6 µM of the analyte. At saturation (concentrations of

448 Chemical Reviews, 2008, Vol. 108, No. 2

phosphate of 6 µM or higher), an 18-fold increase (at average) in fluorescence intensity was observed for the mutants investigated. A homogeneous assay for sulfate was reported that is based on fluorescently labeled sulfate binding protein.339 The fluorescence of the labels was quenched upon binding sulfate. The LODs varied from 30 to 200 nM depending on the label used. Sumner et al.340 have discovered that the wild form of red fluorescent protein (RFP) can reversibly bind Cu+ and Cu2+ ions with high selectivity and sensitivity. In fact, the quenching of fluorescence of the RFP by the heavy metal ions occurred even at 10 nM concentrations of the analytes. The RFP was found to be ∼107-fold more selective for copper over Mg2+ and Ca2+, with its fluorescence being virtually unaffected even by high concentrations of those. Fluorescence was recovered on addition of EDTA, albeit not to the initial level. A nanobiosensor for Cu2+ also was designed on the basis of RFP.341 The protein was immobilized in 80nm polyacrylamide nanobeads together with an inert reference dye. The ratio of the intensities under 488-nm excitation served as the analytical parameter. The signal change caused by Cu2+ returned to 95% of its original value within 3 min when EDTA was added, thus showing an almost complete reversibility of the sensor. As in the case of the homogeneous assay in solution, no interference by other metal ions was observed. The sensitivity to Cu2+ was independent of pH in the range from 6 to 8.5. However, the nanobiosensor was found to be much less responsive to Cu2+ at low pH. The analytical range for Cu2+ is from 0.2 to 50 µM, but sensitivity can be fine-tuned by varying the concentration of the nanobeads, thus generating a larger signal change per nanoparticle at the same concentration of the analyte.

5.3. Receptor-Based Biosensors for Gaseous Species A number of optical affinity biosensors make use of heme proteins, natural compounds that can weakly bind gases such as oxygen and carbon dioxide, and much more strongly bind carbon monoxide and nitrogen monoxide. Blyth et al.342 showed that the heme proteins cytochrome c, myoglobin, and hemoglobin (Hb) enable semiquantitative detection of CO and NO in aqueous medium. The heme proteins immobilized into a sol-gel matrix exhibited a distinct change of their absorption spectra upon coordination of NO and CO. Although the effect was reversible, desorption of gases took up to 2 h. However, fast regeneration was accomplished by using other reagents. For example, cytochrome c could be regenerated from its complex with NO by reduction with sodium dithionite, washing with buffer, and addition of potassium ferricyanide. Cytochrome c embedded in a sol-gel was demonstrated to sense NO in a gas phase, with a response time of 200 s and full recovery within 300 s.343 The sensor operated in the dynamic range from 1 to 25 ppm of NO. Reversible and fast micro- and nanosensors for NO were developed by Barker et al.344 Cytochrome c′ was immobilized on gold nanobeads (L 50 nm). Two ways of optical interrogation were reported. The first is to measure the intrinsic fluorescence of cytochrome c′, which changes in the presence of NO. The second is to measure the increase in the efficiency of FRET from the label (Oregon Green) to cytochrome (which is enhanced when the latter binds NO because of better spectral overlap between the emission spectrum of Oregon Green and the excitation spectrum of cytochrome). The response to NO is linear in the concentra-

Borisov and Wolfbeis

tion range from 20 µM to 1 mM. The sensors operate in a fully reversible way and have response and recovery times of <1 s. Although the LOD of the sensor was relatively high (20 µM), the limits of quantification are very low because only small volumes are required. Cytochrome c′ also was entrapped in polyacrylamide; however, binding of NO was found to be irreversible. The performance of such microsensors was further improved by using a reference fluorescent indicator and employing ratiometric measurements.345 The microsensor incorporated labeled with Oregon Green cytochrome c′ along with polystyrene nanobeads (L 40 nm) labeled with a fluorescent reference dye whose red emission (λmax ) 685 nm) allowed for ratiometric (two-wavelength) measurements. Compared to the previous work,344 the LOD of the sensor was improved to as low as 8 µM. Immobilization of the reference spheres was, however, not reproducible, and the ratio of fluorescence intensities and calibration plots, therefore, varied from sensor to sensor. The ratiometric sensors were employed to measure extracellular NO released by macrophages. Blyth et al.346 observed that cytochrome c′, when immobilized into a sol-gel, undergoes irreversible conformational changes (in 2-4 days), which lowers binding affinity of the metalloprotein. However, after these changes have occurred, the protein remains selective for NO and the calibration plots are well-reproducible. Analogous biosensors were prepared with the heme domain of soluble guanylate cyclase, the only protein receptor known for signal transduction involving in vivo produced NO and having many similarities to cytochrome c′, including a very low affinity for oxygen and a high affinity for nitric oxide.347 The LOD of the sensor was 1 µM. Reversible binding of oxygen to hemoglobin (Hb) was exploited by Zhujun and Seitz348 to design an oxygen biosensor. They showed that Hb immobilized on a Sephadex ion-exchange resin can sense oxygen in the dynamic range from 0 to 160 Torr. The shelf life of this reflectance-based biosensor is very short due to fast irreversible degradation of the immobilized Hb (within 2 days at room temperature or within 7 days at 4 °C). The sensor is, thus, hardly an alternative to conventional optical oxygen sensors, which are based on stable quenchable luminescent indicators.

5.4. Receptor-Based Biosensors for Toxins An evanescent-wave biosensor for R-bungarotoxin was designed by immobilizing the nicotinic acetylcholine receptor on an optical fiber.349 Fluorescein-labeled R-bungarotoxin was used in the competitive assay format. As little as 1 nM of the toxin could be detected within 5 min. Although the nonspecific binding was totally eliminated by addition of bovine serum albumin, the sensor was inhibited by agonists such as acetylcholine, nicotine, and carbamylcholine and by antagonists such as pancuronium and D-tubocurarine. No regeneration of the biosensor was possible. Song and Swanson50 developed biosensors for cholera toxin (CT). The bioreceptor ganglioside GM1 was incorporated into a biomimetic membrane surface (composed, e.g., of 9-octadecenoyl phosphatidylcholine), which, in turn, was spread onto glass microbeads. The labeled receptor molecules are homogeneously distributed in the lipid bilayer but aggregate in the presence of CT, which has five binding sites for GM1. As a result, fluorescence self-quenching is observed. Alternatively, the receptor molecules are labeled

Optical Biosensors

with a fluorescent donor, and a fluorescent acceptor dye, FRET, is observed in the presence of CT. Generally, the sensors respond to the concentration of toxin from 0 to 10 nM. Sensitivity and dynamic range can be tuned by varying the total concentration of the labeled GM1 in the membrane. Limits of detection as low as 0.05 nM and small dynamic range are associated with samples having low concentration of the receptor, while lower sensitivity and large dynamic range are found for samples with high concentration of the receptor. Boradipyrroles also were found to be viable labels in both types of the biosensor. No interference by albumin was observed in this case, whereas a significant nonspecific drop in fluorescence was observed for GM1 labeled with fluorescein. Many bacterial toxins, viruses, and bacteria target carbohydrate moieties on the surface of a cell so as to attach and gain entry into the cell. Ngundi et al.350 designed a monosaccharide-based array biosensor for detection of protein toxins. Arrays of N-acetyl galactosamine (GalNAc) and N-acetylneuraminic acid (Neu5Ac) derivatives were immobilized on the surface of a planar waveguide (similar to ref 307) and were used as receptors for protein toxins. These arrays were probed with fluorescently labeled bacterial cells and protein toxins. While Salmonella typhimurium, Listeria monocytogenes, Escherichia coli, and staphylococcal enterotoxin B did not bind to either of the monosaccharides, both cholera toxin and tetanus toxin bound to GalNAc and Neu5Ac and could be detected at concentrations down to 100 ng/mL. In conclusion, it can be stated that biosensors based on ligand-receptor interactions are (a) often highly specific; (b) sensitive in giving rather low limits of detection; (c) characterized by virtually irreversible response; (d) sensitive to environmental effects; (e) prone to poisoning; and (f) tedious to fabricate, in particular in terms of genetic modification and isolation of proteinic receptors.

6. Nucleic Acid Biosensors Such sensors (also referred to as DNA biosensors) represent an attractive alternative approach to immunological sensing of species such as bacteria. They take advantage of the exceptional long-term stability of nucleic acids and the high selectivity of the interaction of complementary chains of polynucleotides. Nucleotides and their polymers also can be synthesized easily. Typical examples of DNA biosensors are described in the following (see Figure 11).

Figure 11. Two fundamental forms of nucleic acid-based sensors: (a) conventional DNA sensor using a fluorescently labeled counter strand or a fluorogenic intercalator; (b) molecular beacon DNA sensor.

Chemical Reviews, 2008, Vol. 108, No. 2 449

6.1. Single DNA Sensors on Solid Supports and on Fiber-Optics The feasibility of optical nucleic acid biosensing was demonstrated by Graham et al.351 16- and 20-base oligonucleotides, but also long (204-base) oligonucleotide chains, were attached to the surface of an optical fiber as shown schematically in Figure 11a. Fluorescein-labeled complementary chains were shown to interact with complementary chains within ∼1 min as determined by TIRF spectroscopy (see section 4.3.2). Regeneration of the biosensor was accomplished by raising the temperature from 65 to 80 °C, resulting in complete dissociation of the bound duplex within 10-15 min. The analytical range of the sensor was from 0 to 200 nM of nucleotides. Piunno et al.352,353 immobilized a DNA sequence on an optical fiber by first activating the surface with a long-chain aliphatic spacer arm terminated with a nucleoside to which a longer chain was attached through automated step-by-step DNA synthesis. Detection of hybridized DNA at the fiber surface was achieved by treating it with a solution of the intercalator ethidium bromide. The sensor was regenerated by exposing it to a 85 °C hybridization buffer for 5 min. Total analysis time was <1 h, and the LOD was 86 ng/mL. The sensor showed reproducible results within 3 months of storage. Watts et al.354 immobilized biotinylated oligonucleotide sequences on a solid surface via streptavidin and used a resonant mirror technique (see section 4.3.2) for direct and rapid detection of hybridization. The lowest detectable concentration of the target 40-base nucleotide was 9.2 nM. Hybridization at the sensor surface was followed for 15 min, although a positive response was obvious within 30-60 s. Abel et al.355 compared the performances of DNA biosensors operating in the direct and competitive assay formats, respectively. A biotinylated capture probe was immobilized on a glass surface via avidin or streptavidin. A complementary fluorescein-labeled 16-base oligonucleotide could then be determined with an LOD of 0.2 pM. A competitive assay (using labeled and unlabeled nucleotides) resulted in a much higher LOD (1.1 nM). The use of poly(acrylic acid) sodium salt and Tween 20 reduced the nonspecific binding to 1-2% of the amount of specific binding. The signal loss during long-time measurements, i.e., after consecutive hybridization assays, can be described by a single-exponential function and, thus, compensated for. After 200 cycles, the net signal had decreased by 50%, corresponding to a signal variation of only 2.4% after correction for this signal loss. By using a 50% (w/w) aqueous urea solution for regeneration of the biosensor, the duration of an assay cycle was reduced to 3 min. Pilevar et al.356 used a near-IR cyanine dye (λexc ) 787 nm, λem ) 807 nm) as a label for an oligonucleotide sequence in order to make measurements outside the background fluorescence from natural compounds, which is substantial when using fluorescein labels. The feasibility of detecting bacterial cells using rRNA as the target was demonstrated in a solid-phase sandwich-type of assay where Helicobacter pylori rRNA was used along with IR dye-labeled detector oligonucleotide probe. The result indicates that this biosensor is capable of detecting H. pylori RNA at picomolar concentrations. A biosensor for detection of L-adenosine was developed by Kleinjung et al.357 An L-adenosine specific RNA was attached to an optical fiber via an avidin-biotin link.

450 Chemical Reviews, 2008, Vol. 108, No. 2

Fluorescein-labeled L-adenosine was used in a competitive assay format. The sensor responds to concentrations from 1 nM to 100 µM. Bagby et al.358 found that an intercalating thiazole orange derivative (TOMEHE) gives a 10-fold larger signal change over the commonly used ethidium bromide, thus providing higher sensitivity for hybridization events. TOMEHE, however, also showed a significant response to single-strained DNA and concentration-dependent phenomena at high loading with the dye. This limits the dynamic range over which TOMEHE can be used. A capillary sensor (see section 4.3.2.) for DNA also was reported.359 A capture DNA was attached to the inner walls of a capillary via biotin-streptavidin chemistry. The complimentary DNA sequence labeled with Alexa Fluor 532 can be sensed with a detection limit of 30 pg/mL.

6.2. DNA Arrays Ferguson et al.360 created a fiber array biosensor capable of simultaneous measurements of 7 DNA sequences. The optical fibers (each 200 µm in L) carrying the immobilized oligonucleotide probes were bundled to form a multiplex DNA sensor. The fluorescence intensity of all the fibers was imaged with a CCD camera. Up to 7 DNA sequences could be detected within 10 min with an LOD of 10 nM. As in the case of individual sensors, the array can be stored for prolonged periods (months) without loss of activity. Walt and co-workers demonstrated later361 that the actual detection limits of the array microsensor can be much smaller if microspheres are being used. Small sample volumes (10 µL for a 500 µm array) and higher local concentrations of the DNA enable further amplification. Fewer sensor numbers in the array also increase the signal because more target molecules hybridize per microsphere. By examining multiple identical sensors simultaneously, the signal-to-noise ratio can be improved by allowing incoherent noise to be signal averaged. The authors demonstrated that as few as 600 target DNA molecules (10-21 mol) can be detected. Although DNA at higher concentration can be assayed within 10-30 min, 17 h were necessary to complete hybridization and to achieve the lowest detection limits. The same group developed a method for encoding a set of randomly ordered functionalized microbeads (each bearing alkaline phospatase, avidin, or biotin) using luminescent dyes.362 This method formed the basis for a smart array system suitable for simultaneous detection of numerous DNA sequences.363 Polymer microbeads (L 3.1 µm) were dyed with various fluorophores such as Cy-5 and europium(III) complexes and functionalized with different oligonucleotide probes. Because microbeads with different concentrations of a single fluorophore are optically distinguishable, a total of 100 different beads could be prepared. A mixture of the beads was distributed over a distal end of a fiber bundle (L 500 µm, 6 000 individual fibers) so that each microbead occupied a single well (Figure 12). The position of the recognition elements was decoded by imaging with a CCD chip because each type of the beads has a characteristic emission wavelength and luminescence intensity. The signal was monitored after hybridization to fluorescein-labeled complementary oligonucleotides. Only 10 min are needed to determine 100 pM of oligonucleotides, but up to 17 h are needed for the lowest concentration (10 fM). In continuation of this methodology, Walt and co-workers have designed array biosensors for simultaneous determi-

Borisov and Wolfbeis

Figure 12. Overview of a fiber-optic array system. The functionalized beads occupy the micrometer-sized wells (1 bead/well) located on the tip of the optical fiber. The position of the beads is decoded by imaging its color. When immersed into a sample solution containing labeled DNA, a signal is observed only on the beads bearing the DNA probe complementary to the target in solution. Reproduced with permission from Ferguson, J. A.; Steemers, F. J.; Walt, D. R. Anal. Chem. 2000, 72, 5618. Copyright 2000 American Chemical Society.

nation of Bacillus anthracis, Yersinia pestis, Francisella tularensis, Brucella melitensis, Clostridium botulinum, Vaccinia virus, and a biological warfare agent simulant named Bacillus thuringiensis kurstaki.364 The replacement of the 20mer probes by 50-mer probes allowed for a high specificity of the array. The authors report LODs of “10 fM” (10 femtomolar concentrations) for B. anthracis, Y. pestis, Vaccinia virus, and B. thuringiensis kurstaki, and of 100 fM for B. mellitensis and C. botulinum. This is difficult to interpret since a bacillus does not have a molecular weight. It was also found that overlapping target sequences are partially complementary to the probe sequences. This can result in a nonspecific response, unfortunately. The use of multiple probes (at least two for each analyte) minimizes the potential possibility of false identification. The assay time was 30 min. The above DNA optical fiber array subsequently was coupled to a microfluidic system365 operated at a flow rate of 1 µL/min. This resulted in faster hybridization (15 min, compared to 30 min required for static measurements) and in ∼100-fold lower LODs (10 aM, compared to 1 pM as achieved in static measurements), which makes this approach highly advantageous. The systems described in refs 360367 are quite successful in commercial terms. The target rRNA of an algal bloom species can be determined with a microarray biosensor operating in a kind of “sandwich” assay format.366 The RNA to be analyzed interacts with a long capture probe, and the labeled tracer oligonucleotides then interacts with the residual free end. Fluorescently labeled oligonucleotides (acting as tracers that can capture nucleotide sequences) were coupled to the surface of microbeads positioned in the wells at the tips of optical fibers in an array. As few as 5 cells could be detected within 45 min, and the LOD of the rRNA is 4 × 104 molecules. In similar work,367 the microarray biosensor was used for detection of different Salmonella strains with LODs of 103-104 cfu/mL in pure samples and of 104-105 cfu/mL in the presence of interfering organisms. All investigated Salmonella strains were detectable, albeit with different sensitivities. Other common food pathogens were shown not to interfere at concentrations of 108 cfu/mL. The assay time was 1 h. Another approach of addressing and specifically depositing DNA was demonstrated by Swanson et al.,368 who designed a semiconductor biochip containing a microelectrode array. In order to immobilize a specific capture DNA probe at a

Optical Biosensors

certain location, an electric field was applied that causes the attraction of DNA fragments and their deposition at predetermined electrodes. The target DNA may be guided to the specific locations as well. The assay is performed in a sandwich format by exposing the array to the tracer DNA bearing a fluorescent label. If coupled to an integrated fluidic system, the following steps of the DNA analysis of E. coli can be automatically performed:369 (a) di-electrophoretic collection of bacteria; (b) DNA amplification and electronic DNA hybridization; and (c) fluorescence readout with the help of a CCD camera. The whole assay can be performed in 2.5 h. Rissin and Walt have demonstrated recently370 that the sensitivity of array biosensors can be significantly improved by employing an enzymatic amplification step similar to that known from ELISAs. To prove its feasibility, a biotinylated femtoliter array was incubated with a solution of streptavidin-β-galactosidase (SβG) conjugate, and then with a solution of the enzyme substrate, which generated fluorescent resorufin upon hydrolysis. Since the volume of the wells is very small, the limit of quantification for SβG was found to be 2.6 amol. No signal was measured in control experiments, in which the enzyme was bearing no streptavidin, or, alternatively, the surface of the fiber was not biotinylated.

6.3. Molecular Beacons in DNA Sensors Described first by Tyagi and Kramer,45 molecular beacons (MBs) have become an important tool for studies in genetics, disease mechanisms, and molecular interactions. MBs represent single-stranded types of oligonucleotide probes that possess a stem-and-loop structure (see Figure 11b). The stem is formed by the two ends of an MB containing complimentary nucleotides. A fluorophore attached to one end of the stem and a quencher attached to the other are in close proximity, and little or no fluorescence is observed. The loop portion of the molecule is responsible for reporting the specific complimentary oligonucleotide. Hybridization of a matching oligonucleotide to the loop portion results in conformational reorganization that brings the stem apart so that fluorescence is enhanced. This smart technique has the advantage that no labeling of other species is required (compared to, e.g., the competitive assay format). Following the work of Tyagi and Kramer,45 who used MBs in homogeneous solution, Fang et al.46 immobilized an MB onto the surface of a silica plate via avidin linkage to design a solid-state biosensor. A biotinylated MB was prepared that had a total of 28 bases, including 18 bases complementary to the sequence of interest and 5 base pairs for the stem. Tetramethyl rhodamine was selected as a fluorophore, and a modified azobenzene Dabcyl was selected as a quencher. A significant increase in fluorescence was observed upon addition of the complementary DNA both for the MB contained in homogeneous solution and in the immobilized form. In the control experiment, no effect was observed on addition of the noncomplementary DNA. The results indicated that the immobilized MB could be used to detect target DNA molecules in the subnanomolar range. Liu and Tan47 have investigated DNA sensing in more detail using a similar MB (also labeled with TMR and Dabcyl). The biotinylated MB was immobilized onto the surface of an optical fiber via a streptavidin bridge. A spacer group between the MB and the fiber substantially reduces steric hindrance and increases its mobility. A strong increase in fluorescence intensity was observed only upon hybridiza-

Chemical Reviews, 2008, Vol. 108, No. 2 451

tion with a complementary DNA, while the increase was much lower for the oligonucleotide strand having a single mismatch. The MB could be regenerated by immersing it into 90% formamide plus 10% tris/EDTA buffer mixture for 1 min. While a single regeneration cycle completely recovered the sensing properties of the MB, a few repeated regeneration cycles resulted in a significant drift. A 50% aqueous urea solution can be used for regeneration of DNA sensors as well.371 The authors also showed that higher ionic strength (IS) favored hybridization by decreasing electrostatic repulsion between the loop chain of an MB and the target DNA. Moreover, the initial reaction rate in the presence of divalent cations was ∼20 times higher than in the presence of monovalent ones at the same IS. The best results were obtained by using a high IS created by the divalent cation Mg(II). Under these conditions, 1-100 nM of DNA could be sensed in 10 min. Ueberfeld and Walt40 have designed an MB capable of virtually reversible binding of the target nucleotide. This can be achieved if the free energy of the duplex formation is of the same magnitude as the free energy of the stem formation. To do so, 6 oligonucleotides in the fully complementary 20mer loop were replaced by adenine moieties. The reversibility was indeed achieved, but only when working at a carefully adjusted temperature (38.8 °C), while working at 34 °C resulted in irreversible binding, and working at 43.8 °C resulted in the melting of the loop-target hybrid. In order to measure the efficiency of FRET, the stem of the MB was labeled with a fluorescent donor, while the acceptor dye was positioned between the stem and the loop sequence where it neither hinders stem formation nor inhibits target-loop hybridization. A two-wavelength ratiometric approach was made use of. This has the advantage that the ratio of the fluorescence intensities of the donor and the acceptor, respectively, is independent of the concentration of the labels. Du et al.372,373 attached a fluorescently labeled oligonucleotide to a thin gold surface to create an MB that requires no quencher on the second stem. In the absence of a complementary DNA, the label is in close proximity to the gold surface and its fluorescence is quenched. The intensity increased about 100-fold upon hybridization with a complementary DNA, which was detectable at concentrations from 0.2 to 3 µM. An 8-fold lower sensitivity was observed for a singly mismatched target. The sensor was, however, not suitable for multiple measurements in showing an ∼40% degradation of fluorescence intensity after each regeneration cycle.

6.4. Liposome-Based DNA Assays Some biosensors for nucleic acids make use of liposomes containing thousands of dye molecules and thus generating strong signal for even low nucleic acid concentrations so that quantitative reflectance (or even qualitative visual) measurements are possible. A typical biosensor of this type includes a capture oligonucleotide attached to the surface of a liposome. Biotin-streptavidin binding is employed to attach a reporter oligonucleotide to a liposome loaded with sulforhodamine B. In the presence of the target sequence, a sandwich is formed and the detection zone becomes colored. RNA from B. anthracis spores in concentrations from 0.1 pg/L to 1 ng/L could be detected in 90 min.374 As little as one B. anthracis spore is detected in 12 h. Ba¨umner et al.375 used a similar system for detection of RNA sequences from B. anthracis, C. parVum, and E. coli,

452 Chemical Reviews, 2008, Vol. 108, No. 2

but with the difference that the fluorescein-labeled capture oligonucleotides were immobilized on a polyethersulfone membrane via anti-fluorescein antibodies. Quantification of the RNA was possible between 10 nM and 1 µM by simultaneously incubating the RNA with streptavidin-labeled liposomes, biotinylated capture oligonucleotide, and the membrane containing a target sequence. The assay time was 20-30 min. An analogous RNA biosensor was used for detection of Dengue virus in blood samples.376

6.5. Aptamer-Based DNA Sensing Aptamers are nucleic acid species that have been evolutionary engineered through in vitro selection to bind various molecular targets such as haptens, proteins, nucleic acids, and even cells, tissues, and organisms. Lee and Walt377 have designed an aptamer biosensor for thrombin. The aptamer was immobilized onto nanoporous silica beads that were placed on the distal end of a fiber array. Fluorescein-labeled thrombin was used in a competitive assay format. Although fluorescence signals from the individual aptamer beads showed significant variability, the average signals of 100 beads provided much more precise values. The fiber-optic microarray system had a detection limit of 1 nM for nonlabeled thrombin, and each test could be performed in ∼15 min including the regeneration time. Rupcich et al.378 reported on the immobilization of an aptamer-based system in a sol-gel matrix. Fluorescein was covalently attached to the aptamer, and the quencher (dabcyl) was attached to the complementary oligonucleotide and, thus, was in close proximity of the fluorescein upon hybridization. Binding of the target molecule (ATP) to the aptamer results in a conformational change in the aptamer and formation of a stem-and-loop structure. The quencher-labeled oligonucleotide (QDNA) strand is displaced, and a large increase in fluorescence is observed. A tripartite aptamer complex also was prepared, where the fluorophore was attached to an oligonucleotide sequence complimentary to the stem part of the aptamer. The aptamer was attached to streptavidin to provide a lager molecular mass so to reduce leaching from the sol-gel matrix. Aminopropyltriethoxysilane was added to the sol-gel to promote electrostatic retention of the DNA because it introduces amino groups into the sol-gel, which are positively charged at pH’s below 8. The leaching rate was ∼12%/h. The authors also showed that a QDNA composed of 11 nucleotides is the best compromise between sensitivity and response time. The analytical range of the sensor was from 0.01 to 3 mM ATP. The sensor retained full signaling capability for 1 month but showed no response after a 3 month storage in buffer solution, presumably because of irreversible aging of the sol-gel matrix. To summarize this chapter, one can state that DNA sensors (a) have exceptional stability; (b) possess very low limits of detection; (c) are highly specific; (d) can be easily produced using automated procedures; (e) are self-contained in case of using molecular beacon; (f) are relatively tolerant to heat and frost; (g) are rather sensitive to effects of ionic strength; and (h) have very wide applicability and large potential in the case of aptamer DNA or DNAzyme sensors, thus allowing sensing even of haptens and proteins.

7. Whole-Cell Biosensors This type of biosensor makes use of living cells such as individual microorganisms or tissue, rather than relying on

Borisov and Wolfbeis

using specific biomolecules such as enzymes, proteins, or DNA. Whole-cell biosensors generally exhibit longer shelf lifetimes compared to, e.g., enzymatic biosensors because the active components are contained and produced in the “natural” environment and not in a polymer matrix on the surface of a sensor, which has limited biocompatibility. Whole-cell biosensors often are less costly than the corresponding enzymatic biosensors because some microorganisms can be cultivated and isolated rather easily, which is not the case for many enzymes. On the other side, they often lack specificity for the respective analytes. Whole-cell biosensors mostly are self-contained, do not require the addition of cofactors, and are the biorecognition elements of choice when the total amount of hazardous substances or pollutants is to be determined. Biosensors for determination of biological oxygen demand (BOD) represent a typical example. In contrast to such biosensors, those making use of genetically modified microorganisms can be highly specific. Another disadvantage of whole-cell biosensors is the relatively slow response (tens of minutes to hours) because the analytes need to diffuse through a cell membrane. Such response times are, however, adequate in certain cases. The standard BOD5 test, for example, requires 5 days.

7.1. Catalytic Whole-Cell Biosensors Biosensors for determination of biochemical oxygen demand (BOD) make use of oxygen transducers. As in the case of many enzymatic biosensors, the consumption of oxygen is monitored optically over time or at the endpoint. Trichosporon cutaneum379 and Bacillus subtilis380 bacteria were immobilized into poly(vinyl alcohol) (PVA) and solgel PVA-PVP [(poly(4-vinylpyridine)] networks. A combination of Bacillus licheniformis with Dietzia maris and Marinobacter marinus contained in a sol-gel-PVA matrix381 was used to obtain even lower selectivity and, thus, monitor much more possible pollutants. Compared to the sensor that makes use of B. Licheniformis only, a decrease in LOD from 0.9 to 0.2 mg/L and in response time from 30 to 3.2 min was observed for the multibacteria sensor. In all BOD sensors, the microorganism layer was placed on an oxygensensitive layer of various layouts and materials. These include a quenchable ruthenium(II) complex contained in plasticized PVC,379 in ormosil,381 or in silicone.380 Analytical ranges of the sensors were reported to be from 0 to 110,379 0 to 25,380 and 0.2 to 40 mg/L,381 respectively (expressed as equivalents of a glucose/glutamate BOD standard solution). The first sensor379 possessed moderate stability, and a 30% drop in sensor response was observed after 1 month of storage. The shelf life of the sensors reported later was much higher (a 12% decrease in activity in 1 month for the B. subtilis biosensor380 and only a 5% decrease in 6 months for the multibacteria sensor).381 A biosensor for the organophosphorous pesticide methyl parathion382 was prepared by analogy to the enzyme-type biosensors that made use of organophosphorous hydrolase (OPH). Whole cells of FlaVobacterium sp. containing OPH were immobilized on a glass fiber filter. Hydrolysis of methyl parathion (catalyzed by OHP) results in the formation of p-nitrophenol, which is readily detected by absorbance. The LOD (0.3 µM) and analytical range (4-80 µM) are comparable to the properties of OPH-based enzymatic biosensors. Recombinant E. coli cells immobilized in agarose were placed on a nylon membrane and used for determination of

Optical Biosensors

organophosphorous pesticides.383 These cells are capable of expressing OPH directly on the cell surface, which improves response times because of the low resistance to mass transport of the analytes and products across the cell membrane. The sensor detected as little as 3 µM of paraoxon and 5 µM of coumaphos and could be stored >1 month without a decrease in activity. Evidently, the main advantage of the whole-cell biosensors for pesticides over the enzymatic ones is that no isolation of the enzyme OPH is necessary. This reduces the costs significantly and improves the longterm stability. Arain et al.410 have studied the inhibitory effect of toxic metal ions on the cellular activity of E. coli and P. putida. Microtiter plates were prepared with integrated, fluorescencebased sensors for pH and oxygen, and bacterial respiratorial activity was monitored via the decrease in oxygen partial pressure of the closed system and also via the decrease in pH value. Other applications of such biosensors include respirometry and general toxicity assays.

7.2. External Stimuli-Based Cellular Biosensors Whole-cell biosensors were developed for the determination of various toxicants. For example, Bains384 immobilized Escherichia coli into an agarose membrane and monitored its UV absorbance at the unusual wavelength of 200 nm. In the presence of toxicants (e.g., sodium azide), the cells were stressed so that their metabolism was reduced and a drop in optical density was observed within 15 s. The sensitivity of the system was, however, poor with respect to the requirements of environmental monitoring. Frense et al.385 used algae cells (Scenesdesmus subspicatus) for the determination of environmentally harmful impurities in water. These chlorophyll-containing cells were immobilized on a filter paper that was covered with a thin alginate layer. Pollutants such as atrazine, endrine, and many other pesticides inhibit the electron-transport occurring during photosynthesis. This results in the increase of fluorescence of chlorophyll. The increase of fluorescence is well-related to the concentration of the pesticides, which can be measured at levels of several parts/billion (ppb). The sensor showed comparatively fast response (∼10 min) and good long-term stability in that storage at 4 °C within 6 months did not alter the sensor properties significantly. A similar approach was used by Naessens et al.386 who used algal cells from Chlorella Vulgaris (immobilized on a glass microfiber) for determination of atrazine, simazine, diuron, and other pesticides with high sensitivity (e.g., as little as 5 nM of atrazine when using the sensor in a flow mode). The biosensor showed good storage stability only for a limited period of time (7 days), and significant loss of activity was observed during longer storage. Different bacteria and mutants were found to respond to different pesticide classes387 because they can selectively modify the activity of photosystem II. The microorganisms were immobilized in a BSA-glutaraldehyde network deposited on a porous septum filter that was placed in a flow-through cell. Several flow-through cells were combined as an array to enable simultaneous sensing of several pesticides. The selectivity of the individual sensors, however, remained low enough, because each type of bacteria was sensitive to several classes of pesticides. The long-term stability of the biosensor was rather poor (half-life from 12 to 54 h).

Chemical Reviews, 2008, Vol. 108, No. 2 453

7.3. Genetically Engineered Whole-Cell Biosensors Progress in genetic engineering made possible the creation of a new type of microbial biosensor. It relies on the ability of a cell to respond to an environmentally harmful toxin by expressing specific genes. As a result, the toxin is transported out of the cell. To monitor the process, a reporter gene is fused to the induced gene. In the presence of the toxin, both genes are activated and expressed. A reporter gene usually is responsible for production of species that could be monitored optically, e.g., the green fluorescent protein. In a certain sense, all genetically engineered biosensors are external stimuli biosensors. A variety of biosensors makes use of the lux gene coding for the enzyme luciferase. In contrast to enzymatic bioluminescent sensors, the whole-cell biosensors often are selfcontained because luciferase and other reagents such as the cofactor, flavine mononucleotide, and substrate (a long-chain aldehyde) are produced in vivo. Typical examples of such biosensors are described below. Virta et al.388 developed a mercury biosensor that makes use of Escherichia coli containing the lux gene fused to the mer operon. The latter encodes for resistance to mercury, which is a nonessential and toxic metal for bacteria. The bioluminescence was triggered in the presence of Hg2+. Concentrations as low as 0.1 fM are detectable. A linear dependence is observed up to 0.1 µM of Hg2+. At higher concentrations, the luminescence rapidly drops to zero because of the toxicity of Hg2+ ions. The assay exhibits high selectivity, and no interference by other metals ions (except Cd2+) is observed. Sensitivity to Cd2+ is, however, ∼107fold lower than that to Hg2+ and, therefore, does not really compromise the performance of the assay. Another metal ion biosensor389 makes use of Escherichia coli containing the znt A gene fused to the reporter lac Z gene. While the first is responsible for transporting heavy metal ions out of the cell, the second produces the enzyme β-galactosidase, which cleaves the added substrate fluorescein di-β-D-galactopyranoside. Hence, fluorescence is increased in the presence of heavy metal ions. Individual cells of E. coli were immobilized on an array of 50 000 fibers (L of 2.5 µm) so that each bacterium occupied one individual fiber. Averaging the response from multiple identical sensors improved the signal-to-noise ratio. The LOD for Hg2+ was 100 nM. Unfortunately, no information is given about conceivable interferences by other heavy metal ions. Leth et al.390 have developed a biosensor for copper(II) ion based on a genetically engineered strain of Alcaligenes eutrophus into which was inserted a lux operon from Vibrio fischeri under the control of a copper-induced promoter. As a result, copper ions induce bioluminescence whose intensity is proportional to the concentration of the triggering ion. The cells were immobilized into calcium alginate and agarose gels, which were positioned in a home-made flow-cell, and luminescence intensity was monitored by means of a photodetector. The biosensors based on the two gels showed similar performance, which was highly influenced by the growth medium used. In fact, the analytical range of the sensor for both alginate and agarose was from 0 to 250 µM of Cu2+ when using the Luria-Broth (LODs of 50 and 25 µM, respectively). The use of a modified mineral reconstruction medium resulted in an LOD of 1 µM and an analytical range from 0 to 25 µM for alginate. The authors also showed

454 Chemical Reviews, 2008, Vol. 108, No. 2

that the performance of the biosensor is influenced by the concentration of riboflavin, suggesting the necessity of using a standard composition of nutrient medium. Unfortunately, the activity of the immobilized species was shown to decrease dramatically with time. In fact, a 2 week storage resulted in the complete loss of the activity of the bacteria for both alginate and agarose biogels. After the first 6 days of storage, the activity was, however, almost unchanged for alginate, while it dropped by 7-fold in agarose. Aromatic hydrocarbons are widespread and harmful pollutants that can be successfully detected by whole-cell biosensors. Here, the content of individual hydrocarbons is of less significance than the determination of overall toxicity. For example, Heitzer et al.391 used Pseudomonas fluorescens, which carried the nah G gene fused to a lux reporter gene to design a naphthalene biosensor. The bacteria were physically immobilized in alginate gel that was hardened at elevated temperature in the presence of SrCl2. Exposure of this sensor layer to both naphthalene and its degradation intermediate, salicylate, results in an increase of gene expression and, consequently, an enhancement of bioluminescence. The sensor demonstrated complete reversibility for at least 20 h, but the lowest detectable concentration of naphthalene was 0.12 mM, which is rather high. Ikariyama et al.392 used Pseudomonas putida bearing a xyl R gene (which recognizes benzene and its derivates) fused to the reporter lux gene from firefly. Because firefly luciferase catalyzes a different bioluminescent reaction than bacterial luciferase, addition of the substrate luciferin was necessary. The luminescent signal had a good linear relationship to the concentration of xylenes, which ranged from 0.05 to 1 mM. The response of the biosensor (where P. putida was immobilized onto a polycarbonate membrane) was shown to be much slower than the respective assay in solution and required hours of incubation time to achieve an adequate intensity of bioluminescence. In contrast to the sensors described above, biosensors for toxic chemicals are based on measurement of the reduction of intensity of the bioluminescent reaction when cells experience toxic or lethal conditions. Gil et al.393 immobilized a recombinant E. coli species bearing a lux reporter gene in a solid agar gel located in proximity of the distal end of an optical fiber. The biosensor was used for detection of toxic gases and vapors. As little as 48 ppm of benzene vapor could be detected. The sensor had a response time of ∼10 min and could be stored up to 20 days without degradation of activity. The sensitivity can be improved by increasing the surface that is exposed to vapors and by enhancing the diffusion of vapors, which can be accomplished by addition of glass beads.394 The sensitivity of other strains of bacteria bearing a lux reporter gene also was investigated.395 Shetty et al.396 developed a bioassay for determination of L-arabinose. The binding of the monosaccharide to the ara C regulatory protein was linked to the production of green fluorescent protein (GFP) by the reporter gene in an E. coli strain. The amount of GFP expressed is, thus, directly related to the concentration of L-arabinose. The dynamic range of the assay is from 0.5 µM to 5 mM. The sensitivity to other monosaccharides was ∼10 times lower than to arabinose. A biosensor also was developed where the bacteria were entrapped behind a membrane at the tip of a bifurcated fiber bundle. Although operated similarly to the bioassay, an ∼10fold increase in the LOD was reported.

Borisov and Wolfbeis

Recently, several bioassays were developed for detection of endocrine-disrupting chemicals (EDCs), compounds that affect human health by irregularly modulating endocrine functions. Michelini et al.397 used recombinant Saccharomyces cereVisiae cells that were modified to express the human androgen receptor together with Photinus pyralis luciferase. The assay responds to testosterone in the concentration range from 50 pM to 1 µM. Luciferin needs to be added to the assay solution. Compared to other available methods, the assay is rather fast (150 min of incubation needed for optimal performance against 24 h for other methods). A biosensor also was developed for estrogenic EDCs.398 Genetically modified Saccharomyces cereVisiae cells containing the estrogen receptor were entrapped in hydrogel matrices based on calcium alginate or poly(vinyl alcohol). The LODs for the two EDCs investigated were 0.08 and 0.64 µg/L. The immobilized cells retained their activity for 1 month, however only if stored at -80 °C. A lactate biosensor was reported399 that utilizes a bacterial cytoplasmic membrane isolated from E. coli, which was genetically modified to express its lactate-oxidizing activity. The sensor properties can be tuned by varying the conditions of cultivation. The cytoplasmic membranes were adsorbed on a cellulose disk placed above an oxygen transducersa ruthenium(II)-derived luminescent probe in a silicone matrix. The sensing scheme is based on the consumption of oxygen during oxidation of lactate and is the same as that used in the enzyme based on the use of lactate oxidase. The analytical range of the biosensor is from 0.05 to 5 mM of lactate. In some optical biosensors, even whole tissues have been used as recognition elements. For example, Lundstro¨m et al.400 used fish scales from Labrus ossifagus containing melanophores. Their cells contain pigment granules that are either dispersed or aggregated. Aggregation of the granules is promoted by noradrenaline, which can be monitored optically via the increase in the transmittance of the scale. Addition of the noradrenaline antagonist yohimbine recovers the initial signal. Nanomol quantities of noradrenaline could be measured, and the response time of the sensor was ∼10 min. In summary, it can be stated that cellular biosensors (a) are relatively easy to manufacture; (b) are rather unspecific in the case of catalytic biosensors but fairly specific in the case of external-stimuli biosensors and gene-modified biosensors; (c) possess rather slow response; (d) are selfcontained; (e) are more stable, in general, than enzymatic biosensors but are sensitive to heat and, less so, to frost; and (f) vary over a wide range in terms of sensitivity.

8. Solid Supports for Use in Optical Biosensors, and Other Methods of Immobilization The success of (bio)sensor research and development dependssmore often than anticipatedson the availability of adequate materials. One may differentiate between materials for mechanical sensor supports and materials for use as matrices or membranes that contain the biologically active species, or indicators in the case of catalytic biosensors. These shall be discussed here briefly in addition to the specific examples of immobilization given in section 4.4. There are three kinds of “supports” for optical biosensors. The first one is of the completely inert type. Its only purpose is to serve as a mechanical support to facilitate the handling of planar sensors. The second is of the optical waveguide type and, thus, can serve as an essential compo-

Optical Biosensors

nent in the process of optical interrogation of the sensor material. The third (and most recent) group of supports are the “active” supports such as fluorescent nanoparticles (Qdots), metal films, beads of noble metals, or inorganic or organic micro- and nanoparticles. These can act as microlight sources or quenchers, for example, and thus can actively take part in the spectroscopic scheme. Any of these supports can be preactivated (i.e., made bioreactive) form to enable covalent immobilization of the biocomponent or of other species. Inert supports come in various forms and include films of poly(ethylene terephthalate), which is readily available at low costs and also easy to handle. Other supports include poly(methyl methacrylates) and polycarbonates, with their excellent optical transparency, and polystyrene, which is widely used in microtiterplates (MTPs). In most cases, the chemically responsive material (the sensor “cocktail”) is deposited, or printed, or stamped on such a support, in a groove of this material, or in the wells of a (plastic) MTP. The material, after having been deposited as a thin film on the support, is punched into sensor spots, and these are being placed in disposable sensor devices. The sensor layer is then interrogated by guiding the light beam onto the sensor layer, and reflectivity or fluorescence is measured or interference is measured. It is obvious that the mechanical supports are expected to create no background signal. The support also may act as a waveguide material. Planar waveguides, fiber-optics, and, less often, capillaries have been applied. There are two ways to guide the exciting light to the sensor material. The first is by direct illumination and by collecting luminescence via the waveguide. The second is to use the waveguide for both the exciting beam and for collection of emitted luminescence. Both geometries have their merits (see section 4.3). Waveguide-based sensors are most elegant and, therefore, have found widespread application. Among the third kind of supports, the nanobeads exploit the fact that, because of their intrinsic luminescence, they can act as a donor in FRET assays.55 Metal particles and films, in turn, can act as quenchers or enhancers of luminescence (depending inter alia on the spatial distance between metal and fluorophore and on the kind of metal). A most interesting class of micro- and nanoparticles is represented by the so-called upconverting phosphors (UCPs). They are capable of converting near-infrared light (from lowcost diode lasers) into visible light with fair to high efficiency. Upconversion is not related to 2-photon excitation, which occurs at strong light intensities only. UCPs (mostly oxides, sulfides, andspreferablysfluorides of trivalent lanthanide ions) enable complete elimination of autofluorescence, which commonly impairs the performance of fluorescence-based assays. UCPs are ideal donors for fluorescence resonance energy transfer (FRET)-based assays. UCP-based FRETs have been applied in immunoassays401,402 and in nucleic acid hybridization assays.403,404 Arguably, these methods are at the borderline between solid-phase-based biosensors of the conventional type and of classical solution assays. With respect to materials for use as a bulk matrix for sensors, it is important to remind that the design of such sensors depends on the size of the analyte. Enzyme-based sensors usually digest or metabolize substrates of low or medium molecular weight. Hydrogel matrices, for example, are useful in the case of low-molecular-weight analytes only,

Chemical Reviews, 2008, Vol. 108, No. 2 455

since these can penetrate the bulk matrix, where, in the case of enzyme sensors or whole-cell sensors, metabolism can occur. Immunosensors almost never are designed as bulk sensors even though they could be, at least for small analytes such as atrazine. Immunosensors for large analytes require the antibody to be immobilized on the surface of a support, usually as a nanometer thick coating or film. The situation is similar for DNA sensors where the size and diffusion of the analyte is critical. Whole-cell biosensors (in the majority of cases) have been used for low-molecular-weight analytes and, therefore, have been incorporated into analyte-permeable gels such as from alginates. The surface of a biosensor layer is either covered with a polymer/enzyme matrix (like in most enzymatic biosensors) or directly with biorecognition elements such as antibodies or oligonucleotides. In this case, the surface needs to be made reactive first in order to allow immobilization of a biomolecule. Covalent attachment of a biomolecule to a support is much more commonly used than physical absorption. Numerous cross-linkers and spacers can be used.20,292,296 Proteins such as bovine serum albumin often are also deposited on the surface, so to saturate remaining binding sites. Immobilization of antibodies and oligonucleotides via the (noncovalent) biotin-avidin46,67,68,246,288,299 or biotin-streptavidin47,354,355,371,374 couple is widespread. A support modified with (strept)avidin can be used, in principle, for immobilization of any biotinylated molecule. Alternatively, biotinylated recognition elements can be immobilized via streptavidin onto a surface modified with biotin. Enzymes are often covalently immobilized onto preactivated transparent polyamide or poly(vinylidenedifluoride) membrane supports such as Immunodyne,73,90,140,163,227,228,230,232-234 Biodyne,177,191,225 and Immobilon.164,194,195 Less common supports include eggshell membranes86,154 and swim-bladder membranes88 and were reported to be highly biocompatible. Hydrophilic polymer matrixes are widely used for immobilization of biocomponents (such as enzymes and bioreceptors) and of optical indicators. Sol-gels (whose polarity can vary over a wide range by introducing organic groups to end up with ormosils)401,402 have been often used for immobilization of enzymes,44,84,87,120,131,134,135,380,206,216 bioreceptors,343 and even whole cells.380,381 One major reason for the popularity of sol-gels results from the fact that the activity of biocomponents is retained over a long time. Hydrogels have also become popular36,80,102,118,147,150,212,398 because they do not require a modification of the biological component. Enzymes sometimes are cross-linked with glutaraldehyde and BSA to form a polymer network located on a support or directly in a hydrophilic polymer (e.g., PVA) layer.75,175,176,188,215,217,387 Koncki et al.97 have designed optical biosensors based on the use of thin films of Prussian Blue incorporated into polypyrrole. Other semiconducting organic materials may be used as well. The composite film is deposited on a nonconducting support and used as an optical transducer for flow-through biosensors based on hydrolases and oxidases. Immobilization of glucose oxidase resulted in a glucose biosensor where the film responds to both pH and hydrogen peroxide by a change in its color. Millimolar concentrations can be determined. The biosensor is said to be quite stable owing to the presence of a poly(pyrrolylbenzoic acid) network in the composite material. This organic polymer plays a dual role as a binding agent for inorganic material

456 Chemical Reviews, 2008, Vol. 108, No. 2

and as a functionalized support for strong covalent immobilization of enzyme molecules. Polymer matrixes also can act as a support for the immobilization of indicators used in cellular or enzymatic sensors. Ruthenium(II) polypyridyl complexes are common when oxygen transducers are used (see Table 2). To achieve the desired oxygen sensitivity, they can be adsorbed onto silica beads that are dispersed in a layer of a highly oxygenpermeable silicone layer. Absorption-based and fluorescent pH indicators were employed in enzymatic biosensors using pH and NH3 transducers. Permeability to protons is mandatory in these cases, and the indicators, therefore, often are contained in hydrogels. To prevent leaching of indicators to the sample media, they are sometimes absorbed onto the surface of microbeads.103,194 Dialysis (ultrafiltration) membranes are used in certain sensor types.41,42,48,49,221,222,224 They allow small analytes to diffuse freely in and out of a chamber and interact there with a biorecognition element. The latter is either large enough itself or is conjugated with a large molecule in order to not leach out of the chamber. In contrast to indicators, labels do not respond to substrates or reaction products but render a biomolecule detectable. Fluorescent labels are the most common ones. Fluoresceins, rhodamines, cyanine dyes, and numerous others are commercially available403 and widely used (see Table 3). Ideally, a label should absorb in visible light to reduce background fluorescence, be bright, and be inert. Brightness (defined as the product of molar absorbance and quantum yield) is particularly significant and should exceed 30 000 M-1‚cm-1). Since fluoresceins are viable pH indicators, thorough control of pH is essential for these labels. Luminescent colloidal semiconductor nanocrystals (quantum dots, QDs) also represent viable labels,44,55,56 despite their cell toxicity and difficult surface chemistry, and can largely expand the range of useful fluorophores for biosensors. The group of Seeger have found404 that biotin-functionalized cellulose monolayers can act as a new kind of support and have used it for the fluorescent detection of single molecules via laser-induced confocal single-molecule spectroscopy in glass-bottom microplates. Gold nanobeads can be used to increase the brightness of fluorescent biosensors.37,272,345,405

9. Outlook Optical biosensing has experienced a substantial growth despite the usual critical comments of certain “experts” that expect new technologies to make a breakthrough (mainly in commercial terms) within a few years and despite the overoptimistic presentations of certain researchers, which often does more harm to a new technology than supporting it. Optical biosensor technology is not a matter of spectroscopy only, or of material sciences, or any other single discipline, but rather requires various kinds of scientists to cooperate in order to end up with a viable biosensor scheme and, ideally, commercial products. Optical biosensors have numerous applications, and not each scheme will be applicable to any given analyte. Moreover, methods that may work for a specific analyte in a certain matrix may not even work for the same analyte in another matrix. This fact is but one of the reasons why biosensors, in a commercial sense, are not as successful as was expected initially. The trend toward multianalyte sensing and toward biosensor arrays is obvious, even though certain single sensor

Borisov and Wolfbeis

spots in an array may not be needed in any conceivable application. DNA arrays are technically the most advanced, not the least because DNA is built from 4 nucleotides only, which makes synthetic and surface chemistry comparably more simply than in the case of protein arrays. Enzyme arrays are rather established and have found application in clinical analyzers for glucose, urea, cholesterol, and lactate. Protein arrays are most versatile but also can be most complex, and this has hampered their technical realization. Proteins not only come in a variety of sequences (of 20 different amino acids!) but also in a variety of tertiary and quaternary structures, which makes labeling and immobilization an experience in each single case. Unfortunately, the current terror hype has directed research in protein arrays away from health and environment into other areas, which implies a massive redirection of tax money and appears not to really represent a useful long-term investment.

10. List of Abbreviations and Acronyms Abs. AChE ADH AOx AMPT ASF BL BSA BTB BTP b. luciferase CCA ChOx CLum. CF CFU CDNB Cy-5 2,4-D DDAO DFP f. luciferase EDC ELum. EuTc FITC FRET GAH GDH GFOR GLOx GlDH GOx GPT GSH GST hCG HPOx HPTS IgG I. LDH LOx Lum. LyOx NHS OPH

absorption acetylcholine esterase alcohol dehydrogenase alcohol oxidase 2-(2-acetoxy-5-methylphenylazo)-N-methyl-1,3thiazolium methosulfate African swine fever bioluminescence bovine serum albumine bromothymol blue benzo[a]pyrene tetraol bacterial luciferase chlorendic capronic acid choline oxidase chemiluminescence 5(and 6)-carboxyfluorescein colony-forming units 1-chloro-2,4-dinitrobenzene carboxymethylindocyanine succunimidyl ester 2,4-dinitrophenoxyacetic acid 7-hydroxy-9H-1,3-dichloro-9,9-dimethylacridin-2-one diisopropyl phosphorofluoridate firefly luciferase ethyl-3-[1-dimethylaminopropyl]carbodiimide electroluminescence europium(III) tetracycline fluorescein isothiocyanate fluorescence resonance energy transfer glutaminase glucose dehydrogenase glucose-fructose oxidoreductase glutamate oxidase glutamate dehydrogenase glucose oxidase glutamic-pyruvic transmitase glutathione glutathione-S-transferase chorionic gonadotrophin horseradish peroxidase 8-hydroxypyrene-1,3,6-trisulfonate immunoglobulin intensity lactate dehydrogenase lactate oxidase luminescence lysine oxidase N-hydroxysuccinimide organophosphate hydrolase

Optical Biosensors OR ORP PEG PMMA POx PSA PtOEP RDX Refl. Ru-bipy Ru-phen Ru-dpp

oxidoreductase organophosphorous pesticides poly(ethylene glycol) polymethylmethacrylate peroxidase prostate-specific antigen platinum(II) octaethylporphyrin hexaxydro-1,3,5-trinitro-1,3,5-triazine reflectance ruthenium(II) trisbipyridyl ruthenium(II) tris(1,10-phenanthroline) ruthenium(II) tris(4,7-diphenyl-1,10-phenanthroline) SEB staphylococcal Enterotoxin B SNARF seminaphthofluorescein TCPB 2,4,6-trichlorophenoxybutyrate Ti(IV) reagent titanium(IV) + 24(5-bromopyridyi)azo)5-(Npropyl-N-sulfopropylamino)phenol TMR tetramethylrhodamine TNB trinitrobenzene TNT trinitrotoluene TRITC tetramethylrhodamine-5-isothiocyanate triazine derivative 4-chloro-6-(isopropylamine)-1,3,5-triazine-2-(6aminohexane carboxylic acid) TSH thyroid stimulating hormone XOx xanthine oxidase

11. References (1) Turner, A. P. F., Karube, I., Wilson, G. S., Eds. Biosensors: Fundamentals and Applications; Oxford University Press: Oxford, U.K., 1987. (2) Edmonds, T. E., Ed. Chemical Sensors; Blackie and Son, Ltd.: London, U.K., 1988. (3) Blum, L. J.; Coulet, P. R. Biosensor Principles and Applications; Marcel Dekker Inc.: New York, 1991. (4) Wolfbeis, O. S., Ed. Fiber Optic Chemical Sensors and Biosensors; CRC Press: Boca Raton, FL, 1991; Vol. 1 and 2. (5) Hall, E. A. H. Biosensors; Open University Press: Buckingham, U.K., 1990. (6) Valca´rel, M.; Luque de Castro, M. D. Flow-Through (Bio)Chemical Sensors; Elsevier: Amsterdam, The Netherlands, 1994. (7) Taylor, R. F.; Schultz, J. S., Eds. Handbook of Chemical and Biological Sensors; Inst. Physics Publ.: Bristol, Philadelphia, PA, 1996. (8) Vincenzini, P., Dori, L., Eds. Solid State Chemical and Biochemical Sensors; Techna Publ.: Faenza, Italy, 1999. (9) Ramsden, J. J. J. Mol. Recognit. 1997, 10, 109. (10) Mehrvar, M.; Bis, C.; Scharer, J. M.; Moo-Young, M.; Luong, J. Anal. Sci. 2000, 16, 677. (11) Kuswandi, B.; Andres, R.; Narayanaswamy, R. Analyst 2001, 126, 1469. (12) Schobel, U.; Barzen, C.; Gauglitz, G. Fresenius J. Anal. Chem. 2000, 366, 646. (13) Marazuela, M. D.; Moreno-Bondi, M. C. Anal. Bioanal. Chem. 2002, 372, 664. (14) Monk, D. J.; Walt, D. R. Anal. Bioanal. Chem. 2004, 379, 931. (15) Rogers, K. R.; Mulchandani, A. Affinity Biosensors; Humana Press: Totowa, NJ, 1998. (16) Pickup, J. C.; Hussain, F.; Evans, N. D.; Rolinski, O. J.; Birch, D. J. S. Biosens. Bioelectron. 2005, 20, 2555. (17) Marquette, C. A.; Blum, L. J. Biosens. Bioelectron. 2006, 21, 1424. (18) Leca-Bouvier, B.; Blum, L. J. Anal. Lett. 2005, 38, 1491. (19) Taitt, C. R.; Anderson, G. P.; Ligler, F. S. Biosens. Bioelectron. 2005, 20, 2470. (20) Choi, M. M. F. Microchim. Acta 2004, 148, 107. (21) Wolfbeis, O. S. Anal. Chem. 2004, 76, 3269. (22) Wolfbeis, O. S. Anal. Chem. 2006, 78, 3859-3873. (23) Brogan, K. L.; Walt, D. R. Curr. Opin. Chem. Biol. 2005, 9, 494. (24) Nagl, S.; Schaeferling, M.; Wolfbeis, O. S. Microchim. Acta 2005, 151, 1. (25) Walt, D. R. Science 2005, 308, 217. (26) Walt, D. R. BioTechniques 2006, 41, 529. (27) Wolfbeis, O. S. Anal. Chem. 2006, 78, 3859. (28) Go¨pel, W., Hesse, J., Zemel, J. N., Eds. Sensors: A ComprehensiVe SurVey; VCH Publishers: Weinheim, Germany, 1991; Vol. 2, Part 1, pp 1-27. (29) Hulanicki, A.; Geab, S.; Ingman, F. Pure Appl. Chem. 1991, 63, 1247.

Chemical Reviews, 2008, Vol. 108, No. 2 457 (30) Thevenot, D. R.; Toth, K.; Durst, R. A.; Wilson, G. S. Biosens. Bioelectron. 2001, 16, 121. (31) Hilty, C.; Lowery, T. J.; Wemmer, D. E.; Pines, A. Angew. Chem., Int. Ed. 2005, 45, 70. (32) Valcarcel, M.; Luque de Castro, M. D. Flow-Through (Bio)Chemical Sensors; Elsevier: Amsterdam, The Netherlands, 1994. (33) Parker, C. A. Photoluminescence of Solutions; Elsevier: Amsterdam, The Netherlands, 1981. (34) Yadavalli, V. K.; Koh, W. G.; Lazur, G. J.; Pishko, M. V. Sens. Actuators, B 2004, 97, 290. (35) Xu, H.; Aylott, J. W.; Kopelman, R. Analyst 2002, 127, 1471. (36) Russell, R. J.; Pishko, M. V.; Simonian, A. L.; Wild, J. R. Anal. Chem. 1999, 71, 4909. (37) Simonian, A. L.; Good, T. A.; Wang, S. S.; Wild, J. R. Anal. Chim. Acta 2005, 534, 69. (38) Viveros, L.; Paliwal, S.; McCrae, D.; Wild, J.; Simonian, A. L. Sens. Actuators, B 2006, 115, 150. (39) Tsai, H. C.; Doong, R. A. Anal. Biochem. 2004, 334, 183. (40) Ueberfeld, J.; Walt, D. R. Anal. Chem. 2004, 76, 947. (41) Anderson, F. P.; Miller, W. G. Clin. Chem. 1988, 34, 1417. (42) Hanbury, C. M.; Miller, W. G.; Harris, R. B. Biosens. Bioelectron. 1996, 11, 1129. (43) Ko, S.; Grant, S. A. Biosens. Bioelectron. 2006, 21, 1283. (44) Aoyagi, S.; Kudo, M. Biosens. Bioelectron 2005, 20, 1680. (45) Tyagi, S.; Kramer, F. R. Nat. Biotechnol. 1996, 14, 303. (46) Fang, X.; Liu, X.; Schuster, S.; Tan, W. J. Am. Chem. Sos. 1999, 121, 2921. (47) Liu, X.; Tan, W. Anal. Chem. 1999, 71, 5054. (48) Meadows, D.; Schultz, J. S. Talanta 1988, 35, 145. (49) Astles, J. R.; Miller, W. G. Sens. Actuators, B 1993, 11, 73. (50) Song, X.; Swanson, B. I. Anal. Chem. 1999, 71, 2097. (51) Thompson, R. B.; Ge, Z.; Patchan, M.; Huang, C. C.; Fierke, C. A. Biosens. Bioelectron. 1996, 11, 557. (52) Thompson, R. B.; Patchan, M. W. Anal. Biochem. 1995, 227, 123. (53) Ye, K.; Schultz, J. S. Anal. Chem. 2003, 75, 3451. (54) Chinnayelka, S.; McShane, M. J. Anal. Chem. 2005, 77, 5501. (55) Medintz, I. L.; Clapp, A. R.; Mattoussim, H.; Goldman, E. R.; Fisher, B.; Mauro, J. M. Nat. Mater. 2003, 2, 630. (56) Sandros, M. G.; Gao, D.; Benson, D. E. J. Am. Chem. Soc. 2005, 127, 12198. (57) Zeng, H. H.; Thompson, R. B.; Maliway, B. P.; Fones, G. R.; Moffett, J. W.; Fierke, C. A. Anal. Chem. 2003, 75, 6807. (58) Papkovsky, D. B. Sens. Actuators, B 1993, 11, 293. (59) Papkovsky, D. B.; Olah, J.; Kurochkin, I. N. Sens. Actuators, B 1993, 11, 525. (60) Papkovsky, D. B.; O’Riordan, T. C.; Guilbault, G. G. Anal. Chem. 1999, 71, 1568. (61) Schaeferling, M.; Wu, M.; Wolfbeis, O. S. J. Fluoresc. 2004, 14, 561. (62) D’Auria, S.; Gryczynski, Z.; Gryczynski, I.; Rossi, M.; Lakowicz, J. R. Anal. Biochem. 2000, 283, 83. (63) Dattelbaum, J. D.; Lakowicz, J. R. Anal. Biochem. 2001, 291, 89. (64) Sanz-Vicente, I.; Castillo, J. R.; Galban, J. Talanta 2005, 65, 946. (65) Hoefelschweiger, B. K.; Pfeifer, L.; Wolfbeis, O. S. J. Biomol. Screening 2005, 10, 685. (66) Heideman, R. G.; Kooyman, R. P. H.; Greve, J. Sens. Actuators, B 1991, 4, 297. (67) Stamm, Ch.; Lukosz, W. Sens. Actuators, B 1996, 31, 203. (68) Clerc, D.; Lukosz, W. Sens. Actuators, B 1997, 40, 53. (69) Brecht, A.; Piehler, J.; Lang, G.; Gauglitz, G. Anal. Chim. Acta 1995, 311, 289. (70) Szekacs, A.; Trummer, N.; Adanyi, N.; Varadi, M.; Szendro, I. Anal. Chim. Acta 2003, 487, 31. (71) Ruckstuhl, T.; Rankl, M.; Seeger, S. Biosens. Bioelectron. 2003, 18, 1193. (72) Zourob, M.; Hawkes, J. J.; Coakley, W. T.; Brown, B. J. T.; Fielden, P. R.; McDonnell, M. B.; Goddard, N. J. Anal. Chem. 2005, 77, 6163. (73) Trettnak, W.; Leiner, M. J.; Wolfbeis, O. S. Analyst 1988, 113, 1519. (74) Dremel, B. A.; Schaffar, B. P.; Schmid, R. D. Anal. Chim. Acta 1989, 225, 293. (75) Schaffar, B. P.; Wolfbeis, O. S. Biosens. Bioelectron. 1990, 5, 137. (76) Moreno-Bondi, M. C.; Wolfbeis, O. S.; Leiner, M. J.; Schaffar, B. P. Anal. Chem. 1990, 62, 2377. (77) Dremel, B. A.; Li, S. Y.; Schmid, R. D. Biosens. Bioelectron. 1992, 7, 133. (78) Andres, K. D.; Wehnert, G.; Thordsen, O.; Scheper, T. Sens. Actuators, B 1993, 11, 395. (79) Valencia-Gonzalez, M. J.; Liu, Y. M.; Diaz-Garcia, M. E.; SanzMedel, A. Anal. Chim. Acta 1993, 283, 439. (80) Li, L.; Walt, D. R. Anal. Chem. 1995, 67, 3746. (81) Neubauer, A.; Pum, D.; Sleytr, U. B.; Klimant, I.; Wolfbeis, O. S. Biosens. Bioelectron. 1996, 11, 317. (82) Rosenzweig, Z.; Kopelman, R. Anal. Chem. 1996, 68, 1408.

458 Chemical Reviews, 2008, Vol. 108, No. 2 (83) Rosenzweig, Z.; Kopelman, R. Sens. Actuators, B 1996, 35-36, 475. (84) Wu, X.; Choi, M. M. F.; Xiao, D. Analyst 2000, 125, 157. (85) Wolfbeis, O. S.; Oehme, I.; Papkovskaya, N.; Klimant, I. Biosens. Bioelectron. 2000, 15, 69. (86) Choi, M. M. F.; Wilfred, S. Analyst 2001, 126, 1558. (87) Wu, X. J.; Choi, M. M. F. Anal. Chim. Acta 2004, 514, 219. (88) Zhou, Z.; Qiao, L.; Zhang, P.; Xiao, D.; Choi, M. M. F. Anal. Bioanal. Chem. 2005, 383, 673. (89) Matsubara, C.; Kudo, K.; Kawashita, T.; Takamura, K. Anal. Chem. 1985, 57, 1107. (90) Abdel-Latif, M. S.; Guilbault, G. G. Anal. Chim. Acta 1989, 221, 11. (91) Blum, L. J. Enzyme Microb. Technol. 1993, 15, 407. (92) Kremeskoetter, J.; Wilson, R.; Schiffrin, D. J.; Luff, B. J.; Wilkinson, J. S. Meas. Sci. Technol. 1995, 6, 1325. (93) Zhou, X.; Arnold, M. A. Anal. Chim. Acta 1995, 304, 147. (94) Marquette, C. A.; Blum, L. J. Anal. Chim. Acta 1999, 381, 1. (95) Harms, D.; Meyer, J.; Westerheide, L.; Krebs, B.; Karst, U. Anal. Chim. Acta 1999, 401, 83. (96) Marquette, C. A.; Leca, B. D.; Blum, L. J. Luminescence 2001, 16, 159. (97) Koncki, R.; Lenarczuk, T.; Radomska, A.; Glab, S. Analyst 2001, 126, 1080. (98) Lenarczuk, T.; Wencel, D.; Glab, S.; Koncki, R. Anal. Chim. Acta 2001, 447, 23. (99) Zhu, L.; Li, Y.; Zhu, G. Sens. Actuators, B 2002, 86, 209. (100) Zhu, L.; Li, Y.; Tian, F.; Xu, B.; Zhu, G. Sens. Actuators, B 2002, 84, 265. (101) Wolfbeis, O. S.; Schaeferling, M.; Duerkop, A. Microchim. Acta 2003, 143, 221. (102) Heo, J.; Crooks, R. M. Anal. Chem. 2005, 77, 6843. (103) Trettnak, W.; Leiner, M. J.; Wolfbeis, O. S. Biosens. 1988, 4, 15. (104) McCurley, M. F. Biosens. Bioelectron. 1994, 9, 527. (105) Piletsky, S. A.; Panasyuk, T. L.; Piletskaya, E. V.; Sergeeva, T. A.; El’skaya, A. V.; Pringsheim, E.; Wolfbeis, O. S. Fresenius J. Anal. Chem. 2000, 366, 807. (106) Voelkl, K. P.; Opitz, N.; Luebbers, D. W. AdV. Exp. Med. Biol. 1988, 222, 199. (107) Knobbe, E. T.; Dunn, B.; Gold, M. Proc. SPIE 1989, 906, 39. (108) Cox, J. A.; Hensley, P. M.; Loch, C. L. Microchim. Acta 2003, 142, 1. (109) Wolfbeis, O. S.; Leiner, M. J. P.; Posch, H. E. Microchim. Acta 1986, 90, 359. (110) Alford, P. C.; Cook, M. J.; Lewis, A. P.; McAuliffe, S. G.; Skarda, V.; Thomson, A. J. J. Chem. Soc., Perkin Trans. II 1985, 5, 705. (111) Suchomel, P.; Wolfbeis, O. S. Unpublished results, 2007. (112) Pasic, A.; Koehler, H.; Schaupp, L.; Pieber, T. R.; Klimant, I. Anal. Bioanal. Chem. 2006, 386, 1293. (113) Pasic, A.; Koehler, H.; Klimant, I.; Schaupp, L. Sens. Actuators, B 2007, 122, 60. (114) Liebsch, G.; Klimant, I.; Wolfbeis, O. S. AdV. Mater. 1999, 11, 1296. (115) Mongey, K. F.; Vos, J. G.; MacCraith, B. D.; McDonagh, C. M.; Coates, C.; McGarvey, J. J. J. Mater. Chem. 1997, 7, 1473. (116) Borisov, S. M.; Wolfbeis, O. S. Anal. Chem. 2006, 78, 5094. (117) Borisov, S. M.; Vasylevska, A. S.; Krause, Ch.; Wolfbeis, O. S. AdV. Funct. Mater. 2006, 16, 1536. (118) Freeman, T. M.; Seitz, W. R. Anal. Chem. 1978, 50, 1243. (119) Preuschoff, F.; Spohn, U.; Janasek, D.; Weber, E. Biosens. Bioelectron. 1994, 9, 543. (120) Navas Diaz, A.; Ramos Peinado, M. C.; Torijas Minduez, M. C. Anal. Chim. Acta 1998, 363, 221. (121) Ramos, M. C.; Troijas, M. C.; Navas Diaz, A. Sens. Actuators, B 2001, 73, 71. (122) Schubert, F.; Wang, F.; Rinneberg, H. Microchim. Acta 1995, 121, 237. (123) Kim, S. H.; Kim, B.; Yadavalli, V. K.; Pishko, M. V. Anal. Chem. 2005, 77, 6828. (124) Pastor, I.; Esquembre, R.; Micol, V.; Mallavia, R.; Mateo, C. R. Anal. Biochem. 2004, 334, 335. (125) Wolfbeis, O. S.; Duerkop, A.; Wu, M.; Lin, Z. Angew. Chem., Int. Ed. 2002, 41, 4495. (126) Trettnak, W.; Wolfbeis, O. S. Anal. Chim. Acta 1989, 221, 195. (127) Trettnak, W.; Wolfbeis, O. S. Fresenius J. Anal. Chem. 1989, 334, 427. (128) Chudobova, I.; Vrbova, E.; Kodicek, M.; Janovcova, J.; Kas, J. Anal. Chim. Acta 1996, 319, 103. (129) Sierra, J. F.; Galban, J.; Castillo, J. R. Anal. Chem. 1997, 69, 1471. (130) De Marcos, S.; Sanz, V.; Andreu, Y.; Galban, J. Microchim. Acta 2006, 153, 163. (131) De Marcos, S.; Galindo, J.; Sierra, J. F.; Galban, J.; Castillo, J. R. Sens. Actuators, B 1999, 57, 227. (132) Sanz, V.; Galban, J.; de Marcos, S.; Castillo, J. R. Talanta 2003, 60, 415.

Borisov and Wolfbeis (133) Sanz, V.; de Marcos, S.; Galban, J. Biosens. Bioelectron. 2007, 22, 956. (134) Aylott, J. W.; Richardson, D. J.; Russell, D. A. Analyst 1997, 122, 77. (135) Ferretti, S.; Lee, S. K.; MacCraith, B. D.; Oliva, A. G.; Richardson, D. J.; Russell, D. A.; Sapsford, K. E.; Vidal, M. M. Analyst 2000, 125, 1993. (136) De Marcos, S.; Galban, J.; Albajez, R.; Castillo, J. R. Anal. Chim. Acta 1997, 343, 117. (137) De Marcos, S.; Galban, J.; Alonso, C.; Castillo, J. R. Analyst 1997, 122, 355. (138) Galban, J.; de Marcos, S.; Segura, P.; Castillo, J. R. Anal. Chim. Acta 1994, 299, 277. (139) Blankenstein, G.; Preuschoff, F.; Spohn, U.; Mohr, K. H.; Kula, M. R. Anal. Chim. Acta 1993, 271, 231. (140) Spohn, U.; Preuschoff, F.; Blankenstein, G.; Janasek, D.; Kula, M. R.; Hacker, A. Anal. Chim. Acta 1995, 303, 109. (141) Kar, S.; Arnold, M. A. Anal. Chem. 1992, 64, 2438. (142) Cattaneo, M. V.; Male, K. B.; Luong, J. H. Biosens. Bioelectron. 1992, 7, 569. (143) Cattaneo, M. V.; Luong, J. H. Biotechnol. Bioeng. 1993, 41, 659. (144) Dremel, B. A.; Yang, W.; Schmid, R. D. Anal. Chim. Acta 1990, 234, 107. (145) Wolfbeis, O. S.; Posch, H. E. Fresenius J. Anal. Chem. 1988, 332, 255. (146) Mitsubayashi, K.; Kon, T.; Hashimoto, Y. Biosens. Bioelectron. 2003, 19, 193. (147) Wu, X. J.; Choi, M. M. F. Anal. Chem. 2004, 76, 4279. (148) Trettnak, W.; Wolfbeis, O. S. Anal. Biochem. 1990, 184, 124. (149) Marazuela, M. D.; Cuesta, B.; Moreno-Bondi, M. C.; Quejido, A. Biosens. Bioelectron. 1997, 12, 233. (150) Wu, X. J.; Choi, M. M. F. Anal. Chem. 2003, 75, 4019. (151) Xie, X.; Shakhsher, Z.; Suleiman, A. A.; Guilbault, G. G. Talanta 1994, 41, 317. (152) Li, X.; Rosenzweig, Z. Anal. Chim. Acta 1997, 353, 263. (153) Wu, X. J.; Choi, M. M. F.; Wu, X. M. Analyst 2004, 129, 1143. (154) Xiao, D.; Choi, M. M. F. Anal. Chem. 2002, 74, 863. (155) Opitz, N.; Luebbers, D. W. Talanta 1988, 35, 123. (156) Marazuela, M. D.; Moreno-Bondi, M. C. Anal. Chim. Acta 1998, 374, 19. (157) Kotsira, V. P.; Clonis, Y. D. J. Agric. Food Chem. 1998, 46, 3389. (158) Trettnak, W.; Wolfbeis, O. S. Anal. Lett. 1989, 22, 2191. (159) Blum, L. J.; Gautier, S. M.; Berger, A.; Michel, P. E.; Coulet, P. R. Sens. Actuators, B 1995, 29, 1. (160) Xie, X.; Suleiman, A. A.; Guilbaut, G. G.; Yang, Z.; Sun, Z. Anal. Chim. Acta 1992, 266, 325. (161) Tsafack, V. C.; Marquette, C. A.; Pizzolato, F.; Blum, L. J. Biosens. Bioelectron. 2000, 15, 125. (162) Preuschoff, F.; Spohn, U.; Weber, E.; Unverhau, K.; Mohr, K. H. Anal. Chim. Acta 1993, 280, 185. (163) Hlavay, J.; Guilbault, G. G. Anal. Chim. Acta 1994, 299, 91. (164) Hlavay, J.; Haemmerli, S. D.; Guilbault, G. G. Biosens. Bioelectron. 1994, 9, 189. (165) Rauch, P.; Ferri, E. N.; Girotti, S.; Rauchova, H.; Carrea, G.; Bovara, R.; Fini, F.; Roda, A. A. Anal. Biochem. 1997, 245, 133. (166) Tsafack, V. C.; Marquette, C. A.; Leca, B.; Blum, L. J. Analyst 2000, 125, 151. (167) Gong, Z.; Zhang, Z. Anal. Lett. 1996, 29, 695. (168) Zhujun, Z.; Zhilong, G.; Wangbai, M. Microchem. J. 1995, 52, 131. (169) Marquette, C. A.; Degiuli, A.; Blum, L. J. Biosens. Bioelectron. 2003, 19, 433. (170) Marquette, C. A.; Thomas, D.; Degiuli, A.; Blum, L. J. Anal. Bioanal. Chem. 2003, 377, 922. (171) Corgier, B. P.; Marquette, C. A.; Blum, L. J. Anal. Chim. Acta 2005, 538, 1. (172) Goldfinch, M. J.; Lowe, C. R. Anal. Biochem. 1984, 138, 430. (173) Yerian, T. D.; Christian, G. D.; Ruzicka, J. Anal. Chem. 1988, 60, 1250. (174) Busch, M.; Hoebel, W.; Polster, J. J. Biotechnol. 1993, 31, 327. (175) Kulp, T. J.; Camins, I.; Angel, S. M.; Munkholm, C.; Walt, D. R. Anal. Chem. 1987, 59, 2849. (176) Fuh, M. R. S.; Burgess, L. W.; Christian, G. D. Anal. Chem. 1988, 60, 433. (177) Xie, X.; Suleiman, A. A.; Guilbault, G. G. Biotechnol. Bioeng. 1992, 39, 1147. (178) Hoebel, W.; Papperger, A.; Polster, J. Biosens. Bioelectron. 1992, 7, 549. (179) Healey, B. G.; Walt, D. R. Anal. Chem. 1995, 67, 4471. (180) Yerian, T. D.; Christian, G. D.; Ruzicka, J. Analyst 1986, 111, 865. (181) Gauglitz, G.; Reichert, M. Sens. Actuators, B 1992, 6, 83. (182) Scheper, Th.; Brandes, W.; Maschke, H.; Ploez, F.; Mueller, C. J. Biotechnol. 1993, 31, 345.

Optical Biosensors (183) Busch, M.; Gutberlet, F.; Hoebel, W.; Polster, J.; Schmidt, H. L.; Schwenk, M. Sens. Actuators, B 1993, 11, 407. (184) de Marcos, S.; Hortigu¨ela, R.; Galban, J.; Castillo, J. R.; Wolfbeis, O. S. Microchim. Acta 1999, 130, 267. (185) Koncki, R.; Wolfbeis, O. S. Biosens. Bioelectron. 1999, 14, 87. (186) Brennan, J. D.; Brown, R. S.; Manna, A. D.; Kallury, K. M.; Piunno, P. P.; Krull, U. J. Sens. Actuators, B 1993, 11, 109. (187) Schalkhammer, Th.; Lobmaier, Ch.; Pittner, F.; Leitner, A.; Brunner, H.; Aussenegg, F. R. Sens. Actuators, B 1995, 24-25, 166. (188) Rhines, T. D.; Arnold, M. A. Anal. Chim. Acta 1989, 227, 387. (189) Xie, X.; Suleiman, A. A.; Guilbault, G. G. Anal. Lett. 1990, 23, 2143. (190) Xie, X.; Suleiman, A. A.; Guilbault, G. G. Talanta 1991, 38, 1197. (191) Wolfbeis, O. S.; Li, H. Biosens. Bioelectron. 1993, 8, 161. (192) Kawabata, Y.; Sugamoto, H.; Imasaka, T. Anal. Chim. Acta 1993, 283, 689. (193) Stamm, Ch.; Seiler, K.; Simon, W. Anal. Chim. Acta 1993, 282, 229. (194) Sansubrino, A.; Mascini, M. Biosens. Bioelectron. 1994, 9, 207. (195) Mascini, M. Sens. Actuators, B 1995, 29, 121. (196) Chen, H.; Wang, E. Anal. Lett. 2000, 33, 997. (197) Arnold, M. A. Anal. Chem. 1985, 57, 565. (198) Freeman, T. M.; Bachas, L. G. Biosens. Bioelectron. 1992, 7, 49. (199) Trettnak, W.; Reininger, F.; Zinterl, E.; Wolfbeis, O. S. Sens. Actuators, B 1993, 11, 87. (200) Navas Diaz, A.; Ramos Peinado, M. C. Sens. Actuators, B 1997, 38-39, 426. (201) Choi, J. W.; Kim, Y. K.; Lee, I. H.; Min, J.; Lee, W. H. Biosens. Bioelectron. 2001, 16, 937. (202) Choi, J. W.; Kim, Y. K.; Song, S. Y.; Lee, I. H.; Lee, W. H. Biosens. Bioelectron. 2003, 18, 1461. (203) Choi, J. W.; Kim, Y. K.; Oh, B. K.; Song, S. Y.; Lee, H. L. Biosens. Bioelectron. 2003, 18, 591. (204) Xavier, M. P.; Vallejo, B.; Marazuela, M. D.; Moreno-Bondi, M. C.; Baldini, F.; Falai, A. Biosens. Bioelectron. 2000, 14, 895. (205) Andreou, V. G.; Clonis, Y. D. Biosens. Bioelectron. 2002, 17, 61. (206) Doong, R. A.; Tsai, H. C. Anal. Chim. Acta 2001, 434, 239. (207) Tsai, H. C.; Doong, R. A. Biosens. Bioelectron. 2005, 20, 1796. (208) Vamvakaki, V.; Fournier, D.; Chaniotakis, N. A. Biosens. Bioelectron. 2005, 21, 384. (209) Mulchandani, A.; Pan, S.; Chen, W. Biotechnol. Prog. 1999, 15, 130. (210) Preininger, C. Microchim. Acta 1999, 130, 209. (211) Kuswandi, B. Anal. Bioanal. Chem. 2003, 376, 1104. (212) Walker, J. P.; Asher, S. A. Anal. Chem. 2005, 77, 1596. (213) Narayanaswamy, R.; Sevilla, F. Anal. Lett. 1988, 21, 1165. (214) Wangsa, J.; Arnold, M. A. Anal. Chem. 1988, 60, 1080. (215) Zhang, W.; Chang, H.; Rechnitz, G. A. Anal. Chim. Acta 1997, 350, 59. (216) Li, C. I.; Lin,Y. H.; Shin, C. L.; Tsaur, J. P.; Chau, L. K. Biosens. Bioelectron. 2002, 17, 323. (217) Wang, A.-J.; Arnold, M. A. Anal. Chem. 1992, 64, 1051. (218) Cordek, J.; Wang, X.; Tan, W. Anal. Chem. 1999, 71, 1529. (219) Walters, B. S.; Nielsen, T. J.; Arnold, M. A. Talanta 1988, 35, 151. (220) Williams, A. K.; Hupp, J. T. J. Am. Chem. Soc. 1998, 120, 4366. (221) Scheper, T.; Buckmann, A. F. Biosens. Bioelectron. 1990, 5, 125. (222) Schelp, C.; Scheper, T.; Buckmann, A. F.; Reardon, K. F. Anal. Chim. Acta 1991, 225, 223. (223) Lee, S. J.; Scheper, T.; Buckmann, A. F. Biosens. Bioelectron. 1994, 9, 29. (224) Lee, S. J.; Saleemuddin, M.; Scheper, T.; Loos, H.; Sahn, H. J. Biotechnol. 1994, 36, 39. (225) Gautier, S. M.; Blum, L. J.; Coulet, P. R. Biosens. 1989, 4, 181. (226) Blum, L. J.; Gautier, S. M.; Coulet, P. R. Anal. Lett. 1988, 21, 717. (227) Blum, L. J.; Gautier, S. M.; Coulet, P. R. Anal. Chim. Acta 1989, 222, 331. (228) Gautier, S. M.; Blum, L. J.; Coulet, P. R. Anal. Chim. Acta 1990, 235, 243. (229) Gautier, S. M.; Blum, L. J.; Coulet, P. R. Sens. Actuators, B 1990, 1, 580. (230) Gautier, S. M.; Blum, L. J.; Coulet, P. R. Anal. Chim. Acta 1991, 243, 149. (231) Gautier, S. M.; Blum, L. J.; Coulet, P. R. Anal. Chim. Acta 1991, 255, 253. (232) Gautier, S. M.; Blum, L. J.; Coulet, P. R. Anal. Chim. Acta 1992, 266, 331. (233) Gautier, S. M.; Blum, L. J.; Coulet, P. R. J. Biolumin. Chemilumin. 1990, 5, 57. (234) Blum, L. J.; Gautier, S. M.; Coulet, P. R. J. Biotechnol. 1993, 31, 357. (235) Michael, P. E.; Gautier-Sauvigne, S. M.; Blum, L. J. Talanta 1998, 47, 169. (236) Schubert, F. Sens. Actuators, B 1993, 11, 531. (237) Schaeferling, M.; Wolfbeis, O. S. Chem.sEur. J. 2007, 13, 4342. (238) Schrenkhammer, P.; Rosnizeck, I. D.; Duerkop, A.; Wolfbeis, O. S.; Schaeferling, M. J. Biomol. Screening 2008, in press.

Chemical Reviews, 2008, Vol. 108, No. 2 459 (239) Tromberg, B. J.; Sepaniak, M. J.; Alarie, J. P.; Vo-Dinh, T.; Santella, R. M. Anal. Chem. 1988, 60, 1901. (240) Alarie, J. P.; Bowyer, J. R.; Sepaniak, M. J.; Hoyt, A. M.; Vo-Dinh, T. Anal. Chim. Acta 1990, 236, 237. (241) Bright, F. V.; Betts, T. A.; Litwiler, K. S. Anal. Chem. 1990, 62, 1065. (242) Zhou, Y.; Magill, J. V.; De La Rue, R. M.; Laybourn, P. J. R. Sens. Actuators, B 1993, 11, 245. (243) McCormack, T.; O’Keeffe, G.; MacCraith, B. D.; O’Kennedy, R. Sens. Actuators, B 1997, 41, 89. (244) Bowyer, J. R.; Alarie, J. P.; Sepaniak, M. J.; Vo-Dinh, T.; Thompson, R. Q. Analyst 1991, 116, 117. (245) Zhou, Y.; Laybourn, P. J. R.; Magill, J. V.; De La Rue, R. M. Biosens. Bioelectron. 1991, 6, 595. (246) Sapsford, K. E.; Charles, P. T.; Patterson, C. H.; Ligler, F. S. Anal. Chem. 2002, 74, 1061. (247) Daniels, P. B.; Fletcher, J. E.; O’Neil, P. M.; Stafford, C. G.; Bacarese-Hamilton, T.; Robinson, G. A. Sens. Actuators, B 1995, 26-27, 447. (248) Lepesheva, G. I.; Azeva, T. N.; Knyukshto, V. N.; Chashchin, V. L.; Usanov, S. A. Sens. Actuators, B 2000, 68, 27. (249) Engstroem, H. A.; Andersson, P. O.; Ohlson, S. Anal. Biochem. 2006, 357, 159. (250) Reck, B.; Himmelspach, K.; Opitz, N.; Lu¨bbers, D. W. Analyst 1988, 113, 1423. (251) Barnard, S. M.; Walt, D. R. Science 1991, 251, 927. (252) Anis, N. A.; Eldefrawi, M. E. J. Agric. Food Chem. 1993, 41, 843. (253) Wong, R. B.; Anis, N. A.; Eldefrawi, M. E. Anal. Chim. Acta 1993, 279, 141. (254) Oroszlan, P.; Duveneck, G. L.; Ehrat, M.; Widmer, H. M. Sens. Actuators, B 1993, 11, 301. (255) Owaku, K.; Goto, M.; Ikariyama, Y.; Aizawa, M. Anal. Chem. 1995, 67, 1613. (256) Devine, P. J.; Anis, N. A.; Wright, J.; Kim, S.; Eldefrawi, A. T.; Eldefrawi, M. E. Anal. Biochem. 1995, 227, 216. (257) Shriver-Lake, L. C.; Breslin, K. A.; Charles, P. T.; Conrad, D. W.; Golden, J. P.; Ligler, F. S. Anal. Chem. 1995, 67, 2431. (258) Brummel, K. E.; Wright, J.; Eldefrawi, M. E. J. Agric. Food Chem. 1997, 45, 3292. (259) Bakaltcheva, I. B.; Ligler, F. S.; Patterson, C. H.; Shriver-Lake, L. C. Anal. Chim. Acta 1999, 399, 13. (260) Zhao, C. Q.; Anis, N. A.; Rogers, K. R.; Kline, R. H.; Wright, J.; Eldefrawi, A. T.; Eldefrawi, M. E. J. Agric. Food Chem. 1995, 43, 2308. (261) Schult, K.; Katerkamp, A.; Trau, D.; Grawe, F.; Cammann, K.; Meusel, M. Anal. Chem. 1999, 71, 5430. (262) Tromberg, B. J.; Sepaniak, M. J.; Vo-Dinh, T.; Griffin, G. D. Anal. Chem. 1987, 59, 1226. (263) Sloper, A. N.; Deacon, J. K.; Flanagan, M. T. Sens. Actuators, B. 1990, 1, 589. (264) Shriver-Lake, L. C.; Ogert, R. A.; Ligler, F. S. Sens. Actuators, B 1993, 11, 239. (265) Ogert, R. A.; Brown, J. E.; Singh, B. R.; Shriver-Lake, L. C.; Ligler, F. S. Anal. Biochem. 1992, 205, 306. (266) Vidal, M. I.; Oliva, A. G. Sens. Actuators, B 1997, 38-39, 448. (267) Daneshvar, M. I.; Casay, G. A.; Patonay, G.; Lipowska, M.; Strekowski, L.; Ewans, L., III; Tarazi, L.; George, A. J. Fluoresc. 1996, 6, 69. (268) Zhou, C.; Pivarnik, P.; Auger, S.; Rand, A.; Letcher, S. Sens. Actuators, B 1997, 42, 169. (269) Narang, U.; Anderson, G. P.; Ligler, F. S.; Burans, J. Biosens. Bioelectron. 1997, 12, 937. (270) King, K. D.; Anderson, G. P.; Bullock, K. E.; Regina, M. J.; Saaski, E. W.; Ligler, F. S. Biosens. Bioelectron. 1999, 14, 163. (271) Ligler, F. S.; Breimer, M.; Golden, J. P.; Nivens, D. A.; Dodson, J. P.; Green, T. M.; Haders, D. P.; Sadik, O. A. Anal. Chem. 2002, 74, 713. (272) Hong, B.; Kang, K. A. Biosens. Bioelectron. 2006, 21, 1333. (273) Sheikh, S. H.; Mulchandani, A. Biosens. Bioelectron. 2001, 16, 647. (274) Bier, F. F.; Stocklein, W.; Bocher, M.; Bilitewski, U.; Schmid, R. D. Sens. Actuators, B 1992, 7, 509. (275) Bier, F. F.; Jockers, R.; Schmid, R. D. Analyst 1994, 119, 437. (276) Brecht, A.; Klotz, A.; Barzen, C.; Gauglitz, G.; Harris, R. D.; Quiglei, G. R.; Wilkinson, J. S.; Sztajnbok, P.; Abuknesha, R.; Gascon, J. Anal. Chim. Acta 1998, 362, 69. (277) Rodriguez-Mozaz, S.; Reder, S.; Lopez de Alda, M.; Gauglitz, G.; Barcelo, D. Biosens. Bioelectron. 2004, 19, 633. (278) Cummins, C. M.; Koivunen, M. E.; Stephanian, A.; Gee, S. J.; Hammock, B. D.; Kennedy, I. M. Biosens. Bioelectron. 2006, 21, 1077. (279) Mosiello, L.; Nencini, L.; Serge, L.; Spano, M. Sens. Actuators, B 1997, 38-39, 353.

460 Chemical Reviews, 2008, Vol. 108, No. 2 (280) Barzen, C.; Brecht, A.; Gauglitz, G. Biosens. Bioelectron. 2002, 17, 289. (281) Mallyt, E.; Barzen, C.; Abuknesha, R.; Gauglitz, G.; Barcelo, D. Anal. Chim. Acta 2001, 427, 165. (282) Tschmelak, J.; Kumpf, M.; Ka¨ppel, N.; Proll, G.; Gauglitz, G. Talanta 2006, 69, 343. (283) Homola, J., Ed. Surface Plasmon Resonance Biosensors; SpringerVerlag: Berlin, 2006. (284) Wolfbeis, O. S. Trends Anal. Chem. 1996, 15, 225. (285) Weigl, B. H.; Wolfbeis, O. S. Anal. Chem. 1994, 66, 332. (286) Ruckstuhl, T.; Walser, A.; Verdes, D.; Seeger, S. Biosens. Bioelectron. 2005, 20, 1872. (287) Kuerner, J. M.; Wolfbeis, O. S.; Klimant, I. Anal. Chem. 2002, 74, 2151. (288) Luo, S.; Walt, D. R. Anal. Chem. 1989, 61, 1069. (289) Vermette, P.; Gengenbach, T.; Divisekera, U.; Kambouris, P. A.; Griesser, H. J.; Meagher, L. J. Colloid Interface Sci. 2003, 259, 13. (290) Sato, K.; Sawayanagi, M.; Hosokawa, K.; Maeda, M. Anal. Sci. 2004, 20, 893. (291) Johnson-White, B.; Lin, B.; Ligler, F. S. Anal. Chem. 2007, 79, 140. (292) Schriver-Lake, L. C.; Donner, B.; Edelstein, R.; Breslin, K.; Bhatia, S. K.; Ligler, F. S. Biosens. Bioelectron. 1997, 12, 1101. (293) Preininger, C.; Mencaglia, A.; Baldini, F. Anal. Chim. Acta 2000, 403, 67. (294) Liu, X.; Wang, H.; Herron, J. N.; Prestwich, G. D. Bioconjugate Chem. 2000, 11, 755. (295) Akkoyun, A.; Bilitewski, U. Biosens. Bioelectron. 2002, 17, 655. (296) Tedeschi, L.; Domenici, C.; Ahluwalia, A.; Baldini, F.; Mencaglia, A. Biosens. Bioelectron. 2003, 19, 85. (297) Anderson, G. P.; Jacoby, M. A.; Ligler, F. S.; King, K. D. Biosens. Bioelectron. 1997, 12, 329. (298) Kim, J.; Singh, N.; Lyon, L. A. Angew. Chem., Int. Ed. 2006, 45, 1446. (299) Konry, T.; Novoa, A.; Cosnier, S.; Marks, R. S. Anal. Chem. 2003, 75, 2633. (300) Leshem, B.; Sarfati, G.; Novoa, A.; Breslav, I.; Marks, R. S. Luminescence 2004, 19, 69. (301) Herrmann, S.; Leshem, B.; Landes, S.; Rager-Zisman, B.; Marks, R. S. Talanta 2005, 66, 6. (302) Kulagina, N. V.; Lassman, M. E.; Ligler, F. S.; Taitt, C. R. Anal. Chem. 2005, 77, 6504. (303) Kulagina, N. V.; Schaffer, K. M.; Anderson, G. P.; Ligler, F. S.; Taitt, C. R. Anal. Chim. Acta 2006, 575, 9. (304) Martinez, J. S.; Grace, W. K.; Grace, K. M.; Hartman, N.; Swanson, B. I. J. Mater. Chem. 2005, 15, 4639. (305) Anderson, F. P.; King, K. D.; Gaffney, K. L.; Johnson, L. H. Biosens. Bioelectron. 2000, 14, 771. (306) Petrou, P. S.; Kakabakos, S. E.; Christofidis, I.; Argitis, P.; Misiakos, K. Biosens. Bioelectron. 2002, 17, 261. (307) Feldstein, M. J.; Golden, J. P.; Rowe, C. A.; MacCraith, B. D.; Ligler, F. S. J. Biomed. MicrodeV. 1999, 1, 139. (308) Moreno-Bondi, M. C.; Taitt, C. R.; Shriver-Lake, L. C.; Ligler, F. S Biosens. Bioelectron. 2006, 21 1880. (309) Ngundi, M. M.; Qadri, S. A.; Wallace, E. V.; Moore, M. H.; Lassman, M. E.; Shriver-Lake, L. C.; Ligler, F. S.; Taitt, C. R. EnViron. Sci. Technol. 2006, 40, 2352. (310) Sapsford, K. E.; Ngundi, M. M.; Moore, M. H.; Lassman, Mi. E.; Shriver-Lake, L. C.; Taitt, C. R.; Ligler, F. S. Sens. Actuators, B 2006, 113, 599. (311) Sapsford, K. E.; Taitt, C. R.; Loo, N.; Ligler, F. S. Appl. EnViron. Microbiol. 2005, 71, 5590. (312) Ngundi, M. M.; Shriver-Lake, L. C.; Moore, M. H.; Lassman, M. E.; Ligler, F. S.; Taitt, C. R. Anal. Chem. 2005, 77, 148. (313) Taitt, C. R.; Shubin, Y. S.; Angel, R.; Ligler, F. S. Appl. EnViron. Microbiol. 2004, 70, 152. (314) Shriver-Lake, L. C.; Turner, S.; Taitt, C. R. Anal. Chim. Acta 2007, 584, 66. (315) Sapsford, K. E.; Rasooly, A.; Taitt, C. R.; Ligler, F. S. Anal. Chem. 2004, 76, 433. (316) Taitt, C. R.; Anderson, G. P.; Lingerfelt, B. M.; Feldstein, M. J.; Ligler, F. S. Anal. Chem. 2002, 74, 6114. (317) Tschmelak, J.; Proll, G.; Gauglitz, G. Biosens. Bioelectron. 2004, 20, 743. (318) Tschmelak, J.; Proll, G.; Gauglitz, G. Anal. Chim. Acta 2004, 519, 143. (319) Tschmelak, J.; Kaeppel, N.; Gauglitz, G. Anal. Bioanal. Chem. 2005, 382, 1895. (320) Tschmelak, J.; Proll, G.; Gauglitz, G. Anal. Bioanal. Chem. 2004, 379, 1004. (321) Tschmelak, J.; Proll, G.; Gauglitz, G. Talanta 2005, 65, 313. (322) Rissin, D. M.; Walt, D. R. Anal. Chim. Acta 2006, 564, 34. (323) Schultz, J. S.; Mansouri, S.; Goldstein, I. J. Diabetes Care 1982, 5, 245.

Borisov and Wolfbeis (324) Mansouri, S.; Schultz, J. S. Bio/Technology 1984, 2, 885. (325) Srinivasan, K. R.; Mansouri, S.; Schultz, J. S. Biotechnol. Bioeng. 1986, 28, 233. (326) Meadows, D.; Schultz, J. S. Anal. Chim. Acta 1993, 280, 21. (327) Russell, R. J.; Pishko, M. V.; Gefrides, C. C.; McShane, M. J.; Cote, G. L. Anal. Chem. 1999, 71, 3126. (328) Ballerstadt, R.; Schultz, J. S. Anal. Chem. 2000, 72, 4185. (329) Ballerstadt, R.; Polak, A.; Beuhler, A.; Frye, J. Biosens. Bioelectron. 2004, 19, 905. (330) Ge, X.; Tolosa, L.; Simpson, J.; Rao, G. Biotechnol. Bioeng. 2003, 846, 723. (331) Marvin, J. S.; Hellinga, H. W. J. Am. Chem. Soc. 1998, 120, 7. (332) Gilardl, G.; Zhou, L. Q.; Hibbert, L.; Cass, A. E. G. Anal. Chem. 1994, 66, 3840. (333) Ge, X.; Tolosa, L.; Rao, G. Anal. Chem. 2004, 76, 1403. (334) Ogasawa, F. K.; Wang, Y.; Bobbitt, D. R. Anal. Chem. 1992, 64, 1637. (335) Sanz, V.; De Marcos, S.; Galban, J. Analyst 2007, 132, 59. (336) Thompson, R. B.; Jones, E. R. Anal. Chem. 1993, 65, 730. (337) Lundgren, J. S.; Salins, L. L. E.; Kaneva, I.; Daunert, S. Anal. Chem. 1999, 71, 589. (338) Okoh, M. P.; Hunter, J. L.; Corrie, J. E. T.; Webb, M. R. Biochemistry 2006, 45, 14764. (339) Shrestha, S.; Salins, L. L. E.; Ensor, C. M.; Daunert, S. Biotechnol. Bioeng. 2002, 78, 517. (340) Sumner, J. P.; Westerberg, N. M; Stoddard, A. K.; Hurst, T. K.; Cramer, M.; Thompson, R. B.; Fierke, C. A.; Kopelman, R. Biosens. Bioelectron. 2006, 21, 1302. (341) Sumner, J. P.; Westerberg, N. M; Stoddard, A. K.; Fierke, C. A.; Kopelman, R. Sens. Actuators, B 2006, 113, 760. (342) Blyth, D. J.; Aylott, J. W.; Richardson, D. J.; Russell, D. A. Analyst 1995, 120, 2725. (343) Aylott, J. W.; Richardson, D. J.; Russell, D. A. Chem. Mater. 1997, 9, 2261. (344) Barker, S. L. R.; Kopelman, R.; Meyer, T. E.; Cusanovich, M. A. Anal. Chem. 1998, 70, 971. (345) Barker, S. L. R.; Clark, H. A.; Swallen, S. F.; Kopelman, R. Anal. Chem. 1999, 71, 1767. (346) Blyth, D. J.; Aylott, J. W.; Moir, J. W. B.; Richardson, D. J.; Russell, D. A. Analyst 1999, 124, 129. (347) Barker, S. L. R.; Zhao, Y.; Marletta, M. A.; Kopelman, R. Anal. Chem. 1999, 71, 2071. (348) Zhujun, Z.; Seitz, W. R. Anal. Chem. 1986, 58, 220. (349) Rogers, K. R.; Valdes, J. J.; Eldefrawi, M. E. Anal. Biochem. 1989, 182, 353. (350) Ngundi, M. M.; Taitt, C. R.; McMurry, S. A.; Kahne, D.; Ligler, F. S. Biosens. Bioelectron. 2006, 21, 1195. (351) Graham, C. R.; Leslie, D.; Squirrell, D. J. Biosens. Bioelectron. 1992, 7, 487. (352) Piunno, P. A.; Krull, U. J.; Hudson, R. H.; Damha, M. J.; Cohen, H. Anal. Chim. Acta 1994, 288, 205. (353) Piunno, P. A.; Krull, U. J.; Hudson, R. H.; Damha, M. J.; Cohen, H. Anal. Chem. 1995, 67, 2635. (354) Watts, H. J.; Yeung, D.; Parkes, H. Anal. Chem. 1995, 67, 4283. (355) Abel, A. P.; Weller, M. G.; Duveneck, G. L.; Ehrat, M.; Widmer, H. M. Anal. Chem. 1996, 68, 2905. (356) Pilevar, S.; Davis, C. C.; Portugal, F. Anal. Chem. 1998, 70, 2031. (357) Kleinjung, F.; Klussmann, S.; Erdmann, V. A.; Scheller, F. W.; Furste, J. P.; Bier, F. F. Anal. Chem. 1998, 70, 328. (358) Bagby, D. H.; Piunno, P. A.; Wust, C. C.; Krull, U. J. Anal. Chim. Acta 2000, 411, 19. (359) Dhadwal, H. S.; Kemp, P.; Aller, J.; Dantzler, M. M. Anal. Chim. Acta 2004, 501, 205. (360) Ferguson, J. A.; Boles, T. C.; Adams, C. P.; Walt, D. R. Nat. Biotechnol. 1996, 14, 1681. (361) Epstein, J. R.; Lee, M.; Walt, D. R. Anal. Chem. 2002, 74, 1836. (362) Michael, K. L.; Taylor, L. C.; Schultz, S. L.; Walt, D. R. Anal. Chem. 1998, 70, 1242. (363) Ferguson, J. A.; Steemers, F. J.; Walt, D. R. Anal. Chem. 2000, 72, 5618. (364) Song, L.; Ahn, S.; Walt, D. Anal. Chem. 2006, 78, 1023. (365) Bowden, M.; Song, L.; Walt, D. R. Anal. Chem. 2005, 77, 5583. (366) Ahn, S.; Kulis, D. M.; Erdner, D. L.; Anderson, D. M.; Walt, D. R. Appl. EnViron. Microbiol. 2006, 72, 5742. (367) Ahn, S.; Walt, D. R. Anal. Chem. 2005, 77, 5041. (368) Swanson, P.; Gelbart, R.; Atlas, E.; Yang, L.; Grogan, T.; Butler, W. F.; Ackley, D. E.; Sheldon, E. Sens. Actuators, B 2000, 64, 22. (369) Yang, J. M.; Bell, J.; Huang, Y.; Tirado, M.; Thomas, D.; Forster, A. H.; Haigis, R. W.; Swanson, P. D.; Wallace, R. B.; Martinsons, B.; Krihak, M. Biosens. Bioelectron. 2002, 17, 605. (370) Rissin, D. M.; Walt, D. R. J. Am. Chem. Soc. 2006, 128, 6286. (371) Li, J.; Tan, W.; Wang, K.; Xiao, D.; Yang, X.; He, X.; Tang, Z. Anal. Sci. 2001, 17, 1149.

Optical Biosensors (372) Du, H.; Disney, M. D.; Miller, B. L.; Krauss, T. D. J. Am. Chem. Soc. 2003, 125, 4012. (373) Du, H.; Strohsahl, C. M.; Camera, J.; Miller, B. L.; Krauss, T. D. J. Am. Chem. Soc. 2005, 127, 7932. (374) Hartley, H. A.; Baeumner, A. J. Anal. Bioanal. Chem. 2003, 376, 319. (375) Baeumner, A. J.; Jones, C.; Wong, C. Y.; Price, A. Anal. Bioanal. Chem. 2004, 378, 1587. (376) Baemner, A. J.; Schleisinger, N. A.; Slutzki, N. S.; Romano, J.; Lee, E. M.; Montagna, R. A. Anal. Chem. 2002, 74, 1442. (377) Lee, M.; Walt, D. R. Anal. Biochem. 2000, 282, 142. (378) Rupcich, N.; Nutiu, R.; Li, Y.; Brennan, J. D. Anal. Chem. 2005, 77, 4300. (379) Preininger, C.; Klimant, I.; Wolfbeis, O. S. Anal. Chem. 1994, 66, 1841. (380) Kwok, N.-Y.; Dong, S.; Lo, W.; Wong, K.-Y. Sens. Actuators, B 2005, 110, 289. (381) Lin, L.; Xiao, L.-L.; Huang, S.; Zhao, L.; Cui, J.-S.; Wang, X.-H.; Chen, X. Biosens. Bioelectron. 2006, 21, 1703. (382) Kumar, J.; Kumar Jha, S.; D’Souza, S. F. D. Biosens. Bioelectron. 2006, 21, 2100. (383) Mulchandani, A.; Kaneva, I.; Chen, W. Anal. Chem. 1998, 70, 5042. (384) Bains, W. Biosens. Bioelectron. 1994, 9, 111. (385) Frense, D.; Mu¨ller, A.; Beckmann, D. Sens. Actuators, B 1998, 51, 256. (386) Naessens, M.; Leclerc, J. C.; Tran-Minh, C. Ecotoxicol. EnViron. Saf. 2000, 46, 181. (387) Giardi, M. T.; Guzzella, L.; Euzet, P.; Rouillon, R.; Esposito, D. EnViron. Sci. Technol. 2005, 39, 5378. (388) Virta, M.; Lampinen, J.; Karp, M. Anal. Chem. 1995, 67, 667. (389) Biran, I.; Rissin, D. M.; Ron, E. Z.; Walt, D. R. Anal. Biochem. 2003, 315, 106. (390) Leth, S.; Maltoni, S.; Simkus, R.; Mattiasson, B.; Corbisier, P.; Klimant, I.; Wolfbeis, O. S.; Csoregi, E. Electroanalysis 2002, 14, 35. (391) Heitzer, A.; Malachowsky, K.; Thonnard, J. E.; Beinkowski, P. R.; White, D. C.; Sayer, G. S. Appl. EnViron. Microbiol. 1994, 60, 1487. (392) Ikariyama, Y.; Nishiguchi, S.; Koyama, T.; Kobatake, E.; Aizawa, M.; Tsuda, M.; Nakazawa, T. Anal. Chem. 1997, 69, 2600.

Chemical Reviews, 2008, Vol. 108, No. 2 461 (393) Gil, G. C.; Mitchell, R. J.; Chang, S. T.; Gu, M. B. Biosens. Bioelectron. 2000, 15, 23. (394) Gil, G. C.; Kim, Y. L.; Gu, M. B. Biosens. Bioelectron. 2002, 17, 427. (395) Choi, S. H.; Gu, M. B. Biosens. Bioelectron. 2002, 17, 433. (396) Shetty, R. S.; Ramanathan, S.; Badr, I. H.; Wolford, J. L.; Daunert, S. Anal. Chem. 1999, 71, 763. (397) Michelini, E.; Leskinen, P.; Virta, M.; Karp, M.; Roda, A. Biosens. Bioelectron. 2005, 20, 2261. (398) Fine, T.; Leskinen, P.; Isobe, T.; Shiraishi, H.; Morita, M.; Marks, R. S.; Virta, M. Biosens. Bioelectron. 2006, 21, 2263. (399) Ignatov, S. G.; Ferguson, J. A.; Walt, D. R. Biosens. Bioelectron. 2001, 16, 109. (400) Lundstrom, I.; Gustafsson, A.; Odman, S.; Karlsson, J. O.; Andersson, R. G.; Grundstrom, N.; Sundgren, H.; Elwing, H. Sens. Actuators, B 1990, 1, 533. (401) Rantanen, T.; Paekkilae, H.; Jaemsen, L.; Kuningas, K.; Ukonaho, T.; Loevgren, T.; Soukka, T. Anal. Chem. 2007, 79, 6312. (402) Wang, L.; Yan, R.; Huo, Z.; Wang, L.; Zeng, J.; Bao, J.; Wang, X.; Peng, Q.; Li, Y. Angew. Chem., Int. Ed. 2005, 44, 6054. (403) Wang, L.; Li, Y. Chem. Commun. 2006, 2557. (404) Abrams, W. R.; Barber, C. A.; McCann, K.; Tong, G.; Chen, Z.; Mauk, M. G.; Wang, J.; Volkov, A.; Bourdelle, P.; Corstjens, P. L. A. M.; Zuiderwijk, M.; Kardos, K.; Li, S.; Tanke, H. J.; Niedbala, R. S.; Malamud, D.; Bau, H. Ann. N. Y. Acad. Sci. 2007, 1098, 375. (405) Wolfbeis, O. S.; Reisfeld, R.; Oehme, I. In Structure and Bonding; Springer: Berlin, 1996; Vol. 85, p 51. (406) MacCraith, B. D.; McDonagh, C. J. Fluoresc. 2002, 12, 333. (407) See, for example, the following: www.probes.com; www.sigmaaldrich.com; www.luxcell.com; www.attotec.de; www.chromeon.com. (408) Jung, S.; Angerer, B.; Loescher, F.; Niehren, S.; Winkle, J.; Seeger, S. ChemBioChem 2006, 7, 900. (409) Thanh, N. T. K.; Vernhet, A.; Rosenzweig, Z. In Springer Series on Chemical Sensors and Biosensors; Orellana, G., Moreno-Bondi, M., Eds.; Springer Publishing: New York, 2005; Vol. 3, p 261. (410) Arain, S.; John, G. T.; Kranse, C.; Gerlach, J.; Wolfbeis, O. S.; Klimant, I. Sens. Actuators, B 2006, 113, 639.

CR068105T

Chem. Rev. 2008, 108, 462−493

462

Surface Plasmon Resonance Sensors for Detection of Chemical and Biological Species Jirˇ´ı Homola† Institute of Photonics and Electronics ASCR, Chaberska´ 57, 182 51 Prague 8, Czech Republic Received May 31, 2007

Contents 1. Introduction 2. Surface Plasmons on Planar Structures 2.1. Surface Plasmons on Metal−Dielectric Interface 2.2. Long-Range and Short-Range Surface Plasmons 2.3. Optical Excitation of Surface Plasmons 3. Fundamentals of SPR Sensors 3.1. SPR Sensors 3.2. SPR Affinity Biosensors 3.3. Performance Considerations 3.3.1. Sensitivity 3.3.2. Resolution 3.3.3. Limit of Detection and Minimum Resolvable Surface Coverage 4. Advances in SPR Biosensor Technology 4.1. Optical Platforms Used in SPR Sensors 4.1.1. SPR Sensors Based on Prism Couplers 4.1.2. SPR Sensors Based on Grating Couplers 4.1.3. SPR Sensor Based on Waveguide Couplers 4.2. Biorecognition Elements and Their Immobilization 4.2.1. Biorecognition Elements 4.2.2. Immobilization of Biorecognition Elements 4.2.3. Nonfouling Surfaces 4.3. Summary 5. Applications of SPR Sensors for Detection of Chemical and Biological Species 5.1. Detection Formats 5.2. Food Quality and Safety Analysis 5.2.1. Pathogens 5.2.2. Toxins 5.2.3. Veterinary Drugs 5.2.4. Vitamins 5.2.5. Hormones 5.2.6. Diagnostic Antibodies 5.2.7. Allergens 5.2.8. Proteins 5.2.9. Chemical Contaminants 5.3. Medical Diagnostics 5.3.1. Cancer Markers 5.3.2. Antibodies against Viral Pathogens

462 463 463 464 464 465 465 465 465 465 466 467 467 467 467 470 471 472 472 472 473 475 475 475 477 477 479 479 480 480 480 481 481 481 481 481 482

† Telephone +420 266 773 448; fax +420 284 681 534; e-mail [email protected].

5.3.3. Drugs and Drug-Induced Antibodies 5.3.4. Hormones 5.3.5. Allergy Markers 5.3.6. Heart Attack Markers 5.3.7. Other Molecular Biomarkers 5.4. Environmental Monitoring 5.4.1. Pesticides 5.4.2. 2,4,6-Trinitrotoluene (TNT) 5.4.3. Aromatic Hydrocarbons 5.4.4. Heavy Metals 5.4.5. Phenols 5.4.6. Polychlorinated Biphenyls 5.4.7. Dioxins 5.5. Summary 6. Conclusions 7. Abbreviations 8. Acknowledgment 9. References

483 483 483 484 484 484 484 485 485 485 485 487 487 488 489 489 489 489

1. Introduction Optical sensors based on excitation of surface plasmons, commonly referred to as surface plasmon resonance (SPR) sensors, belong to the group of refractometric sensing devices including the resonant mirror sensor,1,2 the grating coupler sensor,3-5 the integrated optical Mach-Zehnder interferometer,6,7 the integrated Young interferometer,8,9 and the white light interferometer,10,11 which measure changes in the refractive index occurring in the field of an electromagnetic wave supported by the optical structure of the sensor. Since the first demonstration of surface plasmon resonance for the study of processes at the surfaces of metals12 and sensing of gases13 in the early 1980s, SPR sensors have made vast advances in terms of both development of the technology and its applications. SPR biosensors have become a central tool for characterizing and quantifying biomolecular interactions. Moreover, development of SPR sensors for detection of chemical and biological species has gained considerable momentum, and the number of publications reporting applications of SPR biosensors for detection of analytes related to medical diagnostics, environmental monitoring, and food safety and security has been rapidly growing. Over the past decade, thousands of research papers on SPR biosensors have been published. SPR biosensors, as one of the main optical biosensor technologies, have been also extensively featured in biosensor books14-17 and reviews.18-26 Review articles27-30 and books31 focused on SPR biosensor technology have also been published.

10.1021/cr068107d CCC: $71.00 © 2008 American Chemical Society Published on Web 01/30/2008

SPR Sensors

Jirˇ´ı Homola received his M.S. degree in physical engineering in 1988 from the Czech Technical University, Prague, Czech Republic, and his Ph.D. degree from the Academy of Sciences of the Czech Republic in 1993. Since then, he has been with the Institute of Photonics and Electronics, Prague, Czech Republic. From 1997 to 2002, he was with the Department of Electrical Engineering, University of Washington, Seattle, since 2001 as a Research Associate Professor. He is currently Head of Photonics Division and Chairman of Department of Optical Sensors at the Institute of Photonics and Electronics, Prague, and Affiliate Associate Professor at the University of Washington, Seattle. Dr. Homola is a Member of the Editorial Board of Sensors and Actuators B (Elsevier) and Associate Editor of the Journal of Sensors (Hindawi). He has served on technical committees of numerous scientific conferences and is a Member of the Permanent Steering Committees of Advanced Study Course on Optical Chemical Sensors and Europt(r)ode Conference Series. He is also a Member of the NATO CBP Advisory Panel. He received the Otto Wichterle Prize, Czech Republic, in 2003, and the Roche Diagnostics Prize for Sensor Technology, Germany, in 2006. Dr. Homola has edited 2 books, 3 book chapters, and 60 research papers in scientific journals and presented over 100 research papers at international conferences. His research interests are in biophotonics, guided-wave photonics, optical sensors, and biosensors.

This paper reviews advances in SPR sensor technology and its applications since the year 2000. The review focuses on SPR sensors employing conventional (unlocalized) surface plasmons propagating along planar structures and their applications for detection of chemical and biological species. Aspects related to applications of SPR method for the study of biomolecules and their interactions are outside the scope of this review.

2. Surface Plasmons on Planar Structures The first observation of surface plasmons was made in 1902 by Wood, who reported anomalies in the spectrum of light diffracted on a metallic diffraction grating.32 Fano has proven that these anomalies are associated with the excitation of electromagnetic surface waves on the surface of the diffraction grating.33 In 1968 Otto demonstrated that the drop in the reflectivity in the attenuated total reflection (ATR) method is due to the excitation of surface plasmons.34 In the same year, Kretschmann and Raether observed excitation of surface plasmons in another configuration of the attenuated total reflection method.35 These pioneering works of Otto, Kretschmann, and Raether established a convenient method for the excitation of surface plasmons and their investigation and ushered surface plasmons into modern optics. Fundamentals of surface plasmons can be found in numerous review articles and books.36-39

2.1. Surface Plasmons on Metal−Dielectric Interface The simplest geometry in which a surface plasmon can exist consists of a semi-infinite metal with a complex

Chemical Reviews, 2008, Vol. 108, No. 2 463

Figure 1. Spatial distribution of the magnetic intensity for a surface plasmon at the interface between gold and a dielectric (nd ) 1.328) in the direction perpendicular to the interface, λ ) 850 nm.

permittivity, m ) ′m + i′′m, and a semi-infinite dielectric with permittivity, d ) ′d + i′′d, where ′j and ′′j are real and imaginary parts of j (j is m or d). Analysis of Maxwell’s equations with appropriate boundary conditions suggests that this structure can support only a single guided mode of electromagnetic fieldsa surface plasmon. Surface plasmon is a transversally magnetic (TM) mode, and therefore its vector of intensity of magnetic field lies in the plane of metal-dielectric interface and is perpendicular to the direction of propagation. If we use such a Cartesian system of coordinates that the metal occupies the region z < 0 and the surface plasmon propagates along the x-axis, the vector of magnetic intensity H B of the surface plasmon can be written as

H B j ) (0,Hy,0)j ) (0,1,0)A exp[-Rjz + i(βx - ωt)]

(1)

where ω is the angular frequency, t is time, β is the propagation constant, Rj ) xβ2-(ω/c)2j, where j is either m for metal or d for dielectric, and i ) x-1. The vector of the electric field is perpendicular to the vector of the magnetic intensity and can be calculated from Maxwell’s equations and eq 1. A typical profile of the magnetic field of a surface plasmon is shown in Figure 1. The intensity of the magnetic field reaches its maximum at the metal-dielectric interface and decays into both the metal and dielectric. The field decay in the direction perpendicular to the metal-dielectric interface is characterized by the penetration depth, which is defined as the distance from the interface at which the amplitude of the field decreases by a factor of e (where e is the base of the natural logarithm). The penetration depth depends on the wavelength and permittivities of the materials involved. Penetration depth into the dielectric for a surface plasmon propagating along the interface of gold and a dielectric with nd ) 1.32 increases with wavelength and ranges from 100 to 600 nm in the wavelength region from 600 to 1000 nm.31 The propagation constant of a surface plasmon at a metaldielectric interface can be expressed as

βSP )

x

ω c

x

dm 2π )  d + m λ

dm  d + m

(2)

where c is the speed of light in a vacuum and λ is the wavelength in a vacuum.36,37 If the structure is lossless (′′m ) ′′d ) 0), eq 2 represents a guided mode only if the

464 Chemical Reviews, 2008, Vol. 108, No. 2

Homola

Figure 2. Spatial distribution of the magnetic intensity of symmetric and antisymmetric surface plasmons propagating along a thin gold film (m ) -30.5 + i 1.6, thickness ) 20 nm) embedded between two identical dielectrics (nd ) 1.328), λ ) 850 nm.

permittivities ′m and ′d are of opposite signs and ′m < -′d. As the permittivity of dielectric materials is usually positive, this requires that the real part of the permittivity of the metal is negative. Metals such as gold, silver, and aluminum exhibit a negative real part of permittivity in the visible and nearinfrared regions of the spectrum. These metals also exhibit a considerable imaginary part of the permittivity, which causes the propagation constant of a surface plasmon to have a nonzero imaginary part. The imaginary part of the propagation constant is associated with the attenuation of the surface plasmon in the direction of propagation.36,37 The propagation constant is related to the effective index nef and attenuation b as

c 0.2 Im{βSP} nef ) Re{βSP}, b ) ω ln 10

(3)

where Re{ } and Im{ } denote the real and imaginary parts of a complex number, respectively; the attenuation b is in dBcm-1 if β is given in m-1.

2.2. Long-Range and Short-Range Surface Plasmons A planar structure consisting of a thick metal film sandwiched between two semi-infinite dielectrics supports two independent surface plasmons at the opposite boundaries of the metal film. If the metal film is thin, coupling between the surface plasmons at opposite boundaries of the metal film can occur, giving rise to mixed modes of electromagnetic fieldssymmetric and antisymmetric surface plasmons. Characteristics of these surface plasmons can be found from Maxwell’s equations and appropriate boundary conditions.40-42 The symmetric surface plasmon exhibits a propagation constant and attenuation, which both increase with increasing metal film thickness.31 The propagation constant and attenuation of the antisymmetric surface plasmon decrease with increasing thickness of the metal film. The symmetric surface plasmon exhibits a lower attenuation than its antisymmetric counterpart, and therefore it is referred to as a long-range surface plasmon, whereas the antisymmetric mode is referred to as a short-range surface plasmon.40,41 Figure 2 shows the distribution of magnetic intensity of the symmetric and antisymmetric surface plasmons propagating along a thin gold film surrounded by two identical dielectrics. The profiles of magnetic intensity of symmetric and antisymmetric plasmons are symmetric or antisymmetric

Figure 3. Coupling of light to a surface plasmon via (A) prism coupler, (B) waveguide coupler, and (C) grating coupler.

with respect to the center of the metal. The field of the symmetric surface plasmon penetrates much more deeply into the dielectric medium than the field of the antisymmetric surface plasmon or the field of a conventional surface plasmon at a single metal-dielectric interface (Figure 1).

2.3. Optical Excitation of Surface Plasmons A light wave can couple to a surface plasmon at a metaldielectric interface if the component of light’s wavevector that is parallel to the interface matches the propagation constant of the surface plasmon. As the propagation constant of a surface plasmon at a metal-dielectric interface is larger than the wavenumber of the light wave in the dielectric, surface plasmons cannot be excited directly by light incident onto a smooth metal surface. The wavevector of light can be increased to match that of the surface plasmon by the attenuated total reflection or diffraction. This enhancement and subsequently the coupling between light and a surface plasmon are performed in a coupling device (coupler). The most common couplers used in SPR sensors include a prism coupler, a waveguide coupler, and a grating coupler (Figure 3). Prism couplers represent the most frequently used method for optical excitation of surface plasmons.35-37 In the Kretschmann configuration of the attenuated total reflection (ATR) method (Figure 3A),35 a light wave passes through a high refractive index prism and is totally reflected at the base of the prism, generating an evanescent wave penetrating a

SPR Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 465

thin metal film. The evanescent wave propagates along the interface with the propagation constant, which can be adjusted to match that of the surface plasmon by controlling the angle of incidence. Thus, the matching condition

2π n sin (θ) ) Re{βSP} λ p

(4)

can be fulfilled, allowing the evanescent wave to be coupled to the surface plasmon. θ denotes the angle of incidence, np denotes the refractive index of the prism (np > nd), and βSP denotes the propagation constant of the surface plasmon. Surface plasmons can be also excited by a light wave guided in an optical waveguide. This approach is illustrated in Figure 3B. Light propagates in a waveguide in the form of guided modes. The electromagnetic field of a guided mode is concentrated in the waveguiding layer, and a portion of the field propagates, as an evanescent wave, in the low-refractive index medium surrounding the waveguiding layer. When light enters the region of the waveguide containing a metal layer, the evanescent wave excites a surface plasmon at the outer boundary of the metal layer. The coupling condition for the guided mode and the surface plasmon is fulfilled when the propagation constants of the two waves are equal

βmode ) Re{βSP}

(5)

3.2. SPR Affinity Biosensors

where βmode denotes the propagation constant of the waveguide mode. Another approach to optical excitation of surface plasmons is based on the diffraction of light in a grating coupler (Figure 3C). In this method, a light wave is incident from the dielectric medium on a metallic grating. The diffracted light can couple to a surface plasmon if the momentum of diffracted light parallel to the grating surface is equal to the propagation constant of the surface plasmon

2π 2π n sin θ + m ) (Re{βSP} λ d Λ

are classified as sensors with angular, wavelength, intensity, or phase modulation.31 In SPR sensors with angular modulation, a monochromatic light wave is used to excite a surface plasmon. The strength of coupling between the incident wave and the surface plasmon is observed at multiple angles of incidence, typically by employing a convergent light beam. The excitation of surface plasmons is observed as a dip in the angular spectrum of reflected light. The angle of incidence yielding the strongest coupling is measured and used as a sensor output.44 In SPR sensors with wavelength modulation, a surface plasmon is excited by a collimated light wave containing multiple wavelengths, typically a beam of polychromatic light. The excitation of surface plasmons is observed as a dip in the wavelength spectrum of reflected light. The wavelength yielding the strongest coupling is measured and used as a sensor output.45 SPR sensors with intensity modulation are based on measuring the strength of the coupling between the light wave and the surface plasmon at a single angle of incidence and wavelength, and the intensity of light wave serves as a sensor output.13 In SPR sensors with phase modulation the shift in phase of the light wave coupled to the surface plasmon is measured at a single angle of incidence and wavelength of the light wave and used as a sensor output.46

(6)

where m is an integer and denotes the diffraction order and λ is the grating period.43 In the process of optical excitation of surface plasmon, a portion of the energy of the light wave is transferred into the energy of a surface plasmon and dissipated in the metal film, which results in a drop of intensity of the light wave. In addition to the change in the intensity, the light wave exciting a surface plasmon undergoes a change in phase.31

3. Fundamentals of SPR Sensors 3.1. SPR Sensors In principle, SPR sensors are thin-film refractometers that measure changes in the refractive index occurring at the surface of a metal film supporting a surface plasmon. A surface plasmon excited by a light wave propagates along the metal film, and its evanescent field probes the medium (sample) in contact with the metal film. A change in the refractive index of the dielectric gives rise to a change in the propagation constant of the surface plasmon, which through the coupling condition (eqs 4-6) alters the characteristics of the light wave coupled to the surface plasmon (e.g., coupling angle, coupling wavelength, intensity, phase). On the basis of which characteristic of the light wave modulated by a surface plasmon is measured, SPR sensors

SPR affinity biosensors are sensing devices which consist of a biorecognition element that recognizes and is able to interact with a selected analyte and an SPR transducer, which translates the binding event into an output signal. The biorecognition elements are immobilized in the proximity of the surface of a metal film supporting a surface plasmon. Analyte molecules in a liquid sample in contact with the SPR sensor bind to the biorecognition elements, producing an increase in the refractive index at the sensor surface, which is optically measured (section 3.1). The change in the refractive index produced by the capture of biomolecules depends on the concentration of analyte molecules at the sensor surface and the properties of the molecules. If the binding occurs within a thin layer at the sensor surface of thickness h, the sensor response is proportional to the binding-induced refractive index change, which can be expressed as

∆n )

(dndc)Γh

(7)

where (dn/dc) denotes the refractive index increment of the analyte molecules (typically 0.1-0.3 mL/g47,48) and Γ denotes the surface concentration in mass/area.49

3.3. Performance Considerations The main performance characteristics of SPR (bio)sensors include sensitivity, linearity, resolution, accuracy, reproducibility, dynamic range, and limit of detection.31

3.3.1. Sensitivity Sensor sensitivity is the ratio of the change in sensor output to the change in the quantity to be measured (e.g., concentration of analyte). The sensitivity of an SPR affinity biosensor depends on two factorsssensitivity of the sensor output (e.g., resonant angle or wavelength) to the refractive index and efficiency of the conversion of the binding to a change in

466 Chemical Reviews, 2008, Vol. 108, No. 2

Homola

the refractive index (section 3.2).31 The sensitivity of an SPR sensor to a refractive index SRI can be expressed as a product of two terms

SRI )

δY δnef δnef δnd

(8)

where Y denotes sensor output. The first term describes the sensitivity of sensor output to the effective index of a surface plasmon and depends on the method of excitation of surface plasmons and the used modulation approach. The second term describes the sensitivity of the effective index of a surface plasmon to refractive index and is independent of the modulation method and the method of excitation.31 In general, the sensitivity of the effective index of a surface plasmon to refractive index depends on the distribution of the refractive index change. Homola and Piliarik used the perturbation theory50 to calculate the sensitivity of the effective index to the refractive index for two limiting cases: (i) the change in the refractive index that occurs within the whole sample and (ii) the change in the refractive index that occurs only within a very short distance from the sensor surface.31 They showed that the sensitivity of the effective index of a surface plasmon to bulk refractive index change can be expressed as

( ) δnef δnd

B

)

nef3 nd3

>1

(9)

and therefore is always larger than the sensitivity of a free space plane wave in the infinite dielectric medium, which is equal to 1.31 The sensitivity of the effective index of a surface plasmon to surface refractive index change occurring within a layer with a thickness h can be, assuming h , Lpd ) 1/Re{Rd} and |′m| . ′′m, written as

( ) ( ) δnef δnd

δnef h )2 δnd BLpd S

SPR sensors of all the modulation approaches need to measure the intensity of the light wave coupled to a surface plasmon to determine the sensor output. Therefore, their resolution is limited by the noise in the intensity of the detected light. Dominant sources of noise are the fluctuations in the light intensity emitted by the light source, shot noise associated with photon statistics, and noise in conversion of the light intensity into electric signal by the detector.31 To reduce the noise, light intensities are averaged. The averaging involves either averaging of time series of intensity from the same detector (time averaging) or averaging of intensities from multiple detectors (e.g., of a two-dimensional array) measured at a single time (spatial averaging). The time averaging reduces the noise in the intensity of light by a factor of xM, where M is the number of averaged intensities. The spatial averaging used in spectroscopic SPR sensors (averaged spectra are measured in several rows of a 2D detector51,52) or intensity-modulated sensors (averaged area of a 2D detector forms a signal of one measuring channel53-55) is less efficient, as the light fluctuations affect all of the measured intensities in the same way and therefore cannot be eliminated by the spatial averaging. The noise in the light intensity is translated to sensor output noise by a data processing algorithm used to generate the sensor output. Although various methods for processing data from spectroscopic SPR sensors have been developed (centroid method,56,57 polynomial fitting,58,59 and optimal linear data analysis60), the noise in angular or wavelength spectra was found to transform to the noise in the sensor output in a similar fashion.61 Piliarik and Homola investigated the propagation of noise through the centroid data processing algorithm and demonstrated that the noise of the centroid method can be expressed as

1 σth w σRI ) K xN d SRI

(11)

(10)

This suggests that the surface refractive index sensitivity is proportional to the bulk refractive index sensitivity and the ratio of the thickness of the layer within which the surface refractive index change occurs and the penetration depth of the surface plasmon, Lpd. As the penetration depth of a surface plasmon on gold increases with increasing wavelength, the surface refractive index sensitivity of the effective index decreases with the wavelength more quickly than the bulk refractive index sensitivity.31

3.3.2. Resolution Resolution is a key performance characteristic of an SPR sensor and ultimately limits another important performance characteristic of an SPR affinity biosensorsthe limit of detection (LOD). The resolution of an SPR sensor is defined as the smallest change in the bulk refractive index that produces a detectable change in the sensor output. The magnitude of sensor output change that can be detected depends on the level of uncertainty of the sensor outputs the output noise. The resolution of an SPR sensor, rRI, is typically expressed in terms of the standard deviation of noise of the sensor output, σso, translated to the refractive index of bulk medium, rRI ) σso/SRI, where SRI is the bulk refractive index sensitivity.

where N is the number of intensities used for the calculation of the centroid, σth is the total intensity noise at the threshold, d is the difference of intensities at the SPR dip minimum and at the threshold, w is the width of the dip, SRI is the bulk refractive index sensitivity of the sensor, and K is a factor depending on the relative contributions of the sources of noise.31 As follows from eq 11, the noise in the sensor output is mainly determined by the ratio of the noise in the light intensity at the threshold and the depth of the SPR dip. The ratio w/SRI depends only weakly on the choice of coupler and modulation and therefore has only a minor effect on the sensor resolution.31 Although the analysis was performed for the spectroscopic sensors (N > 1), it can be extended to intensity-modulated SPR sensors (N ) 1). Equation 11 also explains why spectroscopic SPR sensors typically exhibit better resolution than their intensity-based counterpartssthe N values in spectroscopic SPR sensors are typically of the order of 100,62-64 which improves resolution by an order of magnitude. Another important conclusion is that in the intensity-modulated SPR sensor σRI is proportional to σth/I, which for most kinds of noise decreases with increasing intensity of light. Ran and Lipson performed theoretical and experimental comparisons of resolution of intensity and phase modulationbased SPR sensors.65 They demonstrated that under identical noise conditions, the performances of SPR sensors based on

SPR Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 467

intensity and phase modulations are comparable. This is due to the facts that in phase modulation-based sensors not the phase but the intensity of a light beam (produced by interference) is measured and that the configurations providing higher phase sensitivity to the refractive index exhibit highest absorption and consequently worse signal-to-noise ratios. They also demonstrated that the refractive index resolution can be improved for both the modulation approaches if the intensity of light coupled to the surface plasmon and received by the detector is increased,65 which is consistent with the results of Piliarik and Homola.31

3.3.3. Limit of Detection and Minimum Resolvable Surface Coverage In the field of SPR sensors, the term resolution usually refers to a bulk refractive index resolution. On the other hand, the limit of detection (LOD) is usually defined as the concentration of analyte that produces sensor output corresponding to 3 standard deviations of sensor output measured for a blank sample.66 The ultimate LOD can be predicted only when the parameters of the interaction between the analyte and biorecognition element and mass transport to the sensor surface are known. However, the smallest detectable surface concentration (minimum resolvable surface coverage) can be determined independently of these factors. The minimum resolvable change of molecular mass captured by the biorecognition elements σΓ depends on the sensor sensitivity and the noise in the sensor output:

σΓ )

σSO h Sh ∂n ∂c vol

( )

(12)

(∂n/∂c)vol denotes the volume refractive index increment of the molecular concentration, and Sh denotes the refractive index sensitivity of sensor output to a refractive index change within the sensitive layer of a thickness h. For thicknesses much smaller than the penetration depth of the surface plasmon (see eq 10), the following relationship between the bulk refractive index resolution and the resolution of surface coverage can be written:

Lpd σΓ ) σRI ∂n 2 ∂c vol

( )

(13)

For an SPR sensor operating at the wavelength of 760 nm (Lpd ) 320 nm) and a typical analyte with a refractive index increment (∂n/∂c)vol ) 0.18cm3/g (DNA or BSA48), eq 13 suggests that a refractive index resolution of σRI ) 10-6 RIU corresponds to a surface coverage resolution of σΓ ) 0.91 pg/mm2.

4. Advances in SPR Biosensor Technology An SPR affinity biosensor consists of a biorecognition element and an SPR transducer. The core of the transducer is an optical platform in which a surface plasmon is optically excited and interrogated and the binding between a biorecognition element (e.g., antibody) immobilized on the surface of the transducer and target analyte in a liquid sample is measured. An SPR biosensor also incorporates a fluidic system that usually consists of a flow cell or cuvette

confining the sample at the sensing surface and a samplehandling system for sample collection and preparation. In the following sections, recent advances in the two most critical elements of the SPR biosensor technologysoptical platforms and biorecognition elements and their immobilizationsare reviewed.

4.1. Optical Platforms Used in SPR Sensors 4.1.1. SPR Sensors Based on Prism Couplers Most of the SPR sensors developed to date, including the first reported SPR sensor,13 use a prism coupler to couple light to a surface plasmon. Prism coupling is convenient and can be realized with simple and conventional optical elements. Moreover, it can be readily combined with any type of modulation. Sensors Based on Intensity Modulation. Research into SPR sensors with intensity modulation focuses mainly on the two important aspectssimproving performance (sensitivity, resolution) and increasing throughput. To increase the sensitivity of intensity-modulated SPR sensors, Lechuga’s group proposed an approach based on combination of the magneto-optic activity of magnetic materials and a surface plasmon resonance in a special multilayer structure.67 They demonstrated an improvement in sensitivity by a factor of 3 compared to a conventional intensity-modulated SPR sensor and a refractive index resolution of 5 × 10-6 RIU.67 A typical example of a high-throughput SPR sensor is the SPR imaging.68,69 In a typical SPR imaging configuration, a beam of monochromatic light passes through a prism coupler and is made incident on a thin metal film at an angle of incidence close to the coupling angle. The intensity of reflected light depends on the strength of the coupling between the incident light and the surface plasmon and therefore can be correlated with the distribution of the refractive index at the surface of the metal film.68,69 Corn’s group has researched SPR imaging for over a decade. In their earlier works, they employed a HeNe laser as a source of illumination.69 However, a highly coherent light source generated images with parasitic interference patterns that were disturbing SPR measurements. In 1997 they improved their SPR imaging instrument by introducing an incoherent light source and a NIR narrow band-pass filter.70 Using this approach, they detected hybridization of short (18-base) oligonucleotides at concentrations as low as 10 nM71 (this was estimated to correspond to a refractive index resolution in the 10-5 RIU range). The use of a white light source and a bandpass filter was also advocated by Yager’s group.72 They demonstrated that by tilting the interference filter, an operating wavelength of the SPR imaging sensor can be tuned.72 Later they demonstrated that their SPR imaging instrument operating at a wavelength of 853 nm can provide a refractive index resolution of 3 × 10-5 RIU.55 In 2005 Corn’s group reported SPR imaging with a special multilayer structure supporting long-range surface plasmons; however, the use of long-range surface plasmons led only to minor sensitivity improvements of 20% (experiment) and 40% (theory) compared to the conventional SPR imaging.73 A dual-wavelength SPR imaging system was reported by Zybin et al.74 In their SPR sensor, they used two sequentially switched-on laser diodes, and the intensities of the reflected light at the two different wavelengths were measured and

468 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 4. Concept of SPR biosensing.

Figure 5. Typical image obtained with an SPR imaging sensor with a polarization control. Bright rectangles correspond to areas of an SPR chip (300 × 300 µm) covered with a monolayer of albumin molecules formed on the surface of gold by microspotting.

the sensor output was defined as the difference of these two signals. A refractive index resolution of 2 × 10-6 RIU was achieved when the signal was averaged over a large beam diameter (6 mm2). Campbell’s group reported an SPR imaging system with a controllable angle of incidence.75,76 This feature allows SPR images to be acquired sequentially at different angles of incidence and selection of the optimum angle of incidence for the SPR measurements. With a HeNe laser as a source of light, their sensor was able to measure simultaneously in 120 sensing channels with a refractive index resolution of 2 × 10-5 RIU. Recently, they claimed an improvement in sensor resolution down to 5 × 10-6 RIU.77 Recently, Piliarik et al. investigated SPR imaging with an elliptically polarized light78 and concluded that a change in the polarization of light induced by the coupling of light to a surface plasmon can be exploited to significantly improve the sensitivity and operating range of SPR imaging sensors. In addition, this approach, as illustrated in Figure 5, provides high-contrast SPR images (with a low background), which are well suited for automated image analysis. Homola’s group developed an SPR imaging approach based on polarization contrast and excitation of surface plasmons on spatially patterned multilayers.53 In this configuration a prism coupler with an SPR chip containing a spatially patterned multilayer structure was placed between two crossed polarizers. The output polarizer blocked all of the light reflected from the (inactive) areas outside the sensing areas, generating high-contrast images. Two types of SPR multilayers with opposite sensitivities to refractive index were employed, and the output signal was defined as a ratio of the intensities generated from the two neighboring

Homola

multilayers. This sensor was shown to be able to detect refractive index changes down to 2 × 10-6 RIU and to detect short oligonucleotides (23-mers) at concentrations as low as 100 pM.79 Currently, commercial SPR imaging instruments are available from GWC Technologies, Inc.80 (Madison, WI), Lumera81 (Bothell, WA),29 IBIS Technologies (Hengelo, The Netherlands),82 and SPRi-Array from GenOptics (Orsay, France).83 Sensors Based on Spectroscopy of Surface Plasmons. In sensors based on spectroscopy of surface plasmons, the angular or wavelength spectrum of a light wave coupled to a surface plasmon is measured and sensor output is related to a change in the angular or wavelength position of the SPR dip. In the early 1990s, an angular modulation-based SPR sensor consisting of a light-emitting diode (LED), a glass prism, and a detector array with imaging optics was introduced.58,84,85 A divergent beam produced by the LED was collimated and focused by means of a cylindrical lens to produce a wedge-shaped beam of light, which was used to illuminate a thin gold film on the back of a glass prism containing several sensing areas (channels). The imaging optics consisted of one imaging and one cylindrical lens ordered in such a way that the angular spectrum of each sensor channel was projected on separate rows of the array detector.86-88 This design has been adopted by Biacore and resulted in a family of commercial SPR sensors with high performance (resolution down to 1 × 10-7RIU) and multiple sensing channels (up to four). In 2004, Thirstrup et al. integrated several optical elements into a single sensor chip.63 In this approach, the cylindrical focusing optics utilized to create a beam of a desired angular span was replaced by a diffraction grating of a special design incorporated into the sensing element.89,90 A wide parallel light beam was diffracted by the focusing grating and focused into a small spot on the SPR measuring surface. The reflected light followed a similar path, producing a parallel beam with an angular spectrum superimposed across the beam. A twodimensional photodetector was used to measure the angular spectrum of the reflected light for several parallel channels. This design offered a compact SPR platform with a resolution of about 5 × 10-7 RIU.89 An SPR sensor with wavelength modulation and parallel channel architecture was reported by Homola’s group.91 In this sensor, a polychromatic light from a halogen lamp was collimated into a large-diameter parallel beam, which was launched in a prism coupler. The light reflected from different sensing channels was collected by different output collimators coupled and transmitted to different inputs of a spectrograph. The SPR sensor of this design was demonstrated to be able to resolve refractive index changes down to 2 × 10-7 RIU.62 An SPR sensor with wavelength division multiplexing (WDM) of sensing channels was proposed by Homola et al.92 In this approach, signals from multiple surface plasmons excited in different areas of a sensing surface are encoded into different regions of the spectrum of the light wave. Two configurations of WDMSPR sensors have been developed.93,94 In the first configuration, a wide parallel beam of polychromatic light is made incident onto a sensing surface consisting of a thin gold film, a part of which is coated with a thin dielectric film. As the presence of the thin dielectric film shifts the coupling wavelength to a longer wavelength

SPR Sensors

(compared to the bare gold), the reflected light exhibits two dips associated with the excitation of surface plasmons in the area with and without the overlayer.93 The second configuration of WDMSPR sensor employs a special prism coupler in which a polychromatic light is sequentially incident on different areas of the sensing surface at different angles of incidence. Due to the different angles of incidence, the surface plasmons in different regions are excited with different wavelengths of the incident light.94 Therefore, the spectrum of transmitted light contains multiple dips associated with surface plasmons in different areas of the sensing surface. The WDMSPR approach was combined with the parallel architecture, yielding an eight-channel SPR sensor with a resolution around 1 × 10-6 RIU.94 An optical sensor based on spectroscopy of long-range surface plasmons was reported by Nenninger et al.95 In that work, a long-range surface plasmon was excited on a special multilayer structure consisting of a glass substrate, a Teflon AF layer, and a thin gold layer. A resolution as low as 2 × 10-7 RIU was achieved.95 Most recently, Homola’s group demonstrated an improved configuration for excitation of long-range surface plasmons and demonstrated an SPR sensor with a resolution as low as 3 × 10-8 RIU.96 Development of portable/mobile SPR sensor platforms suitable for deployment in the field presents an important direction in SPR sensor research. Several miniaturized SPR optical platforms based on spectroscopy of surface plasmons have been developed. A concept of the miniature SPR sensor based on integration of all electro-optical components in a monolithic platform developed by Texas Instruments in the mid-1990s97 was further advanced by researchers at Texas Instruments and the University of Washington. The Spreeta 2000 SPR sensor (Texas Instruments, USA) consists of a plastic prism molded onto a microelectronic platform containing an infrared LED and a linear diode array detector. The LED emits a diverging beam that passes through a polarizer and strikes the sensor surface at a range of angles. The angle at which light is reflected from this surface toward a detector varies with the location on the surface. The initial version of this platform exhibited a refractive index resolution of 5 × 10-6 RIU.98 Baseline noise and smoothness of response of this sensor were investigated by Chinowsky et al.,99 who showed that the baseline noise established under constant conditions was <2 × 10-7 RIU; however, the sensor response to a gradual change in the refractive index revealed departures from the expected sensor output of about 8 × 10-5 RIU. A portable SPR instrument based on the Spreeta 2000 design was reported by Naimushin et al.100 Their instrument incorporated temperature stabilization and was demonstrated to provide a refractive index resolution of 3 × 10-6 RIU. Recently, an SPR instrument based on two Texas Instruments Spreeta devices was adopted for deployment in a surrogate unmanned aerial vehicle and applied for detection of airborne analytes.101 Another compact, portable SPR sensor platform was developed by Kawazumi et al.52 Their system was also based on angular modulation and Kretschamnn geometry. A line-shape light beam from an LED was focused on the sensing surface using a cylindrical lens. Two-dimensional Fourier transform images of the reflected light were measured with a compact CCD camera. SPR measurements were carried out at a fixed angle, and the resonance angles were obtained by analyzing the twodimensional images. The system provided four independent

Chemical Reviews, 2008, Vol. 108, No. 2 469

sensing channels and a refractive index resolution of 10-4 RIU.52 Nowadays numerous SPR sensors based on spectroscopy of surface plasmons are commercially available. They include Biacore systems from Biacore (now part of GE Healthcare, USA),88 Spreeta sensor from Texas Instruments (Dallas, TX),102 Multiskop system from Optrel (Kleinmachnow, Germany),103 SR 7000 platforms from Reichert Analytical Instruments (Depew, NY),104 Plasmonic from Hofmann Sensorsysteme (Wallenfels, Germany),105 Autolab Esprit and Springle SPR systems from Eco Chemie (Utrecht, The Netherlands),106 SPR-20 from DKK-TOA Corp. (Tokyo, Japan),107 BIOSUPLAR 6 from Analytical µ-Systems (Sinzing, Germany),108 and Sensia β-SPR Research Platform (Madrid, Spain).109 SPR Sensors Based on Phase Modulation. The research group of Nikitin demonstrated two SPR sensor platforms based on interferometry.110,111 The first approach was based on the interference of the TM-polarized signal beam with the TE-polarized reference beam,110 whereas the second method was based on a Mach-Zehnder interferometer combining TM-polarized signal and reference beams.111 This configuration was demonstrated in two modes: (a) phase contrast (Zernike phase contrast) increasing the sensor sensitivity and (b) “fringe mode”, in which there was a definite angle between the interfering beams and a pattern of interference fringes was superimposed on the image of the surface. Local variations in the phase of the signal beam resulted in bending and moving of the interference fringes. A refractive index resolution was on the order of 10-7 RIU.111 At the same time as Nikitin et al. published their work,111 a similar configuration of SPR sensor based on the MachZehnder interferometer was reported by Notcovich et al.112 They used their system for measurement of the refractive index of gases and demonstrated a refractive index resolution on the order of 10-6 RIU.112 Wu et al. proposed a phase-modulation SPR sensor based on common-path, heterodyne interferometry.113 Two acoustooptic modulators were used to split the incoming laser light from a HeNe laser into two linearly orthogonally polarized beams with a frequency difference of 60 kHz. These two light beams were merged into one beam by a polarization beam splitter. One portion of the beam was directed to a detector while the other was coupled into an SPR prism coupler. The TE and TM components of light reflected from a thin layer of gold on the base of the prism were recombined using a polarizer, and the output beam was received by a detector. SPR-induced phase shift was determined by an electronic phase meter (lock-in amplifier). The refractive index resolution of this design was estimated to be 2 × 10-7 RIU.113 Alieva and Konopsky developed an SPR sensor based on interference between a surface plasmon supported on a metal film and a bulk wave propagating at grazing angle in the flow cell just above the surface of metal.114 This approach suppresses the sensitivity of the SPR method to variations in the refractive index of a liquid sample, which in SPR biosensors interfere with binding measurements. Naraoka and Kajikawa reported a phase-modulation SPR sensor based on a rotating analyzer method.115 In their approach, a linearly polarized light from a semiconductor laser was coupled to an SPR prism coupler and reflectivity was detected while the rotational angle of the analyzer was scanned. The phase difference between the TE and TM

470 Chemical Reviews, 2008, Vol. 108, No. 2

components of the reflected light were determined from the dependence of the reflectivity on the angle of analyzer. The refractive index resolution of their system was estimated to be below 2 × 10-7 RIU.115 In recent years, SPR sensors with phase modulation have been extensively studied by the researchers at the Chinese University of Hong Kong and City University of Hong Kong. In 2002 Ho et al. reported an SPR sensor based on the Mach-Zehnder interferometer.116 In that work, an optical beam from an unpolarized HeNe laser passed through a prism coupler and the TM-polarized component of the beam excited surface plasmons in two parallel sensing channels, one filled with a sample and the other with a reference solution. TE and TM polarization components of the output beam were split by a polarizing beam splitter. The optical path for TE polarization was modulated by means of a piezoelectric actuator. Finally, TM polarization was converted to TE polarization by a half-waveplate, and the two beams were recombined. The shift between the interference patterns for the measuring and reference sensing channels was measured. Resolution of this sensor was estimated to be about 3 × 10-6 RIU.116 An alternative configuration of an SPR sensor with phase modulation was reported by Wu et al. in 2004.117 In this configuration, one Mach-Zehnder interferometer performed independent interference of TE and TM polarized components of a signal beam emerging from a prism coupler and a reference beam. Subsequently, the output TE and TM beams were separated in a Wollaston prism and directed to two separate detectors. A piezoelectric actuator modulated the optical path in a reference arm of the interferometer, producing a periodic intensity modulation in both TE and TM polarizations. The interference patterns for TE and TM polarizations were processed to reduce the noise and compensate for instabilities in the setup,118 and the sensor output was determined as a mutual shift of the two patterns. Resolution of the sensor was estimated to be 5.5 × 10-8 RIU.117 Another SPR sensor based on measuring the phase difference between TE and TM polarization components of light beam was reported by Ho et al.119 They used a single beam and a photoelastic phase modulator to introduce a carrier frequency so that the phase can be determined by measuring the relative amplitude of the first harmonic signal. Resolution of the sensor was determined to be 1.2 × 10-6 RIU.119 In 2007, a phase modulation-based SPR sensor employing a Michelson interferometer was reported by Yuan et al.120 They used a Michelson interferometer with an SPR prism coupler inserted in one arm of the interferometer. This arrangement allows the TM-polarized component of the light beam to incur a 2-fold phase shift compared with that in the Mach-Zehnder interferometer. Therefore, the sensitivity of the Michelson interferometer-based SPR sensor was twice as high as that of the SPR sensor with the Mach-Zehnder interferometer. This improvement was demonstrated in a single experimental system incorporating both Michelson and Mach-Zehnder interferometers in which the refractive index resolutions were established to be 7.7 × 10-7 and 1.5 × 10-6 RIU, respectively.120

4.1.2. SPR Sensors Based on Grating Couplers Grating couplers have not been used in SPR sensors as widely as the prism couplers. However, their compatibility with mass production (in particular, replication into plastic) makes a grating coupler an attractive approach for fabrication of low-cost SPR sensing structures.

Homola

In 2001 Brockman and Fernandez presented an SPR imaging device based on grating coupling.121 In this approach, a collimated monochromatic light beam (wavelength ) 860 nm) was made incident onto a plastic chip with a gold-coated diffraction grating. An array of 400 sensing channels (spot diameter ) 250 µm) was prepared on the chip by means of spatially resolved functionalization. Upon reflection from the chip, the light was projected onto a twodimensional CCD array.121 This concept was further developed by HTS Biosystems (Hopkinton, MA).122 In 2005, Biacore International AB acquired the FLEXChip technology from HTS Biosystems. Another high-throughput SPR sensor based on grating coupling was reported by Homola’s group.64 This approach was based on angular spectroscopy of surface plasmons on an array of diffraction gratings. A collimated beam of monochromatic light was focused with a cylindrical lens on a row of gold-coated diffraction gratings and reflected under nearly normal incidence. The angular spectra were transformed back to a collimated beam by means of a focusing lens and projected onto a two-dimensional CCD detector. Different rows of gratings were read sequentially by moving the beam splitter and cylindrical lens with respect to the sensor chip. A refractive index resolution of 5 × 10-6 RIU was achieved for simultaneous measurements in over 200 sensing channels.64 Homola’s group reported an SPR sensor based on simultaneous spectroscopy of multiple surface plasmons on a multidiffractive grating.123 A polychromatic light beam was made incident onto a special metallic grating with a grating profile composed of multiple harmonics. The reflected light contained multiple SPR dips, one for each grating period. The use of multiple surface plasmons of different field profiles can provide more detailed information about the refractive index distribution at the sensor surface. This approach was illustrated in a model experiment in which three-surface plasmon spectroscopy was used to determine background refractive index variations and changes in the thickness and refractive index of a bovine serum albumin (BSA) multilayer.123 Recently, an SPR sensor based on twoplasmon spectroscopy on a bi-diffractive grating was investigated in terms of its ability to distinguish contributions to sensor response due to refractive index changes at the sensor surface (i.e., binding) and due to refractive index changes in the whole sample. Theoretical analysis yielding an estimate of an error of such decomposition was reported.124 Recently, Homola’s group reported an SPR biosensor using both longrange and short-range surface plasmons excited simultaneously on a diffraction grating of a special design.125 This approach offers several interesting features such as extended probe depth of the long-range surface plasmon and ability to distinguish sensor response caused by bulk and surface refractive index changes. The sensor was demonstrated to be able to detect changes in the refractive index as small as 3.5 × 10-6 RIU.125 An SPR sensor based on angular spectroscopy of surface plasmons was reported by Unfricht et al. in 2005.126 In their configuration, an LED source was moved by an angle encoder along an arc centered on the chip in order to change the incident angle. The light reflected from the chip surface was detected using a CCD camera that captured sequential images across the range of interrogated angles, and the coupling angle was measured.126

SPR Sensors

Figure 6. Concept of SPR sensor based on simultaneous excitation of surface plasmons by a polychromatic light and the dispersion of light on a special grating coupler.127

A new approach to the development of sensors based on spectroscopy of surface plasmons on diffractive gratings was reported by Telezhnikova and Homola.127 A collimated beam of polychromatic light was made incident on a special diffraction grating. A portion of incident light was coupled to a surface plasmon at the metal-dielectric interface via the second order of diffraction. Simultaneously, the light diffracted into the first diffraction order was dispersed and the light components of different wavelengths were directed to different areas of a position-sensitive detector (Figure 6). The coupling of light into a surface plasmon resulted in a drop in the intensity of diffracted light, which was observed as a narrow dip in the spectrum of diffracted light.127 A refractive index resolution of this sensor was established at 3 × 10-7 RIU. In 2007, Chien et al. reported an SPR sensor in which light was coupled into a metal-dielectric waveguide by a subwavelength grating.128 In this configuration, white light was made incident on the waveguiding layer through the grating structure and was coupled to a surface plasmon supported by a thin metal film. An SPR dip was observed in the spectrum of reflected light. A refractive index resolution of this sensor was established at 1 × 10-6 RIU.128

4.1.3. SPR Sensor Based on Waveguide Couplers Fiber optic SPR sensors present the highest degree of miniaturization of SPR sensors. The first fiber optic SPR sensors were reported in the early 1990s.129-131 In SPR sensors based on side-polished single-mode optical fibers, a fundamental mode of the fiber couples to a surface plasmon at the outer surface of a metal layer deposited on a sidepolished region of the fiber. The coupling results in attenuation of a transmitted light at a fixed wavelength (intensitymodulated SPR sensor) or a characteristic dip in the spectrum of transmitted light (wavelength-modulated SPR sensor). The main challenge for obtaining a stable performance is the sensitivity of polarization of light guided in the fiber to deformations of the fiber. The deformations generate changes in the strength of coupling between the light and a surface plasmon and thus interfere with SPR measurements. In 1999 Homola’s group reported an intensity-modulated fiber optic SPR sensor with a resolution better than 2 × 10-5 RIU.132 Later, they reported a wavelength-modulated version of this sensor with resolutions of 5 × 10-7 RIU (no deformations) and 3 × 10-5 RIU (under moderate deformations).133 In 2003 the sensitivity of the sensor to deformation of the fiber was

Chemical Reviews, 2008, Vol. 108, No. 2 471

dramatically reduced by the introduction of polarizationmaintaining fibers. An SPR sensor based on a side-polished polarization-maintaining fiber was demonstrated to exhibit a refractive index resolution of 2 × 10-6 RIU.134 Chiu et al. reported a fiber optic SPR sensor based on a D-shape optical fiber and heterodyne interferometry.135 Their sensor measured refractive index changes down to 2 × 10-6 RIU. In 2007 Lin et al. extended this approach to multimode fibers and reported an SPR sensor with wavelength modulation based on a side-polished multimode optical fiber.136 The sensor was shown to be able to resolve refractive index changes as small as 3 × 10-6 RIU. An integrated optical SPR sensor with intensity modulation and one sensing channel and one reference channel was reported by Mouvet et al.137 The signal from the sensing channel was normalized to the signal from the reference channel, resulting in an increased stability and a refractive index resolution of 5 × 10-5 RIU.138 A wavelength modulation-based integrated optical SPR sensor was reported by Homola’s group. The sensor was demonstrated to provide a refractive index resolution as low as 1 × 10-6 RIU.139 An SPR sensor based on a strip waveguide consisting of a germanium-doped silicon dioxide waveguiding layer on a silicon substrate and wavelength modulation was reported by Huang et al.140 Their sensor exhibited a sensitivity similar to that reported in ref 139. In conventional waveguide-based SPR sensors, the resonant coupling between a surface plasmon and a waveguide mode occurs for refractive indices of sample considerably higher than the refractive index of a typical aqueous sample. Various approaches to control the operating range of waveguide-based SPR sensor, so that it includes aqueous environments, were proposed. They include an integrated optical waveguide fabricated in low refractive index glass,141 a buffer layer,142 or a high refractive index overlayer.143 Skorobogatiy and Kabashin proposed to overcome this limitation of conventional waveguides by employing a photonic-crystal waveguide.144 In their paper, they theoretically demonstrated the feasibility of a photonic crystal waveguide-based SPR sensor using a single-mode photonic crystal waveguide in which the effective index of a mode confined in the waveguiding layer can be made considerably smaller than the refractive index of the waveguiding layer material, enabling phase matching with a surface plasmon at any wavelength.144 Debackere et al. proposed an interferometric SPR sensor consisting of a silicon-on-insulator (SOI) waveguide and a thin metal layer.145 In their design, a mode guided by a thin silicon film excited two surface plasmons on opposite sides of a thin metal film. These two modes propagated side by side over a short distance (10 µm) and were recombined in a waveguide mode at the end of the metal film. The transmitted intensity depended on the mutual phase delay of the two interfering surface plasmons. As the propagation constant of the surface plasmon propagating at the outer boundary was sensitive to the refractive index of the adjacent medium (sample), changes in the refractive index of sample could be measured by measuring changes in the intensity of transmitted light. As this sensor is based on interference rather than phase-matching, the operating range of the sensor can be conveniently controlled by the geometry of the device.145 Wang et al. reported an alternative approach to integrated optical SPR sensing based on electro-optical modulation.146

472 Chemical Reviews, 2008, Vol. 108, No. 2

Their sensor consisted of titanium-diffused channel waveguide in a lithium niobate substrate in which the propagation constant of a mode was modulated by voltage applied between the electrodes on two sides of the waveguiding layer. In the amplitude modulation mode, the intensity of transmitted light was measured as the electrode voltage was scanned and the slope of the intensity-voltage dependence was correlated with the amount of analyte captured at the sensor surface. The same approach was demonstrated for the wavelength modulation mode in which the resonant wavelength was measured.146

4.2. Biorecognition Elements and Their Immobilization In SPR affinity biosensors, one of the interacting molecules (biorecognition element or target molecule) is immobilized on the solid surface of the SPR sensor and the other is contained in a liquid sample. Which of the molecules is immobilized depends on the used detection format (section 5.1)sin direct, sandwich, and competitive detection formats, the molecule that needs to be immobilized is a biorecognition element; in the inhibition detection format, the immobilized molecules is the target molecule or its derivative. The choice of appropriate biorecognition elements and immobilization method is of critical importance with direct impact on key performance characteristics of the sensor such as sensitivity, specificity, and LOD.

4.2.1. Biorecognition Elements Various kinds of biorecognition elements have been employed in affinity SPR biosensors. Antibodies remain by far the most frequently used biorecognition element. They offer high affinity and specificity against target analyte. Moreover, antibodies against numerous target molecules are now commercially available. Development of high-quality antibodies is, however, a rather expensive and laborious process.147 Recently, single-chain antibody fragments (scFvs) have been also used as biorecognition elements.148 Biotinylated scFv fragments expressed in yeast can be spotted on streptavidin-coated sensor surfaces directly from cell supernatant without the need of purification.149 Another type of biorecognition element that has been employed in SPR sensors are peptides. In comparison with antibodies, peptides, in general, are inexpensive, more stable, and easier to manipulate. However, peptides sometimes lack high affinity and specificity against the target. In SPR biosensors, peptides have been applied mainly for the detection of antibodies, for example, antibodies against hepatitis G,150 herpes simplex virus type 1 and type 2,151 and Epstein-Barr virus,152 and for the detection of heavy metals.153 Recently, aptamers emerged as another promising type of biomolecular recognition element for SPR biosensors.154,155 DNA or RNA aptamers are single-stranded oligonucleotide sequences, which can be produced to bind to various molecular targets such as small molecules, proteins, nucleic acids, and even cells, tissues, and organisms.155,156 Moreover, the synthesis of aptamers is straightforward and reproducible.

4.2.2. Immobilization of Biorecognition Elements In SPR biosensors, one of the interacting molecules (mostly biorecognition element) is immobilized on the sensor surface. The surface chemistry has to be designed in such a way that it enables immobilization of a sufficient number

Homola

of biorecognition elements on the sensing surface while minimizing the nonspecific binding to the surface. In addition, biorecognition elements need to be immobilized on the sensor surface without affecting their biological activity. In principle, the molecules can be immobilized either on the surface or in a three-dimensional matrix. Although immobilization on surfaces is more straightforward to perform, the number of accessible biorecognition elements is limited by the capacity of the surface (too high density of immobilized biorecognition elements can lead to lower response due to sterical hindrance). Immobilization in a threedimensional matrix typically provides more binding sites than immobilization on the surface and a better environment for the preservation of immobilized molecules during prolonged storage.157 The most widely used three-dimensional matrix for immobilization of molecules in a structured environment is the carboxymethylated dextran.158 For two-dimensional (surface) immobilization of biorecognition elements on the sensing (gold) surface, selfassembled monolayers (SAMs) of alkanethiolates or disulfides have been widely used.159 To provide a desired surface concentration of biomolecular recognition elements and a nonfouling background, mixed SAMs of long-chained (n ) 12 and higher) alkanethiolates terminated with a functional group for further attachment of biomolecular recognition elements and oligo(ethylene glycol)-terminated shorterchained alkanethiolates for a nonfouling background have been developed.160,161 The main approaches to immobilization of molecules to the surface of SPR sensors are based on physical absorption and hydrophobic and electrostatic interactions,162 covalent coupling,158 and attachment of tagged molecules by a sitespecific non-covalent interaction between the tag and an immobilized capture molecule via biotin-avidin163 or histidine-chelated metal ion164 interaction or DNA hybridization.165 A more detailed account of immobilization methods is given in refs 31 and 166. As proteins represent the class of molecules that are most frequently used as biorecognition elements, the following section is dedicated to the immobilization of proteins. The most commonly used approach to the immobilization of proteins is via a covalent bond formed between the nucleophilic functional groups supplied by amino acids (e.g., amino groups, lysine; thiol groups, cysteine) of the protein and electrophilic groups (e.g., activated carboxyls, aldehydes) on the sensor surface.158,167 The covalent immobilization is stable; however, as proteins typically contain many functional groups, immobilization via these functional groups results in random orientation of immobilized proteins. Moreover, simultaneous immobilization via multiple functional groups may restrict conformational flexibility of the protein and impair its function. Another approach to the immobilization of proteins is based on biochemical affinity reaction. The most common example of this approach is the immobilization based on avidin-biotin chemistry. In this immobilization method, protein avidin (or a closely related streptavidin) is immobilized on the sensing surface (covalently or via preimmobilized biotin) and provides binding sites for subsequent attachment of a biotin-conjugated protein. The protein can be biotinylated by various methods targeting different groups on the protein. Orientation of the immobilized proteins depends on the orientation of avidin/streptavidin molecules, the biotinylation method used, and the properties of the protein. Alternatively, antibodies can be immobilized via

SPR Sensors

interaction between the Fc region of the antibody and protein A or protein G. This method provides good access to the binding site of the antibody, located on the Fab variable region; however, to control orientation of the antibody, the orientation of protein A itself needs to be controlled.168 Recently, DNA-directed immobilization of antibodies has been described.165 This approach takes advantage of DNA chip technology, which provides an exceptionally stable pattern of single-stranded DNA (ssDNA) sequences and uses it for the immobilization of proteins conjugated with complementary ssDNA sequences via DNA hybridization. This approach provides an elegant and flexible platform both for SPR sensors with a limited number of sensing channels and for high-throughput screening SPR systems. The drawback of this method is that it requires that the protein is conjugated with ssDNA.165 Another bioaffinity immobilization method is based on the interaction between histidinetagged protein and chelated metal ions, for example, nitrilotriacetic acid (NTA) immobilized on the surface and loaded with bivalent metal cations.164 NTA can be attached to the sensing surface covalently via acetic group using EDC/ NHS chemistry.169 Recombinant proteins with affinity tags can be produced by genetic engineering. As the tag can be placed at a defined position on the protein, this approach allows site-specific and thus highly ordered protein immobilization. Moreover, the binding between chelated metal ions and histidine is reversible and the immobilized protein can be released by introduction of a competing ligand (e.g., imidazole) or a chelation agent (e.g., ethylenediaminetetraacetic acid, EDTA). Whereas reusability of the surface is one of the main advantages of this immobilization method, the main drawback is the low affinity of the His-tag to an individual chelator. In 2005 Piehler’s group expanded this immobilization technique by designing supramolecular entities binding to oligohistidine tags with high affinity and stability.170 They designed supramolecular multivalent chelator heads (MCH) containing multiple NTA moieties and investigated their binding with hexahistidine (H6)- and decahistidine (H10)-tagged molecules. It was demonstrated that the binding stability of the complex increases with an increasing number of NTA moieties. An improvement of the stability of the chelator-oligohistidine complex by 4 orders of magnitude compared to that of mono-NTA was achieved.170 Tinazli et al. developed multivalent metal-chelating thiols for attachment of histidine-tagged proteins to the surface of SPR sensors via SAMs.171 Dramatically improved stability of protein binding by these multivalent chelator surfaces was observed compared to mono-NTA/His(6) tag interaction. Regenerability of the surface (removal of the protein) using EDTA was also demonstrated.171 Immobilization and arraying of histidine-tagged proteins by combining molecular and surface multivalency was demonstrated by Valiokas et al.172 They employed SAMs formed by triethylene glycolterminated alkyl thiols171 functionalized with either a single NTA moiety (mono-NTA) or a chelator head group containing two NTA moieties (bis-NTA). The process of immobilization of proteins was observed with SPR imaging.172 Immobilization of peptides can follow strategies similar to those developed for proteins, including electrostatic attraction and amine- or thiol-based covalent coupling. The most straightforward approach to the immobilization of oligonucleotides is based on the use of biotinylated derivatives.173 Small molecules with functional groups (aliphatic amines, thiols, aldehydes, or carboxylic groups) can be

Chemical Reviews, 2008, Vol. 108, No. 2 473

covalently linked to suitable corresponding groups on the sensor surface. Small molecules without suitable functional groups need to be derivatized.174 To deliver molecular recognition elements to different areas of the SPR sensor surface, the immobilization chemistry needs to be spatially controlled. For instance, Campbell’s group demonstrated the microspotting of double-stranded DNA on gold for SPR microscopy using two approaches.175 Both methods use streptavidin and biotinylated oligonucleotides. In the first method, the robotic microspotter was used to deliver nanoliter droplets of dsDNAs onto a uniform layer of streptavidin. In the second method, a streptavidin layer was also microspotted on a mixed-alkanethiolate SAM and, subsequently, microspots of dsDNA were added using microspotting. Homola’s group compared the microspotting technique with the conventional flow-through in situ functionalization.79 In their study, the spatially resolved functionalization based on microspotting applied to immobilization of short oligonucleotides was shown to provide a surface concentration of oligonucleotide probes of about 2.2 × 1012 oligonucleotides per cm2, which was higher by 80% than the surface coverage provided by the flow-through functionalization method.79

4.2.3. Nonfouling Surfaces As the adsorption of proteins is of major concern in numerous important biomedical and biological applications (biocompatible materials for prostheses, tissue engineering, cell culturing, implantable devices, microarrays, etc.), the adsorption of proteins to synthetic surfaces has been the subject of extensive research worldwide. Nonspecific adsorption of proteins to sensing surface presents a key challenge also for affinity biosensors.176,177 This problem is more severe when complex samples such as blood or cell lysate are to be analyzed.178,179 Although the molecular mechanism of protein resistance has not been fully understood, research into protein-resistant surfaces has made significant advances. Various surface coatings displaying low fouling or even nonfouling properties (i.e., exhibiting complete resistance to protein binding and cell colonization) have been proposed. For affinity biosensors it is especially important to create surfaces with low fouling background providing also abundant binding sites for immobilization of biomolecular recognition elements. Hydrophilic polymers such as poly(ethylene glycol) (PEG) and its derivatives have been successfully employed in the design of protein-resistant coatings for SPR sensors. The key factors that influence nonfouling properties of PEG molecules have been considered to be steric-entropy barrier characteristics and a high degree of hydration.180 Recent measurements of interfacial forces have shown that the protein resistance of PEGylated surfaces correlates with a net repulsive force versus distance curve.181 An approach utilizing a poly(Llysine) grafted with poly(ethylene glycol) (PLL-g-PEG) have been used by Pasche et al. to minimize the nonspecific adsorption of proteins.182,183 In their work, they varied the ratio of the number of lysine monomers and the PEG side chains and, with the optimized surface composition, they observed adsorption from blood serum below 2 ng/cm2.182 It was demonstrated that the immobilization of biorecognition elements to PLL-g-PEG surfaces can be performed by introducing biotin to the surface by assembling mixed (PLLg-PEG/PEGbiotin + PLL-g-PEG) from the corresponding mixed solutions. The resulting biotinylated surfaces have been shown to be highly resistant to nonspecific adsorption

474 Chemical Reviews, 2008, Vol. 108, No. 2

from serum while allowing binding of linkage proteins (e.g., streptavidin or avidin) and subsequent attachment of biotinylated biorecognition elements.184-186 PEG-containing molecules were also successfully employed as the secondary blocking agents on the surface with covalently immobilized proteins.187 Oligo(ethylene glycol) (OEG)-terminated alkanethiolates are widely used to form SAMs on gold surfaces of SPR sensors, typically also in combination with alkanethiolates containing various functional groups.188,189 Several heterobifunctional OEG or PEG surfaces have been employed in affinity biosensors. 190Although OEG-terminated SAMs have shown high resistance to nonspecific protein adsorption, the real biomedical applicability of these surfaces has been limited, mainly due to the limited oxidative stability of thiolates and the difficulty of integrating them into biomedical devices.191,192 It has been reported that PEG molecules autoxidize relatively rapidly, especially in the presence of oxygen and transition metal ions.193 Whitesides and co-workers explored alkanethiolates in SAMs with various functional groups and concluded that inertness of the surface is a general property of a group of surfaces rather than a specific property of ethylene glycol groups. They have also investigated the resistance of SAMs terminated with various groups to bacteria and mammalian cells and found that there was very little correlation between the resistance to the adsorption of protein and the adhesion of cells.191 On the basis of experimental investigation of SAMs with different functional groups, Whitesides’s group concluded that important requirements for protein resistance are (1) hydrophilicity, (2) ability to accept a hydrogen bond, (3) inability to donate to a hydrogen bond, and (4) a net neutral charge.194 Subsequently, they hypothesized that a zwitterionic SAM combining positively charged and negatively charged groups might offer a new type of protein-resistant surface.194 Kitano et al. formed a SAM of zwitterionic telomers on a metal surface and demonstrated its ability to reduce the nonspecific adsorption of proteins.195 Subsequently, Jiang’s group extended this material into a dual-functional zwitterionic poly(carboxybetaine methacrylate) (polyCBMA) polymer using reactive carboxyl groups for protein immobilization.196 They demonstrated a polyCBMA polymer with immobilized antibodies against human chorionic gonadotropin (hCG) that, when exposed to high concentrations of lysozyme and fibrinogen, exhibited nonspecific adsorption of <0.3 ng/cm2 (Figure 7).196 Most recently, Jiang’s group has investigated bacterial adhesion to the zwitterionic poly(sulfobetaine methacrylate) (polySBMA) and demonstrated that polySBMA surfaces dramatically reduce bacterial adhesion.197 Chen et el. showed that oligo(phosphorylcholine) SAMs exhibit strong resistance to protein adsorption, specifically to high concentrations of fibrinogen, lysozyme, and bovine serum albumin.198 In the past several few years, numerous SPR biosensors for detection in complex matrices have been reported. The most frequently targeted complex medium is blood serum, which is a key medium for medical diagnostics applications. Most of the reported SPR biosensors were designed to operate in serum diluted by buffer to serum concentrations from 1 to 25%. The detection of antibodies against EpsteinBarr virus (anti-EBNA) in 1% serum was reported by Homola’s group.199 A synthetic peptide, which was used as a biorecognition element, was immobilized on the surface via hydrophobic and electrostatic interactions. Nonspecific

Homola

Figure 7. Adsorption of 1 mg/mL fibrinogen, 1 mg/mL lysozyme, and 20 µg/mL hCG from PBS on (a) polyCBMA-grafted surfaces and (b) polyCBMA-grafted surfaces with immobilized hCG antibodies. Reprinted with permission from ref 196. Copyright 2006 American Chemical Society.

adsorption from 1% serum was found to be negligible, and the LOD for anti-EBNA was 0.2 ng/mL.199 Ayela et al. reported an SPR sensor for the detection of IA-2 autoantibodies also in 1% human serum.200 In that work various types of mixed SAMs were evaluated in terms of specific and nonspecific binding. It was observed that the nonspecific adsorption from serum to surface coated with EG6SAM100%COOH was about an order of magnitude lower than the adsorption to a SAM100%COOH-coated surface. Using an EG6-SAM25%COOH-coated surface, their SPR sensor was able to detect antibody at a concentration of 0.2 nM.200 Cao et al. detected prostate-specific antigen-1antichymotrypsin (PSA-ACTcomplex) in 10% serum. Their sensor employed a mixed SAM of alkanethiolates terminated with EG6-COOH and EG3-OH groups. The COOH group was biotinylated and used for subsequent immobilization of streptavidin and antibody against the PSA-ACTcomplex. Nonspecific adsorption of albumin, IgG, and fibrinogen on the sensing surface was found to be negligible, and the sensor was demonstrated to be able to detect the PSA-ACT complex at concentrations below 50 ng/mL.201 Chung et al. demonstrated detection of antibody against human hepatitis B virus (hHBV) in serum dilutions from 5 to 60%.202 Their sensor employed a thiol monolayer prepared using 11mercaptoundecanoic acid to which hHBV antigen was coupled using the EDAC/NHS coupling chemistry. The level of nonspecific binding at different concentrations of serum

SPR Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 475

was measured; serum concentrations of 5, 10, and 20% produced a nonspecific sensor response corresponding to protein adsorption of 20, 26, and 68 ng/cm2, respectively.202 Miura’s group reported an SPR sensor for the detection of insulin in 10 and 25% serum.203 The sensing surface of their sensor was constructed using a heterobifunctional oligo(ethylene glycol)-dithiocarboxylic acid derivative (OEGDCA) containing dithiol and carboxyl end groups to which insulin was covalently bound. The results observed with OEG-DCA SAMs were further compared to those reported using a SAM of monothiol tethered oligo(ethylene glycol)carboxylic acid (OEG-COOH), and it was concluded that the resistance of the present bare dithiol-tethered OEG-DCA SAM is comparable to or better than that of the monothioltethered OEG-COOH SAM.204 The sensor was able to detect insulin in serum at insulin concentrations down to 6 ng/mL.203 Although these developments present clear progress toward SPR biosensing in serum, an SPR sensor capable of detecting relevant concentrations of analyte in whole serum has not been demonstrated yet.

4.3. Summary The past decade has witnessed development of numerous SPR sensors based on excitation of surface plasmons via prism coupling, waveguide coupling, or diffraction coupling and angular, wavelength, intensity, or phase modulation. SPR sensor platforms based on prism coupling remain by far the most common. SPR sensing platforms providing the highest resolutions are typically based on angular88,127 or wavelength62 spectroscopy of surface plasmons or phase modulation.113,115,117 The best SPR sensor platforms with a limited number of sensing channels (<10) provide a refractive index resolution around 10-7 RIU. High-throughput SPR sensors with a large number of sensing channels (>100) are usually based on intensity modulation (SPR imaging) and offer an order of magnitude worse performance.79 One of the prospective approaches to further improving the resolution of SPR sensors involves long-range surface plasmons. SPR sensors based on wavelength spectroscopy of long-range surface plasmons were demonstrated to be able to deliver resolution as low as 3 × 10-8 RIU.96 However, as the field of long-range surface plasmons extends much farther from the sensing surface than that of conventional surface plasmons, this improvement can be fully harnessed only when large analytes (e.g., bacterial pathogens) are targeted or biorecognition elements are immobilized in a extended coupling matrix. Various types of biorecognition elements and immobilization methods are available to allow the SPR sensing platforms to be tailored for specific detection of chemical and biological substances. Proteins (e.g., antibodies) and peptides are most frequently immobilized via covalent bonds formed between amino groups of the protein and activated carboxyls on a SAM of alkanethiolates or within a dextran matrix. Oligonucleotides can be efficiently immobilized via interaction between avidin or streptavidin immobilized on the sensing surface and biotinylated oligonucleotide. Small molecules are usually conjugated with a larger protein (BSA), which is subsequently (covalently) immobilized on the sensor surface. High-throughput SPR sensors demand immobilization methods capable of accurate spatially controlled delivery of different biorecognition elements to different areas of the sensing surface. This can be achieved by combining the streptavidin-coated surface with a spatially controlled de-

Figure 8. Main detection formats used in SPR biosensors: (A) direct detection; (B) sandwich detection format; (C) competitive detection format; (D) inhibition detection format.

livery of biotinylated biorecognition elements by microspotting.79 An interesting alternative approach is based on the use of a conventional DNA chip and its conversion to a protein chip by incubating the chip with a mixture of proteins conjugated with complementary DNA sequences.165

5. Applications of SPR Sensors for Detection of Chemical and Biological Species SPR biosensors have been applied in numerous important fields including medical diagnostics, environmental monitoring, and food safety and security.

5.1. Detection Formats Various formats for the detection of chemical and biological analytes have been applied in SPR sensors.205,206 The format of detection is chosen on the basis of the size of target analyte molecules, binding characteristics of available biomolecular recognition element, range of concentrations of analyte to be measured, and sample matrix.206 The most frequently used detection formats include (a) direct detection, (b) sandwich detection format, (c) competitive detection format, and (d) inhibition detection format (Figure 8). In the direct detection mode (Figure 8A), the biorecognition element (e.g., antibody) is immobilized on the SPR

476 Chemical Reviews, 2008, Vol. 108, No. 2

sensor surface. Analyte in solution binds to the antibody, producing a refractive index change detected by the SPR sensor. Direct detection is usually preferred in applications, where direct binding of analyte of concentrations of interest produces a sufficient response. The specificity and LOD can be improved by using the sandwich detection format (Figure 8B), in which the sensor surface with captured analyte is incubated with a second antibody. Smaller analytes (molecular weight < 5000) often do not generate a sufficient change in the refractive index and therefore are measured using either competitive or inhibition detection format. Figure 8C shows an example of the competitive detection format, in which the sensing surface is coated with an antibody interacting with the analyte; when a conjugated analyte is added to the sample, the analyte and its conjugated analogue compete for a limited number of binding sites on the surface. The binding response is inversely proportional to the analyte concentration. In the inhibition detection format (Figure 8D) a fixed concentration of an antibody with affinity to analyte is mixed with a sample containing an unknown concentration of analyte. Then, the mixture is injected in the flow cell of the SPR sensor and passed over a sensor surface to which analyte or its analogue is immobilized. Noncomplexed antibodies are measured as they bind to the analyte molecules immobilized on the sensor surface. The binding response is inversely proportional to the concentration of analyte. In recent years, various modifications and extensions of these basic detection formats have been developed in order to expand and improve detection capabilities of SPR biosensors.207 Several detection formats for the detection of multiple analytes have been reported. Chung et al. modified the sandwich assay approach to allow detection of multiple analytes in a single sensing channel.208 They immobilized antibodies against two different analytes on the same area of the sensor surface. After incubation of sample with the sensor surface, solutions containing respective antibody were sequentially injected. Sensor response to each antibody was proportional to a concentration of the respective analyte.208 The same concept was adopted for inhibition detection format by Lechuga’s group.209 They immobilized three analyte derivatives on the sensor surface, and the sensor response to each analyte was determined by incubation of the sensor with a solution containing a respective antibody.209 Enhancement of sensor sensitivity through the “labeling” of a secondary antibody in the sandwich detection format by latex particles210 or gold nanoparticles211 was demonstrated in the 1990s. In 2005 Mitchell et al. used the labeling approach in the inhibition detection format.212 They used gold nanoparticles to improve the sensitivity of the SPR sensor for detection of progesterone. In their study, the (primary) antibody mixed with a sample was conjugated with biotin. Upon incubation of the sample with a sensor surface coated with progesterone, streptavidin conjugated with a gold nanoparticle (10 or 20 nm in diameter) was injected. The LOD for progesterone in buffer was established at 23 pg/ mL; this corresponds to an improvement by a factor of 17 compared to the inhibition format.212 In 2003 Sato et al. described amplification of SPR response to DNA based on the use of DNA-carrying hydrogel microspheres.213 Acrylamide-based microspheres with carboxyl groups were conjugated with DNA probes. Binding of DNA-carrying acrylamide-based microspheres with target DNAs at the sensor surface resulted in a 100-fold increase in sensitivity

Homola

compared to the sensitivity of nonamplified DNA target hybridization. In 2004 Okamura et al. reported the enhancement of an SPR sensor response by means of hydrogel nanospheres and the same protocol.214 Their hydrogel nanospheres were prepared by precipitation polymerization of acrylamide, methylenebisacrylamide, and methacrylic acid.214 Styrene-glycidyl methacrylate (SG) microspheres prepared by soap-free emulsion copolymerization were proposed by Sato et al.215 They used DNA conjugated with SG microspheres to enhance the signal from hybridization of DNA to a complementary DNA strand immobilized on the SPR sensor surface. They immobilized short ssDNA onto the sensor surface to capture (a longer) target DNA. Subsequently, DNA-carrying microspheres were injected to bind to the free portion of the target DNA.215 In 2006 Komatsu et al. proposed an amplification method suitable for intensitymodulated SPR sensors based on dye-doped polymer particles.216 The dye-doped polymer particles can enhance the sensitivity of intensity-modulated SPR sensors by two mechanismssby the shift in the resonant coupling condition due to the refractive index change induced by the presence of the particles and by absorption of light in the dye-doped particles. In a model experiment, the authors compared sensor responses due to the binding of BSA or BSA conjugated with dye-doped polymer particles to anti-BSA immobilized on the sensor surface. The use of particles was demonstrated to provide a 100-fold improvement in sensor sensitivity.216 Recently, several methods for enhancing the sensitivity of SPR biosensors based on enzymatic amplification were developed by Corn’s group. Goodrich et al. reported an enzymatic amplification method for the detection of DNA molecules that utilizes RNA microarrays in conjunction with the enzyme RNase H.217 In this method, a single-stranded RNA microarray is exposed to a solution containing both the complementary DNA and RNase H. The DNA binds to its RNA complement on the surface and forms an RNADNA heteroduplex. RNase H then binds to this heteroduplex, selectively hydrolyzes the RNA probe, and releases the DNA complement back into solution. The released DNA molecule binds to another RNA probe on the surface, so that a single DNA molecule can initiate the destruction of many surfacebound RNA probes. Eventually, all of the RNA probe molecules are destroyed and removed from the surface. The loss of RNA probe molecules from the surface is detected by the SPR method. Using this approach, DNA solutions were detected at levels down to 10 fM.217 Another enzymatic amplification approach for the detection of DNA was demonstrated by Lee et al.218 In the first step of this approach, an ssDNA array is exposed to a solution containing target DNA and enzyme exonuclease III (ExoIII); the target DNA hybridizes to its complementary ssDNA array elements, and ExoIII binds to the dsDNA. ExoIII selectively hydrolyzes the probe DNA strand from the duplex, releasing the target DNA strand back into solution. The released target DNA is then free to bind to another surface-bound ssDNA probe. This cyclic reaction progresses until all ssDNA probes on the surface are destroyed by ExoIII.218 Using this approach, a 16-mer ssDNA was detected down to 10-100 pM, which presents a 102-103 -fold improvement.218 Fang et al. described an amplification method for the detection of RNA based on poly(A) enzyme chemistry and nanoparticle enhancement.219 In this method the target RNA is adsorbed from solution onto a single-stranded LNA microarray. Subsequently, poly(A) tails are introduced to the surface-

SPR Sensors

bound RNAs via the poly(A) polymerase surface reaction. Finally, poly(A) tails are hybridized with T30-DNA-coated Au nanoparticles. This approach was demonstrated to allow detection of RNAs down to 10 fM.219 Recently, Li et al. reported another enzymatic amplification approach.220 In the first step of this method, the target protein binds to the aptamer immobilized on the sensor surface. Then, a horseradish peroxidase (HRP)-conjugated antibody to the target protein is introduced to create an aptamer-protein-antibody sandwich, which is subsequently exposed to the substrate 3,3′,5,5′-tetramethylbenzidine (TMB). Reaction of TMB with HRP gives rise to a dark blue precipitate. This methodology was applied in an SPR sensor for detection of human thrombin. The LOD was demonstrated at a concentration of 500 fM, which corresponds to an enhancement factor of ∼104.220

5.2. Food Quality and Safety Analysis As the acceptance of SPR biosensor technology in food analysis continues to increase, the number of publications on SPR biosensors for the detection of analytes related to food quality and safety increases.221-225 The targeted analytes include pathogens, toxins, drug residues, vitamins, hormones, antibodies, chemical contaminants, allergens, and proteins.

5.2.1. Pathogens In recent years, various pathogens have been targeted by SPR biosensors.226 In particular, they include bacteria, protozoa, fungi, and parasites. Escherichia coli O157:H7 was first detected by SPR by Fratamico et al. in 1998.227 Since then, numerous SPR biosensors for the detection of E. coli O157:H7 have been reported. Choi’s group used the commercial SPR sensor Multiskop (Optrel, Germany) and monoclonal antibodies immobilized on a protein G-coated sensor surface. The sensor was demonstrated to be able to directly detect E. coli O157: H7 at concentrations as low as 104 cells/mL.228 Subsequently, they demonstrated that in conjunction with the immobilization of antibodies via a mixed SAM of alkanethiolates, the same SPR instrument can detect E. coli O157:H7 down to 102 cells/mL.229 Taylor et al. detected E. coli O157:H7 using a custom-built SPR sensor with wavelength modulation and examined the effect of various treatment methods on sensor performance.230 A monoclonal antibody was immobilized on a mixed -COOH- and -OH-terminated SAM of alkanethiolates via amine coupling chemistry. Detection of E. coli O157:H7 was performed in the sandwich detection format using a secondary polyclonal antibody. Detection limits for detergent-lysed bacteria, heat-killed bacteria, and untreated bacteria were determined to be 104, 105, and 106 colonyforming units (cfu)/mL, respectively.230 Detection of E. coli O157:H7 using a commercially available Spreeta sensor (Texas Instruments Co.) was reported by Meeusen et al.231 Biotinylated polyclonal antibody against E. coli O157:H7 was immobilized on the avidinated gold surface. The SPR biosensor was shown to be capable of detecting E. coli O157: H7 in cultures at levels down to 8.7 × 106 cfu/mL in 35 min.231 Another SPR sensor for the detection of E. coli O157: H7 based on the Spreeta sensor was reported by Su and Li.232 In that work, polyclonal E. coli O157:H7 antibodies were immobilized via protein A adsorbed on the sensor surface. The sensor was demonstrated to be able to detect E. coli O157:H7 in an aqueous environment at levels down to 106 cells/mL.232 Subramanian et al. demonstrated an SPR bio-

Chemical Reviews, 2008, Vol. 108, No. 2 477

sensor for the detection of E. coli O157:H7. They used the commercial SPR sensor SR 7000 (Reichert Analytical Instruments) and attachment of E. coli O157:H7 polyclonal antibodies via alkanethiolate SAM and amine coupling chemistry. Detection of E. coli O157:H7 was performed in sandwich format. The detection limit for E. coli O157:H7 was established at 103 cfu/mL.233 Taylor et al. reported SPRbased detection of E. coli O157:H7 in apple juice using a custom-built multichannel SPR sensor with wavelength modulation and sandwich detection format.234 Biotinylated polyclonal antibodies against E. coli O157:H7 were immobilized via streptavidin attached to a mixed SAM of oligo(ethylene glycol) alkanethiolate and biotinylated alkanethiolate. Detection of heat-killed E. coli O157:H7 was performed in buffer, in a mixture of four bacterial species, and in apple juice. The effect of the pH of the apple juice on the sensor response was investigated, and SPR responses were higher for bacteria in apple juice at pH 7.4 than in apple juice at pH 3.7. The LOD was 1.4 × 104 cfu/mL in buffer and about 105 cfu/mL in apple juice with an adjusted pH of 7.4.234 Waswa et al. used two commercial SPR sensorsslaboratory instrument Biacore 2000235 and a miniature SPR sensor Spreeta236sto detect E. coli O157:H7. Immobilization of E. coli O157:H7 antibody for the laboratory SPR system was performed by first attaching protein A using a carboxymethylated dextran layer and amine coupling chemistry and subsequent attachment of the antibody.235 The LOD for E. coli O157:H7 in pasteurized milk was determined to be 25 cfu/mL.235 In the Spreeta sensor, the biotinylated E. coli O157:H7 antibody was attached to a layer of neutravidin molecules adsorbed on the gold surface.236 The detection limit for E. coli O157:H7 in milk, apple juice, and ground beef was estimated to be in the range of 102-103 cfu/mL.236 Salmonella enteritidis was detected using a custom-built SPR sensor with wavelength modulation by Koubova´ et al.162 In that work, a double layer of antibodies was physisorbed on a bare gold surface and cross-linked with gluteraldehyde. Direct detection of heat-killed, ethanol-soaked S. enteritidis at a concentration as low as 106 cfu/mL was demonstrated.162 Bokken et al. demonstrated detection of Salmonella groups A, B, D, and E using the commercial SPR sensor Biacore 3000.237 Antibodies were immobilized in a carboxymethylated dextran layer via amine coupling chemistry, and detection of Salmonella serotypes was performed using the sandwich format. Salmonella serotypes were detectable at a concentration of 1.7 × 105 cfu/mL even in the presence of other bacteria at 108 cfu/mL levels.237 Choi’s group demonstrated the detection of Salmonella typhimurium using the commercial SPR sensor Multiskop and monoclonal antibodies immobilized via protein G attached to an alkanethiolate SAM on the sensor surface. The LOD was 102 cfu/mL.238 An SPR sensor for the detection of Salmonella paratyphi was demonstrated by the same group.239 They used the same SPR instrumentsa Multiskopsand a similar method for the attachment of monoclonal antibodies via protein G. Detection of S. paratyphi was shown down to concentrations of 102 cfu/mL.239 In 2006 detection of Salmonella in food matrices was reported by two groups. Taylor et al. reported SPRbased detection of Salmonella choleraesuis serotype typhimurium in apple juice using a custom-built multichannel SPR sensor with wavelength modulation and sandwich detection format.234 Biotinylated polyclonal antibodies against Salmonella were immobilized via streptavidin attached to a mixed SAM of oligo(ethylene glycol) alkanethiolate and

478 Chemical Reviews, 2008, Vol. 108, No. 2

biotinylated alkanethiolate. Detection of Salmonella was performed in buffer, in a mixture of four bacterial species, and in apple juice. The effect of the pH of the apple juice on the sensor response was investigated, and SPR responses were higher for bacteria in apple juice at pH 7.4 than in apple juice at pH 3.7. The LOD for S. choleraesuis was 4.4 × 104 cfu/mL in buffer and about 104 cfu/mL in apple juice with an adjusted pH of 7.4.234 Waswa et al. detected S. enterica serovar Enteritidis in milk using the commercial SPR sensor Biacore 2000.235 In that work, polyclonal S. enterica antibody was immobilized by first attaching protein A using a carboxymethylated dextran layer and amine coupling chemistry and subsequent attachment of the antibody to protein A.235 The LOD for Salmonella in pasteurized milk was determined to be 23 cfu/mL.235 In 2007 Mazumdar et al. also reported the detection of Salmonella in milk.240 They used the commercial SPR sensor Plasmonic and sandwich detection format. Polyclonal capture antibody was immobilized by self-assembly on the hydrophobic sensing surface formed by alkylsilanes. Milk spiked with S. typhimurium cells, killed by thimerosal (1%, w/w), was incubated with the sensing surface for 15 min and then switched with a solution containing the second antibody. The LOD for S. typhimurium cells in milk was at 105 cells/mL.240 Listeria monocytogenes was detected by Koubova´ et al.162 They used a custom-built SPR sensor with wavelength modulation and a double layer of antibodies adsorbed on a bare gold surface and cross-linked with glutaraldehyde. Heatkilled Listeria bacteria were detected at levels down to 107 cfu/mL.162 Leonard et al. used the commercial SPR sensor Biacore 3000 and competitive format to detect the L. monocytogenes.241 A polyclonal anti-goat antibody was immobilized in a carboxymethylated dextran layer using amine coupling chemistry. Solutions of known concentrations of L. monocytogenes were incubated with rabbit anti-Listeria antibodies. Cells and bound antibodies were then centrifuged out of solution, and the unbound antibodies remaining in solution were detected by the SPR sensor. The LOD was determined to be 105 cells/mL.241 Detection of L. monocytogenes in apple juice was reported by Taylor et al.234 They used a custom-built multichannel SPR sensor with wavelength modulation and sandwich detection format. Biotinylated polyclonal antibodies against L. monocytogenes were immobilized via streptavidin attached to a mixed SAM of oligo(ethylene glycol) alkanethiolate and biotinylated alkanethiolate. Detection of L. monocytogenes was performed in buffer, in a mixture of four bacterial species, and in apple juice. In the considered range of concentrations of L. monocytogenes in apple juice, the sensor response was higher for bacteria in apple juice with an adjusted pH of 7.4 than for those in buffer or natural apple juice with a pH of 3.7. The LOD for L. monocytogenes was determined to be about 3 × 103 cfu/mL for detection in both buffer and apple juice with a pH of 7.4.234 Campylobacter jejuni was detected by Taylor et al.234 They used a custom-built multichannel SPR sensor with wavelength modulation and sandwich detection format. Biotinylated polyclonal antibodies against C. jejuni were immobilized via streptavidin attached to a mixed SAM of oligo(ethylene glycol) alkanethiolate and biotinylated alkanethiolate. Detection of heat-killed C. jejuni was performed for buffer solutions containing only C. jejuni as well as a mixture of C. jejuni and other bacteria and apple juice spiked with C. jejuni. The LOD for C. jejuni was 1 × 105 cfu/mL in buffer

Homola

and about 5 × 104 cfu/mL in apple juice.234 Staphylococcus aureus was detected by means of an SPR biosensor by Subramanian et al.242 They used the commercial SPR sensor SR 7000 and detected S. aureus directly or in sandwich detection format. Alkane monothiol and dithiol dendritic tether-based SAMs were examined for subsequent attachment of S. aureus antibodies using amine coupling chemistry. The LOD was determined to be 107 cfu/mL for direct detection and 105 cfu/mL for sandwich format for both sensing surfaces.242 Balasubramanian et al. reported an SPRbased detection of S. aureus using lytic phage as a biorecognition element.243 In that work a commercial SPR sensor, Spreeta, was used as a detection platform, and lytic phage was immobilized on the sensor surface by direct physical adsorption. The LOD for S. aureus in buffer was found to be 104 cfu/mL.243 Recently, Taylor et al. demonstrated the simultaneous detection of the four above-mentioned bacteriasE. coli O157:H7, C. jejuni, S. typhimurium, and L. monocytogeness on a custom-built multichannel SPR sensor.234 All bacteria were heat-killed and ultrasonicated prior to detection. Simultaneous detection of individual bacteria in the mixtures showed good agreement with detections of individual bacteria in buffer. Detections of individual bacteria and mixtures were also performed in apple juice samples. LODs for all four cases were established at 104, 5 × 104, 5 × 104, and 104 cfu/mL for E. coli O157:H7, C. jejuni, S. typhimurium, and L. monocytogenes, respectively.234 Yersinia enterocolitica was detected by Choi’s group,244 using the commercial SPR sensor Multiskop and monoclonal Y. enterocolitica antibodies immobilized via protein G attached to an alkanethiolate SAM on the sensor surface. The LOD for Y. enterocolitica in buffer was determined to be 102 cfu/mL.244 Vibrio cholerae O1 was detected by means of an SPR biosensor by Choi’s group.245 They used the commercial SPR sensor Multiskop and monoclonal antibodies immobilized via protein G attached to an alkanethiolate SAM on the sensor surface. The LOD for V. cholera O1 in buffer was determined to be about 4 × 105 cfu/mL.245 Protozoan parasite Cryptosporidium parVum oocyst, was detected by an SPR biosensor by Kang et al.246 In that work, the authors used the commercial SPR sensor platform Biacore 2000 and direct detection approach. Immobilization of biotinylated monoclonal antibody against C. parVum oocyst was performed on a mixed alkanethiolate SAM with attached streptavidin. The biosensor was able to detect C. parVum oocyst in buffer directly at a concentration of 106 oocysts/ mL. The authors also explored an alternative detection format consisting of the immunoreaction step between the biotinylated antibody and oocysts followed by the binding step of antibody-oocysts complex on the streptavidin-incubated surface. In this format, the LOD for C. parVum oocyst in buffer was established at 102 oocysts/mL.246 Detection of a fungal pathogen, Fusarium culmorum, in wheat using an SPR sensor was demonstrated by Zezza et al.247 Detection of F. culmorum was based on extraction of DNA from a sample, amplification of specific DNA fragment of F. culmorum, and subsequent detection of the amplicon using the SPR method via hybridization with complementary sequence immobilized on the SPR sensor (Biacore X) surface. The detection limit for the F. culmorum amplicon was 0.25 ng/µL. In 30 ng of durum wheat DNA, the smallest

SPR Sensors

detectable amount of specific F. culmorum DNA was 0.06 pg.247

5.2.2. Toxins Toxins implicated in food safety include mainly toxins produced by bacteria, fungi, and algae. Staphylococcal enterotoxin B (SEB) was detected by an SPR sensor by Nedelkov et al. in 2000.248,249 They used the commercial SPR sensor Biacore X and SEB antibody immobilized in a carboxymethyldextran layer on the sensor surface via amine coupling chemistry. SEB was detected directly in milk and mushroom samples at levels down to 1 ng/mL. The SPR detection was followed by identification of the bound toxin by matrix-assisted laser desorption/ ionization time-of-flight (MALDI-TOF) mass spectrometry.248,249 In 2002 Homola’s group reported detection of SEB using a fiber optic SPR sensor.250 A double layer of antibodies was physisorbed on the surface of the SPR sensor and cross-linked with glutaraldehyde. SEB was detected directly, and the limit of detection for SEB in buffer was established at 10 ng/mL.250 Naimushin et al. detected SEB using a prototype of an SPR sensor developed by Texas Instruments Co. and antibodies immobilized on the sensor surface via a gold binding peptide.251 SEB was detected in buffer and seawater using direct detection or sandwich format with one or two amplification antibodies. The LOD for direct detection was 0.2 nM (5.6 ng/mL) in buffer and 1 nM (28 ng/mL) in seawater. Using a one-step amplification, concentrations of 20 pM (0.6 ng/mL) and 50 pM (1.4 ng/mL) were detected in buffer and seawater, respectively. The use of a second amplification antibody was shown to improve the LOD in buffer to 100 fM (2.8 pg/mL).251 Homola et al. reported the detection of SEB in buffer and milk.91 In that work, a custom-built wavelength modulation SPR sensor was employed, and polyclonal SEB antibody was immobilized on a mixed SAM of alkanethiolate using amine coupling chemistry. Detection of SEB was performed directly or using sandwich detection format. The LOD for direct detection of SEB in buffer was 5 ng/mL. Using a secondary antibody the LOD was improved to 0.5 ng/mL for both buffer and milk.91 In 2003 Medina reported the detection of SEB using an inhibition detection format.252 Sample containing SEB was incubated with a known concentration of SEB antibody for 20-30 min, and then the mixture was analyzed by the SPR sensor. The LOD for SEB in milk was established at 0.3 ng/mL. The sensing surface was demonstrated to be regenerable for repeated use by 100 mM hydrochloric acid.252 Detection of SEB in another food matrix was demonstrated by Medina, who detected SEB in ham tissue.253 In that work the commercial SPR sensor Biacore 1000 was used, and polyclonal antibody was immobilized in a carboxymethyldextran layer on the sensor surface via amine coupling chemistry. The LOD for sandwich detection format was determined to be 2.5 ng/mL in both buffer and ham tissue extract.253 Medina also demonstrated detection of Staphylococcal enterotoxin A (SEA) in raw eggs using the commercial SPR sensor Biacore 1000 and competitive detection format.254 SEA was immobilized in a carboxymethyldextran layer on the sensor surface via amine coupling chemistry. Homogenized raw egg samples were clarified by centrifugation. Anti-SEA was added to the sample, allowing SEA to bind with anti-SEA. The bound complex was separated from the free antibody by centrifugation. The supernatant was injected

Chemical Reviews, 2008, Vol. 108, No. 2 479

over the SEA-coated surface. Using this approach, SEA was detected in whole egg at concentrations down to 1 ng/mL.254 Domoic acid (DA) was detected using an SPR biosensor by Lotierzo et al.255 They used the commercial SPR sensor Biacore 3000 and a molecularly imprinted polymer photografted on a gold chip as a biorecognition element. Detection was performed in a competitive binding format in which free DA competed with its conjugate with horseradish peroxidase. The sensor was demonstrated to be able to detect DA in buffer at a concentration as low as 5 ng/mL.255 In 2005 Yu et al. demonstrated detection of DA using a custom-built SPR and inhibition detection format.256 DA was immobilized on a mixed SAM of OEG-containing alkanethiolates using amine coupling chemistry. The effect of regeneration and storage on the performance of the SPR sensor was investigated. The LOD of DA in buffer was established at 0.1 ng/ mL.256 Traynor et al. demonstrated SPR-based detection of DA in extracts of shellfish species.257 They used the commercial SPR sensor Biacore Q, and immobilization of DA in a carboxymethyldextran layer on the sensor surface was performed by amine coupling chemistry. Detection of DA was performed in inhibition format. Detection limits for DA in mussels, oysters, and cockles were determined to be about 1, 4.9, and 7 µg/g, respectively.257 Stevens et al. detected domoic acid in clam extracts using a portable SPR biosensor employing Spreeta 2000 modules and inhibition detection format.258 Polyclonal DA antibodies were immobilized on the sensor surface via a gold binding peptide. Detection was performed in buffer and in diluted clam extracts. The sensor was able to detect DA at 3 ng/mL in both buffer and diluted clam extracts.258 Aflatoxin B1 was detected using the commercial SPR sensor Biacore 1000 by Daly et al.259 In that work, aflatoxin B1 was conjugated to BSA and immobilized on carboxymethylated dextran using amine coupling chemistry. Detection was performed using inhibition detection format. Detection of aflatoxin B1 in buffer was demonstrated at levels down to 3 ng/mL.259 Dunne et al. demonstrated an SPR sensor for aflatoxin B1 using scFvs as a biorecognition element.148 Detection was performed in the commercial SPR sensor Biacore 3000 using inhibition detection format. Aflatoxin B1 derivative was immobilized on the carboxymethylated dextran layer on the sensor surface using amine coupling chemistry. Regeneration protocol was developed enabling at least 75 detection/regeneration cycles. The LOD for aflatoxin B1 in buffer was 375 pg/mL for monomeric scFv and 190 pg/mL for dimeric scFv.148 Detection of deoxynivalenol in buffer and wheat was demonstrated by Tu¨do¨s et al.260 In that work, the commercial SPR system Biacore Q system was used and detection of deoxynivalenol was performed using the inhibition detection format. Deoxynivalenol conjugated to casein was immobilized on a carboxymethylated dextran layer on the sensor surface using amine coupling chemistry. The sensor was demonstrated to detect deoxynivalenol in buffer down to 2.5 ng/mL and showed good agreement with liquid chromatography-tandem mass spectrometry measurements on wheat samples.260

5.2.3. Veterinary Drugs Another important field in which SPR biosensor technology has been increasingly applied is testing for veterinary drug residues (e.g., antibiotics, β-agonists, and antiparasitic drugs) in food.261

480 Chemical Reviews, 2008, Vol. 108, No. 2

An SPR sensor for the detection of antibiotics (penicillins and cephalosporins) in milk was demonstrated by Cacciatore et al.262 Their approach was based on the noncompetitive inhibition of the binding of digoxigenin-labeled ampicillin (DIG-AMPI) to a soluble penicillin-binding protein 2× derivative (PBP 2×*) of Streptococcus pneumoniae by other β-lactam antibiotics. Subsequently, the DIG-AMPI/PBP 2×* complex was detected using the commercial SPR platform Biacore 3000 and digoxigenin antibody immobilized on the sensor chip. The LODs for the selected antibiotics in raw milk were established to be below 2 ng/mL for benzylpenicillin, ampicillin, and amoxicillin, 15 ng/mL for cloxacillin, 50 ng/mL for cephalexin, and 25 ng/mL for cefoperazone.262 Chloramphenicol and chloramphenicol glucuronide residues in various food matrices were detected by Ashwin et al.263 They used the commercial SPR platform Biacore Q and direct detection format. They detected chloramphenicol in extracts from honey, prawns, and dairy products and chloramphenicol glucuronide in extracts of porcine kidney at concentrations below 0.2 µg/kg.263 Detection of chloramphenicol and chloramphenicol glucuronide using an SPR sensor and inhibition assay was performed by Ferguson et al.264 They used the commercial SPR sensor Biacore Q and a chip with immobilized chloramphenicol derivative (Qflex Kit Chloramphenicol, Biacore). A known concentration of drug-specific antibody was mixed with the sample and injected over the surface of a sensor chip on which a chloramphenicol derivative was immobilized. Chloramphenicol and chloramphenicol glucuronide in extracts from food matrices were detected at levels down to 0.005 µg/kg (poultry), 0.02 µg/kg (honey), 0.04 µg/kg (prawn), and 0.04 µg/kg (milk).264 Dumont et al. demonstrated an SPR sensor for the detection of fenicol antibiotic residues in shrimps.265 They used the commercial SPR sensor Biacore Q and inhibition detection format. Analyte molecules were immobilized on carboxymethylated dextran using amine coupling chemistry. Chloramphenicol, florefenicol, florefenicol amine, and thiamphenicol were detected in extract from shrimp at levels down to 1, 0.2, 250, and 0.5 ng/mL, respectively.265 In 2007 Moeller et al. reported an SPR biosensor for the indirect detection of tetracycline in honey and milk.266 Their approach was based on the resistance mechanism against tetracycline in Gram-negative bacterias tetracyclines release Tet repressor protein (TetR) from the tet operator (tetO). Biotinylated single-strain DNA containing the sequence of the tetracycline operator tetO1 was immobilized on a streptavidin-coated sensor chip. The repressor protein TetR was attached to the chip-bound operator tetO. Injection of a solution containing tetracycline allowed tetracycline to bind to TetR. This resulted in the release of a conformationally changed protein, which was continuously flushed away from the sensor surface. The decrease in surface density was measured using the commercial SPR sensor Biacore 3000. The LOD was estimated to be 1 ng/mL for tetracycline in buffer and 15 ng/mL and 25 µg/kg for tetracycline in raw milk and honey, respectively.266 Detection of the antibiotic tylosin was demonstrated by Caldow et al.267 They used the commercially available SPR sensor Biacore Q and inhibition detection format. Immobilization of tylosin on a carboxymethyldextran layer on the sensor surface was performed by amine coupling chemistry. Tylosin was detected in extract from honey at levels down to 2.5 µg/ kg.267

Homola

5.2.4. Vitamins Vitamin B2 (riboflavin) was detected by Caelen et al. using the commercial SPR sensor platform Biacore Q and inhibition detection format.268 A riboflavin derivative was immobilized on the carboxymethylated dextran using amine coupling chemistry. A known concentration of riboflavin binding protein was mixed with a sample, and the amount of unreacted protein was measured using the SPR sensor. Riboflavin was detected in milk-based products, and the LOD was established at 70 ng/mL.268 Haughey et al. reported an SPR sensor for the detection of vitamin B5 (pantothenic acid).269 They used the commercial SPR sensor Biacore Q and inhibition detection format. Vitamin B5 derivative was immobilized in carboxymethylated dextran using amine coupling chemistry. Detection of vitamin B5 was performed in extracts from various foods (e.g., infant formula, cereal, pet food, egg powder). The LOD was 4.4 ng/mL.269 An SPR sensor for the detection of vitamin B8 (biotin) and vitamin B9 (folic acid) was demonstrated by Indyk et al.270 They used the commercial SPR sensor Biacore Q, a biotin sensor chip, and inhibition detection format. In their experiments, they detected biotin and folic acid in infant formulas and milk powders at concentrations as low as 2 ng/mL.270 SPRbased detection of vitamin B12 (cobalamine) was demonstrated by Indyk et al.271 They used the commercial SPR sensor Biacore Q and inhibition detection format. Vitamin B12 was immobilized in carboxymethylated dextran layer on the sensor surface using amine coupling chemistry. The LOD for cobalamine in milk or infant formula or an extract from beef was determined to be 0.06 ng/mL.271

5.2.5. Hormones The steroid hormone progesterone in cow’s milk was detected by Gillis et al.272,273 They used the commercial SPR sensor platform Biacore 2000 and inhibition assay. Progesterone derivative was immobilized in the carboxymethylated dextran layer on the sensor surface using amine coupling chemistry. A known concentration of monoclonal antibody was incubated with sample (buffer or cow’s milk), and the amount of unreacted antibody was detected by the SPR sensor. In their earlier work in 2002, Gillis et al. established the LOD for progesterone in raw bovine milk at 3.6 ng/mL.272 Optimization of the assay reported in 2006 allowed Gillis et al. to detect progesterone in buffer and bovine milk down to 60 pg/mL and 0.6 ng/mL, respectively.273 Mitchell et al. used the commercial SPR sensor Biacore 2000 and inhibition detection format combined with gold nanoparticles and proteins to improve the sensitivity of the detection.212 Progesterone was immobilized to a dextran layer through covalent immobilization using an OEG linker attached to the progesterone. Detection formats employing gold particles conjugated with streptavidin and attached to biotinylated monoclonal antibody in both label prebinding and sequential binding formats were explored. Prelabeling format allowed detection of progesterone down to 143 pg/mL, and sequential binding formats yielded a LOD of 23.1 pg/mL. Secondary antibody labeling produced an 8-fold signal enhancement and a LOD of 20.1 pg/mL, whereas the use of secondary antibody conjugated with a gold nanoparticle improved the LOD to 8.6 pg/mL.212

5.2.6. Diagnostic Antibodies Detection of Mycoplasma hyopneumoniae antibody in pig serum using the SPR method was demonstrated by Kim et

SPR Sensors

al.274 They used the commercial Autolab Esprit SPR system and a recombinant 30 kDa fragment of P97 adhesin as an antigen. The performance of the SPR biosensor was compared with that of enzyme-linked immunosorbent assay (ELISA) using 70 pig serum samples. There was found to be a strong positive correlation between these two methods and, in terms of the LOD, the SPR sensor outperformed ELISA by a factor of 10.274 Classical swine fever virus (CSFV) antibody in pig serum was detected by Cho and Park.275 They used the commercial Autolab Esprit SPR system and recombinant gp55 protein as an antigen. They used the same immobilization method and methodology as in their previous study.274 One hundred and seventy pig serum samples were analyzed by the SPR sensor and ELISA. It was determined that the LOD of the SPR sensor was better by a factor of 10 than that of ELISA.275

5.2.7. Allergens Direct detection of peanut allergens by means of SPR sensor was demonstrated by Mohammed et al.276 They used a miniature commercial SPR sensor Spreeta and peanutspecific antibodies adsorbed on the sensor surface. The LOD for the peanut allergen in buffer was estimated to be 700 ng/mL.276 Food allergens were also detected by Malmheden Yman et al.277 They used the commercial Biacore Q SPR instrument and affinity-purified antibodies raised against egg white, protein conalbumin, sesame seed protein, peanut protein, hazelnut protein (corylin), and crab meat, which were immobilized on the carboxymethylated dextran by the amine coupling method. Detection of allergens was performed directly and by sandwich detection format. The second antibody in the sandwich assay was demonstrated to improve both the sensitivity and specificity of the detection. Peanut proteins in chocolate diluted only 10 times before the analysis were detected down to 1 µg/g. Conalbumin in pasta was detected at levels as low as 0.3 µg/g. Sesame seed protein was detected down to 0.125 µg/mL, corresponding to 12.5 µg/g in solid food (e.g., bread). Tropomyosin in pasta was detected at the level of 10 µg/g.277

5.2.8. Proteins Simultaneous detection of three caseins in milk using an SPR method was demonstrated by Dupont and MullerRenaud.278 They used the commercial SPR sensor platform Biacore 3000 and sandwich assay format employing two monoclonal antibodies directed against the N- and C-terminal extremities of each of the caseins, respectively. Antibody against C-terminal extremities of each of the caseins was immobilized in a separate sensing channel of the four-channel SPR sensor via carboxymethylated dextran matrix and amine coupling. Three major caseins (Rs1, β, and κ) were detected in milk samples. The LODs were estimated at 85 ng/mL for β-casein, 870 ng/mL for Rs1-casein, and 470 ng/mL for κ-casein.278 Indyk et al. detected proteins such as immunoglobulin G, folate-binding protein, lactoferrin, and lactoperoxidase in bovine milk using the commercial SPR biosensor Biacore Q.279 Respective monoclonal antibodies were immobilized on the dextran matrix using amine coupling chemistry. The detection was performed directly in milk samples diluted in buffer. The LODs were established at 16.8 ng/mL for immunoglobulin G, 0.7 ng/mL for folate-binding protein, 1.1 ng/mL for lactoferrin, and 75 ng/mL for lactoperoxidase.279

Chemical Reviews, 2008, Vol. 108, No. 2 481

5.2.9. Chemical Contaminants Detection of 4-nonylphenol in shellfish using an SPR biosensor was demonstrated by Samsonova et al.280 They used the commercial SPR platform Biacore Q and inhibition detection format. 9-(p-Hydroxyphenyl)nonanoic acid was immobilized on dextran matrix using amine coupling chemistry. Using monoclonal antibodies, a detection limit of 2 ng/mL in buffer was achieved. The detection was performed in <3 min including a 30 s regeneration step. The sensor was regenerated by 100 mM sodium hydroxide in 10% acetonitrile. In shellfish samples, 4-nonylphenol was detected at concentrations down 10 ng/g.280 A suspected carcinogen, insulin-like growth factor-1 (IGF-1), can occur in milk of cows treated by recombinant bovine somatotropin treatment. Guidi et al. detected IGF-1 using the commercial Biacore SPR sensor platform and inhibition detection format.281 Recombinant IGF-1 was immobilized to a carboxymethylated dextran matrix via amine coupling chemistry. Polyclonal antibody was incubated with a sample for 2 h, and then the sample was injected in the flow cell of the SPR sensor. On the basis of the reported data, the LOD for IGF-1 in buffer and milk can be estimated to be below 10 ng/mL.281

5.3. Medical Diagnostics Fast, sensitive, and specific detection of molecular biomarkers indicating normal biologic processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention presents an important goal for modern bioanalytics.282,283 SPR biosensors have been demonstrated to hold promise for the detection of analytes related to medical diagnostics such as cancer markers, allergy markers, heart attack markers, antibodies, drugs, and hormones.

5.3.1. Cancer Markers The prostate-specific antigen (PSA) is a marker for prostate cancer.284 Detection of PSA in PBS buffer using the commercial SPR sensor Ibis II has been reported by Besselink et al.285 In their work, monoclonal antibodies against PSA were immobilized on the sensor surface via amine coupling chemistry. After incubation of the sensing surface with sample containing PSA, the sensor response was amplified with rabbit anti-PSA polyclonal antibodies followed with either biotinylated goat anti-rabbit IgG and streptavidin-coated latex microspheres or goat anti-rabbit IgG-coated colloidal gold. The detection format employing gold particle enhancement provided a LOD as low as 0.15 ng/mL. Huang et al. investigated the detection of PSA using direct and sandwich detection formats and the commercial SPR sensor Biacore 2000.286 PSA-receptor molecules consisting of a single-domain antigen-binding fragment were covalently immobilized on the sensor surface via a mixed alkanethiolate SAM. PSA concentrations as low as 10 ng/ mL were detected in buffer. Sandwich detection format involving a biotinylated secondary antibody and streptavidinmodified gold nanoparticles improved the LOD for PSA below 1 ng/mL. Recently, the determination of a complex of PSA with R1-antichymotrypsin (PSA-ACT) in both HBS buffer and human serum was demonstrated by Cao et al.201 using the commercial SPR sensor Biacore 2000. Mixed alkanethiolates were optimized to provide a stable surface for sequential attachment of biotin, streptavidin, and biotinylated antibodies against PSA-ACT. The PSA-ACT complex in HBS buffer and human serum was detected

482 Chemical Reviews, 2008, Vol. 108, No. 2

directly at concentrations as low as 20.7 and 47.5 ng/mL, respectively. The LOD for the PSA-ACT complex was improved by employing sandwich detection format and PSA polyclonal antibody to 10.2 and 18.1 ng/mL for detection in the HBS buffer and serum, respectively.201 The quantitation of a pancreatic cancer marker, carbohydrate antigen (CA 19-9) was performed by Chung et al. using a miniature commercial SPR sensor Spreeta.287 The antibody against CA 19-9 was immobilized on the sensor surface via a SAM of alkanethiolates. The sensor was shown to be able to detect CA 19-9 at a concentration of 410.9 U/mL directly and 66.7 U/mL using a sandwich assay.287 Protein vascular endothelial growth factor (VEGF), which plays a role in breast cancer, lung cancer, and colorectal cancer, was detected using an SPR imaging method and RNA aptamer microarray. The adsorption of proteins onto the RNA microarray was detected by the formation of a surface aptamer-protein-antibody complex. The sensor response was amplified using a localized precipitation reaction catalyzed by the enzyme horseradish peroxidase conjugated to the antibody. The sensor was demonstrated to be able to detect VEGF at a concentration of 1 pM.220 Yang et al.288 measured levels of interleukin-8 (IL-8) protein in the saliva of healthy individuals and patients with oropharyngeal squamous cell carcinoma using the commercial SPR sensor Biacore X. The sandwich detection format using two monoclonal antibodies recognizing different epitopes on the IL-8 was used. A monoclonal antibody against IL-8 was immobilized in the dextran layer via amine coupling chemistry. Saliva samples were first centrifuged to clarify the supernatants. The supernatants were then aspirated and separated from the cellular pellet. The detection limit for IL-8 was determined to be 2.5 pM (∼0.02 ng/mL) for detection in buffer and 184 pM (∼1.5 ng/mL) for detection in saliva samples. Carcinoembryonic antigen (CEA), a marker related to colorectal cancer, was detected by the SPR method of Tang et al.289 They used the commercial SPR sensor Autolab Springle and protein A adsorbed on the SPR sensor surface for subsequent attachment of carcinoembryonic antibody. The LOD of 0.5 ng/mL was achieved, and the sensor was demonstrated to be regenerable for repeated use.289 Tian et al. demonstrated an SPR biosensor for the detection of fibronectin, a glycoprotein implicated in carcinoma development.212 In that work, a research SPR instrument employing an acousto-optic tunable filter and wavelength modulation was used. Fibronectin antibody was immobilized on the self-assembled alkanethiolate monolayer using amine coupling chemistry. Fibronectin was detected directly, in sandwich detection format, and with additional amplification using colloidal gold. Regenerability of the sensor was demonstrated. The LODs for fibronectin in buffer were established at 2.5, 0.5, and 0.25 µg/mL for direct detection, sandwich detection, and colloidal Au-enhanced detection, respectively.212

5.3.2. Antibodies against Viral Pathogens SPR biosensors for the detection of hepatitis virus specific antibodies were reported by several research groups. In 2003 Rojo et al. described the SPR biosensor-based detection of antibodies against hepatitis G in serum.150 They used the commercial SPR sensor Biacore 1000 and synthetic peptide as a biorecognition element, which was immobilized in the dextran layer on the sensor surface via amine coupling

Homola

chemistry. Threshold measurements on sera of chronic hepatitis C patients as well as control samples from healthy patients obtained with the sensor were consistent with those obtained by ELISA. Chung et al. used the commercial SPR sensor Spreeta for the detection of antibodies against human hepatitis B virus (hHBV). The hHBV antigen was immobilized on the SAM of alkanethiolates via amine coupling chemistry. Antibodies against hHBV were detected in 5% serum in PBS. The LOD for direct detection was established at 9.2 nM. Amplification methods based on sandwich detection and avidin-biotinylated antibodies were shown to yield amplification factors of 7 and 14, respectively. Using peroxidase-antiperoxidase complex, 17-fold amplification of the sensor response was obtained and the LOD was lowered to 0.64 nM.202 Wittekindt et al. demonstrated an SPR sensor for the detection of antibodies against herpes simplex virus type 1 and type 2 (HSV-1, HSV-2) in human sera.151 They used the commercial SPR sensor Biacore X and SPR chips coated with streptavidin on which two biotinylated peptides, used as biorecognition elements, were immobilized. Human serum samples (diluted 1:100 in HBS buffer) were tested using the SPR biosensor and immunoblotting (reference method). A good agreement between the SPR biosensor and immunoblotting was obtained (correlation of 83 and 86% for antibodies against HSV-1 and HSV-2, respectively).151 Direct detection of antibodies against Epstein-Barr virus (anti-EBNA) in 1% human serum was reported by Homola’s group.152 A short synthetic peptide was used as biorecognition element and was immobilized on the surface of a wavelengthmodulated SPR biosensor via hydrophobic and electrostatic interactions. The LOD for anti-EBNA was determined to be 0.2 ng/mL. A procedure for the regeneration of the sensor was developed and demonstrated to allow at least 10 repeated anti-EBNA detection experiments without a significant loss in sensor sensitivity.152 Regnault et al. reported the SPR biosensor-based detection of anti-protein S antibodies following Varicella-Zolter virus infection.290 In that work, protein S was immobilized in the layer of dextran via amine coupling chemistry and the experiments were performed using the commercial SPR sensor Biacore X. A high sensor response was observed to diluted plasma (1:5) of an infected patient, whereas samples from healthy patients generated a minimum response.290 Antibodies against human respiratory syncytial virus (RSV) in sera were detected by the commercial SPR biosensor Biacore 2000 by McGill et al.291 Monoclonal antibodies against the virus glycoproteins (F- and G-glycoproteins) were covalently attached to the dextran matrix via amine coupling chemistry and then used to immobilize the respective virus glycoproteins. Serum samples isolated from patients’ respiratory tracts were diluted in HBS buffer (1:10) and filtered. The SPR biosensor was demonstrated to be able to recognize the antigenic differences between the two different genotypes of the virus (G- and F-virus glycoproteins).291 Abad et al. demonstrated SPR-based detection of isotypespecific anti-adenoviral antibodies in patients dosed with an adenoviral-based gene therapy vector. The antibodies were detected using the commercial SPR instrument Biacore 3000 and intact virus immobilized in the thin layer of dextran on the surface of the sensor by amine coupling chemistry.292 Patient serum samples or ascites fluid samples were diluted 1:10 with HEPES prior to the analysis.

SPR Sensors

5.3.3. Drugs and Drug-Induced Antibodies Dillon et al. demonstrated the SPR biosensor-based detection of morphine-3-glucuronide (M3G), the main metabolite of heroin and morphine.293 They used the commercial SPR sensor Biacore 1000 and inhibition detection format. M3G-ovalbumin conjugate was immobilized on the dextran matrix via amine coupling chemistry. Two polyclonal antibodies were produced, purified, and tested for the detection of M3G. Regeneration protocols were developed for both polyclonal antibodies and allowed for approximately 60 cycles for the first antibody and 50 cycles for the second antibody. The LOD for M3G in buffer and in urine (diluted 1:250) was found to be <1 ng/mL for both antibodies.293 Detection of the oral anticoagulant warfarin by the SPR method was performed by Fitzpatrick and O’Kennedy.294 They used the commercial SPR sensor Biacore 3000 and inhibition detection format. 4′-Aminowarfarin or 4′-azowarfarin-BSA was immobilized on a dextran matrix via amine coupling chemistry. Detection of warfarin was performed in plasma ultrafiltrate (diluted 1:100). The sensor was demonstrated to detect warfarin at concentrations down to 4 ng/mL and to be regenerable for more than 70 detection cycles.294 Gobi et al. reported SPR sensor-based detection of insulin.203 In that work the commercial SPR sensor SPR670 and inhibition format were employed. Insulin was covalently bound to the activated monolayer of heterobifunctional OEG-dithiocarboxylic acid derivative. After 5 min of incubation of sample with a known concentration of monoclonal anti-insulin antibody, the mixture was injected in the SPR sensor and the concentration of the unreacted antibody was measured. A regeneration protocol was developed, and the chip was shown to be reusable for more than 25 detection cycles without an appreciable change in the sensor activity. The LOD for insulin in buffer was 1 ng/mL. The lowest detectable concentration of insulin in the serum samples spiked with insulin was 6 ng/mL.203 Rini et al. reported an SPR sensor for threshold measurement of antibodies against granulocyte macrophage colony stimulating factor (GM-CSF) used in therapies for various kinds of cancer.295 Rini et al. used the commercial SPR instrument Biacore 2000 and SPR chips with GM-CSF antigen immobilized on the carboxymethylated dextran via amine coupling chemistry. Antibodies against GM-CSF were induced in prostate cancer patients by repeated administration of GM-CSF, and their presence was measured in diluted sera (1:5). The SPR measurements showed the presence of GMCSF reactive antibodies for all prostate cancer patients treated with GM-CSF, which was in agreement with reference ELISA measurements.295 An SPR biosensor for the detection of antibody against insulin was demonstrated by Kure at al.296 Insulin antibodies can cause insulin resistance or hypoglycemia in diabetic patients treated with human insulin. In that work the commercial SPR biosensor Biacore 2000 was employed and purified human insulin, as a biorecognition element, was immobilized on the sensor surface via amine coupling chemistry. In calibration experiment in buffer, monoclonal antibodies against insulin were detected at concentrations as low as 0.6 µg/mL. Serum samples were pretreated to remove insulin and filtered before SPR measurements. Insulin antibodies were detected in eight diabetic patient serum samples and determined to fall in the range of 2.91-16.3

Chemical Reviews, 2008, Vol. 108, No. 2 483

µg/mL. No insulin antibodies were detected in the control group.296

5.3.4. Hormones A marker of pregnancyshuman chorionic gonadotropin hormone (hCG)shas been a frequent target of optical biosensor technologies. Jiang’s group reported SPR sensorbased detection of hCG exploiting a wavelength-modulated SPR sensor and DNA-directed antibody immobilization method.165 The immobilization consisted of non-covalent attachment of streptavidin to a biotinylated SAM of alkanethiolates followed with the binding of biotinylated oligonucleotides to available streptavidin binding sites. Antibodies chemically modified with oligonucleotides with a complementary sequence were finally attached to this surface via DNA hybridization. The detection limit for direct detection of hCG in buffer was determined to be 0.5 ng/ mL. Recently, detection of hCG in urine was performed using a sequential detection method developed by Chung et al.208 They used the commercial SPR sensor Spreeta and immobilized two molecular recognition elements (antibody against HCG and antibody against human albumin) in a single sensing channel of Spreeta sensor using SAM of alkanethiolates and amine coupling chemistry. Amplification polyclonal anti-hCG antibodies were used to increase the sensor response. In 10-fold diluted urine, the detection limit for hCG was established to be 46 mIU/mL.208 SPR-based detection of 17β-estradiol was demonstrated by Miyashita et al.297 They used the commercial SPR sensor Biacore X and inhibition detection format. Estradiol-BSA conjugate was immobilized on carboxymethylated dextran layer by amine coupling chemistry. The binding of unreacted antibody to 17β-estradiol conjugates at the surface of the sensor was measured. The 17β-estradiol was detected down to 0.47 nM (∼0.14 ng/mL).297 Teramura and Iwata demonstrated an SPR sensor for the detection of R-fetoprotein (AFP) in human plasma.298 They used a research SPR sensor system with angular modulation and a SAM of tri(ethylene glycol), and carboxyl groupterminated hexa(ethylene glycol) was employed for covalent attachment of monoclonal AFP-antibody. Detection was performed in sandwich detection format. The SPR signal shift was further enhanced by applying a polyclonal antibody against the second antibody. The polyclonal antibody against the second antibody was demonstrated to amplify the sensor response to AFP by a factor of 7. The LOD for AFP in blood plasma was estimated to be in nanograms per milliliter levels.298

5.3.5. Allergy Markers Measurement of immunoglobulin E (IgE) antibody levels plays an important role in the diagnostics of allergies. Imato’s group reported direct detection of IgE antibody using an SPR biosensor.199 In that work, anti-IgE antibody was immobilized on the surface of the commercial SPR sensor SPR 20 by physical adsorption. A sample containing IgE antibody was mixed with an anti-IgE(H) antibody solution to form an antiIgE(H) complex via the Ce2 domain of the IgE antibody. The solution was introduced in the SPR sensor and the immunocomplex of the IgE-anti-IgE(H) reacted with the anti-IgE(D) antibody immobilized on the sensor chip via the Ce3 domain of the IgE antibody. The LOD for the IgE antibody was about 10 ppb.199

484 Chemical Reviews, 2008, Vol. 108, No. 2

Homola

The same group reported SPR-based detection of histamine (β-imidazole ethylamine)sa protein involved in allergic reactions. Their SPR sensor was based on the commercial SPR-20 sensor and inhibition detection format. Histamine was immobilized on the sensor surface using a self-assembled alkanethiolate monolayer and amine coupling chemistry. A regeneration protocol was developed, and it was demonstrated that the sensor can be used for more than 10 detection/ regeneration cycles. The limit of detection was 3 ppb.299

Hwang et al. reported an SPR sensor for the detection of hepatitis B surface antigen (HBsAg), an early indicator of hepatitis B.307 In their work, they used the commercial SPR sensor Biacore 3000 and immobilized anti-HBsAg polyclonal antibody to the dextran layer on a sensor chip surface using amine coupling chemistry. HBsAg was detected at concentrations as low as ∼1 µg/mL.307

5.3.6. Heart Attack Markers

Analytes of environmental concern225,308 targeted by SPR biosensors include, in particular, pesticides, aromatic hydrocarbons, heavy metals, phenol, polychlorinated biphenyls, and dioxins.

Detection of a marker of cardiac muscle injury, troponin (cTn I), in serum was demonstrated by Wei et al.300 In that work, biotinylated antibodies against cTn I were immobilized on the avidin layer created using amine coupling chemistry on an activated SAM of alkanethiolates. cTn I was detected directly and in sandwich detection format. The LODs were determined to be 2.5 and 0.25 ng/mL for direct and sandwich detection format, respectively.300 Detection of cTn I was also demonstrated by Booksh’s group.301 They used a miniature fiber optic SPR sensor on which human anti-cardiac troponin I was immobilized via a dextran layer and amine coupling chemistry. The LOD for cTnI in buffer was established at 3 ng/mL.301

5.3.7. Other Molecular Biomarkers Detection of antibodies against glucose 6-phosphate isomerase (GPI) in synovial fluids of rheumatoid arthritis and osteoarthritis patients (diluted 1:100 in Hepes) using a Biacore 2000 is presented in the work of Kim et al.302 Recombinant human GPI proteins produced from E. coli were immobilized on the dextran sensor surface via amine coupling chemistry. The synovial fluid samples from rheumatoid arthritis patients showed a significantly higher level of binding to the recombinant GPI proteins than samples from osteoarthritis patients. SPR-based detection of anti-glutamic acid decarboxylase (GAD) antibodies for diagnosing type I diabetes mellitus was reported by Sim’s group.303,304 They used the commercial SPR sensor Biacore 2000 and biotinylated GAD coupled to streptavidin molecules anchored covalently on a mixed SAM of hydroxyl- and carboxyl-terminated alkanethiolates. Optimization of SAM composition was carried out. The SPR sensor employing an optimized SAM was shown to be able to detect antibody in HBS-EP buffer in sub-micromolar levels.304 Detection of c-reactive protein (CRP), a human blood serum marker for inflammatory processes, using SPR biosensor technology was demonstrated by Meyer.305 They used the commercial SPR sensor Plasmonic and sandwich detection format. Biotinylated monoclonal antiCRP antibody C6 was immobilized on the biotin-coated sensor surface via streptavidin. Buffer spiked with CRP was injected in the SPR sensor cuvette and then replaced by a solution containing the secondary antiCRP antibody C2. The sandwich assay was completed typically within 30-60 min. The LOD for CRP in buffer was determined to be 1 µg/mL.305 SPR-based detection of the cystatin C marker of the glomerular filtration rate (GFR), a critical measure of normal kidney function, was demonstrated by Corn’s group.306 They used an SPR imaging instrument with antibodies immobilized on the alkanethiolate-modified sensor surface using carbonyldiimidazole surface reaction. The sensor was shown to be able to detect cystatin C at 1 nM levels.306

5.4. Environmental Monitoring

5.4.1. Pesticides Following the demonstration of SPR biosensors for the detection of atrazine309 and simazine137,138 in the 1990s, various pesticides have been targeted by SPR sensor technology. Lechuga’s group reported a portable SPR sensor for the detection of atrazine in water using inhibition format.310 An atrazine derivative was immobilized on the alkanethiolate SAM formed on the gold-coated sensor surface. The sample was incubated with polyclonal antibodies, and then the mixture was analyzed by the SPR sensor. The LOD was established at 20 pg/mL. A measurement/regeneration cycle required about 25 min.310 An alternative approach to the detection of atrazine based on specifically expressed mRNA in Saccharomyces cereVisiae bacteria exposed to atrazine was reported by Lim et al.311 The cells were brought into contact with the analyzed sample and disrupted, and the amount of expressed P450 mRNA was measured by an SPR biosensor with complementary oligonucleotide probes immobilized on the sensor surface using streptavidin-biotin chemistry. The LOD (1 pg/mL) presents a substantial improvement compared to previous works.18,309 Lechuga’s group demonstrated SPR biosensors for the detection of organophosphate pesticide chlorpyrifos and carbaryl.312-314 They used inhibition detection format in which a pesticide derivative with BSA was covalently immobilized on a self-assembled alkanethiolate monolayer formed on the SPR sensor surface. Typical limits of detection were around 50 pg/mL for chlorpyrifos312,314 and 1 ng/mL for carbaryl.312,313 A protocol for regeneration of the sensing surface was developed, and the sensor was able to perform ∼200 detection cycles without degradation in performance.312 The detection cycle was completed in 20 min. The sensors were tested in ground, river, and drinking water samples without the observation of significant matrix effects. Lechuga’s group also demonstrated an SPR sensor for the detection of dichlorodiphenyltrichloroethane (DDT).315 They used a portable SPR sensor now commercially available from Sensia (Spain) and inhibition detection format. DDT derivative was immobilized on a self-assembled alkanethiolate monolayer on the SPR sensor surface. Two monoclonal antibodies, specific to DDT and specific to DDT and its metabolites, were used in the inhibition detection format. Regeneration of the sensor surface was developed, and 270 detection cycles were performed. The LOD in distilled water was established at 15 pg/mL for the DDT-specific assay and at 31 pg/mL for the DDT group-selective assay. Nearly the same performance was achieved when the sensor was used to analyze river water samples.315 Detection of three pesticidessDDT, carbaryl, and chlorpyrifossusing a single

SPR Sensors

channel of an SPR sensor and inhibition detection format was demonstrated by Lechuga’s group.209 DDT, carbaryl, and chlorpyrifos derivatives were attached to carboxylic terminal groups on a SAM of alkanethiolates. A sample was mixed with antibody against one target and injected in the flow cell of the sensor. After the sensor response had been established, a sample mixed with antibody against another target was injected with or without previous regeneration of the sensing surface. The LODs for this multianalyte detection approach were established at 18 pg/mL for DDT, 50 pg/mL for carbaryl, and 52 pg/mL for chlorpyrifos. These detection limits were comparable with those obtained using singleanalyte functionalizations.209 Miura’s group applied the SPR method to the detection of 2,4-dichlorophenoxyacetic acid (2,4-D).316,317 Initially, they used inhibition detection format and a conjugate of 2,4-D derivative and BSA (2,4-D-BSA) immobilized on the sensor by physisorption. The LOD was established at 0.5 ng/mL.316 In their later study, they used a SAM of alkanethiolates for covalent attachment of 2,4-D-BSA conjugate on the sensor surface. This functionalization approach led to an improvement in the LOD to 10 ppt.317 Regeneration of the sensor for up to 30 detection cycles was demonstrated using pepsin.317

5.4.2. 2,4,6-Trinitrotoluene (TNT) SPR biosensors for the detection of TNT, which is a prime constituent of most of landmines and also exhibits toxic, mutagenic, and carcinogenic effects, have been extensively researched by Miura’s group. They used inhibition detection format involving various conjugates and antibodies. The conjugates used in their experiments included 2,4,6-trinitrophenol-BSA (TNP-BSA).318,319 TNP-ovalbumin (TNPOVA),318 2,4,6-trinitrophenyl-keyhole limpet hemocyanin (TNP-KLH),320 and TNP-β-Alanine-ovalbumin (TNPβ-Ala-OVA).321 The antibodies (monoclonal and polyclonal) were homemade321 or commercial. Most of their sensors delivered a LOD below 10 pg/mL (10 ppt);318,320-322 the best LOD (1 ppt) was achieved using the immunoreaction between a homemade polyclonal anti-2,4,6-trinitrophenylkeyhole limpet hemocyanine antibody with a physically immobilized TNP-OVA.318 The sensor surface was regenerated by pepsin. The detection cycle was typically completed in <20 min. The inhibition-based SPR biosensors of TNT were found to exhibit low cross-sensitivity to other similar compounds such as 1,4-DNT, 1,3-DNB, 2A-4,6-DNT, and 4A-4,6-DNT.323 Larsson et al. studied the effect of composition of molecular coatings on the performance of the SPR biosensor for the detection of TNT.324 They used the commercial SPR platform Biacore2000 and inhibition detection format. Two types of thiols, OEG-alkylthiols terminated with a hydroxyl group and a TNT analogue (2,4-dinitrobenzene), were self-assembled on the surface of an SPR chip. The ratio of TNT analogues and hydroxyl-terminated OEGthiols was optimized to provide highly selective and sensitive biochips with minimum nonspecific binding. The LOD for TNT in buffer was demonstrated to be <10 pg/mL.324

5.4.3. Aromatic Hydrocarbons An SPR biosensor for the detection of 2-hydroxybiphenyl (HBP, a metabolite of BaP) was demonstrated by Miura’s group. The sensor was based on the inhibition immunoassay format. An antibody against HBP was mixed with a sample and, after incubation, the unreacted antibody was detected

Chemical Reviews, 2008, Vol. 108, No. 2 485

using an SPR biosensor functionalized with an HBP conjugate. HBP-BSA was immobilized on the surface of the SPR sensor by physical adsorption.325,326 The LOD using the commercial SPR biosensor SPR-20 was demonstrated to be 0.1 ppb (ng/mL). Most recently, this detection approach was combined with a portable SPR sensor, and the LOD of 0.1 ppb (ng/mL) was reproduced also on this sensor platform. A simple regeneration procedure using a pepsin solution allowed more than 30 measurement cycles without appreciable deterioration of sensor response. The measurement cycle was completed in 20 min. The same group also demonstrated detection of benzo[a]pyrene (BaP). They used the commercial SPR biosensor SPR-20 and inhibition assay. BAP molecules were attached to the sensor surface either by immobilizing BaP-BSA conjugate on the sensor via physical adsorption327,328 or by immobilizing BaP analogue on the sensor surface with a mixed SAM of alkanethiolates.329 The biomolecular coating incorporating a BaP analogue was determined to yield a more sensitive sensor with a LOD for BaP in buffer as low as 0.05 ppb.329 In 2005, Kawazumi et al. reported the simultaneous detection of benzo[a]pyrene and 2-hydroxybiphenyl using a compact, portable SPR instrument.52 They employed inhibition detection format and BSA-BaP and BSA-HBP conjugates immobilized on the surface of a dual-channel SPR sensor via physical adsorption. The sensor was shown to be able to detect BaP and HBP in buffer down to parts per billion levels.52

5.4.4. Heavy Metals An SPR sensor for direct detection of Cu2+ ions was reported by Ock et al.330 Their sensor employed a thin polymer layer containing squarylium dye (SQ), which changes its refractive index absorption properties when interacting with Cu2+ ions. Owing to anomalous dispersion accompanied with this absorption, a substantial refractive index change can be observed when SQ dye is exposed to Cu2+ ions. The sensor responded to Cu2+ in buffer at levels as low as 1 × 10-12 M.330 Another SPR sensor for the detection of heavy metals was demonstrated by Wu et al., who used a Biacore X instrument with rabbit metallothinein coupled to the dextran matrix on the sensor surface.286 Metallothinein is a protein that can be found in cells of many organisms and is known to bind to metals (especially cadmium and zinc). Model experiments in which metallothein was used as a receptor demonstrated the potential of this sensor to directly detect Cd, Zn, and Ni in buffer at concentrations down to 100 ng/mL. Forzani et al. demonstrated another approach to the direct detection of Cu2+ and Ni2+ ions. In their work they used a differential SPR sensor coated by properly selected peptides specifically binding metal ions. Detection limits for Cu2+ and Ni2+ in deionized water were 32 and 178 pM, respectively.153

5.4.5. Phenols An SPR biosensor for the detection of bisphenol A (BPA) was developed by Soh et al.332 They used the commercial SPR-20 sensor and inhibition detection format. The sensor surface was modified with a thiol monolayer on which BPA molecules were immobilized through BPA succinimidyl ester. Using a monoclonal antibody, detection of BPA in buffer at concentrations as low as 10 ng/mL was achieved.332 Detection time was approximately 30 min, and the sensor was demonstrated to be regenerable using 0.01 M hydrochloric acid. Another SPR biosensor for the detection of BPA

486 Chemical Reviews, 2008, Vol. 108, No. 2

Homola

Table 1. Overview of SPR Biosensors for Food Quality and Safety Analysis analyte

limit of detection

sensor system

detection matrix

Multiskop Multiskop custom-built

buffer buffer detergent-lysed bacteria heat-killed bacteria untreated bacteria BPV solution aqueous solution PBST solution heat-killed bacteria in buffer apple juice, pH 7.4 pasteurized milk milk; apple juice; ground beef

104 cells/mL 102 cells/mL 104 cfu/mL 105 cfu/mL 106 cfu/mL 8.7 × 106 cfu/mL 106 cells/mL 103 cfu/mL 1.4 × 104 cfu/mL 105 cfu/mL 25 cfu/mL 102-103 cfu/mL

Biacore 2000

heat-killed bacteria in buffer HBS-EP solution buffer buffer milk buffer apple juice, pH 7.4 pasteurized milk

106 cfu /mL 1.7 × 105 cfu/mL 102 cfu/mL 102 cfu/mL 105 cells/mL 4.4 × 104 cfu/mL 104 cfu/mL 23 cfu/mL

custom-built Biacore 3000 custom-built

heat-killed bacteria in buffer PBS solution apple juice, pH 7.4; buffer

custom-built

SR 7000

detection format

ref

Escherichia coli O157:H7

Spreeta Spreeta SR 7000 custom-built Biacore 2000 Spreeta Salmonella spp. S. enteritidis S. group A, B, D, E S. typhimurium S. paratyphi S. choleraesuis S. enterica Lysteria monocytogenes

custom-built Biacore 3000 Multiskop Multiskop Plasmonic custom-built

direct direct sandwich

228 229 230

direct direct sandwich sandwich

231 232 233 234

direct direct

235 236

direct sandwich direct direct sandwich sandwich

162 237 238 239 240 234

direct

235

107 cfu /mL 105 cells/mL 3 × 103 cfu/mL

direct competitive sandwich

162 241 234

heat-killed bacteria in buffer apple juice, pH 7.4

1 × 105 cfu/mL 5 × 104 cfu/mL

sandwich

234

PBST soluiton

105 cfu/mL 107 cfu/mL 104 cfu /mL

sandwich direct direct

242

104-5 × 104 cfu/mL

sandwich

234

102 cfu/mL

direct

244

4 × 105 cfu/mL

direct

245

106 oocyst/mL 102 oocyst/mL

direct

246 246

Campylobacter jejuni

Staphylococcus aureus

Spreeta buffer mixture of E. coli, C. jejuni, S. typhimurium, and L. monocytogenes cutom-built heat-killed bacteria in buffer; apple juice Yersinia enterocolitica Multiskop buffer Vibrio cholerae O1 Multiskop buffer Cryptosporidium parVum Biacore 2000 buffer

243

Fusarium culmorum

Staphylococcal enterotoxins SEB

Biacore X

wheat

0.25 ng/µL

direct, PCR amplicon

247

Biacore X fiber optic Spreeta

1 ng/mL 10 ng/mL 5.6 ng/mL 28 ng/mL 0.6 ng/mL 1.4 ng/mL 5 ng/mL 0.5 ng/mL 0.3 ng/mL 2.5 ng/mL 1 ng/mL

direct direct direct direct sandwich sandwich direct sandwich inhibition sandwich competitive

248, 249 250 251

Biacore 1000 Biacore 1000 Biacore 1000

milk buffer buffer seawater buffer seawater buffer; milk buffer; milk milk buffer; ham tissue extract raw eggs

Biacore 3000 custom-built Biacore Q Spreeta

buffer buffer shellfish species extract buffer; clam extracts

5 ng/mL 0.1 ng/mL 1-7 µg/g 3 ng/mL

competitve inhibition inhibiton inhibition

255 256 257 258

Biacore 1000 Biacore 3000

buffer buffer

3 ng/mL 190 pg/mL

inhibition inhibition

259 148

Biacore Q

buffer

2.5 ng/mL

inhibition

260

custom-built

SEA domoic acid

91 252 253 254

aflatoxin B1

deoxynivalenol

SPR Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 487

Table 1 (Continued) analyte antibiotics benzylpenicillin, ampicillin, amoxicillin Row70cloxacillin cephalexin cefoperazon chloramphenicol, chloramphenicol glucuronide

sensor system Biacore 3000

Biacore Q

tylosin vitamins B2 (riboflavin) B5 (pantothenic acid) B8 (biotin), B9 (folic acid) B12 (cobalamine) hormones progesterone diagnostic antibodies Mycoplasma hyopneumoniae classical swine fever virus allergens peanut allergens peanut proteins conalbumin seasame seed protein tropomyosin proteins β-casein Rs1-casein κ-casein IgG folate-binding protein lactoferrin lactoperoxidase chemical contaminants 4-nonylphenol IGF-1

raw milk

honey extract; prawn; dairy products; porcine kidney poultry honey prawn milk shrimps

limit of detection

detection format

ref

2 ng/mL

inhibition

262

15 ng/mL 50 ng/mL 25 ng/mL 0.2 µg/kg

direct

263

inhibition

264

inhibition

265

indirect

266

inhibition

267 268

Biacore Q

buffer raw milk honey honey extract

0.005 µg/kg 0.02 µg/kg 0.04 µg/kg 0.04 µg/kg 1 ng/mL 0.2 ng/mL 250 ng/mL 0.5 ng/mL 1 ng/mL 15 ng/mL 25 µg/kg 2.5 µg/kg

Biacore Q Biacore Q Biacore Q Biacore Q

milk-based products various foods milk powder; infant formulas milk; infant formulas

70 ng/mL 4.4 ng/mL 2 ng/mL 0.06 ng/mL

inhibition inhibition inhibition inhibition

269 270 271

Biacore 2000

60 pg/mL 0.6 ng/mL 8.6 pg/mL

inhibition

273

Biacore 2000

buffer milk buffer

imhibition, amplification

212

Autolab Esprit SPR Autolab Esprit SPR

pig serum pig serum

direct direct

274 275

Spreeta Biacore Q

buffer chocolate pasta bread pasta

700 ng/mL 1 µg/g 0.3 µg/g 12.5 µg/g 10 µg/g

direct direct sandwich

276 277

Biacore 3000

milk

sandwich

278

Biacore Q

milk

85 ng/mL 870 ng/mL 470 mg/mL 16.8 ng/mL 0.7 ng/mL 1.1 ng/mL 75 ng/mL

direct

279

Biacore Q

buffer shellfish buffer, milk

2 ng/mL 10 ng/g 10 ng/mL

inhibition

280

inhibition

281

Biacore Q

chloramphenicol florephenicol florephenicol amine thiamphenicol tetracycline

detection matrix

Biacore Q

Biacore 3000

Biacore

was demonstrated by Matsumoto’s group.334 Their sensor also used inhibition detection format and BPA-OVA conjugate immobilized on the sensor surface by physical adsorption. The sensor was shown to be able to detect BAP at 1 ng/mL (1 ppb) levels.334 Imato’s group developed an SPR biosensor for the detection of 2,4-dichlorophenol based on competitive detection format.333 They used the commercial SPR sensor SPR20 functionalized with monoclonal antibodies against 2,4dichlorophenol immobilized on the sensor surface via gold binding peptide and protein G. Detection was based on the competition between the analyte present in sample and added BSA-2,4-dichlorophenol conjugate. The LOD was established at 20 ng/mL.333

5.4.6. Polychlorinated Biphenyls Karube’s group demonstrated the detection of PCB 3,3′,4,4′,5-pentachlorobiphenyl using the commercial SPR sensor Biacore 2000 and competitive detection format.333 In

their work, the sample was mixed with a conjugate of PCBHRP and flowed across the sensor surface with polyclonal antibodies immobilized in the dextran matrix. The presence of PCB was detected as a decrease in the binding of PCBHRP conjugate. The LOD was determined to be 2.5 ng/mL. The sensor surface was regenerable using 0.1 M hydrochloric acid. The detection was completed in 15 min.335

5.4.7. Dioxins Karube’s group also demonstrated an SPR biosensor for the detection of 2,3,7,8-TCDD.335 They used the commercial SPR sensor Biacore 2000 and competitive detection format. Monoclonal antibody was immobilized in the dextran layer on the sensor chip by amine coupling chemistry. The sample was mixed with a of 2,3,7,8-TCDD-HRP conjugate and injected into the sensor. A LOD of 0.1 ng/mL was attained, and the sensor was shown to be regenerable using 0.1 M hydrochloric acid.335 Detection was completed in 15 min.

488 Chemical Reviews, 2008, Vol. 108, No. 2

Homola

Table 2. Overview of SPR Biosensors for Medical Diagnostic analyte cancer markers prostate-specific antigen (PSA)

sensor system Ibis II Biacore 2000

3% BSA in PBS buffer

PSA-ACT complex (R1-antichymotrypsin)

Biacore 2000

carbohydrate antigen (CA 19-9) vascular endothelial growth factor (VEGF) interleukin-8 (IL-8)

Spreeta

carcinoembryonic antigen (CEA) fibronectin

Autolab Springle

antibodies against viral pathogens hepatitis virus specific

anti-hepatits simplex virus anti-EBNA anti-protein S anti-RSV anti-adenoviral drugs and drug-induced antibodies morphine-3-glucuronide warfarin insulin anti-GM-CSF anti-insulin hormones hCG 17β-estradiol R-fetoprotein allergy markers IgE histamine heart attack markers troponin (cTn I) other molecular biomarkers Glc-6-P isomerase anti-GAD c-reactive protein cystatin C HBsAg

limit of detection

detection matrix

detection format

ref

HBS buffer HBS buffer human serum human serum buffer buffer buffer

0.15 ng/mL 10 ng/mL 1 ng/mL 20.7 ng/mL 10.2 ng/mL 47.5 ng/mL 18.1 ng/L 410.9 U/mL 66.7 U/mL 1 pM

sandwich direct sandwich direct sandwich direct sandwich direct sandwich sandwich

buffer human saliva buffer

2.5 pM (∼0.02 ng/mL) 184 pM (∼1.5 ng/mL) 0.5 ng/mL

sandwich

288

direct

289

custom-built

buffer buffer buffer

2.5 µg/mL 0.5 µg/mL 0.25 µg/mL

direct sandwich amplification

212

Biacore 1000 Spreeta

human serum 5% serum in PBS

150 202

Biacore X custom-built Biacore X Biacore 2000 Biacore 3000

serum in HBS (1:100) 1% human serum plasma 1:5 HBS Buffer ascites fluid in Hepes (1:10)

direct direct sandwich amplified direct direct direct direct direct

Biacore 1000 Biacore 3000 SPR-670

urine in buffer (1:250) plasma (1:100) buffer serum serum (1:5) buffer serum

1 ng/mL 4 ng/mL 1 ng/mL 6 ng/mL 0.6 µg/mL 2.91 µg/mL

inhibition inhibition inhibition inhibition direct direct direct

custom-built Spreeta Biacore X custom-built

buffer urine HBS-EP buffer plasma

0.5 ng/mL 46 mIU/mL 0.47 nM (∼0.14 ng/mL) ng/mL

direct sandwich inhibition sandwich

165 208 297 298

SPR 20 SPR 20

buffer buffer

10 ppb 3 ppb

direct inhibition

199 299

2.5 ng/mL 0.25 ng/mL 3 ng/mL

direct sandwich direct

300

custom fiber-optic

serum serum buffer

direct

302

direct sandwich direct direct

303, 304 305 306 307

cutom-built SPRI Biacore X

Biacore 2000 Biacore 2000

Biacore 2000 Biacore 2000 Plasmonic custom-built SPRI Biacore 3000

synovial fluids in Hepes (1:100) HBS-EP buffer buffer buffer buffer

5.5. Summary Over the past 5 years, more than 100 SPR biosensors for the detection of a variety of chemical and biological analytes were demonstrated. Most of these biosensors are based on prism coupling and angular or wavelength spectroscopy of surface plasmons. Commercial SPR systems have played an important role in the development of detection applications due to their increasing spread and the availability of special SPR platforms and kits dedicated to specific applications (e.g., Biacore Q for food analysis). Data collected in Tables 1-3 illustrate recent applications of SPR biosensors and achieved levels of performance. The performance figures should be compared with caution as performance of an SPR

9.2 nM 4.4 nM 0.64 nM 0.2 ng/mL

<µM 1 µg/mL 1 nM 1 µg/mL

285 286 201

287 220

151 152 290 291 292 293 294 203 295 296

301

biosensor is a result of a multitude of factors (performance of optical platform, characteristics of the employed biorecognition element, suitability and degree of optimization of the immobilization method, detection format, and methodology), and thus low performance of one part of the biosensor (e.g., optical platform) can be compensated for by high performance of another component (e.g., biorecognition elements). Clearly, analytes implicated in food safety have received the most attention (Table 1). Bacterial pathogens such as E. coli and Salmonella were the most frequently targeted analytes. Detection limits below 102 bacteria/mL were reported. A great deal of research has been devoted to the

SPR Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 489

Table 3. Overview of SPR Biosensors for Environmental Monitoring analyte pesticides atrazine chlorpyrifos carbaryl DDT DDT carbaryl chloropyrifos 2,4-dichlorophenoxyacetic acid

sensor system

detection matrix

Sensia

water

custom-built custom-built Sensia

water water water water

SPR-20 custom-built

limit of detection

detection format inhibition

buffer buffer

20 pg/mL 1 pg/mL 50 pg/mL 1 ng/mL 15 pg/mL 18 pg/mL 50 pg/mL 52 pg/mL 0.5 ng/mL 10 ppt

Biacore 2000 Biacore 2000

buffer buffer

SPR-20 SPR-20 SPR-20

ref

inhibition inhibition inhibition inhibition

310 311 312, 314 312, 313 315 209

inhibition inhibition

316 317

10 pg/mL 10 pg/mL

inhibition inhibition

318, 320-322 324

buffer buffer buffer

0.1 ng/mL 0.01 ng/mL 0.05 ng/mL

inhibition inhibition inhibition

325, 326 327, 328 329

custom-built Biacore X custom-built

buffer buffer deionized water

∼pM ∼µM 32 pM 178 pM

direct direct direct

330 331 153

SPR-20 Spreeta SPR-20 Biacore 2000 Biacore 2000

buffer buffer buffer buffer buffer

10 ng/mL 1 ng/mL 20 ng/mL 2.5 ng/mL 0.1 ng/mL

inhibition inhibition competitive competitive competitive

332 334 333 335 335

2,4,6-trinitrotoluene (TNT) aromatic hydrocarbons 2-hydroxybiphenyl benzo[a]pyrene heavy metals Cu2+ ions Cd, Zn, Ni Cu2+ Ni2+ phenols bisphenol A 2,4-dichlorphenol PCB 2,3,7,8-TCDD

development of SPR biosensors for other significant groups of analytes such as Staphylococcal enterotoxins (best demonstrated LODs < 1 ng/mL) and antibiotics (best LODs < 1-10 ng/mL depending on the substance). Several analytes have been detected also in complex food matrices. In the field of medical diagnostics (Table 2), the most attention has been paid to the development of SPR sensors for the detection of cancer markers (best LODs < 1-100 ng/mL) and antibodies (best LODs < 1-100 ng/mL). However, most of the detection experiments were performed in buffers rather than in clinical samples. The development of SPR biosensors for environmental monitoring (Table 3) has focused mainly on the detection of pesticides. The best LODs ranged from 1 to 100 pg/mL, depending on the analyte. Detection experiments were performed in buffers or real-world water samples.

6. Conclusions In the past 5 years, SPR biosensor technology has made substantial advances in terms of both sensor hardware and biospecific coatings. SPR biosensors have been applied for the detection of a variety of chemical and biological analytes. We envision that the performance of SPR biosensor technology will continue to evolve and that advanced SPR sensor platforms combined with novel biospecific surfaces with high resistance to the nonspecific binding will lead to robust SPR biosensors enabling rapid, sensitive, and specific detection of chemical and biological analytes in complex samples in the field. These biosensors will benefit numerous important sectors such as medical diagnostics, environmental monitoring, and food safety and security.

7. Abbreviations ATR, attenuated total reflection; CCD, charge-coupled device; DNA, deoxyribonucleic acid; ELISA , enzyme-linked

immunosorbent assay; LED, light-emitting diode; LOD, limit of detection; MALDI-TOF, matrix-assisted laser desorption/ ionization time-of-flight mass spectrometry; RIU, refractive index unit; RNA, ribonucleic acid; scFvs, single-chain antibody fragment; SAM, self-assembled monolayer; SP, surface plasmon; SPR, surface plasmon resonance; WDM, wavelength division multiplexing.

8. Acknowledgment I gratefully acknowledge the financial support of the Grant Agency of the Academy of Sciences of the Czech Republic (IAA400500507, KAN200670701).

9. References (1) Cush, R.; Cronin, J. M.; Stewart, W. J.; Maule, C. H.; Molloy, J.; Goddard, N. J. Biosens. Bioelectron. 1993, 8, 347. (2) Buckle, P. E.; Davies, R. J.; Kinning, T.; Yeung, D.; Edwards, P. R.; Pollard-Knight, D.; Lowe, C. R. Biosens. Bioelectron. 1993, 8, 355. (3) Clerc, D.; Lukosz, W. Sens. Actuators B 1997, 40, 53. (4) Brandenburg, A.; Gombert, A. Sens. Actuators B 1993, 17, 35. (5) Lukosz, W.; Clerc, D.; Nellen, P. M. Sens. Actuators A 1990, 25, 181. (6) Heideman, R. G.; Lambeck, P. V. Sens. Actuators B 1999, 61, 100. (7) Heideman, R. G.; Kooyman, R. P. H.; Greve, J. Sens. Actuators B 1993, 10, 209. (8) Ymeti, A.; Kanger, J. S.; Greve, J.; Besselink, G. A. J.; Lambeck, P. V.; Wijn, R.; Heideman, R. G. Biosens. Bioelectron. 2005, 20, 1417. (9) Schmitt, K.; Schirmer, B.; Hoffmann, C.; Brandenburg, A.; Meyrueis, P. Biosens. Bioelectron. 2007, 22, 2591. (10) Gauglitz, G.; Brecht, A.; Kraus, G.; Mahm, W. Sens. Actuators B 1993, 11, 21. (11) Schmitt, H.-M.; Brecht, A.; Piehler, J.; Gauglitz, G. Biosens. Bioelectron. 1997, 12, 809. (12) Gordon, J. G.; Ernst, S. Surf. Sci. 1980, 101, 499. (13) Nylander, C.; Liedberg, B.; Lind, T. Sens. Actuators 1982, 3, 79. (14) Ligler, F. S.; Taitt, C. A. R. Optical Biosensors: Present and Future; Elsevier: Amsterdam, The Netherlands, 2002. (15) Malhotra, B. D.; Turner, A. P. F. PerspectiVes in Biosensors; JAI: Amsterdam, The Netherlands, 2003.

490 Chemical Reviews, 2008, Vol. 108, No. 2 (16) Gorton, L. Biosensors and Modern Biospecific Analytical Techniques, 1st ed.; Elsevier: Amsterdam, The Netherlands, 2005. (17) Vo-Dinh, T. Nanotechnology in Biology and Medicine: Methods, DeVices, and Applications; CRC Press: Boca Raton, FL, 2007. (18) Nakamura, H.; Karube, I. Anal. Bioanal. Chem. 2003, 377, 446. (19) Ince, R.; Narayanaswamy, R. Anal. Chim. Acta 2006, 569, 1. (20) Bally, M.; Halter, M.; Voros, J.; Grandin, H. M. Surf. Interface Anal. 2006, 38, 1442. (21) Rebecca, L.; Rich, D. G. M. J. Mol. Recognit. 2006, 19, 478. (22) Rebecca, L.; Rich, D. G. M. J. Mol. Recognit. 2005, 18, 431. (23) Rebecca, L.; Rich, D. G. M. J. Mol. Recognit. 2005, 18, 1. (24) Rebecca, L.; Rich, D. G. M. J. Mol. Recognit. 2003, 16, 351. (25) Rebecca, L.; Rich, D. G. M. J. Mol. Recognit. 2002, 15, 352. (26) Rebecca, L.; Rich, D. G. M. J. Mol. Recognit. 2001, 14, 273. (27) Homola, J.; Yee, S. S.; Gauglitz, G. Sens. Actuators B 1999, 54, 3. (28) Homola, J. Anal. Bioanal. Chem. 2003, 377, 528. (29) Boozer, C.; Kim, G.; Cong, S.; Guan, H.; Londergan, T. Curr. Opin. Biotechnol. 2006, 17, 400. (30) Phillips, K. S.; Cheng, Q. Anal. Bioanal. Chem. 2007, 387, 1831. (31) Homola, J. Surface Plasmon Resonance Based Sensors; Springer: Berlin, Germany, 2006. (32) Wood, R. W. Proc. Phys. Soc. London 1902, 18, 269. (33) Fano, U. J. Opt. Soc. Am. 1941, 31, 231. (34) Otto, A. Z. Phys. 1968, 216, 398. (35) Kretschmann, E.; Raether, H. Z. Naturforsch 1968, 2135-2136, 2135. (36) Boardman, A. D. Electromagnetic Surface Modes; Wiley: Chichester, U.K., 1982. (37) Raether, H. Springer Trends Mod. Phys. 1988, 111, 1. (38) Pitarke, J. M.; Silkin, V. M.; Chulkov, E. V.; Echenique, P. M. Rep. Prog. Phys. 2007, 70, 1. (39) Ishimaru, A.; Jaruwatanadilok, S.; Kuga, Y. Prog. Electromagn. Res.-Pier 2005, 51, 139. (40) Stegeman, G. I.; Burke, J. J.; Hall, D. G. Opt. Lett. 1983, 8, 383. (41) Burke, J. J.; Stegeman, G. I.; Tamir, T. Phys. ReV. B 1986, 33, 5186. (42) Sarid, D. Phys. ReV. Lett. 1981, 47, 1927. (43) Hutley, M. C. Diffraction Gratings; Academic Press: London, U.K., 1982. (44) Matsubara, K.; Kawata, S.; Minami, S. Appl. Spectrosc. 1988, 42, 1375. (45) Zhang, L. M.; Uttamchandan, D. Electron. Lett. 1988, 24, 1469. (46) Brockman, J. M.; Nelson, B. P.; Corn, R. M. Annu. ReV. Phys. Chem. 2000, 51, 41. (47) Ho, C.; Robinson, A.; Miller, D.; Davis, M. Sensors 2005, 5, 4. (48) Tumolo, T.; Angnes, L.; Baptista, M. S. Anal. Biochem. 2004, 333, 273. (49) de Feijter, J. A.; Benjamins, J.; Veer, F. A. Biopolymers 1978, 17, 1759. (50) Snyder, A. W.; Love, J. D. Optical WaVeguide Theory; Chapman and Hall: London, U.K., 1983. (51) O’Brien, M. J.; Perez-Luna, V. H.; Brueck, S. R. J.; Lopez, G. P. Biosens. Bioelectron. 2001, 16, 97. (52) Kawazumi, H.; Gobi, K. V.; Ogino, K.; Maeda, H.; Miura, N. Sens. Actuators B 2005, 108, 791. (53) Piliarik, M.; Vaisocherova´, H.; Homola, J. Biosens. Bioelectron. 2005, 20, 2104. (54) Berger, C. E. H.; Beumer, T. A. M.; Kooyman, R. P. H.; Greve, J. Anal. Chem. 1998, 70, 703. (55) Fu, E.; Chinowsky, T.; Foley, J.; Weinstein, J.; Yager, P. ReV. Sci. Instrum. 2004, 75, 2300. (56) Kukanskis, K.; Elkind, J.; Melendez, J.; Murphy, T.; Miller, G.; Garner, H. Anal. Biochem. 1999, 274, 7. (57) Goddard, N. J.; Pollardknight, D.; Maule, C. H. Analyst 1994, 119, 583. (58) Sjo¨lander, S.; Urbanitzky, C. Anal. Chem. 1991, 63, 2338. (59) Stenberg, E.; Persson, B.; Roos, H.; Urbaniczky, C. J. Colloid Interface Sci. 1991, 143, 513. (60) Chinowsky, T. M.; Jung, L. S.; Yee, S. S. Sens. Actuators B 1999, 54, 89. (61) Tobisˇka, P.; Homola, J. Sens. Actuators B 2005, 107, 162. (62) Nenninger, G. G.; Piliarik, M.; Homola, J. Meas. Sci. Technol. 2002, 13, 2038. (63) Thirstrup, C.; Zong, W.; Borre, M.; Neff, H.; Pedersen, H. C.; Holzhueter, G. Sens. Actuators B 2004, 100, 298. (64) Dosta´lek, J.; Homola, J.; Miler, M. Sens. Actuators B 2005, 107, 154. (65) Ran, B.; Lipson, S. G. Opt. Express 2006, 14, 5641. (66) Thomsen, V.; Schatzlein, D.; Mercuro, D. Spectroscopy 2003, 18, 112. (67) Sepulveda, B.; Calle, A.; Lechuga, L. M.; Armelles, G. Opt. Lett. 2006, 31, 1085. (68) Jordan, C. E.; Frutos, A. G.; Thiel, A. J.; Corn, R. M. Anal. Chem. 1997, 69, 4939. (69) Jordan, C. E.; Corn, R. M. Anal. Chem. 1997, 69, 1449.

Homola (70) Nelson, B. P.; Frutos, A. G.; Brockman, J. M.; Corn, R. M. Anal. Chem. 1999, 71, 3928. (71) Nelson, B. P.; Grimsrud, T. E.; Liles, M. R.; Goodman, R. M.; Corn, R. M. Anal. Chem. 2001, 73, 1. (72) Fu, E.; Foley, J.; Yager, P. ReV. Sci. Instrum. 2003, 74, 3182. (73) Wark, A. W.; Lee, H. J.; Corn, R. M. Anal. Chem. 2005, 77, 3904. (74) Zybin, A.; Grunwald, C.; Mirsky, V. M.; Kuhlmann, J.; Wolfbeis, O. S.; Niemas, K. Anal. Chem. 2005, 77, 2393. (75) Shumaker-Parry, J. S.; Campbell, C. T. Anal. Chem. 2004, 76, 907. (76) Shumaker-Parry, J. S.; Aebersold, R.; Campbell, C. T. Anal. Chem. 2004, 76, 2071. (77) Campbell, C. T.; Kim, G. Biomaterials 2007, 28, 2380. (78) Piliarik, M.; Katainen, J.; Homola, J. Optical Sensing Technology and Applications; Prague, Czech Republic; SPIE: Bellingham, WA, 2007; p 658515. (79) Piliarik, M.; Vaisocherova´, H.; Homola, J. Sens. Actuators B 2007, 121, 187. (80) http://www.gwctechnologies.com. (81) http://www.lumera.com. (82) http://www.ibis-spr.nl/. (83) http://www.genoptics-spr.com/. (84) Lofas, S.; Malmqvist, M.; Ronnberg, I.; Stenberg, E.; Liedberg, B.; Lundstrom, I. Sens. Actuators B 1991, 5, 79. (85) Liedberg, B.; Lundstrom, I.; Stenberg, E. Sens. Actuators B 1993, 11, 63. (86) Karlsson, R.; Stahlberg, R. Anal. Biochem. 1995, 228, 274. (87) Nice, E. C.; Catimel, B. Bioessays 1999, 21, 339. (88) http://www.biacore.com. (89) Pedersen, H. C.; Thirstrup, C. Appl. Opt. 2004, 43, 1209. (90) Pedersen, H. C.; Zong, W. Y.; Sorensen, M. H.; Thirstrup, C. Opt. Eng. 2004, 43, 2505. (91) Homola, J.; Dosta´lek, J.; Chen, S. F.; Rasooly, A.; Jiang, S. Y.; Yee, S. S. Int. J. Food Microbiol. 2002, 75, 61. (92) Homola, J.; Lu, H. B.; Yee, S. S. Electron. Lett. 1999, 35, 1105. (93) Homola, J.; Lu, H. B. B.; Nenninger, G. G.; Dosta´lek, J.; Yee, S. S. Sens. Actuators B 2001, 76, 403. (94) Dosta´lek, J.; Vaisocherova´, H.; Homola, J. Sens. Actuators B 2005, 108, 758. (95) Nenninger, G. G.; Tobisˇka, P.; Homola, J.; Yee, S. S. Sens. Actuators B 2001, 74, 145. (96) Slavı´k, R.; Homola, J. Sens. Actuators B 2007, 123, 10. (97) Melendez, J.; Carr, R.; Bartholomew, D. U.; Kukanskis, K.; Elkind, J.; Yee, S.; Furlong, C.; Woodbury, R. Sens. Actuators B 1996, 35, 212. (98) Elkind, J. L.; Stimpson, D. I.. A. A. S. D. U. B.; Melendez, J. L. Sens. Actuators B 1999, 54, 182. (99) Chinowsky, T. M.; Quinn, J. G.; Bartholomew, D. U.; Kaiser, R.; Elkind, J. L. Sens. Actuators B 2003, 91, 266. (100) Naimushin, A.; Soelberg, S.; Bartholomew, D.; Elkind, J.; Furlong, C. Sens. Actuators B 2003, 96, 253. (101) Naimushin, A. N.; Spinelli, C. B.; Soelberg, S. D.; Mann, T.; Stevens, R. C.; Chinowsky, T.; Kauffman, P.; Yee, S.; Furlong, C. E. Sens. Actuators B 2005, 104, 237. (102) http://www.ti.com/. (103) http://www.optrel.de/. (104) http://www.reichertspr.com/. (105) http://www.plasmonic.de/. (106) http://www.ecochemie.nl/. (107) http://dkktoa.net/index.html. (108) http://www.biosuplar.com/. (109) http://www.sensia.es/. (110) Nikitin, P. I.; Beloglazov, A. A.; Kochergin, V. E.; Valeiko, M. V.; Ksenevich, T. I. Sens. Actuators B 1999, 54, 43. (111) Nikitin, P. I.; Grigorenko, A. N.; Beloglazov, A. A.; Valeiko, M. V.; Savchuk, A. I.; Savchuk, O. A.; Steiner, G.; Kuhne, C.; Huebner, A.; Salzer, R. Sens. Actuators A 2000, 85, 189. (112) Notcovich, A. G.; Zhuk, V.; Lipson, S. G. Appl. Phys. Lett. 2000, 76, 1665. (113) Wu, C.-M.; Jian, Z.-C.; Joe, S.-F.; Chang, L.-B. Sens. Actuators B 2003, 92, 133. (114) Alieva, E. V.; Konopsky, V. N. Sens. Actuators B 2004, 99, 90. (115) Naraoka, R.; Kajikawa, K. Sens. Actuators B 2005, 107, 952. (116) Ho, H. P.; Lam, W. W.; Wu, S. Y. ReV. Sci. Instrum. 2002, 73, 3534. (117) Wu, S. Y.; Ho, H. P.; Law, W. C.; Lin, C. L.; Kong, S. K. Opt. Lett. 2004, 29, 2378. (118) Ho, H. P.; Law, W. C.; Wu, S. Y.; Lin, C. L.; Kong, S. K. Biosens. Bioelectron. 2005, 20, 2177. (119) Ho, H. P.; Law, W. C.; Wu, S. Y.; Liu, X. H.; Wong, S. P.; Lin, C. L.; Kong, S. K. Sens. Actuators B 2006, 114, 80. (120) Yuan, W.; Ho, H. P.; Wong, C. L.; Kong, S. K.; Lin, C. L. IEEE Sensors J. 2007, 7, 70. (121) Brockman, J. M.; Fernandez, S. M. Am. Lab. 2001, 33, 37.

SPR Sensors (122) Baggio, R.; Carven, G. J.; Chiulli, A.; Palmer, M.; Stern, L. J.; Arenas, J. E. J. Biol. Chem. 2005, 280, 4188. (123) Adam, P.; Dosta´lek, J.; Homola, J. Sens. Actuators B 2006, 113, 774. (124) Adam, P.; Dosta´lek, J.; Telezhnikova, O.; Homola, J. 2007; p 65851Y. (125) Vala, M.; Dosta´lek, J.; Homola, J. 2007; p 658522. (126) Unfricht, D. W.; Colpitts, S. L.; Fernandez, S. M.; Lynes, M. A. Proteomics 2005, 5, 4432. (127) Telezhnikova, O.; Homola, J. Opt. Lett. 2006, 31, 3339. (128) Chien, F. C.; Lin, C. Y.; Yih, J. N.; Lee, K. L.; Chang, C. W.; Wei, P. K.; Sun, C. C.; Chen, S. J. Biosens. Bioelectron. 2007, 22, 2737. (129) Van Gent, J.; Lambeck, P. V.; Kreuwel, H. J. M.; Gerritsma, G. J.; Sudholter, E. J. R.; Reinhoudt, D. N.; Popma, T. J. A. Appl. Opt. 1990, 29, 2843. (130) Jorgenson, R. C.; Yee, S. S. Sens. Actuators B 1993, 12, 213. (131) Homola, J. Sens. Actuators B 1995, 29, 401. (132) Slavı´k, R.; Homola, J.; C ˇ tyrok, J. Sens. Actuators B 1999, 54, 74. (133) Slavı´k, R.; Homola, J.; C ˇ tyrok, J.; Brynda, E. Sens. Actuators B 2001, 74, 106. (134) Piliarik, M.; Homola, J.; Manikova´, Z.; C ˇ tyrok, J. Sens. Actuators B 2003, 90, 236. (135) Chiu, M. H.; Wang, S. F.; Chang, R. S. Opt. Lett. 2005, 30, 233. (136) Lin, H. Y.; Tsai, W. H.; Tsao, Y. C.; Sheu, B. C. Appl. Opt. 2007, 46, 800. (137) Mouvet, C.; Harris, R.; Maciag, C.; Luff, B.; Wilkinson, J.; Piehler, J.; Brecht, A.; Gauglitz, G.; Abuknesha, R.; Ismail, G. Anal. Chim. Acta 1997, 338, 109. (138) Harris, R. D.; Luff, B. J.; Wilkinson, J. S.; Piehler, J.; Brecht, A.; Gauglitz, G.; Abuknesha, R. A. Biosens. Bioelectron. 1999, 14, 377. (139) Dosta´lek, J.; C ˇ tyrok, J.; Homola, J.; Brynda, E.; Skalsk, M.; Nekvindova´, P.; Spirkova´, J.; Sˇ kvor, J.; Schrofel, J. Sens. Actuators, B 2001, 76, 8. (140) Huang, J. G.; Lee, C. L.; Lin, H. M.; Chuang, T. L.; Wang, W. S.; Juang, R. H.; Wang, C. H.; Lee, C. K.; Lin, S. M.; Lin, C. W. Biosens. Bioelectron. 2006, 22, 519. (141) Harris, R. D.; Wilkinson, J. S. Sens. Actuators B 1995, 29, 261. (142) Lavers, C. R.; Wilkinson, J. S. Sens. Actuators B 1994, 22, 75. (143) Ctyroky, J.; Homola, J.; Skalsky, M. Electron. Lett. 1997, 33, 1246. (144) Skorobogatiy, M.; Kabashin, A. V. Appl. Phys. Lett. 2006, 89. (145) Debackere, P.; Scheerlinck, S.; Bienstman, P.; Baets, R. Opt. Express 2006, 14, 7063. (146) Wang, T. J.; Lin, W. S.; Liu, F. K. Biosens. Bioelectron. 2007, 22, 1441. (147) Tomizaki, K. Y.; Usui, K.; Mihara, H. ChemBioChem 2005, 6, 782. (148) Dunne, L.; Daly, S.; Baxter, A.; Haughey, S.; O’Kennedy, R. Spectrosc. Lett. 2005, 38, 229. (149) Scholler, N.; Garvik, B.; Quarles, T.; Jiang, S.; Urban, N. J. Immunol. Methods 2006, 317, 132. (150) Rojo, N.; Ercilla, G.; Haro, I. Curr. Protein Pept. Sci. 2003, 4, 291. (151) Wittekindt, C.; Fleckenstein, B.; Wiesmuller, K.; Eing, B. R.; Kuhn, J. E. Virol. J. Methods 2000, 87, 133. (152) Vaisocherova´, H.; Mrkvova´, K.; Piliarik, M.; Jinoch, P.; Steinbachova´, M.; Homola, J. Biosens. Bioelectron. 2007, 22, 1020. (153) Forzani, E. S.; Zhang, H. Q.; Chen, W.; Tao, N. J. EnViron. Sci. Technol. 2005, 39, 1257. (154) Win, M. N.; Klein, J. S.; Smolke, C. D. Nucleic Acids Res. 2006, 34, 5670. (155) Wang, Z.; Wilkop, T.; Xu, D.; Dong, Y.; Ma, G.; Cheng, Q. Anal. Bioanal. Chem. 2007, 389, 819. (156) Jayasena, S. D. Clin. Chem. 1999, 45, 1628. (157) Zhang, S. G. Nat. Mater. 2004, 3, 7. (158) Lofas, S.; Johnsson, B.; Edstrom, A.; Hansson, A.; Lindquist, G.; Hillgren, R. M. M.; Stigh, L. Biosens. Bioelectron. 1995, 10, 813. (159) Knoll, W.; Liley, M.; Piscevic, D.; Spinke, J.; Tarlov, M. J. AdV. Biophys. 1997, 34, 231. (160) Jung, L. S.; Nelson, K. E.; Stayton, P. S.; Campbell, C. T. Langmuir 2000, 16, 9421. (161) Nelson, K. E.; Gamble, L.; Jung, L. S.; Boeckl, M. S.; Naeemi, E.; Golledge, S. L.; Sasaki, T.; Castner, D. G.; Campbell, C. T.; Stayton, P. S. Langmuir 2001, 17, 2807. (162) Koubova´, V.; Brynda, E.; Karasova´, L.; Sˇ kvor, J.; Homola, J.; Dosta´lek, J.; Tobisˇka, P.; Rosick, J. Sens. Actuators, B 2001, 74, 100. (163) Busse, S.; Scheumann, V.; Menges, B.; Mittler, S. Biosens. Bioelectron. 2002, 17, 704. (164) Zhen, G. L.; Falconnet, D.; Kuennemann, E.; Voros, J.; Spencer, N. D.; Textor, M.; Zurcher, S. AdV. Funct. Mater. 2006, 16, 243. (165) Ladd, J.; Boozer, C.; Yu, Q.; Chen, S.; Homola, J.; Jiang, S. Langmuir 2004, 20, 8090. (166) Rusmini, F.; Zhong, Z.; Feijen, J. Biomacromolecules 2007, 8, 1775. (167) Oshannessy, D. J.; Brighamburke, M.; Peck, K. Anal. Biochem. 1992, 205, 132.

Chemical Reviews, 2008, Vol. 108, No. 2 491 (168) Johnson, C. P.; Jensen, I. E.; Prakasam, A.; Vijayendran, R.; Leckband, D. Bioconjugate Chem. 2003, 14, 974. (169) Wegner, G. J.; Lee, N. J.; Marriott, G.; Corn, R. M. Anal. Chem. 2003, 75, 4740. (170) Lata, S.; Reichel, A.; Brock, R.; Tampe, R.; Piehler, J. J. Am. Chem. Soc. 2005, 127, 10205. (171) Tinazli, A.; Tang, J.; Valiokas, R.; Picuric, S.; Lata, S.; Piehler, J.; Liedberg, B.; Tampe, R. Chemistry 2005, 11, 5249. (172) Valiokas, R.; Klenkar, G.; Tinazli, A.; Tampe, R.; Liedberg, B.; Piehler, J. ChemBioChem 2006, 7, 1325. (173) Vaisocherova´, H.; Zı´tova´, A.; Lachmanova´, M.; Sˇ tcˇpa´nek, J.; Rosenberg, I.; Kra´lı´kova´, Sˇ .; Liboska, R.; Rejman, D.; Homola, J. Biopolymers 2006. (174) Hermanson, G. T. Bioconjugate Techniques; Academic Press: San Diego, CA, 1996. (175) Shumaker-Parry, J. S.; Zareie, M. H.; Aebersold, R.; Campbell, C. T. Anal. Chem. 2004, 76, 918. (176) Joos, T. O.; Schrenk, M.; Hopfl, P.; Kroger, K.; Chowdhury, U.; Stoll, D.; Schorner, D.; Durr, M.; Herick, K.; Rupp, S.; Sohn, K.; Hammerle, H. Electrophoresis 2000, 21, 2641. (177) Calvo, K. R.; Liotta, L. A.; Petricoin, E. F. Biosci. Rep. 2005, 25, 107. (178) Huang, R. P. Frontiers Biosci. 2003, 8, D559. (179) Figeys, D. Proteomics 2002, 2, 373. (180) Blattler, T. M.; Pasche, S.; Textor, M.; Griesser, H. J. Langmuir 2006, 22, 5760. (181) Pasche, S.; Textor, M.; Meagher, L.; Spencer, N. D.; Griesser, H. J. Langmuir 2005, 21, 6508. (182) Pasche, S.; De Paul, S. M.; Voros, J.; Spencer, N. D.; Textor, M. Langmuir 2003, 19, 9216. (183) Pasche, S.; Textor, M.; Meagher, L.; Spencer, N. D.; Griesser, H. J. Langmuir 2005, 21, 6508. (184) Huang, N. P.; Voros, J.; De Paul, S. M.; Textor, M.; Spencer, N. D. Langmuir 2002, 18, 220. (185) Fei, X.; Guoliang, Z.; Marcus, T.; Wolfgang, K. Biointerphases 2006, 1, 73. (186) Marie, R.; Dahlin, A. B.; Tegenfeldt, J. O.; Hook, F. Biointerphases 2007, 2, 49. (187) Frederix, F.; Bonroy, K.; Reekmans, G.; Laureyn, W.; Campitelli, A.; Abramov, M. A.; Dehaen, W.; Maes, G. J. Biochem. Biophys. Methods 2004, 58, 67. (188) Feng, C. L.; Zhang, Z.; Forch, R.; Knoll, W.; Vancso, G. J.; Schonherr, H. Biomacromolecules 2005, 6, 3243. (189) Mancebo, H. S.; Lee, G.; Flygare, J.; Tomassini, J.; Luu, P.; Zhu, Y.; Peng, J.; Blau, C.; Hazuda, D.; Price, D.; Flores, O. Genes DeV. 1997, 11, 2633. (190) Harbers, G. M.; Emoto, K.; Greef, C.; Metzger, S. W.; Woodward, H. N.; Mascali, J. J.; Grainger, D. W.; Lochhead, M. J. Chem. Mater. 2007, 19, 4405. (191) Ostuni, E.; Chapman, R. G.; Liang, M. N.; Meluleni, G.; Pier, G.; Ingber, D. E.; Whitesides, G. M. Langmuir 2001, 17, 6336. (192) Bearinger, J. P.; Terrettaz, S.; Michel, R.; Tirelli, N.; Vogel, H.; Textor, M.; Hubbell, J. A. Nat. Mater. 2003, 2, 259. (193) Li, L.; Chen, S.; Jiang, S. J. Biomater. Sci., Polym. Ed. 2007, 18, 1415. (194) Holmlin, R. E.; Chen, X. X.; Chapman, R. G.; Takayama, S.; Whitesides, G. M. Langmuir 2001, 17, 2841. (195) Kitano, H.; Kawasaki, A.; Kawasaki, H.; Morokoshi, S. J. Colloid Interface Sci. 2005, 282, 340. (196) Zhang, Z.; Chen, S.; Jiang, S. Biomacromolecules 2006, 7, 3311. (197) Cheng, G.; Zhang, Z.; Chen, S. F.; Bryers, J. D.; Jiang, S. Y. Biomaterials 2007, 28, 4192. (198) Chen, S.; Liu, L.; Jiang, S. Langmuir 2006, 22, 2418. (199) Vaisocherova´, H.; Mrkvova´, K.; Piliarik, M.; Jinoch, P.; Steinbachova´, M.; Homola, J. Biosens. Bioelectron. 2007, 22, 1020. (200) Ayela, C.; Roquet, F.; Valera, L.; Granier, C.; Nicu, L.; Pugniere, M. Biosens. Bioelectron. 2007, 22, 3113. (201) Cao, C.; Kim, J. P.; Kim, B. W.; Chae, H.; Yoon, H. C.; Yang, S. S.; Sim, S. J. Biosens. Bioelectron. 2006, 21, 2106. (202) Chung, J. W.; Kim, S. D.; Bernhardt, R.; Pyun, J. C. Sens. Actuators B 2005, 111, 416. (203) Gobi, K. V.; Iwasaka, H.; Miura, N. Biosens. Bioelectron. 2007, 22, 1382. (204) Lahiri, J.; Isaacs, L.; Tien, J.; Whitesides, G. M. Anal. Chem. 1999, 71, 777. (205) Habauzit, D.; Chopineau, J.; Roig, B. Anal. Bioanal. Chem. 2007, 387, 1215. (206) Shankaran, D. R.; Gobi, K. V. A.; Miura, N. Sens. Actuators B 2007, 121, 158. (207) Seydack, M. Biosens. Bioelectron. 2005, 20, 2454. (208) Chung, J. W.; Bernhardt, R.; Pyun, J. C. Immunol. J. Methods 2006, 311, 178.

492 Chemical Reviews, 2008, Vol. 108, No. 2 (209) Mauriz, E.; Calle, A.; Manclus, J. J.; Montoya, A.; Lechuga, L. M. Anal. Bioanal. Chem. 2007, 387, 1449. (210) Kubitschko, S.; Spinke, J.; Bruckner, T.; Pohl, S.; Oranth, N. Anal. Biochem. 1997, 253, 112. (211) Lyon, L. A.; Musick, M. D.; Smith, P. C.; Reiss, B. D.; Pena, D. J.; Natan, M. J. Sens. Actuators B 1999, 54, 118. (212) Mitchell, J. S.; Wu, Y.; Cook, C. J.; Main, L. Anal. Biochem. 2005, 343, 125. (213) Sato, Y.; Ikegaki, S.; Suzuki, K.; Kawaguchi, H. J. Biomater. Sci. Polym. Ed. 2003, 14, 803. (214) Okumura, A.; Sato, Y.; Kyo, M.; Kawaguchi, H. Anal. Biochem. 2005, 339, 328. (215) Sato, Y.; Sato, Y.; Okumura, A.; Suzuki, K.; Kawaguchi, H. J. Biomater. Sci. Polym. Ed. 2004, 15, 297. (216) Komatsu, H.; Miyachi, M.; Fujii, E.; Citterio, D.; Yamada, K.; Sato, Y.; Kurihara, K.; Kawaguchi, H.; Suzuki, K. Sci. Technol. AdV. Mater. 2006, 7, 150. (217) Goodrich, T. T.; Lee, H. J.; Corn, R. M. Anal. Chem. 2004, 76, 6173. (218) Lee, H. J.; Li, Y.; Wark, A. W.; Corn, R. M. Anal. Chem. 2005, 77, 5096. (219) Fang, S. P.; Lee, H. J.; Wark, A. W.; Corn, R. M. J. Am. Chem. Soc. 2006, 128, 14044. (220) Li, Y.; Lee, H. J.; Corn, R. M. Anal. Chem. 2007, 79, 1082. (221) Patel, P. D. J. AOAC Int. 2006, 89, 805. (222) Lazcka, O.; Campo, F. J. D.; Munoz, F. X. Biosens. Bioelectron. 2007, 22, 1205. (223) Leonard, P.; Hearty, S.; Brennan, J.; Dunne, L.; Quinn, J.; Chakraborty, T.; O’Kennedy, R. Enzyme Microb. Technol. 2003, 32, 3. (224) Mello, L. D.; Kubota, L. T. Food Chem. 2002, 77, 237. (225) Baeumner, A. J. Anal. Bioanal. Chem. 2003, 377, 434. (226) Bergwerff, A. A.; Van Knapen, F. J. AOAC Int. 2006, 89, 826. (227) Fratamico, P. M.; Strobaugh, T. P.; Medina, M. B.; Gehring, A. G. Biotechnol. Tech. 1998, 12, 571. (228) Oh, B. K.; Kim, Y. K.; Bae, Y. M.; Lee, W. H.; Choi, J. W. J. Microb. Biotechnol. 2002, 12, 780. (229) Oh, B. K.; Lee, W.; Lee, W. H.; Choi, J. W. Biotechnol. Bioprocess Eng. 2003, 8, 227. (230) Taylor, A. D.; Yu, Q. M.; Chen, S. F.; Homola, J.; Jiang, S. Y. Sens. Actuators B 2005, 107, 202. (231) Meeusen, C. A.; Alocilja, E. C.; Osburn, W. N. Trans. ASAE 2005, 48, 2409. (232) Su, X. L.; Li, Y. Trans. ASAE 2005, 48, 405. (233) Subramanian, A.; Irudayaraj, J.; Ryan, T. Biosens. Bioelectron. 2006, 21, 998. (234) Taylor, A. D.; Ladd, J.; Yu, Q. M.; Chen, S. F.; Homola, J.; Jiang, S. Y. Biosens. Bioelectron. 2006, 22, 752. (235) Waswa, J. W.; Debroy, C.; Irudayaraj, J. J. Food Process Eng. 2006, 29, 373. (236) Waswa, J.; Irudayaraj, J.; DebRoy, C. LWT-Food Sci. Technol. 2007, 40, 187. (237) Bokken, G.; Corbee, R. J.; van Knapen, F.; Bergwerff, A. A. FEMS Microbiol. Lett. 2003, 222, 75. (238) Oh, B. K.; Kim, Y. K.; Park, K. W.; Lee, W. H.; Choi, J. W. Biosens. Bioelectron. 2004, 19, 1497. (239) Oh, B. K.; Lee, W.; Kim, Y. K.; Lee, W. H.; Choi, J. W. J. Biotechnol. 2004, 111, 1. (240) Mazumdar, S. D.; Hartmann, M.; Kampfer, P.; Keusgen, M. Biosens. Bioelectron. 2007, 22, 2040. (241) Leonard, P.; Hearty, S.; Quinn, J.; O’Kennedy, R. Biosens. Bioelectron. 2004, 19, 1331. (242) Subramanian, A.; Irudayaraj, J.; Ryan, T. Sens. Actuators B 2006, 114, 192. (243) Balasubramanian, S.; Sorokulova, I. B.; Vodyanoy, V. J.; Simonian, A. L. Biosens. Bioelectron. 2007, 22, 948. (244) Oh, B. K.; Lee, W.; Chun, B. S.; Bae, Y. M.; Lee, W. H.; Choi, J. W. Colloids Surf. A 2005, 257-58, 369. (245) Jyoung, J. Y.; Hong, S. H.; Lee, W.; Choi, J. W. Biosens. Bioelectron. 2006, 21, 2315. (246) Kang, C. D.; Lee, S. W.; Park, T. H.; Sim, S. J. Enzyme Microb. Technol. 2006, 39, 387. (247) Zezza, F.; Pascale, M.; Mule, G.; Visconti, A. J. Microbiol. Methods 2006, 66, 529. (248) Nedelkov, D.; Rasooly, A.; Nelson, R. W. Int. J. Food Microbiol. 2000, 60, 1. (249) Nedelkov, D.; Nelson, R. W. Appl. EnViron. Microbiol. 2003, 69, 5212. (250) Slavı´k, R.; Homola, J.; Brynda, E. Biosens. Bioelectron. 2002, 17, 591. (251) Naimushin, A. N.; Soelberg, S. D.; Nguyen, D. K.; Dunlap, L.; Bartholomew, D.; Elkind, J.; Melendez, J.; Furlong, C. E. Biosens. Bioelectron. 2002, 17, 573. (252) Medina, M. B. J. Rapid Methods Autom. Microbiol. 2005, 13, 37. (253) Medina, M. B. J. Rapid Methods Autom. Microbiol. 2003, 11, 225.

Homola (254) Medina, M. B. J. Rapid Methods Autom. Microbiol. 2006, 14, 119. (255) Lotierzo, M.; Henry, O. Y. F.; Piletsky, S.; Tothill, I.; Cullen, D.; Kania, M.; Hock, B.; Turner, A. P. F. Biosens. Bioelectron. 2004, 20, 145. (256) Yu, Q. M.; Chen, S. F.; Taylor, A. D.; Homola, J.; Hock, B.; Jiang, S. Y. Sens. Actuators B 2005, 107, 193. (257) Traynor, I. M.; Plumpton, L.; Fodey, T. L.; Higgins, C.; Elliott, C. T. J. AOAC Int. 2006, 89, 868. (258) Stevens, R. C.; Soelberg, S. D.; Eberhart, B. T. L.; Spencer, S.; Wekell, J. C.; Chinowsky, T. M.; Trainer, V. L.; Furlong, C. E. Harmful Algae 2007, 6, 166. (259) Daly, S. J.; Keating, G. J.; Dillon, P. P.; Manning, B. M.; O’Kennedy, R.; Lee, H. A.; Morgan, M. R. A. J. Agric. Food Chem. 2000, 48, 5097. (260) Tudos, A. J.; Lucas-van den Bos, E. R.; Stigter, E. C. A. J. Agric. Food Chem. 2003, 51, 5843. (261) Haughey, S. A.; Baxter, C. A. J. AOAC Int. 2006, 89, 862. (262) Cacciatore, G.; Petz, M.; Rachid, S.; Hakenbeck, R.; Bergwerff, A. A. Anal. Chim. Acta 2004, 520, 105. (263) Ashwin, H. M.; Stead, S. L.; Taylor, J. C.; Startin, J. R.; Richmond, S. F.; Homer, V.; Bigwood, T.; Sharman, M. Anal. Chim. Acta 2005, 529, 103. (264) Ferguson, J.; Baxter, A.; Young, P.; Kennedy, G.; Elliott, C.; Weigel, S.; Gatermann, R.; Ashwind, H.; Stead, S.; Sharman, M. Anal. Chim. Acta 2005, 529, 109. (265) Dumont, V.; Huet, A. C.; Traynor, I.; Elliott, C.; Delahaut, P. Anal. Chim. Acta 2006, 567, 179. (266) Moeller, N.; Mueller-Seitz, E.; Scholz, O.; Hillen, W.; Bergwerff, A. A.; Petz, M. Eur. Food Res. Technol. 2007, 224, 285. (267) Caldow, M.; Stead, S. L.; Day, J.; Sharman, M.; Situ, C.; Elliott, C. J. Agric. Food Chem. 2005, 53, 7367. (268) Caelen, I.; Kalman, A.; Wahlstrom, L. Anal. Chem. 2004, 76, 137. (269) Haughey, S. A.; O’Kane, A. A.; Baxter, G. A.; Kalman, A.; Trisconi, M. J.; Indyk, H. E.; Watene, G. A. J. AOAC Int. 2005, 88, 1008. (270) Indyk, H. E.; Evans, E. A.; Caselunghe, M. C. B.; Persson, B. S.; Finglas, P. M.; Woollard, D. C.; Filonzi, E. L. J. AOAC Int. 2000, 83, 1141. (271) Indyk, H. E.; Persson, B. S.; Caselunghe, M. C. B.; Moberg, A.; Filonzi, E. L.; Woollard, D. C. J. AOAC Int. 2002, 85, 72. (272) Gillis, E. H.; Gosling, J. P.; Sreenan, J. M.; Kane, M. J. Immunol. Methods 2002, 267, 131. (273) Gillis, E. H.; Traynor, I.; Gosling, J. P.; Kane, M. J. AOAC Int. 2006, 89, 838. (274) Kim, T. J.; Cho, H. S.; Park, N. Y.; Lee, J. I. J. Vet. Med. Ser. B: Infect. Dis. Vet. Public Health 2006, 53, 87. (275) Cho, H. S.; Park, N. Y. J. Vet. Med. Sci. 2006, 68, 1327. (276) Mohammed, I.; Mullett, W. M.; Lai, E. P. C.; Yeung, J. M. Anal. Chim. Acta 2001, 444, 97. (277) Malmheden Yman, I.; Eriksson, A.; Johansson, M. A.; Hellenas, K. E. J. AOAC Int. 2006, 89, 856. (278) Dupont, D.; Muller-Renaud, S. J. AOAC Int. 2006, 89, 843. (279) Indyk, H. E.; Filonzi, E. L.; Gapper, L. W. J. AOAC Int. 2006, 89, 898. (280) Samsonova, J. V.; Uskova, N. A.; Andresyuk, A. N.; Franek, M.; Elliott, C. T. Chemosphere 2004, 57, 975. (281) Guidi, A.; Laricchia-Robbio, L.; Gianfaldoni, D.; Revoltella, R.; Del Bono, G. Biosens. Bioelectron. 2001, 16, 971. (282) Pejcic, B.; De Marco, R.; Parkinson, G. Analyst 2006, 131, 1079. (283) Andreescu, S.; Sadik, O. A. Pure Appl. Chem. 2004, 76, 861. (284) Healy, D. A.; Hayes, C. J.; Leonard, P.; McKenna, L.; O’Kennedy, R. Trends Biotechnol. 2007, 25, 125. (285) Besselink, G. A.; Kooyman, R. P.; van Os, P. J.; Engbers, G. H.; Schasfoort, R. B. Anal. Biochem. 2004, 333, 165. (286) Wu, L. P.; Li, Y. F.; Huang, C. Z.; Zhang, Q. Anal. Chem. 2006, 78, 5570. (287) Chung, J. W.; Bernhardt, R.; Pyun, J. C. Sens. Actuators B 2006, 118, 28. (288) Yang, C. Y.; Brooks, E.; Li, Y.; Denny, P.; Ho, C. M.; Qi, F. X.; Shi, W. Y.; Wolinsky, L.; Wu, B.; Wong, D. T. W.; Montemagno, C. D. Lab Chip 2005, 5, 1017. (289) Tang, D. P.; Yuan, R.; Chai, Y. Q. Bioprocess Biosyst. Eng. 2006, 28, 315. (290) Regnault, V.; Boehlen, F.; Ozsahin, H.; Wahl, D.; de Groot, P. G.; Lecompte, T.; de Moerloose, P. J. Thromb. Haemostasis 2005, 3, 1243. (291) McGill, A.; Greensill, J.; Marsh, R.; Craft, A. W.; Toms, G. L. J. Med. Virol. 2004, 74, 492. (292) Abad, L. W.; Neumann, M.; Tobias, L.; Obenauer-Kutner, L.; Jacobs, S.; Cullen, C. Anal. Biochem. 2002, 310, 107. (293) Dillon, P. P.; Daly, S. J.; Manning, B. M.; O’Kennedy, R. Biosens. Bioelectron. 2003, 18, 217. (294) Fitzpatrick, B.; O’Kennedy, R. J. Immunol. Methods 2004, 291, 11.

SPR Sensors (295) Rini, B.; Wadhwa, M.; Bird, C.; Small, E.; Gaines-Das, R.; Thorpe, R. Cytokine 2005, 29, 56. (296) Kure, M.; Katsura, Y.; Kosano, H.; Noritake, M.; Watanabe, T.; Iwaki, Y.; Nishigori, H.; Matsuoka, T. Intern. Med. 2005, 44, 100. (297) Miyashita, M.; Shimada, T.; Miyagawa, H.; Akamatsu, M. Anal. Bioanal. Chem. 2005, 381, 667. (298) Teramura, Y.; Iwata, H. Anal. Biochem. 2007, 365, 201. (299) Li, Y.; Kobayashi, M.; Furui, K.; Soh, N.; Nakano, K.; Imato, T. Anal. Chim. Acta 2006, 576, 77. (300) Wei, J.; Mu, Y.; Song, D.; Fang, X.; Liu, X.; Bu, L.; Zhang, H.; Zhang, G.; Ding, J.; Wang, W.; Jin, Q.; Luo, G. Anal. Biochem. 2003, 321, 209. (301) Masson, J. F.; Obando, L.; Beaudoin, S.; Booksh, K. Talanta 2004, 62, 865. (302) Kim, J. Y.; Lee, M. H.; Jung, K. I.; Na, H. Y.; Cha, H. S.; Ko, E. M.; Kim, T. J. Exp. Mol. Med. 2003, 35, 310. (303) Lee, J. W.; Sim, S. J.; Cho, S. M.; Lee, J. Biosens. Bioelectron. 2005, 20, 1422. (304) Choi, S. H.; Lee, J. W.; Sim, S. J. Biosens. Bioelectron. 2005, 21, 378. (305) Meyer, M. H. F.; Hartmann, M.; Keusgen, M. Biosens. Bioelectron. 2006, 21, 1987. (306) Lee, H. J.; Nedelkov, D.; Corn, R. M. Anal. Chem. 2006, 78, 6504. (307) Hwang, S. Y.; Yoo, C. H.; Jeon, J. Y.; Choi, S. C.; Lee, E. K. Biotechnol. Bioprocess Eng. 2005, 10, 309. (308) Rodriguez-Mozaz, S.; de Alda, M. J. L.; Barcelo, D. Anal. Bioanal. Chem. 2006, 386, 1025. (309) Minunni, M.; Mascini, M. Anal. Lett. 1993, 26, 1441. (310) Farre, M.; Martinez, E.; Ramon, J.; Navarro, A.; Radjenovic, J.; Mauriz, E.; Lechuga, L.; Marco, M. P.; Barcelo, D. Anal. Bioanal. Chem. 2007, 388, 207. (311) Lim, T. K.; Oyama, M.; Ikebukuro, K.; Karube, I. Anal. Chem. 2000, 72, 2856. (312) Mauriz, E.; Calle, A.; Manclus, J. J.; Montoya, A.; Escuela, A. M.; Sendra, J. R.; Lechuga, L. M. Sens. Actuators B 2006, 118, 399. (313) Mauriz, E.; Calle, A.; Abad, A.; Montoya, A.; Hildebrandt, A.; Barcelo, D.; Lechuga, L. M. Biosens. Bioelectron. 2006, 21, 2129. (314) Mauriz, E.; Calle, A.; Lechuga, L. M.; Quintana, J.; Montoya, A.; Manclus, J. J. Anal. Chim. Acta 2006, 561, 40. (315) Mauriz, E.; Calle, A.; Manclus, J. J.; Montoya, A.; Hildebrandt, A.; Barcelo, D.; Lechuga, L. M. Biosens. Bioelectron. 2007, 22, 1410.

Chemical Reviews, 2008, Vol. 108, No. 2 493 (316) Gobi, K. V.; Tanaka, H.; Shoyama, Y.; Miura, N. Sens. Actuators B 2005, 111, 562. (317) Kim, S. J.; Gobi, K. V.; Tanaka, H.; Shoyama, Y.; Miura, N. Chem. Lett. 2006, 35, 1132. (318) Shankaran, D. R.; Matsumoto, K.; Toko, K.; Miura, N. Electrochemistry 2006, 74, 141. (319) Shankaran, D. R.; Gobi, K. V.; Sakai, T.; Matsumoto, K.; Imato, T.; Toko, K.; Miura, N. IEEE Sensors J. 2005, 5, 616. (320) Shankaran, D. R.; Kawaguchi, T.; Kim, S. J.; Matsumoto, K.; Toko, K.; Miura, N. Anal. Bioanal. Chem. 2006, 386, 1313. (321) Matsumoto, K.; Torimaru, A.; Ishitobi, S.; Sakai, T.; Ishikawa, H.; Toko, K.; Miura, N.; Imato, T. Talanta 2005, 68, 305. (322) Shankaran, D. R.; Matsumoto, K.; Toko, K.; Miura, N. Sens. Actuators B 2006, 114, 71. (323) Kumbhat, S.; Shankaran, D. R.; Kim, S. J.; Gobi, K. V.; Joshi, V.; Miura, N. Chem. Lett. 2006, 35, 678. (324) Larsson, A.; Angbrant, J.; Ekeroth, J.; Mansson, P.; Liedberg, B. Sens. Actuators B 2006, 113, 730. (325) Gobi, K.; Tanaka, H.; Shoyama, Y.; Miura, N. Biosens. Bioelectron. 2004, 20, 350. (326) Kim, S. J.; Gobi, K. V.; Harada, R.; Shankaran, D. R.; Miura, N. Sens. Actuators B 2006, 115, 349. (327) Miura, N.; Sasaki, M.; Gobi, K. V.; Kataoka, C.; Shoyama, Y. Biosens. Bioelectron. 2003, 18, 953. (328) Gobi, K.; Miura, N. Sens. Actuators B 2004, 103, 265. (329) Gobi, K. V.; Kataoka, C.; Miura, N. Sens. Actuators B 2005, 108, 784. (330) Ock, K.; Jang, G.; Roh, Y.; Kim, S.; Kim, J.; Koh, K. Microchem. J. 2001, 70, 301. (331) Wu, C.-M.; Lin, L.-Y. Biosens. Bioelectron. 2004, 20, 864. (332) Soh, N.; Watanabe, T.; Asano, Y.; Imato, T. Sensor Mater. 2003, 15, 423. (333) Soh, N.; Tokuda, T.; Watanabe, T.; Mishima, K.; Imato, T.; Masadome, T.; Asano, Y.; Okutani, S.; Niwa, O.; Brown, S. Talanta 2003, 60, 733. (334) Matsumoto, K.; Sakai, T.; Torimaru, A.; Ishitobi, S. J. Fac. Agric. Kyushu UniV. 2005, 50, 625. (335) Shimomura, M.; Nomura, Y.; Zhang, W.; Sakino, M.; Lee, K. H.; Ikebukuro, K.; Karube, I. Anal. Chim. Acta 2001, 434, 223.

CR068107D

Chem. Rev. 2008, 108, 494−521

494

Nanostructured Plasmonic Sensors Matthew E. Stewart,† Christopher R. Anderton,† Lucas B. Thompson,† Joana Maria,‡ Stephen K. Gray,§ John A. Rogers,†,‡ and Ralph G. Nuzzo*,†,‡ Department of Chemistry, University of Illinois at Urbana−Champaign, Urbana, Illinois 61801, Department of Materials Science and Engineering, University of Illinois at Urbana−Champaign, Urbana, Illinois 61801, and Chemistry Division and Center for Nanoscale Materials, Argonne National Laboratory, Argonne, Illinois 60439 Received June 2, 2007

Contents 1. Introduction 2. Theoretical Considerations: Optical Properties of Metal Nanoparticles and Nanoholes 3. Synthesis and Fabrication of Plasmonic Nanostructures 3.1. Solution-Phase Syntheses 3.2. Top-Down Lithography 3.3. Unconventional Lithographic Techniques 3.3.1. Nanosphere Lithography 3.3.2. Colloidal Lithography 3.3.3. Soft Lithography 4. Applications of Plasmonic Nanostructures in Sensing and Chemical Imaging 4.1. Colorimetric Sensing Based on Particle−Particle Coupling 4.2. Nanoparticle-Enhanced Surface Plasmon Resonance 4.3. Exploiting Rayleigh Scattering for Sensing and Imaging 4.4. Label-Free Optical Detection Based on Changes in Refractive Index 4.4.1. Nanoparticle Dispersions 4.4.2. Surface-Immobilized Nanoparticles 4.4.3. Periodic Nanohole Arrays 4.4.4. Random Nanohole Arrays 4.5. Surface-Enhanced Spectroscopies 4.5.1. Surface-Enhanced Raman Scattering 4.6. Plasmonics for Detection Beyond the Diffraction Limit 4.6.1. Plasmon-Enhanced Fluorescence 5. Concluding Remarks 6. Acknowledgments 7. References

494 495 498 498 499 500 500 500 501 502 502 503 504 506 506 506 508 512 512 513 514 515 515 516 516

1. Introduction Surface plasmons (SPs) are coherent oscillations of conduction electrons on a metal surface excited by electromagnetic radiation at a metal-dielectric interface. The growing field of research on such light-metal interactions * To whom correspondence should be addressed. Phone: 217-244-0809. Fax: 217-244-2278. E-mail: [email protected]. † Department of Chemistry, University of Illinois at Urbana-Champaign. ‡ Department of Materials Science and Engineering, University of Illinois at Urbana-Champaign. § Argonne National Laboratory.

is known as ‘plasmonics’.1-3 This branch of research has attracted much attention due to its potential applications in miniaturized optical devices, sensors, and photonic circuits as well as in medical diagnostics and therapeutics.4-8 Plasmonics is also a highly active area due, in part, to recent advances in nanofabrication methodologies.9-12 These methodologies have led to the realization of metal nanostructures composed of nanoparticles (NPs),13 nanoholes,14 and other components15 with precisely controlled shapes, sizes, and/ or spacings.16,17 Such exquisite synthetic control in combination with advances in theory and the emergence of quantitative electromagnetic modeling tools has provided a better understanding of the optical properties of isolated and electromagnetically coupled nanostructures of various sizes and shapes. In addition to control over the geometry and optical properties of nanostructures,18,19 various strategies for modifying the surfaces of these materials make it possible to effect the selective binding and detection of specific targets for chemical and biological sensing.20,21 Detection schemes based on techniques that utilize plasmons experience enhancements that are commensurate with the magnitudes of the associated electric fields. These enhancements lead to new competencies for chemical sensing that are both useful and extraordinarily sensitiveswith detection levels in some cases reaching that of single molecules.22 As the field of plasmonic-based sensing grows, it is understood that the explicit control of nanostructured components will continue to provide techniques with unparalleled sensitivity, improved ease of fabrication, and thus enhanced utility outside of the laboratory. Two types of surface plasmon resonances (SPRs) are used in surface-based sensing: (i) propagating surface plasmon polaritons (SPPs) and (ii) nonpropagating localized SPRs (LSPRs) (The terms ‘propagating’ and ‘nonpropagating’ are used here to describe evanescently confined surface plasmons. It should be kept in mind that a ‘nonpropagating’ LSPR is intimately coupled to ordinary or nonevanescent propagating light since the LSPR is excited by and scatters such light.). SPPs can be excited on thin metal films using grating or prism couplers.23 These plasmons propagate tens to hundreds of micrometers along the metal surface with an associated electric field that decays exponentially from the surface (normal to the dielectric-metal interface).24 Changes in the refractive index above the metal shifts the plasmon resonance condition, which can be detected as intensity, wavelength, or angle shifts in sensing applications.25 SPR sensors that utilize propagating SPPs are covered in an article by Homola in this issue of Chemical ReViews. LSPRs are

10.1021/cr068126n CCC: $71.00 © 2008 American Chemical Society Published on Web 01/30/2008

Nanostructured Plasmonic Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 495

Matthew Stewart received his B.S. degree in Chemistry from Wittenberg University (Springfield, OH) in 2002, graduating summa cum laude with both departmental and university honors. In the summer of 2003 he worked as a flavor chemist at Givaudan, developing yeast blends, conducting accelerated aging experiments, and testing production lots for quality control. He joined Professor Ralph Nuzzo’s group at the University of Illinois at Urbana−Champaign in the fall of 2003 to pursue his Ph.D. degree in Analytical Chemistry. His research interests include plasmonics, microfabrication, nanostructured materials, biomolecular−surface interactions, and microanalytical systems.

Lucas B. Thompson was born on March 27, 1981, in Cleveland, OH. Lucas received his B.A. degree from The College of Wooster (Wooster, OH) in 2003. While at Wooster, he computationally modeled the reconstruction of adlayers on metal surfaces under the supervision of Professor Wingfield V. Glassey. In the fall of 2003 he joined Professor Ralph Nuzzo’s group at The University of Illinois at Urbana−Champaign to pursue his Ph.D. degree in Materials Chemistry. His research interests include collecting and analyzing cellular release with microfluidic devices, SPR as an analytical tool for selectively sensing biomolecules, and nanoparticle-enhanced polymer and hydrogel sensors.

Christopher Anderton was born in Colorado Springs, CO. He received his Bachelors of Science degree in Chemistry with highest distinction at the University of Colorado, Colorado Springs (UCCS). While at UCCS he worked under Professor David Weiss, researching flavonoid activity in many commercially available teas and green tea supplements. He also completed two NSF Research Experiences for Undergraduates under Professor Donald Dittmer at Syracuse University and under Professor Francisco Zaera at the University of California, Riverside. He is a third year analytical chemistry graduate student at the University of Illinois at Urbana−Champaign under Professor Ralph Nuzzo. His research interests include plasmonics, plasmonic enhancement from nanomanipulations, and microanalytical systems.

Joana Maria was born and raised in Lisbon, Portugal. She graduated with a degree in Engineering Physics from the New University of Lisbon and is currently pursuing her Ph.D. degree in Materials Science and Engineering under the guidance of Professor John A. Rogers at the University of Illinois at Urbana−Champaign. She does research in soft optical lithography and plasmonic sensors.

nonpropagating plasmon excitations that can be resonantly excited on metal NPs and around nanoholes or nanowells in thin metal films. The spectral position and magnitude of the LSPR depends on the size, shape, composition, and local dielectric environment.26,27 This latter property has been exploited for label-free optical sensing where adsorbateinduced refractive index changes near or on plasmonic nanostructures are used to monitor binding events in real time.28 Electromagnetic field enhancements also accompany these plasmonic resonances, which are used for performing surface-enhanced spectroscopies. This review focuses mostly, although not exclusively, on LSPRs and their use in chemical29 and biological30 sensing and surface-enhanced spectroscopies.31

2. Theoretical Considerations: Optical Properties of Metal Nanoparticles and Nanoholes Gold (Au) and silver (Ag) metal NPs are frequently studied because they can exhibit strong SPRs in the visible wavelength range.32 At these wavelengths, their optical properties are best described by a complex, wavelength-dependent dielectric constant

(λ) ) r(λ) + ii(λ)

(1)

where  ) m2 and m ) n + ik is the complex refractive index given as a function of the refractive index, n, and the absorption coefficient, k. Noble-metal NPs can support LSPRs33 when the incident photon frequency is resonant with the collective oscillation of the conduction electrons confined in the volume of the NPs (Figure 1).34 The simplest type of LSPR is a dipolar LSPR, which can be viewed as following in the limit of the particle’s diameter, d, being much smaller than the wavelength of the incident light, λ (d , λ). The

496 Chemical Reviews, 2008, Vol. 108, No. 2

Stewart et al.

Stephen Gray was born in Sherbrooke, Que´bec, Canada. He received his B.Sc. (Hons) degree in Chemistry from Ottawa’s Carleton University in 1977. Carrying out undergraduate thesis research work with Professor James Wright inspired him to pursue a career in theoretical chemistry. He went on to obtain his Ph.D. degree in Chemistry from the University of California at Berkeley in 1982 with Professor William Miller as his Ph.D. advisor. This was followed by post-doctoral work with Professor Mark Child at Oxford University and Professor Stuart Rice at The University of Chicago. He was an Assistant Professor of Chemistry at Northern Illinois University in DeKalb from 1986 to 1990 before joining the scientific staff at Argonne National Laboratory. His research interests include the quantum dynamics of chemical reactions and electrodynamics of metallic nanostructures.

Ralph G. Nuzzo is the William H. and Janet G. Lycan Professor of Chemistry at the University of Illinois at Urbana−Champaign, where he also holds an appointment as a Professor of Materials Science and Engineering. He received his AB degree with High Honors and Highest Distinction in Chemistry from Rutgers College in 1976, where he was also recognized as a Henry Rutgers Scholar, awarded the Merck Prize for undergraduate research, and elected to Phi Beta Kappa. He earned his Ph.D. degree in Organic Chemistry from the Massachusetts Institute of Technology in 1980. He accepted the position of Member of Technical Staff in Materials Research at Bell Laboratories in Murray Hill, NJ, in 1980, where he was named a Distinguished Member of the Staff in Research in 1987sa title held until he left to join the Illinois faculty in 1991. He is a fellow of the American Academy of Arts and Sciences, the World Innovation Foundation, and the American Vacuum Society. In 2006 he was recognized by the Wall Street Journal Innovators Award for Semiconductors and the Adamson Award of the American Chemical Society in 2003 for original discoveries leading to the development of self-assembled monolayers. He currently serves as a Senior Editor of Langmuir as well as a member of numerous advisory boards.

John A. Rogers is Founder Professor of Engineering at the University of Illinois at Urbana−Champaign with primary appointment in the Department of Materials Science and Engineering, where his research includes materials for large area and unusual format electronics. From 1995 to 1997, he was a Junior Fellow in the Harvard University Society of Fellows. He joined Bell Laboratories as a Member of the Technical Staff in the Condensed Matter Physics Research Department at the end of 1997 and served as Director of this department from the end of 2000 to the end of 2002. He has been on the faculty at Illinois since January 2003. He has published more than ∼180 papers and is co-inventor on ∼70 patents and patent applications, more than 40 of which are licensed or in active use. His research has been recognized with many awards including most recently the Xerox Distinguished Lectureship (2006), the Dorn Lectureship at Northwestern University (2007), the 2007 Daniel Drucker Eminent Faculty Award, the highest honor from the University of Illinois College of Engineering, and the 2007 Baekeland Award from the American Chemical Society for outstanding achievement by a chemist under the age of 40. Rogers was elected a Fellow of the American Physical Society in 2006. He serves or has recently served on several Editorial Boards, including those for Applied Physics Letters, Journal of Applied Physics, and Nano Letters. He is Associate Editor of IEEE Transactions on Nanotechnology and SPIE Journal of Microlithography, Microfabrication and Microsystems.

Figure 1. Schematic illustration of a localized surface plasmon of a metal sphere showing the displacement of the electron charge cloud relative to the nuclei. Reprinted with permission from ref 34. Copyright 2007 by Annual Reviews.

conduction electrons inside the particle will all move in phase upon plane-wave excitation. This leads to the buildup of polarization charges on the particle surface that will act as a

restoring force, allowing a resonance to occur at a specific frequency known as the particle dipole plasmon frequency.33,35 A resonantly enhanced field builds up inside the NP, which in the small particle limit is homogeneous throughout its volume, while a dipolar field is produced outside. This results in strong light scattering, the appearance of intense surface plasmon absorption bands, and the enhancement of the near-field in the immediate vicinity of the particle surface. The spectroscopic responses of larger metallic NPs are modified due to the excitation of higher order modes such as quadrupoles and retardation and skin depth effects.33,35 The bandwidth, peak height, and position of the absorption maximum depend on the particle material, size, and geometry (Figure 2) and the dielectric function of the surrounding environment.33,36,37 Classical Mie theory38 corresponds to the rigorous analytical solution of Maxwell’s equations for the optical properties of a spherical particle. It assumes that the particle and the surrounding medium are homogeneous.33,35 Solutions have

Nanostructured Plasmonic Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 497

Figure 2. (a) Dark-field microscopy image and corresponding SEM images and (b) light scattering spectra of Au nanocrystals of different shapes. Reprinted with permission from ref 54. Copyright 2003 American Institute of Physics.

subsequently been developed for other particle shapes, including more general spheroidal forms;32,39-41 such solutions are also often referred to as “Mie Theory”. These analyses are greatly simplified when the particles are much smaller than the wavelength of light since only the lowest (dipolar) order of the Mie theory scattering coefficients need to be retained. In the long wavelength, electrostatic dipole regime, the extinction E(λ)sthe sum of absorption and scattering cross sectionssof a spheroid metallic NP is given by the following equation33

E(λ) ∝

[

i

]

(r + χmed)2 + i2

(2)

where med is the dielectric constant of the surrounding medium, λ is the excitation wavelength, χ is a form factor that describes the NP’s aspect ratio (χ ) 2 for a sphere and increases directly with the aspect ratio of the NP), and r and i are the real and imaginary parts of the dielectric function of the metallic NP, respectively. For a spherical metal NP with |r| . i, eq 2 has a maximum when r(ω) ≈ -2med and this is the LSPR. The position of this resonance red shifts with an increase in magnitude of the dielectric constant of the medium surrounding the NP due to the buildup of polarization charges on the dielectric side of the interface, which is responsible for the weakening of the total restoring force. For nonspherical metallic NPs, the surface plasmons are unevenly distributed around them, manifesting in a shape dependence of the LSPR absorption spectra.42 The plasmon resonance of metallic nanorods, for example, splits into two peaks: (i) a strongly red-shifted long axis or longitudinal mode (L ) polarization parallel to the long axis) and (ii) a slightly blue-shifted transverse mode (T ) polarization perpendicular to the long axis). As the aspect ratio of a nanorod increases, the separation between the two plasmon bands becomes more pronounced (Figure 3).16,43-51 Triangular metallic NPs exhibit multiple plasmon resonances, a longitudinal (bulk) plasmon mode, and very large field enhancements at their sharp tips.52,53 Although a generalization of the quasistatic approach to metallic NPs of arbitrary shape has been suggested and demonstrated good results,54

numerical methods such as the T-matrix method,55 the discrete dipole approximation,56,57 or finite-difference timedomain simulations58,59 usually have to be used to calculate the optical properties (resonant frequencies; local field enhancement at the NP surface; absorption, extinction, and scattering efficiencies) for these more complex cases. For larger metallic NPs beyond the Rayleigh approximation (d > 30 nm), the dipolar resonance red shifts and suffers a substantial broadening. The red shift arises due to a reduction of the depolarization field caused by retardation effects60 where the conduction electrons do not all move in phase, as is true for smaller NPs, which leads to a reduced depolarization field at the particle center generated by the surrounding polarized matter. Radiative losses61 also start to contribute significantly to the plasmon damping, eventually dominating it totally for Au and Ag NPs with diameters d > 100 nm, and for this reason can impact the analytical sensitivity measurements. The latter effect causes significant broadening of the resonance peak. Scattering processes at the NP’s surface are thought to begin to contribute to the total damping for NPs smaller than the free-electron scattering length.33 The depolarization field and additional damping mechanisms for large and small particles can be seen as lowest order corrections to the quasistatic theory leading to decreases in the total enhancement of the excitation field. Mie theory only applies to noninteracting NPs well separated in the solid state or present at low concentration in solution. Recent advances in particle synthesis and fabrication techniques, however, have allowed the assembly of ordered arrays of interacting metallic NPs, leading to interesting new optical responses.3,62-68 The plasmon resonances of interacting particles are split and shifted, depending on the polarization of the incident light, relative to those of noninteracting NPs.65-67 In such cases, each NP with a diameter much smaller than the wavelength of the exciting light acts as an electric dipole. Two types of electromagnetic interactions prevail in this context, depending on the spacing d between adjacent NPs: (i) near-field coupling and (ii) farfield dipolar interactions.62 Far-field dipolar interactions with a d-1 dependence dominate at NP spacings on the order of the wavelength (λ) of the exciting light,69 while near-field dipolar interactions with a dependence of d-3 dominate for spacings much smaller than λ.63 These distance dependences have important consequences for sensing based on the LSPR peak shifts caused by changes in the electromagnetic interactions that occur upon aggregation (or dissociation) of NPs, a feature that has been broadly exploited for colorimetric sensing.70 LSPRs analogous to metal NP LSPRs can be excited around nanoscale holes in thin metal films. This is not surprising in view of Babinet’s principle, which relates the diffraction properties of particles to holes.71 For example, Prikulis et al.72 demonstrated remarkable correlations between the light scattering of holes in metal films and light scattering by disk-shaped NPs. Nanoholes tend to exhibit somewhat broader scattering features due to the fact that, in addition to the possibility of exciting LSPRs, the holes can serve as point sources for SPP waves in the thin film.73,74 As noted in the Introduction, propagating SPP waves form the basis of the SPR sensors reviewed by Homola in this issue. In the case of periodic hole arrays in thin metal films that are reviewed by us here (section 4.4.3), the periodic analog of SPPs and related diffractive phenomena such as

498 Chemical Reviews, 2008, Vol. 108, No. 2

Stewart et al.

Figure 3. Transmission electron micrographs (top), optical spectra (left), and photographs (right) of aqueous solutions of Au nanorods of various aspect ratios. The seed sample has an aspect ratio of 1. Samples a, b, c, d, and e have aspect ratios of 1.35 ( 0.32, 1.95 ( 0.34, 3.06 ( 0.28, 3.50 ( 0.29, and 4.42 ( 0.23, respectively. Scale bars: 500 nm for a and b, 100 nm for c-e. Reprinted with permission from ref 16. Copyright 2005 American Chemical Society.

Wood’s anomalies75-77 are intimately entangled with LSPRs and so must also be factored into our discussion. We will consider these influences and their consequences for sensing applications in the sections below.

3. Synthesis and Fabrication of Plasmonic Nanostructures Formation of metal nanostructures has been an active area of research due in part to their growing importance in diverse applications including photonics and optoelectronics,2,3,6,78-82 electronics,83-86 chemical and biosensing,14,87-92 and medical diagnostics and therapies.93-97 It is now well appreciated that the optical, electronic, and catalytic properties of metal nanostructures can be tuned very broadly by controlling their size, shape, and composition.25,35,98-101 This has resulted in a wealth of literature on synthetic methodologies for generating isotropic and anisotropic nanostructures with well-controlled sizes and shapes from a variety of materials.16,17,93,100-104 Nanostructures are typically formed using either so-called top-down or bottom-up approaches.10 Top-down techniques involve using various forms of conventional lithographic techniques to pattern nanostructures (e.g., onto planar substrates),65-67,105 whereas bottom-up methods exploit the interactions of atoms, molecules, or more complex mesoscale objects, in conjunction with the controlling influences of process kinetics, to “assemble” nanostructures either on substrates or in solution.72,106-111 The following sections will describe in more detail the use of these approaches for synthesizing representative metal nanostructures that support propagating and/or localized plasmons and exhibit interesting/useful optical properties.

3.1. Solution-Phase Syntheses Bottom-up solution-phase synthesis is a versatile approach to forming NPs that allows control over their size,112-117 shape,17,101 composition,98,118-120 and structure (e.g., solid or shell).95,121,122 This approach generally involves the reduction of metal salts in a solution containing an appropriate stabilizer to control the growth and suppress the aggregation of the NPs.123,124 The stabilizerscommonly ligands, surfactants, ions, organic acids, or polymerssadsorbs or coordinates to the surface of the NPs and inhibits aggregation by Coulombic

repulsion20,114,125 and/or steric hindrance.124,126,127 Reduction of the metal salt can be carried out electrochemically,15,128-132 photochemically,114,133-136 sonochemically,137-139 or using chemical reductants such as citrate,114,117,140 hydrides,141,142 alcohols,143,144 hydrogen,104 hydroxylamine,145,146 or hydrazine.147,148 The specific choice of reductant, stabilizer, temperature, and relative concentrations of the reagents can all affect the size and shape of the NPs. Recent work has investigated biosynthetic approaches149-153 and other environmentally friendly methods of synthesizing NPs.126,154,155 This topic has been reviewed in a recent issue of Chemical ReViews.156 Solution-phase synthesis tends to yield approximately spherical particles since the lowest surface energy shape is that of a sphere.4,100 (These particles have facets and should be more correctly called ‘quasi-spheres’,4 but will be referred to here as spherical particles for simplicity.) One of the most commonly used procedures for making spherical Au NPs is the citrate reduction of HAuCl4 in an aqueous solution, which was first reported by Turkevitch et al. in 1951.157 In this protocol the citrate acts as both a reducing agent and an electrostatic stabilizer, and the size of the NPs can be tuned by controlling the citrate to HAuCl4 ratio.113,158 Citrate can also be employed in the synthesis of Ag particles.114 Another broadly adopted solution-phase synthesis that yields highly stable, monodisperse thiol-protected particles with controllable sizes is the Brust-Schiffrin method.159 This approach uses a biphasic synthesis (an aqueous and organic phase) and tetraoctylammonium bromide as a phase-transfer agent to reduce the [AuCl4]-1 anion with NaBH4 in the presence of alkanethiols to yield thiol-stabilized particles that are one to several nanometers in size.159 The particle sizes can be controlled between 1.5 and 5.2 nm by adjusting the temperature, reduction rate, and thiol to Au ratio.13,112,156,160,161 The thiols on these so-called monolayer-protected cluster (MPCs) can be easily switched to introduce functionality,162,163 such as thiolated oligonucleotides164 or proteins,165 through simple place-exchange reactions.166-169 Other ligands such as amines and phosphines can also be used to stabilize and control the growth of NPs,170-174 and more recent singlephase adaptations of the Brust method have eliminated the need for phase-transfer agents.170,175-177

Nanostructured Plasmonic Sensors

Well-defined NPs can also be fabricated using seedmediated growth techniques.99,102,145,146,178-182 Nanorods are commonly prepared using this approach,100,102,178,179 although they can also be prepared electrochemically49,183-185 and photochemically.133,186 In seed-mediated syntheses, a stable growth solution is prepared containing a metal salt, a mild reducing agent (e.g., ascorbic acid), a surfactant molecule (e.g., cetyl-trimethylammonium bromide, CTAB), and possibly other additives.45,46,100,179,187 A NP ‘seed’, the nucleating agent, is added to the solution and the metal salt is reduced directly on the surface of the seed with minimal nucleation occurring in solution.178 The surfactant acts as an organic micellar template for anisotropic growth. The morphology, aspect ratios, and synthetic yields of the seeding approach are controlled by the composition and concentration of the surfactant, additives, seeds, metal salts, and reducing agent.45,46,100,179,188-191 Seeding growth can also be used to grow Au and Ag nanorods directly on surfaces,192-194 where the initial seed concentration affects the resulting nanorod aspect ratios and size distributions.195 As noted above, nanorods exhibit two distinct plasmon resonance modes: one associated with electron oscillations parallel to the longitudinal axis and the other with electron oscillations parallel to the transverse axis.16,43-51 The longitudinal plasmon mode can be tuned by adjusting the aspect ratio of the nanorod, making it a particularly useful structure for applications in photonics and biotechnology.16,43-51 Au nanoshells are another type of metal nanostructure that have highly tunable plasmon resonances.95,196 Halas and coworkers formed such core-shell structures by seeded metallization of colloidal silica spheres.197 The silica particles are functionalized with a monolayer of an amine-terminated silane (aminopropyltriethoxysilane (APTES)), which is subsequently used to bind small colloidal Au particles. After this initial seeding, more Au is deposited by an electroless plating procedure. The thickness of the final Au shell can be controlled by adjusting the initial Au seed coverage and the amount of reductant used in the plating process.197,198 The plasmon resonance frequency is tuned by adjusting the relative sizes of the core and shell dimensions (Figure 4).95,196 This synthetic approach has also been used to form a socalled nanorice structure,199 prolate spheroidal NPs consisting of a hematite core and a Au shell. The polyol process is a highly versatile synthetic procedure that can be used to form metal and alloy NPs with a variety of shapes, sizes, and optical properties.17,101,144 This synthesis uses a polyol such as ethylene glycol as both a solvent and reducing agent (at elevated temperatures) for a metal salt precursor. Xia and co-workers used the polyol synthesis to produce a variety of Ag nanostructures with well-defined shapes (including cubes, rods, wires, or spheres) by adjusting the relative amounts of the capping agent (poly(vinyl pyrrolidone)) and the precursor salt (AgNO3) in solution.17,101,144,200,201 Nanoplates and nanobelts are formed using a different capping agent (such as sodium citrate).17 Hollow Au or porous Ag/Au alloy nanostructures can also be produced using Ag NPs as a physical template in a wellknown galvanic replacement reaction.121,202 For example, Au nanocages have been generated by simply adding HAuCl4 to a suspension of Ag nanocubes.203 Adjusting the volume of the HAuCl4 solution added to the Ag nanocube suspension allows formation of a variety of different nanostructures such as hollow Au/Ag alloy nanoboxes and/or Au nanocages. The LSPR of these structures can be tuned from the visible to

Chemical Reviews, 2008, Vol. 108, No. 2 499

Figure 4. Visual demonstration of the tunability of metal nanoshells (top), and optical spectra of Au shell-silica core nanoshells (the labels indicate the corresponding Au shell thickness). Reprinted with permission from ref 95. www.tcrt.org.

near-infrared by varying the amount of HAuCl4 solution added to the Ag nanocube suspension.17,203 Surfactant aggregates such as micelles, reversed micelles, and microemulsions are also used to make restricted volume nanoreactors for the in-situ synthesis of NPs.204-209 The interested reader is referred to recent review articles on this synthetic approach204,205 as well as to a recent Chemical ReView article for more detailed descriptions of NP syntheses.13,156,210

3.2. Top-Down Lithography The size, shape, and interparticle spacing of surface-bound metallic nanostructures can be exquisitely controlled using scanning beam lithographies, such as electron beam lithography (EBL)67,105 and focused ion beam (FIB) lithography.211 This type of control is highly important for making reproducible substrates with tunable optical properties for conducting systematic studies of SERS212-214 and plasmonenhanced fluorescence.215 In EBL a tightly focused beam of electrons is scanned across a thin layer of resist (a radiationsensitive polymer), which makes it either more or less soluble in an organic developer solution. The patterned resist is then used as a sacrificial mask in subsequent etching or deposition processes to generate nanostructured metallic patterns with well-controlled geometries. EBL can be used to attain sub20 nm resolution using specialized resists such as hydrogen silsesquioxane (HSQ)216,217 or NaCl crystals218 or using more traditional organic resists such as poly(methylmethacrylate) (PMMA) in conjunction with ultrasonically assisted development.219 FIB is a related technique that uses a focused beam of ions (typically Ga+) to perform both additive and subtractive patterning by physical or chemically assisted processes. These include (1) FIB milling,211,220,221 (2) ion-assisted etching,222,223 and (3) FIB-induced deposition.224,225 FIB is capable of forming patterns with ∼10 nm resolution using either PMMA226 or inorganic resists.227 Various metallic nanostructures such as circular slits,221 nanoholes,228,229 slit

500 Chemical Reviews, 2008, Vol. 108, No. 2

Stewart et al.

Figure 5. Schematic illustration of hexagonally close pack colloidal crystal mask (left), and a representative AFM image of a triangular NP array (right). Reprinted with permission from ref 109. Copyright 2001 American Chemical Society.

gratings,230,231 and V-grooves6,232 have been fabricated by FIB for research in the growing field of plasmonics. While scanning beam lithographies are capable of precise control over the size, shape, and spacing of metallic nanostructures, more recent research has focused on unconventional lithographic techniques that are capable of patterning large areas in parallel at low cost.

3.3. Unconventional Lithographic Techniques 3.3.1. Nanosphere Lithography Nanosphere lithography (NSL) is an inexpensive and versatile hybrid bottom-up procedure, popularized by Van Duyne, for fabricating periodic arrays of metallic nanostructures on surfaces.109,233 This unconventional approach to nanofabrication is a variant of a technique originally named “natural lithography” where monolayers of nanospheres are used as deposition or etch masks.234,235 NSL extended natural lithography with the development of double-layer colloidal masks, which provide a degree of control over the size and shape of the resulting NPs.109 Single-layer NSL begins with the deposition of a singlelayer colloidal crystal mask of hexagonally close-packed latex or silica spheres on an appropriate substrate, which yields defect-free domains that are approximately 10-100 µm2 in size.109,236,237 The colloidal crystal contains triangular void spaces that are created between three neighboring particles, which make these crystals useful as masks in subsequent deposition or etching processes.91,109,238 In the case of additive lithography, a metal or other material is deposited through the mask at normal incidence by physical vapor deposition to produce a metallic film over nanosphere (MFON) structure.31,239,240 This approach has been used to produce Ag FON surfaces that are useful structures for applications based on surface-enhanced Raman spectroscopy (SERS).240-242 Alternatively, the nanosphere mask can be removed by sonicating in solvent to yield arrays of surface-bound triangular NPs with P6mm symmetry. These structures can be used for LSPR sensing and SERS (Figure 5).109,243-246 Nanostructured films composed of periodic spherical voids can also be formed using self-assembled polystyrene (PS) colloidal crystal masks. These ‘nanovoid arrays’ support delocalized Bragg and localized Mie plasmons247,248 and are formed by electrochemical deposition of Au through a singlelayer self-assembled colloidal template. The thickness of the metal is controlled by monitoring the current passed through the plating solution and the plating time.18,249 This allows

Figure 6. SEM images of nanovoid arrays with void diameters of 600 nm at three normalized thicknesses t-)0.2, 0.5, and 0.9 (left), and schematic illustrations of the surface at each thickness (right). Reprinted with permission from ref 247. Copyright 2006 The American Physical Society.

control over the geometry and thus the spectroscopic/ plasmonic properties of the nanostructured films.18,250 After completing the metal deposition step, the PS spheres are dissolved in tetrahydrofuran to yield a ‘nanovoid array’ (Figure 6). These structures have been used with some success as substrates for SERS.249-252 Self-assembled colloidal crystals also can be used as an etch mask to form arrays of submicrometer triangular pits in an underlying substrate.238 Anchored arrays of triangular metal NPs110 and metal films over nanowells253 can be formed using this subtractive processing technique. The anchored NPs are formed by depositing metal at normal incidence by physical vapor deposition on the colloidal mask and underlying etched substrate. The spheres are subsequently removed by sonication in an appropriate solvent, leaving behind an array of substrate-embedded triangular NPs.110 Alternatively, metal deposition can be performed after the spheres are removed from the etched substrate. This results in structured metallic surfaces that support a single, narrow plasmon resonance that exhibits sensitivity to external changes in refractive index.253 The attractive features of NSL include its low cost, versatility, and ability to produce well-ordered sub-100 nm array structures. The geometry and shapesand thus the optical propertiessof the particles can be tuned by moving the sample during metal deposition,254 annealing the latex colloidal mask prior to metal deposition,255 as well as changing the thickness of the deposited metal, size of the colloidal spheres, number of colloidal layers, or angle of the metal deposition.109,256,257 The size and shape of the NPs can also be changed in a controllable manner using electrochemistry to oxidize the substrate-bound particles.258 The surfacebound NPs formed by NSL can be released from the substrate into solution by adding surfactant and sonicating to generate both isolated particles and dimers.111

3.3.2. Colloidal Lithography Colloidal lithography72 is a versatile technique that can be used to form random arrays of nanoholes,259 nanodisks,260

Nanostructured Plasmonic Sensors

Figure 7. Schematic depiction of nanoring fabrication. (a) (1) Polystyrene colloidal particles are deposited by electrostatic selfassembly onto the substrate in a dispersed layer. (2) A 20-40 nm thick Au film is evaporated onto the particle-coated substrate at normal incidence. (3) Argon-ion beam etching is used to remove the Au film. During the etching, secondary sputtering creates a Au shell around the base of the polystyrene particles. (4) The remainder of the polystyrene particles are removed by UV-ozone treatment, resulting in free-standing Au nanorings on the substrate. SEM images of Au nanorings (150 nm diameter) made using a 40 nm thick sacrificial Au layer at normal incidence (b) and 80° tilt (c). Reprinted with permission from ref 383. Copyright 2007 American Chemical Society.

and nanorings.108 This unconventional nanofabrication technique involves adsorbing polystyrene (PS) particles onto a substrate via electrostatic self-assembly. The distance between the self-assembled spheres is governed by the particleparticle repulsion, which can be controlled by adjusting the electrolyte concentration of the colloidal solution.107 The randomly adsorbed particles are then used as a mask for subsequent evaporation and/or etching processes wherein the size of the sphere dictates the size of the resulting nanostructures. The main difference between colloidal lithography and NSL is that the colloidal spheres do not form an hcp monolayer on the substrate in colloidal lithography. Random nanoholes in Au films can be formed by evaporating metal on top of an assembled colloidal mask followed by lift off of the PS particles by tape stripping or boiling the sample in ethanol.259 Nanorings are formed on substrates in a similar process with the exception that prior to particle removal an Ar+-ion beam is used to etch the Au film, during which secondary sputtering creates Au shells around the bottom of the PS particles.108 The remains of the particles are then removed by an UV-ozone treatment, which leaves free-standing Au rings (Figure 7). Nanodisks can be formed in two ways.260,261 In one approach, nanodisks are formed by assembling a colloidal mask on top of a Au film followed by etching of the Au and removal of the colloidal mask.260 In the second approach, a colloidal mask is self-assembled on a PMMA film followed by deposition of a thin Au film. The Au-capped PS particles are then removed by tape stripping, leaving behind holes in the Au film. The exposed PMMA is then etched from the holes, and metal is deposited through the holes onto the substrate.

Chemical Reviews, 2008, Vol. 108, No. 2 501

Figure 8. Fabrication and structural characterization of large-area hole arrays. Preparation of free-standing films of subwavelength hole arrays (a). SEM image of a portion of a free-standing 100 nm Au film perforated with 250 nm holes (b). Scale bar: 500 nm. Optical micrograph of a free-standing film placed on a glass substrate (c). SEM images of representative areas of the film illustrating the uniformity of the nanohole array (d and e). The holes are spaced 1.6 × 2.4 µm. Scale bar: 2 µm. Reprinted with permission from ref 270. Copyright 2005 American Chemical Society.

The metal-coated PMMA film is removed by lift off, leaving a random array of metal disks on the substrate.261

3.3.3. Soft Lithography Soft lithography refers to a set of microfabrication techniques that use a structured elastomer as a stamp, conformable photomask, or mold to pattern a material of interest.262-264 The most commonly used elastomer, Sylgard 184 poly(dimethylsiloxane) (PDMS), has a low modulus, which limits the utility of this elastomer for patterning in the nanometer regime.265,266 The development of siloxane-based elastomers with larger moduli, such as the socalled hard PDMS (h-PDMS)266-268 and UV-curable PDMS (hV-PDMS),269 has extended the patterning ability of soft lithography to the nanometer regime. Composite stamps consisting of a thin layer of structured h-PDMS supported by a thicker planar layer of compliant PDMS are typically used for patterning at the nanoscale due to the mechanical instabilities and difficulty of achieving conformal contact with nonplanar surfaces using h-PDMS stamps. In one embodiment of soft lithography for fabricating plasmonic structures, Odom et al. used high-resolution composite PDMS stamps as conformable phase masks to generate large-area, free-standing 2D nanohole arrays in Au.270,271 A schematic illustration of the fabrication process is shown in Figure 8. In this process, an array of posts of positive photoresist with diameters of ∼250 nm is patterned first on a Si(100) wafer by phase-shifting photolithography using a conformable composite PDMS photomask.266,271,272

502 Chemical Reviews, 2008, Vol. 108, No. 2

Stewart et al.

thane film with a composite h-PDMS/PDMS stamp presenting surface relief features in the geometry of a square array of cylindrical posts. The composite PDMS stamp is pressed into the liquid polyurethane film and cured by UV light passed through the stamp. The stamp is then removed, and a thin layer of Au (∼50 nm) is uniformly deposited by e-beam evaporation onto the raised and recessed regions of the structured polyurethane film. This creates a Au film with an array of nanoscale holes (top surface) that is physically separated from a second level of Au disks at the bottom of the wells (Figure 9).

4. Applications of Plasmonic Nanostructures in Sensing and Chemical Imaging 4.1. Colorimetric Sensing Based on Particle−Particle Coupling

Figure 9. Plasmonic crystal fabrication process: (a) imprint, (b) cure, (c) remove stamp, and (d) Au deposition (top). SEM of a crystal and a high-magnification SEM (inset) that shows the upper and lower levels of Au (bottom). Reprinted with permission from refs 8 (top) and 77 (bottom). Copyright 2005 OSA (top) and Copyright 2006 The National Academy of Sciences of the USA (bottom).

A thin layer of chromium (Cr) is then deposited by electronbeam (e-beam) evaporation followed by removal of the photoresist posts. This yields an array of holes patterned in a thin film of Cr, which acts as an etch mask and as a sacrificial release layer to generate the free-standing nanostructured Au films. The exposed Si is then anisotropically etched using a KOH/isopropyl alcohol (IPA) solution to produce pyramidal-shaped voids directly below the Cr nanoholes. A layer of Au can then be deposited by e-beam evaporation to form a 2D nanohole array Au film on the top surface and pyramidal mesoscale Au particles in the lower pyramidal voids. The top Au nanohole array can be released by selectively etching the Cr film to form a free-standing structure (Figure 8), and the Au pyramids can be released by etching the Si with KOH/IPA. The material and chemical functionality of these nanostructures are easily controlled by depositing other metals or a combination of metals in a layerby-layer mode using e-beam evaporation.270 The periodic grating structure of the nanohole array allows the direct coupling of light to SPPs on these films.270,273 This fabrication protocol has been used to produce pyramidal metallic particles274 and particle arrays.275 Imprint lithography is another unconventional lithographic technique holding broad utility for patterning materials at the nanoscale over large areas.8,77,276-278 This rapidly emerging technology is used to replicate features on a hard or soft stamp in a thermoplastic or UV-curable polymeric material by embossing or molding (Figure 9). A metal film can be deposited on the resulting polymeric replicas to produce plasmonic structures useful for chemical and biosensing applications.8,77,279,280 In a recent example, large-area spatially coherent arrays of quasi-3D plasmonic crystals were formed by soft nanoimprint lithography and used for multispectral sensing and imaging of molecular binding events. The arrays were fabricated by embossing a thin UV-curable polyure-

Colorimetric detection is perhaps one of the most powerful and simple nanosensing methods available. In an exemplary model of this approach, Mirkin et al.70 reported an assay using oligonucleotide-functionalized Au NPs that exhibit strong red shifts upon aggregation in the presence of a complementary nucleotide (Figure 10).281 The color change in this case results from particle-particle plasmonic coupling as well as aggregate scattering281 and provides a “litmus test” method for determining nucleic acid targets. The optical properties of these assays are due to the aforementioned resonantly exited LSPRs of the NPs. The enhanced electronic fields are confined within a small area around the NPs (typically on the order of the particle radius) and decay approximately exponentially thereafter.65,66 As the distance between the NPs decreases, near-field coupling begins to dominate, leading to a strong enhancement of the localized electric field within the interparticle spacing producing pronounced red shifts of the LSPR frequency.65-67 Most reports of colorimetric assays exploit the LSPRs that develop on spherical Au NPs, but the method is also amenable to nonspherical particles21,282 as well as particles of other noble metals.283,284 Some benefits to using elongated particles include their inherent higher sensitivity to changes occurring in the local dielectric environment285 and the capability they afford for multiplexed detection schemes.286 Methods exploiting nucleotide interactions have been the most reported of these particle-particle coupling systems. This field of research has advanced rapidly, and biological analyses using oligonucleotide-modified Au NPs have been developed that achieve limits of detection (LOD) for a variety of analytes in the low-picomolar to mid-femtomolar range.281 Han et al. used oligonucleotide-functionalized Au NPs to determine the relative binding strengths of a variety of duplex and triplex DNA-binding molecules.287,288 Such data can offer insights into the activity of an array of possible anticancer drugs due to the correlations that can be made between ligand binding strength and biological activity. Kanaras and coworkers289 demonstrated the use of DNA-Au NP interactions for the determination of the enzymatic cleavage of DNA, in which a recognition site for the restriction endonulease EcoRI was designed to monitor the enzymatic cleavage activity, and consequent disassembly of the closely coupled Au NPs, via blue to red color shifts. Peptide nucleic acid (PNA) and aptamer-modified Au NPs have also been used to control assembly rates and aggregate sizes of Au NPs for sensing applications.87,290-293 The dissolution of aggregated NPs can also be used to monitor binding events.

Nanostructured Plasmonic Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 503

Figure 10. In the presence of complementary target DNA, oligonucleotide-functionalized Au NPs will aggregate (left), resulting in a change in the color of the solution from red to blue (right). Reprinted with permission from ref 281. Copyright 2005 American Chemical Society.

Figure 11. Schematic representation of colorimetric detection of adenosine. Absorbance spectra of the adenosine sensor before (blue) and 10 s after (red) addition of adenosine. Reprinted with permission from ref 87. Copyright 2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim.

In a specific example, Liu and Lu87 employed aptamerfunctionalized Au NP aggregates specific to adenosine and cocaine that become unstable when the target was bound, leading to a color shift from purple to red (Figure 11).87 Functionalized Au NPs that do not exploit nucleotide interactions also have been reported that provide a viable route to colorimetric detection.284,294-303 In a prototypical example, the ubiquitous biotin-avidin system linked to Au NPs was utilized to create a colorimetric assay for kinase inhibitors that could be applied to microplates amenable to parallel screening.301 Rapid detection of cholera toxin has been demonstrated using lactose-stabilized Au NPs, in which the cholera toxin binds to a lactose derivative inducing NP aggregation, which is detected visually by a red to purple color shift.299 In an expansion of the current methods for colorimetric detection, cation-specific functionalized Au NPs have been developed.88,282,297,304,305 The detection of Pb2+ is an area that has received much attention due to its physiological impacts. Demonstrating the use of DNAzymes in Au NP assembly for the detection of Pb2+, Liu and Lu304 optimized the DNA configuration for NP alignment, the NP size, Pb2+-specific DNAzyme concentration, and salt concentration, leading to a colorimetric detection method for Pb2+ that can be performed in the field in less than 10 min. In contrast to the previously mentioned method of sensing lead, Lin et al.297 used a crown ether specific to Pb2+ for colorimetric detection of Pb2+ using Au NPs ligated with a mixed monolayer of crown ether thiols and carboxylic acids.306 Recently, Lee and co-workers88 demonstrated a detection method for the mercuric ion (Hg2+) analogous to Liu and Lu,305 which takes advantage of the Hg2+ affinity toward T-T mismatches in DNA. Aside from the cation determination methods mentioned here, additional methods have been reported for determining other relevant charged molecules.296,302,303 All of these methods employ the plasmonic coupling inherent in aggregated NP systems and demonstrate that

detection of desirable biological compounds and other relevant molecules can be accomplished without the use of complex instrumentation. These methods are interesting in that they are sensitive at relevant time scales and that the NPs are easily functionalized to provide chemical and biological selectivity. The optical properties of NPs also have been exploited to enhance the sensitivity of conventional SPR systems as described below.

4.2. Nanoparticle-Enhanced Surface Plasmon Resonance As mentioned above, SPR spectroscopy is a surfacesensitive technique that can be used to detect refractive index changes that occur within the evanescent field of propagating SPPs excited at metal-dielectric interfaces.307 A change in refractive index shifts the plasmon resonance condition, which can be detected as intensity, wavelength, or angle shifts to provide quantitative information about the binding event.307-313 Small refractive index changes caused by the binding of low molecular weight analytes or small quantities of a larger analyte can challenge the sensitivity/detection limits of SPR spectroscopy.314-319 The sensitivity/detection limits can be improved by coupling the molecular recognition of analyte at the surface of the metal with another event that leads to larger changes in the SPR signal.314-319 This signal enhancement can be achieved using competitive inhibition or sandwich assays320-323 and enzymatic amplifaction.312,313,324,325 The SPR signal also can be enhanced by labeling the target analyte with dielectric or plasmonic NPs. These labels increase the refractive index at the metal surface and can electromagnetically couple to the flat metal film in the case of plasmonic NPs, leading to larger SPR shifts.314-319,326 Figure 12a shows a possible architecture for performing plasmonic NP-enhanced SPR immunoassays.314 In this approach a surface-immobilized antibody (e.g., anti-human

504 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 12. Schematic of plasmonic NP-enhanced SPR immunoassay (a). In situ SPR curves of an evaporated Au film modified with a-h-IgG(γ) (solid line) followed by sequential exposure to a 0.045 mg/mL solution of h-IgG (‚‚‚) and a 8.5 mg/mL solution of a-h IgG(Fc) (- -) (b). A film modified with a-h-IgG(γ) (solid line) followed by sequential exposure to a 0.045 mg/mL solution of h-IgG (‚‚‚) and a solution of a-h-IgG(Fc)-10-nm Au colloid conjugate (- -) (c). Reprinted with permission for ref 314. Copyright 1998 American Chemical Society.

immunoglobulin G (a-h-IgG)) is sequentially exposed to antigen (e.g., h-IgG) and a solution of secondary antibodyconjugated Au NPs (e.g., a-h-IgG/Au NPs). The presence of the Au NP leads to an enhancement of the measured SPR shift as shown in Figure 12b and c. Figure 12b shows that the shift of the SPR peak upon binding the secondary antibody alone is small compared to the shift observed when the secondary antibody is conjugated to a Au NP.314 The particle size, composition, and surface coverage as well as the substrate metal and the distance between the substrate and NP can affect the observed signal enhancement.314,315,327-329 This approach has been used not only for performing immunoassays314,330 but also for detecting DNA hybridization,319,331 protein conformational changes,332 small molecule binding interactions,317,318 and single nucleotide polymorphisms.316 Recently, Fang et al. reported the detection of attomoles of microRNA using NP-enhanced SPR imaging.333 In addition to flat film SPR, NPs can also be used to enhance the peak shifts obtained from transmission LSPR sensing using Au island films.334-336 This platform has been used with responsive polymer brushes, surface-immobilized singlestranded DNA, and molecularly imprinted polymers to detect pH changes,334 DNA hybridization,336 and cholesterol,335 respectively.

4.3. Exploiting Rayleigh Scattering for Sensing and Imaging Biosensing based on the optical scattering properties of plasmonic NPs is regarded as a potentially more powerful, yet remains a less widely exploited, modality of NP sensing than the extinction-based colorimetric assays described above.337,338 A single 80 nm Au NP, for example, exhibits a light scattering power equivalent to the emission of ∼106 fluorescein molecules.339,340 Unlike molecular fluorophores, however, plasmonic NPs do not undergo photobleaching or blinking. These are enabling distinctions that facilitate long-term single-particle measurements and tracking.285,341-343 These optical properties have been exploited in a variety of applications where NPs are used for sensing285,344 and as labels for immunoassays and DNA

Stewart et al.

Figure 13. Colorimetric detection of nucleic acids using scattered light (a). Step 1: DNA-Au NP probes (A and B) are hybridized to a DNA target in solution. Step 2: The samples are spotted onto a glass slide, which is illuminated with white light in the plane of the slide. The evanescent induced scatter from the Au NPs is visually observed. Individual 40 to 50 nm diameter Au probes scatter green light, whereas complexed probes scatter yellow to orange light because of a plasmon band red shift (b). Reprinted with permission from ref 352. Copyright 2004 Nature Publishing Group.

microarrays,345-348 imaging contrast agents,95,349 and molecular rulers.341,350,351 As noted in section 4.1, the changes in extinction (absorption and scattering) that occur due to NP aggregation can be used to detect the presence of a specific target DNA or protein.70,287,288,301 Monitoring only the changes in the scattering properties of NPs upon target-induced aggregation can also be used to detect target DNA.344 This method of sensing was demonstrated by immobilizing two different oligonucleotide fragments (A′ and B′, both 15 bases long), separately, on 13 nm Au NPs to create two sets of probe particles that aggregate in the presence of a complementary 30 base long target DNA (AB). Changes in the scattered light intensity upon target-induced aggregation were measured using a commercially available spectrofluorimeter. This approach was used to detect target DNA at picomolar concentrations and allowed the detection of single-nucleotide polymorphisms (SNPs) without the need for temperature control. Even lower concentrations of target DNA were detected by Muller et al.352 using 50 nm Au particles in conjunction with a light-scattering-based spot test. In this work, a target sequence was added to a solution containing oligonucleotidemodified Au NPs to induce hybridization and aggregation of the probes. An aliquot of the solution was then spotted on a glass slide (a planar waveguide) that was illuminated using a planar fiber optic illuminator (Figure 13a). The evanescently coupled light was scattered from the particles at the surface of the waveguide and imaged with a complementary metal-oxide-semiconductor (CMOS) camera. Au NPs not exposed to target DNA scattered green light (control), whereas Au NPs aggregated in the presence of the target DNA scattered orange light due to particle-particle coupling (Figure 13b). This approach allowed the detection of femtomolar concentrations of target DNA without the need for PCR or signal amplification.352 The high sensitivity and simple readout make this approach highly promising for use in point-of-care molecular diagnostics.

Nanostructured Plasmonic Sensors

Figure 14. Schematic of DNA hybridization to microarrays and detection using Ag-amplified Au NP probes. Reprinted with permission from ref 346. Copyright Elsevier B.V.

Wavelength-ratiometric plasmon light scattering is another sensing method that measures changes in scattered light to detect target-induced aggregation or dissociation of NPs in solution.353,354 In this approach, the ratio of the scattered light intensity at two wavelengths is used to quantify a target analyte, such as glucose.353 Other ‘ratiometric’ methods of biosensing include measuring the ratio of scattered light intensity355 or polarization356 at two different angles as a function of target-induced aggregation or dissociation of plasmonic NPs. The benefits of ratiometric approaches are that the measurements become independent of source and/ or detector fluctuations and NP concentration, and in this way improve the analytical stability of protocols relative to single-wavelength measurements. Fluorescent tagging is the most common method of labeling targets for optical detection; however, the demand for greater sensitivity and simplicity (by removing the need and thus cost and complexity of PCR oligonucleotide target amplification) has led to research on and development of alternative labels, such as plasmonic NPs.345-348 As an example, the large scattering cross section of plasmonic NPs has also led to their use as alternative labels in microarraybased technologies.345-348 Microarrays are made by immobilizing biologically relevant moieties, such as DNA or proteins, as discrete spots, typically 10-500 µm in size, on a substrate.345 These spatially arranged capture agents are then used to bind targets through specific interactions such as hybridization (nucleic acids) or ligand-receptor binding (proteins) to allow multiplexed detection of multiple analytes in a complex solution. Plasmonic NP labels have been shown to be very promising for improving the sensitivity of microarray-based analyses.345-348 Muller et al.347 recently used 15 nm oligonucleotide-modified Au NPs to label captured target DNA on a microarrayed glass slide using a three-component sandwich assay (Figure 14). Once the Au NPs were immobilized, a Ag amplification step was used to increase the target signal by electroless reduction of Ag ions to metal at the surface of the Au NPs.346 The glass slide was then used as a waveguide, and evanescently coupled light scattered from the NPs was imaged. The Ag-amplified Au NPs provided an approximately 1000-fold increase in sensitivity compared to the Cy3 fluorescent labels commonly used in microarray analyses.346 This protocol was used to detect femtomolar concentrations of target sequences in human genomic DNA samples without prior PCR amplification.347 A similar procedure was used to perform multiplexed SNP genotyping in total human genomic DNA without prior

Chemical Reviews, 2008, Vol. 108, No. 2 505

complexity reduction or target amplification, which is a major advancement toward point-of-care diagnostic medical applications.348 It is interesting to note that these light scattering measurements were performed using commercially available systems357 that can achieve attomolar and zeptomolar sensitivity when used in conjunction with immobilized transcyclopentane-modified peptide nucleic acid capture strands358 or bio-bar-code techniques,359 respectively. The large scattering cross section339 and ability to tailor the scattered wavelength of Au NPs,95 in conjunction with their biocompatibility198 and the availability of well-characterized surface conjugation chemistries,360 also makes them attractive candidates for use as contrast agents for imaging applications.95,349 Such optical measurements provide a promising route to noninvasive, high-resolution diagnostic imaging of tissues and cells with high sensitivity and chemical specificity.95,350,361 For example, cancer cells can be labeled by conjugating plasmonic NPs to antibodies that target a protein that is overexpressed by cancerous cells. AntiHER293,94 and anti-EGFR362-364 are commonly used antibodies that target epidermal growth factor receptors (EGFRs),96,364,365 which are transmembrane glycoproteins that are overexpressed in many types of cancers such as cervical, bladder, breast, lung, and oral cancers.365 The cells or tissues are incubated with the antibody-conjugated plasmonic NPs, and the labeled cells are then examined using an appropriate form of optical imaging (e.g., dark field microscopy, twophoton luminescence, etc.).97,362-364 Both spherical Au NPs362-364 and nanorods96,97 have been conjugated with antibodies for use in imaging a variety of cancer cells. Recently, Sokolov et al. used multifunctional Au NPs that incorporated both cytosolic delivery and targeting moieties for real-time intracellular imaging of biomarkers in live cells.366 The authors used water-soluble 20 nm Au NPs formed by citrate reduction that were functionalized with (i) TAT-HA2 peptides, (ii) anti-actin antibodies, and (iii) 5000 molecular weight PEG-SH (poly(ethylene glycol)thiol)). The first component serves two purposes: (a) the TAT protein transduction domain induces endocytic uptake of the functionalized Au NPs into the cell, and (b) the HA2 protein destabilizes the endosomal lipid membrane, which causes release of the NPs into the cytosol.367 The second component allows the NPs to bind to actin, and the third component improves the biocompatibility of the multifunctional particles. NIH3T3 fibroblasts were labeled with these probes and imaged using dark field reflectance microscopy. Actin labeling by these intriguing plasmonic probes was observed as an increase in red scattering due to dipole-dipole coupling between the NPs, and live cell imaging was used to track actin rearrangement. While the scattering properties of NPs are useful for labeling and optically detecting cancer cells, their large optical absorption cross section can also be exploited for photothermal therapeutic treatments.93,94,96 The treatment begins with the labeling of cancerous cells with molecular probe-conjugated plasmonic NPs. The labeled cells are then irradiated with light, which is absorbed by the NPs and converted to heat. The local heating causes irreversible damage and kills the cancerous cells. Research has focused on developing nanostructures with absorption peaks in the near-infrared (NIR) due to the transmissivity of blood and tissue at these wavelengths.198,368 Au nanoshells,94 nanocages,93 and nanorods96 with NIR absorption peaks have all been used to demonstrate this mode of photothermal treat-

506 Chemical Reviews, 2008, Vol. 108, No. 2

ment, although spherical Au NPs with absorption bands in the visible region of the electromagnetic spectrum have also been used.369

4.4. Label-Free Optical Detection Based on Changes in Refractive Index Plasmons are exceptionally sensitive reporters for chemical phenomena that influence the refractive index of the local environment of a probe. SPR spectroscopy307,370,371 and imaging24,28,313,333,372,373 are well-known label-free optical detection methods that use this property to monitor surface binding events in real time (as discussed in a separate contribution in this issue). In these techniques, changes in the dielectric environment at the surface of a flat or periodically structured noble-metal film shift the observed SPR resonance, which can be measured using angular interrogation, wavelength interrogation, or intensity measurements. This well-established technique enables determination of kinetic and thermodynamic data for a wide variety of molecular binding events, especially those involving biomolecular targets.371,374,375 Progress is currently being made to develop nanostructured plasmonic materials for performing similar analyses. Changes in the dielectric environment of nanostructured metals, such as NPs, results in measurable shifts of the LSPR peak position and/or magnitude that can be used to perform label-free chemical or biosensing in real time.342,343 A variety of noble-metal nanostructures such as NPs in solution26,286 or immobilized on surfaces,5,29,376 nanoholes and nanohole arrays,8,14,77,90,377,378 and nanoisland films379-381 have been used in this way. Substrate-bound nanostructures offer several attractive features as platforms for chemical sensing. These include (i) the shape, size, composition, and spacing of the NPs can be readily controlled to provide tunable peak positions and widths65,77,105,109,256 and (ii) the NPs are free of the capping agents or stabilizers used in solution-phase NP synthesis, making their surfaces readily accessible for functionalization with specific receptors or ligands.77,92,382,383 In the following discussion we will provide some examples of chemical and biosensing based on refractive index changes near the surface of nanostructured metals in solution and on substrates, where the LSPR is monitored using extinction or scattering spectroscopy. The refractive index sensitivity of each sensing platform will also be given where validated data is available. This sensitivity is commonly defined in terms of the change in an experimentally measurable parameter (typically peak position or magnitude) per ‘refractive index unit’ (RIU), which corresponds to a change of 1 in the refractive index. These measurements are usually performed by taking spectra of a plasmonic nanostructure in solutions of increasing refractive index while monitoring peak position or intensity changes.

4.4.1. Nanoparticle Dispersions Au nanostructures are known to exhibit strong plasmonic bands that are dependent on their shape, size, and surrounding media.26,286,384 Ghosh et al.26 studied the effects of changing solvents and ligands on the LSPR of Au NPs dispersed in solution. It was found that the surface plasmon absorption maximum of the Au NPs varied between 520 and 550 nm, depending on the refractive index and chemical nature of the surrounding solvent. The authors found that the LSPR peak red shifted linearly with the refractive index of the solvent when using solvents that do not possess active

Stewart et al.

functional groups that could complex with the surface of the Au NPs. A nonlinear relationship between the LSPR peak position and refractive index was found, however, when using solvents with nonbonding electrons capable of complexing to the surface of the Au NPs. Interestingly, these authors found that the LSPR peak position blue shifted ∼3 nm for every one carbon atom when the NPs were dispersed in alcohols with varying linear carbon chain lengths.26 This rather atypical trend reverses in the presence of more strongly coordinating ligands. In such cases, the magnitude of the red shift caused by stabilizing ligands, such as alkyl amines or thiols, increases when the headgroup of the ligand interacts more strongly with the surface of the Au NPs. In a recent study,286 a dispersion of gold nanorods (GNRs) with different aspect ratios was used to perform a multiplexed bioanalytical sensing measurement in solution. This work exploited the fact that small changes in the aspect ratio of GNRs lead to drastic changes in their optical properties (the longitudinal plasmon mode red shifts with increasing aspect ratio, as described in section 2). A series of GNRs with aspect ratios (length/width) of 2.1, 4.5, and 6.5sannotated as GNR 1, 2, and 3, respectivelyswere functionalized with a different recognition molecule (Figure 15a) and dispersed into a single solution.286 The solution yielded a composite absorption spectrum with three pronounced longitudinal LSPR maxima, each one corresponding to one of the three types of GNR (Figure 15b). The longitudinal peaks, in order of decreasing energy, correspond to GNR 1 > GNR 2 > GNR 3. Targets that were complementary to the different recognition molecules on the GNRs were then added to the solution to initiate binding events that caused a selective red shift of one or more of the longitudinal peaks. For example, when target 1 (complementary to the receptors on GNR 1) was added to the solution, a red shift in the longitudinal peak corresponding to GNR 1 was observed with smaller shifts in the peaks corresponding to GNR 2 and 3 (Figure 15b). When targets 1 and 2 (complementary to the receptors on GNR 1 and 2, respectively) were added to the solution, a shift in the peaks corresponding to GNR 1 and 2 was seen, while only a small shift in the LSPR maximum associated with GNR 3 occurred (Figure 15c). All three peaks red shifted when their corresponding targets were added to the solution (Figure 15d). To our knowledge, this report is the first to take full advantage of the multiplexing potential offered by GNRs and demonstrates the potentially enabling qualities offered by the ability to tailor the optical properties of metallic nanostructures.

4.4.2. Surface-Immobilized Nanoparticles NPs synthesized in solution can also be immobilized on surfaces for potential ‘on-chip’ sensing applications and thereby engender more useful multiplexing capabilities.5,376 Thin Cr or titanium (Ti) films are conventionally applied to substrates such as glass to promote adhesion of Au or Ag to the substrate. These metal layers are known to attenuate and broaden plasmon resonance bands246,385,386 and lead to markedly adverse effects on conventional SPR and LSPR sensing with noble metals.385,387 For this reason, it is beneficial to use organic adhesion layers of amine- or mercapto-terminated silanes to immobilize Au or Ag NPs to oxide-bearing substrates such as silica (Figure 16).5,376,385,388,389 The LSPR of these Ag or Au colloidal monolayers can be measured with a commercially available UV-Vis spectrometer in a simple collinear transmission configuration,5,376,390 and binding events can be monitored

Nanostructured Plasmonic Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 507

Figure 15. Schematic of a GNR molecular probe (a). Multiplexing detection of various targets using GNR molecular probes: one target (b), two targets (c), and three targets (c). Reprinted with permission from ref 286. Copyright 2007 American Chemical Society.

Figure 16. Schematic representation of the applied biosensing principle based on Au or Ag NPs deposited on a quartz substrate. From left to right: light source, quartz substrate, mercaptosilane adhesion layer, Au or Ag NPs, self-assembled monolayer of functional thiols, antibodies and antigens. The resulting absorbance spectra increase upon binding of analytes to the NPs is shown on the right-hand side of the figure. Reprinted with permission from ref 385. Copyright 2003 American Chemical Society.

in real time by integrating the functionalized substrate with a flow cell.5,376 Changes in refractive index occurring near the surface of the immobilized NPs (caused, for example, by the adsorption of a protein from a dilute solution) shifts the LSPR peak position and magnitude.5,376,385,391-393 These devices exhibit peak position sensitivities of ∼167, ∼76, and ∼252 nm/RIU for spherical Ag391 and Au NPs5 and Au nanorods394 respectively, and changes in extinction at an off-peak wavelength of ∼0.4-1.2 and ∼0.8 per RIU for spherical Au NPs376 and Au nanorods, respectively.394 This sensitivity can be enhanced using a multireflection attenuated total reflection setup.389 Surface-immobilized spherical NPs have recently been used to detect the binding of antibodies to BSA and human serum albumin (HSA)392 and for the selective detection of phosphopeptides on titania-coated Au NPs in complex samples at nanomolar concentrations.393 The at-

tractive features of this approach to sensing are the relatively easy and low-cost fabrication process and the simple optical setup. This approach has also been used to immobilize NPs on optical fibers,388,395 which have shown the ability to detect streptavidin (using a model biotin-avidin assay) and staphylococcal enterotoxin B (using a model antibody-antigen immunoassay) at picomolar concentrations.29,395 Surfaceimmobilized core shell NPs, such as spherical silica-Au (core-shell, Figure 4) and rice-shaped hematite-Au (core-shell) NPs, have been shown to exhibit bulk refractive index sensitivities of ∼555396 (dipole resonance) and ∼800 nm/RIU,199 respectively. Nanostructures for refractive index sensing have also been formed directly on substrates using NSL,7,337,382,397 colloidal lithography,259,383,398,399 soft nanoimprint lithography,77 or metal thin film evaporation.379,400,401 Arrays of nanostructures offer the advantage of tunability of the wavelength response.

508 Chemical Reviews, 2008, Vol. 108, No. 2

For example, arrays of triangular Ag NPs formed on substrates by NSL (Figure 5) exhibit an LSPR extinction band that can be tuned from the near-UV to the mid-IR by simply changing the size and shape of the NPs.257 The peak position λmax is sensitive to the local dielectric environment and red shifts linearly with increasing solvent refractive index with a sensitivity of ∼200 nm/RIU.397 The distance dependence of the LSPR resonance of these triangular NPs has been systematically studied using SAMs,402 layer-by-layer self-assembly,403 and atomic layer deposition.404 These studies provided several important insights into applicationrelevant parameters including (i) the sensing volume can be systematically tuned by controlling the composition, size, and shape of the NPs, (ii) the sensor response is linear with respect to analyte coverage/quantity when the binding occurs at short distances from the surface of the NPs, and (iii) the sensor response varies in a complex, nonlinear fashion when binding occurs at large distances from the surface of the NPs.403 The refractive index sensitivity of triangular Ag NP arrays has been exploited in sensing applications where peak shifts were used to monitor binding events at the surface of the NPs.7,92,243,382,402,403 The measured peak shifts were used in conjunction with a simple mathematical formalism developed for flat film SPR spectroscopy to quantitatively determine target analyte concentrations.405 This allowed the quantitative detection of Concanavalin A,382 streptavidin,7 and antibiotin243 as well as the thermodynamic evaluation of their binding constants. In a recent and notable example, triangular NP arrays were used to detect biomarkers for Alzheimer’s disease in both synthetic and human patient samples.92 In this work, synthetic amyloid-β-derived diffusible ligands (ADDL) were detected at femtomolar concentrations using a sandwich assay.92 The sensitivity of these types of measurements is enhanced if the molecular resonance of the analyte overlaps with the intrinsic LSPR of the NPs or if the analyte is labeled with a marker that has a resonance that overlaps with the intrinsic LSPR of the NPs.406,407 One of the advantages of using plasmonic nanostructures for sensing is their relatively small footprintsone that is more amenable to miniaturization than flat film SPR detection. For example, LSPR sensing has been demonstrated at the single-NP level using spherical (Au343 and Ag408), triangular (Ag),285,408,409 disk-like (Au),410 and cubic (Ag) NPs.342 Scattering-based spectroscopies must be used to characterize the optical properties of single NPs103,285,343,408-412 since the absorbance of individual NPs is close to the shot noisegoverned limit of detection.285,337 As an example, McFarland et al. have shown that a LSPR peak shift of ∼40 nm occurs upon binding of ∼100 zeptomoles of 1-hexadecanethiol to a single triangular Ag NP (as measured by resonant Rayleigh scattering spectroscopy). This high sensitivity and small transducer size suggests that single NPs could be useful for the analysis of precious or limited-volume samples. Rubinstein developed the so-called transmission LSPR (T-LSPR) spectroscopy method.379 In this protocol one measures the changes that occur in the extinction band of the LSPRs of discontinuous Au or Ag films upon analyte binding, which is monitored in transmission mode using a standard spectrophotometer. Discontinuous and random island films for this plasmonic measurement are prepared by direct evaporation of an ultrathin (e10 nm nominal thickness) layer of the desired metal onto a transparent substrate like quartz, mica, or polystyrene. Random Au island

Stewart et al.

films, for example, display a SP extinction peak at 550800 nmsone whose shape, intensity, and position depend on the island morphology, which in turn is determined by the evaporation conditions and postdeposition treatment. Rubinstein and co-workers demonstrated the potential of this method by measuring changes in the position and intensity of the SP extinction band that result from the binding of various molecules to Au islands.379,400,401,413 A linear relationship was shown to exist between the surface coverage of the adsorbing molecules, ones bound either directly to the Au or through a receptor layer, and the plasmon intensity or wavelength changes.379,401,413 In an interesting modification of typical RI-based measurement methods, Au island films were modified with a biotinylated monolayer and used to monitor the binding of avidin based on changes in plasmon intensity (rather than the more commonly used wavelength shift) since the binding event caused only a small change in the peak position. T-LSPR spectroscopy was shown to be widely applicable with a sensitivity (under optimized conditions)380 that is comparable to that of conventional forms of SPR sensing.379 The instability of T-LSPR sensors, however, is a source of concern. Changes in the optical properties of metal island films due to morphological changes occurring upon immersion in organic solvents and aqueous solutions has been noted to appreciably introduce uncertainties into NP-based sensing measurements. To obtain metal island films with stable and reproducible optical properties, new design schemes for stabilizing the structures of the evaporated film have been devised. The most useful of these reported to date consists of depositing an ultrathin silica layer (1.5 nm thick) on the metal island film by a sol-gel procedure.381 Au nanorings formed directly on substrates by colloidal lithography (Figure 7) have recently been used for chemical and real-time biosensing.383 These plasmonic nanostructures exhibit a bulk refractive index sensitivity of ∼880 nm/RIU and show a peak shift of ∼5.2 nm per CH2 unit when SAMs of varying alkanethiol chain lengths were formed on the nanorings by chemisorption. This peak shift per CH2 unit (short-range refractive index sensitivity) for the Au nanorings is greater than the peak shifts reported for triangular Ag NPs (∼3.1 nm per CH2 unit) and nanoprisms (∼4.4 nm per CH2 unit).402,409 The greater sensitivity of Ag plasmonic nanostructures compared to Au plasmonic nanostructures suggests that greater sensitivities could be achieved by forming Ag nanorings.403 Real-time, label-free biosensing was demonstrated using these structures by monitoring the optical changes that occur due to the nonspecific binding of biotinBSA to the Ag nanorings followed by the specific binding of NeutrAvidin (NA).

4.4.3. Periodic Nanohole Arrays The enhanced transmission of light through periodic arrays of subwavelength holes in metal films has generated considerable interest since it was first reported by Ebbesen et al. in 1998.75 Nanohole arrays are typically formed by the serial process of the FIB milling of holes in a thin film of Au supported on a transparent substrate. While Ebbesen et al. attributed their enhanced transmission to surface plasmons,75 several authors have pointed out alternatives ranging from waveguide modes to even more novel surface waves.414-416 In actual fact, and depending on the specific details of the plasmonic structures involved, a

Nanostructured Plasmonic Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 509

variety of mechanisms can be operative. For this reason, it is necessary to carefully analyze each situation to determine what sorts of resonance and diffractive effects are possible. In most cases involving thin Au films with periodic nanoholes, however, surface plasmons do play a key role.76,273 The features in transmission spectra of nanohole arrays are thought to arise from a combination of LSPRs, Bloch wave SPPs (BW-SPPs), and Wood’s anomalies.76 An approximate relation for the allowed wavelengths of BW-SPPs excited by normal incidence illumination on a square array of subwavelength holes in a metal film is given by76,77,417,418

λ)

P

xnx2 + ny2

Re

x

(λ)med

(λ) + med

(3)

where (λ) is the wavelength-dependent relative dielectric constant of the metal, med is the relative dielectric constant of an adjacent medium, P is the nanohole lattice spacing, and nx and ny are integer scattering or diffraction orders from the array. Another important equation in analyzing the features of transmission spectra is that for the positions of Wood’s anomalies (WAs)75,76

λ)

Pxmed

xnx2 + ny2

(4)

which is written here for the case of normal incidence for simplicity. The WAs in this case may be thought of as light moving parallel to the metal surface, i.e., light that is diffracted 90° relative to the normal incident direction. Note that in one classification scheme of WAs this particular WA is often referred to, more precisely, as a Rayleigh anomaly.419 Equations 3 and 4 predict a variety of discrete wavelength positions for the BW-SPPs and WAs. Originally a correlation between transmission minima and WAs and transmission maxima and BW-SPPs was noted;386,417 however, this may not always be the case. At a naı¨ve level, for example, one might associate both the BW-SPP and WA positions with transmission minima, not maxima, because they represent wavelengths where incident light is channeled into light moving completely counter to the direction of the transmitted light. Indeed, careful numerical studies made in the absence of experimental imperfections and uncertainties have shown a definite correlation between transmission minima and eqs 3 and 4.76 It remains a fact, however, that nonradiating BW-SPPs and WAs are an idealization. One can regard the BW-SPPs or WAs as being analogous to zero-order quantum mechanical bound states embedded in a zero-order continuum. In the present case a zero-order continuum state would be light that is predicted to be transmitted by the hole/ film structure disregarding the possibility of zero-order bound states. The full problem is described by a coupling of the zero-order bound and continuum states. A Fano resonance line shape (generally an asymmetric line shape with a distinct minimum and maximum) can often describe each BW-SPP or WA feature very well.76,419,420 It often turns out that the same order BW-SPP and WA have wavelengths that are close to one another, which can complicate this picture. Coupled with the fact that the WA features can be narrow and difficult to resolve experimentally, it sometimes follows that only a minimum/maximum resonance feature associated with a particular BW-SPP is seen, along with possibly a dip or kink in the region where the WA would be ex-

pected.76,421 Finally, we note that the features discussed here can occur alongside or be superimposed on generally broader LSPR resonance features. With all these factors taken into account and playing varying roles depending on the specific details of the experiment, it is perhaps not surprising that there has been some confusion in the literature regarding the mechanisms of enhanced transmission. The involvement of LSPRs, BW-SPPs, and WAs in the optical response of nanohole arrays suggests that the features in the transmission spectra should be sensitive to changes in the local refractive index, making these nanostructures useful for sensing applications. Indeed, the spectral features of 2D Au nanohole arrays have been shown to depend on the external dielectric environment,14,422,423 which has been used to perform label-free detection of SAM formation on Au followed by subsequent adsorption of bovine serum albumin (BSA).14 The sensitivity of these structures to changes in the external dielectric environmentsmeasured as peak shift per RIUsis ∼333-400 nm/RIU.14,377 The advantages of plasmonic nanostructures over flat film SPR systems are their relatively small footprints (i.e., small patterned areas on the order of micrometers14,377 or single holes259 or NPs285,342) and simple optical setups (i.e., normal incidence transmission or reflection).14,342,377 These advantages were recently exploited by integrating a microfluidic device with a nanohole array for detecting refractive index changes and surface binding events ‘on-chip’.377 Plasmonic quasi-3D periodic nanohole arrays have been formed on surfaces using a type of soft nanoimprint lithography (see section 3.3.3 and Figure 9).8,77 These multilayered structures consist of arrays of nanoscale holes in Au films with a second level of Au disks at the bottom of the embossed wells. The normal incidence transmission spectrum of these structures is complex (Figure 17a, blue spectrum) and shows high transmission despite the fact that no ‘line of sight’ exists through the sample (i.e., unlike 2D nanohole arrays, the cylindrical holes or perforations are ‘capped’ with Au disks in a quasi-3D geometry). Rigorous 3D finite-difference time domain (FDTD) simulations were performed to assist in the interpretation of the experimental results (Figure 17a). (See also refs 76 and 424 for related FDTD studies of hole arrays.) The calculated electromagnetic field distributions in and around the metal nanostructures are shown in Figure 17b and c. The two largest features in the spectra (Figure 17a) are labeled B and C. Peak B is associated with excitations of LSPRs on the rims of the nanoholes in the upper Au film near the air/Au interface, while peak C involves overlapping Wood’s anomaly and BW-SPPs excitations on the Au disk/polymer side of the device. The intensity associated with peak C also extends vertically up to the hole opening, showing a strong coupling between disk and hole, a feature that is absent in random and ordered 2D nanohole arrays. Quantitative modeling of the experimental spectra required consideration of fine structural details in this region of strong coupling. Good agreement between the experimental and theoretical spectra required the addition of small (20-30 nm), isolated grains of Au on the sidewalls of the nanoholes, just above the edges of the Au disks at the bottom of the nanowells (Figure 17a, red spectrum). The transmission properties of the quasi-3D nanohole array are sensitive to the nature of the adjacent dielectric medium at the crystal surface (Figure 18a). The bulk refractive index sensitivity of these devices was determined by passing

510 Chemical Reviews, 2008, Vol. 108, No. 2

Stewart et al.

difference spectra, as referenced to the spectrum at time t ) 0, illustrate changes in transmission due to both peak shifts and intensity changes throughout the wavelength range as solutions of increasing PEG concentration were injected into the flow cell. As shown in Figure 18b, the transmission can increase or decrease depending on the measurement wavelength. The particular crystal geometry used in this study (hole diameter and periodicity of ∼480 and 780 nm, respectively) exhibited the largest multiwavelength response in the near-infrared region (900-1250 nm) as shown in Figure 18c. The response over all wavelengths, including both positive and negative transmission changes, was calculated using the following equation

R ) ∫|∆(%T(λ))dλ

Figure 17. Correlation of transmission spectral features with hole/ disk plasmonic excitations. (a) Normal incidence transmission spectrum of a quasi-3D plasmonic crystal (blue), and rigorous electrodynamics modeling of the spectrum for an ideal crystal (green) and one that includes subtle isolated nanoscale grains of Au near the edges of the Au disks (red). (b) Computed electromagnetic field distribution associated with the resonance at 883 nm (labeled B in a). The intensity is concentrated at the edges of the nanoholes in the upper level of the crystal. (c) Field distribution associated with the resonance at 1138 nm (labeled C in a), showing strong coupling between the upper and lower levels of the crystal. Reprinted with permission from ref 77. Copyright 2006 The National Academy of Sciences of the USA.

solutions of increasing concentration (0-7.6 wt %) of polyethylene glycol (PEG) through a fluid flow cell containing a plasmonic crystal. Changes in both peak positions and intensities were observed over a wide spectral range as the refractive index of the PEG solution was increased. The most sensitive peak at ∼1023 nm linearly red shifted with a sensitivity of ∼700-800 nm/RIU and linearly increased in intensity with a sensitivity of ∼2.5-3.5 Abs/RIU as the refractive index of the solution was increased (determined using several devices).77 SPR sensing is typically performed by following the response of an individual peak or wavelength to binding events at the surface of the sensor. This method of analysis, however, does not fully capture the sensitivity of the quasi3D plasmonic sensor since it does not take advantage of all the peak shifts and intensity changes occurring at multiple plasmonic resonances over the spectral range created by the coherent couplings of the LSPRs, BW-SPPs, and Wood’s anomaly responses exhibited by this type of system. This wide spectral response was exploited using a type of full, multispectral analysis as described below. Multispectral analysis of a PEG calibration of a quasi-3D plasmonic crystal is shown in Figure 18a. A series of

(5)

This total or integrated response, R, has units of ∆%T‚nm (Figure 18d) and changes linearly with the refractive index of the PEG solution with a sensitivity of ∼22,000 ∆%T‚ nm/RIU (inset of Figure 18d). This multispectral analysis also improves the signal-to-noise ratio by a factor of 3-10 times that of a single-wavelength response.77 The linear response of the integrated metric was used to perform quantitative sensing and imaging of binding events. The biotin-avidin ligand-receptor conjugate was used to illustrate the utility of these devices for performing quantitative sensing (Figure 19). The device was first exposed to a solution of biotinylated BSA (bBSA), which led to an increase and plateau of the integrated response of the sensor (Figure 19b) upon formation of a bBSA monolayer. This layer rendered the surface of the sensor inert to further nonspecific adsorption, which was demonstrated by the lack of response after rinsing the bBSA monolayer with buffer and exposing it to a solution of nonfunctionalized BSA. Subsequent exposure to avidin, however, led to a response due to a specific binding interaction between the avidin and the initial bBSA monolayer. The surface-immobilized avidin was then used to complete the assay by binding a layer of bBSA to the remaining free biotin binding sites on the avidin layer (inset Figure 19b). This resulted in a response that was smaller than that observed for the initial bBSA adsorption step, an observation that follows the patterns of layerdependent mass coverage generated in assays of this sort.77 The integrated response can be converted to an effective protein thickness using a mathematical formalism.405 Although this model was developed for quantifying binding events measured using flat film SPP-based SPR sensors (i.e., it assumes a uniform plasmon evanescent field), it provides approximate protein coverages that agree with literature values.425-427 One notable advantage of soft nanoimprint lithography is the ability to pattern over large areas in parallel with high spatial uniformity and low defect densities, which facilitates large-area imaging for multiplexed microarray-based analyses. The high-quality, large-area patterning capability along with the capacities for quantitative biosensing was combined to perform quantitative imaging. To this end, five lines of nonspecifically adsorbed fibrinogen were patterned on the surface of a plasmonic crystal using a microfluidic device (Figure 20a). Figure 20b shows changes in transmission measured relative to an interchannel region on the crystal that did not come in contact with the protein. The spectral image shows five stripes with the expected geometries, each corresponding to a line of nonspecifically adsorbed fibrino-

Nanostructured Plasmonic Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 511

Figure 18. Optical response of a plasmonic crystal to sequential injections of increasing concentrations of aqueous PEG solutions. The color contour plot of the change in transmission (T) as a function of wavelength and time (with the corresponding injection sequence overlaid on the plot) (a), change in T as a function of time during the injection sequence, evaluated at several wavelengths (b), absolute value of the change in T as a function of wavelength evaluated at different times (c), and integrated multispectral plasmonic response as a function of time (d). (Inset) A linear correlation to the change in refractive index. Reprinted with permission from ref 77. Copyright 2006 The National Academy of Sciences of the USA.

Figure 19. Plasmonic crystal used in a biotin-avidin assay. The color contour plot of the change in T as a function of wavelength and time (a). The overlaid injection sequence corresponds to PBS (1), bBSA (2), BSA (3), and avidin (4). The integrated multispectral plasmonic response and corresponding effective thickness of the biotin-avidin-biotin assay (schematically illustrated in the upper inset) (b). The noise limited refractive index resolution of the crystals corresponds to submonolayer coverages (lower inset). Reprinted with permission from ref 77. Copyright 2006 The National Academy of Sciences of the USA.

gen. Analysis of the step edges (Figure 20c, inset) shows a width of ∼20 µm, which is only slightly larger than the ∼17 µm resolution limit of the imaging optics. The ∼3 µm of additional width in the plasmonic image can be associated with the propagation lengths of plasmons on these structures.428 The integrated spatial response of the spectral image could be converted to an effective protein thickness of ∼7 nm using the PEG calibration and the mathematical model described above.77,405 This thickness is consistent with the molecular dimensions of fibrinogen.429,430 The micrometer-

Figure 20. Spatial imaging of fibrinogen nonspecifically adsorbed to the surface of a crystal. A schematic illustrating the use of a multichannel PDMS microfluidic network to pattern the surface of a crystal (shown here with the multicolored appearance that characterizes these crystals) (a), spectroscopic difference image of fibrinogen lines patterned on a crystal (b), and spatially resolved integrated response and corresponding effective thickness illustrating binding events in the geometry of the microfluidic channels (c). (Inset) A measured step edge between a fibrinogen line and bare area of the crystal (blue symbols) and a fitted step edge with a Gaussian width of ∼20 µm. Reprinted with permission from ref 77. Copyright 2006 The National Academy of Sciences of the USA.

scale imaging resolution and the large area defect-free (and thus uniform response) aspects of the crystals suggest a promising platform for performing parallel diagnostic bioassays. This was further demonstrated by performing quantitative 2D imaging on these quasi-3D devices.279

512 Chemical Reviews, 2008, Vol. 108, No. 2

Stewart et al.

4.4.4. Random Nanohole Arrays Random nanohole arrays formed by colloidal lithography have been used in enabling ways for the sensing of chemical and biological targets.90,259,398,399 Randomly arranged holes do not exhibit the long-range higher-order diffractive coupling effects that are observed in the highly ordered, periodic nanohole arrays discussed above.72,75,90,422 They instead exhibit LSPRs that are concentrated at the edges of the holes with decay lengths on the order of 10-20 nm, which have been used for sensing binding events occurring near or in the holes.90,259,398,399 The refractive index sensitivity of these random nanohole arrays depends on the metal thickness398 and the detection modality. As noted above, refractive index sensitivities are often reported as a change in a given measurable parameter per RIU. Sensitivities of ∼100-270 nm/RIU,72,90,398,399 0.23-1 Abs/RIU,398,399 and ∼70 nm/ RIU259 have been reported for measurement of changes in the extinction peak position, extinction peak magnitude, and scattering peak position, respectively. In one notable example of biosensing carried out with such devices, Dahlin et al. used random nanohole arrays to detect membrane-mediated recognition of bioanalytes using surface-immobilized phospholipid bilayers (SPBs). These SPBs were supported on the silica surface within the interior of the nanoholes (Figure 21). This was accomplished by first exposing the nanohole array to bBSA, which preferentially adsorbs to the Au rather than the SiO2,431 making it inert to lipid vesicle fusion. The SPBs were then formed on the SiO2 substrate at the bottom of the nanoholes by performing vesicle adsorption in the presence of Ca2+, which promotes the vesicle fusion process.432 The top of Figure 21a shows schematically the steps leading to the adsorption of the protein NA within the nanoholes containing the biotin-modified lipids (left). In the absence of biotin-modified lipids the NA binds only to the bBSA on the Au surface, whereas in the presence of biotinmodified lipids the NA binds to both the SPB and the Au surface. The bottom panel of Figure 21 shows the measured changes in extinction as a function of time at ∼725 nm for the two cases, with the signal being more than a factor of 3 greater when NA binds to lipids inside the holes. This demonstrates that the holes are highly sensitive regions for detecting protein-binding events. The top of Figure 21b shows schematics illustrating the binding of cholera toxin to SPBs containing ganglioside GM1 glycolipids (left) and hybridization of 15-base single-stranded DNA (ssDNA) to SPBs presenting a complementary strand (right). The bottom of Figure 21b shows the measured changes in extinction as a function of time at ∼725 nm upon binding of the cholera toxin and complementary and noncomplementary ssDNA. These examples demonstrate the potential for using random nanohole arrays as a platform for real-time plasmonic-based label-free sensing. The figure of merit (FOM) for this system appears to be quite good in that analytically discriminable signals could be obtained from zeptomole quantities of protein. Related nanohole arrays also have been used to detect cancer biomarkers where the detected signal (peak shift) was estimated to arise from the binding of picograms of an antigen (a specific tumor biomarker, cancer antigen 19-9) to its corresponding surface-immobilized antibody.90

4.5. Surface-Enhanced Spectroscopies The optical techniques discussed thus far are amenable to chemical functionalizations that engender capacities for molecule-specific sensing, ones that the methods alone lack.

Figure 21. Detection of lipid-membrane-mediated binding events using SPB in random arrays of gold nanohole. (a) Temporal variation in extinction measured at the longer wavelength inflection point (725 nm) of the LSPR peak upon addition of ∼0.3 µM NA. In both cases, Au is modified with biotin-BSA. In one case, biotinmodified SPB patches cover the SiO2 regions (NAAu+SiO2), while in the other (top right illustration in a), unmodified SPB patches cover the SiO2 regions (NAAu) (top left illustration in a). (b) Variation in extinction measured upon addition of ∼0.5µM cholera toxin (CT) to GM1-modified (5 wt %) SPBs (purple curve, illustration of SPB-modified nanohole shown in upper left of b). (Inset) Magnification of changes in extinction versus time upon addition of a 15 base long noncomplementary (0.2 µM, blue) and a fully complementary strand (0.2 µM, red) to SPB patches modified with a DNA construct carrying two cholesterol moieties at its one end and a 15-base-long single strand available for hybridization at the other. Reprinted with permission from ref 399. Copyright 2005 American Chemical Society.

It is useful then to consider cases where other properties besides explicit forms of chemical recognition can be used to discriminate a molecularly specific event. Techniques such as Raman scattering and fluorescence have the capability of providing molecule-specific data, and perhaps most interesting in the context of this review, surface plasmons are able to markedly enhance the sensitivity of these techniques. The utility of Raman spectroscopy, for example, stems from its ability to probe molecular vibrations, but the method suffers from relatively weak signal intensity. In most cases the number of inelastically scattered photons (which directly corresponds with the signal level) is extremely small, corresponding to scattering cross sections of 10-30-10-25 cm2.22 As a result, even though Raman scattering is a type of spectroscopy that can give very specific molecular information, it lacks the intrinsic sensitivity required of a viable technique for high-throughput detection. Plasmonic architectures provide one general method for enhancing Raman signals to levels that can enable many analytical applications, ones ranging from DNA sequencing to forensics.337,433,434 The sections below examine several of the more interesting prospects emerging from current research.

Nanostructured Plasmonic Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 513

4.5.1. Surface-Enhanced Raman Scattering The first discovery of surface-enhanced Raman scattering (SERS) showed that it was possible for a roughened noblemetal surface to enhance the intensity and sensitivity of Raman spectroscopy.435,436 Using the SERS effect in a quantitative form is intricately tied to the ability to fabricate nanostructured surfaces with well-defined morphologies. New methods to create nanostructured materials are constantly evolving, and research centered around SERS has followed that trendswork that also has increased the sensitivity to levels required for advanced sensing applications. Signal enhancements of up to 1014 over normal Raman scattering have been observed in special cases,437 a value sufficient to study analytes even at the single-molecule level. The electromagnetic theory of SERS437-440 suggests that the enhancement effect is a result of the creation of surface plasmons that can transfer energy to the bound molecule through the associated electric field. We direct the interested reader to a number of reviews that have already been published on this subject.437-440 The ability to enhance the Raman signal via SERS is fundamentally linked to the precise details of the structured nature of the surface as well as the choice of the metal itself, which includes but is not limited to Ag, Au, and Cu.438 These metals are used because they have the appropriate values of both the real and imaginary parts of the dielectric constant, which allows surface plasmons to propagate at wavelengths of interest for spectroscopic applications. As the discussions above have revealed, the plasmonic excitations that a system can support depend very sensitively on the metal structure, a factor that in turn drastically affects its overall ability to enhance the Raman scattering cross section. The advances made in nanomaterial synthesis and NP fabrication have greatly empowered competencies in this latter regard. Such forms of modern SERS substrates span a vast geometrical range encompassing colloids, templated colloidal crystal films, electrochemically roughened electrodes, deposited metal island films, nanohole arrays, and lithographically defined thin films. The ability to use these different structures as SERS substrates allows a variety of experimental setups including solution- and surface arraybased sensing.435,441-446 Solution-based SERS has been shown to provide a viable option for the determination of binding constants and monitoring DNA/RNA mononucleotide recognition.447 For the latter, detection of mononucleotides at a submicromolar level (3 ppb) was achieved for 2′-deoxyadenosine 5′-monophosphate (dAMP) in the presence of MgSO4, an aggregating agent (Figure 22).447 The data indicates that it is essential for the analyte to be bound within close proximity to the NPs. Solution-based sensing with NPs has also been applied to the field of cellular biology.448 In a notable example taken from recent research, it was demonstrated that it is possible to introduce 60 nm Au NPs into a single osteosarcoma cell and use their optical responses to map the distribution of the cellular constituents.449 It is possible to further enhance the SERS signal of an analyte by altering the physical properties of the NPs.450-452 For example, using Au octahedrons instead of Au nanospheres allows for a more than 3× greater enhancement in the spectrum of 2-napthalenethiol.450 Substrate-bound nanostructures can also be used to enhance Raman signals.441 Nanosphere lithography, for example, has been used to fabricate two intriguing forms of surface-immobilized structures for SERS: (i) triangular NP

Figure 22. Effect of aggregating agents to the SERS signal of dAMP: 1000 ppm dAMP mixed with Ag colloid (a); mixture after aggregation with 0.1 M MgCl2 (b); 0.1 ppm dAMP mixed with MgSO4-aggregated (0.1 M) Ag colloid (c). Spectra d, e, and f used the same conditions as c except 0.03, 0.01, and 0.003 ppm dAMP. The inset shows the calibration plot of the dAMP (4 s accumulation times). Reprinted with permission for ref 447. Copyright 2006 American Chemical Society.

Figure 23. Ambient contact-mode atomic force microscope image of 200 nm Ag over 542 nm diameter polystyrene spheres. Array of spheres (10 µm × 10 µm) and image (600 nm × 600 nm) of one sphere showing substructure roughness (a and b, respectively). AgFON electrode SER spectrum of 50 mM pyridine in 0.1 M KCl at -0.7 V vs Ag/AgCl (c). Reprinted with permission from ref 453. Copyright 2002 American Chemical Society.

arrays (Figure 5) and (ii) metal film over nanosphere (MFON) surfaces (Figure 23). Triangular NP arrays have been shown to be quite sensitive with SERS enhancement factors of ∼108.337 Analytes detected in this way include Alzheimer’s precursors,93 glucose,433 and Concanavalin

514 Chemical Reviews, 2008, Vol. 108, No. 2

A.382 The highly corrugated MFON surfaces have been used to detect a key component of an analogue to anthrax stimulant at a concentration lower than the infectious dosage.242,445 The MFON substrates are also useful as electrodes with an improvement in stability and reproducibility over other types of SERS active electrodes.453 The latter are critical parameters since most SERS active substrates are not particularly stable. It is notable, therefore, that recent work has shown that a sub-1 nm thick alumina film deposited over a MFON surface can extend the stability of the sensor up to seven times, corresponding to a shelf life of at least nine months.242 Electrochemistry can be used to roughen an electrode surface for SERS applications, but this method generally yields substrates with irreproducible enhancement factors.453 An alternative method that can be used to form SERS-active electrode structures is to use a template to create nanowires. As an example, Au nanopillar films can be formed by physical/electrodeposition of Au in an anodized aluminum oxide (AAO) template followed by dissolution of the AAO template.454 These nanopillar substrates are highly reproducible (to within 20% of the SERS intensity profile) and capable of detecting thionine as a model analyte at concentrations near 10-8 M.454 Nanopillars of Ag can also be fabricated in the form of aligned pillars of controllable length, giving SERS enhancement factors on the order of 108.455 Nanodots,443 nanohole arrays,444 and even nanostructured multicore optical fibers456 have been fabricated for use in SERS applicationsswork that evidence a broadening research landscape in response to requirements for SERS substrates that are both stable and ultrasensitive. 4.5.1.1. Single-Molecule Surface-Enhanced Raman Scattering. Highly specific single-molecule detection is a challenging goal for chemical sensing.22,457,458 It is interesting, therefore, to consider how the marked enhancements afforded by SERS compare to this most demanding analytical FOM. Kniepp and co-workers elucidated the fact that singlemolecule SERS (SMSERS) detection levels can be achieved both in solution and on surfaces, the latter using patterned metal substrates. Each protocol requires the use of a proximal probe in which incident light is restricted to encompass a probe volume of a few femto- to picoliters. This makes it possible to examine limited volume samples with analyte concentrations as low as 10-12-10-14 M. Solution-based SMSERS, for example, has been used to detect pseudoisocyanine (PIC) in an aqueous solution at a concentration of 10-14 M, corresponding to an enhancement factor on the order of 1014.22 Ag colloids also have been used to detect yeast cytochrome c at near single-molecule levels.459 Patterned Ag surfaces have been shown to give enhancements that make it possible to detect single molecules of enkephalin, in this case by monitoring the ring-breathing mode of phenylalanine at 1000 cm-1.22 The small sample probe volume has other (potentially beneficial) impacts as diffusion brings single molecules into and then out of the probed volume at very low analyte concentrations due to Brownian motion. This leads to a variation in the Raman signal that can be correlated to the number of molecules in the probed volume via models based on Poisson statistics.434

4.6. Plasmonics for Detection Beyond the Diffraction Limit One limitation of SMSERS (or for that matter any optical technique) is that the spatial resolution of the probed area is

Stewart et al.

dictated by the spot size of the incident light. Optical detection methods are constrained by the wave nature of light (which determines the diffraction limit); however, it is particularly interesting that plasmonic nanostructures make it possible to probe systems with a spatial resolution that is well beyond the diffraction limit of light. Two methods that use the enhanced local electric fields associated with plasmons for detection at length scales beyond the diffraction limit are apertureless near-field optical microscopy (aNSOM) and tip-enhanced Raman spectroscopy (TERS). Both of these methods have enabled the scientific community to optically probe surfaces with resolutions previously only afforded by techniques such as electron microscopy and surface probe microscopy (SPM). Near field optical microscopy (NSOM), with a typical aperture, is a technique that affords capabilities of sub-100 nm resolution; however, it is important to note that the transmission of light through the fiber tip is typically quite low. One way to improve the resolution and limited transmission is to move to a fully metallized tip that has no aperture (a-NSOM). The apertureless tip is then capable of supporting surface plasmons that lead to enhancement factors of the electric field of 10-1000 times, and the ability to sharpen the tip affords resolution limits on the order of 10 nm.460 For example, tips coated with 15-25 nm of Ag were used to detect Alexa 532 dye molecules with a resolution of 15 nm.461 The ability to garner optical images from nearfield microscopy is a useful technique for imaging nanostructures with sub-100 nm resolution but lacks the specific chemical information afforded by techniques such as TERS. The basis of the SERS effect is tied to the nanoscaled structures present on a metal surface, and it is interesting to note that the SERS effect also can be observed using a metallized SPM tip. The surface of the probe tip must be coated with a metal (e.g., Ag, Au, Cu) that exhibits strong SERS enhancements. In this case the Raman scattering is constrained to within a few nanometers of the tip.462 The enhancement factors realized to date (102-104) are still moderately low and considerably less than the predicted maximum values (1010).462 One particularly interesting aspect of the advances coming from this research is the ability to combine the spatial resolution of SPM with the molecule-specific detection afforded by SERS. An example of this is illustrated by the data shown in Figure 24. In this case the nanoscale roughness, achieved with a 2 nm abrupt step on the surface of a metal-coated sharp TERS tip, can lead to an increase in Raman scattering cross section over a standard proximal probe by over an order of magnitude for benzenethiol adsorbed on Au.463 The tips of silicon nanowires have also been modified with Au droplets to improve detection limits.464 The ability to create tips of varying geometrical parameters allows the spatial resolution of the technique to be optimized. These methods also can be used to characterize features of crystalline materials such as GaN with high spatial resolution and enhancement factors as large as 104.465 The ability to break the diffraction limit while gaining chemically specific information constitutes an area of particular opportunity for progress in research, one whose progress will be dictated directly by capabilities to create better tips and to optimize plasmonic excitation and correlated collection optics.462

Nanostructured Plasmonic Sensors

Chemical Reviews, 2008, Vol. 108, No. 2 515

are attracting growing attention in research.468,472,476 In one specific example, Au nanohole arrays exhibited an enhancement of up to 2 times over flat metal films when the fluorophores were beyond the inherent quenching distance.468 Surface-immobilized Ag NP arrays have also been used as a medium to enhance the fluorescence of common dyes with leading work directed to determine how the geometry of the NP films affects the enhancement factor.215 It was found that the observed enhancements (of some 10-20 times) were dependent on the width and height of the particles with fluorescein and Cy3 exhibiting a maximum enhancement for particle sizes in the range of 85-95 nm.215 We believe that the ability to enhance fluorescence through the use of plasmons is an area of research that is not yet mature and will continue to grow in importance and impact as the method is optimized. Figure 24. TERS mapping on a rough Au surface. An STM image of the sample is shown in a. TERS data was collected at the positions indicated by the arrows. The cross section of the topography image is shown in b, and the TERS collection sites are labeled with crosses. (c) Corresponding TERS sequence. The numbers denote the sites where the spectra were collected. Reprinted with permission from ref 463. Copyright 2007 American Chemical Society.

4.6.1. Plasmon-Enhanced Fluorescence Plasmon-enhanced fluorescence (PEF) is a relatively new technique that can be used to increase sensitivity and detection limits. This is a capability that involves integrating the measurement within a device form factor that can exploit the enhanced electrical fields associated with surface plasmons. Fluorescence microscopy and spectroscopy are perhaps the most widely adopted methods used to study complex biomolecular systems.457,466-468 The use of surface plasmon modes to enhance fluorescence has recently attracted considerable interest as a way to extend the already considerable FOM of this spectroscopic measurement.215,356,469,470 Fluorophore quenching occurs in close proximity to metallic surfaces (typically < 50 Å);471 however, at longer distances ranging from 5 to 200 nm an interesting coupling between the fluorophore and the enhanced local field near the metal structure occurs, resulting in both increased absorption cross section and radiative decay rates.470 The increase of the radiative decay rate is central to a number of other phenomena observed in PEF including increased photostability, enhanced wavelength-tunable emission, and the ability to probe fluorescence resonance energy transfer (FRET) at larger distances than previously possible.470 These enhancements have opened new opportunities to improve detection and imaging schemes. A number of different metallic structures such as nanohole arrays, NPs, flat thin films, and deposited nanostructured films have been used for PEF.215,472-474 Au NPs, for example, have been used to sense mouse IgG at concentrations as low as 7 fM using a sandwich assay format coupled to an optical fiber.469 Sensing of metal ions such as Cu2+ has also been accomplished using fluorophore-labeled Au NPs with a detection limit of 1 µM.475 This particular sensing scheme has advantages over other methods because it shows increased quantum yields even in the presence of Cu2+, which generally quenches fluorescence. In the same vein, Au NPs have also been used to detect the presence of Hg2+ in pond water at concentrations as low as 2.0 ppb.473 Nanohole arrays can also be used for PEF detection, and such applications

5. Concluding Remarks The rich literature summarized in this review develops a compelling story about both the health of and the prospect for technological impacts following from research in the interdisciplinary field of plasmonics. Whether in the form of refractive-index-based detection schemes for analytes of extremely low concentrations or as optics for chemically sensitive imaging, the reports appearing from many laboratories at this time demonstrate an accelerating pace of underlying progress, one driven by rapidly improving capacities for nanoscale materials synthesis and methods of fabrication. An interesting question to ask at this point is one related to context. Specifically, will the most important impact going forward with respect to chemical analysis come from the development of systems that embed “new physics” (such as those emerging within the vastly interesting topic of metamaterials)477-480 or from compelling applications of systems that exploit what is currently known? Our hope is that the answer comes from both. There is no doubt that the rapid progress being made in both the fabrication and theory will greatly enhance progress in the field to render the enabling science in a form that begins to become a true form of predictive design for function. Essentially all applications in sensing will benefit from this advent. Still though, what might happen in the context of “new physics” and the opportunities it might engender? An interesting (and nascent) model here is provided by an area only touched on briefly in this reviewsuse of plasmonics as devices for subwavelength “imaging”. It is well appreciated that the spatial resolution in optical microscopy is limited to λ/2, according to the Rayleigh criterion. This limit is caused by the loss of the evanescent field intensity in the far field, which carries high spatial frequency information. Complex methods can overcome this limit and have been devised in such notable forms as nearfield scanning optical microscopy.460 The complexity of the methods used there carry with them limitations that have tended to limit its widespread use for biological imaging, especially cellular imaging. Metal nanohole arrays have been shown to inherently give anomalous high transmission of light,75 and this, in conjunction with “a perfect lens” based on negative index metamaterials,481 is theorized to make possible subwavelength resolution microscopy down to limits as small as 25 nm. Recent reports have demonstrated482-485 that a surface plasmon’s near-field dimensions are smaller than free-space radiation at the same frequency. Srituravanich et al. examined

516 Chemical Reviews, 2008, Vol. 108, No. 2

this effect by exposing photoresist through a nanohole array, obtaining features with sizes of 90 nm spaced by 170 nm.485 Fields with dimensions as small as 25 nm were obtained using slightly more complex plasmonic structures.486 It is therefore most intriguing that wide field optical microscopy with subwavelength resolution, based on the concepts of plasmon-coupled transmission and optical near fields, has been proposed along with methods to implement it.487-489 More recently, two different types of magnifying metamaterials-based superlenses that can be integrated into a conventional far-field optical microscope were reported.490,491 Liu et al. demonstrated a magnifying optical hyperlens consisting of a curved periodic stack of Ag (35 nm) and Al2O3 (35 nm) deposited on a half-cylindrical cavity in quartz.491 This setup was able to image a pair of 35 nm lines spaced 150 nm apart. In a later report, Smolyaninov et al. demonstrated a superlens design based on a multilayer photonic metamaterial consisting of alternating layers of positive and negative refractive index.490 With this design a resolution of 70 nm was reported. These are advances that serve to illustrate the inspirational opportunities for progress coming from work in this field.

6. Acknowledgments We gratefully acknowledge the support of aspects of our work by the National Science Foundation (CHE 04-02420) and the Department of Energy (DEFG02-91ER45439). S.K.G. was supported by the Office of Basic Energy Sciences, Division of Chemical Sciences, Geosciences, and Biosciences, U.S. Department of Energy (DE-AC0206CH11357).

7. References (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23)

Atwater, H. A. Sci. Am. 2007, 296, 56. Ozbay, E. Science 2006, 311, 189. Maier, S. A.; Atwater, H. A. Appl. Phys. Lett. 2005, 98, 011101/1. Hu, M.; Chen, J.; Li, Z.-Y.; Au, L.; Hartland, G. V.; Li, X.; Marquez, M.; Xia, Y. Chem. Soc. ReV. 2006, 35, 1084. Nath, N.; Chilkoti, A. Anal. Chem. 2002, 74, 504. Bozhevolnyi, S. I.; Volkov, V. S.; Devaux, E.; Laluet, J.-Y.; Ebbesen, T. W. Nature 2006, 440, 508. Haes, A. J.; Van Duyne, R. P. J. Am. Chem. Soc. 2002, 124, 10596. Malyarchuk, V.; Hua, F.; Mack, N. H.; Velasquez, V. T.; White, J. O.; Nuzzo, R. G.; Rogers, J. A. Opt. Express 2005, 13, 5669. Rothschild, M.; Bloomstein, T. M.; Efremow, N., Jr.; Fedynyshyn, T. H.; Fritze, M.; Pottebaum, I.; Switkes, M. MRS Bull. 2005, 30, 942. Stewart, M. E.; Motala, M. J.; Yao, J. T., L. B.; Nuzzo, R. G. Proc. ImechE, Part N: J. Nanoeng. Nanosyst. 2007, 220, 81. Tseng, A. A. Small 2005, 1, 594. Watt, F.; Bettiol, A. A.; Van Kan, J. A.; Teo, E. J.; Breese, M. B. H. Int. J. Nanosci. 2005, 4, 269. Daniel, M.-C.; Astruc, D. Chem. ReV. 2004, 104, 293. Brolo, A. G.; Gordon, R.; Leathem, B.; Kavanagh, K. L. Langmuir 2004, 20, 4813. Huang, C.-J.; Chiu, P.-H.; Wang, Y.-H.; Yang, C.-F. J. Colloid Interface Sci. 2006, 303, 430. Murphy, C. J.; Sau, T. K.; Gole, A. M.; Orendorff, C. J.; Gao, J.; Gou, L.; Hunyadi, S. E.; Li, T. J. Phys. Chem. B 2005, 109, 13857. Wiley, B.; Sun, Y.; Mayers, B.; Xia, Y. Chem. Eur. J. 2005, 11, 454. Bartlett, P. N.; Baumberg, J. J.; Coyle, S.; Abdelsalam, M. E. Faraday Discuss. 2004, 125, 117. Gates, B. D.; Xu, Q.; Stewart, M.; Ryan, D.; Willson, C. G.; Whitesides, G. M. Chem. ReV. 2005, 105, 1171. Genevieve, M.; Vieu, C.; Carles, R.; Zwick, A.; Briere, G.; Salome, L.; Trevisiol, E. Microelectron. Eng. 2007, 84, 1710. Wang, C.; Ma, Z.; Wang, T.; Su, Z. AdV. Funct. Mater. 2006, 16, 1673. Kneipp, K.; Kneipp, H.; Bohr, H. G. Top. Appl. Phys. 2006, 103, 261. Raether, H. 1988.

Stewart et al. (24) Wark, A. W.; Lee, H. J.; Corn, R. M. Anal. Chem. 2005, 77, 3904. (25) Lee, K.-S.; El-Sayed, M. A. J. Phys. Chem. B 2006, 110, 19220. (26) Ghosh, S. K.; Nath, S.; Kundu, S.; Esumi, K.; Pal, T. J. Phys. Chem. B 2004, 108, 13963. (27) Hutter, E.; Fendler, J. H. AdV. Mater. 2004, 16, 1685. (28) Ly, N.; Foley, K.; Tao, N. Anal. Chem. 2007, 79, 2546. (29) Chau, L.-K.; Lin, Y.-F.; Cheng, S.-F.; Lin, T.-J. Sens. Actuators B 2006, 113, 100. (30) Zhao, J.; Zhang, X.; Yonzon, C. R.; Haes, A. J.; Van Duyne, R. P. Nanomedicine 2006, 1, 219. (31) Zhang, X.; Shah, N. C.; Van Duyne, R. P. Vib. Spectrosc. 2006, 42, 2. (32) Bohren, C. F.; Huffman, D. R. Absorption and Scattering of Light by Small Particles; John Wiley and Sons: New York, 1983; p 530. (33) Kreibig, U.; Vollmer, M. Optical Properties of Metal Clusters; Springer: Berlin, 1995; Vol. 25, p 532. (34) Willets, K. A.; Van Duyne, R. P. Ann. ReV. Phys. Chem. 2006, 58, 267. (35) Kelly, K. L.; Coronado, E.; Zhao, L. L.; Schatz, G. C. J. Phys. Chem. B 2003, 107, 668. (36) Underwood, S.; Mulvaney, P. Langmuir 1994, 10, 3427. (37) Mulvaney, P. Langmuir 1996, 12, 788. (38) Mie, G. Ann. Phys. 1908, 25, 377. (39) Calander, N.; Willander, M. J. Appl. Phys. 2002, 92, 4878. (40) Asano, S.; Yamamoto, G. Appl. Opt. 1975, 14, 29. (41) Voshchinnikov, N. V.; Farafonov, V. G. Astrophys. Space Sci. 1993, 204, 19. (42) Sun, Y. G.; Xia, Y. N. Proc. SPIE 2003, 5221, 164. (43) Link, S.; El-Sayed, M. A. Ann. ReV. Phys. Chem. 2003, 54, 331. (44) Perez-Juste, J.; Pastoriza-Santos, I.; Liz-Marzan, L. M.; Mulvaney, P. Coord. Chem. ReV. 2005, 249, 1870. (45) Perez-Juste, J.; Liz-Marzan, L. M.; Carnie, S.; Chan, D. Y. C.; Mulvaney, P. AdV. Funct. Mater. 2004, 14, 571. (46) Kou, X.; Zhang, S.; Tsung, C.-K.; Yang, Z.; Yeung, M. H.; Stucky, G. D.; Sun, L.; Wang, J.; Yan, C. Chem. Eur. J. 2007, 13, 2929. (47) Hu, M.; Wang, X.; Hartland, G. V.; Mulvaney, P.; Perez-Juste, J.; Sader, J. E. J. Am. Chem. Soc. 2003, 125, 14925. (48) Link, S.; Mohamed, M. B.; El-Sayed, M. A. J. Phys. Chem. B 1999, 103, 3073. (49) Yu, Y.-Y.; Chang, S.-S.; Lee, C.-L.; Wang, C. R. C. J. Phys. Chem. B 1997, 101, 6661. (50) Lee, K.-S.; El-Sayed, M. A. J. Phys. Chem. B 2005, 109, 20331. (51) El-Sayed, M. A. Acc. Chem. Res. 2001, 34, 257. (52) Kottmann, J.; Martin, O.; Smith, D.; Schultz, S. Opt. Express 2000, 6, 213. (53) Hao, E.; Bailey, R. C.; Schatz, G. C.; Hupp, J. T.; Li, S. Nano Lett. 2004, 4, 327. (54) Kuwata, H.; Tamaru, H.; Esumi, K.; Miyano, K. Appl. Phys. Lett. 2003, 83, 4625. (55) Barber, P. W.; Hill, S. C. Light Scattering by Particles: Computational Methods; World Scientific Publishing: Singapore, 1990; Vol. 2. (56) Yang, W.-H.; Schatz, G. C.; Van Duyne, R. P. J. Chem. Phys. 1995, 103, 869. (57) Sosa, I. O.; Noguez, C.; Barrera, R. G. J. Phys. Chem. B 2003, 107, 6269. (58) Shlager, K. L.; Schneider, J. B. IEEE Antenn. Propag. Mag. 1995, 37, 39. (59) Gray, S. K.; Kupta, T. Phys. ReV. B 2003, 68, 045415. (60) Meier, M.; Wokaun, A. Opt. Lett. 1983, 8, 581. (61) Wokaun, A.; Gordon, J. P.; Liao, P. F. Phys. ReV. Lett. 1982, 48, 957. (62) Maier, S. A.; Brongersma, M. L.; Kik, P. G.; Atwater, H. A. Phys. ReV. B 2002, 65, 193408/1. (63) Maier, S. A.; Kik, P. G.; Atwater, H. A. Appl. Phys. Lett. 2002, 81, 1714. (64) Maier, S. A.; Kik, P. G.; Atwater, H. A.; Meltzer, S.; Harel, E.; Koel, B. E.; Requicha, A. A. G. Nat. Mater. 2003, 2, 229. (65) Gunnarsson, L.; Rindzevicius, T.; Prikulis, J.; Kasemo, B.; Kall, M.; Zou, S.; Schatz, G. C. J. Phys. Chem. B 2005, 109, 1079. (66) Su, K.-H.; Wei, Q.-H.; Zhang, X.; Mock, J. J.; Smith, D. R.; Schultz, S. Nano Lett. 2003, 3, 1087. (67) Rechberger, W.; Hohenau, A.; Leitner, A.; Krenn, J. R.; Lamprecht, B.; Aussenegg, F. R. Opt. Commun. 2003, 220, 137. (68) Krenn, J. R.; Dereux, A.; Weeber, J. C.; Bourillot, E.; Lacroute, Y.; Goudonnet, J. P.; Schider, G.; Gotschy, W.; Leitner, A.; Aussenegg, F. R.; Girard, C. Phys. ReV. Lett. 1999, 82, 2590. (69) Lamprecht, B.; Schider, G.; Lechner, R. T.; Ditlbacher, H.; Krenn, J. R.; Leitner, A.; Aussenegg, F. R. Phys. ReV. Lett. 2000, 84, 4721. (70) Mirkin, C. A. L.; Robert, L.; Mucic, R. C.; Storhoff, J. J. Nature 1996, 382, 607. (71) Born, M.; Wolf, E. 1999.

Nanostructured Plasmonic Sensors (72) Prikulis, J.; Hanarp, P.; Olofsson, L.; Sutherland, D.; Kaell, M. Nano Lett. 2004, 4, 1003. (73) Wannemacher, R. Opt. Commun. 2001, 195, 107. (74) Yin, L.; Vlasko-Vlasov, V. K.; Rydh, A.; Pearson, J.; Welp, U.; Chang, S.-H.; Gray, S. K.; Brown, D. B.; Kimball, C. W. Appl. Phys. Lett. 2004, 85, 467. (75) Ebbesen, T. W.; Lezec, H. J.; Ghaemi, H. F.; Thio, T.; Wolff, P. A. Nature 1998, 391, 667. (76) Chang, S.-H.; Gray, S. K.; Schatz, G. C. Opt. Express 2005, 13, 3150. (77) Stewart, M. E.; Mack, N. H.; Malyarchuk, V.; Soares, J. A. N. T.; Lee, T.-W.; Gray, S. K.; Nuzzo, R. G.; Rogers, J. A. PNAS 2006, 103, 17143. (78) Yin, L.; Vlasko-Vlasov, V. K.; Pearson, J.; Hiller, J. M.; Hua, J.; Welp, U.; Brown, D. E.; Kimball, C. W. Nano Lett. 2005, 5, 1399. (79) Dionne, J. A.; Lezec, H. J.; Atwater, H. A. Nano Lett. 2006, 6, 1928. (80) Satuby, Y.; Orenstein, M. Opt. Express 2007, 15, 4247. (81) Weeber, J. C.; Gonzalez, M. U.; Baudrion, A.-L.; Dereux, A. Appl. Phys. Lett. 2005, 87, 221101/1. (82) Bozhevolnyi, S. I.; Volkov, V. S.; Devaux, E.; Ebbesen, T. W. Phys. ReV. Lett. 2005, 95, 046802. (83) Ko, S. H.; Park, I.; Pan, H.; Grigoropoulos, C. P.; Pisano, A. P.; Luscombe, C. K.; Frechet, J. M. J. Nano Lett. 2007, ACS ASAP. (84) Herderick, E. D.; Tresback, J. S.; Vasiliev, A. L.; Padture, N. P. Nanotechnology 2007, 18, 155204/1. (85) Bluemel, A.; Klug, A.; Eder, S.; Scherf, U.; Moderegger, E.; List, E. J. W. Org. Electron. 2007, 8, 389. (86) Ko, S. H.; Pan, H.; Grigoropoulos, C. P.; Luscombe, C. K.; Frechet, J. M. J.; Poulikakos, D. Appl. Phys. Lett. 2007, 90, 141103/1. (87) Liu, J.; Lu, Y. Angew. Chem., Int. Ed. 2006, 45, 90. (88) Lee, J.-S.; Han, M. S.; Mirkin, C. A. Angew. Chem., Int. Ed. 2007, 46, 4093. (89) Storhoff, J. J.; Elghanian, R.; Mucic, R. C.; Mirkin, C. A.; Letsinger, R. L. J. Am. Chem. Soc. 1998, 120, 1959. (90) Gao, D.; Chen, W.; Mulchandani, A.; Schultz, J. S. Appl. Phys. Lett. 2007, 90, 073901/1. (91) Haes, A. J.; Haynes, C. L.; McFarland, A. D.; Schatz, G. C.; Van Duyne, R. P.; Zou, S. MRS Bull. 2005, 30, 368. (92) Haes, A. J.; Chang, L.; Klein, W. L.; Van Duyne, R. P. J. Am. Chem. Soc. 2005, 127, 2264. (93) Chen, J.; Wang, D.; Xi, J.; Au, L.; Siekkinen, A.; Warsen, A.; Li, Z.-Y.; Zhang, H.; Xia, Y.; Li, X. Nano Lett. 2007, 7, 1318. (94) Loo, C.; Lowery, A.; Halas, N.; West, J.; Drezek, R. Nano Lett. 2005, 5, 709. (95) Loo, C.; Lin, A.; Hirsch, L.; Lee, M.-H.; Barton, J.; Halas, N.; West, J.; Drezek, R. Technol. Cancer Res. Treat. 2004, 3, 33. (96) Huang, X.; El-Sayed, I. H.; Qian, W.; El-Sayed, M. A. J. Am. Chem. Soc. 2006, 128, 2115. (97) Durr, N. J.; Larson, T.; Smith, D. K.; Korgel, B. A.; Sokolov, K.; Ben-Yakar, A. Nano Lett. 2007, 7, 941. (98) Chen, H. M.; Hsin, C. F.; Liu, R.-S.; Lee, J.-F.; Jang, L.-Y. J. Phys. Chem. C 2007, 111, 5909. (99) Lu, L.; Kobayashi, A.; Tawa, K.; Ozaki, Y. Chem. Mater. 2006, 18, 4894. (100) Murphy, C. J.; Sau, T. K.; Gole, A.; Orendorff, C. J. MRS Bull. 2005, 30, 349. (101) Wiley, B.; Im, S. H.; Li, Z.-Y.; McLellan, J.; Siekkinen, A.; Xia, Y. J. Phys. Chem. B 2006, 110, 15666. (102) Xiong, Y.; Cai, H.; Wiley, B. J.; Wang, J.; Kim, M. J.; Xia, Y. J. Am. Chem. Soc. 2007, 129, 3665. (103) Nehl, C. L.; Liao, H.; Hafner, J. H. Nano Lett. 2006, 6, 683. (104) Evanoff, D. D.; Chumanov, G. J. Phys. Chem. B 2004, 108, 13948. (105) Grand, J.; Adam, P.-M.; Grimault, A.-S.; Vial, A.; de la Chapelle, M. L.; Bijeon, J.-L.; Kostcheev, S.; Royer, P. Plasmonics 2006, 1, 135. (106) Whitesides, G. M.; Love, C. J. Sci. Am. 2001, 285, 38. (107) Hanarp, P.; Sutherland, D. S.; Gold, J.; Kasemo, B. Colloids Surf., A 2003, 214, 23. (108) Aizpurua, J.; Hanarp, P.; Sutherland, D. S.; Kall, M.; Bryant, G. W.; Garcia de Abajo, F. J. Phys. ReV. Lett. 2003, 90, 057401. (109) Haynes, C. L.; Van Duyne, R. P. J. Phys. Chem. B 2001, 105, 5599. (110) Hicks, E. M.; Lyandres, O.; Hall, W. P.; Zou, S.; Glucksberg, M. R.; Van Duyne, R. P. J. Phys. Chem. B 2007, 111, 4116. (111) Haes, A. J.; Zhao, J.; Zou, S.; Own, C. S.; Marks, L. D.; Schatz, G. C.; Van Duyne, R. P. J. Phys. Chem. B 2005, 109, 11158. (112) Hostetler, M. J.; Wingate, J. E.; Zhong, C.-J.; Harris, J. E.; Vachet, R. W.; Clark, M. R.; Londono, J. D.; Green, S. J.; Stokes, J. J.; Wignall, G. D.; Glish, G. L.; Porter, M. D.; Evans, N. D.; Murray, R. W. Langmuir 1998, 14, 17. (113) Frens, G. Nat.: Phys. Sci. 1973, 241, 20. (114) Pillai, Z. S.; Kamat, P. V. J. Phys. Chem. B 2004, 108, 945. (115) Jacob, J. A.; Kapoor, S.; Biswas, N.; Mukherjee, T. Colloids Surf., A 2007, 301, 329.

Chemical Reviews, 2008, Vol. 108, No. 2 517 (116) He, P.; Zhu, X. Mater. Res. Bull. 2007, 42, 1310. (117) Sun, Y.; Zhang, L.; Zhou, H.; Zhu, Y.; Sutter, E.; Ji, Y.; Rafailovich, M. H.; Sokolov, J. C. Chem. Mater. 2007, 19, 2065. (118) Liz-Marzan, L. M. Langmuir 2006, 22, 32. (119) Mallin, M. P.; Murphy, C. J. Nano Lett. 2002, 2, 1235. (120) Steinbruck, A.; Csaki, A.; Festag, G.; Fritzsche, W. Plasmonics 2006, 1, 79. (121) Lu, X.; Tuan, H.-Y.; Chen, J.; Li, Z.-Y.; Korgel, B. A.; Xia, Y. J. Am. Chem. Soc. 2007, 129, 1733. (122) Selvakannan, P. R.; Sastry, M. Chem. Commun. 2005, 1684. (123) Roucoux, A.; Schulz, J.; Patin, H. Chem. ReV. 2002, 102, 3757. (124) Khomutov, G. B.; Koksharov, Y. A. AdV. Colloid Interface Sci. 2006, 122, 119. (125) Ott, L. S.; Finke, R. G. Inorg. Chem. 2006, 45, 8382. (126) Anad, M.; Bell, P. W.; Fan, X.; Enick, R. M.; Roberts, C. B. J. Phys. Chem. B 2006, 110, 14693. (127) Hunter, R. J. Foundations of Colloid Science; Oxford University Press: Oxford, 2001. (128) Starowicz, M.; Stypula, B.; Banas, J. Electrochem. Commun. 2006, 8, 227. (129) Plieth, W.; Dietz, H.; Anders, A.; Sandmann, G.; Meixner, A.; Weber, M.; Kneppe, H. Surf. Sci. 2005, 597, 119. (130) Vosgroene, T.; Meixner, A. J.; Anders, A.; Dietz, H.; Sandmann, G.; Plieth, W. Surf. Sci. 2005, 597, 102. (131) Lee, W.; Scholz, R.; Nielsch, K.; Goesele, U. Angew. Chem., Int. Ed. 2005, 44, 6050. (132) Huang, C.-J.; Chiu, P.-H.; Wang, Y.-H.; Yang, C.-F.; Feng, S.-W. J. Colloid Interface Sci. 2007, 306, 56. (133) Kim, F.; Song, J. H.; Yang, P. J. Am. Chem. Soc. 2002, 124, 14316. (134) Zhu, J.; Shen, Y.; Xie, A.; Qiu, L.; Zhang, Q.; Zhang, S. J. Phys. Chem. C 2007, 111, 7629. (135) Sakamoto, M.; Tachikawa, T.; Fujitsuka, M.; Majima, T. AdV. Funct. Mater. 2007, 17, 857. (136) McGilvray, K. L.; Decan, M. R.; Wang, D.; Scaiano, J. C. J. Am. Chem. Soc. 2006, 128, 15980. (137) Park, J.-E.; Atobe, M.; Fuchigami, T. Ultrason. Sonochem. 2006, 13, 237. (138) Su, C.-H.; Wu, P.-L.; Yeh, C.-S. J. Phys. Chem. B 2003, 107, 14240. (139) Jiang, L.-P.; Wang, A.-N.; Zhao, Y.; Zhang, J.-R.; Zhu, J.-J. Inorg. Chem. Commun. 2004, 7, 506. (140) Rodriguez-Gonzalez, B.; Mulvaney, P.; Liz-Marzan, L. M. Z. Phys. Chem. 2007, 221, 415. (141) Pal, A.; Shah, S.; Devi, S. Colloids Surf., A 2007, 302, 483. (142) Kariuki, N. N.; Luo, J.; Maye, M. M.; Hassan, S. A.; Menard, T.; Naslund, H. R.; Lin, Y.; Wang, C.; Engelhard, M. H.; Zhong, C.-J. Langmuir 2004, 20, 11240. (143) Toshima, N.; Yonezawa, T. New J. Chem. 1998, 22, 1179. (144) Wiley, B.; Herricks, T.; Sun, Y.; Xia, Y. Nano Lett. 2004, 4, 1733. (145) Yuan, H.; Ma, W.; Chen, C.; Zhao, J.; Liu, J.; Zhu, H.; Gao, X. Chem. Mater. 2007, 19, 1592. (146) Srnova-Sloufova, I.; Vlckova, B.; Bastl, Z.; Hasslett, T. L. Langmuir 2004, 20, 3407. (147) Youk, J. H.; Park, M.-K.; Locklin, J.; Advincula, R.; Yang, J.; Mays, J. Langmuir 2002, 18, 2455. (148) Shi, X.; Ganser, T. R.; Sun, K.; Balogh, L. P.; Baker, J., Jr. Nanotechnology 2006, 17, 1072. (149) Shahverdi, A. R.; Minaeian, S.; Shahverdi, H. R.; Jamalifar, H.; Nohi, A.-A. Process Biochem. 2007, 42, 919. (150) Rangnekar, A.; Sarma, T. K.; Singh, A. K.; Deka, J.; Ramesh, A.; Chattopadhyay, A. Langmuir 2007, 23, 5700. (151) Willner, I.; Baron, R.; Willner, B. AdV. Mater. 2005, 18, 1109. (152) Gericke, M.; Pinches, A. Gold Bull. 2006, 39, 22. (153) Klaus, T.; Joerger, R.; Olsson, E.; Granqvist, C.-G. PNAS 1999, 96, 13611. (154) Raveendran, P.; Fu, J.; Wallen, S. L. J. Am. Chem. Soc. 2003, 125, 13940. (155) McLeod, C. M.; Gale, W. F.; Roberts, C. B. Langmuir 2004, 20, 7078. (156) Dahl, J. A.; Maddux, B. L. S.; Hutchison, J. E. Chem. ReV. 2007, 107, 2228. (157) Turkevitch, J.; Stevenson, P. C.; Hillier, J. Discuss. Faraday Soc. 1951, 11, 55. (158) Hayat, M. A. Colloidal Gold: Principles, Methods and Applications; Academic Press: San Diego, 1989. (159) Brust, M.; Walker, M.; Bethell, D.; Schiffrin, D. J.; Whyman, R. J. Chem. Soc., Chem. Commun. 1994, 801. (160) Donkers, R. L.; Lee, D.; Murray, R. W. Langmuir 2004, 20, 1945. (161) Jimenez, V. L.; Georganopoulou, D. G.; White, R. J.; Harper, A. S.; Mills, A. J.; Lee, D.; Murray, R. W. Langmuir 2004, 20, 6864. (162) Nerambourg, N.; Werts, M. H. V.; Charlot, M.; Blanchard-Desce, M. Langmuir 2007, 23, 5563. (163) Tracy, J. B.; Kalyuzhny, G.; Crowe, M. C.; Balasubramanian, R.; Choi, J.-P.; Murray, R. W. J. Am. Chem. Soc. 2007, 129, 6706.

518 Chemical Reviews, 2008, Vol. 108, No. 2 (164) Ackersonn, C. J.; Sykes, M. T.; Kornberg, R. D. PNAS 2005, 102, 13383. (165) Becker, C. F. W.; Marsac, Y.; Hazarika, P.; Moser, J.; Goody, R. S.; Niemeyer, C. M. Chembiochem 2007, 8, 32. (166) Hostetler, M. J.; Templeton, A. C.; Murray, R. W. Langmuir 1999, 15, 3782. (167) Kassam, A.; Bremner, G.; Clark, B.; Ulibarri, G.; Lennox, R. B. J. Am. Chem. Soc. 2006, 128, 3476. (168) Shaffer, A. W.; Worden, J. G.; Huo, Q. Langmuir 2004, 20, 8343. (169) Montalti, M.; Prodi, L.; Zaccheroni, N.; Baxter, R.; Teobaldi, G.; Zerbetto, F. Langmuir 2003, 19, 5172. (170) Jana, N. R.; Peng, X. J. Am. Chem. Soc. 2003, 125, 14280. (171) Leff, D. V.; Brandt, L.; Heath, J. R. Langmuir 1996, 12, 4723. (172) Gomez, S.; Philippot, K.; Colliere, V.; Chaudret, B.; Senocq, F.; Lecante, P. Chem. Commun. 2000, 1945. (173) Weare, W. W.; Reed, S. M.; Warner, M. G.; Hutchison, J. E. J. Am. Chem. Soc. 2000, 122, 12890. (174) Yamamoto, M.; Nakamoto, M. Chem. Lett. 2003, 32, 452. (175) Templeton, A. C.; Chen, S.; Gross, S. M.; Murray, R. W. Langmuir 1999, 15, 66. (176) Chen, S.; Kimura, K. Langmuir 1999, 15, 1075. (177) Rowe, M. P.; Plass, K. E.; Kim, K.; Kurdak, C.; Zellers, E. T.; Matzger, A. J. Chem. Mater. 2004, 16, 3513. (178) Jana, N. R.; Gearheart, L.; Murphy, C. J. AdV. Mater. 2001, 13, 1389. (179) Iqbal, M.; Chung, Y.-I.; Tae, G. J. Mater. Chem. 2007, 17, 335. (180) Brown, K. R.; Walter, D. G.; Natan, M. J. Chem. Mater. 2000, 12, 306. (181) Brown, K. R.; Natan, M. J. Langmuir 1998, 14, 726. (182) Zou, X.; Ying, E.; Dong, S. Nanotechnology 2006, 17, 4758. (183) Gu, G. H.; Kim, J.; Kim, L.; Suh, J. S. J. Phys. Chem. C 2007, 111, 7906. (184) Wang, H.-W.; Shieh, C.-F.; Chen, H.-Y.; Shiu, W.-C.; Russo, B.; Cao, G. Nanotechnology 2006, 17, 2689. (185) Wang, Z. L.; Mohamed, M. B.; Link, S.; El-Sayed, M. A. Surf. Sci. 1999, 440, L809. (186) Niidome, Y.; Nishioka, K.; Kawasaki, H.; Yamada, S. Colloids Surf., A 2005, 257-258, 161. (187) Jana, N. R.; Gearheart, L.; Murphy, C. J. Chem. Commun. 2001, 617. (188) Gole, A.; Murphy, C. J. Chem. Mater. 2004, 16, 3633. (189) Sau, T. K.; Murphy, C. J. Langmuir 2004, 20, 6414. (190) Nikoobakht, B.; El-Sayed, M. A. Chem. Mater. 2003, 15, 1957. (191) Ha, T. H.; Koo, H.-J.; Chung, B. H. J. Phys. Chem. C 2007, 111, 1123. (192) Mieszawska, A. J.; Jalilian, R.; Sumanasekera, G. U.; Zamborini, F. P. J. Am. Chem. Soc. 2005, 127, 10822. (193) Mieszawska, A. J.; Slawinski, G. W.; Zamborini, F. P. J. Am. Chem. Soc. 2006, 128, 5622. (194) Mieszawska, A. J.; Zamborini, F. P. Chem. Mater. 2005, 17, 3415. (195) Lee, K.-H.; Huang, K.-M.; Tseng, W.-L.; Chiu, T.-C.; Lin, Y.-W.; Chang, H.-T. Langmuir 2007, 23, 1435. (196) Neeves, A. E.; Birnboim, M. H. J. Opt. Soc. Am. B 1989, 6, 787. (197) Oldenburg, S. J.; Averitt, R. D.; Westcott, S. L.; Halas, N. J. Chem. Phys. Lett. 1998, 288, 243. (198) Halas, N. MRS Bull. 2005, 30, 362. (199) Wang, H.; Brandl, D. W.; Le, F.; Nordlander, P.; Halas, N. J. Nano Lett. 2006, 6, 827. (200) Wiley, B.; Sun, Y.; Chen, J.; Cang, H.; Li, Z.-Y.; Li, X.; Xia, Y. MRS Bull. 2005, 30, 356. (201) Sun, Y.; Mayers, B.; Xia, Y. Nano Lett. 2003, 3, 675. (202) Sun, Y.; Mayers, B.; Xia, Y. AdV. Mater. 2003, 15, 641. (203) Chen, J.; Wiley, B.; Li, Z.-Y.; Campbell, D.; Saeki, F.; Cang, H.; Au, L.; Lee, J.; Li, X.; Xia, Y. AdV. Mater. 2005, 17, 2255. (204) Pileni, M.-P. Nat. Mater. 2003, 2, 145. (205) Landfester, K. Annu. ReV. Mater. Res. 2006, 36, 231. (206) Salzemann, C.; Urban, J.; Lisiecki, I.; Pileni, M.-P. AdV. Funct. Mater. 2005, 15, 1277. (207) Maillard, M.; Giorgio, S.; Pileni, M.-P. J. Phys. Chem. B 2003, 107, 2466. (208) Hou, G.; Zhu, L.; Chen, D.; Jiang, M. Macromolecules 2007, 40, 2134. (209) Chaudhari, V. R.; Haram, S. K.; Kulshreshtha, S. K.; Bellare, J. R.; Hassan, P. A. Colloids Surf., A 2007, 301, 475. (210) Cushing, B. L.; Kolesnichenko, V. L.; O’Connor, C. J. Chem. ReV. 2004, 104, 3893. (211) Ohno, T.; Bain, J. A.; Schlesinger, T. E. J. Appl. Phys. 2007, 101, 083107/1. (212) Wang, S.; Pile, D. F. P.; Sun, C.; Zhang, X. Nano Lett. 2007, 7, 1076. (213) Billot, L.; Lamy de la Chapelle, M.; Grimault, A. S.; Vial, A.; Barchiesi, D.; Bijeon, J. L.; Adam, P. M.; Royer, P. Chem. Phys. Lett. 2006, 422, 303.

Stewart et al. (214) De Jesus, M. A.; Giesfeldt, K. S.; Oran, J. M.; Abu-Hatab, N. A.; Lavrik, N. V.; Sepaniak, M. J. Appl. Spectrosc. 2005, 59, 1501. (215) Corrigan, T. D.; Guo, S.; Phaneuf, R. J.; Szmacinski, H. J. Fluoresc. 2005, 15, 777. (216) Yamazaki, K.; Namatsu, H. Jpn. J. Appl. Phys. 2004, 43, 3767. (217) Word, M. J.; Adesida, I.; Berg, P. R. J. Vac. Sci. Technol. B 2003, 21, L12. (218) Isaacson, M.; Murray, A. J. Vac. Sci. Technol. 1981, 19, 1117. (219) Hinsberg, W.; Houle, F.; Sanchez, M.; Hoffnagle, J.; Wallraff, G.; Medeiros, D.; Gallatin, G.; Cobbc, J. Proc. SPIE 2003, 5039, 1. (220) Fu, Y.; Ngoi, B. K. A.; Zhou, W.; Loh, T. F. Int. J. Nanosci. 2006, 5, 529. (221) Chang, C.-K.; Lin, D.-Z.; Yeh, C.-S.; Lee, C.-K.; Chang, Y.-C.; Lin, M.-W.; Yeh, J.-T.; Liu, J.-M. Appl. Phys. Lett. 2007, 90, 061113/1. (222) Cabrini, S.; Carpentiero, A.; Kumar, R.; Businaro, L.; Candeloro, P.; Prasciolu, M.; Gosparini, A.; Andreani, C.; De Vittorio, M.; Stomeo, T.; Di Fabrizio, E. Microelectron. Eng. 2005, 78-79, 11. (223) Taniguchi, J.; Ohno, N.; Takeda, S.; Miyamoto, I.; Komuro, M. J. Vac. Sci. Technol. B 1998, 16, 2506. (224) Reyntjens, S.; Puers, R. J. Micromech. Microeng. 2000, 10, 181. (225) Ishida, M.; Fujita, J.; Ichihashi, T.; Ochiai, Y.; Kaito, T.; Matsui, S. J. Vac. Sci. Technol. B 2003, 21, 2728. (226) Kubena, R. L.; Ward, J. W.; Stratton, F. P.; Joyce, R. L.; Atkinson, G. M. J. Vac. Sci. Technol. B 1991, 9, 3079. (227) Gierak, J.; Vieu, C.; Schneider, M.; Launois, H.; Ben Assayag, G.; Septier, A. J. Vac. Sci. Technol. B 1997, 15, 2373. (228) Przybilla, F.; Genet, C.; Ebbesen, T. W. Appl. Phys. Lett. 2006, 89, 121115/1. (229) Dintinger, J.; Klein, S.; Ebbesen, T. W. AdV. Mater. 2006, 18, 1267. (230) Gonzalez, M. U.; Weeber, J.-C.; Baudrion, A.-L.; Dereux, A.; Stepanov, A. L.; Krenn, J. R.; Devaux, E.; Ebbesen, T. W. Phys. ReV. B 2006, 73, 155416/1. (231) Lopez-Tejeira, F.; Rodrigo, S. G.; Martin-Moreno, L.; Garcia-Vidal, F. J.; Devaux, E.; Ebbesen, T. W.; Krenn, J. R.; Radko, I. P.; Bozhevolnyi, S. I.; Gonzalez, M. U.; Weeber, J. C.; Dereux, A. Nat. Phys. 2007, 3, 324. (232) Volkov, V. S.; Bozhevolnyi, S. I.; Devaux, E.; Laluet, J.-Y.; Ebbesen, T. W. Nano Lett. 2007, 7, 880. (233) Malinsky, M. D.; Kelly, K. L.; Schatz, G. C.; Van Duyne, R. P. J. Phys. Chem. B 2001, 105, 2343. (234) Fischer, U. C.; Zingsheim, H. P. J. Vac. Sci. Technol. 1981, 19, 881. (235) Deckman, H. W.; Dunsmuir, J. H. Appl. Phys. Lett. 1982, 41, 377. (236) Hulteen, J. C.; Treichel, D. A.; Smith, M. T.; Duval, M. L.; Jensen, T. R.; Van Duyne, R. P. J. Phys. Chem. B 1999, 103, 3854. (237) Hulteen, J. C.; Van Duyne, R. P. J. Vac. Sci. Technol. A 1995, 13, 1553. (238) Whitney, A. V.; Myers, B. D.; Van Duyne, R. P. Nano Lett. 2004, 4, 1507. (239) Freunscht, P.; Van Duyne, R. P.; Schneider, S. Chem. Phys. Lett. 1997, 281, 372. (240) Hulteen, J. C.; Young, M. A.; Van Duyne, R. P. Langmuir 2006, 22, 10354. (241) Stuart, D. A.; Yuen, J. M.; Shah, N.; Lyandres, O.; Yonzon, C. R.; Glucksberg, M. R.; Walsh, J. T.; Van Duyne, R. P. Anal. Chem. 2006, 78, 7211. (242) Zhang, X.; Zhao, J.; Whitney, A. V.; Elam, J. W.; Van Duyne, R. P. J. Am. Chem. Soc. 2006, 128, 10304. (243) Riboh, J. C.; Haes, A. J.; McFarland, A. D.; Yonzon, C. R.; Van Duyne, R. P. J. Phys. Chem. B 2003, 107, 1772. (244) Schmidt, J. P.; Cross, S. E.; Buratto, S. K. J. Chem. Phys. 2004, 121, 10657. (245) Haynes, C. L.; Van Duyne, R. P. J. Phys. Chem. B 2003, 107, 7426. (246) Haes, A. J.; Hall, W. P.; Chang, L.; Klein, W. L.; Van Duyne, R. P. Nano Lett. 2004, 4, 1029. (247) Kelf, T. A.; Sugawara, Y.; Cole, R. M.; Baumberg, J. J.; Abdelsalam, M. E.; Cintra, S.; Mahajan, S.; Russell, A. E.; Bartlett, P. N. Phys. ReV. B 2006, 74, 245415/1. (248) Kelf, T. A.; Sugawara, Y.; Baumberg, J. J.; Abdelsalam, M.; Bartlett, P. N. Phys. ReV. Lett. 2005, 95, 116802. (249) Baumberg, J. J.; Kelf, T. A.; Sugawara, Y.; Cintra, S.; Abdelsalam, M. E.; Bartlett, P. N.; Russell, A. E. Nano Lett. 2005, 5, 2262. (250) Cintra, S.; Abdelsalam, M. E.; Bartlett, P. N.; Baumberg, J. J.; Kelf, T. A.; Sugawara, Y.; Russell, A. E. Faraday Discuss. 2006, 132, 191. (251) Mahajan, S.; Abdelsalam, M.; Sugawara, Y.; Cintra, S.; Russell, A.; Baumberg, J.; Bartlett, P. Phys. Chem. Chem. Phys. 2007, 9, 104. (252) Abdelsalam, M. E.; Bartlett, P. N.; Baumberg, J. J.; Cintra, S.; Kelf, T. A.; Russell, A. E. Electrochem. Commun. 2005, 7, 740. (253) Hicks, E. M.; Zhang, X.; Zou, S.; Lyandres, O.; Spears, K. G.; Schatz, G. C.; Van Duyne, R. P. J. Phys. Chem. B 2005, 109, 22351. (254) Kosiorek, A.; Kandulski, W.; Chudzinski, P.; Kempa, K.; Giersig, M. Nano Lett. 2004, 4, 1359.

Nanostructured Plasmonic Sensors (255) Kosiorek, A.; Kandulski, W.; Glaczynska, H.; Giersig, M. Small 2005, 1, 439. (256) Haynes, C. L.; McFarland, A. D.; Hulteen, J. C.; Van Duyne, R. P. J. Phys. Chem. B 2002, 106, 1898. (257) Jensen, T. R.; Malinsky, M. D.; Haynes, C. L.; Van Duyne, R. P. J. Phys. Chem. B 2000, 104, 10549. (258) Zhang, X.; Hicks, E. M.; Zhao, J.; Schatz, G. C.; Van Duyne, R. P. Nano Lett. 2005, 5, 1503. (259) Rindzevicius, T.; Alaverdyan, Y.; Dahlin, A.; Hoeoek, F.; Sutherland, D. S.; Kaell, M. Nano Lett. 2005, 5, 2335. (260) Hanarp, P.; Kaell, M.; Sutherland, D. S. J. Phys. Chem. B 2003, 107, 5768. (261) Langhammer, C.; Yuan, Z.; Zoric, I.; Kasemo, B. Nano Lett. 2006, 6, 833. (262) Weibel, D. B.; DiLuzio, W. R.; Whitesides, G. M. Nat. ReV. Microbiol. 2007, 5, 209. (263) Rogers, J. A.; Nuzzo, R. G. Mater. Today 2005, 8, 50. (264) Xia, Y.; Whitesides, G. M. Ann. ReV. Mater. Sci. 1998, 28, 153. (265) Delamarche, E.; Schmid, H.; Michel, B.; Biebuyck, H. AdV. Mater. 1997, 9, 741. (266) Odom, T. W.; Love, J. C.; Wolfe, D. B.; Paul, K. E.; Whitesides, G. M. Langmuir 2002, 18, 5314. (267) Hua, F.; Sun, Y.; Gaur, A.; Meitl, M. A.; Bilhaut, L.; Rotkina, L.; Wang, J.; Geil, P.; Shim, M.; Rogers, J. A. Nano Lett. 2004, 4, 2467. (268) Schmid, H.; Michel, B. Macromolecules 2000, 33, 3042. (269) Choi, K. M.; Rogers, J. A. J. Am. Chem. Soc. 2003, 125, 4060. (270) Kwak, E.-S.; Henzie, J.; Chang, S.-H.; Gray, S. K.; Schatz, G. C.; Odom, T. W. Nano Lett. 2005, 5, 1963. (271) Henzie, J.; Barton, J. E.; Stender, C. L.; Odom, T. W. Acc. Chem. Res. 2006, 39, 249. (272) Odom, T. W.; Thalladi, V. R.; Love, J. C.; Whitesides, G. M. J. Am. Chem. Soc. 2002, 124, 12112. (273) Gao, H.; Henzie, J.; Odom, T. W. Nano Lett. 2006, 6, 2104. (274) Henzie, J.; Kwak, E.-S.; Odom, T. W. Nano Lett. 2005, 5, 1199. (275) Henzie, J.; Shuford, K. L.; Kwak, E.-S.; Schatz, G. C.; Odom, T. W. J. Phys. Chem. B 2006, 110, 14028. (276) Truskett, V. N.; Watts, M. P. C. Trends Biotechnol. 2006, 24, 312. (277) Chou, S. Y.; Krauss, P. R.; Zhang, W.; Guo, L.; Zhuang, L. J. Vac. Sci. Technol. B 1997, 15, 2897. (278) Stewart, M. D.; Johnson, S. C.; Sreenivasan, S. V.; Resnick, D. J.; Willson, G. C. J. Microlith., Microfab., Microsyst. 2005, 4, 011002/ 1. (279) Malyarchuk, V.; Stewart, M.; Nuzzo, R. G.; Rogers, J. A. Appl. Phys. Lett. 2007, 90, 203113/1. (280) Alvarez-Puebla, R. A.; Cui, B.; Bravo-Vasquez, J.-P.; Veres, T.; Fenniri, H. J. Phys. Chem. C 2007, 111, 6720. (281) Rosi, N. L.; Mirkin, C. A. Chem. ReV. 2005, 105, 1547. (282) Nakashima, H.; Furukawa, K.; Kashimura, Y.; Torimitsu, K. Chem. Commun. 2007, 1080. (283) Reynolds, R. A.; Mirkin, C. A.; Letsinger, R. L. J. Am. Chem. Soc. 2000, 122, 3795. (284) Schofield, C. L.; Haines, A. H.; Field, R. A.; Russell, D. A. Langmuir 2006, 22, 6707. (285) McFarland, A. D.; VanDuyne, R. P. Nano Lett. 2003, 3, 1057. (286) Yu, C.; Irudayaraj, J. Anal. Chem. 2007, 79, 572. (287) Han, M. S.; Lytton-Jean, A. K. R.; Mirkin, C. A. J. Am. Chem. Soc. 2006, 128, 4954. (288) Han, M. S.; Lytton-Jean, A. K. R.; Oh, B.-K.; Heo, J.; Mirkin, C. A. Angew. Chem., Int. Ed. 2006, 45, 1807. (289) Kanaras, A. G.; Wang, Z.; Brust, M.; Cosstick, R.; Bates, A. D. Small 2007, 3, 590. (290) Chakrabarti, R.; Klibanov, A. M. J. Am. Chem. Soc. 2003, 125, 12531. (291) Ellington, A. D. S., J. W. Nature 1990, 346, 818. (292) Huang, C. C.; Huang, Y. F.; Cao, Z.; Tan, W.; Chang, H. T. Anal. Chem. 2005, 77, 5735. (293) Tuerk, C. G., L. Science 1990, 249, 505. (294) Baron, R.; Zayats, M.; Willner, I. Anal. Chem. 2005, 77, 1566. (295) Dykman, L. A.; Bogatyrev, V. A.; Khlebtsov, B. N.; Khlebtsov, N. G. Anal. Biochem. 2005, 341, 16. (296) Kuong, C.-L.; Chen, W.-Y.; Chen, Y.-C. Anal. Bioanal. Chem. 2007, 387, 2091. (297) Lin, S.-Y.; Wu, S.-H.; Chen, C.-h. Angew. Chem., Int. Ed. 2006, 45, 4948. (298) Lu, C.; Zu, Y.; Yam, V. W. W. Anal. Chem. 2007, 79, 666. (299) Schofield, C. L.; Field, R. A.; Russell, D. A. Anal. Chem. 2007, 79, 1356. (300) Suzuki, D.; Kawaguchi, H. Langmuir 2006, 22, 3818. (301) Wang, Z.; Levy, R.; Fernig, D. G.; Brust, M. J. Am. Chem. Soc. 2006, 128, 2214. (302) Watanabe, S.; Seguchi, H.; Yoshida, K.; Kifune, K.; Tadaki, T.; Shiozaki, H. Tetrahedron Lett. 2005, 46, 8827. (303) Wu, Z.-S.; Zhang, S.-B.; Guo, M.-M.; Chen, C.-R.; Shen, G.-L.; Yu, R.-Q. Anal. Chim. Acta 2007, 584, 122.

Chemical Reviews, 2008, Vol. 108, No. 2 519 (304) Liu, J.; Lu, Y. J. Am. Chem. Soc. 2004, 126, 12298. (305) Liu, J.; Lu, Y. J. Am. Chem. Soc. 2003, 125, 6642. (306) Lin, S. Y.; Chen, C. h.; Lin, M. C.; Hsu, H. F. Anal. Chem. 2005, 77, 4821. (307) Homola, J. Anal. Bioanal. Chem. 2003, 377, 528. (308) Homola, J.; Yee, S. S.; Gauglitz, G. Sens. Actuators B 1999, 54, 3. (309) Fu, E.; Ramsey, S. A.; Chen, J.; Chinowsky, T. M.; Wiley, B.; Xia, Y.; Yager, P. Sens. Actuators B 2007, B123, 606. (310) Liu, X.; Song, D.; Zhang, Q.; Tian, Y.; Liu, Z.; Zhang, H. Sens. Actuators B 2006, B117, 188. (311) Liu, X.; Song, D.; Zhang, Q.; Tian, Y.; Ding, L.; Zhang, H. TrAC 2005, 24, 887. (312) Lee, H. J.; Li, Y.; Wark, A. W.; Corn, R. M. Anal. Chem. 2005, 77, 5096. (313) Li, Y.; Lee, H. J.; Corn, R. M. Anal. Chem. 2007, 79, 1082. (314) Lyon, L. A.; Musick, M. D.; Natan, M. J. Anal. Chem. 1998, 70, 5177. (315) Hutter, E.; Cha, S.; Liu, J.-F.; Park, J.; Yi, J.; Fendler, J. H.; Roy, D. J. Phys. Chem. B 2001, 105, 8. (316) Li, Y.; Wark, A. W.; Lee, H. J.; Corn, R. M. Anal. Chem. 2006, 78, 3158. (317) Mitchell, J. S.; Wu, Y.; Cook, C. J.; Main, L. Anal. Biochem. 2005, 343, 125. (318) Matsui, J.; Akamatsu, K.; Hara, N.; Miyoshi, D.; Nawafune, H.; Tamaki, K.; Sugimoto, N. Anal. Chem. 2005, 77, 4282. (319) He, L.; Musick, M. D.; Nicewarner, S. R.; Salinas, F. G.; Benkovic, S. J.; Natan, M. J.; Keating, C. D. J. Am. Chem. Soc. 2000, 122, 9071. (320) Teramura, Y.; Iwata, H. Anal. Biochem. 2007, 365, 201. (321) Gobi, K. V.; Matsumoto, K.; Toko, K.; Ikezaki, H.; Miura, N. Anal. Bioanal. Chem. 2007, 387, 2727. (322) Aizawa, H.; Tozuka, M.; Kurosawa, S.; Kobayashi, K.; Reddy, S. M.; Higuchi, M. Anal. Chim. Acta 2007, 591, 191. (323) Shankaran, D. R.; Gobi, K. V.; Miura, N. Sens. Actuators B 2007, 121, 158. (324) Goodrich, T. T.; Lee, H. J.; Corn, R. M. J. Am. Chem. Soc. 2004, 126, 4086. (325) Goodrich, T. T.; Lee, H. J.; Corn, R. M. Anal. Chem. 2004, 76, 6173. (326) Teramura, Y.; Arima, Y.; Iwata, H. Anal. Biochem. 2006, 357, 208. (327) He, L.; Smith, E. A.; Natan, M. J.; Keating, C. D. J. Phys. Chem. B 2004, 108, 10973. (328) Hutter, E.; Fendler, J. H.; Roy, D. J. Phys. Chem. B 2001, 105, 11159. (329) Lyon, A. L.; Pena, D. J.; Natan, M. J. J. Phys. Chem. B 1999, 103, 5826. (330) Lee, W.; Oh, B.-K.; Kim, Y.-W.; Choi, J.-W. J. Nanosci. Nanotechnol. 2006, 6, 3521. (331) Hayashida, M.; Yamaguchi, A.; Misawa, H. Jpn. J. Appl. Phys. 2005, 44, L1544. (332) Kang, T.; Hong, S.; Choi, I.; Sung, J. J.; Kim, Y.; Hahn, J. S.; Yi, J. J. Am. Chem. Soc. 2006, 128, 12870. (333) Fang, S.; Lee, H. J.; Wark, A. W.; Corn, R. M. J. Am. Chem. Soc. 2006, 128, 14044. (334) Tokareva, I.; Minko, S.; Fendler, J. H.; Hutter, E. J. Am. Chem. Soc. 2004, 126, 15950. (335) Tokareva, I.; Tokarev, I.; Minko, S.; Hutter, E.; Fendler, J. H. Chem. Commun. 2006, 3343. (336) Hutter, E.; Pileni, M. P. J. Phys. Chem. B 2003, 107, 6497. (337) Stuart, D. A.; Haes, A. J.; Yonzon, C. R.; Hicks, E. M.; Van Duyne, R. P. IEE Proc.-Nanobiotechnol. 2005, 152, 13. (338) Aslan, K.; Lakowicz, J. R.; Geddes, C. D. Curr. Opin. Chem. Biol. 2005, 9, 538. (339) Yguerabide, J.; Yguerabide, E. E. Anal. Biochem. 1998, 262, 137. (340) Yguerabide, J.; Yguerabide, E. E. Anal. Biochem. 1998, 262, 157. (341) Reinhard, B. M.; Sheikholeslami, S.; Mastroianni, A.; Alivisatos, A. P.; Liphardt, J. PNAS 2007, 104, 2667. (342) Sherry, L. J.; Chang, S.-H.; Schatz, G. C.; Van Duyne, R. P.; Wiley, B. J.; Younan, X. Nano Lett. 2005, 5, 2034. (343) Raschke, G.; Kowarik, S.; Franzl, T.; Sonnichsen, C.; Klar, T. A.; Feldmann, J.; Nichtl, A.; Kurzinger, K. Nano Lett. 2003, 3, 935. (344) Du, B.-A.; Li, Z.-P.; Liu, C.-H. Angew. Chem., Int. Ed. 2006, 45, 8022. (345) Bally, M.; Halter, M.; Voros, J.; Grandin, H. M. Surf. Interface Anal. 2006, 38, 1442. (346) Taton, A. T.; Mirkin, C. A.; Letsinger, R. L. Science 2000, 289, 1757. (347) Storhoff, J. J.; Marla, S. S.; Bao, P.; Hagenow, S.; Mehta, H.; Lucas, A.; Garimella, V.; Patno, T.; Buckingham, W.; Cork, W.; Muller, U. R. Biosens. Bioelectron. 2004, 19, 875. (348) Bao, P. Y.; Huber, M.; Wei, T.-F.; Marla, S. S.; Storhoff, J. J.; Muller, U. R. Nucleic Acids Res. 2005, 33, e15. (349) Alivisatos, P. Nat. Biotechnol. 2004, 22, 47. (350) Liu, G. L.; Yin, Y.; Kunchakarra, S.; Mukherjee, B.; Gerion, D.; Jett, S. D.; Bear, D. G.; Gray, J. W.; Alivisatos, A. P.; Lee, L. P.; Chen, F. F. Nat. Nanotechnol. 2006, 1, 47.

520 Chemical Reviews, 2008, Vol. 108, No. 2 (351) Reinhard, B. M.; Siu, M.; Agarwal, H.; Alivisatos, A. P.; Liphardt, J. Nano Lett. 2005, 5, 2246. (352) Storhoff, J. J.; Lucas, A. D.; Garimella, V.; Bao, Y. P.; Muller, U. R. Nat. Biotechnol. 2004, 22, 883. (353) Aslan, K.; Lacowicz, J. R.; Geddes, C. D. Anal. Chem. 2005, 77, 2007. (354) Roll, D.; Malicka, J.; Gryczynski, I.; Gryczynski, Z.; Lakowicz, J. R. Anal. Chem. 2003, 75, 3440. (355) Aslan, K.; Holley, P.; Davies, L.; Lakowicz, J. R.; Geddes, C. D. J. Am. Chem. Soc. 2005, 127, 12115. (356) Aslan, K.; Lakowicz, J. R.; Geddes, C. D. Appl. Phys. Lett. 2005, 87, 234108/1. (357) http://www.nanosphere.us/, 2007. (358) Pokorski, J. K.; Nam, J.-M.; Vega, R. A.; Mirkin, C. A.; Appella, D. H. Chem. Commun. 2005, 2101. (359) Nam, J.-M.; Stoeva, S. I.; Mirkin, C. A. J. Am. Chem. Soc. 2004, 126, 5932. (360) Katz, E.; Willner, I. Angew. Chem., Int. Ed. 2004, 43, 6042. (361) Lin, A. W. H.; Lewinski, N. A.; West, J. L.; Halas, N. J.; Drezek, R. A. J. Biomed. Opt. 2005, 10, 064035/1. (362) Sokolov, K.; Follen, M.; Aaron, J.; Pavlova, I.; Malpica, A.; Lotan, R.; Richards-Kortum, R. Cancer Res. 2003, 63, 1999. (363) Tkaczyk, T. S.; Rahman, M.; Mack, V.; Sokolov, K.; Rogers, J. D.; Richards-Kortum, R.; Descour, M. R. Opt. Express 2004, 12, 3745. (364) El-Sayed, I.; Huang, X.; El-Sayed, M. A. Nano Lett. 2005, 5, 829. (365) Aaron, J. S.; Oh, J.; Larson, T. A.; Kumar, S.; Milner, T. E.; Sokolov, K. V. Opt. Express 2006, 14, 12930. (366) Kumar, S.; Harrison, N.; Richards-Kortum, R.; Sokolov, K. Nano Lett. 2007, 7, 1338. (367) Wadia, J. S.; Stan, R. V.; Dowdy, S. F. Nat. Med. 2004, 309, 121. (368) Pissuwan, D.; Valenzuela, S. M.; Cortie, M. B. Trends Biotechnol. 2006, 24, 62. (369) El-Sayed, I. H.; Huang, X.; El-Sayed, M. A. Cancer Lett. 2006, 239, 129. (370) Pattnaik, P. Appl. Biochem. Biotechnol. 2005, 126, 79. (371) Myszka, D. G. Curr. Opin. Biotechnol. 1997, 8, 50. (372) Brockman, J. M.; Nelson, B. P.; Corn, R. M. Annu. ReV. Phys. Chem. 2000, 51, 41. (373) Piliarik, M.; Vaisocherova, H.; Homola, J. Sens. Actuators B 2007, B121, 187. (374) Schuck, P. Ann. ReV. Biophys. Biomol. Struct. 1997, 26, 541. (375) Hahnefeld, C.; Drewianka, S.; Herberg, F. W. Methods Mol. Med. 2004, 94, 299. (376) Nath, N.; Chilkoti, A. Anal. Chem. 2004, 76, 5370. (377) De Leebeeck, A.; Kumar, L. K. S.; de Lange, V.; Sinton, D.; Gordon, R.; Brolo, A. G. Anal. Chem. 2007, 79, 4094. (378) Tetz, K. A.; Pang, L.; Fainman, Y. Opt. Lett. 2006, 31, 1528. (379) Kalyuzhny, G.; Vaskevich, A.; Schneeweiss, M. A.; Rubinstein, I. Chem. Eur. J. 2002, 8, 3850. (380) Doron-Mor, I.; Cohen, H.; Barkay, Z.; Shanzer, A.; Vaskevich, A.; Rubinstein, I. Chem. Eur. J. 2005, 11, 5555. (381) Ruach-Nir, I.; Bendikov, T. A.; Doron-Mor, I.; Barkay, Z.; Vaskevich, A.; Rubinstein, I. J. Am. Chem. Soc. 2007, 129, 84. (382) Yonzon, C. R.; Jeoung, E.; Zou, S.; Schatz, G. C.; Mrksich, M.; Van Duyne, R. P. J. Am. Chem. Soc. 2004, 126, 12669. (383) Larsson, E. M.; Alegret, J.; Kaell, M.; Sutherland, D. S. Nano Lett. 2007, 7, 1256. (384) Nath, N.; Chilkoti, A. J. Fluoresc. 2004, 14, 377. (385) Frederix, F.; Friedt, J.-M.; Choi, K.-H.; Laureyn, W.; Campitelli, A.; Mondelaers, D.; Maes, G.; Borghs, G. Anal. Chem. 2003, 75, 6894. (386) Thio, T.; Ghaemi, H. F.; Lezec, H. J.; Wolff, P. A. J. Opt. Soc. Am. B 1999, 16, 1743. (387) Ekgasit, S.; Thammacharoen, C.; Yu, F.; Knoll, W. Appl. Spectrosc. 2005, 25, 661. (388) Mitsui, K.; Handa, Y.; Kajikawa, K. Appl. Phys. Lett. 2004, 85, 4231. (389) Weikel, A. L.; Conklin, S. D.; Richardson, J. N. Sens. Actuators B 2005, B110, 112. (390) Taue, S.; Nishida, K.; Sakaue, H.; Takahagi, T. e-J. Surf. Sci. Nanotechnol. 2007, 5, 74. (391) Malynych, S.; Chumanov, G. J. Opt. A: Pure Appl. Opt. 2006, 8, S144. (392) Fujiwara, K.; Watarai, H.; Itoh, H.; Nakahama, E.; Ogawa, N. Anal. Bioanal. Chem. 2006, 386, 639. (393) Lin, H.-Y.; Chen, C.-T.; Chen, Y.-C. Anal. Chem. 2006, 78, 6873. (394) Marinakos, S. M.; Chen, S.; Chilkoti, A. Anal. Chem. 2007, ASAP Article. (395) Cheng, S.-F.; Chau, L.-K. Anal. Chem. 2003, 75, 16. (396) Tam, F.; Moran, C.; Halas, N. J. Phys. Chem. B 2004, 108, 17290. (397) Jensen, T. R.; Duval, M. L.; Kelly, K. L.; Lazarides, A. A.; Schatz, G. C.; Van Duyne, R. P. J. Phys. Chem. B 1999, 103, 9846. (398) Dahlin, A. B.; Tegenfeldt, J. O.; Hook, F. Anal. Chem. 2006, 78, 4416.

Stewart et al. (399) Dahlin, A.; Zaech, M.; Rindzevicius, T.; Kaell, M.; Sutherland, D. S.; Hoeoek, F. J. Am. Chem. Soc. 2005, 127, 5043. (400) Lahav, M.; Vaskevich, A.; Rubinstein, I. Langmuir 2004, 20, 7365. (401) Kalyuzhny, G.; Schneeweiss, M. A.; Shanzer, A.; Vaskevich, A.; Rubinstein, I. J. Am. Chem. Soc. 2001, 123, 3177. (402) Haes, A. J.; Zou, S.; Schatz, G. C.; Van Duyne, R. P. J. Phys. Chem. B 2004, 108, 6961. (403) Haes, A. J.; Zou, S.; Schatz, G. C.; Van Duyne, R. P. J. Phys. Chem. B 2004, 108, 109. (404) Whitney, A. V.; Elam, J. W.; Zou, S.; Zinovev, A. V.; Stair, P. C.; Schatz, G. C.; Van Duyne, R. P. J. Phys. Chem. B 2005, 109, 20522. (405) Jung, L. S.; Campbell, C. T.; Chinowsky, T. M.; Mar, M. N.; Yee, S. S. Langmuir 1998, 14, 5636. (406) Zhao, J.; Das, A.; Zhang, X.; Schatz, G. C.; Sligar, S. G.; Van Duyne, R. P. J. Am. Chem. Soc. 2006, 128, 11004. (407) Haes, A. J.; Zou, S.; Zhao, J.; Schatz, G. C.; Van Duyne, R. P. J. Am. Chem. Soc. 2006, 128, 10905. (408) Mock, J. J.; Smith, D. R.; Schultz, S. Nano Lett. 2003, 3, 485. (409) Sherry, L. J.; Jin, R.; Mirkin, C. A.; Schatz, G. C.; Van Duyne, R. P. Nano Lett. 2006, 6, 2060. (410) Sonnichsen, C.; Geier, S.; Hecker, N. E.; von Plessen, G.; Feldmann, J.; Ditlbacher, H.; Lamprecht, B.; Krenn, J. R.; Aussenegg, F. R.; Chan, V. Z.-H.; Spatz, J. P.; Moller, M. Appl. Phys. Lett. 2000, 77, 2949. (411) Mock, J. J.; Barbic, M.; Smith, D. R.; Schultz, D. A.; Schultz, S. J. Chem. Phys. 2002, 116, 6755. (412) Novo, C.; Gomez, D.; Perez-Juste, J.; Zhang, Z.; Petrova, H.; Reismann, M.; Mulvaney, P.; Hartland, G. V. Phys. Chem. Chem. Phys. 2006, 8, 3540. (413) Kalyuzhny, G.; Vaskevich, A.; Ashkenasy, G.; Shanzer, A.; Rubinstein, I. J. Phys. Chem. B 2000, 104, 8238. (414) Cao, Q.; Lalanne, P. Phys. ReV. Lett. 2002, 88, 057403. (415) Lezec, H.; Thio, T. Opt. Express 2004, 12, 3629. (416) Sarrazin, M.; Vigneron, J.-P. Phys. ReV. B 2005, 71, 075404/1. (417) Ghaemi, H. F.; Thio, T.; Grupp, D. E.; Ebbesen, T. W.; Lezec, H. J. Phys. ReV. B 1998, 58, 6779. (418) Dintinger, J.; Degiron, A.; Ebbesen, T. W. MRS Bull. 2005, 30, 381. (419) Sarrazin, M.; Vigneron, J.-P.; Vigoureux, J.-M. Phys. ReV. B 2003, 67, 085415/1. (420) Genet, C.; van Exter, M. P.; Woerdman, J. P. Opt. Commun. 2003, 225, 331. (421) Benabbas, A.; Halte, V.; Bigot, J.-Y. Opt. Express 2005, 13, 8730. (422) Kim, T. J.; Thio, T.; Ebbesen, T. W.; Grupp, D. E.; Lezec, H. J. Opt. Lett. 1999, 24, 256. (423) Stark, P. R. H.; Halleck, A. E.; Larson, D. N. Methods 2005, 37, 37. (424) Muller, R.; Malyarchuk, V.; Lienau, C. Phys. ReV. B 2003, 68, 205415/1. (425) Huang, T. T.; Sturgis, J.; Gomez, R.; Geng, T.; Bashir, R.; Bhunia, A. K.; Robinson, P. J.; Ladisch, M. R. Biotechnol. Bioeng. 2003, 81, 618. (426) Brewer, S. H.; Glomm, W. R.; Johnson, M. C.; Knag, M. K.; Franzen, S. Langmuir 2005, 21, 9303. (427) Haussling, L.; Ringsdorf, H.; Schmitt, F.-J.; Knoll, W. Langmuir 1991, 7, 1837. (428) Kim, D. S.; Hohng, S. C.; Malyarchuk, V.; Yoon, Y. C.; Ahn, Y. H.; Yee, K. J.; Park, J. W.; Kim, J.; Park, Q. H.; Lienau, C. Phys. ReV. Lett. 2003, 91, 143901. (429) Snopok, B. A.; Kostyukevych, K. V.; Rengevych, O. V.; Shirshov, Y. M.; Venger, E. F.; Kolesnikova, I. N.; Lugovskoi, E. V. Semicond. Phys. Quantum Elect. Optoelect. 1998, 1, 121. (430) Donahue, J. P.; Patel, H.; Anderson, W. F.; Hawiger, J. PNAS 1994, 91, 12178. (431) Svedhem, S.; Pfeiffer, I.; Larsson, C.; Wingren, C.; Borrebaeck, C.; Hook, F. ChemBioChem 2003, 4, 339. (432) Reviakine, I.; Brisson, A. Langmuir 2000, 16, 1806. (433) Dieringer, J. A.; McFarland, A. D.; Shah, N. C.; Straurt, D. A.; Whitney, A. V.; Yonzon, C. R.; Young, M. A.; Zhang, X.; Van Duyne, R. P. Faraday Discuss. 2006, 132, 9. (434) Kneipp, K.; Kneipp, H. Appl. Spectrosc. 2006, 60, 322A. (435) Jeanmarie, D. L.; Van Duyne, R. P. J. Electroanal. Chem. 1977, 84, 1. (436) Albrecht, M. G.; Creighton, J. A. J. Am. Chem. Soc. 1977, 99, 5215. (437) Stockman, M. I. Top. Appl. Phys. 2006, 103, 47. (438) Schatz, G. C.; Young, M. A.; Van Duyne, R. P. Top. Appl. Phys. 2006, 103, 19. (439) Moskovits, M. Top. Appl. Phys. 2006, 103, 1. (440) Otto, A. J. Raman Spectrosc. 2006, 37, 937. (441) Anderson, D. J.; Moskovits, M. J. Phys. Chem. B 2006, 110, 13722. (442) Kuncicky, D. M.; Prevo, B. G.; Velev, O. D. J. Mater. Chem. 2006, 16, 1207. (443) Li, Z.; Tong, W. M.; Stickle, W. F.; Neiman, D. L.; Williams, R. S. Langmuir 2007, 23, 5135.

Nanostructured Plasmonic Sensors (444) Brolo, A. G.; Arctander, E.; Gordon, R.; Leathem, B.; Kavanagh, K. L. Nano Lett. 2004, 4, 2015. (445) Zhang, X.; Young, M. A.; Lyandres, O.; Van Duyne, R. P. J. Am. Chem. Soc. 2005, 127, 4484. (446) Lim, J. K.; Kim, I.; Kim, K.; Shin, K. S.; Kang, W.; Choo, J.; Joo, S. Chem. Phys. 2006, 330, 245. (447) Bell, S. E. J.; Sirimuthu, N. M. S. J. Am. Chem. Soc. 2006, 128, 15580. (448) Chowdhury, M. H.; Campbell, C. J.; Theofanidou, E.; Lee, S. J.; Baldwin, A.; Sing, G.; Yeh, A. T.; Crain, J.; Ghazal, P.; Cote´, G. L. Proc. SPIE 2006, 6099, 609905. (449) Tang, H.; Yang, X. B.; Kirkham, J.; Smith, D. A. Anal. Chem. 2007, 79, 3646. (450) Zhang, J.; Gao, Y.; Alvarez-puebla, R. A.; Buriak, J. M.; Fenniri, H. AdV. Mater. 2006, 18, 3233. (451) Orendorff, C. J.; Gearheart, L.; Jana, N. R.; Murphy, C. J. Phys. Chem. Chem. Phys. 2006, 8, 165. (452) Wang, T.; Hu, X.; Dong, S. J. Phys. Chem. B 2006, 110, 16930. (453) Dick, L. A.; McFarland, A. D.; Haynes, C. L.; Van Duyne, R. P. J. Phys. Chem. B 2002, 106, 853. (454) Ruan, C.; Eres, G.; Wang, W.; Zhang, Z.; Gu, B. Langmuir 2007, 23, 5757. (455) Shanmukh, S.; Jones, L.; Driskell, J.; Zhao, Y.; Dluhy, R.; Tripp, R. A. Nano Lett. 2006, 6, 2630. (456) White, D. J.; Mazzolini, A. P.; Stoddart, P. R. J. Raman Spectrosc. 2007, 38, 377. (457) Willis, R. C. Anal. Chem. 2007, 79, 1785. (458) Cornish, P. V.; Ha, T. ACS Chem. Biol. 2007, 2, 53. (459) Delfino, I.; Bizzarri, A. R.; Cannistrato, S. Chem. Phys. 2006, 326, 356. (460) Kim, J.; Song, K.-B. Micron 2007, 38, 409. (461) Frey, H. G.; Bolwein, C.; Brandenburg, A.; Ros, R.; Anselmetti, D. Nanotechnology 2006, 17, 3105. (462) Pettinger, B. Top. Appl. Phys. 2006, 103, 217. (463) Zhang, W.; Cui, X.; Yeo, B.-S.; Schmid, T.; Hafner, C.; Zenobi, R. Nano Lett. 2007, 7, 1401. (464) Becker, M.; Sivakov, V.; Andra¨, G.; Geiger, R.; Schreiber, J.; Hoffman, S.; Michler, J.; Milenin, A. P.; Werner, P.; Christiansen, S. H. Nano Lett. 2007, 7, 75. (465) Matsui, R.; Verma, P.; Ichimura, T.; Inouye, Y.; Kawata, S. Appl. Phys. Lett. 2007, 90, 061906. (466) Huebsch, N. D.; Mooney, D. J. Biomaterials 2007, 28, 2424. (467) Wei, D.; Wang, Y.; Luo, Q.; Bi-Feng, L. Anal. Bioanal. Chem. 2006, 386, 444. (468) Brolo, A. G.; Kwok, S. C.; Moffitt, M. G.; Gordon, R.; Riordon, J.; Kavanagh, K. L. J. Am. Chem. Soc. 2005, 127, 14936.

Chemical Reviews, 2008, Vol. 108, No. 2 521 (469) Hseih, B.; Chang, Y.-F.; Ng, M.-Y.; Liu, W.-C.; Lin, C.-H.; Wu, H.-T.; Chou, C. Anal. Chem. 2007, 79, 3487. (470) Lacowicz, J. R. Plasmonics 2006, 1, 5. (471) Lakowicz, J. R.; Geddes, C. D.; Gryczynski, I.; Malicka, J.; Gryczynski, Z.; Aslan, K.; Lukomska, J.; Matveeva, E.; Zhang, J.; Badugu, R.; Huang, J. J. Fluoresc. 2004, 14, 425441. (472) Garrett, S. H.; Smith, L. H.; Barnes, W. L. J. Mod. Opt. 2005, 52, 1105. (473) Huang, C. C.; Chang, H. T. Anal. Chem. 2006, 78, 8332. (474) Nagamura, T.; Yamamoto, M.; Terasawa, M.; Shiratori, K. Appl. Phys. Lett. 2003, 83, 803. (475) He, X.; Liu, H.; Li, Y.; Wang, S.; Li, Y.; Wang, N.; Xiao, J.; Xu, X.; Zhu, D. AdV. Mater. 2005, 17, 2811. (476) Brolo, A. G.; Kwok, S. C.; Cooper, M. D.; Moffitt, M. G.; Wang, C.-W.; Gordon, R.; Riordon, J.; Kavanagh, K. L. J. Phys. Chem. B 2006, 110, 8307. (477) Shelby, R. A.; Smith, D. R.; Schultz, S. Science 2001, 292, 77. (478) Lezec, H. J.; Dionne, J. A.; Atwater, H. A. Science 2007, 316, 430. (479) Soukoulis, C. M.; Kafesaki, M.; Economou, E. N. AdV. Mater. 2006, 18, 1941. (480) Shalaev, V. M. Nat. Photonics 2007, 1, 41. (481) Pendry, J. B. Phys. ReV. Lett. 2000, 85, 3966. (482) Hohng, S. C.; C., Y. Y.; Kim, D. S.; Malyarchuk, V.; Muller, R.; Lienau, C. H.; Park, J. W.; Yoo, K. H.; Kim, J.; Ryu, H. Y.; Park, Q. H. Appl. Phys. Lett. 2002, 81, 3239. (483) Luo, X.; Ishihara, T. Appl. Phys. Lett. 2004, 84, 4780. (484) Dragnea, B.; Szarko, J. M.; Kowarik, S.; Weimann, T.; Feldmann, J.; Leone, S. R. Nano Lett. 2003, 3, 3. (485) Srituravanich, W.; Fang, N.; Sun, C.; Luo, Q.; Zhang, X. Nano Lett. 2004, 4, 1085. (486) Liu, Z.-W.; Wei, Q.-H.; Zhang, X. Nano Lett. 2005, 5, 957. (487) Smolyaninov, I. I.; Elliot, J.; Zayats, A. V.; Davis, C. C. Phys. ReV. Lett. 2005, 94, 057401. (488) Garini, Y.; Kutchoukov, V. G.; Bossche, A.; Alkemade, P. F. A.; Docter, M.; Verbeek, P. W.; van Vliet, L. J.; Young, I. T. Proc. SPIE 2004, 5327, 115. (489) Docter, M. W.; Young, I. T.; Kutchoukov, V. G.; Bossche, A.; Alkemade, P. F. A.; Garini, Y. Proc. SPIE 2005, 5703, 118. (490) Smolyaninov, I. I.; Hung, Y.-J.; Davis, C. C. Science 2007, 315, 1699. (491) Liu, Z.-w.; Lee, H.; Xiong, Y.; Sun, C.; Zhang, X. Science 2007, 315, 1686.

CR068126N

Chem. Rev. 2008, 108, 522−542

522

Microcantilevers: Sensing Chemical Interactions via Mechanical Motion Karen M. Goeders,† Jonathan S. Colton,‡ and Lawrence A. Bottomley*,† School of Chemistry & Biochemistry and George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, Georgia 30332 Received September 21, 2007

Contents 1. Introduction 2. Theory of Operation 2.1. Static Deflection 2.2. Dynamic Response 2.2.1. Mass−Spring−Dashpot System. 2.2.2. Quality Factor 2.2.3. Plane Strain 2.2.4. Bending Mode Frequency Response 2.2.5. Effect of Damping Due to Viscous Fluids 2.2.6. Effect of Air Damping on Quality Factor 2.3. Lateral and Torsional Mode Frequency Responses 2.4. Temperature Effects 2.4.1. Effect on Material Properties 2.4.2. Effect on Geometry 3. Detection Schemes 3.1. Optical Lever 3.2. Interferometer 3.3. Piezoresistive 3.4. Capacitive 4. Design, Materials, and Fabrication 4.1. Design Considerations 4.2. Fabrication of Silicon-based Cantilevers 4.2.1. Film Deposition 4.2.2. Photolithography 4.2.3. Etching 4.2.4. Doping 4.3. Fabrication of Polymeric Cantilevers 5. Chemical Selectivity 6. Chemical Applications 6.1. Volatile Organics 6.2. Chemical Warfare Agents 6.3. Explosives 6.4. Toxic Metal Ions 7. Biological Applications 7.1. Cells 7.2. Viruses 7.3. Antigen−Antibody Interactions 7.4. DNA Hybridization 7.5. Enzymes 8. Recommendations for Future Work * To whom correspondence should be addressed. [email protected]. † School of Chemistry & Biochemistry. ‡ George W. Woodruff School of Mechanical Engineering.

522 522 523 525 525 525 525 526 526 527 527 527 528 528 528 528 529 529 529 529 530 530 530 530 530 530 531 531 533 533 534 534 534 534 534 534 535 536 537 538 E-mail:

8.1. Guidelines for Reporting Sensor Performance 8.2. Experimental Design Considerations 8.3. Fruitful Areas for Further Research 8.3.1. More Selective Coatings 8.3.2. Increased Sensitivity and Faster Response 9. References

538 538 539 539 539 539

1. Introduction Micromechanical devices comprise emerging sensor platforms with straightforward sensing mechanisms. Molecular adsorption onto the sensing element, typically a cantilever, shifts its resonance frequency and changes its surface forces (surface stress). Adsorption onto the sensing element composed of two chemically different surfaces produces a differential stress between the two surfaces and induces bending. The analyte that induces the mechanical response may be physi- or chemisorbed onto the cantilever in a reversible or irreversible process. Devices that respond to chemical stimuli in this manner are more commonly referred to as microcantilever sensors.1-6 A compelling feature of microcantilever sensors is that they can be operated in air, vacuum, or liquid. The rapid growth in microcantilever-based sensor technology parallels advancements in micromachining methodologies and is in response to the need for more sensitive and selective detection of airborne and waterborne toxic and pathogenic substances. The purpose of this review is to critically examine the current state of theory, modes of detection, design considerations, and innovative applications of this sensing platform. Each will be addressed separately in the following sections. At the conclusion of this review, we will identify areas that warrant further investigation and suggest guidelines for reporting the performance of microcantilever sensors to facilitate comparison of microcantilevers with other sensing platforms.

2. Theory of Operation Microcantilever sensors rely on their deflection to indicate sensing. This section describes the theory for the mechanical response of microcantilevers in the bending, lateral, and torsional modes when used as sensors. This discussion is divided into the two modes of microcantilever deflection, static and dynamic, that are used in sensing applications. The means for detecting deflection are discussed in a separate section. The static mode of deflection occurs when an adsorbed species causes differential surface stresses on the oppo-

10.1021/cr0681041 CCC: $71.00 © 2008 American Chemical Society Published on Web 01/30/2008

Sensing Chemical Interactions via Mechanical Motion

Karen Meloy Goeders is pursuing her doctorate in analytical chemistry from the Georgia Institute of Technology. She came to the School of Chemistry & Biochemistry after completing her B.S. degree in chemistry from Louisiana State University. Her dissertation research focuses on developing microcantilever array technology for enzymatic assays.

Chemical Reviews, 2008, Vol. 108, No. 2 523

Dr. Lawrence A. Bottomley is a professor of chemistry at the Georgia Institute of Technology. He obtained his B.S. in chemistry at California State University, Fullerton, and his Ph.D. in analytical chemistry at the University of Houston. His current research interests include the biological and nanotechnological applications of scanning probe microscopy, electroanalytical chemistry, and microcantilever array sensing.

Figure 1. Microcantilever geometry and nomenclature.

Dr. Jonathan Colton is a professor of mechanical engineering at the Georgia Institute of Technology. He obtained his S.B., S.M., and Ph.D. in Mechanical Engineering at the Massachusetts Institute of Technology. He serves as the director of the Center for Polymer Processing. He is a fellow of the American Society of Mechanical Engineers and of the Society of Plastics Engineers and is a registered professional engineer in the state of Georgia. His research interests include polymer and polymer composites processing, bioMEMS sensors, biomedical devices, and dielectric materials.

site surfaces of the microcantilever. The equations that describe the static deflection of microcantilevers will be presented, and the response of the deflection to surface stress will be discussed. In the dynamic mode of detection, the frequency of vibration of the beam changes as species are adsorbed onto the microcantilever. The equations describing the vibration of the beam in air will be presented. The damping effects of measurement in a viscous gas and liquid will be described next. Thermal effects also will be discussed. A microcantilever can be modeled as a cantilever beam (thickness, t; width, w; and length, L), which is built in (fixed) at one of its ends (see Figure 1).7 Note that, in Figure 1, z denotes the deflection in the thickness direction along the beam length and time [i.e., z(x,T)] and does not indicate the origin of the coordinate system. The discussion first will be limited to pure bending of a beam; lateral and torsional (twisting) motions will be discussed at the end of this section. Figure 2 shows the bending (a), lateral (b), and torsional (c) deflections of a built-in beam.7

Figure 2. Schematic of the first bending (a), lateral (b), and torsional (c) modes of a resonating cantilever. The heavy lines denote the undeformed cantilever; the shaded regions denote the deformed cantilever.

2.1. Static Deflection Static deflection is used to determine the amount of material adsorbed onto a microcantilever. The more material that is adsorbed, the more the microcantilever will deflect.8-10 Deflection results from two mechanisms: added mass and surface stress from adsorbed species.11-13 However, the surface stress may not necessarily correlate with the amount

524 Chemical Reviews, 2008, Vol. 108, No. 2

Goeders et al.

Figure 3. End-loaded beam. Figure 5. Fixed beam subjected to surface stresses.

t is the thickness of the coating, and d is the thickness of the beam. The radius of curvature, R, of a microcantilever under the influence of surface stresses on its top and bottom surfaces is typically reported as eq 518 Figure 4. Beam loaded by a uniform load (wo/unit length).

of material adsorbed. The deflection of the free end of the beam depends on the type of loading to which the beam is subjected. If a concentrated load, F, is applied to the free end of a rectangular beam (see Figure 3), then the deflection of the free end of the beam, δ, is given by eq 114

δ)

FL3 3EI

(1)

where E is the Young’s elastic modulus of the beam material, L is the length of beam, and I is the second moment of the beam’s cross-sectional area. I is a function of the beam width, w, and thickness, t, and is equal to wt3/12. The resistance to bending deformation (i.e., stiffness) in bending of a rectangular beam (k) is given by eq 2.15

k)

3EI L3

(2)

An example of this type of deflection is a tipped microcantilever used in atomic force microscopy to measure surface geometry. Equation 2 is commonly used to calculate the force constant of imaging probes. However, since the dynamic etches used to create these probes modify the shape and dimensions of both the tip and the beam, use of this equation to compute k results in significant error. Poggi et al.16 have presented an improved method for determining beam stiffness that takes into account the actual geometry of the cantilever. If the beam is uniformly loaded along its length by a load per unit length, wo, (see Figure 4), the deflection of its free end is given by eq 3.14

δ)

woL4 8EI

(3)

An example of this type of deflection is when a species is adsorbed uniformly to a cantilever’s surfaces. For example, if one assumes that the species of interest absorbs only on one surface of a microcantilever, a surface stress results on that side. The difference in the stresses on the top and bottom surfaces of the cantilever generates a deflection that is independent of that due to the adsorbed mass. Stoney’s equation17 has been used to relate the difference in surface stresses on each surface of a beam to its deflection (see Figure 5)

Pt 1 )6 2 r Ed

(4)

where r is the radius of curvature of the beam, P is the surface stress due to a coating on one surface of the beam,

1-ν 1 ) 6 2 (∆σ1 - ∆σ2) R Et

(5)

where V is the Poisson’s ratio of the material; it is included to reflect the plane strain condition of the microcantilever (see discussion of the Searle parameter below). ∆σ1 and ∆σ2 are the stresses that act on the top and bottom surfaces of the beam, and t is the thickness of the beam. The deflection of such a beam can be calculated using the geometric relation R-1 ) 2∆z/L2, where ∆z is the beam’s end deflection. Only a difference in absorption between the top and bottom surfaces will cause deflection; equal absorption to both top and bottom surfaces will counteract each other, resulting in no deflection. One will note that the units on the two sides of eq 5 do not match (left-hand side (LHS) ) distance-1 vs right-hand side (RHS) ) distance-2); this is due to the fact that the surface stresses in Stoney’s equation are reported on the basis of a per unit thickness of the layer that causes the deflection in the beam. As is convention in the literature, the term stress is used and its units are N/m. In practice, j ftf, where σ j f is the average normal stress acting on ∆σs ) σ the cross-sectional area residing in a plane that is normal to the neutral axis of the beam-coating composite and tf is the coating thickness. Hence, ∆σs is visualized as the normal force per unit width acting on a normal section of the coating. Stoney’s equation (eqs 4 and 5) is an exact solution for plate bending that is unrestrained at all edges and assumes no interaction between adsorbed species. Hence, a more accurate equation is necessary. For a cantilever with length L (x), width w (2y), and thickness t (z) (see Figure 1), Sader19 has developed a more complete solution for the deflection of a point on a fixed cantilever beam under the influence of a surface stress, wcant(X, Y), as shown in eq 620

{

wcant(X, Y) ) ΩL2 X2 + 2νX[τ1-1 + τ2-2]

() w

-

L

[12-1 + 2υ(τ1-2 + τ2-2 + τ1-1τ2-1) 2 w2 ∑di(12-1 + 2ντi -2) × exp(-τiXLw-1)] L + i)1

()

2

Y2[1 - ∑di exp(-τiXLw-1)] i)1

}

(6)

where X ) xL-1, Y ) yL-1, di ) τ3-i(τ3-i - τi)-1, Ω ) ∆σst/ [4D(1 + υ)], where ∆σs is the surface stress and D ) Et3{12(1 - υ2)}-1 is the cantilever bending rigidity. The τi are defined by eq 7.

τi ) 2x3[5(1 - ν + (-1)ix10(1 - ν)(2 - 3ν))] (7)

Sensing Chemical Interactions via Mechanical Motion

Chemical Reviews, 2008, Vol. 108, No. 2 525

Figure 6. Mass-spring-dashpot system.

For the case where the deflection at the end of the tip is desired, x ) L and y ) w/2. Klein provides an expression for the error in using Stoney’s equation for a multilayer laminate.21 Stoney’s and Sader’s equations can be used to relate the deflections of microcantilevers to the surface stresses resulting from adsorbed species. Typically, the surface stress is calculated following measurement of deflection. It is difficult to determine a priori the surface stresses due to molecular attachment. In the following section on dynamic methods, we will develop expressions for the effect of surface stress on the dynamic response of a microcantilever.

2.2. Dynamic Response In this section, the dynamic response of a simple massspring-dashpot system first is described to introduce the reader to the appropriate concepts. Then, the quality factor, a well-used metric of microcantilever performance, is discussed. Next, the vibration of a fixed microcantilever in air and shifts in resonant frequency resulting from added mass layers and changes in surface stress are presented. The damping effects of a liquid on vibrational frequency then are presented, which correlates to the submersion of a microcantilever in viscous gas and liquid media. Finally, the effect of temperature is discussed.

2.2.1. Mass−Spring−Dashpot System A mass-spring-dashpot system, such as that in Figure 6, has the equation of motion shown in eq 821

m

d2x dx + c + kx ) F(t) 2 dt dt

(8)

where m is the mass, c is the dashpot constant, k is the spring constant, and x is the displacement of the mass. F(T) represents the generalized force that is exciting the system. The natural frequency of vibration (ωnat) (i.e., with no excitation force, F(t) ) 0) of the undamped beam (i.e., with no dashpot, c ) 0) is given by eq 9.

ωnat ) xk/m

(9)

When damping is present (i.e., one includes the effect of the dashpot), one can use the natural frequency to rewrite the equation of motion with no excitation force (F(t) ) 0) as eq 10. 2

dx dx + 2ζωnat + ωnat2x ) 0 2 dt dt

(10)

The parameter ζ is known as the system’s damping and is defined for this system in eq 11.

ζ)

c c ≡ 2ωnatm 2xkm

(11)

Figure 7. Generic frequency response curves.

Similarly, a freely vibrating microcantilever beam (i.e., due to thermal excitation) will resonate at its natural frequency (eq 9). Thus, as material absorbs onto the beam forming a coating, the microcantilever’s vibrational frequency will decrease and, depending upon the thickness of the adsorbed layer, its spring constant may change.

2.2.2. Quality Factor The quality factor of a microcantilever characterizes the shape of its frequency response curve (e.g., a plot of the displacement amplitude versus frequency) near a resonance mode.22 Accordingly, each resonance mode has its own quality factor. Mathematically, the ith mode quality factor, Qi, is defined as the ratio of the resonance frequency of the ith mode, fi, to the full width of the resonance peak evaluated at the half-maximum (FWHM ) full width half-maximum) of the peak. The quality factor indicates the narrowness of a resonant peak. Figure 7 shows generic frequency response curves and their quality factors and fi values.7 The definition and value of the quality factor for a lightly damped onedegree-of-freedom system, such as an AFM microcantilever, is given by eq 1221

Qi )

fi ωi 1 ) ) ∆ω 2ζ FWHM

(12)

where ωi ) 2πfi and ∆ω ) 2πFWMH. The quality factor depends on the cantilever geometry and the fluid in which the cantilever is immersed.23-30 Increased damping effects lead to a lower Q value. A higher Q value is desired because it lowers the minimum detectable resonance shift (i.e., it increases the frequency resolution). For a quality factor of 10, the minimum detectable resonance frequency shift is roughly 25 Hz, whereas a quality factor of 100 allows for a frequency resolution below 10 Hz.

2.2.3. Plane Strain The discussions in this paper assume that microcantilevers are in a plane strain situation.18,31 The Searle parameter32 is defined as βb ) w2κb/t in a bending mode (Figure 2a) and βl ) t2κl/w in a lateral mode (Figure 2b), where κb and κl are the principal curvatures in the bending and lateral modes, respectively, and dictate the deformational situation a beam is undergoing. A Searle parameter value < 1 indicates a plane stress situation, whereas a Searle parameter value > 100 indicates a plane strain situation.33 From the Euler-Bernoulli beam theory,34 the expression for the maximum curvature

526 Chemical Reviews, 2008, Vol. 108, No. 2

Goeders et al.

of a beam subjected to a transverse deflection at its end (causing a bending or lateral deformation) of δ is 3δ/L2. This implies that the maximum Searle parameter value for a bending mode is βbmax ) 3δw2/L2t and for a lateral mode is βlmax ) 3δt2/L2w. Assuming an end deflection at thermal resonance of 100 nm for both the bending and lateral modes (the lateral mode will have a smaller deflection)35 and using the L, w, and t values of 500, 100, and 0.8 µm, typical of silicon microcantilevers,7,18,31 the maximum Searle parameters for the bending and lateral modes are βbmax ) 0.015 and βlmax ) 7.7 × 10-9. These are ,1, indicating a plane stress situation; hence, equations derived in this article are valid for plane strain situations.

2.2.4. Bending Mode Frequency Response An unloaded beam freely vibrating in a bending mode (see Figure 2a) in a vacuum will have a number of resonance frequenciessfrequencies at which it will naturally vibrate under thermally induced excitation. The following discussion is taken from McFarland and co-workers.31 The general expression for the ith mode resonance of the beam, fi, is given by eq 13

fi )

( )x

1 Ri 2π L

2

EI Fbwt

(13)

Equation 19 shows the effects of the mass (+m) and the thickness (+k)

fi+m,+k )

( )x

t Ri 4π L

2

Iads )

EI 3Fb

(15)

Ri2 2πx3

x

k Mb

(16)

where Mb is the mass of the microcantilever beam. If a layer of molecules or other species is added to (coats) a beam, the mass of the composite beam will increase by ∆M; hence, the new resonance frequencies (+m) due to the mass increase will be given by eq 17. The following discussion is taken from McFarland20 and can be used to determine the sensitivity of a microcantilever’s response to added mass.

fi+m )

Ri2 2πx3

x

k Mb + ∆M

(17)

Such an added layer also will increase the second moment of the microcantilever’s cross-sectional area, I, making the beam stiffer. Equation 18 shows this effect (+k).

fi+k )

Ri2 2πx3

x

3EadsIads k + 3 Mb LM b

(18)

b

[( )

]

wtads3 tads + wtads - tb - ycm 12 2

2

(20)

The centroid of the cross section including the adsorbed layer is given by eq 21

Ebtb2 + Eads(2tadstb + tads2) ycm ) 2Eadstads + 2Ebtb

(21)

where tb is the beam thickness and tads is the thickness of the adsorbed layer. The effect of surface stress (+∆σ) effect is given by eqs 22 and 23 +∆σ

)

fi

(Ri∆σ)2 2πx3

[

Ri∆σ ) Ri 1 +

x

k Mb

]

2σL3 π2EI3

(22)

1/4

(23)

where σ is the surface stress. Equations 22 and 23 can be rearranged to allow calculation of surface stress based on the resonant frequency for a microcantilever of rectangular cross section (L, w, and t) as shown in eq 24.7,31

[( ) ]

fi+∆σ 2 π2Ewt3 -1 fi 24L3

σ)

Equation 13 can be modified further by inserting eq 2, resulting in eq 16

fi )

2πx3

(14)

The solutions to eq 14 can be found in the report by Han and co-workers.36 For a rectangular beam with I ) wt3/12, eq 15 can be used to determine fi

fi )

3EadsIads k + 3 (19) Mb + ∆M L (M + ∆M)

where the adsorbed layer (denoted by subscript “ads”) has elastic modulus Eads and second moment Iads are given by eq 20 via the parallel axis theorem.

where Fb is the cantilever material density and Ri is obtained numerically from the frequency relation (eq 14).

cosh Ri cos Ri + 1 ) 0

x

Ri2

(24)

One can combine eqs 19, 22, and 23 to arrive at the most general case of a change in frequency due to adsorbed mass (+m), to increased stiffness due to a change in thickness (+k), and to surface stress (+∆σ); see eq 25.18,20,37

fi+m,+k,+∆σ )

x

(Ri∆σ)2 2πx3

3EadsIads k + 3 (25) Mb + ∆M L (M + ∆M) b

2.2.5. Effect of Damping Due to Viscous Fluids For most cases, air is assumed not to affect the operation of the cantilever; hence, the equations derived above are valid for microcantilevers operating in air. If the microcantilever is used in a liquid or gas that does influence its operation, then damping effects will influence the response of the microcantilever.38 The effect of damping can be estimated by eq 26, which relates f vi , the ith mode frequency of a beam with density Fb oscillating in a vacuum, to f Di , the frequency when oscillating in, and hence damped by, a fluid of density F for a Reynolds number . 1.39 This is an inertiaresistance dominated situation, so the resistance is roughly proportional to the acceleration of the cantilever.40

Sensing Chemical Interactions via Mechanical Motion

f Di f vi

(

) 1+

)

πFw 4Fbt

Chemical Reviews, 2008, Vol. 108, No. 2 527

-1/2

(26)

It should be noted that, while the damping effects in air have only a minute effect on the resonance frequency, the effect on the quality factor can be quite dramatic, with Q jumping from the order of 10-100 in air to the order of 1 000-10 000 in vacuum. For flows where the Reynolds number is ,1 (i.e., damping mainly due to the viscosity of the fluid surrounding the beam), an expression (eq 27) analogous to eq 26 can be derived23,39

f Di f

v i

)

x

1-

1 4Qi2

(27)

where Qi is the ith frequency quality factor for the damped microcantilever (i.e., submerged in a viscous gas or fluid).

2.2.6. Effect of Air Damping on Quality Factor The effect of air damping on the quality factor of microcantilevers is discussed in more detail by Newell.41 He presents a number of expressions for the quality factor for a microcantilever damped by air. The first case is where the pressure is so low that air damping is negligible. Here, the quality factor is independent of pressure and must be determined empirically. The second case is where air damping is the dominant mechanism but the air molecules are so far apart that they do not interact with each other. In this case, the quality factor is given by eq 28

[ ] ( )

R0T M0 π 3/2 Ftfi Qi ) 2 1 P

()

1/2

(28)

where F is the density of the microcantilever, t is the thickness of the microcantilever, fi is the resonant frequency of the microcantilever, R0 is the universal gas constant, M0 is the molar mass of the air, T is the temperature, and P is the air pressure. The final case is where the air molecules do interact with each other, and here one assumes that the air acts as a viscous fluid. Since viscosity will be independent of pressure, the quality factor also is independent of pressure. If one uses Stokes’ law for damping, eq 29 results

[

]( )

w(EF)1/2 t Qi ) 24 µ L

2

(29)

where w is the width of the microcantilever, t is its thickness, L is its length, and µ is the viscosity of air. Thus, the value of the quality factor is strongly influenced by the media that surrounds the microcantilever.

2.3. Lateral and Torsional Mode Frequency Responses Until now, only pure bending of a microcantilever beam has been discussed. Lateral (Figure 2b) and torsional (Figure 2c) motions can be modeled in similar manners to those presented above. The results presented here are for undamped microcantilevers. The frequency for a freely vibrating beam deflecting in the jth lateral mode is described by eqs 30 and 317

f lj )

( )x

l w Rj 4π L

2

EI 3Fb

(30)

cosh Rlj cos Rlj + 1 ) 0

(31)

where the superscript l signifies the lateral mode of deformation. In a similar manner, the sth torsional resonant modes for a freely vibrating beam can be modeled as eq 327

f ts )

x

2s - 1 4L

Gξ FbIp

(32)

where the superscript t denotes the torsional mode of deformation, G is the shear modulus (G ) E/(2(1 + υ)), Ip is the polar moment of the cross section defined by eq 3342 for a rectangular cross section beam, and ξ is defined by eq 34. For the approximate solution of eq 34, the reader is referred to the paper by McFarland et al.7

Ip )

(

1 (tw3 + wt3) 12

(33)

)

nπw 1 w 192 ∞ 1 tanh ξ ) t4 ∑ 3 t 2t π5 n)1n5

(34)

These equations for lateral and torsional modes of deflection can be used to derive situation-specific resonant frequencies in an analogous manner for the bending mode equations shown above, e.g., for added mass, added thickness, and fluid viscosity effects.

2.4. Temperature Effects Thermomechanical noise (vibration due to thermal agitation) is a consequence of a microcantilever being in thermal equilibrium with its environment. This discussion, taken from Newell41 and Yasumura et al.,28 is for an undamped microcantilever; one utilizes the material discussed above to include damping. Energy dissipation in a microcantilever causes the stored mechanical energy to be converted into heat. The interaction of a microcantilever with the many microscopic degrees of freedom in its environment will subject the microcantilever to constant random excitation. The relationship between energy dissipation and random thermal excitation is embodied in the “fluctuation-dissipation theorem” of statistical mechanics. The net result is that, the lower the mechanical Q of the system is, the larger is the noise force. The mean square vibration amplitude associated with a mode of oscillation at temperature T can be determined from the equipartition theorem as shown in eq 35

1 1 k T ) k〈z2〉 2 B 2

(35)

where kB is Boltzmann’s constant, k is the cantilever stiffness, and z is the microcantilever’s deflection. If one assumes that the noise spectrum is white (i.e., frequency independent), then the spectral density SF ) 4kkbT/ω0Q and the force noise (F) in a bandwidth (B) is given by eq 36

Fmin )

x

4kkBTB ω0Q

(36)

528 Chemical Reviews, 2008, Vol. 108, No. 2

Goeders et al.

where ω0 is the cantilever resonance frequency and is equal to 2πf0. For a simple rectangular cantilever, the minimum detectable force can be expressed by eq 37

Fmin )

( ) wt2 LQ

1/2

(kBTB)1/2(EF)1/2

(37)

where w is the microcantilever’s width, t is its thickness, L is its length, E is its Young’s modulus, and F is its density. Similarly, the mean square root deflection is given by eq 38.

zrms )

() ( )() kT kB

1/2

)

2kT wE

1/2

L3 t3

1/2

(38)

Equations 37 and 38 can be used to design the sensitivity of microcantilevers, but one can see that very high quality factors are necessary for ultrasensitive devices.

2.4.1. Effect on Material Properties Many of the material properties of microcantilevers depend on temperature. For example, as the temperature increases, the elastic modulus decreases. The temperature dependence of the elastic modulus of silicon in the high-temperature limit has been modeled semiempirically by eq 3943

( )

E(T) ) E0 - BT exp -

T0 T

(39)

where E0 is the Young’s modulus at 0 K. The constants B > 0 and To > 0 are temperature independent. For aluminum oxide, E0 is reported as approximately 4.6 × 1012 dyn/cm2, B as 4.41 × 108 dyn/cm2, and T0 as 373 K.43 The frequency shift of silicon microcantilevers and variations in the Q-factor over a range of temperatures has been studied.44

2.4.2. Effect on Geometry Temperature also affects the geometry of a microcantilever, with an increase in temperature generally being related to an increase in dimensions through a parameter termed the coefficient of thermal expansion (CTE). For example, the thermal expansion of silicon is on the order of 3.2 parts per million (ppm) per °C and those of a polymer are on the order of 50-100 ppm/°C. As one can see from the discussion in this section, it is critical that calibration and operation of the microcantilever be performed at the same temperature and that the temperature is controlled within very tight tolerances. In the absence of temperature control, differential measurements utilizing pairs of coated and uncoated microcantilevers must be performed.

3. Detection Schemes Vertical, lateral, or torsional movement of a cantilever changes its position. This movement ranges from several angstroms to a micrometer or more, depending upon the dimensionality of the cantilever and the magnitude of surface stress. In this section, methods for measurement of cantilever deflection are examined.

3.1. Optical Lever The optical lever method, illustrated schematically in Figure 8, is the most widely utilized method of quantifying

Figure 8. Schematic of the experimental setup with liquid cell, optical readout of cantilever deflections, and sample liquid exchange system: VCSEL ) vertical cavity surface emitting lasers, PSD ) position sensitive detector. Reproduced with permission from Arntz, Y. et al. Nanotechnology 2003, 14, 86. Copyright 2003 Institute of Physics Publishing.

static and dynamic cantilever deflections. The method involves reflection of a beam of light off the cantilever onto a segmented photodiode or a position-sensitive detector (PSD). Light emitting diodes (LED) and laser diodes are the sources typically used to generate the beam of light. Photodiodes, divided into two or four segments, transduce the light energy striking each segment into an electrical signal that can be compared, amplified, and displayed. Motion of the cantilever changes the position of the reflected light beam on the photodiode and, consequently, the level of light energy incident on each segment. Quad-type photodiodes can, in principle, measure all modes of deflection (bending, lateral motion, and twisting) simultaneously. Typically, the reflected beam is centered on photodiode so that each segment has the same level of illumination at the beginning of each experiment. Then, as the cantilever bends, the laser spot changes location on the photodiode array. By comparing the outputs of the segments, the location of the centroid of the reflected laser spot, and, hence, the deflection of the microcantilever, can be determined. Segmented photodiodes are employed extensively in atomic force microscopes. PSDs are monolithic PIN (positive intrinsic negative) photodiodes with uniform resistance in one or two dimensions. Incident light on the photosensitive region of the PSD generates two photocurrents, each inversely proportional to the distance of the spot from the end of the region. The difference in photocurrents is converted to a voltage, amplified, converted, and displayed. PSDs possess high position resolution and fast response speed and require simple operating circuits. Establishing the relationship between output signal of either segmented photodiodes or PSDs with the magnitude of deflection requires careful calibration. Optical lever detection is currently the most sensitive method for measuring deflection; vertical deflections as small as a few angstroms can be reliably measured with this technique. An intrinsic limitation of this technique is that the laser diode, positioning system, and detector must be external to the air or fluid stream passing by the cantilever. Their dimensions are large in comparison to the microcantilever. Also, this technique is ineffective when the sample passing over the cantilever absorbs or scatters light, e.g., smoky air streams,45 and fluids with suspended particles.46 The optical lever technique is well-suited for detection of cantilever arrays. A number of formats have been published; two have been commercialized. One approach is to have multiple beam sources and detectors, one pair for each cantilever in the array. While this approach enables simultaneous measurement of all cantilevers in the array, the

Sensing Chemical Interactions via Mechanical Motion

integrated source, detector, and signal-processing system is quite complex to design and expensive to manufacture. Sequential reflection of a light beam off each cantilever in the array onto a single detector significantly reduces the complexity of the system and dramatically lowers its cost. One way to achieve this is to scan a single laser source across all of the beams in a cantilever array.47,48 Another way is to sequentially illuminate each element in an array of LEDs or vertical cavity surface emitting lasers.11,49-56 A third approach is to illuminate all cantilevers in the array with a single collimated beam, reflecting the light onto the image plane of a charge-coupled device camera.57,58 Deflection of each cantilever in the array is computed from changes in reflection spot location in images acquired over time.

3.2. Interferometer Interferometric detection of cantilever deflection is based on constructive and destructive interferences that occur when a collimated beam of light reflects off two surfaces displaced from one another.59 In the majority of applications of this technique, cantilevers containing a deformable diffraction grating consisting of a reference and movable set of interdigitated fingers were used. These can be intrinsic to a given cantilever or between cantilever pairs. Chemisorption onto the movable set displaces them relative to the reference fingers and alters the intensity of the diffracted orders is altered. The order intensity is measured with a photodiode array. This technique is capable of measuring very small deflections (as small as 0.01 Å)60 but has a very limited dynamic range. As with the optical lever technique, the interferometric detection technique is ineffective when the sample stream absorbs or scatters the incident or reflected beams. Interferometric detection is being used for hightemperature vibration sensors,61 while Gimzewski and coworkers62 used strobed interferometric microscopy to study the different resonance modes of cantilevers in arrays.

3.3. Piezoresistive The electrical conductivity of a piezoresistive material changes when stress is applied to it. Thus, when a piezoresistive element is integrated onto the cantilever during fabrication, cantilever bending is proportional to the change in resistance. The change in resistance is measured with a Wheatstone bridge, often located at the base of the cantilever.63,64 Piezoresistive elements fabricated onto or into cantilevers comprise either semiconductor or metallic strain gauges. A semiconductor strain gauge is smaller and lower in cost than a metallic foil resistance sensor described below. While the higher unit resistance and sensitivity of semiconductor sensors are definite advantages, their greater sensitivity to temperature variations and tendency to drift are disadvantages in comparison to metallic foil sensors. Another disadvantage of semiconductor strain gauges is that the resistance-to-strain relationship is nonlinear, varying 10-20% from a straightline equation, although this limitation can be overcome through software compensation. Metallic foil strain gauges measure the change in resistance of a metal as it is stretched. By appropriate calibration, the relation between the strain and the change in resistance can be determined and used to determine the strain in the substrate. The gauge factor (GF) of a material is used to characterize its strain sensitivity and is defined by eq 40

Chemical Reviews, 2008, Vol. 108, No. 2 529

GF )

∆R/R ∆L/L

(40)

where ∆R is the change in resistance, R is the initial resistance, ∆L is the change in length, and L is the initial length. The numerator is also known as the strain. The relation between strain and resistance change is linear. The application of thin, narrow gold traces to a microcantilever similarly can be used to measure its deflection. Piezoresitive detection is the second most common technique used for measuring cantilever deflection, even though its sensitivity is less than that of the optical lever.55 It is applicable to cantilever arrays of almost any size. The read-out electronics can be integrated onto the chip containing the cantilever array. This technique is unaffected by lightabsorbing or scattering components in the analyte stream. Because current is flowing through the cantilevers while measurements are being made, local heating can occur. It can be manipulated by changing the amount of current flowing through the resistive layer.60 Other drawbacks to this technique are thermal, electronic, and conductance fluctuation noise, thermal drifts, nonlinearity in piezoresponse, and poor sensitivity.45

3.4. Capacitive In this detection mode, the cantilever acts as one of the parallel plates of a capacitor. As the cantilever deflects, the distance between the two plates changes and this changes the capacitance of the system. The advantage of capacitive detection is in the simplicity of the associated electronics.65 This technique is not one of the more common ones used because of a number of limitations. To accurately record cantilever deflection, the dielectric material between the conductive plates must be constant throughout the experiment. The presence of analyte within the gap often changes its effective dielectric constant. Additionally, if the parallel plates are brought in too close proximity, they may stick together, which terminates the collection of useful data until they become separated. This phenomenon is frequently encountered when solvent vapor is passed over the cantilever and the solvent condenses onto the surfaces. Also, although the capacitive cantilevers can be integrated onto a microchip,66 scaling down the size of the capacitive cantilever will lower its overall sensitivity because the capacitance of a capacitor is directly proportional to its surface area. For gas sensing, Amirola and co-workers67,68 used capacitive detection of gaseous molecules and found the limit of detection (for their specific cantilever set up) to be 50 ppm for toluene and 10 ppm for octane. Verd and co-workers report sensitivity on the order of 10-8 g for their specific capacitive cantilever system.69

4. Design, Materials, and Fabrication Fabrication of cantilevers is an attractive option for groups with the appropriate resources, facilities, and time available. Creating cantilevers in-house allows greater flexibility in the design of the cantilever to enhance its suitability for the intended application. Only recently have commercial sources for cantilever arrays become available (see, for example, the following websites: http://www.concentris.ch; http://www. micromotive.de/Octosensis_e.php; and http://www.cantion. com). This section examines the interplay between the shape of the cantilever, the material from which it is made, and

530 Chemical Reviews, 2008, Vol. 108, No. 2

the fabrication methods required to achieve cantilevers with desired mechanical properties.

4.1. Design Considerations The shape of the cantilever often depends upon the detection technique. For example, square pads on the end of cantilever beams are used with capacitive detection systems to increase sensitivity because the measured capacitance is proportional to the surface area of the parallel plates.70 Piezoresistive cantilevers are often U-shaped, with components of the Wheatstone bridge circuit manufactured at their base points. For optical detection schemes, rectangular, paddle, and T-shaped cantilevers are quite common. However, as pointed out by Mertens and co-workers48,71 and previously in the Theory of Operation section, the actual cantilever deflection does not agree exactly with Stoney’s equation, especially when the shape of the beam is different from that which Stoney used in deriving his equation. The implicit assumption behind the equation is that surface stress will cause a uniform curvature of the beam. Because the cantilever beam is clamped on one end, the surface-stress induced curvature is not uniform, and this can cause the beam to twist.48 To reduce cantilever torsion caused by the additional stress at the clamped end, Plaza et al. used T-shaped microcantilever arrays.72 The “T” allows the major part of the beam to be mechanically decoupled from the twist-inducing stress at the clamped end.

4.2. Fabrication of Silicon-based Cantilevers The microfabrication process for silicon-based (i.e., silicon, silicon nitride, and silicon dioxide) cantilevers comprises four main techniques that, when used in combination, yield multiple cantilever chips with the desired shape and mechanical properties. These techniques are film deposition, photolithography, etching, and doping,73 the same as those commonly used in fabricating integrated circuits. The intent of this section is to provide an overview of each technique. For in-depth information, the reader is referred to the review by Hierlemann et al.73 and Madou’s text.74 The fabrication process typically begins with a polished monocrystalline wafer of silicon or silicon-on-insulator (SOI).75,76 SOI wafers are composed of a thick bottom later of single-crystal silicon, a middle silicon oxide layer, and a top layer of single-crystal silicon or silicon nitride.77 SOI wafers are useful because the buried oxide layer acts as an etch stop during the fabrication process. The thin top layer of single-crystal silicon (or silicon nitride) is commonly used as the material of the actual cantilever, so it is important that the defects in this layer are minimized.75-77

4.2.1. Film Deposition Deposition of thin films onto the wafer is carried out by spin-coating, either chemical (CVD) or physical (PVD) vapor deposition, and electroplating. Spin-coating is useful for the formation of polymer thin films, most commonly photoresist, whose utility is described in the next section. The wafer to be coated with the polymer is placed onto a vacuum chuck, which holds it in place. An aliquot of photoresist is dropped onto the wafer, and then the wafer is rotated at thousands of rotations per minute to distribute the polymer evenly over the wafer. Generally, spin-coated polymer thin films have thicknesses of 1-2 µm.74

Goeders et al.

CVD is used to deposit silicon oxide and silicon nitride layers that can be used as insulation, masks, and etchstops.73,74 In CVD, gaseous reactants are introduced into the vacuum chamber containing a heated wafer substrate. A thin layer is deposited onto the heated substrate via thermally induced reaction. Depending on the material deposited, CVD films can range in thickness from 20-1500 nm.73 Metals are generally deposited using PVD, i.e., sputtering and evaporation.73,74 In a PVD process, a thin film accumulates on the substrate from a heated reservoir of material in a linear alignment. Metal also can be deposited through electroplating. Metals are useful as reflective surfaces, electrode material, electronic interconnects, thermistors, and chemically reactive binding sites (because alkanethiols covalently bind to gold).

4.2.2. Photolithography Photolithography is the process used to transfer a pattern onto the wafer. First, a thin film of a UV-active polymer in a volatile solvent (i.e., photoresist) is placed on the wafer by spin-coating. Excess solvent is evaporated by heating the wafer in an oven. Next, a glass plate with transparent and opaque regions (mask) that contains the desired pattern is placed close to the wafer; then the mask and wafer are exposed to UV light. Depending upon the tone of photoresist used, UV light exposure initiates chemical bonding between adjacent polymer strands (cross-linking) or chemical bond cleavage along a strand. The reaction is completed as the wafer is placed in the oven for the postbake. Placement of the exposed photoresist wafer into a developer solution dissolves away the uncross-linked polymer and products of the chemical bond cleavage reaction. Etching is the final step that transfers the pattern from the photoresist onto the wafer. The remaining photoresist protects the underlying wafer from the etchant. After the wafer has been etched, the remaining photoresist is removed.

4.2.3. Etching Etching is a process used to remove parts of a thin film or the wafer. There are many different etching reagents; both liquid and dry etchants are available. The specific chemicals used for etching are chosen so that they preferentially etch one type of material over another. This way, thin film layers of various materials can prevent certain features of the wafer from being etched and transfer the desired pattern onto the wafer.

4.2.4. Doping Doping refers to the process of introducing specific impurities into the silicon lattice to alter the electrical conductivity of the silicon. Ion implantation and thermal diffusion are two methods of doping. The type of doping describes whether the dopant contains more or less valence electrons than silicon. For example, incorporation of boron or gallium into the silicon lattice results in p-type doping. These elements have one less electron than silicon; at their location in the lattice, a “hole” is momentarily created. Similarly, incorporation of phosphorus or arsenic into the lattice results in N-type doping. These elements have one more valence electron than silicon; at their location in the lattice, an unpaired electron resides. The unpaired electron and hole are charge carriers and can move about the lattice. Thus, the resistivity of the silicon wafer is determined by

Sensing Chemical Interactions via Mechanical Motion

the dopant type and concentration. Doping is commonly employed in piezoresistive cantilevers. The actual fabrication sequence depends upon the intended use and detection scheme. Silicon-based cantilevers used in optical detection schemes require fewer fabrication steps than those used in piezoresistive, piezoelectric, or capacitive detection schemes. For optical detection schemes, sequences of film deposition, photolithography, and etching are used. A reflective coating often is evaporated onto the surface of the finished cantilever beam to enhance the reflectivity of the beam. Silicon cantilevers used in piezoresistive, piezoelectric, and capacitive detection schemes require more fabrication steps because the detection mechanism is integrated onto the cantilever or the chip holding the cantilevers. Piezoresistive cantilevers require doping in specific areas to create the resistors of a Wheatstone bridge. Detailed information concerning the fabrication sequence and process optimization is readily available in the literature for piezoresistive cantilevers,78-82 piezoelectric cantilevers,83-90 and capacitive cantilevers.68-70,91,92 The number of steps can be reduced by use of silicon-on-insulator (SOI) wafers as the starting substrate.76,77,93 SOI wafers are composed of a thick bottom later of single-crystal silicon, a middle silicon oxide layer, and a top layer of single-crystal silicon. These wafers are commercially available in a variety of layer thicknesses and dopant levels. SOI wafers are useful because the buried oxide layer acts as an etch-stop during the fabrication process. The thin top layer of single-crystal silicon is commonly used as the material of the actual cantilever, so it is important that the defects in this layer are kept at a minimum.

4.3. Fabrication of Polymeric Cantilevers Microcantilevers fabricated from polymers inherently possess readily tailorable mechanical and chemical properties. To alter the stiffness of silicon-based cantilevers, their geometry must be changed or a rigid coating must be applied to the surface. In contrast, the stiffness of polymeric cantilevers requires only a change in material. In this way, microcantilevers with the same geometry but different properties can be produced. This reduces manufacturing costs and simplifies the apparatus required for detection. The materials used for polymer-based cantilevers span a wide range of thermosets, thermoplastics, and polymeric composites. Examples of polymer composites include silver nanoparticles and SU-8,94 carbon nanotubes, poly(m-phenylenevinylene-co-2,5-dioctoxy-p-phenylenevinylene),95 and many other combinations. Polymeric microcantilevers can be fabricated in a variety of ways; the method used is determined by the type of polymer to be used. For example, microcantilevers have been fabricated out of SU-8, a photopolymerizable epoxy-acrylate polymer. The process for fabricating SU-8 cantilevers is quite similar to that used for silicon-based cantilevers. A thin film of SU-8 is deposited onto a wafer by spin-coating. Photolithography then is used to define the regions that will comprise both the cantilever and the chip to which it is attached. The unwanted material is removed and the polymer cantilevers are released from the substrate by immersion in appropriate solvent. SU-8 cantilevers have been made into arrays for optical lever96-98 and piezoresistive99,100 detection schemes. Calleja et al. compared the deflection of silicon nitride cantilevers to SU-8 cantilevers with similar dimensions.96 When a surface stress change of 1 mN/m was applied to the

Chemical Reviews, 2008, Vol. 108, No. 2 531

Figure 9. Photograph of polymer microcantilevers produced by injection molding.

SU-8 cantilever, a deflection of 11 nm was observed using optical lever detection. When the same stress was applied to the silicon nitride cantilever, a deflection of only 1.2 nm was measured. On the basis of a minimum detectable deflection of 0.5 nm, Calleja et al. concluded that SU-8 cantilevers could be used to detect surface stress changes as small as 60 µN/m. Injection molding, an economical, mass production technique, has also been used to fabricate microcantilevers out of thermoplastic polymers.15,18,37,101 In this process, a molten polymer is forced under pressure into a steel cavity (mold); the shape of the cavity defines the dimensions of both the base and the cantilever(s), as shown in Figure 9. Microcantilevers with thicknesses down to 2 µm and lengths of up to 500 µm have been produced. Because of the small size of microcantilevers, the mold must be heated to the temperature of the molten polymer to ensure mold filling. Any thermoplastic polymer material can be formed into microcantilevers with injection molding; examples include polystyrene, polypropylene, liquid crystal polymer, polymethylmethacrylate, and nanoclay-filled nylon. Cantilevers with tips also have been molded in this manner.101 Injection-molded microcantilevers have been shown to be of equal caliber to commercial silicon microcantilevers. McFarland and coworkers15,18,37,101 detail the fabrication of injection-molded microcantilevers. Despite their advantages over silicon-based cantilever arrays, polymeric cantilever arrays are not commercially available.

5. Chemical Selectivity To achieve selectivity in response, one or more surfaces of the cantilever must be modified to promote binding of desired analytes to the surface and inhibit interfering substances from doing so. A variety of approaches have been used to impart selectivity to microcantilever sensors. The efficacy of a specific approach depends, to a large extent, on the complexity of the sample matrix in which the sensor is used and the chemical reversibility of analyte binding to the cantilever coating. For detection of a gaseous analyte in an air stream, metallic or ceramic films with a high affinity for the analyte are useful. When all surfaces of the cantilever are coated with a selective thin metallic or ceramic film, then the concentration of analyte in the air stream is proportional to the change in frequency. When only one side of the cantilever is laden with the selective thin film, then the concentration of the analyte in the air stream is proportional to the extent of deflection. Thin metallic or ceramic films are applied to the

532 Chemical Reviews, 2008, Vol. 108, No. 2

desired surfaces of the cantilever using the film-deposition technique described above. To prevent delamination, an intermediate adhesion-promoting layer is often employed. As an example, if a thin film of gold is to be applied to one side of a cantilever by evaporation, then a thin underlayer of titanium or chromium is used to promote adhesion of the gold film onto the silicon cantilever. Mercury vapor in air chemisorbs onto gold films with high affinity.5,102-104 Known interfering substances include water vapor and volatile organic compounds, which bind to either the gold or mercury surface (e.g., thiols and nitriles). Thus, high fidelity detection of mercury vapor in air is possible with a single gold-coated cantilever sensor only when the air stream has been dried and scrubbed of these interfering substances. The chemical selectivity of metallic and ceramic surfaces is significantly diminished in fluid. To enhance selectivity, several research groups have self-assembled monolayer films onto one or more faces of the cantilever. Reactive terminally substituted thiols, silanes, and siloxanes are commonly used to impart specific chemical functionality to the surface. The choice of reactive group depends upon the cantilever surface composition; the choice of terminal group depends upon the specific chemical interaction desired to attract the analyte to the cantilever surface. While this approach is effective in creating densely packed films on the surface, the analyte binding capacity is limited. Construction of multilayer films, through self-assembly, dip-coating, or spin-coating, is one means of increasing the capacity of the chemically selective film. Examples of this approach in the literature include the use of trialkoxysilanes,105-107 cyclodextrins,108-110 hydrogels,111-118 and polymers119,120 as chemically selective coatings. While the capacity of the film increases with increasing film thickness, as the film thickness increases, the following occur: • The rate of analyte transport into and out of the film diminishes, thereby slowing the temporal response of the sensor. • The added mass changes the effective spring constant of the cantileversthe degree of change depends upon the uniformity of coverage of the film on the cantilever. • The viscoelastic response of the film impacts the temporal response of the cantilever and its Q factor. • The number of compounds that partition into the coating increases, thereby reducing the chemical selectivity of the film. Thus, there is a clear tradeoff between film capacity and both detection specificity and temporal response. The impact of this tradeoff is minimized through the use of arrays in which each cantilever in the array has a different coating. A variety of coatings are available; the identification of the most appropriate coatings for detection of specific analytes in either gas or fluid streams has been aided by the application of chemometrics to this field.42,49,121-125 Perhaps the most promising area for development and application of cantilever sensing technology is biology. Deliberate attachment of biological molecules to a cantilever surface opens the possibility of highly selective interactions between the capture molecule and its binding partner. The large number of highly selective binding pairs in biology suggests that, through judicious selection of the coating, cantilever sensor systems can be designed to detect single analytes in complex media with high fidelity. The challenges in creating chemically selective biofilms lie in the following: controlling the spatial distribution and orientation of

Goeders et al.

Figure 10. Immersion of a cantilever array into an array of glass microcapillaries filled with food coloring for demonstration purposes. Reproduced with permission from Bietsch et al. Nanotechnology 2004, 15, 873. Copyright 2004 Institute of Physics Publishing.

capturing agent; minimizing the nonspecific binding by components in the sample matrix to the cantilever; maximizing the chemical sensitivity and dynamic range in sensor response; and extending the time before the capture agent denatures, thereby eliminating selectivity in cantilever response. Various approaches for immobilizing capturing agents on cantilever surfaces have been published. Most involve covalent attachment of the biomolecule directly to the cantilever surface or through a hetero-bifunctional linker molecule (e.g., alkanethiol or -siloxane). The use of a linker facilitates uniform distribution of biomolecule on the surface and minimizes denaturing caused by interactions with the cantilever surface. Blocking agents (e.g., polyethylene glycol and bovine serum albumin) typically are employed to reduce nonspecific binding. Further details regarding biomolecular coatings on cantilevers are provided below in the Biologicial Applications section. Coating individual cantilevers within arrays can be challenging. One way is to insert the desired cantilever into a capillary filled with reagent using a micromanipulator.51 The capillary must have an internal diameter larger than the width of the beam, and the wall of the capillary must be thin enough to fit between the cantilever beams in the array. The capillary is held in place for an allotted amount of time required for functionalization and then retracted. When several different cantilevers within the array require functionalization, this approach becomes time-consuming and tedious. Three approaches have proven useful for functionalizing multiple cantilevers: capillary arrays, inkjet printing, and contact printing. All cantilevers within the array can be simultaneously inserted into an array of capillaries (or smallvolume reaction wells) using an appropriately designed micromanipulator (see Figure 10).126,127 All sides of the cantilever are wetted using this approach. Thus, if only one side of the cantilever is to be modified, then the reaction chemistry of the fluid within the capillary must be designed to react only with the desired region (e.g., photochemically initiated reaction). Inkjet printing is also useful for coating individual cantilevers within an array.128,129 Commercial micro-inkjet printing systems are available from several manufacturers (e.g., Cantisens and Microdrop Technologies). Micro-inkjet printing affords efficient and controlled functionalization of only one side of the cantilever (see Figure

Sensing Chemical Interactions via Mechanical Motion

Figure 11. Inkjet printing of individual droplets onto a cantilever array: (a) schematic and (b) image from a video camera. A positioning system allows accurate placement of single droplets onto selected cantilevers. Reproduced with permission from Bietsch et al. Nanotechnology 2004, 15, 873. Copyright 2004 Institute of Physics Publishing.

11), and it is faster than the capillary array. Coatings also can be applied by contact-printing methods using dip-pen lithography130-134 or specially designed stamps.135,136 Commercial dip-pen lithography systems are available (e.g., BioForce, NanoInk, and Nanonics Imaging).

6. Chemical Applications Numerous applications of chemomechanical sensors in environmental monitoring, medical diagnostics, and chemical detection in air and flowing liquids have been published. Several reviews have been published over the past decade.55,60,137-144 This section highlights recent and innovative applications. Because of editorial restrictions on the number of citations, the work cited herein illustrates only some of the applications currently being explored. Our selections are, without doubt, subjective.

6.1. Volatile Organics In a series of papers emanating from IBM Zurich, the University of Basel, and the Paul-Scherrer-Institute, the efficacy of cantilever arrays for detection of specific analytes in complex gaseous mixtures has been demonstrated. Lang and co-workers11,50,145 showed that the diffusion of various alcohols into polymethylmethacrylate coating induces resonance frequency shifts and differential bending of cantilevers. Baller et al.49 coated each cantilever in the eight-cantilever array with a specific polymer layer to transduce a physical process or a chemical reaction into a nanomechanical response. Chemisorption of the analyte induced polymer swelling; the kinetics of the swelling process was related to the vapor pressure and the solubility characteristics of the analyte in the polymers. The array format enabled the use of some cantilevers as reference sensors. Baller et al. distinguished different mixtures of alcohols using principal component analysis (PCA) with mixtures that had been previously characterized. They could not determine the mixing ratio of individual analytes directly from the cluster positions of the mixture’s constituents in the PCA plot because desorption kinetics of analyte mixtures do not depend on the mixing ratio in a predictable way. A year later, the same group demonstrated the simultaneous detection of

Chemical Reviews, 2008, Vol. 108, No. 2 533

deflection and resonant frequency shifting of cantilevers within the array to the same analytes.42 Using artificial neural network analysis of the deflection and resonant frequency shift data, they demonstrated the utility of polymer-coated cantilevers for both qualitative and quantitative analysis of gaseous mixtures with well-defined composition. In related work, Betts and co-workers119 evaluated two polymeric chromatographic stationary phases as cantilever coatings for select vapor phase analytes. Fagan et al.107 evaluated sol-gels as cantilever coatings for nonpolar vapor phase analytes. Maute and colleagues146-149 used polydimethylsiloxane, polyetherurethane, zinc phthalocyanine, and ethyl cellulose as coatings for the detection of volatile organic compounds in the gas phase. Differential chemisorption of analytes into each polymer film and principal component analysis of the response of each cantilever to the analyte provided a means of qualitative and quantitative determination. Improved performance was found when using higher resonant modes for detection. Headrick et al.109 used focused ion milling of the cantilever surface to create submicron channels across the width of one side of the cantilever. Responses of the nanostructured, coated cantilevers to a series of volatile organic compounds were compared to smooth, coated cantilevers. The results showed that roughened cantilevers were more sensitive, i.e., exhibited an increase in differential stress to the analytes investigated. Lange et al.150 compared the performance of cantilever arrays to thickness shear mode resonators and to surface acoustic wave devices to detect volatile organic compounds in vapor phase. From parallel analyses performed by these transducers on a mixture of n-octane and toluene, it was shown that the limit of detection achieved with cantilever sensors is comparable to that of other acoustic wave-based gas sensors. To enable improved quantification of analyte mixtures, Kurzawski et al.151 evaluated the performance of a singlechip, three-transducer, complementary metal oxide semiconductor gas sensor microsystem. This system comprised a mass-sensitive cantilever, a thermoelectric calorimetric sensor, and an interdigitated capacitive sensor. Each sensing element was coated with various partially selective polymers and then was exposed to different volatile organic compounds. The sensitivities of the three different polymercoated transducers to defined sets of gaseous analytes were determined. These workers have demonstrated that each transducer responds to fundamentally different molecular properties. Thus, the response of each transducer to an analyte provides orthogonal data from which analytes present in the mixture can be quantified, using appropriate signalprocessing and pattern-recognition techniques. Fadel et al.26,152 investigated the analysis of gaseous mixtures using piezoresistive cantilevers of millimeter dimensions. They showed that the choice of the cantilever dimensions and the polymer thickness for gas detection requires compromises concerning sensitivity, response time, quality factor, and resonant frequency. Their comparison between millimeter-size and micrometer-size cantilevers shows the importance of noise in the design of an integrated sensor. The Ziegler group8,153 showed that electrostatic or magnetic actuation of the cantilever results in the enhancement of the quality factor by over 3 orders of magnitude for commercial cantilevers. With actuation, cantilever sensors possess a 1000-

534 Chemical Reviews, 2008, Vol. 108, No. 2

fold higher mass sensitivity compared to quartz crystal microbalances.

6.2. Chemical Warfare Agents Cantilever technology has been applied to detect chemical warfare agents.154-159 Most published reports have centered on detection of the nerve agent simulant dimethylmethylphosphonate in their studies. Limits of detection were in the 0.5-20 ppb range, depending upon the coating used and the detection mode.160-164 In only a few instances has the selectivity of detection been investigated.165 Assessment of the practical utility of this sensing technology for chemical warfare agent detection awaits a systematic study of the selectivity of coatings to common components found in the atmospheres of cities and on the battlefield.

6.3. Explosives Two approaches have been explored in developing cantilever sensors for detection of explosives: deflagration of particles placed on the cantilever155,166-168 and chemisorption of vapor into thin coatings on cantilevers. The low volatility of most explosives limits the utility of the latter approach and sensitivities obtained to date are less than those of competitive technologies.169-171

6.4. Toxic Metal Ions The concentration of a variety of metal ions in solution has been determined using cantilever technology. For metal ions that chemisorb (or amalgamate) with the metallic coatings used to increase reflectivity for optical lever measurements of cantilever detection, quantization of metal ion concentration is straightforward (e.g., detection of Hg2+ with gold-coated cantilevers).172 For other metals, cantilever deflection can be induced through ion exchange of the analyte onto thin film coatings. For example, the concentrations of Cr2+, Ca2+, Cs+, and CrO42- can be determined via ionexchange with ω-modified alkanethiol monolayers selfassembled onto gold-coated cantilevers.123,173-175 In some instances, very low limits of detection are obtained (e.g., Cs+ and CrO42-), whereas in others (e.g., Ca2+), the limit of detection, dynamic range, and selectivity of the method are not competitive with ion-selective electrode technology.173,175 Hydrogel coatings can also be used in quantifying metal ion concentrations.116,117 Monolayers composed of alkanethiols modified with crown ethers have proven to be an effective way to improve selectivity and sensitivity for specific ion detection.175

7. Biological Applications 7.1. Cells In 2001, Ilic and co-workers176 first reported the detection of Escherichia coli cells using a cantilever array. Selective binding to the cantilever was achieved with the anti-E. coli antibodies immobilized onto the cantilever. Resonant frequency shifts correlated with the number of cells bound to the surface. The sensitivity of the method was sufficient to detect the binding of a single cell. These findings were confirmed by Zhang and Ji.177 Campbell and Mutharasan178 extended the work of Ilic and co-workers, showing that a composite self-excited cantilever made of a PZT film and glass of a few millimeters in length and coated with anti-E.

Goeders et al.

coli antibodies can be used to detect E. coli O157:H7 in fluid. Sensitivity achieved was in the order of tens of nanograms, with a limit of detection of only 700 bacterial cells/mL.179 Gfeller et al.180,181 demonstrated that an oscillating cantilever can be used as a sensor for active bacterial growth. Their approach was elegant in its simplicity. E. coli cells were deposited onto cantilevers coated with a thin nutritive agarose layer and kept in a humid environment. Within minutes, the cells started to grow and assimilate water, protein, salts, and carbohydrates from the nutritive layer. To regain equilibrium with the humid environment, the nutritive layer absorbed water; the resultant mass increase produced a commensurate shift in resonant frequency. When they compared the observed frequency shifts due to additional mass loading onto the cantilevers with a conventional bacterial growth curve, all characteristic bacterial growing phases were observed. By incorporation or omission of antibiotics in the cantilever coating, they demonstrated the utility of their approach for rapidly assessing antibiotic resistance. This new application of cantilever array technology offers numerous advantages over conventional bacterial detection methods including rapid real-time detection, labelfree and small analyte volume, and high sensitivity. Ramos and co-workers182 recently showed that the response of oscillating cantilevers to bacteria adsorption depends on the added mass, the site of immobilization of the cell on a cantilever, and the stiffness of the bacterial cells. They predicted that detection sensitivities can be increased by an order of magnitude or more by monitoring higher vibrational modes or scaling down cantilever size. However, the mechanical properties of adsorbed molecules became increasingly important as the size of the resonator was decreased. Taken collectively, these reports portend of the use of microcantilever-based sensors for detection of pathogenic bacteria in medical diagnostics and monitoring of our food supply.

7.2. Viruses Ilic et al.183 first reported on the use of cantilever arrays to detect immunospecific binding of viruses, captured from liquid. Baculovirus particles bound selectively to an AcV1 antibody monolayer immobilized onto the cantilever surface. The resonant frequency shift resulting from the adsorbed mass of the virus particles distinguished solutions of virus concentrations varying between 105 and 107 pfu/mL. Single virus particle detection was achieved using specially designed cantilevers. Similar findings were reported by Gupta and coworkers184 using vaccinia virus, a member of the PoxViridae family and the virus that forms the basis of the smallpox vaccine.185 Ji and co-workers159 showed the utility of antibody-antigen binding interactions for detection of biowarfare agents ricin and tularemia. Dhayal and co-workers130 demonstrated the utility of peptide-functionalized silicon cantilever arrays for detection of whole B. subtilis spores (a nonpathogenic B. anthracis simulant) in liquids. Real-time detection was achieved by monitoring stress changes in the cantilever due to spore binding. Estimates for the induced stress per binding event were obtained. They also observed a higher sensitivity to resonant frequency shifts by monitoring with the fifth mode of vibration. There results suggest that real-time detection of multiple pathogenic organisms can be realized using peptide-funtionalized microcantilever arrays. Campbell and Mutharasan186,187 investigated the detection of pathogen Bacillus anthracis spores in liquid under both

Sensing Chemical Interactions via Mechanical Motion

stagnant and flow conditions. They reported the detection of B. anthracis spores at a very low concentration (300 spores/mL) using piezoelectrically excited millimeter-sized cantilever sensors coated with antibody specific to B. anthracis. High selectivity was demonstrated by detecting B. anthracis spores in the presence of another Bacillus spore (Bacillus thuringiensis) at ratios up to 1/500. More complicated spore mixtures also have been examined.188 In these, the presence of non-antigenic Bacillus species reduced the binding kinetics of B. anthracis spores but did not alter the steady-state response of the sensor. Nugaeva and co-workers189 explored the use of cantilever arrays for selective immobilization and rapid detection of fungal spores. Cantilever arrays were exposed to either the mycelial form Aspergillus niger or the unicellular yeast form Saccharomyces cereVisiae, as models to explore their utility for growth detection of eukaryotic organisms using cantilever arrays. These workers exploited the specific biomolecular interactions of surface-grafted proteins (concanavalin A, fibronectin, or immunoglobulin G) with the molecular structures on the fungal cell surface to achieve selective immobilization of the spores. They found that these proteins have different affinities and efficiencies to bind the spores. Maximum spore immobilization, germination, and mycelium growth were observed on the immunoglobulin G functionalized cantilever surfaces. They also found that spore immobilization and germination of the mycelial fungus A. niger and yeast S. cereVisiae led to shifts in resonance frequency within a few hours, in contrast to conventional techniques that require several days. Measured frequency shifts were proportional to the mass of single fungal spores, and this biosensor could detect the target fungi in a range of 103-106 CFU/mL. This work exemplifies an important application of cantilever array technology in medical and agricultural diagnostics and food- and water-quality monitoring.

7.3. Antigen−Antibody Interactions Raiteri and colleagues60 reviewed the working principles behind cantilever-based sensors based on antigen-antibody interactions. The reader is referred to this review for a critical analysis of the literature in this area up through the year 2000. Several reports have appeared over the last 7 years that utilize antigen-antibody binding for selectivity. For example, Arntz et al.51 presented continuous label-free detection of two cardiac biomarker proteins, creatine kinase and myoglobin, using anti-creatine kinase and anti-myoglobin antibodies covalently anchored to a cantilever array. Binding of the antigen to the anchored antibodies generated sufficient surface stress to enable detection via cantilever deflection. Both myoglobin and creatine kinase could be detected independently using cantilevers functionalized with the corresponding antibodies, in unspecific protein background. These workers showed the utility of reference cantilevers to eliminate thermal drift, undesired chemical reactions (i.e., nonspecific binding), and turbulence from injection of liquids into the cell. They achieved a sensitivity detection of myoglobin below 20 mg mL-1. Grogan et al.190 investigated the activity, stability, lifetime, and reusability of monoclonal antibodies to myoglobin covalently immobilized onto cantilever surfaces. Sucrose was shown to be an effective stabilizing agent for the immobilized antibody layer; with it, the immobilized antibody was found to have a stable active lifetime for up to 7 weeks.

Chemical Reviews, 2008, Vol. 108, No. 2 535

Alvarez and co-workers191 reported the use of a synthetic hapten conjugated with bovine serum albumin as a bioselective layer for cantilever-based detection of the pesticide dichlorodiphenyltrichloroethane (DDT). Exposure to a solution of a specific monoclonal antibody to the DDT hapten derivative results in deflection of the cantilever. Specific detection is achieved by performing competitive assays in which the cantilever is exposed to a mixture of the monoclonal antibody and DDT. Backmann and colleagues192 showed that single-chain Fv (scFv) antibody fragments can be used as receptors to detect antigens by the static deflection of cantilevers. The authors reported that the performance of the microcantilever-based immunosensor was comparable with surface plasmon resonance. By simultaneously tracking deflection of sensing and reference cantilevers, the differential deflection signal revealed specific antigen binding and was proportional to the concentration of antigen in solution. Dutta et al.193 reported the first demonstration of chiral discrimination using microcantilever sensors. Stable, reusable protein bioaffinity phases based on unique enantioselective antibodies were created by covalently linking monoclonal anti-D- and anti-L-R-amino acid antibodies to nanostructured cantilever surfaces. The temporal response of the cantilever (∆deflection/∆time) was linearly proportional to the concentration of chiral amino acid and allowed quantization of enantiomeric purity up to an enantiomeric excess of 99.8%. Hwang and co-workers194,195 have fabricated a selfactuating and self-sensing piezoelectric cantilever for labelfree detection of a prostate-specific antigen. Cantilevers were coated with parylene-c, deposited by chemical vapor deposition, to electrically insulate the oscillator circuitry for use in fluids.196,197 Specificity in detection of PSA was achieved through its binding to a PSA antibody that was immobilized via host-guest interactions with a proprietary calixcrown self-assembled monolayer. The resonance frequency shift of the cantilever was proportional to antigen concentration. This strategy also was used for detection of C-reactive protein.198,199 Kang and co-workers200 reported the assay of myoglobin concentration using PZT cantilevers coated with biotinylated myoglobin antibodies immobilized onto the surface through streptavidin conjugation. Most of the published works in this area focus on demonstrating that specific antigen-antibody pair interactions lead either to mass increases that can be sensed by shifts in cantilever resonance or to changes in surface stress that produce measurable cantilever deflections. Little attention is directed to performing cantilever-based immunoassays in a clinically relevant setting. A noteworthy exception is the work of Wu and colleagues.201 These authors report the detection of two forms of prostate-specific antigen (PSA) over a wide range of concentrations from 0.2 ng/mL to 60 mg/mL in a background of human serum albumin (HSA) and human plasminogen (HP) at 1 mg/mL. Prostate-specific antigen is a particularly useful marker for early detection of prostate cancer and in patient monitoring for disease progression. In serum, this biomarker exists in two forms: uncomplexed and complexed with the serum protease inhibitor R1antichymotrypsin. Early diagnosis of prostate cancer requires an accurate measure of both the total concentration and the ratio of the complexed to uncomplexed forms of the antigen in serum. In addition, the clinically useful range spans from 0.01 to >10 ng/mL. The dose-response curve they obtained

536 Chemical Reviews, 2008, Vol. 108, No. 2

Goeders et al.

Figure 12. Steady-state cantilever deflections as a function of uncomplexed (fPSA) and complexed (cPSA) prostate specific antigen concentrations for three different cantilever geometries. Reprinted with permission from Wu et al. Nat. Biotechnol. 2001, 19, 856. Copyright 2001 Nature Publishing Group.

with cantilevers of differing length is shown in Figure 12. This curve was obtained under static conditions that included thermal regulation. In contrast to the conventional enzyme-linked immunosorbent assay for this antigen, the cantilever-based assay required no labels and was performed in a single reaction without additional reagents. A logical extension of this work involves the use of an array of microcantilevers to perform multiple assays. High fidelity, clinically relevant detection of this biomarker for prostate cancer would be anticipated via the coating of individual cantilevers within the array with antibodies selective for different epitopes on this antigen. A number of such antibodies are now commercially available.

7.4. DNA Hybridization Fritz and his colleagues125 pioneered the use of cantilevers for detection of nucleic acid hybridization. Deflection of each cantilever in the array was measured using the optical beam deflection technique. 5′-Thio-modified synthetic DNA oligonucleotides with different base sequences were covalently immobilized on the gold-coated side of the cantilevers the array. When solutions containing the complementary oligo were injected into the liquid cell, hybridization resulted in a change in surface stress between the functionalized gold and the nonfunctionalized Si surface, bending the cantilever. This is shown schematically in Figure 13. This work stimulated interest in exploiting the sensitivity of chemomechanical detection of DNA hybridization. A crucial test for any DNA hybridization sensor is its ability to discern mismatches. Fritz and colleagues125 observed a small but measurable difference in surface stress between a pair of complementary oligos and a pair with a single base mismatch between two DNA sequences that can be detected. Hansen and co-workers202 further evaluated the capability of cantilever sensors for detecting single base mismatches. They found that the direction of cantilever bending, whether tensile or compressive, depended up the number and location of mismatch sites along the strand pairs. Wu et al.203 showed that the magnitude of cantilever deflection during hybridization depends upon the ionic strength of the matrix. McKendry et al.127 systematically examined the impact of single strand extensions on cantilever deflection during DNA hybridization. In all cases, compressive surface stress results from hybridization, regardless of whether the complementary

Figure 13. Schematic illustration of the hybridization assay. Each cantilever is functionalized on one side with a different oligo (red or blue). (A) The differential signal is set to zero. (B) After injection of the first complementary DNA strand (green), hybridization occurs on the cantilever laden with the matching sequence (red), increasing the differential signal. (C) Injection of the second complementary DNA oligo (yellow) causes the cantilever functionalized with the second oligo (blue) to bend. Reprinted with permission from Fritz et al. Science 2000, 288, 316. Copyright 2000 American Association for the Advancement of Science.

oligomers have the same number of nucleotides. They found that hybridization between two complementary 12-mers generated an average compressive surface stress of 2.7 mN/m. Hagan et al.204 presented an explanation of cantilever deflections resulting from adsorption and subsequent hybridization of DNA molecules. Using an empirical model, they predicted deflections upon hybridization that are consistent with experimental results. They asserted that hydration forces, not conformational entropy or electrostatics, are the dominant contributors to deflections arising from DNA hybridization. They showed that predicted deflections before and after hybridization strongly depend on surface coverage as well as the degree of disorder on the surface. The latter point was experimentally verified by Alvarez and co-workers.205

Sensing Chemical Interactions via Mechanical Motion

In a follow-up report from the Majumdar group, Stachowiak et al.206 provided experimental evidence that the surface stresses resulting from hybridization depend on oligo length, grafting density, and hybridization efficiency. At low ionic strength, the osmotic pressure of counterions dominates the intermolecular forces, while at higher ionic strength, the grafting density is independent of the ionic strength and hydration interactions dominate. They also showed that, regardless of the length and grafting density of the singlestranded probe DNA, surface stress was related exponentially to the density of hybridized DNA. The same group207,208 observed surface stress changes in response to thermal dehybridization, or melting, of double-stranded DNA oligonucleotides that were grafted on one side of a microcantilever beam. Changes in surface stress occur when one complementary DNA strand melts and diffuses away from the other, resulting in alterations in the degree of hydration and electrostatic interactions between the remaining neighboring surface-grafted DNA molecules. They were able to distinguish changes in the melting temperature of dsDNA as a function of salt concentration and oligomer length. Recent effort has focused on improving sensor performance. Several groups have evaluated piezoresistive detection of DNA hybridization as an alternative to optical methods.128,209,210 While piezoresistive detection is less sensitive than the optical lever method, piezoresistive methods are sufficiently sensitive to detect hybridization and single base mismatches. Improvement in sensitivity is anticipated with continued optimization of piezoresistive cantilevers. Others have focused on fabricating cantilevers from polymers in hopes of lowering the limit of detection through reduction in cantilever spring constant without significantly changing the active area.96,211 Su et al.212 used gold nanoparticle modified oligos to improve the mass sensitivity of resonant frequency-based microcantilever detection of nucleic acid hybridization. Their method is capable of detecting DNA concentrations as low as 0.05 nM. As clearly pointed out by Alvarez and coworkers,205 detection of nucleic acid hybridization requires reference cantilevers sensitized with noncomplementary DNA to decouple the molecular recognition signal from nonspecific binding events and matrix effects. This highlights the need for an improved understanding of the mechanisms responsible for surface stress due to the biomolecular interactions. Such knowledge is crucial for the development of immobilization procedures in which the geometry of the receptor molecules is addressed to generate high interaction forces between neighboring molecules during molecular recognition.

7.5. Enzymes Subramanian et al.213 reported on the first microcantileverbased enzymatic assay. The enzyme glucose oxidase was immobilized onto a gold-coated silicon cantilever with glutaraldehyde following coating of the gold surface with poly-L-lysine. Quantifiable deflection of the cantilever was observed in the presence of analyte. Their analysis of the heat of the enzymatic reaction and the thermal sensitivity of the cantilever suggested that cantilever deflection is not simply a result of reaction-generated heat but appears to result from surface-induced stresses. They offered two hypotheses to account for the surface stress: entropic effects due to the continual binding of glucose at the active site of the enzyme and changes in the local chemical environment that result from glucose conversion to gluconic acid and peroxide.

Chemical Reviews, 2008, Vol. 108, No. 2 537

Figure 14. Schematic illustration of protein binding and enzymatic assays with cantilevers.

Yan and co-workers sought to clarify the source of cantilever deflection when immobilized glucose oxidase is exposed to glucose.214,215 Their immobilization strategy differed from that of Subramanian et al. in that the enzyme was electrostatically immobilized within an alternately charged polyelectrolyte multilayer structure that comprised poly(sodium 4-styrenesulfonate) and polyethyleneimine. The multilayer approach provided improved performance. They proposed that cantilever bending results from both a conformational change of the enzyme in the presence of glucose and from protonation of the polymer multilayer structure as a consequence of the enzyme-catalyzed oxidation of glucose to gluconic acid. Pei et al.216 further characterized the performance of a glucose oxidase-based cantilever sensor. They cross-linked the enzyme to bovine serum albumen chemisorbed onto the surface. They concluded that the deflection response of the cantilever cannot be due to the heat of the enzymatic reaction and attributed the deflection mechanism to changes in the local chemical environment of the coating layer. They noted that the poor reproducibility of results for this enzyme-based glucose sensor is likely due to the corrosive nature of peroxide produced by the enzymatic reaction. Clearly, the mechanism behind the surface-induced stress observed for this type of glucose sensor remains unknown. Bottomley and co-workers reported the use of microcantilevers as sensors of enzymatic function.217 Exposure of cantilevers coated with enzyme substrates to enzymes capable of changing substrate mass, conformation, and charge results in measurable deflection of the cantilever as shown in Figure 14. Enzyme inhibitors also can be identified using this approach. Stevenson and colleagues218 monitored the restriction and ligation of cantilevers coated with DNA. An oligo containing the Hind III restriction site was immobilized on the cantilever and then digested with that enzyme; strand scission produced cantilever bending and left behind a shortened oligo with a single-stranded sticky end. Exposure of a second oligo with a compatible end to the DNA on the cantilever in the presence of ligase resulted in the extension of the immobilized oligo and commensurate cantilever deflection in the opposite direction. The authors point out that, since most DNA restriction and ligation enzymes require dithiothreitol to retain their activity, immobilization of the oligo through thiol linkages must be avoided. Otherwise, displacement of the thiolated DNA from the gold surface by dithiothreitol will produce cantilever deflection and complicate detection of the restriction and ligation events. Liu and colleagues219 presented a new approach to track enzyme action with cantilevers. Their technique relies on

538 Chemical Reviews, 2008, Vol. 108, No. 2

the detection of bead detachment from the cantilever due to the enzyme cleavage of the linker tethering the bead to the cantilever. To illustrate this principle, they used the enzymatic action of Botulinum neurotoxin type B on its substratum, the synaptic protein synaptobrevin 2. Nickel-agarose beads were functionalized with recombinant synaptobrevin 2 conjugated to six consecutive His residues at its C terminus. To suspend the bead off the cantilever tip, they used proteinprotein interaction (synaptobrevin 2 with another synaptic protein, syntaxin 1A). In the presence of zinc ion, the neurotoxin cleaves synaptobrevin 2, leading to the detachment of the bead from the tip. Since the mass of the bead is many times larger than that of the immobilized protein, its detachment chemically amplifies the mass loss of the protein fragment. The bead detachment technique is general and can also be used for any cleavage reaction. For example, Weizmann et al.220 utilized the endonuclease scission of magnetic beads functionalized with sequence-specific DNAs to detect single base mismatch specificity of the endonucleases. Magnetic beads were used to reduce thermal motion and amplify the mechanical motion of the cantilever to enzymatic action. In a subsequent report, they extended this approach to the development of enzyme-based AND or OR logic gates.221 The bead detachment technique is not limited to cleavage reactions; it is also suitable for displacement reactions, such as in receptor-ligand pairs, where the introduction of one chemical leads to the displacement of another.

8. Recommendations for Future Work 8.1. Guidelines for Reporting Sensor Performance To facilitate comparison with other sensing platforms, we suggest the following figures of merit be included in all future publications regarding the performance of microcantilever sensors: detector sensitiVity, limit of detection, dynamic range, and sensitiVity of the analysis. Within the microcantilever community, the term sensitiVity is used to describe several different parameters. Some workers use this term to describe the minimum concentration of analyte that can be detected. Others use it to describe the performance characteristics of the sensing technique used to measure shifts in cantilever resonance frequency or changes in cantilever deflection. A third group uses this term to describe the slope of the calibration curve. Multiple usages of the same word can lead to confusion and misinterpretation on the part of the reader. We suggest more explicit terminology be used in reporting results. The term “limit of detection” should be used to describe the minimum concentration of analyte that can be reliably detected. Convention within the analytical chemical community is that this is the concentration that gives a sensor response signal equal to three times the background noise level. The performance of the sensor to changing concentrations of analyte should be reported and distinguished from the sensitivity of the device used to measure cantilever movement. We suggest that the term “detector sensitiVity”, the measured change in signal per unit value of the sensor response, be employed in characterizing device performance. For optical level and piezoresistive detection of cantilever deflection, the detector sensitivities would have units of V/nm and (∆R/R)/nm, respectively. This term is a function of the properties of the cantilever, the deflection measurement technique, and the signal amplification techniques employed.

Goeders et al.

This figure of merit would be of benefit to those trying to compare cantilever designs, materials, and detection techniques. We suggest that the performance of the sensor to changing concentrations of analyte be labeled “sensitiVity of the analysis”. This parameter is determined from the slope of the linear region of the dose-response curve. As acquisition of this parameter requires exposure of the coated cantilever to varying concentrations of analyte and measurement of the system’s response, a measure of the chemical reversibility of analyte binding to the coating is readily obtainable and also should be reported. Similarly, in the course of determining the chemical sensitivity, the experimenters should determine and report the “dynamic range” of the sensor for the specific analyte under study. While this issue may seem obvious to the majority of readers, the omission of this information in many of the published papers has made it difficult for us, during the course of writing this review, to evaluate the scientific contribution of many papers and include them in the context of important, new applications of microcantilever sensor technology.

8.2. Experimental Design Considerations In the first few years following the invention of this sensor technology,1,3-6,222,223 experimentation with a single cantilever was commonplace. Since then, it has become well-established that cantilevers respond to small changes in temperature, viscosity, and ionic strength of the medium in which they are immersed, as well as to the flow dynamics of the cell that houses the cantilever chip. Thus, the utility of single cantilever experimentation in fluid streams is, at best, questionable. In many instances, interpretation of results of present-day research involving single-cantilever experimentation are based largely on the assumption of fixed conditions between sequential experiments. Often, experiments expressly designed to test the validity of the assumption are unreported. Microcantilever arrays are the preferred format. They enable control experiments to be performed simultaneously with analyses and provide more reliable control of empirical factors such as thermal drift, changes in viscosity, and solution flow dynamics. They also provide a straightforward means and correct for nonspecific adsorption and nonspecific chemical reactions that may occur on or within the chemically selective coating. In addition, multiple targets can be detected simultaneously, leading to high-throughput measurements and producing distinct recognition patterns from complex mixtures.139 We recommend that all future work with microcantilever sensor technology be performed using cantilever arrays. The field of microcantilever sensors has matured to the point where reports of new applications of this technology should include performance testing under relevant conditions with measures of the fidelity and selectivity of detection. For example, if a new chemically selective cantilever coating is developed that provides a means for detecting a volatile analyte in air, then the report should include the performance of this coating when exposed to a variety of air samples (e.g., compressed air, laboratory air, auto exhaust, etc.). Similarly, reports of new biological applications should include results of tests carried out in the fluids where the analyte is typically found (e.g., sputum, serum, urine, and cell lysate). While there is some value to disseminating results of analyses carried out with pristine solutions, reports of successful detection of specific analytes in complex mixtures signifi-

Sensing Chemical Interactions via Mechanical Motion

Chemical Reviews, 2008, Vol. 108, No. 2 539

cantly advances the field and provides strong impetus for increased commercial participation in the development of this sensor technology.

8.3. Fruitful Areas for Further Research To become competitive with existing commercial sensing technologies (e.g., quartz crystal microbalance, surface plasmon resonance, and surface acoustic wave), microcantilever sensors must provide faster, cheaper, more sensitive, rugged, and reliable analyses. In addition, the microcantilever sensing system must be easy to operate and field deployable. On the basis of these benchmarks and the present state-ofthe-art, there is a need for more research in the following areas.

8.3.1. More Selective Coatings The quartz crystal microbalance and surface acoustic wave device are two commercialized sensing technologies that rely on changes in mass for detection. To compete with these technologies, analyses based on shifts in cantilever resonance either should be performed on short, stiff cantilevers with resonance frequencies in the MHz range, or by tracking shifts in one of the higher resonance modes of conventional cantilevers. The latter is preferred as the small dimensions of short, stiff cantilevers reduce the capacity of the sensor and, thus, the dynamic range of detection. We suggest increased effort in the development of new, highly selective coatings that give rise to large changes in surface stress upon analyte binding. It seems likely that these coatings will utilize highly specific biomolecular interactions. Also needed are novel packaging approaches to increase the shelf life of existing biomolecular coatings.

8.3.2. Increased Sensitivity and Faster Response To compete favorably with benchmark sensing technologies, the speed and sensitivity of analysis with microcantilever sensors must be improved. Shortening the temporal response of cantilevers to analyte passing over the selective layer requires additional insight into the analyte-binding mechanism(s) as well as a dramatic reduction in the volume of the compartment in which the cantilever array is housed. The latter will require careful attention to the mass transport of analyte to the sensor surface and modeling of the fluid dynamics of analyte flow through the compartment and about the cantilevers. Thus, incorporation of arrays in microfluidics cartridges would seemingly be one way to shorten the response time. Another approach is being pursued by the Manalis group at MIT.224-226 They have achieved significantly enhanced sensitivity and very low limits of detection for fluidborne analytes using specially designed cantilevers that have integrated microfluidic channels within them (see Figure 15). The analyses are performed using optical lever detection with the cantilever under vacuum and sample flowing through the interior of the cantilever. This approach eliminates both the damping normally encountered when the cantilever is immersed in fluid and light scattering or absorption by the fluid sample, which negatively impacts optical lever detection. Selectivity in detection is achieved by precoating the walls of the microfluidics channel.225 This very recent advance suggests that the present shortcomings which impede many applications of microcantilever technology will be removed through innovations in the design of cantilevers,

Figure 15. Schematic illustration of cantilevers with integrated microchannels developed by Manalis and co-workers. (a) A suspended microchannel translates mass changes into changes in resonance frequency. Fluid continuously flows through the channel and delivers biomolecules, cells, or synthetic particles. (b) While bound and unbound molecules both increase the mass of the channel, species that bind to the channel wall accumulate inside the device, and, as a result, their number can greatly exceed the number of free molecules in solution. This enables specific detection by way of immobilized receptors. Reprinted with permission from Burg et al. Nature 2007, 446, 1066. Copyright 2007 Nature Publishing Group.

detection devices, and sample delivery systems; intelligent design of coating-layer composition and high-throughput methods for their application; and incorporation of chemometric methods of analysis for processing data acquired with cantilever arrays. With more advances such as these, microcantilever technology will enable rapid detection of harmful agents that may be present in the air we breathe and the fluids we ingest.

9. References (1) Gimzewski, J. K.; Gerber, C. H.; Mayer, E.; Schlitter, R. R. Chem. Phys. Lett. 1994, 217, 589. (2) Gimzewski, J. K.; Gerber, C.; Meyer, E.; Schlittler, R. R. NATO ASI Ser., Ser. E: Appl. Sci. 1995, 286, 123. (3) Chen, G. Y.; Thundat, T.; Wachter, E. A.; Warmack, R. J. J. Appl. Phys. 1995, 77, 3618. (4) Thundat, T.; Chen, G. Y.; Warmack, R. J.; Allison, D. P.; Wachter, E. A. Anal. Chem. 1995, 67, 519. (5) Thundat, T.; Wachter, E. A.; Sharp, S. L.; Warmack, R. J. Appl. Phys. Lett. 1995, 66, 1695. (6) Wachter, E. A.; Thundat, T. ReV. Sci. Instrum. 1995, 66, 3662. (7) McFarland, A. W.; Poggi, M. A.; Bottomley, L. A.; Colton, J. S. J. Micromech. Microeng. 2005, 15, 785. (8) Then, D.; Vidic, A.; Ziegler, C. Sens. Actuators, B 2006, B117, 1. (9) Then, D.; Ziegler, C. Encycl. Nanosci. Nanotechnol. 2004, 1, 499. (10) Abadal, G.; Davis, Z. J.; Helbo, B.; Borrise, X.; Ruiz, R.; Boisen, A.; Campabadal, F.; Esteve, J.; Figueras, E.; Perez-Murano, F.; Barniol, N. Nanotechnology 2001, 12, 100. (11) Lang, H. P.; Berger, R.; Battiston, F.; Ramseyer, J. P.; Meyer, E.; Andreoli, C.; Brugger, J.; Vettiger, P.; Despont, M.; Mezzacasa, T.; Scandella, L.; Guentherodt, H. J.; Gerber, C.; Gimzewski, J. K. Appl. Phys. A 1998, 66, S61. (12) Rasmussen, P. A.; Grigorov, A. V.; Boisen, A. J. Micromech. Microeng. 2005, 15, 1088. (13) Lu, P.; Lee, H. P.; Lu, C.; O’Shea, S. J. Phys. ReV. B: Condens. Matter 2005, 72, 085405/1. (14) Crandall, S. H.; Dahl, N. C. An Introduction to the Mechanics of Solids; McGraw-Hill, Inc.: New York, 1972. (15) McFarland, A. W.; Poggi, M. A.; Bottomley, L. A.; Colton, J. S. Nanotechnology 2004, 15, 1628.

540 Chemical Reviews, 2008, Vol. 108, No. 2 (16) Poggi, M. A.; McFarland, A. W.; Colton, J. S.; Bottomley, L. A. Anal. Chem. 2005, 77, 1192. (17) Stoney, G. G. Proc. R. Soc. London, Ser. A 1909, 82, 172. (18) McFarland, A. W.; Colton, J. S. J. Micromech. Microeng. 2005, 15, 1060. (19) Sader, J. E. J. Appl. Phys. 2001, 89, 2911. (20) McFarland, A. W. Ph.D. thesis, Georgia Institute of Technology, Atlanta, GA, 2005. (21) Ginsberg, J. Mechanical Structural Vibrations; John Wiley and Sons: New York, 2001. (22) Petersen, K. E.; Guarnieri, C. R. J. Appl. Phys. 1979, 50, 6761. (23) Sandberg, R.; Moelhave, K.; Boisen, A.; Svendsen, W. J. Micromech. Microeng. 2005, 15, 2249. (24) Tamayo, J. J. Appl. Phys. 2005, 97, 044903/1. (25) Blom, F. R.; Bouwstra, S.; Elwenspoek, M.; Fluitman, J. H. J. J. Vac. Sci. Technol., B 1992, 10, 19. (26) Fadel, L.; Dufour, I.; Lochon, F.; Francais, O. Sens. Actuators, B 2004, 102, 73. (27) Tamayo, J.; Humphris, A. D. L.; Owen, R. J.; Miles, M. J. Biophys. J. 2001, 81, 526. (28) Yasumura, K. Y.; Stowe, T. D.; Chow, E. M.; Pfafman, T.; Kenny, T. W.; Stipe, B. C.; Rugar, D. J. Microelectromech. Syst. 2000, 9, 117. (29) Yang, J.; Ono, T.; Esashi, M. Sens. Actuators, A 2000, A82, 102. (30) Lei, F. H.; Nicolas, J. L.; Troyon, M.; Sockalingum, G. D.; Rubin, S.; Manfait, M. J. Appl. Phys. 2003, 93, 2236. (31) McFarland, A. W.; Poggi, M. A.; Doyle, M. J.; Bottomley, L. A.; Colton, J. S. Appl. Phys. Lett. 2005, 87, 053505. (32) Searle, G. F. C. Experimental elasticity, a manual for the laboratory; University Press: Cambridge, U.K., 1920. (33) Kaldor, S. K.; Noyan, I. C. Appl. Phys. Lett. 2002, 80, 2284. (34) Timoshenko, S. Vibration Problems in Engineering, fourth ed.; Wiley: New York, 1974. (35) Timoshenko, S. P.; Goodier, J. N. Theory of Elasticity; McGrawHill, Inc.: New York, 1970. (36) Han, S.; Benaroya, H.; Wei, T. J. Sound Vib. 1999, 225, 935. (37) McFarland, A. W.; Colton, J. S. J. Microelectromech. Syst. 2005, 14, 1375. (38) Sader, J. E. J. Appl. Phys. 1998, 84, 64. (39) Sader, J. E.; Larson, I.; Mulvaney, P.; White, L. R. ReV. Sci. Instrum. 1995, 66, 3789. (40) Lindholm, U. S.; Kana, D. D.; Chu, W.-H.; Abramson, H. N. J. Ship Res. 1965, 9, 11. (41) Newell, W. E. Science 1968, 161, 1320. (42) Battiston, F. M.; Ramseyer, J. P.; Lang, H. P.; Baller, M. K.; Gerber, C.; Gimzewski, J. K.; Meyer, E.; Guntherodt, H. J. Sens. Actuators, B 2001, B77, 122. (43) Wachtman, J. B., Jr.; Tefft, W. E.; Lam, D. G. J.; Apstein, C. S. Phys. ReV. B 1961, 122, 1754. (44) Gysin, U.; Rast, S.; Ruff, P.; Meyer, E.; Lee, D. W.; Vettiger, P.; Gerber, C. Phys. ReV. B: Condens. Matter 2004, 69, 045403. (45) Shekhawat, G.; Tark, S. H.; Dravid, V. P. Science 2006, 311, 1592. (46) Bottomley, L. A.; Poggi, M. A.; Shen, S. X. Anal. Chem. 2004, 76, 5685. (47) Alvarez, M.; Tamayo, J. Sens. Actuators, B 2005, B106, 687. (48) Mertens, J.; Alvarez, M.; Tamayo, J. Appl. Phys. Lett. 2005, 87, 234102/1. (49) Baller, M. K.; Lang, H. P.; Fritz, J.; Gerber, C.; Gimzewski, J. K.; Drechsler, U.; Rothuizen, H.; Despont, M.; Vettiger, P.; Battiston, F. M.; Ramseyer, J. P.; Fornaro, P.; Meyer, E.; Guntherodt, H. J. Ultramicroscopy 2000, 82, 1. (50) Lang, H. P.; Berger, R.; Andreoli, C.; Brugger, J.; Despont, M.; Vettiger, P.; Gerber, C.; Gimzewski, J. K.; Ramseyer, J. P.; Meyer, E.; Guntherodt, H. J. Appl. Phys. Lett. 1998, 72, 383. (51) Arntz, Y.; Seelig, J. D.; Lang, H. P.; Zhang, J.; Hunziker, P.; Ramseyer, J. P.; Meyer, E.; Hegner, M.; Gerber, C. Nanotechnology 2003, 14, 86. (52) Archibald, R.; Datskos, P.; Devault, G.; Lamberti, V.; Lavrik, N.; Noid, D.; Sepaniak, M.; Dutta, P. Anal. Chim. Acta 2007, 584, 101. (53) Backmann, N.; Zahnd, C.; Huber, F.; Bietsch, A.; Pluckthun, A.; Lang, H. P.; Guntherodt, H. J.; Hegner, M.; Gerber, C. Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 14587. (54) Huber, F.; Hegner, M.; Gerber, C.; Guntherodt, H. J.; Lang, H. P. Biosens. Bioelectron. 2006, 21, 1599. (55) Carrascosa, L. G.; Moreno, M.; Alvarez, M.; Lechuga, L. M. TrAC Trends Anal. Chem. 2006, 25, 196. (56) Jeon, S.; Thundat, T. Appl. Phys. Lett. 2004, 85, 1083. (57) Lim, S. H.; Raorane, D.; Satyanarayana, S.; Majumdar, A. Sens. Actuators, B 2006, 119, 466. (58) Yue, M.; Lin, H.; Dedrick, D. E.; Satyanarayana, S.; Majumdar, A.; Bedekar, A. S.; Jenkins, J. W.; Sundaram, S. J. Microelectromech. Syst. 2004, 13, 290.

Goeders et al. (59) Helm, M.; Servant, J. J.; Saurenbach, F.; Berger, R. Appl. Phys. Lett. 2005, 87, 064101/1. (60) Raiteri, R.; Grattarola, M.; Butt, H. J.; Skladal, P. Sens. Actuators, B 2001, B79, 115. (61) Nieva, P. M.; McGruer, N. E.; Adams, G. G. J. Micromech. Microeng. 2006, 16, 2618. (62) Reed, J.; Wilkinson, P.; Schmit, J.; Klug, W.; Gimzewski, J. K. Nanotechnology 2006, 17, 3873. (63) Linnemann, R.; Gotszalk, T.; Hadjiiski, L.; Rangelow, I. W. Thin Solid Films 1995, 264, 159. (64) Linnemann, R.; Gotszalk, T.; Rangelow, I. W.; Dumania, P.; Oesterschulze, E. J. Vac. Sci. Technol., B 1996, 14, 856. (65) Britton, C. L.; Jones, R. L.; Oden, P. I.; Hu, Z.; Warmack, R. J.; Smith, S. F.; Bryan, W. L.; Rochelle, J. M. Ultramicroscopy 2000, 82, 17. (66) Forsen, E.; Abadal, G.; Ghatnekar-Nilsson, S.; Teva, J.; Verd, J.; Sandberg, R.; Svendsen, W.; Perez-Murano, F.; Esteve, J.; Figueras, E.; Campabadal, F.; Montelius, L.; Barniol, N.; Boisen, A. Appl. Phys. Lett. 2005, 87. (67) Amirola, J.; Rodiguez, A.; Castaner, L. Proc. SPIE-Int. Soc. Opt. Eng. 2003, 5116, 92. (68) Amirola, J.; Rodriguez, A.; Castaner, L.; Santos, J. P.; Gutierrez, J.; Horrillo, M. C. Sens. Actuators, B 2005, B111-B112, 247. (69) Verd, J.; Abadal, G.; Teva, J.; Gaudo, M. V.; Uranga, A.; Borrise, X.; Campabadal, F.; Esteve, J.; Costa, E. F.; Perez-Murano, F.; Davis, Z. J.; Forsen, E.; Boisen, A.; Barniol, N. J. Microelectromech. Syst. 2005, 14, 508. (70) Chatzandroulis, S.; Tserepi, A.; Goustouridis, D.; Normand, P.; Tsoukalas, D. Microelectron. Eng. 2002, 61-62, 955. (71) Mertens, J.; Finot, E.; Thundat, T.; Fabre, A.; Nadal, M. H.; Eyraud, V.; Bourillot, E. Ultramicroscopy 2003, 97, 119. (72) Plaza, J. A.; Zinoviev, K.; Villanueva, G.; Alvarez, M.; Tamayo, J.; Dominguez, C.; Lechuga, L. M. Appl. Phys. Lett. 2006, 89, 094109/ 1. (73) Hierlemann, A.; Brand, O.; Hagleitner, C.; Baltes, H. Proc. IEEE 2003, 91, 839. (74) Madou, M. Fundamentals of Microfabrication: The Science of Miniaturization, second ed.; CRC Press: Boca Raton, FL, 2002. (75) Jastrzebski, L.; Corboy, J. F.; Soydan, R. J. Electrochem. Soc. 1989, 136, 3506. (76) Jastrzebski, L.; Corboy, J. F.; McGinn, J. T.; Pagliaro, R., Jr. J. Electrochem. Soc. 1983, 130, 1571. (77) Jenkins, N. E.; DeFlores, L. P.; Allen, J.; Ng, T. N.; Garner, S. R.; Kuehn, S.; Dawlaty, J. M.; Marohn, J. A. J. Vac. Sci. Technol., B 2004, 22, 909. (78) Minne, S. C.; Manalis, S. R.; Quate, C. F. Bringing Scanning Probe Microscopy Up to Speed; Kluwer: Boston, MA, 1999. (79) Nesterov, V.; Brand, U. J. Micromech. Microeng. 2006, 16, 1116. (80) Okamoto, H.; Akazaki, T.; Ueki, M.; Yamaguchi, H.; Namatsu, H. Physica E 2006, 32, 512. (81) Yu, X. M.; Thaysen, J.; Hansen, O.; Boisen, A. J. Appl. Phys. 2002, 92, 6296. (82) Zuo, G.; Li, X.; Li, P.; Yang, T.; Wang, Y.; Cheng, Z.; Feng, S. Anal. Chim. Acta 2006, 580, 123. (83) Campbell, G. A.; Mutharasan, R. Biosens. Bioelectron. 2006, 22, 35. (84) Campbell, G. A.; Mutharasan, R. Anal. Chem. 2006, 78, 2328. (85) Shih, W. Y.; Li, X. P.; Gu, H. M.; Shih, W. H.; Aksay, I. A. J. Appl. Phys. 2001, 89, 1497. (86) Detzel, A. J.; Campbell, G. A.; Mutharasan, R. Sens. Actuators, B 2006, B117, 58. (87) Yi, J. W.; Shih, W. Y.; Mutharasan, R.; Shih, W.-H. J. Appl. Phys. 2003, 93, 619. (88) Yi, J. W.; Shih, W. Y.; Shih, W. H. J. Appl. Phys. 2002, 91, 1680. (89) Lee, Y.; Lim, G.; Moon, W. Sens. Actuators, A 2006, A130-A131, 105. (90) Mortet, V.; Haenen, K.; Potmesil, J.; Vanecek, M.; D’Olieslaeger, M. Phys. Status Solidi A 2006, 203, 3185. (91) Davis, Z. J.; Abadal, G.; Helbo, B.; Hansen, O.; Campabadal, F.; Perez-Murano, F.; Esteve, J.; Figueras, E.; Verd, J.; Barniol, N.; Boisen, A. Sens. Actuators, A 2003, A105, 311. (92) Forsen, E.; Nilsson, S. G.; Carlberg, P.; Abadal, G.; Perez-Murano, F.; Esteve, J.; Montserrat, J.; Figueras, E.; Campabadal, F.; Verd, J.; Montelius, L.; Barniol, N.; Boisen, A. Nanotechnology 2004, 15, S628. (93) Jenkins, D. F. L.; Clegg, W. W.; Cattan, E.; Remiens, D. J. Electroceram. 2001, 7, 5. (94) Jiguet, S.; Bertsch, A.; Hofmann, H.; Renaud, P. AdV. Eng. Mater. 2004, 6, 719. (95) Curran, S.; Ajayan, P. M.; Blau, W.; Carroll, D. L.; Coleman, J. N.; Dalton, A.; Davey, A. P.; McCarthy, B. AdV. Mater. 1998, 10, 1091. (96) Calleja, M.; Nordstroem, M.; Alvarez, M.; Tamayo, J.; Lechuga, L. M.; Boisen, A. Ultramicroscopy 2005, 105, 215.

Sensing Chemical Interactions via Mechanical Motion (97) Calleja, M.; Tamayo, J.; Nordstrom, M.; Boisen, A. Appl. Phys. Lett. 2006, 88. (98) Ransley, J. H. T.; Watari, M.; Sukumaran, D.; McKendry, R. A.; Seshia, A. A. Microelectron. Eng. 2006, 83, 1621. (99) Johansson, A.; Calleja, M.; Rasmussen, P. A.; Boisen, A. Sens. Actuators, A 2005, A123-A124, 111. (100) Johansson, A.; Blagoi, G.; Boisen, A. Appl. Phys. Lett. 2006, 89, 173505. (101) McFarland, A. W.; Poggi, M. A.; Bottomley, L. A.; Colton, J. S. Nanotechnology 2005, 16, 1249. (102) Hu, Z.; Thundat, T.; Warmack, R. J. Proc. Electrochem. Soc. 1999, 99-23, 347. (103) Rogers, B.; Bauer, C. A.; Adams, J. D. MEM Syst. 2003, 5, 663. (104) Rogers, B.; Manning, L.; Jones, M.; Sulchek, T.; Murray, K.; Beneschott, B.; Adams, J. D.; Hu, Z.; Thundat, T.; Cavazos, H.; Minne, S. C. ReV. Sci. Instrum. 2003, 74, 4899. (105) Chang, C. P.; Wang, Y. S.; Huang, R. S. Jpn. J. Appl. Phys., Part 1 2004, 43, 5491. (106) Leichle, T.; Silvan, M. M.; Belaubre, P.; Valsesia, A.; Ceccone, G.; Rossi, F.; Saya, D.; Pourciel, J.-B.; Nicu, L.; Bergaud, C. Nanotechnology 2005, 16, 525. (107) Fagan, B. C.; Tipple, C. A.; Xue, Z. L.; Sepaniak, M. J.; Datskos, P. G. Talanta 2000, 53, 599. (108) Dutta, P.; Senesac, L. R.; Lavrik, N. V.; Datskos, P. G.; Sepaniak, M. J. Sens. Lett. 2004, 2, 238. (109) Headrick, J. J.; Sepaniak, M. J.; Lavrik, N. V.; Datskos, P. G. Ultramicroscopy 2003, 97, 417. (110) Tipple, C. A.; Lavrik, N. V.; Culha, M.; Headrick, J.; Datskos, P.; Sepaniak, M. J. Anal. Chem. 2002, 74, 3118. (111) Lavrik, N. V.; Tipple, C. A.; Datskos, P. G.; Sepaniak, M. J. Proc. SPIE-Int. Soc. Opt. Eng. 2001, 4560, 152. (112) Hilt, J. Z.; Gupta, A. K.; Bashir, R.; Peppas, N. A. Mater. Res. Soc. Symp. Proc. 2002, 729, 173. (113) Bashir, R.; Hilt, J. Z.; Elibol, O.; Gupta, A.; Peppas, N. A. Appl. Phys. Lett. 2002, 81, 3091. (114) Hilt, J. Z.; Gupta, A. K.; Bashir, R.; Peppas, N. A. Biomed. MicrodeV. 2003, 5, 177. (115) Zhang, Y.; Ji, H.-F.; Snow, D.; Sterling, R.; Brown, G. M. Instrum. Sci. Technol. 2004, 32, 361. (116) Liu, K.; Ji, H. F. Anal. Sci. 2004, 20, 9. (117) Zhang, Y. F.; Ji, H. F.; Brown, G. M.; Thundat, T. Anal. Chem. 2003, 75, 4773. (118) Zhang, Y. F.; Ji, H. F.; Snow, D.; Sterling, R.; Brown, G. M. Instrum. Sci. Technol. 2004, 32, 361. (119) Betts, T. A.; Tipple, C. A.; Sepaniak, M. J.; Datskos, P. G. Anal. Chim. Acta 2000, 422, 89. (120) Sayer, M.; Barrow, D.; Zou, L.; Kumar, C. V. R. V.; Noteboom, R.; Knapik, D. A.; Schindel, D. W.; Hutchins, D. A. Mater. Res. Soc. Symp. Proc. 1993, 310, 37. (121) Chapman, P. J.; Vogt, F.; Dutta, P.; Datskos, P. G.; Devault, G. L.; Sepaniak, M. J. Anal. Chem. 2007, 79, 364. (122) Senesac, L. R.; Dutta, P.; Datskos, P. G.; Sepaniak, M. J. Anal. Chim. Acta 2006, 558, 94. (123) Dutta, P.; Chapman, P. J.; Datskos, P. G.; Sepaniak, M. J. Anal. Chem. 2005, 77, 6601. (124) Porter, T. L.; Eastman, M. P.; Pace, D. L.; Bradley, M. Sens. Actuators, A 2001, A88, 47. (125) Fritz, J.; Baller, M. K.; Lang, H. P.; Rothuizen, H. V. P.; Meyer, E.; Guntherodt, H.-J.; Gerber, C.; Gimzewski, J. K. Science 2000, 288, 316. (126) Zhang, J.; Lang, H. P.; Huber, F.; Bietsch, A.; Grange, W.; Certa, U.; McKendry, R.; Guntherodt, H. J.; Hegner, M.; Gerber, C. Nature Nanotechnol. 2006, 1, 214. (127) McKendry, R.; Zhang, J.; Arntz, Y.; Strunz, T.; Hegner, M.; Lang, H. P.; Baller, M. K.; Certa, U.; Meyer, E.; Guntherodt, H.-J.; Gerber, C. Proc. Natl. Acad. Sci. U.S.A. 2002, 99, 9783. (128) Mukhopadhyay, R.; Lorentzen, M.; Kjems, J.; Besenbacher, F. Langmuir 2005, 21, 8400. (129) Bietsch, A.; Zhang, J. Y.; Hegner, M.; Lang, H. P.; Gerber, C. Nanotechnology 2004, 15, 873. (130) Dhayal, B.; Henne, W. A.; Doorneweerd, D. D.; Reifenberger, R. G.; Low, P. S. J. Am. Chem. Soc. 2006, 128, 3716. (131) Belaubre, P.; Guirardel, M.; Garcia, G.; Pourciel, J. B.; Leberre, V.; Dagkessamanskaia, A.; Trevisiol, E.; Francois, J. M.; Bergaud, C. Appl. Phys. Lett. 2003, 82, 3122. (132) Wu, S.-Y.; Berkenbosch, R.; Lui, A.; Green, J.-B. D. Analyst (Cambridge, U. K.) 2006, 131, 1213. (133) Zou, J.; Bullen, D.; Wang, X.; Liu, C.; Mirkin, C. A. Appl. Phys. Lett. 2003, 83, 581. (134) Banerjee, D. BioMEMS Biomed. Nanotechnol. 2006, 1, 265. (135) Chisholm, R. A.; Qiu, W.; Green, J.-B. D. Langmuir 2007, 23, 7891. (136) Green, J.-B. D. Anal. Chim. Acta 2003, 496, 267.

Chemical Reviews, 2008, Vol. 108, No. 2 541 (137) Thundat, T.; Oden, P. I.; Warmack, R. J. Microscale Thermophys. Eng. 1997, 1, 185. (138) Thundat, T.; Oden, P. I.; Warmack, R. J. Proc. Electrochem. Soc. 1997, 97-5, 179. (139) Lang, H. P.; Hegner, M.; Meyer, E.; Gerber, C. Nanotechnology 2002, 13, R29. (140) Raiteri, R.; Grattarola, M.; Berger, R. Materials Today (Oxford, U. K.) 2002, 5, 22. (141) Datskos, P. G.; Lavrik, N. V.; Sepaniak, M. J. NATO Sci. Ser., II 2004, 181, 331. (142) Datskos, P. G.; Lavrik, N. V.; Sepaniak, M. J. Intro. Nanoscale Sci. Technol. 2004, 417. (143) Datskos, P. G.; Thundat, T.; Lavrik, N. V. Encycl. Nanosci. Nanotechnol. 2004, 5, 551. (144) Lavrik, N. V.; Sepaniak, M. J.; Datskos, P. G. ReV. Sci. Instrum. 2004, 75, 2229. (145) Lang, H. P.; Baller, M. K.; Berger, R.; Gerber, C.; Gimzewski, J. K.; Battiston, F. M.; Fornaro, P.; Ramseyer, J. P.; Meyer, E.; Guntherodt, H. J. Anal. Chim. Acta 1999, 393, 59. (146) Maute, M.; Raible, S.; Prins, F. E.; Kern, D. P.; Ulmer, H.; Weimar, U.; Gopel, W. Sens. Actuators, B 1999, B58, 505. (147) Maute, M.; Raible, S.; Prins, F. E.; Kern, D. P.; Weimar, U.; Gopel, W. Microelectron. Eng. 1999, 46, 439. (148) Kim, B. H.; Maute, M.; Prins, F. E.; Kern, D. P.; Croitoru, M.; Raible, S.; Weimar, U.; Gopel, W. Microelectron. Eng. 2000, 53, 229. (149) Kim, B. H.; Prins, F. E.; Kem, D. P.; Raible, S.; Weimar, U. Sens. Actuators, B 2001, 78, 12. (150) Lange, D.; Hagleitner, C.; Hierlemann, A.; Brand, O.; Baltes, H. Anal. Chem. 2002, 74, 3084. (151) Kurzawski, P.; Hagleitner, C.; Hierlemann, A. Anal. Chem. 2006, 78, 6910. (152) Fadel, L.; Lochon, F.; Dufour, I.; Francais, O. J. Micromech. Microeng. 2004, 14, S23. (153) Vidic, A.; Then, D.; Ziegler, C. Ultramicroscopy 2003, 97, 407. (154) Lechuga, L. M.; Tamayo, J.; Calle, A.; Calleja, M.; Dominquez, C. NATO Sec. Sci. Ser., Ser. B 2005, 1, 175. (155) Pinnaduwage, L. A.; Gehl, A.; Hedden, D. L.; Muralidharan, G.; Thundat, T.; Lareau, R. T.; Sulchek, T.; Manning, L.; Rogers, B.; Jones, M.; Adams, J. D. Nature (London, U. K.) 2003, 425, 474. (156) Pinnaduwage, L. A.; Ji, H.-F.; Thundat, T. IEEE Sens. J. 2005, 5, 774. (157) Pinnaduwage, L. A.; Wig, A.; Hedden, D. L.; Gehl, A.; Yi, D.; Thundat, T.; Lareau, R. T. J. Appl. Phys. 2004, 95, 5871. (158) Ji, H. F.; Yan, X.; Lu, Y.; Du, H.; Thundat, T. Proc. SPIE-Int. Soc. Opt. Eng. 2006, 6223, 622307/1. (159) Ji, H. F.; Yan, X. D.; Zhang, J.; Thundat, T. Expert ReV. Mol. Diagn. 2004, 4, 859. (160) Pinnaduwage, L. A.; Gehl, A. C.; Allman, S. L.; Johansson, A.; Boisen, A. ReV. Sci. Instrum. 2007, 78, 055101. (161) Voiculescu, I.; Zaghloul, M. E.; McGill, R. A.; Houser, E. J.; Fedder, G. K. IEEE Sens. J. 2005, 5, 641. (162) Voiculescu, I. R.; Zaghloul, M. E.; McGill, R. A.; Vignola, J. F. Proc. Inst. Mech. Eng., Part C 2006, 220, 1601. (163) Zuo, G. M.; Li, X. X.; Li, P.; Yang, T. T.; Wang, Y. L.; Cheng, Z. X.; Feng, S. L. Anal. Chim. Acta 2006, 580, 123. (164) Yang, Y.; Ji, H.-F.; Thundat, T. J. Am. Chem. Soc. 2003, 125, 1124. (165) Karnati, C.; Dua, H.; Ji, H.-F.; Xua, X.; Lvov, Y.; Mulchandani, A.; Mulchandani, P.; Chen, W. Biosens. Bioelectron. 2007, 22, 2636. (166) Pamula, V. K.; Fair, R. B. Proc. SPIE-Int. Soc. Opt. Eng. 1999, 3710, 321. (167) Pamula, V. K.; Fair, R. B. Proc. SPIE-Int. Soc. Opt. Eng. 2000, 4038, 547. (168) Thundat, T.; Pinnaduwage, L.; Lareau, R. NATO Sci. Ser., II 2004, 159, 249. (169) Rogers, B.; Whitten, R.; Adams, J. D. Proc. SPIE-Int. Soc. Opt. Eng. 2006, 6201, 62010G/1. (170) Rogers, B.; Whitten, R.; Adams, J. D. Proc. SPIE-Int. Soc. Opt. Eng. 2005, 5986, 59860O/1. (171) Rogers, B.; Whitten, R.; Adams, J. D. Proc. SPIE-Int. Soc. Opt. Eng. 2006, 6394, 639409/1. (172) Xu, X. H.; Thundat, T. G.; Brown, G. M.; Ji, H. F. Anal. Chem. 2002, 74, 3611. (173) Ji, H.-F.; Thundat, T. Biosens. Bioelectron. 2002, 17, 337. (174) Ji, H.-F.; Thundat, T.; Dabestani, R.; Brown, G. M.; Britt, P. F.; Bonnesen, P. V. Anal. Chem. 2001, 73, 1572. (175) Ji, H.-F.; Finot, E.; Dabestani, R.; Brown, G. M.; Britt, P. F. Chem. Commun. (Cambridge) 2000, 457. (176) Ilic, B.; Czaplewski, D.; Craighead, H. G.; Neuzil, P.; Campagnolo, C.; Batt, C. Appl. Phys. Lett. 2000, 77, 450. (177) Zhang, J.; Ji, H. F. Anal. Sci. 2004, 20, 585. (178) Campbell, G. A.; Mutharasan, R. Biosens. Bioelectron. 2005, 21, 462. (179) Campbell, G. A.; Mutharasan, R. Anal. Sci. 2005, 21, 355.

542 Chemical Reviews, 2008, Vol. 108, No. 2 (180) Gfeller, K. Y.; Nugaeva, N.; Hegner, M. Appl. EnViron. Microbiol. 2005, 71, 2626. (181) Gfeller, K. Y.; Nugaeva, N.; Hegner, M. Biosens. Bioelectron. 2005, 21, 528. (182) Ramos, D.; Tamayo, J.; Mertens, J.; Calleja, M.; Zaballos, A. J. Appl. Phys. 2006, 100. (183) Ilic, B.; Yang, Y.; Craighead, H. G. Appl. Phys. Lett. 2004, 85, 2604. (184) Gupta, A.; Akin, D.; Bashir, R. Appl. Phys. Lett. 2004, 84, 1976. (185) Johnson, L.; Gupta, A. K.; Ghafoor, A.; Akin, D.; Bashir, R. Sens. Actuators, B 2006, B115, 189. (186) Campbell, G. A.; Mutharasan, R. Biosens. Bioelectron. 2006, 22, 78. (187) Campbell, G. A.; Mutharasan, R. Biosens. Bioelectron. 2006, 21, 1684. (188) Campbell, G. A.; Mutharasan, R. Anal. Chem. 2007, 79, 1145. (189) Nugaeva, N.; Gfeller, K. Y.; Backmann, N.; Lang, H. P.; Dueggelin, M.; Hegner, M. Biosens. Bioelectron. 2005, 21, 849. (190) Grogan, C.; Raiteri, R.; O’Connor, G. M.; Glynn, T. J.; Cunningham, V.; Kane, M.; Charlton, M.; Leech, D. Biosens. Bioelectron. 2002, 17, 201. (191) Alvarez, M.; Calle, A.; Tamayo, J.; Lechuga, L. M.; Abad, A.; Montoya, A. Biosens. Bioelectron. 2003, 18, 649. (192) Backmann, N.; Zahnd, C.; Huber, F.; Bietsch, A.; Pluckthun, A.; Lang, H.-P.; Guntherodt, H.-J.; Hegner, M.; Gerber, C. Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 14587. (193) Dutta, P.; Tipple, C. A.; Lavrik, N. V.; Datskos, P. G.; Hofstetter, H.; Hofstetter, O.; Sepaniak, M. J. Anal. Chem. 2003, 75, 2342. (194) Hwang, K. S.; Lee, J. H.; Park, J.; Yoon, D. S.; Park, J. H.; Kim, T. S. Lab Chip 2004, 4, 547. (195) Lee, J. H.; Hwang, K. S.; Park, J.; Yoon, K. H.; Yoon, D. S.; Kim, T. S. Biosens. Bioelectron. 2005, 20, 2157. (196) Wee, K. W.; Kang, G. Y.; Park, J.; Kang, J. Y.; Yoon, D. S.; Park, J. H.; Kim, T. S. Chem. Sens. 2004, 20, 216. (197) Wee, K. W.; Kang, G. Y.; Park, J.; Kang, J. Y.; Yoon, D. S.; Park, J. H.; Kim, T. S. Biosens. Bioelectron. 2005, 20, 1932. (198) Lee, J. H.; Yoon, K. H.; Hwang, K. S.; Park, J.; Ahn, S.; Kim, T. S. Biosens. Bioelectron. 2004, 20, 269. (199) Lee, J. H.; Kim, T. S.; Yoon, K. H. Appl. Phys. Lett. 2004, 84, 3187. (200) Kang, G. Y.; Han, G. Y.; Kang, J. Y.; Cho, I.-H.; Park, H.-H.; Paek, S.-H.; Kim, T. S. Sens. Actuators, B 2006, B117, 332. (201) Wu, G.; Datar, R. H.; Hansen, K. M.; Thundat, T.; Cote, R. J.; Majumdar, A. Nat. Biotechnol. 2001, 19, 856. (202) Hansen, K. M.; Ji, H. F.; Wu, G. H.; Datar, R.; Cote, R.; Majumdar, A.; Thundat, T. Anal. Chem. 2001, 73, 1567. (203) Wu, G. H.; Ji, H. F.; Hansen, K. M.; Thundat, T.; Datar, R.; Cote, R.; Hagan, M. F.; Chakraborty, A. K.; Majumdar, A. Proc. Natl. Acad. Sci. U.S.A. 2001, 98, 1560.

Goeders et al. (204) Hagan, M. F.; Majumdar, A.; Chakraborty, A. K. J. Phys. Chem. B 2002, 106, 10163. (205) Alvarez, M.; Carrascosa, L. G.; Moreno, M.; Calle, A.; Zaballos, A.; Lechuga, L. M.; Martinez, A. C.; Tamayo, J. Langmuir 2004, 20, 9663. (206) Stachowiak, J. C.; Yue, M.; Castelino, K.; Chakraborty, A.; Majumdar, A. Langmuir 2006, 22, 263. (207) Biswal, S. L.; Raorane, D.; Chaiken, A.; Birecki, H.; Majumdar, A. Anal. Chem. 2006, 78, 7104. (208) Biswal, S. L.; Raorane, D.; Chaiken, A.; Majumdar, A. Jala 2006, 11, 222. (209) Marie, R.; Thaysen, J.; Christensen, C. B. V.; Boisen, A. Spec. Publ.s R. Soc. Chem. 2004, 297, 485. (210) Gunter, R. L.; Zhine, R.; Delinger, W. G.; Manygoats, K.; Kooser, A.; Porter, T. L. IEEE Sens. J. 2004, 4, 430. (211) Zhang, X. R.; Xu, X. F. Appl. Phys. Lett. 2004, 85, 2423. (212) Su, M.; Li, S. U.; Dravid, V. P. Appl. Phys. Lett. 2003, 82, 3562. (213) Subramanian, A.; Oden, P. I.; Kennel, S. J.; Jacobson, K. B.; Warmack, R. J.; Thundat, T.; Doktycz, M. J. Appl. Phys. Lett. 2002, 81, 385. (214) Yan, X. D.; Ji, H. F.; Lvov, Y. Chem. Phys. Lett. 2004, 396, 34. (215) Yan, X. D.; Xu, X. H. K.; Ji, H. F. Anal. Chem. 2005, 77, 6197. (216) Pei, J. H.; Tian, F.; Thundat, T. Anal. Chem. 2004, 76, 292. (217) Bottomley, L. A.; Ghosh, M.; Shen, S.; Saul, R.; Kossek, S.; Pace, G. W. U. S. Patent 7,141,385, 2006. (218) Stevenson, K. A.; Mehta, A.; Sachenko, P.; Hansen, K. M.; Thundat, T. Langmuir 2002, 18, 8732. (219) Liu, W.; Montana, V.; Chapman, E. R.; Mohideen, U.; Parpura, V. Proc. Natl. Acad. Sci. U.S.A. 2003, 100, 13621. (220) Weizmann, Y.; Elnathan, R.; Lioubashevski, O.; Willner, I. Nano Lett. 2005, 5, 741. (221) Weizmann, Y.; Elnathan, R.; Lioubashevski, O.; Willner, I. J. Am. Chem. Soc. 2005, 127, 12666. (222) Barnes, J. R.; Stephenson, R. J.; Welland, M. E.; Gerber, C.; Gimzewski, J. K. Nature 1994, 372, 79. (223) Barnes, J. R.; Stephenson, R. J.; Woodburn, C. N.; O’Shea, S. J.; Welland, M. E.; Rayment, T.; Gimzewski, J. K.; Gerber, C. ReV. Sci. Instrum. 1994, 65, 3793. (224) Burg, T. P.; Godin, M.; Knudsen, S. M.; Shen, W.; Carlson, G.; Foster, J. S.; Babcock, K.; Manalis, S. R. Nature 2007, 446, 1066. (225) Burg, T. P.; Manalis, S. R. Appl. Phys. Lett. 2003, 83, 2698. (226) Burg, T. P.; Mirza, A. R.; Milovic, N.; Tsau, C. H.; Popescu, G. A.; Foster, J. S.; Manalis, S. R. J. Microelectromech. Syst. 2006, 15, 1466.

CR0681041

Chem. Rev. 2008, 108, 543−562

543

Radionuclide Sensors for Environmental Monitoring: From Flow Injection Solid-Phase Absorptiometry to Equilibration-Based Preconcentrating Minicolumn Sensors with Radiometric Detection Jay W. Grate,*,† Oleg B. Egorov,†,§ Matthew J. O’Hara,† and Timothy A. DeVol‡ Pacific Northwest National Laboratory, P.O. Box 999, Richland Washington 99352, and Environmental Engineering and Earth Sciences, Clemson University, 342 Computer Court, Anderson, SC 29625−6510 Received March 7, 2007

Contents 1. Introduction 2. Background 2.1. Flow Injection and Sequential Injection Analysis 2.2. Preconcentrating Minicolumn Sensors 2.3. Bead Injection and Renewable Surface Sensing 2.4. Automated Radiochemical Separation and Analysis 3. Radionuclide Sensors 3.1. Challenges of Radionuclide Sensing in Water 3.2. Minicolumn Sensors Based on Extractive Scintillating Resins 3.3. Composite Bed Scintillating Minicolumn Sensors 3.4. Sensor Regeneration or Renewal 3.5. Equilibration-Based Sensing 3.6. Chromatographic Theory for Equilibration-Based Sensing 3.7. Engineered Radiometric Preconcentrating Minicolumn Sensors for Groundwater Measurements 3.8. Planar Dual-Functionality Radionuclide Sensors 3.9. Planar Radionuclide Sensors Based on Diodes 3.10. Fiber-Based Sensor 3.11. Whole-Column Chromatographic Sensor 3.12. Dual-Functionality Sensor for Tritiated Water in Air 4. Discussion 5. Acknowledgment 6. References

543 545 545 546 547 547 548 548 549 553 554 555 556 558 558 559 559 559 560 560 561 561

1. Introduction The development of in situ sensors for ultratrace detection applications in process control and environmental monitoring * To whom correspondence should be addressed. E-mail: [email protected]. Telephone: 509-376-4242. Fax: 509-376-5106. † Pacific Northwest National Laboratory. ‡ Clemson University. § Present address: Isoray Medical, Inc., 350 Hills St., Suite 106, Richland, Washington 99354.

remains a significant challenge. Such sensors must meet difficult detection limit requirements while selectively detecting the analyte of interest in complex or otherwise challenging sample matrixes. Nowhere are these requirements more daunting than in the field of radionuclide sensing for R- and β-emitting radionuclides in water. The detection limit requirements can be extremely low. Nevertheless, a promising approach to radionuclide sensing based on preconcentrating minicolumn sensors has been developed. In addition, a method of operating such sensors, which we call equilibration-based sensing, has been developed that provides substantial preconcentration and a signal that is proportional to analyte concentration, while eliminating the need for reagents to regenerate the sorbent medium following each measurement. While this equilibration-based sensing method was developed for radionuclide sensing, it can be applied to nonradioactive species as well, given a suitable on-column detection system. By replacing costly sampling and laboratory analysis procedures, in situ sensors could have a significant impact on monitoring and long-term stewardship applications. The preconcentrating minicolumn sensor relies on a solid phase, typically a packed bed of particles or beads, to collect and concentrate the analyte species of interest within a detector. A portion, or ideally all, of the solid phase is within the detected volume. Typically the solid phase is packed in a small column with fluid flow parallel to the column axis. This can be a straight column or a column that has been coiled to fit within the detection system. Some sensors, however, are prepared with the solid phase in a disk or plate geometry with radial flow from the center to the periphery. Optical or luminescent methods predominate for the detection of analyte species, or their reaction products, captured on the solid phase. The radionuclide sensors described in this review are primarily preconcentrating minicolumn sensors that rely on the detection of scintillation photons from a dualfunctionality column. The column contains selectively sorbent functionality and scintillating properties in the same material, or in materials that are in close proximity to one another. The preconcentrating minicolumn sensor is shown schematically in Figure 1a, along with the radionuclide sensor concept in Figure 1b, where the scintillation photons are detected with a pair of photomultiplier tubes. The fluidic format is an efficient means of collecting analyte species from much larger sample volumes and for concentrating them on the solid phase for on-column detection. This precon-

10.1021/cr068115u CCC: $71.00 © 2008 American Chemical Society Published on Web 01/03/2008

544 Chemical Reviews, 2008, Vol. 108, No. 2

Grate et al. rations, and equilibration sensing. His research interests include radiochemical analysis, selective radionuclide sensing, nuclear waste process monitoring, medical radioisotope production, and laboratory automation.

Jay W. Grate is a Laboratory Fellow at the Pacific Northwest National Laboratory (PNNL) and an Affiliated Professor with the Chemistry Department of the University of Washington. He received a B.A. in Chemistry at Rollins College and received his Ph.D. in Chemistry from the University of California, San Diego. After postdoctoral research at the University of California Irvine, he joined the Naval Research Laboratory in 1984, moving to PNNL in 1992. He spent a sabbatical at the Scripps Research Institute prior to joining PNNL. Dr. Grate’s research has focused on chemically interactive polymers and nanomaterials, chemical vapor sensors, radiochemical separations and sensing, and bioanalytical fluidics for biothreat detection. His work integrates aspects of the chemical sciences, material sciences, and measurement sciences into new microanalytical principles, methods, and systems. His work in the radioanalytical field has entailed the development of new radiochemical analysis methods using sequential injection separations, fully automated radiochemical process monitors, and sensors for radionuclides in water. He has published over 100 papers in peer-reviewed journals and numerous book chapters, and he is author or coauthor on over a dozen patents, several of which have been licensed. He received an R&D 100 Award in 2004 and an American Chemical Society Regional Industrial Innovation Award in 2007.

Prior to joining IsoRay Medical Inc. in 2005 as a Director of Research and Development, Dr. Egorov has worked as a senior research scientist at the Pacific Northwest National Laboratory. He received his Ph.D. in Analytical Chemistry from the University of Washington in 1998. In addition to his Ph.D., Dr. Egorov has an M.S. degree specializing in Radiochemistry from Moscow State University in Moscow, Russia, and an M.S. in Environmental Analytical Chemistry from the University of Washington. His research at PNNL has specialized in microfluidic systems and their application toward automation of radionuclide separations and analysis, where he has authored or co-authored several key publications, including invited review articles and book chapters. Dr. Egorov pioneered the application of flow-injection techniques for automating radiochemical analyses of nuclear wastes and process monitoring, renewable surface sensing and sepa-

Matthew J. O’Hara is a scientist at the Pacific Northwest National Laboratory. He received B.A. degrees in Chemistry and in Geology from the University of Montana in 1996, and a Masters degree in Business Administration from Washington State University in 2004. He has been involved in scientific research in the fields of radioanalytical chemistry and laboratory automation for over 10 years. The primary focus of his research has been the selective preconcentration, separation, and detection of actinides and radioactive fission products from various matrices using automated fluid handling and detection systems. The instrumentation he has developed has targeted specific Rand β-emitting radionuclides with the objective of activity quantification ranging from ultralow activities in environmental waters to high-level activity in nuclear waste samples. His research has resulted in the development of medical isotope separation systems, sensors for groundwater monitoring, and prototype process monitors for Hanford’s nuclear waste treatment plant.

Timothy A. DeVol is a Professor of Environmental Engineering and Earth Science at Clemson University. Dr. DeVol earned a B.S. in engineering physics from The Ohio State University, and a M.S. and Ph.D. in nuclear engineering from the University of Michigan. Dr. DeVol has been teaching and conducting research at Clemson University since 1993. He teaches courses in radiation and health physics, ionizing radiation detection, and radioactive waste management. His major research interests are in the detection and measurement of ionizing radioactivity in the environment, environmental radiochemistry, and statistical analysis of monitoring data. Dr. DeVol has published over 40 papers in peer-reviewed journals and has made over 130 scientific presentations.

Radionuclide Sensors for Environmental Monitoring

Chemical Reviews, 2008, Vol. 108, No. 2 545

Throughout the narrative, we will focus on sensors for Tc as the prototypical examples for illustrating the detection principles. These are the most mature radionuclide sensors to date, and 99Tc is an important radionuclide to detect in environmental monitoring. It is generated from the thermal fission of 235U with a high production yield of 6% and is a significant radioactive contaminant at U.S. Department of Energy sites associated with nuclear weapons production. It has a long radioactive half-life of 2.13 × 105 years, and it is highly mobile in the environment in the Tc(VII) oxidation state as the pertechnetate oxyanion, TcO4-. Hence, this contaminant will persist in the environment, and it must be monitored as it is unlikely to stay in one place. As a portion of the background material, we will cite selected material on preconcentrating minicolumn sensors using transduction mechanisms other than radioactivity to detect analytes ranging from metal ions to organic pharmaceuticals and nutrients. This work has largely occurred from 1985 to the present. These types of fluidic sensors were developed within the fields of flow injection and sequential injection analysis; therefore, these topics will be introduced in the background material. In addition, the method of “bead injection” has been developed where the sorbent material is delivered to the detection flow cell for each measurement and then released. Bead injection represents a renewable surface preconcentrating sensor. Before turning our attention entirely to radionuclide sensors for water monitoring, we will also provide some background on automated radiochemical analysis. Finally, this review is not concerned with assays that use radionuclides as labels, such as the scintillation proximity assay (SPA).19-21 This method is used for studies of binding interactions of biologically relevant compounds, using scintillating microspheres and radiolabeled molecules of high specific activity. It is designed to discriminate between bound and unbound molecules. Although this assay combines chemical selectivity with scintillation, its purpose is not focused on environmental radiochemical analysis applications. 99

Figure 1. Schematic diagrams for (a) preconcentrating minicolumn sensors and (b) radionuclide sensors based on dual-functionality materials in preconcentrating minicolumn sensors. The column may also be a coil within the detection zone.

centration increases sensitivity and reduces detection limits. In addition, concentrating the analyte in a smaller volume can simplify and reduce the size of the detection method used. These features are all desirable for environmental sensors, where analytes are typically present at very low concentrations and the sensor should ideally be suitable for at site or in situ deployment. In typical use, the fluidic system containing the sensor processes a sample aliquot of defined volume, and the analytical signal is taken within a defined time range of the process. In this regard, the sensor may be regarded as part of an assay system that determines the quantity of the analyte in that particular sample volume. However, as we shall illustrate with the equilibration-based radionuclide sensor to be described below, such a sensor can also function as a true sensor whose signal goes up and down with the ambient concentration. As long as the sensor is allowed to reach equilibrium, the signal is not dependent on the volume of the sample processed through the flow cell; it depends on the concentration of the analyte and its interaction with the solid phase. The aim of this review is to cover radionuclide sensors for R- and β-emitting radionuclides that combine some form of selective sorption with a radiometric detection method andsas a primary aimsto comprehensively review preconcentrating minicolumn sensors for radionuclide detection. This work that has largely occurred from 1995 to the present.1-18 As a secondary aim, we will cover radionuclide sensors that combine sorption and scintillation in formats other than minicolumn sensors. We are particularly concerned with the detection of R- and β-emitting radionuclides in liquids, which presents particular challenges as we shall describe below. We will not cover systems to detect γ rays or the radionuclides that emit them, since γ rays can readily pass through condensed media to radiation detectors and the γ-ray energy spectrum provides considerable selectivity. Nonetheless, preconcentrating sensor methodologies to be described in this paper can also offer advantages for lowlevel sensing of γ-ray emitters where preconcentration is required.

2. Background 2.1. Flow Injection and Sequential Injection Analysis A large number of sensors fitting our definition of preconcentrating minicolumn sensors were developed as detectors for flow injection analysis systems.22-24 Flow injection consists of a fluidic analysis approach where sample and reagent solutions are driven in a continuous forward flow paradigm through a progression of mixing, reaction, and/or separation steps to a flow-through detector. The prototypical assay was colorimetric with an optical absorbance detector. Figure 2a provides a schematic diagram of a flow injection system. The original fluid drive was typically one or more peristaltic pumps coupled with at least one valve for sample injection. By incorporating an appropriate solid phase within the detector, creating an optosensor, the analyte or reaction products could be captured, preconcentrated, and focused within the detection zone for greater sensitivity. Sequential injection arose as a subsequent approach to flow-based analysis; a schematic diagram is shown in Figure 2b.25-28 Sequential injection relies on programmed bidirectional flow. The preferred fluid drive is a digital syringe pump coupled to a multiposition valve with a holding coil

546 Chemical Reviews, 2008, Vol. 108, No. 2

Grate et al.

Figure 2. Schematic diagrams of prototypical (a) flow injection analysis, (b) sequential injection analysis, and (c) sequential injection separation systems.

in between. In a typical analysis, samples and reagents are pulled as zones into the holding coil using reverse flow, stepping the multiposition valve from one position to another for each solution. These zones are then propelled forward though the multiposition valve to the analysis system consisting of a reactor or separator and detector. In a simple colorimetric assay, the zones intermix by dispersion, generating reaction products for detection. Sequential injection systems are fully automated with a computer providing precise control of volumes, flow rates, and timing. The detector in some examples has been a preconcentrating minicolumn optosensor. Because sequential injection systems provide such a versatile system for fluid handling, and they scale well for handling milliliter size to microliter size samples, they have been used in a great variety of analytical approaches beyond the simple example of mixing, reaction, and detection just given. Sequential injection separations (Figure 2c) and bead injection represent two prominent examples, which will be described in more detail below. The use of solid phases in sequential injection systems has recently been reviewed.29

2.2. Preconcentrating Minicolumn Sensors Early work on preconcentrating minicolumn sensors in flow-based analysis was reviewed in 1993.30 Another review on such sensors appeared in 2004.31 Our own surveys indicate there are over 100 papers on such sensors. Generally, these sensors are columns containing sorbent solid phases and fit the general idea shown in Figure 1a, with all or part of the column in the detection zone. In the fluidic systems shown in Figure 2, a and b, the sensor serves as the detector. Many such sensors are spectrophotometric, measuring absorbance of the packed bed in the visible or UV wavelengths. Accordingly, they have been described using terms such as optosensors, optosensing, solid-phase absorptiometry, and ion-exchanger phase absorptiometry. The flow cell may be a modified cuvette fitting in a conventional spectrophotometer, or it may be a flow cell configured with fiber optics. Luminescent methods, such as fluorescence, phosphorescence, and chemiluminescence, and luminescent methods using energy transfer processes have also been widely employed. Even photoacoustic detection has been adapted to flow cells containing sorptive solid phases.32 The solid phases are typically ion exchangers, ligandloaded complexing resins, or hydrophobic phases such as C18-modified silica. Molecularly imprinted polymers have also been used. Ligand-loaded complexing resins have been reviewed.33 The detected analyte may be the “native” ionic

or molecular species, a reaction product or complex formed upstream of the flow cell, or such products formed upon interaction with the solid-phase material. The range of species detected using these methods are extremely broad, including metal ions (including the rare earths), complexes, inorganic anions such as iodide and phosphate, pharmaceuticals and metabolites, nutrients and other food components, oxygen, and aromatic hydrocarbons and phenolics. Typically, the analyte or analyte reaction product from a certain volume of sample is captured on the solid phase, measured, and then released using a suitable reagent after the completion of the measurement. In some cases the interaction is weak enough that the species migrates down the column and is detected as it traverses the optical path. For example, Yoshimura used an ion-exchange resin in an optosensor flow cell to capture copper ions from a 0.17-mL injected sample volume.34 Using a 0.014 M nitric acid concentration, where the distribution ratio was D ) 62000 [mol of copper sorbed/kg of resin]/[mol of copper/L of solution], the copper ions were completely retained and were detected with a spectrophotometer at 800 nm. The sensor was regenerated by perfusion with 2 M nitric acid solution. On the other hand, if the copper ions were captured from 0.28 M nitric acid, where the distribution coefficient was D ) 340, the ions were eluted in a “fairly short time” in additional carrier solution. Recently, an alternative methodology for operation of a preconcentrating minicolumn sensor with optical absorption detection has been described.18 Hexavalent chromium ions were accumulated in an equilibration-based sensing approach, where the entire bed of the anion exchange column sensor was equilibrated with the analyte in the sample by flowing an excess of sample through the column. Once the column was fully equilibrated, the entering and exiting chromium ion concentrations were the same. At trace concentrations (e.g., on the linear portion of a sorption isotherm), the amount retained on the column at equilibration is proportional to the sample concentration. Because it is dynamic equilibrium, pumping a sufficient volume of a blank solution through the column will eventually elute the analyte. This approach can be regarded as a true sensor whose response can go up and down with analyte concentration, rather than representing an assay on a specific volume of sample solution. Furthermore, reagents are not required to regenerate the sensor. The concept of equilibration-based preconcentrating minicolumn sensors will be described in more detail below for sensing the radionuclide 99Tc (as pertechnetate, 99TcO4-).

Radionuclide Sensors for Environmental Monitoring

2.3. Bead Injection and Renewable Surface Sensing As an alternative to methods that elute the analyte from the solid phase, as just described, a methodology has been developed to automatically provide a fresh solid phase for each measurement. Renewable surface sensing using solid phases has also been dubbed “bead injection”.28,35-40 In this approach, the solid phase is again within the observed region of a detector, and most detection methods are optical or microscopic. However, fluidic procedures and specialized flow cells have been developed so that the solid phase is delivered as a liquid suspension to the flow cell, captured within the flow cell, perfused with the sample for interaction and measurement, and then removed or released from the flow cell, all under computer control. In this way, a fresh solid phase with a new surface can be provided for each sample measurement, hence the phrase “renewable surface sensing”. These approaches can be used in both separation and sensing for a variety of species; however, they have been particularly useful in bioanalytical measurements where sensitive biochemical interfaces are involved. Nevertheless, renewable surface separations41,42 and sensing3 have both been described in the field of radiochemistry as well. A variety of flow cells have been designed for implementing renewable surface techniques, including a “jet ring cell” with a moveable tubing end in contact with a transparent plate,35,38,40 a machined flow cell with a moveable solid rod intersecting the flow channel,35,38,43 and a rotating rod design where the angled end of a solid rod intersects an angular flow path in one position but allows beads to pass when rotated 180°.44,45 Methods that can direct the flow and beads toward one channel with a frit or another channel without a frit have also been developed.41,42,46 These will not be described in detail here. Some have been reviewed together in connection with nucleic acid-based analyses.46

2.4. Automated Radiochemical Separation and Analysis Radiochemical analysis is concerned with the determination of radionuclides from a variety of sample matrixes. If the radionuclide of interest cannot be determined nondestructively by detection of a γ-ray emission and identification from the γ-ray energy spectrum, then chemical separations are normally a necessary and critical aspect of the analysis. The radionuclides of interest must be separated from the sample matrix and concentrated for determination by either radiometric or mass spectrometric techniques. The classical methods for performing such separations, including precipitation, solvent extraction, and manual ion exchange, are tedious and time-consuming. Significant advances in separation materials for columnbased separations have simplified radiochemical analysis. At the same time, fluidic and in some cases robotic methods have been developed to automate column-based radiochemical separation and detection. The fluidic automation methods have been based on flow injection and sequential injection methods as described briefly above. In particular, the coupling of sequential injection fluidics to small separation columns lead to “sequential injection separations”, as shown in Figure 2c. In this approach, the sequential injection fluidics provide fluid handling to deliver samples, reagents, and eluants; the column provides selective separations based on the sorbent material in the column; and the analytes that are separated from the matrix and eluted from the column are

Chemical Reviews, 2008, Vol. 108, No. 2 547 Scheme 1

detected downstream. Typically, the separation is based on using a separation chemistry where the radionuclide(s) are retained under the sample load conditions, the sample matrix and unretained radionuclides are removed in a wash step, and then the retained radionuclide(s) are released in one or more steps using a change in solution conditions that greatly reduces the affinity of the radionuclide for the separation material. These separations can rapidly separate individual radionuclides or groups and tolerate significant sample loading. The separation materials for automated radiochemical separations can be conventional ion-exchange resins or more recent extraction chromatography or solid-phase extraction materials. These same separation materials can be used in the development of radionuclide sensors to be described below. Extraction chromatographic materials47 for radionuclide separations, using selective or semiselective extractants impregnated on macroreticular polymer supports, have been developed by Horwitz and co-workers and commercialized by Eichrom Technnologies, Inc.48-52 The uptake properties and chemical selectivities of these materials are well characterized in the literature, and hence the selectivities of sensors can be rationally designed and understood. The chemical structures of the extractants in some of these resins are shown in Scheme 1. A variety of solid-phase extraction materials dubbed “SuperLig” have been developed using “molecular recognition” ligands on solid supports and commercialized by IBC Advanced Technologies (American Fork, Utah).53-56 These ligands are covalently bound to various polymeric or silica gel supports. Automated sequential injection separation can be illustrated with data for 99Tc analysis, the same radionuclide we will use to illustrate the principles of radionuclide sensors. A system similar to that in Figure 2c was set up with a separation column containing the extraction chromatographic resin known as TEVA-resin.9,42,57 TEVA-resin is a macroreticular polymer support impregnated with Aliquat 336, a liquid anion exchanger consisting of a mixture of long-chain quaternary ammonium ions. This resin is known to retain 99 Tc(VII) as pertechnetate under neutral to weakly acidic conditions and to release it in strongly acidic conditions.50 As shown in Figure 3, the majority of the radionuclides in a nuclear waste sample pass through the column in the wash step following sample injection, resulting in a large

548 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 3. Detector traces from a sequential injection separation system set up to isolate 99Tc as pertechnetate from a nuclear waste sample, using TEVA-resin as the separation material. Figure reprinted with permission from reference 57. Copyright 1998 American Chemical Society.

transient peak as detected by a flow-through scintillation detector. Pertechnetate is retained until a strongly acid eluant solution releases it, resulting in a small 99Tc peak seen in the inset. Flow injection and sequential injection separations using extraction chromatographic separations have been developed for a variety of radionuclides.3,41,57-65 Automated radiochemical separation methods have been reviewed.9,42

3. Radionuclide Sensors 3.1. Challenges of Radionuclide Sensing in Water Prior to the development of preconcentrating minicolumn sensors for radiochemical sensing, there had been very little development of radiochemical sensors suitable for rapid and selective quantification of β- and R-emitting radionuclides in water or process streams. Thus, although there were many radioactivity detectors, there were not any selective radiochemical sensors. This state of affairs is evident, for example, in a review of “Emerging Technologies for Detecting and Measuring Contaminants in the Vadose Zone” in the Handbook of Vadose Zone Characterization and Monitoring, published in 1994.66 This review contained a section on “Radiochemical Sensors”, yet it was notably lacking in any examples of such sensors. Instead, it discussed various ways of detecting and analyzing for radionuclides and heavy metals, including general radioactivity detection techniques, inductively coupled plasma mass spectrometry (ICPMS), inductively coupled plasma atomic emission spectroscopy (ICPAES), neutron activation analysis (NAA), and X-ray fluorescence (XRF) spectrometry. Radioactivity detection and instrumental analysis techniques such as these are not radiochemical sensors and have significant limitations for field analysis. The article correctly noted that radioactivity detection “usually requires some form of sample preparation to concentrate the radionuclides prior to counting to achieve a reasonable degree of sensitivity”. The article further stated that “much of current analytical work is still done in fixed chemical laboratories using conventional radiochemical analysis” and that “conventional detection techniques... are confined to fixed or mobile laboratories”. Thus, the conventional analytical methods for R- and β-emitting radionuclides in water consist of concentration, separation, and source

Grate et al.

preparation methods prior to either radioactivity counting or mass spectrometry, activities that are largely performed in centralized laboratories. A radionuclide sensor for water monitoring must succeed at achieving results similar to those of multistep laboratory procedures, all in a compact sensor package that operates automatically. The detection and quantification of R- and β-emitting radionuclides in water present a number of basic challenges: the required detection limits are typically extremely low, the particles emitted have short penetration ranges in condensed media, and the R/β decay events in condensed media provide limited spectroscopic information for distinguishing one radionuclide from another (i.e., for selectivity). Detection limit requirements determined by regulations such as drinking water standards or maximum contaminant levels,3 typically defined in radioactivity units, translate into chemical detection limits that are well below parts per billion (ppb) levels. For example, the 33 Bq/L (900 pCi/L) drinking water standard for 99Tc67-69 translates to 0.05 µg/L , which is the same as 0.05 ppb. Required mass detection limits for other radionuclides such as 90Sr, 129I, and various transuranic actinides are from 1 to 6 orders of magnitude lower. Consequently, chemical detection with a sensor is simply not feasible; radiometric detection methods are required for measurement at and below the standards-based requirements. In addition, chemical detection alone does not distinguish between stable isotopes (that may be natural) and radioactive isotopes of concern. Uranium is an exception to this conclusion, since detection at the required tens of ppb is feasible with chemical sensing approaches, such as stripping voltammetry.70-72 Taking radiometric detection as a given, the properties of R and β emissions in water must then be considered. In contrast to γ rays, which are characterized by relatively long mean-free paths through solid and liquid media, β and especially R particles are characterized by short ranges and rapid energy dispersion in condensed media. For example, the ranges in water for β particles emitted by 90Y (Emax ) 2282 keV), 90Sr (Emax ) 546 keV), and 99Tc (Emax ) 294 keV) are 1.1 cm, 1.8 mm, and 750 µm, respectively. The range of a 5.5 MeV R particle emitted by 241Am is only 47 µm in water. Furthermore, the energies of R and β particles detected in liquids do not provide well-resolved energy spectra that can be used for selective radionuclide identification. β particles are emitted with broad energy spectra, the β spectra of different radionuclides are not well separated, and the particles lose energy as they travel through water or other condensed media. Although R particles are emitted with characteristic energies, detection by scintillation does not provide adequate energy resolution for selective individual detection of R emitters, and again, the particles lose energy as they travel through water. High-resolution R spectroscopy requires preparation of very thin counting sources placed in a close proximity to a solid-state diode detector, typically in vacuum. Even then, radiochemical separations are required to overcome R energy peak overlap problems and interferences from the sample matrix. The challenges listed above lead to a number of requirements for radiometric sensors for R and β emitters in water. (1) Due to the short radiation travel distances, the species of interest must be spatially localized within a detector volume of well-defined geometry in close proximity to the transducing medium. Localization can be achieved by sorbing the species in a material that is in close proximity to the transducing medium. Typically this transducing medium is a scintillating material, although semiconductor diodes may

Radionuclide Sensors for Environmental Monitoring

also be used. (2) Due to insufficient energy information for discrimination, the method for localizing the analyte must also be selective for particular species and separate them from potentially interfering radionuclides. (3) Due to the challenging detection limit requirements, the species must be collected from a large sample volume and preconcentrated. The preconcentrating minicolumn sensor configuration meets these requirements. The flow-based sensor comprising the sensing material in combination with the radioactivity detection method captures the analyte from the matrix, does so according to its selectivity, and can achieve very low detection limits. This approach achieves the same preconcentration and separation results that would conventionally be obtained as the result of a multistep, manual procedure. In terms of the automated radiochemical methods described above, the preconcentrating minicolumn sensor combines the separation column and the radioactivity detection, as shown in series in Figure 2c, into a single functional unit as shown in Figure 1b.

3.2. Minicolumn Sensors Based on Extractive Scintillating Resins Using a preconcentrating minicolumn sensor for radionuclide detection via scintillation requires that the column have dual functionality. It must provide chemical selectivity for capture and separation of the radionuclide of interest, and it must scintillate. Given the short ranges of the R or β particles, these functions are most readily achieved by (1) creating a column packing where the packing medium has dualfunctionality or (2) combining and intimately mixing scintillating media with selectively sorbent media in one column. The latter approach, which we call a composite bed column, will be described in the next section. The creation of polymeric beads with both ion-exchange and scintillating properties was reported over 40 years ago;73 however, there was practically no follow-up related to radiochemical analysis. The collection of radionuclides from the sample and subsequent counting were performed manually in separate steps; consequently, this did not yet represent a sensor. In 1994, the oxidation of a scintillating plastic to create ion-exchange sites was described.74 The purpose of this work was to create a tool for studying ion exchange, where the β particles from 45Ca (simulating Ca in hard water) would result in a signal when they were absorbed. The creation of dual-functionality materials for radiochemical analysis and sensing was pursued in the late 1990s by independent teams at Clemson University and Pacific Northwest National Laboratory. An investigation by DeVol et al. into the adsorption of uranium ions onto CaF2:Eu scintillator particles (Eu-doped CaF2) represented a combination of sorption and scintillation in a flow cell.1 Detection efficiencies of 60% were reported, and scintillation pulse height spectra were measured. Detection efficiency, Ed, is the ratio of observed counts to the number of decay events that occur within the detector. Subsequently, these authors described scintillating glass beads coated with organic extractants for the detection of radionuclides.2 These investigators also reported impregnation of polymer beads with extractants and scintillators for radionuclide sensing.75 Similarly, Egorov and co-workers described impregnation of both extractant and scintillating fluors in polymer beads, and characterized the analytical performance of the material in minicolumn sensors for 99Tc.3 Both groups now have a number of publications, sometimes jointly, on dual-functionality, preconcentrating minicolumn sensors for radionuclides.2-18

Chemical Reviews, 2008, Vol. 108, No. 2 549 Scheme 2

The separation properties of extractive scintillating resins were modeled after the extraction chromatographic resins used for radiochemical separations. Scintillating properties could be obtained by co-impregnating the resins with fluor molecules. For example, the fluor 2,5-diphenyloxazole, PPO, has been used as a primary fluor which captures energy deposited in the polymeric material and subsequently emits light. It has been combined with 1,4-bis(2-methylstyryl)benzene, bis-MSB,3 or with 1,4-bis(4-methyl-5-phenyloxazol2-yl)benzene, DM-POPOP,4 as a secondary fluor to shift the emitted wavelength. Alternatively, resins have also been prepared with 2-(1-napthyl)-5-phenyloxazole, R-NPO, as the primary fluor, without a secondary fluor. The chemical extractant is chosen according to the radionuclide to be retained and detected. Structures of the fluors used in developing extractive scintillating resins are shown in Scheme 2. Egorov et al. described a sensor for technetium based on co-impregnating macroreticular acrylic polymer beads with Aliquat 336, PPO, and bis-MSB.3 The Aliquat 336 is a liquid anion exchanger, as noted above, for pertechnetate separations. Its use for an automated SI separation of 99Tc was shown in Figure 3. Characterization results for the dualfunctionality material are shown in Figure 4. Analyte retention, shown as the capacity factor, k′,76 in the upper plot, is very high in low acid to neutral conditions, which is favorable for uptake from groundwater. At higher acid concentrations, pertechnetate is released; hence, acidic solutions can be used to regenerate the sensor. These results are consistent with the known uptake characteristics of Aliquat 336. The lower plot in Figure 4 shows the instrumental pulse height spectra of 99Tc obtained using the selective sensor material (trace A) in a static liquid scintillation spectrometer. This result is compared with the 99Tc spectrum in liquid scintillation cocktail (trace B). The luminosity of the sensor material is lower than that of the liquid scintillator, but the detection efficiency remains sufficiently high (56%) for practical analytical applications. This sensor material was packed into a minicolumn flow cell which was placed between the photomultiplier tubes of a Packard Radiomatic 515A flow-through scintillation detector. In this configuration the detection efficiency Ed (observed counts divided by the radioactive events from 99Tc quantitatively captured on the column) was 45%. The sensor flow cell was configured as part of a computer controlled sequential injection fluidic system for sample and reagent delivery. Detector traces illustrating selective 99Tc sensing are shown in Figure 5. Analyte capture and measurable scintillation light output are observed upon injection of an aliquot of 99Tc standard (duplicate traces shown as A) in dilute acid. The signal persists as the sensor column is washed with dilute

550 Chemical Reviews, 2008, Vol. 108, No. 2

Grate et al.

Figure 4. Analyte uptake and scintillation properties of a dualfunctionality sensor material for 99Tc. (Top) Plot of the sensor material capacity factor as a function of nitric acid concentration. (Bottom) Pulse height spectra for the dual-functionality sensor (A), and liquid scintillation spectrometer (B). Figures reprinted with permission from reference 3. Copyright 1999 American Chemical Society.

acid. By contrast, radioactive species that do not have a high affinity for the sensor material appear as only transient peak signals, and are promptly removed from the system using a small volume of wash solution. This is illustrated with 137 Cs in Figure 5 (trace C). In the presence of interferences, the light output integrated over time after the wash step provides a quantitative measure of the 99Tc in the sample. If interferences are not a problem, the analyte can be quantified from either the slope of the uptake signal or from the steady-state signal as shown in the calibration traces in the middle plot of Figure 5. Each standard was measured using a freshly packed column followed by release of the packing after the measurement, as in a renewable column sensor. The lower plot in Figure 5 illustrates detection of 99Tc in actual groundwater from the Hanford nuclear site. The water was acidified to pH 2 and analyzed with and without a spike. The detection limit for this sensor was estimated to be 6.2 Bq/L (167 pCi/L) or 0.0098 ppb, based on a 50-mL sample size and a 30-min signal accumulation. This detection limit is well below the 33 Bq/L (900 pCi/L) drinking water standard for 99Tc.67-69,77 These results illustrate the extremely low detection limits that can be achieved with a preconcentrating minicolumn radiometric sensor. Because the sensor material exhibits high binding affinity toward pertechnetate, large sample volumes can be preconcentrated using a small sensor column. DeVol et al. investigated a wide range of dual-functionality column packing materials for on-line and off-line radionuclide measurements.4 Subsequent studies examined these materials in greater detail and introduced additional materials.5-7,12-14,17 These studies included a number of extrac-

Figure 5. (Top) Sensor response to 99Tc(VII) analyte and a potentially interfering species (137Cs) unretained by the sensor material. Flow rate 1 mL min-1, injected sample volume 0.1 mL. Following the injection the sensor bed is washed with 10 mL of 0.02 M nitric acid. (Middle) Calibration traces for 99Tc sensing. (a) Sample load step (10 mL), (b) sensor wash step (10 mL), and (c) ejection of the sensing material from the column. (Bottom) Detector traces from the analysis of acidified Hanford groundwater (GW 1) sample. Flow rate used was 2 mL min-1; (a) sample load step (50 mL); (b) sensor wash step using 5 mL of 0.05 M HNO3; and (c) 30-min stopped-flow counting interval. Time zero corresponds to the beginning of the sample load step. Figures reprinted with permission from reference 3. Copyright 1999 American Chemical Society.

tive scintillating materials, extractant-coated glass scintillator particles, and a heterogeneous mixture of plastic scintillator beads with extraction chromatographic resin. The extractive scintillator materials were prepared by impregnating acrylic or styrene-based polymer beads with PPO and DM-POPOP, followed by impregnation with the extractants of interest. Extractive scintillating materials developed for radionuclide sensors are summarized in Table 1.

Radionuclide Sensors for Environmental Monitoring

Chemical Reviews, 2008, Vol. 108, No. 2 551

Table 1. Scintillating Extractive Resins for Flow-Cell Minicolumn Sensors fluor or scintillator PPO, bis-MSB PPO, DM-POPOP R-NPO PPO, 9,10 diphenylanthracene PPO, DM-POPOP PPO, bis-MSB PPO, DM-POPOP PPO, DM-POPOP PPO, DM-POPOP PPO, DM-POPOP R-NPO PPO, 9,10 diphenylanthracene GS-20 scintillating glass GS-20 scintillating glass

extractant Aliquat 336 Aliquat 336 Aliquat 336 MnO2 “ABEC” Me-PEG-2000 HDEHP crown ether DtBuCH18C6 crown ether DtBuCH18C6 CMPO/TBP CMPO/TBP H2DEH[MDP] H2DEH[MDP] H2DEH[MDP] crown ether DtBuCH18C6

Materials for 99Tc detection were developed based on two extractants, Aliquat 336 and a monomethylated-polyethylene glycol (“ABEC”). The latter is known to bind pertechnetate from certain high ionic strength solutions which is useful in the analysis of nuclear waste or waste-processing streams.78 Resins with bifunctional organophosphorus extractant octyl(phenyl)-N,N-diisobutylcarbamoylmethylphosphine oxide (CMPO) in tri-n-butyl phosphate (TBP) were investigated for actinide retention and sensing. These were modeled after Eichrom TRU-resin.48,79 In later work,13 another material for actinide detection was developed based on bis(2-ethylhexyl)methane-diphosphonic acid, which is abbreviated as H2DEH[MDP] and also known as Dipex. This material is modeled after Eichrom Actinide Resin.80 For 90Sr sensing, the crown ether4,4′(5′)-bis(tert-butylcyclohexano)-18-crown-6(DtBuCH18C6) was used. This material was modeled after Eichrom Sr-resin, which contains the crown ether in 1-octanol solution impregnated in a polymeric resin.49,81-85 This material selectively binds strontium from nitric acid solutions as Sr(NO3)2(DtBuCH18C6). In separate work, 2-ethylhexylphosphoric acid, HDEHP, was used in a prototypical 90Sr sensor with uptake from 0.001 M HCl.3 These various extractive, scintillating resins were evaluated for the efficiency with which radionuclides were captured, the efficiency with which the radionuclides were recovered from the column by elution, and the detection efficiencies.4 Values from 30 to 100% were found for the detection efficiencies in the initial study,4 which are all suitable for development of sensors. A number of these were further investigated and tested against groundwater, synthetic groundwater samples, or nuclear waste samples.5-7,12-14,17 For example, further studies of 99Tc sensing were carried out using Aliquat 336 as the sorbent.5 The fluors, PPO and bis-MSB, were diffused into the macroporous acrylic-based (Amberchrom CG-71t2) polymer beads in a separate step prior to impregnation with the extractant. This resin in minicolumn format was used to analyze six contaminated groundwater samples from the Hanford site. The samples were acidified to pH 2 prior to analysis, and quantification was carried out using standard addition methods. Each measurement involved analysis of the following solutions: (1) reagent blank (0.01 M HCl or 0.02 M HNO3, depending on acid used for sample acidification), (2) acidified groundwater sample, and (3) spiked acidified groundwater sample. All solution delivery steps were performed at 2-mL min-1 flow rate, delivering 50-mL sample aliquots, and washing the flow cell with 5 mL of 0.02 M HNO3. Then the flow was stopped, and the count rate was determined over a 30min counting interval. Analysis results obtained using the flow-cell sensor and standard radiochemical methods were in excellent agreement. The minimum detectable concentra-

support macroreticular acrylic polymer macroporous styrenic polymer macroporous styrenic polymer polyvinyltoluene macroporous styrenic polymer macroreticular acrylic polymer macroporous acrylic polymer macroporous styrenic polymer macroporous styrenic polymer macroporous acrylic polymer macroporous styrenic polymer polyvinyltoluene GS-20 scintillating glass GS-20 scintillating glass

analyte 99Tc 99Tc 99Tc

U 99Tc 90Sr 90Sr 90Sr actinides actinides actinides actinides actinides 90Sr

refs 3,5 4,5 14 14 4,5 3 6 4,6 4,7, 12,17 4 13 13 2,4 2,4,6

Figure 6. 99Tc sensing in Paducah groundwater, acidified to 0.1 M nitric acid, using a preconcentrating minicolumn sensor in portable detection instrumentation. Volumetric flow rate of ∼1 mL min-1 results in a measured slope of 8.26 cps/L. Results are replotted from data published in reference 87.

tion (MDC) of the flow-cell procedure was calculated with the Currie equation,86 using average background levels and analyte loading and detection efficiencies. For the 50-mL groundwater samples and 30-min counting interval, the MDC was reported to be 6 Bq/L. An additional pertechnetate-selective resin was prepared by co-immobilization of the R-NPO fluor with Aliquat 336 extractant within an inert macroporous polystyrene resin. This resin was used in combination with portable and transportable instruments to demonstrate these sensors as potential field screening tools.14,87,88 A Hidex Triathler field portable scintillation counter was modified with a sensing flow cell containing a small coil of Teflon tubing containing the resin. The detection efficiency measured for this sensor instrument was ∼30%. Figure 6 illustrates the Triathler instrument response obtained during loading of a 400-mL quantity of acidified 99Tc-contaminated groundwater from the Paducah site. These data were collected by recording the count rate of the detector in 100-s intervals while the groundwater was continually pumped through the flow cell. A 99Tc activity of 26.8 Bq/L was determined from the slope of the count rate, in reasonable agreement with an independent radiochemical measurement of 22.0 Bq/L. In addition to the experiments with the Triathler, a minicolumn, flow-cell detection system was designed around an Eberline E-600 survey meter and a modified photomultiplier tube (PMT) housing. The advantage of this configuration is portability, but the disadvantage is the low detection efficiency of about ∼2%. Preconcentrating minicolumn sensors for radiostrontium were prepared using a crown ether chemistry (DtBuCH18C6) to concentrate the radioactive ions of interest.6 The polymeric resins (both acrylic and styrenic resins were investigated) were impregnated with PPO, DM-POPOP, and DtBuCH18C6

552 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 7. Repeated loading and regeneration of a Sr-selective extractive scintillating resin. 89Sr was loaded in 4 M nitric acid and released in water.

in 1-octanol. Packed flow cells were monitored, and the signal was quantified using a commercially available scintillation detection system, the IN/US Beta-Ram model 1. Strontium ions were captured from 4 M nitric acid solutions with efficiencies of 99-100% and released in water, which is consistent with the known extraction chromatographic separation chemistry of DtBuCH18C6. Regeneration capability was demonstrated through multiple loading/elution cycles with 89Sr and 90Sr. Repeated capture and release of 89Sr are shown in Figure 7. There was only a slight degradation in the detection efficiency (59.7 ( 2.97% for 89Sr) over time. The detection efficiency for 90Sr was lower than that for 89 Sr because 90Sr has a weaker β energy. Detection of 89Sr was demonstrated at concentration as low as 120 Bq/L. Lower minimum detectable concentrations would require (1) that the detection efficiency be higher, (2) that the detection system background count rate be lower, (3) the use of longer count times, and/or (4) replacement of the extractant with one which has a higher distribution coefficient so that analyte can be collected from a larger sample volume without breakthrough. For uptake of 90Sr from weakly acidic to neutral solutions, a scintillating extractant resin was created based on HDEHP extractant. In this case, both 90Sr and its daughter product 90 Y were extracted from 0.001 M HCl.3 However, they could be individually determined by selective elution of the 90Sr using 0.2 M HCl, leaving the 90Y on the column. Finally, 90 Y was eluted with 4 M HCl to regenerate the column. These steps are shown in Figure 8. Assuming 100% efficiencies of the capture and elution procedures, the individual absolute detection efficiencies, Ed, for 90Sr (Emax ) 546 keV) and 90Y (Emax ) 2282 keV) were determined to be 46 and 99% respectively. Thus, co-retention and selective elution steps can be used to quantify individual radionuclide species. The method of selective uptake of a group of radionuclides with selective elution steps has also been demonstrated in actinide sensing. The extractive scintillating resin containing fluors PPO and DM-POPOP was impregnated with CMPO in TBP.7 The resin was packed in an FEP Teflon tubing flowcell coil and placed into a dual photomultiplier tube coincidence detection system to obtain pulse height spectra and time-series data. Loading and elution experiments were conducted with 241Am (9.8 Bq), 239Pu (7.4 Bq), and 233U (10.2 Bq) as illustrated in Figure 9, where on-line sensor detection results are shown in plot A (top) along with the eluted radionuclides in plot B, as determined by fraction collection and liquid scintillation counting. The data shown in Figure 9, a and b, are complementary and were obtained essentially simultaneously using the same column. The intervals when

Grate et al.

Figure 8. Detector traces for 90Sr/90Y separation experiments using HDEHP-based scintillating extraction resin. The sample was 100 µL of 1.57 × 104 dpm 90Sr/90Y standard in 0.001 M HCl. Flow rate was 1 mL min-1; (a) sample injection and sensor wash step using 6 mL of 0.001 M HCl; (b) 90Sr elution step using 6 mL of 0.2 M HCl; (c) 90Y elution using 6 mL of 4 M HCl. Figure reprinted with permission from reference 3. Copyright 1999 American Chemical Society.

Figure 9. Detection of actinides 241Am, 239Pu, and 233U in the mixed actinide solution with selective elution in a dual functionality sensor (Plot A), and the chromatogram of the eluents as determined by fraction collection and counting (Plot B). Results are replotted from data published in reference 7.

solutions were pumped through the column are indicated by the lettered boxes along the x-axis between the plots, where interval “a” is the sample load in 2 M nitric acid. The captured actinides produce a counting plateau corresponding to the total actinide count rate of 24.88 cps for a 10-min counting time. Trivalent actinides were eluted with 4 M hydrochloric acid (e.g., 241Am during interval “b”). A solution 0.02 M TiCl3 in 4 M hydrochloric acid was applied during interval “c” to reduce the plutonium and elute it from the resin. After a column rinse with 2 M hydrochloric acid during interval “d”, uranium-233 was eluted with 0.1 M ammonium bioxalate during interval “e”. The count rate difference divided by the detection efficiencies for 241Am, 239Pu, and 233 U (96.5%, 77.5% and 96.6%, respectively) constitute the

Radionuclide Sensors for Environmental Monitoring

Chemical Reviews, 2008, Vol. 108, No. 2 553

Table 2. Composite Bed Minicolumn Sensors scintillator BC-400 BC-400 GS-20 BC-400 yttrium silicate (YSO) BC-400 BC-400 BC-400

selective chemistry Aliquat 336 crown ether DtBuCH18C6 crown ether DtBuCH18C6 H2DEH[MDP] H2DEH[MDP] anion exchange, strongly basic anion exchange, weakly basic SuperLig 620 solid-phase extraction material

measured activity and were within 10% of the expected values. One advantage of this system is that quantification can be accomplished at low activities because long count times can be obtained by extending the time between eluents. This sensor resin was also applied to a digested high-level waste sludge and high-activity drain tank samples where the agreement between the on-line and off-line analyses was within 35%. The extractive scintillating resin and detection system just described was subsequently applied to low-level uranium concentration determination in acidified groundwater.12 Using Dipex extractant, a scintillating extraction resin containing R-NPO fluor was demonstrated for monitoring natural uranium in groundwater.13 This resin was packed into a flow cell designed for a modified Hidex Triathler. The average detection efficiencies were 51.7 ( 2.6% and 65.8 ( 10.1% for natural uranium and 241Am, respectively. The resin was stable for sample load volumes of up to 1000 mL, resulting in rapid real-time quantification of natural uranium in groundwater down to 30 µg/L, which is sufficient to meet the standard established in the U.S. Safe Drinking Water Act. In summary, extraction chromatographic materials containing organic scintillator fluors were developed and demonstrated in preconcentrating minicolumn sensors for uptake and detection of a variety of radionuclides, including 99 Tc, 90Sr, actinides, and natural uranium. The uptake characteristics are similar to those of the extraction chromatographic materials without the fluors, and detection efficiencies were good. Accurate detection in Hanford groundwater samples was demonstrated for 99Tc, and it could be detected to below drinking water standard limits. Detection of uranium in groundwater to drinking water standard levels was demonstrated. and samples up to at least one liter volume could be preconcentrated. Nevertheless, extractive scintillating materials have some potential drawbacks related to their stabilities under repeated or long-term use. The impregnated extractants can leach out. Under some conditions, chemiluminescence signals from the scintillating fluors are observed, which can interfere with the radiochemical measurement. In addition, the scintillating fluors are quenched or damaged and may lose their scintillating properties in the presence of some of the acid reagents used. It was noted, for example, that attempts to elute pertechnetate from resins that were co-impregnated with Aliquat 336 and the PPO/bis-MSB fluor combination, using 4 M nitric acid, resulted in a significant chemiluminescence signal and loss of scintillating properties in subsequent sensing experiments.3 Fluor impregnation methods based on diffusion or synthesis yield sensors with greater stability for sequential sensing experiments, but long-term stability remains a concern. The best combination of fluor and support for the extractive scintillating resin of the variations tested is the

solid phase TEVA resin Sr-resin Sr-resin Actinide resin Actinide resin AGMP1 AG 4-X4 silica gel-based solid-phase extraction material

analyte 99

Tc Sr 90 Sr actinides actinides 99 Tc 99 Tc 90 Sr 90

refs 4,5 6 6 13 13 8-11 16,18 10,18

R-NPO fluor impregnated into the macroporous polystyrene resin. For the combination of fluor and support to respond like a scintillator there needs to be good energy transfer from the support, which is the bulk of the scintillator, to the fluor. This π-electron energy transfer is more efficient with the polystyrene support. The procedure for producing the scintillating resin results in some fluor diffusing into the polystyrene and some just being retained within the pores of the resin. In the latter case, the fluor can interact with reagents and eluants, and may be leached out. Of the fluors evaluated, R-NPO resulted in little to no leaching from the polystyrene resin.

3.3. Composite Bed Scintillating Minicolumn Sensors Composite bed columns, consisting of a heterogeneous mixture of scintillating particles and chemically selective particles, represent an alternative to the extractive scintillating resins described above. Composite bed column materials are listed in Table 2. In most cases, the scintillating component consists of Bicron BC-400 beads, which are poly(vinyltoluene) scintillating plastic beads. These nonporous scintillating beads were found to have high chemical stability in sample and regeneration solutions. A further advantage of the composite bed sensing approach is that it facilitates the use of existing extraction chromatographic, ion-exchange, or solid-phase extraction materials for the chemically selective sorbent component of the bed. Given an intimate mixture of the sorbent particles and the scintillating particles, the R or β emission from a radionuclide captured on a sorbent particle has a reasonable probability of colliding with a neighboring scintillating bead, provided the travel distance of the radiation in the condensed phases is large enough in comparison to the distance across column pore spaces (in between particles) and particle diameters. Higher ratios of scintillating particles to sorbent particles increase this collision probability, increasing the detection efficiency, at the cost of reduced sorbent material for radionuclide capture in the composite bed. This approach is particularly well suited to detection of radionuclides that emit β particles, which have longer ranges in condensed media than typical R particles. For example, the β particle from 99 Tc has an estimated maximum range in water of 750 µm, which is greater than typical sorbent particle sizes in the 20200 µm size range. A number of composite bed columns were demonstrated for mixtures of extraction chromatographic resins and BC400 beads (100-200 µm), including those based on TEVA resin for 99Tc sensing,4,5 Sr-resin for 90Sr sensing,6 and Actinide resin for actinide measurements.13 Composite bed approaches were also developed using conventional anionexchange materials for 99Tc sensing.8-11,16,18

554 Chemical Reviews, 2008, Vol. 108, No. 2

For example, a composite bed flow cell was prepared by mixing equal masses of TEVA extraction chromatographic resin (100-150 µm) and BC-400 plastic scintillating beads (100-200 µm).5 The mixture was packed into a flow cell, and responses to 99Tc were quantified using an IN/US BetaRam model 1. The detection efficiency of 99Tc ranged from 7.5% to 16.4% during the subsequent performance tests. The uptake of 99Tc was quantitative. In addition to characterizing responses to 99Tc, tests to evaluate potential inference by 137 Cs, 90Sr, and 239Pu were investigated. Typically, the column was first exposed to a 99Tc standard, followed by an injection of a potential radioactive interference. A statistically significant increase in the sensor count rate after loading and washing the interference, while the 99Tc remained captured, was defined as an interference. For example, after loading the sensor with 5 mL of 24 Bq/mL 99Tc standard, resulting in a measurable steady-state signal, 1 mL of a 7200 Bq/mL 137 Cs standard was loaded and subsequently washed with additional carrier solution (2 M HCl for standards and wash). A transient peak was detectible while the 137Cs progressed through the flow cell, but no interference was detectable after washing. A trace amount of 137Cs was detectable in the eluant when the 99Tc was released in 8 M nitric acid. Interference trials with 90Sr and 239Pu at 54 and 240 Bq/mL, respectively, resulted in no detectable interference nor were these radionuclides detected in the final 99Tc eluant. At a higher activity of 24,000 Bq/mL, 90Sr resulted in a slight interference. Composite bed columns using conventional ion-exchange resins were demonstrated using AG MP-1, a strongly basic anion-exchange resin that has very high uptake affinity and good selectivity toward Tc(VII) ions in basic to weakly acidic media. The weight/volume distribution coefficient (Kd) for 99 Tc(VII) on AG MP-1 was 2.5 × 105 mL/g for 99Tc(VII) in unacidified Hanford groundwater. As a result, small volumes of the sorbent material can be used to preconcentrate analyte from large volumes of groundwater. A composite sensor bed was prepared by mixing 200-400 mesh AG MP-1 material (particle size about 40-70 µm) with BC400 plastic scintillator beads (particle size 100-250 µm) at a 30:1 weight ratio of scintillator to sorbent.8,10 The total bed volume was just 50 µL. The absolute detection efficiency of the composite sensor column was 34%, and analyte loading efficiency was 97%. This sensor was demonstrated to be effective in capturing 99Tc(VII) in unacidified Hanford groundwater; up to 60 mL of the groundwater sample could be preconcentrated without analyte breakthrough using this very small sensor column.9,11 The 99Tc(VII) selective composite bed sensor can be regenerated using a small volume of 2 M nitric acid solution, resulting in rapid elution of the retained analyte without loss of the scintillation properties.8,10 Composite bed sensor columns have also been prepared using the weakly basic anion-exchange resin AG 4-X4.16,18 The uptake affinity of this material for pertechnetate in weakly acidic to weakly basic conditions is substantially lower than that of the AG MP-1. Nevertheless, it has sufficient uptake to collect and preconcentrate pertechnetate, and it offers other practical advantages. The weakly basic anion exchanger is less prone to irreversible uptake of soil organic matter, such as humic acids, and pertechnetate can be readily eluted with sodium carbonate solutions, an environmentally benign reagent that provides an alternative to the use of strong acids for regeneration.16 It should be noted that high ratios of scintillator to sorbent (as was used in the AG MP-1 studies) are not necessary to

Grate et al.

Figure 10. Detection efficiency for 99Tc as a function of the dry scintillator to sorbent ratio for a composite bed sensor column. Scintillator is BC-400 (100-250 µm); sorbent is AG 4-X4 (100200 mesh).

obtain satisfactory detection efficiencies. This high ratio was motivated in part by the extremely high affinity of the strongly basic anion exchanger AG MP-1; dilution with a high ratio of scintillator reduces the overall retention volume of the column for such a high-affinity sorbent. The effect of ratio on detection efficiency was examined in more detail for weakly basic anion exchanger AG 4-X4. Various mixtures of 100-200 mesh (75-150 µm) AG 4-X4 and 100-250 µm beads of BC-400 were prepared. Detection efficiencies from 20 to 60% were found for scintillator to sorbent ratios of 1.5 to 14 (dry weight to dry weight) as shown in Figure 10. Sensing results using composite bed sensors based on AG 4-X4 will be described in more detail in connection with equilibration-based sensing below. BC-400 plastic scintillator beads have been mixed with solid-phase extraction material SuperLig 620 for the development of 90Sr sensors.18 Using a 1:1 weight ratio, a detection efficiency of 63% was reported. It was anticipated and observed that a higher detection efficiency could be obtained for 90Sr than 99Tc because the former has a higher-energy β emission.

3.4. Sensor Regeneration or Renewal Much like the preconcentrating minicolumn sensors developed in flow injection analysis, the radionuclide sensors described above are based on quantitative capture of the analyte from a certain volume of sample. Typically, the sensor is regenerated by passing a solution over the sensor such that the distribution coefficient between the stationary phase and mobile phase is significantly decreased and the analyte is released from the column and washed away. Then the sensor is ready for another sample. The chemistry used for radionuclide elution and sensor regeneration is typically taken directly from the elution conditions developed for extraction chromatographic purification of radionuclides in radiochemical analysis. Examples of analyte elution and sensor regeneration are shown in Figures 7 and 8 for 90Sr sensing. As described above, pertechnetate that is retained on anion-exchange materials at neutral to low pH can be released using more acidic solutions. However, an alternative approach based on the use of sodium carbonate was demonstrated for pertechnetate release from weakly basic anion-exchange materials.16 This environmentally benign reagent was proposed to be compatible with groundwater sensing applications. The alternative to elution or regeneration reagents is to replace the entire column packing using renewable, surfacesensing techniques as described above. This approach was proposed in one of the early examples of radionuclide sensing.3 No liquid reagent solutions would be required for

Radionuclide Sensors for Environmental Monitoring

Chemical Reviews, 2008, Vol. 108, No. 2 555

elution in this approach, but it would require a supply of sensing material to be delivered in suspension to the flow cell. One advantage of this approach is that some of the typical concerns associated with reusing sensor materials or layers, such as long-term stability, reversibility, degradation, fouling, and potential analyte carryover from sample to sample, are also alleviated.

3.5. Equilibration-Based Sensing As just noted, the preconcentrating minicolumn sensors for flow injection and for radionuclides relied on quantitative capture of analyte and subsequent regeneration with reagents. While this approach provided new radionuclide sensors for water monitoring and succeeded at meeting stringent detection limit requirements, the use of reagents to regenerate the sensor column for each and every measurement is a potential drawback for in situ monitoring applications. In addition, as discussed above, this approach represents an assay on a sample volume, as opposed to a sensor that responds to changes in sample concentration. To address these issues, a new modality for the preconcentrating minicolumn sensor was developed that we call “equilibration-based” sensing.8-11,16,18 The equilibration-based approach sets out to deliberately achieve full breakthrough conditions where the analyte concentration exiting the column is the same as the analyte concentration entering the column. Under these conditions, the sensing material in the column has equilibrated with the analyte concentration in solution. At low concentrations typical of trace detection applications, the linear portion of the sorption isotherm applies, and the amount captured on the column material is proportional to the analyte concentration. The sorbent material is not “saturated”; it is equilibrated. The amount of analyte collected has not reached the total capacity of the sorbent (as would be the case at higher concentrations corresponding to the plateau of the sorption isotherm). The key features of the equilibration-based approach are: (1) a steady-state response once the sorbent phase is equilibrated, (2) a response that varies with the analyte activity or concentration, and (3) reversibility of the response because it is based on dynamic equilibrium. When a sample containing a different analyte concentration or activity is pumped through the column, the phase will re-equilibrate, and the signal will go up or down accordingly. If this sample is blank, the signal will go down as if the column were regenerated with a reagent. Thus, in principle at least, no consumable reagents are required in this equilibration-based approach. The responses of a 99Tc sensor using this approach are shown in Figure 11. All three key features just mentioned are apparent in this figure. Steady-state responses are obtained while pumping 225 mL samples through the column, and a final blank shows the reversibility. The sensor response levels and the calibration curve in the inset illustrate how the signal varies with sample activity. The activity of the lowest level standard is equivalent to the drinking water standard for 99Tc. The column in this case had internal dimensions of 4 mm i.d. × 29 mm length for a bed volume ) 0.364 mL. The experimentally determined retention volume, Vr , was 81 mL. Retention volume is equivalent to the sample volume containing the same quantity of analyte as the fully equilibrated sorbent phase in the column. Comparison of the

Figure 11. Responses of a composite bed (1:4 w/w ratio AG 4-X4: BC-400) preconcentrating minicolumn sensor to samples at increasing activities of 99Tc standards (in 0.01 M nitric acid) in the pertechnetate form, followed by a final blank sample. The data were obtained by delivering 225-mL aliquots of the following solutions at 2-mL min-1 syringe pump flow rate: (1) blank sample, (2) 0.033 Bq/mL; (3) 0.13 Bq/mL; (4) 0.34 Bq/mL; (5) 0.65 Bq/mL; (6) blank sample. Figure reprinted with permission from reference 18. Copyright 2006 American Chemical Society.

Figure 12. Detector trace showing responses of the 90Sr sensor column (1:1 w/w ratio SuperLig 620:BC-400) to 90Sr standards in Hanford groundwater acidified to pH ∼2.1 with nitric acid: (1) 1.02 Bq/mL; (2) 0.33 Bq/mL; (3) 0.10 Bq/mL; (4) blank sample. Figure reprinted with permission from reference 18. Copyright 2006 American Chemical Society.

retention volume with the column bed volume illustrates the high degree of preconcentration achieved with this sensor (a factor greater than 200). The experimentally determined column theoretical plates, N, was 12, and the detection efficiency, Ed, was 38%. Whereas the 99Tc column sensor was based on anionexchange chemistry, a sensor for 90Sr was developed using a solid-phase extraction sorbent, SuperLig 620, which consists of a silica support with covalently bound crown ether ligands. A composite bed sensor was created with a 1:1 ratio of the SuperLig sorbent to BC-400 scintillating plastic beads. The sorbent is capable of 90Sr uptake from neutral or acidic solutions. However, because the daughter product 90Y is also slightly retained at neutral conditions and unretained under acidic conditions, experiments to illustrate equilibration-based sensing were carried out in samples acidified to pH 2 with nitric acid. The sensor responses to sequential 450-mL sample volumes of acidified groundwater containing 90Sr are shown in Figure 12. The detection efficiency Ed of this composite bed sensor was 63%, and the column theoretical plates were just N ) 3.4. These results illustrate how a different sorbent chemistry can be used to tailor this sensor concept to different radionuclide analytes, and that even with a column of rather low theoretical plates, a satisfactory sensor can be created.

556 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 13. Schematic illustration of column chromatography concepts. A discrete injection and subsequent Gaussian peak illustrates conventional chromatography in the upper pair of plots. A step function input with a subsequent sigmoidal breakthrough profile is shown in the lower pair of plots illustrating frontal chromatography.

3.6. Chromatographic Theory for Equilibration-Based Sensing The characteristics of the preconcentrating minicolumn sensor in equilibration-based sensing mode, and the attainment of the steady-state response, can be understood in terms of concepts from chromatography. In conventional chromatography with a discrete injection, the analytes are sorbed to the stationary phase in a dynamic process and migrate down the column until they elute with a typically Gaussian peak. This discrete input and the peak-shaped output are shown schematically in Figure 13. The peak shape results from a normal distribution of the velocities with which individual molecules traverse the column length, all starting at essentially the same time. Individual molecules or ions have different net velocities due to random factors including the path through the packed bed, diffusion in various directions while in the mobile phase, and variations in the amount of time each molecule or ion spends being immobilized in the stationary phase. In frontal chromatography, and in the operation of the preconcentrating minicolumn sensor in equilibration-based mode, the input is a step change in analyte concentration, as shown in the lower plots of Figure 13. As analytes migrate down the column, the step input is transformed into a sigmoid-shaped concentration profile in the column effluent.89 In this mode, the analyte molecules or ions still traverse the column with a distribution of individual velocities, but they do not all start at the same time. The sigmoid shaped effluent concentration profile is typically represented by an integral of the Gaussian distribution function.89,90 The inflection point of the sigmoidal breakthrough profile corresponds to the maximum in the Gaussian distribution. In practical terms, this inflection point also corresponds to the retention volume, Vr. These characteristics are shown in detail for the 99Tc sensor in Figure 14.18 The 99Tc activity in the input is first stepped from 0 to 1 Bq/mL for a 250-mL sample volume and then stepped back down from 1 to 0 Bq/mL. This sample volume is clearly more than that which is necessary to equilibrate the column in each step. The sensor response shown in the upper trace indicates the amount of 99Tc captured on the column. Initially the 99Tc is captured quantitatively, resulting

Grate et al.

Figure 14. (A) Trace showing sensor equilibration with 1 Bq/mL 99Tc solution followed by a reequilibration to a blank solution in a matrix of 0.01 M nitric acid. Sensor cell (dimensions 4 mm i.d. × 29 mm height) is composed of a 1:4 dry w/w ratio AG 4-X4/BC400. (B) Fractions collected immediately downstream of the sensor cell show the breakthrough profile of 99Tc from the sensor column. Both the Gaussian model (black lines in each plot) and low plate model (not shown) provide a good fit to the observed data. Figure reprinted with permission from ref 18. Copyright 2006 American Chemical Society.

in a linear increase in the signal. As 99Tc begins to break through, the response begins to level out. The lower plot shows the sigmoid-shaped output profile, i.e., the effluent concentration profile as a function of time or volume, as determined by analyzing collected fractions on the sensor output. The amount accumulating on the column, leading to the sensor response, is the total amount that has been delivered to the column minus the total amount that has broken through and exited the column. When the output concentration equals the input concentration, the amount accumulated on the column stops increasing. After the 250-mL sample load, the input was stepped from the 1 Bq/mL 99Tc activity back down to zero. The sensor signal begins to drop and the effluent concentration also begins to drop, until the retained 99Tc is completely removed from the column. Frontal chromatography theory in equation form has been adapted to derive theoretical equations that express the effluent concentration as a function of volume or time (i.e., the breakthrough curve), and similarly the sensor response as a function of the solution volume or time. The derivation and equations are given in detail in ref 18. The total amount that has exited the column is the integral of the effluent concentration profile, as shown in the first half of Figure 14b, which is itself the integral of a normal distribution (an integral of an integral). The amount that has accumulated on the sensor column, as a function of volume or time, is the amount delivered to the column minus the amount that has exited. While the plot in Figure 14 illustrates step changes from zero to one and one down to zero, the equations can be derived for step changes from any arbitrary concentration to another, and converted to radiometric count rates on the sensor. In equation form, the function f(V) is used to represent the normalized analyte concentration profile in the column

Radionuclide Sensors for Environmental Monitoring

Chemical Reviews, 2008, Vol. 108, No. 2 557

effluent, i.e., the breakthrough curve as shown schematically in the lower right plot in Figure 13 and the first half of the lower plot in Figure 14. For an input step with concentration changing from initial concentration C0 ) 0 to subsequent concentration C1 (C1 > 0), the effluent concentration as a function of volume, Cef(V), can be expressed as

Cef(V) ) C1f(V)

(1)

where f(V) goes from 0 to 1.89 In conventional linear frontal chromatography with a sufficient number of theoretical plates, the shape of function f(V) is usually assumed to be represented by an integral of a Gaussian distribution:89-91

1 1 f(V) ) + erf 2 2

(( ) x ) V -1 Vr

N 2

(2)

where erf(x) is the error function defined as erf(x) ≡ 2/xπ ∫x0 e-t2dt , and N is the number of theoretical plates. The parameter Vr is the analyte retention volume, as usual. The amount of the analyte that has exited the sensor column after delivering sample volume V, Mel(V), is given by the integral of the effluent concentration,

Mel(V) ) C1 ∫0 f(V)dV V

(3)

Then the amount of analyte, Ms,C1(V), present on the sensor column as a function of the sample volume can be calculated as the amount delivered C1V minus the amount that has exited the column:

Ms,C1(V) ) C1V - C1 ∫0 f(V)dV ) C1V - C1F(V) V

(4)

where the integral of the normalized breakthrough profile is expressed by the simplified notation:

F(V) ) ∫0 f(V)dV V

(5)

For an arbitrary step input with prior and subsequent analyte concentrations equal to C0 and C1 the amount of analyte present on the sensor column as a function of the sample volume with analyte concentration C1 can be expressed as:

Ms,C0,C1(V) ) C0Vr + V(C1 - C0) - (C1 - C0)F(V) (6) where, V is the volume of the sample solution with concentration C1. In radiometric detection, the number of radioactive decay events per second is proportional to the number of analyte atoms, while the fraction of the total decay events being detected is expressed as the absolute detection efficiency, Ed. Therefore, the radiometric sensor response can be expressed by using the following equations:

Rc/s,eq ) EdAaVr Rc/s,C0,C1 ) Ed[A0Vr + V(A1 - A0) - (A1 - A0)F(V)]

(7) (8)

Equation 7 expresses the radiometric count rate, Req, in counts/second (c/s) of a sensor column that is fully equilibrated with a sample containing analyte activity, Aa , in Bq/mL. Equation 8 gives the sensor count rate, Rc/s,C0,C1, as

a function of sample volume for an arbitrary activity step with initial and subsequent activities A0 and A1, respectively. Recalling that F(V) is an integral of f(V) (see eq 5), which depends on the column parameters Vr and N (see eq 2), it follows that the sensor response as a function of volume depends on the column chromatographic parameters, i.e., the retention volume Vr and number of theoretical plates N, and the detection efficiency Ed. In turn, the retention volume Vr depends on the volume of the stationary phase, Vs, and the analyte partition coefficient, K ) Cs/Ca, where Cs and Ca are the concentrations of the analyte in the sorbent and aqueous phases at equilibrium, respectively. It is the dependence on the partition coefficient that accounts for the capture of the analyte from the aqueous phase due to interactions between the analyte and sorbent, while variations in the partition coefficient among analytes and potential interferences provide the sorbent selectivity. The sensor responses in Figure 14 for each step change in input concentration were fit to the models using a nonlinear least-squares optimization with the detection efficiency, Ed , retention volume, Vr , and column theoretical plates, N, as regression parameters. A Gaussian function was used to model the normal distribution. These fits are shown as solid lines in Figure 14, where fits were determined for the step change from zero to the sample concentration, and separate fits were determined for the step change from sample concentration to zero. The experimental data can be fit extremely well, and one obtains important sensor and column chromatographic parameters. The fit parameters were in excellent agreement with independent experimental determinations of these parameters. Fits to the model were also used to create the solid lines in Figure 11 for the 99Tc sensor. Strictly speaking, the Gaussian distribution assumption is valid only for columns with relatively high numbers of theoretical plates.90 The sensor columns do not have high numbers of theoretical plates. However, for plate numbers greater than 5 or 10, the Gaussian model provides satisfactory fits and extracts parameters that are in agreement with other measurements. For lower numbers of theoretical plates, the Gaussian model still provides reasonable looking fits to the sensor response, but the fit parameters are unreasonable. Lo¨vkvist et al. reviewed several breakthrough profile equations and proposed an alternative function that can be used to describe the breakthrough profile in low plate number frontal chromatographic systems.90,91 This function was also integrated into the theory for the equilibration-based column sensors and dubbed the Low Plate Model.18 When applied to experimental data from columns with plate numbers of about 3, it was found to give good fits and parameters. It is not surprising that the composite bed minicolumn sensors generally have low numbers of theoretical plates, given that they have short lengths and the bed itself contains the sorbent diluted by the plastic scintillator beads. In addition, the sorbent materials have relatively big particle sizes, and linear flow velocities through the column are high. Nevertheless, such columns do make effective sensors and can be modeled. In general, extremely low numbers of theoretical plates (e.g., less than two) are undesirable because the volume in excess of the retention volume that is necessary to equilibrate the column becomes rather large, i.e., the sigmoid-shaped elution profile becomes extended horizontally. Thus, the

558 Chemical Reviews, 2008, Vol. 108, No. 2

Grate et al.

Figure 15. Sensor flow cell and PMTs of the first-generation prototype 99Tc(VII) preconcentrating minicolumn sensor for equilibration-based sensing.

design of the sensor requires a balance between theoretical plates and the volumes required to equilibrate it, and these must be balanced with the required detection limits. The fact that sensors can be characterized in terms of chromatographic parameters and modeled theoretically can help to guide decisions about sensor design.

3.7. Engineered Radiometric Preconcentrating Minicolumn Sensors for Groundwater Measurements The preconcentrating minicolumn sensor for radiometric detection of 99Tc has been engineered into a sensor probe in a rod-shaped form suitable for monitoring in groundwater via well bore holes. Images of the sensor flow cell with two PMTs is shown in Figure 15, while images of first- and second-generation prototypes are shown in Figure 16. The first-generation prototype, which incorporates an anticoincidence shield around the sensing flow cell to reduce background counts, has been described in some detail.16 Using AG 4-X4 anion exchanger and BC-400, the performance of this sensor in equilibration-based sensing mode was investigated with chemically unmodified Hanford groundwater that was spiked with known amounts of 99Tc(VII). Sample volumes of 150 mL, pumped at 3 mL min-1, were used to equilibrate the column, followed by a 30-min counting interval. A linear calibration curve was obtained. Actual contaminated Hanford groundwater samples were also analyzed with results that were in agreement with independent laboratory analysis results, using the method of standard addition for calibration. It was shown in this study that a hydroxyapatite prefilter could be used to extend the life of the sensor by reducing fouling by colored impurities (likely soil organic matter) in the groundwater. No change in sensor response was evident after pumping a volume equivalent to 36 samples of 150 mL each. For a radionuclide sensor in equilibration-based sensing mode, the time required to obtain a measurement depends on the amount of sorbent material, the flow rate, the volume of sample required to equilibrate the sorbent phase, and the time to devote to counting once the sensor has equilibrated. Although one may spend minutes to hours equilibrating the column and counting the captured radionuclides, fast response time is typically not required for subsurface monitoring applications and long-term environmental stewardship. Changes occur slowly in the subsurface, and the interval between taking data points is long. The Hanford Site’s groundwater-monitoring program typically requires the sampling and analysis of monitoring wells on a quarterly basis. A deployed sensor could be equilibrated and the signal determined and then sealed until the next measurement. At a later date, a new sample could be introduced and

Figure 16. (Left) First-generation package for the engineered 99Tc(VII) sensor. Hands at the bottom where the water intake is located provide a scale. (Right) Second generation prototype, with a water intake, filtration, and pumping components at the bottom in clear plastic, linked to the sensor module and electronics in black with a universal joint to provide some flexibility. The diameters of the two prototypes are the same, the second is longer overall.

equilibrated to get another measurement. A reversible equilibration-based sensor requiring no reagents could potentially work in the field for tens of measurements or years of use if it is sufficiently selective within the sample matrix and it does not become fouled with organic matter or bacteria.

3.8. Planar Dual-Functionality Radionuclide Sensors The column format for preconcentrating radionuclide sensors is convenient but not required. DeVol et al. have described a radial flow format for a disc-shaped composite bed with scintillation detection.15 The flow-cell was based on a planar fountain cell design,92 shown in Figure 17, which introduces solution flow from the center of the resin bed to the periphery. The sensor was interfaced with a single photomultiplier tube (PMT). This design initially concentrates most of the retained activity in the center, where the PMT has the highest sensitivity. For 99Tc capture and detection, a number of sorptive scintillating media were investigated, including (1) an extractive scintillator (dual-functionality beads) combining a porous polystyrene resin with the extractant Aliquat-336 and fluor R-NPO, (2) a composite bed of plastic scintillator beads (BC-400) and Tc-selective TEVA resin, and (3) a composite bed of inorganic scintillator particles (CaF2:Eu) with either TEVA resin or strongly basic anion-exchange resin (Dowex1 × 8-400(Cl-)). These sensors were operated in a quantitative capture mode (as opposed to the equilibration-based modality just discussed), and the capture efficiencies with these materials in this flow-cell

Radionuclide Sensors for Environmental Monitoring

Figure 17. Planar flow cell (fountain cell) coupled to a PMT detector for a dual functionality preconcentrating radiometric sensor. The upper image with a gray background shows a side crosssectional view, while the lower image represents a top crosssectional view.

configuration were all above 98%. The detection efficiencies ranged from 10% for the extractive scintillator resin to 50% for the CaF2:Eu/Dowex. These results show that this alternative geometry provides both effective capture and reasonable detection efficiencies, even using just one photomultiplier tube. Planar configurations offer flexibility in the geometry of the selective materials and transducing materials, as well as creating opportunities for placing permeable membranes between the sample and the separation/transduction materials. Such a membrane could allow diffusion of ions to the resin bead while stopping suspended particles, thus preventing sensor fouling.

3.9. Planar Radionuclide Sensors Based on Diodes Semiconductor diode detectors represent an alternative to scintillation for radiometric detection. The principles of R and β radiation detection using silicon semiconductor diodes have been described in detail previously.93,94 The diode device necessarily leads to planar sensor formats. For sensors in liquids, the requirement to capture the R- and β-emitting radionuclides close to the transducing medium remains. This can be achieved by placing a selective thin film on the diode surface and using a flow cell like the fountain cell in Figure 17. Diode detectors offer a number of potential benefits. First, they offer superior energy resolution compared to scintillation detectors. For R particles the resolution can be as good as 20 keV FWHM, as opposed to several hundred keV for scintillation detectors. Thus, sensors based on diode detectors have the potential to offer R energy spectra of radionuclides, which could be useful for distinguishing actinides that are co-retained in a semiselective film. Second, diodes for R detection offer much lower background noise levels than scintillation approaches for R detection. Third, diodes offer the potential to discriminate R particle detection from β and γ radiation. Diode detectors with a selective layer to capture R emitters have been described recently.10,95 Polymeric layers containing

Chemical Reviews, 2008, Vol. 108, No. 2 559

extractants were applied to thicknesses from 0.25 to 5 µm. Thin polymeric films containing supported extractants have been extensively used in preparation of ion-selective electrodes. Typically, high-molecular weight PVC is the matrix, but other materials such as polyacrylates have been utilized.96,97 Extractant-loaded PVC membranes for radioactive cesium and strontium have also been described.98 Similar layers can be used to create radionuclide sensors. Using a surface passivated ion-implanted silicon diode, actinide sensors were demonstrated using bis(2-ethylhexyl)phosphoric acid (HDEHP) as the extractant and plasticizer in a PVC film on the surface.10,95 This extractant effectively binds actinides from aqueous solutions of low acidity, and the capture of 241Am on the diode from dilute acid was demonstrated. The analyte remained on the diode surface during a wash step, but the diode sensor can be regenerated by eluting retained 241Am with 4 M nitric acid. The absolute analytical efficiency was ∼30%. Using layers of 0.25 µm thickness, characteristic R energy peaks were obtained for 241 Am and 233U with peak widths of 35 keV FWHM. In an alternative format, DeVol et al. used a planar silicon diode in combination with ion-exchange beads for the detection of 99Tc.99 The sample was mixed with a small volume of these beads, which were then allowed to settle on the diode surface at the bottom of the flow cell. This process captured 99Tc from solution and brought it into close proximity to the diode. The experimental results indicated that this approach could detect to levels below the drinking water standards. In these experiments, the passivated ionimplanted planar silicon semiconductor detector was spray coated with a layer of Teflon AF to make the surface resistant to contact with aqueous solutions.

3.10. Fiber-Based Sensor Scintillating fibers have a significant history in radiation detection. One example of their use as a radiation detector in environmental monitoring is the direct detection of 90Sr and 238U in soil using a “blanket” of scintillating fibers, a method developed by Schilk and co-workers and marketed by Beta-Scint Company.100 This radiation detector responds to high-energy β particles from 90Y and 234mPa as an indication of 90Sr and 238U contamination in soil. Scintillating fibers were combined with a selective sorbent resin to create dual-functionality sensors for 137Cs in aqueous samples.101 The resin contained phenol-formaldehyde oligomers grafted on a polystyrene backbone with diphosphonate ligands along the chain. Particles of this resin (3-5 µm diameter) were bonded to Bicron BCF-12 scintillating fiber with an epoxy resin layer approximately one micrometer thick. In typical experiments, the fiber was allowed to equilibrate with a volume of sample. Radiometric measurements were made separately, after removing the fiber from the sample. Detection of 137Cs in alkaline simulated Hanford tank waste samples was demonstrated with selectivity over a number of other metals.

3.11. Whole-Column Chromatographic Sensor Whole-column radiometric detection was demonstrated in the chromatographic retention and separation of positronemitting analytes.102 Link and Synovec created a column flow cell out of BC-400 scintillating plastic materials and packed it with Dionex C14 media. Thus, the device as a whole was a dual-functionality sensor although the packing itself served

560 Chemical Reviews, 2008, Vol. 108, No. 2

only as the sorbent functionality. The planar column geometry was just 0.5 mm thick, while positrons from the analyte were emitted with a distribution of energies with an energy maximum of 0.96 MeV, and a maximum range of 3.5 mm. Using two photomultiplier tubes, a detection efficiency of 96% could be obtained. A flow-through scintillation detector without a packing was placed downstream from the packed column in order to compare on-column radiometric detection with downstream radiometric detection. Injected analytes on this system were concentrated by the sorbent medium and then released using a gradient elution, resulting in chromatographic migration down the column. Counts were recorded for the entire elution period. A plateau in the detector count rate was observed while the entire quantity of injected analyte was within the column sensor. By contrast, the downstream flow-through detector produced typical transient peaks of much lower area. The wholecolumn chromatographic sensor improved detection limits by a factor 50 compared to the downstream flow-through detector. Moreover, upon injection of samples with two or three components, a chromatographic separation was observed, enabling detection and quantification of multiple components in a single injected sample.

3.12. Dual-Functionality Sensor for Tritiated Water in Air While all the selectively sorbent sensors described above were directed toward the detection of radionuclides in water, the combination of sorptive and scintillating properties in a material has also been used to develop a novel sensor for tritiated water in air.103 Tritium has a low-energy β emission. The sensor material in this case is a Eu-doped zeolite of molecular sieve type 13X, where the zeolite structure provides the sorptive properties for water. The zeolite is also intrinsically a scintillator whose scintillation efficiency is increased by the Eu doping. A disk of the zeolite material was combined with a PMT and an air flow system for tritiated water monitoring. Because tritiated hydrogen gas is not adsorbed like water, the detector can distinguish between the oxidized (tritiated water) and elemental (tritiated hydrogen gas) forms.

4. Discussion The key concept of the preconcentrating minicolumn sensor is that analytes can be efficiently collected from a large sample volume into a small detection volume on a solid-phase sorbent in a flow-cell. The macroscopic quantity of sorbent has a capacity for a significant accumulation of analyte. By filling such a flow cell with a radionuclideselective material and a method for transducing radiation events into detectable light pulses, this approach has been demonstrated to provide very sensitive radionuclide sensors for water monitoring. These sensors potentially have applications in monitoring radionuclides in nuclear fuel reprocessing or nuclear waste processing plants, or for radionuclides in environmental matrixes such as groundwater. Using quantitative capture of radionuclide analytes with subsequent regeneration of the columns, these sensors were similar to the optosensors originally developed as detectors for flow injection analysis systems. These sensors assayed the quantity of analyte in a particular volume of solution, or the rate of analyte uptake as a function of sampled volume. With the development of equilibration-based sensing approach, the preconcentrating minicolumn sensors for

Grate et al.

Figure 18. Optical detector response to the hexavalent chromium column sensor in response to chromium standards in Hanford groundwater matrix; data was obtained by pumping at ∼1.5-mL min-1 net flow rate: (1) 45-mL blank sample, (2) 60 mL of 7.1 ppb Cr(VI); (3) 60 mL of 10.6 ppb Cr(VI); (4) 60 mL of 14.2 ppb Cr(VI); (5) 100-mL blank sample. Figure reprinted with permission from ref 18. Copyright 2006 American Chemical Society.

radionuclides became true sensors whose signals rise and fall with the ambient analyte concentration, albeit with a time constant determined in part by the time necessary to equilibrate the column. In this approach, the signal magnitude is no longer dependent on the volume of the sample once the minimum equilibration volume has been delivered. Although developed primarily for radionuclide sensors, the method can also be applied to other aqueous analytes where an on-column detection method of sufficient sensitivity is available. For example, hexavalent chromium can be captured in an anion exchange-based preconcentrating minicolumn sensor and observed with spectrophotometric detection.18 (This was mentioned briefly in the Background section on Preconcentrating minicolumn sensors.) This sensor can be operated in the equilibration sensing modality as shown in Figure 18. The three key features of this sensing modality are seen, with steady-state responses, reversibility, and a signal that varies with sample concentration. Equilibration-based sensing is well-known for sorbent thin films on microsensor device surfaces, but has not been widely practiced for minicolumn-type sensors. Thin films minimize the equilibration time by creating a short diffusion distance within the sorbent material, but the amount of analyte that can be captured is limited by the small amount of sorbent material. A thin film sensor, for example, may have a film mass in the scale of a few micrograms or less, while a preconcentrating minicolumn sensor may have tens of milligrams of sorbent. Thus, a thin film may have a very low capacity compared to a packed column. In addition, a thin film sensor creates a potentially long diffusion distance from the solution to the sorbent film for very low analyte concentrations. The preconcentrating minicolumn sensor overcomes long diffusion distances from solution to sorbent by actively pumping the sample through the column so that all portions of the solution pass within close proximity to the sorbent. The equilibration-based approach maximizes the extent of preconcentration for a given column material and geometry, and the performance can be modeled using chomatographic concepts. This approach can also alleviate the need for column regeneration solutions. All these features are advantageous for environmental sensing and monitoring applications. The additional time required to reach equilibration of

Radionuclide Sensors for Environmental Monitoring

a macroscopic amount of sorbent material is not a significant disadvantage in groundwater monitoring because changes in the subsurface occur slowly, and monitoring intervals are long. Thus, one obtains high sensitivity and operational simplicity in return for a response time penalty that is not important. A reversible equilibration-based sensor requiring no reagents could work in the field for tens of measurements or years of use if all other practical issues with regard to selectivity and stability in the particular application scenario are addressed. To date, the preconcentrating minicolumn sensor represents the best available sensing approach for meeting the daunting detection limit requirements for R- or β-emitting radionuclides in groundwater.

5. Acknowledgment We gratefully acknowledge sustained funding from U.S. DOE Office of Science Environmental Management Science Program and the Environmental Remediation Science Program. Funding for the development of the engineered sensor probe was provided by the DOE Environmental Management Advanced Monitoring Systems Initiative. We thank Dr. John Hartman at the Pacific Northwest National Laboratory (PNNL) for leading efforts in the development of this prototype. J.W.G. acknowledges the William R. Wiley Environmental Molecular Sciences Laboratory, a U.S. DOE scientific user facility operated for the DOE by PNNL. The Pacific Northwest National Laboratory is a multiprogram national laboratory operated for the U.S. Department of Energy by Battelle Memorial Institute.

6. References (1) DeVol, T. A.; Keillor, M. E.; Burggraf, L. W. IEEE Trans. Nucl. Sci. 1996, 43, 1310-1315. (2) DeVol, T. A.; Roane, J. E.; Harvey, J. T. Scintillating Extraction Chromatographic Resin for Quantification of Aqueous Radioactivity. In IEEE Nucl. Sci. Symp. Conf. Rec.; IEEE: New York, NY, 1997; pp 415-419. (3) Egorov, O. B.; Fiskum, S. K.; O’Hara, M. J.; Grate, J. W. Anal. Chem. 1999, 71, 5420-5429. (4) DeVol, T. A.; Roane, J. E.; Williamson, J. M.; Duffey, J. M.; Harvey, J. T. Radioact. Radiochem. 2000, 11, 34-46. (5) DeVol, T. A.; Egorov, O. B.; Roane, J. E.; Paulenova, A.; Grate, J. W. J. Radioanal. Nucl. Chem. 2001, 249, 181-189. (6) DeVol, T. A.; Duffey, J. M.; Paulenova, A. J. Radioanal. Nucl. Chem. 2001, 249, 295-301. (7) Roane, J. E.; DeVol, T. A. Anal. Chem. 2002, 74, 5629-5634. (8) Egorov, O.; O’Hara, M. J.; Grate, J. W. Spectrum 2002, Reno, NV. (9) Grate, J. W.; Egorov, O. B. Automated Radiochemical Separation, Analysis and Sensing. In Handbook of RadioactiVity Analysis, 2nd ed.; L’Annunziata, M. F., Ed.; Academic Press: Boston, 2003; pp 1129-1164. (10) Egorov, O. B.; O’Hara, M. J.; Addleman, R. S.; Grate, J. W. ACS Symp. Ser. 2004, 868, 246-270. (11) Grate, J. W.; Egorov, O. B.; O’Hara, M. J. ACS Symp. Ser. 2004, 904, 322-341. (12) Roane, J. E.; DeVol, T. A. J. Radioanal. Nucl. Chem 2005, 263, 51-57. (13) Hughes, L.; DeVol, T. A. Nucl. Instrum. Methods Phys. Res., Sect. A 2003, 505, 435-438. (14) Ayaz, B.; DeVol, T. A. Nucl. Instrum. Methods Phys. Res., Sect. A 2003, 505, 458-461. (15) Hughes, L. D.; DeVol, T. A. Anal. Chem. 2006, 78, 2254-2261. (16) Egorov, O. B.; O’Hara, M. J.; Grate, J. W.; Knopf, M.; Anderson, G.; Hartman, J. J. Radioanal. Nucl. Chem. 2005, 264, 495-500. (17) Fjeld, R. A.; Roane, J. E.; Leyba, J. D.; Paulenova, A.; DeVol, T. A. ACS Symp. Ser. 2004, 868, 105-119. (18) Egorov, O. B.; O’Hara, M. J.; Grate, J. W. Anal. Chem. 2006, 78, 5480-5490. (19) Bosworth, N.; Towers, P. Nature 1989, 341, 167-168. (20) Cook, N. D. Drug DiscoVery Today 1996, 1, 287-294.

Chemical Reviews, 2008, Vol. 108, No. 2 561 (21) L’Annunziata, M. F. In Handbook of RadioactiVity Analysis; L’Annunziata, M. F., Ed.; Academic Press: San Diego, 1998; pp 556-565. (22) Valcarcel, M.; Luque de Castro, M. D. Flow-Through (Bio)Chemical Sensors; Elsevier: Amsterdam, 1994. (23) Ruzicka, J.; Hansen, E. H. Flow Injection Analysis, 2nd ed.; WileyInterscience: New York, 1988; Vol. 62, p 498. (24) Fang, Z. Flow Injection Separation and Preconcentration; VCH: Weinheim, 1993. (25) Ruzicka, J.; Marshall, B. D. Anal. Chim. Acta 1990, 237, 329. (26) Ruzicka, J. Anal. Chim. Acta 1992, 261, 3-10. (27) Ivaska, A.; Ruzicka, J. Analyst 1993, 118, 885-889. (28) Ruzicka, J. Collect. Czech. Chem. Commun. 2005, 70, 1737-1755. (29) Miro, M.; Hansen, E. H. Trends Anal. Chem. 2006, 25, 267-281. (30) Yoshimura, K.; Matsuoka, S. Lab. Rob. Autom. 1993, 5, 231-244. (31) Miro, M.; Frenzel, W. Trends Anal. Chem. 2004, 23, 11-20. (32) Yoshimura, K.; Yamada, S. Talanta 1992, 39, 1019-1024. (33) Torre, M.; Marina, M. L. Crit. ReV. Anal. Chem. 1994, 24, 327361. (34) Yoshimura, K. Anal. Chem. 1987, 59, 2922-2924. (35) Ruzicka, J.; Scampavia, L. Anal. Chem. 1999, 71, 257A-263A. (36) Ruzicka, J. Anal. Chim. Acta 1995, 308, 14-19. (37) Egorov, O.; Ruzicka, J. Analyst 1995, 120, 1959-1962. (38) Ruzicka, J. Analyst 1994, 119, 1925-1934. (39) Pollema, C. H.; Ruzicka, J. Anal. Chem. 1994, 66, 1825-1831. (40) Ruzicka, J.; Pollema, C. H.; Scudder, K. M. Anal. Chem. 1993, 65, 3566-3570. (41) Egorov, O.; O’Hara, M. J.; Grate, J. W.; Ruzicka, J. Anal. Chem. 1999, 71, 345-352. (42) Grate, J. W.; Egorov, O. B. Anal. Chem. 1998, 70, 779A-788A. (43) Dockendorff, B.; Holman, D. A.; Christian, G. D.; Ruzicka, J. Anal. Commun. 1998, 35, 357-359. (44) Bruckner-Lea, C. J.; Stottlemyre, M. S.; Holman, D. A.; Grate, J. W.; Brockman, F. J.; Chandler, D. P. Anal. Chem. 2000, 72, 41354141. (45) Grate, J. W.; Bruckner-Lea, C. J.; Jarrell, A. E.; Chandler, D. P. Anal. Chim. Acta 2003, 478, 85-98. (46) Chandler, D. P.; Brockman, F. J.; Holman, D. A.; Grate, J. W.; Bruckner-Lea, C. J. Trends Anal. Chem. 2000, 19, 314-321. (47) Cortina, J. L.; Warshawsky, A. Ion Exch. SolVent Extr. 1997, 13, 195-293. (48) Horwitz, E. P.; Chiarizia, R.; Dietz, M. L.; Diamond, H.; Nelson, D. M. Anal. Chim. Acta 1993, 281, 361-372. (49) Horwitz, E. P.; Chiarizia, R.; Dietz, M. L. SolVent Extr. Ion Exch. 1992, 10, 313-336. (50) Horwitz, E. P.; Dietz, M. L.; Chiarizia, R.; Diamond, H.; Maxwell, S. L.; Nelson, M. R. Anal. Chim. Acta 1995, 310, 63-78. (51) Dietz, M. L.; Horwitz, E. P. LC-GC 1993, 11, 424-426, 428, 430, 434, 436. (52) Maxwell, S. L. Radioact. Radiochem. 1997, 8, 36-44. (53) Izatt, R. M.; Bradshaw, J. S.; Bruening, R. L. Pure Appl. Chem. 1996, 68, 1237-1241. (54) Izatt, R. M. J. Inclusion Phenom. Mol. Recognit. Chem. 1997, 29, 197-220. (55) Izatt, R. M.; Bradshaw, J. S.; Bruening, R. L. Pure Appl. Chem. 1995, 68, 1237-1241. (56) Izatt, R. M.; Bradshaw, J. S.; Bruening, R. L.; Bruening, M. L. Am. Lab 1994, 26, 28C. (57) Egorov, O. B.; O’Hara, M. J.; Ruzicka, J.; Grate, J. W. Anal. Chem. 1998, 70, 977-984. (58) Egorov, O. B.; O’Hara, M. J.; Farmer, O. T., III; Grate, J. W. Analyst 2001, 126, 1594-1601. (59) Grate, J. W.; Fadeff, S. K.; Egorov, O. Analyst 1999, 124, 203210. (60) Grate, J. W.; Egorov, O. B.; Fiskum, S. K. Analyst 1999, 124, 11431150. (61) Grate, J. W.; Egorov, O. Anal. Chem. 1998, 70, 3920-3929. (62) Egorov, O.; Grate, J. W.; Ruzicka, J. J. Radioanal. Nucl. Chem. 1998, 234, 231-235. (63) Grate, J. W.; Strebin, R. S.; Janata, J.; Egorov, O.; Ruzicka, J. Anal. Chem. 1996, 68, 333-340. (64) Egorov, O. B.; O’Hara, M. J.; Grate, J. W. Anal. Chem. 2004, 76, 3869-3877. (65) Egorov, O.; O’Hara, M. J.; Grate, J. W. J. Radioanal. Nucl. Chem. 2005, 263, 629-633. (66) Koglin, E. N.; Poziomek, E. J.; Kram, M. L. In Handbook of Vadose Zone Characterization & Monitoring; Wilson, L. G., Everett, L. G., Cullen, S. J., Eds.; Lewis Publishers: Ann Arbor, 1994. (67) Hartman, M. J., Morasch, L. F., Webber, W. D., Eds. Hanford Site Groundwater Monitoring for Fiscal Year 2005, Pacific Northwest National Laboratory, PNNL-14670, 2006.

562 Chemical Reviews, 2008, Vol. 108, No. 2 (68) Hartman, M. J.; Dresel, P. E.; editors, Hanford Site Groundwater Monitoring for Fiscal Year 1997, Pacific Northwest National Laboratory, PNNL-11793 UC-402, 403, 702, 1998. (69) The drinking water standard for a β emitter in water as established by the EPA is 4 mrem/year. Using the dose conversion factor from National Bureau of Standards Handbook 69 (U.S. Department of Commerce, as amended August 1963) and the other parameters established by the EPA, one can calculate an equivalent concentration of 33 Bq/L (900 pCi/L) assuming 99Tc is the sole β emitter. (70) Wang, J.; Lu, J.; Wang, J.; Luo, D.; Tian, B. Anal. Chim. Acta 1997, 354, 275-281. (71) Wang, J. Stripping Analysis: Principles, Instruments, and Applications; VCH Publishers, Inc.: New York, 1985. (72) Olsen, K. B.; Wang, J.; Setiadji, R.; Lu, J. EnViron. Sci. Technol. 1994, 28, 2074-2079. (73) Heimbuch, A. M.; Gee, H. Y.; DeHaan, A. J.; Leventhall, L. Radioisotope Sample Measurement Techniques in Medicine and Biology; International Atomic Energy Agency Symposium, Vienna, May 24-28, 1965. (74) Li, M.; Schlenoff, J. B. Anal. Chem. 1994, 66, 824-829. (75) DeVol, T. A.; Roane, J. E.; Williamson, J. M.; Duffey, J. M.; Harvey, J. T. 44th Annual Conference on Bioassay, Analytical, and Environmental Radiochemistry, Alburquerque, NM, Novermber 15-20, 1998. (76) Capacity factors, k′, were calculated as k′) A × Dw × (Vs/Vm ) after equilibrating known quantities of sensor beads with solutions of known volume and 99-Tc activity . The ratio Vs/Vm is the stationary/ mobile phase volume ratio. Weight distribution ratios, Dw, were calculated according to the formula Dw ) ((A0 - As)/W)/(As/V), where A0 is the activity of the blank solution after equilibration, As is the activity of the test solution after equilibration with beads, W is the weight of the beads (g), and V is the volume of the equilibrated solution (mL). Finally, the coefficient A is the conversion factor from Dw to volume distribution ratio). Further details are in the cited work. (77) “National Primary Drinking Water Regulations; Radionuclides; Final Rule,“ Federal Register, Vol. 65, No. 236, December 7, 2000, pp 76708-76753. (78) Rogers, R. D.; Bond, A. H.; Griffin, S. T.; Horwitz, E. P. SolVent Extr. Ion Exch. 1996, 14, 919-946. (79) Horwitz, E. P.; Dietz, M. L.; Diamond, H.; LaRosa, J. J.; Fairman, W. D. Anal. Chim. Acta 1990, 238, 263-271. (80) Horwitz, E. P.; Chiarizia, R.; Dietz, M. L. React. Funct. Polym. 1997, 33, 25-36. (81) Dietz, M. L.; Horwitz, E. P.; Nelson, S. M.; Wahlgren, M. Health Phys. 1991, 61, 871-877.

Grate et al. (82) Horwitz, E. P.; Dietz, M. L.; Fisher, D. E. Anal. Chem. 1991, 63, 522-525. (83) Horwitz, E. P.; Dietz, M. L.; Chiarizia, R. J. Radioanal. Nucl. Chem. Articles 1992, 161, 575-583. (84) Vajda, N.; Ghods-Esphahani, A.; Cooper, E.; Danesi, P. R. J. Radional. Nucl. Chem. 1992, 162, 307-323. (85) Jeter, H. W.; Brob, B. Radioact. Radiochem. 1994, 5, 8-16. (86) Currie, L. A. Anal. Chem. 1968, 40, 586-591. (87) DeVol, T. A.; Roane, J. E.; Leyba, J. D. In LSC 2001 AdVances in Liquid Scintillation Spectrometry; Mobius, S., Noakes, J. E., Schonhofer, F., Eds.; University of Arizona: Tuscon, 2002; pp 415-424. (88) DeVol, T. A.; Roane, J. E.; Hughes, L.; Ayaz, B. Spectrum 2002: 9th Biennial International Conference on Nuclear and Hazardous Waste Management, Reno, NV, 2002. (89) Reilley, C.; Hildebrand, G. P.; Ashley, J. W. Anal. Chem. 1962, 34, 1198-1213. (90) Lovkvist, P.; Jonsson, J. A. Anal. Chem. 1987, 59, 818-821. (91) Lovkvist, P.; Jonsson, J. A. J. Chromatogr. 1986, 356, 1-8. (92) Scudder, K. M.; Pollema, C. H.; Ruzicka, J. Anal. Chem. 1992, 64, 2657-2660. (93) Fettweis, P. F.; Verplancke, J.; Venkataraman, R.; Young, B. M.; Schwenn, H. In Handbook of RadioactiVity Analysis, 2nd ed.; L’Annunziata, M. F., Ed.; Academic Press: Boston, 2003; pp 239346. (94) Knoll, G. F. Radiation Detection and Measurement; John Wiley & Sons: New York, 1989. (95) Addleman, R. S.; O’Hara, M. J.; Grate, J. W.; Egorov, O. B. J. Radioanal. Nucl. Chem. 2005, 263, 291-294. (96) Edmonds, T. E., Ed. Chemical Sensors; Chapman and Hall: New York, 1988. (97) Janata, J.; Bezegh, A. Anal. Chem. 1988, 60, 62R. (98) Rais, J.; Mason, C. V.; Abney, K. D. Sep. Sci. Technol. 1997, 32, 951-969. (99) Hughes, L. D.; DeVol, T. A. J. Radional. Nucl. Chem. 2006, 267, 287-295. (100) Schilk, A. J.; Knopf, M. A.; Thompson, R. C.; Hubbard, C. W.; Abel, K. H.; Edwards, D. R.; Abraham, J. R. Nucl. Instrum. Methods Phys. Res., Sect. A 1994, 353, 477-481. (101) Headrick, J.; Sepaniak, M.; Alexandratos, S. D.; Datskos, P. Anal. Chem. 2000, 72, 1994-2000. (102) Link, J. M.; Synovec, R. E. Anal. Chem. 1999, 71, 2700-2707. (103) Campi, F.; Edwards, A. H.; Ossiri, A.; Pacenti, P.; Terrani, S. Health Phys. 1998, 75, 179-182.

CR068115U

Chem. Rev. 2008, 108, 563−613

563

Higher-Order Chemical Sensing Andreas Hierlemann* ETH Zu¨rich, Laboratory of Physical Electronics, Wolfgang-Pauli-Strasse 16, 8093 Zu¨rich, Switzerland

Ricardo Gutierrez-Osuna Texas A&M University, Department of Computer Science, College Station, Texas 77843-3112 Received April 3, 2007

Contents 1. Introduction 1.1. Integrated versus Discrete Sensor Arrays 1.1.1. Materials and Fabrication Processes 1.1.2. Performance 1.1.3. Auxiliary Sensors/Smart Features 1.1.4. Connectivity 1.1.5. Sensor Response Time 1.1.6. Package 1.1.7. Summary 1.2. Outline 2. Arrays and Systems Comprising Identical Transducers 2.1. Parameter Variations 2.2. Dynamic Methods and Transient Signals 2.3. Extracting Information in Different Domains 3. Arrays and Systems Comprising Different Transducers 3.1. Metal- and Metal-Oxide-Based Gas Sensors 3.2. Polymer-Based Gas Sensors 3.3. Gas Sensor Arrays Relying on Different Transducer and Sensitive-Material Types 3.4. Liquid-Phase Chemo- and Biosensors 3.5. Cell-Based Biosensors 4. Operational Considerations for Higher-Order Devices 4.1. Setup and Manifold Considerations 4.2. Multitransducer Operation Example 5. Sensor-Based Microanalytical Systems 6. Are More Sensors Better? 6.1. Characteristics of High-Dimensional Vector Spaces 6.2. Orthogonality versus Independence 6.3. Cross-sensitivity and Diversity 6.4. Multiple Roles of Redundancy 7. Data Preprocessing 7.1. Baseline Correction 7.2. Scaling 7.2.1. Global Techniques 7.2.2. Local Techniques 7.2.3. Nonlinear Transforms

563 565 565 565 565 565 565 565 566 566 566 566 567 569 570 570 571 574 577 578 579 579 580 581 583 584 584 585 585 586 586 587 588 588 588

* To whom correspondence should be addressed. Voice: +41 44 633 3494. Fax: +41 44 633 1054. E-mail: [email protected].

8. Drift Compensation 8.1. Univariate Drift Compensation 8.2. Multivariate Drift Compensation 9. Feature Extraction from Sensor Dynamics 9.1. Transient Analysis 9.1.1. Oversampling Procedures 9.1.2. Ad hoc Transient Parameters 9.1.3. Model-Based Parameters 9.1.4. Comparative Studies 9.2. Temperature-Modulation Analysis 10. Multivariate Calibration 10.1. Multiway Analysis 10.2. Dynamical Models 11. Array Optimization 11.1. Sensor Selection 11.2. Feature Selection 11.3. Optimization of Excitation Profiles 12. Conclusion and Outlook 13. References

588 589 589 591 591 592 593 593 595 596 599 599 602 604 604 605 607 608 609

1. Introduction Desired properties of a chemical sensor include high sensitivity, a large dynamic range, high selectivity or specificity to a target analyte, related low cross-sensitivity to interferents, perfect reversibility of the physicochemical detection or sensing process (short sensor recovery and response times), and long-term stability of the sensor and sensing material.1-9 Unfortunately, a sensor exhibiting all these properties is a largely unrealizable ideal. Sensor sensitivity, selectivity, speed of response, and reversibility are determined by the thermodynamics and kinetics of sensor material/analyte interactions. In particular, high sensitivity and specificity on the one hand and perfect reversibility on the other hand impose contradictory constraints on the sensor design: high sensitivity and selectivity are typically associated with strong interactions, whereas perfect reversibility requires weak interactions. Consequently, it is necessary to compromise, and, in most cases, sensors showing partial selectivity to only some of the detected species are used to ensure reversibility. The output of an individual sensor consists of, e.g., a certain current value measured at a fixed potential or a resistance value of a certain material in response to a chemical stimulus.1-9 This means that, usually, one feature per sensor is monitored at a time, preferably

10.1021/cr068116m CCC: $71.00 © 2008 American Chemical Society Published on Web 01/17/2008

564 Chemical Reviews, 2008, Vol. 108, No. 2

Andreas Hierlemann received his diploma in chemistry in 1992 and the Ph.D. degree in physical chemistry in 1996 from the University of Tu¨bingen, Germany. After working as a Postdoc at Texas A&M University, College Station, TX (1997), and Sandia National Laboratories, Albuquerque, NM (1998), he is currently associate professor at the Physical Electronics Laboratory at ETH Zurich in Switzerland. The focus of his research activities is on CMOS-based microsensors and interfacing CMOS electronics with electrogenic cells.

Ricardo Gutierrez-Osuna received a B.S. degree in Electrical Engineering from the Polytechnic University of Madrid, Spain, in 1992, and M.S. and Ph.D. degrees in Computer Engineering from North Carolina State University in 1995 and 1998, respectively. From 1998 to 2002, he served on the faculty at Wright State University. He is currently an associate professor of Computer Engineering at Texas A&M University. Dr. GutierrezOsuna is a recipient of the National Science Foundation Career Award for his research on machine olfaction with chemical sensors arrays. His research interests include pattern recognition, neuromorphic computation, chemical sensor arrays, and machine perception.

during an equilibrium-type or steady-state-type situation, in which a certain analyte concentration can unequivocally be correlated to the resulting sensor response. Individual sensors and the analysis of the respective individual signals or features, however, show limited selectivity performance in most practical applications, as mentioned above. Therefore, arrays of several sensors are commonly used, which effectively extends the “feature space”. A feature space is an abstract space in which each sample (e.g., sensor measurement value) is represented as a point in n-dimensional space, whose dimension is determined by the number of features evaluated. Features are the individual measurable heuristic properties of the phenomena being observed, in our case, e.g., sensor measurement values. The acquired information is then processed using pattern recognition and multicomponent analysis tools.10-12 Increasing the measurement or feature space dimensionality is an attractive approach to obtain a substantial improvement in analytical capabili-

Hierlemann et al.

ties,13,14 provided that the additional dimensions carry complementary information and provided that suitable dataanalysis techniques are used, i.e., techniques to handle small numbers of samples. The notion of “order” has been frequently used in instrumental analysis to categorize the dimensionality of the feature space and, consequently, the richness of the information obtained by a certain device or analytical instrument15,16 and can, within certain limitations, also be applied to sensor or sensor array configurations.17 A zero-order device or sensor would be a single, moreor-less specific sensor. This sensor would be aimed at the detection of a certain target compound. However, it would not be possible to perform any reliable analyte quantification in the presence of interferents. Even worse, there would not be any evidence to let the operator know that the measurement had been influenced by interferents. A first-order device is then a sensor array, the sensors of which differ in one domain, i.e., an array of identical transducers such as chemoresistors featuring different sensitive materials or layers or an array of chemoresistors featuring the same sensitive material but operated at different temperaturessprovided that temperature has a distinct effect on the sensor selectivity. This is the case for, e.g., hightemperature metal-oxide-based sensors.18-22 A prerequisite for the successful use of such first-order devices is the establishment of a calibration model that includes the signals of the species of interest and of all possible interferents. Then, multicomponent analysis and outlier detection will be possible, but any unexpected interferent will invariably upset the respective predictions. In other words, the calibration of a first-order array allows interferences to be detected but not to be compensated for. This problem can be addressed by using second-order or higher-order sensor devices. These devices utilize analyte characteristics in at least two domains, which should be ideally orthogonal, or, inasmuch as possible, independent from each other (see section 6.2 for a definition of these terms). Under certain conditions, the calibration of secondorder instruments allows the target analytes to be quantified even in the presence of unknown interferents; this property is commonly referred to as the “second-order advantage”.23 Additional potential benefits of second-order devices include the ability to perform calibration with a single mixture sample and recover the response profiles of the individual target analytes.24 Higher-order sensor devices can, according to a paper published by Go¨pel,25 be generated by making use of various features to be exploited in chemical sensing. Though the estimated overall number of features may be overly optimistic (Figure 1), in particular with regard to “independent features”, it may be interesting to briefly summarize the findings of this paper. The number of sensitive materials to convert a chemical into, e.g., electrical information is estimated to be on the order of 108, the number of suitable transducers (e.g., chemoresistor, microbalance, optical fiber, etc.) is estimated to be ∼101, the number of transducer geometries (e.g., electrode distance, shape, etc.) is estimated to be on the order of 102, and the features that can be added through variation of modulation of external (e.g., gas switching strategies, use of filters, catalysts, etc.) and internal parameters (e.g., light frequency, operation temperature, bias voltage, etc.) is estimated to be on the order of 102. The distinctive way to modulate these parameters (e.g., stepwise, sinusoidal, ramp, etc.) is assumed to account for an additional

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 565

Figure 1. “Hyperspace” of chemical sensor features with more than 1021 independent features according to Go¨pel. Redrawn with persmission from ref 25. Copyright 1998 Elsevier.

102 features, and, finally, multiple modulations that can be performed simultaneously (e.g., simultaneous variation of bias voltage, temperature, and gas flow) are assumed to account for another 106 features. While the above considerations are of hypothetical nature, and the number of viable parameters and realizable variations will be massively lower, the exercise shows that a large variety of parameters is available that can be used to identify or quantify a specific analyte or complex mixture. Higher-order (i.e., “hyphenated”) instrumentation, such as the combination of gas chromatography (GC) and mass spectrometry (MS), is vastly used in modern laboratory analytical chemistry. Although the data-analysis methods applied to sensor and analytic-instrument data are mostly the same, there is no direct analogy between the preprocessing of a sample to separate it into multiple, less complex samples that are then characterized by an analytical method and the attempts to enhance the selectivity of a set of sensors by modulation of physical parameters. Yet, it is striking that higher-order methods are still uncommon in the field of chemical sensors. This holds particularly true, since the higher-order advantage may be important due to unpredictable changes in the sample matrix composition. This situation is arguably a consequence of the requirements for sensors or sensor systems, which include low costs, small physical size, and ease of use. The progress in micro- and nanotechnology, microelectronics, and in data-processing speed and capability will help to address many, if not all, of these issues:26-30 rather complex and versatile microsensor and microanalysis systems operable directly through standard interfaces from a laptop or palmtop by means of standard software are already available, as will be also demonstrated in this article.

1.1. Integrated versus Discrete Sensor Arrays In this context, we want to address the advantages and disadvantages of integrated or even monolithic multisensor arrays or systems versus the use of sets of discrete sensors and electronics, particularly, since both types of sensor arrays or systems will appear in the following sections. There is a number of aspects that have to be taken into account, which will be briefly discussed here.31

1.1.1. Materials and Fabrication Processes For monolithic designs or integrated systems, the selection of materials is restricted to a few, e.g., CMOS-technologyrelated materials and CMOS-process-compatible materials, as well as to a set of specific fabrication steps32 (CMOS technology, complementary-metal-oxide-semiconductor technology, is an industrial standard fabrication technology for integrated circuits on silicon microchips). High-temperature steps (e.g., >400 °C) can be detrimental to metallizations (metal oxidation, diffusion) and may alter semiconductor

characteristics. For hybrid or discrete devices, any material, or the optimum sensor material, can be used, and a wealth of fabrication techniques is available.

1.1.2. Performance Microsensors frequently also generate “microsignals”, and perform pronouncedly better in monolithic designs owing to the fact that the signals can be conditioned at the site of generation (filters, amplifiers, etc.), e.g., by means of onchip electronics, so that the influence of parasitic and crosstalk effects can be reduced.33,34 On-chip analog-to-digital conversion is another feature that helps to generate a stable sensor output that can be easily transferred to off-chip units. For hybrid or discrete microsensors, it is sometimes very difficult to read out rather minute analog sensor signals.

1.1.3. Auxiliary Sensors/Smart Features Temperature or flow sensors can be monolithically cointegrated with the chemical sensors on the same chip. Calibration, control, and signal processing functions as well as self-test features can be realized on-chip. For hybrid designs, additional devices and off-chip components are required.

1.1.4. Connectivity The number of electrical connections prominently contributes to the overall system costs (failure probability and packaging costs). The monolithic implementation of, e.g., an array of gas sensors (see also later in this article) with multiplexer structures and interface units requires only a few connections.33,34 A hybrid/discrete approach will require many more connections, since each sensor has to be individually addressed and since there are no interface units available on the sensor side.

1.1.5. Sensor Response Time The response time of, e.g., a gas sensor array is, in most cases, determined by the volume of the measurement chamber and the flow rate (other relevant processes include also analyte diffusion or dissociation). Using the monolithic or integrated approach and a suitable packaging technique, such as flip-chip packaging, the volume of the measurement chamber can be kept very small as a consequence of a smallsize, flat and planar sensor or sensor array. Therefore, parameter modulation, such as flow variations or dynamic protocols, can be easily realized. For hybrid or discrete arrays, the volume is dependent on sensor geometries and array arrangements.

1.1.6. Package To package monolithic designs, microelectronics-derived packaging techniques can be modified and adapted, such as

566 Chemical Reviews, 2008, Vol. 108, No. 2

flip-chip technology or simple epoxy-based packaging methods. Hybrid implementations require complex packages to reduce sensor interference (e.g., high-frequency acousticwave sensors), to minimize electric crosstalk, and to optimize the critical connections. This further complicates the already difficult task of chemical sensor packaging.

1.1.7. Summary The main disadvantages of integrated or monolithic arrays or systems include the restriction in materials and the limited choice of fabrication processes and steps. On the other hand, integrated systems offer unprecedented advantages over hybrid or discrete arrays, especially with regard to signal quality, device performance, increased functionality, and available packaging solutions.35 These advantages, in our opinion, clearly outweigh the drawbacks and limitations.31 In the case of well-established physical sensors, such as acceleration and pressure sensors, a trend toward integrated monolithic solutions can be identified for large production volumes and severe cost restrictions.

1.2. Outline The topic of this review is “higher-order devices”, i.e., neither single sensors nor homogeneous transducer arrays featuring only different coatings will be treated any further. The latter constitute, according to the text in the introduction above, a first-order system, since analyte exposure generates a one-dimensional data vector (row or column). In some publications, an array of sensors with different coatings is referred to as a zero-order array,9 with an array being itself a first-order device. We here prefer to use a categorization according to the data output format (0th order, single value; first order, vector; second order, matrix; third order, tensor; etc.). We decided to apply a rather broad scope in this review in order to give the reader a comprehensive overview on strategies to increase the information that can be extracted from sensor systems or arrays. In section 2, devices consisting of arrays of identical transducers (with different coatings) will be described, for which an extension to a higher order has been realized by adding additional dimensions such as time (sensor dynamics and transients) or by varying internal and external operation parameters such as temperature modulation or the use of a catalyst for analyte decomposition. In this section, it will also be shown how different information, e.g., in the physical and chemical domains, can be extracted from an array of identical transducers. Section 3 focuses on different monolithic and discrete sensor arrays making use of more than one transduction principle. In section 4, we will briefly describe practical measurement and setup considerations for using multitransducer sensor arrays and for recording transients or applying parameter variations. Section 5 includes sensor-based complex microanalytical systems, consisting of preconcentration, separation, and detection stages. In section 6, we will discuss the relations between dimensionality, information, cross-selectivity, and redundancy, concepts that are important when dealing with higher-order sensor systems. Section 7 presents a review of two important data preprocessing procedures for chemosensors: baseline correction and scaling. Section 8 will be dedicated to methods for drift compensation. In section 9, we will review computational methods to extract information from transient and temperature-modulated responses of chemosensors. Section 10 is dedicated to multivariate

Hierlemann et al.

calibration for higher-order sensing, with an emphasis on multiway and dynamical models. Section 11 will include methods to optimize arrays for specific applications, including sensor selection, feature selection, and optimization of temperature programs. This article will be concluded by a short summary and outlook. Before embarking upon the subject of multisensor arrays, there is a last note on terminology. The term “electronic nose” has been very popular for more than a decade to describe multisensor and multitransducer arrays.6,36,37 We believe, however, that this term can be very misleading for several reasons. Following a definition of Gardner and Bartlett, an “electronic nose” is “an instrument comprising an array of electronic chemical sensors with partial specificity and an appropriate pattern recognition system, capable of recognizing simple or complex odors”.6,36,37 The majority of such “electronic noses” may be capable of differentiating between analytes or analyte mixtures from the headspace of different foods or beverages, but, in most cases, the sensor response patterns cannot be directly correlated with human olfactory perception. Another more important point concerns the general applicability implied by the term “electronic nose”. Most of the sensor systems perform well in certain key applications, but there are few systems, if any, that exhibit the enormously broad applicability spectrum, at once including the sensitivity and discriminating power of a human or animal nose. In addition, successful sensor systems have to be designed and optimized with the key application in mind to guide the selection of coatings, transduction mechanisms, etc. As yet, however, there is no universally applicable system that invariably provides satisfactory performance under all circumstances. Therefore, we will use this term sparingly and only in quotation marks. Similar considerations apply to “electronic tongues”, ion-sensitive or lipid-film based sensors in liquid phase. An article on “electronic noses” by Weimar38 is included in this issue.

2. Arrays and Systems Comprising Identical Transducers In this section, we will treat arrays of identical transducers with, e.g., different coatings, for which an extension to a higher order has been realized by adding additional dimensions in the feature space. These additional dimensions may include time (sensor dynamics and transients), or the variation of internal and external sensor operation parameters (temperature variation, use of a catalyst for analyte decomposition). It will also be shown how information in the physical and chemical domains can be extracted from microarrays and microsystems. A short glance at differential or ratiometric methods concludes this section.

2.1. Parameter Variations Following Go¨pel and Weimar,25,39 the parameters that can be varied during sensor operation include internal parameters, such as sensor temperature, electrode bias voltage, or measurement frequencies, and external parameters, such as the use of filters or catalysts to change the gas composition. In looking at the literature, it is evident that one type of internal variation, the sensor operation temperature variation, is very popular in particular for conductometric metal-oxidebased sensors.40 This development has been fueled by the appearance of microhotplates with low thermal mass,41 which allow for millisecond-scale temperature variations, so that

Higher-Order Chemical Sensing

Figure 2. Scanning-electron micrograph of microhotplates. The suspended plates exhibit a polysilicon heater, an aluminum plane for homogeneous heat distribution, and electrodes for measuring the resistance of a semiconductor metal oxide. Reprinted with permission from ref 51. Copyright 1998 American Chemical Society.

the temperature variations are faster or on the same time scale as the chemical processes occurring during gas/metaloxide interaction. This enables effective gas discrimination via the use of analyte-specific dedicated temperature profiles. More details on temperature-modulation strategies can be found in section 9.2 of this article, as well as in the articles by Benkstein and Semancik42 and Nakamoto43 in this issue. A variety of hotplate structures has been used including membranes,18,41,44-47 spiderlike structures48-51 (see Figure 2), and bridgelike structures.52 The most recent developments in temperature-variable microhotplates include standalone CMOS-based microsystems featuring temperature-control loops, transistor heaters, digital circuitry, and standard interfaces, which allow for the application of any arbitrary temperature profile to three differently coated hotplates via standard software and a USB interface.53,54 Besides microhotplate-based systems, there have been also static approaches to temperature variation using 38 metaloxide sensor elements (the array features a noble metal doping gradient) on a 4 × 8 mm2 bulk silicon substrate equipped with 4 meander heaters, which create temperature gradients between 3 and 7 °C per mm in the array area, producing a temperature difference of 50 °C over the array.55-57 This system consumes up to 6 W at operation temperatures between 300 and 400 °C, which is 3-4 times the power consumption of microhotplate-based systems per detection spot or sensor. The gas reactions at the metal-oxide surfaces and, hence, the sensor selectivity or sensitivity patterns, are highly depending on the operation temperature:1,58-66 carbon monoxide is usually detected at lower operation temperatures using a tin-oxide-based sensitive layer, whereas higher temperatures are used for monitoring, e.g., methane. The variation of the operation temperature of a single sensor or a small set of sensors can lead to a degree of selectivity that would otherwise require arrays of fixed-temperature sensors and, thus, effectively extends the feature space of single sensors or small arrays. With regard to static, fixed-temperature sensors, the fast temperature variation of microhotplates, which generates a large set of “virtual” sensors, is clearly preferable due to the almost infinite number of possible and target-analytespecific temperature-variation profiles (sinusoidal, ramp, rectangular); the arbitrarily selectable temperature interval,

Chemical Reviews, 2008, Vol. 108, No. 2 567

within which the variations can be realized; and the massively lower instrumental complexity and overall power consumption. A temperature-modulation example is shown in Figure 3,47 which shows a sinusoidal modulation of the operation temperature of a single tin-oxide sensor between 200 and 400 °C that produces characteristic frequency-dependent resistance features. Resistance changes of the micromachined sensor upon exposure to CO, NO2, and a mixture of CO and NO2 in synthetic air at 50% relative humidity are displayed. By applying this temperature-modulation profile and by using fast Fourier transformation techniques for feature extraction and data evaluation, a single sensor could be used to qualitatively and quantitatively analyze a binary mixture.47 The authors ascribe the possibility to differentiate between the two gases, CO and NO2, to the different reaction kinetics of the two gases at the sensor surface and, in particular, to the presence of oxygen species at the surface at low temperature as a consequence of the fast temperature modulation. These surface oxygen species would not exist on the surface under equilibrium conditions at the lower temperatures in the cycling range (200 °C).47 Additionally, an example for external parameter variation will be given, which includes the use of a catalyst located upstream of the sensor array in the analyte gas inlet.67-69 The noble metal catalyst is heated to different temperatures and decomposes (oxidizes) the incoming analyte molecules or promotes chemical reactions of those. The resulting reaction products are then detected by an array of, e.g., electrochemical sensors.67-69 By varying the catalyst temperature, the sensor responses can be modified, and operation regimes can be optimized for the detection of specific target compounds. The catalysts included, e.g., rhodium or platinum filaments.67-69 A test analyte pattern for 8 different electrochemical sensors (4 CO sensors, 2 hydrogen sulfide sensors, 1 sulfur dioxide sensor, and 1 nitrogen monoxide sensor) and 7 different catalyst temperature steps is shown in Figure 4.67 The sensor-response patterns vary according to temperature and sensor type upon exposure to the 19 analytes. Again, the use of the catalyst generates virtual sensors and efficiently extends the feature space.

2.2. Dynamic Methods and Transient Signals The sensor-signal evolution over time can be used to extend the feature space of a sensor array. The information content that can be extracted from a transient signal f(t) is considerably higher than that from a steady-state signal f; whereas the steady-state signal is given by a single number f, the transient signal f(t) provides a series of measurement values at discrete time intervals t. More detailed information on transient analysis can be found in section 9 of this article and also in the article by Nakamoto43 in this review issue. An example of transient signals is displayed in Figure 5, which shows how the creation of exposure steps and transients of varying length can help to discriminate between methanol and ethanol using a polymer-coated capacitive device.70 It is noteworthy that the recording of transient signals stringently requires a dedicated gas manifold70,71 (permanent gas flows, crossover valves, small dead volumes between valves and measurement chamber, and small-volume chamber; for more information, see section 4.1 of this article), so that the recorded dynamics represent the real sensor dynamics and not those of the manifold or of the gas exchange in the measurement chamber. The time required for a full exchange

568 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 3. Sinusoidal modulation of the operation temperature of a tin oxide sensor between 200 and 400 °C (bottom) leads to characteristic frequency-dependent resistance features (upper part). Changes of the resistance (R sensor) of the micromachined sensor upon exposure to 50 ppm CO, 1 ppm NO2, and a mixture of 50 ppm CO and 1 ppm NO2 in synthetic air (50% relative humidity). Reprinted with permission from ref 47. Copyright 1997 Elsevier.

Figure 4. Analyte test data of a sensor array consisting of 8 electrochemical sensors detecting the analyte gas reaction products at 7 different catalyst temperatures (30, 100, 200, 500, 600, 750, 900 °C) of an upstream Pt filament. The analytes included a set of 10 alcohols, 2 ketones, ammonia, an amine, 2 sulfides, and 3 aldehydes, all of which are characteristic for fish freshness. Reprinted with permission from ref 67. Copyright 1994 Elsevier.

Figure 5. Sensor signals for a series of concentration steps of decreasing lengths from 160 down to 1 s. The capacitor was coated with a 4-µm-thick layer of poly(epichlorohydrin). The envelope of the response profile is highlighted in gray. It is analyte-specific and depends on the analyte absorption and desorption times in the respective polymer. Reprinted with permission from ref 70. Copyright 2006 American Chemical Society.

of the measurement chamber volume is often underestimated. A recording of the setup and manifold dynamics using sensors with very fast response times, e.g., sensors coated with very thin sensitive layers,70 is, therefore, recommendable.

A wealth of parameters can be extracted from a sensorsignal-versus-time representation and can be used as input to multicomponent-analysis or pattern-recognition algorithms. These parameters can include simple parameters like pulse heights, derivatives, and integrals calculated directly from the response curves or coefficients estimated from different models of the transient response like polynomial functions, exponential functions, or autoregressive models that have been fitted to the response curves.72 For most types of sensors (metal oxides and polymer-based sensors), the respective response times to reach equilibrium state are on the order of tens of seconds. In general, there are two different mechanisms determining the transient sensor response upon a sharp analyte concentration increase or decrease:73 (a) diffusion within the sensitive layer, whereat the diffusion processes in the measurement chamber should be significantly faster, and (b) surface or bulk reaction kinetics of the sensitive material. A nonlinear diffusionreaction model for thick-film metal-oxide sensors has been proposed by Gardner,73 and similar models have been used by other authors.74-77 The sensors included, in most cases,

Higher-Order Chemical Sensing

Taguchi-type thick-film sensors, and it was found that the response time is predominantly diffusion-limited (porous thick-film layer) and not reaction-limited.73 In many publications, polymer-based sensors are used, for which no reactions occur, so that bulk dissolution processes, i.e., analyte molecule diffusion into and out of the sensitivematerial matrix, determine the transient characteristics.71,78-80 Mass-sensitive devices, such as thickness-shear-mode resonators, have been used by several authors to identify, e.g., the aromas of alcohols,80 a variety of organic volatiles,71,79 or wine aroma compounds.78 The polymeric layers are usually <1 µm thick. The temporal or transient characteristics of sensor responses upon different analytes can also be induced by applying modulation techniques similar to the ones described in the preceding section on parameter variation: the exposure interval of the sensors to the analytes can be varied by actuating valves and by switching between analyte-loaded and pure carrier gas as displayed in Figure 5.70 The sensor signals in Figure 5 are given in Hertz, since on-chip electronics convert the minute capacitive signals into the frequency domain.81 A poly(epichlorohydrin)-coated (4 µm thickness) capacitive sensor has been used in these experiments, since measurements can be made very rapidly with this transducer (no extended gate time needed as for resonant sensors), and since the dynamic sensor signal neither relies on secondary effects like analyte-induced conductivity changes (conducting polymers) nor is influenced by accompanying effects such as analyte-induced film plasticization (acoustic-wave-based devices). Methanol and ethanol exposure steps of varying duration (from 160 s exposure duration down to 1 s exposure duration) were applied to the sensor. By applying long exposure intervals, all analytes reach absorption equilibrium and maximum signal amplitude, whereas for short intervals, this holds true only for fastdiffusing analytes. The sensor responses to methanol shown in Figure 5 reach saturation and sorption equilibria, even for relatively short exposure duration. For ethanol, which is a larger molecule with a smaller diffusion constant, the sensor signals do not reach equilibrium for medium or short exposure durations; as a result, the amplitude of the ethanol response begins to decrease much earlier in comparison to methanol. In conclusion, variations in the exposure interval can be used to facilitate the discrimination of analytes that belong to even the same homologous series.70 Additionally, modulation techniques to produce transients and to reveal the temporal sensor signal characteristics can be combined with any other parameter modulation in the preceding section, such as temperature modulations or the use of a catalyst.

2.3. Extracting Information in Different Domains In this last subsection on arrays of identical transducers, we will detail examples on how to extract, e.g., physical sensor data such as temperature or magnitude of flow from chemical sensors in addition to the chemical information they provide. A very simple multipurpose sensor/actuator structure offering three sensor operation modes (temperature, conductivity, and amperometric measurements) and two actuator operation modes (local heating and pH gradient control) was proposed by Langereis et al.82 and is displayed in Figure 6. The temperature can be measured along two different resistive paths between pads A and B or between C and D. By short-circuiting A and B as well as C and D, an

Chemical Reviews, 2008, Vol. 108, No. 2 569

Figure 6. Sketch of the multipurpose sensor structure: temperature between pads A and B or C and D, conductivity between pads A, B and C, D, and amperometric working electrode with all pads (A-D) short-circuited against an additional reference electrode. Reprinted with permission from ref 82. Copyright 1998 Elsevier.

interdigitated structure for conductivity measurements results, as has been shown for various potassium nitrate concentrations. By short-circuiting all four pads (A-D), the resulting electrode can be used for amperometric measurements against an additional reference electrode in a two-electrode setup, as has been demonstrated for different hydrogen peroxide concentrations. The heating can be performed by applying a voltage between A and B, or C and D, and by using one of the meanders or both as resistive heaters. One meander can be used as a temperature sensor. Finally, by supplying a current to all four pads (A-D) against an additional counter electrode, the aqueous environment can be electrolyzed and the local pH can be either increased or decreased (production of protons or hydroxide ions), as has been demonstrated by a coulometric titration of an aqueous acetic acid solution. In another approach, multifunctional modules have been realized on the basis of an array of ion-sensitive field-effect transistors (ISFETs).83-86 A schematic is shown in Figure 7.86 The setup is arranged around a flow-through cell hosting a so-called “hybrid module” (2 ISFETs, 1 Pt-wire counter electrode, and 1 gold generator electrode), and an Ag/AgCl reference electrode, all connected to external measurement equipment. The ISFET sensors are either pH-ISFETs with sensitive Ta2O5 films (55-58 mV/pH-unit) or enzymemodified ISFETs (penicillinase adsorbed on Ta2O5). The temperature is measured by a differential measurement of the two ISFETs operated at different working points. A flowvelocity measurement has been realized by using the generator electrode to electrochemically generate ions (H+ ions, electrolysis) and by measuring the ion concentration downstream upon their arrival at one of the ISFETs (pH-ISFET). By placing the ion generator electrode between two ISFETs, the flow direction and flow velocity can be determined. If the solution is not pumped through the setup, the diffusion of generated ions away from the generator electrode can be measured, and diffusion coefficients can be determined. An extended version including two of the “hybrid chips” in series has been detailed by the same authors, and, in the same paper, the use of an ISFET as a liquid-level sensor has been described.85 The multiparameter detection systems as described above have been used to detect, besides the physical parameters (temperature, flow, and diffusion), potassium ion concentrations (limit of detection (LOD) ) 5 µM) via valinomycincontaining poly(vinyl chloride) (PVC) membranes on the Ta2O5 gates85 and pH changes, since also the penicillin sensor (LOD ) 5 µM) detected the concentration of H+ ions resulting from an enzymatic penicillin hydrolysis.83-86 Examples for other enzyme-based ISFETs are given by the same authors.87 An open question that remains is, whether or not the concept of using chemosensors with their well-known drift- and stability problems for measuring physical param-

570 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 7. Schematic of a multiparameter detection system with a hybrid module in a flow-through setup: PCB, printed circuit board; GE, generator electrode; CE, counter electrode; ISFET, ion-sensitive field-effect transistor; PenFET, penecillinase-modified ISFET; RE, reference electrode; PC, personal computer. Redrawn with permission from ref 86. Copyright 2003 Elsevier.

eters (temperature, flow, and liquid level) will prove itself in the long run, in particular, since rather simple fully integrated temperature/flow units are commercially available. As will be seen later in this review, the co-integration of temperature sensors has been realized with the aim to enhance the reliability of the chemical sensor signals.81,88

3. Arrays and Systems Comprising Different Transducers In this section, different monolithic and discrete sensor arrays making use of more than one transduction principle will be detailed. The information gained from the different transducers should be as orthogonal or complementary as possible (see discussion in section 6.2), i.e., different analyteinduced changes in the properties of the coating materials, such as resistivity and work function changes, or different properties of the analyte molecules themselves, such as dielectric coefficients and mass, should be exploited. The application of different transduction principles for monitoring changes in the same or in highly correlated physical properties upon analyte dosage will not provide significantly more information than applying only one transducer. In this section, several multisensor arrays will be presented, which include discrete transducers and fully integrated complex microsystems. We have categorized the different systems with regard to the thermodynamic phase (gas and liquid) they are operating in and then subcategorized them according to the types of sensitive materials.

3.1. Metal- and Metal-Oxide-Based Gas Sensors In the case of metal-oxide-based sensors, several approaches have been made to extract more than only resistance/conductance or impedance values. A rather obvious possibility is to monitor gas-reaction-induced temperature changes on the heated stage of the metal-oxide-coated sensor. The reaction of, e.g., CO, methane, or alcohols at heated metal-oxide surfaces featuring catalytic metals such as Pd or Pt leads to changes in the heat budget, which either increase or decrease the temperature of the heated structure.89-91 To explain the occurring temperature effects, all processes involved in the gas interaction process and contributing to heat budget changes have to be considered: adsorption, dissociation, surface reaction, and desorption of the products. The particular thermal gas signature is dependent on these different contributions. CO was found to provide negative

Figure 8. Simultaneously recorded sensor temperature and resistance traces upon dosage of CO at concentrations between 7 and 200 ppm to a Pt-doped tin oxide sensor at 50% relative humidity. Reprinted with permission from ref 91. Copyright 1999 Elsevier.

calorimetric signals (temperature decrease) upon surface reaction with Pt- or Pd-doped tin oxide, though the oxidation reaction and formation of CO2 is generally exothermic.54,90,91 Exemplary resistance changes (resistance decreases) and simultaneously recorded temperature changes (temperature decreases) are displayed in Figure 8.91 The thermal signature can also be recorded for temperature-controlled microhotplate devices by monitoring the changes in the heating power (or in the source-gate voltage in the case of using a heating transistor54) for maintaining a preset temperature. The recording of work function data and catalytic activity in addition to the metal-oxide resistance has been reported on by several authors.92-94 A setup schematic for such measurements is shown in Figure 9 for the example of tin oxide.94 It includes a two-electrode resistance arrangement (Taguchi-type sensor), a Kelvin probe for the work function measurements (Kelvin probe relies on the displacement of one of the two surfaces in a periodic oscillation so that a sinusoidal current is produced, which is proportional to the

Higher-Order Chemical Sensing

Figure 9. Setup schematic for measurements of conductance, change in work function, and catalytic activity of tin-oxide-based sensors. The catalytic activity is assessed by measuring the concentration of the formed CO2 via spectroscopy or electrochemical sensors. Reprinted with permission from ref 94. Copyright 1990 Elsevier.

work function difference between the two surfaces), and spectroscopic methods or electrochemical sensors to, e.g., detect the concentration of CO2 formed through the catalytic reaction of CO at the tin-oxide-sensor surface. Since the setup and experimental efforts are quite substantial, this combination of methods has been used prevailingly to reveal details of surface reactions and sensor mechanisms for, e.g., CO.93 Conductivity or impedance measurements provide information on band bending effects and/or changes in the relative position of the Fermi level; the work function measurements additionally provide information on electron affinities. By conducting both types of measurements simultaneously, the different contributions to the overall work function and resistance characteristics can be sorted out, and mechanistic details of the surface and interface reactions can be revealed. Specific signatures for different gases like CO, methane, hydrogen, and H2S have been found and reported on by several authors.92,94 Because of the large cross-sensitivity of the metal oxides to water, differently doped metal-oxide sensors have been combined with a commercially available humidity sensor to yield a sensor system for more reliable carbon monoxide detection.95 Simply implementing a commercially available humidity sensor, however, may not always be a good solution. Humidity sensors may not perform well over a wide dynamic range, and they may exhibit large cross-sensitivities to other analytes. In an example of a metal-based gas sensor approach, catalytic palladium nickel metal resistors (thin-film metal resistors) have been combined with catalytic metal-gate fieldeffect transistors (FETs), FET-type heaters, and a temperature diode in a single-chip integrated system for the detection of hydrogen.96,97 The FET can detect hydrogen already at rather low concentrations (0.0001-1%), whereas the resistor is aimed at measuring in the range of higher concentrations (up to 20%). The system can be utilized in a multitude of sensing applications, and the respective calibration models have been presented.96

3.2. Polymer-Based Gas Sensors In analogy to the metal-oxide-based sensors above, different transduction principles have also been used for organic materials, such as conducting polymers, to elucidate sensing mechanisms. The question whether charge transfer or sorption characteristics drive the polymer-analyte interaction for the combination polypyrrole/methanol has been addressed

Chemical Reviews, 2008, Vol. 108, No. 2 571

by the simultaneous use of mass-sensitive thickness-shearmode resonators, a Kelvin probe to measure work function changes, and UV/vis spectrophotometrical methods to monitor optical absorption characteristics.98 Methanol transients have been investigated, and it has been found that analyte sorption is the driving force of the interaction.98 The simultaneous use of two different transduction mechanisms, chemomechanical transduction by thickness-shear-mode resonators and chemoelectric transduction in conductometric measurements (resistance measurement over a 50 µm wide gap on one of the faces of the quartz crystal) has been shown for polypyrrole films.99 Frequency decreases and resistance increases upon analyte sorption (various alcohols) have been observed.99 An array of eight solid-state field-effect-transistor-based sensors for simultaneous potentiometric and impedance sensing in the gas phase using the conducting polymer polyaniline has been studied by Polk et al.100 The sensor platform consisted of two different chips, a chemicalsensing chip (CSC), and an electronic service chip (ESC), with the latter intended to be flip-chip bonded to the center area of the sensor chip. Two different measurands, the work function and the impedance or resistance changes upon exposure of the polyaniline gate material to ammonia, have been simultaneously recorded, as is displayed in Figure 10. More recently, arrays of discrete chemical sensors relying on optical fibers (silica optical fiber, 1310 nm wavelength) and thickness-shear-mode resonators (10 MHz, AT-cut, quartz) with carbon-nanotube-based sensitive materials deposited using Langmuir-Blodgett techniques have been used.101,102 The authors find a significant improvement in the identification of the organic vapors (alcohols, acetone, toluene, and ethyl acetate) by combining the optical and mass-sensitive sensor responses.101,102 Several authors have combined capacitive and mass-sensitive sensors to detect sulfur dioxide,103 or capacitive, mass-sensitive, and calorimetric sensors (discrete devices104-106 or integrated microsystems81,88,107) to detect a wide range of organic volatiles. An example of a polymer-based integrated microsystem in CMOS technology (complementary-metal-oxide-semiconductor technology; standard fabrication technology for microelectronics) is shown in Figure 11. The single-chip gas-detection system comprises three polymer-coated transducers (capacitive, mass-sensitive, and calorimetric) that record changes upon analyte absorption. The absorption of the analyte in the polymeric coating alters the physical properties of the polymer film, such as its mass or volume, which is detected by the mass-sensitive cantilever; it changes the composite dielectric constant as detected by the capacitive transducer, or a certain amount of heat is generated during the absorption process (heat of analyte condensation or vaporization), which can be detected by the Seebeck-effect-based calorimetric transducer (aluminumpolysilicon thermopile). The three different transducers require different operation conditions, the mass-sensitive and the capacitive sensors rely on steady-state signals during sorption equilibria, whereas the calorimetric sensor needs sharp concentration gradients and a switching mechanism, since it only produces a signal upon sudden concentration changes (no signal at equilibrium state). A strategy to deal with these different operation requirements will be presented in section 4.2. The polymer-coated cantilever responds to any analyte dosing with frequency decreases (increasing oscillating mass), and the calorimetric sensor shows two transients per exposure, a positive one at the analyte

572 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 11. Micrograph of the single-chip CMOS gas sensor microsystem. The three different transducers (capacitive, masssensitive, and calorimetric) and the additional temperature sensor are marked. The driving and signal-conditioning circuitry of the different sensors and the digital interface are integrated on chip. The total size of the chip is 7 × 7 mm2. Reprinted with permission from ref 107. Copyright 2006 American Chemical Society.

Figure 10. (a) Schematic of the transducer setup to perform impedance and work function measurements. Work function and impedance response of one of the eight sensor modules coated with polyaniline upon exposure to different concentrations of ammonia (100 ppm to 1%) at 22 °C. The graph in (b) shows the potentiometric response (FET gate voltage changes), whereas the graph in (c) shows the impedance response. Graphics kindly provided by Prof. Jiri Janata, GeorgiaTech, Atlanta, GA.

concentration onset (analyte condensation into the polymer matrix) and a negative one upon switching off the analyte (vaporization of the analyte). The responses of the capacitive sensor can, in the case of thick polymer layers (>1.2 µm, larger than half the periodicity of the electrodes), be tuned according to the ratio of the dielectric constants of analyte and polymer. If the dielectric constant of the polymer is lower than that of the analyte, the capacitance will be increased; if the polymer dielectric constant is larger than that of the analyte, the capacitance will be decreased.108 This effect is shown in Figure 12 for two analytes featuring a larger (ethanol, 24.3) and smaller (toluene, 2.36) dielectric coefficient than that of the sorptive polymer (4.8) and has been

Figure 12. Sensor responses of capacitive sensors coated with a 1.4-µm-thick poly(etherurethane) layer (PEUT) upon exposure to various concentrations of ethanol and toluene. The analyte concentrations included 500-2500 ppm, up and down. The dielectric coefficient of ethanol (24.3) is larger than that of PEUT (4.8), so that positive capacitance changes occur upon ethanol dosage, and the dielectric coefficient of toluene (2.36) is lower, producing negative signals.

previously detailed and substantiated by simulations.109 It offers the possibility to pick polymers according to their dielectric properties in order to differentiate selected analytes. Moreover, a blinding-out of selected analytes (same dielectric coefficient as that of the polymer) and the use of polymer blends is possible. Another parameter that allows fine-tuning is the layer thickness,108 as the relative thickness of the polymer layer with respect to the extension of the electric

Higher-Order Chemical Sensing

field lines is decisive for the observed capacitance changes, as will be shown in section 3.3. All three transducer responses are simultaneously used to characterize the analyte or analyte mixture. Methanol, e.g., provides comparably low signals on mass-sensitive transducers because of its high saturation vapor pressure and low molecular mass. On the other hand, methanol has a dielectric constant of 33 and provides rather high signal intensities on capacitors. Drastic changes in the thermovoltages on the thermopiles are, e.g., measured upon exposure to chlorinated hydrocarbons, which have a low dielectric constant and, thus, provide only low signal intensity on capacitors. The simultaneous probing and recording of changes in different polymer properties upon gas exposure produces additional dimensions in the feature space and provides more comprehensive and complementary information about the analyte or the analyte mixture at hand. Since physisorption processes of organic volatiles in polymers are strongly temperaturedependent, a temperature sensor has to be integrated in such a system to enable reliable quantitative measurements. As a rule of thumb, a temperature increase of 10 °C decreases the fraction of analyte molecules absorbed into the polymer by ∼50%, which results in a drastic sensor signal reduction. The temperature sensor in the microsystem exhibits an accuracy of 0.1 K at operation temperatures between -40 and 120 °C. The sensor front-end circuitry that has been integrated on the chip includes all the sensor-specific driving circuitry and signal-conditioning circuitry. The analog/digital conversion is done on-chip as well. This leads to achieving a unique signal-to-noise ratio, since noisy connections are avoided, and since a robust digital signal is generated onchip and then transmitted to off-chip units.33,110 The sensor system has been used to demonstrate that the different transducers indeed provide complementary information on the various organic volatiles and that this information can be used for an analyte characterization according to the respective physical properties.107 The sensitivity values for a set of analytes and polymers have been evaluated. These sensitivity values have been normalized with regard to the partition coefficients (divided by the partition coefficients) so that all thermodynamic effects related to analyte absorption were accounted for and that the characteristics of the different transducers should then become clearly visible.107 The partition coefficient is a dimensionless thermodynamic equilibrium constant and is characteristic for a given volatile/ polymer combination; it is inversely proportional to the saturation vapor pressure or proportional to the boiling temperature and vaporization enthalpy.111 A selection of normalized sensitivity values is shown in Figure 13. The normalization of the sensitivity values with respect to the partition coefficient allows the detection or transduction process to be split into two parts (a) the absorption or partitioning, which is the same for all transducers for a given polymer, and (b) the transducer-specific part, which includes the measurand detected by the respective transducer such as sorption heat (calorimetric), molecular mass (mass-sensitive), and dielectric properties (capacitive).107 For a selected set of analytes, the characteristic properties of which are sufficiently different, there should be a systematic order in the normalized sensitivity values with respect to the transducer-specific measurand. This is obviously the case and is clearly demonstrated in Figure 13 for any given polymer: The order in the normalized sensitivity values of the calorimeter approximately reflects the

Chemical Reviews, 2008, Vol. 108, No. 2 573

Figure 13. Bar graphs representing normalized sensitiVity values at 30 °C. Four different analytes (n-octane, toluene, propan-1-ol, and trichloromethane) were detected with three different polymers (ethyl cellulose, EC; poly(epichlorohydrin), PECH; and poly(etherurethane), PEUT). The analytes have been ordered with regard to the decisive molecular property for the respective transducer: (a) decreasing heat of vaporization for the calorimeter, (b) decreasing molecular weight for the cantilever, and (c) decreasing analyte dielectric coefficient for the capacitor Reprinted with permission from ref 107. Copyright 2006 American Chemical Society.

vaporization heat of the respective analytes (propan-1-ol, 48.4 kJ/mol; n-octane, 41.6 kJ/mol; toluene, 38.0 kJ/mol; and trichloromethane, 31.5 kJ/mol), the order in the cantilever values is according to the analyte molecular mass (trichloromethane, 119.38 g/mol; n-octane, 114.23 g/mol; toluene, 92.14 g/mol; and propan-1-ol, 60.10 g/mol), and the order of the capacitive values reflects the analyte dielectric properties or dielectric coefficients (propan-1-ol, 20.45; trichloromethane, 4.81; toluene, 2.38; and n-octane, 1.95).

574 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 14. System schematic of the modular sensor system (MOSES): independent sensor modules (mass-sensitive, calorimetric, conductometric, temperature, and humidity) and gas intake and sampling units communicate via a digital bus with the overall system controller. Several alternative gas intake units such as a headspace sampler or a purge-and-trap system can be chosen. The system can be extended by additional modules. Redrawn with permission from ref 119. Copyright 1998 American Chemical Society.

The simultaneous recording from the different transducers causes a unique response pattern for each volatile compound. Because of their fundamentally different transduction principles, the sensors do (within experimental error) indeed respond to the diverse physical properties of the analytes, such as the molecular weight, the dielectric constant, and the heat of vaporization, so that they provide orthogonal information on a given analyte. To what extent the different responses are then also independent in the feature space, however, cannot be determined a priori (see also the respective discussion in section 6.2). Finally, combinations of the above-mentioned different transducers (mass-sensitive and optical112,113 or calorimetric114) coated with chiral receptors (e.g., cyclodextrins or amino-acid-derived compounds) dissolved in or bound to polymers have been successfully used to discriminate enantiomers.

3.3. Gas Sensor Arrays Relying on Different Transducer and Sensitive-Material Types Modular sensor systems including different types of polymer-based transducers, metal-oxide-based transducers, noble-metal-gate field-effect transistors, and electrochemical cells have been used as “electronic noses” by different groups to, e.g., qualitatively determine the quality of paper or packaging materials,115,116 to identify odors and flavors,116,117 or to assess food products.118 Please note that a more detailed article on the concept and performances of “electronic noses” is included in this issue.38 Holmberg et al. used an array of 10 noble-metal-gate CHEMFET devices (Pd, Ir, and Pt as gate metals) operated between 150 and 190 °C, 4 metal-oxide base conductometric sensors (Taguchi sensors), and an infrared-based carbon dioxide sensor to differentiate various types of cardboard papers.115 An examination of the sensor correlation matrix revealed that many sensor responses were strongly correlated

and that a subset of 7 sensors (5 CHEMFETs and 2 conductometric sensors) or, after data preprocessing, even of only 4 sensors (2 CHEMFETs and 2 conductometric sensors) showed the best discrimination performance. The authors concluded that the success in their application critically depended on the way of gathering the samples, the selection of sensors, and the data-preprocessing method.115 Data-preprocessing strategies will be covered in detail in section 7. The concept of a modular sensor system (MOSES) featuring an open architecture and the possibility to add new sensor modules was introduced at the University of Tu¨bingen in the late 1990s.119 A schematic of this system is displayed in Figure 14. Arrays of different discrete transducers are located in the respective sensor modules (mass-sensitive, electrochemical, calorimetric, and conductometric modules), which, along with temperature and humidity sensors and gas intake and sampling units, communicate via a digital bus with the overall system controller. The system can be extended by additional modules or modified in any arbitrary way to accommodate the sensors and sampling units needed for a specific application.119 According to the authors, the modularity offers the best prospects to select sensors and features from a potentially large variety and to optimize the individual sensors or components of the system. Moreover, it provides great flexibility in the feature selection for specific applications: The information content of each feature can be analyzed with due regard to the application at hand, and the total number of features can then be optimized and reduced accordingly (for more details about feature extraction issues, see section 11 of this review). In most applications, a metal-oxide-based chemoresistor array (8 sensors), operated at temperatures between 200 and 500 °C, and a polymer-based 30 MHz thickness-shear-mode resonator array (8 sensors), operated at room temperature,

Higher-Order Chemical Sensing

have been used. The polymer-based sensors are more stable in the long term and show less drift in comparison to the metal-oxide sensors. Moreover, metal-oxide-based and polymer-based sensors show considerable differences in the response time: the thickness-shear-mode resonators reach equilibrium values 10-15 s after the dosing of the respective analyte, whereas metal-oxide-based chemoresistors need at least 60-90 s, with both transients being slower.116 While this feature could be used to advantage as described in section 2.2, in most cases only the equilibrium signals or response maxima were evaluated. Principal-component analysis plots of an application example are shown in Figure 15.117 Principal-component analysis (PCA) is an orthogonal linear transformation method that arranges the data in a new coordinate system such that the greatest variance by any projection of the data comes to lie on the first coordinate (called the first principal component), the second greatest variance comes to lie on the second coordinate, and so on.10-12 The new coordinates are orthogonal to each other. PCA can be used for dimensionality reduction in a data set by retaining those characteristics of the data set that contribute most to its variance.10-12 Figure 15 shows the results of investigations on an artificially rancidized vegetable oil. To have reproducible and defined sample composition, a vegetable oil was contaminated with 100 ppm of different aldehydes (pentan1-al, hexan-1-al, heptan-1-al, octan-1-al, and nonan-1-al), since aldehydes have been identified as the key components causing rancid taste and smell of degraded edible oils. Figure 15 shows three principal-component analysis (PCA) plots for only the set of metal-oxide-based sensors (MOX), for only the set of thickness-shear-mode resonators (TSMRs), and for a combination of both sets. The added different aldehydes are indicated; “blank” means that the oil is in its original state and has not been manipulated. The metal-oxidebased sensors provide a discrimination of most oils according to the added aldehydes, but the noncontaminated oil and the nonan-1-al-contaminated oil cannot be differentiated. The polymer-based TSMRs cannot really discriminate the shortchain (C5-C7) aldehydes. However, the use of both arrays simultaneously leads to a clear separation and relatively small scattering within the different clusters.117 Other examples investigated with the same array configuration, and with an additional electrochemical module in selected cases, include textile materials,117 odors of plastic materials, coffees, olive oils, whiskey, and tobacco samples.116 In all cases detailed above, and in many other cases, the data analysis of sensor-array or “electronic-nose” data is limited to the drawing of PCA plots, which might be sufficient for easy problems or problems with a small data set, where the advantage of using a multitransducer array is rather large and obvious. PCA plots are not very representative for higher-dimensional measurement or feature spaces, simply because all the data are projected onto a two-dimensional plane irrespective of the original dimensionality. Thus, multitransducer arrays may also be beneficial even if this is not immediately apparent from the respective PCA plots. The important criterion is that a quantitative indicator of the array performance, such as the test set error for some classifiers, is lowered.118 Feature selection and the selection of good or optimized sensor subsets for a given application, in this case, the analysis of cured meat products (salami, ham, corned beef, salmon, roast beef, and different packaging materials), has been performed using an extended MOSES array (7

Chemical Reviews, 2008, Vol. 108, No. 2 575

Figure 15. Principal-component-analysis (PCA) plots showing the first two principal components, PC1 and PC2. Discrimination of homologous aldehydes (100 ppm) added to a vegetable oil matrix using (a) eight polymer-based thickness-shear-mode resonators (TSMR), (b) eight metal-oxide-based sensors (MOX), or (c) a combination of both arrays. The PCA plot of the MOX sensors (b) shows an overlap of the clusters of nonan-1-al contamination and the pure oil. In the case of the TSMRs (a), the clusters of the lowmolecular-weight aldehydes (pentan-1-al to heptan-1-al) overlap. Only by using both arrays simultaneously, all different contaminated oils and the pure oil can be discriminated. Reprinted with permission from ref 117. Copyright 2000 Elsevier.

polymer-based sensors, 8 metal-oxide-based sensors, and 4 electrochemical cells).118 The findings of the authors include

576 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 16. Monolithic multitransducer system including four polymer-based sensors relying on two capacitive and two gravimetric transducers, two metal-oxide-based conductometric sensors on temperature-controlled microhotplates (temperature modulation possible), the respective driving and signal processing electronics, and a digital communication interface. Reprinted with permission from ref 122. Copyright 2007 Elsevier.

that (i) subsets of selected sensors perform better than the whole array in most of the applications (test set error lowered by ∼25%), (ii) selected sensors from different classes (different transducer types) show significantly better performance than sensors selected only from a single class, and (iii) subsets that outperform the whole array may be as small as only two different sensors, such as one TSMR and one electrochemical cell.118 Several modular multitransducer systems based on discrete sensors are commercially available (see, e.g., refs 120 and 121). More details on feature selection can be found in section 11.2. A monolithically integrated multitransducer array in CMOS technology for the detection of organic and inorganic gases has been recently presented.122 The system comprises two polymer-based sensor arrays based on capacitive and gravimetric transducers (magnetically actuated cantilevers123), a temperature sensor, a metal-oxide sensor array located on microhotplates (thermal time constant ∼20 ms with metaloxide coating), the respective driving and signal processing electronics, and a digital communication interface (Figure 16). The chip has been fabricated in industrial 0.8 µm, CMOS technology with subsequent post-CMOS micromachining. The system has been developed in the framework of a “toolbox strategy” relying on microelectronics standard technology (CMOS), which was identified as the most promising platform technology to achieve major progress.124,125 The toolbox strategy was chosen as a consequence of the fact that the sensor market is strongly fragmented and that there exist a large variety of applications with different specifications and sensor requirements. The components of the toolbox, such as transducers, sensor modules, and circuit modules, can be developed one by one. Thereafter, specific components that meet the respective applications needs can be selected and assembled into a customized system. The simultaneous detection of organic and inorganic target analytes with the single-chip multitransducer system has been demonstrated in ref 122. Different organic volatiles have been discriminated according to their dielectric properties and molecular mass in analogy to the results presented in the context of Figures 12 and 13 in the preceding section. Another application possibility concerns the detection of carbon monoxide (CO) or other inorganic gases on a

background of changing humidity or alcohol content. For this scenario, the microhotplates and the capacitive sensor, which acts in this case as a humidity or alcohol sensor, have been used.122 The microhotplates can be covered with any metal oxide and can be temperature-modulated using any arbitrary waveform. The magnetically actuated cantilevers (Lorentz force123) can be used to monitor organic volatiles or interferents. Because of its high dielectric coefficient, humidity will have a major impact on any organic-volatile measurement of the capacitor. However, there is a possibility to measure organic volatiles with capacitive transducers even on a background of humidity or changing relative humidity. This method relies on the use of two differently thick polymer coatings on two identical capacitor structures and has been detailed previously:108 The signal difference of two capacitors with different layer thicknesses in the range of 0.8-4 µm is almost insensitive to water but retains sensitivity to low-dielectric-constant analytes like toluene or n-octane. Such differential or ratiometric methods have also been used, e.g., for conducting polymers,126,127 and constitute a very useful approach in dealing with interferents, cross-sensitivities, or low signal levels. It is very often more effective to purposefully select or deselect sensors or to use signal ratios or differential values instead of increasing the array size or the transducer diversity. In summary, this system offers great flexibility and can be used for various applications. The respective system configuration can be selected, and all parameters (sensor selection, differential or single sensor signal measurement, and temperature modulation of the hotplates) can be varied and set by means of standard software on a computer communicating with the digital control circuitry on the chip.122 At the end of this section on multitransducer systems and “electronic noses”, it is noteworthy that a shortcoming of many multisensor-array or electronic-nose papers, besides the predominant use of PCA score plots, is that the qualitative sensor results are not scientifically explained or substantiated by a thorough chemical gas-phase or headspace composition analysis, so that it is not clear, which compounds or which chemical effects lead to a discrimination of the different samples. A more detailed analysis of the contributions of

Higher-Order Chemical Sensing

the different sensors, and of the underlying surface reactions and physicochemistry of the different types of sensors, would be desirable. Varying humidity or alcohol content, e.g., may be more effective in changing the sensor array response to different food, perfume, or wine samples than the presence of aroma or odor components at very low concentration levels, which still are perceivable in human olfaction, but which are no more detectable using chemical sensors. Moreover, sample-to-sample variability, sample deterioration, and the strong influence of the sample preparation and sampling procedure on the sensor results, in particular for natural products, are often underestimated, and the corresponding information is missing in many papers. Information on sampling methods and how these influence sensor array results can be found in dedicated papers.128,129 There are also a number of multitransducer sensor systems, besides the already mentioned MOSES II sytem,130 commercially available, such as the GDA 2 (electrochemical cells, metal-oxide sensors, ion-mobility spectrometer, and photoionization detector) from Airsense Analytics,131 the FOX 4000 (metal-oxide sensors and polymer-based sensors: thickness-shear-mode resonators, conducting polymers) and the RQ Box from Alpha M.O.S,132 the Hazmatcad Plus (surface-acoustic-wave devices and electrochemical cells) and the CW Sentry 3G (surface-acoustic-wave devices and electrochemical sensor array) from Microsensor Systems,133 or microanalytics-based systems from RAE Systems.134

3.4. Liquid-Phase Chemo- and Biosensors A liquid-phase chemical microanalysis system aimed at applications in liquid-phase chromatography has been developed by Norlin et al.135 The system includes a multisensor chip, a micromachined flow-through cell, and optical fiber interfaces to monitor pressure, flow rate, temperature, conductivity, UV-absorption, and fluorescence. A schematic of the microanalysis system is shown in Figure 17a.135 The multisensor chip hosts integrated sensors for pressure, temperature, fluid flow, and conductivity; a flow-cell chip (silicon) defines the measurement chamber or liquid volume (5 µL) and features ports for the optical fibers to monitor fluorescence and UV-absorption. A close-up of the sensor structures is shown in Figure 17b.135 The substrate of the sensor chip is quartz. The temperature sensor is a simple Pt thermoresistor. The pressure sensor consists of a closed cavity under a polysilicon membrane; the pressure-induced strain in the membrane is measured with piezoresistors (doped polysilicon). For conductivity measurements, planar Pt electrodes (size: 500 µm × 1000 µm; 400 µm gap) are used. The principle of the fluid-flow sensor is to locally heat the fluid with a heating resistor (polysilicon) and to measure the temperature difference between two points up- and downstream from the heater using aluminum/polysilicon thermopiles (thermoelectric or Seebeck effect). Two laterally connected optical fibers enable UV light to be introduced into and collected from the liquid volume (path length ) 9 mm). The fluorescence measurements are performed by using a bundle of seven optical fibers connected laterally to the chamber or from below. The excitation light from a laser diode (wavelength ) 630 nm) is guided through the central fiber, and the six outer fibers transmit the fluorescent signal back to a photodetector. Initial results for the pressure sensor, the thermistor, and the flow sensor have been shown as well as conductivity measurements for NaCl solutions with concentrations between 0.001 and 1 mol/L and UV absorption signals for relative acetone contents between 0.1 and 1%.135

Chemical Reviews, 2008, Vol. 108, No. 2 577

Figure 17. (a) Schematic of the microanalysis system including a multisensor chip, a micromachined flow-through cell, and optical fiber interfaces. (b) Micrographs of the different single sensors, the temperature sensor, the pressure sensor, the conductivity electrodes, and the thermoelectric flow sensor. Reprinted with permission from ref 135. Copyright 1998 Elsevier.

A similar array of sensors including three temperature sensors (microelectronic proportional-to-absolute-temperature sensors), three pressure sensors (thin-silicon-membrane gauge-type sensors with piezoresistive readout), two ISFETs (silicon nitride and silicon oxide ISFET with Pt counterelectrode to monitor pH), and some basic circuitry (multiplexer, differential measurement electronics for chemical sensors, and temperature compensation for pressure sensors) was realized on a CMOS chip and is intended to be part of a drug delivery microsystem.136 The ISFET sensors were intended to control the pH value of the liquid to be delivered. Besides test results of the pressure and temperature sensors, the sensitivities of the ISFET sensors were determined to be 20-30 mV/pH for the silicon-oxide ISFET and 52 mV/pH for the silicon-nitride ISFET. A nonlinearity in the differential signal of both ISFETs was assessed to be due to the nonlinearity in the silicon-oxide ISFET.136 A multisensor array of discrete ISFETS, light-addressable potentiometric sensors (LAPS, p-silicon with SiO2 and Ta2O5 on top), and miniaturized ion-selective electrodes (ISE, p-silicon, SiO2, and metal electrode: 15 nm Ti, 30 nm Pt, and 250 nm Au) with a chalcogenide glass material (CdSAgIAs2S3) as the sensitive layer (200-1300 nm thickness) to detect heavy metal ions in aqueous solution was presented by Kloock et al.137 The sensitive material was deposited on the transducer structures by means of pulsed laser deposition, and the three different transducers were then compared in their sensitivity to Cd2+ ions. The sensitivities of all three potentiometric transducers are in the range of 22-25 mV per decade Cd2+, and the lower detection limit varied between 6 × 10-7 and 5.7 × 10-7 mol/L.137 The different transducers may, according to the authors, be combined in a future handheld “electronic-tongue” system.

578 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 18. Micrograph of the CMOS multiparameter biochemical sensor chip, which includes 6 ISFETS (1-6), an (amperometric) oxygen sensor (7), and a conductometric sensor (8a,b). The onchip circuitry includes an EPROM, a multiplexer and counter, a driver unit, a conductometric and potentiostatic circuit, and a heater. Reprinted with permission from ref 138. Copyright 2001 IEEE.

The benefit of using different transducers, however, is not obvious, since all three transducers provide very similar information. Significant differences exist in the transducer fabrication and signal readout complexity. A biochemical microsensor system aimed at continuous monitoring of ions, dissolved gases, and biomolecules in liquid phase, such as blood, has been presented recently (Figure 18)138 and is based on an earlier design by Gumbrecht et al.139,140 The eight integrated chemical sensors comprise six ion-sensitive field-effect transistors (ISFETs ) 1-6 in Figure 18), one oxygen sensor (7 in Figure 18) and one conductometric sensor (8a and 8b in Figure 18), all of which can be operated in parallel.138 An Ag/AgCl reference electrode is also integrated on the CMOS chip to obviate the need for external references. The eight sensors can continuously monitor ions, dissolved gases, and biomolecules via enzymatic reactions that produce charged particles. A flow channel (polyimide) restricts the liquid-phase access to the sensor area. The six ISFETs allow for direct contact of the electrolyte with the gate oxide. Either the gate oxide itself is pHsensitive or the ISFET can be used as a “SeVeringhaus”type pH-FET to measure dissolved carbon dioxide (detection of carbon dioxide via dissolution in water, formation of “carbonic acid”, and monitoring of the pH change). The gate oxide can also be covered with different ion-selective membranes to achieve sensitivity to a range of target ions, such as potassium. All six ISFETs or only a subset can be used. The idea was to make a standard chip to reduce manufacturing costs and to then modify the chip with selective coatings according to user needs. The integrated amperometric sensor can be used as a Clark-type oxygen sensor, which is based on a two-step reduction of gaseous oxygen in aqueous solution via hydrogen peroxide to hydroxyl ions. The conductometric sensor consists of two parallel sensors (8a), which share one common electrode (8b). A sinusoidal ac potential is applied to the electrodes, and the current, which depends on the solution composition (concentration of charged particles or ions), is recorded. The full system has been produced in a 1.2 µm single-metal, single-poly CMOS process, and the chip size is 4.11 × 6.25 mm2.138 The chip, operated at 5 V, hosts all driving circuitry of the sensors such as ISFET buffer amplifiers, a potentiostatic setup for the amperometric sensor, and the

Hierlemann et al.

circuitry necessary to perform a four-point conductometric measurement on-chip. In addition, the chip exhibits a temperature-control unit to keep the system temperature at a preset value (physiological conditions). This temperature-control unit includes a temperature sensor and a transistor heater. A single-bit EPROM (electrically programmable read-only memory) was implemented on-chip to make sure that the chip is used only once and then is disposed of, which is a crucial feature for medical applications.138 First tests including amperometric oxygen measurements, the assessment of potassium concentrations with ISFETs (by directly connecting the ISFET buffer to a plotter), and conductometric measurements with a buffer solution have been performed.138 Disposable electrochemical multisensor systems for fast blood analysis are marketed by, e.g., companies like Abbott (formerly I-STAT).141 Sodium, potassium, chloride, calcium, pH, and carbon dioxide are measured by ion-selectiveelectrode potentiometry. Concentrations are calculated from the measured potential through the Nernst equation. Urea is first hydrolyzed to ammonium ions in a reaction catalyzed by the enzyme urease. The ammonium ions are also monitored by means of an ion-selective electrode. Glucose is measured amperometrically. Oxidation of glucose, catalyzed by the enzyme glucose oxidase, produces hydrogen peroxide. The liberated hydrogen peroxide is oxidized at an electrode to produce an electric current, the intensity of which is proportional to the glucose concentration. Oxygen is also measured amperometrically. The oxygen sensor is similar to a conventional Clark-electrode. Oxygen permeates through a gas-permeable membrane from the blood sample into an internal electrolyte solution, where it is reduced at the cathode. The oxygen reduction current is proportional to the dissolved oxygen concentration. Hematocrit is determined conductometrically. The measured conductivity, after correction for electrolyte concentration, is related to the hematocrit.

3.5. Cell-Based Biosensors Whole living cells can be used to sensitively detect the presence of certain chemicals in their environment.142-148 The cell reacts upon exposure to a chemical in a cell-specific response, which can include changes in the cell electricalactivity pattern in the case of electroactive cells (neuronal cells and heart cells). The cellular responses can be monitored by a suitable set of different sensors, with the cell itself acting as a transducer and constituting a very sensitive and selective recognition system for different chemicals. It has to be noted that the cellular environment of living cells in in vitro situations differs considerably from their native environment in vivo. An example of a multiparameter sensor chip to monitor the cell-culture temperature, the cell-metabolism products, the cell electrical activity, and the cell adhesion to the sensor surface has been developed by a group at the University of Rostock.149-151a The aim was to develop a sensor system that allows for the measurement of chemical or metabolic parameters as well as electrical signals with the same sensor chip. The developed system, the concept of which is illustrated in Figure 19a, provides online monitoring of cellular reactions under well-controlled experimental conditions. The system includes cell-potential field-effect transistors (CPFET, sensitive gate areas of 6 × 1 µm2) and palladium electrodes

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 579

Figure 19. (a) Cell monitoring system concept: thermoregulated cell culture chamber with fluid handling system and different microsensors (ISFET, ion-selective field-effect transistor; ENFET, enzyme FET; ISE, ion-selective electrode; CPFET, cell potential FET; TD, temperature diode; CCD, charge-coupled device; SPR, surface plasmon resonance). Reprinted with permission from ref 142. Copyright 1999 Elsevier. (b) Extracellular recordings from one of the chip electrodes (∼40 superimposed neuronal action potentials). Reprinted with permission from ref 149. Copyright 2002 University of Prague. (c) Extracellular acidification measurements in a neuronal network on a silicon chip as performed with ISFETs in a flow-through system. The acidification was measured during the time when the pump was off. When the pump was on, the medium was completely replaced with fresh medium. Output signal of four ISFETs on one sensor chip (ISFETs 1 and 2 with UDS ) 0.2 V and ISFETs 3 and 4 with UDS ) 0.4 V ; IDS was 10 mA). The pump cycle was 5 min “pump on” and 10 min “pump off”. During the pump-off period, the pH of the medium decreased significantly due to the acidification through the presence of the cells. In the pump-on period, fresh medium is pumped through the chamber. Reprinted with permission from ref 142. Copyright 1999 Elsevier.

(10 µm diameter) to measure the electrical cell activity as shown in Figure 19b,149 a sensor to monitor the temperature of the cell culture, and ion-sensitive field effect transistors (ISFETs) to monitor the pH in the cellular microenvironment, recordings of which are shown in Figure 19c.142 The ISFETs allow for monitoring local acidification and respiration in in vitro cell networks. The interdigitated electrodes are used to measure the cell adhesion by means of impedance measurements.151,151a The quality of the contact between the electrically active cells and the transducers is of pivotal importance for applications in basic and biomedical research. According to the authors, impedimetric measurements using interdigitated electrode structures have been found to provide information on the cell density and number, the cell adhesion, and the cellular morphology, since an ac current between the electrodes is influenced by the presence and structural properties of living cells growing on these electrode structures. More details on how different chemicals trigger cellular responses of prevailingly electrogenic cells can be found in the literature.142-148,152

4. Operational Considerations for Higher-Order Devices 4.1. Setup and Manifold Considerations An often underestimated issue concerns the gas test setup and manifold for sensor measurements. The manifold for, e.g., any type of gas sensors relying on fast steep concentration gradients and interval analyte dosing (thermopile sensors), or for performing dynamic measurements and applying modulation techniques, has to be carefully designed, so that the dynamics of the transient sensor signal reflect the sensorspecific analyte diffusion and reaction characteristics rather than the gas flow dynamics of the setup and the measurement chamber. This means that all gas switching processes must be fast in comparison to the analyte-specific diffusion and reaction dynamics. To this end, a manifold and flow setup as shown in Figure 20 can be used. The most important features include a crossover flow architecture by use of a fast crossover 4-way valve, matched flow resistances of the

580 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 20. Schematic of the gas manifold as designed for fast transient signal recording.

two output gas lines of the 4-way valve, as well as a small tubing volume between the valve and the sensor measurement chamber.70,71 The crossover flow architecture implies that there are two input gas lines, one supplying pure carrier gas and the other supplying carrier gas with defined doses of the analyte, and two output gas lines, one leading to the measurement chamber and the other leading directly to the exhaust. This architecture offers the advantage that both input flows and both output flows are continuously flowing, and that the buildup time of a certain analyte concentration does not influence the dynamic sensor response. With the dosing line being routed to the exhaust (sensors exposed to pure carrier gas), the desired analyte concentration can be adjusted by means of flow controllers. After sufficient time for concentration stabilization, the crossover valve switches the dosing line to the sensors (carrier gas to the exhaust), which then experience a sudden steep concentration gradient. Using the crossover architecture, it is, hence, possible to rapidly switch between pure carrier gas and carrier gas containing a defined concentration of a certain analyte. The valve must be very fast, e.g., a pneumatically driven 4-way crossover valve with a switching time of <0.5 s, which is commercially available.70 The 4-way crossover gas switching functionality can also be obtained with a pair of appropriately connected 3-way valves, wired in parallel so that a single switch activates both valves simultaneously.153 The fast switching of the valves may generate pressure waves in the direction of the measurement chamber but also backward in the direction of the supply lines and the flowcontrollers. The system is open on the measurement-chamber side, and no effect on the sensor signal is usually observed. On the side of the flow controllers, additional measures have to be taken since pressure-wave-induced artifacts can otherwise be observed: flow controllers are very sensitive to pressure transients occurring either at their inlet or their outlet, so that an additional empty glass bubbler (large diameter and volume) has to be mounted in between the flow controller for the carrier gas in the dosing line to eliminate these artifacts (Figure 20). The glass bubbler acts as an expansion chamber or accumulator commonly used in pneumatic systems. Moreover, when switching the 4-way valve, any pressure difference in the two output flow lines affects the gas flow

Figure 21. Operation mode as developed for micromachined multisensor chips: operation states of valves and corresponding gas concentrations in the chamber (lines 1-2), timing of the signal recording for the different transducers, and resulting sensor signals (lines 3-6). For details, see text.

dynamics and, consequently, influences the preset concentrations. Therefore, the output line without measurement chamber has to be designed to exhibit a flow resistance as similar as possible to that with the measurement chamber, and the two output lines of the 4-way valve should feed into the same exhaust line after the measurement chamber. The overall gas volume between the valve and the sensors has to be minimized, taking into account the target overall gas flow. The time span between switching the valve and the moment, at which the gas reaches the sensor, should be as short as possible. The overall flow rate also may influence the dynamic sensor signals if it is rather low or may influence the operating temperature of high-temperature sensors if very high. The optimum flow rate for a given flow setup has to be assessed in prestudies.

4.2. Multitransducer Operation Example Since multitransducer systems include different types of transducers that require different operation regimes, such as the recording of steady-state or transient signals, it is necessary to apply dedicated operation protocols, which enable reliable qualitative and quantitative measurements and allow for the extraction of a maximum of information. An example strategy will be described here that has been developed for the polymer-based multitransducer unit described in section 3.2 and that meets the operational requirements of the different transducers. The signal baseline is established by purging with filtered ambient air or clean air from the gas manifold. The operation state of the valve and the resulting analyte concentration in the measurement chamber is displayed in Figure 21, which additionally shows the timing of the signal recording from the different transducers as well as prototypical sensor signals. The gas manifold that can be used to perform the respective measurements has been described in the previous section (4.1).

Higher-Order Chemical Sensing

Line 1 is used to indicate the valve status. “0” represents the basic state of the valve, when pure carrier gas flows through the measurement chamber. In state “1”, fractions of the carrier gas pass one or more vaporization units or bubblers, and analyte molecules are present in the gas stream: analyte-loaded gas is flowing over the sensors. In line 2, the corresponding analyte gas-phase concentrations are displayed. In the beginning of a measurement sequence, there is no analyte gas in the measurement chamber, which is purged with pure carrier gas. Baseline signals of the capacitive and mass-sensitive transducers are recorded, the measurement timing of which is displayed in line 3. The valve is then switched to the analyte line for, e.g., 30 s, which leads to an instantaneous analyte concentration increase since analyte-loaded gas is now flowing through the measurement chamber. Equilibrium state capacitive and mass-sensitive signals in analyte-loaded air are then recorded. The resulting sensor signals (mass changes or capacitance changes) are schematically shown in line 4. The valve is then switched back to pure carrier gas, which generates a sharp decrease in analyte concentration. The last switching would not be necessary for the equilibrium-based sensors, but it is necessary to get the second calorimetric transient, as shown in line 6. As already described in section 3.2, the calorimetric sensor relies on transients and provides signals exclusively upon concentration changes. Therefore, the calorimetric recording has to be performed at high temporal resolution (1 kHz) in two short intervals covering both flanks of the concentration signal (line 2), i.e., at the maximum gradient of the analyte concentration. The two transient signals of the calorimetric transducer (positive upon analyte absorption, negative upon analyte desorption) are displayed in line 6. Usually, the areas of the respective peaks (absorption and desorption peaks) are integrated and then averaged to obtain the final value.

5. Sensor-Based Microanalytical Systems In this section, we briefly describe more complex miniaturized analytical systems based on gas sensor arrays, which resemble most closely higher-order analytical instruments. The sensor arrays act as detector units in those systems. In most cases, preconcentration (see also the article of Grate et al. in this issue154) and/or separation stages have been combined with the sensor array for better analytical performance of the resulting system.155-166 The preconcentration stages lower the detection limits for the sensors through enrichment of the target analytes in a sorptive matrix. After some time allowed for the analyte enrichment, a sharp heating pulse is applied to the sorptive material so that all the analyte molecules, which were absorbed during a userdefined time span, desorb at once. In this manner, considerably higher analyte concentrations hit the subsequent separation (micro-GC) and/or detection unit (sensor array).156,157 The preconcentration stages can be classified into two groups: (i) dynamic headspace or purge-and-trap systems and (ii) solid-phase microextraction methods using fibers coated with absorbing materials.167 Nanoporous carbon, solgels, ceramic matrices, polymers, and commercial packing materials are commonly used as absorption matrixes. In comparison to sensors without preconcentrators, improvements in the lower detection limit range between 1 and 3 orders of magnitude can be achieved, so that the lower ppb range (relevant for many applications) becomes accessible. Preconcentrators with commercial material such as Tenax

Chemical Reviews, 2008, Vol. 108, No. 2 581

TA have been used in conjunction with mass-sensitive Rayleigh surface-acoustic-wave devices to detect the BTX compounds (benzene, toluene, and xylene) in the low ppm and sub-ppm range,161 or with thickness-shear-mode resonators (temperature of the preconcentrator was modulated) for apple and banana flavors.164 A two-step preconcentrator to enrich organic volatiles and to remove water vapor from the sample air (first and second stages feature a hydrophobic coating, which enriches organic volatiles and which lets water pass) was used for analyzing exhaled air or human breath with the help of carbon-black/polymer-coated chemoresistors.162 Low levels of organic volatiles in human breath could be detected.162 Several groups have used sensors as detectors at the end of standard desktop chromatographic units.165,166 The use of bulky chromatographic units to boost the discrimination performance of small and cheap sensors, however, defeats the purpose of having small and portable units, in particular since the performance of the sensors is, in most cases, not superior to that of a standard flame-ionization detector (FID). It also has been proposed to combine a metal-oxide-based chemoresistor (zinc-oxide pellet) with a 80 mm long fused silica capillary to record diffusion-dependent sensor responses and to identify certain target analytes.168 Miniaturized gas chromatographic units were first presented in the late 1970s169 and, then, in the mid-1990s.170 In most cases, they have been realized as spirals (column lengths ) 0.6-0.9 m; widths ) 100-200 µm; and depths ) 200-400 µm) micromachined into a planar silicon substrate (∼1 cm2) with a glass plate bonded to the silicon substrate to close the column (see Figure 22 and Figure 24). More recently, rather long (up to 3 m) square-type micromachined columns on 3.3 × 3.3 cm2 dies have been presented.160 Within this review, we will not give more details on micromachined gas chromatographic units but will describe two approaches to miniaturized, sensor-based, lowpower microsystems potentially capable of comprehensive environmental vapor analysis. A hybrid microsystem developed in a broad-based effort at the University of Michigan155-157,159,160,163 contains the following components (Figure 22):160 a sample inlet with particulate filter, an on-board calibration-vapor source, a multistage preconcentrator/focuser, a dual-column separation module with pressure- and temperature-programmed separation tuning, an array of microsensors for analyte recognition and quantification, and a pump and valves to direct the sample flow. MEMS (micro-electromechanical system) processing technologies have been used to fabricate the system with the ultimate goal of creating a fully operational micro-instrument that occupies only 1-2 cm3, requires an average (battery) power of just a few mW per analysis, provides rapid determinations of mixtures of at least 30 vapors of arbitrary composition at low- or sub-part-per-billion (ppb) levels, has an embedded microcontroller, and can be remotely interrogated through an RF-MEMS (RF ) radio frequency) wireless communication link.160 The calibrationvapor source, shown in Figure 22a, is designed to generate calibrant vapor at a constant rate by passive diffusion from a liquid reservoir. Analysis of this ‘‘internal standard’’, along with vapors captured from the environment, provides the means to compensate for aging, drift, or other factors that might affect analytical performance. The calibration-vapor source is a 3-layer structure, whose base contains a deep porous-Si (PS) reservoir for retaining the volatile-liquid calibrant, a glass spacer layer with a central aperture that

582 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 22. Schematic and components of the Michigan analytical microsystem: (a) calibration-vapor source before (left) and after (right) assembly; (b) 3-stage adsorbent micropreconcentrator prior to loading and sealing (top left), with close-up SEM images of each section loaded with adsorbents (lower left); (c) 3 m separation-column chip (lower right) with close-up views of the channel cross sections prior to (top right) and after (top left) sealing; (d) detector assembly with 4-chemiresistor-array chip (right), Macor lid (white square structure), and sealed detector with connecting capillaries mounted on a custom fixture (left). Reproduced with permission from ref 160. Copyright 2005 Royal Society of Chemistry.

Figure 23. Seven-vapor chromatograms of the Au-6-phenoxyhexane-1-thiolate-coated chemiresistor showing the effect of micropreconcentrator-desorption and column-elution flow rates and flowrate ratios on resolution and analysis time. For split-flow operation, a portion of the flow through the micropreconcentrator was diverted around the separation column: (a) no split flow, 1.3 mL/min; (b) 4:1 split ratio, 5.1 mL/min (micropreconcentrator)/1.3 mL/min (column); (c) 8:1 split ratio, 5.8 mL/min (micropreconcentrator)/ 0.75 mL/min (column). Vapors: 0, water; 1, toluene; 2, n-butyl acetate; 3, m-xylene; 4, n-nonane; 5, mesitylene; 6, n-decane; 7, octamethylcyclotetrasiloxane. Reproduced with permission from ref 160. Copyright 2005 Royal Society of Chemistry.

defines the headspace region, and a Si cap that contains an etched diffusion channel and exit port. The three-stage micropreconcentrator (Figure 22b) is designed to capture organic vapors quantitatively and to thermally desorb them into a much smaller volume, thereby increasing the effective concentration to facilitate detection as well as providing a sharp injection plug to facilitate highspeed chromatographic separations.156,157 The preconcentrator

is manually packed with porous, carbon-based adsorbents (total mass ≈ 5 mg) in order to increase the specific surface area. Adsorbents are loaded through a stencil mask to maintain segregated sections of each material. As can be seen in Figure 22c, a large single-substrate column was used, which consisted of a convolved squarespiral silicon channel (150 µm wide, 240 µm deep, and 3 m long) on a square die, 3.3 cm on a side, capped with an anodically bonded Pyrex glass cover plate. Figure 22c shows a sealed column, with the inset providing a closer view of the channel cross section. A polydimethylsiloxane stationary phase (thickness of ∼1 µm) was employed and was deposited dynamically from a dilute pentane solution. The detection unit (Figure 22d) included an integrated array of four chemiresistors, designed to produce a set of partially selective responses to vapors eluting from the separation column. The response pattern can then be combined with the retention time to identify the vapor, and the magnitude of the responses from the sensors can be used to quantify the vapor concentration. Each sensor consists of 20 pairs of interdigital Au/Cr electrodes (1.4 mm long, 15 µm wide, and spaced by 15 µm) on a Si substrate. Intersensor spacings are ∼1 mm. The chemiresistor array employs interfacial films of Au-thiolate monolayer-protected nanoclusters, whose resistances are shifted to different extents upon vapor sorption.171,172 There are many parameters in this complex system that influence its performance and have to be optimized with due regard to the target analytes and the analysis problem, such as gas flow velocity, flow rates, temperatures and temperature programs of GC column and preconcentrator, preconcentration time, and column and preconcentrator materials.

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 583

Figure 24. Optical photographs of a monolithic microanalysis system. (a) Front side surface micromachining is shown: dual pivotal-plate resonator sensors are evident as are multiple oblong through-wafer access ports, a preconcentrator in the lower left, and a gas chromatography resistive heater and circular coating ports in the lower right. (b) Reverse side deep etching: the spiral GC is on the lower left. (c) Close-up of the pivot-plate resonator, rotated 90° with respect to images (a) and (b). The direction of the magnetic field, set up by miniature magnets, is indicated by an arrow. Current lines follow the perimeter of the paddle and the two torsional suspension beams. Reprinted with permission from ref 158. Copyright 2006 IEEE.

For sample collection, 0.25 L of analyte-loaded air was drawn at 25 mL/min through the preconcentrator, where the vapors were trapped. The preconcentrator was then heated to 280 °C, and the desorbed vapor mixture was passed to the column and sensors at flow rates below 10 mL/min for separation and detection. A sample measurement is shown in Figure 23.160 Figure 23a shows a 7-vapor chromatogram from one of the sensors (Au-6-phenoxyhexane-1-thiolatecoated sensor), illustrating that symmetric peak shapes and adequate separations can be achieved at 1.3 mL/min with the column temperature ramped from 25 to 80 °C at 1.4 °C/s. The separation required only 75 s. For this test, the entire desorbed sample volume was transferred to the separation column. A fraction of the sample flow can be diverted around the column and sensor array, since it was shown that sharper injection pulses are obtained at higher desorption flow rates (i.e., 0.3 mL/min) through the preconcentrator, and since it was also shown that the flow restriction imposed by the 3 m column length constrains the maximum flow rate through the column to values of <3 mL/min. The split ratio was adjusted by varying the length, and thereby the flow resistance, of the bypass. Figure 23b shows the chromatogram obtained with a 4:1 ratio: the preconcentrator flow rate (5.1 mL/min) was four times that passing through the column (1.3 mL/min). Retention times were increased slightly and all peaks became sharper and better separated than without a flow split (compare Figure 23a). Increasing the split ratio to 8:1 and reducing the column flow rate to 0.75 mL/min yielded the chromatogram shown in Figure 23c. The separation is improved substantially due to the narrower injection band and due to the operation of the column at a lower velocity. However, the time required for the separation increased by ∼50%, and the magnitudes of all peaks are reduced because of the smaller fraction of the desorbed sample being passed through the column and because of the slight increase in dilution associated with the higher desorption flow rate.160 The system is capable of separating, recognizing, and quantifying mixtures of moderate complexity (e.g., 11 vapors) in <1.5 min. The needed preconcentration time ranges from ∼1 min (industrial work places, analyte concentration in the single ppm range) to 10 min or more for less-contaminated office or residential environments (ppb range).160 Development efforts in the field of microanalytical systems have taken another step forward in devising extremely compact monolithic systems with all components realized on the same silicon substrate.158,173 There are advantages to

hybrid systems, such as modular replacement of components, and the fact that the thermal isolation of the individual components is much easier to accomplish in hybrid systems, which is important, since the individual components often have different operation temperatures. However, the manifolds previously described often have cold transfer lines interconnecting the components. This can cause collection or condensation of analyte in the transfer lines, ultimately reducing sensitivity. Although the size of the manifold channels may be subminiature, there is still excess dead volume present. Moreover, the assembly of the hybrid system can add to the cost of the completed system, and physical isolation strategies and system timing can be used to mitigate thermal isolation issues for the monolithic system.158 A monolithic “MicroChemLab” system on a 5 × 6 mm2 size chip developed at the Sandia National Laboratories, Albuquerque, NM, is shown in Figure 24. The length of the spiral GC column is 8.1 cm in one design and, in another, 11.8 cm. The 8.1 cm long, 50 µm wide GC column is integrated with a preconcentrator and a novel magnetically actuated pivot-plate resonator sensor pair. The pivot-plate resonator is potentially more sensitive than the magnetically actuated flexural-plate-wave transducer used before158 and is also actuated by making use of Lorentz forces. The pivot-plate resonator consists of a central paddle supported by two torsional beams. An alternating current passing through the transducer lines interacts via the Lorentz force with an orthogonal, in-plane magnetic field, causing the paddle to oscillate around the torsional beams (Figure 24c).158 The monolithic chip design also incorporates a surfacemicromachined bypass valve, intended to switch the flow between the sampling and separation/detection portions of the overall analysis system. The valve consists of an electrostatically actuated silicon nitride flap situated over the bypass port. Machined glass lids, baseplates, and packages have been fabricated to coat and test the monolithic system, which is work in progress.158

6. Are More Sensors Better? In the introduction to this chapter, we suggested that increasing the measurement-data dimensionality, either by adding more sensors or by extracting additional features, could offer substantial benefits with respect to the analytical capabilities of the instrument. The issue of whether or not “more sensors are better” is an ongoing debate in the chemical-sensor-array community.174-177 Providing a general

584 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

answer to this question is difficult, if at all possible. On the one hand, the use of multiple sensors is central to the “electronic nose” paradigm; arrays of cross-selective sensors (i.e., first-order arrays) do provide more analytical capability and power than the individual sensors. Further, while it is evident that adding “orthogonal” sensors can improve the selectivity of the instrument, the use of redundant sensors can also be beneficial, e.g., in terms of increasing the faulttolerance and sensitivity of the array. On the other hand, increasing the dimensionality of the feature space can have detrimental effects in terms of increased computational complexity, higher levels of noise, and an increased risk of overfitting (i.e., the modeling of noise in the training set), even if the additional dimensions are orthogonal. In the following subsections, we will provide a more detailed treatment of these issues.

6.1. Characteristics of High-Dimensional Vector Spaces Humans have an uncanny ability to perceive patterns in the three-dimensional world in which we live. We can understand speech (a first-order signal) under much degraded acoustic conditions, recognize a familiar face (a second-order signal) at a large distance, or appreciate the gracefulness of a ballerina (a third-order signal) already upon a short glance. Unfortunately, our capabilities in the three-dimensional space do not scale up to higher dimensionality. To illustrate this point, we will highlight a few geometric and statistical characteristics of high-dimensional hyperspaces that defeat intuition.178 Consider a hypersphere of radius r, defined in d dimensions. It can be shown that the volume of the hypersphere is given by179

Vd(r) )

2πd/2rd d dΓ 2

(1)

()

where Γ() is the gamma function, an extension of the factorial function to complex and noninteger numbers. Using elementary calculus, the fraction of this volume that is contained in an outermost shell of thickness, , can be computed as

fd )

Vd(r) - Vd(r - ) Vd(r)

)

rd - (r - )d r

d

(

)1- 1-

d r (2)

)

It then follows that, as the dimensionality of the hypersphere increases, so does the fraction of the volume concentrated in the outermost shell. Likewise, it can be shown that the volume of a hypercube tends to be concentrated in the corners. Thus, high-dimensional spaces tend to be mostly empty, and the data tend to be concentrated in a low-dimensional manifold. The latter suggests that data can be projected onto a low-dimensional subspace without a significant loss of information. Unfortunately, finding an optimal projection becomes increasingly harder with more dimensions. According to the central limit theorem, any sum of independent and identically distributed random variables tends to be more normally distributed than the variables themselves, even if these are markedly non-Gaussian. Thus, as the dimensionality of the feature space increases, lowdimensional projections of the data have the tendency to

Figure 25. Performance of a statistical pattern classifier as a function of the feature-vector dimensionality, n, for a fixed dataset size, m. Reprinted with permission from ref 182. Copyright 1968 IEEE.

become normally distributed, which may destroy any natural clustering of the data in a high-dimensional space.178 In addition, computation in higher-dimensional spaces increases the amount of data that is required to effectively train the models. It has been shown that the number of training samples should grow linearly with the feature space dimensionality for linear models,180 in a quadratic fashion for Gaussian models, and exponentially for nonparametric models.181 What this means is that, for a defined dataset size, there is an optimum number of dimensions, beyond which the performance degrades;182 see Figure 25. Therefore, on the basis of statistical considerations, and assuming a given number of training samples, the smallest number of sensors that can provide the necessary chemical discrimination is better.

6.2. Orthogonality versus Independence One of the potential advantages of higher-order sensor arrays, such as arrays based on different transducers, is their ability to produce “orthogonal” features.88,183 In this context, two features are said to be orthogonal if they convey information about, e.g., different physicochemical properties of the target compounds. Thus, orthogonality is a geometric property defined in chemical space, where each dimension represents a unique molecular chemical or physical property. It is important to note, however, that sensor orthogonality is neither necessary nor sufficient to ensure higher analytical power of an array. In fact, the addition of an orthogonal sensor may even lower the performance of the array through the introduction of noise, if the information provided in the respective added feature is irrelevant to the discrimination and quantification of target compounds or, worse, if the feature is sensitive to the chemical background or to interferents. Consider, for instance, the problem of developing a new sensor array for CO, an example given by Stetter and Penrose.177 One may be tempted to combine an optical infrared detector with a metal-oxide-covered conductometric device. Both sensors can be considered to be orthogonal, since the IR sensor measures molecular vibrations and the metal-oxide-based sensor relies on electronic effects. By adding a metal-oxide-based sensor, however, we may obtain little additional information. More importantly, since metaloxides are very sensitive to a broad variety of gases, we may have rendered the array more vulnerable to interferences. On the other hand, two sensors are said to be independent if the knowledge of the response of one sensor upon exposure to a target compound does not provide any information about

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 585

Figure 26. Array performance as a function of the relative proportion of broadly tuned (large) and narrowly tuned (small) sensors. The results show that the maximum performance is obtained when the array contains a mixture of “large” and “small” sensors. Dashed vertical lines indicate the performance of the array when all sensors have the same degree of selectivity. Reprinted with permission from ref 186. Copyright 2002 Oxford University Press.

the response of the other sensor.183a In other words, independence is a statistical property defined in the feature space, where each dimension represents a certain feature or sensor.184 Thus, in contrast to orthogonality, sensor independence cannot be ensured unless the sensor array has been designed for an a priori known set of target analytes. In this context, more sensors are better, in the sense that increasing the number of sensors in the array also increases the odds that a subset of independent sensors can be found for a wider range of applications.

6.3. Cross-sensitivity and Diversity The inherent cross-sensitivity of chemical sensors is commonly seen as both beneficial, to the extent that it broadens the detection range of the array, and detrimental, in that it makes the instrument vulnerable to interferences. Common sense seems to indicate that, if one were able to develop sensors that are specific to only one of the target compounds, the resulting array would be more accurate than a similar array of cross-sensitive sensors. Quite the contrary has been suggested by a number of theoretical results in computational neuroscience (see, e.g., Brown and Ba¨cker185 and references therein) and machine olfaction.175 According to these studies, arrays of broadly tuned sensors provide a more accurate representation of a stimulus than arrays of highly specific sensors, assuming that the stimulus is of high dimensionality (e.g., large number of target compounds). In

fact, using computational models, Alkasab et al.186 have estimated that an optimum configuration should include arrays in which each individual sensor responds to 25-35% of the target compounds. Several authors (see, e.g., refs 187 and 188) have also reported that the overall performance of large sensor arrays can be improved by allowing the individual sensors to have different degrees of selectivity by combining, e.g., broadly tuned and narrowly tuned sensors; see Figure 26. This theoretical result is particularly relevant in the case of higher-order devices, since different transduction principles and sensitive layers can be combined to produce sensor arrays of very distinct and diverse sensitivity and selectivity patterns. Experimental results on arrays combining selective and partially selective sensors are also consistent with the above theoretical predictions.189 Therefore, from this perspective, one can argue that more sensors are better, provided that the respective selectivity profiles increase the diversity in the array.

6.4. Multiple Roles of Redundancy Biological olfactory systems rely on a diverse and highly redundant population of sensory neurons to gather information about the stimulus; see ref 190 and references therein. Depending on the animal species, it has been estimated that 100-1000 different types of receptors are involved in the coding of chemical information at the olfactory epithelium. Each type of receptor is expressed on a large number of

586 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 27. Effect of computing the average response of an array of homogeneous tin-oxide sensors: the variance of the array output decreases with the square root of the number of sensors. Reprinted with permission from ref 192. Copyright 2002 Wiley-VCH, Weinheim.

sensory neurons, with each neuron being specialized on one or a few receptor types. This massive degree of redundancy serves multiple purposes. First, it allows the system to cope with the massive turnover of sensory neurons, since the distribution of sensory neurons can be considered to be stationary over time with respect to the developmental stage of the individual neurons. Second, the integration of the response from multiple neurons can be used to average out uncorrelated noise, which effectively increases the sensitivity of the instrument. More specifically, theoretical estimates and experimental results show that signal integration improves the detection threshold by a factor of xn, where n is the number of identical (or identically responding) sensors in the array;191 this result is illustrated in Figure 27 for an array of nominally identical tin-oxide sensors.192 Third, by relying on a large population of sensors, the system becomes more robust and fault-tolerant. Thus, from this perspective, more sensors (of the same type) are better, provided that their noise characteristics are also independent.

7. Data Preprocessing The term “data preprocessing” broadly refers to any transformation performed on the raw sensor data prior to building the main analysis model. The goal of data preprocessing is typically two-fold: (i) reduction of noise or removal of information that is known to be irrelevant to the analysis problem and model (e.g., interferences, drift) and (ii) numerical preconditioning of the data, such as scaling or normalization.192 The selection of a suitable datapreprocessing approach can have a significant impact on the performance of the analysis model,193 but, unfortunately, the data-preprocessing approach is highly dependent on the sensor technology (e.g., metal-oxide chemiresistors vs quartz crystal microbalance), the type of analysis (e.g., classification vs regression), the type of model (e.g., nearest-neighbors vs multiway), and the type of noise present in the data (e.g., baseline drift vs concentration effects). Thus, there is only a handful of general guidelines as to how to select the appropriate preprocessing technique (see, e.g., ref 194), and, in practice, a suitable technique must be selected empirically.195-197 Data preprocessing is particularly important in the case of higher-order sensor arrays, since these devices can employ

a number of different transducer types and/or take advantage of the dynamical responses of the sensors. In the first case, a separate preprocessing technique may need to be applied to each type of transducer and then globally to the multivariate response of the array. In the latter case, it is important to ensure that the preprocessing technique does not destroy the higher-order structure of the data (e.g., trilinearity).198 Preprocessing techniques can be grouped into three categories: (1) baseline correction, (2) scaling, and (3) dynamic feature extraction. Baseline correction and scaling will be reviewed only briefly here, since they have been extensively covered in the literature.192-196,199-201 Somewhat related to baseline correction is the issue of drift compensation. Due to the potentially large impact of drift on the analytical performance of the sensor array, computational methods to handle sensor drift will be treated separately in section 8. An emphasis will also be placed on dynamic feature extraction, since it constitutes one of the easiest ways to realize “higher-order” sensing; dynamic techniques will also be reviewed separately in section 9.

7.1. Baseline Correction The objective of baseline correction techniques is to remove background noise from the raw sensor responses and to increase the contrast. Three types of baseline correction techniques are widely used: differential, relative, and fractional techniques.193,202 Differential techniques subtract a baseline value from the sensor response and can be used to remove additive noise or interferences. Differential techniques are typically used for piezoelectric sensors,203,204 where the response is a frequency or phase shift with respect to a reference analyte (and/or an uncoated reference sensor), and for MOSFET sensors,205 where the response is a voltage shift in the I(V) curve. Relative techniques compute the ratio between the sensor response to the sample and the sensor baseline value and, therefore, can be used to reduce multiplicative noise. The relative technique is commonly used with metal-oxide devices, since their resistance upon exposure to a sample, RS, is related to the baseline resistance, R0, i.e., RS ) R0[C]-β.206 Fractional techniques subtract the baseline value and then divide by the baseline value, which yields a per-unit response. It has been shown202 that the use of fractional changes in conductance provides the best

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 587

Table 1. Dynamic Parameters That Can Be Extracted from Sensor Response Curves72 parameter

description gasOn (1/5)‚∑T)gasOn-4s

baseline

(sensor value) sensor value (averaged over 5 s) at gasOff-baseline sensor value (averaged over 5 s) 30/90 s after gasOn/Off-baseline max (sensor value)-baseline min/max difference between two samples during measurement (sensor value 10 s after gasOn/Off-baseline)/10 (response - 90 s on response)/30 gasOff ∑T)gasOn (sensor value - baseline)

final response, response 30/90 s on/off response maximum response min/max derivative on/off derivative plateau derivative on integral off integral

gasOff+119s ∑T)gasOff (sensor value - baseline)

short on/off integral

gasOn/Off+9s ∑T)gasOn/Off (sensor value - baseline) response/on integral time from gasOn for sensor value to reach baseline + 0.9 × response or baseline + 0.6 × response time from gasOff for sensor value to reach baseline + 0.1 × response or baseline + 0.4 × response Y ) A3x3 + A2x2 + A1x + A0 On: Y ) (sensor value - baseline) and x ) time from gasOn to gasOff Off: Y ) (response + baseline - sensor value), x ) time from gasOff to gasOff + 240 s Y ) A(1 - exp(-(x/T)), where Y and x are defined like in the polynomial fit Y ) A0 + A1 exp(-x/T1) +A2 exp(-x/T2), where Y and x are defined like in the polynomial fit y(t) ) a1‚y(t - 1) + a2‚y(t - 2) + b‚u(t - 1) On: y(t) ) (sensor value - baseline), t ) time from gasOn - 5 s to gasOff and u(t) ) 0 if test gas off and 1 if test gas on Off: y(t) ) (response + baseline - sensor value), t ) time from gasOff - 5 s to gasOff + 240 s and u(t) ) 1 if test gas off and 0 if test gas on

response/on integral T0-90% T0-60% T100-10% T100-40% polynomial on/off 1. exponential on/off 2. exponential on/off ARX on/off

Table 2. Summary of Baseline Correction and Scaling Techniquesa type

name

transform ) ) )

application notes

(ref) x(k) i - xi (k) (ref) (xi /xi ) (ref) (ref) (x(k) i - xi )/xi

baseline correction baseline correction baseline correction

differential relative fractional

x(k) i x(k) i x(k) i

baseline correction

MSC

x(k) ) (x(k) - a(k))/b(k)

global scaling

feature norm

(k) x(k) i ) (xi - min[xi])/(max[xi] - min[xi])

global scaling

autoscaling

(k) x(k) i ) (xi - mean[xi])/std[xi]

global scaling global scaling

mean centering whitening

(k) x(k) i ) xi - mean[xi] x ) Λ-1/2MTx

local scaling

vector norm

(k) 2 ) (x(k) x(k) i i / ∑i(xi ) )

local scaling

SNV

x(k) i

nonlinear transform nonlinear transform

logarithm square-root

(k) x(k) i ) log(xi )

nonlinear transform

Box-Cox

x(k) i )

nonlinear transform

Horner-Hierold

(k) (ref) -1/β i x(k) i ) (xi /xi )

x

)

x(k) i )

(x(k) i

- mean[x(k])/std[x(k]

xx(k)i

{

λ ((x(k) i ) - 1)/λ λ * 0

ln( (k) i )

removal of additive noise/drift removal of multiplicative noise has been shown to work well for metal-oxide chemoresistors removal of information correlated with a reference sample; a(k, b(k are estimated for each sample makes signal magnitudes comparable across sensors but can amplify noise and is sensitive to outliers makes signal magnitudes comparable across sensors but can amplify noise removal of common-mode signal across samples yields uncorrelated, unit-variance features, but can also amplify noise reduction of concentration dependence; useful for qualitative (discriminative) analyses reduces within-class scattering but makes the data “closed” linearization and dynamic range compression linearization compensates for nonlinearities and compresses the dynamic range of the sensor

λ)0

linearization of metal-oxide chemoresistors; parameter βi estimated from the data

a x(k) denotes the response of sensor i to sample k. x(ref) denotes the response of sensor i to a reference sample. Notation: mean() and std() denote i i the sample mean and sample standard deviation.

pattern-recognition performance for (n-type) MOS chemoresistors. Fractional changes in resistance are also commonly employed with conducting-polymer chemoresistors.207,208 The above techniques operate on a sensor-by-sensor basis. Instead, baseline effects on the data set may be treated by means of multivariate techniques, such as multiplicative scatter correction (MSC).209-211 Developed to remove light scattering and particle-size issues in near-infrared spectroscopy,212 MSC has so far received little attention in the “electronic nose” or sensors community.213 Given a feature vector x(k) and a reference sample x(ref), MSC computes the

regression model x(k) ) a(k) + b(k)x(ref), and then uses the regression parameters (scalars a(k) and b(k)) to rescale the feature vector by subtracting the intercept a(k) and dividing by the slope of the estimated regression b(k): x(k) ) (x(k) a(k))/b(k). Thus, MSC can be used to correct for both multiplicative and additive effects. Table 2 summarizes the various forms of baseline correction techniques.

7.2. Scaling The objective of scaling techniques is to either eliminate irrelevant information from the sensor data (e.g., concentra-

588 Chemical Reviews, 2008, Vol. 108, No. 2

tion), or to precondition the data (e.g., decorrelating features). Scaling techniques can be grouped into global or local techniques, depending on whether they operate on a featureby-feature basis or on a sample-by-sample basis.194

7.2.1. Global Techniques Global techniques transform the data on a feature-byfeature basis across an entire database. The most common techniques are feature normalization and autoscaling. Feature normalization scales each feature to the range [0, 1] by subtracting the minimum value and then dividing by the overall measurement range of the sensor response, both computed across the entire database. Feature normalization makes full use of the input dynamic range but is very sensitive to outliers, since the range is determined by extreme values in the sensor data. In contrast, autoscaling normalizes each feature by subtracting the sample mean value and then dividing by the standard deviation, both computed across the entire database. Autoscaling cannot provide tight boundaries for the input range but is more robust to outliers than feature normalization. Moreover, robust statistics may be used to reduce the sensitivity to outliers.214 Multivariate techniques can also be used to globally scale the data. For instance, the whitening transform180 may be used to produce uncorrelated and unit-variance features. The procedure consists of first projecting the data along the eigenvectors of the covariance matrix and then normalizing bythe corresponding eigenvalues, i.e., x ) Λ-1/2MTx, where M contains the eigenvectors (arranged as columns) and Λ is a diagonal matrix with the corresponding eigenvalues. The whitening transform is closely related to principal-components analysis (PCA), with the key difference to PCA being that PCA only uses the eigenvectors corresponding to the largest eigenvalues (for dimensionality-reduction purposes). Note that the whitening transform is equivalent to autoscaling if the sensors/features are independent and zero-mean. Global methods are typically used to ensure that sensor response amplitudes are comparable, preventing subsequent pattern-recognition procedures from being overwhelmed by sensors with arbitrarily large values. For instance, nearestneighbor procedures180,215,216 are extremely sensitive to feature weighting, and multilayer perceptrons, the most common type of feedforward neural networks, may saturate for excessively large input values. However, it must be noted that these techniques can amplify noise since all the sensors (including those which may not provide any useful information) are weighted equally.217

7.2.2. Local Techniques Local techniques transform the data on a sample-by-sample basis across the feature vector. Local techniques include vector normalization and standard normal variate correction. In vector normalization, the response of each individual sensor is normalized (i.e., divided) by the L2 norm of the vector (|x|L2 ) x∑ixi2). This forces the distribution of samples to be located on a hypersphere of unit radius. Vector normalization can be used to remove concentration effects, provided that all sensors in the array have the same concentration dependence, e.g., xi ) kif([C]). This is the case for surface and bulk acoustic wave sensors,206 electrochemical cells and fluorescent indicators,184 carbon-black sensors,208 and metal-oxide sensors. Similar concentration-removal effects can be achieved by normalizing each sensor with the

Hierlemann et al.

L1 norm (|x|L1 ) ∑i|xi|) or with the response of a reference sensor. In the case of metal-oxide sensors, the concentration removal requires that the exponent β of the power-law dependence be the same for all sensors. However, an alternative normalization technique has been recently developed that allows this condition to be relaxed.218 Vector normalization is beneficial for discrimination problems but should be avoided in concentration-estimation problems or whenever the vector norm is known to carry relevant information. For hybrid array data, vector normalization should be performed separately on groups of sensors with the same concentration dependence, such as sensors of the same type, possibly followed by a second normalization across the entire array. The standard normal variate (SNV) transform219 normalizes each sensor response by first subtracting the average across the array (for a given sample) and by then dividing through the standard deviation across the array (for a given sample). Thus, SNV can be thought of as an autoscaling of each feature vector. SNV is commonly used in near-infrared spectroscopy to effectively reduce in-class variance but has also been applied to chemical sensor transients.198 Care must be taken in employing local transforms, as they render the data set “closed,” i.e., SNV forces the sum of the features to become zero, whereas vector normalization renders the sum-square equal to one. Closure can introduce spurious positive correlations between the sensors featuring the highest response levels and spurious negative correlations between sensors exhibiting the lowest response levels.220 This issue is particularly relevant in the case of hybrid arrays, where each sensor type may have an intrinsically different range of signal magnitude. It is then advisable to first scale each sensor using a global technique, see, e.g., refs 221 and 222.

7.2.3. Nonlinear Transforms Various transformations have been proposed to compensate for nonlinearities in the data, such as concentration dependencies, or saturation effects. They include logarithms, square-roots,200 and the Box-Cox transform.223 Of particular interest for metal-oxide sensors is a linearization transform proposed by Horner and Hierold,224 which we describe here for illustration purposes. The method assumes a resistanceconcentration dependence that can be described by q

Ri ) R0i(1 + ∑(Aij[Cj])mj)-βi

(3)

j)1

where R0i is the sensor resistance in air, [Cj] is the concentration of gas j, q is the number of gases, and Aij, mj, and βi are model parameters. Suitable values for these parameters can be found by fitting the model to experimental data, {Ri,[Cj]}, by means of a nonlinear optimization technique (Levenberg-Marquardt). Once these parameters have been estimated, the following nonlinear transformation can be applied to linearize the sensor response with respect to the analyte concentration:

ri ) (Ri/R0i)-1/βi; [cj] ) [Cj]mj

(4)

8. Drift Compensation The most serious limitation of current sensor arrays is the inherent drift of individual sensors, which results in a slow,

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 589

Figure 28. Illustration of the multiplicative signal correction method of Haugen et al.229 Response of an individual sensor to the calibrant and target gases (a) before calibration, (b) after short-term correction (within sequence), and (c) after long-term correction (between sequences).

random temporal variation of the sensor response when exposed to the same analyte under identical conditions. As a result of drift, learned sensor response patterns may become obsolete over time, so continuous recalibration may be required. Following Holmberg and Artursson,225 drift-like effects can be attributed to a number of sources. First, there are issues related to the sensor itself, such as aging (e.g., reorganization of the sensing layer) and poisoning (e.g., irreversible binding); only aging and poisoning are strictly considered as drift. These effects are very difficult to compensate for and have been the subject of many investigations, as will be detailed below. Second, drift-like effects can occur also in the measurement system due to, e.g., fluctuations in flow rate, temperature, pressure, or humidity content in the sensing chamber, or analyte condensation in the manifold. These types of artifacts can be most effectively addressed by measuring the variables that are known to fluctuate and by then compensating for the fluctuations in software. This includes, e.g., removing any variance due to fluctuating parameters from the sensor response. An effective compensation may pose a major challenge, when small environmental perturbations induce large changes in the sensor response. In addition, issues related to experimental procedures can give rise to effects that are often confounded with drift, such as memory effects (hysteresis, systematic errors due to fixed sampling sequences), short-term effects (system warm-up, thermal trends), or even the degradation of the samples themselves. These types of errors can be addressed with a proper experimental design, whereas the previously discussed two sources of drift will typically require some form of signal processing. The modulation of the sensor operation temperature has been used to generate features that are significantly more stable than isothermal features. Along these lines, Roth et al.226 alternated the temperature of a CO2 gas sensor coated with an organic material and showed that the normalized slope of the sensor response remained much more stable over time in comparison to constant-temperature measurements. Aigner et al.227 derived similar conclusions for Si-planarpellistors. A number of drift-compensation algorithms have been developed over the past decade, which can be grouped into

univariate and multivariate techniques. These two types will be reviewed below.

8.1. Univariate Drift Compensation Drift compensation may be performed for each sensor individually. At the simplest level, one may employ the baseline-correction techniques described in section 7.1; differential measurements can be used to remove additive (baseline) drift, whereas multiplicative (sensitivity) drift can be compensated for by conducting relative measurements using a reference gas (clean or purified air). Differential measurements can be made with respect to a calibration gas,228 which must be chemically stable over time, and whose behavior should be highly correlated with the target analytes.229,230 A practical calibration method that operates on a sensor-by-sensor basis has been developed by Haugen et al.,229 in which drift compensation is performed on two time scales: (i) within a single measurement sequence and (ii) between measurement sequences. At each time scale, the method models temporal variations in a calibration gas by means of a multiplicative correction factor, which is then applied to the target samples. The process is illustrated in Figure 28. A multiplicative correction scheme has also been used by Sisk and Lewis.230 More interestingly, these authors have shown that event-driven calibration provides superior performance with respect to periodic calibration. The events may be triggered when, e.g., unlabeled samples start to fall outside the decision boundaries of the classifiers, when outliers are detected, or after interruptions in the data collection. Needless to say, event-driven calibration is also most cost-effective, since it is only performed upon demand.

8.2. Multivariate Drift Compensation Alternatively, one may correct for drift on the entire array data as a whole, rather than on a sensor-by-sensor basis. The advantage of this approach is that the procedure can exploit correlations between the sensors. The majority of these methods is based on adaptive modeling, system identification,

590 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

orthogonal signal correction, or blind signal deconvolution techniques, as will be detailed below. Adaptive models are an interesting alternative for the problem of drift compensation. The basic idea behind these techniques is to model the distribution of training examples with a codebook (i.e., a collection of cluster centers) and then to adapt this codebook upon the presentation of the test data: an incoming (unknown) sample is assigned to the “closest-matching” class and is then used to adapt the class parameters. A variety of adaptive models have been used, which update one cluster center per class,231 a single Kohonen self-organizing map (SOM)232 for all the classes,233-235 or a separate SOM per class.236,237 A potential problem of these approaches is that they rely on correct classification; misclassification errors will eventually cause the model to lose track of the class patterns. In addition, all analytes need to be sampled frequently to prevent their patterns from drifting away too much. System-identification techniques have also been used to model sensor dynamics and then predict drift effects. Holmberg et al.231,238 have investigated a number of models (e.g., AR, ARMA, Box-Jenkins) to generate a prediction for the common-mode component of the drift for each sensor using the remaining sensors as inputs to the model,

where xs(k) is the response of sensor s at time k and xi(k) is the response of all other sensors at time k. Model parameters {ai, bk, ck} can be learned off-line231 or online by applying a recursive least-squares procedure.238 For classification purposes, a separate dynamical model is built for each analyte class, and unknown analytes are assigned to the class, whose model displays the lowest prediction error. Nonlinear extensions of this approach, such as Volterra series or artificial neural networks, have been explored by Marco and coworkers.239-241 Finally, Perera et al.242 have developed a novelty-detection method based on recursive dynamic PCA243,244 that can operate under drift conditions. Approaches based on orthogonal signal correction245 have been also successfully employed. As illustrated in Figure 29, the basic idea behind these methods is to subtract from the sensor-array response the components that account for as much of the variance as possible but which are uncorrelated with analyte information (mixture concentrations in multicomponent analysis or class labels in discrimination problems). Along these lines, Artursson et al.246 have developed a drift-compensation method that first estimates the main direction of drift by computing the first principal component of the samples from a calibration gas. This direction is then removed from the multivariate sensor response by subtracting the corresponding bilinear component,

xcorrected ) x - (x‚vcal)vTcal

(6)

where vcal is the first eigenvector of the calibration data, xcal. A related procedure has been proposed by GutierrezOsuna.247 Here, the experimenter first identifies a set of variables (y) whose variance can be attributed to drift or interferents. These variables can include, e.g., the response to a purging or reference gas, the date and time when the sample was collected, or measurements from temperature,

Figure 29. (a) Illustration of orthogonal signal correction; (b) principal-components analysis of the responses of an array of metaloxide sensors to various food items. Notice that drift-related and class-related information are nearly orthogonal. Reprinted with permission from ref 217. Copyright 2002 IEEE.

pressure, and humidity sensors. Any variance in the measurement vector (x) that can be explained by y is then due to drift or interferences and should be removed. This can be done by means of regression/deflation methods as shown in eq 7. This technique is also closely related to “target rotation”.248

xcorrected ) x - Wy where W ) argmin|x - Wy|2

(7)

Kermit and Tomic249 have approached drift-compensation as a linear, blind-source-separation problem. In this approach, the sensor array response can be modeled as the weighted sum of a number of independent “sources”, such as driftrelated noise and discriminatory information. The authors use independent-component analysis (ICA),250 an extension of principal-component analysis aimed at finding statistically independent projections of the data. As described in section 6.2, two variables x and y are said to be independent if their joint probability density function (PDF) p(x,y) is equal to the product of their marginal PDFs: p(x, y) ) p(x)p(y), in other words, if knowledge of the value of one variable

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 591

Figure 30. (a) Three principal components and (b) three independent components extracted from the response of a hybrid array to six different alcohols (headspace of 0.5% propanol, 1% propanol, 2% propanol, 0.5% butanol, 1% butanol, and 2% butanol in aqueous solution). Samples are sorted according to classes and time stamps within each class. The estimated probability density functions (PDFs) of each component are plotted at the right of each component. Note that most PCA and ICA projections have markedly non-Gaussian PDFs. Reprinted with permission from ref 249. Copyright 2003 IEEE.

does not provide any information about the value of the other variable. In contrast, two variables are said to be uncorrelated if the expected value (i.e., the average) of their product is equal to the product of their expected values: E(xy) ) E(x)E(y), where E[ ] is the expectation operator. Principalcomponent analysis finds uncorrelated projections, whereas independent-component analysis finds independent projections, which is a more restrictive criterion. To find the desired solution, ICA uses higher-order statistics (i.e., entropy), whereas PCA relies on second-order statistics (i.e., covariance). A clarification is in place at this point: “higher-order” statistics should not be confounded with “higher-order” sensing; the latter refers to the way in which the data are structured. Experimental data in the study of Kermit and Tomic were obtained from a hybrid array with 10 MOSFET and 12 metaloxide sensors, all of which were exposed to the dynamic headspace of 6 analyte solutions (0.5% propanol, 1% propanol, 2% propanol, 0.5% butanol, 1% butanol, and 2% butanol). Ninety measurements were made, 15 per solution, and processed off-line with fastICA.250 The left panels of Figure 30 show the first three principal components, where samples have been ordered first by label (e.g., the first 15 samples are those from the first class) and then by time of presentation to the array. The right panel of Figure 30 shows the corresponding independent components. The first ICA captures information about the concentration of the analytes (notice the six distinct steps, which correspond to concentrations of 0.5%, 1%, 2%, 0.5%, 1%, and 2%), whereas the second ICA source captures information about the drift (notice the trend for the 15 measurements from each analyte), and the third ICA source captures information about the identity of the analytes (i.e., low for the first 45 samples (propanol) and high for the last 45 samples (butanol)). Thus, ICA is able to separate the three sources of information in the sensor response: analyte identity, analyte concentration, and drift effects. In contrast, PCA is only able to separate concentration information (first principal component), but analyte identity and drift are mixed together in the second and third principal components. It is important to note,

however, that the ICA model proposed here computes a solution off-line, i.e., after all the data have been collected. The question remains, though, whether or not these results will hold, when the ICA demixing matrix (equivalent to the eigenvectors in PCA) is tested on data that have not been included in the training set.

9. Feature Extraction from Sensor Dynamics As described in sections 2.1 and 2.2, one may achieve “higher-order” sensing by exploiting the dynamic properties of the sensors for analytical purposes. In this review, we will concentrate on two strategies that have been extensively used in the literature: the analysis of the transient response of the sensors to sudden changes in the sample concentration (or temperature) and the modulation of the operating temperature of metal-oxide chemoresistors.

9.1. Transient Analysis When performing data analysis of chemical sensor arrays, it is, in most cases, convention to assume that the information of interest is contained in the quasi-steady-state (or final) response of each sensor. While this approach yields measurements that are simple to conduct and evaluate, it ignores useful information that may be contained in the transient response of the sensor (see, e.g., Table 1). The transient response is the result of dynamic processes that take place when the sensors interact with the target sample. These dynamic processes are unique for each sensor-analyte pair and, therefore, are potentially very useful for analytical purposes. They are typically triggered by modulating an internal parameter of the sensor, such as the operating temperature, or an external one, such as the gas-phase composition of the sample.251 The most straightforward but not necessarily the most robust approach consists of analyzing the evolution of the sensor response upon dosing the sample. One of the earliest accounts of this approach is the work of Mu¨ller and Lange,252 who showed that a single cross-selective sensor may be used to discriminate a number of target compounds at different

592 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 31. Gas sensor transient response to a short odor pulse (a). Transient analysis approaches: (b) oversampling, (c) parameter extraction, and (d) model fitting. Reprinted with permission from ref 194. Copyright 2002 Wiley-VCH, Weinheim.

concentrations (something that cannot be achieved using only the steady-state response). In their landmark study, the authors extracted two parameters from the transient response of a zeolite-covered metal-oxide sensor: the initial slope (S1), which was shown to be proportional to the concentration of the gas, and the steady-state response (S2), which was proportional to the square root of the concentration. As a result, the authors showed that the variable S1/S22 could be used to discriminate different simple gases regardless of their concentrations. Though this concentration-independent parameter may be different for other sensors (see, e.g., Vilanova et al.253 for a different case), the study of Mu¨ller and Lange is important because it illustrates that more than one parameter may be extracted from the sensor response. While the transient response will depend on the odor delivery system (see, e.g., the discussion in sections 2.2 and 4.1, as well as in Chapter 6 in ref 206), transient parameters have, in some cases, been shown to be more repeatable than static descriptors; see ref 254. In addition, transient analysis can reduce the data acquisition and calibration time.255 If the initial sensor transients contain sufficient discriminatory information, one may avoid the lengthy acquisition time required to reach steady state. As a consequence, the sensors also require less time to recover to their baseline, a process that can be particularly slow when the target analytes are present at high concentrations. By reducing the duration of the analyte pulse, and by thus minimizing irreversible binding, the lifetime of the sensors can also be increased. Furthermore, in the case of dynamic headspace analysis, a steady-state response may not even be attainable, since the volatiles in the headspace may be depleted faster than they can be released from the sample. In these cases, the transient response to a short concentration pulse, as illustrated in Figure 31a, may provide sufficient information.256 The remainder of this section will provide an in-depth review of different computational methods that have been proposed to extract information from the transient responses of gas sensors. These methods can be grouped into three broad categories: (i) oversampling, (ii) parameter extraction, and (iii) model-based methods, as illustrated in parts b-d of Figure 31. Outputs from these methods can then be treated using conventional pattern classification, regression, and clustering techniques.217 Alternatively, the entire transient response may be processed with suitable classification or

regression models; the reader is referred to section 10 for a discussion of these methods.

9.1.1. Oversampling Procedures The most straightforward approach to capture transient information is to oversample the sensor transient at different time intervals during the odor exposure and/or odor recovery phase, as illustrated in Figure 31b. The term “oversampling” is used here to emphasize that the sensor is sampled more frequently than at steady state; the opposite term “decimation” is sometimes used in the literature to emphasize that the sensor transient is first measured with a very fine time scale, and then a subset of those measurements is used as a feature vector (i.e., the finely sampled transient is said to have been decimated). Leaving aside terminology, when using oversampling/decimation techniques the dynamic information is represented implicitly, in the correlation of these measurement values, rather than explicitly, as is the case for the other two approaches. Nanto et al.257 characterized the transient response of thickness-shear-mode resonators by means of nine parameters, which correspond to the sensor response values at defined times {t ) 1, 2, 3, ..., 8, 14 min}, normalized with respect to the maximum sensor response during the transient. The authors show that a multilayer perceptron trained on these parameters was able to discriminate among different types of wines and liquors using a single sensor. Saunders et al.258 used the transient response of thickness-shear-mode resonators during the odor exposure and recovery times. The authors extracted 50 measurements from these transients and normalized them with respect to the baseline frequency and the maximum response of the sensor during the transient, and used then as input features into a bank of multilayer perceptrons (one per sensor). The normalized transient responses (termed “kinetic signatures” in their article) were shown to be very consistent for each sensor across repeated trials, despite a drift in the baseline and in the maximum response parameters. Hongmei et al.259 employed a similar kinetic-signature procedure to simultaneously determine the concentration of sulfur dioxide and relative humidity using a single piezoelectric quartz thickness-shear-mode resonator. White et al.260 used an array of fiber-optic sensors to identify single analytes, binary mixtures, and the relative component concentrations. Analytes were delivered to the distal end of the fibers using a

Higher-Order Chemical Sensing

short 2 s pulse, and the transient response was resampled to yield 10 measurement values, each representing the average sensor response of 6 consecutive time points. Their results show that multilayer perceptrons trained on the oversampled transient significantly outperform those trained on only the integral response of each sensor transient. Kermani et al.261 proposed a time-windowing technique to extract transient information. Their method relies on four overlapping bellshaped kernel functions, which are used to compute a weighted integral response of the sensor at different times during the sensor transient. Using an array of 15 metal-oxide sensors, their method was shown to significantly outperform the steady-state and the transient integral on a number of odor databases.196 A family of five uniform time-windows was used by Brahim-Belhouari262 to extract information from the transients of an array of eight SnO2 microhotplates. However, while the time-windowed features outperformed steady-state features, the authors showed that similar performance could be obtained by combining steady-state signals with the slope of each transient, measured during the first minute of the sensor exposure.

9.1.2. Ad hoc Transient Parameters Alternatively, a wide range of parameters may be extracted from the transient response of a gas sensor, such as rise times, derivatives and integrals, computed at different time points during the exposure and recovery phases, as shown in Figure 31c. With little computational expense, these methods can provide a more compact representation of the information contained in the sensor transients. As discussed earlier, a combination of the initial slope of the transient and the steady-state response was used by Mu¨ller and Lange252 to discriminate multiple analytes at different concentrations. More recently, Llobet et al.254 characterized the transient response of metal-oxide sensors by means of the conductance rise time, measured from 20% to 60% of the total conductance change (G(0) - G(∞)). An important result of this study is that the rise time appears to be significantly more repeatable than the steady-state response. Moreover, an analysis of variance also showed that the response time was independent of the analyte concentration (toluene and oxylene in the range 25-100 ppm) and only depended on the vapor/sensor pair. Roussel et al.263 evaluated a large number of ad hoc features for the purposes of discriminating off-odors in wines with an array of five tin-oxide sensors. Different features were computed from the transient response and their first- and second-order derivatives, including the response values at different time intervals and the response maxima/minima, yielding a total of 29 features per sensor. Features were evaluated with respect to three different criteria: repeatability across trials (within-class variance), discrimination results (ratio of between-class to within-class variance), and correlation with other features. Their results show that (1) the best features include the maximum sensor response values, the maximum slope during the exposure transient, and the minimum slope during recovery, and that (2) the most suitable features are the same for all five sensors. Paulsson et al.264 performed a feature-selection study for various preprocessing and transient-analysis techniques on experimental data from a real-life application: the evaluation of breath alcohol contents using a hybrid array of MOSFETs, chemoresistors, and an infrared sensor. The sensor features included the final response value, the maximum response values, the response integral, and the maximum response

Chemical Reviews, 2008, Vol. 108, No. 2 593

slope. However, their results showed no systematic advantage for using any of these feature types. Employing concepts from dynamic systems, Martinelli et al.265 proposed to extract transient information from the phase plot of each sensor. In their article, the state variables were the sensor response and its derivative. A single transient feature was extracted from each sensor: the area circumscribed by the phase plot of each sensor during the adsorption and desorption processes. The method was validated on an experimental database containing the transient response of thickness-shear-mode resonators when exposed to the headspace of apples with different degrees of internal defects. The results showed that phase-space transient features outperform steady-state features in terms of both uncorrelatedness and discrimination capabilities. In a subsequent study, Martinelli et al.266 proposed to extract additional information from the phase space, arguing that the evaluation of the area of the phase plot does not take into account information that may be present in the shape of the trajectory. For this purpose, they computed a number of higher-order geometric moments267 from the phase plot of the sensor transient. In this study, the phase space was defined by the sensor response and a time-delayed version, i.e., [s(t), s(t τ)]. The use of such “dynamic moments” was shown to yield better results for two experimental databases in comparison to only using steady-state information. However, the authors acknowledged that the optimum time delay (τ) is applicationspecific and, more importantly, that the dynamic moments tend to be rather sensitive to small changes in the sensor dynamics. Similar results and conclusions for dynamic moments have been reported on by Vergara et al.,268 using metal oxide sensors to detect the rancidity of crisps (potato chips). In a related study, Pardo and Sverbeglieri269 compared five different features: the steady-state response, the phasespace area,266 and the transient integral, with the latter two computed for both the exposure and the recovery process. While their results are not unequivocal as to which type of feature is best, and while the evaluation was performed on a small data set (coffee ripening), the authors suggest that the phase-space area during the recovery process outperforms steady-state and transient integral information and that features calculated during the recovery interval (either phasespace area or integral) consistently provide better performance than those calculated for the exposure interval.

9.1.3. Model-Based Parameters Transient information can also be captured by fitting an analytical model to the experimental transient, and then using the resulting parameters as features. Various types of models have been used for this purpose, such as autoregressive and polynomial methods, but multiexponential models are most common due to the exponential nature of the transient response, as shown in Figure 32. Transients are generally modeled by a sum of exponential functions of the following form: M

f(t) ) ∑Gi e-t/τi

(8)

i)1

Although conceptually easy, the task of modeling a curve with a set of exponential functions with real exponents is ill-conditioned. Unlike the familiar sinusoidal functions used in Fourier analysis, exponential functions do not provide an orthogonal expansion. Therefore, if one tries to determine

594 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 32. Transient responses of an array of conducting-polymer sensors. Reprinted with permission from ref 283. Copyright 1999 Elsevier.

the coefficients {Gi, τi, i ) 1, ..., M} from finite-time and finite-precision samples of the transient, the distribution function of time constants will not be unique. An additional problem is the determination of M, the number of exponential components that should be used for the fit. This issue has been known for over 40 years, when Lanczos270 demonstrated that three-exponential curves with similar time constants could be fitted accurately with two-exponential models with significantly different amplitudes and time constants. The task of performing a multiexponential fit according to eq 8 is of importance in a variety of disciplines in science, such as gas relaxation kinetics, fluorescence, radioactivity, and nuclear magnetic resonance.271 A number of methods has been developed, which can be grouped into three classes:272 (a) Stepwise or exhaustive methods, which extract the different exponential components in a sequential manner, as in the case of the “graphical” peeling-off.273 These methods can be considered as nonglobal, because each component is extracted independently of the rest. (b) Global approximation or least-squares methods, which approximate the experimental transient using a defined number of exponential components by minimizing a figure of merit of the fit. These methods are not aimed at component detection. (c) Global detection or integral transforms methods. These methods exhibit similarities to (a) and (b): like (a), they are true component detectors, and like (b), they are global, because all model parameters are extracted simultaneously. Representative examples of these methods include the Gardner transform,274 multiexponential transient spectroscopy,271,275 and the Pade-Laplace/Pade-Z transform.276 In the context of modeling chemical-sensor transients, the vast majority of multiexponential approaches rely on global approximation, arguably due to the broader availability of optimization tools. One of the earliest reports on multiexponential modeling is by Vaihinger et al.,69 who showed that two or more exponentials were required to provide an accurate fit to experimental data from amperometric sensors. Their results suggest that the extracted time-constants are gas-specific but concentration-independent, whereas the corresponding amplitudes are concentration-dependent. Vilanova et al.253 used a diffusion-reaction model developed

Hierlemann et al.

by Gardner73 to fit transients of metal-oxide sensors exposed to individual gases. Their method provides a single time constant, which is also shown to be gas-specific and concentration-independent. In ref 277, a general diffusionreaction model is applied to gas mixtures in the low-tomedium concentration range, where interactions between gas species can be ignored. This new model is shown to provide a good fit to the transient response of binary mixtures and yields time constants (one per gas in the mixture) that are also concentration-independent. Eklo¨v et al.72 performed curve-fitting to transients of Pt-MOSFET sensors with oneexponential and two-exponential models (see Table 1). While the two-exponential model provided a better overall fit to the experimental transient, the model parameters were shown to be unstable. In contrast, parameters from the singleexponential model had rather high signal-to-noise ratios, comparable to those of the “simple” parameters mentioned already in section 2.2. Galdikas et al.278 used an array of ten metal-oxide sensors to monitor the freshness of poultry. Samples of poultry meat were stored in a room at 17 °C and 45% RH (relative humidity) and monitored continuously with the sensor array. The authors analyzed the steady-state response of the sensors, as well as the time constants of a biexponential fit to the sensor transients. The steady-state response did not show any significant changes until after 16 h, whereas the smallest of the two time constants started to show significant changes after 2-3 h, which could be used to provide an early detection of food spoilage. Nakamoto et al.256 used two-exponential models to fit the recovery phase of thickness-shear-mode resonators upon short pulses of various odorants. Parameters of the exponential component with the largest contribution to the response of each sensor were then selected as features. In comparison to the maximum response values of the sensors upon a concentration step, the transient parameters were shown to have better discrimination properties. Di Nucci et al.279 used oneexponential models to approximate the exposure and recovery transients of thickness-shear-mode resonators to various odorants. Their exponential parameters were shown to provide more discriminatory information than the steadystate response of the sensors, a result that is consistent with those reported in ref 72. Baumbach et al.280 used a biexponential model to extract information from the transient response of semiconductor microsensors upon steps in their operating temperature. One exponential component was used to explain temperature effects, which were relatively fast owing to the low thermal mass of their microsensors. This term had a fixed (i.e., gas-independent) time constant. The second exponential component was used to explain the slower effects, which were due to the interaction between the gases and the sensing layer. This term had a gasdependent time constant. The authors showed that a simple decision tree could be used to discriminate CO, H2, and their mixture using the parameters of the biexponential model. Global detection techniques have only in a few cases been used to build multiexponential models for sensor transients. Nakamura et al.281 proposed a system-identification method to estimate the parameters of the exponential components: an autoregressive (AR) model was fitted to the sensor transient, L

x(k) + ∑aix(k - i) ) e(i)

(9)

i)1

where x(k) is the sensor response at time t ) kT, L is the

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 595

order of the model, and e(i) is the residual error of the model. While AR coefficients {ai} Li)1 could be used as features, these parameters (also known as linear predictive coefficients in speech processing) have poor interpolation properties. Instead, by computing the real roots of the characteristic equation, AR coefficients were converted into time constants and amplitudes of a multiexponential model. The results show that, in the case of single gases, 2-3 real roots (exponential components) could be found using the AR model, whereas 3-4 real roots could be found for gas mixtures. However, only one of these exponential components appeared to be stable from run to run (for both single gases and mixtures), which, again, is a hint to the illconditioned nature of multiexponential models. Artursson et al.282 also used multiexponential models to extract information from an electronic tongue based on pulse voltammetry. Their model consisted of two exponential components, which were assumed to represent the two types of currents present during the measurement: Faradaic and capacitive currents. Model parameters were found in a linear-least-squares fashion through a reparametrization of the biexponential model into a homogeneous differential equation. The resulting time constants were then converted into the coefficients of the corresponding characteristic equation; this step ensured that the final features were invariant to the optimization algorithm. These final features were shown to provide better class separability than the original data, while also filtering out experimental noise and providing near-lossless compression. Rather than finding the discrete coefficients of the multiexponential model in eq 8, one may instead attempt to recover the spectrum of time constants, G(τ):

f(t) ) ∫0 G(τ) e-t/τ dτ ∞

(10)

As pointed out by Samitier et al.,271 spectral methods have several advantages. First, the number of exponential terms does not need to be known a priori: the individual exponential components will be detected as peaks in the spectrum. Second, spectral methods are global methods, since all the components are obtained simultaneously in the spectrum. Third, the width of the peaks can be used to infer the resolution power of the spectral method, e.g., wider peaks suggest that two or more exponential components with similar time constants have not been resolved. Multiexponential transient spectroscopy (METS) is one such spectral method, which has been shown to be suitable for modeling gas sensor transients.271,275 METS recovers a spectrum of time constants through a multiple differentiation of the experimental transient on a logarithmic scale; higher spectral resolution can be achieved at higher orders of differentiation at the expense of amplifying experimental noise. Samitier et al.271 applied METS to the transient response of electrochemical fuel cells; their results showed that the amplitude of the spectral peak was proportional to the concentration of the gases (ethanol, methanol, and 2-propanol in their study), whereas the location of the peak, i.e., the time constant, was gas-specific. The Gardner transform274 can also be used to recover a pseudo-spectrum g(τ), in which the amplitude and time constants are coupled: g(τ) ) G(τ)τ; this condition biases the Gardner transform toward exponential curves for which the product of the amplitude and the time constant of the components are on the same order of magnitude, see, e.g., ref 283.

Alternatively, one may employ a fine-grained set of time constants, {τi}, and solve eq 8 for the amplitudes by using least-squares: N-1

M

k)0

i)1

{Gi} ) arg min[ ∑ (f(k) - ∑Gi e-kT/τi)2] with Gi

M (11) fixed {τi} i)1

The resulting distribution of amplitudes {Gi} can be treated as a spectral representation of the transient. This approach, also known as the exponential series,284,285 has the advantage that the minimization problem in eq 11 is linear in the amplitudes, so it can be solved efficiently. In practice, regularization techniques need to be used to ensure a smooth distribution of amplitudes, e.g., by adding an identity matrix to the data covariance matrix that results from solving eq 11 through least-squares.286 An alternative approach to model sensor transients has been recently proposed by Carmel et al.287 Their model is derived from a simple physical description of the measurement system,

fi(t) ) Ri∫0 hi(u)k(t - ti - u) du ∞

(12)

where Ri is a sensor-specific constant, ti is the time it takes a gas molecule to travel from the gas inlet to the surface of the ith sensor, hi(u) represents the probability that a gas molecule absorbed in the ith sensor at time t is still present at time t + u, and k(t) is the shape of the injected stimulus (e.g., a step or a pulse in concentration). In ref 287, the Lorentzian decay function h(t) ) τi2/(t2 + τi2) was found to provide a good fit to the exposure and recovery transients of both thickness-shear-mode resonators and metal-oxide sensors. Assuming a pulse function of duration T for the injected stimulus, eq 12 can be transformed into the following:

{

fi(t) ) 0

( ) [ ( )

t - ti βiτi tan-1 τi t ti t - ti - T - tan-1 βiτi tan-1 τi τi

(

)]

t < ti >ti e t e ti + T t > ti + T

(13)

From this equation, the model parameters {βi, T, ti, τi} that best fit the experimental transient can be found through a simplex optimization procedure.288 Doing this for every sensor-analyte pair yields a 4-dimensional feature vector that captures the shape of both the exposure and the recovery transient. In ref 287, the model was validated on experimental data of a hybrid sensor array exposed to 30 different odorants. The results showed that the Lorentzian model parameters {βi, T, ti, τi} provide significantly better recognition performance than standard transient features. Carmel et al.289 have also shown that the Lorentzian parameters are robust with respect to distortions in the sensor transients, a feat yet to be matched by multiexponential models. The Lorentzian model has also been generalized for the use with sensor transients containing multiple peaks.290

9.1.4. Comparative Studies Eklo¨v et al.72 provided a systematic investigation of transient parameters, including simple features such as pulse

596 Chemical Reviews, 2008, Vol. 108, No. 2

heights, integrals, and derivatives at various times during the exposure and desorption phases, and model-based parameters obtained by fitting the experimental transient by means of three types of analytical models (see also Table 1): multiexponential, autoregressive (ARX), and polynomial. The data set consisted of the response of Pt-MOSFET sensors exposed to mixtures of hydrogen and ethanol. Several conclusions can be derived from this study. First, most of the simple parameters have relatively high signal-to-noise ratios, including those from one-exponential models. Second, time-critical parameters such as derivatives, time constants, short integrals, and ARX models tend to be very much influenced by the exact timing of the gas delivery, which renders them unsuitable for pattern-recognition purposes. Third, the selection of model-based parameters based on their fitting performance can be misleading; ARX and twoexponential models provide the bests fits but also have very low signal-to-noise ratios. The main conclusion of the study is that a combination of simple parameters (final response, windowed response, derivatives, and integrals) can provide a performance comparable to parameters obtained through computationally intensive fitting procedures. In a follow-up study, Eklo¨v et al.291 performed a feature subset selection on the same database to identify the most relevant parameters. Features were selected using a sequential forward procedure (see section 11.2), where the selection criterion was the root-mean-square reconstruction error from a multilinear regression model. Their results indicated that discriminatory information is broadly distributed in the exposure and desorption transients, with 7 of the top 10 features being “simple” transient parameters. Delpha et al.292 compared the performance of parameters of a two-exponential model to the dynamic slope of the transient on a database consisting of six Taguchi sensors. The array was exposed to humid air at different relative humidity levels, to dry Forane 134a (a refrigerant gas), and to humid Forane 134a (different relative humidity levels). The dynamic slope was computed using the sensor response between 1 and 5 min after introduction of the sample, whereas the biexponential parameters were computed from the entire transient, once the sensors had reached steady state in a 60 min long exposure. The prediction performance of the biexponential parameters was 60% on the test data but increased to 100% when combined with the dynamic slope. Although no performance results were provided for the dynamic slope alone, the authors concluded that the biexponential and dynamic slope parameters provide complementary information. Distante et al.293 compared several transformation and feature extraction techniques using experimental data from an array of metal-oxide sensors exposed to concentration pulses of acetone, hexanal, and pentanone, each in humid and dry air. In this study, the authors advocate the use of a discrete-wavelet-transform (DWT) technique to extract transient information. Unlike the Fourier transform, which is only localized in frequency only, wavelets are localized in space and frequency, which renders them more suitable for the analysis of transient signals since they capture both spectral and temporal information. DWT coefficients were compared to those of a fast Fourier transform (FFT) as well as with feature vectors containing the integral and derivative in several locations of the transient. Their results show that the DWT provides the best performance, with integral features being a very close second. Derivative and FFT features

Hierlemann et al.

Figure 33. Time constants and signal amplitudes extracted from a 5-6V thermal transient of a TGS2620 sensor by the Pade-Z method: A, acetone (10-4 volume %); I, isopropyl alcohol (10-1 volume %); M, ammonia (1 volume %). Four samples of each analyte were extracted. Reprinted with permission from ref 286. Copyright 2003 Elsevier.

appear to be only marginally worse than the previous two techniques. Gutierrez-Osuna et al.286 compared the performance of METS,275 the Pade-Z transform,276 transient oversampling,196 the exponential series in eq 11, and steady-state isothermal responses. All these methods were evaluated using a data set of Taguchi sensors exposed to various concentrations of acetone, isopropyl alcohol, and ammonia under a stepwise change in temperature. As shown in Figure 33, the Pade-Z was able to recover several stable multiexponential models for the three analytes, though with different numbers of exponentials. Since most pattern-recognition techniques assume fixed-length feature vectorssbut see rational kernels294 for an exceptionsthe Pade-Z models were transformed into the coefficients of a fixed-length Taylor series expansion. Experimental results show that the exponential series method provides the best performance, whereas METS, Pade-Z, and transient oversampling show comparable performance, which is still better than that of using the steadystate response. Shafiqul Islam et al.295 compared a number of “simple” parameters, such as levels, slopes, and integrals at different times, to the coefficients of a third-order polynomial fit of the sensor transient. The experimental data set consisted of responses of an array of thickness-shear-mode resonators to various solvent exposures. While the simple parameters provided better separability than the polynomial coefficients, a combination of these two types of features appeared to improve the overall performance of the array. Altogether, these studies indicate that models that provide the best curvefitting results do not necessarily contain the most analytical information. Simple parameters should be used first, since they tend to have higher signal-to-noise ratios, though complex parameters can sometimes provide complementary information.

9.2. Temperature-Modulation Analysis It has long been known that the selectivity of metal-oxide sensors is greatly influenced by the operating temperature of the device, since the reaction rates of different volatile compounds and the stability of surface-adsorbed oxygen species are a function of the temperature.296 As a result,

Higher-Order Chemical Sensing

modulation of the operating temperature can give rise to gasspecific temporal signatures, which provide a wealth of discriminatory and quantitative information. One of the earliest reports on the use of temperature modulation is a 1975 patent by Le Vine, in which a sensor was operated at two temperatures: a low temperature, at which the sensor was preferentially selective to CO, and a higher temperature, at which the sensor was less selective, and which was used to purge the sensor of CO.297 However, it is the modeling work of Clifford and Tuma65 and the algorithms of Sears et al.298 that are often credited for bringing the concept of temperature modulation to the attention of the sensors community. An excellent account of early work on temperature modulation in the 1980s and 1990s was written by Lee and Reedy.22 Hence, we will focus our review on later work (1998-2007), with an emphasis on computational methods for extracting information from temperature-modulated signals. Temperature-modulation approaches for MOS sensors can be broadly classified into two categories: (i) thermal transients and (ii) temperature modulation. In thermal transients, the sensor is driven by a step or pulse waveform in the heater voltage, and the discriminatory information is contained in the thermochemical transient induced by the fast change in temperature. Thermal transients have the advantage that one does not need to wait for the sensor to stabilize following power-up, which allows for an immediate evaluation of the signal. In addition, by intermittently powering down the sensor, a significant reduction in power consumption can be achieved. Data analysis for thermal transients resembles that of concentration transients, so that the methods described in section 9 will be generally applicable here as well (see, e.g., ref 286 for an example of multiexponential methods for thermal transients). For temperature modulation, however, the sensor is subjected to a continuous, sometimes periodic, heater voltage variation. To help resolve the various peaks in sensitivity that may occur during such a cycle, a slowly varying sine wave is often used.299 If the heater waveform is slow enough to allow the sensor to settle at the respective temperatures, the behavior of the sensor at each temperature may be treated as a “pseudo-sensor” by virtue of the relationship between operating temperature and sensor selectivity. It is broadly accepted that temperature cycling is the most promising approach to temperature modulation22 and will, therefore, be the focus of this section. Information from the temperature-modulation response can be captured in a variety of ways, but there are three general approaches that parallel those of transient analysis. First, the sensor response can be oversampled at a number of points during the modulation pattern to form a feature vector.300,301 Second, a number of “simple” features can be extracted from the response, such as maxima/minima, and their corresponding occurrence times.301-303 Finally, transform methods such as the fast Fourier transform (FFT) or the discrete wavelet transform (DWT) may be used to convert the temporal response into the frequency or time-frequency domain, respectively. Most of the early work on transient analysis relied on the FFT; see, e.g., refs 19 and 304-308. Recent work, however, indicates that the DWT is a much better choice for processing temperature-modulated patterns, which are markedly nonlinear and nonstationary. The interested reader is referred to ref 309 for a brief introduction to wavelet analysis or to ref 310 for a more thorough presentation.

Chemical Reviews, 2008, Vol. 108, No. 2 597

Corcoran et al.301 performed a systematic comparison of temperature-modulation parameters. An array of eight Taguchi sensors was exposed to the headspace of three types of loose-leaf teas while modulating the operating temperature with a triangular waveform (period ) 240 s, temperature range ) 250-500 °C). Three types of features were extracted from the sensor conductance measurements: single temperature measurements (STM), dynamic parameters (dynamic parameter method, DPM), and total signature differences (TSDs). In the STM, a single measurement was extracted from the temperature-modulated response, yielding an 8-dimensional feature vector; the results of this method served as a benchmark. In the DPM, eight different “simple” measurements were obtained, including the derivative maxima/ minima and their occurrence times, resulting in a 64dimensional feature vector. In the TSD method, the temperature-modulated response was oversampled at 26 different times, yielding a 208-dimensional vector. In addition, a genetic algorithm (GA) was used to select a feature subset from the DPM feature vector using a measure of betweento-within-class scatter as a figure of merit.217 Validation on unseen test data using multilayer perceptrons showed that, despite its relatively high dimensionality, the TSD method provided the best overall performance. DPM features ranked second, whereas STM features performed worst, as expected. Feature subsets from the GA procedure ranked (on average) between STM and DPM features. These results suggest that there was more information in the temperature-modulated response than could be captured by using the simple DPM features. Gutierrez-Osuna et al.300 have investigated the effect of the modulation frequency on the information content and the stability of the sensor patterns. Two metal oxide sensors were exposed to four analytes at dilution levels close to their isothermal detection threshold. The sensors were heated using sinusoidal heater voltage variations of different frequencies (125 mHz, 250 mHz, 500 mHz, 1 Hz, 2 Hz, and 4 Hz) and then exposed to the four analytes during 10 consecutive days. The authors showed that the classification performance decreased monotonically with increasing frequency, since the sensors approached isothermal behavior. Normalization of the raw temperature-modulated response patterns in the [0, 1] range was shown to minimize drift effects at low modulation frequencies, where sufficient discriminatory information is preserved in the shape of the response, but not at high frequencies, at which information tends to be contained in the dc response of the sensor. Building upon extensive prior work,311-315 Nakata et al.316 analyzed the nonlinear properties of a TGS sensor exposed to various target gases under sinusoidal temperature variation. The purpose of this study was to investigate the effect of the sinusoidal dc offset (T0) and the modulation frequency on the sensor response. The effect of T0 is shown in Figure 34, which indicates that the optimum temperature range is dependent on the gas species to be detected. FFT analysis showed that the concentrations and the kinetics of the different gas species were reflected in the higher-order harmonics of the signals.317 Thus, the authors argue that the nonlinear characteristics of chemical sensors should not be viewed as a drawback but rather as a property to be exploited for discrimination purposes. In a subsequent study, Nakata and Ojima318 showed that these higher-order harmonics could be used to estimate the concentration of a target analyte, even in the presence of water vapor. More recently, Nakata et al.319,320 have proposed a method to increase the discrimina-

598 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 34. Dynamic response of a semiconductor sensor to (a) methane, (b) ethane, and (c) propane in different temperature ranges with a modulation frequency of f ) 0.04 Hz. Reprinted with permission from ref 316. Copyright 1998 Elsevier.

tion capabilities of the nonlinear sensor responses by applying a second-harmonic perturbation to the temperature-modulation program. For a given heater voltage modulation, V1(t) ) Va0 + Va1 cos 2πf0t, the authors showed that superimposing a second heater voltage function of the form V2(t) ) Va1/2 cos(4πf0t + θ2) can have a unique effect on the nonlinear sensor response to each target analyte. Thus, by properly selecting the phase shift θ2 of the second-harmonic heater voltage modulation V2(t), the sensor response can be optimized for different analytes. Other authors have also investigated temperature-modulation procedures in recent years. Fort et al.18 have compared the performance of chemical transients, temperature transients, and temperature modulation. In this study, an array of eight metal-oxide sensors was exposed to the headspace of water solutions containing basic constituents of wine. Principal-components analysis suggests that the chemical transients can only be used to detect the presence of esters. In contrast, a PCA of the first, third, and fourth harmonics of the temperature-modulated response shows a clear discrimination of the different solutions. These results suggest that temperature modulation provides maximum discriminatory power.321 Schu¨tze et al.303 used a single semiconductor gas sensor to discriminate six model substances (benzene, diethyl ether, isopentane, methyl butyl ether, methyl alcohol, and propylene oxide). The sensor was operated using two different temperature programs (each consisting of several steps in temperature during a period of 20 s). Then, a number of “simple” features was extracted, such as signal levels at different temperatures and response slopes after a temperature change. The resulting feature vector was processed in a hierarchical fashion, so that different types of features were used to discriminate subgroups of target gases. Most interestingly, the authors showed that a division of the temperaturemodulated conductance-value pattern by its average value almost entirely eliminated the effects of relative humidity in the sample and also improved the repeatability of the responses over a period of several months. Huang et al.322 investigated the effects of temperature modulation, frequency, and waveform on the response patterns of thick-film tinoxide sensors exposed to various gases (butanone, acetone, ethanol, methanol, formaldehyde, and cyclohexanone). The authors compared the sensor responses to temperature pulse trains in five different temperature ranges (25-100, 100150, 150-200, 200-250, and 250-300 °C). In the lowtemperature ranges, the sensor response were shown to be monotonic (first-order response) and did not carry much information, since most reactions occur at the surface level. At high temperatures, response patterns became complex and characteristic of the target gases, as they increasingly

Figure 35. Temperature-programmed sensing: (a) temperature pulse amplitude (20-450 °C), (b) pulse duration (10-300 ms), and (c) delay (5 ms). Notice that the conductance is measured immediately after the sensor returns to room temperature. Reprinted with permission from ref 21. Copyright 1998 Elsevier.

involved bulk reactions. The authors also compared several temperature-modulation waveforms, including rectangular, triangular, sawtooth, sinusoidal, and trapezoidal shapes. Each waveform gave rise to a unique sensor-response pattern, which the authors ascribed to characteristic changes in the actual surface temperatures of the sensor. For well over a decade, Semancik and co-workers at the National Institute of Standards and Technology have used temperature programming for microhotplate-based gas sensors.21,51,323-328 While a review of this technology will be available in an article by Benkstein and Semancik in this issue,42 it is noteworthy that this research group uses a unique approach to measuring the conductivity of the sensors. As illustrated in Figure 35, conductance is always measured at room temperature, so that thermally controlled chemical effects can be separated from temperature-dependent changes in the sensing material; this is possible because of the very low thermal time constant of the microhotplates, which has been estimated to be on the order of a few milliseconds. A great deal of work on temperature modulation has been performed by Llobet and co-workers during the past few years268,329-334 (see also section 11.3). In ref 335, the authors compared the DWT and the FFT for the purpose of extracting information from the response of a tin-oxide microhotplate sensor exposed to mixtures of CO and NO2. The temperature of the sensor was modulated between 243 and 405 °C by means of a 50 mHz sinusoidal waveform. Four temperaturemodulation cycles were used to compute the FFT, from which the amplitudes of the first six harmonics were used as features. In contrast, a single temperature-modulation cycle was used to compute the DWT. Experimental results show that the DWT leads to improved separability as compared to the FFT. In addition, the DWT coefficients can be obtained from a single modulation cycle, whereas the FFT require a larger number of cycles to accurately estimate the spectral content of the signal (a more in-depth analysis of these results may also be found in ref 330). Later studies have also suggested (through simulation) that DWT features are more

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 599

Figure 36. (a) Transient or temperature-modulated response of a sensor array naturally leads to (b) a 2D matrix per sample, and a 3D matrix for all the samples in the dataset. (c) Unfolding the data leads to the traditional 2D data structure, where each row represents a sample and each column represents a feature.

robust than FFT features to additive noise and additive drift.334 Ding et al.336 have also compared the DWT and FFT for the purpose of extracting information from temperaturemodulated signals. Two commercial metal-oxide sensors were exposed to CO, H2, and CH4 at concentrations ranging from 50 to 1000 ppm under a 20 mHz sinusoidal temperature-modulation signal. Sensor signals were first normalized to the [0, 1] range and then processed with the DWT and FFT. The results showed that the DWT features are gasdependent and fairly stable across various concentrations; more importantly, these features were shown to be repeatable across sensor responses recorded 4 months apart. In contrast, FFT harmonics were shown to be noisier and had a more pronounced concentration dependence.

10. Multivariate Calibration Once dynamic features have been applied using the techniques reviewed in section 9, the experimenter will usually build a calibration model to obtain the dependent variables, such as class labels or concentrations, from those features. A number of pattern-recognition techniques are available at this point, which include various statistical methods (nearestneighbor or quadratic classifiers), multilinear regression methods (partial least-squares (PLS) or principal-components regression), and neural networks (multilayer perceptrons, radial basis functions, or support vector machines), to mention but a few. These models have been extensively reviewed in a number of recent articles and book chapters.195,199,217,337,338 For this reason, we will focus our attention on calibration techniques that are particularly well-suited to handle the raw time-dependent response of the sensor, without the need of a preceding feature-extraction stage.

10.1. Multiway Analysis The transient (or temperature-modulated) response of a chemosensor array is naturally represented as a two-dimensional matrix, where each row corresponds to the response of a sensor over time (or operating temperature), and each column represents the response of the array at a particular time or temperature. When the time-dependent response of the array is recorded for multiple samples, then the data set is naturally represented as a 3D matrix, a tensor, as shown

in Figure 36b. While it is possible to unfold this data set into a 2D structure, where each row represents a sample and each column represents a variable (see Figure 36c), this “unfolding” adds extra degrees of freedom to the model, because it treats the response of each sensor at a given time, [x1(t), x2(t), ..., xN(t)]T, as an independent variable, where in reality these measurement data were collected at the same time or at the same temperature. Preserving the multiway structure of the data in Figure 36b can lead to a more parsimonious, i.e., simpler, solution, which is likely to be more robust and easier to interpret. It may also provide the “second-order” advantage discussed in section 1 so that target analytes can be quantified even in the presence of unknown interferents.339,340 Despite these potential advantages, however, multiway methods have only recently received attention in the “electronic nose” literature.197,198 A number of decomposition methods have been developed to analyze multiway data, with the most common being parallel factor analysis (PARAFAC),341,342 Tucker3,343 and unfold-PCA. As illustrated in Figure 37b, PARAFAC decomposes a 3-way data matrix X (a tensor) in a trilinear fashion, F

xijk ) ∑aifbjfckf + eijk

(14)

f)1

where F is the number of factors in the decomposition. Figure 37b shows the case for F ) 2. The solution to this decomposition, i.e., the loading matrices A, B, and C, is commonly found by a method known as alternating leastsquares (ALS), which works as follows: first, two of the loading matrices (say B and C) are initially set to a good starting value,339 and the third matrix (A, in this case) is estimated by least-squares regression from X, B, and C. This process is subsequently repeated for matrix B and then C, and the cycle (reestimate A, then B, then C) is repeated until convergence occurs. It can be shown339 that ALS will improve the solution with every iteration. The algorithm can be computationally intensive, but several acceleration strategies have been devised.339 More importantly, PARAFAC is also known to produce a unique solution under certain rank constraints (e.g., the sum of linearly independent columns in matrices A, B, and C must be larger than or equal to F +

600 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 37. (a) Bilinear decomposition with unfold-PCA. The threeway data matrix is first unfolded into a two-way matrix (see Figure 36 for an example) and then modeled as a sum of terms, with each term being the outer product of two vectors. The term E is the residual error. (b) Trilinear decomposition with PARAFAC. In this case, the three-way data matrix is modeled with three factors, one per dimension. The term E is also a residual error. (c) PARAFAC illustrated in compact form. (d) PARAFAC2 is an extension of PARAFAC, which allows each sensor, k, to have its own time loadings, Bk. (e) Tucker 3 is an extension of PARAFAC, which introduces a core matrix G to allow different interactions among loadings. Adapted from ref 347.

2). Unlike PCA, where the loadings (eigenvectors) and scores (principal components) can be rotated without increasing the reconstruction error, there is only one rotation of the PARAFAC loadings that provides a minimum error: using PCA, one can replace the top eigenvectors with linearly independent combinations of these (which constitutes a rotation) and will still capture the same percentage of the total variance in the data. Unlike two-way data, however, centering and scaling (see section 7) must be done carefully to preserve the trilinearity of the data. The reader is referred to Gurden et al.344 for a discussion of preprocessing strategies for multiway data. Tucker3 (named after Ledyard R. Tucker, who proposed the model in 1966345) provides a more flexible decomposition of the data matrix X, where the main difference with PARAFAC is the addition of a “core” matrix G, which defines how the individual loadings in the different modes (A, B, and C) interact: D

xijk )

E

F

∑ ∑∑aidbjeckfgdef + eijk

(15)

d)1e)1f)1

Finally, unfold-PCA first converts the tensor X into a 2D matrix (see parts b and c of Figure 36) and then performs a bilinear decomposition: F

xij ) ∑aifbjf + eij

(16)

f)1

The decompositions performed by each of the three methods are graphically summarized in Figure 37 (parts a-c and e). It can be shown346 that PARAFAC is a “constrained” version of Tucker3, which, in turn, is a constrained version of unfold-PCA. Here, “constrained” means that there are

fewer degrees of freedom to fit the data. An example by Bro339 will help one to understand this hierarchy of models. Consider an experiment in which 10 samples have been collected from 20 sensors, with each sample measurement extending over 100 s. These data can be represented by a 10 × 100 × 20 matrix. Assume that we seek to decompose these data into five factors. For unfold-PCA, this will produce a model with 10 500 parameters (!), whereas Tucker3 will require 775 parameters and PARAFAC will require only 650 parameters. Clearly, unfold-PCA will provide the best fit to the data in terms of mean-square-error (in fact, PCA does provide optimal reconstruction in the mean-square sense180), whereas Tucker3 and PARAFAC will produce larger errors. But, as has been pointed out in section 9.1, curve-fitting accuracy does not necessarily lead to good analytical performance. If PARAFAC returns results that are reasonable, then it is very likely that the extra degrees of freedom in Tucker3 and unfold-PCA will be used to model noise in the data.339 Thus, all things being equal in terms of curve-fit (and sometimes things not being equal), the simpler model should always be preferred. In the context of chemical sensor arrays, however, PARAFAC may be too restrictive, since it is unable to model shifted profiles or different shapes; this may occur, for instance, if the sensors are placed at different locations along the manifold or have intrinsically different dynamics. In these cases, the additional flexibility of Tucker3 may be helpful.347 However, this comes at a price: unlike PARAFAC, Tucker3 is sensitive to rotational ambiguities, i.e., a unique solution does not exist. Alternatively, an extension of PARAFAC known as PARAFAC2348 may be used in some cases. As illustrated in Figure 37c, PARAFAC2 allows each sensor to have a unique set of time loadings so that PARAFAC2 can deal with non-trilinear data (as Tucker3 does), while a unique solution is ensuredsprovided that some constraints on the Bk matrices are met.197 Shaffer et al.349-351 provided one of the first studies of multiway analysis methods for sensor-based instruments. In this work, the authors developed a second-order instrument that consisted of an array of five surface-acoustic-wave sensors and a preconcentrator unit. Samples of four nerve agents (ethyl N,N-dimethylphosphoramidocyanidate, GA; O-ethyl-S-(2 isopropylaminoethyl)methyl phosphonothiolate, VX; pinacolylmethylphosphofluoridate, GD; isopropylmethylphosphonofluoridate, GB) and one nontoxic simulant (dimethylmethylphosphonate, DMMP) were absorbed in a preconcentrator unit in the presence of several interferents (water, bleach, ammonia, sulfur dioxide, isopropanol, dichloroethane, diesel exhaust, and jet fuel) and then subsequently rapidly desorbed by heating the sorbent column, a process that additionally provided some chromatographic separation of the mixture components. The response of the instrument to a mixture of water, gasoline, and one nerve agent (GA) is shown in Figure 38a. To analyze these responses, the authors compared three types of score plots: (1) PCA performed on the peak signal amplitude of each sensor response, (2) PCA of the peak signal amplitude and the peak location, and (3) unfold-PCA on the entire sensor transient. As shown in Figure 38 (parts b-d), combining peak amplitude and peak location provides better discrimination performance than using peak amplitudes alone; unfold-PCA further decreases the spread of the nerve agent VX and the dimethylmethylphosphonate DMMP clusters, but it also seems to impair the discrimination of the GA samples. A

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 601

Figure 38. (a) Response of an array of SAW sensors to the thermal desorption of a preconcentrated ternary mixture of one nerve agent (GA) and two interferents (water and gasoline). Score plots of (b) PCA of the peak amplitudes, (c) PCA of the peak amplitudes and their locations, and (d) unfold-PCA on the entire sensor transient. Training and test data are depicted as solid squares and open circles, respectively, whereas mixtures of GA with interferents are depicted as crosses. GA, ethyl-N,N-dimethylphosphoramidocyanidate; VX, O-ethyl-S-(2isopropylaminoethyl)methylphosphonothiolate; DMMP, dimethylmethylphosphonate; GD, pinacolylmethylphosphofluoridate; GB, isopropylmethylphosphonofluoridate. Reprinted with permission from ref 351. Copyright 1998 Wiley, New York.

visual comparison of PCA scatter plots, however, can be misleading; a more objective measure of performance is the predictive accuracy on test data. To this end, the authors compared the performance of two classifiers (nearestneighbors and linear discriminants) on three feature vectors: (1) the peak amplitudes, (2) the peak amplitudes and their locations, and (3) the entire sensor transient. Using the entire sensor transient provided the highest performance (96-100% correct classification), closely followed by peak amplitudes and locations (94-98%), and then peak amplitudes (81-83%). A second comparison of four models was performed: unfold-PCA, multiway-PCA (PARAFAC), unfoldPLS, and multiway-PLS. This comparison, however, failed to show any advantage of PARAFAC and multiway-PLS over their unfolded counterparts. Interestingly, the location of the peaks in Figure 38a is sensor-dependent, and the authors report a shift of those peak locations with increasing analyte concentration; both results suggest that the PARAFAC model may have been too restrictive for these data. Furthermore, none of these four models performed better than a direct classification using the raw data. It is quite possible, though, that the lack of improvement may have been a result of ceiling effects, since the raw data could already be classified with 96-100% success. Skov and Bro197 analyzed the transient response of an array of 12 metal-oxide sensors exposed to three kinds of lic-

orice (good, bad, and fabricated bad). The authors applied various types of baseline compression and scaling, and compared the performance of three decomposition methods: PCA on the steady-state signals of the sensors, PARAFAC, and PARAFAC2. For each of the PARAFAC models, a two-factor decomposition was performed. Figure 39a shows the loadings of the PARAFAC decomposition, whereas parts b and c of Figure 39 shows the loadings of PARAFAC2 (two loadings per sensor). While the loadings of PARAFAC are easier to interpret, those of PARAFAC2 indicate that the starting time of the transient responses of some sensors may be shifted, which renders the PARAFAC model too inflexible (interestingly, the “electronic nose” used in this study contained two sensor chambers, which might explain why some sensor transients appear to be shifted in time). This interpretation can be confirmed by analyzing the scores in parts d and e of Figure 39, which show that PARAFAC2 provides much better separability of the three types of licorice. Figure 39f shows the scores when only the steady-state signal of each sensor is used as a feature. While PARAFAC2 seems to return more compact clusters, it also appears that the steady-state signals already contain sufficient information to solve the discrimination problem. Padilla et al.198 have also used PARAFAC to analyze the transient responses of gas sensors. In this study, an array of 13 metal-oxide sensors was exposed to the headspace of

602 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 39. (a) Loadings 1 and 2 of a two-factor PARAFAC decomposition. These loadings define matrix B in Figure 37c. (b) Loading 1 (one per sensor, as illustrated in Figure 37d) in a two-factor PARAFAC2 decomposition. (c) Loading 2 of the same two-factor PARAFAC2 decomposition. Loadings 1 and 2 in (b and c) represent the matrix Bk in Figure 37d. Scores of the samples with (d) PARAFAC, (e) PARAFAC2, and (f) PCA; the PCA decomposition used only the maximum response of each sensor. Separability in (d) is rather poor, which indicates that the PARAFAC model is too restrictive to explain the data. In contrast, PARAFAC2 provides significantly better separability. Note that PARAFAC2 is only marginally better than PCA; this result suggests that the peak response of the sensors already contains most of the discriminatory information. Reprinted with permission from ref 197. Copyright 2005 Elsevier.

potato chips with different amounts of flavor agents. To check for trilinearity, the 3-dimensional data set (samples × sensors × time) was unfolded onto each one of the three dimensions, and the number of factors was computed for each unfolded matrix by means of singular value decomposition. Each of the three matrices appeared to have the same number of factors, which suggested that the data were trilinear.347 The dataset was preprocessed by means of differential baseline correction and standard normal variate normalization methods (see section 7); these techniques were found to preserve the trilinearity of the data, an important safety check before applying PARAFAC. Using a coreconsistency diagnosis proposed by Bro and Kiers,352 the authors determined that the dataset was best described using a three-factor model. The corresponding scores (matrix A in Figure 37c) were then used to predict the concentration of the flavor additives by means of an inverse-least-squares regression model. A correlation coefficient of 0.902 between true concentrations and predictions on calibration data was found using the PARAFAC-ILS model; predictions on a test data set were comparably accurate.

10.2. Dynamical Models Dynamical models may also be used to process information directly from the sensor transient, i.e., without the use of a feature extraction stage. Various types of recurrent neural networks, as well as hidden Markov models, have been used for this purpose, as will be reviewed in this section. Pardo et al.241 investigated various approaches to model the nonlinear inverse dynamics of a gas sensor array. The overarching goal of this study was to build a model that could predict the inputs (concentration pulses) to a gas sensor

system from the sensor responses, in particular for rapid variations of the input concentrations. This inverse problem is known to be ill-posed because of the collinearity across sensors, nonlinearities in the steady-state and the response dynamics of the sensors, and long-term drift, and the fact that the sensors and the flow manifold act as low-pass filters. An array of four thickness-shear-mode resonators with GC stationary-phase coatings was exposed to mixtures of toluene and octane, which were delivered as odor pulses of Gaussiandistributed concentrations. It has to be noted here again that, due to the very short response time of polymer-coated thickness-shear-mode resonators, there is the risk of recording the dynamic gas manifold characteristics rather than those of the sensors. Several models were explored, which included static models, linear autoregressive models, Elman networks, Wiener series expansions, radial-basis function (RBF) networks, and multilayer perceptrons (MLP). In the static models, the concentration inputs were predicted directly from the sensor outputs on a sample-by-sample basis, whereas in the linear autoregressive models, concentration inputs were predicted from a short history of the sensor outputs. Elman networks353 are recurrent neural networks whose hidden units have feedback connections, which serve as a short-term memory to enable the model to “remember” preceding inputs. Wiener series expansions are a parametric model with finite memory that approximates a nonlinear system by a series of functionals, with the advantage of this model being that the parameters can be estimated through least-squares methods.239,240 Finally, radial-basis-function networks and multilayer perceptrons are feedforward neural networks, which act as nonlinear regression models;216 short-term memory in these models was implemented by means of tapped-delay inputs. Results of this study are summarized

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 603

Figure 40. (a) Average prediction capability of various inverse dynamical models. (b) True concentration of toluene (dotted line) vs predictions of the Wiener series expansions (solid line); residual errors are shown as dashed lines. Reprinted with permission from ref 241. Copyright 1998 IEEE. (c) Structure of the time-delay neural network as used in the study of Zhang et al.357 Reprinted with permission from ref 357. Copyright 2003 Elsevier.

in parts a and b of Figure 40. The best results were obtained by the Wiener series, closely followed by the RBF network. Surprisingly, the Elman network did not perform better than the simple linear model. As expected, the static model

showed the poorest performance, since it does not account for the dynamics of the system. Roppel et al.354 used an Elman network to classify analytes using the transient responses of an array of 15 metal-oxide

604 Chemical Reviews, 2008, Vol. 108, No. 2

sensors. Sensor transients were converted into a binary pattern by means of an adaptive-threshold method and then passed on to an Elman network. A network with 15 input units (one per sensor), 15 hidden units, and 9 outputs (one per analyte class) was trained on a set of 27 samples (3 samples of each analyte). For validation purposes, the 27 training samples were presented in a different (random) order. Since the network is time-dependent, this validation procedure can give some indication of the degree to which the network is able to generalize. However, no results were reported on the generalization performance of the network with respect to previously unseen transient signals. More recently, Tan and Wilson355 have used hidden Markov models (HMMs)356 for outlier detection. HMMs are the “gold standard” in automatic speech-processing applications because of their ability to model nonstationary time series. The goal of the study by Tan and Wilson was to determine whether or not HMMs could be used to discriminate “normal” transient responses of a sensor from “unhealthy” ones. Training data consisted of the transient responses of 10 polymer-coated sensors to a concentration step of 5 different analytes. Ten HMMs (1 per sensor) were trained on multiple transient responses to each of the 5 analytes. For validation purposes, each HMM was then tested on transient responses of a different sensor to each of the analytes. HMMs were shown to be able to distinguish “normal” responses (transients of the specific training sensor) from “unhealthy” ones (transients of any other sensor). Zhang et al.357 used a time-delay neural network (TDNN) to classify four different types of spices using the transient responses of an array of 12 conductive-polymer sensors. As shown in Figure 40c, a TDNN is a feed-forward network that has local memory in the form of a tapped-delay line (a first-in-first-out buffer that stores previous values, a very simple form of (short-term) memory) at the inputs and the hidden units. In this study, the tapped delay was replaced by a gamma memory,358 which can be thought of as a cascade of low-pass filters (see insert in Figure 40c). The TDNN was compared to a conventional MLP and a linear-discriminant-function (LDF) method, both trained on the steady-state responses of each sensor. The TDNN was able to correctly classify 100% of the samples in a separate test set, whereas the MLP and LDF provided 63% and 59% correct classification. While these results cannot be extrapolated to other data sets, the large improvement in the classification rate suggests that the TDNN-gamma model is well-suited to exploit differences in the transient responses of gas sensors.

11. Array Optimization As described in the previous sections, a wide variety of sensors and feature extraction methods are available to the experimenter when approaching a new sensing problem. Which of the sensors or features should be selected? How should the experimenter proceed to find the “optimal” combination? Both of these questions are intimately related and have been extensively covered in the literature under the notion of “array optimization” and “feature subset selection”.

11.1. Sensor Selection A number of theoretical studies have addressed the issue of array optimization with nonspecific sensors. One of the earliest investigations was performed by Zaromb and Stetter over 20 years ago.359 The authors assumed an array of S

Hierlemann et al.

sensors, each capable of operating in M distinct modes, to develop a theoretical estimate of the minimum number of parameters P ()S sensors × M modes) that would be required to discriminate mixtures of up to A analytes from a pool of n different analytes. Assuming the sensors to be noiseless and binary (i.e., response/no response), this estimate was shown to be A n! 2P - 1 g ∑ i)1(n - i)!i!

(17)

A rule of thumb is proposed in this study, according to which sensors and operating modes should be selected so that each of the P parameters does not respond to more than P/A individual compounds. While the assumption of noiseless and binary sensor responses is clearly simplistic, the rule of thumb is qualitatively similar to the simulation results of Alkasab et al.,186 which have been discussed in section 6.3. Niebling and Mu¨ller360 proposed an “inverse” feature space to design sensor arrays. In this inverse feature space, each of the n analytes is represented as a separate dimension, and each of the s sensors is represented as a point in this n-dimensional space. The authors show that this visual representation enables the experimenter to detect potential discrimination problems and to design new sensors to address these problems. Gardner and Bartlett37 proposed a computational model for cross-selective sensors that also considers the effects of noise and errors. An upper limit of the number of analytes that can be discriminated by a given array was estimated by the ratio between the total volume of the sensor space and the volume made up by the sensor errors. A measure of performance was proposed, which was essentially equivalent to the classical Fisher’s ratio (i.e., the ratio of between-class distance to within-class variance). More recently, Pearce and Sanchez-Montanes175 have improved the model of Gardner and Bartlett by incorporating the concept of hypervolume of accessible sensor space (VS), which is defined as the volume in sensor space that contains the sensor-array response to a set of analytes. As shown in Figure 41a for a three-odor, two-sensor problem, collinearity limits the number of possible sensor responses. Therefore, the maximum number of analyte mixtures that can be discriminated by the array is limited by the ratio between VS and VN, the hypervolume defined by the accuracy of the sensor array response, as illustrated in Figure 41b. Assuming that errors/noise do not exhibit any correlation with the analyte stimulus, the authors show that the geometric interpretation in Figure 41 can be expressed by means of the Fisher information matrix (FIM), defined as

b) ) ∫p(x b|c b) Jij(c

(

)(

)

∂ ∂ ln p(x b|c b) ln p(x b|c b) dx b (18) ∂ci ∂cj

where b c is a vector containing the concentration of the analytes, b x is the response of the sensor array to the stimulus b c, and p(x b|c b) is the conditional probability of observing the sensor response b x upon a given stimulus b c. The FIM is important because it provides a lower bound (i.e., best-case case) on the accuracy with which the stimulus, b c, can be predicted from the sensor response, b x. This lower limit has been determined as S

b))ii var(c b′|c b) g ∑(J-1(c i)1

(19)

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 605

Figure 41. (a) Visualization of a three-odor-to-two-sensor transformation. (b) The maximum number of feature vectors that can be discriminated is the ratio between the hypervolume of the accessible sensor space (VS) and the accuracy of the sensor array response. Reprinted with permission from ref 175. Copyright 2003 Wiley-VCH, Weinheim.

where b c′ is the prediction of b c from b x through a calibration model. To use these theoretical constructs in practice, one would (1) assume a parametric density p(x b|c b) for each individual sensor, (2) estimate the parameters from experimental data (i.e., by measuring the sensor array responses to a number of analyte mixtures), (3) compute the FIM using eq 18, and (4) compute the expected accuracy of the array from eq 19. This accuracy estimate would then be used as a “figure of merit” to select an optimal array configuration from a pool of cross-selective sensors. Once this “optimal” array has been found, further improvements can be obtained by replicating the array a number of times; see Di Natale et al.361 and Wilson362 for an authoritative discussion of redundancy in sensor arrays.

11.2. Feature Selection In most cases, however, array optimization is approached empirically by defining alternative figures of merit that can be computed more conveniently. This approach is typically referred to as feature subset selection in the pattern-recognition and machine-learning literature. A number of empirical figures of merit can be used for this purpose, which can be grouped into two categories: filters and wrappers.363 A filter

is a measure of the information content provided by a given combination of features, where “information” can be associated with variance (e.g., assessed through the PCA eigenvalues), interclass discrimination (Fisher discrimination, e.g., measured with the LDA eigenvalues), or correlation (e.g., between the feature vectors and the dependent variables), to name but a few. The advantage of this method is that the “figure of merit” is independent of the type of calibration model used to process these features. In contrast, wrappers evaluate each combination of features by the predictive accuracy of the calibration model trained on that particular feature subset, measured by statistical resampling or crossvalidation of a dataset. Each approach has a number of advantages and disadvantages.364 The wrapper approach usually achieves better predictive accuracy since the feature subset can be tuned with respect to the particular bias of the calibration model. In addition, the wrapper has a mechanism to avoid overfitting, since the feature subsets are evaluated according to their performance on test data. Wrappers are, however, computationally intensive, since the calibration model must be continuously retrained. Filters usually find a more general feature subset that works well on a wider range of calibration models, and they are computationally attractive,

606 Chemical Reviews, 2008, Vol. 108, No. 2

but it is difficult to design filters that correlate well with the final predictive accuracy of the calibration model. Owing to their respective pros and cons, both wrappers72,118,264,365-367 and filters368-372 have been used in different applications in the field of “electronic noses”. Once a measure of performance or “figure of merit” has been designed, an “optimal” subset of features must be found. One may be tempted to exhaustively evaluate all possible combinations of features and to then select the global optimum. However, due to combinatorial explosion, exhaustive search is unfeasible for all but very small problems (see, e.g., ref 118 for an exhaustive evaluation). Thus, several methods have been devised that explore the space of all possible feature combinations in a more efficient fashion.180,373 These search strategies can be assigned to three categories:374 (i) exponential, (ii) sequential, and (iii) random strategies. Exponential techniques perform a search whose complexity grows exponentially with the number of states. Among these, branch and bound375 is guaranteed to find the optimal feature subset of a given size if the evaluation function is monotonic. Monotonicity assumes that the addition of a new feature always improves the information content of the subset. This assumption is, however, violated in practical problems, since the addition of features does increase the risk of overfitting. Sequential-search algorithms are strategies that reduce the number of states to be visited during the search by applying local search. The most popular methods include sequential forward selection (SFS) and sequential backward selection (SBS). SFS starts from the empty set and sequentially adds features, whereas SBS starts from the full set and sequentially removes features. These two algorithms, however, have a tendency to become trapped in local minima since they cannot backtrack from there (i.e., SFS cannot remove a feature once it is added, and SBS cannot add a feature once it is removed). More recently, sequential-floating methods with backtracking capabilities have become popular since they do not require monotonicity and often lead to optimal or near-optimal solutions in a fraction of the computation time required by branch and bound. Random search algorithms are an attempt to overcome the computational costs of exponential methods and to avoid the tendency of sequential methods to become trapped in local minima. Among these techniques, simulated annealing376 and genetic algorithms377 are most widely used. Simulated annealing (SA) is based on the annealing process of thermal systems. Starting from an initial solution, SA updates the current solution in a local fashion (e.g., adding or removing a feature). If the new solution is better, it is accepted; if it is worse, it can still be chosen with a probability, P, which depends on a global temperature parameter T. The temperature is initially set to a high value, which allows SA to perform a global search, but T is gradually decreased, which allows the algorithm to converge to a final solution. Genetic algorithms (GAs), on the other hand, are inspired by the process of natural selection. Starting from a random population of solutions, a GA will generate a new population of solutions by means of mutation operations (adding or removing features) and crossover operations (combining features from two parent solutions). Members for the new population are selected probabilistically based on their fitness; better solutions have a higher probability of making it to the new population, but “lessfit” solutions are also allowed in order to promote diversity. Because of their ability to perform global optimization and

Hierlemann et al.

the ever-increasing computational capabilities of personal computers, the tendency in recent years has been to move toward genetic algorithm methods.261,301,366,368,378-384 If the number of potential features is large, the selection procedure can be computationally intensive. Therefore, it may be advantageous to initially “weed out” poor features with a filter and then use a wrapper-based selection on a reduced set of features. It must be noted, however, that the prescreening step may remove features that provide limited but complementary information. Gualdron et al.385 proposed a two-stage selection algorithm, where individual features are first evaluated by their ratio of between-class to withinclass variance. A threshold is set, and only those features whose ratio is higher than the threshold are retained for further selection. The results showed that the performance of this two-step method is comparable to that of a one-step selection procedure, in which all features undergo full subset selection, but it requires only 25% of the computation time. On the basis of this work, Llobet et al.386 developed an improved feature selection procedure for mass-spectrometrybased “electronic nose” instruments. Their method evolves in three stages. During the first stage, every possible pair of features (each being a mass-to-charge ratio) is evaluated according to their Fisher’s discriminant ratio (between-class to within-class scattering), and only the top 30% features are selected. Evaluating features in pairs prevents features with low but complementary information from being thrown away. During the second stage, the Pearson’s correlation between every pair of features is computed, and a collinearity threshold is set so that only the top 20% of the features (the most uncorrelated) are preserved. During the third stage, stochastic methods (simulated annealing and genetic algorithms) are used to perform a suitable subset selection. The overall method was validated on an experimental database of various kinds of Iberian ham. The three-stage algorithm selected 14 out of 111 m/z ratios as features and yielded 95% classification performance on test data, which compared favorably to the 88% achieved by a classifier trained on the entire feature set. In addition to these techniques, a high-level view of the information provided by the different sensors/features may be obtained from a loadings plot of, e.g., PCA, LDA, or PLS. In a loadings plot, each feature is displayed as a point, typically in a 2D or 3D representation. The farther a feature is located from the origin, the more information the feature provides for the analysis (e.g., variance in PCA, discrimination in LDA, correlation with the dependent variable in PLS). Boilot et al.379 performed a sensor fusion from four “electronic nose” instruments, an electronic olfactometer based on a temperature-modulated metal-oxide sensor (INRA), an array of 7 thickness-shear-mode resonators (ROMA), a second array of 8 thickness-shear-mode resonators (UPM), and an array of 32 conductive-polymer sensors (WARWICK). The four instruments were used to measure the headspace of various analyte samples (apple, pear, and peach juices), and a total of 72 features was extracted from the instruments. Figure 42a shows the PCA loadings plot of these features. The analysis of this plot can provide insights on how analyte information is detected by the instruments. First, sensors of the same instrument tend to cluster together, which suggests that they provide correlated information. Second, sensors from the UPM and ROMA instruments also tend to cluster together, a reasonable result since both instruments are based on the same sensor technology. Third, the spread

Higher-Order Chemical Sensing

Chemical Reviews, 2008, Vol. 108, No. 2 607

Figure 42. (a) PCA loadings plot of 72 features extracted from four different “electronic-nose” instruments. Reprinted with permission from ref 379. Copyright 2003 Elsevier. (b) PCA loadings plot of the “simple” transient parameters in the study by Eklo¨v et al.72 Shadowed areas mark groups of features that provide redundant information. Reprinted with permission from ref 72. Copyright 1997 Elsevier.

of the sensors within each instrument is an indication of the degree of collinearity of the sensors; e.g., the conductivepolymer sensors seem to provide very similar information, possibly due to their large inherent cross-sensitivity to the humidity present in the samples. Figure 42b shows the loadings plot of the “simple” parameters in the study of Eklo¨v et al.,72 which was reviewed in section 9.1.4. From this plot, it is possible to identify a number of highly correlated parameters, such as max/on derivative and short-on integral, 1-exp-OnTime constant and TO-60%, off integral and 30soff response. These features describe the same properties of the response curve, and only one of them is thus needed.

11.3. Optimization of Excitation Profiles Much less attention has been paid to the optimization of temperature-modulation profiles for metal-oxide sensors. While a number of articles report on empirical studies with various temperature waveforms (e.g., rectangular, sine, sawtooth, and triangular) and stimulus frequencies,300,304,322 only a handful of studies have approached the problem in a systematic fashion. Kunt et al.21 developed an optimization method for microhotplate devices that works in two stages.

First, a dynamical model of the sensor is developed from experimental data; the model predicts the next conductance value of the sensor (yi+1) from the previous values of the y+1 , as well as from the next and conductance {yk} i-n k)i u+2 previous values of the temperature set points {uk} i-n k)i+1 ,

yi+1 ) F(yi, yi-1, ..., yi-ny+1, ui+1, ui, ..., ui-nu+2) (20) where ny and nu represent the model order. A suitable model F( ) is built from experimental data using a Wavelet network.387 This model can be used to simulate the sensor response to different temperature programs. In the second stage, an optimization routine is used to find the “optimal” program {ui}Ti)1 that maximizes the distance between the (simulated) temperature-modulated sensor responses to two target gases: T {ui} i)1 ) arg maxd(ygas1,ygas2) u1,u2,...,uT

(21)

This procedure is subject to a continuity constraint (|ui+1 ui| e 40 °C) to avoid drastic changes between consecutive

608 Chemical Reviews, 2008, Vol. 108, No. 2

Hierlemann et al.

Figure 43. (a) Normalized conductance response to methanol (solid) and ethanol (dashed) upon applying a linear temperature ramp as shown in (b). (c) Normalized conductance response to methanol (solid) and ethanol (dashed) upon application of the “optimal” temperature program shown in (d); predictions from the model in eq 20 are shown as circles (methanol) and crosses (ethanol). Note the dramatic improvement in discrimination between (a) and (c). Reprinted with permission from ref 21. Copyright 1998 Elsevier.

Figure 44. Optimization of the temperature-modulation frequency using pseudo-random binary sequences. Reprinted with permission from ref 333. Copyright 2005 IEEE.

temperatures. Further improvements in smoothness are achieved by means of a wavelet-based distance that uses only the lower scales of the decomposition (lower scales capture the general shape of the sensor response, whereas higher scales capture its details). When applied to the discrimination of methanol and ethanol, the optimization routine of Kunt et al. returned the temperature program shown in parts c and d of Figure 43. Whereas the sensor responses to ethanol and methanol upon applying a simple linear ramp are highly overlappingssee parts a and b of Figure 43sthe response patterns upon applying the optimal temperature program are nearly orthogonal. Vergara et al.333 have proposed a system-identification method for determining suitable temperature-modulation programs for specific target gases. Their method is based on pseudo-random binary sequences (PRBS) and maximum length sequences (MLSs). PRBS-MLSs are square-wave signals with several interesting properties: (1) they are repeatable, which ensures that the respective results are reproducible, (2) they have a flat power spectrum over a large frequency range, which renders them very suitable for system identification, and (3) they have a maximum length, so that the impulse response of the system can be estimated from the cross-correlation. This method is illustrated in Figure 44 and works as follows. First, a PRBS-MLS is used to drive the sensor heater, while the sensors are exposed to various target compounds (NH3, NO2, and mixtures). For each

individual target compound, the impulse response h(t) is computed as the cross-correlation between the excitation signal (PRBS) and the sensor response, and the spectral components are computed from the FFT of h(t). Second, each individual frequency is ranked on the basis of its information content (between-class to within-class scatter ratio), and a subset of the most informative frequencies is selected. The authors show that this procedure can be used to discriminate and quantify various gases and their mixtures using one sensor and three modulating frequencies. This method was extended in ref 388 to multilevel pseudo-random sequences (ML-PRS), which are better suited than binary sequences to estimate the linear dynamics of a system with nonlinearities. In a subsequent investigation, Vergara et al.268 used the dynamic moments of the sensor’s phase plot,266 which we reviewed in section 9.1.2, to extract information from MLPRS responses. Their results show that similar or better results than those in ref 388 can be obtained with the dynamic moments, while using only a small fraction of the ML-PRS response. Collectively, these studies have demonstrated that temperature profiles with very short time scales can be found that provide a maximum discrimination for a given set of analytes.

12. Conclusion and Outlook It can be concluded from the contents of this articlesand many references hereinsthat the use of various transducer types or inhomogeneous transducer arrays is, indeed, beneficial with regard to the performance of such sensor arrays. In many cases, the data analysis of sensor-array or “electronic nose” instruments has been limited to an empirical qualitative analysis or the drawing of PCA plots. While useful for rapid visualization purposes, PCA plots are not very representative for higher-dimensional measurement/feature spaces, simply because (i) the data are projected onto a two-dimensional plane irrespective of the original or intrinsic dimensionality and (ii) PCA only captures directions of maximum variance, which do not necessarily contain analytical information. A quantitative indicator of the array performance, such as predictive accuracy (e.g., classification rate or mean-square-

Higher-Order Chemical Sensing

error) on unseen test data, should be used as a figure of merit. A careful selection of sensor arrays, feature subsets, and excitation profiles for a given application can further improve the sensor-array performance. Another shortcoming of multisensor-array or “electronic nose” papers is that many of the studies have been performed on food samples or their headspaces, the analyte composition of which has been rather complex, hardly known, and highly variable. Moreover, the qualitative sensor results have not been scientifically explained or substantiated by a chemical gas-phase or headspace composition analysis. Thus, it is not always clear which compounds or chemical effects lead to a discrimination of the different samples. Moreover, sampleto-sample variability, sample deterioration, and the strong influence of the sample preparation and sampling procedure on the sensor results, in particular for natural products, are often underestimated, and the corresponding information is missing in many papers. On the technological side, the progress in micro- and nanotechnology, microelectronics, and in data-processing speed and capability will dramatically influence the development of chemical sensors and sensor systems in the near future: rather complex and versatile microsensor and microanalysis systems operable directly through standard interfaces from a laptop or palmtop by means of standard software are emerging, as has been demonstrated in this article. The end-user is interested in reliable, user-friendly, and affordable sensor systems irrespective of the internal system complexity, which, in most cases, will not be evident to the user anyway. Therefore, we think that a concept of versatile adaptiVe (micro)sensor systems can be most successful. AdaptiVe sensor systems may be devices that include various transducer types, auxiliary sensors, eventually separation and preconcentration units, which can respond or adapt their operation to occurring analysis situations or events. In the event that, e.g., a certain target analyte or a major interferent is detected, the sensor selection, sensor operation mode, feature extraction, and data treatment would be adapted to this situation, and the protocols would be executed in a way that the best-possible target analyte detection is achieved or that the interferent can be recognized and its influence on the sensor system output can be minimized or suppressed. In dealing with interferents, cross-sensitivities, or low signal levels, it may be very effective to purposefully select or deselect sensors or to use signal ratios or differential values instead of merely increasing the array size or the transducer diversity.

13. References (1) Sensors, A ComprehensiVe SurVey; Go¨pel, W., Hesse, J., Zemel, J. N., Eds.; VCH Verlagsgesellschaft mbH: Weinheim, Germany 1991; Vol. 2 (2) Janata, J. Principles of Chemical Sensors; Plenum Press: New York, 1989. (3) Janata, J. Anal. Chem. 2001, 73, 150A. (4) Janata, J.; Josowicz, M.; Devaney, D. M. Anal. Chem. 1994, 66, R207. (5) Edmonds, T. E. Chemical sensors; Chapman and Hall: New York, 1988. (6) Gardner, J. W.; Bartlett, P. N. Electronic noses: principles and applications; Oxford University Press: Oxford, U.K., 1999. (7) Kress-Rogers, E. Handbook of biosensors and electronic noses: Medicine, food, and the enVironment; CRC Press: Boca Raton, FL, 1997. (8) Madou, M. J.; Morrison, S. R. Chemical sensing with solid state deVices; Academic Press: Boston, MA, 1989. (9) Janata, J.; Josowicz, M.; Vanysek, P.; DeVaney, D. M. Anal. Chem. 1998, 70, 179.

Chemical Reviews, 2008, Vol. 108, No. 2 609 (10) Massart, D. L.; Vandeginste, B. G. M.; Deming, S. N.; Michotte, Y.; Kaufman, L. Chemometrics: A Textbook; Elsevier: Amsterdam, The Netherlands, 1988. (11) Hierlemann, A.; Schweizer-Berberich, M.; Weimar, U.; Kraus, G.; Pfau, A.; Go¨pel, W. In Sensors Update; Baltes, H., Go¨pel, W., Hesse, J., Eds.; VCH: Weinheim, Germany, 1996; Vol. 2, p 119. (12) Brereton, R. G. MultiVariate Pattern Recognition in Chemometrics; Elsevier: Amsterdam, The Netherlands, 1992. (13) Weimar, U.; Go¨pel, W. Sens. Actuators, B 1998, 52, 143. (14) Lorber, A.; Kowalski, B. R. J. Chemom. 1988, 2, 67. (15) Lemmo, A. V.; Jo¨rgenson, J. W. Anal. Chem. 1993, 65, 1576. (16) Takatera, K.; Watanabe, T. Anal. Chem. 1993, 65, 759. (17) Lin, Z. H.; Booksh, K. S.; Burgess, L. W.; Kowalski, B. R. Anal. Chem. 1994, 66, 2552. (18) Fort, A.; Gregorkiewitz, M.; Machetti, N.; Rocchi, S.; Serrano, B.; Tondi, L.; Ulivieri, N.; Vignoli, V.; Faglia, G.; Comini, E. Thin Solid Films 2002, 418, 2. (19) Heilig, A.; Barsan, N.; Weimar, U.; Schweizer-Berberich, M.; Gardner, J. W.; Go¨pel, W. Sens. Actuators, B 1997, 43, 45. (20) Janata, J. Proc. IEEE 2003, 91, 864. (21) Kunt, T. A.; McAvoy, T. J.; Cavicchi, R. E.; Semancik, S. Sens. Actuators, B 1998, 53, 24. (22) Lee, A. P.; Reedy, B. J. Sens. Actuators, B 1999, 60, 35. (23) Booksh, K. S.; Kowalski, B. R. Anal. Chem. 1994, 66, 782A. (24) Smilde, A. K.; Tauler, R.; Saurina, J.; Bro, R. Anal. Chim. Acta 1999, 398, 237. (25) Go¨pel, W. Sens. Actuators, B 1998, 52, 125. (26) Elwenspoek, M.; Jansen, H. V. Silicon micromachining; Cambridge University Press: Cambridge, U.K., 1998. (27) Gad-el-Hak, M. The MEMS handbook; CRC Press: Boca Raton, FL, 2002. (28) Kovacs, G. T. A. Micromachined transducers sourcebook; WCB: New York, 1998. (29) Madou, M. Fundamentals of microfabrication; CRC Press: Boca Raton, FL, 1997. (30) Sze, S. M. Semiconductor Sensors; Wiley & Sons: New York, 1994. (31) Hierlemann, A.; Brand, O.; Hagleitner, C.; Baltes, H. Proc. IEEE 2003, 91, 839. (32) Brand, O. Proc. IEEE 2006, 94, 1160. (33) Hagleitner, C.; Hierlemann, A.; Baltes, H. In Sensors Update, Vol. 12; Baltes, H., Korvink, J., Fedder, G., Eds.; Wiley VCH: Weinheim, New York, 2003; pp 51-120. (34) Hagleitner, C.; Hierlemann, A.; Brand, O.; Baltes, H. In Sensors Update, Vol. 11; Baltes, H., Korvink, J., Fedder, G., Eds.: Wiley VCH: Weinheim, New York, 2002; pp 101-155. (35) Lemmerhirt, D. F.; Wise, K. D. Proc. IEEE 2006, 94, 1138. (36) Gardner, J. W.; Bartlett, P. N. Sens. Actuators, B 1994, 18, 211. (37) Gardner, J. W.; Bartlett, P. N. Sens. Actuators, B 1996, 33, 60. (38) Ro¨ck, F.; Barsan, N; Weimar, U. Chem. ReV. in press (cr068121q). (39) Weimar, U.; Go¨pel, W. Sens. Actuators, B 1998, 52, 143. (40) Lee, A. P.; Reedy, B. J. Sens. Actuators, B 1999, 60, 35. (41) Demarne, V.; Grisel, A. Sens. Actuators 1988, 13, 301. (42) Benkstein, K. D.; Semancik, S. Chem. ReV. in press. (43) Nakamoto, T.; Ishida, H. Chem. ReV. in press (cr068117e). (44) Demarne, V.; Balkanova, S.; Grisel, A.; Rosenfeld, D.; Levy, F. Sens. Actuators, B 1993, B14, 497. (45) Graf, M.; Barrettino, D.; Taschini, S.; Hagleitner, C.; Hierlemann, A.; Baltes, H. Anal. Chem. 2004, 76, 4437. (46) Graf, M.; Jurischka, R.; Barrettino, D.; Hierlemann, A. J. Micromech. Microeng. 2005, 15, 190. (47) Heilig, A.; Barsan, N.; Weimar, U.; Schweizer-Berberich, M.; Gardner, J. W.; Go¨pel, W. Sens. Actuators, B 1997, 43, 45. (48) Cavicchi, R. E.; Suehle, J. S.; Kreider, K. G.; Gaitan, M.; Chaparala, P. IEEE Electron DeVice Lett. 1995, 16, 286. (49) Cavicchi, R. E.; Suehle, J. S.; Kreider, K. G.; Gaitan, M.; Chaparala, P. Sens. Actuators, B 1996, 33, 142. (50) Kunt, T. A.; McAvoy, T. J.; Cavicchi, R. E.; Semancik, S. Sens. Actuators, B 1998, 53, 24. (51) Semancik, S.; Cavicchi, R. Acc. Chem. Res. 1998, 31, 279. (52) Mu¨ller, G.; Friedberger, A.; Kreisl, P.; Ahlers, S.; Schulz, O.; Becker, T. Thin Solid Films 2003, 436, 345. (53) Graf, M.; Frey, U.; Taschini, S.; Hierlemann, A. Anal. Chem. 2006, 78, 6801. (54) Graf, M.; Mu¨ller, S. K.; Barrettino, D.; Hierlemann, A. IEEE Electron DeVice Lett. 2005, 26, 295. (55) Althainz, P.; Goschnick, J.; Ehrmann, S.; Ache, H. J. Sens. Actuators, B 1996, 33, 72. (56) Menzel, R.; Goschnick, J. Sens. Actuators, B 2000, 68, 115. (57) Sysoev, V. V.; Kiselev, I.; Frietsch, M.; Goschnick, J. Sensors 2004, 4, 37. (58) Barsan, N.; Stetter, J. R.; Findlay, M.; Go¨pel, W. Sens. Actuators, B 2000, 66, 31.

610 Chemical Reviews, 2008, Vol. 108, No. 2 (59) Cavicchi, R. E.; Semancik, S.; Walton, R. M.; Panchapakesan, B.; DeVoe, D. L.; Aquino, Class, M.; Allen, J. D.; Suehle, J. S. Proc. SPIE 1999, 3857, 38. (60) Cavicchi, R. E.; Suehle, J. S.; Chaparala, P.; Kreider, K. G.; Gaitan, M.; Semancik, S. Technical Digest Solid-State Sensor and Actuator Workshop; Hilton Head Island, SC, 1994; p 53. (61) Geistlinger, H. Sens. Actuators, B 1993, B17, 47. (62) Semancik, S.; Cavicchi, R. E.; Wheeler, M. C.; Tiffany, J. E.; Poirier, G. E.; Walton, R. M.; Suehle, J. S.; Panchapakesan, B.; DeVoe, D. L. Sens. Actuators, B 2001, B77, 579. (63) Wlodek, S.; Colbow, K.; Consadori, F. Sens. Actuators, B 1991, 3, 63. (64) Wlodek, S.; Colbow, K.; Consadori, F. Sens. Actuators, B 1991, 3, 123. (65) Clifford, P. K.; Tuma, D. T. Sens. Actuators 1982, 3, 255. (66) Simon, T.; Barsan, N.; Bauer, M.; Weimar, U. Sens. Actuators, B 2001, 73, 1. (67) Schweizer-Berberich, M.; Vaihinger, S.; Go¨pel, W. Sens. Actuators, B 1994, 18-19, 282. (68) (a)Stetter, J. R.; Zaromb, S.; Findlay, M. W. Sens. Actuators 1984, 6, 269. (b) Otagawa, T.; Stetter, J. R. Sens. Actuators 1987, 11, 251. (69) Vaihinger, S.; Go¨pel, W.; Stetter, J. R. Sens. Actuators, B 1991, 4, 337. (70) Kummer, A. M.; Burg, T. P.; Hierlemann, A. Anal. Chem. 2006, 78, 279. (71) Gouws, G. J.; Gouws, D. J. Sens. Actuators, B 2003, 91, 326. (72) Eklo¨v, T.; Martensson, P.; Lundstro¨m, I. Anal. Chim. Acta 1997, 353, 291. (73) Gardner, J. W. Sens. Actuators, B 1990, 1, 166. (74) Fort, A.; Machetti, N.; Rocchi, S.; Santos, M. B. S.; Tondi, L.; Ulivieri, N.; Vignoli, V.; Sberveglieri, G. IEEE Trans. Instrum. Meas. 2003, 52, 921. (75) Llobet, E.; Brezmes, J.; Vilanova, X.; Sueiras, J. E.; Correig, X. Sens. Actuators, B 1997, 41, 13. (76) Llobet, E.; Vilanova, X.; Brezmes, J.; Sueiras, J. E.; Correig, X. Sens. Actuators, B 1998, 47, 104. (77) Yamanaka, T.; Ishida, H.; Nakamoto, T.; Moriizumi, T. Sens. Actuators, A 1998, 69, 77. (78) Di Nucci, C.; Fort, A.; Rocchi, S.; Tondi, L.; Vignoli, V.; Di Francesco, F.; Santos, M. B. S. IEEE Trans. Instrum. Meas. 2003, 52, 1079. (79) Islam, A.; Ismail, Z.; Ahmad, M. N.; Saad, B.; Othman, A. R.; Shakaff, A. Y. M.; Daud, A.; Ishak, Z. Sens. Actuators, B 2005, 109, 238. (80) Nanto, H.; Kondo, K.; Habara, M.; Douguchi, Y.; Waite, R. I.; Nakazumi, H. Sens. Actuators, B 1996, 35, 183. (81) Hagleitner, C.; Lange, D.; Hierlemann, A.; Brand, O.; Baltes, H. IEEE J. Solid State Circuits 2002, 37, 1867. (82) Langereis, G. R.; Olthuis, W.; Bergveld, P. Sens. Actuators, B 1998, 53, 197. (83) Poghossian, A.; Berndsen, L.; Schultze, J. W.; Lu¨th, H.; Scho¨ning, M. J. Proceedings of the Meeting of the Electrochemical Society, San Francisco, CA, 2001; p 143. (84) Poghossian, A.; Lu¨th, H.; Schultze, J. W.; Scho¨ning, M. J. Electrochim. Acta 2001, 47, 243. (85) Poghossian, A.; Scho¨ning, M. I. Electroanalysis 2004, 16, 1863. (86) Poghossian, A.; Schultze, J. W.; Scho¨ning, M. J. Sens. Actuators, B 2003, 91, 83. (87) Scho¨ning, M. J.; Poghossian, A. Electroanalysis 2006, 18, 1893. (88) Hagleitner, C.; Hierlemann, A.; Lange, D.; Kummer, A.; Kerness, N.; Brand, O.; Baltes, H. Nature 2001, 414, 293. (89) Liu, W. L.; Ma, W. C.; Li, L.; Li, J. M. Sens. Actuators, B 2000, 66, 222. (90) Takada, T. Sens. Actuators, B 2000, 66, 1. (91) Heilig, A.; Barsan, N.; Weimar, U.; Go¨pel, W. Sens. Actuators, B 1999, 58, 302. (92) Mizsei, J.; Lantto, V. Sens. Actuators, B 1991, 4, 163. (93) Schierbaum, K. D.; Weimar, U.; Go¨pel, W.; Kowalkowski, R. Sens. Actuators, B 1991, 3, 205. (94) Weimar, U.; Schierbaum, K. D.; Go¨pel, W.; Kowalkowski, R. Sens. Actuators, B 1990, 1, 93. (95) Cardinali, G. C.; Dori, L.; Fiorini, M.; Sayago, I.; Faglia, G.; Perego, C.; Sberveglieri, G.; Liberali, V.; Maloberti, F.; Tonietto, D. Analog Integ. Circuits Signal Process. 1997, 14, 2756. (96) Hughes, K. L.; Miller, S. L.; Rodriguez, J. L.; McWhorter, P. J. Sens. Actuators, B 1996, 37, 75. (97) Rodriguez, J. L.; Hughes, R. C.; Corbett, W. T.; McWhorter, P. J. Technical Digest of IEEE International Electron DeVices Meeting, San Francisco, CA, 1992; p 521. (98) Topart, P.; Josowicz, M. J. Phys. Chem. 1992, 96, 8662. (99) Kunugi, Y.; Nigorikawa, K.; Harima, Y.; Yamashita, K. J. Chem. Soc., Chem. Commun. 1994, 873.

Hierlemann et al. (100) Polk, B. J.; Smith, J. A.; DeWeerth, S. P.; Zhou, Z. P.; Janata, J.; Domansky, K. Electroanalysis 1999, 11, 707-711. (101) Penza, M.; Cassano, G.; Aversa, P.; Antolini, F.; Cusano, A.; Consales, M.; Giordano, M.; Nicolais, L. Sens. Actuators, B 2005, 111-112, 171. (102) Penza, M.; Cassano, G.; Aversa, P.; Cusano, A.; Cutolo, A.; Giordano, M.; Nicolais, L. Nanotechnology 2005, 16, 2536. (103) Endres, H. E.; Mickle, L. D.; Ko¨stlinger, C.; Drost, S.; Hutter, F. Sens. Actuators, B 1992, 6, 285. (104) Haug, M.; Schierbaum, K. D.; Gauglitz, G.; Go¨pel, W. Sens. Actuators, B 1993, 11, 383. (105) Schierbaum, K. D.; Gerlach, A.; Haug, M.; Go¨pel, W. Sens. Actuators, A 1992, 31, 130. (106) Zhou, R.; Hierlemann, A.; Weimar, U.; Go¨pel, W. Sens. Actuators, B 1996, 34, 356. (107) Kurzawski, P.; Hagleitner, C.; Hierlemann, A. Anal. Chem. 2006, 78, 6910. (108) Kummer, A. M.; Hierlemann, A.; Baltes, H. Anal. Chem. 2004, 76, 2470. (109) Kummer, A. Ph.D. Thesis, ETH Zu¨rich, Switzerland, 2004. (110) Hagleitner, C.; Hierlemann, A.; Brand, O.; Baltes, H. In Sensors Update, Vol 11; Baltes, H., Fedder, G. K., Korvink, J. G., Eds.; Wiley VCH: Weinheim, Germany, 2002; pp 101-155. (111) Hierlemann, A.; Ricco, A. J.; Bodenho¨fer, K.; Dominik, A.; Go¨pel, W. Anal. Chem. 2000, 72, 3696. (112) Bodenho¨fer, K.; Hierlemann, A.; Seemann, J.; Gauglitz, G.; Christian, B.; Koppenhoefer, B.; Go¨pel, W. Anal. Chem. 1997, 69, 3058. (113) Bodenho¨fer, K.; Hierlemann, A.; Seemann, J.; Gauglitz, G.; Koppenhoefer, B.; Go¨pel, W. Nature 1997, 387, 577. (114) Lerchner, J.; Kirchner, R.; Seidel, J.; Wahlisch, D.; Wolf, G.; Konig, W. A. Thermochim. Acta 2006, 445, 98. (115) Holmberg, M.; Winquist, F.; Lundstro¨m, I.; Gardner, J. W.; Hines, E. L. Sens. Actuators, B 1995, 27, 246. (116) Ulmer, H.; Mitrovics, J.; Noetzel, G.; Weimar, U.; Go¨pel, W. Sens. Actuators, B 1997, 43, 24. (117) Ulmer, H.; Mitrovics, J.; Weimar, U.; Go¨pel, W. Sens. Actuators, B 2000, 65, 79. (118) Pardo, M.; Kwong, L. G.; Sberveglieri, G.; Brubaker, K.; Schneider, J. F.; Penrose, W. R.; Stetter, J. R. Sens. Actuators, B 2005, 106, 136-143. (119) Mitrovics, J.; Ulmer, H.; Weimar, U.; Go¨pel, W. Acc. Chem. Res. 1998, 31, 307. (120) http://www.alpha-mos.com/en/index.php. (121) http://www.appliedsensor.com/. (122) Li, Y.; Vancura, C.; Barrettino, D.; Graf, M.; Hagleitner, C.; Kummer, A.; Zimmermann, M.; Kirstein, K. U.; Hierlemann, A. Sens. Actuators, B 2007, 126, 431-440. (123) Vancura, C.; Ruegg, M.; Li, Y.; Hagleitner, C.; Hierlemann, A. Anal. Chem. 2005, 77, 2690. (124) Hierlemann, A.; Baltes, H. Analyst 2003, 128, 15. (125) Hierlemann, A.; Brand, O.; Hagleitner, C.; Baltes, H. Proc. IEEE 2003, 91, 839. (126) Cole, M.; Ulivieri, N.; Garcia-Guzman, J.; Gardner, J. W. Microelectron. J. 2003, 34, 865. (127) Garcia-Guzman, J.; Ulivieri, N.; Cole, M.; Gardner, J. W. Sens. Actuators, B 2003, 95, 232. (128) Mielle, P.; Marquis, F. Sens. Actuators, B 1999, 58, 526. (129) Mielle, P.; Souchaud, M.; Landy, P.; Guichard, E. Sens. Actuators, B 2006, 116, 161. (130) http://www.gsg-analytical.com/english/moses2.htm. (131) http://www.airsense.com. (132) http://www.alpha-mos.com. (133) http://microsensorsystems.com. (134) http://www.raesystems.com/global. (135) Norlin, P.; O ¨ hman, O.; Ekstro¨m, B.; Forssen, L. Sens. Actuators, B 1998, 49, 34. (136) Go¨tz, A.; Gracia, I.; Cane, C.; Morrissey, A.; Alderman, J. J. Microelectromech. Syst. 2001, 10, 569. (137) Kloock, J. P.; Moreno, L.; Bratov, A.; Huachupoma, S.; Xu, J.; Wagner, T.; Yoshinobu, T.; Ermolenko, Y.; Vlasov, Y. G.; Scho¨ning, M. J. Sens. Actuators, B 2006, 118, 149. (138) Lauwers, E.; Suls, J.; Gumbrecht, W.; Maes, D.; Gielen, G.; Sansen, W. IEEE J. Solid State Circuits 2001, 36, 2030. (139) Gumbrecht, W.; Peters, D.; Schelter, W.; Erhardt, W.; Henke, J.; Steil, J.; Sykora, U. Sens. Actuators, B 1994, 19, 704. (140) Schelter, W.; Gumbrecht, W.; Montag, B.; Sykora, U.; Erhardt, W. Sens. Actuators, B 1992, 6, 91. (141) http://www.abbottpointofcare.com/istat/www/products/cartridgesfamily.htm. (142) Baumann, W. H.; Lehmann, M.; Schwinde, A.; Ehret, R.; Brischwein, M.; Wolf, B. Sens. Actuators, B 1999, 55, 77.

Higher-Order Chemical Sensing (143) (a) Gross, G.; Rhoades, B. K.; Azzazy, H. M. E.; Ming, Chi, W. Biosens. Bioelectron. 1995, 10, 553. (b) Gross, G. W.; Rhoades, B.; Jordan, R. Sens. Actuators, B 1992, B6, 1. (144) Gross, G. W.; Harsch, A.; Rhoades, B. K.; Go¨pel, W. Biosens. Bioelectron. 1997, 12, 373. (145) McConnell, H. M.; Owicki, J. C.; Parce, J. W.; Miller, D. L.; Baxter, G. T.; Wada, H. G.; Pitchford, S. Science 1992, 257, 1906. (146) Kovacs, G. T. A. Proc. IEEE 2003, 91, 915. (147) Pancrazio, J. J.; Bey, P. P., Jr.; Cuttino, D. S.; Kusel, J. K.; Borkholder, D. A.; Shaffer, K. M.; Kovacs, G. T. A.; Stenger, D. A. Sens. Actuators, B 1998, B53, 179. (148) Pancrazio, J. J.; Whelan, J. P.; Borkholder, D. A.; Ma, W.; Stenger, D. A. Ann. Biomed. Eng. 1999, 27, 687. (149) Baumann, W.; Schreiber, E.; Krause, G.; Stu¨we, S. Proc. Eurosensors XVI, Prag 2002, 4. (150) Ehret, R.; Baumann, W.; Brischwein, M.; Lehmann, M.; Henning, T.; Freund, I.; Drechsler, S.; Friedrich, U.; Hubert, M. L.; Motrescu, E.; Kob, A.; Palzer, H.; Grothe, H.; Wolf, B. Fresenius J. Anal. Chem. 2001, 369, 30. (151) Lehmann, M.; Baumann, W.; Brischwein, M.; Ehret, R.; Kraus, M.; Schwinde, A.; Bitzenhofer, M.; Freund, I.; Wolf, B. Biosens. Bioelectron. 2000, 15, 117. (b) Ehret, R.; Baumann, W.; Brischwein, M.; Schwinde, A.; Wolf, B. Med. Biol. Eng. Comput. 1998, 36, 365. (152) Ziegler, C. Fresenius J. Anal. Chem. 2000, 366, 552. (153) Martin, S. J.; Ricco, A. J.; Ginley, D. S.; Zipperian, T. E. IEEE Trans. UFFC 1987, 34 (2), 142. (154) Grate, J. W.; Egorov, O. B.; O’Hara, M. J.; DeVol, T. A. Chem. ReV. in press (cr068115u). (155) Potkay, J. A.; Driscoll, J. A.; Agah, M.; Sacks, R. D.; Wise, K. D. Proc. IEEE Micro Electro Mechanical Systems MEMS, 2003, Kyoto, Japan 2003, 395-398. (156) Tian, W. C.; Chan, H. K. L.; Lu, C. J.; Pang, S. W.; Zellers, E. T. J. Microelectromech. Syst. 2005, 14, 498. (157) Tian, W. C.; Pang, S. W.; Lu, C. J.; Zellers, E. T. J. Microelectromech. Syst. 2003, 12, 264. (158) Lewis, P. R.; Manginell, R. P.; Adkins, D. R.; Kottenstette, R. J.; Wheeler, D. R.; Sokolowski, S. S.; Trudell, D. E.; Byrnes, J. E.; Okandan, M.; Bauer, J. M.; Manley, R. G.; Frye-Mason, G. C. IEEE Sens. J. 2006, 6, 784. (159) Lu, C. J.; Jin, C. G.; Zellers, E. T. J. EnViron. Monit. 2006, 8, 270. (160) Lu, C. J.; Steinecker, W. H.; Tian, W. C.; Oborny, M. C.; Nichols, J. M.; Agah, M.; Potkay, J. A.; Chan, H. K. L.; Driscoll, J.; Sacks, R. D.; Wise, K. D.; Pang, S. W.; Zellers, E. T. Lab Chip 2005, 5, 1123. (161) Bender, F.; Barie, N.; Romoudis, G.; Voigt, A.; Rapp, A. Sens. Actuators, B 2003, 93, 135. (162) Cho, S. M.; Kim, Y. J.; Heo, G. S.; Shin, S.-M. Sens. Actuators, B 2006, 117, 50. (163) Lu, C. J.; Whiting, J.; Sacks, R. D.; Zellers, E. T. Anal. Chem. 2003, 75, 1400. (164) Nakamoto, T.; Sukegawa, K.; Sumitomo, E. IEEE Sens. J. 2005, 5, 68. (165) Rock, F.; Gurlo, A.; Weimar, U. Anal. Chem. 2005, 77, 2762. (166) Wanekaya, A. K.; Uematsu, M.; Breimer, M.; Sadik, O. A. Sens. Actuators, B 2005, 110, 41. (167) Pillonel, L.; Bosset, J. O.; Tabacchi, R. Lebensm.-Wiss. Technol. 2002, 35, 1. (168) Hossein-Babaei, F.; Hemmati, M.; Dehmobed, M. Sens. Actuators, B 2005, 107, 461. (169) Terry, S. C.; Jerman, J. H.; Angell, J. B. IEEE Trans. Electron DeVices 1979, 26, 1880. (170) Reston, R. R.; Kolesar, E. S. J. Microelectromech. Syst. 1994, 3, 134. (171) Cai, Q. Y.; Zellers, E. T. Anal. Chem. 2002, 74, 3533. (172) Wohltjen, H.; Snow, A. W. Anal. Chem. 1998, 70, 2856. (173) Mowry, C. D.; Morgan, C. H.; Manginell, R. P.; Kottenstette, R. J.; Lewis, P. R.; Frye-Mason, G. C.; Baca, Q. In Chemical and Biological Early Warning Monitoring for Water, Food, and Ground; Jensen, J. L., Burggraf, L. W., Eds.; 2002; Vol. 4575, p 83. (174) (a) Wilson, D. M.; Gaffod, S. D. IEEE Sens. J. 2002, 2, 169. (b) Zubritsky, E. Anal. Chem. 2000, 72, 421A. (175) Pearce, T. C.; Sanchez-Montanes, M. In Handbook of Machine Olfaction; Pierce, T. C., Schiffman, S., Nagle, H. T., Gardner, J. W., Eds.; Wiley VCH: Weinheim, New York, 2003; pp 347-375. (176) Li, J.; Hermann, B.; Hobbs, S.; Severin, E. In 7th International Symposium on Olfaction and Electronic Nose (ISOEN), Brighton, U.K., 2000; p 81. (177) Stetter, J. R.; Penrose, W. R. Sensors Update, Vol. 10; Baltes, H., Korvink, J., Fedder, G., Eds.; Wiley VCH: Weinheim, New York, 2002; p 189. (178) Jimenez, L. O.; Landgrebe, D. A. IEEE Trans. Syst. Man Cybern. C 1998, 28, 39.

Chemical Reviews, 2008, Vol. 108, No. 2 611 (179) Kendall, M. G. A course in the geometry of n dimensions; Hafner Pub. Co.: New York, 1961. (180) Fukunaga, K. Introduction to statistical pattern recognition, 2nd ed.; Academic Press: Boston, MA, 1990. (181) Hwang, J.-N.; Lay, S.-R.; Lippman, A. IEEE Trans. Signal Process. 1994, 42, 2795-2810. (182) Hughes, G. IEEE Trans. Inf. Theory 1968, 14, 55. (183) (a) Ulmer, H.; Mitrovics, J.; Weimar, U.; Go¨pel, W. Sens. Actuators, B 2000, 65, 79. (b) Ricco, A. J.; Crooks, R. M.; Osbourn, G. C. Acc. Chem. Res. 1998, 31, 289. (184) Pearce, T. C. Neurocomputing 2000, 32-33, 941. (185) Brown, W. M.; Ba¨cker, A. Neural Comput. 2006, 18, 1511. (186) Alkasab, T. K.; White, J.; Kauer, J. S. Chem. Senses 2002, 27, 261. (187) Sanchez-Montanes, M. A.; Pearce, T. C. Biosystems 2002, 67, 229. (188) Wilke, S. D.; Eurich, C. W. Neural Comput. 2002, 14, 155. (189) Ciosek, P.; Wroblewski, W. Talanta 2007, 71, 738. (190) Firestein, S. Nature 2001, 413, 211. (191) Pearce, T. C.; Verschure, P. F. M. J.; White, J.; Kauer, J. S. In Neural Computation Architectures Based on Neuroscience; Wermter, S. A. J., Willshaw, D., Eds.; Springer-Verlag: New York, 2001; p 461. (192) Wilson, D. M.; Garrod, S.; Hoyt, S.; McKennoch, S.; Booksh, K. S. In Sensors Update, Vol. 10; Wiley VCH: Weinheim, New York, 2002; p 77. (193) Gardner, J. W.; Craven, M.; Dow, C.; Hines, E. L. Meas. Sci. Technol. 1998, 9, 120. (194) Gutierrez-Osuna, R.; Nagle, H. T.; Kermani, B.; Schiffman, S. S. In Handbook of Machine Olfaction: Electronic Nose Technology; Pearce, T. C., Schiffman, S. S., Nagle, H. T., Gardner, J. W., Eds.; VCH: Weinheim, Germany, 2002; pp 105-132. (195) Jurs, P. C.; Bakken, G. A.; McClelland, H. E. Chem. ReV. 2000, 100, 2649. (196) Gutierrez-Osuna, R.; Nagle, H. T. IEEE Trans. Syst. Man Cybern. B 1999, 29, 626. (197) Skov, T.; Bro, R. Sens. Actuators, B 2005, 106, 719. (198) Padilla, M.; Montoliu, I.; Pardo, A.; Perera, A.; Marco, S. Sens. Actuators, B 2006, 116, 145. (199) Scott, S. M.; James, D.; Ali, Z. Microchim. Acta 2007, V156, 183. (200) Vaihinger, S.; Go¨pel, W. In Sensors: A ComprehensiVe SurVey; Go¨pel, W., Hesse, J., Zemel, J. N., Eds.; VCH Publishers: Weinheim, Germany, 1991; Vol. 2. (201) Hierlemann, A.; Schweizer-Berberich, M.; Weimar, U.; Kraus, G.; Pfau, A.; Go¨pel, W. In Sensors Update, Vol. 2; Baltes, H., Go¨pel, W., Hesse, J., Eds.; VCH: Weinheim, Germany, 1996; pp 119180. (202) Gardner, J. W. Sens. Actuators, B 1991, 4, 109. (203) Hierlemann, A.; Weimar, U.; Kraus, G.; Schweizer-Berberich, M.; Go¨pel, W. Sens. Actuators, B 1995, 26, 126. (204) Grate, J. W.; Martin, S. J.; White, R. M. Anal. Chem. 1993, 65, 940A. (205) Sundgren, H.; Winquist, F.; Lukkari, I.; Lundstro¨m, I. Meas. Sci. Technol. 1991, 2, 464. (206) Gardner, J. W.; Bartlett, P. N. Electronic noses: principles and applications; Oxford University Press: Oxford, New York, 1999. (207) Persaud, K. C.; Khaffaf, S. M.; Payne, J. S.; Pisanelli, A. M.; Lee, D.-H.; Byun, H.-G. Sens. Actuators, B 1996, 36, 267. (208) Severin, E. J.; Doleman, B. J.; Lewis, N. S. Anal. Chem. 2000, 72, 658. (209) Geladi, P.; MacDougall, D.; Martens, H. Appl. Spectrosc. 1985, 39, 491. (210) Isaksson, T.; Naes, T. Appl. Spectrosc. 1988, 42, 1273. (211) Naes, T.; Isaksson, T.; Kowalski, B. Anal. Chem. 1990, 62, 664. (212) Zeaiter, M.; Roger, J. M.; Bellon-Maurel, V. Trends Anal. Chem. 2005, 24, 437. (213) Trihaas, J.; Nielsen, P. V. J. Food Sci. 2005, 70, E44. (214) Daszykowski, M.; Kaczmarek, K.; Vander Heyden, Y.; Walczak, B. Chemom. Intell. Lab. Syst. 2007, 85, 203. (215) Duda, R. O.; Hart, P. E.; Stork, D. G. Pattern classification, 2nd ed.; Wiley: New York, 2001. (216) Bishop, C. M. Pattern recognition and machine learning; Springer: New York, 2006. (217) Gutierrez-Osuna, R. IEEE Sens. J. 2002, 2, 189. (218) Goschnick, J.; Magapu, V.; Kiselev, I.; Koronczi, I. Sens. Actuators, B 2006, 116, 85. (219) Barnes, R. J.; Dhanoa, M. S.; Lister, S. J. Appl. Spectrosc. 1989, 43, 772. (220) Johansson, E.; Wold, S.; Sjoedin, K. Anal. Chem. 1984, 56, 1685. (221) Khalheim, O. M. Anal. Chim. Acta 1985, 177, 71. (222) Guo, Q.; Wu, W.; Massart, D. L. Anal. Chim. Acta 1999, 382, 87. (223) Box, G. E. P.; Cox, D. R. J. R. Stat. Soc. B 1964, 26, 211. (224) Horner, G.; Hierold, C. Sens. Actuators, B 1990, 2, 173. (225) Holmberg, M.; Artursson, T. In Handbook of Machine Olfaction: Electronic Nose Technology; Pearce, T. C., Schiffman, S. S., Nagle, H. T. Gardner, J. W., Eds.; 2002; pp 325-346.

612 Chemical Reviews, 2008, Vol. 108, No. 2 (226) Roth, M.; Hartinger, R.; Faul, R.; Endres, H. E. Sens. Actuators, B 1996, 36, 358. (227) Aigner, R.; Auerbach, F.; Huber, P.; Mu¨ller, R.; Scheller, G. Sens. Actuators, B 1994, 18, 143. (228) Fryder, M.; Holmberg, M.; Winquist, F.; Lu¨ndstrom, I. Proc. Transducers’ 95 and Eurosensors IX, Stockholm, Sweden, 1995; p 683. (229) Haugen, J.-E.; Tomic, O.; Kvaal, K. Anal. Chim. Acta 2000, 407, 23. (230) Sisk, B. C.; Lewis, N. S. Sens. Actuators, B 2005, 104, 249. (231) Holmberg, M.; Winquist, F.; Lundstro¨m, I.; Davide, F.; DiNatale, C.; D’Amico, A. Sens. Actuators, B 1996, 36, 528. (232) Kohonen, T. Biol. Cybern. 1982, 43, 59. (233) Davide, F. A. M.; Di Natale, C.; D’Amico, A. Sens. Actuators, B 1994, 18, 244. (234) Marco, S.; Ortega, A.; Pardo, A.; Samitier, J. IEEE Trans. Instrum. Meas. 1998, 47, 316. (235) Marco, S.; Pardo, A.; Ortega, A.; Samitier, J. Proc. IEEE Instrumentation and Measurement Technology Conference, Ottawa, Canada, 1995; p 904. (236) Zuppa, M.; Distante, C.; Siciliano, P.; Persaud, K. C. Sens. Actuators, B 2004, 98, 305. (237) Distante, C.; Siciliano, P.; Persaud, K. C. Pattern Anal. Appl. 2002, 5, 306. (238) Holmberg, M.; Davide, F. A. M.; Di, Natale, C.; D’Amico, A.; Winquist, F.; Lundstro¨m, I. Sens. Actuators, B 1997, 42, 185. (239) Davide, F. A. M.; Natale, C. D.; D’Amico, A.; Hierlemann, A.; Mitrovics, J.; Schweizer, M.; Weimar, U.; Go¨pel, W.; Marco, S.; Pardo, A. Sens. Actuators, B 1995, 27, 275. (240) Marco, S.; Pardo, A.; Davide, F. A. M.; Natale, C. D.; D’Amico, A.; Hierlemann, A.; Mitrovics, J.; Schweizer, M.; Weimar, U.; Go¨pel, W. Sens. Actuators, B 1996, 34, 213. (241) Pardo, A.; Marco, S.; Samitier, J. IEEE Trans. Instrum. Meas. 1998, 47, 644. (242) Perera, A.; Papamichail, N.; Barsan, N.; Weimar, U.; Marco, S. IEEE Sens. J. 2006, 6, 770. (243) Ku, W.; Storer, R. H.; Georgakis, C. Chemom. Intell. Lab. Syst. 1995, 30, 179. (244) Russell, E. L.; Chiang, L. H.; Braatz, R. D. Chemom. Intell. Lab. Syst. 2000, 51, 81. (245) Wold, S.; Antti, H.; Lindgren, F.; Ohman, J. Chemom. Intell. Lab. Syst. 1998, 44, 175. (246) Artursson, T.; Eklo¨v, T.; Lundstro¨m, I.; Mårtensson, P.; Sjo¨stro¨m, M.; Holmberg, M. J. Chemom. 2000, 14, 711. (247) Gutierrez-Osuna, R. Proc. 7th International Symposium On Olfaction & Electronic Noses (ISOEN), Brighton, U.K., 2000; p 137. (248) Esbensen, K.; Lindqvist, L.; Lundholm, I.; Nisca, D.; Wold, S. Chemom. Intell. Lab. Syst. 1987, 2, 161. (249) Kermit, M.; Tomic, O. IEEE Sens. J. 2003, 3, 218. (250) Hyvarinen, A. IEEE Trans. Neural Networks 1999, 10, 626. (251) Weimar, U.; Go¨pel, W. Sens. Actuators, B 1998, 52, 143. (252) Mu¨ller, R.; Lange, E. Sens. Actuators 1986, 9, 39. (253) Vilanova, X.; Llobet, E.; Alcubilla, R.; Sueiras, J. E.; Correig, X. Sens. Actuators, B 1996, 31, 175. (254) Llobet, E.; Brezmes, J.; Vilanova, X.; Sueiras, J. E.; Correig, X. Sens. Actuators, B 1997, 41, 13. (255) Di Natale, C.; Marco, S.; Davide, F.; D’Amico, A. Sens. Actuators, B 1995, 25, 578. (256) Nakamoto, T.; Iguchi, A.; Moriizumi, T. Sens. Actuators, B 2000, 71, 155. (257) Nanto, H.; Tsubakino, S.; Ikeda, M.; Endo, F. Sens. Actuators, B 1995, 25, 794. (258) Saunders, B. W.; Thiel, D. V.; Mackay-Sim, A. Analyst 1995, 120, 1013. (259) Hongmei, W.; Lishi, W.; Wanli, X.; Baogui, Z.; Chengjun, L.; Jianxing, F. Anal. Chem. 1997, 69, 699. (260) White, J.; Kauer, J. S.; Dickinson, T. A.; Walt, D. R. Anal. Chem. 1996, 68, 2191. (261) Kermani, B. G.; Schiffman, S. S.; Nagle, H. T. IEEE Trans. Biomed. Eng. 1999, 46, 429. (262) Brahim-Belhouari, S.; Bermak, A.; Guangfen, W.; Chan, P. C. H. 2004 IEEE Region 10 Conference (TENCON 2004), 2004; Vol. 1, p 693. (263) Roussel, S.; Forsberg, G.; Steinmetz, V.; Grenier, P.; Bellon-Maurel, V. J. Food Eng. 1998, 37, 207. (264) Paulsson, N.; Larsson, E.; Winquist, F. Sens. Actuators, A 2000, 84, 187. (265) Martinelli, E.; Falconi, C.; D’Amico, A.; Di Natale, C. Sens. Actuators, B 2003, 95, 132. (266) Martinelli, E.; Pennazza, G.; Natale, C. D.; D’Amico, A. Sens. Actuators, B 2004, 101, 346. (267) Schalkoff, R. J. Pattern recognition: Statistical, structural, and neural approaches; J. Wiley: New York, 1992.

Hierlemann et al. (268) Vergara, A.; Llobet, E.; Martinelli, E.; Di Natale, C.; D’Amico, A.; Correig, X. Sens. Actuators, B 2007, 122, 219. (269) Pardo, M.; Sberveglieri, G. Sens. Actuators, B 2007, 123, 437. (270) Lanczos, C. Applied analysis; Prentice Hall: Englewood Cliffs, NJ, 1956. (271) Samitier, J.; Lopez-Villegas, J. M.; Marco, S.; Camara, L.; Pardo, A.; Ruiz, O.; Morante, J. R. Sens. Actuators, B 1994, 18, 308. (272) Claverie, P.; Denis, A.; Yeramian, E. Comput. Phys. Rep. 1989, 9, 247. (273) Sobol, W. T. Comput. Methods Programs Biomed. 1993, 39, 243. (274) Gardner, D. G.; Gardner, J. C.; Laush, G.; Meinke, W. W. J. Chem. Phys. 1959, 31, 978. (275) Marco, S.; Samitier, J.; Morante, J. R. Meas. Sci. Technol. 1995, 6, 135. (276) Yeramian, E.; Claverie, P. Nature 1987, 326, 169. (277) Llobet, E.; Vilanova, X.; Brezmes, J.; Sueiras, J. E.; Correig, X. Sens. Actuators, B 1998, 47, 104. (278) Galdikas, A.; Mironas, A.; Senuliene, D.; Strazdiene, V.; Setkus, A.; Zelenin, D. Sens. Actuators, B 2000, 69, 258. (279) Di Nucci, C.; Fort, A.; Rocchi, S.; Tondi, L.; Vignoli, V.; Di Francesco, F.; Santos, M. B. S. IEEE Trans. Instrum. Meas. 2003, 52, 1079. (280) Baumbach, M.; Kammerer, T.; Sossong, A.; Schu¨tze, A. Proc, IEEE Sensors 2004, 3, 1388. (281) Nakamura, M.; Sugimoto, I.; Kuwano, H.; Lemos, R. Sens. Actuators, B 1994, 20, 231. (282) Artursson, T.; Spangeus, P.; Holmberg, M. Anal. Chim. Acta 2002, 452, 255. (283) Gutierrez-Osuna, R.; Nagle, H. T.; Schiffman, S. S. Sens. Actuators, B 1999, 61, 170. (284) Siemiarczuk, A.; Wagner, B. D.; Ware, W. R. J. Phys. Chem. 1990, 94, 1661. (285) Ware, W. R.; Doemeny, L. J.; Nemzek, T. L. J. Phys. Chem. 1973, 77, 2038. (286) Gutierrez-Osuna, R.; Gutierrez-Galvez, A.; Powar, N. Sens. Actuators, B 2003, 93, 57. (287) Carmel, L.; Levy, S.; Lancet, D.; Harel, D. Sens. Actuators, B 2003, 93, 67. (288) Hillier, F. S.; Lieberman, G. J. Introduction to operations research, 8th ed.; McGraw-Hill: Boston, MA, 2005. (289) Carmel, L. Sens. Actuators, B 2005, 106, 95. (290) Haddad, R.; Carmel, L.; Harel, D. Sens. Actuators, B 2007, 120, 467. (291) Eklo¨v, T.; Martensson, P.; Lundstro¨m, I. Anal. Chim. Acta 1999, 381, 221. (292) Delpha, C.; Siadat, M.; Lumbreras, M. Sens. Actuators, B 2001, 77, 517. (293) Distante, C.; Leo, M.; Siciliano, P.; Persaud, K. C. Sens. Actuators, B 2002, 87, 274. (294) Cortes, C.; Haffner, P.; Mohri, M. J. Mach. Learn. Res. 2004, 5, 1035. (295) Shafiqul Islam, A. K. M.; Ismail, Z.; Ahmad, M. N.; Saad, B.; Othman, A. R.; Shakaff, A. Y. M.; Daud, A.; Ishak, Z. Sens. Actuators, B 2005, 109, 238. (296) Madou, M. J.; Morrison, S. R. Chemical sensing with solid state deVices; Academic Press: Boston, MA, 1989. (297) Le Vine, H. D. U.S. Patent 3,906,473, 1975. (298) Sears, W. M.; Colbow, K.; Consadori, F. Sens. Actuators 1989, 19, 333. (299) Sears, W. M.; Colbow, K.; Consadori, F. Semicond. Sci. Technol. 1989, 4, 351. (300) Gutierrez-Osuna, R.; Korah, S.; Perera, A. In Proceedings of the 8th International Symposium on Olfaction and Electronic Noses, Washington, DC, 2001; p 212. (301) Corcoran, P.; Lowery, P.; Anglesea, J. Sens. Actuators, B 1998, 48, 448. (302) Gramm, A.; Schu¨tze, A. Sens. Actuators, B 2003, 95, 58. (303) Schu¨tze, A.; Gramm, A.; Ruhl, T. IEEE Sens. J. 2004, 4, 857. (304) Ortega, A.; Marco, S.; Perera, A.; Sundic, T.; Pardo, A.; Samitier, J. Sens. Actuators, B 2001, 78, 32. (305) Cai, K.; Maekawa, T.; Takada, T. IEEE Sens. J. 2002, 2, 230. (306) Kato, Y.; Mukai, T. Sens. Actuators, B 2007, 120, 514. (307) Yea, B.; Osaki, T.; Sugahara, K.; Konishi, R. Sens. Actuators, B 1997, 41, 121. (308) Kato, K.; Kato, Y.; Takamatsu, K.; Udaka, T.; Nakahara, T.; Matsuura, Y.; Yoshikawa, K. Sens. Actuators, B 2000, 71, 192. (309) Nason, G. P. IEE Colloquium on Applied Statistical Pattern Recognition, Brimingham, U.K., 1999; p 1. (310) Strang, G.; Nguyen, T. WaVelets and filter banks; WellesleyCambridge Press: Wellesley, MA, 1996. (311) Nakata, S.; Yoshikawa, K.; Kawakami, H. Physica D 1992, 59, 169. (312) Nakata, S.; Nakamura, H.; Yoshikawa, K. Sens. Actuators, B 1992, 8, 187.

Higher-Order Chemical Sensing (313) Yoshikawa, K.; Yoshinaga, T.; Kawakami, H.; Nakata, S. Physica A 1992, 188, 243. (314) Nakata, S.; Yoshikawa, K.; Shoji, M.; Kawakami, H.; Ishii, T. Biophys. Chem. 1989, 34, 201. (315) Nakata, S.; Akakabe, S.; Nakasuji, M.; Yoshikawa, K. Anal. Chem. 1996, 68, 2067. (316) Nakata, S.; Ozaki, E.; Ojima, N. Anal. Chim. Acta 1998, 361, 93100. (317) Nakata, S.; Nakasuji, M.; Ojima, N.; Kitora, M. Appl. Surf. Sci. 1998, 135, 285. (318) Nakata, S.; Ojima, N. Sens. Actuators, B 1999, 56, 79. (319) Nakata, S.; Okunishi, H.; Nakashima, Y. Sens. Actuators, B 2006, 119 (2), 556. (320) Nakata, S.; Okunishi, H.; Inooka, S. Anal. Chim. Acta 2004, 517, 153. (321) Fort, A.; Machetti, N.; Rocchi, S.; Serrano Santos, M. B.; Tondi, L.; Ulivieri, N.; Vignoli, V.; Sberveglieri, G. IEEE Trans. Instrum. Meas. 2003, 52, 921. (322) Huang, X.; Meng, F.; Pi, Z.; Xu, W.; Liu, J. Sens. Actuators, B. 2004, 99, 444. (323) Meier, D. C.; Evju, J. K.; Boger, Z.; Raman, B.; Benkstein, K. D.; Martinez, C. J.; Montgomery, C. B.; Semancik, S. Sens. Actuators, B 2007, 121 (1), 282. (324) Dable, B. K.; Booksh, K. S.; Cavicchi, R.; Semancik, S. Sens. Actuators, B 2004, 101, 284. (325) Semancik, S.; Cavicchi, R. E.; Gaitan, M.; Suehle, J. U.S. Patent 5,345,213, 1994. (326) Cavicchi, R. E.; Suehle, J. S.; Chaparala, P.; Poirier, G. E.; Kreider, K. G.; Gaitan, M.; Semancik, S. Proc. 5th International Meeting on Chemical Sensors, Rome, Italy, 1994; p 1136 (327) Cavicchi, R. E.; Suehle, J. S.; Kreider, K. G.; Gaitan, M.; Chaparala, P. IEEE Electron DeVice Lett. 1995, 16, 286. (328) Semancik, S.; Cavicchi, R. E.; Gaitan, M.; Suehle, J. S. U.S. Patent 5345213, 1995. (329) Ionescu, R.; Llobet, E.; Al-Khalifa, S.; Gardner, J. W.; Vilanova, X.; Brezmes, J.; Correig, X. Sens. Actuators, B 2003, 95, 203. (330) Ionescu, R.; Llobet, E. Sens. Actuators, B 2002, 81, 289. (331) Llobet, E.; Vilanova, X.; Brezmes, J.; Lopez, D.; Correig, X. Sens. Actuators, B 2001, 77, 275. (332) Vergara, A.; Llobet, E.; Brezmes, J.; Ivanov, P.; Cane, C.; Gracia, I.; Vilanova, X.; Correig, X. I. Sens. Actuators, B 2007, 123 (2), 1002. (333) Vergara, A.; Llobet, E.; Brezmes, J.; Vilanova, X.; Ivanov, P.; Gracia, I.; Cane, C.; Correig, X. IEEE Sens. J. 2005, 5, 1369. (334) Llobet, E.; Brezmes, J.; Ionescu, R.; Vilanova, X.; Al-Khalifa, S.; Gardner, J. W.; Barsan, N.; Correig, X. Sens. Actuators, B 2002, 83, 238. (335) Llobet, E.; Ionescu, R.; Al-Khalifa, S.; Brezmes, J.; Vilanova, X.; Correig, X.; Barsan, N.; Gardner, J. W. IEEE Sens. J. 2001, 1, 207. (336) Ding, H.; Ge, H.; Liu, J. Sens. Actuators, B 2005, 107, 749755. (337) Bermak, A.; Belhouari, S.; Shi, M.; Martinez, D. In The Encyclopedia of Sensors; Grimes, C. A., Dickey, E. C., Pishko, M. V., Eds.; American Scientific Publishers: Valencia, CA, 2006; p 1. (338) Hines, E. L.; Boilot, P.; Gardner, J. W.; Gongora, M. A. In Handbook of Machine Olfaction: Electronic Nose Technology; Pearce, T. C., Schiffman, S. S., Nagle, H. T., Gardner, J. W., Eds.; 2002; pp 133160. (339) Bro, R. Chemom. Intell. Lab. Syst. 1997, 38, 149. (340) Bro, R. Anal. Chim. Acta 2003, 500, 185. (341) Harshman, R. A. UCLA Working Papers in Phonetics 1970, 16. (342) Carroll, J. D.; Chang, J. Psychometrika 1970, 35, 283. (343) Kroonenburg, P. M. Three-Mode Principal Component Analysis: Theory and Applications; DSWO Press: Leiden, The Netherlands, 1983. (344) Gurden, S. P.; Westerhuis, J. A.; Bro, R.; Smilde, A. K. Chemom. Intell. Lab. Syst. 2001, 59, 121. (345) Tucker, L. Psychometrika 1966, 31, 279. (346) Kiers, H. A. L. K. Psychometrika 1991, 56, 449. (347) de Juan, A.; Tauler, R. J. Chemom. 2001, 15, 749. (348) Harshman, R. A. UCLA Working Papers in Phonetics 1972, 22, 30. (349) Shaffer, R. E.; Rose Pehrsson, S. L.; McGill, A. NRL Formal Report 6110-98-9879; 1998. (350) Shaffer, R. E.; Rose-Pehrsson, S. L.; McGill, A. Anal. Chim. Acta 1999, 384, 305.

Chemical Reviews, 2008, Vol. 108, No. 2 613 (351) Shaffer, R. E.; Rose-Pehrsson, S. L.; Mcgill, R. A. Field Anal. Chem. Technol. 1998, 2, 179. (352) Bro, R.; Kiers, H. A. L. J. Chemom. 2003, 17, 274. (353) Elman, J. L. Cognit. Sci. 1990, 14, 179. (354) Roppel, T.; Dunman, K.; Padgett, M.; Wilson, D.; Lindblad, T. 23rd International Conference on Industrial Electronics, Control and Instrumentation IECON 97, 1997; Vol. 1, p 218. (355) Tan, G.; Wilson, D. Proc. IEEE Sensors 2003, 1, 139. (356) Rabiner, L. R. Proc. IEEE 1989, 77, 257. (357) Zhang, H.; Balaban, M. O.; Principe, J. C. Sens. Actuators, B 2003, 96, 385. (358) de Vries, B.; Principe, J. C. Neural Networks 1992, 5, 565. (359) Zaromb, S.; Stetter, J. R. Sens. Actuators 1984, 6, 225. (360) Niebling, G.; Mu¨ller, R. Sens. Actuators, B 1995, 25, 781. (361) Di Natale, C.; D’Amico, A.; Davide, F. A. M. Sens. Actuators, A 1993, 37-38, 612. (362) Wilson, D. M.; Gaffod, S. D. IEEE Sens. J. 2002, 2, 169. (363) John, G.; Kohavi, R.; Pfleger, K. Proceedings of the EleVenth International Conference on Machine Learning (ICML-94), 1994; p 121. (364) Hall, M. A.; Smith, L. A. Proceedings of the 1997 International Conference on Neural Information Processing and Intelligent Information Systems, 1997; p 855. (365) Pardo, A.; Marco, S.; Ortega, A.; Perera, A.; Sundic, T.; Samitier, J. Proceedings of International Symposium on Olfaction and Electronic Noses (ISOEN), Brighton, U.K., 2000; p 83. (366) Gardner, J. W.; Boilot, P.; Hines, E. L. Sens. Actuators, B 2005, 106, 114. (367) Pardo, M.; Sberveglieri, G. Sens. Actuators, B 2007, 123 (1), 437. (368) Corcoran, P.; Anglesea, J.; Elshaw, M. Sens. Actuators, A 1999, 76, 57. (369) Kermani, B. G.; Schiffman, S. S.; Nagle, H. T. IEEE Trans. Instrum. Meas. 1998, 47, 728. (370) Stella, R.; Barisci, J. N.; Serra, G.; Wallace, G. G.; De Rossi, D. Sens. Actuators, B 2000, 63, 1. (371) Xiao-bo, Z.; Jiewen, Z.; Shou-yi, W. Sens. Actuators, B 2002, 87, 437. (372) (a) Chaudry, A. N.; Hawkins, T. M.; Travers, P. J. Sens. Actuators, B 2000, 69, 236. (b) Osbourn, G. C.; Bartholomew, J. W.; Ricco, A. J.; Frye, G. C. Acc. Chem. Res. 1998, 31, 297. (373) Devijver, P. A.; Kittler, J. Pattern recognition: A statistical approach; Prentice/Hall International: Englewood Cliffs, NJ, 1982. (374) Doak, J. An eValuation of feature selection methods and their application to Computer Security; Tech Report CSE-92-18; University of California at Davis: Davis, CA, 1992. (375) Narendra, P. M.; Fukunaga, K. IEEE Trans. Comput. 1977, 26, 917. (376) Kirkpatrick, S.; Gelatt, C. D.; Vecchi, M. P. Science 1983, 220, 671. (377) Michalewicz, Z. Genetic algorithms + data structures ) eVolution programs, 3rd rev. and extended ed.; Springer-Verlag: Berlin, New York, 1996. (378) Llobet, E.; Brezmes, J.; Gualdron, O.; Vilanova, X.; Correig, X. Sens. Actuators, B 2004, 99, 267. (379) Boilot, P.; Hines, E. L.; Gongora, M. A.; Folland, R. S. Sens. Actuators, B 2003, 88, 80. (380) Marth, M.; Maier, D.; Stahl, U.; Rapp, M.; Wessa, T.; Honerkamp, J. Sens. Actuators, B 1999, 61, 191. (381) Buratti, S.; Ballabio, D.; Benedetti, S.; Cosio, M. S. Food Chem. 2007, 100, 211. (382) Roussel, S.; Bellon-Maurel, V.; Roger, J.-M.; Grenier, P. J. Food Eng. 2003, 60, 407. (383) Pavlou, A. K.; Magan, N.; Sharp, D.; Brown, J.; Barr, H.; Turner, A. P. F. Biosens. Bioelectron. 2000, 15, 333. (384) Srivastava, A. K.; Shukla, K. K.; Srivastava, S. K. Microelectron. J. 1998, 29, 921. (385) Gualdron, O.; Llobet, E.; Brezmes, J.; Vilanova, X.; Correig, X. Sens. Actuators, B 2006, 114, 522. (386) Llobet, E.; Gualdron, O.; Vinaixa, M.; El-Barbri, N.; Brezmes, J.; Vilanova, X.; Bouchikhi, B.; Gomez, R.; Carrasco, J. A.; Correig, X. Chemom. Intell. Lab. Syst. 2007, 85, 253. (387) Zhang, Q.; Benveniste, A. IEEE Trans. Neural Networks 1992, 3, 889. (388) Vergara, A.; Llobet, E.; Brezmes, J.; Ivanov, P.; Vilanova, X.; Gracia, I.; Cane, C.; Correig, X. Sens. Actuators, B 2005, 111-112, 271.

CR068116M

Chem. Rev. 2008, 108, 614−637

614

Very High Density Sensing Arrays Christopher N. LaFratta and David R. Walt* Department of Chemistry, Tufts University, 62 Talbot Avenue, Medford, Massachusetts 02155 Received April 3, 2007

Contents 1. Introduction 1.1. Terminology 2. Ensembles 2.1. Electrical Ensembles 2.1.1. Fabrication 2.1.2. Applications 2.2. Optical Ensembles 3. Very High Density Arrays 3.1. Directed Arrays 3.1.1. Photolithography 3.1.2. Dip-Pen Nanolithography 3.1.3. Chemical Synthesis by Photolithography 3.2. Randomly Ordered Arrays 3.2.1. Introduction 3.2.2. Analyte-Specific Sensing Arrays 3.2.3. Cross-Reactive Sensing Arrays 3.3. Suspension Arrays 3.3.1. Introduction 3.3.2. Protein Detection 3.3.3. Nucleic Acid Detection 4. Future Directions 4.1. Substrates and Materials 4.1.1. New Materials 4.1.2. Functional Materials 4.2. Novel Array Designs 4.2.1. Molecular Arrays 4.2.2. Liquid Arrays 4.3. New Tools and Devices 4.3.1. Optical 4.3.2. Surface Readout 4.4. Novel Applications of Very High Density Sensing Arrays 4.4.1. Next-Generation Sequencing 4.5. Issues 5. Acknowledgments 6. References

614 615 616 616 617 617 619 620 620 620 620 621 623 623 623 628 629 629 630 631 631 632 632 632 632 632 632 632 632 633 633 633 633 634 634

1. Introduction Historically, measurements have been made by taking a sample and analyzing it for a single analyte. When multiple analytes must be measured from a single sample, the sample conventionally is divided into appropriate aliquots and each * To whom correspondence should be addressed. Phone: (617) 627-3470. Fax: (617) 627-3443. E-mail: [email protected].

aliquot is analyzed for a single analyte. This approach is exemplified by many of the clinical analyzers employed in today’s modern hospital laboratories. Entire diagnostic panels are obtained from a single milliliter of blood by dividing the sample into dozens of channels that are analyzed for individual analytes such as Na+, K+, glucose, cholesterol, creatinine, uric acid, lactate dehydrogenase, and other clinically relevant analytes. As more sophisticated scientific instrumentation developed, the ability to detect multiple substances simultaneously became routine. Multiple metals can be measured using atomic emission and absorption spectrometers. Hyphenated methods such as GC-MS enable the separation of a sample into its components followed by identification of each component. While these instruments have tremendous capabilities, they tend to be large, power-hungry, and require routine maintenance. Consequently, most of these instruments are relegated to a central laboratory staffed by scientists and technicians. At the other extreme, simple colorimetric tests have been around for over a century. For example, litmus paper for determining acidity or basicity morphed into pH dipsticks in which different dyes were impregnated onto different pieces of filter paper, glued onto a plastic backing, and cut into strips containing multiple pH indicators. Similarly, dipsticks for measuring multiple analytes in swimming pools have been developed. These test strips were probably the first type of array, even though they were never recognized as such. In the last several decades, sensors have become a staple of analytical research and been used increasingly for making quantitative measurements. Initially, single sensors were used to measure a single analyte. Eventually multiple sensors were bundled either to provide a multianalyte measurement capability or to obtain spatial measurements of a single analyte. During the same time frame, scientists began to work at the microscale and have since progressed to the nanoscale. Opportunities in nanoscience and nanotechnology, as well as improved microscale capabilities, are driving feature sizes down. As scientists have become more comfortable working at these scales, the thinking has shifted with the developing capability to put increasing functionality into smaller and smaller spaces. This capability is exemplified by the computer chip industry. A direct outgrowth of small feature sizes has been the movement of the analytical community toward arrays. By developing methods to place different substances in different locations on a given substrate, it has become possible to produce arrays that contain many sensors or probes in a small area. These multifunctional devices enable multiple measure-

10.1021/cr0681142 CCC: $71.00 © 2008 American Chemical Society Published on Web 01/30/2008

Very High Density Sensing Arrays

Christopher LaFratta was born in Malden, MA, in 1979. He received his B.S. degree in Chemistry from the University of Massachusetts, Dartmouth in 2001. He then joined the laboratory of Professor John Fourkas at Boston College, where his research focused on various aspects of multiphoton fabrication. Having moved with Professor Fourkas to the University of Maryland, College Park, he received his Ph.D. degree in Chemistry from UMCP in 2006. He then joined the laboratory of Professor David Walt at Tufts University, where he was awarded a NIH/NIGMS postdoctoral fellowship. His current research involves the development of a microarraybased optoelectronic chemical sensing platform.

Chemical Reviews, 2008, Vol. 108, No. 2 615

This review deals with arrays containing the smallest feature sizes and the highest densities; hence, its title Very High Density Sensing Arrays. The motivation to create very high density sensing arrays is driven by several fronts. Increasing data are needed to solve increasingly complex problems. The field of Systems Biology has created a need to collect an enormous amount of data to understand the interactions and connections between biological pathways. The completions of the Human Genome Project and the HapMap Project have provided a rich database of human variation. Only by collecting millions of pieces of data from many thousands or millions of individuals will scientists be able to uncover the causes of disease and recommend changes in lifestyle to avoid them. Very high density sensing arrays enable the collection of large amounts of data. Fortunately, the tools exist for both collecting and processing large amounts of high-resolution data rapidly. For example, electronic components such as CCD chips, CMOS devices, and high-density integrated circuits provide the ability to collect enormous amounts of data on short time scales. Data storage capacity has increased to enable these data to be collected and stored. In addition, the ability to process data rapidly has kept pace. Without these corresponding improvements in data storage and processing capability, there would be no need to collect more data. As discussed in this review, very high density arrays are beginning to change the way we make measurements, process the data from these measurements, and use the information that can be extracted from these data. Very high density sensing arrays will have applications in many fields including diagnostics, the environment, industrial processing, fundamental science, and many others.

1.1. Terminology

David R. Walt is Robinson Professor of Chemistry at Tufts University and a Howard Hughes Medical Institute Professor. He received his B.S. degree in Chemistry from the University of Michigan and Ph.D. degreevin Chemical Biology from SUNY at Stony Brook. After postdoctoral studies at MIT, he joined the chemistry faculty at Tufts. He served as Chemistry Department Chairman from 1989 to 1996. He serves on many government advisory panels and boards and on editorial advisory board for numerous journals. From 1996−2003 he was Executive Editor of Applied Biochemistry and Biotechnology. He is the Scientific Founder and Director of Illumina Inc. and Quanterix Corp. He has received numerous national and international awards and honors for his fundamental and applied work in the field of optical sensors and arrays and is a fellow of the American Association for the Advancement of Science. He has published over 200 papers, holds over 50 patents, and has given hundreds of invited scientific presentations.

ments to be made simultaneously by simply bringing a sample into contact with the array. As the feature sizes of arrays have decreased, the term “microarray” has become commonplace. Microarrays, with probes spotted onto a solid support, have revolutionized molecular biology and ushered in the ‘-omics’ era. These arrays traditionally have densities of ∼50 spots/mm2, with spots on the order of 100 µm in diameter, although some smaller spots have been reported.1 A number of good reviews have been written about these ‘high density’ microarrays,2-4 but here we will focus on newer array technologies enabling significantly higher densities.

Martin and co-workers first used the term ensemble to describe the electrodes they fabricated by filling the pores of a membrane with metal. The term nanoelectrode ensemble (NEE) was used, rather than nanoelectrode array, because the pores of the template were not “evenly spaced”.5 Here we will use the term ensemble to mean a grouping of like sensor elements that respond collectively. The response of individual sensor elements in an ensemble, therefore, cannot be queried; instead, a collective signal is obtained from the entire ensemble. Unlike ensembles, the components of an array need not be identical and each can provide its own signal for detection. Arrays can be made by many different techniques, but in all cases the resulting array will fall into one of two categories: directed arrays or randomly ordered arrays. An experiment that uses a 96-well plate is an example of a directed array because the materials inside each well are known and purposefully placed. This situation is not the case for a randomly ordered array, where the elements selfassemble into a pattern that is not preordained. A random array can be made by filling a template with different elements, such as filling microwells with beads having different surface chemistries. Here, the location of each bead in the array is random. It is then necessary to ‘decode’ the position and identity of each element to use it as a sensing array. Another way to distinguish arrays from one another is based on the density of their elements. This article will focus on ‘very high density’ arrays, which qualitatively implies a density too high to manage without a computer. Quantita-

616 Chemical Reviews, 2008, Vol. 108, No. 2

LaFratta and Walt

Figure 1. Schematic of electrode ensembles of different size and density showing (A) radial diffusion, (B) overlapping radial diffusion, and (C) planar diffusion. (D and F) Cyclic voltammograms for the diffusion scenario in A and C, respectively. (E) Electrode ensemble with metal electrodes represented by yellow circles surrounded by gray insulating material. Microelectrode radius, Rb, and diffusion zone radius, R0, are shown.

tively, we define a very high density array to be one that contains more than 1000 elements per mm2. This specification dictates that each feature on the array is spaced approximately 30 µm apart or less. Data obtained from even a modestly sized (1 cm2) array would contain more than 100 000 array features. Such data sets, with hundreds of thousands to millions of data points, necessitate computer processing; drawing meaningful correlations among data of this size is beyond human capability. This definition of very high density arrays based on element size invites a discussion of another class of array, known as suspension arrays, into this review. The elements of a suspension are not fixed in position but float freely in solution, and the size of the elements is generally about 5 µm or less. As in randomly ordered arrays, the elements of a suspension array must be ‘decoded’ to identify their content. Thus, this review will not cover many microarrays used for gene expression analysis made by ink jet or contact printing since spot sizes are typically larger than 50 µm. Instead, we will focus primarily on ensembles with nanometer-sized elements as well as directed and randomly ordered arrays with sub-30 µm sized features.

2. Ensembles 2.1. Electrical Ensembles The definition of an ensemble is a group of identical elements that are addressed in unison. This definition arose from electrode ensembles in which a collection of electrodes shares a common electrical connection. Nanoelectrode ensembles (NEEs) were pioneered by Martin and co-workers over 20 years ago.5 Motivated by the unique diffusional properties of nanoscale electrodes, they invented a technique, now known as template synthesis, to quickly and reproducibly create millions of nanoscale electrodes by depositing metals inside nanoporous membranes. NEEs created in templates can have densities upward of 1011 electrodes/cm2, easily putting them in the very high density category.6

Before NEEs were created, there was considerable interest within the electrochemical community in experimentally validating the diffusion properties of ions as the size of electrodes decrease and approach the dimensions of the electric double layer.7,8 Models predicted substantially higher mass transfer rates in microelectrodes due to radial diffusion, which would enable ultrafast electrochemical measurements, compared to measurements using bulk electrodes that operate via planar diffusion.9,10 Smaller electrode sizes also promised access to microenvironments not accessible to larger electrodes, such as cells. The ability to make measurements in this realm drove researchers to taper microelectrodes to smaller and smaller sizes. A persistent issue that arose, however, was the decreasing current as electrodes shrank. NEEs successfully addressed both the issue of decreasing electrode size, by using nanoporous templates, and the issue of increasing the overall current, by multiplexing the nanoelectrodes.11 The diffusional properties of NEEs are influenced by the nanoelectrode density and voltage scan rate of the experiment.12 If the electrodes are tightly packed and the scan rate is relatively slow, then their individual radial diffusion profiles overlap, resulting in planar diffusion (Figure 1). In this ‘total overlap’ regime, the NEE will respond similar to bulk electrodes during cyclic voltammetry (CV) measurements, except with a lower background signal. Microelectrodes spaced far apart and scanned quickly maintain radial diffusion and show a characteristic sigmoidal CV trace. The ideal case is to have the electrodes at an intermediate density to maintain radial diffusion while using the electrode area as efficiently as possible.13 Many researchers have used the condition (C1)

R0 > 6Rb, (C1) where R0 is one-half the interelectrode distance and Rb is the electrode radius (Figure 1b), as a guide to designing arrays for the ideal case. This equation, which can be traced

Very High Density Sensing Arrays

Chemical Reviews, 2008, Vol. 108, No. 2 617

Table 1. Comparison of Microelectrode Radius with Its Diffusion Zone Radius Comparison of Microelectrode Radius, Rb, with the Simulated Critical Domain Radius, R0c, and the Diffusion Radius Obtained from C1a Rb (µm) 10 1 0.1 0.01

R0c (µm) 50 34 15 5

R0C1 (µm) 60 6 0.6 0.06

a The formal definition of R0c is defined in ref 9, but for our purposes here it can be considered as one-half the minimum distance between electrodes that avoids significant overlap of their diffusion profiles. Adapted with permission from ref 13. Copyright 2005 Elsevier B.V.

back to work by Saito in 1968,14 is accurate for moderately sized electrodes, near 10 µm in radius, but more recent models predict it underestimates the size of the diffusion zone from smaller electrodes.13 The recent simulations by Compton and co-workers could not be simplified by a single equation, but results are given in Table 1 for electrodes whose sizes vary over 4 orders of magnitude. Using varying densities and sizes, electrode ensembles operating in both ‘total overlap’ and ‘radial’ diffusion modes have been successfully used in chemical sensing.15,16

2.1.1. Fabrication Several methods now exist to create NEEs, including methods based on self-assembly,17,18 but the most commonly used techniques are based on the template methods of Martin and co-workers.5,19,20 In these methods, the template is either an anodized alumina membrane or a polycarbonate tracketch membrane. The anodized alumina is created electrochemically in a two-step process starting from Al foil. By varying time, anodization potential, and the electrolyte solution composition, pores of different length and diameter can be made.21-23 Unlike track-etch membranes, alumina membranes can be created using equipment readily available in an electrochemistry lab. Track-etch membranes are created by exposing thin polymer films to nuclear fission fragments in the chamber of a nuclear reactor. These subatomic particles tear through the film, leaving straight nanometer-size tracks that can be chemically amplified by etching to create monodispersed pores with controlled diameters.24 Both types of membrane are commercially available with pore sizes between 10 nm and 10 µm and densities over the range from 104 to 1011 pores/cm2. Templates are typically filled either by electrodeposition or electroless deposition. During electrodeposition, one side of the membrane is sputter coated with a metal that serves as an electrode and then used to electrodeposit additional metal from an electroplating solution in contact with the open pores on the opposite side of the membrane.25 This versatile method can be used to deposit both metals and conducting polymers.26 Electroless deposition is performed by sensitizing the pores with Sn2+ and then using the physisorbed Sn2+ to electrolessly reduce ammoniacal Ag(NO3).15 The silver nanoparticle seeds then catalyze electroless deposition of other metals, such as gold, which grows inward from the pore walls. Thus, metal tubules are electrolessly grown until they become solid cylinders. Along with these two methods, other filling mechanisms have also been demonstrated such as chemical polymerization,27 sol-gel deposition,28 and chemical vapor deposition (CVD).29 Alumina and track-etch membranes have also been used for a number of alternative applications besides NEEs. For

example, they have been used as stencil masks to etch an array of wells30-32 or deposit materials in a geometry dictated by the pore locations.33 The filling material is not exclusive to metals and polymers; carbon,34,35 semiconductors,36 and Li+-intercalation materials37-39 have also been prepared. Carbon nanotubes (CNTs) deposited in alumina membranes have been used as IR detectors,40 and electrodeposited Ge nanowires have been used as photoresistors.41 The pores can also be used to synthesize nanomaterials, such as tubules and wires, that can be released by dissolving the membrane.42,43 While these alternative technologies have not yet been used in chemical sensing ensembles, their common starting point of a very high density template makes them noteworthy.

2.1.2. Applications The first chemical sensing application of NEEs took advantage of their increased diffusion rates, demonstrating a lower detection limit for several standard electrochemical species such as Ru(NH3)63+/2+, Mo(CN)84-/3-, and TMAFc1+/2+ ([(trimethylamino)methyl]ferrocene).15,44 The detection limit in a voltammetric experiment is dependent on the ratio of analytical signal to background (S/B). Signal is caused by Faradaic current that occurs at the electrode during the redox reaction of the analyte.45 In a NEE, at appropriately high scan rates, the signal can be as high as macroelectrodes due to the radial diffusion zone around each nanoelectrode and the large number of electrodes in the ensemble. Background is predominantly caused by a double-layer charging current at the electrode-solution interface and proportional to the area of the conductive portion of the electrode. For most electrodes, the conductive area is equal to the total area, but for NEEs the conductive area is only about 0.1% of the entire electrode surface. Therefore, since the signal is the same but the background is several orders of magnitude lower, the S/B for a NEE is significantly higher than for conventional electrodes. Other groups have used this increase in S/B to detect more interesting electrochemically active species using NEEs. Ugo and co-workers, for example, utilized NEEs to detect submicromolar concentrations of iodide and cytochrome c (cyt c).46 The iodide concentration in table salt was measured using CV with a detection limit more than 10 times lower compared to a bulk electrode. Cyt c was of particular interest because electrochemical studies of cyt c usually require a promoter or mediator to avoid electrode poisoning due to adsorption. The high sensitivity of the NEE and differential pulsed voltammetry method used enabled the concentration of the cyt c to be reduced low enough to avoid poisoning, thereby allowing detection without a promoter. Recently, several groups have used very high density ensembles to electrically detect DNA hybridization events. Kelley and co-workers used NEEs to electrically detect DNA hybridization. The NEEs were made by electroless deposition of gold into polycarbonate track-etch membranes. The polycarbonate membranes of these standard 2-D NEEs were then O2 plasma etched to yield 3-D brush-like electrodes.47 This change in geometry allowed more DNA to bind to each electrode, decreasing the NEEs detection limit into the attomole range.48 DNA hybridization was measured by CV using Ru(NH3)63+ and Fe(CN)6-3. Ru(III) was reduced to Ru(II) by electron transfer through dsDNA, which was formed by the hybridization of target ssDNA to selfassembled ssDNA probes on the electrode surface (Figure

618 Chemical Reviews, 2008, Vol. 108, No. 2

LaFratta and Walt

Figure 2. (A) Schematic of Ru(III)/Fe(III) electrocatalysis at a DNA-modified Au NEE. (B) Scanning electron micrograph of a NEE. The white spots are the tips of the Au nanowires protruding the polycarbonate membrane. The individual wires extend ∼200 ( 10 nm from the membrane surface. Representative cyclic voltammograms for an 18-mer duplex DNA-modified bulk (C) Au electrode and (D) NEE. Solutions contain 40 µM Ru(NH3)63+ and 0 (blue) and 32 µM (red) Fe(CN)6-3. Scan rate for all CV experiments was 100 mV/s. Background subtraction was performed so that all scans could be directly compared. Reprinted with permission from ref 48. Copyright 2005 American Chemical Society.

2). Ru(III) was catalytically regenerated from Ru(II) by conversion of excess of Fe(II) to Fe(III) in solution, which amplifies the response by a factor greater than 10.49 Another method of DNA detection has been demonstrated by Andreu et al. using a gold nanowire ensemble produced using an anodized alumina membrane.50 After electrodepositing Au wires in the membrane, the alumina template was dissolved leaving a high surface area Au electrode. Probes of ssDNA were self-assembled onto the electrode, and the surface charge was measured by chronocoulometry using Ru(NH3)63+ as a redox marker. The measurement was repeated after hybridization to target DNA, which caused an increase in the surface charge. The detection limit has not yet been determined for this method, but the authors report a 25 bp sequence was detected at micromolar concentrations.50 Meyyappan and co-workers also demonstrated DNA detection with NEEs; however, they used a different fabrication method to create a NEE composed of multiwalled carbon nanotubes (MWCNTs). The MWCNTs were grown on a lithographed Si wafer from patterned Ni catalyst spots using plasma-enhanced chemical vapor deposition (PECVD).51 The

nanotubes were nominally 5 µm in length and 80 nm in diameter with a density of 2 × 109 nanotubes/cm2. The wafer and MWCNTs were then coated by CVD with an insulating layer of SiO2, which was later polished to expose the nanotube tips. These tips were electrochemically etched, leaving hydroxyl and carboxylic acids groups that could be used to covalently bind analytes such as DNA. DNA hybridization was detected using the catalytic redox species tris(2,2′-bipyridyl)ruthenium(II), Ru(bpy)32+, to oxidize guanine residues in dsDNA. Using AC voltammetry and an 18 bp probe sequence, 300 bp PCR amplicon targets were detected with sensitivity approaching that of laser-based fluorescence techniques.52 Very high density ensembles have also been made without photolithography or template methods. Walt and co-workers demonstrated a method of creating electrode ensembles starting from a fiber optic bundle with a density higher than 106 fibers/cm2.53 The cladding of a fiber optic bundle was etched anisotropically, resulting in pointed fiber optic cores that were then sputter coated with gold. This bulk gold surface was then covered with an insulating paint, leaving

Very High Density Sensing Arrays

Chemical Reviews, 2008, Vol. 108, No. 2 619

Figure 3. Illustration of patterning by nanosphere lithography (NSL). A colloidal monolayer is created and used as a mask for metal deposition in the voids between the nanospheres. Deposition of ∼50 nm of metal followed by mask removal creates a triangular nanoparticle ensemble useful for LSPR. Deposition of ∼200 nm of metal results in a textured metallic film used in SERS studies. Reprinted with permission from ref 61. Copyright 2005 Elsevier B.V.

∼1 µm diameter Au tips periodically protruding the insulation. The fiber optic bundles were used as electrode ensembles to optically detect the electrogenerated chemiluminescence signal from Ru(bpy)32+ and tripropylamine in phosphate buffer.53 In a method analogous to the insulated fiber optic tips, Lowe et al. showed that an ensemble of conical CNTs grown on a Pt electrode could be partially insulated to create a NEE as demonstrated by sigmoidal CV traces showing steady-state diffusion, a characteristic of NEEs.54 Very high density ensembles, characterized by their bulk addressability, have proven useful in electroanalytical chemistry, especially for the measurement of DNA hybridization. Inspired by the diffusion properties of ever smaller microelectrodes, the template methods developed by Martin to create NEEs have led to numerous devices with improved S/B ratios and lower limits of detection.

2.2. Optical Ensembles Metal nanoparticle ensembles have also been used for optical sensing. Metal nanoparticles and thin films have distinct optical properties due to surface plasmon resonance (SPR), which is a collective oscillation of surface electrons excited by electromagnetic radiation. Radiation impinging on a thin metal surface can excite the surface electrons into a propagating wave known as a surface plasmon polariton. A similar oscillation occurs in nanoparticles, and this phenomenon is called localized surface plasmon resonance (LSPR) because it is confined to the surface of isolated particles.55 In both cases, the resonant frequency depends on the dielectric constant of the surrounding material and in the case of the LSPR it also depends on the size and shape of the nanoparticles. Islands of metal nanoparticles that have been functionalized with capture probes have been used as sensors.56 Wavelength-shift LSPR sensors can detect analyte binding because the act of binding changes the dielectric constant near the surface of the film, and therefore, the resonant wavelength also changes. Van Duyne and coworkers pioneered nanosphere lithography (NSL)57 to make nanoparticle arrays for LSPR sensors, but other methods also exist.58,59 NSL uses hexagonally packed polymer nanospheres

on a glass surface to create a stencil mask. Depositing metal onto this monolayer results in small, 20-1000 nm, metallic triangles in the voids between the spheres (Figure 3).60 An LSPR sensing substrate results after removing the spheres, leaving a homogeneous ensemble of uniform metal triangles, the size and shape of which can be tuned. Deposition of a thicker film onto the spheres results in a highly textured metal surface, which is useful for surface-enhanced Raman scattering (SERS).61 SERS surfaces can enhance the Raman effect by ∼106. This enhancement is caused in part by the local electromagnetic field resulting from LSPR, which then induces a dipole in molecules that are in proximity (<4 nm) to the metal surface, thereby raising the effective Raman cross-section of the molecule.61 SERS surfaces have also been used as sensors by measuring the Raman spectrum of analytes bound to patterned substrates.62 Both LSPR and SERS substrates have been used for sensing in either transmissive or reflective modes, and in both cases the signal originates from the entire ensemble. Wavelength-shift LSPR sensing has been used to detect several biomolecules including proteins, DNA, and the biomarker for Alzheimer’s disease, amyloid-beta-derived diffusible ligand (ADDL). The carbohydrate-binding protein concanavalin A (con A) was one of the earliest analytes detected using LSPR by exposure to a mannose-coated Ag pattern made by NSL.63 Binding of 0.19 µM con A could be monitored in real time by tracking the resonant wavelength. A slightly modified platform was used for DNA detection, consisting of Au-coated silica nanoparticles on top of a Au-coated glass slide. Peptide nucleic acid (PNA) bound probes were able to detect as little as 0.677 pM complementary DNA.64 This platform was also used for immunoassays by attaching protein A to the Au-coated silica beads and then using protein A to bind C-reactive protein, fibrinogen, and immunoglobulins IgA, IgD, IgG, and IgM.65 These six proteins were spotted to create a low density array that was able to detect 100 pg/mL of antigen by wavelengthshift LSPR. The biomarker ADDL was measured in the cerebrospinal fluid of Alzheimer’s patients in a LSPR sandwich assay.66 After ADDL bound to the nanoparticle

620 Chemical Reviews, 2008, Vol. 108, No. 2

LaFratta and Walt

3. Very High Density Arrays

sors. These detectors can reach several megapixel resolution on chips smaller than 1 cm2 and be manufactured at modest cost. A digital micromirror device (DMD) is a dense array of actuatable mirrors about 10 µm on a side. DMDs are finding application in photolithography as dynamic replacements for photomasks in addition to their use in displays. Spatial light modulators (SLMs), which use an array of liquid crystals pixels, can also modulate light in two dimensions. While these four microelectronic devices have not yet been used directly as transducers for chemical sensing, they are widely used for spatial control of chemical processes (DMD and SLM) and capturing spectroscopic data and images (CCD and CMOS). Photolithography can also be used to fabricate arrays of chemically sensitive field effect transistors (CHEMFETs).116 Over their 35 year history, CHEMFETs have been used to detect several analytes, including K+,117 Na+,118 urea,119 enzymes,120,121 and DNA.122,123 CHEMFETs can, in principle, be arrayed at high density, enabling multiplexed detection; however, there are very few examples of densely arrayed CHEMFETs in the literature.124,125 Recent advances in nanoparticle synthesis have enabled the creation of semiconducting nanowires, such as silicon nanowires (SiNWs) and carbon nanotubes (CNTs), that can be used as the channel material in place of bulk semiconducting crystals. There are many reports of SiNW- and CNT-FET sensors, including several good reviews.125,127,128,267 There are also some papers that are critical of the feasibility and selectivity of these devices.134,268-273 The most common analytes detected up to now include small cations, such as H+ and Ca2+,129 andbiomoleculessuchasDNAandantigens.135,136,138-144 These nanowire channels typically have dimensions of tens of nanometers in diameter and several microns in length. This <1 µm2 footprint makes these devices capable of very high density arrays; however, nanowire FETs have thus far been arrayed with densities of only ∼50 FETs/mm2.143 For this reason, a detailed description of these arrays is beyond our scope.

3.1. Directed Arrays

3.1.2. Dip-Pen Nanolithography

3.1.1. Photolithography

Since the 1980s, there has been a push to extend photolithography beyond the scope of electronics and use its microscale patterning capability to create sensing devices.75 Some of the first microscopic devices made were cantilevers for use in scanning probe microscopies, such as scanning tunneling microscopy (STM) and atomic force microscopy (AFM). Cantilevers, usually comprised of silicon, can be made in a variety of shapes and sizes but are typically ∼100 µm long, 10 µm wide, and 2 µm thick. Cantilevers are also used in dip-pen nanolithography (DPN),76,77 which is an adaptation of AFM. DPN uses cantilevers with atomically sharp tips to pattern ‘inks’ onto substrates. Following the first demonstration of using DPN to pattern alkanethiol inks on gold substrates,78 various other materials, including polymers,79 proteins,80 peptides,81 and sols,82 have been patterned on metallic, semiconducting, and insulating substrates. Patterned feature sizes are approximately 100 nm wide and have been made as small as 15 nm.83 DPN has enabled the creation of very dense sensing arrays. Protein arrays of rabbit immunoglobulin gamma (IgG) and lysozyme have been patterned directly by DPN and shown to retain their biomolecule recognition capabilities.84 To avoid denaturing the proteins, the cantilever tips have been modified with a hydrophilic monolayer, such as 2-[meth-

ensemble, a resonant wavelength shift of 28.5 nm was observed, with an additional shift of 15.4 nm being measured after binding of the secondary antibody. Compared to the control sample, with shifts of 2.9 and 4.3 nm respectively, the biomarker was readily identified. SERS ensembles have been used to detect glucose, viruses, and warfare agent simulants. Using a mixed monolayer of hydrophilic and hydrophobic groups on a SERS substrate, glucose has been detected both in vivo and in vitro. This real-time sensing measured glucose using its spectral signature rather than indirectly, as in most electrochemical sensors.67 Antibody-captured feline calicivirus particles on a gold substrate were detected in a sandwich assay by attaching Au nanoparticles coated with a Raman-active reporter molecule.68 Virus concentrations as low as 106/mL could be detected. The anthrax biomarker calcium dipicolinate (CaDPA) has been detected in the spores of B. subtilis, which is B. anthracis simulant.69,70 An infectious dose is 104 spores and as few as 1400 spores could be detected using the SERS platform. Likewise, a mustard gas simulant was also detected well below the harmful limit. This measurement was performed on a portable Raman spectrometer, enabling measurements in the field. Tan and Vo-Dinh also report a field deployable SERS spectrometer, which has been used to detect simulants for warfare agents such as sarin, soman, tabun, and sulfur mustard (HD).71 Ensemble sensors probed optically have also been made from hydrogels. A colloidal crystal, made of 100-nm diameter polymer spheres, has been coated in a hydrogel that is responsive to heavy metal ions and glucose.72 Upon exposure to analyte, the hydrogel swells changing the lattice spacing of the crystal, which is measured by the change in the diffraction pattern of the crystal. Similarly, the swelling of microlens made of hydrogels can be monitored by measuring their change in focal length.73 These microlenses were demonstrated by detecting antibody-antigen binding.74

Arrays that contain elements whose identity is purposefully mapped to specific positions are known as directed arrays. In order to create very high density directed arrays, photolithography is used, sometimes in conjunction with electronbeam or soft lithography. Photolithography is the workhorse of the microelectronics industry and used to pattern intricate circuits on the micrometer and submicrometer scales with feature sizes below 100 nm. Using UV radiation and photomasks, photolithography patterns light-sensitive resists atop semiconducting or insulating substrates. The developed photoresist protects the underlying chip in specific regions for subsequent processing steps, such as etching, coating, or doping. Complex circuits, such as microprocessors and memory, are created by repeated cycles of protecting and processing. These techniques of photoprotecting/deprotecting have also been used for rapid parallel chemical synthesis, as will be described below. Electronic devices that consist of an array of components, such as the array of capacitors that make up dynamic random access memory (DRAM), can have densities >100 000 per mm2. Many other microelectronic arrays also have very high densities, including charge-coupled devices (CCDs) and complementary metal-oxide-semiconductor (CMOS) sen-

Very High Density Sensing Arrays

Figure 4. Schematic representation of the sandwich immunoassay format used to detect HIV-1 p24 antigen with an anti-p24 antibody nanoarray made by DPN. The HIV-1 p24 antigen was sandwiched between anti-p24 antibody bound to the MHA patterned surface and gold nanoparticle probes coated with anti-p24 antibody. The change in height due to the nanoparticle binding event could be detected by AFM. Reprinted with permission from ref 86. Copyright 2004 American Chemical Society.

oxypoly(ethyleneoxy)propyl]trimethoxysilane, and maintained at 60-90% relative humidity during patterning. IgG has also been arrayed indirectly by patterning 16mercaptohexadecanoic acid (MHA) on a gold substrate. The patterned MHA then selectively binds IgG.85 This strategy was used in a sandwich immunoassay for detection of human immunodeficiency virus type 1 (HIV-1) as shown in Figure 4.86 Nanoparticle binding has also been used for oligonucleotide detection by patterning the ssDNA using DPN and exposing this array to gold nanoparticles labeled with the complementary DNA strand.87 Thus far, arrays made by DPN have only been fabricated for the detection of two analytes,88 although the technique could potentially be applied to massively parallel multiplexed detection. A major step toward this goal has already been taken by Mirkin and co-workers with their creation of a cantilever array containing 55 000 cantilevers in a 1 cm2 array (Figure 5).76,89,90 This array has been used to pattern 88 000 000 gold dots, each 100 nm in diameter, and will undoubtedly be used for highly multiplexed sensing in the future.

Chemical Reviews, 2008, Vol. 108, No. 2 621

Figure 5. Optical micrograph of a small section of a DPN array containing 55 000 cantilevers. Scale bar ) 100 µm. (Inset) Electron micrograph of the cantilever tips. Reprinted with permission from Nature (http://nature.com), ref 76. Copyright 2004 Nature Publishing Group.

3.1.3. Chemical Synthesis by Photolithography

Figure 6. (A) Light-directed oligonucleotide synthesis. A substrate coated with a covalently bound linker molecule containing a photolabile protecting group (orange squares) is locally exposed to light through a photomask. The exposed regions are deprotected and then reacted with protected nucleotides. The process is repeated, deprotecting and reacting different sites with different nucleotides, to synthesize arbitrary DNA probes at each site. (B) Schematic illustration of a photomask used to expose an array. Reprinted with permission from Nature (http://nature.com), ref 99. Copyright 1999 Nature Publishing Group.

In 1991, Fodor and co-workers pioneered the application of photolithography to combinatorial chemical synthesis to create very high density biomolecule arrays.91 The starting point for an oligonucleotide array is a quartz substrate modified with photochemically removable protecting groups.92,93 Areas of the substrate are then activated by exposure to UV radiation through a photomask. Next, the substrate is incubated with hydroxyl-protected deoxynucleosides, which results in addition of the first base to the activated areas. A different mask is then used to expose and deprotect other regions of the substrate, enabling those areas to react with the next protected deoxynucleoside. The process of deprotection and reaction is repeated resulting in the synthesis of different oligonucleotides in different locations on the array (Figure 6). All 4n combinations of an n-mer oligonucleotide can be synthesized in 4 × n steps. Oligonucleotides synthesized photolithographically are generally less than 30

bases long with densities >250 000 features/cm2.4 While feature size is usually about 20 µm on edge, features as small as 8 µm have been demonstrated.94 One of the drawbacks of this fabrication technique is the need for possibly 100 photomasks to create the desired array of sequences.95 Patterning the light using a DMD, in place of photomasks, is one method to alleviate this problem.96-98 Feature sizes as low as 4 µm have been reported using DMDs to pattern oligonucleotides.96 Oligonucleotide arrays have been used for gene expression and genotyping.99 Genes are expressed in cells by first translating genomic DNA into messenger RNA (mRNA) followed by transcription of mRNA into functional proteins. To test for expressed genes in a sample, typically mRNA is reverse transcribed into complementary DNA (cDNA) containing a fluorescent label. The cDNA is more stable than RNA and can also be amplified with polymerase chain

622 Chemical Reviews, 2008, Vol. 108, No. 2

LaFratta and Walt

Figure 7. (A) Gene expression monitoring using an array containing 40 000 human genes and expressed sequence tags. The optical micrograph shows a substrate, 1.28 × 1.28 cm, containing features less than 22 × 22 µm. (B) The oligonucleotide probes are chosen based on composition design rules and a uniqueness criterion. Use of perfect match (PM) and mismatched (MM) probes greatly reduces background and crosshybridization signals, increasing accuracy and reproducibility. Reprinted with permission from Nature (http://nature.com), ref 99. Copyright 1999 Nature Publishing Group.

reaction (PCR). Alternatively, the mRNA can be fluorescently labeled and then randomly fragmented into 50-100 bp segments and hybridized to the array. The array contains thousands of 25-mer oligonucleotides sequences, called probes, known to be complementary to genes.100,101 Each gene from a sample, which spans hundreds of bases, is covered on the array using multiple 25-mer sequences. This strategy offers a type of redundancy because while the probe sequences are not the same, several probes encode for the same gene. Expressed genes are identified based on the intensity and location of the fluorescent signal. To quantify nonspecific hybridization and background signals, the perfect match (PM) probes on the array are placed next to mismatch (MM) probes, which are identical to the PM except for one nucleotide in the center of the sequence, which is different (Figure 7).101 Thus, cross-hybridized signals can be subtracted from the PM signal. This lithographed platform, developed by Affymetrix Inc. under the name GeneChip, has been used for genome-wide expression analysis for over 10 years.102 The density of this array enables comprehensive analysis of cell functions by monitoring thousands of genes in a sample simultaneously.102 Gene expression studies are widely used to identify and study diseases, such as cancer, as well as study basic biological functions.103-106 Genotyping can also be performed using lithographed arrays. Genotyping refers to identification of genetic differences, such as single-nucleotide polymorphisms (SNPs), which account for phenotypic differences between and within species. A SNP is an alteration in a single nucleotide in genomic sequence that occurs in at least 1% of a population. For example, in the sequence AATTGAT, a SNP of the sequence would be AATCGAT. Theoretically, there are about 11 million known SNPs in the human genome.107 Genetic variation can affect an individual’s response to a disease and environmental factors, such as toxins or drugs. Genotyping will help identify genetic disease markers and accelerate new therapies. SNPs can be identified using two probe sequences that vary in only one position. The probe that forms the most stable duplex will result in the highest fluorescent signal and identify which allele is present in the sample. On the GeneChip platform, a single SNP is queried with 40 probes. A quartet of four probes represents the PM and MM for both alleles with the SNP position in the center. Two more pairs of quartets, with the SNP position shifted

Figure 8. Section of a genotyping array that shows the fluorescence intensity pattern for a set of probes that interrogates a single locus. The upper half of the probe blocks interrogate the A alleles and the lower half interrogate the B alleles. Each half has pairs of probes centered on polymorphic position and offset one and four bases to either side. The pairs consist of a PM and a MM to the reference sequence for the specific allele. The presence of the AA homozygote, the AB heterozygote, and the BB homozygote is shown. Reprinted with permission from Nature (http://nature.com), ref 99. Copyright 1999 Nature Publishing Group.

(1, (4 from the center, make up 20 probes. The remaining 20 probes are the anti-sense version of the first 20 probes (Figure 8).94,108 Presently Affymetrix arrays can simultaneously detect over 900 000 SNPs. Similar arrays have been used for a broad range of purposes from cancer research to drug development.109-112 It is not the purpose of this review to cover the many sophisticated applications of gene chips or DNA microarrays. For a more comprehensive overview of such applications, a number of excellent reviews are available.113-115 It is important to note here that while the original purpose of DNA microarrays was to use the specificity of hybridization to determine the sequences present in a genetic sample, this approach is no longer the preferred one. Modern genotyping experiments now implement a two-phase approach. In the first phase, a series of complex biochemistry and molecular biology steps is employed to interrogate many different genetic sequences simultaneously and prepare the resulting sample for hybridization to the array. The hybridization step that follows is simply used as a readout for the assay. In several implementations of this two-phase approach, the goal is to interrogate many SNPs in parallel and the strategy is to convert the small single-base differences into molecular signals that allow easy discrimination. To this end, the arrays

Very High Density Sensing Arrays

Chemical Reviews, 2008, Vol. 108, No. 2 623

rarely need single-base specificity because the assay converts a single-base mismatch into a multiple-base readout difference. More information about this strategy can be found in the section below on Randomly Ordered Arrays.

3.2. Randomly Ordered Arrays 3.2.1. Introduction A randomly ordered array is classified as such because its elements are randomly self-assembled into a pattern. The identity of a probe at any location in the array is therefore not known a priori. This is fundamentally different than directed arrays, where array elements are synthesized or placed in known locations. A template is used to structure self-assembled arrays. Sensor elements, such as microspheres with different surface chemistries, randomly fill the template to create a multiplexed sensing platform. After identifying the surface chemistry of each element, in a process called registration, analytes are detected based on which array elements they interact with. Our lab has pioneered the use of fiber optic bundles as templates for randomly ordered array sensors.146-150 A fiber optic bundle is a collection of individual fiber optic cores that share a common cladding. Each core can act as a waveguide to transmit an optical signal without cross-talk between adjacent fibers. Typically, a fiber bundle contains between 5000 and 50 000 waveguides with individual fibers in the array ranging in diameters between 3 and 7 µm. These bundles are coherent, such that the position of a particular fiber at one end of the bundle corresponds to its position at the other end. The different glass compositions of the core and clad materials cause them to etch at different rates. When treated with an acidic etching solution, the core etches faster than the clad and creates an array of uniform microwells. These femtoliter-sized microwell chambers can then be loaded with a variety of microsensors or probes, living cells, or they can be used to house reactive species, such as enzymes. For bead-based sensing applications, indicator or probe molecules are covalently attached to polymer or porous silica beads that can then be loaded into the wells. Each bead type is prepared in a separate reaction scheme, and the different bead types are then pooled before loading into the array. The wells are sized to ensure that there is only one bead per well. For most applications, the transduction mechanism is based on fluorescence as it allows for simple optical instrumentation. Detection is performed using a microscope objective to launch excitation light into the proximal end of the fiber and detecting the epi-fluorescence from the beads housed on the distal face of fiber (Figure 9). Parallel detection involving thousands of beads is accomplished using a CCD camera. The platform can be spectrally multiplexed using different combinations of excitation and emission wavelengths (Figure 10). The fiber optic, randomly ordered, addressable array format has several advantages over traditional patterned microarrays, where elements are preregistered by position. The primary benefit is the ease with which randomly ordered arrays can be created. Directed arrays made by ink jet printing, screen printing, or photolithography typically require several fabrication steps where the probability of fabrication errors increase in proportion to the number of processing steps involved. Bead-based random arrays are quickly produced via self-assembly from a few microliters of bead stock solution, which contains ∼109 beads/mL (a dry bead

Figure 9. Schematic illustration of a typical epi-fluorescence microscope setup for imaging fiber optic bundle arrays. Reprinted with permission from ref 150. Copyright 2001 Elsevier B.V.

powder contains ∼1012 beads/g). Also, new bead pools can be created from any number of existing stock solutions, allowing flexibility as experimental needs change. High sensor density (∼25 000 mm-2) and small array size (∼1 mm2) enables the measurement of small sample volumes. The high sensor density enables hundreds, possibly thousands, of duplicate sensor probes, which practically eliminates false positive and false negative results. The signalto-noise ratio, S/N, also is improved, since S/N is proportional to the square root of the number of samples measured, and there are so many duplicate samples available. Finally, because each bead type is prepared in a batch reaction, all the beads of a particular type have virtually identical properties, minimizing array to array variability. Since the microspheres randomly self-assemble into the wells, a registration process must be performed to map the position of the different bead types after array fabrication. There are two ways array registration can be performed. In one method, beads are encoded by fluorescent dyes, which can be used to identify each bead type and the sensing chemistry they are associated with. Single or multiple fluorescent labels at varying concentrations can be used to create optical barcodes to distinguish multiple bead types.151,152 These labels must be different from the dyes used during the analytical measurement. Each element of the array can be rapidly decoded using image-processing software. A second class of methods involves using the analytical properties of the chemistries attached to the beadssthis approach involves decoding the beads. For example, when different DNA sequences are attached to the different bead types comprising an array, the sequences on each bead can be decoded by sequentially hybridizing fluorescently labeled complementary oligonucleotide sequences using a combinatorial algorithm.153 This method requires that binding is reversible as it is important that the fluorescent DNA be removed in order for the array to be used for analytical purposes. Both registration strategies, optical barcoding and nucleic acid encoding, have been used to encode randomly ordered sensing arrays.

3.2.2. Analyte-Specific Sensing Arrays Analyte-specific sensors respond with high selectivity to a given species in a lock-and-key configuration. Classic examples of the lock-and-key mechanism include DNA base pairing and antibody-antigen binding. For analyte-specific probes in a randomly ordered array, knowing the specificity

624 Chemical Reviews, 2008, Vol. 108, No. 2

LaFratta and Walt

Figure 10. Multiplexed detection using a 1-mm diameter fiber optic bundle containing ∼50 000 individual 3-µm optical fibers, each capable of containing an oligonucleotide-functionalized bead. The remaining images show a portion of the fiber bundle and the response of each bead type as well as their collective response. Each bead type is marked using a different color. The blue circle is a positional marker and the same for all images. Reprinted with permission from ref 157. Copyright 2005 American Chemical Society.

of a particular array element is critical for analyte identification. Using optical and nucleic acid encoding strategies, analytes ranging from salivary proteins to biowarfare agents have been detected. The following sections describe randomly ordered arrays used for nucleic acid, protein, and cellbased sensing. 3.2.2.1. Nucleic Acid Detection: Using Optical Barcoding. The optical barcode identification method was demonstrated by Ferguson et al. using 13 different ssDNA probes ranging from 10 to 22 base pairs in length.151 The beads were encoded using combinations of a Europium dye trapped inside the beads and two externally bound dyes, Cy5 and TAMRA, at different concentrations. Amine-terminated DNA probes were attached to amine-functionalized polymer microspheres using a two-step approach shown in Figure 11. First, the amine functionality of the microspheres was increased by a factor of 10 by coupling polyethyleneimine (PEI) to the bead using glutaraldehyde. Second, the amineterminated DNA probes were reacted with cyanuric chloride154 and then covalently bonded to the microspheres. By exciting and monitoring fluorescence at three different optical channels, the concentration of each dye in every bead was determined, thus identifying the ssDNA attached to that bead. Using only 4 µL of solution, target DNA could be detected at concentrations of 100 pM in 10 min and down to 10 fM if allowed to hybridize for longer times (17 h).151 After analyte detection, the array could be regenerated by dipping the fiber in 90% formamide solution to dehybridize captured targets. The probes were regenerated and reused 100 times with negligible deterioration.

Figure 11. (A) Reaction scheme used to attach encoding dye and probes to microspheres. PEI is used to increase the number of functional groups on the bead surface. (B) Depiction of seven bead types, self-assembled into the wells of an etched fiber optic bundle. Dipping the fiber into labeled target solution produces a response only from the beads with the complementary DNA sequence. Reprinted with permission from ref 151. Copyright 2000 American Chemical Society.

The detection limit of this fiber optic microarray was measured in another experiment, performed under slightly modified conditions. The array was reduced to 3 ssDNA probes about 21 bp long, hybridizations times were fixed at 12 h, and the sample volume was increased to 10 µL.155 A

Very High Density Sensing Arrays

smaller number of sensing beads was used, with only ∼10 beads of each DNA probe present in the array, in order to concentrate the small number of target molecules. These conditions allowed detection of 100 aM target DNA samples, equivalent to approximately 600 molecules, using a standard white light source, CCD camera, and microscope optics. Another experiment demonstrated 10 aM detection by integrating the fiber array into a microfluidic channel for sample delivery.156 In the microfluidic system, beads containing two 50-mer oligonucleotide probes were placed in the array and exposed to 50 µL of fluorescently labeled analyte at a flow rate of 1 µL/min. Fluorescently labeled targets have been used in microsphere arrays for the detection of biowarfare agents (BWAs) and bacterial typing. Six BWAs, including B. anthracis and C. botulinum, were detected using 50-mer species-specific probes bound to polymer beads. Cy3-modified reverse primers were used to PCR amplify autoclaved samples of BWAs.154 Using a multiplexed array, these fluorescently labeled targets could then be detected at 10 fM concentrations after 30 min of hybridization using only 50 µL of sample. Twelve strains of the bacteria E. coli have also been typed using a similar detection scheme.157 Fluorescein-labeled reverse PCR primers were used to amplify specific polymorphic regions between 100 and 250 bp in size. Six probe sequences, 33-46 bp long, were each designed to hybridize to a single allele at different polymorphic loci. In principle, these six probes should be able to distinguish 26 ) 64 strains in a binary response format; however, due to allele overlaps, only 12 strains were demonstrated. Detection of unlabeled targets has also been shown using molecular beacon probes. Molecular beacons (MB) are hairpin-shaped oligonucleotides with one end terminated by a fluorophore and the other by a fluorescence quencher.158 Upon binding of the hairpin section to a target sequence, the fluorophore and quencher separate significantly increasing the fluorescence. Biotinylated MB probes for three different genes were bound to streptavidin-coated beads and used to detect unlabeled cystic fibrosis related targets in a random array. The beads contained a unique concentration of internal encoding dye, but all MB probes used the same fluorophore-fluorescein and quencher-4-(4-dimethylaminophenylazo) benzoic acid. Randomly ordered microsphere arrays have also been used for multiplexed sandwich assays to detect other dangerous pathogens, such as foodborne bacteria and harmful algae blooms (HABs). The foodborne pathogen Salmonella spp. was detected in concentrations between 103 and 104 cfu/ mL.159 These samples did not require fluorescent labeling and consisted of chromosomal DNA extracted from lysed cells that had been treated with RNase. Microspheres in the array contained six capture probes, between 20 and 35 bp in length, specific to five different virulence genes of Salmonella spp. After hybridization of the chromosomal DNA, fluorescently labeled oligonucleotide signal probes complementary to a second site on the bound DNA were added. Use of two hybridization events in this sandwich assay format was hypothesized to increase specificity. This system could accurately detect Salmonella spp. even in a mixture of genealogically close organisms, such as E. coli, in 1 h. The same basic scheme was also used in the detection of HABs, such as Alexandrium fundyense, which are associated with toxic blooms in the Gulf of Maine.160 The HAB measurements relied on ribosomal RNA (rRNA) instead of

Chemical Reviews, 2008, Vol. 108, No. 2 625

DNA for detection but still incorporated sequence-specific capture and signal probes. rRNA is present in thousands of copies per cell and therefore requires no amplification. As few as five cells of HAB were detected without any amplification in 45 min, even in the presence of three other closely related HAB strains. This rapid and specific detection mechanism requires minimal sample processing and should be broadly applicable to a number of pathogenic species. 3.2.2.2. Nucleic Acid Detection: Using Nucleic Acid Encoding. Fluorescent barcoding is limited to the number of distinct optical signatures that can be distinguished. Epstein and co-workers sought to increase the number of different bead types in an array using a combinatorial decoding scheme in which the oligonucleotides attached to the bead were used as an intrinsic identifier.161 This approach could also be used as a method for sequencing oligonucleotides attached to beads. Gunderson et al. also demonstrated a combinatorial nucleic acid decoding method.153 This strategy identifies bead location based on sequential hybridization to known, fluorescently labeled targets. They showed that 1520 bead types, each labeled with a unique oligonucleotide between 22 and 24 bases long, could be identified in only eight hydridization-dehybridization cycles. Three possible fluorescent states (red, green, or neither) were prepared for all 1520 complementary strands and pooled in eight combinatorial groups. This procedure gives 38 ) 6561 unique fluorescent responses for all eight pools, more than enough to decode the 1520 sequences. Using a redundancy of about 30 duplicate beads per fiber, this decoding strategy was able to identify nearly 50 000 beads with an error rate of <1 × 10-4 per bead. The fiber optic random array combined with the nucleic acid decoding strategy of Gunderson et al. has been commercialized by Illumina Inc. and used to study gene expression and genome-wide SNP genotyping. Gene expression and RNA profiling studies have been performed by direct hybridization162 and DASL163 (cDNA-mediated annealing, selection, extension and ligation), respectively (Figure 12). Direct hybridization is a standard method for analyzing intact RNA using oligonucleotide probes concatenated to decoding sequences on microspheres. On the basis of the work of Eberwine and co-workers,164 direct hybridization relies on whole genome amplification of RNA. DASL uses probe sequences approximately 50 bases long and is performed by an extension-ligation reaction of two target specific sequences that bind to either side of a gene. One of these target-specific sequences also contains an encoding segment, which can bind to specific beads on the array, enabling the expressed gene to be identified. DASL can study over 500 genes at a time, and since it uses relatively short probe sequences and only 100 ng of total RNA, it is ideal for partially degraded formalin-fixed, paraffin-embedded samples. Whole-genome SNP genotyping has been performed by three techniques: (i) an allele-specific extensionligation reaction analogous to DASL, called GoldenGate, (ii) an enzyme-based assay named Infinium I, which uses an allele-specific primer extension, and (iii) the enzyme-based Infinium II that uses single-base extension reactions. These platforms have been used to genotype over 60% of the SNP loci for the HapMap project using ‘tag’ SNPs in the human genome.165 A ‘tag’ SNP is one that is highly correlated to nearby SNPs, thus reducing the total number of SNPs necessary for identification. These large-scale whole genome

626 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 12. (A) Direct hybridization using a matrix of 96 fiber bundles. The 1.4-mm diameter optical fiber bundle contains >50 000 beads housed in wells at one end of the bundle. Each bead contains a 25-nucleotide identification sequence and a 50nucleotide gene-specific probe. Reprinted with permission from ref 162. Copyright 2004 Cold Springs Harbor Laboratory Press. (B) Schematic of DASL, a cDNA-based assay for RNA profiling. Using biotinylated oligo-d(T)18 and random hexamers, RNA is converted to cDNA and immobilized to a streptavidin-coated solid support. Two oligonucleotides are designed to query each target site of the cDNA. The upstream oligonucleotide consists of a gene-specific sequence and a universal PCR primer (P1). The downstream oligonucleotide consists of a gene-specific sequence, address sequences, and a universal PCR primer (P2). The upstream oligonucleotide hybridizes to the target and extends and ligates to the corresponding downstream oligonucleotide creating a PCR template that can be amplified using P1 and P2. The PCR products are fluorescently labeled and detected, using their address sequence, on beads in an array. Reprinted with permission from ref 163. Copyright 2004 Cold Springs Harbor Laboratory Press.

association studies, which cost about $0.001 per SNP,166 have the potential to revolutionize the identification of diseaseassociated loci, proteins, and pharmacogenomic responses. 3.2.2.3. Protein Detection. Randomly ordered microarrays have also been used for the detection of proteins by several methods. One of the first approaches tried in our laboratory used aptamer-coated microspheres in a competitive binding study to detect the coagulation protein thrombin.167 Aptamers are short oligonucleotides or peptides designed by an evolutionary protocol to bind specific target molecules. For the thrombin study, an anti-thrombin aptamer was bound to silica microspheres and the competitive binding curve was calibrated by measuring the fluorescent response to solutions containing a standard amount of fluorescently labeled thrombin and various concentrations of unlabeled thrombin. The system could detect 1 nM unlabeled thrombin in about

LaFratta and Walt

15 min. In a different experiment, the interaction of proteins and carbohydrates was probed in a multiplexed array.168 The beads were coated with one of five synthetic carbohydrates and exposed to the fluorescently labeled protein cyanovirin N (CVN). The affinity of CVN for three of the carbohydrates was in agreement with previously reported microcalorimetry studies. Immunoassays have been duplexed using bead-based arrays. Szurdoki et al. reported detection of the clinically important drugs digoxin and theophylline using a competitive binding assay.169 Catalyzed reporter deposition (CARD) based on horseradish peroxidase (HRP) was used to enzymatically amplify the signal. Digoxin, in the range of 0.10.4 ppb, and theophylline, between 0.3 and 1.0 ppm, were detected simultaneously. Sandwich immunoassays have also been duplexed using a microsphere array for the measurement of immunoglobulin A (IgA) and lactoferrin, two immune system proteins found in saliva.170 Mouse monoclonal capture antibodies for IgA and lactoferrin were immobilized on beads and placed in the etched wells of fiber bundle. Samples containing IgA and lactoferrin were then incubated on the sensing array for 60 min and detected using another pair of IgA and lactoferrin antibodies that were fluorescently labeled. The detection range was from 385 pM to 10 nM for lactoferrin and 700 pm to 100 nM for IgA with little cross-reactivity, suggesting multiplexed immunoassays should be possible. The etched wells of a fiber optic bundle have also been used in a different format to determine the concentration of extremely dilute solutions of enzymes. For these measurements, the wells are used as a very dense array of microscopic reaction chambers. If the ratio of enzyme molecules to the number of wells is reduced, the Gaussian distribution describing the number of molecules per well reduces to the Poisson distribution. In this regime, the concentrations can be controlled so that only 1 or 0 enzymes will be in a well (Figure 13). Thus, a digital readout of the concentration can be made by comparing the number of wells containing one enzyme to those with zero enzymes. Two strategies were used to confine single enzymes within the wells for low-concentration measurements. In both strategies, single molecules of the enzyme β-galactosidase were observed by observing catalysis of the substrate resorufin-β-D -galactopyranoside (RDG), which yields the yellow fluorescent compound, resorufin, after enzymatic hydrolysis. One method used a mixture of enzyme and substrate confined into single wells by pressing the fiber into an elastomeric gasket. This strategy was able to measure the concentration of β-galactosidase down to 72 fM.171 Another method used biotinylated wells to capture streptavidinmodified β-galactosidase. The captured enzymes were then exposed to substrate using a similar gasket seal. This system was able to detect concentrations as low as 17 fM after an enzyme incubation time of 1 h.172 3.2.2.4. Cell Sensing. Cell-based biosensors offer an advantage over traditional receptor-based biosensors in that they measure function as well as binding. For example, chemical and biochemical sensors operate on the basis of molecular recognition and give a signal when the molecular receptor is occupied. Cell-based biosensors, on the other hand, report on bioavailability, access to target receptors, and binding. For example, a toxin must be able to traverse the cell membrane, maneuver its way to its cellular target, bind the receptor, and cause its biochemical downstream

Very High Density Sensing Arrays

Chemical Reviews, 2008, Vol. 108, No. 2 627

Figure 13. Monitoring the activity of β-galactosidase. (A) Background image of a portion of the fiber. (B) Portion of the fiber with a 1:5 enzyme to well ratio. (C) Portion of the fiber with a 1:80 enzyme to well ratio. (Pseudocolor added using IPlab software). Reprinted with permission from ref 171. Copyright 2006 American Chemical Society.

effect such as secondary messenger generation, gene transcription, proteolysis, etc. Furthermore, the chemical form of the analyte must be one that elicits a cellular response. For example, heavy metals can exist in numerous oxidation states and be associated with a multiplicity of ligands that can affect their bioavailability. By simply measuring metal binding to a receptor, one may be misled about the true toxicity of a sample. Consequently, cell-based biosensors provide additional information that cannot be gleaned from a simple binding event. Cell-based biosensors in an array format offer an additional advantage over traditional high-throughput methods. Typically cellular assays, widely used in drug screening, are performed in 96-, 384-, or 1536-well plates and measure an overall response from wells containing thousands of cells. These cells, each a highly complex system, are slightly different from one another, and their physiological and genetic variabilities are masked in their collective response. In a cellular array, where each cell can be monitored continuously, the detailed stochastic nature of individual cells is revealed. Other methods to measure individual cells, such as flow cytometry,173 can also reveal stochastic variation between cells but lack the ability to track cells over time. Only an array platform enables the monitoring of multiple cells before, during, and after exposure to various stimuli. Arrays of single cells most often confine cells to wells, which are made by photolithography174 or etching fiber optic bundles.175 Soft lithography has also been widely applied to create lower density arrays of cells using either wells or patterned surface chemistries to maintain the position of a cell.176-180 Cells have also been pneumatically trapped using an array of small orifices etched through the surface of a SOI wafer.181 For well-based arrays, the cells are randomly assembled by sedimentation and sustained by a reservoir of nutrients held above them. Mammalian cells can maintain viability for 24 h or more,175 and bacterial cells have been shown to be viable for more than 14 days when arrayed.182 While many studies involve a homogeneous cell array, analyses of mixed populations of cells have also been demonstrated. The different cells have been identified by three labeling strategies: (i) lipophilic dyes,175 (ii) fluorescently labeled lectin,183 and (iii) genetic encoding.182,184 Lipophilic dye molecules are composed of a fluorophore and a long hydrophobic chain and embed themselves into the lipid bilayer of the cytoplasmic membrane. Three lipophilic dyes, PKH 26, PKH 67, and DiIC18, have been used to label mouse fibroblast cells.175 Five fluorescent dye conjugates have also been used for labeling using the lectin concanavalin A (con A). These lectins bind to mannoproteins present on

Figure 14. Scanning electron micrograph of single S. cereVisiae cells distributed in the wells of an etched fiber bundle. Adapted with permission from ref 183. Copyright 2002 American Chemical Society.

cell walls and were used to label five different strains of yeast.183 The third approach, genetic encoding, uses genetically engineered cells to express fluorescent proteins, such as green fluorescent protein (GFP). This method has the advantage that it is a built-in indicator of transcription and translation and can therefore elucidate gene expression profiles while helping to distinguish different cell types contained in a multiplexed array.182,184 Cell noise, or the variation with which identical cells respond to their environment, has been studied in two systems, S. cereVisiae and E. coli, on an array platform.183,184 Dye-conjugated lectins were used to label three strains of yeast to test in vivo protein-protein interactions in the yeast two hybrid (Y2H) system.183 Yeast cells were engineered to transcribe the reporter gene lacZ upon protein interaction. The three yeast strains, positive (interacting proteins), negative (noninteracting proteins), and wild type, were randomly assembled into the wells of a fiber bundle (Figure 14). After decoding the array and adding a fluorogenic substrate, highly stochastic responses were obtained for the positive control strain. Further studies of this system confirm that a range of responses from ostensibly identical cells.185 Cell noise in bacteria was analyzed by arraying two strains of E. coli carrying the fusions recA::gfp and lacZ::gfp.184 For both recA and lacZ the expression became noisier with time. In the induced state, lacZ showed 5 times greater noise compared to recA, possibly due to its more complex gene network. The information-rich data of these studies, showing the stochastic nature of gene translation and transcription

628 Chemical Reviews, 2008, Vol. 108, No. 2

dynamics, could only be collected from cellular arrays. Cellular arrays have also been used for toxicity and drug screenings and identify and isolate antigen-specific B-cells. Genetically engineered E. coli was used to measure as low as 100 nM Hg2+, based on the expression of reporter genes.186 In another study, the effectiveness of the antimigratory drug nocodazole was verified by monitoring rates with which individual cells traversed the optical cores of a fiber optic bundle.187 A third report used a very high density array (>140 000 wells/cm2) of single lymphocytes to identify antigen-specific B-cells.188 The response from each cell was measured after exposure to an antigen, and responsive cells were isolated from the array by a micromanipulation pipet. These three studies exemplify the benefit of functional biosensorssbecause cells are alive, they can measure things beyond just binding. Cell-based sensors also measure bioavailabilty, access to key cellular components, and show the effect on the overall biological system. Despite the aforementioned benefits of cellular arrays, there are several issues that have limited their commercialization.189 One issue is well sizessince cells come in many shapes and sizes, there is no universal well size that works for all cell types. An assortment of well sizes would likely be necessary for different experiments. Likewise, for bacteria and yeast cells, wells should be only a few micrometers in diameter. This size is somewhat challenging because it coincides with the smallest feature sizes achievable on a typical mask aligner used for photolithography. Another concern is the ability of cells to communicate with each other when confined to wells. Work needs to be performed to prove that collections of separated cells respond in the same way as a collection of unconstrained cells. Finally, it is extremely difficult to isolate individual cells from the array for further analysis, such as gene expression. Technical improvements in array manufacturing and cell manipulation, as well as further studies into the biological consequences of isolating cells, will likely lower the barriers of commercializing cellular arrays.

3.2.3. Cross-Reactive Sensing Arrays A sensing element that has a broad range of specificity and responds to a wide variety of analyte species is known as a cross-reactive sensor. Evaluation of an analyte by a cross-reactive array is based on the overall response pattern of the array for the unknown substance in comparison to the response pattern from known controls.190 The scheme is based loosely on principles of the mammalian olfactory system. In some mammalian olfaction systems, millions of olfactory receptor neurons respond to a given odor with each neuron expressing only one type of receptor out of a repertoire of ca. 1000 receptors. The various responses of these neurons are sent to the brain for processing where the pattern is recognized based on previous exposure to similar mixtures, thereby creating an odor memory. Every receptor (and neuron) responds to many different vapors but differentially such that with 1000 different receptors a nearly infinite number of patterns can be generated. This combinatorial advantage enables sensor arrays to be created that do not rely on traditional “lock and key” binding. Specificity is encoded in the response pattern rather than in any specific sensor; hence, the term “distributed specificity” has been applied to this approach.191,192 An “artificial nose” based on the principles of cross-reactivity employs many semi-selective sensing elements. In this approach, a pattern

LaFratta and Walt

recognition algorithm must be trained first to recognize the vapors of interest. When the sensor array is subsequently exposed to a vapor in the database, the algorithm compares the response of an unknown vapor to the responses from prior training. This technology could have broad applications ranging from monitoring food and air quality to detecting explosives.193 Sensors with broad response use physical and chemical properties common to all molecules such as polarity or hydrophobicity. Analyte molecules can span the continuum of a given physicochemical property. Using a single sensor is insufficient to detect very similar analytes because their physicochemical properties may be similar. Multiple sensors, responsive to different properties or in different ranges, must be used to distinguish between molecules with similar properties. Transduction mechanisms for a variety of physicochemical properties have been demonstrated to create cross-reactive sensors. One transduction mechanism is the adsorption of gas in metals,194 metal oxides,195 semiconductors,196 or conducting polymers,197 which changes their conductive properties, enabling electrical measurements of vapor samples. Solvatochromic dyes,152 ion-selective electrodes,198 and surface acoustic wave sensors199 have also been used in cross-reactive detection. In very high density fiber optic bead arrays, solvatochromic dyes such as Nile Red are adsorbed onto the surface or entrapped in various polymer or porous silica beads.152 When a vapor is sorbed into the polymer beads, the fluorescent reporter shifts wavelengths and/or intensity based on polarity changes in the bead during exposure to analyte vapors. The different bead types have different polarities and consequently exhibit different spectral and sorption properties. A time trace of these changes is collected from all the beads in the array using the optical system with image acquisition software. Other factors also influence the time trace, such as the porosity of the bead, its ability to swell, and its hydrophobicity. By using a system like the one shown in Figure 9 to monitor fluorescence traces versus time over different bead types, characteristic responses profiles are generated for different analytes. The response pattern of known analytes, depicted in Figure 15, can then be used to train pattern recognition software, such as artificial neural networks, in order to classify the response of unknowns.200 As discussed above, the different bead types in this array are distributed randomly. Because like elements of the array respond the same, they can be readily identified by exposing them to a known test vapor because the response pattern from a given sensor type to a given vapor is reproducible. This ‘self-encoding’ mechanism allows the random array to be decoded if desired. A separate decoding step is not necessary, however, because cross-reactive arrays make an identification using response patterns and, as long as the sensors and responses are reproducible, can be compared to a response library.201 There are many advantages to having a very high density cross-reactive sensor array. One benefit is analogous to the NEEs where small individual signals measured by each element of the ensemble are integrated to give a large collective response. Since there can be many thousands of beads with the same sensing dye, their responses can be combined to amplify what may be a very small fluorescence signal or signal change.152 Another advantage is the increased sensitivity and rapid response time, typically a few seconds, of the microspheres due to their small size and high surface to volume ratio. The microsphere platform also improves

Very High Density Sensing Arrays

Chemical Reviews, 2008, Vol. 108, No. 2 629 Table 2. Odor Discrimination Accuracy for 100 Odor Exposures When All Sensor Responses Are Combined (Nondecoded Arrays)a classification rate (%)

Figure 15. Schematic illustration of a self-encoded bead array. A mixture of sensor beads is prepared by combining beads from three stock solutions. A drop of the mixture is placed on the etched end of a fiber optic bundle. The beads are identified and categorized by the characteristic responses to a test vapor pulse. Since the analytical signal of each bead also identifies the bead and maps its position in the array, the beads are self-encoding. Reprinted with permission from ref 152. Copyright 1999 American Chemical Society.

array to array reproducibility because millions of identical beads are created and stored together. This reproducibility allows a training database to be carried over from array to array despite the differences in location of the beads between two arrays.202 By summing all identical bead types in an array, slight bead-to-bead variations are also eliminated. While these features make this system attractive, it still suffers from some drawbacks. One problem is sensor poisoning upon exposure to reactive analytes. Another drawback is photobleaching of the dye over long periods of time. Strategies such as illuminating subsections of the array and using an adaptive light exposure scheme by beginning the experiment at low illumination levels and gradually increasing to compensate for photobleaching have been developed to avoid this problem.203 Fiber optic cross-reactive sensing arrays have been demonstrated in artificial nose applications to detect nitroaromatic explosive-like compounds (NACs) and complex vapor mixtures such as distinguishing brands of coffee and living/ dead bacteria. NACs like 1,3-dinitrobenzene (DNB) and 2,6dinitrotoluene have been detected at ppb levels, even in the presence of volatile organic compounds, such as toluene and benzene, at levels thousands of times higher.204 The sensors were shown to have a shelf life of at least 10 months, and their responses were highly reproducible. The complex odor samples consisted of three varieties of coffee bean along with acetone, toluene, and DNB.205 Using discriminant functional analysis, these six samples could be identified with 100% accuracy at high concentration levels and 85% accuracy at lower levels. Larger numbers of vapors have also been classified. More specifically, 20 odor compounds consisting of several alcohols, alkanes, aromatics, and several twocomponent mixtures were distinguishable with greater than 90% accuracy using between 6 and 18 sensor types (Table 2).201 The ability to ‘learn’ the profile of a large number of vapors and distinguish chemically similar species rapidly, in high backgrounds, is necessary to realize the goal of realtime vapor detection systems for applications ranging from monitoring food quality to national security.

array type

trial I

trial II

single sensor single sensor single sensor 03-bead random 06-bead random 09-bead random 12-bead random 15-bead random 18-bead random

74 86 76 80 86 93 95 85 97

74 98 94 98 97 94 94 98 96

a Three distinct response patterns are obtained for fluorescence vs time traces for three different beads types after exposure to the same vapor. All arrays in trials I and II employed different microsensor types, even for the 01-bead arrays. The only arrays with the same sensor composition for I and II were the 18-bead arrays (see ref 201). Twenty different odor exposures (5 replicates each): (1) air carrier gas, (2) acetone, (3) n-heptane, (4) ethanol, (5) toluene, (6) water, (7) ethanol/ heptane mixture 1:1 (v/v), (8) methanol/ethanol mixture 1:1 (v/v), (9) benzene, (10) 1-propanol, (11) aqueous 90 ppb 1,3-dinitrobenzene, (12) 1,3-dinitrobenzene (s), (13) methanol/1-propanol mixture 1:2 (v/v), (14) methanol, (15) 1-butanol, (16) 3-pentanol, (17) p-xylene, (18) ethanol/ 1-pentanol mixture 1:3 (v/v), (19) cyclohexanone, and (20) 1-pentanol. Since there were 100 observations, the number of misclassifications is apparent from the classification rate ((97%) 3 mistakes; (86%) 14 mistakes). Reprinted with permission from ref 201. Copyright 2003 American Chemical Society.

3.3. Suspension Arrays 3.3.1. Introduction A third class of very high density arrays is suspension arrays. Unlike the previously mentioned arrays, suspension arrays are not in a fixed 2-D pattern. Instead, the array elements, which typically consist of microspheres, are free floating or suspended in solution. While the term “array” is probably a misnomer, this terminology is used to describe assays performed on microparticles in solution. This type of array is considered very high density because the typical element size is about 5 µm in diameter. Because the elements are not in a fixed pattern, it is impossible to analyze the array with the same detection methods that are used for planar arrays, such as imaging. Instead, the microparticles are scanned individually. The methods used for scanning microparticles in solution have their roots in flow cytometry. Flow cytometry is a well-established method of counting, sorting, and examining microparticles. Initially developed in the 1970s for cell counting and sorting, flow cytometry works by hydrodynamically focusing particles from a sample into a narrow stream where the particles move in single file.206 The particles pass through the beam of a laser, and the scattered light and/or any resulting fluorescence is detected (Figure 16). Particles can be scanned at rates up to 100 000 particles per second, and several lasers or detectors can be used simultaneously for multiparameter analysis. For multiplexed suspension array assays, standard cytometry equipment can be used, but the microcarriers must be encoded to identify which analyte they are sensing. Several reviews have been written on the topic of suspension arrays207-209 and their encoding,210,211 but in general there are two main encoding optionssspectral or graphical. The more established encoding method is spectral and usually done by fluorescent labeling. Dye labeling with multiple

630 Chemical Reviews, 2008, Vol. 108, No. 2

LaFratta and Walt

Figure 16. Schematic illustration of a flow cytometer used in a suspension array. The sample microspheres are hydrodynamically focused by a sheath fluid and passed through two laser beams. The fluorescence caused by laser 1 is detected at two wavelengths to identify the encoding dyes to determine which analyte the microsphere captures. Laser 2 excites fluorescence at a third wavelength and is used to quantify the bound analyte.

fluorophores in several concentrations has been commercialized for the encoding of up to 100 bead types by Luminex Corp. (Austin, TX). Luminex uses orange and red fluorescent dyes for encoding and a third dye, often green, for analysis. The 5.6 µm polystyrene beads can be identified and measured at a rate of about 1000 per second. With this throughput, a 100-plex assay consisting of ∼200 duplicates of each bead type could be read in about 30 s. This acquisition rate would enable nearly 3000 100-plex assays to be performed per day.209 While these numbers are impressive, the maximum number of about 100 encoding combinations is limited compared to the multiplexing available in planar arrays. The other encoding method, graphical encoding, promises a much higher degree of multiplexing. Graphical encoding is done by imprinting an identifying code into a particle based on its size, shape, or composition (Figure 17). For example, silica nanowires that are composed of thick and thin segments of various sizes can be visibly distinguished from one another.212,213 Likewise, nanowires made of silver and gold layers can be identified by the pattern length and frequency of the different metal segments.42 More elaborate graphical encoding schemes have recently been examined by Doyle and co-workers.214 By combining photolithography and microfluidics, Doyle created microcarriers that resemble computer punch cards. These pill-shaped microparticles are on the order of 100 µm in width and 300 µm in length; the encoding region occupies about one-half of the microparticle area with the remaining portion being used for analysis. This scheme has the potential to produce millions of encoding combinations for highly multiplexed detection but has thus far only been demonstrated on a small scale with several analytes. While still in its infancy, graphical encoding seems very promising and will likely be the focus of many future applications. The established microsphere-based technology has been used by many groups for various multiplexed analysis with several thousand Luminex systems in place for both research and clinical applications.215 Protein detection has been performed for a wide range of applications. As will be

Figure 17. (A) Schematic of an alumina template used to create shape-encoded silica nanotubes. Adapted with permission from ref 212. Copyright 2006 American Chemical Society. (B) Dark-field optical micrograph of silica nanotubes prepared using template shown in A; the larger diameter segments are more reflective and therefore look brighter. Reprinted with permission from ref 213. Copyright 2006 American Chemical Society. (C) Optical and (D) FE-SEM micrographs of a single Au-Ag multistriped particle. The gold sections are ∼550 nm in length, and the silver sections range from 60 to 240 nm in length. Reprinted with permission from ref 42. Copyright 2001 American Association for the Advancement of Science. (E) Optical micrograph of dot-coded polymer microparticles. One-half of the particle is for encoding, while the other half is used for analyte detection. Scale bar ) 100 µm. Reprinted with permission from ref 214. Copyright 2007 American Association for the Advancement of Science.

described, suspension arrays have also been used for nucleic acid analysis, such as genotyping and gene expression.

3.3.2. Protein Detection The use of suspension arrays for protein detection was proposed as early as 1977 and has been used extensively for immunoassays for over 20 years.206,216-218 Fulton et al.

Very High Density Sensing Arrays

completed pioneering work in the area of multiplexed immunoassays by testing canine serum for IgG and IgE antibodies specific to 16 grass allergens simultaneously.219 Similarly, Carson and Vignal detected 15 cytokines, including IL-2, IL-4, and IFN-γ, using only 100 µL of sample.220 Using a more traditional enzyme-linked immunosorbent assay (ELISA) would require 100 µL for each of the 15 cytokines in the assay. Other studies confirmed that suspension arrays are more reproducible, have a greater dynamic range, and require less preparation time than conventional ELISA.221 Immunoassays performed on suspension arrays have been used to measure allergens, toxins, and even explosives. The binding affinity of 17 plant lectins, such as peanut and wheat germ agglutinin, for 13 different glycoproteins was measured on a suspension array.222 Microspheres were prepared by coating their surface with one of the 13 glycoproteins, and then the 13 bead types were exposed to biotinylated lectins. After incubation, the lectins were labeled using R-phycoerythrin-conjugated streptavidin, and the binding was measured through flow cytometry. The determined affinities were in good agreement with previously reported values.222 In another study, a library of single-domain antibodies (SdAb) from llamas was created and screened using a suspension array to find SdAbs that bound toxins, such as ricin and cholera.223 The antigen binding arms, or V domains, of SdAbs are particularly interesting; because they are the smallest natural binding domains, they are inherently thermostable and because they can often refold after denaturation. Error-prone PCR SdAb genes from three llamas were mutated to create SdAbs that would selectively bind one of six toxins.223 In another screening study, six antibodies were designed to bind TNT and other nitroaromatics and tested in a competitive binding study. The best antibody could detect between 0.1 ppb to 10 ppm of TNT.224 Suspension array-based immunoassays have also been used for exploring the detection and mechanisms of viruses, such as influenza and HIV. For example, the influence of HIV on plasmacytoid dendritic cells (pDCs) has been studied using a suspension array system made by BD biosciences (San Jose, CA).225 It was shown that HIV-1 g120 interferes with pDCs ability to secrete type I IFNs. Similarly, experiments using an unusually severe influenza virus from 1918 were completed on non-human primates. Results from a suspension array-based immunoassay revealed that this strain has the ability to modulate the innate immune response of the host, which could be a common trait among virulent influenza viruses like avian H5N1.226

3.3.3. Nucleic Acid Detection In 2001, Yang et al. described the use of a suspension array for gene expression.227 Using fluorescently encoded microspheres, they were able to quantify the presence of 20 RNA sequences in each sample. Sample RNA was amplified by PCR with biotinylated primers and then captured by cRNA immobilized on microspheres. Following streptavidin-phycoerythrin labeling, the beads were analyzed on a flow cytometer. One advantage of this method is that large numbers of different cRNA beads are made and can be aliquoted for use in many experiments. This method fills a niche not served by other methods of gene expression, such as high density lithography arrays, because it is a fast and cost-effective way to test a relatively small number of genes in a large number of samples.

Chemical Reviews, 2008, Vol. 108, No. 2 631

More recently, gene expression has been performed using suspension arrays to compare the expression of micro RNAs (miRNAs) between cancerous and healthy cells.228 Suspension arrays were chosen for this study because the short size of miRNAs (∼21 nucleotides), and the similarity between miRNA family members often causes cross-hybridization on planar glass slide arrays. A study was performed comparing the extent of cross-hybridization to each of the two different array formats, and the suspension arrays performed better than the planar arrays for all 11 miRNAs tested. Overall, results from the analysis of 217 mammalian miRNAs from 334 samples found a general trend toward down regulation of miRNAs in tumor cells compared to healthy cells. The researchers also observed that poorly differentiated tumors could be classified by miRNA analysis more effectively than by mRNA profiling. Suspension arrays have also been used to multiplex the detection of mRNA using a sandwich assay with amplification technology involving branched DNA (bDNA).229 In this method, capture probes on microspheres bind to multiple locations on the target mRNA. Highly branched DNA labeled with biotin can then bind to the mRNA to form the sandwich. A streptavidin-phycoerythrin conjugate is then used to tag the bDNA, indicating the presence of the mRNA. This technique does not amplify or purify the target mRNA and can be used to analyze crude cell lysates or tissue homogenates. Flagella et al. multiplexed their assay to simultaneously detect 10 mRNA sequences with sensitivity down to 25 000 RNA transcripts.229 Several methods of SNP genotyping have also been demonstrated using suspension arrays. A direct hybridization technique was used by coating four types of fluorescently encoded microspheres with four oligonucleotides that varied by only a single base.230 The labeled target then bound to only one of the four bead types identifying the SNP; eight SNPs have been detected in this way, requiring 32 bead types. Two other methods, known as oligonucleotide ligation assays (OLA)231 and single base chain extension (SBCE) assays,232,233 have also been used. Highly analogous to the GoldenGate and Infinium assays developed by Illumina Inc., OLA and SBCE use microspheres coated with oligonucleotides that act as address encoders, known as ZipCodes, for binding amplified and labeled product. Unlike the Illumina encoding method, however, ZipCodes are not sufficient to identify the microsphere; instead, the microspheres are still fluorescently encoded, and the ZipCode acts as an intermediate linker to associate a particular bead with a particular SNP call. These methods have been used to identify over 50 SNPs simultaneously.232 While fluorescently encoded suspension arrays are far below their planar array counterparts in terms of the number of SNPs they can call, they may fill a niche where a small number of SNPs need to be rapidly genotyped among a large number of samples.

4. Future Directions There are many promising materials and technologies that one day may enable the preparation of very high density sensing arrays. In some cases, substrates have been created with feature sizes suitable for implementation with very high density sensing arrays. In other cases, functional materials containing both array characteristics and the ability to transduce signals exist. Other technologies exist that may one day enable the readout of very high density sensing arrays at scales that cannot be achieved using today’s technologies.

632 Chemical Reviews, 2008, Vol. 108, No. 2

4.1. Substrates and Materials 4.1.1. New Materials Materials scientists are developing a significant number of new substrates that offer potential platforms for very high density sensing arrays. One of the most promising of these substrates is anodically etched alumina, which has been used for the NEEs. Martin and co-workers recently used the alumina nanopore membrane as a mask by overlaying it on a polymer during a plasma etching process. After removing the alumina mask, a regular array of nanopores is created.31 Silica can be deposited in these nanopores to produce nano test tubes.43 The diameter of these test tubes is approximately 85 nm, and the depth can be controlled by the length of plasma etching. These alumina membranes can also be etched to produce an array of conical nanopores, which have been used as synthetic resistive-pulse sensors for stochastic measurement of biomedical analytes.234 As the fabrication of these conical pores becomes more reproducible, use of artificial-nanopore biosensors will likely become more widespread.235 A wide variety of other promising materials exist with the potential for creating very high density sensing arrays. Such materials include wire ensembles (e.g., carbon nanotubes, metal wires),236-238 colloidal crystal arrays,239,240 self-assembled nanostructures,241,242 polymeric and silica microsphere monolayers,57 and metal nanoparticle arrays.243 All of these materials offer attractive and tantalizing substrates for creating a variety of different very high density array architectures. The ability to capitalize on these materials will depend on the ingenuity of materials scientists, chemists, and life scientists.

4.1.2. Functional Materials At a higher level of sophistication are functional materials in which, in addition to an array format substrate, some form of function is integrated. For example, zinc oxide nanowire ensembles have been created.244 ZnO nanowires have been grown epitaxially on an alumina substrate using gold particles as a catalyst. The resulting nanowires exhibit a piezoelectric effect such that mechanical stimulation of the wires leads to an electrical signal. This approach integrates both array fabrication with a transduction mechanism. At present, this approach remains relegated to an ensemble as all the nanowires are connected to a single readout device. Another approach to functional sensors employs molecular valves. In this approach, rotaxanes are attached at the openings of mesoporous silica nanopores.245-247 The rotaxanes can be switched to one of two conformations using an electrochemical or redox reaction resulting in opened or closed nanopores. Consequently, the rotaxanes act as nanovalves to open or close a channel. While exhibiting a functional response, all nanovalves are comprised of the same rotaxanes resulting in a uniform response of the entire material. In addition, the valves do not exhibit selectivity in the types of molecules that are released or allowed to enter from the pores. By integrating chemical selectivity into such nanovalve arrays, it may be possible to create extremely high density sensor arrays. Recently Aizenberg and co-workers reported very high density ensembles of hydrogel nanocolumns that were responsive to humidity.248 The shapes of the nanocolumns could be controlled by the stress field in the hydrogel. The regular pattern of nanoscale features combined with an

LaFratta and Walt

intrinsic responsivity suggests that mechanical transducers with built-in response mechanisms can be fabricated. A host of hydrogel sensors already exists that are responsive to a variety of chemicals such as ions, glucose, and neurotransmitters that could be integrated into such arrays.72-74 In this context, the ability to create new materials or readout mechanisms that do not require labels is one of the major future goals for any sensing method. By using the intrinsic signals of the materials upon binding analytes or developing new readout methods that can detect analyte binding, it should be possible to simplify array design and increase the level of multiplexing significantly.

4.2. Novel Array Designs 4.2.1. Molecular Arrays Perhaps the ultimate in density will be when single molecules can serve as the array elements. For example, Bayley and co-workers have been developing elegant methods for engineering R-hemolysinsa pore protein that in its natural form punctures red blood cell membranes.249 The engineered forms of R-hemolysin can be designed with molecular specificity to allow specific molecules to traverse the pores. By measuring the conductivity of the membranes, stochastic binding events can be measured from single analyte molecules binding to the pore.249 If these pores can be arranged in an array format and measured individually, they will offer an unprecedented density of molecular sensors.234 Seeman and co-workers recently reported their ability to tile DNA structures with pendant arms that enable molecular attachment and recognition.250,251 This approach offers a spectacular demonstration of self-assembly and offers the potential for creating molecular arrays with the ability to direct multiple and different receptors to defined sites.

4.2.2. Liquid Arrays A revolutionary approach to creating very high density sensing arrays involves creation of liquid or “virtual” arrays. In this approach, optical traps are employed to capture microspheres or cells in liquids. Optical traps, also called optical tweezers, are created by focusing a high-intensity laser to a small spot. Because of the refractive index differences between particles and the liquids in which they reside, momentum can be imparted to the particle such that the particle is confined to the focal point of laser beam. Recently, arrays of optical traps have been created using holography252 or optical fibers.253 By integrating microfluidics with these systems, it is possible to trap many particles simultaneously in either two or even three dimensions. Individual traps can be controlled to either hold or release a given particle or cell. One can imagine using such microsphere arrays to analyze samples, release the microspheres when they are exhausted, and then create another array out of fresh microspheres without the need for any substrate.

4.3. New Tools and Devices 4.3.1. Optical One of the existing limitations for very high density sensing arrays is the ability to read out the individual sensing elements at the requisite resolution. A number of optical methods for breaking the diffraction limit of light have been developed recently and offer the potential to be used for array

Very High Density Sensing Arrays

readout. One of the earliest methods for breaking the diffraction limit is near-field scanning microscopy (NSOM). This method is still relegated to the laboratory and a relatively slow technique. More rapid techniques using more conventional microscope platforms have been developed recently. For example, the stimulated emission depletion (STED) approach involves illuminating a sample with a highly focused laser beam to excite a fluorescent dye while simultaneously illuminating with a doughnut-shaped beam to deexcite fluorophors outside the region of interest.254,255 Using STED microscopy, resolutions of 20 nm can be achieved, potentially enabling the ability to read extremely high-density arrays or ensembles. Another method is the sub-diffraction-limit imaging by stochastic optical reconstruction microscopy (STORM).256,257 In this approach, only a fraction of the fluorophors in an image field is excited. By building up a series of fluorescence images, each with sub diffraction resolution of multiple sites within the field, it is possible to achieve resolutions of 20 nm.

4.3.2. Surface Readout STM, AFM, and related methods have been the cornerstone of surface analysis over the past decade. With the ability to scan more rapidly using less expensive and smaller systems, the ability to integrate surface readouts with very high density sensing arrays in an inexpensive format is on the horizon. The work of Mirkin and co-workers in preparing a very high density of cantilevers for dip-pen nanolithography should make such readout devices practical.90 There has been a revolution in device fabrication over the last several decades. For example, CCD chips are now commonplace in digital cameras; CMOS devices are in children’s toys; microfluidics and MEMS systems are pervasive. These devices will undoubtedly enable a transformation of the very high density array field over the next few years. CMOS devices are of particular note as they are inexpensive and possess on-chip processing. They are megapixel devices with all the integrated circuitry. One can imagine that these devices may be used directly for fabricating sensor arrays by simply attaching different chemistries at different pixel locations.

4.4. Novel Applications of Very High Density Sensing Arrays In the most optimistic scenario very high density sensing arrays containing thousands to millions of individually addressable nanoscale elements will be accessible. Assuming that the requisite chemistries for performing molecular recognition of thousands of different species is developed, such arrays will have the capability for performing a high level of multiplexed sensing or analysis. Such arrays will have tremendous functionality, be inexpensive because the materials costs will be low (due to the small amounts of material required), and should provide a universal platform for low-cost analysis. These arrays may have specific sensors, cross-reactive sensors as discussed above, or both. Advances in cell-based sensing arrays could revolutionize functional sensing (vide supra) by enabling rapid and highcontent screening for new drug candidates including absorption, distribution, metabolism, excretion, and toxicity.258 The ability to array different types of cells in precise locations also offers the possibility to design tissue mimics and

Chemical Reviews, 2008, Vol. 108, No. 2 633

understand how different cell types communicate and affect one another. Langer and co-workers developed a controlled release drug delivery array in which reservoirs can be filled with drugs and sealed.259 The back of the array has microcircuitry that allows the release of drugs from different regions of the chip upon electrical actuation. As the array reservoirs become smaller and methods are developed for loading different drugs in different regions of the chip, the ability to control release on a much finer scale will be possible. In addition, sensors may one day be integrated into the array to enable simultaneous analyte sensing and multidrug controlled release. Such arrays could find use as implantable monitoring devices for detecting an oncoming illness (e.g., infectious disease or heart attack) and autonomously take action by releasing drugs or nutrients to prevent their occurrence. Other manifestations of very high density sensing arrays will enable tremendous advances in fundamental science. For example, arrays of many individual cells can be simultaneously interrogated using sensitive patch clamp techniques, which will enable functional sensing for drug discovery applications.260 Arrays of fully functional genes that can be translated into proteins localized to the region where they are translated will enable studies of protein-protein interactions as well as biochemical pathways.261 Maerkl and Quake recently demonstrated the ability to integrate a fluidic delivery system with a DNA microarray to measure transcription factor binding constants.262 As arrays become higher density and multifunctional, the ability to collect fundamental chemical, biochemical, and biological information will increase.

4.4.1. Next-Generation Sequencing One of the most exciting contemporary areas in life sciences technology is the field of next-generation sequencing. Over the last several years the cost of de novo sequencing has been reduced more than 2 orders of magnitude as a result of new technologies. In most of these technologies, a single molecule of DNA is amplified either on a bead or after binding to a surface. In the former approach, each bead represents a “clone” of a particular sequence and the library of beads is then spread onto a substrate or confined in microwells. In the latter approach, each single DNA molecule is replicated manyfold and confined to a small spot on a substrate. Using a series of biochemical steps such as elongation, ligation, dye attachment, and/or hybridization, the sequences of the immobilized DNA can be determined. Sequence determination is conducted in parallel on many thousands to millions of DNA strands simultaneously. Most of this work is being carried out by commercial entities. All of these approaches employ random arrays in which the positions of particular DNA molecules are undetermined. In some cases, the array format is identical to existing very high density sensing arrays such as the 454 sequencing technology that relies on fiber optic microwell arrays as described above.263 In other cases, such as the Solexa264 or Agencourt265 approaches, single DNA molecules are amplified and deposited randomly on a planar substrate. In perhaps the highest density approach, Helicos is pursuing similar technology in which random arrays of single DNA molecules are sequenced with no amplification.266

4.5. Issues One of the most significant issues with very high density sensing arrays is the lack of methods to prepare arrays

634 Chemical Reviews, 2008, Vol. 108, No. 2

containing different features in different locations. For example, molecular receptors if configured properly can be used to create sensors. Even if very high density array substrates are available, with present day technology it is painstaking to put different receptors in different locations of an array in a precise and registered manner. Although dip-pen nanolithography offers a set of tools for accomplishing this task, it is not amenable to all array formats. A related issue is connecting the arrays such that individual signals from each array element can be detected. With optical methods, direct connections are not necessary; however, with electrical and/or mass measurements, a direct connection to the transducer is required. Direct connections to the different array elements must also be to an external readout device. Readout devices for both optical and electrical measurements must be small and inexpensive for most very high density array applications; otherwise, the use of such arrays will be relegated to research laboratories. Despite the issues discussed above and the challenges remaining before very high density sensing arrays achieve their full potential, the transformation has been remarkable in terms of the short time it has taken to move from single measurements to the high density high-content array formats in use today. The opportunities presented by new materials, devices, and tools, coupled with the clever designs of scientists working at the micro- and nanoscales, promises rapid advances in very high density sensing arrays that will permanently transform the fields of measurement science and life sciences.

5. Acknowledgments We would like to acknowledge the TEACRS program of the NIH/NIGMS as well as the Howard Hughes Medical Institute for supporting this work. Assistance from members of the Walt lab, including Timothy Blicharz, Ragnhild Whitaker, Ryan Hayman, and Dr. David Rissin, is also acknowledged.

6. References (1) Tehan, E. C.; Higbee, D. J.; Wood, T. D.; Bright, F. V. Anal. Chem. 2007, 79, 5429. (2) Schena, M.; Heller, R. A.; Theriault, T. P.; Konrad, K.; Lachenmeier, E.; Davis, R. W. Trends Biotechnol. 1998, 16, 301. (3) Bowtell, D. D. L. Nat. Genet. 1999, 21, 25. (4) Schena, M. Microarray Analysis; John Wiley & Sons Inc.: Hoboken, NJ, 2003. (5) Penner, R. M.; Martin, C. R. Anal. Chem. 1987, 59, 2625. (6) AlMawlai, D.; Coombs, N.; Moskovits, M. J. Appl. Phys. 1991, 70, 4421. (7) Wightman, R. M. Anal. Chem. 1981, 53, 1125A. (8) Dayton, M. A.; Ewing, A. G.; Wightman, R. M. Anal. Chem. 1980, 52, 2392. (9) Sˇ tulik, K.; Amatore, C.; Holub, K.; Marecek, V.; Kutner, W. J. Pure Appl. Chem. 2000, 72, 1483. (10) Zoski, C. G. Electroanalysis 2002, 14, 1041. (11) Bard, A. J.; Crayston, J. A.; Kittlesen, G. P.; Varco Shea, T.; Wrighton, M. S. Anal. Chem. 1986, 58, 2321. (12) Hulteen, J. C.; Menon, V. P.; Martin, C. R. J. Chem. Soc., Faraday Trans. 1996, 92, 4029. (13) Davies, T. J.; Compton, R. G. J. Electroanal. Chem. 2005, 585, 63. (14) Saito, Y. ReV. Polarogr. 1968, 15, 177. (15) Menon, V. P.; Martin, C. R. Anal. Chem. 1995, 67, 1920. (16) Davies, T. J.; Ward-Jones, S.; Banks, C. E.; del Campo, J.; Mas, R.; Munoz, F. X.; Compton, R. G. J. Electroanal. Chem. 2005, 585, 51. (17) Jeoung, E.; Galow, T. H.; Schotter, J.; Bal, M.; Ursache, A.; Tuominen, M. T.; Stafford, C. M.; Russell, T. P.; Rotello, V. M. Langmuir 2001, 17, 6396. (18) Thurn-Albrecht, T.; Schotter, J.; Kastle, G. A.; Emley, N.; Shibauchi, T.; Krusin-Elbaum, L.; Guarini, K.; Black, C. T.; Tuominen, M. T.; Russell, T. P. Science 2000, 290, 2126.

LaFratta and Walt (19) Martin, C. R. Chem. Mater. 1996, 8, 1739. (20) Martin, C. R. Science 1994, 266, 1961. (21) O’Sullivan, J. P.; Wood, G. C. Proc. R. Soc. London A 1970, 317, 511. (22) Li, A.-P.; Muller, F.; Birner, A.; Nielsch, K.; Gosele, U. AdV. Mater. 1999, 11, 483. (23) Masuda, H.; Fukuda, K. Science 1995, 268, 1466. (24) Fleisher, R. L.; Price, P. B.; Walker, R. M. Nuclear tracks in solids: Principles and applications; University of California Press: Berkeley, CA, 1975. (25) Tierney, M. J.; Martin, C. R. J. Phys. Chem. 1989, 93, 2878. (26) Parthasarathy, R. V.; Martin, C. R. Nature 1994, 369, 298. (27) Wilson, J. N.; Bangcuyo, C. G.; Erdogan, B.; Myrick, M. L.; Bunz, U. H. F. Macromole. 2003, 36, 1426. (28) Lakshmi, B. B.; Patrissi, C. J.; Martin, C. R. Chem. Mater. 1997, 9, 2544. (29) Che, G.; Jirage, K. B.; Fisher, E. R.; Martin, C. R.; Yoneyama, H. J. Electrochem. Soc. 1997, 144, 4296. (30) Kang, M.; Trofin, L.; Mota, M. O.; Martin, C. R. Anal. Chem. 2005, 77, 6243. (31) Kang, M.; Yu, S.; Li, N.; Martin, C. R. Small 2005, 1, 69. (32) Chen, L.; Yin, A.; Im, J. S.; Nurmikko, A. V.; Xu, J. M.; Han, J. Phys. Status Solidi A: Appl. Res. 2001, 188, 135. (33) Masuda, H.; Yasui, K.; Watanabe, M.; Nishio, K.; Rao, T. N.; Fujishima, A. Chem. Lett. 2000, 29, 1112. (34) Che, G.; Lakshmi, B. B.; Martin, C. R.; Fisher, E. R.; Ruoff, R. S. Chem. Mater. 1998, 10, 260. (35) Li, J.; Papadopoulos, C.; Xu, J. M.; Moskovits, M. Appl. Phys. Lett. 1999, 75, 367. (36) Klein, J. D.; Herrick Ii, R. D.; Palmer, D.; Sailor, M. J.; Brumlik, C. J.; Martin, C. R. Chem. Mater. 1993, 5, 902. (37) Lee, Y. H.; Leu, I. C.; Liao, C. L.; Chang, S. T.; Wu, M. T.; Yen, J. H.; Fung, K. Z. Electrochem. Solid-State Lett. 2006, 9, 207. (38) Zhou, Y.; Shen, C.; Li, H. Solid State Ionics 2002, 146, 81. (39) Li, N. C.; Patrissi, C. J.; Che, G. L.; Martin, C. R. J. Electrochem. Soc. 2000, 147, 2044. (40) Xu, J. M. Infrared Phys. Technol. 2001, 42, 485. (41) Polyakov, B.; Daly, B.; Prikulis, J.; Lisauskas, V.; Vengalis, B.; Morris, M. A.; Holmes, J. D.; Erts, D. AdV. Mater. 2006, 18, 1812. (42) Nicewarner-Pena, S. R.; Freeman, R. G.; Reiss, B. D.; He, L.; Pena, D. J.; Walton, I. D.; Cromer, R.; Keating, C. D.; Natan, M. J. Science 2001, 294, 137. (43) Hillebrenner, H.; Buyukserin, F.; Kang, M.; Mota, M. O.; Stewart, J. D.; Martin, C. R. J. Am. Chem. Soc. 2006, 128, 4236. (44) Zoski, C. G.; Yang, N.; He, P.; Berdondini, L.; Koudelka-Hep, M. Anal. Chem. 2007, 79, 1474. (45) Hulteen, J. C.; Martin, C. R. J. Mater. Chem. 1997, 7, 1075. (46) Ugo, P.; Pepe, N.; Moretto, L. M.; Battagliarin, M. J. Electroanal. Chem. 2003, 560, 51. (47) Gasparac, R.; Taft, B. J.; Lapierre-Devlin, M. A.; Lazareck, A. D.; Xu, J. M.; Kelley, S. O. J. Am. Chem. Soc. 2004, 126, 12270. (48) Lapierre-Devlin, M. A.; Asher, C. L.; Taft, B. J.; Gasparac, R.; Roberts, M. A.; Kelley, S. O. Nano Lett. 2005, 5, 1051. (49) Lapierre, M. A.; O’Keefe, M.; Taft, B. J.; Kelley, S. O. Anal. Chem. 2003, 75, 6327. (50) Andreu, A.; Merkert, J. W.; Lecaros, L.; Broglin, B. L.; Brazell, J. T.; El-Kouedi, M. Sens. Actuators, B 2006, 114, 1116. (51) Koehne, J.; Li, J.; Cassell, A. M.; Chen, H.; Ye, Q.; Ng, H. T.; Han, J.; Meyyappan, M. J. Mater. Chem. 2004, 14, 676. (52) Koehne, J.; Chen, H.; Li, J.; Cassell, A. M.; Ye, Q.; Ng, H. T.; Han, J.; Meyyappan, M. Nanotechnology 2003, 14, 1239. (53) Szunerits, S.; Tam, J. M.; Thouin, L.; Amatore, C.; Walt, D. R. Anal. Chem. 2003, 75, 4382. (54) Lowe, R. D.; Mani, R. C.; Baldwin, R. P.; Sunkara, M. K. Electrochem. Solid-State Lett. 2006, 9, 43. (55) Willets, K. A.; Van Duyne, R. P. Ann. ReV. Phys. Chem. 2007, 58, 267. (56) Zhao, J.; Zhang, X.; Yonzon, C. R.; Haes, A. J.; Van Duyne, R. P. Nanomedicine 2006, 1, 219. (57) Zhang, X.; Whitney, A. V.; Zhao, J.; Hicks, E. M.; Van Duyne, R. P. J. Nanosci. Nanotechnol. 2006, 6, 1920. (58) Hankus, M. E.; Li, H.; Gibson, G. J.; Cullum, B. M. Anal. Chem. 2006, 78, 7535. (59) Li, J.; Kamata, K.; Watanabe, S.; Iyoda, T. AdV. Mater. 2007, 19, 1267. (60) Haynes, C. L.; Van Duyne, R. P. J. Phys. Chem. B 2001, 105, 5599. (61) Yonzon, C. R.; Stuart, D. A.; Zhang, X.; McFarland, A. D.; Haynes, C. L.; Van Duyne, R. P. Talanta 2005, 67, 438. (62) Vo-Dinh, T. Trends Anal. Chem. 1998, 17, 557. (63) Yonzon, C. R.; Jeoung, E.; Zou, S.; Schatz, G. C.; Mrksich, M.; Van Duyne, R. P. J. Am. Chem. Soc. 2004, 126, 12669. (64) Endo, T.; Kerman, K.; Nagatani, N.; Takamura, Y.; Tamiya, E. Anal. Chem. 2005, 77, 6976.

Very High Density Sensing Arrays (65) Endo, T.; Kerman, K.; Nagatani, N.; Hiepa, H. M.; Kim, D. K.; Yonezawa, Y.; Nakano, K.; Tamiya, E. Anal. Chem. 2006, 78, 6465. (66) Haes, A. J.; Chang, L.; Klein, W. L.; Van Duyne, R. P. J. Am. Chem. Soc. 2005, 127, 2264. (67) Stuart, D. A.; Yuen, J. M.; Shah, N.; Lyandres, O.; Yonzon, C. R.; Glucksberg, M. R.; Walsh, J. T.; Van Duyne, R. P. Anal. Chem. 2006, 78, 7211. (68) Driskell, J. D.; Kwarta, K. M.; Lipert, R. J.; Porter, M. D.; Neill, J. D.; Ridpath, J. F. Anal. Chem. 2005, 77, 6147. (69) Zhang, X.; Young, M. A.; Lyandres, O.; Van Duyne, R. P. J. Am. Chem. Soc. 2005, 127, 4484. (70) Zhang, X.; Zhao, J.; Whitney, A. V.; Elam, J. W.; Van Duyne, R. P. J. Am. Chem. Soc. 2006, 128, 10304. (71) Yan, F.; Vo-Dinh, T. Sens. Actuators, B 2007, 121, 61. (72) Holtz, J. H.; Asher, S. A. Nature 1997, 389, 829. (73) Serpe, M. J.; Kim, J.; Lyon, L. A. AdV. Mater. 2004, 16, 184. (74) Kim, J.; Singh, N.; Lyon, L. A. Angew. Chem., Intl. Ed. 2006, 45, 1446. (75) Petersen, K. E. Proc. IEEE 1982, 70, 420. (76) Salaita, K.; Wang, Y.; Mirkin, C. A. Nat. Nanotechnol. 2007, 2, 145. (77) Ginger, D. S.; Zhang, H.; Mirkin, C. A. Angew. Chem., Int. Ed. 2004, 43, 30. (78) Piner, R. D.; Zhu, J.; Xu, F.; Hong, S.; Mirkin, C. A. Science 1999, 283, 661. (79) Liu, X.; Zhang, Y.; Goswami, D. K.; Okasinski, J. S.; Salaita, K.; Sun, P.; Bedzyk, M. J.; Mirkin, C. A. Science 2005, 307, 1763. (80) Nam, J.-M.; Han, S. W.; Lee, K.-B.; Liu, X.; Ratner, M. A.; Mirkin, C. A. Angew. Chem., Int. Ed. 2004, 43, 1246. (81) Jiang, H.; Stupp, S. I. Langmuir 2005, 21, 5242. (82) Su, M.; Liu, X.; Li, S.-Y.; Dravid, V. P.; Mirkin, C. A. J. Am. Chem. Soc. 2002, 124, 1560. (83) Hong, S.; Zhu, J.; Mirkin, C. A. Science 1999, 286, 523. (84) Lee, K.-B.; Lim, J.-H.; Mirkin, C. A. J. Am. Chem. Soc. 2003, 125, 5588. (85) Lee, K.-B.; Park, S.-J.; Mirkin, C. A.; Smith, J. C.; Mrksich, M. Science 2002, 295, 1702. (86) Lee, K.-B.; Kim, E.-Y.; Mirkin, C. A.; Wolinsky, S. M. Nano Lett. 2004, 4, 1869. (87) Zhang, H.; Li, Z.; Mirkin, C. A. AdV. Mater. 2002, 14, 1472. (88) Park, S.-J.; Taton, T. A.; Mirkin, C. A. Science 2002, 295, 1503. (89) Lenhert, S.; Sun, P.; Wang, Y.; Fuchs, H.; Mirkin, C. A. Small 2007, 3, 71. (90) Salaita, K.; Wang, Y.; Fragala, J.; Vega, R. A.; Liu, C.; Mirkin, C. A. Angew. Chem., Int. Ed. 2006, 45, 7220. (91) Fodor, S. P.; Read, J. L.; Pirrung, M. C.; Stryer, L.; Lu, A. T.; Solas, D. Science 1991, 251, 767. (92) McGall, G. H.; Barone, A. D.; Diggelmann, M.; Fodor, S. P. A.; Gentalen, E.; Ngo, N. J. Am. Chem. Soc. 1997, 119, 5081. (93) Fodor, S. P.; Rava, R. P.; Huang, X. C.; Pease, A. C.; Holmes, C. P.; Adams, C. L. Nature 1993, 364, 555. (94) Matsuzaki, H.; Dong, S.; Loi, H.; Di, X.; Liu, G.; Hubbell, E.; Law, J.; Berntsen, T.; Chadha, M.; Hui, H.; Yang, G.; Kennedy, G. C.; Webster, T. A.; Cawley, S.; Walsh, P. S.; Jones, K. W.; Fodor, S. P. A.; Mei, R. Nat. Meth. 2004, 1, 109. (95) Dufva, M. Biomol. Eng. 2005, 22, 173. (96) Singh-Gasson, S.; Green, R. D.; Yue, Y.; Nelson, C.; Blattner, F.; Sussman, M. R.; Cerrina, F. Nat. Biotechnol. 1999, 17, 974. (97) Nuwaysir, E. F.; Huang, W.; Albert, T. J.; Singh, J.; Nuwaysir, K.; Pitas, A.; Richmond, T.; Gorski, T.; Berg, J. P.; Ballin, J.; McCormick, M.; Norton, J.; Pollock, T.; Sumwalt, T.; Butcher, L.; Porter, D.; Molla, M.; Hall, C.; Blattner, F.; Sussman, M. R.; Wallace, R. L.; Cerrina, F.; Green, R. D. Gen. Res. 2002, 12, 1749. (98) Naiser, T.; Mai, T.; Michel, W.; Ott, A. ReV. Sci. Instrum. 2006, 77, 063711/1. (99) Lipshutz, R. J.; Fodor, S. P. A.; Gingeras, T. R.; Lockhart, D. J. Nat. Gen. 1999, 21, 20. (100) Harrington, C. A.; Rosenow, C.; Retief, J. Curr. Opin. Microbiol. 2000, 3, 285. (101) Lockhart, D. J.; Dong, H.; Byrne, M. C.; Follettie, M. T.; Gallo, M. V.; Chee, M. S.; Mittmann, M.; Wang, C.; Kobayashi, M.; Horton, H.; Brown, E. L. Nat. Biotechnol. 1996, 14, 1675. (102) Wodicka, L.; Dong, H.; Mittmann, M.; Ho, M.-H.; Lockhart, D. J. Nat. Biotechnol. 1997, 15, 1359. (103) Zaharik, M. L.; Nayar, T.; White, R.; Ma, C.; Vallance, B. A.; Straka, N.; Jiang, X.; Rey-Ladino, J.; Shen, C.; Brunham, R. C. Immunology 2007, 120, 160. (104) Dame, T. M.; Orenzoff, B. L.; Palmer, L. E.; Furie, M. B. J. Immunol. 2007, 178, 1172. (105) Takaishi, S.; Wang, T. C. Cancer Sci. 2007, 98, 284. (106) Chang, W. L. W.; Coro, E. S.; Rau, F. C.; Xiao, Y.; Erle, D. J.; Baumgarth, N. J. Immunol. 2007, 178, 1457. (107) Kruglyak, L.; Nickerson, D. A. Nat. Gen. 2001, 27, 234.

Chemical Reviews, 2008, Vol. 108, No. 2 635 (108) Kennedy, G. C.; Matsuzaki, H.; Dong, S.; Liu, W.-m.; Huang, J.; Liu, G.; Su, X.; Cao, M.; Chen, W.; Zhang, J.; Liu, W.; Yang, G.; Di, X.; Ryder, T.; He, Z.; Surti, U.; Phillips, M. S.; Boyce-Jacino, M. T.; Fodor, S. P. A.; Jones, K. W. Nat. Biotechnol. 2003, 21, 1233. (109) Yuan, E.; Haghighi, F.; White, S.; Costa, R.; McMinn, J.; Chun, K.; Minden, M.; Tycko, B. Cancer Res. 2006, 66, 3443. (110) Britta Stordal, G. P. R. D. Genes Chromosomes Cancer 2006, 45, 1094. (111) Cao, X.; Eu, K. W.; Kumarasinghe, M. P.; Li, H. H.; Loi, C.; Cheah, P. Y. J. Med. Genet. 2006, 43, e13. (112) Calhoun, E. S.; Hucl, T.; Gallmeier, E.; West, K. M.; Arking, D. E.; Maitra, A.; Iacobuzio-Donahue, C. A.; Chakravarti, A.; Hruban, R. H.; Kern, S. E. Cancer Res. 2006, 66, 7920. (113) Brown, P. O.; Botstein, D. Nat. Gen. 1999, 21, 33. (114) Lockhart, D. J.; Winzeler, E. A. Nature 2000, 405, 827. (115) Fan, J.-B.; Chee, M. S.; Gunderson, K. L. Nat. ReV. Genet. 2006, 7, 632. (116) Bergveld, P. Sens. Actuators, B 2003, B88, 1. (117) Reinhoudt, D. N.; Engbersen, J. F. J.; Brzozka, Z.; van der Vlekkert, H. H.; Honig, G. W. N.; Holterman, H. A. J.; Verkerk, U. H. Anal. Chem. 1994, 66, 3618. (118) Rochefeuille, S.; Jimenez, C.; Tingry, S.; Seta, P.; Desfours, J. P. Mater. Sci. Eng. C: Biomimetic Supramol. Syst. 2002, C21, 43. (119) Sant, W.; Pourciel, M. L.; Launay, J.; Do Conto, T.; Martinez, A.; Temple-Boyer, P. Sens. Actuators, B 2003, 95, 309. (120) van der Schoot, B. H.; Bergveld, P. Biosensors 1988, 3, 161. (121) Caras, S.; Janata, J. Anal. Chem. 1980, 52, 1935. (122) Han, Y.; Offenhausser, A.; Ingebrandt, S. Surf. Interface Anal. 2006, 38, 176. (123) Uslu, F.; Ingebrandt, S.; Mayer, D.; Bocker-Meffert, S.; Odenthal, M.; Offenhausser, A. Biosens. Bioelectron. 2004, 19, 1723. (124) Yeow, T. C. W.; Haskard, M. R.; Mulcahy, D. E.; Seo, H. I.; Kwon, D. H. Sens. Actuators, B 1997, 44, 434. (125) Allen, B. L.; Kichambare, P. D.; Star, A. AdV. Mater. 2007, 19, 1439. (126) Deleted in proof. (127) Patolsky, F.; Zheng, G.; Lieber, C. M. Anal. Chem. 2006, 78, 4260. (128) Patolsky, F.; Timko, B. P.; Zheng, G.; Lieber, C. M. MRS Bull. 2007, 32, 142. (129) Cui, Y.; Wei, Q.; Park, H.; Lieber, C. M. Science 2001, 293, 1289. (130) Deleted in proof. (131) Deleted in proof. (132) Deleted in proof. (133) Deleted in proof. (134) Carlen, E. T.; van der Berg, A. Lab Chip 2006, 7, 19. (135) Wang, W. U.; Chen, C.; Lin, K.-h.; Fang, Y.; Lieber, C. M. PNAS 2005, 102, 3208. (136) Hahm, J.-i.; Lieber, C. M. Nano Lett. 2004, 4, 51. (137) Deleted in proof. (138) Li, Z.; Chen, Y.; Li, X.; Kamins, T. I.; Nauka, K.; Williams, R. S. Nano Lett. 2004, 4, 245. (139) Li, Z.; Rajendran, B.; Kamins, T. I.; Li, X.; Chen, Y.; Williams, R. S. Appl. Phys. A: Mater. Sci. Process. 2005, 80, 1257. (140) Stern, E.; Klemic, J. F.; Routenberg, D. A.; Wyrembak, P. N.; TurnerEvans, D. B.; Hamilton, A. D.; LaVan, D. A.; Fahmy, T. M.; Reed, M. A. Nature 2007, 445, 519. (141) Bunimovich, Y. L.; Shin, Y. S.; Yeo, W.-S.; Amori, M.; Kwong, G.; Heath, J. R. J. Am. Chem. Soc. 2006, 128, 16323. (142) Wang, D.; Sheriff, B. A.; Heath, J. R. Nano Lett. 2006, 6, 1096. (143) Patolsky, F.; Zheng, G.; Hayden, O.; Lakadamyali, M.; Zhuang, X.; Lieber, C. M. PNAS 2004, 101, 14017. (144) Zheng, G.; Patolsky, F.; Cui, Y.; Wang, W. U.; Lieber, C. M. Nat. Biotechnol. 2005, 23, 1294. (145) Deleted in proof. (146) Steemers, F. J.; Walt, D. R. Mikrochim. Acta 1999, 131, 99. (147) Walt, D. R. Curr. Opin. Chem. Biol. 2002, 6, 689. (148) Epstein, J. R.; Walt, D. R. Chem. Soc. ReV. 2003, 32, 203. (149) Walt, D. R. Science 2000, 287, 451. (150) Epstein, J. R.; Leung, A. P. K.; Lee, K.-H.; Walt, D. R. Biosens. Bioelectron. 2003, 18, 541. (151) Ferguson, J. A.; Steemers, F. J.; Walt, D. R. Anal. Chem. 2000, 72, 5618. (152) Dickinson, T. A.; Michael, K. L.; Kauer, J. S.; Walt, D. R. Anal. Chem. 1999, 71, 2192. (153) Gunderson, K. L.; Kruglyak, S.; Graige, M. S.; Garcia, F.; Kermani, B. G.; Zhao, C.; Che, D.; Dickinson, T.; Wickham, E.; Bierle, J.; Doucet, D.; Milewski, M.; Yang, R.; Siegmund, C.; Haas, J.; Zhou, L.; Oliphant, A.; Fan, J.-B.; Barnard, S.; Chee, M. S. Gen. Res. 2004, 14, 870. (154) Song, L.; Ahn, S.; Walt, D. R. Anal. Chem. 2006, 78, 1023. (155) Epstein, J. R.; Lee, M.; Walt, D. R. Anal. Chem. 2002, 74, 1836. (156) Bowden, M.; Song, L.; Walt, D. R. Anal. Chem. 2005, 77, 5583. (157) Shepard, J. R. E.; Danin-Poleg, Y.; Kashi, Y.; Walt, D. R. Anal. Chem. 2005, 77, 319.

636 Chemical Reviews, 2008, Vol. 108, No. 2 (158) Steemers, F. J.; Ferguson, J. A.; Walt, D. R. Nat. Biotechnol. 2000, 18, 91. (159) Ahn, S.; Walt, D. R. Anal. Chem. 2005, 77, 5041. (160) Ahn, S.; Kulis, D. M.; Erdner, D. L.; Anderson, D. M.; Walt, D. R. Appl. EnViron. Microbiol. 2006, 72, 5742. (161) Epstein, J. R.; Ferguson, J. A.; Lee, K.-H.; Walt, D. R. J. Am. Chem. Soc. 2003, 125, 13753. (162) Kuhn, K.; Baker, S. C.; Chudin, E.; Lieu, M.-H.; Oeser, S.; Bennett, H.; Rigault, P.; Barker, D.; McDaniel, T. K.; Chee, M. S. Gen. Res. 2004, 14, 2347. (163) Fan, J.-B.; Yeakley, J. M.; Bibikova, M.; Chudin, E.; Wickham, E.; Chen, J.; Doucet, D.; Rigault, P.; Zhang, B.; Shen, R.; McBride, C.; Li, H.-R.; Fu, X.-D.; Oliphant, A.; Barker, D. L.; Chee, M. S. Gen. Res. 2004, 14, 878. (164) Gelder, R. N. V.; von Zastrow, M. E.; Yool, A.; Dement, W. C.; Barchas, J. D.; Eberwine, J. H. PNAS 1990, 87, 1663. (165) Shen, R.; Fan, J.-B.; Campbell, D.; Chang, W.; Chen, J.; Doucet, D.; Yeakley, J.; Bibikova, M.; Wickham Garcia, E.; McBride, C.; Steemers, F.; Garcia, F.; Kermani, B. G.; Gunderson, K.; Oliphant, A. Mut. Res. Fund. Mole. Mech. of Mutagen. 2005, 573, 70. (166) Steemers, F. J.; Gunderson, K. L. Biotechnol. J. 2007, 2, 41. (167) Lee, M.; Walt, D. R. Anal. Biochem. 2000, 282, 142. (168) Adams, E. W.; Ueberfeld, J.; Ratner, D. M.; O’Keefe, B. R.; Walt, D. R.; Seeberger, P. H. Angew. Chem., Int. Ed. 2003, 42, 5317. (169) Szurdoki, F.; Michael, K. L.; Walt, D. R. Anal. Biochem. 2001, 291, 219. (170) Rissin, D. M.; Walt, D. R. Anal. Chim. Acta 2006, 564, 34. (171) Rissin, D. M.; Walt, D. R. Nano Lett. 2006, 6, 520. (172) Rissin, D. M.; Walt, D. R. J. Am. Chem. Soc. 2006, 128, 6286. (173) Bigos, M.; Baumgarth, N.; Jager, G. C.; Herman, O. C.; Nozaki, T.; Stovel, R. T.; Parks, D. R.; Herzenberg, L. A. Cytometry 1999, 36, 36. (174) Deutsch, A.; Zurgil, N.; Hurevich, I.; Shafran, Y.; Afrimzon, E.; Lebovich, P.; Deutsch, M. Biomed. MicrodeV. 2006, 8, 361. (175) Taylor, L. C.; Walt, D. R. Anal. Biochem. 2000, 278, 132. (176) Fink, J.; Thery, M.; Azioune, A.; Dupont, R.; Chatelain, F.; Bornens, M.; Piel, M. Lab Chip 2007, 7, 672. (177) Khademhosseini, A.; Langer, R.; Borenstein, J.; Vacanti, J. P. PNAS 2006, 103, 2480. (178) Chen, C. S.; Mrksich, M.; Huang, S.; Whitesides, G. M.; Ingber, D. E. Science 1997, 276, 1425. (179) Ostuni, E.; Chen, C. S.; Ingber, D. E.; Whitesides, G. M. Langmuir 2001, 17, 2828. (180) Kandere-Grzybowska, K.; Campbell, C. J.; Mahmud, G.; Komarova, Y.; Soh, S.; Grzybowski, B. A. Soft Matter 2007, 3, 672. (181) Greve, F.; Lichtenberg, J.; Hierlemann, A. IEEE Sens. 2004, 80. (182) Kuang, Y.; Biran, I.; Walt, D. R. Anal. Chem. 2004, 76, 2902. (183) Biran, I.; Walt, D. R. Anal. Chem. 2002, 74, 3046. (184) Kuang, Y.; Biran, I.; Walt, D. R. Anal. Chem. 2004, 76, 6282. (185) Dragoy Whitaker, R.; Walt, D. R. Anal. Biochem. 2007, 360, 63. (186) Biran, I.; Rissin, D. M.; Ron, E. Z.; Walt, D. R. Anal. Biochem. 2003, 315, 106. (187) DiCesare, C.; Biran, I.; Walt, D. R. Anal. Bioanal. Chem. 2005, 382, 37. (188) Yamamura, S.; Kishi, H.; Tokimitsu, Y.; Kondo, S.; Honda, R.; Rao, S. R.; Omori, M.; Tamiya, E.; Muraguchi, A. Anal. Chem. 2005, 77, 8050. (189) www.molecular-cytomics.com. (190) Albert, K. J.; Lewis, N. S.; Schauer, C. L.; Sotzing, G. A.; Stitzel, S. E.; Vaid, T. P.; Walt, D. R. Chem. ReV. 2000, 100, 2595. (191) Dickinson, T. A.; White, J.; Kauer, J. S.; Walt, D. R. Trends Biotechnol. 1998, 16, 250. (192) Stitzel, S. E.; Stein, D. R.; Walt, D. R. J. Am. Chem. Soc. 2003, 125, 3684. (193) Albert, K. J.; Myrick, M. L.; Brown, S. B.; James, D. L.; Milanovich, F. P.; Walt, D. R. EnViron. Sci. Technol. 2001, 35, 3193. (194) Lundstrom, I.; Shivaraman, S.; Svensson, C.; Lundkvist, L. Appl. Phys. Lett. 1975, 26, 55. (195) Persaud, K.; Dodd, G. Nature 1982, 299, 352. (196) McAlpine, M. C.; Ahmad, H.; Wang, D.; Heath, J. R. Nat. Mater. 2007, 6, 379. (197) Pearce, T. C.; Gardner, J. W.; Friel, S.; Bartlett, P. N.; Blair, N. Analyst 1993, 118, 371. (198) Legin, A. V.; Vlasov, Y. G.; Rudnitskaya, A. M.; Bychkov, E. A. Sens. Actuators, B 1996, B34, 456. (199) Amati, D.; Arn, D.; Blom, N.; Ehrat, M.; Saunois, J.; Widmer, H. M. Sens. Actuators, B 1992, B7, 587. (200) Burns, J. A.; Whitesides, G. M. Chem. ReV. 1993, 93, 2583. (201) Albert, K. J.; Walt, D. R. Anal. Chem. 2003, 75, 4161. (202) Stitzel, S. E.; Cowen, L. J.; Albert, K. J.; Walt, D. R. Anal. Chem. 2001, 73, 5266. (203) Bencic-Nagale, S.; Walt, D. R. Anal. Chem. 2005, 77, 6155. (204) Albert, K. J.; Walt, D. R. Anal. Chem. 2000, 72, 1947.

LaFratta and Walt (205) Albert, K. J.; Walt, D. R.; Gill, D. S.; Pearce, T. C. Anal. Chem. 2001, 73, 2501. (206) Horan, P. K.; Wheeless, L. L., Jr. Science 1977, 198, 149. (207) Vignali, D. A. A. J. Immunol. Methods 2000, 243, 243. (208) Nolan, J. P.; Sklar, L. A. Nat. Biotechnol. 1998, 16, 633. (209) Nolan, J. P.; Sklar, L. A. Trends Biotechnol. 2002, 20, 9. (210) Wilson, R.; Cossins, A. R.; Spiller, D. G. Angew. Chem., Int. Ed. 2006, 45, 6104. (211) Finkel, N. H.; Lou, X.; Wang, C.; He, L. Anal. Chem. 2004, 76, 352A. (212) He, B.; Son, S. J.; Lee, S. B. Anal. Chem. 2007, 79, 5257. (213) He, B.; Son, S. J.; Lee, S. B. Langmuir 2006, 22, 8263. (214) Pregibon, D. C.; Toner, M.; Doyle, P. S. Science 2007, 315, 1393. (215) www.luminexcorp.com. (216) McHugh, T. M.; Stites, D. P.; Casavant, C. H.; Fulwyler, M. J. J. Immunol. Methods 1986, 95, 57. (217) McHugh, T. M.; Miner, R. C.; Logan, L. H.; Stites, D. P. J. Clin. Microbiol. 1988, 26, 1957. (218) Saunders, G. C.; Jett, J. H.; Martin, J. C. Clin. Chem. 1985, 31, 2020. (219) Fulton, R. J.; McDade, R. L.; Smith, P. L.; Kienker, L. J.; Kettman, J. R., Jr. Clin. Chem. 1997, 43, 1749. (220) Carson, R. T.; Vignali, D. A. A. J. Immunol. Methods 1999, 227, 41. (221) Dasso, J.; Lee, J.; Bach, H.; Mage, R. G. J. Immunol. Methods 2002, 263, 23. (222) Yamamoto, K.; Ito, S.; Yasukawa, F.; Konami, Y.; Matsumoto, N. Anal. Biochem. 2005, 336, 28. (223) Goldman, E. R.; Anderson, G. P.; Liu, J. L.; Delehanty, J. B.; Sherwood, L. J.; Osborn, L. E.; Cummins, L. B.; Hayhurst, A. Anal. Chem. 2006, 78, 8245. (224) Anderson, G. P.; Moreira, S. C.; Charles, P. T.; Medintz, I. L.; Goldman, E. R.; Zeinali, M.; Taitt, C. R. Anal. Chem. 2006, 78, 2279. (225) Martinelli, E.; Cicala, C.; Van Ryk, D.; Goode, D. J.; Macleod, K.; Arthos, J.; Fauci, A. S. PNAS 2007, 104, 3396. (226) Kobasa, D.; Jones, S. M.; Shinya, K.; Kash, J. C.; Copps, J.; Ebihara, H.; Hatta, Y.; Kim, J. H.; Halfmann, P.; Hatta, M.; Feldmann, F.; Alimonti, J. B.; Fernando, L.; Li, Y.; Katze, M. G.; Feldmann, H.; Kawaoka, Y. Nature 2007, 445, 319. (227) Yang, L.; Tran, D. K.; Wang, X. Genome Res. 2001, 11, 1888. (228) Lu, J.; Getz, G.; Miska, E. A.; Alvarez-Saavedra, E.; Lamb, J.; Peck, D.; Sweet-Cordero, A.; Ebert, B. L.; Mak, R. H.; Ferrando, A. A.; Downing, J. R.; Jacks, T.; Horvitz, H. R.; Golub, T. R. Nature 2005, 435, 834. (229) Flagella, M.; Bui, S.; Zheng, Z.; Nguyen, C. T.; Zhang, A.; Pastor, L.; Ma, Y.; Yang, W.; Crawford, K. L.; McMaster, G. K.; Witney, F.; Luo, Y. Anal. Biochem. 2006, 352, 50. (230) Armstrong, B.; Stewart, M.; Mazumder, A. Cytometry 2000, 40, 102. (231) Iannone, M. A.; Taylor, J. D.; Chen, J.; Li, M.-S.; Rivers, P.; SlentzKesler, K. A.; Weiner, M. P. Cytometry 2000, 39, 131. (232) Chen, J.; Iannone, M. A.; Li, M.-S.; Taylor, J. D.; Rivers, P.; Nelsen, A. J.; Slentz-Kesler, K. A.; Roses, A.; Weiner, M. P. Genome Res. 2000, 10, 549. (233) Cai, H.; White, P. S.; Torney, D.; Deshpande, A.; Wang, Z.; Marrone, B.; Nolan, J. P. Genomics 2000, 66, 135. (234) Bayley, H.; Martin, C. R. Chem. ReV. 2000, 100, 2575. (235) Wharton, J. E.; Jin, P.; Sexton, L. T.; Horne, L. P.; Sherrill, S. A.; Mino, W. K.; Martin, C. R. Small 2007, 3, 1424. (236) Shao, M.-W.; Shan, Y.-Y.; Wong, N.-B.; Lee, S.-T. AdV. Funct. Mater. 2005, 15, 1478. (237) Star, A.; Tu, E.; Niemann, J.; Gabriel, J.-C. P.; Joiner, C. S.; Valcke, C. PNAS 2006, 103, 921. (238) Sirbuly, D. J.; Tao, A.; Law, M.; Fan, R.; Yang, P. AdV. Mater. 2007, 19, 61. (239) Zhao, X.; Cao, Y.; Ito, F.; Chen, H.-H.; Nagai, K.; Zhao, Y.-H.; Gu, Z.-Z. Angew. Chem., Int. Ed. 2006, 45, 6835. (240) Kuncicky, D. M.; Prevo, B. G.; Velev, O. D. J. Mater. Chem. 2006, 16, 1207. (241) Claridge, S. A.; Goh, S. L.; Frechet, J. M. J.; Williams, S. C.; Micheel, C. M.; Alivisatos, A. P. Chem. Mater. 2005, 17, 1628. (242) Cha, J. N.; Zhang, Y.; Wong, H. S. P.; Raoux, S.; Rettner, C.; Krupp, L.; Deline, V. Chem. Mater. 2007, 19, 839. (243) McFarland, A. D.; Van Duyne, R. P. Nano Lett. 2003, 3, 1057. (244) Wang, Z. L.; Song, J. Science 2006. (245) Angelos, S.; Johansso, E.; Stoddart, J. F.; Zink, J. I. AdV. Funct. Mater. 2007, 17, 2261. (246) Nguyen, T. D.; Liu, Y.; Saha, S.; Leung, K. C.; Stoddart, J. F.; Zink, J. I. J. Am. Chem. Soc. 2007, 129, 626. (247) Stoddart, J. F.; Zink, J. I. AdV. Funct. Mater. 2007, 17, 685. (248) Sidorenko, A.; Krupenkin, T.; Taylor, A.; Fratzl, P.; Aizenberg, J. Science 2007, 315, 487. (249) Gu, L. Q.; Cheley, S.; Bayley, H. Science 2001, 291, 636. (250) Seeman, N. C. Chem. Biol. 2003, 10, 1151. (251) Seeman, N. C. Nature 2003, 421, 427.

Very High Density Sensing Arrays (252) Grier, D. G.; Roichman, Y. Appl. Opt. 2006, 45, 880. (253) Tam, J. M.; Biran, I.; Walt, D. R. Appl. Phys. Lett. 2004, 84, 4289. (254) Willig, K. I.; Rizzoli, S. O.; Westphal, V.; Jahn, R.; Hell, S. W. Nature 2006, 440, 935. (255) Klar, T. A.; Jakobs, S.; Dyba, M.; Egner, A.; Hell, S. W. PNAS 2000, 97, 8206. (256) Rust, M. J.; Bates, M.; Zhuang, X. Nat. Meth. 2006, 3, 793. (257) Bates, M.; Huang, B.; Dempsey, G. T.; Zhuang, X. Science 2007, 317, 1749. (258) Smith, C. Nature 2007, 446, 219. (259) Staples, M.; Daniel, K.; Cima, M. J.; Langer, R. Pharmacol. Res. 2006, 23, 847. (260) Sigworth, F. J.; Klemic, K. G. Biophys. J. 2002. (261) Buxboim, A.; Bar-Dagan, M.; Frydman, V.; Zbaida, D.; Morpurgo, M.; Bar-Ziv, R. Small 2007, 3, 500. (262) Maerkl, S. J.; Quake, S. R. Science 2007, 315, 233. (263) Margulies, M.; Egholm, M.; Altman, W. E.; Attiya, S.; Bader, J. S.; Bemben, L. A.; Berka, J.; Braverman, M. S.; Chen, Y. J.; Chen, Z. Nature 2005, 437, 376.

Chemical Reviews, 2008, Vol. 108, No. 2 637 (264) Bentley, D. R. Curr. Opin. Genet. DeV. 2006, 16, 545. (265) Malek, J. A.; Wierzbowski, J. M.; Tao, W.; Bosak, S. A.; Saranga, D. J.; Doucette-Stamm, L.; Smith, D. R.; McEwan, P. J.; McKernan, K. J. Nucleic Acids Res. 2004, 32, 1059. (266) Harris, T.; Buzby, P. R.; Jarosz, M.; Dimeo, J. J.; Gill, J. (Helicos Biosciences Corp., U.S.A.). WO Patent Application, 2006. (267) Teo, B. K.; Sun, X. H. Chem. ReV. 2007, 107, 1454. (268) Collins, S.; Janata, J. Anal. Chim. Acta 1982, 136, 93. (269) Janata, J.; Blackburn, G. F. Ann. N. Y. Acad. Sci. 1984, 428, 286. (270) Janata, J. Principles of Chemical Sensors; Plenum Press: New York, 1989. (271) Bradley, K.; Gabriel, J.-C. P.; Star, A.; Gruener, G. Appl. Phys. Lett. 2003, 83, 3821. (272) Minot, E. D.; Janssens, A. M.; Heller, I.; Heering, H. A.; Dekker, C.; Lemay, S. G. Appl. Phys. Lett. 2007, 91, 093507/1. (273) Stern, E.; Wagner, R.; Sigworth, F. J.; Breaker, R.; Fahmy, T. M.; Reed, M. A. Nano Lett. 2007, 7, 3405.

CR0681142

Chem. Rev. 2008, 108, 638−651

638

Chemical Sensors with Integrated Electronics Segyeong Joo and Richard B. Brown* College of Engineering, University of Utah, 1692 Warnock Engineering Building, 72 Central Campus Drive, Salt Lake City, Utah 84112 Received August 29, 2007

Contents 1. Introduction 2. Electrochemical Sensors 2.1. Potentiometric Sensors 2.1.1. ChemFETs 2.1.2. ISEs 2.2. Conductometric Sensors 2.2.1. Resistive Sensors 2.2.2. Capacitive Sensors 2.3. Voltammetric Sensors 3. Optical Sensors 4. Mass-Sensitive Sensors 5. Integration of Different Transducers on a Single Chip 6. Conclusions and Outlook 7. Acknowledgments 8. References

638 638 639 639 641 642 642 645 645 646 647 647 649 650 650

1. Introduction Building on the invention of the transistor by Brattain and Bardeen,1 an enormous amount of research in semiconductors and integrated circuits (ICs) has established the modern complementary metal-oxide-semiconductor (CMOS) technology. The combination of accumulated techniques in semiconductor processing technology and the good mechanical properties of silicon led to the development of microelectromechanical systems (MEMS) in the early 1970s.2 MEMS technology produced a variety of micrometer-sized mechanical structures, such as overhanging, suspended, and freestanding elements fabricated with silicon processing techniques2-4 that have been utilized in sensors and actuators such as accelerometers, pressure sensors, and micromirrors for video projection systems. Silicon-based liquid chemical sensors, the first of which were reported in 1970,5 are considered a branch of the MEMS family tree, though they often do not include any micromechanical components. Solid-state liquid chemical sensors can also be fabricated on glass, ceramic, or plastic substrates. Use of a semiconducting substrate, such as silicon, opens the possibility of integrating electronics with the chemical sensors (or other MEMS devices) on a single silicon chip to reduce noise, improve manufacturing control, reduce size, and enable batch fabrication of the whole system. Chemical sensors present a particularly challenging application for integrated electronics. Integrated circuits are * To whom correspondence should be addressed. Phone: 801-585-7498. Fax: 801-581-6892. E-mail: [email protected].

generally packaged in such a way as to protect the circuit from any liquid, and even from humidity, but liquid chemical sensor chips must be in contact with the solution under test. Certain restrictions are placed on the materials and processing steps for chemical sensors if they are to incorporate CMOS, the dominant electronics technology. If contaminated with alkali or other ions, CMOS suffers from drifting transistor threshold voltages, so the manufacturing process must be cleaner than it would need to be to fabricate just the sensor. High temperatures or high electric fields during the manufacturing process can also destroy the circuits.6,7 There are materials issues to consider; in addition, the metals used in some chemical sensors are not compatible with the interconnect metals used in CMOS. However, with the use of barrier layers it is possible to fabricate a sensor on top of a fully passivated CMOS circuit or to connect a separately fabricated CMOS circuit to a sensor chip. Combining CMOS and MEMS technologies, one can fabricate a sensor system that incorporates the sensor, amplifier, signal processing, analogto-digital converter, and microcontroller on a chip. Such a system-on-a-chip (SoC) provides computerized control of the sensor, less noise from the interconnection wires between sensor and data acquisition system, low power operation, less reagent use, and high reproducibility. On the other hand, if a sensor’s lifetime is relatively short, as is the case with many chemical sensors, the additional cost for including electronics may overwhelm these advantages. In this review, enough background information is provided to put each sensor type into perspective, but because an excellent survey of the literature on sensors with integrated CMOS electronics was done in 2002,8,9 our focus is on papers published since then. The current review does not cover sensors having a single metal-oxide-semiconductor (MOS) transistor or hybrid systems in which separately fabricated chipssgenerally one sensor chip and one CMOS circuit chipsare connected by wire-bonding or other means. The review is organized by the operating principle of the sensor system. Section 2 describes integrated electrochemical sensors, including potentiometric, conductometric, and voltammetric sensors. Sections 3 and 4 cover optical- and masssensitive sensors, respectively. Section 5 surveys systems that integrate several sensors having different operating principles on a single chip. Section 6 offers concluding comments and a brief outlook for CMOS-integrated sensors.

2. Electrochemical Sensors Electrochemical sensors are the oldest and most widely available group in the solid-state chemical sensor field. Many solid-state electrochemical sensors have been commercialized, such as glucose monitors for diabetes and ion sensors for blood electrolytes. Electrochemical sensors detect charge

10.1021/cr068113+ CCC: $71.00 © 2008 American Chemical Society Published on Web 01/10/2008

Chemical Sensors with Integrated Electronics

Segyeong Joo was born in 1976 in Busan, Korea. He received his B.S. degree in 1999 with a major in Electrical Engineering and a minor in Chemistry and his M.S. degree in 2001 in Electrical Engineering from the Korea Advanced Institute of Science and Technology (KAIST). In 2006, he received his Ph.D. degree in Biomedical Engineering from Seoul National University (SNU) on developing a rapid field-free electroosmotic micropump incorporating charged microchannel surfaces. He worked on the instrumentation of a medical analysis system and developing a microfluidic component for BioMEMS in the Medical Research Center at SNU Hospital as a research engineer from 2004 to 2006. He is currently a postdoctoral fellow in the College of Engineering at the University of Utah. His fields of interest are a chemical/biosensor system with integrated electronics and its medical instrumentation.

Chemical Reviews, 2008, Vol. 108, No. 2 639

perform these functions can be advantageously integrated with the sensor electrodes. Electrochemical sensors are classified as potentiometric, conductometric, or voltammetric sensors based upon their analytical principles of operation.10,13 Potentiometric sensors measure an equilibrium potential difference between a sensing electrode and a reference electrode. Ideally, there is no current flow through the electrodes at equilibrium. In general, the potential difference shows a linear relationship with the logarithm of the activity of the analyte, as in the Nernst equation. Conductometric sensors quantitate the changes of electrical properties between two electrodes. Resistive sensors measure the resistivity change due to chemical reactions, while capacitive sensors detect the capacitance change due to a dielectric-constant modification. Voltammetric sensors measure the current from the charge transport of an electrochemical reaction on a sensing (working) electrode when a varying potential or a constant potential (amperometric detection) is applied between the working electrode and the solution. Voltammetric sensors use an auxiliary (counter) electrode to control the solution potential and as an electron source or sink for the counter reaction to the one at the working electrode. For more stable analysis, most voltammetric sensors also use a nonpolarizable reference electrode to monitor the solution potential.12

2.1. Potentiometric Sensors

Richard B. Brown received his B.S. (with highest honors) and M.S. degrees in Electrical Engineering from Brigham Young University in 1976. Following graduation, he designed computers, analog-to-digital converters, and computer peripheral devices for companies in California and Missouri. In 1985 he received his Ph.D. degree in Electrical Engineering from the University of Utah and joined the faculty of the University of Michigan’s Department of Electrical Engineering and Computer Science. He has conducted major research projects in the development of sensors, circuits, and microprocessors. At Michigan he developed the highly respected integrated circuit design (VLSI) program and served as Associate Chair and Interim Chair of EECS. He holds 15 patents and has consulted in the areas of solid-state sensor and microprocessor design. He is a founder of Sensicore, i-SENS, and Mobius Microsystems. He was honored to receive the second ECE Distinguished Young Alumnus Award from the University of Utah in 2003 and was appointed the eleventh Dean of the College of Engineering at the University of Utah in July 2004.

transport between chemical phases or changes of electrical properties that arise due to chemical reactions on the sensor.10-12 These electrochemical sensors require specialized electrodes, depending upon the sensing mechanism, that come in contact with the solution under test and generate the signal that is correlated with the concentration of the target analyte. In every case, the detected signal also requires amplification or signal processing. The circuitry required to

Many papers describing the integration of CMOS circuits with potentiometric sensors are based on chemically sensitive field-effect transistors (ChemFETs). The structure of a ChemFET is the same as that of the normal metal-oxidesemiconductor field-effect transistor (MOSFET) except for the transistor gate, which incorporates the means of transduction from a chemical concentration to a voltage. Another popular type of integrated potentiometric sensor is based upon the ion-selective electrode (ISE), a relatively simple sensor that is widely used in the detection of ions in aqueous solutions. The miniaturized, solid-state versions of these sensors mimic the internal filling solutions of ISEs with hydrogel layers between the selective membrane and the electrode. An even simpler version that is called a solidcontact ISE eliminates the hydrogel layer, resulting in a structure that is more like a coated wire. Light-addressable potentiometric sensors (LAPS)14 also employ CMOS-like fabrication techniques, but integration of CMOS electronic circuits with LAPS has not yet been reported. In this section, potentiometric sensor systems that include integrated electronic circuits are surveyed.

2.1.1. ChemFETs FET-based potentiometric sensors were introduced in 1970.5 Figure 1 shows the general structure of an n-channel MOSFET and an ion-sensitive field-effect transistor (ISFET), the most common type of ChemFET. In place of the gate and gate oxide of a general MOSFET, the ISFET has ionic solution with a reference electrode immersed in the solution and an insulating layer appropriate for detecting a specific analyte. More than 650 papers describing the development of ChemFET sensors have been published.15 Many of these papers address structural issues, selection of the insulation/ detection layer, and improvement of sensing accuracy. ChemFETs fabricated on a semiconducting substrate have the possibility of being integrated with electronic circuits

640 Chemical Reviews, 2008, Vol. 108, No. 2

Joo and Brown

Figure 1. Schematic diagrams of general (a) n-channel MOSFET and (b) ISFET. VGS is the gate-source voltage, and VDS is the drainsource voltage.

Figure 2. Photograph of diode-temperature-compensated pH ISFET sensor. Reprinted with permission from ref 21. Copyright 2001 Elsevier.

on the same chip, and thus in the 1980s, several papers reported the fabrication of ChemFETs with integrated circuits.16-18 Furthermore, Bausells et al.19 demonstrated ISFETs fabricated in a commercial CMOS foundry process, which is seen as a practical path toward commercialization. The encapsulation challenges of integrating active electronic circuits with chemical sensors still requires special attention, and standard CMOS processes allow little opportunity for engineering of the gate material or sensor structure. In an effort to compensate for the intrinsic thermal instability of ISFETs,20 Chin et al.21 integrated a temperature sensor and readout circuit (see Figure 2) with a pH-sensitive ISFET. A simple p-n diode was used for temperature detection. Integrating this temperature sensor reduced the temperature coefficient of the output voltage, which is normally between 1 and 12 mV/°C depending on the measuring conditions, to 0.16 mV/°C. Morgenshtein et al.22 showed an interface circuit for body effect elimination and temperature compensation using a complementary ISFET/ MOSFET pair technique. They also introduced the Wheatstone-bridge readout interface for ISFETs23 to improve noise performance and temperature compensation. Yang et al.24

Figure 3. (a) Schematic view of Lundstrom-FET/suspended gate FET double sensor, and (b) photograph of the fabricated chip, where the solid-lined rectangles on the center and right surround the suspended gate FET region, the solid-lined rectangle on the left side surrounds the Lundstrum-FET region, the dashed rectangle in the center indicates the suspended gate area, and the small dottedline rectangles indicate integrated circuits. Reprinted with permission from ref 27. Copyright 2005 Elsevier.

presented a pH-ISFET integrated with an on-chip differential measurement circuit; the sensitivity was 53.67 mV/pH. It is common to modify the gate structure of ChemFETs to improve the performance or measure specific chemicals. Integration of readout circuitry for a specifically modified ChemFET is reported in several papers.25-27 An ISFET pH sensor with a discrete tin dioxide/aluminum (SnO2/Al) gate structure and an integrated readout circuit showed a linear sensitivity of 58 mV/pH within the pH range of 2-10.25 A differential read-out architecture was used for the extended gate approach to reduce leakage and drift.26 A hydrogen sensor was implemented with a suspended gate FET28 and a Lundstrom-FET29 integrated on a same chip with appropriate circuits.27 Figure 3 is the schematic diagram of this sensor and a photograph of the fabricated chip.27 The suspended gate FET has a gas-sensitive membrane separated a specific distance from the active FET layer, while in the LundstromFET, the gas-sensitive layer is deposited directly on the transistor gate insulator.30 Another application for integrated electronics on ChemFET chips is for read-out of an array of sensors. Milgrow et al.31 presented a scalable 2 × 2 array of pH-sensitive ISFETs with a memory-like (row and column) decoder for the sensor

Chemical Sensors with Integrated Electronics

Chemical Reviews, 2008, Vol. 108, No. 2 641

Figure 4. (a) Schematic diagram of a single pixel of the pH image sensor for charge storing, transfer, and readout, and (b) photograph of a fabricated pH image sensor array and enlarged view of a single pixel. Reprinted with permission from ref 33. Copyright 2006 Elsevier.

array and an on-chip analog-to-digital converter (ADC). This group extended their work to a 16 × 16 array of ISFETs the following year for direct extracellular imaging.32 Hizawa et al.33 also developed a two-dimensional pH image sensor chip with a 10 × 10 pH-ISFET array. They measured the amount of charge which is stored in the potential well under the sensing region since the depth of this potential well is determined by the pH of the sample. Figure 4 shows the schematic diagram of the pH image sensor and a photograph of the fabricated sensor array.33 By incorporating chargetransfer readout circuits, the chip was able to produce pH images at up to 30 frames per second,34 a rate that can be used for two-dimensional monitoring of chemical reactions. An example of the system-on-a-chip (SoC) concept with chemical sensors is the digital pH meter introduced by Professor Cummings’ group.35,36 These systems use an ISFET with a floating gate as the transducer and include a microcontroller unit (MCU), static random access memory (SRAM) for program and data, ADC, and programmable voltage reference on a single chip. Figure 5 shows the structure of the floating electrode CMOS ISFET and the fabricated system.35,36 The chips had sensitivities of 43 mV/pH in ref 35 and 48 mV/pH in ref 36.

As mentioned above, one of the challenges of integrating electronics with ChemFETs is that the circuitry must be protected from the ion-containing solution because the ions can migrate in silicon dioxide and cause shifts in transistor characteristics. Most sensors fabricated in CMOS processes employ the standard silicon nitride passivation layer as a barrier to ion migration and also as an ion-sensitive layer.19 A comparison of characteristics of several ChemFET-based sensors with integrated circuitry is presented in Table 1.

2.1.2. ISEs The history of ISEs is much longer than that of ISFETs. The first ISE, the glass electrode,37,38 was invented more than 100 years ago. Since then, many other types of ion-selective membranes have been developed and characterized, including polymeric membranes having various ionophores, which extend the sensing capabilities of ISEs to a wide range of ions.39,40 Electrochemical analysis systems based on ISEs are widely used to measure ionic concentrations in biomedical, food processing, water quality, and pollution monitoring applications.41 The ISE structure was miniaturized using thin-film (photolithographic), thick-film (screen printing), and/or automatic dispensing methods to pattern metals,

642 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 5. (a) Schematic diagram of the cross-section through a p-type floating-electrode CMOS ISFET. (b) Photograph of the integrated SoC pH sensor. Reprinted with permission from (a) ref 36 and (b) ref 35. Copyright (a) 2005 and (b) 2004 IEEE. (b)

insulators, and chemical-selective membranes on semiconductor, ceramic, or plastic substrates. Some of the earliest chemical sensors that were integrated with more extensive electronic circuits were miniaturized ISEs fabricated on silicon.42,43 The line dividing ChemFETs from ISEs having integrated electronics is thin; the ISEs put a conductor between their chemically selective membrane and the transistor gate (like an extended gate ISFET) and replace the single transistor of the ChemFET with an operational amplifier configured as a voltage follower. The active ISE configuration significantly improves thermal stability and photoinduced junction currents. Miniaturized ISEs typically use polymeric membranes as the sensing element, but these have been used in many ChemFETs as well. The membranes are composed of a polymer, ionophore, and usually a plasticizer and lipophilic additives. These components are typically dissolved in a solvent and dispensed onto the sensor surface, where a membrane forms as the solvent evaporates.44 Membrane deposition must be done after the CMOS processing is completed. While passive ISEs tend to be slower responding to chemical changes than ChemFETs, active miniaturized ISEs recover the speed. Multiple-ion sensing chips can be made by depositing several different ion-selective membranes on a single chip. Membrane lifetime is an issue for integrated ISE sensors. The primary causes of membrane failure are detachment of the membrane from the surface, causing an electrolyte shunt

Joo and Brown

around the membrane,45 or loss of plasticizer, carrier, or ionophore from the polymeric film due to leaching into the sample solution.40 The integration of circuitry with a sensor that has a short lifetime is not cost effective, so lengthening the lifetime of ISEs is a priority. This might be accomplished by modifying the composition of the ISE membrane to slow the leaching process, developing a practical method for replacing ISE membranes after they lose sensitivity, or forming an array of ISE membranes on the sensor chip with a method for sequentially exposing a fresh membrane to the solution when the old membrane begins to lose sensitivity. There are fewer papers about the integration of CMOS circuits with ISE sensors than with ChemFETs, but the published papers in this field show very notable results. Integration of an on-chip CMOS buffer with a silicone rubber-based ISE showed a 7.5 times faster response time than that of a conventional ISE that was 225 times larger.46 In addition, these ISEs survived more than 120 days with a response within 5% of ideal and a detection limit of 10 µM. Figure 6 shows the schematic view of the cross-section of an ISE sensor and scanning electron micrographs of an integrated ISE sensor.46 The ISEs were postprocessed on the foundry-fabricated CMOS circuit. In 2005, this group improved the detection limit by 25 times and shortened the response time by 200 times.47 Depositing the ISE membranes onto active ISE chips is challenging because the membrane sites are usually made quite small in order to keep the chip size small since the CMOS wafers cost the same whether they contain few or many sensor chips. Use of a microdispenser and thin wells to define the membrane boundaries facilitates the deposition of very small volumes of membrane cocktail and enables the repeatable formation of small, uniformly thick membranes.46,47

2.2. Conductometric Sensors Conductometric sensors measure the impedance change between two electrodes before and after analyte exposure. In general, conductometric sensors consist of two electrodes with a sensitive layer between them.10,13 If the resistance of the sensitive layer changes when it reacts with the analyte, the structure is a chemoresistor. If the capacitance changes upon exposure to the analyte, the device is a chemocapacitor. In this section, various conductometric sensor systems which are integrated with CMOS circuits will be surveyed.

2.2.1. Resistive Sensors Most of CMOS-integrated resistive sensors are gas sensors based on a metal oxide sensing layer. Various metal oxides have been used for these applications,48,49 and the deposition methods of many different metal oxides for CMOS integration are well known.50 The conductivity change due to reaction of the metal oxide layer with the gas is a measure of the concentration of the gas. In general, when the gas takes electrons from the surface of a metal oxide (oxidizing gas), the conductance of the metal oxide layer decreases, and when the gas gives electrons to the metal oxide (reducing gas), the conductance of the metal oxide layer increases. To activate these reactions, high temperature (>200 °C) is required, and thus, CMOS metal oxide gas sensors must have microhotplates to increase the temperature of the sensitive metal oxide layer. The paper by Barsan et al.51 presents a complex conduction model of metal oxide gas sensors.

Chemical Sensors with Integrated Electronics

Chemical Reviews, 2008, Vol. 108, No. 2 643

Table 1. Characteristics of Several ChemFET Sensors with Integrated Circuits target

detection range

sensitivity

Yang et al.24

H+

pH 4-12.5

53.7 mV/pH

amplifier

Chin et al.25

H+

pH 2-10

58 mV/pH

amplifier

Milgrew et al.32

H+

pH 3-7.4

46 mV/pH

amplifier, decoder, ADC

Hizawa et al.33

H+

pH 4-9.1

229 mV/pH

charge-transfer, source follower

5 mm, 1-metal, 1-poly

Hammond et al.36

H+

7 pH unit

48 mV/pH

MCU, SRAM, ADC

0.6 mm, 3-metal (Austria Micro Systems)

authors

a

functions of integrated circuits

CMOS process detail (Foundry Company) 0.35 mm, 4-metal, 2-poly (Charterd) 0.5 mm, 2-metal, 2-poly (UMC) 0.35 mm, 3-metal, 2-poly (Austria Micro Systems)

area of chip

power consumptiona

5 mm2

N/A (Vdd ) 3.3 V)

3.24 mm2

N/A

12.3 mm2 (16 × 16 array) 26 mm2 (10 × 10 array) 15 mm2

∼60 mW (Vdd ) 3.3 V) N/A ∼30 mW (Vdd ) 3.3 V)

N/A indicates the information was not available in the reference.

Figure 6. (a) Cross-section of an ISE sensor, and (b) SEM images of four ISEs and an active ISE. Reprinted with permission from ref 46. Copyright 2004 Transducers Research Foundation.

Useful electronics to integrate with a metal oxide gas sensor include a readout circuit for the sensor, a temperature monitoring and control circuit, a heater driving circuit, and an interface circuit for the output connection. A system developed by Graf et al.52 is an example of such a monolithic metal oxide gas sensor for carbon monoxide (CO). This sensor used tin oxide (SnO2) as the gas-sensitive material and integrated a polysilicon heater, polysilicon temperature sensor, and platinum electrodes on the sensing region. Figure 7 is a picture of the microhotplate and the complete system.52 This research group has demonstrated several similar sensors for CO detection.53-56 Another application of electronics for resistive sensors is in controlling arrays of metal oxide sensors. Afridi et al.57 fabricated four microhotplate gas sensors on a chip with a polysilicon heater and two gas-sensitive materials: SnO2 and titanium oxide (TiO2). They also integrated op-amps, MOSFET switches, a decoder for selecting the gas sensors,

and bipolar junction transistors (BJTs) for switching the heaters. The sensor was tested using hydrogen, carbon monoxide, and methanol. The same group extended their work to the integration of feedback control for the heater circuits, including an analog-to-digital converter (ADC), digital gain control amplifier, and digital-to-analog converter (DAC).58 Bota et al.59 also introduced a chip with four SnO2 gas sensors with a heater controller circuit that used pulse-width modulation (PWM). Barrettino et al.60 extended their sensor system of ref 52 to an array of three sensors having logarithmic conversion and offset correction in their readout electronics and a proportional integral derivative (PID) controller for each of the sensors in the array. This group also introduced an array of three SnO2-based gas sensors for not only CO but also methane gas.61 Guo et al.62 introduced a 4 × 4 array of SnO2 gas sensors with on-chip multiplexing and readout circuits for measuring methane (CH4), hydrogen (H2), ethanol, and CO. The base of the microhotplate (MHP) was

644 Chemical Reviews, 2008, Vol. 108, No. 2

Joo and Brown

Figure 7. Photograph of (a) the microhotplate with heater, temperature sensor, and sensing electrodes and (b) the integrated sensor chip with microhotplate sensor and CMOS circuitry for signal acquisition and processing. Reprinted with permission from ref 52. Copyright 2004 American Chemical Society.

made with a multilayer stack consisting of 800 nm densified low-temperature oxide (LTO), 1 µm low-stress silicon nitride, and 2 µm LTO. Figure 8 shows the structure and SEM image of this sensor and fabricated sensor array chip.62 Fabrication of metal oxide resistive sensors requires bulk micromachining (etching of the silicon substrate) after fabrication of the CMOS circuits. Since the thermal conductance of silicon is high, the sensor structure must be thermally isolated from the other components on the chip so that the circuits can be cool while the sensing site is hot. This is accomplished by etching the substrate to suspend the sensor structure, which is attached to the rest of the chip by thin tethers of material. Resistive sensors based on cantilever deflection also require silicon substrate etching to form the cantilever structure. When a microcantilever is coated with a polymer layer that expands when it absorbs a specific analyte, the analyteinduced stress of the polymer causes deformation (bending) of the cantilever.63 Zimmermann et al.64 applied this phenomenon to detect ethanol and humidity. Their system has a polymer-coated microcantilever with integrated piezoresistors, an ADC, and a serial digital interface. With the analyte-induced resistance change being detected with an onchip Wheatstone bridge, the system showed sensitivities of 6 nV/ppm/V for ethanol and 5 µV/%RH/V for humidity. Another type of resistive sensor, made of carbon black polymer, has been shown previously to be able to detect

Figure 8. (a) Structure and SEM picture of the sensor element, and (b) photograph of a fabricated chip with a 4 × 4 gas sensor array and on-chip decoders, switches, and differential readout circuit (DRC). Reprinted with permission from ref 62. Copyright 2007 Elsevier.

various organic solvent vapors.65 Unlike the metal oxide gas sensors, these devices operate at room temperature. Recent papers have described carbon black polymer as a sensing material for the development of an olfaction chip or electronic nose.66-68 These papers integrate multiple polymer sensors with CMOS circuits for signal processing and classification of analytes. After training with known analytes, the sensor developed by Tang et al.66,67 was able to distinguish eight odors with three polymer sensors using integrated learning and classifying circuits. Koickal et al.68 showed an olfaction chip with five different polymer sensors and an on-chip spike-time-dependent learning circuit. Dai et al.69 described a resistive humidity sensor with integrated microheater and amplifier circuits. They used nanowire tungsten trioxide (WO3) as a humidity-sensitive material. The integrated microheater controlled the operating temperature up to 75 °C, and the operational amplifiers were

Chemical Sensors with Integrated Electronics

Chemical Reviews, 2008, Vol. 108, No. 2 645

Table 2. Characteristics of Several Resistive Sensors with Integrated Circuitsa

authors Graf et al.52

Guo et al.62

Zimmermann et al.64 Koickal et al.68 Dai et al.69 a

detection method SnO2

target

sensitivity

CO

functions of integrated circuits

N/A (detection temperature limit ) control, ADC, (0.1 ppm) DAC, digital interface SnO2 ethanol high readout circuit, CH4 high decoder H2 medium CO low polymer-coated ethanol 6 nV/ppm/V amplifier, ADC, cantilever with humidity 5 µV/%RH/V DAC, digital piezoresistor interface carbon black ethanol 0.00012%/pm signal processing, polymer toluene 0.00644%/pm on-chip learning nanowire WO3 humidity 4.5 mV/%RH amplifier

CMOS process detail (Foundry Company)

area of chip

power consumption ∼100 mW (Vdd ) 5.5 V)

0.8 µm, 2-metal, 2-poly (Austria Micro Systems)

17 mm2

5 µm, 1-metal, 1-poly

∼15.5 mW 3.24 mm2 (4 × 4 array)

0.8 µm, 2-metal, 2-poly

14 mm2 (2 sensors)

∼50 mW (Vdd ) 5 V)

0.6 µm (Austria Micro Systems) 0.35 µm (TSMC)

50 mm2 (70 sensors) 15 mm2

N/A (Vdd ) 5 V) ∼30 mW (Vdd ) 3.3 V)

N/A indicates the information was not available in the reference.

used for signal processing. Chow et al.70 integrated multiwalled carbon nanotubes (MWCNT) with CMOS circuits for flow and chemical vapor sensing. They formed MWCNT on a foundry fabricated CMOS chip. A circuit for current control and an ADC were designed to measure the resistance change of the MWCNT due to flow variation or a chemical vapor initiated reaction on the surface of the MWCNT. The characteristics of various resistive sensors with integrated circuits are summarized in Table 2.

2.2.2. Capacitive Sensors Humidity sensors are the prototypical application for CMOS integrated capacitive sensors since the high dielectric constant of water (78.4 at 25 °C, liquid phase)71 causes a large capacitance change when water is absorbed into a membrane placed between electrodes. The most common type of capacitive humidity sensor uses a polyimide membrane as the sensing material and interdigitated electrodes for measuring the capacitance.72,73 Qui et al.72 integrated signal processing and a calibration circuit to form a capacitive humidity sensor on a chip. Dai73 used an integrated ring oscillator to convert the capacitance change to an oscillator frequency change. Kummer et al.74 demonstrated configurable electrodes on a capacitive gas sensor. Two polymers, poly(etherurethane) and poly(dimethylsiloxane), used as gas-sensitive layers, were coated on interdigitated electrodes. On-chip integrated circuits were used for control of the configurable electrodes and signal acquisition. This sensor system was able to detect low concentrations of volatile organic compounds (n-octane and toluene) in humid air. Stagni et al.75 developed a label-free deoxyribonucleic acid (DNA) sensor array utilizing the capacitance change of probe DNA-coated interdigitated gold electrodes when hybridization of the prove DNA and target DNA occurs. They integrated circuits for signal processing of each of 128 sensors in the array and an ADC for data conversion on the same chip, showing the possibility of label-free DNA detection.

must be formed after fabrication of any CMOS circuits. The reference electrode also presents challenges for miniaturized electrochemical sensors. A small, stable, long-lived reference electrode is needed. Silver/silver chloride (Ag/AgCl) has been used as a reference electrode in many electrochemical applications, but it maintains a fixed potential only when the chloride concentration is fixed, and its lifetime is limited because silver chloride can be dissolved in aqueous solutions. A key component to be integrated with voltammetric sensors is the potentiostat, which applies the potential between electrodes and measures the current from the resulting reaction. Many papers report the development of CMOS potentiostats for general electrochemical sensors or a specific application.76-85 Zhang et al.86 developed an electrochemical sensor array with an integrated potentiostat for monitoring electrochemical reactions using cyclic voltammetry. The system has a 3 × 3 array of gold working electrodes, a gold counter electrode, and a Ag/AgCl reference electrode on top of CMOS rangeprogrammable amperometric readout amplifier circuits for cyclic voltammetry, which can measure currents in the range from 10 pA to 10 µA.

2.3. Voltammetric Sensors

Martin et al.87 introduced a microinstrument for trace detection of heavy metals that was designed as an in situ rainwater analyzer. For heavy metal detection, they used subtractive anodic stripping voltammetrs since this method reduces the signal from interfering analytes such as dissolved oxygen. Amplifiers, a pseudo-differential potentiostat, and data conversion circuits were integrated for signal acquisition and processing. The sensor, having a gold working electrode, platinum counter electrode, and Ag/AgCl reference electrode, was fabricated on top of the CMOS circuits. The system detected 0.3 ppb of lead. The passivation layer for protecting the circuit from solution uptake can endure more than 100 days in saturated salt solution. However, the integrated Ag/AgCl reference electrode failed after 10 h in solution due to poor adhesion.

In voltammetry, the electrodes, which are typically made of gold, silver, platinum, palladium, carbon, or graphite, are critical to the operation of the sensor. Deposition of these metals is not CMOS compatible, and thus, the electrodes

Several recent publications describe on-chip electrical DNA detection utilizing a chronocoulometric detection method by Dr. Thewes’ group.88-91 These detection systems measure the redox-cycling current at the electrodes in the

646 Chemical Reviews, 2008, Vol. 108, No. 2

Joo and Brown

Figure 10. Photograph of luminescence detection SoC with pixel array, ADC, and DSP SIMD array. Reprinted with permission from ref 97. Copyright 2006 IEEE.

Figure 9. (a) Schematic plot of the electrode configuration, schematic illustration of the redox-cycling process, and (b) chip microphotograph. Reprinted with permission from ref 90. Copyright 2004 IEEE.

time domain by applying an appropriate potential when the labeled target DNA is hybridized with the probe DNA. If the probe DNA is not hybridized with the target DNA, there is no electron exchange on the electrodes. The CMOS potentiostat89 and in-pixel ADC were monolithically integrated.90 Figure 9 shows the electrode configuration, an illustration of the redox-cycling detection, and a photograph of the fabricated chip, which has a 16 × 8 sensor array, along with circuits for calibration current generation, a potentiostat, and decoders.90

3. Optical Sensors Optical-based chemical sensors detect the intensity of photon radiation that arrives at a sensor. The intensity can be modulated by absorption, or the direction of wave propagation or wavelength can be modulated by scattering, refraction, or reflection. The detected photons could be emitted by fluorescence, phosphorescence, or chemi-/bioluminescence. Each of these photon sources has a known relationship to a given chemical reaction, facilitating chemical sensing with high selectivity. Fiber-optic sensors92 are the most common type of optical sensors, but they are not suitable for CMOS integration (even with on-chip waveguides) because the indirect band gap of silicon6 makes it difficult to generate the light source. Direct band gap semiconductors such as gallium arsenide (GaAs) and indium phosphate (InP) (III-V semiconductors) are

suitable for photon generation and detection, but these technologies are more expensive than silicon and have not been used for integrating more complex electronics with optical sensors. Light detection, on the other hand, can be done with silicon-based devices using a photodiode or phototransistor. Image sensors fabricated in CMOS technology are expanding their share of the image sensor market because, unlike charge-coupled devices (CCDs), they can incorporate pixel-level analog and digital processing circuits.93 This capability is also advantageous for optical chemical sensors utilizing CMOS photodetectors. Fluorescence- and phosphorescence-based sensors require bulky excitation sources, so they are not as compatible with miniaturized solid-state photosensors and electronics, but many papers report bioluminescence detectors having integrated CMOS circuitry.94-98 Simpson et al.94 showed a bioluminescent bioreporter integrated circuit (BBIC) having integrated signal processing circuits; the sensor was able to detect 4 × 105 cells/mL using the Pseudomonas fluorescens 5RL bacterial cell as a bioreporter. The following year, this group modified their signal processing circuit to reduce leakage current in the photodiode and lower the detection limit.95 Vijarayaghavan et al.98 also showed a BBIC for liquid- and air-based detection of salicylate and naphthalene using the P. fluorescens 5RL bacterial cell as the bioreporter. Iordanov et al.96 developed a 4 × 5 photodiode array with front-end electronic circuits for bioluminescence detection and a 5 × 5 photodiode array for fluorescence detection. Luciferin and luciferase were mixed with different concentrations of adenosine triphosphate (ATP), and the decay of luminescence intensity was monitored over time. For fluorescence detection, an enzyme of Protein Tyrosine Phosphate was mixed with 6,8-difluoro-4-methylumbelliferyl phosphate (DiFMUP) since the reaction product 6,8-difluoro-4-methylumbelliferone (DiFMU) exhibits an excitation/emission maxima of ∼358/455 nm. The fluorescence intensity increased with time when the mixture was exposed to an ultraviolet (UV) light source since the concentration of product (DiFMU) increased as the reaction progressed. A system-on-a-chip (SoC) approach for optical sensors was demonstrated by Eltoukhy et al.97 The chip included a CMOS bioluminescence detection sensor with an 8 × 16 photodiode array, a 128-channel 13-bit ADC, and a column-level singleinstruction multiple-data (SIMD) digital signal processing (DSP) circuit. Figure 10 shows the fabricated bioluminescence chip.97 A mixture of luciferin and luciferase with ATP was used as a bioluminescence source. The system was able to detect emission rates below 10-6 lux over 30 s of integration time at room temperature. Photodetectors for optical sensors can be fabricated concurrently with electronic circuits as part of the CMOS process flow, avoiding the need for complex postprocessing

Chemical Sensors with Integrated Electronics

Chemical Reviews, 2008, Vol. 108, No. 2 647

steps. Selection of a photodetector for a given optical sensor can be a challenge because of the variety of semiconductor photodetectors: p-n photodiode, p-i-n photodiode, avalanche photodiode, and phototransistor. The characteristics of these photodetectors are summarized in a review article by Yotter et al.99 that provides the information needed to select the most appropriate detector for a given application.

4. Mass-Sensitive Sensors Mass-sensitive sensors detect the change of mass on a sensing layer. In the case of chemical sensors, the mass changes arise from absorption, evaporation, deposition, or erosion due to chemical reactions. Several sensing structures have been employed to detect these mass changes, such as the thickness shear mode (TSM) resonator, quartz crystal microbalance (QCM), and surface acoustic wave (SAW) device.100 Most of these devices are not suitable for integration with CMOS circuits. The mass-based sensors that do incorporate CMOS circuits use the resonant frequency shift of a cantilever beam to detect the change in mass.101-105 The resonant frequency is very sensitive to the beam’s mass. The cantilever is electrostatically actuated, and the resulting vibration of the cantilever changes the capacitance between the cantilever and a sensing electrode. The integrated electronics are used to monitor the frequency through this capacitance change.101-105 Verd et al.102 developed an integrated submicroelectromechanical resonator with readout circuits. Figure 11 shows the schematic of the cantilever-based mass-sensitive sensor and a photograph of the fabricated chip.102 The cantilever, driver electrode, and polarization capacitor (nanocapacitor) were fabricated in the “nanoarea” after the CMOS circuitry was fabricated and passivated. The cantilever dimensions were 40 µm (length) × 840 nm (width) × 600 nm (thickness). Cantilever structures, which are essential to this type of sensors, are not CMOS compatible. The above papers used two different ways of fabricating cantilevers on CMOS circuits. For polysilicon cantilevers,101-104 a region of polysilicon was patterned in the CMOS circuit layout. After completion of CMOS processing, the passivation layer in the region of the polysilicon cantilever was removed and the polysilicon was patterned and etched to form a cantilever structure. Finally, the oxide under the polysilicon layer was removed to release the cantilever. Li et al.105 fabricated a vertical cantilever composed of a stack of metal and dielectric materials. They covered the circuit and cantilever areas with a metal layer that was used as a mask. Dielectric material between the cantilevers was anisotropically etched down to the silicon surface. Finally, the exposed silicon was etched to release the cantilever.

5. Integration of Different Transducers on a Single Chip Integration of two or more types of sensors on a single chip offers a number of potential advantages. Signals from one transducer could help compensate or calibrate other sensors, and simultaneous detection could increase the selectivity or sensitivity of the system. With different transducer technologies available, a broader set of chemicals can be detected. If the analyte of interest is detectable by two different sensor types, the confidence in the measurement is significantly increased. From a circuit point of view, it might be possible to share circuit blocks such as an ADC,

Figure 11. (a) Schematic drawing of a cantilever-based CMOS integrated mass-sensitive sensor, and (b) a photograph of a fabricated chip with resonator and readout circuit. Reprinted with permission from ref 102. Copyright 2005 IEEE.

DAC, and digital modules between the sensor types, thereby minimizing the total chip size and number of output pads. As a simple example of integrating two different types of transducers on a chip, Covington et al.106 made a CMOS integrated sensor with both a ChemFET and a chemoresistor on the same site. They used a carbon black polymer composite membrane for both a resistive gas-sensing membrane and the gas-sensitive membrane of a ChemFET. This integration improved the discrimination when detecting ethanol and toluene vapor in air. Sawada et al.107 presented an interesting device that can sense both photons and ions on the same pixel for the nearsimultaneous detection of a photosignal and ion concentration. They used different mechanisms for photosensing and ion sensing and detected each parameter in a separate time period. For ion sensing, the depth of a surface potential well under the sensing region is converted to charge, and this charge is moved to a floating diffusion by a charge-transfer technique.34 The amount of charge, which is proportional to the pH of the solution over the sensing area, is measured using an integrated source-follower circuit. For photon sensing, the electrons generated by incident photons are transferred to the floating diffusion and the accumulated

648 Chemical Reviews, 2008, Vol. 108, No. 2

Joo and Brown

Figure 12. (a) Concept, schematic diagram, and (b) photograph of the device in ref 107. Reprinted with permission from ref 107. Copyright 2005 Elsevier.

charge is measured. Figure 12 shows the concept, schematic diagram, and photograph of the device.107 The performance of each sensing mechanism was tested using solutions of various pH and under several light intensities. The output of each sensor was unaffected by the other parameter, i.e., there was no cross-talk between the two sensors. A programmable CMOS electrochemical sensor array for biomolecular detection was developed by Hassibi et al.108 The digitally programmed system has a 5 × 10 sensor array that can perform impedance spectroscopy, voltammetry, potentiometry, and field-effect sensing. The chip was fabricated in a standard CMOS processes with no postprocessing, and the circuits for electrode configuration, array control, and signal readout for each analysis method are fully integrated on a single chip. Figure 13 shows the pixel

configuration of the sensor and a photograph of the fabricated chip.108 However, application of the system is limited to biosensor or nonfaradaic electroanalytical techniques because the system was fabricated by a standard CMOS process. The working electrode of the system was made of aluminum/ 1% silicon (Al/1%Si), which is commonly used for interconnect in CMOS circuits, rather than platinum, gold, or silver, which would have been much better electrochemically. Therefore, applications are limited to the conditions in which aluminum is not disruptive or corroded (within (200 mV of the applied bias). Nevertheless, the authors believe that this device could be applied to many biochemical detection platforms, including DNA and protein assays. For biomolecular sensing, a CMOS integrated system with on-chip optical and electrochemical dual-image CMOS

Chemical Sensors with Integrated Electronics

Figure 13. (a) SEM picture and top view of the micrograph picture of the differential transducer structure, and (b) microphotograph of fabricated chip and package. Reprinted with permission from ref 108. Copyright 2006 IEEE.

sensors was presented by Tokuda et al.109 The system can measure two-dimensional voltammetry with an 8 × 8 electrochemical sensor array and perform simultaneous image analysis using a 128 × 128 CMOS image sensor array with on-chip readout circuits. Figure 14 shows the concept of the dual-image sensor and a photograph of the fabricated chip.109 In addition to the separate optical and electrochemical detection, the integrating image sensor and electrochemical sensor make possible on-chip electrochemiluminescence imaging. The chip uses a modified three-transistor CMOS active pixel sensor110 for light sensing and has deposited gold on the working electrode site. For voltammetric tests, a twoelectrode configuration is used with an external Ag/AgCl reference electrode. The performance of simultaneous detection of optical and electrochemical sensing was tested using an agarose gel island in a saline droplet.

6. Conclusions and Outlook From the papers reviewed here, it is clear that integration of electronics with chemical sensors provides a number of performance advantages. The system signal-to-noise ratio can be improved by buffering and converting signals close to the transducers. System size and power dissipation can be reduced by integrating the instrumentation with the sensors. Sensing systems, rather than just sensors, can be batch

Chemical Reviews, 2008, Vol. 108, No. 2 649

Figure 14. (a) Concept of on-chip biomolecular detection using optical and electrochemical dual-image CMOS sensors, and (b) photograph of fabricated optical and electrochemical dual-image CMOS sensor chip. Reprinted with permission from ref 109. Copyright 2007 Elsevier.

fabricated. Multiple and complex analyses can be performed by precise on-chip control circuits. The chemical sensors that called for non-CMOS-compatible metals, insulators, or membranes realized these through appropriate postprocessing. Ongoing research in integrating circuits with chemical sensors will bring remarkable advances in the field of sensing and lead to true systems-on-chips. Realization of complete SoC chemical sensors calls for more circuits to be integrated, including not only signal processing circuits for data readout and conversion but also circuits such as microcontrollers, digital signal processing circuits, memory, and circuits for wired or wireless communications. Recently developed lowpower wireless communication standards such as Bluetooth111 and ZigBee112 could be good options for sensor systems requiring remote monitoring or in vivo biochemical analysis. Applying microfluidics techniques or lab-on-a-chip (LOC) technology with CMOS integrated chemical sensors, whether monolithic or hybrid, is another promising research direction. LOC devices are widely used in chemical and biological analysis.113 Linder et al.114 and Ghafar-Zadeh et al.115 have shown examples of such devices. An on-chip processor could control the microfluidic flow of sample, reagents, and calibration solutions, reducing the volumes of these needed, and performing on-chip calibration and reconditioning of the sensor. Merging the techniques for sample transport, filtering, and mixing from microfluidics with CMOS integrated chemical sensors may lead to a great synergy.

650 Chemical Reviews, 2008, Vol. 108, No. 2

There are challenges in fabricating CMOS integrated chemical sensors. Development of such sensors requires expertise in chemical/biological analysis, circuit design, semiconductor chip fabrication, and packaging. The cost of these sensors makes them ill suited for disposable or shortterm use. Still, integration of CMOS circuits with chemical sensors is a solution for applications requiring fast, accurate, complex analyses, and the functionality per unit cost of semiconductor processing continues to improve.

7. Acknowledgments This work was supported in part by the Engineering Research Centers Program of the U.S. National Science Foundation under Award Number EEC-9986866, by a U.S. National Institutes of Health NIBIB grant (R21-EB005022), and by the U.S. National Science Foundation Award Number BES-0529385. The authors also express gratitude for support by the Korea Research Foundation Grant funded by the Korean Government (MOEHRD, KRF-2006-214-D00084).

8. References (1) Bardeen, J.; Brattain, W. H. Phys. ReV. 1948, 74, 230. (2) Liu, C. Foundations of MEMS; Prentice Hall: Upper Saddle River, 2005. (3) Gardner, J. W.; Varadan, V.; Awadelkarim, O. O. Microsensors MEMS and Smart DeVices; Wiley: New York, 2001. (4) Franssila, S. Introduction to Microfabrication; Wiley: New York 2004. (5) Bergveld, P. IEEE Trans. Biomed. Eng. 1970, BME-17, 70. (6) Sze, S. M. Semiconductor DeVices: Physics and Technology, 2nd ed.; Wiley: New York, 2002. (7) Pierret, R. F. Semiconductor DeVice Fundamentals; AddisonWiley: Reading, 1995. (8) Hierlemann, A.; Baltes, H. Analyst 2003, 128, 15. (9) Hierlemann, A.; Brand, O.; Hagleitner, C.; Baltes, H. Proc. IEEE 2003, 91, 839. (10) Janata, J. Principles of Chemical Sensors; Plenum: New York, 1989. (11) Skoog, D. A.; West, D. M.; Holler, F. J.; Crouch, S. R. Analytical Chemistry: An Introduction, 7th ed.; Harcourt: Orlando, 2000. (12) Bard, A. J.; Faulkner, L. R. Electrochemical Methods, 2nd ed.; Wiley: New York, 2001. (13) Madou, M. J.; Morrison, S. R. Chemical Sensing with Solid State DeVices; Academic Press: San Diego, 1989. (14) Hafeman, D. G.; Parce, J. W.; McConnell, H. M. Science 1988, 240, 1182. (15) Bergveld, P. Sens. Actuators B 2003, 88, 1. (16) Sibbald, A. Sens. Actuators 1985, 7, 23. (17) Fung, C. D.; Fu, C. W. Proceedings of the 3rd International Conference on Solid-State Sensors and Actuators (Transducers ’85), Philadelphia, PA; IEEE: Washington, DC, 1985; p 55. (18) Wong, H.-S.; White, M. H. IEEE Trans. Electron DeVices 1989, 36, 479. (19) Bausells, J.; Carrabina, J.; Errachid, A.; Merlos, A. Sens. Actuators B 1999, 57, 56. (20) Gui-Hua, W.; Dun, Y.; Yao-Lin, W. Sens. Actuators 1987, 11, 221. (21) Chin, Y.-L.; Chou, J.-C.; Sun, T.-P.; Chung, W.-Y.; Hsiung, S.-K. Sens. Actuators B 2001, 76, 582. (22) Morgenshtein, A.; Sudakov-Boreysha, L.; Dinnar, U.; Jakobson, C. G.; Nemirovsky, Y. Sens. Actuators B 2004, 97, 122. (23) Morgenshtein, A.; Sudakov-Boreysha, L.; Dinnar, U.; Jakobson, C. G.; Nemirovsky, Y. Sens. Actuators B 2004, 98, 18. (24) Yang, H.; Sun, H.; Han, J.; Wei, J. Proceedings of the 5th International Workshop on System-on-Chip for Real-Time Applications, Banff, Canada; IEEE: Washington, DC, 2005; pp 180-183. (25) Chin, Y.-L.; Chou, J.-C.; Sun, T.-P.; Liao, H.-K.; Chung, W.-Y.; Hsiung, S.-K. Sens. Actuators B 2001, 75, 36. (26) Chodavarapu, V. P.; Titus, A. H.; Cartwright, A. N. Electron. Lett. 2005, 41, 698. (27) Wilbertz, C.; Frerichs, H.-P.; Freund, I.; Lehmann, M. Sens. Actuators A 2005, 2, 123-124. (28) Janata, J. U.S. Patent 4,411,741, 1983. (29) Lundstrom, I.; Shivaraman, S.; Svensson, C.; Lundkvist, L. Appl. Phys. Lett. 1975, 26, 55. (30) Eisele, I.; Doll, T.; Burgmair, M. Sens. Actuators B 2001, 78, 19.

Joo and Brown (31) Milgrew, M. J.; Hammond, P. A.; Cumming, D. R. S. Sens. Actuators B 2004, 103, 37. (32) Milgrew, M. J.; Riehle, M. O.; Cumming, D. R. S. Sens. Actuators B 2005, 347, 111-112. (33) Hizawa, T.; Sawada, K.; Takao, H.; Ishida, M. Sens. Actuators B 2006, 117, 509. (34) Sawada, K.; Shimada, T.; Ohshina, T.; Takao, H.; Ishida, M. Sens. Actuators B 2004, 98, 69. (35) Hammond, P. A.; Ali, D.; Cumming, D. R. S. IEEE Sens. J. 2004, 4, 706. (36) Hammond, P. A.; Ali, D.; Cumming, D. R. S. IEEE Trans. Biomed. Eng. 2005, 52, 687. (37) Cremer, M. Z. Biol. 1906, 47, 562. (38) Haber, F.; Klemensiewicz, Z. Z. Phys. Chem. 1909, 67, 365. (39) Arnold, M. A.; Meyerhoff, M. E. Anal. Chem. 1984, 56, 20R. (40) Bakker, E.; Buhlmann, P.; Pretsch, E. Chem. ReV. 1997, 97, 3083. (41) http://www.nico2000.net/Book/Guide1.html (42) Lauks, I.; Van der Spiegel, J.; Sansen, W.; Steyaert, M. Proceedings of the 3rd International Conference on Solid-State Sensors and Actuators (Transducers ’85), Philadelphia, PA; IEEE: Washington, DC, 1985; pp 122-124. (43) Brown, R. B.; Huber, R. J.; Petelenz, D.; Janata, J. Proceedings of the 3rd International Conference on Solid-State Sensors and Actuators (Transducers ’85), Philadelphia, PA; IEEE: Washington, DC, 1985; pp 125-127. (44) Koryta, J.; Stulik, K. Ion-SelectiVe Electrodes, 2nd ed.; Cambridge University Press: London, 1983. (45) Cha, G. S.; Liu, D.; Meyerhoff, M. E.; Cantor, H. C.; Midgley, A. R.; Goldberg, H. D.; Brown, R. B. Anal. Chem. 1991, 63, 1666. (46) Martin, S. M.; Ha, J.; Kim, J. W.; Strong, T. D.; Cha, G. S.; Brown, R. B. Proceedings of the Technical Digest of Solid-State Sensor, Actuator, and Microsystems Workshop, Hilton Head, SC; Transducer Research Foundation, Inc.: Cleveland, OH, 2004; pp 396-399. (47) Martin, S. M.; Strong, T. D.; Brown, R. B. Proceedings of the Materials Research Society Spring Meeting, San Francisco, CA; Materials Research Society: Warrendale, PA, 2005; p D3.3. (48) Eranna, G.; Joshi, B. C.; Runthala, D. P.; Gupta, R. P. Crit. ReV. Solid State Mater. Sci. 2004, 29, 111. (49) Korotcenkov, G. Mater. Sci. Eng. B 2007, 139, 1. (50) Graf, M.; Gurlo, A.; Barsan, N.; Weimar, U.; Hierlemann, A. J. Nanoparticle Res. 2006, 8, 823. (51) Barsan, N.; Weimar, U. J. Electroceram. 2001, 7, 143. (52) Graf, M.; Barrettino, D.; Taschini, S.; Hagleitner, C.; Hierlemann, A.; Baltes, H. Anal. Chem. 2004, 76, 4437. (53) Barrettino, D.; Graf, M.; Zimmermann, M.; Hagleitner, C.; Hierlemann, A.; Baltes, H. Analog Integr. Circ. Sig. Process. 2004, 39, 275. (54) Graf, M.; Barrettino, D.; Zimmermann, M.; Hierlemann, A.; Baltes, H.; Hahn, S.; Barsan, N.; Weimar, U. IEEE Sens. J. 2004, 4, 9. (55) Barrettino, D.; Graf, M.; Song, W. H.; Kirstein, K.-U.; Hierlemann, A.; Baltes, H. IEEE J. Solid-State Circ. 2004, 39, 1202. (56) Graf, M.; Barrettino, D.; Kirstein, K.-U.; Hierlemann, A. Sens. Actuators B 2006, 117, 346. (57) Afridi, M. Y.; Suehle, J. S.; Zaghloul, M. E.; Berning, D. W.; Hefner, A. R.; Cavicchi, R. E.; Semancik, S.; Montgomery, C. B.; Taylor, C. J. IEEE Sens. J. 2002, 2, 644. (58) Afridi, M.; Hefner, A.; Berning, D.; Ellenwood, C.; Varma, A.; Jacob, B.; Semancik, S. Solid-State Electron. 2004, 48, 1777. (59) Bota, S. A.; Dieguez, A.; Merino, J. L.; Casanova, R.; Samitier, J.; Cane, C. Analog Integr. Circ. Sig. Process. 2004, 40, 175. (60) Barrettino, D.; Graf, M.; Taschini, S.; Hafizovic, S.; Hagleitner, C.; Hierlemann, A. IEEE Sens. J. 2006, 6, 276. (61) Frey, U.; Graf, M.; Taschini, S.; Kirstein, K.-U.; Hierlemann, A. IEEE J. Solid-State Circ. 2007, 42, 441. (62) Guo, B.; Bermak, A.; Chan, P. C. H.; Yan, G.-Z. Solid-State Electron. 2007, 51, 69. (63) Sepaniak, M.; Datskos, P.; Lavrik, N.; Tipple, C. Anal. Chem. 2002, 74, 568A. (64) Zimmermann, M.; Volden, T.; Kirstein, K.-U.; Hafizovic, S.; Lichtenberg, J.; Hierlemann, A. Proceedings of the 31st European Solid-State Circuits Conference, Grenoble, France; IEEE: Washington, DC, 2005; pp 343-346. (65) Lonergan, M. C.; Severin, E. J.; Doleman, B. J.; Beaber, S. A.; Grubbs, R. H.; Lewis, N. S. Chem. Mater. 1996, 8, 2298. (66) Tang, K.-T.; Goodman, R. M. Proceedings of the 5th World MultiConference on Systemics, Cybernetics and Informatics (SCI 2001), Orlando, FL; International Institute of Informatics and Systems (IIIS): Orlando, FL, 2001; pp 534-539. (67) Tang, K.-T.; Goodman, R. M. Proceedings of the IEEE Custom Integrated Circuits Conference, San Jose, CA; IEEE: Washington, DC, 2006; pp 273-276. (68) Koickal, T. J.; Hamilton, A.; Tan, S. L.; Covington, J. A.; Gardner, J. W.; Pearce, T. C. IEEE Trans. Circ. Syst. I 2007, 54, 60.

Chemical Sensors with Integrated Electronics (69) Dai, C.-L.; Liu, M.-C.; Chen, F.-S.; Wu, C.-C.; Chang, M.-W. Sens. Actuators B 2007, 123, 896. (70) Chow, C. T.; Sin, M. L. Y.; Leong, P. H. W.; Li, W. J.; Pun, K. P. Proceedings of the 2nd IEEE International Conference on Nano/ Micro Engineered and Molecular Systems, Bangkok, Thailand; IEEE: Washington, DC, 2007; pp 1209-1214. (71) Murrell, J. N.; Jenkins, A. D. Properties of Liquids and Solutions, 2nd ed.; Wiley: Chichester, England, 1994. (72) Qiu, Y. Y.; Azeredo-Leme, C.; Alcacer, L. R.; Franca, J. E. Sens. Actuators A 2001, 92, 80. (73) Dai, C.-L. Sens. Actuators B 2007, 122, 375. (74) Kummer, A. M.; Hierlemann, A. IEEE Sens. J. 2006, 6, 3. (75) Stagni, C.; Guiducci, C.; Benini, L.; Ricco, B.; Carrara, S.; Samori, B.; Paulus, C.; Schienle, M.; Augustyniak, M.; Thewes, R. IEEE J. Solid-State Circ. 2006, 41, 2956. (76) Turner, R. F. B.; Harrison, D. J.; Baltes, H. P. IEEE J. Solid-State Circ. 1987, 22, 473. (77) Steyaert, M. S. J.; Sansen, W. M. C.; Zhongyuan, C. IEEE J. SolidState Circ. 1987, 22, 1163. (78) Kakerow, R. G.; Kappert, H.; Spiegel, E.; Manoli, Y. Proceedings of the 8th International Conference on Solid-State Sensors and Actuators (Transducers ’95), Stockholm, Sweden; IEEE: Washington, DC, 1995; pp 142-145. (79) Bandyopadhyay, A.; Mulliken, G.; Cauwenberghs, G.; Thakor, N. Proceedings of the IEEE International Symposium on Circuits and Systems, Phoenix, AZ; IEEE: Washington, DC, 2002; pp II-740II-743. (80) Narula, H. S.; Harris, J. G. Proceedings of the International Symposium on Circuits and Systems, Vancouver, Canada; IEEE: Washington, DC, 2004; pp I-457-I-460. (81) Martin, S. M.; Gebara, F. H.; Strong, T. D.; Brown, R. B. Proceedings of the International Symposium on Circuits and Systems, Vancouver, Canada; IEEE: Washington, DC, 2004; pp IV-892-IV-895. (82) Ahmadi, M. M.; Jullien, G. A. Proceedings of the 5th International Workshop on System-on-Chip for Real-Time Applications, Banff, Canada; IEEE: Washington, DC, 2005; pp 184-189. (83) Gore, A.; Chakrabartty, S.; Pal, S.; Alocilja, E. C. IEEE Trans. Circ. Syst. I 2006, 53, 2357. (84) Genov, R.; Stanacevic, M.; Naware, M.; Cauwenberghs, G.; Thakor, N. V. IEEE Trans. Circ. Syst. I 2006, 53, 2371. (85) Ayers, S.; Gillis, K. D.; Lindau, M.; Minch, B. A. IEEE Trans. Circ. Syst. I 2007, 54, 736. (86) Zhang, J.; Huang, Y.; Trombly, N.; Yang, C.; Mason, A. Proceedings of the IEEE Sensors, Irvine, CA; IEEE: Washington, DC, 2005; pp 385-388. (87) Martin, S. M.; Gebara, F. H.; Larivee, B. J.; Brown, R. B. IEEE J. Solid-State Circ. 2005, 40, 2777. (88) Frey, A.; Hofmann, F.; Peters, R.; Holzapfl, B.; Schienle, M.; Paulus, C.; Schindler-Bauer, P.; Kuhlmeier, D.; Krause, J.; Eckstein, G.; Thewes, R. Microelectron. Reliab. 2002, 42, 1801. (89) Paulus, C.; Wieder, H.; Musewaid, C.; Lossau, H.; Thewes, R. Proceedings of the IEEE Sensors, Toronto, Canada; IEEE: Washington, DC, 2003; pp 1329-1332. (90) Schienle, M.; Paulus, C.; Frey, A.; Hofmann, F.; Holzapfl, B.; Schindler-Bauer, P.; Thewes, R. IEEE J. Solid-State Circ. 2004, 39, 2438.

Chemical Reviews, 2008, Vol. 108, No. 2 651 (91) Augustyniak, M.; Paulus, C.; Brederlow, R.; Persike, N.; Hartwich, G.; Schmitt-Landsiedel, D.; Thewes, R. Proceedings of the IEEE International Solid-State Circuits Conference, San Francisco, CA; IEEE: Washington, DC, 2006; pp 59-68. (92) Wolfbeis, O. S. Anal. Chem. 2006, 78, 3859. (93) El Gamal, A.; Eltoukhy, H. IEEE Circ. DeV. Mag. 2005, 21, 6. (94) Simpson, M. L.; Sayler, G. S.; Patterson, G.; Nivens, D. E.; Bolton, E. K.; Rochelle, J. M.; Arnott, J. C.; Applegate, B. M.; Ripp, S.; Guillorn, M. A. Sens. Actuators B 2001, 72, 134. (95) Bolton, E. K.; Sayler, G. S.; Nivens, D. E.; Rochelle, J. M.; Ripp, S.; Simpson, M. L. Sens. Actuators B 2002, 85, 179. (96) Iordanov, V. P.; Iliev, B. P.; Bossche, A.; Bastemeijer, J.; Sarro, P. M.; Young, I. T.; van Dedem, G. W. K.; Vellekoop, M. J. Proceedings of the IEEE Sensors, Vienna, Austria; IEEE: Washington, DC, 2004; pp 810-813. (97) Eltoukhy, H.; Salama, K.; El Gamal, A. IEEE J. Solid-State Circ. 2006, 41, 651. (98) Vijayaraghavan, R.; Islam, S. K.; Zhang, M.; Ripp, S.; Caylor, S.; Bull, N. D.; Moser, S.; Terry, S. C.; Blalock, B. J.; Sayler, G. S. Sens. Actuators B 2007, 123, 922. (99) Yotter, R. A.; Wilson, D. M. IEEE Sens. J. 2003, 3, 288. (100) Grate, J. W. Chem. ReV. 2000, 100, 2627. (101) Davis, Z. J.; Abadal, G.; Helbo, B.; Hansen, O.; Campabadal, F.; Perez-Murano, F.; Esteve, J.; Figueras, E.; Verd, J.; Barniol, N.; Boisen, A. Sens. Actuators A 2003, 105, 311. (102) Verd, J.; Abadal, G.; Teva, J.; Gaudo, M. V.; Uranga, A.; Borrise, X.; Campabadal, F.; Esteve, J.; Costa, E. F.; Perez-Murano, F.; Davis, Z. J.; Forsen, E.; Boisen, A.; Barniol, N. J. Microelectromech. Syst. 2005, 14, 508. (103) Forsen, E.; Abadal, G.; Ghatnekar-Nilsson, S.; Teva, J.; Verd, J.; Sandberg, R.; Svendsen, W.; Perez-Murano, F.; Esteve, J.; Figueras, E.; Campabadal, F.; Montelius, L.; Barniol, N.; Boisen, A. Appl. Phys. Lett. 2005, 87, 043507. (104) Villarroya, M.; Verd, J.; Teva, J.; Abadal, G.; Forsen, E.; PerezMurano, F.; Uranga, A.; Figueras, E.; Montserrat, J.; Esteve, J.; Boisen, A.; Barniol, N. Sens. Actuators A 2006, 132, 154. (105) Li, Y.-C.; Ho, M.-H.; Hung, S.-J.; Chen, M.-H.; Lu, M. S.-C. J. Micromech. Microeng. 2006, 16, 2659. (106) Covington, J. A.; Tan, S. L.; Gardner, J. W.; Hamilton, A.; Koickal, T.; Pearce, T. Proceedings of the IEEE Sensors, Toronto, Canada; IEEE: Washington, DC, 2003; pp 1120-1123. (107) Sawada, K.; Ohshina, T.; Hizawa, T.; Takao, H.; Ishida, M. Sens. Actuators B 2005, 106, 614. (108) Hassibi, A.; Lee, T. H. IEEE Sens. J. 2006, 6, 1380. (109) Tokuda, T.; Tanaka, K.; Matsuo, M.; Kagawa, K.; Nunoshita, M.; Ohta, J. Sens. Actuators A 2007, 135, 315. (110) Fossum, E. R. Proc. SPIE 1993, 1900, 2. (111) http://www.bluetooth.com. (112) http://www.zigbee.org. (113) Dittrich, P. S.; Manz, A. Nat. ReV. Drug DiscoVery 2006, 5, 210. (114) Linder, V.; Koster, S.; Franks, W.; Kraus, T.; Verpoorte, E.; Heer, F.; Hierlemann, A.; Rooij, N. F. d. Biomed. MicrodeV. 2006, 8, 1387. (115) Ghafar-Zadeh, E.; Sawan, M.; Therriault, D. Sens. Actuators A 2007, 134, 27.

CR068113+

Chem. Rev. 2008, 108, 652−679

652

Wireless Sensor Networks and Chemo-/Biosensing Dermot Diamond,* Shirley Coyle, Silvia Scarmagnani, and Jer Hayes Adaptive Sensors Group, National Centre for Sensor Research, School of Chemical Sciences, Dublin City University, Dublin 9, Ireland Received June 2, 2007

Contents

1. Introduction

1. Introduction 2. Internet-Scale Sensing and Control 3. WSN Platforms 3.1. Building Blocks of Autonomous Sensing Platforms 3.2. Linking the Sensor into Communications Infrastructure 3.3. Wireless Communications Options 3.4. Examples of Mote-Based Environmental Sensing Deployments 3.4.1. Example 1: Vineyard Monitoring 3.4.2. Example 2: Tree Microclimate 3.4.3. Example 3: Habitat Monitoring 3.4.4. Example 4: Intruder Detection over a Very Wide Area 3.4.5. Example 5: Volcanic Activity 3.4.6. Example 6: Soil Moisture 3.5. Discussion and Conclusions 4. Body Sensor Networks 4.1. Wearable Sensors 4.2. Functionalized Fabrics 4.2.1. Metal Fibers 4.2.2. Conductive Inks 4.2.3. Inherently Conducting Polymers 4.2.4. Optical Fibers 4.2.5. Coating with Nanoparticles 4.2.6. Integrated Components 4.2.7. Wearable Actuators 4.2.8. Interconnects and Infrastructure 4.3. Applications of Wearable Sensors 4.4. Wearable Chemosensing 4.5. Applications in Personalized (p)Health 4.6. Conclusions 5. Materials SciencesThe Future 5.1. Microfluidics and Lab-on-a-Chip Devices 5.2. Controlling Liquid Movement in Surfaces and on Channels 5.3. Controlling Binding Processes at Sensor Surfaces 5.4. Bead-Based Systems 6. Overall Conclusions 7. Abbreviations 8. Acknowledgments 9. References

652 652 654 654 656 656 657 657 657 658 658 658 659 659 661 661 662 662 662 662 662 662 663 663 664 665 667 668 670 670 670 671 672 675 677 677 678 678

* To whom correspondence should be addressed. Fax: 00-353-1-7007995. E-mail: [email protected].

The concept of ‘wireless sensor networks’ or WSNs is relatively new, probably less than 10 years old, and a logical extension of the greater ‘networked world’ through which a large proportion of the world’s population is already connected, for example, through mobile phones and other digital communication platforms. It envisages a world in which the status of the real world is monitored by large numbers of distributed sensors, forming a sensor ‘mesh’, that continuously feeds data into integration hubs, where it is aggregated, correlations identified, information extracted, and feedback loops used to take appropriate action. The entire system, in its ultimate manifestation, will be composed of interlocking layers of sensors that can be characterized in terms of their fit into a hierarchical model based on complexity (and therefore dependability) with feedback equally divided into layers of complexity (e.g., local vs aggregated). University engineering groups and electronics companies such as INTEL have driven much of the early research in this area. Given the diversity of technologies and disciplines involved and the ubiquitous nature of its impact in a wide variety of application sectors, it is impossible to cover everything in appropriate detail, even in a comprehensive review such as this. We therefore apologize in advance to those readers whose work or area of interest is not included. Our particular emphasis in this review is to give a general overview of aspects of the area we feel are important to readers of Chemical ReViews, and hence, we will focus particularly on both the opportunities for researchers involved in chemo-/ biosensing and the challenges that they must confront in order to ensure there is an appropriate fit between chemo-/ biosensing and communications technologies. Hence, the review is organized into several sections: (1) Developments in low-power wireless communications focusing on so-called ‘motes’ rather than mobile phone technologies as these have been designed specifically with wireless sensing in mind; applications of mote-based networks will focus on environmental deployments; (2) Wearable sensors and applications in personal health monitoring; (3) Futuristic concepts in chemo-/biosensing focused on control of surface binding and fluid movement.

2. Internet-Scale Sensing and Control Early champions of the concept of Internet-scale control were the TJ Watson-based IBM researchers Alex Morrow and Ron Ambrosio.1 According to their vision of ‘InternetScale Control’ the future world will operate on the basis of complex interlocking control loops that range from localized sensor-actuator systems to platforms that aggregate information from multiple heterogeneous sources. In the latter, specialized software routines trawl through huge information

10.1021/cr0681187 CCC: $71.00 © 2008 American Chemical Society Published on Web 01/24/2008

Wireless Sensor Networks and Chemo-/Biosensing

Dermot Diamond received his PhD from Queen’s University Belfast (Chemical Sensors, 1987), and was vice president for Research at Dublin City University (DCU), Ireland, (2002−2004). He has ublished over 150 peer reviewed papers in international science journals, is a named inventor in 12 patents, and is coauthor and editor of two books, ‘Spreadsheet Applications in Chemistry using Microsoft Excel’ (1997) and ‘Principles of Chemical and Biological Sensors’, (1998) both published by Wiley. He is currently the Director of the National Centre for Sensor Research, one of the largest sensor research efforts worldwide (see www.ncsr.ie) and a Science Foundation Ireland Principle Investigator (Adaptive Information Cluster award, see www.adaptiveinformation.ie). He is a member of the editorial advisory boards of the international journals Talanta (Elsevier) and The Analyst (RCS). In 2002, he was awarded the Inaugural Silver Medal for Sensor Research by the Royal Society of Chemistry. He was awarded a DSc in July 2002 by Queen’s University Belfast.

Shirley Coyle received her BEng in Electronic Engineering in 2000 from Dublin City University, Ireland. She then worked in the Information and Communications division in Siemens Ltd. for 2 years before commencing a PhD study in the field of Biomedical Engineering. The focus of this research was to develop a brain computer interface using optical brain imaging techniques. She received her PhD from the National University of Ireland Maynooth in 2005. Her research interests combine her biomedical engineering background with a longstanding interest in apparel design - wearable sensors and smart textiles for healthcare management. She has worked on the EU FP6 ‘Biotex’ project, which is a Europeanwide multipartner research effort to merge sensing capabilities with fabrics and textiles. She is currently studying for a diploma in fashion design at the Grafton Academy of Dress Designing.

repositories searching for patterns and correlations, which can form the basis of responses to multiple action points. This is represented in a simplistic manner in Figure 1. Conventionally, the engineers who dominated this area over the past decade promote this as the merging of the ‘real and digital’ worlds. However, introducing chemo-/biosensing extends this vision to the merging of the ‘molecular and digital’ worlds, with chemical sensors, biosensors, and analytical devices providing a window between these worlds.2

Chemical Reviews, 2008, Vol. 108, No. 2 653

Silvia Scarmagnani studied pharmaceutical chemistry in the University of Padua where in 2006 she received her Master Degree (Honors) in “Pharmaceutical’s Chemistry and Tecnology”. She carried out her master thesis (based on the synthesis of Antitumor Agents derived from Hydroxybenzaldehyde) in collaboration with Cardiff University, United Kingdom. In 2006 she started her PhD in Dublin City University, Ireland, where she is currently investigating the development of adaptive surfaces for optical sensing using molecular photoswitches under the supervision of Prof. Dermot Diamond.

Jer Hayes received a B.A. (Hons) in Psychology from University College Dublin in 1997 and completed an MSc. in Computer Science in 2003. He has researched Natural Language Processing, esp. in relation to semantics. More recently he worked on a project testing and further developing a wireless sensor network for monitoring the temperature of fish catches from ship to shore and onto the processing plant which was funded by Bord Iacaigh Mhara. He has also worked on wireless sensor networks as applied to water purification process monitoring and gas detection. He is currently involved in a desk-study for the Marine Institute investigating data management and communication issues for marine sensor systems.

In principle, if this vision is realized, it holds that the digital world can sense, interpret, and control the real world at the molecular level. Interestingly, it also means that the digital world approaches the complexity of the real world, and each can be regarded as a mirror of the other. This raises the interesting concept of ‘soft-sensors’ (i.e., software code whose function is to seek out specific patterns in data) which, in some ways, mimic the behavior of real sensors, for example, in terms of selectivity (detect a specific pattern) and transduction (generate a signal). Hence, the real world will be mapped to the digital world by vast numbers of networked sensors of various levels of complexity and capability which are autonomic in nature in that they are self-sustaining for extended periods of deployment. However, the cost of reliable autonomous chemo-/ biosensing is still far too great for massively scaled-up deployments, even for obvious applications in environmental

654 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 1. Concept of internet-scale control: conventionally control loops operate at a localized level with sensors monitoring one of more key parameters at one or more locations (bottom). Actuators are used to control the system being monitored on the basis of various algorithms. When the sensed information is passed through to the Internet, it is aggregated with other information streams emanating from a wide variety of sources. Specialized software seeks to identify patterns and correlations across the resulting hugely diverse data reservoirs which can be used to modify the system at the Internet scale.

monitoring or the rapid detection of bio-/chemowarfare agents. This cost base is due to the complexity of the processes that occur during chemo-/biosensing and particularly the need to include regular recalibration because of, for example, changes in the chemistry of the sensing surface that inevitably occur through exposure to the real world.3 In this review, we will examine developments in wireless sensor platforms that are helping to drive the area forward and discuss how these platforms will stimulate demand for compatible approaches to chemo-/biosensing in areas like environmental monitoring and wearable sensing for vital signs monitoring. Potential routes to delivering reliable autonomous chemo-/biosensing platforms capable of some degree of scale up will be examined, like microfluidics. Finally, we will highlight the critical role of fundamental materials science research in bridging the very significant gap between what the chemo-/biosensor community can currently offer and what is needed to realize this vision.

3. WSN Platforms 3.1. Building Blocks of Autonomous Sensing Platforms In engineering parlance, a sensor node is the smallest component of a sensor network that has integrated sensing and communication capabilities. It contains basic networking capabilities through wireless communications with other nodes as well as some data storage capacity and a microcontroller that performs basic processing operations. They usually come with several on-board transducers for temper-

Diamond et al.

ature, light level, etc., and increasingly a sensor board that usually slots onto the controller board. This allows the user to interface other sensors, including chemo-/biosensors to the mote, provided the signal is presented in the appropriate form for the controller. They also include a power supply, usually provided by an on-board battery. Ultimately, the goal is that WSNs will evolve into longlived, open, ubiquitous, multipurpose networked systems that continuously feed sensed data into the networked world. However, in order for the required massive scale up in numbers to happen, these devices must be completely selfsustaining over extended periods of time (up to years). In recent years, there has been a focus on power consumption as the small lithium button batteries commonly employed have limited lifetime and regular manual replacement is unrealistic. The sensor nodes within a wireless sensor network are also commonly referred to as “motes”. Much of the early research into mote platforms happened in California, led by people like Deborah Estrin and David Culler at Berkeley and Kris Pister (originally at UCLA but now at Berkeley). The most widely used motes in recent years have been those provided by Crossbow Technologies Inc., based in San Jose, CA, which is a spin off from the Berkeley groups (www.xbow.com). The importance of this research was recognized by the establishment of the ‘Centre for Embedded Networked Sensors’ (CENs) in 2002 through the NSF Science and Technology Centre program.4 Pister is also CTO of the company Dust Networks, which is making rapid headway in the commercialization of mote-based sensing. See the website www.dustnetworks.com for more information. According to Wang et al., the hardware requirements for wireless sensors include robust radio technology, a low-cost and energy-efficient processor, flexible signal inputs/outputs for linking a variety of sensors, a long-lifetime energy source, and a flexible, open source development platform.5 They also outline a number of software requirements for a wireless sensor node which include a small footprint capable of running on low-power processors, small memory requirement, and high modularity to aid software rescue. Thus, the basic components of a sensor node are a microcontroller, radio transceiver, set of transducers, and power source, and the software which runs on these nodes must be small and allow for efficient energy use. With some motes, such as those provided by uParts,6 a number of sensors are already built onto the mote and further sensors cannot easily be added. However, many motes have the capability to add specific expansion boards, which allow a wide variety of sensors to be attached. The motes listed in Table 1 share a number of common features such as the use of low-cost energy-efficient reduced instruction set computer (RISC) processors with a small program and data memory size (about 128 kb) and battery power supply. The typical interfaces are common on-board I/O buses and devices, e.g., the Universal Asynchronous Receiver-Transmitter (UART), timers, and analog-to-digital converters. A number of motes are displayed in Figure 2. In comparison to the current generation of laptops, motes have tiny amounts of memory and use low-powered processors, and they are therefore very challenged in terms of computational capability. In the simplest case, motes are programmed before deployment to perform measurements at a particular sampling rate and return the captured data in a prearranged format. In more sophisticated deployments, the motes are programmed

PIC 18F6720 at 20 MHz

ATmega128L

ATmega128L

ATmega128L at 8mhz

Intel PXA271 Scale processor at 3-416 MHz

8 MHz Texas Instruments MSP430 microcontroller

8 MHz Texas Instruments MSP430 microcontroller ATmega128L.

Particle

Mica2

Mica2Dot

Micaz

Imote2

Tmote Sky

Tmote Invent

BTnode(rev2)

PIC12F675 at 4 MHz

uPart

processor

TI MSP430 at 4.6 MHz ATmega128L

uNode Tyndall25

sensor platform

128 kB flash 4 kB SRAM, 4 kB EEPROM

10k RAM, 48k flash

flash memory 128 kB, measurement (serial) flash 512 kB, configuration EEPROM 4 kB flash memory 128 kB, measurement (serial) flash 512 kB, configuration EEPROM 4 kB flash memory 128 kB, measurement (serial) flash 512 kB, configuration EEPROM 4 kB SRAM memory 256 kB, SDRAM memory 32 MB, flash memory 32 MB 10k RAM, 48k flash

4 kB RAM, 1 kB EEPROM, 512 kB FLASH

n/a

10k RAM, 48k flash 128 kB

memory

radio

250 kbps 2.4 GHz IEEE 802.15.4 Chipcon wireless transceiver Ericsson Bluetooth module

Chipcon wireless transceiver 2.4 GHz IEEE 802.15.4

CC2420 IEEE 802.15.4 radiotransceiver 2.4 GHz band

2.4 GHz IEEE 802.15.4

868/916, 433, or 315 MHz multichannel transceiver

868/915 MHz transceiver, 50kbs 2.4 GHz ISM band (nRF2401 from Nordic VLSI) transmitter in 868, 914 MHz band communication or 433, 310/ 315 MHz band rf communication through RFM TR1001, 125 kb bandwidth, 868.35 ISM band Europe 315, 433, or 868/916 MHz multichannel radio transceiver

Table 1. Comparison of Currently Available Motes and Motes in Real-World Applications interface

integrated ADC, DAC, supply voltage supervisor, and DMA controller integrated humidity, temperature, and light sensors integrated light, temperature, acceleration, and sound sensors 10-bit analog-digital converter, I2C bus, and two hardware UART

USB client mini-BB), SB host UART 3×, PIOs, 2 C, SDI0, SPI 2×, 2 S, AC97, camera

51 pin expansion connector supports analog inputs, digital I/O, I2C, SPI and UART interfaces

10 bit ADC other interfaces DIO, 18 pin expansion board

on-board sensors: movement, light sensor, temperature, 1 LED, power regulation for unit 21 pin multipurpose connector with I2C, SPI, serial (625 kbps), parallel bus, analog input lines, interrupt input lines, digital I/O lines 10 bit ADC; other interfaces DIO, I2C, SPI

12 bit ADCs (8), DACs (2), and GPIO (8) modular architecture

power supply

rechargeable lithium ion battery 3 cell battery pack

2 AA batteries

3 AAA batteries

2 AA batteries

3 V coin cell

2 AA batteries

1 AAA battery

coin cell

coin cell coin cell

Wireless Sensor Networks and Chemo-/Biosensing Chemical Reviews, 2008, Vol. 108, No. 2 655

656 Chemical Reviews, 2008, Vol. 108, No. 2

Diamond et al.

or loss of data packets, which tends to increase with the number of motes involved in relaying the information. In practical terms, a random spatial deployment of sensor nodes without information on their location can be problematic as the time series data must be tied to a particular place to be of any real use. At present, locations are recorded manually in most deployments, which obviously inhibits large-scale deployments. Increasingly, GPS chips are being included on motes, but this comes with a cost and energy requirement that must be built into deployment considerations.

3.3. Wireless Communications Options

Figure 2. In the above image the following are displayed: (1) Teco Particle, (2) Teco uPart, (3) Micaz, (4) Mica2Dot, (5) coin cell battery, and (6) AA battery. The batteries offer a way to compare the relative size of the motes. Indeed, the actual size of a mote is generally constrained by the size of the power supply (the battery).

to facilitate sampling rates that adapt to external events and function cooperatively in terms of finding the optimum route for returning data to remote base stations. A new standard for smart sensors is also under development (IEEE 1451).7 This “Smart Transducer Interface Standard” will enable Transducer Electronic Data Sheets (TEDS) to be attached to compatible transducers, which stores the following information: sensor identification (ID), calibration, measurement range, and manufacture-related information. While such data sheets are commonplace for transducers like thermistors (and include circuits for various specific applications), they tend to be less popular with chemo-/biosensors due to the increased complexity of their behavior and response characteristics.

3.2. Linking the Sensor into Communications Infrastructure Due to the limited computing power of sensor motes, they often employ an operating system called TinyOS,8 although more recently, products that are fully ‘C’ compliant have become available, and these are generally preferred by experienced programmers. TinyOS is an operating system written in the nesC programming language,9 which is a dialect of C10 specifically designed for restricted operating environments as exists on sensor motes where there is limited memory and processor power available. TinyOS has an extensive component library that includes network protocols, sensor drivers, and data acquisition tools, which can be used either as is or modified for custom applications. The operating system supports a large number of sensor boards and can be used with the most popular mote sensor platforms. With respect to WSN deployments, they can either be deterministic or self-organizing. In a self-organizing deployment the routes to pass information between the nodes are determined by the network itself. In theory, multihop routing is more desirable as the transmission power of a radio is theoretically proportional to the distance squared (or even higher orders in the presence of obstacles), and multihop routing will therefore consume less overall energy than direct communication to a remote base station. However, this must be balanced against the increased incidence of data corruption

The communication standards for sensor networks in practice breakdown into general ISM band multichannel RF, the ZigBee protocol (IEEE 802.15.4), and the Bluetooth protocol (IEEE 802.15.1).11 In addition to these low-power, relatively short distance platforms, other forms of wireless communication can be used such as short-range and pointto-point infrared (IrDA), wireless local area networks (i.e., 802.11 wireless LAN as embodied in Internet hotspots and laptop computers), and GSM mobile phone technology, with the latter being used over longer distances. Deployments typically include a gateway node or base station that allows data transfer with other communications networks. In some scenarios the gateway node or base station is connected directly to a PC or laptop and the communication capabilities of the attached machine are used, for example, to transfer data to a web site. In other cases where WSNs are deployed in remote locations, a base station with built in communication capabilities such as GSM (where a signal is available) is used to simultaneously harvest the information from local motes (e.g., using Zigbee) and act as a gateway to the Internet.12 Briefly, both Bluetooth (802.15.1) and ZigBee (802.15.4) run in the 2.4 GHz unlicensed frequency band, use low power, and have a small form factor. The ZigBee standard is intended for consumer electronics, PC peripherals, medical monitoring, toys, and security and automation applications in buildings/homes. These applications require a technology with the ability to easily add or remove network nodes. ZigBee was developed largely for in-door use with rf signals being able to pass through most walls and ceilings,13 while Bluetooth was initially oriented toward user mobility and replacement of short cables (e.g., between phone and headset). However, Bluetooth can also support ad-hoc networks over a short range. With the IEEE 802.15.4 protocol and the IEEE 802.15.1 protocol becoming more widespread there will be movements toward more interoperability between WSNs based on different physical sensor boards using the same communications protocol. Both Bluetooth and ZigBee have been designed for short range, although this range can be extended up to 75 m for ZigBee and 100 m for Bluetooth using more specialized chipsets and antennae. However, these are essentially short-range systems and will tend to be confined to situations that conform to this limitation such as within rooms or vehicles. WSNs can be deployed using a number of network topologies such as ‘star’, ‘tree’, and ‘mesh’. These network topologies determine the way in which nodes receive and transmit messages to each other. In a star topology there is a central hub to which all other nodes send and receive data. The hub needs to be more sophisticated than the other nodes to carry out message and data handling. In this type of network topology the central hub is a base station or gateway

Wireless Sensor Networks and Chemo-/Biosensing

node. In the tree topology, there again is a central root node but it communicates with only one level beneath it in the hierarchy and the nodes at this lower level in turn only communicate with a parent node and child nodes. This tree topology is less common than the mesh topology, where nodes in the network can communicate with any other node that lies within range (the nearest neighbors). As sensor nodes can be in contact with more than one neighbor there are usually multiple routing paths between nodes. The shortest distance is usually the favored route to the base station, but mesh networks can use alternative pathways where nodes fail and so are somewhat more robust. In reality a single network can be composed of several subnetworks which are composed of different topologies. For example, using Bluetooth a maximum of 8 nodes, out of a total of 256 devices, can actively communicate in a star-shaped cluster, called a piconet. In a piconet the central hub of the star topology is called a ‘master’, while the other nodes are called ‘slaves’. However, piconets can also be interconnected via ‘bridge nodes’, and the resulting linked piconets together form a ‘scatternet’. A bridge is a node which participates in more than one piconet on a time-sharing basis. The mesh network topology is appropriate for ad-hoc networks where nodes enter and leave the network at different times (e.g., when nodes are mobile).14 ZigBee can also support star topologies and mesh topologies. With ZigBee technology, the central hub is the coordinator and this node needs to store information about the network and act as a bridge to other networks. The other types of nodes in a ZigBee network are ‘router nodes’, which just pass data, and ‘end-device nodes’. An end device can only communicate with its parent in the network. ZigBee operates in two main modes: nonbeacon mode and beacon mode. In the former, the router nodes periodically transmit beacons to each node, which wakes up each device and allows this device to return data if needed. This mode results in low power consumption as the end device can be maintained in a low-power sleep mode unless needed, and this can result in significant energy savings. In contrast, in nonbeacon mode any device can communicate with the coordinator and the coordinator must therefore always be awake to listen for communications. This requires more power for the coordinator device and may result in data loss, for example, when multiple end devices attempt to communicate with the coordinator at the same time.11

3.4. Examples of Mote-Based Environmental Sensing Deployments In this section we will briefly examine six real-world deployments of mote-based wireless sensor networks for environmental sensing. We have chosen these deployments as they give a flavor of how current WSN technology is actually being used, as opposed to futuristic views on what WSNs could be used for, which at times can be misleading and over-optimistic. These deployments also highlight some advantages and some issues with deploying WSNs in the real world. The examples are (1) vineyard monitoring, (2) tree microclimate, (3) natural habitat monitoring, (4) intruder detection over a very wide area, (5) volcanic activity, and (6) soil moisture monitoring.

3.4.1. Example 1: Vineyard Monitoring The vineyard deployment involved a sensor network comprising 64 Mica2 motes which were employed to monitor

Chemical Reviews, 2008, Vol. 108, No. 2 657

temperature over a hectare section of a vineyard for 30 days.15,16 The motes were deployed in a grid and configured as a multihop network with a maximum depth of eight nodes. The sensor nodes were static (being placed 1 m off the ground), and the routing of messages across the network was determined before the network was deployed. Two pathways for upstreaming data were chosen. The network was composed of 16 backbone nodes, and associated with each backbone node were 3 sensor nodes. The backbone nodes could send packets up to 25 m, while the sensor nodes sent packets up to 15 m. Data were recorded every 5 min, and during the deployment two arctic fronts moved across the vineyard. Between the sampling points, sensor nodes remained in sleep mode to conserve energy. All nodes had 43 amp hours of battery power, but the backbone nodes had to be changed every 6 weeks. However, one of the interesting aspects of the study was the difference in the expected success of data delivery and the actual data delivery in the field. Beckwith et al.15 suggest that in-lab performance resulted in 99% of packets being delivered. The predicted performance for an eighthop packet getting through was 92%, but over the course of the real deployment the actual success rate was 77%. This performance was based on sending the same data multiple times (five times from each sensor node), the performance deteriorating at higher transmission frequencies. Beckwith et al.15 also reported that nodes would occasionally leave the network, i.e., they would lose contact with the rest of the network. If this happens to a backbone node then all the data from the associated sensor nodes can be lost. Despite these issues, the vineyard deployment allowed collection of dense information on the temperature of a vineyard over an extended period of time. They discovered that the regions of highest temperature changed from day-to-day throughout the vineyard. This type of information is important as it can identify regions within the vineyard that will be more susceptible to mildew attack and can therefore be used to determine a targeted and optimized spraying regime to minimize product loss and hence maximize yield.

3.4.2. Example 2: Tree Microclimate In this case, a network of 33 Mica2Dot motes was deployed in a 70 m tall redwood tree to monitor the surrounding microclimate over a period of 44 days.17 The sensor nodes monitored temperature, relative humidity, and photosynthetically active radiation, with the choice of phenomena measured guided by the biological research priorities. For example, data on temperature and relative humidity can be fed into transpiration models for redwood forests, and the photosynthetically active radiation data provides information about energy available for photosynthesis in redwood forests. The sensor node was based on a Mica2Dot that had sensors for temperature, relative humidity, solar radiation (direct and ambient), and barometric pressure on board. The whole sensor node was encased in a specially designed housing to protect the components from physical damage during the deployment. To keep the WSN running for as long as possible without having to change batteries, the sensor nodes were activated for only 4 s to take measurements and data was transmitted at 5 min intervals for a period of 44 days. As the data was sampled every 5 min, large quantities of data were collected. Tolle et al.17 suggest that each sensor

658 Chemical Reviews, 2008, Vol. 108, No. 2

node acquired 50 450 data points and that 1.7 million data points in total should have been collected by the deployed network. However, only 820 700 data points were collected, meaning that only 49% of the possible data points were actually received. This suggests that WSNs should have a large degree of redundancy built in so that not every data point is required for decision making. Despite these losses in data gathering, Tolle et al.17 were able generate rich information on the mesoclimate of the redwood tree that previously had not be accessible. Arising from this study it was also apparent that some sensor nodes returned anomalous readings.17 These sensors either never produced readings in the expected normal range or produced readings that did not tally with other sensors. It was found that battery failure correlated strongly with these anomalous findings.

3.4.3. Example 3: Habitat Monitoring Szewczyk et al.18 incrementally deployed two sensor networks of increasing scale and complexity in a wildlife preserve in order to monitor the distribution and abundance of sea birds on Great Duck Island (Maine). It was assumed that passive infrared (PIR) sensors could directly measure heat from a seabird in a burrow and that temperature and humidity sensors could measure variations in the ambient conditions of the burrow, which would indicate the length of occupancy. Sensor nodes were deployed in various groupings referred to as ‘patches’ which involved either a line, a grid region, or a volume of nodes for 3-D monitoring. Each sensor patch had a gateway that sent data back via a transit network to a remote base station. The base station was located on a PC and provided database services and Internet connectivity. The sensor nodes were based on the Mica2Dot mote with two classes of sensor node deployed: a ‘weather’ mote and a ‘burrow’ mote. The weather mote was used to monitor the microclimate around a burrow and measured humidity, temperature, and atmospheric pressure. The sensors onboard the burrow mote measured temperature and humidity and had PIR sensors to detect burrow occupancy. This mote had to have a small form factor so it could be placed in a burrow. Two network topologies were employed, namely, single hop and multihop. The single-hop network was deployed in an elliptic shape and covered 57 m. No routing was performed by the nodes, and data was passed straight through to the gateway system. The gateway system was composed of two motes with one in contact with the sensor nodes and other in contact with the base station. Data was sent every 5 min. The second sensor network was a multihop network which was kite-shaped, 221 m long, with a maximum width of 71 m, narrowing to 8 m. This network sampled data through to the gateway system every 20 min as a result of routing beacons transmitted by the gateway node to seed the network discovery process. Both networks operated on different radio frequencies with the single-hop network using the 433 MHz band and the multihop using the 435 MHz band. This was done to eliminate potential interference between the networks. During the 115 day deployment, the networks produced in excess of 650 000 observations. It is difficult to judge the relative effectiveness of the two networks as a performance breakdown in relation to each network is not given, which is unfortunate. However, in relation to lab-based predictions compared against real-world applications, Szewczyk et al.18

Diamond et al.

were able to accurately predict the lifetime of the singlehop network but not the multihop network. In particular, the impact of multihop traffic on power source consumption was underestimated. They also state that the quality of a mote’s sensor readings was strongly dependent upon the mote power availability, which is in accordance with the previous case.17 Consequently, it is vital to ensure that sensing motes have adequate power that can cope with extended periods of deployment.

3.4.4. Example 4: Intruder Detection over a Very Wide Area Arora et al.19 outlined the biggest current deployment of a WSN with respect to the number of sensor nodes and area covered. The central idea behind the project was to deploy a dense wireless sensor network that would be a virtual “tripwire” over a large area. The WSN would detect, track, and categorize “intruders” that enter the area covered by the network. The project involved two demonstrations with the first comprising 90 Mica2 motes that were deployed over a 25 m × 10 m grassy area. The second used over 1000 ‘XSM’ motes as sensor nodes and 300 ‘XSS’ gateway motes. These XSM (extreme scale motes) do not appear to be publicly available and so are not listed in section 5.1, although they were commercially available previously under the trade name ‘MSP410CA Mote Security Package’. They are based on an Atmel ATmega128L microcontroller, a Chipcon CC1000 radio operating at 433 MHz, and a 4 Mb serial flash memory. The mote has four PIR, two magnetometer and acoustic sensors, and the entire device is housed in a rugged weatherproof package. The sensors nodes were deployed in such a way that more than one (up to five) would be triggered if an intruder (a person) entered the area covered by the WSN. More would be triggered if a larger object such as a vehicle entered the area. The coverage area was large compared to other WSN deploymentss1.3 km by 300 m. This deployment involved two tiers, the 1000 sensor nodes and 300 gateway nodes. The PIR sensor surface charge varies in response to the received infrared radiation emitted from a body, giving an indication that someone is present. However, a polyethylene film was placed on the PIR windows to reduce the effect of sunlight and increase the robustness of the sensor. The raw data from the sensors also had to be analyzed in such a way as to isolate sensor signals from the slower background variations rising from temperature-based drift using a digital band-pass filter.

3.4.5. Example 5: Volcanic Activity This network consisted of 16 nodes equipped with seismic and acoustic sensors deployed over a 3 km aperture on the Volca´n Reventador in northern Ecuador.20 The network was deployed for 3 weeks, and the data collected was routed over a multihop network and a long-distance radio link to a laptop sited at a remote observatory. The volcano is active, and at the time prior to deployment, seismic activity such as tremors and shallow rock fracturing had been recorded. The network consisted of 16 sensor nodes with each sensor node equipped with seismic and acoustic sensors. The nodes were built around the Moteiv TMote Sky wireless sensor network node and included a seismometer, microphone, and custom hardware interface board. These sensors draw a lot of power, and consequently, sensor nodes were powered by D cell batteries. Over the course of the 3 week trial, batteries were

Wireless Sensor Networks and Chemo-/Biosensing

Chemical Reviews, 2008, Vol. 108, No. 2 659

Table 2. Summary of Real-World WSN Deployments project

no. of motes

vineyard 64 (48 sensor nodes) macroscopic forest 33 Exscal Great Duck Island

soil moisture volcano eruptions

1000* 98 (62 burrow motes, 26 weather motes)

phenomena measured

sensor platform

temperature temperature, humidity, photosynthetically active radiation PIR burrow motes: temperature and humidity sensors, infrared temperature sensor

weather motes: temperature, humidity, barometric pressure 9 (4 were sensor nodes) soil moisture seismic waves

types of sensors used

longevity

Mica2 Mica2dot

thermistors PIR

30 days 44 days

XMS Mica2dot

PIR Sensirion SHT11, burrow motes: 52 days Intersema MS5534A barometer, TAOS TSL2550 light sensors, Hamamatsu S1087 photodiodes Intersema MS5534A weather motes: 120 days

Mica2 Tmote Sky

soil moisture sensor seismometer

28 days 3 weeks

Table 3. Comparison of Commercially Available Batteries type

typical lowest voltage output, V

typical highest voltage output, V

lowest capacity

highest capacity, Ah

alkaline lithium zinc carbon lead acid lithium rechargeable nickel cadmium nickel metal hydride

1.5 1.5 1.5 2 3 1.2 1.2

15 9 9 12 15 24 24

18 mAh 2.2 mAh 405 mAh 1 Ah 1 mAh 1.25 mAh 12 mAh

27 35 16.5 70 6.8 4.5 10

changed between 4 and 5 times. During the duration of the deployment, the network detected over 200 seismic events.

3.4.6. Example 6: Soil Moisture Cardell-Oliver et al.21 reported a wireless sensor network deployment that monitored soil moisture. The network also monitored rainfall and adjusted the frequency of measurements accordingly, i.e., when a heavy rainfall occurs the measurement frequency is increased. The purpose of the study was to monitor changes in the spatial distribution of soil moisture over time. The WSN was built around Mica2 motes as the sensor nodes and base station. Three motes had two soil moisture probes (the Echo20 soil moisture probe) attached as sensors. Another mote was connected to a tipping bucket rain gauge, while a fifth mote was used as a base station (which had GMS capabilities) and another four were used for routing. As the network needed to react to events such as a sudden rain fall, the network could not just sleep and wake up to sample at predetermined times over the course of the deployment. Rather the sensor nodes had to wake up and check regularly if an event had occurred. For this to happen every node on the network (base station, router nodes, and sensor nodes) had to be awake at the same time.21 The WSN was structured in such a way that the soil moisture readings from the sensor nodes were transmitted over five sensor network hops before reaching the base station. If any of these single-network hops failed then a reading would be lost. Over the first 13 days a total of 434 soil moisture messages were triggered but only 277 were logged, which is an overall delivery rate of 63.8%. However, Cardell-Oliver et al.21 report that despite these losses, the deployment met the ultimate goal of providing useful data on dynamic responses of soil moisture to rainfall. Table 2 compares these deployments in relation to how many sensors were used, what was “sensed”, what type of sensors were used, and the longevity of deployment. The deployments vary in the number of motes deployed with the Exscal project having the largest scale,19 while the period of the deployments ranges from 3 weeks to 120 days.

Information on the loss of packets is also summarized where this is available. Glasgow et al.22 in their discussion of realtime water quality monitoring describe a 92% data accuracy rate of one project as disappointing. From this perspective, the effectiveness of these WSN deployments is also disappointing but not unexpected. With wireless communications in the ISM bands used by these motes environmental factors will interfere with the signal. These performance rates of packet delivery are an indicator of problems that will affect WSNs in scaled-up deployments, as this will become much worse as the complexity and scale of the network increases.

3.5. Discussion and Conclusions These deployments share a number of things in common. Generally in undertaking the deployment the researchers have to choose the wireless platform to use and the appropriate sensors to attach to these platforms. When the hardware has been chosen the researchers have to decide where the sensor nodes will be physically located and how they will operate cooperatively, e.g., single hop versus multihop. If a multihop architecture is chosen then the most appropriate routing algorithm must also be decided upon. A number of issues become apparent from these deployments. For example, it is clear we are still a long way from the vision of large-scale deployments over wide geographical areas for long-term monitoring applications of any kind. It is also clear that massive scale up can only happen if the motes are essentially self-sustaining in all requirements. In terms of energy sources, at present, batteries are currently extensively employed in sensor networks. Table 3 compares the characteristics of a variety of available batteries. While battery performance has clearly improved in recent years and with power efficient sensors (like thermistors) it may be possible to achieve several years of autonomous operation, for chemo-/biosensing platforms battery power supply is at best only an interim solution as the power demand is much greater (see discussion below). For scale up, the inescapable conclusion is that each mote must incorporate a local energy

660 Chemical Reviews, 2008, Vol. 108, No. 2

Diamond et al.

Table 4. Typical Power Consumption of Sensors and Sensor Components sensor/sources

typical power consumption (mW)

thermistors light-dependent resistors (LDRs) LEDs laser diodes metal oxides IR gas sensors electrochemical: pH electrodes

<1 250 30 225 280 6000 50

examples Radionics 813-806 NORP12 Standard SIGMA/635/1 Figaro TGS2611 KT sensor (PPM 4022H) Global water 407228 pH/mV

Table 5. Power Consumption of Components of an Autonomous Reagent-Based Analyzer component

power consumption

photodiode microprocessor (normal operation) pump GSM: idle mode TX burst mode

33 mW (10 mA, 3.3 V) 19.8 mW (6 mA, 3.3 V) 3.6 W (300 mA, 12 V) 300 mW (25 mA, 12 V), 30,W (2500 mA, 12 V)

generation or scavenging capability that can provide enough power for the device to carry out all functions. Sources include small solar panels, wind/water turbines, and vibration energy, and it is not surprising that research to integrate appropriately scaled local energy sources to motes is rapidly gaining momentum.23 Outdoor solar energy is a promising source with current technologies providing energy densities of ca. 7.5 mW/cm2, a figure that is likely to increase significantly given the tremendous research activity in this area. For more background on the potential of local energy scavenging/harvesting to power autonomously deployed sensors, including wearable devices, the reader is referred to recent excellent reviews.24,25 In terms of distributed chemo-/biosensing, the energy demands and reliability of the sensor will be very significant limiting factors. Table 4 summarizes the power demand of some sensors and sensing components. Immediately, one can understand why temperature sensing is the first parameter that researchers focus on and why it is often built into sensing motes as a ‘give away’. These are very low-cost (ca. $15), rugged, reliable, and consume virtually no power. Light detectors like LDRs are also popular, as they are also lowcost (ca. $2-5), rugged, and reliable, although they consume considerably more power. Chemosensors consume considerably more energy than thermistors and are much more expensive. For example, a single IR gas sensor (Table 4) will consume the same energy per unit time as several thousand thermistors, and the Figaro gas sensor range is priced from ca. $20 to $2000 per sensor. Interestingly, the cost of laser sources for optical sensing continues to drop. We use LEDs for many optical sensing applications because of their low cost, reliability, and relatively low power demand. However, for applications where sensitivity is an absolute requirement, laser diodes may be substituted for LEDs, bearing in mind that the power demand is increased by an order of magnitude. With autonomous chemo-/biosensing platforms there are other demands on the power of the system beyond that of the sensor itself. In particular, incorporation of regular recalibration further increases the energy demand of autonomous chemo-/biosensing platforms beyond that normally encountered in physical sensing platforms. Table 5 compares

the energy demand of various components used in an autonomous phosphate analyzer employing reagent-based chemistry in a microfluidic manifold that is typical of many devices of its kind.26 From this it is clear that liquid pumping and GSM communications dominate the power demand. The wireless communications power issue is the subject of very considerable research activity worldwide, driven by huge commercial opportunities (e.g., personal communications, mobile computing, RFID, etc.), and it is likely that solutions will be found that will be employed in autonomous analytical devices. However, the issue of liquid handling is much more specific to analytical science, and this will be the key limiting factor inhibiting chemo-/biosensor network deployments for applications involving liquid-phase measurements (e.g., water quality monitoring). The solution is either to use communities of very simple (noncalibrated) devices that do not employ liquid handling or to develop completely new ways to move liquids around in fluidic manifolds, for example, through fundamental breakthroughs in materials science that will facilitate very low energy liquid movement within a microfluidic manifold. The importance of fundamental materials science to the future of chemo-/biosensor networks is discussed in more detail below. Another factor that influences greatly the power demand of an instrument is the duty cycle (i.e., the time it is ‘on’ or in active mode compared to the time it is ‘off’ or in sleep mode). This in turn is related to how long it takes for the instrument to perform a complete measurement (including calibration and diagnostics) and how often this measurement must be performed. For more sophisticated and power-hungry instruments, the overall power demand can be reduced by maximizing the ‘off’ time relative to the ‘on’ time. This can be achieved by minimizing the number of measurements (a somewhat contradictory tactic as the function of the device is to perform measurements) and reducing the measurement time. This latter factor is determined by characteristics such as manifold dead volume, flow rate, and detector/electronics ‘wake up’ time. The employment of microfluidics is therefore an attractive option as this greatly reduces dead volume and analysis times, leading to a much more efficient system in terms of power demand. It is likely that power-hungry systems will be used to provide highly reliable validation measurements in future hierarchical deployments with the duty cycle linked to information provided from the sensor network (i.e., sampling frequency is increased at times when a dynamic event is predicted to be imminent by the network). Considering the above discussion, it is understandable why mote-based deployments focus almost entirely on relatively simple transducers (thermistors, photodetectors, movement/ vibration sensors, etc.) that require little power and are reliable over long periods of time. Another critical difference between chemo-/biosensors and these transducers is that transducers are invariably completely encased within a rugged encapsulant (e.g., epoxy) and yet continue to function, whereas chemo-/biosensors must be exposed directly to the real world (which in many cases can be a very hostile environment), and this constitutes a significant point of weakness for the entire device. One strategy to overcome this is to separate out the sensing and sampling functions using a fluidic manifold. For example, in water quality applications, water samples may be drawn in through an appropriate filter into the device where it is analyzed and stored for later disposal. However, for extended use, this means that all reagents must be stable in storage for at least

Wireless Sensor Networks and Chemo-/Biosensing

months (preferably several years) and the quantities used must be minimized, for example, using microfluidics (see below). This strategy allows the sensing to be performed within the relatively benign environment of the microfluidic manifold rather than in the sample.27 Clearly, microanalytical instruments are much more sophisticated sensing devices than those employed in current WSN research, and despite the relatively small reagent consumption, they will have to be serviced or collected eventually, which acts against the realization of very large scale deployments. An alternative strategy that has great promise is to incorporate the chemical sensing film into RFID tags such that the response generated in the film modulates the response characteristics of the tag. In contrast to mote-based sensing, this platform has no inherent networking communications overhead, does not require a battery, and is very low cost and yet capable of very sensitive measurements to the parts per billion level of polar volatiles.28 This type of imaginative approach to distributed chemical sensing is much more likely to lead to large-scale deployments than the current manifestations of mote platforms.

4. Body Sensor Networks 4.1. Wearable Sensors Traditionally our clothes function as a protective layer, offer comfort and aesthetic appeal, and serve communication and cultural purposes. Moving beyond these functions, textiles provide a pervasive platform upon which to build sensor networks. Through integration of novel technologies our clothing will be equipped with information and power transmission capabilities, sensory functions, and an infrastructure for embedded microsystems.29,30 This additional functionality has important applications in areas like personalized healthcare, athletic performance, threat detection, and future fashion trends. Given that most people these days are connected to digital communications via a mobile phone, the issue is focused on how to realize a body sensor network (BSN) that is linked to the external world via the mobile phone; for an excellent overview of this area, see ref 31. This means that the distances involved are much smaller and can be covered using a combination of low-power wireless communications and integrated conducting tracks embedded in the fabrics. There are major forcing driving this area forward, principally based around the concept of personal health or pHealth. This concept has arisen as a response to the unsustainable increase in healthcare costs worldwide due mainly to an epidemic of ‘life-style’ diseases arising from unhealthy diet and lack of exercise, such as obesity, cardiovascular disease, diabetes and chronic respiratory disease. There is a very significant focus on research related to pHealth in the European Union 7th Framework Programme, and an annual series of conferences in pHealth has recently commenced in an effort to structure this effort across many diverse disciplines (see, for example, the 4th pHealth conference,website‘pHealth2007’athttp://phealth2007.med.auth.gr/ ). Body sensor networks, or wearable sensors, is a major theme within pHealth as, in principle, it allows an individual’s vital signs and physical activity to be tracked and monitored remotely. This is a technology that is certain to make a major impact on society within the next 5 yearss the first commercial consumer products are beginning to appear, like the ‘Nike-iPod sport kit’ launched in May 2006 (see www.apple.com/pr/library/2006/may/23nike.html). At

Chemical Reviews, 2008, Vol. 108, No. 2 661

the annual MAC-World event earlier this year, it was announced that this product had realized sales of 500 000 units in the first 90 days since launch, which translates to a market of ca. $100 million for this product alone. The kit enables users to log activity through an accelerometer sensor embedded in an exercise trainer shoe that transmits data (such as number of steps, distance covered) wirelessly to an iPod (see www.apple.com/pr/library/2006/may/23nike.html). At the moment, this product is being used by joggers, but given the enormous impact of inactivity on healthcare costs, it is clear that its use will quickly be extended to encourage individuals at risk of obesity-linked diseases to exercise regularly. This in turn will create a framework within which demand for more specific information offered by chemo-/ biosensors will rise rapidly, but once again, as for environmentally deployed mote-based sensing, the sensor configuration, form factor, and operating specifications will have to meet with the expectations of the users. As an intermediary interface, wearable sensors have the potential to monitor and respond to both the wearer and his or her environment. Given the intimate connection with the wearer, garments may accommodate sensors placed in close proximity to the skin that look into the physiology of the body, e.g., breathing rate, body temperature, and heart rate. Conversely, clothes act as our protective shield, and sensors may be configured to look outward to the environment and identify any potential hazards that may endanger the wearer, e.g., the presence of toxic gases for emergency disaster workers.32 In order for the technology to be accessible it must remain innocuous and impose minimal intrusion on the daily activities of the wearer, and consequently, the sensors must monitor in a discreet manner that is easy to use. Therefore, wearable technologies should be soft, flexible, and washable to meet the expectations of normal clothing. Data transmission must be wireless to allow free movement of the wearer, and a means of feedback must be provided to the wearer through visual, tactile, or auditory means. The type and frequency of such feedback is largely application dependent and dynamic in nature, and context awareness is therefore a crucial feature in generating user feedback. Context awareness involves knowledge about the user (motion, activity, gestures, health status), the environment (weather, lighting, noise), and social influences (other people sharing information with an individual). Control systems are needed to perform data fusion from multiple sensors, data filtering, feature extraction, and classification approaches to take these factors into account in decision making. Smart fabrics (in which sensing capabilities are integrated with the textile) are capable of sensing environmental conditions or stimuli arising from various sources (mechanical, thermal, photonic, chemical, electrical, or magnetic). They may also be able to respond to signals through actuators creating a fabric with inherent motor functionality. Conventional electronic components are hard and brittle, which lends to an inharmonious integration into the soft textile substrate. While electronic components are being continuously miniaturized, in order to preserve the inherent mechanical properties of clothing, the ultimate objective is to directly integrate the sensing capability into the textiles themselves, i.e. the fabric becomes the sensor.

662 Chemical Reviews, 2008, Vol. 108, No. 2

4.2. Functionalized Fabrics The first stages in wearable computing research involved adorning the wearer with conventional electronic devices. However, this resulted in problems related to washability, comfort, and flexibility.33,34 For sensing technologies to be adopted by the clothing industry, aesthetic properties such as the handle, drape, and comfort of the fabric and garment must not be compromised. The cost factor is also a major challenge as the sensors must be priced appropriately in line with the application. Conductive textiles (i.e., fabrics with integrated conductive polymer threads35 or conductively plated fabrics and fabrics with embedded metallic fibers36), while originally developed for their antistatic behavior, can form the interconnection substructure required for wearable circuitry. Materials such as metallics, optical fibers, and conducting polymers may be integrated into the textile structure, thus supplying electrical conductivity and sensing and data transmission capabilities. The successful development of electronic textiles must take into account traditional textile manufacturing techniques in addition to the properties of the finished garment, e.g., metal threads tend to be heavier than most textile fibers, and their brittle characteristics can damage spinning machinery over time.36 Woven optical fiber textiles must consider bending of the fiber during the manufacturing process and also with the end product, as bending and mechanical damage cause light to escape and hence signal loss.37 Polymer electronic devices may provide a solution to overcome the stiffness of inorganic crystals such as silicon as they are light, elastic, resilient, mechanically flexible, inexpensive, and easy to process and have the potential to provide a variety of functions required for such systems including sensing, actuating, computation, and energy generation/storage.38

4.2.1. Metal Fibers Metal threads are made up of metal fibers which are very thin metal filaments (diameters ranging from 1 to 80 µm). The fibers are produced either through a bundle-drawing process or shaved off the edge of thin metal sheeting.39 Bekintex produces a range of yarns made of stainless steel blends for intelligent textile applications. These vary in composition from 100% continuous conductive steel fibers to feltings or composites of polyester with short steel fibers interspersed throughout. By varying the ratio of the two constituent fibers, different resistivity of the yarns may be achieved.

4.2.2. Conductive Inks A conductive layout can be screen printed using conductive inks to add conductivity to specific areas of a garment. Carbon, copper, silver, nickel, and gold may be added to conventional printing inks to make them conductive. Printed areas can be subsequently used as switches or pressure pads for activation of circuits. This technique is supported by recent advances in digital printing, but further development is required for pre- and post-treatments, ink performance, agitation of the ink reservoir for an equal distribution of the conductive particles, and drying of the printed output.36

4.2.3. Inherently Conducting Polymers Materials with inherent piezoresistive properties may be used as simple strain gauge or pressure-sensitive devices. Inherently conducting polymers (ICPs) are a class of

Diamond et al.

polymers with multiple functions in sensing and actuation40 with the electrical conductivity arising from their extended conjugated π-bond structure. Some commonly used ICPs are polyacetylene, polypyrrole, polyaniline, and polythiophene with polypyrrole (PPy) being popular due to its high mechanical strength coupled with high elasticity, general stability, and electroactivity in both organic and aqueous solvents. It has been used in the development of fabricresistive sensors by depositing thin layers of PPy onto fabrics with high elasticity such as nylon Lycra using an in situ chemical polymerization process. Conductivity changes result from external deformation of the material, thereby creating a flexible strain gauge41 that can be easily integrated with a wide variety of garments. The major advantage of this approach is that the sensors retain the natural characteristics of the underlying material. PPy-coated fabrics are reported to have an average gauge factor (GF) of -12 and temperature coefficient of resistance (TCR) of 0.018 °C-1, making them suitable for strain gauge implementations. The gauge factor is a measure of the sensitivity to strain, i.e., the ratio of its relative change in resistance to the applied strain. Polyurethane foam coated with PPy has been used to produce a soft, compressible, conducting foam which may be used as a pressure sensor. When the foam is compressed under an external force, the degree of contact between regions coated with conducting PPy film increases. This results in a shortening of the overall conducting pathlength, which increases the bulk conductivity. However, these polymer sensors exhibit a variation in resistance over time and a relatively long response time.38 Another approach to developing piezoresistive polymer sensors employs carbon-loaded elastomers.42 The sensors are fabricated by coating with a conductive mixture of silicone and carbon black powder. The sensing component pattern is applied by masked smearing, and the same polymer/conductor composite is also used for the connection tracks between sensors and an acquisition electronic unit, thereby avoiding the stiffness of conventional metal wires.38 They have been used to develop a number of sensorized garments in this way for high-performance sensing with a reported GF of 2.5, similar to metals, and TCR value of 0.08 °C-1.

4.2.4. Optical Fibers Optical fibers may be used to transmit data signals, transmit light for optical sensing, detect deformations in fabrics due to stress and strain, and perform chemical sensing.43 Optical fibers have the advantage of not generating heat and are insensitive to EM radiation. Bending of the fibers is a problem in the manufacturing process and also with the end product as mechanical damage causes signal loss. Therefore, the textile structure must be designed to minimize bending to prevent light from being lost.37 For a good description of the integration process, see El-Sherif et al.43 Commercially available Luminex fabric is a textile incorporating woven optical fibers capable of emitting light. While this has aesthetic appeal for the fashion industry, it is also used in safety vests and has the potential to be used for data transmission and a platform for building optical sensing capabilities.44

4.2.5. Coating with Nanoparticles Nanotechnology is being widely applied within the textile industry to improve the performance and functionality of textiles. While conventional methods of imparting different properties to fabrics often do not lead to permanent effects,

Wireless Sensor Networks and Chemo-/Biosensing

Chemical Reviews, 2008, Vol. 108, No. 2 663

Figure 3. Strips of polypyrrole actuators are arranged in a flower structure. The actuators are shown to bend significantly, their bending direction depending on the voltage polarity ((1 V). The large displacement that can be obtained using low mass, low power, and low voltage polymer actuators makes them attractive in the development of biologically inspired robots. Photos courtesy of Y. Wu, Intelligent Polymer Research Institute, University of Wollongong.

nanotechnology can provide high durability for fabrics.45 This is due to the large surface area to volume ratio and high surface energy of nanoparticles. Coating with nanoparticles can enhance the textiles with properties such as antibacterial, water repellence, UV protection, and self-cleaning while still maintaining breathability and tactile properties of the textile.46 NanoTex has a range of products using such coatings to resist spills, repel and release stains, and resist static.47 These textile enhancements are increasingly an inherent property of the fabric, improving the performance and durability of everyday apparel and interior furnishings.

4.2.6. Integrated Components Organic semiconductors (polymers and oligomers) having the electrical properties of semiconductors and mechanical properties of plastics are good candidates for developing electronic and optoelectronic flexible components, e.g., transistors, LEDs, that are compatible with a flexible textile substrate.38,48 For example, Bonfiglio et al. developed a structure for organic FETs (field effect transistors) based on thin film technology that is suitable for integration with flexible substrates. The approach employs an insulating film coated with gold on one side and a thin film of organic semiconductor on the opposite side. This was then glued to the textile ribbon with source and drain contacts created by simply crossing the ribbon with two parallel gold wires.49 Another approach Bonfiglio’s group has taken is to build organic field effect transistors in a fiber format rather than the conventional planar geometry, and they demonstrated that the resulting cylindrical organic thin film transistor could be used in a textile process such as weaving or knitting.49 Organic LEDs consist of multilayer structures where organic emitters are embedded between an evaporated metal electrode and a film of indium tin oxide coupled to a plastic or glass substrate. Philips recently released a light-emitting fabric, Lumalive, featuring flexible arrays of fully integrated colored LEDs. These light-emitting textiles can present dynamic messages, graphics, or multicolored surfaces50 and could also provide a basic substrate for building wearable optical chemo-/biosensors.

4.2.7. Wearable Actuators In order to create fabrics with motor functions flexible actuating devices are needed. One method of achieving this is through the use of shape memory fabrics. These are materials that are able to return to a preprogrammed shape under an external stimulus, normally temperature. Two approaches are used to make these materials, shape memory alloys (SMAs) or shape memory polymers (SMPs). In the

case of shape memory alloys, at a low temperature the structure of the materials changes to a martensite phase, where they can be easily deformed. Upon heating, the structure changes to an austenite phase, and the programmed shape is recovered as the material “remembers” and restores its original shape. SMAs may be spun in combination with traditional fibers to create bicomponent yarns which can then be woven or knitted. SMPs on the other hand have higher extensibility, superior processability, lower weight, and a softer handle and therefore may be considered more suitable for the clothing industry. By coating a fabric with SMP it is possible to make “non-iron” garments that can return to the original shape when heated. Mitsubishi Heavy Industries have applied this technology to create a fabric, DiAPLEX, for the outerwear industry. DiAPLEX has an ultrathin nonporous SMP membrane that has the ability to open up its microstructure when the temperature increases, allowing heat and water vapor molecules to escape through the membrane. A similar approach has been taken by the Centre for Biomimetics at the University of Bath, where SMPs have been molded into biomimetic structures. A fabric surface mimicking the pine cone structure contains scales on the fabric surface that open and close in response to relative humidity or heat.51 Another approach to thermoregulation has been demonstrated by Corpo Nove’s shape memory shirt which incorporates SMAs to cause the sleeves to shorten when the temperature increases.52 Another technique is to use ICPs which have the ability to function in both sensing and actuation modes. The actuation property of ICPs results from a volume change of the polymer which accompanies oxidation and reduction (Figure 3). An applied positive potential leads to removal of electrons from the polymer backbone and incorporation of dopant ions to maintain electrical neutrality. The positive charges on the polymer backbone provide Coulombic repulsion forces between polymer chains, and this, together with the incorporation of dopant ions and associated water of hydration, leads to an increase in the polymer volume. This process can be reversed in a controlled fashion to produce usable mechanical work.53 ICP-based mechanical actuators can achieve average stresses ∼10-20 times those generated from natural muscle,54 realize strains of over 20% comparable to that of natural muscle,55 and achieve fast free-standing beam actuation with an operational frequency up to 40 Hz.56 The most attractive feature of ICPs is their ability to emulate biological muscles with high toughness, large actuation strain, and inherent vibration damping.53 This similarity has led to the term “Artificial Muscle” being applied to these materials, and in principle, they offer the potential of developing

664 Chemical Reviews, 2008, Vol. 108, No. 2

biologically inspired robots.57 While performance and longterm stability of these actuators still needs to be improved, there is potential to develop rehabilitation and orthotic devices.57,58 Researchers at the Intelligent Polymer Research Institute in the University of Wollongong employ ICP actuators to assist the insertion of the Cochlear Implant electrode. In this application, a bilayer polypyrrole actuator is used to steer or bend the Cochlear Contour electrode in a controllable manner during surgery as it is being inserted into the ear.58

4.2.8. Interconnects and Infrastructure The fundamental components within smart textiles are sensors, actuators, and control units. Obviously reliable links between the various components of a smart garment system are essential for wearable electronics, between the sensors and actuators incorporated in the garment, and also to external control and communication devices. In cases where conventional electronics are embedded into fabrics, interconnection strategies have included soldering, bonding, stapling, and joining.33 (i) Soldering: Surface-mount components have been soldered directly onto conductive fabric such as metallic organza. The problem with this technique is the toxicity of solder compounds, making them unsuitable for constant contact with a user’s body. Aside from that, while it may give good electrical contact, the mechanical strength of the connection does not withhold typical strains exerted within a flexible garment. (ii) Bonding: Components are bonded to a substrate using conductive adhesives. Adhesives are more suitable than solder, being nontoxic, highly conductive, highly durable, and moderately flexible (iii) Stapling: Component leads grips a sewn conductive trace, within a conductive stitched circuit, by being pressed into shape around it. However, when the substrate flexes, it is likely to stretch open pins that have been formed into clasps and accelerate wear and tear of the fabric substrate. (iv) Joining: Component leads are joined directly to a stitched fabric circuit, where components are formed with a single conductive thread per pin. The threads may be attached to the substrate by covering stitches of conductive threads, known as e-broidery.33 Infineon used a similar approach to connect their components. They use narrow fabrics with woven copper threads that are coated with silver and polyester. Where electrical connections are made the coating of the thread is removed using a laser technique and soldered to bonding wires connected to a component or alternatively to a thin flexible PCB.59 The electronic device is then encapsulated for protection and to allow laundering. This approach has been used to integrate an MP3 player into clothing using the conductive fabric bands to provide connections to headphones, battery power, and keypad. Conductive textiles have been shown to be suitable for data transmission. Starlab Research Laboratories are working on fabric area networks (FANs). Magnetic induction with textile coils can effectively bridge distances less than 2 cm, allowing nodes of a communication network to be interspersed within the wearer’s garments and accessories. Antennas are routed to the trouser pockets, shirt pockets, cuffs of the trousers, sleeves, back of the shirt, and other locations. These antennas can then be used to communicate with transponder chips that are embedded in the wallet, shoes,

Diamond et al.

pens, watches, accessories, or personal items at the backpack, allowing intergarment communication within the body area network.60 Winterhalter et al. demonstrated the use of fine filament copper wires (<50um) within narrow woven fabrics to create a fabric version of the universal serial bus (USB) and also a radiating conductor for body-borne antenna applications. These devices were developed with military applications specifically in mind, where high performing and durable materials are needed. The fabric USB was tested before and after exposure to abrasion and cyclic loading and met selected test requirements of USB 2.0, while the fabric antenna demonstrated communication ranges compatible with military communications equipment.61 To accommodate communication using wireless technologies, flexible antennas need to be integrated into the textile material. The Wearable Computing Laboratory at ETH Zu¨rich has developed textile patch antennas for Bluetooth applications in wearable computing using the frequency range around 2.4 GHz.62 The antenna is a planar structure with a maximal thickness of 6 mm, based on the technology used to create printed microstrip antennas. The textile antennas use an electrically conductive fabric (metallic plated) for the ground planes as well as for the antenna patches. A fabric substrate, using wool/polyamide spacer fabric, is used as the dielectric between the antenna patch and the ground plane. Textile antennas have also been investigated at the University of Ghent to allow communication outside the garment, e.g., between the garment and a central control unit. A singlefeed rectangular-ring textile antenna is proposed for wireless body LANs operating in the 2.4-2.483 GHz band. The conductive parts of the planar antenna consists of FlecTron fabric, which is a thin, flexible, and lightweight copper-plated nylon woven fabric, whereas fleece fabric is used as nonconductive antenna substrate.63 Power requirement is a critical issue for wearable sensor networks. Lithium ion batteries may be used, and in coin cell format they may be integrated relatively easily for many applications, but the ideal format is to use flexible batteries that are integrated into the garment. For example, Powerpaper is thin and flexible with zinc anode and manganese dioxidebased cathode layers. The cells are made of proprietary inks that can be printed or pasted onto virtually any substrate to create a battery that is thin and flexible.64 Alternatively, fiber cells may be more suitable for textile integration. ITN energy systems are working with the U.S. agency DARPA to make batteries with fibers containing the anode, cathode, and electrolyte by coating a fiber with thin film layers, consisting of the same materials typically used in flat batteries, such as LiCoO2 as the cathode, lithium as the anode, and LiPON as a solid electrolyte. To be more energy efficient and have the potential of an endless power supply, harvesting energy from various available energy sources would be of huge benefit. Starner summarized the harvesting of energy from the wearer’s everyday actions, mainly leg motions and body heat;65 see also an overview of energy scavenging by Yeatman and Mitcheson.66 Piezoelectric transducers (converting mechanical energy), thermoelectric transducers (converting thermal gradients), and pyroelectric transducers (converting body heat) can provide power that is always available, thus removing the need for battery recharging. Solar cells may also be employed as garments have a relatively large surface area that can be employed for solar energy harvesting. Photovoltaic (PV) technology commonly

Wireless Sensor Networks and Chemo-/Biosensing

Chemical Reviews, 2008, Vol. 108, No. 2 665

Table 6. Physiological Signals and Possible Sensing Platforms for Wearable Applications physiological parameter

signal source

ECG EMG breathing rate

electrical activity of the heart electrical activity of a muscle thoracic volume changes

blood pressure

force exerted by circulating blood on the walls of blood vessels

pulse oximetry GSR body motion

oxy-hemoglobin concentration in blood skin electrical conductivity limb movements/gestures posture

employs thin film materials based on amorphous silicon (ASi) or copper indium gallium diselenide (CIGS). The former, which is the most common solar cell material for commercial application, suffers from being relatively heavy, fragile, rigid, and costly to ship and handle.30 Recently CIGS thin film technology has been used in the development of a solarpowered jacket to connect and charge portable devices such as cell phones, PDAs, and MP3 players, thus providing power on demand.67 Wearable networks must also provide user input/output peripherals. Switches that are fabric based and fully washable and maintain the comfort and feel of textiles are being integrated into garments to control electronic devices such as MP3 players. Eleksen use capacitive pressure sensors to create control buttons on fabric and have even developed a fabric keyboard for PDAs that can be rolled out, easily stored, and transported. SOFTswitch also manufacture fabric switches, keypads, and pressure sensors utilizing a quantum tunneling composite (QTC) provided by Peratech. QTC is a material that is normally a very good electrical insulator but turns into a conductor under compression, tension, or twisting. QTC contains tiny metal particles insulated within a polymer. When deformation occurs the metal particles are brought closer together and electrons pass through the insulation by a process called quantum tunneling. The transition from insulator to conductor follows a smooth and repeatable curve with the resistance dropping exponentially. In addition to providing user input through such textile peripherals, output display is another important aspect for the wearer and must be provided in a comfortable accessible format. On-garment visual feedback to the wearer can be provided via flexible displays based on liquid crystal technology or organic LEDs.50,68

4.3. Applications of Wearable Sensors Monitoring vitals signs such as heart rate and breathing rates can give an insight to the wearer’s health. By detecting limb movements it is possible to study body kinematics, an important physiology tool in sports applications to improve technique and performance, and also as a rehabilitation tool, e.g., post stroke patients. In sports science, wearable sensors enable coaches and athletes to understand how an athlete’s body responds to exercise and track improvements in performance. In the healthcare sector wearable sensors allow continuous monitoring of a patient’s condition, providing the opportunity to identify threats before an event happens, e.g., assessing cardiovascular disease. This is a rapidly growing research area with many new technologies and applications emerging. Table 6 lists a number of physiological signals of valuable diagnostic importance that have been successfully monitored

suitable sensors for wearable applications woven/knitted metal electrodes70,72 woven/knitted metal electrodes91 electrical impedance plethysmography,42,76 piezoresistive strain/ pressure sensors151 volume clamp method with finger cuff actuator,152 measurement of pressure oscillations,153 pulse transit time91 optical sensors (LEDs and photodiode)154 and optical fibers82 woven metal electrodes75 piezoresistive strain/pressure sensors,155 accelerometers, gyroscopes,78 optical fiber sensors70

using wearable sensors. As part of any physiological assessment breathing rate is an important parameter as it is closely linked to our physiological and psychological state, even though we are often unaware of changes in our personal breathing pattern. For sports applications breathing rate is equally an important measure, giving a good indication of physical exertion or assessing metabolic capabilities. Further to breathing rate, respiration can be monitored by measuring changes in thoracic volume. This can be done using wearable strain gauge sensors or electrical impedance plethysmography. Electrical impedance plethysmography integrates two conductive wires into a garment, one around the ribcage and the other around the abdomen. Motions of the chest wall cause changes of the self-inductance of the two loops. Expansion of the thoracic cavity during inhalation causes an increase in conductivity of the material. This allows the rate and amplitude of breathing to be monitored, e.g., deep breathing manifests as a large change in signal amplitude and slow breathing rate. Electrical signals such as electrocardiograph (ECG) and electromyography (EMG) may be measured by integrating textile-based electrodes into fabric. ECG typically uses silver chloride electrodes coupled to the skin with gel to measure electrical potentials. Flexible conductive yarns, fully metal yarns, or natural/synthetics blended with conductive fibers have been knitted into garments to develop textile electrodes, referred to as textrodes.69,70 Carbon or conducting polymerbased yarns are not currently used as they are not conductive or sensitive enough for this application.38 Figure 4 shows a comparison of fabric electrodes against a standard system in a study by Paradiso et al.69,70 and shows good correlation between the two systems. EMG uses surface electrodes to detect stimulation signals of the muscle fibers and has also been measured using metal-based yarn electrodes.70 However, a major issue (particularly with ECG electrodes) is the need to use a gel to provide a good contact between the sensor and the skin. This restricts the ease of use, and in recent years, much effort has been focused on the development of ‘dry’ electrodes and the use of hygrogel membranes that can be integrated into a garment and used for ECG monitoring without having to manually apply a gel.42,71,72 While electrodes may be integrated as detectors, another possibility is to use the electrodes to stimulate muscles. Electrodes for functional electrical stimulation (FES) have been integrated into fabrics to provide actuation stimuli to muscles of spinal cord injured and stroke subjects in order to generate or improve lost motor function, e.g., for walking or hand gripping movements.73 Strain sensors made from piezoelectric materials may be used in biomechanical analysis to realize wearable kinesthetic

666 Chemical Reviews, 2008, Vol. 108, No. 2

Diamond et al.

Figure 4. Comparison of ECG signals obtained with subject walking on the spot with standard and fabric electrodes. (Reprinted with permission from Paradiso, R.; Loriga, G.; Taccini, N. IEEE J. Inf. Technol. B. 2005, 9, xxxx. Copyright 2005 IEEE.)

Figure 5. (a) SEM of uncoated polyurethane foam (×100). (b) SEM of polyurethane foam coated with polypyrrole (×100). (c) Foambased polypyrrole pressure sensors integrated into an insole for gait analysis; the sensors are connected to a microprocessor and Crossbow wireless transmitter.

interfaces able to detect posture, improve movement performance, and reduce injuries.42 Garments integrating piezoresistive electroactive polymers (EAPs) and conductor-loaded rubbers with strain-sensing capabilities have been demonstrated for continuous monitoring of body kinematics and vital signs.38,74 Such devices may be used to teach athletes the correct way to perform movement skills by providing real-time feedback about limb orientation. For example, the Intelligent Knee Sleeve is a biofeedback device using PPy fabric sensors that monitor the wearer’s knee joint motion during jumping and landing and uses an audible feedback signal to reinforce the correct landing technique signs.38,74 Likewise, a foam-based PPy sensor has been used to detect joint movement and breathing function75,76 and applied to examine pressure distribution within a shoe insole77 (Figure 5). The insoles each have two sensors to monitor the

difference between the heel strike and front foot lift off. An example of the measured response during walking is shown in Figure 6. This provides a way to characterize different movement activities, such as distinguishing normal walking steps from shuffling of the feet, which can help with the early diagnosis of conditions such as Parkinson’s disease. Pressure-sensitive fabrics may also be used to prevent pressure sores which can lead to ulcers, which can happen if a person remains in the same position for long periods of time, or where sensation/feeling has been lost in a limb, which is a common problem for people suffering from diabetes. The monitoring of body posture and gesture has been demonstrated using fabric strain gauges; another approach to measure posture has been demonstrated by Dunne et al., where a fiber-optic bend sensor has been integrated into a

Wireless Sensor Networks and Chemo-/Biosensing

Chemical Reviews, 2008, Vol. 108, No. 2 667

Figure 6. Normalized response for two PPy-coated foam sensors placed at the heel (RT Heel Sensor) and toe (RT Toe Sensor) position of an insole.156 The time difference between heel strike and toe lift off is demonstrated during walking motion and may be used to analyze gait patterns for rehabilitation purposes or for the assessment of conditions such as Parkinson’s disease.

garment for long-term monitoring of spinal posture in the working environment where poor seated posture is an increasingly significant source of back problems.78 Another method to monitor body posture and motion is to use triaxial accelerometers mounted on different parts of the human body, as demonstrated by Farella et al., where the accelerometers form nodes of a wireless and wearable network for posture recognition.79 Accelerometers have been widely used in body sensor networks for a wide range of applications, investigating aspects such as sleep-wake rhythm and hyperactivity disorder to locomotor activity rhythms in Alzheimer’s disease.29 Optical components may be used to investigate blood oximetry. The absorption of light depends on oxy-hemeoglobin saturation. Tissue may be probed using LEDs and photodiodes that are coupled to the skin. Asada et al. demonstrated a reflectance prototype oximetry measurement system including a rf data transmission unit miniaturized in a finger-ring configuration.80 The possibility of measuring blood pressure using photoplethysmography techniques has also been investigated.81 There is also potential for monitoring emotional, sensorial, and cognitive activities, as demonstrated by the MARSIAN smart glove. This monitors the autonomic nervous system, which is responsible for our body’s involuntary vital functions, by measuring physiological parameters of the skin.82,83

4.4. Wearable Chemosensing Most research to date in the field of wearable electronics has focused on physical transducers for reasons explained previously. These issues, including sampling procedures, calibration, safety, and power, also apply to wearable sensors with the added constraint that the sensors must not compromise the functionality or comfort of the garment. As for normal chemo-/biosensing, a sample must be delivered to an active surface on the sensor for a reaction to occur and the signal to be generated. If the wearable sensor is monitoring the external environment, the sample is likely to be volatile (e.g., presence of hazardous/toxic gas), whereas

if the sensor is monitoring the body’s physiology, a liquid sample must be delivered to the sensor. For wearable chemo-/ biosensors, obtaining a sample is an issue as this ideally should happen naturally, and for this reason attention has focused on sweat rather than blood. In the case of body fluids, sample collection and delivery ideally should be incorporated within the layers of the garment, whereas external gaseous sampling requires the sensor to be positioned at the outer edge of the garment interface, where contact between the sensor and the sample will be optimal. The sensor itself must be robust, miniature, flexible, washable, and ideally textile based. The overall assembly of sensor must be safe for the wearer’s health, and the use of toxic or hazardous reagents should obviously be avoided. BioTex and ProeTex are two EU-funded projects that aim to develop textile-based sensors to be integrated within a garment for monitoring chemical and biological targets.3,84 For Biotex, the goal is to place the sensor within a fluid handling platform that employs the inherent capillary action of certain fabrics. For example, polyamide Lycra has natural moisture-wicking properties and is used in conjunction with a super absorbent material to control fluid transport in the fabric. The configuration of the device is shown in Figure 7. It consists of a collection layer with a hydrophilic channel surrounded by hydrophobic silicone-coated regions. This channel contains an inlet for sweat, a pH-sensitive region, and an absorbent region. A cover is held 5 mm above the channel by moulded silicone, which prevents contamination, blocks stray light, and positions the optical sensing components. The color of the immobilized pH-sensitive dyes is monitored by reflectance colorimetry using LEDs with the measurements controlled by a Mica2dot mote which is also responsible for wireless transmission of the data to a remote base station.84 Much work has been carried out to develop wearable sensors for the management of diabetes. This condition requires the patient to monitor glucose levels closely and at regular intervals, typically using finger-pricking sequential blood tests. A wearable glucose sensor would be a real

668 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 7. Textile-based device for collection and analysis of sweat. The device uses the inherent moisture wicking and absorbance properties of fabric to control fluid movement. A pH indicator is used to demonstrate colorimetric sensing capabilities using a paired LED configuration to perform reflectance measurements from the fabric surface.

benefit to the millions of people suffering from this disease provided it could meet the analytical requirements, provide hypo- and hyperglycaemic alarms, and meet the wearable sensor requirements summarized above. The Glucowatch (Figure 8) was one of the first approaches to noninvasive glucose monitoring and approved by the FDA in 2001. The device extracts glucose through intact skin via reverse iontophoresis, where it is detected by an amperometric biosensor. In a clinical evaluation study the device was reported to give accurate measurements continuously (3 per h) over a 24 h period.83 Another approach developed by Badagu et al. is a disposable contact lens embedded with newly developed boronic acid-based fluorophores.85 The contact lens changes color in relation to the amount of sugar in tears and can be monitored by the wearer simply looking into a mirror and comparing the color to a precalibrated color strip. Kudo et al. developed flexible biosensor for glucose measurement using functional polymers. The biosensor utilizes the physical and chemical functions of hydrophobic polydimethyl siloxane (PDMS) and hydrophilic 2-methacryloyloxyethyl phosphorylcholine (MPC) copolymerized with dodecyl methacrylate (DMA). The glucose sensor was constructed by immobilizing glucose oxidase (GOD) onto a flexible hydrogen peroxide electrode (Pt working electrode and Ag/AgCl counter/reference electrode). The electrodes were fabricated and integrated with the functional polymers using microelectromechanical systems (MEMS) techniques.86 SCRAM (secure continuous remote alcohol monitoring) is a tamper- and water-resistant bracelet containing an electrochemical sensor that is attached to the ankle using a durable strap. Targeted at drunk-driving offenders, the device captures transdermal alcohol readings from continuous samples of vaporous or insensible perspiration collected from the air above the skin. A correlation exists between the alcohol concentration in blood and perspiration, although there is a time difference in the response. The system contains a flash memory chip to store alcohol readings, a circumvention detection device to monitor body temperature and detect tampers, and uplink features that can transfer these readings via a wireless radiofrequency (rf) signal to the SCRAM modem. At scheduled times set by the court or probation agency, the anklet will transfer these data to the modem.

Diamond et al.

Fiber optic sensors with modified cladding materials have been shown to be suitable for detecting hazardous battlefield situations and may be easily integrated into soldiers’ uniforms. The modified cladding materials are sensitive to different environmental conditions and cause a change in the refractive index, which affects the propagation of the transmitted light signal. El-Sherif et al. demonstrated this using a thermochromic agent, segmented polyurethanediacetylene copolymer, and a photochemical polymer, PANi, as cladding agents.37 This type of sensing has many uses in military applications to provide protection and advance warning to the troops regarding chemical and biological warfare threats or above-normal temperatures. Another chemosensor developed for military applications is Chameleon, developed by Morphix Technologies. It is a wearable armband integrating colorimetric sensing techniques to detect toxic gases like ammonia, basic gases, chlorine/fluorine, hydrogen sulfide, iodine, phosgene, phosphine, and sulfur dioxide via a range of disposable cassettes.87 Collins and Buckley demonstrated the chemical-sensing properties of conductive polymers coated onto woven fabric materials.88 Thin films of conductive polymers, polypyrrole or polyaniline, were coated onto poly(ethylene terephthalate) or nylon threads woven into a fabric mesh. The conductive polymer overlayers were grown by chemical polymerization or oxidative coupling of the monomer, pyrroles, or aniline. These sensors were assessed for their ability to detect hazards that may endanger the health of the wearer. Low ppm detection limits were demonstrated for toxic gases such as ammonia and nitrogen dioxide as well as the chemical warfare simulant dimethyl methylphosphonate, DMMP. The chemical-sensing fabrics were reported to have the following advantages over the use of microelectronic devices: (i) potentially low cost and low operating power; (ii) commercial availability of the coated fabric; (iii) expected wide dynamic range due to the large surface area of a conductive polymercoated piece of fabric; (iv) ease of measurement; and (v) demonstrated stability of the conductivities of coated fabrics. It was claimed that these electroactive polymer-coated fabrics could form the basis of future wearable, lightweight, clothingintegrated sensor systems to detect external hazardous environments.

4.5. Applications in Personalized (p)Health The concept of personalized health, or pHealth, seeks to empower the individual with the ability to manage and assess their own state of health and healthcare needs. Wearable sensors enabling monitoring of long-term and short-term healthcare needs have a huge potential for disease prevention in medicine and early diagnosis. They facilitate assessment of the impact of clinical intervention by gathering data in the home and community setting. This can overcome the problem of infrequent clinical visits that may fail to detect transient events that may be of important diagnostic importance. Another problem with hospital-based healthcare is that it is impractical to measure physiological responses during normal periods of rest, activity, and sleep and monitor circadian variations in physiological signals without committing people to specialist facilities for days or weeks.89 Early diagnosis through long-term trend analysis reduces the potential severity of an illness, and in the case of rehabilitation, a wearable sensing system could monitor the recovery process and detect complications as they arise. Providing feedback on health and well being through a body-sensor network in principle allows the patient to take

Wireless Sensor Networks and Chemo-/Biosensing

Chemical Reviews, 2008, Vol. 108, No. 2 669

Figure 8. Glucowatch uses reverse iontophoresis to pull glucose through the skin into two gel collection discs (A). The glucose level is measured, and a reading is displayed to the user (B). (Reprinted with permission from ref 83. Copyright 2001 Elsevier.)

Figure 9. On-body sensors and electronics for monitoring vital body signssa garment with textile ECG electrodes developed by the MyHeart project to prevent heart failure. (Permission to use this from Philips.)

care of his/her own health at home. For people at risk of disease, the system should facilitate responses to reduce individual risk factors, e.g., hypertension, obesity, diabetes, stress, and physical inactivity. In the occurrence of complications, the wearer may be alerted to contact their physician or a message can be sent automatically.70 For those suffering from chronic illnesses, wearable sensors could facilitate selfmanagement of the disease by facilitating personalized drug treatments and allowing self-administration of medications, e.g., in the case of Parkinson’s disease, medication is often over- or underestimated based on population averages rather than individual statistics. Another potential application is epilepsy seizure detection by combining contextual activity regarding general activity with physiological measurements.90 Depression is also another chronic illness where monitoring is needed to assess the effect of treatment. Overall the aim is to enable a personalized medication and rehabilitation schedule through continuous long-term monitoring and profiling of individuals. There are a number of personalized healthcare systems that have reached or are nearing commercialization. Vivometrics developed the Lifeshirt which is a Lycra vest with sensors to detect respiration, ECG, posture, and movement. Textronics have released the NuMetrex heart-rate-monitoring sports bra that incorporates conductive knitted sensors that link wirelessly to a Polar heart-rate-monitoring watch. The Smartshirt by Sensatex, which evolved from the wearable motherboard shirt initially developed at Georgia Institute of

Technology, integrates optical and conductive fibers with embedded sensing elements to monitor vital signs. The EU projects WEALTHY and MyHeart have developed sensorized cotton/Lycra shirts which integrate carbon-loaded elastomer strain sensors and fabric bioelectrodes, enabling the monitoring of respiration, electrocardiogram (ECG), electromyogram (EMG), and body posture and movement.70,91 Figure 9 shows the system developed by MyHeart for on-line monitoring of vital signs. Foster-Miller Inc. is developing and commercializing an ambulatory physiological status monitoring system to monitor heart rate, respiration rate, posture, activity, and skin temperature.92 BodyMedia are working on a body-monitoring system in the form of an armband as a physiological monitoring and calorie management system. The target users of all of these products include doctors, researchers, fitness professionals, nutritionists, and general consumers. As mentioned previously, the first mass consumer-wearable sensor product is the Nike-Apple iPod Sports kit in which a sensor placed in an exercise shoe relays information to an iPod. Given the market potential for wearable sensors evidenced by this product, it is clear that a series of new products will be launched in the next couple of years, and there are obvious opportunities for wearable chemo-/biosensors that can provide important health-related information within this scenario. The Institute for Soldier Nanotechnologies (ISN, see http:// web.mit.edu/ISN/) is an interdepartmental research center based at MIT whose research mission is to use nanotech-

670 Chemical Reviews, 2008, Vol. 108, No. 2

nology to dramatically improve the survival of soldiers. Their vision of the battle suit of the future is a bulletproof jumpsuit no thicker than ordinary Spandex that monitors vital signs, eases injuries, communicates automatically, and allows medics to conduct a remote triage of combat casualties to help them respond more rapidly and safely. The EU Framework project ‘ProeTex’ is developing textile- and fiber-based integrated smart wearables for emergency disaster intervention personnel in order to improve their safety and efficiency (see http://www.proetex.org/). By empowering them with wearable sensing and transmission systems, it is envisaged that vital signs, activity, position, and external environment can be monitored with the information and status indicators relayed to the individual and also to a central monitoring unit. The system monitors the vital signs of emergency disaster workers and warns of potential dangers, allowing communication between the individuals in the operations area and also with the support services outside the operations area. The aim is to provide active measures to prevent or reduce injury from burns, flashover, hyperthermia, explosion, burns by acids and corrosive substances, electrocution, and detection of NRBC (Nuclear, Radiological, Biological, Chemical) risks.

4.6. Conclusions Wearable sensors provide the ability to enable innocuous sensing of the wearers and their environs. The challenge is to create sensors, driving electronics, and communication networks that are small, robust, reliable, and flexible so as not to impair the comfort of the wearer. While miniaturization of components has been of immense benefit in wearable sensing applications, the major problem of interconnections still remains, where conventional silicon and metal components are highly incompatible with the soft textile substrate. To help in overcoming this there has been much research and development of functionalized fabrics that can provide conductivity and sensing components to our clothing. Nanotechnology is making a huge impact within the textile industry and novel materials, such as conducting polymers, allow the fabrics to retain their natural flexible feel, which is imperative for “smart clothing” to become a realistic part of day to day life. Body sensor networks allow the wearer to simultaneously monitor their personal health and status of their immediate environment and share this information with others via the Internet. Developments will be very rapid in the coming years, and as the user base expands, there will be a demand to integrate additional information from chemo-/biosensors, both in wearable formats and through small analytical kits or instruments (e.g., personal diagnostics). It is important therefore that the chemo-/biosensors research community is aware of these developments and emerging opportunities and focus on ensuring that technological barriers to this integration are overcome. In the next section we shall examine the role of materials science in making this happen.

5. Materials SciencesThe Future Materials science is a very broad interdisciplinary field that spans many areas of scientific research with recent exciting developments arising from our increasing ability to control the behavior of bulk materials through manipulation at the molecular level. The sections above have outlined

Diamond et al.

developments in sensor networks in two main areas: environmentally-deployed mote-based sensor networks and wearable sensors. Both are linked in that they arise from the concept of pervasive computing, every person will be continuously linked to digital communications systems through wireless communications (infrastructure essentially already in place), and through this people will be increasingly linked to their environment, for example, to networks of objects and services, themselves (personal vital signs monitoring), and other people (for general background on this, see http://www.itu.int/osg/spu/publications/internetofthings/). However, the real challenge is to link ‘molecular’ sensing through chemo-/biosensors to the networked world. From the previous sections it is clear that while certain building blocks for performing ‘pervasive’ sensing are being assembled, we are still a long way from the realization of largescale sensor networks, even for relatively simple targets like temperature. The conclusion from those sections is clear: large-scale deployments of sensors can only happen when the sensor nodes are essentially self-sustaining in terms of all consumables (energy, reagents, etc.) for many years. Chemo-/biosensing is inherently much more complex than the type of sensing currently employed in these demonstration deployments and invariably involves liquid handling using conventional technologies (pumps, valves, etc.) which are very energy intensive. Furthermore, chemo-/biosensing requires intimate binding events or reactions to occur either in solution (e.g., with reagent-based systems) or at the surfaces of sensing devices. Thus, chemo-/biosensors are prone to degradation as they depend on active materials and surfaces that interact with the sample, and therefore, they must be regularly calibrated to compensate for changes in their operating characteristics over time. While considerable advances have been achieved, for example, with the move toward microfluidics and lab-on-a-chip devices (see below), the scale of the improvements required to meet the demands of pervasive sensing are such that a complete rethink of ways to perform chemo-/biosensing is probably needed. For the remainder of this review we shall focus on two critical areas where revolutionary breakthroughs could have a potentially revolutionary impact: liquid handling and control of surface interactions.

5.1. Microfluidics and Lab-on-a-Chip Devices The concept of ‘micro-total analysis systems’ or µTAS was introduced by Manz et al. around 1990,97 and it rapidly became known as lab-on-a-chip (LOAC). At its core the concept involved integrating multiple separate operations (sampling, sample processing, reagent addition, calibration, detection, etc.) into a compact, integrated, microfluidic manifold, which had channel dimensions typically in the range 10-500 µm, to both speed up sample throughput and improve precision. This has become a huge area of research in its own right, and for general information the reader is referred to other recent excellent reviews.98-100 In this section, we will restrict ourselves to the potential of LOAC devices for distributed sensor networks based on wearable and motebased environmental platforms. One of the main limitations of LOAC devices arises from difficulties in finding liquid handling/transport approaches that are truly compatible with scaled down microfluidic manifolds. Most published work involves the use of benchscale pumps and external valves with emphasis being on the

Wireless Sensor Networks and Chemo-/Biosensing

improved sample-handling capabilities of the integrated microfluidics system and reduced reagent consumption/waste generation. Furthermore, attempts to produce integrated systems typically involve the use of scaled-down conventional pumps/valves produced by micromachining. While this has produced some very elegantly engineered systems, there is a growing realization that the next evolution of these devices requires completely new approaches to integration of liquid-handling functions and that the solution to this issue lies in fundamental materials science and new ways of thinking about microfluidics. For example, Whitesides and co-workers demonstrated valve effects using distortions in soft polymers,101 while others propose more biomimetic solutions in which the microfluidic manifold is regarded as a primitive mimic of a biological circulation system, and biologically inspired routes to generating fluid circulation are employed, for example, using the ‘artificial muscle’ function exhibited by some redox-active soft polymers. Wallace et al. produced a pump based on the expansion/ contraction cycle accompanying oxidation/reduction of polypyrrole deposited on a water-permeable tube. Application of an oxidizing potential at one end of the tube generated a peristalsis-type expansion wave that migrated along the surface, carrying water with it.102 Recently, a biomimetic soft polymer pump was demonstrated based on polypyrrole Nafion actuators compressing a soft-walled polyurethane/ polydimethylsiloxane chamber.3 These pumps offer advantages in terms of relatively low energy requirements and are ultimately more compatible with microfluidics as their soft nature makes them much less prone to physical malfunction due to, for example, ingress of microparticulates, than conventional microengineered pumps and valves. For an introduction to biomimetics related to sensors, actuators, and artificial muscles, see ref 23 and the references therein. LOAC devices, in principle, offer a route to incorporation of sophisticated chemo-/bioprocessing steps in a compact, low-power platform, which is attractive for remote sample processing (phase extraction, separation steps, reagent additions, calibration, etc.). Therefore, these devices offer a compromise between existing lab-based instruments and the vision of tiny, completely self-sustaining sensors capable of massive scale up. The limiting factor is now probably the need for reagents and waste storage/disposal. Ideally, these devices should be able to generate their own reagents on demand (e.g., electrochemical generation of protons or hydroxide ions to provide localized control of pH), but we are still a long way from this idealic vision of completely self-sustaining chemo-/biosensing devices.

5.2. Controlling Liquid Movement in Surfaces and on Channels Providing the reagents within the LOAC device is only part of the story however. Reagents must be stable in longterm storage to be of any use (at least months, ideally years).27 Furthermore, the manifold design must incorporate structures and features that control the flow characteristics by providing effective mixing regions to overcome the dominant laminar flow behavior. For example, a recent report described how efficient mixing can be achieved in LOAC manifolds converting the flowing liquid into discrete droplets that can be manipulated by electrowetting. The interfacial tension of droplets is controlled with the application of a voltage across an electrode array and the droplets behave as micromixing chambers while moving across the array.98

Chemical Reviews, 2008, Vol. 108, No. 2 671

Other approaches exploit colloidal particles99 and flexible elastomers, that can open and close a microfluidic channel.100 A hydrophilic microfluidic network with geometric valves that can slow down, stop, and accelerate liquid volumes within the microfluidic system has also been demonstrated.103 Clearly, a fundamental issue with all chemo-/biosensing platforms is how to control the movement of liquids in a way that is virtually energy free, so that functions like calibration can be incorporated. While considerable improvements can be obtained through downscaling analytical instruments using microfabrication approaches and this will lead to more widespread deployments of autonomous analytical instruments, for example, in environmental monitoring104 or hazard/threat detection,105,106 for truly pervasive chemo-/biosensing we need to go beyond downscaling of conventionally engineered devices like pumps and valves. Control of the wetting properties of materials has been the focus of considerable research effort for many years, particularly by groups linked with the textile industry.93,94 The ubiquitous presence of water in nature makes this study extremely relevant, yet despite the fact that many critical processes in nature, including sensing, occur at surfaceaquo interfaces, the fundamental structure of water and its behavior at surface interfaces is still not fully understood.95 Classically, materials are divided into two categories in terms of their degree of interaction with water: hydrophilic and hydrophobic.96 Surface tension can enable liquids to move spontaneously over surfaces without any application of external forces.107 Wetting and adhesion properties between liquids and solids can be described by the measure of the contact angle between the two surfaces: the larger the wetting tendency, the smaller the contact angle or the surface tension. Surface tension changes, caused by evaporation of the alcohol, are responsible for formation of drops of wine on the surface of a glass.108 This is generally known as the Marangoni effect,109 and it describes the spontaneous movement of liquid from a low surface tension to a high surface tension environment. Unbalanced surface tension forces can drive liquid motion without any external intervention of any kind. It has been demonstrated that the speed of liquid flow driven by surface tension can be increased by a factor of hundreds or thousands times compared to the typical Marangoni effect by creating a radial surface tension gradient through the deposition of hydrophobic molecules in the center of a hydrophilic surface and subsequently forming water droplets by condensing steam on the hydrophobic region.110 The drops driven by the gradient forces from one side and coalescence forces from the other side were found to move at speeds of up to 1.5 m/s. While the immediate applications of this effect are found in heat exchangers, it has an important role in clinical conditions such as in the treatment of respiratory distress syndrome (RDS), in which the lungs of prematurely born infants are not sufficiently developed to produce enough surfactant to control liquid distribution, and hence effective gas exchange is impaired.111 These phenomena demonstrate that it is possible to use purpose-designed chemistries to control water behavior (including transport) across or even through surfaces. For example, by generating a spatial surface tension gradient on a surface the movement of a water droplet uphill without the assistance of any external energy has been demonstrated.112 The gradient surface was created by diffusing decyltrichlorosilane vapor over a silicon wafer; evaporation of the silane vapor created a gradient of

672 Chemical Reviews, 2008, Vol. 108, No. 2

concentration that decreased along the length of the surface. As a consequence, the edge of the wafer surface closest to the vapor source became hydrophobic and the farthest hydrophilic. When a water droplet is placed on the obliquely positioned silicon wafer, with the hydrophobic part lower than the hydrophilic part, the droplet begins to move spontaneously ‘uphill’ toward the hydrophilic end. This arises because of unbalanced forces acting on the liquid-solid contact line on the two opposite sides of the drop, driving the droplet upward against gravity toward the hydrophilic end. Water repellency and in general material wettability are very important properties which depend on a material’s surface free energy and surface roughness.113 Control of these properties is important for many applications, such as variable focus liquid lens114 or self-cleaning surfaces.115 Using an electrode, droplet movement can be achieved using direct electrical control of the surface tension.116,117 For example, it has been demonstrated that a droplet of a polarizable and conductive liquid can be moved when placed between two parallel plates, the upper consisting of a single continuous ground electrode and the bottom consisting of an array of independently controlled electrodes.116 In another case, controlled droplet motion has been demonstrated on an open surface. A droplet placed on the surface between two dielectric-coated coplanar electrodes starts to move toward the more positive electrode when a sufficient voltage is applied. The greater positive polarity of one electrode relative to the other is responsible for the phenomenon, which is known as asymmetric electrowetting-on-dielectric oscillation (AEWOD). It is claimed that through this effect unidirectional and oscillatory transport of a droplet on an electrode surface can be performed.117 Another exciting possibility is light-driven liquid movement on a photoresponsive surface.118 In one example, azobenzene derivatives were switched between the cis and trans isomers using light with corresponding changes from hydrophilic to hydrophobic surface characteristics. For a surface modified with an azobenzene derivative (o-carboxymethyllated calixresorcinarene), switching was achieved using UV irradiation to form the cis-azobenzene (hydrophilic) isomer and blue light to form the trans-azobenzene (hydrophobic) isomer.119 When an olive oil droplet was placed on the surface, asymmetrical photoirradiation caused a gradient in the surface polarity that induced droplet motion, whose direction and velocity were found to be tunable by varying the light intensity (Figure 10). In the examples reported above, external control of the surface wettability by means of light or electrochemical potential provides a mechanism for controlling liquid transport on surfaces at relatively high flow rates without the need for high voltages/currents and avoiding the use of conventional pumps, valves, or channels. Hence, changes in surface tension can provide a driving force to move liquids without application of an external mechanical force. Another consequence of surface tension is capillary movement. This depends on the diameter of the channel and the contact angle between the liquid and the channel. The smaller the radius of the channel and larger the contact angle, the more a liquid will be affected by capillary forces. When capillary forces become dominant, the flow regime is typically laminar, and the high ratio between area and volume makes the physical and chemical properties of the liquid-surface interface

Diamond et al.

Figure 10. Photographs of light-driven motion of an olive oil droplet on a silica plate modified with an azobenzene derivative. The oil droplet moves because of a surface energy gradient generated by asymmetrical irradiation with 436 nm light perpendicular to the surface. (A-C) The sessile contact angles were changed from 18° (A) to 25° (C), confirming photonic modification of the surface polarity. (D) The direction of movement of the droplet could be controlled by varying the direction of the photoirradiation. (Reprinted with permission from ref 118. Copyright 2000 AAAS.)

critical for determining the overall characteristics of the system.

5.3. Controlling Binding Processes at Sensor Surfaces For all chemo-/biosensors, control of surface structure at the molecular scale is the ultimate goal as this in turn determines all observable macroscale behavior, such as chemo-/bioactivity, selectivity, sensitivity, response time. In the past, materials science has focused on generation of surfaces that had a particular function, for example, very passive, protective surfaces such as Teflon, or very active surfaces generated by immobilization of specific binding sites such as synthetic ligands, or bioreceptors such as enzymes or antibodies. More recently, the concept of ‘switchable’ or ‘adaptive’ surfaces has emerged in which the surface typically can be switched between two or more modes that can have very distinctly different characteristics. Key to this developing area has been ways of immobilizing functional molecules on sensor substrates. In the following section, we shall examine strategies for producing functionalized surfaces that can be electrochemically, optically, or chemically switched between different modes of behavior (e.g., passive or nonbinding surface and active or binding surface), using external chemical, electrochemical, or photonic stimuli. Control of binding behavior can be effected through molecules whose conformational, electron distribution (polarity), charge, wetting, and optical properties can be changed using light, electrostatic/magnetic field, electrochemical, or chemical stimuli.120 For surfaces exhibiting such behavior, it raises the intriguing possibility of having a material that can be maintained in a passive mode that is relatively unaffected by exposure to the sample environment over time, switched to the active mode only when a measurement is needed, and subsequently switched back to passive mode. In principle, this could open the way to new types of chemo-/ biosensors that can maintain their operating characteristics

Wireless Sensor Networks and Chemo-/Biosensing

Chemical Reviews, 2008, Vol. 108, No. 2 673

Figure 11. Self-assembled monolayer of pseudorotaxane on Au and dethreading and rethreading of the cucurbituril (CB(6)) macrocyclic cage on varying the pH. (Reprinted with permission from ref 122. Copyright 2003 Wiley-VCH Verlag GmbH & Co.)

for much longer times, hence reducing the need for complex calibration routines. Electrochemical modulation of host-guest complexes at solid-liquid interfaces has been demonstrated using selfassembled monolayers (SAMs) based on pseudorotaxanes. Pseudorotaxanes are a class of supramolecular compounds composed of a molecular thread, encircled by a molecular bead, typically a macrocyclic cage, free to dissociate from the molecular thread121,122 (Figure 11). pH-dependent, reversible dethreading and rethreading of the ring has been demonstrated when the thread component of the pseudorotaxanes is anchored to a gold surface by means of a disulfide pentacycle at the end of the chain and the macrocyclic cage is a cucurbituril (CB(6)), a compound comprising six glycoluril units. Under acidic conditions, the macrocyclic cage can bind an appropriate guest to form the pseudorotaxane supramolecular compound, while under alkaline conditions there is a dethreading of the macrocyclic guest and consequently dissociation of the pseudorotaxane complex. On the surface of an electrode, depending on the pH, the SAM can block or allow accessibility of electroactive species such as iron hexacyanate(III).122 On the surface of an electrode the pseudorotaxane-SAM layer acts as an “ion gate”, and its conductance can be changed under pH control. It has also been demonstrated that the complexation properties of a pseudorotaxane-functionalized self-assembled monolayer can be electrochemically controlled by adjusting the redox state of the guest species.121 In this system the guest is a stable tetrathiafulvalene anchored to a gold electrode and the hosts are electron-deficient and electron-rich macrocyclic compounds. Electrochemical reduction of the tetrathiafulvalene leads to formation of a pseudorotaxane with the electron-rich macrocyclic host, while oxidation of the tetrathiafulvalene in the presence of the electron-deficient macrocyclic host leads to formation of a pseudorotaxane. Although reversibility is still an issue, this system offers the possibility of electrochemical- or pH-based control of interactions at the molecular scale. In a similar way, reversible control of the surface properties can be performed with electrical switching. Electrical potential stimulation can be used to control the surface properties of SAMs based on ionizable alkanethiolate on a gold surface.123 They are lowdensity ionizable SAMs (LD-SAM) of (16-mercapto)hexadecanoic acid (MHA) where on one side of the long alkane chain the thiol groups anchor the molecule to a gold surface while on the other side the carboxylic function provides a hydrophilic cap over the hydrophobic chain. Upon applying

a positive potential, the negatively charged carboxylate groups bend toward the gold surface, exposing the hydrophobic chain of the MHA to the surrounding medium. Otherwise, by applying a negative potential the negatively charged carboxylate groups undergo electronic repulsion toward the gold surface and the MHA adopts a straight conformation, exposing hydrophilic groups to the surrounding medium (Figure 12). A simple conformation change induced by electronic potential therefore can induce significant changes in the overall surface binding properties. In order to ensure that there is enough space for the MHA chains to bend and lift up, the assembling process of the SAM includes a wrapping-unwrapping step of the alkanethiolate with molecules that can provide a spacer function to maintain the MHA chains at a certain distance. The LD-SAM switchable surfaces have been demonstrated use to control protein assembly of two kinds of fluorescent avidin with two different isoelectric points, one positive and one neutral. This could open the way to controlled protein adsorption-release in capillaries, channels, or protein chips.124 Magnetic switching of surface orientation has been demonstrated with the development of a magnetic tunable electrochemical reactivity on catalytic nickel nanowires.125 Considering the nickel’s magnetic properties and electrocatalytic action toward carbohydrates and alcohols, it is possible to control its reactivity by magnetic modulation. Placing the nickel nanowire on the surface of a carbon electrode, a vertical orientation of the magnetic field induced a big increase of the current signal and full accessibility to the catalytic site, whereas with the opposite orientation, a decrease of the signal is detected and accessibility to the catalytic site is blocked. Light-driven processes can also be used to induce reversible conformational transitions in molecules that can be used to produce surfaces with switchable behavior.126 It commonly involves switching a leuco form to a colored form by exposure to UV radiation, accompanied by changes in the UV-vis spectrum and hence changes in the color of the material (photochromism). Organic photochromic compounds are molecules of considerable interest as they are expected to offer new functional materials such as erasable photomemories and photoresponsive devices that take advantage of polarity and geometrical changes induced by irradiation.127 Among the class of the organic photochromic compounds, spiropyrans are some of the most attractive due to their high photosensitivity and very clear color change. Spiropyrans exist in a closed, uncharged, inactive, nonplanar, and

674 Chemical Reviews, 2008, Vol. 108, No. 2

Diamond et al.

Figure 12. Representation of the wrapping-unwrapping step of the alkanethiolate with molecules that can provide a molecular event between to maintain the MHA chains at a certain distance and then the transition of the MHA chains between straight (hydrophilic) and bent (hydrophobic) molecular conformations. (Reprinted with permission from ref 123. Copyright 2003 AAAS.)

Figure 13. Spiropyrans reversible conversion between a closed, uncharged, inactive, nonplanar, and colorless “spiropyran” form (SP) by exposure to visible or green light and open, planar, active, highly conjugated, highly colored “merocyanine” form (MC) by exposure to ultraviolet (UV) light. The merocyanine form can bind metal ions through the phenolate anionic site and amino acids due to complementary zwitterionic association.157 The guests can be subsequently expelled by irradiation with visible or green light, and the spiropyran reverts to the closed form. Light-modulated control of ion binding at spiropyran-modified surfaces is therefore possible.

colorless “spiropyran” form (SP) (leuco dye) which is converted by exposure to ultraviolet (UV) light to an open, planar, active, highly conjugated, highly colored “merocyanine” form (MC) (Figure 13). The SP closed form presents an absorption spectrum in the UV range with an actinic band situated near 320-380 nm. Absorption in this region leads to formation of the MC open form with a maximum absorbance wavelength between 560 and 570 nm (depending on the solvent used). The reverse process is stimulated by visible light (particularly in the green region, around 525 nm) or occurs slowly in the absence of light because the SP form is thermodynamically favored. The exciting feature of this system is the ability of the merocyanine form to reversibly bind metal ions through the negatively charged phenolate group. On binding the metal ion the absorbance maximum shifts to around 430 nm and the absorbance at 560 nm decreases.128 Hence, the system is inherently self-indicating as the color changes from deep purple (merocyanine form) to pink (metal ion complex). It has been demonstrated that when the spiropyran is covalently immobilized onto an optically transparent polymeric surface of PMMA (polymethylmethacrylate), it can be reversibly switched many times between the SP and MC forms using UV/green LEDs.128 The negatively charged phenolate group of the MC form provides a weak and relatively nonselective binding site for a range of cations (Figure 14) that can be activated and deactivated at a user-selected location using the LEDs. As in the liquid-phase experiments mentioned above, ion-binding results in a shift in absorbance, which corresponds to a color change from purple to pink (λmax ca. 430 nm). Then, upon irradiation of the complex with a green LED or visible light, the metal ion can be expelled and the MC reverted to the inactive SP form. This cycle can be

repeated multiple times using UV and green LEDs to control the surface binding of a range of metal ions, including Co2+ and Cu2+.128,129 In principle, therefore, these are surfaces that can be switched reversibly between active and passive states. In the SP form no binding with metal ions occurs and the system is inherently self-indicating due to the color changes that accompany switching between SP and MC isomers and formation of the MC-metal ion complexes. Furthermore, the bound species can be reversibly expelled on demand simply by illumination of the surface with green or visible light, and the system once again indicates its status through return to the colorless SP form. Due to the zwitterionic nature of MC, amino acids are potential guests through complementary two-point electrostatic binding. L-Tryptophan, L-tyrosine, and L-dopa have been demonstrated to bind reversibly with merocyanine zwitterions.130 This raises the prospect of surfaces that can reversibly bind biomolecules on demand using light to control the system. Recently combining the photochromic behavior of spiropyrans and the swelling properties of hydrogel matrix it has been demonstrated that the volume of a hydrogel copolymer of spirobenzopyran and poly(Nisopropylacrylamide) (pNIPAAm) can be reversibly reduced and increased when the gel is alternatively exposed to blue light or kept in the dark.131 In addition to switchable microrelief formation properties, the hydrogel exhibits changes in its electrostatic properties. When the hydrogel is placed in a slightly acidic bath the spiropyran moiety is neutral when irradiated with blue light (because of the prevalence of the closed form) and positively charged when kept in the dark (because of the prevalence of the open form in which both the phenolate and indoline nitrogen are protonated). After microrelief formation, negatively charged

Wireless Sensor Networks and Chemo-/Biosensing

Chemical Reviews, 2008, Vol. 108, No. 2 675

Figure 14. Schematic representation of a spiropyran derivative covalently immobilized onto an optically transparent polymeric surface of PMMA (polymethylmethacrylate). The colorless inactive SP form (top) can be reversibly switched using UV light/green LEDs to the pink zwitterionic MC active form (bottom). The negatively charged phenolate group of the MC form provides a binding site for a range of cations including Cu2+, Cr3+, and Co2+.

fluorescent nanoparticles are adsorbed on the nonirradiated region because of the positive overall charge of the photochromic compound. These are examples of materials whose function and properties can be radically changed and controlled using external stimuli. They will have important applications in microfluidics, separation science, controlled sampling and release, soft actuators, and sensing and could lead to development of radically different platforms for realizing autonomous analytical devices suitable for scaled-up sensor network deployments

5.4. Bead-Based Systems Bead-based systems are a particular kind of biphase system in which solid particles in suspension can be moved as a fluid but easily separated from the liquid phase. Beads provide a higher surface area than flat surfaces for chemical reactions, and bead-based analytical approaches offer significant advantages for techniques that require reactions on surfaces.122 The low density of the polymer matrix allows binding kinetics comparable to those of solution-based systems, and their large surface area and greater density permit rapid and highly efficient binding of target species. Micro- or nanoparticles can be chemically derivatized with a wide range of specific ligands or specific recognition groups. General specificity particles are largely produced as substrates to attach a variety of affinity ligands which allow a broad range of direct applications, such as fluid flowcitometry analysis,132 immunoassay-diagnostics,133,134 cell

biology,135,136 and controlled release.130 They are also generating increasing interest in the photonics community due to their high quality-factor (Q-factor) morphology-dependent resonances and sensitivity to refractive index and size changes. When biomolecules are adsorbed on their surface, a change on their effective size and refractive index occurs, and it is claimed that a biosensor based on this phenomenon can detect a single molecule.137 Configured as optical microcavity resonators, applications include microlasers, narrow optical filters, optical switching, high-resolution spectroscopy,137 and Raman sources.138 A variety of materials, including organic and inorganic polymers, have been used as substrates for micro- and nanospheres. Common organic substrates are polystyrene or polymethacrylate. Polystyrene microspheres are often used for protein binding due to the high degree of nonspecific protein adsorption (hydrophobic interaction is the most likely mechanism involved in adsorption of proteins). Inorganic substrates include metals, silica, or alumina. Silica microspheres are naturally hydrophilic, so relatively little nonspecific protein absorption should occur. Chromophore-modified microspheres are available in a wide variety of colors as are beads modified with fluorochromes, fluorophors, and scintillators. Superparamagnetic or radioactive microspheres are available with a range of different mean magnetic and radioactivities. Bead-based approaches have been applied in fluidic systems in numerous ways, for separations, sample extraction, analyte detection, and controlled release. Silica beads

676 Chemical Reviews, 2008, Vol. 108, No. 2

coated with octadecylsilane (ODS) packed in a microfluidic system enabled on-chip solid-phase electrochromatography for extraction of analytes with different polarity to be developed. The ODS-coated beads were mobilized in the cavity of a chromatography bed as part of a microfluidic system using electroosmotic flow. Two wires within a sample channel formed a cavity in which the beads were trapped. Beads were similarly packed in a chamber and reversedphase capillary electrochromatography performed. A fluorescent nonpolar analyte and fluorescein were loaded on the chromatographic bed, and complete separation was achieved after 20 s.139 Bead-based techniques have been successfully adopted also for miniature biosamplers, such as filterless bioseparators, that allow separation of target biomolecules or bacteria from a liquid.140 Superparamagnetic particles have been coated with antibodies and controlled in a microfluidic channel with a planar electromagnet. When the magnetic field is applied, beads are held in the channel, and when antigens are injected, the bound antigens are retained while others are washed out with the flow. Subsequently, magnetic beads are released to the sensing chamber for detection, and the filterless bioseparator is ready for another separation. Immunoassay, thanks to its high specificity, is one of the most important detection methods for clinical diagnosis and biochemical studies. Employing mobile beads in an immunosorbent assay leads to a very flexible system with attractive characteristics compared to conventional approaches.141 For example, a bead-based system provides simple and effective separation of the free and bound forms and allows heterogeneous immunosorbent assays to be performed within a microchannel. Polystyrene beads preabsorbed with s-IgA (antigen) were introduced in a microchannel, reacted with colloidal gold conjugated anti-s-IgA antibody, and detected by a thermal lens microscope. With this method the overall analysis time was reduced from 24 h for the conventional homogeneous liquid-phase immunoassay to 1 h, avoiding numerous washing and solution removal procedures. A bead system has also been employed for gene expression analysis. A so-called ‘nanobarcode’-based microbead assay has been developed to give an accurate and reproducible gene expression profile.142 Four different quantum dot nanocrystal fluorescence emitters with different emission maxima were mixed with a polymer and coated in magnetic microbeads to generate nanobarcode beads called Qbeads. Gene-specific oligonucleotide probes were conjugated to the surface of each Qbead to create a panel which was able to decode an RNA target on the basis of the spectral profiles and intensity ratios of the four Qbeads. Unbound RNA can be easily washed out when the beads are exposed to a magnetic field. The gene expression data obtained with the beads was found to have a high correlation with reference Affymetrix Genechip microarray data. This nanobarcode system opens the way for a huge variety of applications in biology, chemistry, and medical diagnosis and can potentially code more than one million combinations. Similarly, a bifunctional system based on mesoporous silica beads embedded with both semiconductor quantum dots for optical encoding and iron oxide for magnetic separation has been developed.143 Quantum dots have attractive properties such as size-tunable light emission, resistance against photobleaching, and simultaneous excitation properties, while beads embedded with iron oxide have been used for biological separation, capture of rare cells,

Diamond et al.

Figure 15. Magneto-controlled reversible translocation of the functionalized magnetic particles between the organic phase above the aqueous electrolyte and the electrode surface (A). The electrode surface is blocked with functionalized magnetic particles attracted to the electrode by the external magnet. (B) The magnetic particles are retracted from the electrode surface, and the electrode surface is electrochemically active. (Reprinted with permission from ref 144. Copyright 2004 American Chemical Society.)

proteins, and nucleic acids. Functionalized magnetic nanoparticles have also been successfully used to control and switch the hydrophilicity/hydrophobicity of an electrode surface.144 A two-phase system, made up of water and toluene as the liquid phase and magnetic nanoparticles embedded with hydrophobic alkyl chains as a suspended solid phase, was placed in contact with an Au electrode. Applying an external magnetic field on the electrode surface induced physical attraction of the particles to the electrode surface turning the surface hydrophobic. On the other hand, applying a magnetic field to the upper toluene phase induces migration of the particles to this layer, generating a hydrophilic surface at the gold electrode (Figure 15). Electrochemical switching of the conductive support surface has been demonstrated by performing magnetoswitchable bioelectrocatalytic oxidation of glucose in the presence of glucose oxidase and ferrocene dicarboxylic acid. Removal of the magnetic particles from the electrode surface allows oxidation of ferrocene dicarboxylic acid at the electrode surface and bioelectrocatalyzed oxidation of glucose by glucose oxidase. This process is blocked by attraction of the particles to the electrode. Hence, by means of a physical change (bead movement) it is possible to control a reactive interface using the chemical and mobility properties of a bead-based system. Light excitation is another mechanism for controlling the properties of microparticle systems. Light modulation of spiropyran-coated gold nanoparticles provides a potential system for the controlled release of amino acids.130 A spiropyran-modified derivative with a thiol chain was selfassembled onto a gold nanoparticle surface, and the photoswitching of the Au-SP nanoparticle in the presence of various amino acid derivatives has been investigated. The spiropyran was excited with UV light and converted to the merocyanine form. Subsequently, it was demonstrated that the presence of certain amino acid derivatives stabilized the MC form due to formation of a stable complex between MC and the amino acid derivatives. Furthermore, irradiation of the complex with visible light triggers reversion of the MC to the SP form and release of the guest species. With this approach it may be possible to use light-modulated binding to control a localized amino acid concentration, which could

Wireless Sensor Networks and Chemo-/Biosensing

form the basis of new approaches to drug delivery or biological separations.

6. Overall Conclusions Because of space limitations, this review has been largely restricted to two potential areas of application for wireless sensor networks: remote environmental monitoring and wearable sensors or body sensor networks. With respect to the environment, we focused on recent developments in mote-based wireless sensor networks and highlighted limitations with the current manifestations of these platforms, particularly with respect to integration of chemo-/biosensing capabilities and the essential requirement for these devices to be completely self-sustaining in all respects if the promised massive scale up to true pervasive sensing is to be realized. However, in addition to the need for improvements in the engineering aspects of motes, there are very significant challenges for the chemo-/biosensor research community to deliver sensing platforms that are appropriate for integration into scaled-up deployments in terms of sustainability, cost, reliability, etc. While microfluidics offers a route to intermediate scale up in the short-to-medium term (5 years), particularly for autonomous environmental sensing, they still require reagents and generate waste and are still not sufficiently reliable for sustained use over many months/ years. Consequently, examples of real deployments of chemo-/biosensor networks are few and far between. Consequently, we are only beginning to understand how signals obtained from groups or communities of simple, low-cost sensors may provide a higher degree of certainty in event detection (e.g., through tracking the dynamics of response patterns to validate a decision) and to what extent this may compensate for the lack of sophisticated calibration procedures.145 In general, we can speculate that in the medium term (5 years) we will see a growth in the use of reagentless and ‘noncontact’ approaches that can generate molecular or chemical information about a sample. For example, methods such as UV-vis, IR, and Raman spectroscopies provide direct access to such information. While IR/Raman spectrometers are too expensive (typically $20K upward) to be considered for scaled-up deployments, IR detectors tuned to the specific absorbances of target gases are available commercially, and the cost of these is likely to decrease rapidly.146,147 Direct UV measurement of nitrate in wastewater, lakes, rivers, and marine environments is a good example of a reagentless approach that has become increasingly employed for long-term monitoring applications.148 Such nonselective measurements can be effective under very specific constraints (e.g., there are no coabsorbing species present in the sample) or where reagent-based approaches cannot be employed (e.g., time scale is very long, months/ years between servicing).149 Another approach is to employ noncontact ac conductance/impedance measurements. The advantage here is that the electrodes can be completely encased in a protective coating while still probing the local environment for changes in conductivity that are related to the ionic strength of liquid samples, which in turn is related to changes in the chemical composition. This approach does not provide the molecular selectivity accessible through more sophisticated chemo-/biosensors. In other words, it can detect that changes may be happening in the local chemical environment without being able to identify what species is causing the change. In the recent literature it has been combined with microcapillary electrophoresis as a detector,

Chemical Reviews, 2008, Vol. 108, No. 2 677

but it can function independently of a separation platform if general information about the gross chemical composition is the key operational requirement.150 Applications in wearable sensors/body sensor networks are certain to happen in the near future (2 years) as the first commercial products are already appearing and a very significant user base is rapidly emerging, initially through consumer exercise/fitness products, but these will quickly evolve into personal health applications and drive demand for chemo-/biosensing products capable of providing specific information on disease status that will complement the more generic health information available from wearable sensors. However, the key to large-scale deployments of chemo-/ biosensors lies in fundamental materials science. Radically new approaches to low-power transport of water across surfaces, through materials, and along microchannels are needed, for example, based on light- or electrochemicalmediated control of surface charge/polarity. In a similar manner, control of surface binding behavior is an exciting prospect as is control of surface docking and release of molecular targets or multifunctional beads. Biomimetics will gain popularity, leading to the development of microfluidic platforms with circulatory systems incorporating polymeric structures that provide the function of pumps and valves through muscle-like actuation behavior. Accordingly, there is a wonderful opportunity to link research in molecular materials with microsystems engineering and developing new platforms capable of performing reliable chemo-/biosensor measurements in scaled-up deployments that will have a major impact on individuals and society.

7. Abbreviations AEWOD Ag AgCl A-Si BSN CB(6) CENs CIGS CTO DARPA DMA DMMP EAPs ECG EM EMG ETH EU FANs FDA FETs GF GOD GPS GSM ICP IEEE ISM ISN LAN LD-SAM LED LDR LOAC MC

asymmetric electrowetting-on-dielectric oscillation silver silver chloride amorphous silicon body sensor network cucurbituril Centre for Embedded Networked Sensors copper indium gallium diselenide Chief Technology Officer defense advanced research projects agency dodecyl methacrylate dimethyl methylphosphonate electroactive polymers electrocardiograph electromagnetic (radiation) electromyography Eidgeno¨ssische Technische Hochschule (Zu¨rich) European Union fabric area networks Food and Drug Administration field effect transistors Gauge factor glucose oxidase global positioning system global system for mobile communications inherently conducting polymers Institute of Electrical and Electronics Engineers Industrial, scientific, and medical (radio band) Institute for Soldier Nanotechnologies local area network low-density ionizable SAMs light-emitting diode light-dependent resistor lab-on-a-chip merocyanine form

678 Chemical Reviews, 2008, Vol. 108, No. 2 MEMS MHA MIT MPC µTAS NRBC NSF RFID ODS PANi

microelectromechanical systems (16-mercapto) hexadecanoic acid Massachusetts Institute of Technology hydrophilic 2-methacryloyloxyethyl phosphorylcholine micrototal analysis systems Nuclear, Radiological, Biological, Chemical National Science Foundation radiofrequency identification octadecylsilane polyaniline

8. Acknowledgments The authors wish to thank the following for their support: European Union grant BioTex (FP6-2004-IST-NMP-2), Science Foundation Ireland (SFI 03/IN.3/1361 and SFI 07/ RFP/MASF81Z), and the Marine Institute (AT/04/01/06).

9. References (1) Ambrosio, R. Creating An Expanded DER Industry Washington, DC 2001. (2) Diamond, D. Anal. Chem. 2004, 76, 278A. (3) Ramirez-Garcia, S.; Diamond, D. J. Intell. Mater. Syst. Struct. 2007, 18, 159. (4) CENS http://research.cens.ucla.edu. (5) Wang, N.; Zhang, N.; Wang, M. Comput. Electron. Agr. 2006, 50, 1. (6) Beigl, M.; Krohn, A. T.; Riedel, T.; Zimmer, C.; Decker, M.; Isomura, M. IPSN-2006, 2006; p 366. (7) Lee, K. IMTC-2000 Baltimore, MD, U.S.A. 2000 p 525. (8) Levis, P.; Madden, S.; Gay, D.; Polastre, J.; Szewczyk, R.; Woo, A.; Brewer, E.; Culler, D. NSDI-2004, San Francisco, U.S.A. 2004, p 1. (9) nesC 1.1 Language Reference Model http://nescc.sourceforge.net/ papers/nesc-ref.pdf. (10) Kernighan, B. W.; Ritchie, D. M. The C Programming Language (Second Edition); Prentice Hall, Inc. 1988. (11) Kredo, K.; Mohapatra, P. Comput. Netw. 2007, 51, 961. (12) Hayes, J.; Crowley, K.; Diamond, D. EFFOST-INTRADFOOD, Valencia Spain 2005. (13) Adams, J. http://www.sensorsmag.com/articles/0603/14/main.shtml 2003. (14) Kallo´, C. K.; Chiasserini, C.; Jung, S.; Brunato, M.; Gerla, M. Ad Hoc Netw. 2007, 5, 340. (15) Beckwith, R.; Teibel, D.; Bowen, P. LCNS-2004, 2004; p 471. (16) Beckwith, R.; Teibel, D.; Bowen, P. IEEE Sensors-2004, 2004; p 561. (17) Tolle, G.; Szewczyk, R.; Culler, D.; Turner, N.; Tu, K.; Burgess, S.; Dawson, T.; Buonadonna, P.; Gay, D.; Hong, W. SenSys-2005, 2005; p 51. (18) Szewczyk, R.; Mainwaring, A. M.; Polastre, J.; Anderson, J.; Culler, D. E. SenSys-2004, 2004; p 214. (19) Arora, A.; Ramnath, R.; Ertin, E.; Sinha, P.; Bapat, S.; Naik, V.; Kulathumani, V.; Zhang, H.; Cao, H.; Sridharan, M.; Kumar, S.; Seddon, N.; Anderson, C.; Herman, T.; Trivedi, N.; Zhang, C.; Nesterenko, M.; Shah, R.; Kulkarni, S.; Aramugam, M.; Wang, L.; Gouda, M.; Choi, Y.; Culler, D.; Dutta, P.; Sharp, C.; Tolle, G.; Grimmer, M.; Ferriera, G. K. P. RTCSA-2005, 2005; p 102. (20) Werner-Allen, G.; Lorincz, K.; Ruiz, M. C.; Marcillo, O.; Johnson, J. B.; Lees, J. M.; Welsh, M. IEEE Internet Comput. 2006, 10, 18. (21) Cardell-Oliver, R.; Smettem, K.; Kranz, M.; Mayer, K. Int J. of Distributed Sensor Networks 2005, 1, 149. (22) Glasgow, H. B.; Burkholder, J. M.; Reed, R. E.; Lewitus, A. J.; Kleinman, J. E. J. Exp. Mar. Biol. Ecol. 2004, 300, 409. (23) Myers, R.; Vickers, M.; Kim, H.; Priya, S. Appl. Phys. Lett. 2007, 90. (24) Paradiso, J. A.; Starner, T. IEEE PerVasiVe Comput. 2005, 4, 18. (25) Stephen, N. G. J. Sound Vibrat. 2006, 293, 409. (26) Diamond, D.; Lau, K. T.; Brady, S.; Cleary, J. Talanta 2008, accepted for publication. (27) Sequeira, M.; Bowden, M.; Minogue, E.; Diamond, D. Talanta 2002, 56, 355. (28) Potyrailo, R. A.; Morris, W. G. Anal. Chem. 2007, 79, 45-51. (29) Troester, G. In IMIA Yearbook of Medical Informatics; Haux, R.; Kulikowski, C., Eds.; Schattauer: Stuttgart 2005. (30) Lam Po Tang, S.; Stylios, G. K. Int. J. Clothing Science Tech. 2006, 18, 108. (31) Yang, G.-Z. Body Sensor Networks; Springer 2006.

Diamond et al. (32) Brady, S.; Dunne, L.; Lynch, A.; Smyth, B.; Diamond, D. Stud. Health. Technol. Inform. 2006, 117, 80. (33) Post, E. R.; Orth, M.; Russo, P. R.; Gershenfield, N. IBM Syst. J 2000, 39, 840. (34) Wagner, S.; Bonderover, E.; Jordan, W. B.; Sturm, J. C. Int. J. High Speed Electron Syst. 2002, 12, 391. (35) Bowman, D.; Mattes, B. M. Synth. Met. 2005, 154, 29. (36) Meoli, D.; May-Plumlee, T. J.Textile Apparel Tech. Manage. 2002, 2, 1. (37) El-Sherif, M. A.; Yuan, J.; MacDiarmid, A. J. Intell. Mater. Syst. Struct. 2000, 11, 407. (38) Carpi, F.; De Rossi, D. IEEE T Inf. Technol. B. 2005, 9, 295. (39) Bekaert “What are Metal Fibres? http://www.bekaert.com/bft/,” 2004. (40) Chiang, C. K.; Fincher, C. R. Jr.; Park, Y. W.; Heeger, A. J.; Shirakawa, H.; Louis, E. J.; Gau, S. C.; MacDiarmid, A. G. Phys. ReV. Lett. 1977, 39, 1098. (41) Mazzoldi, A.; Santa, A. D.; Rossi, D. D. In Polymer Sensors and Actuators; Osada, Y. Ed. Springer: Berlin 2000. (42) De Rossi, D.; Carpi, F.; Lorussi, F.; Paradiso, J.; Scilingo, E. P. In Wearable Electronics and Photonics; Tao, X., Ed. Woodhead: Cambridge 2005. (43) El-Sherif, M. A. In Wearable electronics and photonics; Tao, X., Ed. Woodhead Publishing: Cambridge 2005. (44) Luminex http://www.luminex.it/. (45) Wong, Y. W. H.; Yuen, C. W. M.; Leung, M. Y. S.; Ku, S. K. A.; Lam, H. L. I. AutexRJ 2006, 6, 1. (46) Qian, L.; Hinestroza, J. P. J. Textile Apparel Technol. Manage. 2004, 4, 1. (47) Nano-tex “Press Release: Nano-Tex Unveils Anti-Static, Stain Release, Moisture-Wicking Fabric Enhancements, http://www.nanotex.com/news&media/Feb16_2005.pdf,” 2005. (48) Maccioni, M.; Orgui, E.; Cosseddu, P.; Locci, S.; Bonfiglio, A. Appl. Phys. Lett. 2006, 89, 1. (49) Bonfiglio, A.; De Rossi, D.; Kirstein, T.; Locher, I.; Mameli, F.; Paradiso, R.; Vozzi IEEE T. Inf. Technol. B. 2005, 9, 319. (50) Philips “Philips Research Press Release: Philips illuminates IFA2006 with production-ready Lumalive textile garments” 2006. (51) Dawson, C.; Vincent, J. F. V.; Rocca, A. Nature 1997, 390, 668. (52) Marks, P. In New Scientist 2001. (53) Baughman, R. H. Synth. Met. 1996, 78, 339. (54) Della Santa, A.; Rossi, D. D.; Mazzoldi, A. Smart Mater. Struct. 1997, 6, 23. (55) Hara, S.; Zama, T.; Takashima, W.; Kaneto, K. Polym. J. 2004, 36, 933. (56) Wu, Y.; Alici, G.; Spinks, G. M.; Wallace, G. G. Synth. Met. 2006, 156, 1017. (57) Bar-Cohen, Y. Robotics 2000 and Space 2000, Alburquerque, NM U.S.A. 2000; p 1. (58) Zhou, D.; Wallace, G.; Spinks, G. M.; Liu, L.; Cowan, R.; Saunders, E.; Newbold, C. Synth. Met. 2003, 39, 135-136. (59) Jung, S.; Lauterbach, C.; Strasser, M.; Weber, W. ISSCC-2003, San Francisco, U.S.A. 2003, p 386. (60) Cottet, D.; Grzyb, J.; Kirstein, T.; Troester, G. IEEE Trans AdV. Pack. 2003, 26, 182. (61) Winterhalter, C. A.; Tererovsky, J.; Wilson, P.; Slade, J.; Horowitz, W.; Tierney, E.; Sharma, V. IEEE T Inf. Technol. B 2005, 9, 402. (62) Locher, I.; Klemm, M.; Kirstein, T.; Troster, G. IEEE Trans AdV. Pack. 2006, 29, 777. (63) Tronquo, A.; Rogier, H.; Hertleer, C.; Van Langenhove, L. Electron. Lett. 2006, 42, 142. (64) Power Paper http://www.powerpaper.com/. (65) Starner, T. IBM Syst. J 1996, 35, 618. (66) Yeatman, E.; Mitcheson, P. In Body Sensor Networks; Yang, G.-Z., Ed. Springer 2006. (67) Scottevest http://www.scottevest.com 2007. (68) Tsang, P. T. S. K.; Moore, A. J.; Farringdon, J. Koninklijke Philips Electronics N.V. (Eindhoven, NL) U.S. 2004. (69) Catrysse, M.; Puers, R.; Hertleer, C.; Van Langenhove, L.; Van Egmond, H.; Matthys, D.; Leuven, K. U. Transducers ’03, Boston, U.S.A. 2003, p 1758. (70) Paradiso, R.; Belloc, C.; Loriga, G.; Taccini, N. In Personalised Health Management Systems; Nugent, C., Mc Cullagh, P. J., McAdams, E., Lymberis, A., Eds. IOS Press, 2005. (71) Muehlsteff, J.; Such, O. IEEE EMBS-2004, San Francisco, CA 2004, p 2212. (72) Scilingo, E. P.; Gemignani, A.; Paradiso, R.; Taccini, N.; Ghelarducci, B.; De Rossi, D. IEEE T Inf. Technol. B 2005, 9, 345. (73) Kirstein, T.; Lawrence, M.; Troester, G. International workshop on a new generation of wearable systems for e-Health, Pisa, Italy 2003, p 201. (74) Munro, B. J.; Steele, J. R.; Campbell, T. E.; Wallace, G. G. Stud. Health Technol. Inform. 2004, 108, 271.

Wireless Sensor Networks and Chemo-/Biosensing (75) Dunne, L. E.; Tynan, R.; O’Hare, G. M. P.; Smyth, B.; Brady, S.; Diamond, D. IFAWC-2005, Zurich, Switzerland 2005. (76) Brady, S.; Carson, B.; O’Gorman, D.; Moyna, N.; Diamond, D. BSN2006 Boston 2006, p 31. (77) Brady, S.; Coyle, S.; Wu, Y.; Wallace, G.; Diamond, D. MRS Spring Meeting, Symposium S, San Francisco 2006. (78) Dunne, L.; Walsh, P.; Smyth, B.; Caulfield, B. ISWC06, Montreaux, Switzerland 2006, p 65. (79) Farella, E.; Acquaviva, A.; Benini, L.; Ricco`, B. EUROMEDIA2005, Toulouse, France 2005, p 110. (80) Asada, H. H.; Shaltis, P.; Reisner, A.; Rhee, S.; Hutchinson, R. C. IEEE Eng. Med. Biol. Mag. 2003, 22, 28. (81) Asada, H. H.; Reisner, A.; Shaltis, P.; McCombie, D. IEEE EMBS 2005, Shanghai, China 2005, p 4156. (82) Axisa, F.; Gehin, C.; Delhomme, G.; Collet, C.; Robin, O.; Dittmar, A. IEEE EMBS 2004, San Francisco, CA, U.S.A. 2004, p 2161. (83) Tierney, M. J.; Tamada, J. A.; Potts, R. O.; Jovanovic, L.; Garg, S. Biosens. Bioelectron. 2001, 16, 621. (84) Coyle, S.; Wu, Y.; Lau, K.; Brady, S.; Wallace, G. G.; Diamond, D. BSN-2007, Aachen, Germany 2007. (85) Badagu, R.; Lakovicz, J. R.; Geddes, C. D. J. Fluoresc. 2003, 13, 371. (86) Kudo, H.; Iguchi, S.; Yamada, T.; Kawase, T.; Saito, H.; Otsuka Mitsubayashi, K. Biomed. MicrodeVices 2007, 9, 1. (87) Morphix Technologies “http://www.morphtec.com,” 2007. (88) Collins, G. E.; Buckley, L. J. Synth. Met. 1996, 78, 93. (89) Engin, M.; Demirel, A.; Engin, E. Z.; Fedakar, M. Measurement 2005, 37, 173. (90) Sung, M.; Marci, C.; Pentland, A. J. Neuroeng. Rehabil. 2005, 2, 1. (91) Loriga, G.; Taccini, N.; De Rossi, D.; Paradiso, J. EMBCO5, Shanghai, China 2005, p 7349. (92) Foster-Miller Inc. “Ambulatory physiological status monitor [Online] http://www.foster-miller.com/literature/documents/DS06-031WPSM.pdf,” 2007. (93) Patnaik, A.; Rengasamy, R. S.; Kothari, V. K.; Ghosh, A. Textile Progress 2006, 38, 1. (94) Pan, N.; Zhong, W. Textile Progress 2006, 38, 1. (95) Gopalakrishnan, S.; Liu, D.; Allen, H. C.; Kuo, M.; Shultz, M. J. Chem. ReV. 2006, 106, 1155. (96) Verdaguer, A.; Sacha, G. M.; Bluhm, H.; Salmeron, M. Chem. ReV. 2006, 106, 1478. (97) Manz, A.; Grabner, R.; Widmer, H. N. Sensor Actuat. B-Chem 1990, 244. (98) Paik, P.; Pamula, V. K.; Pollack, M. G.; Fair, R. B. Lab Chip 2003, 3, 28. (99) Terray, A.; Oakey, J.; Marr, D. W. M. Science 2002, 296, 1841. (100) Vestad, T.; Marr, D. W. M.; Oakey, J. J. Micromech. and Microeng. 2004, 14, 1503. (101) Weibel, D. B.; Kruithof, M.; Potenta, S.; Sia, S. K.; Lee, A.; Whitesides, G. M. Anal. Chem. 2005, 77, 4726. (102) Wu, Y.; Zhou, D.; Spinks, G. M.; Innis, P. C.; Megill, W. M.; Wallace, G. G. Smart Mater. Struct. 2005, 14, 1511. (103) Vuk, S.; Milicevic, N.; Griss, P. Transducers ’05, 2005, p 1565. (104) Diamond, D.; Lau, K. manuscript submitted to Talanta 2007. (105) Angerer, J.; Bird, M. G.; Burke, T. A.; Doerrer, N. G.; Needham, L.; Robison, S. H.; Sheldon, L.; Zenick, H. Toxicol. Sci. 2006, 93, 3. (106) Pearson, G. S.; Roberts, B. Defence Sci. J. 2001, 51, 377. (107) Grunze, M. Science 1999, 283, 41. (108) Gugliotti, M. J. Chem. Edu. 2004, 81, 67. (109) Scriven, L. E.; Sterling, C. V. Nature 1960, 187, 186. (110) Wasan, D. T.; Nikolov, A. D.; Brenner, H. Science 2001, 291, 605. (111) Bull, J. L.; Grotberg, J. Exp. Fluids 2003, 34, 1. (112) Chaudhury, M. K.; Whitesides, G. M. Science 1992, 256, 1539. (113) Chen, T.-H.; Chuang, Y.-J.; Chieng, C.-C.; Tseng, F.-G. J. Micromech. Microeng. 2007, 17, 489. (114) Kuiper, S.; Hendriks, B. H. W. Appl. Phys. Lett. 2004, 85, 1128. (115) Cheng, Y. T.; Rodak, D. E.; Wong, C. A.; Hayden, C. A. Nanotechnology 2006, 17, 1359. (116) Pollack, M. G.; Fair, R. B.; Shenderov, A. D. Appl. Phys. Lett. 2000, 77, 1725. (117) Tsu-Te, W.; Po-Wen, H.; Shih-Kang, F. MEMS-2006, 2006 p 174.

Chemical Reviews, 2008, Vol. 108, No. 2 679 (118) Ichimura, K.; Oh, S.-K.; Nakagawa, M. Science 2000, 288, 1624. (119) Sekkat, Z.; Wood, J.; Geerts, Y.; Knoll, W. Langmuir 1995, 11, 2856. (120) Liu, Y.; Mu, L.; Liu, B.; Kong, J. Chem-Eur J. 2005, 11, 2622. (121) Bryce, M. R.; Cooke, G.; Duclairoir, F. M. A.; John, P.; Perepichka, D. F.; Polwart, N.; Rotello, V. M.; Stoddart, J. F.; Tseng, H.-R. J. Mater. Chem. 2003, 13, 2111. (122) Kim, K.; Jeon, W. S.; Kang, J.-K.; Lee, J. W.; Jon, S. Y.; Kim, T.; Kim, K. Angew. Chem. Int. Edit. 2003, 115, 2395. (123) Lahann, J.; Mitragotri, S.; Tran, T.-N.; Kaido, H.; Sundaram, J.; Choi, I. S.; Hoffer, S.; Somorjai, G. A.; Langer, R. Science 2003, 299, 371. (124) Liu, Y.; Mu, L.; Liu, B.; Zhang, S.; Yang, P.; Kong, J. Chem. Commun. 2004, 1194. (125) Wang, J.; Scampicchio, M.; Laocharoensuk, R.; Valentini, F.; Gonzalez-Garcia, O.; Burdick, J. J. Am. Chem. Soc. 2006, 128, 4562. (126) Stitzel, S.; Byrne, R.; Diamond, D. J. Mater. Sci. 2006, 41, 5841. (127) Ito, T.; Hiramatsu, M.; Hirano, I.; Ohtani, H. Macromolecules 1990, 23, 4528. (128) Byrne, R. J.; Stitzel, S. E.; Diamond, D. J. Mater. Chem. 2006, 16, 1332. (129) Shao, N.; Zhang, Y.; Cheung, S.; Yang, R.; Chan, W.; Mo, T.; Li, K.; Liu, F. Anal. Chem. 2005, 77, 7294. (130) Ipe, B. I.; Mahima, S.; Thomas, K. G. J. Am. Chem. Soc. 2003, 125, 7174. (131) Szilagyi, A.; Sumaru, K.; Sugiura, S.; Takagi, T.; Shinbo, T.; Zrinyi, M.; Kanamori, T. Chem. Mater. 2007, 19, 2730. (132) Wedemeyer, N.; Potter, T. Clin. Genet. 2001, 60, 1. (133) Haukanes, B.-I.; Kvam, C. Nat. Biotech. 1993, 11, 60. (134) Sinha, V. R.; Goyal, V.; Bhinge, J. R.; Mittal, B. R.; Trehan, A. Crit. ReV. Ther. Drug. Carrier Syst. 2003, 20. (135) Margel, S.; Beitler, U.; Ofarim, M. J. Cell Sci. 1982, 56, 157. (136) Windler-Hart, S.; Chen, K.; Chenn, A. BMC Cell Biol. 2005, 6, 14. (137) Demir, A.; Serpenguzel, A. IEE Proc-Nanobiotech 2005, 152, 105. (138) Lu, Y.; Liu, G. L.; Lee, L. P. Nano Lett. 2005, 5, 5. (139) Oleschuk, R. D.; Shultz-Lockyear, L. L.; Ning, Y.; Harrison, D. J. Anal. Chem. 2000, 72, 585. (140) Choi, J.-W.; Ahn, C. H.; Bhansali, S.; Henderson, H. T. Sensor Actuat. B-Chem 2000, 68, 34. (141) Sato, K.; Tokeshi, M.; Odake, T.; Kimura, H.; Ooi, T.; Nakao, M.; Kitamori, T. Anal. Chem. 2000, 72, 1144. (142) Eastman, P. S.; Ruan, W.; Doctolero, M.; Nuttall, R.; deFeo, G.; Park, J. S.; Chu, J. S. F.; Cooke, P.; Gray, J. W.; Li, S.; Chen, F. F. Nano Lett. 2006, 6, 1059-1064. (143) Sathe, T. R.; Agrawal, A.; Nie, S. Anal. Chem. 2006, 78, 5627. (144) Katz, E.; Sheeney-Haj-Ichia, L.; Basnar, B.; Felner, I.; Willner, I. Langmuir 2004, 20, 9714. (145) Shepherd, R.; Beirne, S.; Lau, K. T.; Corcoran, B.; Diamond, D. Sensor Actuat. B-Chem. 2006, 121, 142. (146) Bernstein, R. W.; Ferber, A.; Johansen, I.; Moe, S. T.; Rogne, H.; Wang, D. T. Optical MEMS 2000, Hawaii, U.S.A. 2000, p 137. (147) Lammel, G.; Schweizer, S.; Renaud, P. MEMS-2001, Interlaken, Switzerland 2001, p 578. (148) Gruber, G.; Winkler, S.; Pressl, A. Water Sci. Technol. 2004, 50, 73. (149) Stanley, M. A.; Maxwell, J.; Forresteral, M.; Doherty, A. P.; Maccraith, B. D.; Diamind, D.; Vos, J. G. Anal. chim. acta 1994, 299, 81. (150) Hohercakova, Z.; Opekar, F.; Stulik, K. Electroanalysis 2005, 17, 1924. (151) FinaPres Medical Systems http://www.finapres.com. (152) Shaltis, P.; Reisner, A.; Asada, H. IEEE-EMBS 2005, Shanghai, China 2005, p 3970. (153) Ma, T.; Zhang, Y. T. EMBC05, Shanghai, China 2005, p 996. (154) Biotex http://www.biotex-eu.com/. (155) Knight, J.; Bristow, H.; Anastopoulou, S.; Baber, C.; Schwirtz, A.; Arvanitis, T. Pers. Ubiquitous Comput. 2007, 11, 117. (156) Brady, S. Ph.D. Thesis, Dublin City University 2006. (157) Byrne, R.; Diamond, D. Nat. Mater. 2006, 5, 421.

CR0681187

Chem. Rev. 2008, 108, 680−704

680

Chemical Sensing in Spatial/Temporal Domains Takamichi Nakamoto*,† and Hiroshi Ishida‡ Graduate School of Science and Engineering, Tokyo Institute of Technology, 2-12-1 Ookayama, Meguro-ku, Tokyo 152-8552, Japan, and Department of Mechanical Systems Engineering, Tokyo University of Agriculture and Technology, 2-24-16 Nakacho, Kogahei, Tokyo 184-8588, Japan Received May 14, 2007

Contents 1. Introduction 2. Spatial Domain 2.1. Plume Behavior and Analysis Method 2.2. Plume Observation Method 2.3. Gas-Distribution Measurement 2.3.1. Measurement of Spatial Gas-Distribution Change by a Large Sensor Array 2.3.2. Gas-Distribution Measurement by Packed Sensor Array 2.3.3. Gas-Distribution Measurement by Sparse Sensor Array 2.3.4. Measurement of Continuous Distribution of Gas Using Optical Method 2.4. Presentation of Virtual Odor Source 3. Analysis of Temporal Response 3.1. Analysis of Sensor Dynamics 3.2. Sensor Dynamics Model When Instantaneous Gas Concentration Is Available 3.3. Extraction of Time Constant 3.3.1. Diffusion Model and Its Modification 3.3.2. AR Model 3.3.3. System Identification Model 3.4. Frequency Analysis 3.5. Temporal Data for Preconcentrator 4. Sensing in Both Spatial and Time Domains 4.1. Observing Change in Spatial Chemical Distribution with Time 4.2. Correlating Signal Features in Time Domain with Spatial Locations 4.3. Frequency Analysis of the Chemical Signals in Plumes 5. Conclusion 6. References

680 681 681 682 683 683 683 686 690 691 691 691 692 693 693 694 695 695 697 698 698 699 701 702 703

1. Introduction Sensors are devices to receive physical or chemical signals and to convert them into electrical signals. Physical signals are carried by waves such as electromagnetic, optical, and acoustic ones. Sensing technology for physical signals has been well-understood and has already been established. On the other hand, sensing technology for chemical signals, which are carried by chemical substances, is not matured. * Corresponding author. Tel.: +81-3-5734-2579. Fax: +81-3-5734-2828. E-mail: [email protected]. † Tokyo Institute of Technology. ‡ Tokyo University of Agriculture and Technology.

It is thought that chemical substances are moved by molecular diffusion. However, diffusion velocities of gas molecules are too slow to transport chemical signals under many conditions. The transport of chemical substances onto chemical sensors is actually governed by fluid dynamics. Fluid dynamics offers two aspects such as a signal in spatial domain and one in time domain. A signal in spatial domain is tightly coupled with a plume, a flowing trail of a chemical substance. Although the electromagnetic and acoustic waves mainly propagate straight and their behaviors are easily predicted, it is difficult to predict the behavior of the plume. Thus, it is helpful to see the plume dynamics so that people can understand the plume behavior in spatial domain. Then, the gas distribution can be measured using a homogeneous sensor array. The two types of sensor arrays such as sparse and packed sensor arrays are available. The sparse sensor array can show the global behavior of the plume, whereas the packed one reveals the local detailed behavior of the plume. Although the measurement of gas distribution is the typical method to reveal the chemical-signal behavior in spatial domain, one of the recent topics is the plume generated in a virtual environment, where people perceive sensory stimuli even if they do not stay in the actual environment. In virtual reality, people can perceive an object with smell. The direction to an odor source, the feeling of approaching or going away from it, might be realized even if the actual smell source is not in front of people; it is difficult for a chemical sensor to follow the true dynamic concentration change of the chemical substance. Generally, temporal behavior of a chemical sensor has not been well-studied in comparison with steady-state response. However, the temporal signal sometimes has information of chemical substance. Thus, the technique to know the sensor dynamics such as time constant is required. In some cases, time constant must be obtained even if the concentration profile is irregular and is not known. In some cases, the peaks of the chemical signal over time provide information of chemical substance. The temporal data from preconcentrator is also useful to obtain information on the chemical substance. In addition to raising the sensitivity, the preconcentrator with variable temperature can be used to enhance the pattern separation among odor samples. The sensing in both spatial and time domains is complicated. Although there have been many works in a single domain, a limited number of works have been addressed to combining both domains. The straightforward method to understand the combination of both domains is to observe change in spatial distribution with time. Another approach

10.1021/cr068117e CCC: $71.00 © 2008 American Chemical Society Published on Web 01/26/2008

Chemical Sensing in Spatial/Temporal Domains

Chemical Reviews, 2008, Vol. 108, No. 2 681

ature are explained. The final part is the sensing in both spatial and time domains. Observation and analysis techniques for dynamic behavior in the liquid phase are described.

2. Spatial Domain

Takamichi Nakamoto received his B.E. and M.E. degrees from the Tokyo Institute of Technology in 1982 and 1984, respectively. In 1991, he earned his Ph.D. degree in Electrical and Electronic Engineering from the same institution. He worked for Hitachi in the area of VLSI design automation from 1984 to 1987. In 1987, he joined the Tokyo Institute of Technology as a research associate. He has been an Associate Professor at the Department of Electrical and Electronics Engineering, Tokyo Institute of Technology, since 1993. From 1996 to 1997, he was a visiting scientist at the Pacific Northwest Laboratories, Richland, WA. His research interests cover chemical sensing systems, acoustic wave sensors, olfaction in virtual reality, and LSI design.

Hiroshi Ishida was born in Morgantown, WV, in 1970. He received M.E. and Ph.D. degrees in electrical and electronic engineering from the Tokyo Institute of Technology, Tokyo, Japan, in 1994 and 1997, respectively. From 1997 to 2004, he was a research associate of the Tokyo Institute of Technology. From 1998 to 2000, he visited the School of Chemistry and Biochemistry, Georgia Institute of Technology, Atlanta, GA, as a Postdoctoral Fellow. In 2004, he joined the Department of Mechanical Systems Engineering at the Tokyo University of Agriculture and Technology, Tokyo, Japan, where he is currently an associate professor. In 2007, he was a visiting researcher in the AASS Research Centre, O¨ rebro University, Sweden. His research interests are in biomimetic electronics with emphasis on chemical sensors and their applications in robotics.

is to see the correlation of signal features in time domain with several locations. The frequency analysis of the signals also provides us with useful information about an odor-source location. This paper covers dynamic behavior of a chemical sensor both in spatial and temporal domains. First, the spatial domains such as plume behavior and method of gasdistribution measurement are described. Moreover, presentation of an odor source in a virtual environment is explained. Another part is temporal-response behavior of a gas sensor. The sensor dynamics model and the analysis method are described. Then, the analysis in the frequency domain and temporal data from a preconcentrator with variable temper-

We focused on the chemical sensor signals in spatial domain. First, the plume behavior and its observation method using an optical tracer are described. Then, the methods of the gas-distribution measurements using the packed sensor array for obtaining the direction to the gas source, the sparse sensor array for obtaining the global information, and the optical method for obtaining the information remotely are explained. Moreover, the virtual environment where people can perceive spatial information of odor is introduced in this section.

2.1. Plume Behavior and Analysis Method Gas molecules are carried by air flow and distributed by turbulence. The transport of chemical substances can be visualized by a tracer such as smoke. Smoke from a chimney moves in a downwind direction, and the smoke density is dispersed and made thin by turbulence. The chemical substances emanating from the chimney are moving in the same manner as the smoke because the molecular diffusion velocities are smaller than the wind velocity. In environmental chemistry, air or water is analyzed at sites and in a laboratory. The pollutant is carried by a plume in air or water flow, and its concentration fluctuates because of the turbulence. The typical shape of the plume is illustrated in Figure 1. The plume spreads gradually from an odor source along the downwind direction. Thus, the concentration gradually decreases according to the distance from the odor source along the wind direction. However, the concentration gradient is steep across the wind direction. It is difficult to determine the direction to the odor source using only the concentration gradient because the concentration gradient along the wind direction is small and often within the noise level.1 Note that the plume shape in Figure 1 is the averaged one over time. Since its actual shape is highly fluctuated due to the turbulence, the noise level is high when we measure the concentration gradient along the wind direction. The plume behavior is governed by the fluid dynamics, which is solved using the Navier-Stokes equation. Although the numerical approach is often used to obtain the gasconcentration distribution, it consumes a long time and plenty of computational resources. Thus, a number of models for plume were proposed. The simple mathematical model of the time-averaged plume shape is available and can be expressed as the solution of the Fickian turbulent diffusion equation when the time-averaged wind speed is constant and the wind turbulence is isotropic and homogeneous.2 The instantaneous plume is very thin. The time-averaged plume shape is wider because the fluctuated plume is integrated over time, as is shown in Figure 2. Semiempirical Gaussian plume models assume a Gaussian distribution of mean concentration in the plane perpendicular to the plume center line.3 The growth of the plume width and the height are determined by the parameters called dispersion coefficients. The values of those constants for large-scale outdoor plumes under various atmospheric stability conditions were obtained from a number of experiments.4 Although the Gaussian plume models are widely used for assessing the distribution

682 Chemical Reviews, 2008, Vol. 108, No. 2

Nakamoto and Ishida

Figure 1. Typical plume shape.

Figure 2. Plume fluctuation.

of gaseous pollutants, they lack an important feature of real plumes, i.e., meanderings. Puff models assume that a plume is composed of a series of puffs released from the source over time. A Gaussian concentration distribution is generally assumed in each puff, but now the position of each puff is free to move according to the local wind direction. A variety of plume models are also proposed and used to describe other features of real plumes, e.g., rise of a buoyant plume and chemical reaction in plumes,5 but providing a detailed description of each model is beyond the scope of this review.

2.2. Plume Observation Method Study of the dynamic behavior of gas flow is laborious work because air turbulence is complicated even if the fluiddynamics analysis is performed on a supercomputer. Visualization of dynamic gas distribution using a tracer is a more realistic way to understand the dynamic behavior of the plume. A variety of simulations for localizing an odor source can be performed if the real-time image of the gas flow is obtained through visualization. An optical tracer can be used to visualize the plume. The tracer is a visible light particle that behaves in the same manner as that of the fluid. When the tracer is emitted from the source together with the odor, we can know the odor distribution when we visually observe the tracer distribution. In the gas phase, titanium tetrachloride and dry ice are used. However, they are toxic or dangerous when sufficient smoke required for the charge-coupled device (CCD) camera is used. Another candidate is oil mist, often generated by a smoke machine. It is typically used for entertainment in TV show, theater, concert, etc. Since plenty of smoke is required to obtain the clear image using a video camera, the smoke machine is appropriate from that point of view. However, the oil is deposited over the place close to the smoke machine after the experiment. The white smoke of joss sticks can be used as a tracer.6 Smoke particles from burning joss sticks are so small that the fall velocity is much smaller than typical wind velocity. The diameter of a joss stick particle is ∼1 µm, and it is easy to track the behavior of the gas molecule under the environment of typical wind velocity. It does not track the gas molecule under the environment without wind.

Figure 3. Plume observation method. Reprinted with permission from ref 6. Copyright 1998 Elsevier Science.

The advantage of joss stick (incense) is that it works as both tracer and gas molecule. When we observe the plume behavior together with gas sensor response, only the single source is required. Although joss stick is good for visualizing plume, hollow microfiber polymer particles are sometimes used.7 Since the polymer particle does not produce the signal by itself, a gas sensor does not respond to it. Thus, it can be mixed with any vapor. Next, the visualization system is described. The image of the 2D optical tracer distribution corresponding to the 2D gas distribution at the light sheet is captured by a video camera. Since it is difficult to obtain a real-time 3D image, the light sheet is illuminated to obtain the two-dimensional image, as is illustrated in Figure 3. Smoke is spouted from the nozzle in a wind tunnel, and the height of the light sheet is adjusted so that most of the smoke can be visualized. The image of the light scattered by smoke particles is captured by the video camera. That image is recorded by a VCR (video cassette recorder) and is transferred to a computer. Although there are many products of video cameras, the video camera without AGC (auto gain control) was selected because the light intensity was proportional to the gas concentration when AGC was off. We can currently obtain the real-time image in a digital format using a DVC (digital video camera), although the system in Figure 3 is a little old. The light sheet was generated by the strong illumination through the slit (60 mm in width, 2 mm in height). Although the xenon lamp (500 W) was previously used as the light source, most of light energy was discarded. A combination of a semiconductor laser with cylindrical lens or high-power

Chemical Sensing in Spatial/Temporal Domains

Chemical Reviews, 2008, Vol. 108, No. 2 683

Figure 4. Instantaneous smoke image visualizing gas concentration. Reprinted with permission from ref 1. Copyright 1999 American Chemical Society.

Figure 5. Large sensor array and visualization system. Reprinted with permission from ref 9. Copyright 1992 Elsevier Science.

light-emitting diode (LED) array is a good candidate to realize bright light sheet. An example of instantaneous smoke image visualizing gas concentration is shown in Figure 4. This is the image just above the semiconductor gas sensor in a wind tunnel. The smoke of joss stick emitted into the wind tunnel was carried by the wind toward the exhaust, and the aerial smoke trail was formed in the downwind direction. Since the shape of this trail was disordered because of the wind turbulence, the instantaneous gas concentration at the sensor place changed rapidly. The temporal behavior of the sensor will be discussed in section 3.1.

2.3. Gas-Distribution Measurement Gas distribution can be measured using many sensors. There are two types of gas-distribution measurement systems. One is a sensor array packed into a small region, and the other is an array of sensor nodes located away from each other. We discuss both types of sensor arrays for gasdistribution measurements.

2.3.1. Measurement of Spatial Gas-Distribution Change by a Large Sensor Array The initial approach to measure the spatial gas distribution was performed by Yamasaki and Hiranaka.8,9 They made the sensor array composed of the same tin oxide gas sensors as is illustrated in Figure 5. 8 × 8 sensors were spatially placed, and the distance between the two was 20 cm. The outputs of the sensors were digitally processed to form images of the spatial gas distribution on a computer screen. The image of the scene was taken by a video camera, and these two images were overlaid and displayed on a screen for easily understanding the gas field. The gas-distribution image was formed by linear or bilinear interpolation. The vapor from the liquid source placed at the center of the sensor array is visualized as is shown in parts a-c of Figure 6. Figure 6a shows the overlaid image of ethanol

Figure 6. Visualization result using a large sensor array: (a) overlaid image of ethanol vapor, (b) overlaid image of ethyl ether vapor, and (c) odor from a human body. Reprinted with permission from ref 9. Copyright 1992 Elsevier Science.

vapor. The gas distribution is expressed using grayscale, and the white cloud indicates the position of the vapor source. Figure 6b shows the ethyl ether diffusing in the vertical direction. Figure 6c shows the distribution of odor from a human body. The odor seemed to come from his socks. It was found that the spatial gas distribution could be easily grasped because of the overlaid image from the video camera.

2.3.2. Gas-Distribution Measurement by Packed Sensor Array 2.3.2.1. Strategy. The first type of gas sensor array is a two-dimensionally packed sensor array as is illustrated in Figure 7. This sensor array is used to measure local gas distribution. The temporal change of local gas distribution provides the information of the direction toward the gas source.

684 Chemical Reviews, 2008, Vol. 108, No. 2

Nakamoto and Ishida

Figure 7. 2D packed sensor array.

The strategy to search for the odor source using a gas sensor together with wind direction sensor is often used.10,11 Upon the basis of that strategy, a robotic system can trace the plume in the upwind direction. However, the capability of this system is limited by the low sensitivity of the anemometric sensors used for obtaining the wind direction. The typical wind velocity in an ordinary room, a few cm/s, is not detectable by the sensors on the system. The capability of searching for the gas/odor source can be improved if the direction toward the source is obtained not using the anemometric sensor. One of the methods to obtain the direction toward the odor source is to use the image flow of the visualized gas cloud. When we use a two-dimensional homogeneous gas sensor array, the gas-concentration distribution over the sensor array is visualized as a grayscale image. The structure of the plume in a turbulent wind field looks like a gas cloud composed of many fragments, as is shown in Figure 7. Then, the gas flow direction is obtained from the movement of the visualized gas cloud. This method is effective under the environment of low wind speed. Moreover, redundant information from the sensor array may make the system more reliable under the environment with large fluctuation due to the air turbulence. Although the instantaneous and local gas flow is not always the same as the global one, the approach of the sensor array enhances the robustness of the flow direction estimation by obtaining the averaged direction from the redundant information. 2.3.2.2. Study of Response-Speed Influence by Simulation. The requirement for gas sensors to realize a gas flow imaging system is both sensitivity and response speed. The recovery speed from the response is also important. The first trial was to make the semiconductor gas sensor array.12 5 × 5 gas sensors were placed within the area (55 mm × 55 mm). However, the gas flow observed using the sensor array was not so clear, and the brightness of the whole image changed because of the slow recovery speed. The influence of the sensor speed can be understood when the simulation is performed. The gas concentration at each sensor location is obtained using the plume-observation technique in Figure 3. Smoke of joss sticks was introduced into the wind tunnel as a tracer to simulate a gas field, and the video data captured by a CCD camera were transferred to the computer. Then, an array of virtual semiconductor gas sensors was then assumed to be placed on the visualized gas distribution in the wind tunnel. Then, the transfer function from gas concentration to a sensor response was applied at each point in the array. As a result, an image from a virtual gas sensor array was obtained.13

Figure 8. Example of gas flow images. Reprinted with permission from ref 13. Copyright 2000 Elsevier Science.

The transfer function consists of two terms with different time constants as described later in section 3.2. The two time constants change in response and recovery phases. The obtained time constants were 0.33 and 0.15 s in the response phase and 15.12 and 0.1 s in the recovery phase when the semiconductor gas sensor (TGS800, Figaro) was used. Examples of images obtained from a 10 × 10 virtual sensor array are shown in Figure 8 parts c and d with the original smoke images captured by a CCD camera in Figure 8 parts a and b. The gas flow direction can be obtained by comparing the successive images. Although the convention effect might influence the plume, it was not clearly observed in these figures. 2.3.2.3. Method of Estimating Gas-Flow Direction. Here, the method of estimating the gas-flow direction using a 2D packed sensor array is described. The gas flow vector v ) (Vx, Vy)T is obtained by using the optical flow constraint equation expressed as

∂c ∂c ∂c V + V + )0 ∂x x ∂y y ∂t

(1)

where c is gas concentration. The flow velocity is estimated by applying the least-squares method to the discrete form of eq 1. When an array is composed of N sensors, the following equations are valid assuming that u and V are constant within the array.

| | | ∂c ∂c ∂c V + | V )- | ∂x| ∂y ∂t

∂c ∂c ∂c Vx + Vy ) ∂x 1 ∂y 1 ∂t 1 2

x

2

y

(2)

2

|

|

|

|

|

∂c ∂c ∂c V + V )∂x N x ∂y N y ∂t N where ∂c/∂x|i and ∂c/∂y|i are the concentration gradients along x and y at sensor i.

Chemical Sensing in Spatial/Temporal Domains

[ ][]

Chemical Reviews, 2008, Vol. 108, No. 2 685

When X and y are defined as

| | | | | |

∂c ∂c ∂x 1 ∂y 1 ∂c ∂c X ) ∂x 2 ∂y 2 , y ) l l ∂c ∂c ∂x N ∂y N

| | l ∂c - | ∂t

∂c ∂t 1 ∂c ∂t 2 -

(3)

N

eq 2 can be expressed as

y ) Xv

(4)

Then, the estimated vˆ can be obtained using the least-squares method.14 -1 T

vˆ ) (X X) X y T

(5)

∂c/∂x|i and ∂c/∂y|i are approximated using the difference between two sensor responses. It was found from the simulation that the direction of the gas flow was correctly obtained when the time constant of response recovery was less than the array size divided by the wind velocity.13 The important factor to obtain the clear image of the gas flow is the temporal resolution since it determines the maximum flow velocity that the system can follow. Although the response time of the semiconductor gas sensor is sufficiently short (<1 s), slow recovery (>30 s) was a serious problem. Thus, QCM (quartz crystal microbalance) gas sensors15-19 were employed in the next version of the array.20 Their faster recovery enabled the visualization of the gas flow up to 5 cm/s. The temporal resolution of the system is determined not only by the recovery time of the sensors but also by the sampling rate of the sensor responses. The responses of QCM sensors are given in the form of frequency shifts. A multichannel reciprocal counter21 is implemented in the system to achieve higher sampling rates of the sensor responses than the conventional frequency counter.22 The sampling rate here was 8 samples/s, whereas it was 1 sample/s in the conventional frequency counter, i.e., binary modulus counter. 2.3.2.4. Experiment. The aim of the experiment here is to evaluate the capability of estimating gas-flow direction by a packed sensor array and to investigate the possibility of searching for an odor source by that array. The photo of sensor array and multichannel frequency countercircuit is shown in Figure 9. Twenty-one QCM gas sensors coated with sensing film (phosphatidylcholine) were mounted on the top board. In order to make a compact array, SMD (surface-mounted device)-type miniaturized AT-cut quartz resonators (27.8 MHz) were employed. Each sensor spans 4mm × 8 mm including an internally installed oscillator, and the distance between the sensors is 1.27 cm. On the bottom board, an FPGA (field-programmable gate array) chip was mounted. The 21-channel frequency shift measurement circuit was implemented into the FPGA. The sensor responses along the wind direction are shown in Figure 10. The sensor array was placed 20 cm downstream from the source of triethylamine in the wind tunnel. Sensor 1 was at the upwind edge followed by successive sensors. The frequency shift of each sensor is normalized by the maximum value during the measurement. Since the timing for the sensors to reach their maximum values agrees with

Figure 9. Photo of QCM gas sensor array and frequency counter circuit.

Figure 10. Response curves of 5 QCM sensors in a sensor array in a wind tunnel. Reprinted with permission from ref 22. Copyright 2002 Elsevier Science.

the sequence along the wind direction, the direction of the gas flow can be obtained. It should be noted that the conventional frequency counter with 1 sample/s cannot capture the difference in timing between sensors 1 and 5 because that difference is at most 1 s. The sequence of visualized image in the wind tunnel is shown in Figure 11. The white pixel means that the sensor at that place has a large response, whereas the dark one has a small response. Four corners of the image are eliminated because no sensor is placed at those places. This figure reveals that the gas flows from left to right. Figure 12 shows a histogram of the angular deviation of the estimated flow from the actual mean direction of the wind. The gas-flow direction can be estimated according to eq 5. The estimation was performed for 100 s. The deviation of the estimated directions in the range between -67° and -22°, one between -22.5° and 22.5°, and one between 22.5° and 67.5° are classified into -45°, 0°, and 45°, respectively. 74% of the estimated directions in Figure 12 are within the deviation of 67.5°. Although it may seem to be a large deviation, it is reasonable considering that the instantaneous wind direction itself was highly fluctuating. The deviation in Figure 12 is tolerable for plume tracking since the direction to track is repeatedly measured to approach the source. Once the gas flow direction is obtained, the system can track it down to the source. An experiment was performed in the laboratory room to show the system’s capability in source localization. A plastic bottle with a small hole on its

686 Chemical Reviews, 2008, Vol. 108, No. 2

Nakamoto and Ishida

Figure 14. The locations of 10 NO2 sensors in Sapporo, Japan. Reprinted with permission from ref 26. Copyright 2003 Elsevier Science.

Figure 11. Sequence of visualized images in a wind tunnel. Reprinted with permission from ref 22. Copyright 2002 Elsevier Science.

image was dark, and the gas flowed from lower left to upper right. Thus, the gas leak position was localized. The plumetracking technique based on single-point measurement cannot determine the source location until the field is thoroughly investigated for a long time.23 The gas-flow imaging system enables the plume tracking even when the wind speed is too low to be detected using an anemometric sensor. This system is applicable to most of the situation where the wind speed is <30 cm/s. Thus, it can be used in ordinal domestic or industrial buildings.

2.3.3. Gas-Distribution Measurement by Sparse Sensor Array

Figure 12. Histogram of errors in estimated direction obtained in wind tunnel.

Figure 13. Visualized images around the gas leakage point. Thick arrows show direction of gas flows observed in visualized images. Reprinted with permission from ref 22. Copyright 2002 Elsevier Science.

side was prepared and the triethylamine vapor was leaked from that hole, as is shown in Figure 13. The gas flowed from left to right at locations a and b. Then, the sensor array was placed at location c. Since the gas flow from left to right was observed at location c, the sensor array was moved to location d. The upper part of the

2.3.3.1. Gas-Distribution Measurement for Monitoring Air Pollution. The packed two-dimensional sensor array is used to raise the estimation accuracy of the direction to the gas source within a short time. However, the sensor nodes are sparsely placed in a huge area when we monitor gas distribution in space. In this section, the measurement by a sparse sensor array is described. In most of the cases, the gas distribution is measured to monitor air pollution such as NO2, SPM, Ox, SO2, and CO. Those are measured periodically by government stations, and the data are available from the web sites in almost real time24,25 in Japan. However, the gas analyzer used in the station is very expensive, and the number of the stations is limited. Thus, the distance between the stations is long. The stations can be densely placed if cheap and reliable sensors are available. Maruo and co-workers measured NO2 distribution in Sapporo, Japan, using a porous glass substrate impregnated with a Saltzman reagent, an LED, and a photo detector.26,27 The sensor element was a porous glass substrate with sulfanilamide (SFA) and N,N-dimethyl-1-naphthylamine (DMNA). A specific absorption peak appears at 525 nm, and the absorbance increases as exposure time or concentration increases. The interference from NO in the analysis of NO2 was not significant. A photo detector received the light from a LED through the porous glass substrate. Its voltage was read by a microcomputer through an A/D (analog to digital) converter. The pump was not used because the wind blew constantly, and its performance was not influenced by the wind speed with the range between 0.5 and 5 m/s. Those sensors were placed around the roadway intersection, as is shown in Figure 14. The measurement area is an

Chemical Sensing in Spatial/Temporal Domains

Chemical Reviews, 2008, Vol. 108, No. 2 687

Figure 16. Monitoring stations around the measurement site. Reprinted with permission from ref 28. Copyright 2005 Elsevier Science.

Figure 15. Measured NO2 concentrations at all sites (July 2001). Reprinted with permission from ref 26. Copyright 2003 Elsevier Science.

inner-city area with two main roads of high traffic density (L1 and L2) and two tall buildings (B1 and B2), a large parking area (P), and markets (M). Approximately 24 000 vehicles pass these two main roads. They set a network composed of 10 sensor nodes to look at spatial and temporal variations of NO2 concentrations every hour. The measurement data during 24 h are shown in Figure 15. It shows the averages of 9 values from 9 different days. The typical pattern consists of a high level in the morning and in the evening and a low level during the night and in the afternoon. The proximity of road traffic increased NO2 level. The small peak early in the morning was related to the peculiarity of this area where many trucks waited for the markets to open near the road early in the morning. At all the nodes 8, 9, 6, and 10 behind the tall building with southeast wind, the NO2 level was low in the middle of night and the highest around noon. They said that the phenomenon of “street canyon” due to the tall buildings might occur. It was found that the concentration distribution was influenced by tall buildings and wind direction when they checked the wind direction. Tsujita et al. reported the sensor network made up of semiconductor gas sensors to monitor NO2.28 The temperature and humidity sensors were also included in a sensor node to compensate for the sensor response. Moreover, a new autocalibration method was proposed to achieve maintenancefree operation for the sensor network. The lack of long-term stability is a serious problem of a gas sensor, which causes the deterioration of the measurement accuracy over time. Frequent recalibration is not realistic. The network connectivity can be used not only for collecting the measurement data but also for the calibration

and diagnosis of the sensors. The measured concentration can be easily compared through the network with those measured at nearby sensor nodes at government monitoring station. Although the different concentrations are usually monitored at different sites, the pollutant concentration in the whole local area becomes uniform in a certain weather condition. Each sensor in a network can be calibrated at that condition. The case study was performed in the area shown in Figure 16. The semiconductor gas sensor and NOx analyzer were placed at the university (Tokyo Institute of Technology). The four stations around that university were the government monitoring stations with gas analyzers. The hourly mean data measured at those stations are available at the web site. The autocalibration procedure is as follows. In normal circumstances, NO2 concentration in an urban area is high. In that case, the concentration within this local area is not uniform. However, it would be reasonable to consider that the concentration at the monitoring site is almost zero when the concentrations of all the stations around the monitoring site were near zero. This unusual low concentration might be obtained due to small traffic on a national holiday or extremely large dilution of NO2 in a stormy day. The gas sensor output around the baseline can be calibrated on this special case by assuming the uniformity of the concentration. A sensor at a node connected to a network is automatically calibrated using this assumption. The raw sensor data and the data reported at the environmental monitoring stations were collected for several months. The initial calibration of the gas sensor was performed immediately before starting the long-term measurement. Two months after the gas sensor was calibrated, the measurement error increased to ∼40 ppb. Figure 17 includes the occasion when the calibration was performed. Dashed line shows that calibrated sensor value after adjusting the sensor baseline at 1445 h to the average of the concentrations reported from surrounding stations. The sensor output after the baseline adjustment agreed well with NO2 concentration obtained from NOx analyzer. This autocalibration method was effective to maintain the accuracy of the sensor system in an atmospheric-monitoring network. Another point of view in a sensor placed in the ambient air is a vapor-sampling method. In the case of a packed sensor array, sensors are directly exposed to the ambient air. On the other hand, the air is sometimes sucked by a pump at each sensor node in the case of a sparse sensor array.

688 Chemical Reviews, 2008, Vol. 108, No. 2

Nakamoto and Ishida

Figure 19. Discoloration of gas detector tube exposed to methyl mercaptan.

Figure 17. Comparison of NO2 concentration with gas sensor and NOx analyzer when autocalibration was performed. Reprinted with permission from ref 28. Copyright 2005 Elsevier Science.

Figure 18. Photo and block diagram of sensor node for hydrogen detection. Reprinted with permission from ref 31. Copyright 2005 IEE of Japan.

However, sampling schemes for sensors to detect vapors are important. Settles reviewed the fluid dynamics about sniffers for the point-detection devices.29 It was also reported that the vapor discrimination capability was enhanced when a sensor chamber mimicking a nasal cavity was used.30 2.3.3.2. Gas-Distribution Measurement for Monitoring Hydrogen Leak. Gas distribution is also measured to check the leakage of combustible or toxic gases. Gas distribution should be monitored real time to localize the leakage location. A wireless network for detecting hydrogen leak was proposed.31 Hydrogen filling stations are expected to become common in the near future when fuel-cell cars are widely used. Thus, a sophisticated hydrogen monitoring system is required. The employed sensor is a FET (field effect transistor) gas sensor32 with a catalytic metal gate (Pd), which works at 100 °C. Heating the sensor reduced the humidity influence and the response time. They constructed a prototype sensor network to detect hydrogen using 10 sensor nodes. A photograph and block diagram of a sensor node is shown in Figure 18. Each sensor node has two sensors, a readout circuit, a controller, a communication unit, and a power supply. The controller

includes a microcomputer with low power consumption and short rise time from standby mode to active mode. The communication unit sends and receives signals at 430 MHz. The communication area was wide enough to cover a hydrogen filling station (at least 100 m2). The function of reducing the power was implemented. Initially, the sensors worked at room temperature without heating. Once the sensor signal exceeded the threshold, the microcomputer woke up to change from standby mode to active mode. Then, it switched the heaters attached to sensors. Each sensor node consumed much power only when hydrogen was detected. Testing showed that the signal came to the access point within 0.3 s after the sensor was exposed to hydrogen. 2.3.3.3. Bad-Smell Sensing Network. The final topic in this subsection is a bad-smell sensing network composed of gas-detector tubes. Recently, the deterioration of the environment caused by bad smell has become one of the problems in daily life. Although the gas chromatograph/mass spectrometry (GC/MS) method is often used to analyze gases, it is time-consuming. Moreover, it is impossible to do the onsite monitoring when GC/MS is used. Gas-detector tubes were used here since they are cheap and easily handled.33 In the gas-detector tube, the fundamental function of the chemical reaction between the analyte and the reagent system is to form color compounds that make the reaction visible.34 A gas concentration is visually obtained by reading the length affected by the color change. Although the discolorationlayer length has been manually read, the optical sensor enables the automatic measurement of the gas detector tube. The automatic measurement increases the accuracy and the reproducibility of the measured data. Moreover, the sensitivity is enhanced by the continuous gas sampling and capturing the image of the gas-detector tube because of the accumulation effect. Three types of image sensors such as A4-size optical scanner,35 one-dimensional CCD sensor,36 and mobile phone camera37 were utilized to capture the image of the gas-detector tube. When each sensor system is equipped with a communication module, the sensing network is realized. The bad-smell sensing network was constructed and was applied to the measurement at the paint factory.38 The photo of the gas-detector tube (Gastec, No. 71) is shown in Figure 19. When methyl mercaptan vapor was flowed to the gas-detector tube, its color changed from white to yellow. The selectivity of the gas-detector tube is larger than that of typical gas sensor, and a few hundreds kinds of gas-detector tubes are commercially available. The disadvantage of the gas-detector tube is its irreversible reaction. Once the discoloration occurs along the whole length, it cannot be used again. Since the brightness change along the detector tube axis is not smooth, the noise reduction technique was applied to the image of the gas-detector tube.35 Using the method in ref 35, the discoloration-layer length was obtained. The result is shown in Figure 20 when several temporal changes of gas concentrations were tried. The gas was supplied from

Chemical Sensing in Spatial/Temporal Domains

Chemical Reviews, 2008, Vol. 108, No. 2 689

Figure 22. Brightness distribution along axis of gas detector tube for 30 ppb methyl mercaptan. Reprinted with permission from ref 36. Copyright 2006 Elsevier Science. Figure 20. Temporal change of discoloration-layer length under several gas concentration profile (methyl mercaptan). Reprinted with permission from ref 35. Copyright 2004 IEE of Japan.

Figure 23. Result of gas distribution measurement at corridor using bad-smell sensing network composed of gas detector tubes.

Figure 21. Photo of gas detector tube system combined with onedimensional CCD image sensor.

the blender based upon high-speed switching of a solenoid valve.39 The concentration was a relative one with the unit [% RC]. 100 [% RC] was the maximum concentration and corresponded to 2-2.5 ppm. The two curves in the left portion correspond to concentrations 100 and 30 [% RC]. It was found that the slope of the curve became larger when the concentration became higher. The two curves in the right portion were obtained when the vapor with the concentrations 10, 0, 30, 0, 100, 0 [% RC] was exposed for each corresponding 4 min. The reproducibility was good because the two curves with the same concentration profiles were almost overlapped. Moreover, the discoloration stopped when the tube was exposed to air at 0 [% RC]. The discoloration stopped at 4.5 cm because that point is the edge of the reagent region. A photo of the gas-detector tube system combined with a one-dimensional CCD sensor is shown in Figure 21. It includes a one-dimensional CCD, LEDs for illumination, a microcomputer, and wireless LAN (local area network) modules. The gas-detector tube was put into the black acrylic box to prevent influence of the light from the outside. Using this system, the gas (methyl mercaptan) with low concentration (30 ppb) was measured as is shown in Figure 22. The accuracy of measuring the length of the discoloration layer was much better than that of typically used manual inspection because it was difficult to check the brightness change within the region of 1 mm by manual inspection. However, a tiny change of the discoloration layer was clearly captured by the CCD image sensor. The sensitivity was improved by 1 order of magnitude compared with the manual inspection. Since this system has the function of communication, the data from the multiple sensor nodes can be collected through the wireless LAN. The six sensor nodes were placed at the

corridor of the building. The measurement area was 2 m × 20 m, and the sample was propion aldehyde. Sensor nodes 5 and 6 were 20 m away from the source, and those nodes were placed outside the building. The petri dish filled with the liquid of propion aldehyde was placed at the corridor, and the concentration just above the petri dish was 18 ppm. Figure 23 shows the gas distribution obtained from six sensor nodes. The gas-detector tube responded to the vapor at the place even 20 m away from the source. Although the highest concentration was measured at sensor node 2, it was ∼200 ppb. The response at node 4 was higher than that at node 3, whereas the response at node 6 was higher than that at node 5. This phenomenon might be caused by the change of the wind direction along the corridor, as is illustrated in Figure 23. This result indicates that it is possible to do the field measurement using a bad-smell sensing network composed of gas-detector tubes. Although the image of a single gas-detector tube is captured using a one-dimensional CCD sensor, the image of multiple gas-detector tubes can be simultaneously captured when two-dimensional image sensor is employed. Moreover, it is better to collect the data even when the sensor nodes are far away from each other. Thus, the mobile phone camera was used because of its communication capability. The photo of the sensor node using the mobile phone camera is shown in Figure 24. The light source and the multiple gas-detector tubes were placed inside the black acrylic box. The mobile phone camera was placed at the top of the acrylic box. It was controlled by a microcomputer, and the image was periodically captured. The measurement can be automatically performed. The concept of a bad-smell sensing network using a gas-detector tube and a mobile phone camera is illustrated in Figure 25. Each sensor node sends the host computer the image file attached to the e-mail. The host computer collects the image files from many nodes, and the gas distribution is obtained after the analysis. Using this system, it is possible to collect the data from many sensor nodes located far away from each other.

690 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 24. Photo of sensor node using mobile phone camera. Reprinted with permission from ref 37. Copyright 2007 Elsevier Science.

Figure 25. Concept of bad-smell sensing network using gas detector tube and mobile phone camera.

2.3.4. Measurement of Continuous Distribution of Gas Using Optical Method 2.3.4.1. Backscatter Gas Absorption Imaging. Backscatter absorption gas imaging (BAGI) is a technique to realize the real-time visualization of gas plume using laserbased remote sensing.40 A field is illuminated by a narrowbandwidth infrared laser. Gas visualization occurs if a plume absorbing the laser radiation is located in that field. Its presence causes a dark plume image in the video picture. The description of the long-range BAGI imager is summarized here.41 The imager operates in a raster-scanned mode to achieve real-time laser-illuminated imaging at a wavelength between 9 and 11 µm tuning range of the CO2 laser. Scanning of both the 18 W continuous-wave CO2 laser beam and the instantaneous field-of-view of the single-element detector is accomplished using a pair of galvanometrically driven mirrors. The imager viewed the release of sulfur hexafluoride gas at the range of 90 m in the test field. Figure 26 shows BAGI images under condition of no gas, 3 ppm SF6, and 40 ppm SF6. The dark plume images were observed when SF6 gas was present. There are a few aspects to be considered when BAGI images are observed. First, gas imaging requires a hard target in the imager field-of-view. In the case of Figure 26, a 12 ft2 panel was placed to serve as a backscattering surface. Second, BAGI requires the spatial contrast in gas concentration. It is not possible to detect uniformly distributed gases. Third, BAGI has a large signal-to-noise ratio compared with the passive gas visualization method because of active laser illumination.

Nakamoto and Ishida

Figure 26. BAGI plume image taken at a range of 90 m. Top left image is taken at no gas release. Top right image is taken with 3 ppm SF6 release; lower image is taken with 40 ppm SF6 release. Reprinted with permission from ref 41. Copyright 1997 SPIE.

Figure 27. Experimental setup for gas-distribution measurement using photoacoustic effect.

Another approach employed the differential absorption mode.42 In this mode, two wavelengths, where the light energy is absorbed and not absorbed by the gas to be detected, are used, respectively. Images are displayed as the natural logarithm of the ratio of the on wavelength absorbed by the gas to the off wavelength not absorbed by the gas. Using this technique, the topographic scene image is removed, whereas the plume image is displayed. Thus, an optical method can be used to obtain a gas plume image. Nagashima et al. proposed the wavelength-differential image obtained by subtracting two images through the variable interferometer with different transmission spectra. They observed the plume of butane using this method.43 2.3.4.2. Measurement of Gas Distribution Using Photoacoustic Effect. This method is a combination of an optical method with an acoustic one. Ochiai et al. studied the measurement system of the gas distribution using photoacoustic effect.44 The experimental setup is illustrated in Figure 27. Adsorption of light at a certain wavelength occurs because of the existence of gas. The absorbed energy is converted to heat, followed by expansion. When the light illumination is periodically performed, the expansion and contraction occur synchronously with light illumination. Then, the sound is generated. The laser beam is twodimensionally scanned. The generated sound is detected by three microphones placed three-dimensionally. The method of localizing a sound source can be used to determine the location of the gas source. Ochiai et al. applied this method to the localization of methane source. This method is fascinating because the gas source can be replaced with a sound source and the problem of the sound-source localization in place of the gas-source one is solved.

Chemical Sensing in Spatial/Temporal Domains

Chemical Reviews, 2008, Vol. 108, No. 2 691

Figure 28. Odor field presentation system. Reprinted with permission from ref 45. Copyright 2006 IEEE.

2.4. Presentation of Virtual Odor Source Recently, several researchers were interested in forming the odor plume in a virtual environment.45 People can perceive the object with smell in that environment. In virtual reality, an olfactory display to present smells is focused on.46 The direction to an odor source, the feeling of approaching or going away from it, may be realized even if the actual odor source is not in front of people. Thus, one of the research studies about an olfactory display is to localize an odor source in a virtual environment. Yamada et al. proposed the wearable olfactory display as is shown in Figure 28. The wearable olfactory display is equipped with a man in the photo. The gas concentration can be controlled using this system, and a tag reader and a radio frequency identification (RFID) tag were used to determine his position. The smell flowed from the slit in the adaptor indicated as the output in this figure. The intensity of the presented odor was controlled by changing the ratio of the air flow rate to that of the odor flow rate. The flow was driven by a DC motor air pump. By changing the concentration of the gas presented from a wearable olfactory display according to the position obtained from the tag reader, a gas field model in a wide space was realized in a virtual environment. When a subject approached the virtual odor source, the intensity of the odor from the slit increased. On the other hand, the odor intensity decreased as the subject was away from the virtual odor source. The gas field model employed here is an isotropic solution, assuming that the diffusion equation without air stream is valid. Although this assumption is not valid in the actual case because gas molecules are mainly carried by the air flow, they used this gas field for the first step to realize the virtual olfactory environment. Several subjects tried to find the virtual odor source within the area 18 m × 9 m. Tracking behaviors of the odor source and temporal changes of gas concentrations are shown in Figure 29. From that figure, both subjects reached the vicinity of the odor source by trial and error, changing their directions on the basis of the sensation of changing odor. They said that the subjects could perceive spatial information of odor presented by the wearable olfactory display. The subjects took two ways to explore the virtual odor source. Subject C chose the strongest point of odor after searching through the entire region. On the other hand, subject D walked randomly and changed the direction so that a stronger perception of the odor could be obtained. This is a very simple case of the presentation of the virtual odor source. In the actual situation, the odor plume is governed by the wind. This behavior is obtained by solving the Navier-Stokes equation. This is the fluid dynamics problem to be solved using a finite-element method, and it

Figure 29. Behavior of tracking virtual odor source and temporal concentration change according to subject’s position. Reprinted with permission from ref 45. Copyright 2006 IEEE.

consumes much time to obtain the exact three-dimensional solution. The tradeoff between the time and the accuracy should be examined when the odor source in the virtual environment is studied.

3. Analysis of Temporal Response In addition to the traditional static method, the dynamic behavior of the gas sensor offers useful information of odor identification. First, the sensor dynamics model and its analysis method are described. Especially, the methods of extracting time constants based on the diffusion model and the autoregressive (AR) model are presented. The application of the system-identification model to the analysis of the gas sensor is also explained. The frequency analysis is often useful for extracting the features from the temporal data. It is also described here. Finally, the temporal-data analysis of the preconcentrator with variable temperature with its interesting feature is shown. Thus, the analysis of the sensor dynamics is reviewed in this section.

3.1. Analysis of Sensor Dynamics It is required to obtain both input and output signals when we determine the model of sensor dynamics. However, it is not always possible to know input signal, i.e., the temporal change of vapor concentration. Thus, a certain situation should be considered to know the input signal. The simplest method to analyze the dynamic behavior is to obtain the step response of the gas sensor.47 They reported that the model parameters derived by fitting the model to the experimental data of the step response represented the type of the gas and its concentration. However, the gas concentration change should be much faster than the time constant of the gas sensor. Moreover, the concentration should be kept constant after the beginning of the vapor exposure. The second method is to obtain the impulse response of a gas sensor.48 When we use the simple vapor supply system as is shown in Figure 30, it is very difficult to keep the concentration constant during the vapor exposure. The air is flowed from the sample bottle, and the vapor at the headspace over the liquid is carried to the sensor. Although the concentration is high at first, it gradually decreases during the vapor exposure because the flow rate typically exceeds the evaporation rate of the sample. Since this dynamics is

692 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 30. Simple method of supplying headspace vapor.

Nakamoto and Ishida

Figure 32. Sensor response measured simultaneously with gas concentration obtained using optical tracer. Reprinted with permission from ref 6. Copyright 1998 Elsevier Science.

In this subsection, three approaches based upon step response, impulse response, and irregular change of the vapor concentration were described. If the linear superposition theorem is valid, the impulse response approach is useful because just the simple vapor supply method is sufficient. However, the final approach should be taken if the nonlinear property is dominant.

3.2. Sensor Dynamics Model When Instantaneous Gas Concentration Is Available Figure 31. Vapor concentration change in pulse vapor supply method.

complicated, it is not easy to model the concentration change at the outlet. Thus, the sensor was exposed to a very short vapor pulse, as is illustrated in Figure 31. The concentration change at the sample bottle can be ignored if the vapor pulse is much shorter than the time constant of the sensor. Although the same vapor supply method was used, the time for vapor exposure is very short: just 1 s. In this situation, we do not have to know the input signal, i.e., vapor concentration. In the third method, the vapor concentration at the sensor is allowed to be dynamically changed. In this case, it is required to know the vapor concentration change at the sensor precisely. One of the methods is to use a sensor with a very fast response.49 Although that sensor has a time constant of a few milliseconds, the gas species to be detected are limited. Thus, the visualization method of the gas flow using an optical tracer described in earlier section was used. In this approach, an optical tracer is effective to obtain the instantaneous vapor concentration at a sensor using the gas-distribution visualized system shown in Figure 3. The brightness change at the CCD camera was measured simultaneously with the tin oxide gas sensor response to the sample vapor from the joss stick, as is shown in Figure 32. The gas sensor (TGS800, Figaro) was placed just below the light sheet, and the mean brightness within 10 × 10 pixels over the area of the gas sensor was obtained. The sampling intervals of gas sensor response and brightness were 0.1 and 0.04 s, respectively. The gas concentration changes very rapidly in the figure since the plume was meandering, as is shown in Figure 4. Therefore, the brightness calculated from the image fluctuated rapidly and resulted in the irregular train of pulses, considerably different from the gas sensor response. A similar phenomenon was observed elsewhere.50 The response and recovery speeds of the gas sensor are not enough to catch up with the concentration pulses in Figure 32. These concentration pulses can be used as input signals of a sensor response model described in the next subsection.

There are several sensor dynamics models.51,52 The simplest model is a linear one governed by the equation below.

dny(t) n

dt

+ a1 b0

dn-1y(t) n-1

dt dmu(t) m

dt

+ ‚‚‚ + an-1

+ b1

dm-1u(t) m-1

dt

dy(t) + any(t) ) dt du(t) + dt bmu(t) (n g m) (6)

+ ‚‚‚ + bm-1

where y(t) is a sensor response, u(t) is the gas concentration, and ai and bi are the coefficients. Laplace transform is performed by

Y(s) ) ∫0 y(t) e-st dt ∞

(7)

In the form of Laplace transform, the transfer function G(s) is

G(s) )

b0sm + b1sm-1 + ‚‚‚ + bm sn + a1sn-1‚‚‚an-1s + an

(8)

When this model was applied to the response of the semiconductor gas sensor as is shown in Figure 32, a secondorder differential equation

d2y(t) 2

dt

+ Ri

dy(t) + βiy(t) ) gils(t) dt

(9)

is assumed, where ls(t) corresponds to the steady-state response to the vapor with the concentration at time t. ls(t) can be in advance obtained from the calibration curve. Ri, βi, and gi are the coefficients. In the case of the semiconductor gas sensor, the rise time of the response is much shorter than the recovery one. Thus, the waveform is divided into two phases such as response and recovery

Chemical Sensing in Spatial/Temporal Domains

Chemical Reviews, 2008, Vol. 108, No. 2 693

Figure 33. Scheme for dividing a waveform into two phases such as response and recovery phases. Reprinted with permission from ref 6. Copyright 1998 Elsevier Science.

Figure 35. Structure of MLP (a) in the training and (b) in the estimation when transient gas sensor response is estimated using an optical tracer.

Figure 34. Comparison of estimated semiconductor gas sensor response with measured one when optical tracer and second-order model were used. Reprinted with permission from ref 6. Copyright 1998 Elsevier Science.

phases, as is illustrated in Figure 33. In eq 9, i is equal to 1 for the response phase, whereas it is 2 for the recovery phase. Equation 9 is transformed into the discrete form

y(k + 1) ) piy(k) + qiy(k - 1) + rils(k)

(10)

where y(k) ) gas sensor response at time k∆t; ls(k) ) transformed steady-state sensor response corresponding to the brightness at time k∆t; ∆t ) sampling interval (0.1 s); pi, qi, and ri ) constants; i ) 1, response phase; and i ) 2, recovery phase. If ls(k) < y(k), y(k) is in the response phase, whereas it is in the recovery phase if ls(k) > y(k). The parameters pi, qi, and ri in eq 10 were estimated for the response phase and the recovery one, respectively. Comparison of the estimated response with the measured one is shown in Figure 34. The estimated gas sensor response agreed well with the experimental one. The model above is effective when we use a slow-speed gas sensor. However, it is not sufficient for a faster-speed gas sensor since the switching between the response and recovery phases is too frequent. The deviation was larger when a QCM (quartz crystal microbalance) gas sensor was used. Thus, another approach is to construct a model using a neural network.53 When we use a neural network, it is not necessary to divide waveform into response and recovery phases because it is a nonlinear technique. A neural network employed here was a MLP (multilayer perceptron) trained with error back-propagation algorithm.54,55 It is possible to realize any continuous function from input to output using a three-layer network, given a sufficient number of hidden units, proper nonlinearities, and weights.56 Thus, a neural network illustrated in Figure 35 was used. The sensor response at k + 1 is obtained using ls(k + 1), ls(k), y(k), and y(k - 1), as is illustrated in Figure 35a. The output of the

Figure 36. Comparison of estimated QCM gas sensor response with measured one when optical tracer and MLP model were used. Reprinted with permission from ref 7. Copyright 2002 IEICE of Japan.

neural network, i.e., the estimated sensor response, is fed back to the input layer when the estimation is performed, as is illustrated in Figure 35b. The experimental result is shown in Figure 36. The QCM gas sensor (29.5 MHz, AT-cut) coated with a sensing film, phosphatidylcholine, was exposed to triethylamine vapor, and a tiny hollow polymer sphere was used as an optical tracer. The sensor data were sampled every 170 ms using a reciprocal counter.21 The numbers of input, hidden, and output layer neurons were 4, 20, and 1, respectively. It was found that the estimated sensor response agreed well with the experimental one even when the sensor with the speed faster than that of the semiconductor gas sensor was used.

3.3. Extraction of Time Constant 3.3.1. Diffusion Model and Its Modification An odor-sensing device often called an electronic nose consists of a sensor array and a pattern-recognition technique. The output pattern of a sensor array with partially overlapping specificities is recognized by a neural network or multivariate analysis. Although there are a variety of gas sensors such as semiconductor gas sensors,57-62 QCM gas sensors,63-72 SAW (surface acoustic wave) gas sensors,73-77 cantilevertype gas sensors,78,79 FPW (flexural plate wave) gas sensors,80 conducting polymer gas sensors,81-84 carbon black polymer gas sensors,85-87 MOS (metal oxide semiconductor) gas sensors,88-90 MS (mass spectrometry),91-93 IMS (ion mobility spectrometry)94 high-speed gas chromatographs,95 optical gas sensors,96-99 and electrochemical gas sensors,100 more in-

694 Chemical Reviews, 2008, Vol. 108, No. 2

Nakamoto and Ishida

ys(t) )

[

∞

ys1 1 -

∑

n)1

8 (2n - 1)2π2

((

2n - 1

exp -

2l

) )] 2

π Dt

( ( ))

m1 1 - exp Figure 37. Gas sorption at sensing film of QCM gas sensor.

formation is required to achieve reliable discrimination. Thus, time constant as well as amplitude information has been studied to use an element of a pattern vector. When we obtain the time constant, a sensor-response model is required. First, the sensor-response model based upon the diffusion model is described. When a QCM gas sensor was considered, vapor diffused into a coating film, as is illustrated in Figure 37. Gas molecules diffuse into a sensing film with the thickness l according to Fick’s law.101 The concentration in the sensing film c(t,x) is governed by

∂c(t,x) ∂2c(t,x) )D ∂t ∂x2

(11)

under the boundary conditions

c(0,x) ) 0

|

∂c )0 ∂t x)l

(12)

where C0 is the concentration in the gas phase. Solving eq 11 under the boundary conditions above, ∞

4C0

∑ (2n - 1)π exp

n)1

((

-

2n - 1 2l

) )

π 2Dt sin

2n - 1

(

( )) (

ys(t) ) m1 1 - exp -

t τ1

( ))

+ m2 1 - exp -

t τ2

(18)

can be used for the curve fitting of a QCM gas sensor.101 Moreover, the time constant in the response phase is different from that in the recovery phase in the same manner as that of the semiconductor gas sensor in the previous section. Since the time constant in the recovery phase is longer than that in the response phase, it is easier to analyze the waveform during the desorption so that many data points can be used for the curve fitting.

In contrast to the method above, the AR (autoregressive) model is also effective to determine the time constant.102 The AR model is typically used to estimate parameters in a dynamic system. It is assumed that ys(k) satisfies the AR model

2l

πx (13)

where L is the order of the model, e(k) is the residual error, and Ri is the scalar coefficient. By solving the equations

∂

l

C0l 1 -

∞

∑

n)1

8 (2n - 1)2π2

( (

exp -

2n - 1 2l

) )) 2

π Dt

(14)

∞

∑

n)1

8 (2n - 1)2π2

∑ birik,

i)1

ri ) exp(-∆t/τi)

(21)

(15) is obtained where τi is the time constant and bi is the scalar coefficient. The Z-transform of eq 21 is

((

exp -

(20)

k

L

Thus,

)1-

∑ e2(k) ) 0

the optimal Ri can be obtained. Assuming that ys(k) is the sum of L exponentials,

ys(k) )

At equilibrium,

ys(∞) ) C0l

(19)

i)1

∂Ri

ys(t) ) ∫0 c(t,x) dx )

ys(∞)

(17)

can be used for the curve fitting where m1 and τ1 are the amount of adsorption and its time constant, respectively, assuming that the speed of the surface adsorption is much faster than that of diffusion. Practically, the term without n ) 1 can be ignored in most cases. Then, the simple equation using two time constants

ys(k) + ∑ Riys(k - i) ) e(k)

is obtained, assuming the diffusion coefficient D is not dependent on c(t,x), the concentration inside the sensing film. The amount of sorption ys(t) is

ys(t)

τ1

L

c(t,x) )

(

t

3.3.2. AR Model

c(t,0) ) C0

C0 -

+

2n - 1 2l

) )

bi

L

2

π Dt

(16)

is obtained. When the diffusion governs the vapor sorption onto the sensing film, the experimental data fits well to the curve of eq 16. However, the curve fitting was sometimes unsuccessful. Not only the diffusion but also the surface adsorption should be taken into account. Thus,

Z[ys(k)] ) ∑ i)1

1 - riz-1

)

B(z-1) A(z-1)

(22)

The denominator of Z[ys(k)] is

A(z-1) ) (1 - r1/z)(1 - r2/z)‚‚‚(1 - rL/z) On the other hand,

(23)

Chemical Sensing in Spatial/Temporal Domains

Chemical Reviews, 2008, Vol. 108, No. 2 695

L

Z[ys(k)] + ∑ Riz-iZ[ys(k)] ) Z[e(k)]

(24)

i)1

using eq 19. Then,

Z[ys(k)] )

Z[e(k)] L

(25)

1 + ∑ Riz-i i)1

Using eqs 22, 23, and 25, it is found that ri is the solution of A(z-1) ) 0. Thereafter, the time constant τi is obtained. Nakamura et al. applied this method to the step response of a QCM gas sensor and obtained the time constants with high accuracy.102 Using this method, the time constants of acetone, 2-butanonone, methanol, ethanol, benzene, and toluene were obtained. The PCA (principal component analysis) result of the steadystate responses of six sensors and those of time constants are shown in parts a and b of Figure 38. It was found that the separation among samples became clear when the time constant information was included. They also proposed the method using linear filter and plural sensors even when the gas concentration was changed.103

3.3.3. System Identification Model The effort to build a mathematical model of a gas sensor as a dynamical system has been done from the viewpoint of system identification. Here, the attempt to model the sensor response to the gas mixture is described. The gas sensor is modeled as a MISO (multi-input single-output) system.104 When the nonlinearity is included, the three methods such as functional expansions, block-structured network model, and neural network were proposed. Functional expansions are valid representations of nonlinear systems under assumptions (stationarity). In the case of a nonlinear time-invariant system, the transfer function can be expressed as a Volterra functional expansion which includes n kernels. The block-structured model consists of interconnections of two different classes of blocks such as dynamic linear blocks and static nonlinear blocks. Figure 39 shows several possible typical topologies for modeling the sensor response to a binary mixture.105,106 These models are easier to implement compared with the kernel representation. Another method is the neural network. The neural network can model a highly nonlinear relationship if there is enough hidden-layer neurons. Time-delayed and recurrent-type neural networks were used to obtain the concentration changes of the binary mixture (octane and toluene).107 Although these methods might be effective to represent the dynamic model of the gas sensor, only the slow dynamics were focused on in the literature.105,106 It is interesting if faster dynamics around a second is studied.

3.4. Frequency Analysis Frequency analysis is useful when we try to extract information from the transient response as much as possible. Amrani et al. reported that the frequency characteristic of dissipation factor of a conducting polymer gas sensor had the information of gas discrimination.108 The vapor in the headspace above the liquid was flowed to the sensor cell, and the dissipation factor was measured using an impedance

Figure 38. PCA diagram of (a) only normalized saturation-mass vector and (b) normalized saturation-mass vector and time-constant vector. Gases are acetone (O), 2-butanone (9), methanol (0), ethanol ([), benzene (]), and toluene (2). Concentration of each sample ranges from 30 to 3000 ppm. Reprinted with permission from ref 102. Copyright 1993 IEE of Japan.

Figure 39. Several possible block-structured models of a sensor response to binary gas mixture. Reprinted with permission from ref 105. Copyright 1996 Elsevier Science.

analyzer. The measurement data of the acetone-methanol binary mixture is shown in Figure 40. It was found that the spectrum changed according to the composition. They said that the spectrum information was useful for the quantification of the multiple components even if only a single sensor was used. The multiexponential models such as Gardner transform, METS (multiexponential transient spectroscopy), PadeLaplace, and Pade-Z were applied to the analysis of conducting polymer gas sensor responses. It was reported that Pade-Laplace and Pade-Z models had better resolution capabilities than the other two methods.109 Fourier analysis and wavelet transform were often used to analyze the signals of the temperature-modulated gas sensors including the normal semiconductor gas sensors and microhotplate sensors.110-112 Spectrum analysis is effective to extract the feature of the waveform. However, most of

696 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 40. Frequency characteristic of dissipation factor of conducting polymer gas sensor. Reprinted with permission from ref 108. Copyright 1998 Elsevier Science.

Nakamoto and Ishida

Figure 42. Concept of robustness enhancement against plume behavior using short-time Fourier transform.

the time constant in the time domain corresponds to the cutoff frequency in the frequency domain. This concept is illustrated in Figure 42. When the time constant is different, a different cutoff frequency is obtained, even under the irregular and rapid change of the gas concentration. Good pattern separation is obtained if the appropriate frequency component is selected. Moreover, the real-time odor classification is required in the actual situation. Since many sampling points are typically required for the spectrum analysis, it takes much time to collect the data. The short-time Fourier transform (STFT) is the pseudo-real-time technique where the data in the moving window are used for the analysis. The spectrum S(m,ω) at time m in the discrete form is ∞

S(m,ω) )

∑

s[n]w[n - m] e-jωn

(26)

n)-∞

where w[n] is the window function and n and m are integers. s[n] is the sensor response signal in the time domain. One of the window functions is Hann Window, expressed as Figure 41. Sensor responses to odors in dynamically changing concentrations (a) apple and (b) Muscat flavors. Reprinted with permission from ref 113. Copyright 2007 Elsevier Science.

those methods have been used to analyze the vapor with the fixed concentration profile such as the step change. Spectrum analysis technique can be used to enhance the robustness against the dynamic plume behavior.113 The irregular change of the gas concentration occurred at the plume, whereas the concentration is stable in the closed system. The irregular changes of the gas concentrations of apple and Muscat flavors are shown in parts a and b of Figure 41. PCA diagram (not shown here) reveals that there was no separation between apple and Muscat flavors when only the magnitudes of the sensor responses were used in the open field with the plume. Although the time constant of apple flavor is different from the Muscat one, it is impossible to extract the time constant under the irregular change of the concentration in the time domain. Thus, the Fourier analysis method was applied. Since the concentration changes irregularly and rapidly, the spectrum of the gas concentration approaches white noise. In this situation, a gas sensor works as a LPF (low-pass filter). Thus,

(

w(n) ) 0.5 1 - cos

, (N2πn - 1))

0 e n e N - 1 (27)

where N is the window width in the discrete form. Several frequency components were used as elements of a vector for training and estimation. However, the discrimination became unstable when too many frequency components were used. Thus, the window width in ref 113 was just 4 s including 32 measurement points. The window moves every 1/8 s. Furthermore, the variable selection based upon the discrimination analysis with Wilks’ lambda114,115 was performed to find the optimal frequency components. Only 2 variables among 64 were selected (4 sensors × 32 points/ 2). This result is shown in Figure 43. There is no information loss in this diagram because the data is two-dimensional after the variable selection. It is clear that the pattern separation is considerably improved when the STFT approach is adopted. Then, LVQ (learning vector quantization) was used to classify the samples.116 The reference vectors of LVQ almost reflected on the data distribution after training, as is shown in Figure 43. As a result, >90% of the recognition probability was achieved after the improvement.

Chemical Sensing in Spatial/Temporal Domains

Figure 43. Pattern vectors obtained using STFT followed by variable selection and reference vector of LVQ after training. Sensing films are Ap-L (Apiezon L) and PEG (polyethylene glycol) 1000. Reprinted with permission from ref 113. Copyright 2007 Elsevier Science.

Chemical Reviews, 2008, Vol. 108, No. 2 697

Figure 45. Response of SAW gas sensor (FPOL) to DMMP. The response is followed through two complete 14 min sampling periods. The numbers of 2 020 and 18 350 indicates the peak heights in Hz of 2 min and 14 min PCT mode responses. Reprinted with permission from ref 117. Copyright 1993 American Chemical Society.

Figure 46. Sensor responses to binary mixture (butyl acetate and hexyl acetate) under gradual increase in temperature of preconcentrator. Reprinted with permission from ref 122. Copyright 2000 Elsevier Science.

Figure 44. Schematic diagram of the sampling system with dual preconcentrator tubes, where PCT indicates preconcentrator tubes and the circles indicate the pump. Reprinted with permission from ref 117. Copyright 1993 American Chemical Society.

3.5. Temporal Data for Preconcentrator In this section, the temporal information includes that of the sample discrimination, whereas the sensor dynamics has been described in the earlier subsection. Here, a preconcentrator with variable temperature is used. The preconcentrator is typically used to enhance the sensor sensitivity.117-120 Grate et al. proposed the system of a SAW sensor array combined with a preconcentrator. The dual preconcentrator tube system is shown in Figure 44. The upper PCT (preconcentrator tube) was operated on a 2 min cycle, while the lower one was operated on a 14 min cycle. The vapor collection during the 14 min cycle enabled more sensitive detection, whereas a longer response time was required. Sample airflow directions are indicated by the arrows, where the solid line indicates direct sampling, the dashed line indicates delivery of the preconcentrated sample from the upper PCT to the sensor array, and the dasheddotted line shows delivery of preconcentrated sample from the lower PCT. The preconcentrator tube consisted of a 1/4 in. o.d. by 1/8 in. i.d. glass tube packed with 40-60 mesh Tenax GC over approximately a 1/4 in. length of the tube.

The coil of nichrome wire wrapped around the glass tube provided the heat for thermal desorption. Figure 45 shows the sensor response to DMMP (dimethyl methylphosphonate) when both 2 min and 14 min cycles were adopted. It was found that the sensor response became larger as the vapor collection time increased. It was also found that the gradual increase in the temperature of the preconcentrator enables the higher-order sensing,121 including both distinguishable waveform and sensor-array output patterns, although the heat pulses with various temperatures were previously applied to the preconcentrator.122 One of the examples is shown in Figure 46.123 The three QCM gas sensors coated with DEGS (diethylene glycol succinate), squalane, and UCON90000 were used together with the adsorbent Tenax-TA. The sample was the binary mixture of hexyl acetate and butyl acetate. The first peak around 30 s should be ignored since the sensor responded to the vapor not accumulated at the preconcentrator. The ramp of the temperature from room temperature to 200 °C started at 60 s and stopped at 180 s. The peak occurred at the desorption temperature of the corresponding compound. It was found that two peaks corresponding to butyl acetate and hexyl acetate were observed. This separation provides the enriched information for the pattern recognition. Figure 47 shows the grayscale images of six sensor responses to apple flavors with various recipes. The waveforms of six sensors for 60 s are shown in the figure. The white portion means a large sensor response, whereas the black portion has no sensor response. The apple flavor was composed of four components: trans-2-hexenyl acetate

698 Chemical Reviews, 2008, Vol. 108, No. 2

Nakamoto and Ishida

The preconcentrator with variable temperature of the gradual ramp is effective to obtain rich information of the sample. The second-order data obtained from the preconcentrator with variable temperature in combination with the sensor array can be regarded as images. Each image of apple flavor with different composition was clearly distinguishable using the preconcentrator with variable temperature.

4. Sensing in Both Spatial and Time Domains

Figure 47. Image of six sensor responses to apple flavors with various recipes: (a) image of typical one, (b) image of enhanced green note, (c) image of enhanced smell of grass, (d) image of enhanced sour sweet, and (e) image of enhanced fruity note. Reprinted with permission from ref 123. Copyright 2005 IEEE.

In this section, we review sensor array systems that involve sensing in both spatial and time domains. As shown in section 3, there are a variety of works devoted to the analysis of chemical sensor data in time domain, since such analysis is helpful in almost all applications of chemical sensors. In any systems equipped with chemical sensors, we have to deal with transient responses even if the intention is just to wait for the sensor signals to reach their steady states. On the other hand, spatially distributed sensor arrays were developed to measure spatial distributions of chemical substances, as described in section 2. However, only a limited number of works were so far addressed to combining the sensing and signal processing both in spatial and time domains. This is partly because working in a single domain is already complicated enough. For sensing in either domain, there is no established method that you can always rely on, and therefore, there are a lot of things to be done. Nonetheless, sensing in both spatial and time domains is extremely beneficial in some applications and provides information that cannot be obtained by sensing in a single domain.

4.1. Observing Change in Spatial Chemical Distribution with Time

Figure 48. Loci of sensor responses to apple flavors with various recipes on PC1-PC2 plane obtained using PCA. Reprinted with permission from ref 123. Copyright 2005 IEEE.

(THA, softly green note), trans-2-hexenal (THL, smell of grass), isobutyric acid (IBA, sour sweet), and ethyl valerate (EVA, fruity note). In parts b-e of Figure 47, the enhanced note indicates that the portion of the corresponding compound in the mixture was twice increased. It was found that those samples with different recipes were easily discriminated using those images. Moreover, the time courses of the sensor responses (sixdimensional data for 60 s) of three samples are projected onto the space obtained from the PCA, as is shown in Figure 48. THA × 2, IBA × 2, and EVA × 2 indicates enhanced green note, enhanced sour sweet note, and enhanced fruity note, respectively. Every sample was measured three times. Since the loci of the six different sensors have different features, those might be identified using a characterrecognition technique.

The most straightforward examples in which both spatial and temporal features of chemical sensor signals are dealt with are the gas sensor arrays for measuring the spatial distribution of a chemical substance. How a chemical substance spreads in the given environment can be analyzed by observing the change in the measured spatial concentration distribution over time. When Yamasaki and Hiranaka reported their gas sensor array system, the demonstrations on measuring the growth of ethanol and ethyl ether gas clouds were presented in their paper.8,9 A sequence of gas distribution maps was obtained by measuring the gas sensor responses at 1 s interval. Although this would be the best way to measure the entire gas distribution in a given environment, the number of sensors required is the problem if a large area is to be covered with high spatial resolution. The problem will be alleviated if the sensor data can be collected through a wireless network. The sensor network technology is reviewed in another paper in this special issue.124 The olfactory video camera was developed in a complementary approach. A highly packed small sensor array was fabricated to measure the gas flow over the sensor array, as shown in section 2.3.2. The direction of the gas flow was estimated by comparing the successive snapshots of the gas concentration distribution. The position of the gas source was localized by reversely tracking the observed gas flow. To make this approach work, care should be taken about the temporal resolution. The response and recovery times of the gas sensors must be short enough to observe clouds of gas passing over the small sensor array. Otherwise, all sensors respond and recover at the same time. The maximum

Chemical Sensing in Spatial/Temporal Domains

Figure 49. (a) Spherical gas sensor array for the measurement of three-dimensional gas flow. Twenty-one gas sensors are placed on a plastic sphere of 17 cm in diameter. (b) Tethered blimp robot. The direction of the gas source is estimated from the responses of 10 gas sensors attached on the 90 cm long balloon. A wheeled tractor robot changes the elevation and the position of the balloon so that it gradually approaches the source location.

measurable speed is higher for a larger sensor array since the difference in time for a gas cloud to reach the upstream and downstream edges of the sensor array becomes larger. However, introduction of a large sensor array in an environment alters the local airflow field around the sensor array. Another problem of the packed gas sensor array is that it can measure the gas flow only when the array is placed along the flow direction. When the flow comes down vertically against the horizontally placed sensor array, for example, a complicated turbulent flow field is created around the sensor array. The sensors no longer respond in an ordered way to the gas clouds. A spherical gas sensor array shown in Figure 49a was fabricated to overcome this problem.125 The spherical shape was chosen because of the symmetry in every direction. As an extension of the same technique, a blimp robot having a spheroidal sensor array was later developed (Figure 49b).126 For those sensor arrays, however, a limited number of sensors were placed rather sparsely because of the difficulty in fabricating a large sensor array using the commercially available gas sensors. The gas flow cannot be measured when gas clouds flowing along the surface of the sensor array are smaller than the spacing between the gas sensors. When the microsensor technology advances to a point where fabrication of a dense array of fast gas sensors is enabled, the spherical sensor array will become a useful tool for locating gas sources by tracking gas plumes threedimensionally. Spatially distributed chemical sensor arrays can be used for applications other than localizing chemical sources. Sawada et al. proposed to use gas sensor units distributed in a house to monitor the activity of a resident.127 Each sensor unit consisted of four different semiconductor gas sensors. The purpose of the monitoring system is to dispatch a medical team to the house when something happens to the resident. If an elderly person living alone in a house becomes seriously sick, he/she may not be able to ask for help by him/herself when the symptom has manifested. Various gaseous chemical components are generated in our daily activities, e.g., cooking, and in our metabolism. No activity in the gas sensor signal means that there is no activity of the resident. Another interesting application of gas sensor arrays is to place a sensor array in a plastic model of a canine nasal cavity.30 The interiors of vertebrate nasal cavities, in which the olfactory receptor cells are distributed, generally have complicated structures. As the inhaled air goes thorough the cavity, gaseous components are separated since the cavity acts in a similar way to a gas chromatography column. Moreover, the complicated flow paths in the nasal cavity cause uneven distribution of odorants to the olfactory receptor cells. Odor molecules with different sizes and diffusion rates are delivered differently to the receptor cells at different locations. Stitzel et al. fabricated a nasal cavity model in

Chemical Reviews, 2008, Vol. 108, No. 2 699

Figure 50. Intermittent and spiky signal from a stationary chemical sensor placed in a plume. Burst length is defined as the time between the leading and trailing edges of a burst of concentration spikes. Burst return period is defined as the time between the leading edges of two successive bursts. Peaks much higher than the mean concentration are often observed even at far downstream locations because of the sporadic and spiky nature of the chemical signal.

which the complicated structure of a canine nasal cavity was precisely replicated based on the geometric data obtained using computed tomography scans of a real nasal cavity. Five fiber-optic vapor sensors were placed in the double-sized model. Although the same types of sensors were used, the time courses of the sensor responses were different when an odor pulse was supplied to the nasal cavity model. A unique spatial and temporal response pattern was obtained for a specific odorant, even though a single sensor type was used. The odor-discrimination capability of electronic noses can be, therefore, improved by placing sensors in the nasal cavity.

4.2. Correlating Signal Features in Time Domain with Spatial Locations The sensor arrays presented in the previous section were designed to measure the spatial features of the chemical signals and to observe their temporal change. There are works that pursue the opposite approach, i.e., to measure the temporal features of the chemical signal and correlate them to the spatial locations with respect to the chemical sources.128-133 A chemical plume has a patchy structure since eddies contained in the turbulent flow stretch and twist the streaks of the chemical substance.134 A series of patches traveling over a stationary chemical sensor is represented as a spiky fluctuating signal in the time domain (Figure 50). The fine structure of the plume is the result of diffusion and turbulent mixing acted on the patches of a chemical substance released from the source. Temporal fluctuations of the signal thus convey some information about how the patches have been transported from their source to the location of the sensor. If such information can be decoded from the sensor signal, it can be exploited to estimate the location of the chemical source from a remote place. Although there is no clear evidence, there is a possibility that animals are using the information encoded in the fluctuating signals when they are in pursuit of smells. The signal fluctuation contains a wide range of frequency components. One of the distinct characteristics of turbulent flow is that it contains a number of eddies of a variety of lengths scales.134 The size of the largest eddies is determined by the geometrical dimension of the flow, which can reach hundreds of meters for large-scale plumes in the fields. The kinematic energy of such large eddies is cascaded to successively smaller eddies to the point at which the eddies get so small that they are damped by the viscosity. The size

700 Chemical Reviews, 2008, Vol. 108, No. 2

of the smallest eddies is, thus, represented by the Kolmogorov length scale, η,

η ≈ (ν3/)1/4 where ν is the kinematic viscosity of the fluid and  is the rate of energy dissipation per unit mass. The Kolmogorov length scale is typically ∼1 cm for the atmosphere128 and was 0.7 mm134 for the open-channel water flow with the depth of 20 cm and the mean velocity of 5 cm/s. Eddies larger than the plume width make the plume meander. On the other hand, eddies smaller than the plume width stir the edges of the plume with surrounding clean fluid medium and, thus, contribute to form the fine internal structure of the plume. Those structural features are observed as signal fluctuations with long and short periods, respectively. Therefore, in order to fully investigate the structure of the plume, a long record of chemical concentration needs to be measured at a high sampling rate. The time scale of the lowest-frequency component is typically several minutes for most of the flow regimes of interest. The time scale for the highest-frequency component can be estimated by calculating the time for an eddy with the size of the Kolmogorov length scale to pass over the sensor. Typical values are 1 ms for the atmosphere128 and 0.1 s for the water flow mentioned above.134 The discussion on the length and time scales so far was made solely on the velocity field. In reality, however, the concentration in a single eddy is not homogeneous. The Batchelor scale, LB, represents the length scale of the smallest concentration patch134 and is defined as

LB ≈ (νD2/)1/4 where D is the molecular diffusion coefficient. The structure of the concentration distribution smaller than the Batchelor scale is immediately faded out by molecular diffusion. The Batchelor length scale is generally much smaller than the corresponding Kolmogorov scale since the molecular diffusion is a slow process. For the above-mentioned open-channel water flow, the Batchelor scale was only 0.02 mm.134 The currently available chemical sensors are not fast enough to resolve the fine-scale structure of chemical plumes. Therefore, for analyzing the plume structure, passive tracers that can be easily detected with high-speed sensors were used instead of real chemical substances. The basic nature of turbulent plumes is the same in airflow and in water flow. A laser-induced fluorescent technique was often used to observe the underwater chemical plumes.130-132 In this technique, an aqueous solution of a fluorescent dye is released in the flow as a tracer, and the laser light sheet is shed to illuminate a cross section of the plume. The density of the dye solution is adjusted to be equal to the background water by adding the appropriate amount of another inert and lighter chemical substance like ethanol. Two-dimensional concentration distribution can be measured by recording the image of the induced florescent light, since its intensity is proportional to the local concentration of the dye. In some works, dopamine was used in conjunction with a high-speed electrochemical sensor, although this is a technique for point measurement.129 The optical visualization technique described in section 2 has been applied so far to the measurement of aerial plumes up to 30 frames/s, although the Kolmogorov time scale of typical aerial plumes is ∼1 ms. Extremely strong illumination is required for video recording with higher speed. Moreover, particles that are large enough to

Nakamoto and Ishida

create bright images are no longer passive because of the significant mismatch of the density between the air and the tracer particles. Therefore, ionized air was used as a tracer for high-speed quantitative measurement.128 Although the time resolution in the order of 1 ms can be achieved using this techniques, it provides the time record of chemical concentration only at a single point. The data on chemical distribution is not available. A high-speed photoionization detector was also used for measuring aerial plumes.133 Since a turbulent plume has a patchy filamentous structure, bursts of concentration spikes are observed when a stationary sensor is placed in the plume (Figure 50). Also, the plume meanders as a whole. Therefore, a single burst starts when the plume comes to the location of the sensor. The burst stops when the plume moves away from the sensor. The result is a series of bursts with periods of no signal between the bursts. Research efforts were made to investigate which features of the fluctuating signals can be used to track chemical plume as animals do. The gradient of instantaneous concentration is chaotic since the plume has a patchy structure. The gradient of time-averaged concentration can be used to track the plume. However, the problem is that the gradient is small, especially in the direction parallel to the flow. Moreover, it takes at least several minutes for the mean of the measured instantaneous concentration to converge to a statistically sound value. Other features of the time-series chemical signals investigated in the literature include burst length,128,133 burst return period,128,133 signal intermittency,128,130,133 peak-to-mean ratio,128,133 and rising slope of the concentration spikes.129,133 However, what features animals use for plume tracking and what feature is the most reliable one for human or robotic searchers are still open questions. How the values of these features change with the sensor location depends not only on the flow characteristics, e.g., the turbulent intensity, but also on various parameters of the experimental setup, e.g., the size of the chemical source and the detection limit of the sensors. Therefore, contradicting results were often reported for different experimental setups. The general conclusions are as follows, although it is difficult to summarize the work done so far for the abovementioned reason. As a chemical plume extends downstream from the source location, the width of the plume itself and that of the plume meandering both increase. This results in the increase in burst length and burst return period at locations farther away from the source, although the increase is not always significant.128 If a pair of sensors is placed across the flow direction, cross-correlation of the signals from the two sensors increases with the distance from the source because of the expansion of the plume width.132 If a chemical substance is continuously released from the source, the signal is also continuous when the sensor is close to the source. As the plume becomes patchy and meandering as it travels, the signals at far locations generally become more intermittent.133 However, as the patches of a plume are carried by the flow, their edges are mixed with the surrounding fluid medium by the small eddies contained in the turbulent flow. Molecular diffusion also makes those patches grow. These effects make the signal less intermittent. Therefore, in some case, the signal intermittency decreases with the distance from the source.133 The response time of the chemoreceptors of the animals is in the order of 0.1 s and is not sufficient to fully resolve the fine structure of chemical plumes.129,133 However, there

Chemical Sensing in Spatial/Temporal Domains

Chemical Reviews, 2008, Vol. 108, No. 2 701

is a possibility that the response characteristics of the chemoreceptors are serving as temporal filters to enhance the reception of specific features of the signals.129 For example, the chemoreceptors show adaptation to sustained stimuli and respond more significantly to changing stimuli. Even if the mean concentration stays the same, the chemoreceptors respond in different ways to chemical stimuli with different intermittencies or different rising slopes of the peaks.

4.3. Frequency Analysis of the Chemical Signals in Plumes As described in the previous section, streaks of a chemical substance released from the source are stretched and twisted by the eddies contained in the turbulent flow as the chemical streaks are carried downstream. The fine structure of the plume is thus created, and the fluctuating signal is obtained from a stationary chemical sensor. There is a possibility that the fluctuating signal contains some information about how the streaks are transported from the source to the location of the sensor. If the flow at some point between the chemical source and the sensor has distinctive characteristics, a particular structure is formed in the plume. The time course of the sensor signal then comes to have a corresponding signature. Detection of such a signature will be quite useful in the search for chemical sources. By analyzing the sensor signal, we might be able to tell on which route the chemical substance was transported. One of the signatures that might be found in the plumes is periodic modulation of the chemical concentration. When a blunt object like a cylinder is immersed in a flow, it is known to periodically generate vortices with alternate rotation in its wake. The vortices are shed into the flow and form two staggered rows known as a Ka´rma´n vortex street.135 In the case of a cylinder, the shedding frequency, fs, is represented as

fs )

SU d

(28)

where U is the flow velocity and d is the diameter of the cylinder. S is a nondimensional parameter called the Strouhal number, which is known to be constant (0.21) for a wide range of Reynolds numbers. When a chemical substance is released in the Ka´rma´n vortex street, an oscillatory plume is created. The introduction of the plume oscillation can be regarded as frequency modulation from the signal-processing perspective. A female moth releases a sexual pheromone, and a male moth tracks a plume of sexual pheromone to find a mate. When a female is perching on a branch of a tree, an oscillatory plume might be generated due to the Ka´rma´n vortex street in the wake of the branch or the trunk of the tree. A chemical plume released from a barrel may have a similar oscillating structure. Mafra-Neto and Carde´ investigated the difference in the moth’s behavior in a continuous plume and in an oscillatory plume.136 A 3 × 3 cm plastic deflector was placed 4 cm downstream from a pheromone-impregnated filter paper to generate an oscillatory plume. A continuous plume was generated in a similar way but without the plastic deflector. It was found that male moths, Cadra cautella, take straighter paths to the pheromone source in the oscillatory plume than in the continuous plume. In order to track a pheromone plume, a male moth surges in the upwind when in contact

with a plume. When the contact is lost, the male starts zigzagging across the wind to find the lost plume. The results of the behavioral observation suggest that a fluctuating signal is required for sustaining the upwind progress toward the source. It should be noted that there is no direct evidence showing that moths detect the periodicity of the chemical signal. There is a possibility that the behavioral change was evoked in response to other properties of the chemical signal, e.g., the intermittency, since such properties also changed when the oscillation was introduced. Justus et al. generated a similar oscillating plume by placing a circular disk immediately downstream from the source location.133 A 3 m long and 1 m wide wind tunnel was prepared, and the wind speed was set to 50 cm/s. In their wind-tunnel setup, 1000 ppm of propene was released from the tip of a pipet, and a disk of 3.5 cm in diameter was placed perpendicular to the flow at 2.5 cm downstream from the pipet. The concentration of the tracer was recorded at various locations in the wind tunnel using a fast-response miniature photoionization detector with a sampling rate of 330 Hz. The frequency analysis was performed on the signals recorded in the oscillatory and continuous plumes. The power spectral density plot showed that the chemical signals in both plumes have a widely distributed spectrum due to the variety in sizes of the eddies contained in the turbulent flow. However, noticeable peaks were found in the power spectral density plot of the signal recorded immediately downstream (100 mm) from the source of the oscillatory plume. The frequencies of the peaks match the rate of the Ka´rma´n vortex generation. Those frequency components decayed rapidly over the distance. At 400 mm from the source, the peaks were almost buried in the background spectrum. Kikas and co-workers proposed the use of an array of chemical sensors to detect frequency modulation introduced into chemical signals.137-139 Signal fluctuation caused by the background turbulence contains a wide range of frequency components. The idea was to use correlation analysis to detect the small additional frequency component induced by the Ka´rma´n vortex street. When an array of sensors is placed in the frequency-modulated plume, the fluctuations of the sensor signals caused by the modulation should be correlated with each other. On the other hand, the fluctuations caused by small eddies in the background turbulent flow are uncorrelated. To find a correlated frequency component, the coherence spectrum140 was calculated as

γAB(f) )

|PAB(f)|2 PA(f)PB(f)

where γAB(f) denotes the coherence between the signals from sensor A and B at frequency f. PA(f) and PB(f) represent the power spectral densities of the signals from sensor A and B, respectively. PAB(f) is the cross-power spectral density between the two signals. Coherence is an equivalent of a correlation coefficient in frequency domain. For a completely correlated signal, coherence has a value of 1. For a completely noncorrelated signals, the coherence becomes zero. The idea was first tested using a benchtop apparatus called the “virtual plume”, which is the combination of a simplified model of chemical transport in flow and real chemical sensors.137-139 A chemical marker was released into water flow as a series of concentration pulses and was delivered to the electrochemical amperometric sensors through tubes

702 Chemical Reviews, 2008, Vol. 108, No. 2

Nakamoto and Ishida

Figure 51. Grayscale calibrated instantaneous images of (a) unmodulated plume and (b) modulated plume. The size of the original images before cropping was 1018 × 1008 pixels for 1 m × 1 m field of view.

of different lengths. The sensors were shown to have sufficiently fast responses, and the 1 Hz pulsation could be detected in the coherence spectrum after the pulses were delivered through a 1 m long tube with a diameter of 0.5 mm. Later, the same coherence analysis was applied to the data of concentration fields measured in real chemical plumes using the laser-induced fluorescent technique. A small amount of fluorescent dye, Rhodamine 6G, was released in a water flow established in a 1.07 m wide, 24.4 m long tilting flume with rectangular cross section and smooth bed. The average velocity in the flume was 5.0 cm/s, and the flow depth was 20.0 cm. Sweeping an argon-ion laser beam in a plane parallel to the bed with a scanning mirror created the illumination sheet. The laser light caused the dye to fluoresce, and a digital CCD camera (8-bit grayscale, with 1018 vertical and 1008 horizontal pixels) captured the emitted light. The light intensity emitted by the dye is directly proportional to the dye concentration and laser intensity. However, the obtained raw images suffer from laser sheet nonuniformity, lens vignette, and pixel variability.141 Therefore, an in situ calibration was performed to convert the raw images into quantitative data of concentration field. For the coherence analysis, 6000 images were captured with 10 frames/s. The field of view was 1 m × 1 m, and therefore, the spatial resolution was roughly 1 mm. The laser sheet was in the same horizontal plane as the plume source, 2.54 cm above the floor. The frequency modulation was performed by placing a circular cylinder of 0.8 cm diameter at 2.54 cm downstream of the chemical source. Figure 51 shows the snapshots of unmodulated and modulated plumes. The periodic meanderings of the plume can be recognized for the modulated plume near the source. Figure 52a shows the power spectral density plot for the concentration on the centerline of the modulated plume at 5 cm downstream from the source. A small peak is recognized at 1.0 Hz, which roughly coincides with the frequency of the Ka´rma´n vortex generation. Figure 52b shows the coherence spectrum between the concentration on the centerline and at 1 cm to the side at 5 cm downstream from the source. Since the frequency modulation was generated by the organized lateral meanderings of the plume, it appears in the coherence spectrum as a correlated signal component at two laterally separated locations. The peak at the modulation frequency manifested itself clearly on the zero background. The peak decayed rapidly when the point of observation was moved downstream and disappeared at ∼10 cm from the source. Therefore, when a dominant single

Figure 52. (a) Power spectral density of the time record of concentration at 5 cm downstream from the source in the modulated plume. (b) Coherence spectrum at the same location. The time record of concentration was taken from the same location as in (a) and the location 1 cm to the side.

peak is found in the coherence between sensors aligned across the flow, it means that the chemical source generating the Ka´rma´n vortex street is in close proximity. If the flow velocity is known, the size of the object can be calculated from the peak frequency. A robot with a visual sensor can start looking around for an object of that specific size. Although the frequency analysis provides us with useful information, its drawback is the need for long data to calculate accurate power spectra from random data. Moreover, most of the currently available chemical sensors are too slow to resolve the concentration fluctuations caused by the Ka´rma´n vortices. Development of more sophisticated signalprocessing algorithms and high-speed chemical sensors is required to implement this technique in real applications.

5. Conclusion In this paper, we described the aspect of chemical sensing in spatial and time domains and then reviewed the sensing related to both domains. Although sensing technology for chemical signals is not matured in comparison with that for physical signals, that technology is gradually proceeding. In the study of spatial domain, the gas distribution can be measured using a homogeneous sensor array. Two types of sensor arrays, such as sparse and packed sensor arrays, are available. The sparse sensor array can show the global behavior of the plume, whereas the packed one reveals the local detailed behavior of the plume. The optical method is also useful to obtain the plume image. An attempt to make the plume generated in a virtual environment, where people perceive sensory stimuli even if they do not stay in the actual environment, is also introduced. Next, a signal in time domain is described. Since the temporal information sometimes includes useful information for discriminating among the vapors, the technique to know the sensor dynamics such as time constant is studied. Frequency analysis is helpful when the useful information is hidden in the temporal data changing irregularly due to the turbulence.

Chemical Sensing in Spatial/Temporal Domains

Then, the sensing in both spatial and time domains is described. The straightforward method to understand the combination of both domains is to observe change in spatial distribution with time. Another approach is to see the correlation of signal features in time domain with several locations. The frequency analysis of the signals also provides us with useful information about an odor-source location. It is an important task for us to fully understand the plume behavior in both spatial and time domains and to establish the measurement method of capturing its behavior. Moreover, a sensor dynamics model is required because a sensor response does not follow the speed of the plume change. A systematic approach including algorithms will become more important as well as the improvement of chemical-sensor capability itself. The current technology is not sufficient to find the toxic or explosive substance immediately. However, the appropriate combination of sensors with signal-processing techniques will make this a field in progress.

6. References (1) Nakamoto, T.; Ishida, H.; Moriizumi, T. Anal. Chem. 1999, 71 (15) 531A. (2) Hinze, J. O.; Turbulence; McGraw-Hill: New York, 1975. (3) Sutton, O. G. Micrometeorology; McGraw-Hill: New York, 1953. (4) Pasquill, E.; Smith, F. B. Atmospheric Diffusion, 3rd ed.; Ellis Horwood: Chichester, U.K., 1983. (5) Pal Arya, S. Air Pollution Meteorology and Dispersion; Oxford University Press: Oxford, U.K., 1999. (6) Yamanaka, T.; Ishida, H.; Nakamoto, T.; Moriizumi, T. Sens. Actuators, A 1998, 69, 77. (7) Tsujita, W.; Nakamoto, T.; Ishida, H.; Moriizumi, T. Trans. IEICE Jpn. 2002, J85-C, 269. (8) Hiranaka, Y.; Yamasaki, H. IEE Jpn., Sensor Symp. 1989, 177. (9) Yamasaki, H.; Hiranaka, Y. Sens. Actuators, A 1992, 35, 1. (10) Ishida, H.; Suetsugu, K.; Nakamoto, T.; Moriizumi, T. Sens. Actuators, A 1994, 45, 153. (11) Ishida, H.; Kagawa, Y.; Nakamoto, T.; Moriizumi, T. Sens. Actuators, B 1996, 33, 115. (12) Ishida, H.; Kushida, N.; Yamanaka, T.; Nakamoto, T.; Moriizumi, T. Trans. IEE Jpn. 1999, 119-E, 194 (in Japanese). (13) Ishida, H.; Yamanaka, T.; Kushida, N.; Nakamoto, T.; Moriizumi, T. Sens. Actuators, B 2000, 65, 14. (14) Sharaf, M. A.; Illman D. L.; Kowalski, B. R. Chemometrics 1986, 54. (15) Sauerbrey, G. Z. Phys. 1959, 155, 289. (16) King, W. H. Anal. Chem. 1964, 36, 1735. (17) Hlavay, J.; Guilbault, G. G. Anal. Chem. 1977, 49, 1890. (18) Kurosawa, S.; Kamo, N.; Matsui, D.; Kobatake, Y. Anal. Chem. 1990, 62, 353. (19) Nakamoto, T.; Moriizumi, T. Jpn. J. Appl. Phys. 1990, 29, 963. (20) Nakamoto, T.; Tokuhiro, T.; Ishida, H.; Moriizumi, T. Tech. Dig. Transducers’99 1999, 1878. (21) Segawa, N.; Tokuhiro, T.; Nakamoto T.; Moriizumi, T. T. IEE Jpn. 2002, 122-E, 16 (in Japanese). (22) Ishida, H.; Tokuhiro, T.; Nakamoto, T.; Moriizumi, T. Sens. Actuators, B 2002, 83, 256. (23) Russell, R. A.; Thiel, D.; Deveza, R.; Mackay-Sim, A. Proc. IEEE Int. Conf. Rob. Autom. 1995, 556. (24) http://w-soramame.nies.go.jp. (25) http://www.kankyo.metro.tokyo.jp. (26) Maruo, Y. Y.; Ogawa, S.; Ichino, T.; Murao, N.; Uchiyama, M. Atmos. EnViron. 2003, 37, 1065. (27) Ohyama, T.; Maruo, Y. Y.; Tanaka, T.; Hayashi, T. Sens. Actuators, B 2000, 64, 142. (28) Tsujita, W.; Yoshino, A.; Ishida, H.; Morrizumi, T. Sens. Actuators, B 2005, 110, 304. (29) Settles, G. S. J. Fluid Dyn. 2005, 127, 189. (30) Stitzel, S. E.; Stein, D. R.; Walt, D. R. J. Am. Chem. Soc. 2003, 125, 3684. (31) Yokosawa, K.; Nakano, S.; Goto, Y.; Tsukada, K. Proc. 22nd Sensor Symp., IEE Jpn. 2005, 435. (32) Lundstrom, I.; Shivaraman, S.; Svensson, C.; Lindkvist, L. Appl. Phys. Lett. 1974, 26, 55. (33) Bather, W. Sens. Update 1998, 4, 82. (34) EnVironmental Analysis Technology Handbook, 5th ed.; GASTEC: Kanagawa, Japan, 2004.

Chemical Reviews, 2008, Vol. 108, No. 2 703 (35) Tanaka, Y.; Yoshioka, M.; Nakamoto, T.; Moriizumi, T. Trans. SM, IEE Jpn. 2004, 124, 321 (in Japanese). (36) Tanaka, Y.; Nakamoto, T.; Moriizumi, T. Sens. Actuators, B 2006, 119, 84. (37) Ninh, H. P.; Tanaka, Y.; Nakamoto, T.; Hamada, K. Sens. Actuators, B 2007, 125, 138. (38) Nakamoto, T.; Tanaka, Y.; Ninh, H. P. Trans. IEE Jpn. 2007, 127, 359. (39) Yamanaka, T.; Matsumoto, R.; Nakamoto, T. Sens. Actuators, B 2002, 87, 457. (40) Kulp, T. J.; Garvis, D.; Kennedy, R.; McRae, T. G. Proc. SPIE 1991, 1479, 352. (41) Kulp, T. J.; Powers, P. E.; Kennedy, R. Proc. SPIE 1997, 3061, 269. (42) Powers, P. E.; Kulp, T. J.; Kennedy, R. Appl. Opt. 2000, 39, 1440. (43) Nagashima, T.; Yamashita, K.; Hatta, S.; Kajihara, R.; Hamakawa, Y.; Okuyama, M. Tech. Dig. Sens. Symp., IEEJ 2001, 109. (44) Ochiai, M.; Kuroki, M. Presented at Technical meeting on chemical sensor, IEE of Japan, 1998; Tokyo, CS-98-47. (45) Yamada, T.; Yokoyama, S.; Tanikawa, T.; Hirota, K.; Hirose, M. Proc. IEEE Virtual Reality 2006, 199. (46) Nakamoto T.; Pham, H. D. M. Proc. IEEE Virtual Reality 2007, 179. (47) Gardner, J. W. Sens. Actuators, B 1990, 1, 166. (48) Nakamoto, T.; Iguchi, A.; Moriizumi, T. Sens. Actuators, B 2000, 71, 155. (49) Tobias, P.; Baranzahi, A.; Spetz, A. L.; Kordina, O.; Janzen, E.; Lundstrom, I. IEEE Electron DeVice Lett. 1997, 18, 287. (50) Murlis, J.; Elkinton, J. S.; Carde, R. T. Annu. ReV. Entomol. 1992, 37, 505. (51) Llobet, E., Pearce, T. C., Schiffman, S. S., Nagle, H. T., Gardner, J. W., Eds.; Handbook of machine olfaction; Wiley-VCH: New York, 2003; p 293. (52) Hines, E. L.; Llobet, E.; Gardner, J. W. IEE Proc. Circuits DeVices Syst. 1999, 146 (6), 297. (53) Tsujita, W.; Nakamoto, T.; Ishida, H.; Moriizumi, T. Trans. IEICE 2002, J85-C, 269 (in Japanese). (54) Rumelhart, D. E.; McClelland, J. L. PDP Research Group Parallel Distributed Processing; MIT Press: Cambridge, MA, 1986; Vol. 1, p. 318. (55) Rumelhart, D. E.; Hinton, G. E.; Williams, R. J. Nature 1986, 323, 533. (56) Duda, R. O.; Hart P. E.; Stork, D. G. Pattern classification; WileyInterscience: New York, 2001; p 287. (57) Shumer, H. V.; Gardner, J. W. Sens. Actuators, B 1992, 8, 1. (58) Weimar, U.; Schierbaum, K. D.; Goepel, W. Sens. Actuators, B 1990, 1, 93. (59) Ionescu, R.; Llobet, E. Sens. Actuators, B 2002, 81, 289. (60) Wilson, D. M.; Dunman, K.; Roppel, T.; Kalim, R. Sens. Actuators, B 2000, 62, 199. (61) Afridi, M. Y.; Suehle, J. S.; Zaghloul, M. E.; Berning, D. W.; Hefner, A. R.; Cavicchi, R. E.; Semancik, S.; Montgomery, C. B.; Taylor, C. J. IEEE Sens. J. 2002, 2, 644. (62) Arnold, C.; Harms, M.; Goschnick, J. IEEE Sens. J. 2002, 2, 179. (63) Ema, K.; Yokoyama, M.; Nakamoto, T.; Moriizumi, T. Sens. Actuators 1989, 13, 476. (64) Davide, F.; Natale, C. C.; D’Amico, A.; Hierlemann, A.; Mitrovics, J.; Schweizer, M.; Weimar, U.; Goepel, W.; Marco, S.; Pardo, A. Sens. Actuators, B 1995, 26-27, 275. (65) Yano, K.; Yoshitake, H.; Bornscheuer, U. T.; Schmid, R. D.; Ikebukuro, K.; Yokoyama, K.; Matsuda, Y.; Karube, I. Anal. Chim. Acta 1997, 340, 41. (66) Nakamura, M.; Sugimoto, I.; Kuwano, H.; Lemos, R. Sens. Actuators, B 1994, 20, 231. (67) Carey, W. P.; Beebe, K. R.; Kowalski, B. R.; Illman, D. L.; Hirschfeld, T. Anal. Chem. 1986, 58, 149. (68) Nakamoto, T.; Fukunishi, K.; Moriizumi, T. Sens. Actuators 1989, 13, 473. (69) Nakamura, K.; Nakamoto, T.; Moriizumi, T. Sens. Actuators, B 1999, 61, 6. (70) Nakamura, K.; Nakamoto, T.; Moriizumi, T. Sens. Actuators, B 2000, 69, 295. (71) Nanto, H.; Subakino, S.; Habara, M.; Kondo, K.; Morita, T.; Douguchi, Y.; Nakazumi, H.; Waite, R. I. Sens. Actuators, B 1996, 34, 312. (72) Muramatsu, H.; Tamiya, E.; Karube, I. Anal. Chim. Acta 1989, 225, 399. (73) Ballantine, D. S.; Rose, S. L.; Grate, J. W.; Wohltjen, H. Anal. Chem. 1986, 58, 3058. (74) Grate, J. W.; Wise, B. M.; Abraham, M. H. Anal. Chem. 1999, 71, 4544. (75) Zellers, E. T.; Han, M. Anal. Chem. 1996, 68, 2409. (76) Stahl, U.; Rapp, M.; Wessa, T. Anal. Chim. Acta 2001, 450, 27.

704 Chemical Reviews, 2008, Vol. 108, No. 2 (77) Bender, F.; Barie´, N.; Romoudis, G.; Voigt, A.; Rapp, M. Sens. Actuators, B 2003, 93, 135. (78) Hagleltner, B.; Hierlemann, A.; Lange, D.; Kummer, A.; Baltes, H. Nature 2001, 414, 293. (79) Battiston, F. M.; Ramseyer, J.-P.; Lang, H. P.; Baller, M. K.; Gerber, C. H; Gimzewski, J. K.; Meyer, E.; Gu¨ntherodt, H.-J. Sens. Actuators, B 2001, 77, 122. (80) Cai, Q.; Park, J.; Heldsinger, D.; Hsieh, M.; Zellers, E. T. Sens. Actuators, B 2000, 62, 121. (81) Hatfield, J. V.; Neaves, P.; Hicks, P. J.; Persaud, K.; Travers, P. Sens. Actuators 1994, 18-19, 221. (82) Gardner, J. W.; Pearce, T. C.; Friel, S.; Bartlett, P. N.; Blair, N. Sens. Actuators, B 1994, 18-19, 240. (83) Hassan, M. E.; Amrani, Dowdeswell, R. M.; Payne, P. A.; Persaud, K. C. Sens. Actuators, B 1997, 44, 512. (84) Slater, J. M.; Paynter, J.; Watt, E. J. Analyst 1993, 118, 379. (85) Doleman, A. J.; Lewis, N. S. Sens. Actuators, B 2001, 72, 41. (86) Ryan, M. A.; Zhou, H.; Buehler, M. G.; Manatt, K. S.; Mowrey, V. S.; Jackson, S. P.; Kisor, A. K.; Shevade, A. V.; Homer, M. L. IEEE Sens. J. 2004, 4, 337. (87) Hayes, A. T.; Martinoli, A.; Goodman, R. M. IEEE Sens. J. 2002, 2, 260. (88) Sundgren, H.; Lundstrom, I.; Winquist, F.; Lukkari, I.; Carlsson, R.; Wold, S. Sens. Actuators, B 1990, 2, 115. (89) Covington, J. A.; Gardner, J. W.; Briand, D.; de Rooij, N. F. Sens. Actuators, B 2001, 77, 155. (90) Davide, F.; Andersson, M.; Holmberg, M.; Lundstrom, I. IEEE Sens. J. 2002, 2, 636. (91) Pillonel, L.; Bosset, J. O.; Tabacchi, R. Eur. Food Res. Technol. 2000, 214, 160. (92) Garrigues, S.; Talou, T.; Nesa, D.; Gaset, A. Sens. Actuators, B 2001, 78, 337. (93) Morvan, M.; Talou, T.; Beziau, J.-F. Sens. Actuators, B 2003, 95, 212. (94) Bindig, U.; Katzung, W.; Mauch, H.; Leonhardt, J.; Muller, G. Tech. Dig. 9th Int. Symp. Olfaction Electronic Nose; 2002, 106. (95) Staples, E. J. Proc. IEEE Ultrason. Symp. 1999, 417. (96) White, J.; Kauer, J. S.; Dickinson, T. A.; Walt, D. R. Anal. Chem. 1996, 68, 2191. (97) Albert, K. J.; Walt, D. R. Anal. Chem. 2003, 75, 4161. (98) Rakow, N. A.; Suslick, K. S. Nature 2000, 406, 710. (99) Natale, C. D.; Salimbeni, D.; Paolesse, R.; Macagnano, A.; D’Amico, A. Sens. Actuators, B 2000, 65, 220. (100) Stetter, J. R.; Jurs, P. C.; Rose, S. L. Anal. Chem. 1986, 58, 860. (101) Nakamura, M.; Sugimoto I.; Kuwano, H. IEEJ Chem. Sens. Meet. 1997 CS-97-37, 21 (in Japanese). (102) Nakamura, M.; Sugimoto, I.; Kuwano H.; Lemos, R. Dig. Tech. Pap., Transducers’93 1993, 434. (103) Nakamura, M.; Sugimoto, I.; Kuwano, H. Dig. Tech. Pap. Transducers’95 1995, 795. (104) Skogestad S.; Postlethwaite, I. MultiVariable Feedback Control; Wiley: New York, 1996; p 1. (105) Marco, S.; Pardo, A.; Davide, F. A. M.; Natale, C. D.; D’Amico, A. Sens. Actuators, B 1996, 34, 213. (106) Davide, F. A. M.; Natale, C. D.; D’Amico, A.; Hierlmann, A.; Mitrovics, J.; Schweizer, M.; Weimar, U.; Goepel, W. Sens. Actuators, B 1995, 24-25, 830. (107) Schweizer-Berberich, M.; Goeppert, J.; Hierleman, A.; Mitrovics, J.; Weimar, U.; Rosenstiel, W.; Gaepel, W. Sens. Actuators, B 1995, 26-27, 232. (108) Amrani, M. E. H.; Dowdeswell, R. M.; Payne, P. A.; Persaud, K. C. Sens. Actuators, B 1997, 44, 512.

Nakamoto and Ishida (109) Guitierrez-Osuna, R.; Troy, Nagle, H.; Schiffman, S. S. Sens. Actuators, B 1999, 61, 170. (110) Nakata, S.; Okunishi, H.; Nakashima, Y. Sens. Actuators, B 2006, 119, 556. (111) Ionescu, R.; Llobet, E. Sens. Actuators, B 2002, 81, 289. (112) Vergara, A.; Llobet, E.; Brezmes, J. ; Ivanov, P.; Cane´, C. ; Gra`cia, I. ; Vilanova, X.; Correig, X. Sens. Actuators, B 2007, 123, 1002. (113) Nimsuk, N.; Nakamoto, T. Sens. Actuators, B 2007, 127, 491. (114) Dillon, W. R.; Goldstein, M. MultiVariate Analysis; Wiley: New York, 1984; p 394. (115) Nakamoto, T.; Sasaki, S.; Fukuda, A.; Moriizumi, T. Sens. Mater. 1992, 4, 111. (116) Kohonen, T. Self-organization and associatiVe memory; SpringerVerlag: New York, 1988; p 199. (117) Grate, J. W.; Rose-Pehrsson, S. L.; Venezky, D. L.; Klusty, M.; Wohltjen, H. Anal. Chem. 1993, 65, 1868. (118) Grooves, W. A.; Zellers, E. T.; Frye, G. C. Anal. Chim. Acta 1998, 371, 131. (119) Kita, J.; Aoyama, Y.; Kinoshita, M.; Nakano, H.; Akamatsu, H. Tech. Dig. IEEJ Sens. Symp. 2000, 301. (120) Bender, F.; Barie, N.; Romoudis, G.; Voigt, A.; Rapp, M. Sens. Actuators, B 2003, 93, 135. (121) Booksh, S.; Kowalski, B. R. Anal. Chem. 1994, 66, 782. (122) Nakamoto, T.; Isaka, Y.; Ishige, T.; Moriizumi, T. Sens. Actuators, B 2000, 69, 58. (123) Nakamoto, T.; Sukegawa, K.; Sumitomo, E. IEEE Sens. J. 2005, 5, 68. (124) Diamond, D. Chem. ReV. 2008, 80. (125) Ishida, H.; Tsuruno, M.; Yoshikawa, K.; Moriizumi, T. Proc. 11th Int. Conf. AdV. Rob. 2003, 369. (126) Ishida, H.; Zhu, M.; Johansson, K.; Moriizumi, T. Conf. Proc., Int. Conf. Electr. Eng. 2004, 3, 117. (127) Sawada, A.; Oyabu, T.; Nanto, H. Trans. IEE Jpn. 2001, 121-E, 434. (128) Murlis, J.; Elkinton, J. S.; Carde´, R. T. Annu. ReV. Entomol. 1992, 37, 505. (129) Moore, P. A.; Atema, J. Biol. Bull. 1991, 181, 408. (130) Liao, Q.; Cowen, E. A. EnViron. Fluid Mech. 2002, 2, 9. (131) Crimaldi, J. P.; Koehl, M. A. R.; Koseff, J. R. EnViron. Fluid Mech. 2002, 2, 35. (132) Weissburg, M. J.; Dusenbery, D. B.; Ishida, H.; Janata, J.; Keller, T.; Roberts, P. J. W.; Webster, D. R. EnViron. Fluid Mech. 2002, 2, 65. (133) Justus, K. A.; Murlis, J.; Jones, C.; Carde´, R. T. EnViron. Fluid Mech. 2002, 2, 115. (134) Roberts, P. J. W.; Webster, D. R. In EnVironmental Fluid Mechanics: Theories and Application; Shen, H. H., Cheng A. H.-D., Wang, K.-H., Teng, M. H., Liu, C. C. K., Eds.; ASCE Press: Reston, VA, 2002. (135) White, F. M. Viscous Fluid Flow, 2nd ed.; McGraw-Hill: New York, 1991. (136) Mafra-Neto, A.; Carde´, R. T. Physiol. Entomol. 1995, 20, 117. (137) Kikas, T.; Ishida, H.; Webster, D. R.; Janata, J. Anal. Chem. 2001, 73, 3662. (138) Kikas, T.; Janata, P.; Ishida, H. Janata, J. Anal. Chem. 2001, 73, 3669. (139) Kikas, T.; Ishida, H.; Janata, J. Anal. Chem. 2002, 74, 3605. (140) Kay, S. M. Modern Spectral Estimation; Prentice Hall: Upper Saddle River, NJ, 1988. (141) Ferrier, A. J.; Funk, D. R.; Roberts, P. J. W. Dyn. Atmos. Oceans 1993, 20, 155.

CR068117E

Chem. Rev. 2008, 108, 705−725

705

Electronic Nose: Current Status and Future Trends Frank Ro¨ck, Nicolae Barsan, and Udo Weimar* Institute of Physical and Theoretical Chemistry, University of Tu¨bingen, Auf der Morgenstelle 15, 72076 Tu¨bingen, Germany Received August 23, 2007

Contents 1. Introduction 2. Technology 2.1. Classical Electronic Noses Based on Chemical Gas Sensors 2.2. New Approaches 2.2.1. Optical Sensor Systems 2.2.2. Mass Spectrometry 2.2.3. Ion Mobility Spectrometry 2.2.4. Gas Chromatography 2.2.5. Infrared Spectroscopy 2.2.6. Use of Substance-Class-Specific Sensors 2.3. Combined Technologies 3. Companies 4. Application Areas 4.1. Food and Beverage 4.2. Environmental Monitoring 4.3. Disease Diagnosis 5. Research and Development Trends 5.1. Sample Handling 5.2. Filters and Analyte Gas Separation 5.3. Data Evaluation 6. Conclusion 7. References

705 705 706 707 707 708 708 709 709 709 710 710 710 712 715 716 718 719 719 720 721 722

nonodorant volatiles, such as the detection of explosives, becomes reachable.5,6 Therefore, the term “electronic nose” may be misleading and makes the uninformed reader believe in system capabilities comparable to those of the human nose. Attempts to avoid this term and to replace it (e.g., by “application-specific sensor system”) have not taken root up to now, and in most of the current literature the term “electronic nose” is still used. In recent years much work has been done to understand the principles of odorant receptors and the organization of the olfactory system.7-9 On each olfactory receptor cell only one type of odorant receptor is located, which can detect a limited number of substances. For a complex odor, composed of multiple odorant molecules, several receptors are activated. The resulting receptor pattern determines our impression of the odor. Keeping in mind the technical limitations of the electronic nose, we should define it as what it is: an attempt to mimic the principles of smelling that gives another view on the whole scene of volatiles compared to its biological inspiration. The sensor data are analyzed to extract features which can be evaluated as a whole to eliminate redundancy and to arrive at a description of the overall mix of volatiles and their intensity. Consequently, in addition to common sensor arrays, new technologies such as flash GC (gas chromatography) or MS (mass spectrometry) devices are also often referred to as electronic noses.

1. Introduction

2. Technology

Dodd and Persaud introduced the idea of an electronic nose as a device to mimic the discrimination of the mammalian olfactory system for smells.1 They used three different metal oxide gas sensors and identified several substances by the steady-state signals of these sensors. One of the initial hopes for work in this area was to instrumentally assess attribute descriptors such as fruity, grassy, earthy, malty, etc. reliably by the results of an electronic nose measurement.2 In other words, capturing the “flavor fingerprint” 3 or “recognizing the odor”. Even if one concentrates solely on the different sensitivity characteristics of technical sensors and biological receptors, it is not surprising that despite 25 years of research this is still not possible. The comparison between an electronic nose and a human nose is in the best case like the comparison of an eye of a bee with a human one.4 It is blind for a part of the visible spectrum but sensitive for other wavelengths. For this reason only in well-defined cases the correlation between human odor impressions and electronic nose data makes sense. On the other side the evaluation of

The term “electronic nose” is often associated with the detection of odors or the attempt to “smell” with a technical device, but as already mentioned, the electronic nose is more and at the same time less, because while it offers the capability to detect some important nonodorant gases, it is not adapted to substances of daily importance in mammalian life such as the scent of other animals, foodstuff, or spoilage. Nevertheless, there are strong drivers to apply it in the field of olfaction because alternatives either are not practicable or are too costly and time-consuming, e.g., human test panels. One of the challenges of the practical application of electronic noses is that the gases of interest are part of a complex background, which may include water vapor, etc. Technical sensors may also be sensitive to these background gases, whereas, for example, humans have no receptors for water vapor; it is not relevant because it is everywhere in the ambient atmosphere. Similarly, we are not able to perceive carbon monoxide, as prior to the ability to deliberately control fire it made no evolutionary sense. This fact, namely, the relation between, on one hand, detectable and not detectable substances and, on the other hand, relevant and not relevant ones is the crucial point for every electronic

* To whom correspondence should be addressed. Phone: +49-7071-2977634. Fax: +49-7071-29-5960. E-mail: [email protected].

10.1021/cr068121q CCC: $71.00 © 2008 American Chemical Society Published on Web 01/19/2008

706 Chemical Reviews, 2008, Vol. 108, No. 2

Frank Ro¨ck got his diploma in chemistry in 2003 with a work dealing with quality assessment of food packaging materials by using gas sensor arrays at the University of Tu¨bingen. There, he is currently finishing his doctoral studies in which he is focusing on the application of chemical sensor systems to solving industrially relevant problems; his main approach is centred on the reduction of cross-interferences by employing various strategies. His contributions have been published in three papers in peerreviewed journals and presented at nine international conferences.

Ro¨ck et al.

Udo Weimar received his diploma in physics in 1989, his Ph.D. in chemistry in 1993, and his Habilitation in 2002 from the University of Tu¨bingen. He is currently the head of the Gas Sensors Group at the University of Tu¨bingen. His research interest focuses on chemical sensors as well as on multicomponent analysis and pattern recognition. He is the author of about 180 scientific papers and short notes. He is responsible for several European projects and for coordinating the Network of Excellence GOSPEL.

achieved by several approaches, which can be related to the configuration of the sensing unit of the electronic nose itself and the sample pretreatment techniques. For the existence of odorous trace components, again two cases should be considered: A limited odor measurement is possible if a correlation exists with substances which are present in higher concentrations, either odorant or odorless concomitant (background) gases (case 5). Otherwise, it is not possible to make a prediction about the odor impression of a sample because the sensitivity of the device to the responsible substances is just not high enough (case 6).

Nicolae Barsan received his diploma in physics in 1982 from the Faculty of Physics of the Bucharest University and his Ph.D. in solid-state physics in 1993 from the Institute of Atomic Physics, Bucharest, Romania. He was a senior researcher at the Institute of Physics and Technnology of Materials, Bucharest, between 1984 and 1995. Since 1995 he has been a researcher at the Institute of Physical Chemistry of the University of Tu¨bingen and actually is in charge of the developments in the field of metal oxide based gas sensors. He has published about 150 papers and contributions to international conferences.

nose application and should be explained in detail for the case of odor detection. Gaseous substances can be either odorous or odorless (Figure 1). We refer here to both true gases and liquids in their vapor phase (“volatiles”). Concerning the technical detection of odors, one has to distinguish between trace components and concentrated gases.10 In the ideal case, the high-concentration substances are responsible for the odor impression and the odorless components which are also present are negligible regarding the measurement results (case 1). Otherwise, the odorless background interferes with the measurement. We can then differentiate between three cases where interfering gases are present: If they are correlated with odorous substances, a limited odor measurement is possible as long as the relation between the concentrations is fixed (case 2). If this is not the case or the odorless gases mask the target compound, an odor measurement is excluded (case 3), unless the measurement system eliminates the effect of the interfering substances (case 4). The latter can be

Figure 1. Schematic demonstrating the possible conditions for a reliable odor measurement. For target analytes not causing the human odor impression but which are of interest for other reasons the same flow diagram is applicable. Therefore, analytical background knowledge is important for the best adaptation of the system. Reprinted with permission from ref 10. Copyright 2003 SpringerVerlag.

2.1. Classical Electronic Noses Based on Chemical Gas Sensors The classical electronic nose, consisting of an array of sensors, is still the most common approach, although new technologies have recently entered this field (Figure 2). There are two reasons for the continuing popularity of sensor arrays. As this is how the field began there is a wide body of experience gained by using them for a diverse set of applications, and the setup of a sensor-based electronic nose resembles most closely the biological model. Every part of

Electronic Nose

Chemical Reviews, 2008, Vol. 108, No. 2 707

ensure that the substances which have to be detected are causing the signal. Early attempts at electronic noses took a “black box” approach to correlating sensor outputs with measurement parameters, blindly hoping that, despite changes in the measurement conditions, the correlation remained reliable. This approach can often be found in the literature and often works well for a limited sample collection or constrictive parameters. There are applications where such approaches can provide reasonable results, but one often faces the risk of focusing on the wrong parameters, such as the age of the test persons or their cigarette consumption instead of the intended lung cancer when analyzing exhaled samples.

2.2. New Approaches Figure 2. Schematic setup of a sensor system. Via sampling, filtering, and preconditioning the analytes are led to the sensing elements. These consist of a sensitive layer and a transducer to transform the chemical information into an electrical one. After the signal is recorded, data pretreatment, and feature extraction, pattern recognition evaluates the data using the calibration data. Reprinted from ref 11. Copyright 1998 American Chemical Society.

the mammalian nose has its technical equivalent. While all of the sensing technologies require a similar approach to data evaluation, the key feature of sensor arrays is their modularity. For the detection of gaseous substances, the counterparts of biological receptors are gas sensors, which, as with biological receptors, provide a certain multiplicity of detection by not being fully selective. The information on the smell or identity of a sample can only be obtained by comparing the signals of several sensors or receptors. One of the main reasons why it has not been possible to make a one to one copy of the human nose is the high specificity of the human receptors. The technical realization is always a tightrope walk between high specificity and reversibility. High specificity demands irreversible interaction between the sensor and target gas. Even after a few million years of evolution, the human receptor cells have a lifetime of only a few weeks.12 This demonstrates the high costs of smelling in nature and the challenges faced in technological development where the lifetime of sensors needs to be much longer. The assortment of different sensor transducer principles is not to be disregarded, and for each sensor type, a variety of sensor specificity tuning possibilities is available.13 For example, for metal oxide sensors different sensitive materials are used, different doping elements are available, different production processes to reach different morphologies of the sensing layer are applied, different electrodes are utilized, different filter layers are attached, and different operating temperatures are possible. Although the metal oxide (MOX) sensor can be considered as one of the standard sensors in the field of electronic noses, the same diversity is found for other transducer principles, be it surface or bulk acoustic wave (SAW, BAW) sensors, metal oxide field effect transistors (MOSFETs), or conducting polymer (CP) sensors. It is important to note that even combining all types of available sensors there are limits to the useful dimensions of the array; instead of obtaining new information about the gaseous composition, increasing the array size amplifies the noise, e.g., by sensitivity toward unimportant information. The best method to arrange a sensor-based electronic nose is not to use as many different sensors as available but to select them with an eye on the desired application and the knowledge of the analytical data. That is the only way to

It can be shown that by using sensors with different transducer principles the gain in useful information correlated with the increase of the sensor set can be further extended.14,15 Sensors with different transducer principles will be selective for different classes of substances and can therefore often provide additional information. Hence, in recent years the original sensor types used for electronic noses were not only enhanced but complemented by other technologies introduced in this field. The range of electronic noses available today is not limited just to devices based on chemoresistors or gravimetric sensors but also includes those based on optical sensors or even systems without a modular setup such as mass spectrometers or flash gas chromatographs. Machine olfaction has benefited from scientific developments in other fields, ranging from optical technologies developed by the telecoms industry to the improvements in analytical chemistry. This trend has also narrowed the gap between the traditional electronic nose used as a black box and classical analytics which aims to quantify each single component of a given sample.

2.2.1. Optical Sensor Systems Optical sensor systems resemble most closely classical sensor-array systems because the dimension of data output can be precisely defined and adapted.16-18 Instead of having transduction principles based on electrical changes in resistance, potential, current, or frequency, the modulation of light properties is measured. In general, optical instruments are more complex but offer a variety of different measuring possibilities. The assortment of applicable technologies is high and ranges from diverse light sources over optical fibers to detectors such as photodiodes and CCD and CMOS cameras.19 Therefore, different operation modes were developed and are deployed using changes in absorbance, fluorescence, optical layer thickness, and polarization. The most direct method measures the absorbance of the analyte gas in a special frequency range. This method is applicable, for example, for carbon dioxide, but is too insensitive (within a justifiable technical effort) for other components in a lower concentration range. Therefore, in other cases, the interaction with a sensitive layer is utilized. The simplest approach is to use color-changing indicators, such as metalloporphyrins, and measure with an LED and a photodetector system their absorbance upon analyte gas exposure. Figure 3 shows how thin films of chemically responsive dyes are used as a colorimetric sensor array. Even more sensitive are the fluorescence methods; they work in a similar setup by detecting not the absorbance but the light emission at a lower wavelength. For reflectometric interfer-

708 Chemical Reviews, 2008, Vol. 108, No. 2

Ro¨ck et al.

Figure 4. Schematic diagram of an ion mobility spectrometer. Ions are generated in an ionization region by electrospray or by a 63Ni source. An ion shutter pulses the ions into the drift tube where they are accelerated by a uniform weak electric field toward a detector. Their progress is impeded by a number of collisions with the drift gas. Larger ions with greater collision cross sections experience more collisions. Therefore, the separation of ions of differing shape and size becomes possible. Reference 21s Reproduced by permission of The Royal Society of Chemistry. Figure 3. Thin films of chemically responsive dyes are used as a colorimetric sensor array. Multiple dyes change their colors depending on intermolecular interactions. By digital subtraction of each single pixel before and after exposure to the sample the difference map of the colorimetric array is obtained. The different colors are caused by the relative change in the red, green, and blue values of each dye and the brightness by its absolute change. Reprinted with permission from ref 17. Copyright 2004 Elsevier.

ence spectroscopy (RifS), the sensitive layers are similar to the polymer layers used for the gravimetric methods (QMB and SAW transducers). However, in this case the changes in the optical layer thickness and not the weight increase are taken as the sensor signal.

2.2.2. Mass Spectrometry Combined with gas chromatographs, mass spectrometers are often applied for lab analytics or as stand-alone devices for the identification of pure chemicals. After ionization of the compounds through thermionic emitted electrons (electron ionization) or through interaction with reagent ions (chemical ionization), the molecule ions and their fragment ions are separated according to their mass-to-charge ratio (m/z). This takes place with an electric and/or magnetic field, and nowadays a variety of mass analyzers are established. To mention only a few of them, the sector instrument is the classical approach with tunable static fields, whereas the quadrupole mass analyzer consists of four parallel metal rods and filters the several ions by oscillating electrical fields. Finally, the ions collide at the electron multiplier, and the current is measured. The disadvantage of all types of mass spectrometers is that their operation requires a vacuum, and therefore, they are not as convenient as the solid-state sensor arrays described previously; it also introduces additional costs. When used as electronic noses, the system is fed with the gaseous sample without previous separationsno chromatographical step. Each m/z ratio can be treated as a separate virtual sensor and analyzed by a pattern recognition algorithm.3,20 Despite its higher technical complexity, this approach is, in general, not better suited for odor detection when compared to the classical electronic noses but has advantages for defined tasks. For example, the mass spectrometer has proved its ability of detecting peptides in a higher mass range and was used for mixtures of peptide pheromones.

2.2.3. Ion Mobility Spectrometry The working principle of ion mobility spectrometry (IMS) is also the filtering of ions as in the case of mass spectrometry (Figure 4). In IMS this is more easily realized, because the aim is not to separate the target molecules exclusively by their differences in the mass/charge ratio, but also on the basis of their different mobilities. This means that, as well as their reduced mass and their charge, the different collision cross sections, determined by size and shape, has a direct influence on the separability of ions. Thereby, the collisions between the ions and the ambient air molecules is utilized, and the measurement can be performed under normal pressure.21 The most common agent for ionization is a radioactive β emitter such as 63Ni or 241Am. After a series of ion-molecule reactions, a sample molecule with a high proton affinity reacts in humid air under proton transfer to a positively charged ion. By doping the drift gas with NH3 vapor, acetone, chlorinated solvents, or others, the selectivity can be modified. Substances with electron-capturing capabilities, such as halogenated compounds, can be detected by potential inversion as negative ions as well. Another often used alternative, for compounds with sufficiently low ionization potential, is UV photoionization. It is appropriate for selective measurements of molecules with an ionization potential of less than 8-12 eV. After ionization of the air sample the ions are pulsed through a shutter into a drift tube, which is isolated from atmospheric air. The drift tube has a uniform weak electric field, which accelerates the ions along the tube. The movement is hindered by collisions, until the ions reach the detector at the end. Depending on the ion impact, a current is generated and measured over the time of flight. For a manageable and calibrated component amount this gives information about the identity and concentration. If the composition is too complex however, this often fails, because of ion-ion interaction or overlapping peaks. In this case, classical electronic nose data evaluation algorithms (adapted from spectroscopy)22,23 can be applied to gain a maximum of information out of the measurements. Compared to mass spectrometry, the virtual sensor array is not given by discrete mass/charge relations, but by the signal integration over definable time intervals.

Electronic Nose

Figure 5. The selective hybrid microsystem consists of a zero grade air unit, a commercial minipump, a minivalve, a silicon micromachined packed GC column, and an MOX sensor as the detector. The analysis time of a certain mixture of volatiles depends on the type of stationary phase, gas flow rate, column length, and temperature of the GC column. Zampolli et al. have shown that within 15 min the complete separation of benzene, toluene, and m-xylene is possible. Reprinted with permission from ref 26. Copyright 2005 Elsevier. By the use of a temperature-controlled capillary column the separation time for microfabricated systems can be decreased.27,28

2.2.4. Gas Chromatography Although it is possible to separate mixtures by using the properties of their ions in electrical or magnetic fields, the most established and widely used technique in analytical chemistry is to separate them by chromatographic methods. In the case of volatiles, gas-liquid chromatography and gassolid chromatography are possible ways. The sample, transported by the mobile phase (gas), is directed over the stationary phase (liquid or solid) and interacts with it. Depending on physical and chemical properties, such as the boiling point, the polarity, H-bonding, polarizability, etc., the affinity of each single substance for the stationary phase is different. The partition behavior determines the retention time of the components and, consequently, the order of elution. Because, compared to sophisticated analytical chemistry, the claim of electronic noses is to be simple and fast in use, GC entered in this field not in the conventional but in the fast or ultrafast mode. To increase the separation speed during analysis, different parameters have to be adapted. For gasliquid chromatography this can be an increase of the carrier gas flow rate, an increase of the temperature-program heating rates, a reduction of the column length, a reduction of the column diameter, a reduction of the thickness of the stationary phase, and the use of a faster carrier gas. Depending on the sample, it is important to avoid using all possibilities at once, because this always results in a decrease of the resolution, the sample capacity, or both. It is also important to note that these optimizations increase the demands on the detector technology used in terms of sensitivity, speed, and dead volume. To simplify the evaluation, the signal over defined time intervals is again integrated and treated as the sensor response of a virtual sensor array.24,25 An example of an electronic nose using chromatography technology is shown in Figure 5.

2.2.5. Infrared Spectroscopy Infrared (IR) spectroscopy can also be considered as an electronic nose.29-31 In a range between 4000 and 200 cm-1,

Chemical Reviews, 2008, Vol. 108, No. 2 709

molecular vibrations and higher energy levels are excited. Through characteristic absorption bands the type of chemical bonds can be determined, and pure chemicals can be identified by their unique fingerprint spectrum. The spectrum corresponding to mixtures is evaluated by classical electronic nose algorithms. For the detection of substances in the gas phase, two affordable methods for mobile devices are established. In photoacoustic infrared spectroscopy, a modulation of the intensity of an IR source causes a temperature variation and the resulting expansion and contraction of the gas will be measured as audible frequencies with a microphone. Alternatively, the absorbed energy of a narrow bandpass infrared beam is measured in filter-based infrared spectroscopy. Commercially available devices (e.g., MIRAN SapphIRe from Thermo Scientific) are mostly used for absolute measurements of concentration either in detection of a single species which has a unique absorbance wavelength or by analysis at multiple wavelengths for a known gas mixture. However, where the constiuents of the gas mixture are unknown, these instruments can also be combined with pattern recognition and used as an electronic nose. Despite confirmed feasibility,29 the infrared-based nose has not become popular and commercially available devices such as the MIRAN SapphIRe from Thermo Scientific can rather be considered as portable analytic tools than as electronic noses.

2.2.6. Use of Substance-Class-Specific Sensors The types of electronic noses discussed all have one characteristic in common in that they measure a set of features, subsequently analyzed by a fixed algorithm to compare samples in a qualitative or quantitative way without targeting the exact identification or concentration of the single compounds. Similarly to human olfaction, the outcome should only be to determine the sample’s identity (orange or apple), to verify variations (compare batches), or to give a prediction on the differences between samples (e.g., intensity of odor correlating with spoilage). In this context, detailed analytical results of the composition are not wanted and often are not available. These facts are reflected in the setup used, where one does not aim to detect one specific substance with one sensor, but one aims to have a broad selectivity and afterward extract the wanted information by comparing the sensor signals. For MS, IMS, and especially GC noses, the number of detectable target molecules per virtual sensor is much more limited. Therefore, an MS nose can detect the presence of high molecular weight substances even without elaborate data evaluation, and a GC nose can differentiate easily between polar and nonpolar substances or between low- and high-volatility compounds, depending on the column used. To follow this line of thought, the next step is to include detectors which are able to detect only one substance/class of interest and not all of the compounds present as with MS and GC. This can either be a classspecific device such as a flame spectrophotometer, which only detects phosphorus-containing compounds, or a standalone device of broad selectivity, such as a thermal conductivity detector measuring nearly every composition change in an air sample. Strictly speaking, these are not independent electronic noses, but they can be integrated into one as a supplementary module providing additional information. The flame photometry detector (FPD) is based on the decomposition of any organic compounds in a hydrogen flame. If phosphorus or sulfur is present, light of a specific

710 Chemical Reviews, 2008, Vol. 108, No. 2

wavelength will be emitted. After the other wavelengths are masked out through filters, a photomultiplier detects the concentration of one of the elements. Because phosphorus and sulfur are present in classical nerve gases, this technology is often used in the military/security application field. Another previously mentioned detector is the photoionization detector (PID). Without coupling it to an ion mobility spectrometer, it is also possible to use it as a standalone detector to measure all volatile organic compounds that have ionization potentials equal to or less than the energy of the UV radiation. For example, by using a 9.5 eV lamp, amines, benzene, and aromatic compounds are detectable. A 10.6 eV lamp additionally detects ammonia, ethanol, and acetone, whereas acetylene, formaldehyde, and methanol are only to be detected by using an 11.7 eV lamp. Single gas detection of oxygen or toxic gases is typically performed by electrochemical cells (ECs). They are designed to detect one special gas, but despite their particular filter, electrodes, and electrolytes, they are often not completely specific. Behind a diffusion barrier the target gas is either oxidized or reduced and determines a current between the sensing and the counter electrode. This current is proportional to the target gas concentration. The third electrode, the reference electrode, has a stable potential and is used to eliminate interferences from side reactions and increase the selectivity of the electrochemical cell. For a nonspecific determination of flammable compounds, flame ionization detectors (FIDs) are used. In a hydrogenoxygen flame the compounds are burned in an electric field, and the increases of ions are detected as an electrical current. Because all organic compounds are detectable, flame ionization detectors are often used in gas chromatographs, but they are available as stand-alone devices as well.

2.3. Combined Technologies The combination of different sensor or detection technologies comes along with an improvement of the selectivity range but determines at the same time an increase of the setup complexity and, accordingly, additional costs for the whole device. Thus, the combination of different technologies is only reasonable for the following two cases: first, for a special problem where a single technology does not achieve satisfactory results and, second, for an all-purpose electronic nose with a maximum of application possibilities. The ideal all-purpose electronic nose does not exist: however, systems that can be applied to more than one application field are available. One example for the latter case is the electronic nose Prometheus produced by Alpha MOS. It combines a sensor array with a fingerprint mass spectrometer. The sensor array consists of 18 different sensors. These are arranged in three separate sensor chambers equipped with six different metal oxide sensors. If desired, the use of conducting polymers or quartz microbalances is also possible. The fingerprint mass spectrometer consists of an electron impact ionizer and a quadrupole mass filter. It can be operated in the single ion mode, or alternatively, the range between 1 and 200 amu will be scanned. The combination of these technologies causes both high selectivity through mass spectrometry and high sensitivity through the use of a sensor array. The system is more flexible in use compared to the individual parts and thereby appropriate for more applications. Another hybrid system is the GDA 2 (Gas Detector Array 2) produced by AIRSENSE Analytics. It consists of an ion-

Ro¨ck et al.

mobility spectrometer with a 63Ni ion source which can be used in the positive and negative modes, a photoionization detector with a 10.2 eV lamp, an electrochemical cell, and two metal oxide sensors. The manufacturer recommends this portable device for detection of hazardous gases and chemical warfare agents. Because a variety of different harmful agents, such as ammonia, benzene, carbon monoxide, chlorocyane, hydrogen cyanide, and phosgene, should be detectable, it is necessary to use sensors and detectors whose sensitivity and selectivity cover the whole range of potential substances and concentrations. This is assured by the use of different technologies.

3. Companies The previous section described how sensing odors using an electronic nose is a significant technical challenge. Instead of attempting to reproduce human odor impression, most commercially available instruments nowadays have other application areas. The classification of odors is not in the fore, but the detection of any volatiles giving information about a characteristic of the sample is. The range of electronic noses on the market spans from military, security, and safety applications, food processing, and medical applications to use in the pharmaceutical industry, and even includes mass markets such as automotive applications or white goods. The border between classical analytical systems, electronic nose technology, and detectors for specific substance classes or even single compounds becomes more and more fuzzy. Some manufacturers call their devices “electronic noses”, whereas others avoid mentioning this term even if their product operates in a similar way. Table 1 gives an overview of electronic noses on the market according to the criteria above, listing their manufacturers and technology basis.

4. Application Areas In the past two decades, the applicability of electronic noses has been tested in every imaginable field where odors or odorless volatiles and gases are thought to play a role.32-35 A typical approach was to prove the ability of a given sensor array to discriminate a sample set in a desired manner (the black box approach). Consequently, researchers were frequently overly hasty in concluding that positive experimental results demonstrated success in the application. As a result one was considered to have reached the target and/or went ahead to the next challenge: the quantification of the sample property of interest. Taking the electronic nose as a black box, without having a feeling for the chemical processes going on and having no idea about the marker substances and interferents, one becomes critically dependent on the sample set. Accordingly, it is very important to be aware of the fact that one can sometimes have a limited or even a biased sample set, and as a consequence, the initial results can look much better than they are in reality. Typical examples have included the determination of the quality of complex food products, see section 4.1, such as coffee, tea, olive oil, or wine.36 Under laboratory conditions for a strongly restricted set of samples, the correlations may succeed: nevertheless, no commercial breakthrough to industry took place. There are many reasons for this approach to fail; one key factor is often a mismatch between the detector sensitivity and the components responsible for the odor.37 For an unrepresentative sample set there is a high risk of discovering bogus correlations with the consequence that for unknown

Electronic Nose

Chemical Reviews, 2008, Vol. 108, No. 2 711

Table 1. Commercially Available Electronic Noses manufacturer

no. of systems sold

Agilent, http://www.chem.agilent.com/ AIRSENSE Analytics, http://www.airsense.com/

180

Alpha MOS, http://www.alpha-mos.com/

500

AltraSens, http://www.altrasens.de/ AppliedSensor, http://www.appliedsensor.com/ Chemsensing, http://www.chemsensing.com/ CSIRO, http://www.csiro.au/ Dr. Foedisch AG, http://www.foedisch.de/

>100 000

Draeger, http://www.draeger-safety.com/ Electronic Sensor Technology, http://www.estcal.com/

Environics, http://www.environics.fi/

9000

Forschungszentrum Karlsruhe, http://www.fzk.de/ Gerstel GmbH & Co. KG, http://www.gerstel.com/ GSG Mess- und Analysengera¨te, http://www.gsg-analytical.com/ Illumina, http://www.illumina.com/ Microsensor Systems Inc., http://microsensorsystems.com/

Owlstone Nanotech, Inc., http://www.owlstonenanotech.com/ Proengin, http://www.proengin.com/ RaeSystemes, http://www.raesystems.com/

model 4440A i-PEN PEN3 GDA 2 FOX 2000 FOX 3000 FOX 4000 Gemini Kronos Heracles RQ Box Prometheus OdourVector Air Quality Module Cybernose OMD 98 OMD 1.10 Multi-IMS MSI150 Pro2i ZNose 4200 ZNose 4300 ZNose 7100 M90-D1-C ChemPro100 SAGAS QCS MOSES II oNose Hazmatcad Hazmatcad Plus Fuel Sniffer CW Sentry 3G SAW MiniCAD mk II VaporLab Tourist Lonestar AP2C TIMs detector ChemRAE UltraRAE Eagel monitor AreaRAE monitor IAQRAE

RST-Rostock, http://www.rst-rostock.de/ Sacmi, http://www.sacmi.eu/ Scensive Technologies Ltd., http://www.scensive.com/ ScenTrak, http://www.cogniscentinc.com/ SMart Nose, http://smartnose.com/ Smith Group, http://www.smithsdetection.com/

<100

Sysca AG, http://www.sysca-ag.de/ Technobiochip, http://www.technobiochip.com/

samples the model will fail. For example, the prediction of the ethanol percentage in the headspace of a wine sample by an electronic nose is easy to accomplish, while, on the contrary, even with elaborated analytical equipment it is not possible to entirely comprehend the quality of wine samples. For a chosen sample set, where the goal is to judge the

250

FF2 GFD1 EOS 835 EOS Ambiente Bloodhound ST214 SMart Nose 2000 Cyranose 320 IONSCAN SENTINEL II CENTURION GID-2A GID-3 SABRE 4000 ADP 2000 CAM Artinose LibraNOSE 2.1

technology quadrupole fingerprint mass spectrometry gas sensor array gas sensor array IMS, PID, EC, 2 MOX sensors 6 MOX sensors (or QMB/CP) 12 MOX sensors (or QMB/CP) 18 MOX sensors (or QMB/CP) gas sensor array quadrupole fingerprint mass spectrometry 2 capillary columns (1-3 m) and 2 FIDs EC, PID, MOX sensors MS and 18 MOX sensors 6 sensors 2 MOX sensors colorimetric array receptor-based array 2 × 6 sensors 2 × 5 MOX sensors ion mobility spectrometry ECs GC and SAW GC and SAW GC and SAW ion mobility spectrometry ion mobility spectrometry 8 SAW sensors 3 MOX sensors modular gas sensor array fluorescence sensorssbead array SAW SAW array and EC SAW SAW and electrochemical sensor array 2 SAW array GC and EC field asymmetric ion mass spectrometry field asymmetric ion mass spectrometry flame spectrophotometer flame spectrophotometer ion mobility spectrometry separation tube and PID GC and EC PID, 2 ECs, 1 catalytic bead sensor, O2 sensor PID, NIRD CO2, EC, polymer-capacitated humidity sensor, thermistor, humidity-temperature sensor 6 MOX, T, humidity 6 MOX, T, humidity gas sensor array gas sensor array 14 conducting polymers fluorescent dye quadrupole fingerprint mass spectrometry gas sensor array ion mobility spectrometry ion mobility spectrometry ion mobility spectrometry ion mobility spectrometry ion mobility spectrometry ion mobility spectrometry ion mobility spectrometry 38 MOX sensors 8 QCM sensors

quality of the wine or the grape variety of the samples by an electronic nose, the ethanol concentration may be fortuitously correlated with those characteristics.38,39 This way we will obtain the right results by dealing with the wrong input data. Admitting that such obvious mistakes are actually avoided, for each application it is still possible to have others

712 Chemical Reviews, 2008, Vol. 108, No. 2

more often than not unknownssubstances not related to the targeted sample characteristics but having a considerable impact on the (classification) result. The conclusion is that, to prove the applicability, analytical background information and/or lots of independent validation measurements are needed (in the ideal case both are to be used). Returning to the wine example, the validation should recognize that characteristics such as ethanol concentration, grape variety, vintage, wine region, or winery are not correlated by chance and outliers in prediction should be critically examined and compared to sample properties. In doing so it is always better to increase the sample set by using new independent samples instead of repetitions or mixtures of old ones to uncover unexpected correlations. For historical reasons, the main research fields for electronic nose technologies are still related to those areas where the human olfaction system is relevant. During recent years many efforts were made in the field of foodstuff and beverages where, in addition to classification, time-dependent processes were investigated.35,36,40 These include unwanted processes such as changes during storage or spoilage as well as the intended ripening or fermentation of particular products. The driver is that electronic noses are by far less expensive when compared to classical analytical systems such as GC/MS or the running costs for human sensory panels. For this reason, the aim has been to replace one or the other established methods or at least to complement them. Besides cost savings, electronic noses promised fast, roundthe-clock operation, which, combined with an automated data evaluation, could at least for some applications replace humans. In addition to the assessment of food, the human nose gives us further important information: It warns us about dangers such as fire or air pollutants and gives us indication of certain diseases such as diabetes or hepatic failure.41-44 Consequently, there are also efforts to mimic this human ability with electronic noses.35,45-49 Because of the different responses and sensitivities to the respective marker molecules, one needs to find which are the appropriate tasks for electronic noses. Therefore, current research also explores the field of marker molecules that are odorless for humans. One further step is to improve on human capability and target instead that of macrosmatic mammals. Even if modern research shows that for some odorants the perception of humans and primates is comparable to that of canines and rodents,50 the ability of the latter is superior in many fields. For instance, dogs are able to identify individuals by their scent, to track them, or to track down hidden narcotic drugs or explosives.51,52 Recently, the capabilities of insects have been investigated, and the feasibility of using honeybees for land mine detection has been demonstrated.53 However, dogs show behavioral variation depending on changes in their mood, and all animals are subject to fatigue. To decrease the complexity of execution, it would be desirable to have an artificial system with the same performance. For this reason, electronic noses are being investigated in the security field for the detection of hazardous substances and explosives. Process control is also a promising application field. Independent of the character of the product, it is important to ensure it always has the same characteristics. Therefore, the application area ranges from control of industrial production lines as in the pharmaceutical industry and in the

Ro¨ck et al.

manufacture of food packaging to the control of composting processes. Besides the control of temperature, humidity, optical appearance, viscosity, etc., the electronic nose adds another dimension in observation and can help minimize the variability between batches.

4.1. Food and Beverage Applications in this field include inspection of the nature and quality of ingredients, supervision of the manufacturing process, and, finally, everything related to shelf life. For instance, it is important to be able to distinguish between different quality classes of the same food, e.g., extravirgin olive oil, virgin olive oil, olive oil, and olive-pomace oil, to avoid fraud and to fulfill customer expectations. Equally important is knowledge of the ingredients of a product to protect the customer from low-quality raw material and/or to avoid breaking the law. For instance, in the European Union one has to make sure that cheese sold as “feta” is only made from goat and/or sheep milk without additions of cow milk to fulfill the protected designation of origin (PDO). To avoid a low-quality product and to reduce defects during the production process, it is desirable to detect irregularities at an early stage and to initiate remedial action as fast as possible. This includes individual adaptation of the treatment of biological raw materials related to their natural variability. Fermentation and roasting processes are examples where the conditions used have a direct influence on the taste and odor of the product and where sophisticated monitoring helps to increase the quality. Because degeneration processes cause off-odors, off-flavors, or in the worst case harmful substances, the detection of spoilage, no matter whether chemical, enzymatic, microbiological, or a combination of these, is an important task in itself and one which opens up the possibility of predicting shelf life. The established methodologies to deal with this challenge are diverse and range from microbiological analysis to sensory test panels to classical analytical approaches. The information they provide is not all the time orthogonal. The question from the electronic nose point of view is which additional information can be obtained by using it and in which fields can it replace the established techniques. To get a feeling on what is feasible, one has to acquire knowledge about the substances detected by the human nose, by classical analytical detectors and by the electronic noses. For that reason, gas chromatography experiments are very helpful because they reduce the problem from the whole bouquet to the single substances. In aroma and odor analysis GC-olfactometry (GC-O) has been established for many years and helps to identify which volatiles are responsible for the respective odor impression (Figure 6). Direct comparison of GC-O results with GC/FID or GC/MS results gives information about which marker substances are detectable without a sensory test panel and about those for which the human nose is the only reliable detection method. This is important to know because measurements with foodstuff such as daidai peel oil,55 green Mexican coffee,54 grapefruit oil,56 cooked asparagus,57 cashew apple nectar,58 tarhana,59 or Croatian Rhine Riesling wine60 have shown that sometimes there is a big discrepancy between substances detectable by the human nose and those detectable with commercial detectors and vice versa. One still needs to keep in mind that, despite this systematic approach, the rules governing the combination of individual chemical compounds in the global aroma of a product are not yet fully understood and

Electronic Nose

Figure 6. Comparison of a GC/FID chromatogram (top) with a time-intensity aromagram (bottom, inverted) of grapefruit oil. Some odorants have been identified by mass spectrometry. It is obvious that the human nose is sensitive to substances the flame ionization detector is not able to detect and vice versa. Reprinted with permission from ref 56. Copyright 2001 John Wiley & Sons, Limited.

Figure 7. HRGC/SOMMSA (high-resolution gas chromatography/ selective odorant measurement by a multisensor array) for sensor evaluation. In this example the sensitivity of a SnO2 sensor at different temperatures (165, 220, 270, and 330 °C) to compounds out of beech wood smoke was tested. The output from the gaschromatographic column is split in two to enable simultaneous measurements with a reference detector (mass spectrometer in this case) and the sensor array to identify relevant compounds. Reprinted with permission from ref 65. Copyright 2003 PCCP Owner Societies. With this approach the choice of adequate sensors/ conditions is possible without costly experiments at the gas mixing system.61-63

the different methodologies allow us only to widen the limited view on the whole scenery. Nevertheless, chromatography has already been successfully used to ensure the appropriateness of chemical sensors to a given problem.61-63 For example, this approach was used to prove the sensitivity of metal oxide sensors to food aroma during baking and roasting processes64 (Figure 7). However, it is also applicable to other problems such as the detection of odorless volatiles or the selection of gas sensors. A very promising application field for the electronic nose is its use in spoilage detection of foodstuffs. The fight against autolysis and against the growth of microorganisms is the main objective for food preservation and can be reached in

Chemical Reviews, 2008, Vol. 108, No. 2 713

Figure 8. Effect of water activity on germination of one isolate of Eurotium spp. at 25 °C on flour wheat-sucrose agar. Water activity levels are (O) 0.90; (4) 0.875, (0) 0.85, (b) 0.825, (2) 0.80, and (9) 0.775. It is shown that the temporal rates of germination depend strongly on the water activity. Reprinted with permission from ref 84. Copyright 1999 Blackwell Publishing.

different ways. The most popular approaches are pasteurization, refrigeration, removal of water, change in pH, the use of packaging under vacuum, the use of food additives, or a combination of these. In all cases, food deterioration cannot be prevented but only postponed. Therefore, the challenge is to detect spoilage at an early stage or, alternatively, to predict it.66 The field is quite complex as both the nature and origin of the foodstuff and the preservation technique used influence the species of bacteria, fungi, or enzymes responsible for spoilage. Due to the variety of different substances that can be produced during spoilage, the biologically evolved human perception is still the best detection method for most applications of off-odor and off-taste detection. To use instrumental analysis, one has to be aware of the relevant substances for each sample type, but despite our knowledge of the formation of free radicals, influence of enzymes, different bacteria which are produced, yeast and mold strains, and their metabolism products, the experience with the electronic noses in detecting them is still at the beginning. First trials with red wine,67 apples,68 mandarins,69 bakery products,70-73 bread,74 wheat,75-77 Crescenza cheese,78 beef,79,80 poultry meat,81 and milk82,83 show that, in principle, differences caused by spoilage are detectable with an electronic nose. For instance, it was shown that it is possible to track the changes in the headspace of an individual food sample during storage. The critical point is the generalization and, closely connected to it, the question of the usability of the electronic nose results without first thoroughly exploring all applications’ variables (different samples, different batches, long-term behavior, etc). Because foodstuff is very heterogeneous, there is no warranty that the results will be reproducible for a sample set varying in an unconsidered parameter. According to Abellana et al., the speed of fungal spoilage depends not only on temperature but also on the water activity in food (Figure 8).84 For simplification, these variables are often kept constant to have a direct correlation between spoilage level and time. Keshri et al. showed that with a Bloodhound BH-114 electronic nose it is possible not only to detect spoilage but even to differentiate and classify the fungal species in the bread analogue.74 However, the question of the validity of their results for different humidities, different corn varieties, variations in baking time, or various bread volume/surface ratios is still unanswered.

714 Chemical Reviews, 2008, Vol. 108, No. 2

Fish spoilage is one of the best investigated deterioration processes with respect to an electronic nose. Knowledge ranges from the very basic post mortem biochemical processes in the fish to the specific volatiles produced and their relationship to the perceived odor. As the oxygen supply stops, the proteolytic mechanisms involved in disorganization of fish muscles are initiated, and hence, the muscles are tenderized.85 The autolytic modifications start with an anaerobic degradation of the stored carbohydrate glycogen to lactic acid, and hence, the pH value drops from close to 7.4 to around 6. The muscle osmotic pressure increases, ATP (adenosine triphosphate) is hydrolyzed, and lipids are oxidized. TMAO (trimethylamine oxide) is reduced to TMA (trimethylamine), nitric oxide and reactive species of oxygen increase, and calcium ions are released into the cytosol. Finally, endogenous enzymes, especially calpains, cause proteolysis of muscle proteins and connective tissue as well as fat hydrolysis. The growth of microorganisms is now supported by the availability of catabolites86 and is dependent on extrinsic and intrinsic factors. The main extrinsic factors are temperature and composition of the atmosphere, whereas the fish species is fundamental for the intrinsic factors. These include the poikilotherm nature of the fish, its aquatic environment, the post mortem pH of the flesh, and the concentration of nonprotein nitrogen and of TMAO. These variables not only determine the absolute microbiological growth but are relevant for the increase of each individual strain and consequently for the proliferation ratio between them. Therefore, for different fish species under different storage conditions (air, vacuum packed, CO2 atmosphere) different spoilage organisms dominate, primarily, Vibrionaceae, Shewanella putrefaciens, Pseudomonas spp., Photobacterium phosphoreum, Lactobacillus spp., and Carnobacterium spp.87 It is not surprising that the sensory descriptors for the metabolites produced by different microorganisms vary.88 For instance, marine temperate-water fish have an offensive fishy, rotten, H2S off-odor, whereas some tropical fish and freshwater fish stored in air can be described with a fruity, sulfydryl off-odor.86 Regarding the volatile spoilage products, Malle et al. showed that the ratio between TVBN (total volatile basic nitrogen) and TMA (trimethylamine) can be used as a quality index for sea fish.89 Because of its restricted precision and limited applicability, it should be only used as an orientating method.90-92 In search of further spoilage markers, Duflos et al. identified 20 common volatiles from whiting, mackerel, and cod.90 For these substances, the contribution to the entire smell of the fish is partially known.93 It was shown that the characteristic spoilage compounds fluctuate significantly from one species to another. Furthermore, there are even quantitative and qualitative differences of volatiles between fish skin and fish muscle for the same species.94 To conclude, much basic research in the field of fish spoilage has been carried out, and useful marker molecules detectable by an electronic nose are known; at the same time, one has to be aware that a lot of different factors influence the smell and the headspace composition of the stored fish (see section 5.1). For that reason the precision of prediction of the electronic nose will increase with the homogeneity of the sample. With these limitations, special attention has to be paid to the comparability of the training set and the real-life samples. The suitability of different electronic noses has been evaluated for fish freshness applications, with transducing principles ranging from electronic noses with electrochemical

Ro¨ck et al.

gas sensors to metalloporphyrin-coated QMB, metal oxide sensors, conducting polymer sensors, computer screen photoassisted based gas sensor arrays, and vapor-phase Fourier transform infrared spectroscopy. It is difficult to compare the different approaches because of the different conditions of the experiments. One exception is the work of Di Natale et al. where for the same sample set of cod fish fillets the commercially available electronic noses FreshSense (Element-Bodvaki)sconsisting of five electrochemical sensorssand LibraNose (Technobiochip)sconsisting of eight thickness shear mode resonatorsshave been tested.95 Data evaluation was done by PLS-DA (partial least-square discriminant analysis), where both systems demonstrated sensitivity to the temporal variations of fish headspace. For the leave-one-out validation the misclassifications of storage times were 33% and 9%, respectively. It was possible to achieve a value of only 4% for the combined input data of both electronic noses. However, this is not surprising as PLSDA is a supervised classification method, so the prediction should be improved by adding additional inputs. Furthermore, these values should not be seen as representative of general application because of possible flaws in the calibration method. In this work eight samples out of three batches were measured for each storage time, but to obtain a reliable prediction model for unknown samples, a segmented crossvalidation instead of the leave-one-out method for the prediction of the freshness of fish would be desirable.96,97 This is the requirement to ensure a correct classification of unknown batches not already comprised in the calibration data set. Otherwise, differences in new batches caused by, for example, different fishing grounds, the fishing season, the fat content in the flesh, or physical damages due to rough handling and bruising will not be taken into account.98,99 A very good study about the potential of electronic noses in this field was presented by Olafsdottir et al.100 Using a fundamental approach, the chemical reason for the sensor response of an electronic nose consisting of four electrochemical gas sensors was identified. The experimental setup used included microbial analysis, determination of TVBN, pH measurements, GC/MS measurements, and GC-O measurements. Thereby, it was possible to determine the increase of the most abundant volatile spoilage compounds over time, including their standard deviations (Figure 9). In addition, the instrumental detectable compounds which influence the odor were identified, and their contribution (intensity, description) to the overall odor was evaluated. At the same time the work demonstrates the discrepancy between substances detectable by humans, the mass spectrometer, and the chemical sensors used and points out the danger that volatile compounds are often not detected until the products are overtly spoiled; the TMA concentration significantly increased in this example, but the electrochemical NH3 sensor was not sensitive enough to contribute relevant information at an early stage of spoilage (Figure 10). In contrast, the CO sensor was suitable to detect incipient spoilage of the Styrofoam-packed chilled cod fillets because of its sensitivity to alcohols, aldehydes, and esters. This success has to be seen in the context of another publication of the authors. For haddock fillets they found that the absolute sensor response was higher.101 This can be interpreted as a proof of the described complexity of spoilage and the need for individual calibration for each product (here fish species) and storage condition.

Electronic Nose

Figure 9. Headspace composition of cod fillets as a function of storage time (storage temperature of 0.5 °C).100 A selection of components out of 25 substances quantified by GC/MS is shown. In addition to alcohols and esters, aldehydes, ketones, acetic acid/ and trimethylamine were detected. The time dependency for each analyte is different and has a high standard deviation (not shown). The signals from electrochemical gas sensors are presented in Figure 10. PAR ) ratio between the peak area of the analyte and the peak area of the external standard. Reprinted from ref 100. Copyright 2005 American Chemical Society.

Figure 10. Response of the electrochemical gas sensors toward cod fillets during storage at 0.5 °C on days 4, 7, 10, 12, and 14. The CO sensor was most sensitive to changes during spoilage because of its sensitivity to alcohols, aldehydes, and esters (Figure 9). Although the trimethylamine concentration increased significantly at day 14, neither the NH3 sensor nor the other ones contributed additional information. Reprinted from ref 100. Copyright 2005 American Chemical Society.

4.2. Environmental Monitoring Environmental monitoring has become more and more important during the past few decades with increased awareness of the effects of pollution on human health and the quality of the environment. Electronic noses have been

Chemical Reviews, 2008, Vol. 108, No. 2 715

investigated for detection of toxic compounds in the ambient atmosphere, at the source (e.g., on industrial premises), and in the headspace of water. In ambient air, toxic compounds are present at concentrations which will not have an immediate effect; nevertheless, the main components, namely, carbon monoxide, nitrogen oxides, sulfur oxides, volatile organic compounds, ammonia, ozone, and particulate matter, are a long-term danger for human health. For that reason, the regulating agencies introduced strict threshold values that have to be observed. This concerns, on one hand, the direct monitoring of emissions at the place where they occur and, on the other hand, monitoring of the concentration limits at the place where people are living and working. The thresholds are not limited only to substances which are known to cause physical damages, but also include compounds with unpleasant odor and, therefore, that reduce the quality of life. Until now, the detection and rating of emissions has been performed using traditional methods including olfactometry measurements realized by a human panel and identification and quantification by analytical instruments. The disadvantage of these techniques is that they are not appropriate for on-site realtime and continuous operation due to their high operating costs. The introduction of the electronic nose for this task issdepending on the target componentssvery challenging. In addition to very complex target mixtures and low detection thresholds, sampling is a major concern. Samples must be representative and independent of variable ambient conditions. Knowledge of spatial and time patterns of concentrations is important, particularly for air pollutants in urban areas where topography and meteorology create a complex pattern that has to be considered to place the electronic nose at the right positions.102 Additionally, changes in temperature and humidity influence the measurement results. To deal with this interference, two methodologies are commonly used. One is sample pretreatment to obtain fixed experimental conditions, and the other is a parametric compensation by additional measuring of the variable parameters and calibration under, e.g., different humidity conditions. From the practical point of view, one can distinguish between the following application areas: (1) the measurement of exhaust gas streams directly at the source of emission; (2) the measurement of ambient outdoor air to characterize broad area pollutant levels; (3) indoor measurements in vehicles, workplaces, and residential buildings;103 (4) the analysis of the headspace over polluted water or contaminated land. This classification can give a first indication about the particularities of the experiments. However, for each single application the sensitivity of the electronic nose to the target substances as well as to potential interfering substances has to be known. This principle is independent of the task and should be applied for the determination of the level of harmful substances, the estimation of odor emissions, and the determination of the general “air quality”. The only difference is the reference data set for calibration, which can be obtained by the approved analytical methods or by artificial mixtures of the critical components. An alternative approach is to try to differentiate between different samples without deeper background knowledge of the occurring substances. This provides an indication of the applicability of an electronic nose but cannot be seen as a proof-of-principle for real life conditions. Consequently, the usability of the gained information is strongly variable and ranges from the first steps toward a new application field

716 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 11. Time evolution of the odor concentration for compost emissions. This was directly calculated from the signal of one single sensor. The reliability of this procedure was surveyed by a sensor array combined with a PCA-based data evaluation. Thus, the presence of non-compost-related interfering gases was detected and taken into account (gap in the curve). These intervals were cut out. The peak at 10:48 can be explained by the turning of a compost row. Reprinted with permission from ref 125. Copyright 2006 Elsevier.

to well-developed prototypes for certain tasks. Using this limited approach, the electronic nose has already been tested for a wide range of applications: to determine odorous emissions from animal production facilities,104 emissions of malodors produced through industrial factories105 or waste disposal sites,106 and emissions at the point of odor production from the decomposition process of kitchen and vegetable waste.107 Applications where the odor impression does not come to the fore are the determination of single solvents or mixtures of them,108 the identification of microorganisms such as bacteria and fungi,109 the detection of leaking of refrigerant gas,110 and the differentiation between automotive fuels.111,112 Very practical examples are the detection of smoke compounds,113 the control of automotive ventilation,114,115 and the determination of indoor air quality.116-119 For headspace measurement of water samples, both very specific and more general cases were considered. Examples of the former are the determination of residues of insecticides120 and the amount of cyanobacteria121 in drinkable water. Examples of the latter are the determination of water pollution122 and sewage facility emissions106 and the general assessment of wastewater samples.123,124 Each application has its relevance, with some of them already further developed because of their extended impact and, consequently, their higher commercial prospects. Examples are the use of sensor arrays for comparing the in-cabin and outdoor air quality for cars for automatic flap-control systems, the use in failureproof fire detection systems, and air quality control for ventilation on demand. However, there are also emerging applications where the electronic nose has the potential to be established. One example is the use as a warning system to detect the emergence of odors from general waste. Nicolas et al. presented a simple approach to estimate the odor emission rate of a compost hall.125 The sensor signal of a single Figaro metal oxide sensor (TGS822) was correlated with the odor concentration measured by olfactometry. In a straightforward way the calibration was directly used to predict the malodorness and the possible odor annoyance for the neighboring area (Figure 11). Knowing that volatiles and gases from other sources also cause a sensor response, Nicolas used a sensor array of six metal oxide sensors to determine time intervals

Ro¨ck et al.

when interferences occur. Using this approach, it is possible to find out when the odor predictions are reliable and when they are influenced, for example, by exhaust gases emitted from trucks or machinery or conversely when an odorneutralizing product is sprayed in the hall. This is, of course, a simple but direct mode of operation which can be improved by compensating for temperature or humidity changes. It shows nevertheless in a clear way that a limited number of different sensors can deliver the desired information if the application characteristics are known. In a more elaborate way, Dickert et al. monitor the composting procedure with the long-term aim to ensure ideal transformation and avoid strong smells from a very early stage.126 With six QCM (quartz crystal microbalance) resonators, coated with different molecularly imprinted polymers, they trace four key analytes, namely, water, 1-propanol, ethyl acetate, and limonene. The concentration pattern of the organic compounds showed strong similarities to GC/MS measurements. Thus, it is possible to determine the state and the advancement of the degradation process throughout its different phases to completion. These two studies represent the first steps on the way to solving the problem of odor monitoring in a robust way, already demonstrating the capability to provide information about odor generation and the process of composting under reproducible conditions. For general applicability, the system should be explored for changing ambient conditions. That means a repetition of the odor-sensor signal calibration curve for other waste compositions and the comparison of odor evaluation through the composting process with the concentration of the key markers.

4.3. Disease Diagnosis Smell has been used to diagnose disease since ancient times and is directly linked to traditional medicine in different cultures. (“You can learn a lot just by smelling your patients with the unaided nose.” -Hippocrates, 430 B.C.) However, as modern diagnostic techniques provide more precise information with physical, chemical, and microbiological methods observation of odors fades into the background and is used only in some obvious cases as a disease indicator. The subjective odor perception of the physician is no longer required, although this ignores a lot of information on the health condition of the patients.43 Hence, there is considerable interest in a reliable device that could use the released volatiles and gases for objective diagnosing of a multiplicity of infections, intoxications, or metabolic diseases. Over recent years laboratory tests and instrumental analysis have been used to increase our knowledge about marker substances, their origin, and their smell. Despite this progress, the standard analytical method, namely, gas chromatography, has not been accepted as an accredited diagnostic tool. Apart from the cost of the equipment, the main reason is the complexity of its use. Because of the measurement time and the need for qualified labor in its operation, it is used neither in diagnosis nor for health condition monitoring. Still, the introduction of an easy to use diagnostic devicesbased on an electronic noseswould open up new fields of application. Several publications and reviews on disease marker substances and their detection reflect the interest in that matter. Independent of the detection methodology used, one important issue is always the question of sampling. The skin, the sputum, the urine, the stool, or the breath can be diseasecorrelated odor sources. This diversity of detection sites makes a universal sampling system, compared to the ability

Electronic Nose

of the human nose, impossible. Therefore, already from the very beginning, one has to adapt the system for the particular needs and the specific disease. Instead of direct measurements requiring complex sample strategies, one can consider the alternative of combining classical microorganism cultivation methods with subsequent analysis by an electronic nose. Of course, the analysis speed advantage for the whole procedure diminishes, but from the academic point of view, the resulting bacterium and fungus cultures are excellent objects to isolate the problem from interferences and, hence, are rewarding investigation subjects. In these in vitro experiments, the electronic nose has shown the ability to detect a variety of fungi and bacteria and, in some studies, to have even the ability to distinguish between them. Furthermore, the particular marker substances were identified by characterization of the gas phase above the microorganisms. Therefore, subsequent studies can fall back on sensors with the required selectivity. In this context the time dependency of incubation time and classification was checked for the in vitro experiments to obtain a reliable and, additionally, a fast classification. Another possibility to accelerate the identification of bacterial strains is to add biochemical precursors to the nutrient media for the liberation of specific odors through the pathogens.127 In clinical research the potential of electronic nose technology has already been tested for a variety of diseases. Swabs, sputum, serum, or urine samples were measured after a short incubation time or in some cases directly. The following list gives an overview on the most recent publications in this domain. (1) Beginning with the identification of bacteria, Parry et al. were able to recognize β-hemolytic streptococci extracted from chronic venous leg ulcers.128 (2) The screening for bacterial vaginosis in vaginal swabs seems to be feasible.129 Newer publications even give the impression that the reliability is comparable to that of present tests and show the possibility of controlling treatment of bacterial vaginosis by tracking the acetic acid concentration with a conducting polymer array.130,131 (3) Common bacterial pathogens of the upper respiratory tract were obtained from in vitro samples and successfully detected by a Cyranose 320 electronic nose.132 The same device was able to identify and classify pathogens from 90 patients suffering from ear, nose, and throat infections with a correct classification of 88.2%.133 (4) In search of the causative agent of tuberculosis, Pavlou et al. was the first to demonstrate proof-of-principle for the detection of mycobacterium tuberculosis in human sputum after incubation with an enzymatic cocktail.134 Furthermore, using the same Bloodhound 114 electronic nose, it was shown that one can distinguish between mycobacterium tuberculosis and other pathogens both in culture and in spiked sputum samples.135 By means of untreated serum, Fend et al. succeeded in diagnosing the agent of tuberculosis in badgers and cows, Mycobacterium boVis, as early as 3 weeks after experimental infection.136 They also used the Bloodhound 114 EN consisting of 14 conducting polymer sensors based on polyaniline. (5) A further field of bacterial disease is urinary tract infections on which first studies have been undertaken to detect the specific volatile pattern.137-139 (6) The analysis of urine by electronic nose technology is also able to detect metabolic disease. Mohamed et al. has predicted type 2 diabetes successfully with accuracy up to

Chemical Reviews, 2008, Vol. 108, No. 2 717

96%, depending on the data evaluation used.140 Furthermore, in renal dysfunction the capacity to remove metabolic products from the blood is limited, and the resulting change in body odor can be detected by an electronic nose.141 On the other hand, the volatile products which are accumulated exist as well in an increased concentration in the headspace of blood. Consequently, an electronic nose can be used for monitoring hemodialysis and to replace the established parameters based on urea concentration.142 In addition to the examples mentioned above, there are still other domains which are relatively unexplored. Examples are typhoid and yellow fever, where the skin has a smell resembling baked brown bread or a butcher’s shop, respectively. The sweat of diphtheria patients smells sweet, and the odor of sweat after a rubella infection has been compared to freshly plucked feathers. Rancid-smelling stools can be an indication of shigellosis, and as the name suggests, in maple syrup urine disease the urine smells of burned sugar.143 A multiplicity of further diseases can be detected by the analysis of breath. For diabetes a sweet, fruity smell is typical, reminiscent of decomposing apples. Uremia patients have a fishy breath, and a feculent odor can be caused by an intestinal obstruction or an esophageal diverticulum. Hepatic failure is the reason for the liberation of mercaptans and dimethyl sulfide, which smell like musty fish or raw liver. The origin of a foul, putrid odor can be a lung abscess or an empyema but just as well an intranasal foreign body.44 The main advantage of breath analysis, besides the detection of diseases directly related to the respiratory tract, is the fact that volatile organic compounds are mainly blood borne and the concentration of biologically relevant substances in exhaled breath closely reflects that in the arterial system. Therefore, breath is predestined for monitoring different processes in the body.144 Apart from the odor impression of specific diseases, much about the biochemical processes and the formation of marker substances is already known.145 In addition, direct sampling is possible without further timeconsuming sample preparation; therefore, breath measurements are suitable for a straightforward and easily achievable diagnosis by the use of an electronic nose. This represents a noninvasive and easily repeatable test that is not disagreeable or embarrassing for the patient compared to blood or urine tests. In spite of the ease of the sampling procedure, special care has to be taken to take the measurements in a reproducible way. In principle, two elements should be considered: First, the different approaches of breath collection should distinguish between pure alveolar gas and the total volume of exhaled breath, which consists of a mixture of dead space air and alveolar air. Additionally, factors such as exhalation speed and ambient temperature have to be standardized.146,147 Second, a correction for exogenous concomitants present in the inhaled ambient air should be carried out.144,145 Without going deeper into the chemical pathway of substances appearing in human breath, examples for analysis done by an electronic nose are the following. (1) The detection of the ethanol content of exhaled breath is the only example not directly connected to a disease,148 but from the practical point of view, the quantification of acetone, which is the marker substance for ketoacidosis, can be solved in a similar approach. This is a possible way to screen for diabetes.149,150 (2) In contrast, the substance of interest for the detection of asthma is an inorganic gas, namely, nitric oxide.151-153 By means of an electronic nose, patients with asthma can

718 Chemical Reviews, 2008, Vol. 108, No. 2

be clearly discriminated from the control group, whereas the accuracy of classification of severity is less reliable.154 Recently, a hand-held device was developed by Aerocrine, the NIOX MINO, which is able to determine the NO concentration in exhaled breath accurately.155,156 (3) Uremia can be reliably detected, whereas between patients with chronic renal insufficiency and chronic renal failure the correct classification is limited to 86.78%.157 (4) Examples for the detection of bacterial and/or fungal respiratory disease are chronic rhinosinusitis158 and the very promising approach to identify ventilator-associated pneumonia in patients in surgical intensive care units.159-161 (5) The reason for halitosis is sulfur-containing gases of oral bacterium origin,162 which is normally evaluated by an organoleptic test. Tanaca et al. and Nonaka et al. presented a clinical assessment of oral malodor by an electronic nose system.163,164 (6) Finally, several groups have undertaken efforts to detect lung cancer.165-170 To conclude, the possibilities for the application of the electronic nose in the medical field are very diverse as the different examples have shown. There is a need for preventive medical checkups to diagnose disease early, to speed up the healing process, to increase the rate of complete recovery, and consequently to save money for the health care system.171 Despite the potential of the electronic nose in this field, for applicability one has to minimize the false positive rate andseven more importantsthe false negative rate. Because humans are a very heterogeneous sample set, one has to know the effect of most common variables on the classification. These can be additional diseases or changes in nutrition,172,173 in medication,174 or in the use of cosmetics. Furthermore, animal experiments suggest the existence of sex-dependent pheromones,175,176 and behavioral studies show that individuals of different genetic backgrounds,177-179 ages,180 menstrual cycles,181,182 or even emotions183 are differentiable. This variability in sweat, urine, saliva,178 or just breath178,184 complicates the implementation of an electronic nose, in the first instance for diseases which are correlated with one of the odor-influencing factors mentioned. This problem should be illustrated in detail by the use of a concrete example. It is known that the breath of lung cancer patients has a defined odor and dogs can be trained to distinguish between the exhaled breath samples of sick and healthy test persons.185 The metabolic pathway for the formation of several biomarkers has been clarified,186 and volatile marker substances in the breath have been identified.187,188 On the basis of chromatography and subsequently selection of the important peaks, a prediction of lung cancer had an accuracy comparable to that of screening chest CT.189,190 Machado et al. used a Cyranose 320 electronic nose consisting of 32 polymer sensors.168 The training set consisted of breath samples from 14 individuals with relatively advanced bronchogenic carcinoma and a control group of 45 individuals consisting of a combination of healthy persons and patients with other diseases. Support vector machine analysis was used to diagnose cancer in an independent validation group. The result was that 10 out of 14 cancer patients were classified correctly and 57 out of 62 individuals of the control group were correctly identified. Repetitions of the misclassified measurements (normally four) were in most cases misclassified again. In a letter to the editor, Phillips criticizes the fact that there is no evidence for the sensitivity of the sensors used to the biomarkers

Ro¨ck et al.

available.191 Therefore, he suspects that there is a possibility that other substances may be responsible for the successful discrimination. One possibility may be compounds directly from tobacco smoke in lung cancer patients. In the study, the target group was also significantly older than the healthy control group. For that reason Phillips speculates that a higher amount of consumed cigarettes during lifetime may be an explanation for the observed results. This case study demonstrates the difficulties that have to be faced in the medical field to construct a robust prediction model. Risk factors, which are often linked to diseases, should not be wrongly treated as a calibration basis.

5. Research and Development Trends After the initial euphoria engendered by the prospect of replicating biological olfaction, the limits of electronic nose technology were realized and linked primarily to the fundamental sensing components192 and the sampling system. For the former, the problem is that, in contrast to nature, the information gained by adding additional sensors rapidly saturates. Therefore, the knowledge content provided by the sensor arrays currently used is far from the one delivered by the receptor cells of the olfactory epithelium. Consequently, an increase in selectivity (i.e., an increase in the number of sensors delivering useful new information) is necessary to enhance the capabilities of the electronic nose. On one hand, this is, of course, possible by the improvement of the individual sensors, which is not the main topic of this review. On the other hand, regarding the electronic nose as a complex system comprising a sampling system, the sensor array itself, the reference data set, and the data evaluation algorithms, there are other starting points for improvement. In this context, a higher sensitivity is often demanded to open up new application fields where trace components are the subject of interest. For instance, the human perception is usually sensitive to odor compounds down to the parts per billion range.193 However, for some substances the detection threshold is even several orders of magnitude lower, as the example of 2,4,6-trichloroanisole shows194 (the target compound for the cork taint in wine quality applications). This benchmark established by human perception is the target for an electronic nose;39,195-199 additionally, it must show its ability when compared to analytical systems.200,201 Besides this well-known application, the detection of explosives is of special interest in recent research and a further example of the need of highly sensitive systems. Because of the low vapor pressure of most explosive substances, the concentrations in the gas phase are in the same range as the previous example or even below.202 Despite the advances in the sensor field based on different transducer principles5 and a multiplicity of different preconcentration possibilities,203,204 one has to point out once more that for real-life applications an increased sensitivity of the system can only be useful if sufficient selectivity is provided. Otherwise, the interferents will cover the target compounds. The established all-around electronic nose systems produced by different companies have finally found their place in basic research and for some particular applications in laboratories. When it comes to mass market applications, a highly optimized system for the specific operating conditions is necessary. This can be a flapcontrol system in the automotive area,114,115 a fire detection system,205 or a quality control device for food packaging206s all of which are either on or close to market.

Electronic Nose

Figure 12. Schematic of an SPME fiber. To take a sample, the fiber is extended through the needle and exposed to the target analytes. After the volatiles have reached equilibrium between the fiber coating and the gaseous phase (or after a strongly defined time), the fiber is withdrawn into the needle. Desorption will take place in a heated inlet under a similar procedure. Reprinted with permission from ref 211. Copyright 2006 Elsevier.

5.1. Sample Handling Sample preparation and sampling are error-prone steps and have to be well considered to achieve reliable results. This begins with a representative selection of samples, continues with their appropriate pretreatment, includes possible preconcentration and separation steps, and ends with a reproducible sample delivery procedure to the sensor array. Each of these steps can cause statistical as well as systematic errors, but besides these possible sources of error, the sample preparation opens up additional opportunities: It has the potential to dramatically increase the sensitivity of the whole system and, in addition, to remove the problems caused by background interferences. Because the original electronic noses concept was to move on from sophisticated analytical instruments and to create a simple and straightforward device, sophisticated sampling procedures were omitted. However, the need to solve ever-demanding applications has brought sample preparation techniques more and more in the focus in the past few years. The fact that samples can be solid, liquid, or gaseous and that their nature differs a lot makes it difficult to give a complete overview of the strategies used. For instance, aqueous samples can be stirred, heated, or salted out, or the pH can be varied to increase the concentration of volatiles in the headspace.207 To make the system even more sensitive and not solely dependent on the direct vapor partitioning, a preconcentration step is inevitable.208 The enrichment of the analytes can be divided into two major categories: active and passive air sampling.203 In active sampling the gaseous sample is drawn through an adsorbent material. To measure the flow rate and the total volume, a flow meter is necessary for this approach. The advantage is that for a given sampling time lower concentrations can be monitored. In contrast, passive sampling is much simpler in implementation, and when a sample is taken, there is no need for additional technical equipment (Figure 12).209 In this case, the analytes follow the concentration gradient according to Fick’s first law to the sorbent. Therefore, the only driving forces are diffusion and the partition coefficient between the two phases. Each method has further advantages and disadvantages, and choosing one of them depends on the particular application.210 In combination with an electronic nose different preconcentration methods have been compared for some specific examples. Schaller et al. analyzed the ripening grades of

Chemical Reviews, 2008, Vol. 108, No. 2 719

Swiss Emmental cheese with the help of a mass-spectrometerbased electronic nose (SMart Nose).212 The extraction methods used are static headspace extraction, purge-and-trap extraction with a mixture of Carbosieve SIII and Carbopack B60/80 as adsorbent materials, and solid-phase microextraction (SPME)211 with a 65 µm CW/DVD-coated fiber. The authors conclude that the static headspace measurement is useful for high levels of volatile compounds for which the two preconcentration methods do not bring an increase of sensitivity. However, both techniques extract approximately the same class of compounds with a higher mass-to-charge ratio. Because of better repeatability, usability, and concentrating ability, in direct comparison they favor the SPME technique to trap middle to high molecular masses. Ampuero et al. confirmed this finding for the classification of the botanical origin of unifloral honeys with the same electronic nose.213 In this study static headspace measurement and solid-phase microextraction were performed under similar conditions. Instead of the classical purge-and-trap technique with continuous gas flow, they used inside-needle dynamic extraction (INDEx) as the active sampling procedure. Compared to SPME, this method has a higher mechanical robustness, needs half of the analysis time, and is simple.213,214 However, SPME showed clearly a better extraction capacity for heavier volatiles with an m/z > 110. One has to note that the benefit of using preconcentration methods for sensor-based electronic noses is often not apparent from the sensor signal itself but becomes visible after data evaluation. Examples are the identification of lampante virgin olive oils,215 the differentiation between apple varieties, the identification of the ripeness of pineapples, and the detection of an off-flavor in sugar with an SPME-SAW sensor array.216 On the basis of a tin dioxide multisensor, Lozano et al. tested the ability of different SPME fiber coatings for wine discrimination.217 Particularly for quantification tasks the influence of the coating thickness has to be considered as even low variations have a strong influence on the analyte response.218,219 Therefore, a lack of interfiber comparability depending on the production process used can adulterate the results.

5.2. Filters and Analyte Gas Separation The comparison between different extraction techniques has already shown that depending on the chosen approach the ratio of the detected compounds changes. This gives the potential to increase not only the sensitivity but also the selectivity to the target compounds of the system by a deliberate choice of sampling conditions. The obvious way is to adapt the polymer coating of the SPME fiber (Figure 13) or the Gerstel Twister, used for stir bar sorptive extraction (SBSE),203,220 or to use an appropriate filling for the adsorbent tubes. In addition, ingenious solutions can be found in the literature for the requirements of specific applications. The following examples show possible approaches and demonstrate that there are no clear boundaries between the separation techniques with or without simultaneous sample preconcentration. (1) Villanueva et al. discriminated red wines, differing only in the variety of grape, by a system based on SPME and a metal oxide sensor array.222 In a two-step desorption process, they first “dried” the polar absorbent fiber at low temperatures to eliminate the influence of water and ethanol. (2) Instead of taking discrete temperature steps, Morris et al. desorbed the volatiles from a Tenax TA bed using a

720 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 13. Chromatogram obtained from a single rose petal.203 Sample preparation was carried out using SPME fibers with different coatings. In the lower trace PDMS (poly(dimethylsiloxane)) was used as the polymer, and in the upper trace PA (polyacrylate) was used. The choice of the fiber coating determines the composition of the detected substances. In this example the ratio between substance 4 and substance 11 is dramatically different. Reprinted with permission from ref 221. Copyright 2000 WileyVCH.

program.223

temperature The bed had previously been exposed to the headspace of groundwater and to urban air. The temperature profile over time means that water is eluted first separately from the interesting volatiles. Instead of obtaining a steady-state sensor signal, a complex spectrum is created that contains information about the boiling point of the particular substances (elution time) and the functional type (peak width). A similar approach was previously used by Strathmann et al.224 (3) Ali et al. used a heated preconcentration tube as a dispersive element for a QCM array.225 Water interferences were eliminated by using the different breakthrough times of water and toluene, the target substance. (4) Investigating off-flavor detection in wine, RagzzoSanchez et al. proposed back-flush gas chromatography to remove water and ethanol from the other volatiles.198 Offflavor-doped wines were discriminated by using FOX 4000 electronic nose data. (5) The group at the University of Tu¨bingen characterized packaging emissions with the help of four metal oxide gas sensors connected to a chromatographic column. For this purpose a very simple packed column was sufficient to separate water from the residual solvents and to determine the total amount of solvent in paper and paperboard in a reliable way (Figure 14).206 (6) The hardware of the zNose is a complete gas chromatograph with an SAW sensor as the detector. A similar approach was used by Zampolli et al. with a micromachined gas chromatographic column connected to a solid-state gas sensor (Figure 5).116 In this case the use of a single sensor means that the convential 2D data evaluation approaches can be used. (7) A further possibility to enhance selectivity was demonstrated using mass transport phenomena across a membrane.226 Organophilic pervaporation can be used to discriminate wine model solutions in the presence of ethanol.

Ro¨ck et al.

Figure 14. Determination of residual solvents from paper and paperboard packaging in the food industry.206 Influences on the results of the variations of the high-humidity concentration have to be excluded. In the graph the separation of the organic solvents (first peak) from water (second peak) by means of a simple chromatographic approach is shown. Within the first minute the peak height of the residual solvents, consisting of ethanol, 2-propanol, 2-butanol, cyclohexanone, 1-ethoxy-2-propanol, and trace components, can be evaluated. Reprinted from ref 206. Copyright 2005 American Chemical Society.

After this pretreatment both conducting polymer-based sensors227 and metal oxide sensors228 are able to overcome the ethanol interference. The examples presented demonstrate different strategies to eliminate interferences and enhance the electronic nose as a whole system. In contrast to sensor-based improvements of the selectivity, they all have the disadvantage of an increase in setup complexity and in analysis time, but the crucial point is that, in contrast to highly selective sensors, reversibility is a feature of most of these approaches. This has practical implications: when the system is being trained on calibration sets, these approaches do not suffer from instrumental drift as in the case of high-selectivity sensors. The stability of the system is preserved, and there is no need for drift correction in the subsequent data analysis. A direct comparison of the improvements and the additional costs brought by the different sampling strategies is difficult. Each application has its own requirements, and the sample preparation cannot be considered in isolation. In the examples shown, the information obtained often increases at the expense of additional time dependency. Therefore, an adapted data evaluation strategy is necessary to maximize the benefit gained.

5.3. Data Evaluation Dodd and Persaud used the ratio of the steady-state sensor responses for data evaluation 25 years ago,1 whereas in current research the data obtained are often so complex that they cannot be manually evaluated. Furthermore, data evaluation is not limited only to pattern recognition; it begins with the data acquisition step.229 This includes the choice of the appropriate sensors, feature selection, scaling, and normalization. Finally, pattern recognition and classification techniques can be model free or model based and supervised or unsupervised. Each of these functions can be performed by a variety of different approaches which are more or less suitable for a specific application. Unfortunately, no general

Electronic Nose

Figure 15. DFA of randomly generated data for a theoretical 24sensor array. A total of 30 data points with a relative standard deviation of 7% were arranged into 3 groups of 10 data points. DFA discriminated them with a confidence interval (shown ellipses) of 95%. Reprinted with permission from ref 230. Copyright 2001 Elsevier.

guidelines to determine the appropriate strategy exist. For this reason, in several publications these factors are a product of chance or, if they were done more systematically, a product of trial and error. In the latter case, however, the danger of overfitting and therefore false classification is high for operators lacking a deeper understanding of this field, as Goodner was able to demonstrate (see Figure 15).230 Additionally, the lack of knowledge on which substances may be encountered hinders an adequate selection of the sensors and the training of the array to each possible analyte. An overview of the analysis of data is given in the review of Scott et al.231 Because of the need to have real experimental data, current research in this field is in most cases specific to the application and the electronic nose used. Therefore, there is a need to compare existing pattern recognition processes on the same data set,232 to adapt and improve existing algorithms,233-235 and to transfer data evaluation methods from other research areas.236-238 The latter is especially important for the new types of electronic nose setups which produce additional time-dependent information.236,237 However, in handling large amounts of data, it is important to consider redundancy. As these new techniques increase the dimensions of the data set the number of theoretical features becomes large, and hence, selection of the right features becomes challenging.20 For electronic noses based on a sensor array these are principally transient sensor response,239-241 temperature modulation of metal oxide sensors,242-245 partial preseparation of the compounds,206,223,225,246 or slight differences in the sensors caused by a gradient over temperature, doping concentration, sensitive layer thickness, or membrane thickness (compare Figure 16).247,248 Modern approaches may also have high-dimensional output data as well, for example, the mass-spectrometerbased Smart Nose with its high amount of mass-to-charge ratios, IMS with the time-dependent measurement,120 or highdensity optical sensor arrays.249,250 However, for any given training set there exists an optimum number of features. In case it is too high, overfitting or computational ill-conditioning will take place and generalization will fail with the consequence of poor validation performance.251 Therefore, a lot of work has been carried out recently to select the best

Chemical Reviews, 2008, Vol. 108, No. 2 721

Figure 16. Measurement results of the gas sensor microarray KAMINA.248 The array consists of a single monolithic metal oxide film separated into 38 segments by a parallel electrode structure. A temperature gradient (a) or a membrane thickness gradient (b) slightly changes the selectivity from segment to segment. The change of conductivity was normalized to the median (inner circle - 100), and the results are depicted as polar plots. As can be seen for both temperature (a) and membrane (b) gradients benzene and propane are difficult to discriminate even without considering the standard deviation. However, propanol can be readily distinguished by comparison of the first and last segments. Reprinted with permission from ref 248. Copyright 2001 Elsevier.

features251-256 or even the most appropriate sensors.257,258 This is progress in the direction of having solid features and consequently reliable results from data evaluation instead of fitting the noise.20

6. Conclusion Since the first attempts to identify a small number of single volatiles with the help of a set of unspecific gas sensors, much work has been carried out within the field of electronic noses. Today it is not only metal oxide sensors of varying selectivities which are available for this task, but also other transducers with electrochemical readouts such as conducting polymers, metal oxide field effect transistors, or amperometric sensors. Furthermore, gravimetric, thermal, and optical sensors which have a completely different transduction principle are also in use. On the basis of this variety of sensors, the electronic nose has proven that it is appropriate for a limited number of well-selected and -characterized applications. It is possible to classify bacteria, to monitor air quality on the space shuttle,259 or to check the spoilage of foodstuff, to mention only a few successful examples. Despite the success in some areas, the efforts to arrive at a universal device that can make fine discrimination of flavors, perfumes, and smells and eventually replace the human nose are disappointing. The initial hope was to approach the ability of human odor sensing by increasing the number of individual sensors. However, the reason for the nose’s unequalled performance has turned out to be not only the high number of different human receptor cells, but their selectivity and their unsurpassed sensitivity for some analyte gases. Therefore, instead of creating redundant information by adding more similar sensors, current research efforts are targeting both these directions. Sensors with new sensitive layers are under development, for instance, based on DNA, molecular imprinted molecules, or even immobilized natural receptors (up to whole cells), which

722 Chemical Reviews, 2008, Vol. 108, No. 2

promise to increase the sensitivity and importantly selectivity.260-263 Moreover, considering the electronic nose as a whole system, there are other possibilities to reach both of these aims. On one hand, the increase in sensitivity can be realized by appropriate sample pretreatment and preconcentration techniques, whereras filters and separation units can be used to increase the selectivity and reduce interfering substances. These strategies are a further step in the evolution of the electronic nose by learning from nature and which should lead to an enlarged field of application areas. Going in this direction, the complexity of the whole system will be obviously increased, but learning from history this step is often inevitable to apply the electronic nose in the desired way. In spite of this divergence from the intended simplicity, the products obtained are still by far less expensive than analytical systems and have the potential for cost-engineering when adapted to one special task. In addition to the classical sensor-array-based approach, electronic noses based on other technologies have become more and more common where, for example, mass and ion mobility spectrometers or flash gas chromatographs are used to detect the components of a gas mixture. Instead of the features given by a sensor array, in these cases, the detector arrays have a virtual character and the multiple features are provided by their specific m/z ratio, their time-of-flight, or their retention time. In spite of having another approach and thus providing a quite different inputsa well-defined concentration profilesthey are as equally unsuccessful in mimicking the sense of smell as their sensor-array counterparts. Neither the sensor-array approach nor instrumental analysis is by definition better. Their suitability for a specific application depends critically on the operating conditions and target species and should be considered on a case by case basis. Without a proper consideration of the problem there is a high risk of obtaining chemical fingerprints without a correlation with the relevant properties of the sample. The electronic nose, in use today, replaces neither complex analytical equipment nor odor panels but supplements both of them. In comparison it might have several advantages regarding mobility, price (TCO), and ease of use. Therefore, it has the potential to enter our daily life far away from wellequipped chemical laboratories and skilled specialists. Keeping its limitations in mind and adapted for a special purpose, this will be the future for the electronic nose for as long as the ability to smelling odors rather than detecting volatiles is still far away over the rainbow.

7. References (1) Persaud, K. C.; Dodd, G. Nature 1982, 299, 352. (2) Walt, D. R. Anal. Chem. 2005, 77, 45a. (3) Halova, J.; Strouf, O.; Zak, P.; Sochozova, A.; Uchida, N.; Yuzuri, T.; Sakakibara, K.; Hirota, M. Quant. Struct.-Act. Relat. 1998, 17, 37. (4) Autrum, H.; Zwehl, V. Z. Vgl. Physiol. 1964, 48, 357. (5) Singh, S. J. Hazard. Mater. 2007, 144, 15. (6) Bunte, G.; Hurttlen, J.; Pontius, H.; Hartlieb, K.; Krause, H. Anal. Chim. Acta 2007, 591, 49. (7) Firestein, S. Nature 2001, 413, 211. (8) Buck, L. B. Angew. Chem. 2005, 117, 6283. (9) Mombaerts, P. Nat. ReV. Neurosci. 2004, 5, 263. (10) Boeker, P.; Hamacher, T.; Mannebeck, D.; Wimmer, P.; Horner, G. GefahrstoffesReinhalt. Luft 2003, 63, 283. (11) Mitrovics, J.; Ulmer, H.; Weimar, U.; Gopel, W. Acc. Chem. Res. 1998, 31, 307. (12) Gaillard, I.; Rouquier, S.; Giorgi, D. Cell. Mol. Life Sci. 2004, 61, 456. (13) James, D.; Scott, S. M.; Ali, Z.; O’Hare, W. T. Microchim. Acta 2005, 149, 1.

Ro¨ck et al. (14) Ulmer, H.; Mitrovics, J.; Noetzel, G.; Weimar, U.; Go¨pel, W. Sens. Actuators, B 1997, 43, 24. (15) Pardo, M.; Kwong, L. G.; Sberveglieri, G.; Brubaker, K.; Schneider, J. F.; Penrose, W. R.; Stetter, J. R. Sens. Actuators, B 2005, 106, 136. (16) Dickinson, T. A.; White, J.; Kauer, J. S.; Walt, D. R. Nature 1996, 382, 697. (17) Suslick, K. S.; Rakow, N. A.; Sen, A. Tetrahedron 2004, 60, 11133. (18) Suslick, K. S. MRS Bull. 2004, 29, 720. (19) Filippini, D.; Lundstrom, I. Appl. Phys. Lett. 2002, 81, 3891. (20) Hajek, P.; Havranek, T. Mechanizing hypothesis formation mathematical foundations for general theory; Springer-Verlag: Berlin, Heidelberg, New York, 1978. (21) Creaser, C. S.; Griffiths, J. R.; Bramwell, C. J.; Noreen, S.; Hill, C. A.; Thomas, C. L. P. Analyst 2004, 129, 984. (22) Massart, D. L.; Vandeginste, B. G. M.; Deming, S. N.; Michotte, Y.; Kaufman, L. Chemometrics: A Textbook; Elsevier: Amsterdam, 1988. (23) Massart, D. L.; Vandeginste, B. G. M.; Buydens, L. M. C.; De Jong, S.; Lewi, P. J.; Smeyers-Verbeke, J. Handbook of Chemometrics and Qualimetrics; Elsevier: Amsterdam, 1998. (24) Mielle, P.; Marquis, F. Semin. Food Anal. 1998, 3, 93. (25) Staples, E. J. Abstr. Pap.sAm. Chem. Soc. 2000, 219, 236. (26) Zampolli, S.; Elmi, I.; Sturmann, J.; Nicoletti, S.; Dori, L.; Cardinali, C. Sens. Actuators, B 2005, 105, 400. (27) Lu, C. J.; Steinecker, W. H.; Tian, W. C.; Oborny, M. C.; Nichols, J. M.; Agah, M.; Potkay, J. A.; Chan, H. K. L.; Driscoll, J.; Sacks, R. D.; Wise, K. D.; Pang, S. W.; Zellers, E. T. Lab Chip 2005, 5, 1123. (28) Kim, H.; Steinecker, W. H.; Reidy, S.; Lambertus, G. R.; Astle, A. A.; Najafi, K.; Zellers, E. T.; Bernal, L. P.; Washabaugh, P. D.; Wise, K. D. TRANSDUCERS; IEEE: Lyon, France, 2007; p 1505. (29) van Kempen, T. A. T. G.; Powers, W. J.; Sutton, A. L. J. Anim. Sci. 2002, 80, 1524. (30) Cozzolino, D.; Smyth, H. E.; Gishen, M. J. Agric. Food Chem. 2003, 51, 7703. (31) Armenta, S.; Coelho, N. M. M.; Roda, R.; Garrigues, S.; de la Guardia, A. Anal. Chim. Acta 2006, 580, 216. (32) Ampuero, S.; Bosset, J. O. Sens. Actuators, B 2003, 94, 1. (33) Strike, D. J.; Meijerink, M. G. H.; Koudelka-Hep, M. Fresenius’ J. Anal. Chem. 1999, 364, 499. (34) Gardner, J. W.; Bartlett, P. N. Sens. Actuators, B 1994, 18-19, 211. (35) Casalinuovo, I. A.; Di Pierro, D.; Coletta, M.; Di Francesco, P. IEEE Sens. J. 2006, 6, 1428. (36) Capone, S.; Distante, C.; Francioso, L.; Presicce, D.; Taurino, A. M.; Siciliano, P.; Zuppa, M. J. Argent. Chem. Soc. 2005, 93, 123. (37) Lacoste, F.; Bosque, F.; Raoux, R. Ocl-Ol. Corps Gras Li. 2001, 8, 78. (38) Maciejewska, A.; Szczurek, A.; Kerenyi, Z. Sens. Actuators, B 2006, 115, 170. (39) Ragazzo-Sanchez, J. A.; Chalier, P.; Chevalier, D.; Ghommidh, C. Sens. Actuators, B 2006, 114, 665. (40) Deisingh, A. K.; Stone, D. C.; Thompson, M. Int. J. Food Sci. Technol. 2004, 39, 587. (41) Liddell, K.; White, J. Br. J. Dermatol. 1975, 92, 215. (42) Doty, R. L. Chem. Senses 1981, 6, 351. (43) Fitzgerald, F. T.; Tierney, L. M. West. J. Med. 1982, 137, 169. (44) Manolis, A. Clin. Chem. 1983, 29, 5. (45) Persaud, K. C. Int. J. Low. Extrem. Wounds 2005, 4, 50. (46) Pavlou, A. K.; Turner, A. P. F. Clin. Chem. Lab. Med. 2000, 38, 99. (47) Turner, A. P. F.; Magan, N. Nat. ReV. Microbiol. 2004, 2, 161. (48) Thaler, E. R.; Hanson, C. W. Expert ReV. Med. DeVices 2005, 2, 559. (49) Di Natale, C.; Pennazza, G.; Martinelli, E.; Santonico, M.; Macagnano, A.; Paolesse, R.; D’Amico, A. J. Argent. Chem. Soc. 2005, 93. (50) Laska, M.; Seibt, A.; Weber, A. Chem. Senses 2000, 25, 47. (51) Lorenzo, N.; Wan, T. L.; Harper, R. J.; Hsu, Y. L.; Chow, M.; Rose, S.; Furton, K. G. Anal. Bioanal. Chem. 2003, 376, 1212. (52) Go¨th, A.; McLean, I. G.; Trevelyan, J. In Mine Detection Dogs: Training, Operations and Odour Detection; MacDonald, J., Lockwood, J. R., Eds.; GICHD: Geneva, 2003. (53) Shaw, J. A.; Seldomridge, N. L.; Dunkle, D. L.; Nugent, P. W.; Spangler, L. H.; Bromenshenk, J. J.; Henderson, C. B.; Churnside, J. H.; Wilson, J. J. Opt. Express 2005, 13, 5853. (54) Cantergiani, E.; Brevard, H.; Krebs, Y.; Feria-Morales, A.; Amado, R.; Yeretzian, C. Eur. Food Res. Technol. 2001, 212, 648. (55) Song, H. S.; Sawamura, M.; Ito, T.; Ido, A.; Ukeda, H. FlaVour Fragrance J. 2000, 15, 323. (56) Lin, J. M.; Rouseff, R. L. FlaVour Fragrance J. 2001, 16, 457. (57) Ulrich, D.; Hoberg, E.; Bittner, T.; Engewald, W.; Meilchen, K. Eur. Food Res. Technol. 2001, 213, 200.

Electronic Nose (58) Valim, M. F.; Rouseff, R. L.; Lin, J. M. J. Agric. Food Chem. 2003, 51, 1010. (59) Gocmen, D.; Gurbuz, O.; Rouseff, R. L.; Smoot, J. M.; Dagdelen, A. F. Eur. Food Res. Technol. 2004, 218, 573. (60) Komes, D.; Ulrich, D.; Lovric, T. Eur. Food Res. Technol. 2006, 222, 1. (61) Barsan, N.; Ionescu, R. Sens. Actuators, B 1994, 19, 470. (62) Hivert, B.; Hoummady, M.; Mielle, P.; Mauvais, G.; Henrioud, J. M.; Hauden, D. Sens. Actuators, B 1995, 27, 242. (63) Mielle, P.; Marquis, F.; Trivier, V. In Sensors and Microsystems; Di Natale, C., D’Amico, A., Siciliano, P., Eds.; World Scientific: London, 2000. (64) Hofmann, T.; Schieberle, P.; Krummel, C.; Freiling, A.; Bock, J.; Heinert, L.; Kohl, D. Sens. Actuators, B 1997, 41, 81. (65) Eberheim, A.; Kohl, D.; Schieberle, P. Phys. Chem. Chem. Phys. 2003, 5, 5203. (66) Dainty, R. H. Int. J. Food Microbiol. 1996, 33, 19. (67) Cynkar, W.; Cozzolino, D.; Dambergs, B.; Janik, L.; Gishen, M. Sens. Actuators, B 2007, 124, 167. (68) Karlshoj, K.; Nielsen, P. V.; Larsen, T. O. J. Agric. Food Chem. 2007, 55, 4289. (69) Gomez, A. H.; Wang, J.; Hu, G. X.; Pereira, A. G. Lebensm.-Wiss. -Technol. 2007, 40, 681. (70) Marin, S.; Vinaixa, M.; Brezmes, J.; Llobet, E.; Vilanova, X.; Correig, X.; Ramos, A. J.; Sanchis, V. Int. J. Food Microbiol. 2007, 114, 10. (71) Needham, R.; Williams, J.; Beales, N.; Voysey, P.; Magan, N. Sens. Actuators, B 2005, 106, 20. (72) Vinaixa, M.; Marin, S.; Brezmes, J.; Llobet, E.; Vilanova, X.; Correig, X.; Ramos, A.; Sanchis, V. J. Agric. Food Chem. 2004, 52, 6068. (73) Needham, R.; Magan, N. Aspects Appl. Biol. 2003, 68, 217. (74) Keshri, G.; Voysey, P.; Magan, N. J. Appl. Microbiol. 2002, 92, 165. (75) Presicce, D. S.; Forleo, A.; Taurino, A. M.; Zuppa, M.; Siciliano, P.; Laddomada, B.; Logrieco, A.; Visconti, A. Sens. Actuators, B 2006, 118, 433. (76) Tognon, G.; Campagnoli, A.; Pinotti, L.; Dell’Orto, V.; Cheli, F. Vet. Res. Commun. 2005, 29, 391. (77) Costello, B. P. J. D.; Ewen, R. J.; Gunson, H.; Ratcliffe, N. M.; Sivanand, P. S.; Spencer-Phillips, P. T. N. Meas. Sci. Technol. 2003, 14, 397. (78) Benedetti, S.; Sinelli, N.; Buratti, S.; Riva, M. J. Dairy Sci. 2005, 88, 3044. (79) Panigrahi, S.; Balasubramanian, S.; Gu, H.; Logue, C. M.; Marchello, M. Sens. Actuators, B 2006, 119, 2. (80) Blixt, Y.; Borch, E. Int. J. Food Microbiol. 1999, 46, 123. (81) Rajamaki, T.; Alakomi, H. L.; Ritvanen, T.; Skytta, E. S.; Smolander, M.; Ahvenainen, R. Food Control 2006, 17, 5. (82) Haugen, J. E.; Rudi, K.; Langsrud, S.; Bredholt, S. Anal. Chim. Acta 2006, 565, 10. (83) Magan, N.; Pavlou, A.; Chrysanthakis, I. Sens. Actuators, B 2001, 72, 28. (84) Abellana, M.; Benedi, J.; Sanchis, V.; Ramos, A. J. J. Appl. Microbiol. 1999, 87, 371. (85) Delbarre-Ladrat, C.; Cheret, R.; Taylor, R.; Verrez-Bagnis, V. Crit. ReV. Food Sci. Nutr. 2006, 46, 409. (86) Gram, L.; Huss, H. H. Int. J. Food Microbiol. 1996, 33, 121. (87) Gram, L.; Dalgaard, P. Curr. Opin. Biotechnol. 2002, 13, 262. (88) Joffraud, J. J.; Leroi, F.; Roy, C.; Berdague, J. L. Int. J. Food Microbiol. 2001, 66, 175. (89) Malle, P.; Poumeyrol, M. J. Food Prot. 1989, 52, 419. (90) Duflos, G.; Coin, V. M.; Cornu, M.; Antinelli, J. F.; Malle, P. J. Sci. Food Agric. 2006, 86, 600. (91) Rodriguez-Jerez, J. J.; M. M.; H.-H.; Roig-Sagues, A. X. Global quality assessment in Mediterranean aquaculture; CIHEAM: Zaragoza, Spain, 2000. (92) Ruiz-Capillas, C.; Gillyon, C. M.; Horner, W. F. A. J. Food Biochem. 2001, 25, 541. (93) Jonsdottir, R.; Olafsdottir, G.; Martinsdottir, E.; Stefansson, G. J. Agric. Food Chem. 2004, 52, 6250. (94) Mansur, M. A.; Bhadra, A.; Takamura, H.; Matoba, T. Fish. Sci. 2003, 69, 864. (95) Di Natale, C.; Olafsdottir, G.; Einarsson, S.; Martinelli, E.; Paolesse, R.; D’Amico, A. Sens. Actuators, B 2001, 77, 572. (96) Naes, T.; Isaksson, T.; Fearn, T.; Davies, T. A user-friendly guide to multiVariate calibration and classification; NIR Publications: Chichester, U.K., 2002. (97) Haddad, R.; Carmel, L.; Harel, D. Sens. Actuators, B 2007, 120, 467. (98) Huss, H. H. Quality and quality changes in fresh fish; FAO: Rome, 1995. (99) Shawyer, M.; Pizzali, A. F. M. The use of ice on small fishing Vessels; FAO: Rome, 2003. (100) Olafsdottir, G.; Jonsdottir, R.; Lauzon, H. L.; Luten, J.; Kristbergsson, K. J. Agric. Food Chem. 2005, 53, 10140.

Chemical Reviews, 2008, Vol. 108, No. 2 723 (101) Olafsdottir, G. In Rapid and on-line instrumentation for food quality assurance; Tothill, I. E., Ed.; Woodhead Publishing: Cambridge, U.K., 2003. (102) Gilbert, R. O. Statistical Methods for EnVironmental Pollution Monitoring; John Wiley & Sons: New York, 1987. (103) Wight, G. D. Fundamentals of air sampling; CRC: Boca Raton, FL, 1994. (104) Nimmermark, S. Water Sci. Technol. 2001, 44, 33. (105) Di Francesco, F.; Lazzerini, B.; Marcelloni, F.; Pioggia, G. Atmos. EnViron. 2001, 35, 1225. (106) Micone, P. G.; Guy, C. Sens. Actuators, B 2007, 120, 628. (107) Biey, E. M.; Verstraete, W. EnViron. Technol. 1999, 20, 331. (108) Zhou, H. Y.; Homer, M. L.; Shevade, A. V.; Ryan, M. A. IEEE Sens. J. 2006, 6, 1. (109) Dutta, R.; Morgan, D.; Baker, N.; Gardner, J. W.; Hines, E. L. Sens. Actuators, B 2005, 109, 355. (110) Delpha, C.; Lumbreras, M.; Siadat, M. Sens. Actuators, B 2004, 98, 46. (111) Feldhoff, R.; Saby, C. A.; Bernadet, P. Analyst 1999, 124, 1167. (112) Sobanski, T.; Szczurek, A.; Nitsch, K.; Licznerski, B. W.; Radwan, W. Sens. Actuators, B 2006, 116, 207. (113) Kohl, D.; Eberheim, A.; Schieberle, P. Thin Solid Films 2005, 490, 1. (114) Kalman, E. L.; Winquist, F.; Lofvendahl, A.; Rudell, B.; Wass, U. Indoor Built EnViron. 2002, 11, 105. (115) Blaschke, M.; Tille, T.; Robertson, P.; Mair, S.; Weimar, U.; Ulmer, H. IEEE Sens. J. 2006, 6, 1298. (116) Zampolli, S.; Elmi, I.; Ahmed, F.; Passini, M.; Cardinali, G. C.; Nicoletti, S.; Dori, L. Sens. Actuators, B 2004, 101, 39. (117) Kuske, M.; Romain, A. C.; Nicolas, J. Build. EnViron. 2005, 40, 824. (118) Ryan, M. A.; Shevade, A. V.; Zhou, H.; Homer, M. L. MRS Bull. 2004, 29, 714. (119) Willers, H.; de Gijsel, P.; Ogink, N.; D’Amico, A.; Martinelli, E.; Di Natale, C.; van Ras, N.; van der Waarde, J. Water Sci. Technol. 2004, 50, 93. (120) Tuovinen, K.; Kolehmainen, M.; Paakkanen, H. Anal. Chim. Acta 2001, 429, 257. (121) Bastos, A. C.; Magan, N. Sens. Actuators, B 2006, 116, 151. (122) Goschnick, J.; Koronczi, I.; Frietsch, M.; Kiselev, I. Sens. Actuators, B 2005, 106, 182. (123) Stuetz, R. M. Water Sci. Technol. 2004, 49, 137. (124) Bourgeois, W.; Gardey, G.; Servieres, M.; Stuetz, R. M. Sens. Actuators, B 2003, 91, 109. (125) Nicolas, J.; Romain, A. C.; Ledent, C. Sens. Actuators, B 2006, 116, 95. (126) Dickert, F. L.; Lieberzeit, P. A.; Achatz, P.; Palfinger, C.; Fassnauer, M.; Schmid, E.; Werther, W.; Horner, G. Analyst 2004, 129, 432. (127) Pavlou, A.; Turner, A. P. F.; Magan, N. Lett. Appl. Microbiol. 2002, 35, 366. (128) Parry, A. D.; Chadwick, P. R.; Simon, D.; Oppenheim, B.; McCollum, C. N. J. Wound Care 1995, 4, 404. (129) Chandiok, S.; Crawley, B. A.; Oppenheim, B. A.; Chadwick, P. R.; Higgins, S.; Persaud, K. C. J. Clin. Pathol. 1997, 50, 790. (130) Hay, P.; Tummon, A.; Ogunfile, M.; Adebiyi, A.; Adefowora, A. Int. J. Stand. Aids 2003, 14, 114. (131) Chaudry, A. N.; Travers, P. J.; Yuenger, J.; Colletta, L.; Evans, P.; Zenilman, J. M.; Tummon, A. J. Clin. Microbiol. 2004, 42, 5170. (132) Lai, S. Y.; Deffenderfer, O. F.; Hanson, W.; Phillips, M. P.; Thaler, E. R. Laryngoscope 2002, 112, 975. (133) Shykhon, M. E.; Morgan, D. W.; Dutta, R.; Hines, E. L.; Gardner, J. W. J. Laryngol. Otol. 2004, 118, 706. (134) Pavlou, A. K.; Magan, N.; Jones, J. M.; Brown, J.; Klatser, P.; Turner, A. R. Biosens. Bioelectron. 2004, 20, 538. (135) Fend, R.; Kolk, A. H. J.; Bessant, C.; Buijtels, P.; Klatser, P. R.; Woodman, A. C. J. Clin. Microbiol. 2006, 44, 2039. (136) Fend, R.; Geddes, R.; Lesellier, S.; Vordermeier, H. M.; Corner, L. A. L.; Gormley, E.; Costello, E.; Hewinson, R. G.; Marlin, D. J.; Woodman, A. C.; Chambers, M. A. J. Clin. Microbiol. 2005, 43, 1745. (137) Aathithan, S.; Plant, J. C.; Chaudry, A. N.; French, G. L. J. Clin. Microbiol. 2001, 39, 2590. (138) Pavlou, A. K.; Magan, N.; McNulty, C.; Jones, J. M.; Sharp, D.; Brown, J.; Turner, A. P. F. Biosens. Bioelectron. 2002, 17, 893. (139) Kodogiannis, V.; Wadge, E. Int. J. Neural Syst. 2005, 15, 363. (140) Mohamed, E. I.; Linder, R.; Periello, G.; Di Daniele, N.; Po¨ppl, S. J.; De Lorenzo, A. Diabetes, Nutr. Metab. 2002, 15, 215. (141) Voss, A.; Baier, V.; Reisch, R.; von Roda, K.; Elsner, P.; Ahlers, H.; Stein, G. Ann. Biomed. Eng. 2005, 33. (142) Fend, R.; Bessant, C.; Williams, A. J.; Woodman, A. C. Biosens. Bioelectron. 2004, 19, 1581. (143) Podebrad, F.; Heil, M.; Reichert, S.; Mosandl, A.; Sewell, A. C.; Bohles, H. J. Inherited Metab. Dis. 1999, 22, 107.

724 Chemical Reviews, 2008, Vol. 108, No. 2 (144) Cao, W. Q.; Duan, Y. X. Crit. ReV. Anal. Chem. 2007, 37, 3. (145) Miekisch, W.; Schubert, J. K.; Noeldge-Schomburg, G. F. E. Clin. Chim. Acta 2004, 347, 25. (146) Jones, A. W. Acta Physiol. Scand. 1982, 114, 407. (147) Schoknecht, G.; Fleck, K.; Kophamel, B. Blutalkohol 1990, 27, 83. (148) Paulsson, N. J. P.; Winquist, F. Forensic Sci. Int. 1999, 105, 95. (149) Ping, W.; Yi, T.; Haibao, X.; Farong, S. Biosens. Bioelectron. 1997, 12, 1031. (150) Yu, J. B.; Byun, H. G.; So, M. S.; Huh, J. S. Sens. Actuators, B 2005, 108, 305. (151) Bates, C. A.; Silkoff, P. E. J. Allergy Clin. Immunol. 2003, 111, 256. (152) Smith, A. D.; Cowan, J. O.; Filsell, S.; McLachlan, C.; MontiSheehan, G.; Jackson, P.; Taylor, D. R. Am. J. Respir. Crit. Care Med. 2004, 169, 473. (153) Smith, A. D.; Cowan, J. O.; Brassett, K. P.; Herbison, G. P.; Taylor, D. R. N. Engl. J. Med. 2005, 352, 2163. (154) Dragonieri, S.; Le Cessie, S.; Mertens, B. J. A.; Schot, R.; Assink, C. J.; Gauw, S. A.; Bel, E. H.; Sterk, P. J. Proceedings of the American Thoracic Society, San Francisco, 2007. (155) Alving, K.; Janson, C.; Nordvall, L. Respir. Res. 2006, 7. (156) Gill, M.; Graff, G. R.; Adler, A. J.; Dweik, R. A. J. Asthma 2006, 43, 731. (157) Lin, Y. J.; Guo, H. R.; Chang, Y. H.; Kao, M. T.; Wang, H. H.; Hong, R. I. Sens. Actuators, B 2001, 76, 177. (158) Mohamed, E. I.; Bruno, E.; Linder, R.; Alessandrin, M.; Di Girolamo, A.; Po¨ppl, S. J.; Puija, A.; De Lorenzo, A. An. Otorrinolaringol. Ibero. Am. 2003, 30, 447. (159) Hockstein, N. G.; Thaler, E. R.; Torigian, D.; Miller, W. T.; Deffenderfer, O.; Hanson, C. W. Laryngoscope 2004, 114, 1701. (160) Hockstein, N. G.; Thaler, E. R.; Lin, Y. Q.; Lee, D. D.; Hanson, C. W. Ann. Otol., Rhinol., Laryngol. 2005, 114, 504. (161) Hanson, C. W.; Thaler, E. R. Anesthesiology 2005, 102, 63. (162) Suarez, F. L.; Furne, J. K.; Springfield, J.; Levitt, M. D. J. Dent. Res. 2000, 79, 1773. (163) Tanaka, M.; Anguri, H.; Nonaka, A.; Kataoka, K.; Nagata, H.; Kita, J.; Shizukuishi, S. J. Dent. Res. 2004, 83, 317. (164) Nonaka, A.; Tanaka, M.; Anguri, H.; Nagata, H.; Kita, J.; Shizukuishi, S. Oral Dis. 2005, 11, 35. (165) Di Natale, C.; Macagnano, A.; Martinelli, E.; Paolesse, R.; D’Arcangelo, G.; Roscioni, C.; Finazzi-Agro, A.; D’Amico, A. Biosens. Bioelectron. 2003, 18, 1209. (166) Yu, H.; Xu, L.; Cao, M.; Chen, X.; Wang, P.; Jiao, J.; Wang, Y. IEEE Sens. J. 2003, 2, 1333. (167) Chen, X.; Cao, M. F.; Li, Y.; Hu, W. J.; Wang, P.; Ying, K. J.; Pan, H. M. Meas. Sci. Technol. 2005, 16, 1535. (168) Machado, R. F.; Laskowski, D.; Deffenderfer, O.; Burch, T.; Zheng, S.; Mazzone, P. J.; Mekhail, T.; Jennings, C.; Stoller, J. K.; Pyle, J.; Duncan, J.; Dweik, R. A.; Erzurum, S. C. Am. J. Respir. Crit. Care Med. 2005, 171, 1286. (169) Mazzone, P. J.; Hammel, J.; Dweik, R.; Na, J.; Czich, C.; Daniel, Laskowski, D.; Mekhail, T. Thorax 2007, 62, 565. (170) Neustein, M. E. M.Sc. Thesis, Drexel University, 2005. (171) Humphrey, L. L.; Teutsch, S.; Johnson, M. Ann. Intern. Med. 2004, 140, 740. (172) Hauser, G. J.; Chitayat, D.; Berns, L.; D.; B. a.; Muhlbauer, B. Eur. J. Pediatr. 1985, 144, 403. (173) Mennella, J. A.; Beauchamp, G. K. J. Pediatr. 1991, 88, 737. (174) Singh, P. B.; Herbert, J.; Roser, B.; Arnott, L.; Tucker, K.; Brown, R. E. J. Chem. Ecol. 1990, 16, 1667. (175) Novotny, M.; Schwende, F. J.; Wiesler, D.; Jorgenson, J. W.; Carmack, M. Experientia 1984, 40, 217. (176) Wyatt, T. D. Pheromones and animal behaViour: communication by smell and taste; Cambridge University Press: Cambridge, U.K., 2003. (177) Thomas, L. Proceedings of the 4th International ConVocation on Immunology, Buffalo, NY; S. Karger: Basel, 1975; p 2. (178) Penn, D. J.; Oberzaucher, E.; Grammer, K.; Fischer, G.; Soini, H. A.; Wiesler, D.; Novotny, M. V.; Dixon, S. J.; Xu, Y.; Brereton, R. G. J. R. Soc. Interface 2007, 4, 331. (179) Ables, E. M.; Kay, L. M.; Mateo, J. M. Physiol. BehaV. 2007, 90, 726. (180) Osada, K.; Yamazaki, K.; Curran, M.; Bard, J.; Smith, B. P. C.; Beauchamp, G. K. Proc. R. Soc. London, B 2003, 270, 929. (181) Singh, D.; Bronstad, P. M. Proc. R. Soc. London, B 2001, 268, 797. (182) Havlicek, J.; Dvorakova, R.; Bartos, L.; Flegr, J. Ethology 2006, 112, 81. (183) Ackerl, K.; Atzmueller, M.; Grammer, K. Neuroendocrinol. Lett. 2002, 23, 79. (184) Phillips, M.; Herrera, J.; Krishnan, S.; Zain, M.; Greenberg, J.; Cataneo, R. N. J. Chromatogr., B 1999, 729, 75. (185) McCulloch, M.; Jezierski, T.; Broffman, M.; Hubbard, A.; Turner, K.; Janecki, T. Integr. Cancer Ther. 2006, 5, 30.

Ro¨ck et al. (186) Hilbe, W.; Dirnhofer, S.; Greil, R.; Woll, E. Eur. J. Cancer PreV. 2004, 13, 425. (187) Gordon, S. M.; Szidon, J. P.; Krotoszynski, B. K.; Gibbons, R. D.; O’Neill, H. J. Clin. Chem. 1985, 31, 1278. (188) Phillips, M.; Gleeson, K.; Hughes, J. M. B.; Greenberg, J.; Cataneo, R. N.; Baker, L.; McVay, W. P. Lancet 1999, 353, 1930. (189) Phillips, M.; Cataneo, R. N.; Cummin, A. R. C.; Gagliardi, A. J.; Gleeson, K.; Greenberg, J.; Maxfield, R. A.; Rom, W. N. Chest 2003, 123, 2115. (190) Phillips, M.; Altorki, N.; Austin, J. H. M.; Cameron, R. B.; Cataneo, R. N.; Greenberg, J.; Rom, W. N.; Schmitt, P. Cancer Biomarkers 2007, 3, 95. (191) Phillips, M. Am. J. Respir. Crit. Care Med. 2005, 172, 1060. (192) Stetter, J. R.; Penrose, W. R. Sens. Update 2002, 10, 189. (193) Cometto-Muniz, J. E.; Cain, W. S.; Abraham, M. H. Exp. Brain Res. 1998, 118, 180. (194) Prescott, J.; Norris, L.; Kunst, M.; Kim, S. Food Qual. Prefer. 2005, 16, 345. (195) Marti, M. P.; Boque, R.; Riu, M.; Busto, O.; Guasch, J. Anal. Bioanal. Chem. 2003, 376, 497. (196) Marti, M. P.; Boque, R.; Busto, O.; Guasch, J. TrAC, Trends Anal. Chem. 2005, 24, 57. (197) Santos, J. P.; Lozano, J.; Aleixandre, A.; Sayago, I.; Fernandez, M. J.; Ares, L.; Gutierrez, J.; Horrillo, M. C. Sens. Actuators, B 2004, 103, 98. (198) Ragazzo-Sanchez, J. A.; Chalier, P.; Ghommidh, C. Sens. Actuators, B 2005, 106, 253. (199) Marti, M. P.; Pino, J.; Boque, R.; Busto, O.; Guasch, J. Anal. Bioanal. Chem. 2005, 382, 440. (200) Buser, H.-R. J. Agric. Food Chem. 1982, 30, 359. (201) Rapp, A. Nahrung 1998, 42, 351. (202) Dionne, B. C.; Rounbehler, D. P.; Achter, E. K.; Hobbs, J. R.; Fine, D. H. J. Energ. Mater. 1986, 4, 447. (203) Baltussen, E.; Cramers, C. A.; Sandra, P. J. F. Anal. Bioanal. Chem. 2002, 373, 3. (204) Namiesnik, J.; Zabiegala, B.; Kot-Wasik, A.; Partyka, M.; Wasik, A. Anal. Bioanal. Chem. 2005, 381, 279. (205) Scorsone, E.; Pisanelli, A. M.; Persaud, K. C. Sens. Actuators, B 2006, 116, 55. (206) Ro¨ck, F.; Gurlo, A.; Weimar, U. Anal. Chem. 2005, 77, 2762. (207) Friant, S. L.; Suffet, I. H. Anal. Chem. 1979, 51, 2167. (208) Landy, P.; Mielle, P.; Ray, J. L. Proceedings of ISOEN, Rome, Italy; Aracne: Rome, 2002; p 19. (209) Seethapathy, S.; Gorecki, T.; Li, X. J. Chromatogr., A, in press. (210) Batterman, S.; Metts, T.; Kalliokoski, P.; Barnett, E. J. EnViron. Monit. 2002, 4, 361. (211) Ouyang, G.; Pawliszyn, J. TrAC, Trends Anal. Chem. 2006, 25, 692. (212) Schaller, E.; Zenhausern, S.; Zesiger, T.; Bosset, J. O.; Escher, F. Analusis 2000, 28, 743. (213) Ampuero, S.; Bogdanov, S.; Bosset, J. O. Eur. Food Res. Technol. 2004, 218, 198. (214) Bosset, J. O.; Pillonel, L.; Altieri, D.; Tabacchi, R. Mitt. Lebensmittelunters. Hyg. 2004, 95, 85. (215) Garcia-Gonzalez, D. L.; Barie, N.; Rapp, M.; Aparicio, R. J. Agric. Food Chem. 2004, 52, 7475. (216) Barie, N.; Bucking, M.; Rapp, M. Sens. Actuators, B 2006, 114, 482. (217) Lozano, J.; Santos, J. P.; Gutierrez, J.; Horrillo, M. C. IEEE Sens. J. 2005, 1339. (218) Shirey, R. TheReporter USA; Supelco: Bellefonte, PA, 2004; Vol. 22.4. (219) Paschke, A.; Popp, P. J. Chromatogr., A 2004, 1025, 11. (220) Baltussen, E.; Sandra, P.; David, F.; Cramers, C. J. Microcolumn Sep. 1999, 11, 737. (221) Vercammen, J.; Sandra, P.; Baltussen, E.; Sandra, T.; David, F. HRC, J. High Resolut. Chromatogr. 2000, 23, 547. (222) Villanueva, S.; Vegas, A.; Fernandez-Escudero, J. A.; Iniguez, M.; Rodriguez-Mendez, M. L.; de Saja, J. A. Sens. Actuators, B 2006, 120, 278. (223) Morris, L.; Caruana, D. J.; Williams, D. E. Meas. Sci. Technol. 2002, 13, 603. (224) Strathmann, S.; Hahn, S.; Weimar, U. Proceedings of ISOEN, Washington, DC; The Electrochemical Society: Pennington, NJ, 2001; p 1087. (225) Ali, Z.; James, D.; O’Hare, W. T.; Rowell, F. J.; Scott, S. M. J. Therm. Anal. Calorim. 2003, 71, 163. (226) Schafer, T.; Vital, J.; Crespo, J. G. J. Membr. Sci. 2004, 241, 197. (227) Pinheiro, C.; Schafer, T.; Crespo, J. G. Anal. Chem. 2005, 77, 4927. (228) Schafer, T.; Serrano-Santos, M. B.; Rocchi, S.; Fuoco, R. Anal. Bioanal. Chem. 2006, 384, 860. (229) http://www.crisp-dm.org/. (230) Goodner, K. L.; Dreher, J. G.; Rouseff, R. L. Sens. Actuators, B 2001, 80, 261. (231) Scott, S. M.; James, D.; Ali, Z. Microchim. Acta 2006, 156, 183.

Electronic Nose (232) Ciosek, P.; Wroblewski, W. Sens. Actuators, B 2006, 114, 85. (233) Wang, M.; Perera, A.; Gutierrez-Osuna, R. IEEE Sens. J. 2004, 2, 591. (234) Raman, B.; Sun, P. A.; Gutierrez-Galvez, A.; Gutierrez-Osuna, R. IEEE Trans. Neural Network 2006, 17, 1015. (235) Raman, B.; Yamanaka, T.; Gutierrez-Osuna, R. Sens. Actuators, B 2006, 119, 547. (236) Skov, T.; Bro, R. Sens. Actuators, B 2005, 106, 719. (237) Bro, R. Crit. ReV. Anal. Chem. 2006, 36, 279. (238) Gutierrez-Galvez, A.; Gutierrez-Osuna, R. Sens. Actuators, B 2006, 116, 29. (239) Gutierrez-Osuna, R.; Nagle, H. T.; Schiffman, S. S. Sens. Actuators, B 1999, 61, 170. (240) Snopok, B. A.; Kruglenko, I. V. Thin Solid Films 2002, 418, 21. (241) Zhang, W. M.; Hu, J. S.; Song, W. G.; Wan, L. J. Sens. Actuators, B 2007, 123, 454. (242) Hiranaka, Y.; Abe, T.; Murata, H. Sens. Actuators, B 1992, 9, 177. (243) Lee, A. P.; Reedy, B. J. Sens. Actuators, B 1999, 60, 35. (244) Burresi, A.; Fort, A.; Rocchi, S.; Serrano, B.; Ulivieri, N.; Vignoli, V. Sens. Actuators, B 2005, 106, 40. (245) Heilig, A.; Barsan, N.; Weimar, U.; Schweizer-Berberich, M.; Gardner, J. W.; Gopel, W. Sens. Actuators, B 1997, 43, 45. (246) Mielle, P.; Souchaud, M.; Landy, P.; Guichard, E. Sens. Actuators, B 2006, 116, 161. (247) Althainz, P.; Goschnick, J.; Ehrmann, S.; Ache, H. J. Sens. Actuators, B 1996, 33, 72. (248) Goschnick, J. Microelectron. Eng. 2001, 57-8, 693. (249) White, J.; Kauer, J. S.; Dickinson, T. A.; Walt, D. R. Anal. Chem. 1996, 68, 2191. (250) Michael, K. L.; Taylor, L. C.; Schultz, S. L.; Walt, D. R. Anal. Chem. 1998, 70, 1242.

Chemical Reviews, 2008, Vol. 108, No. 2 725 (251) Perera, A.; Yamanaka, T.; Gutierrez-Galvez, A.; Raman, B.; Gutierrez-Osuna, R. Sens. Actuators, B 2006, 116, 17. (252) Pardo, M.; Sberveglieri, G. Sens. Actuators, B 2007, 123, 437. (253) Gualdron, O.; Llobet, E.; Brezmes, J.; Vilanova, X.; Correig, X. IEEE Sens. J. 2004, 2, 892. (254) Gualdron, O.; Llobet, E.; Brezmes, J.; Vilanova, X.; Correig, X. Sens. Actuators, B 2006, 114, 522. (255) Llobet, E.; Gualdron, O.; Brezmes, J.; Vilanova, X.; Correig, X. IEEE Sens. J. 2005. (256) Llobet, E.; Gualdron, O.; Vinaixa, M.; El-Barbri, N.; Brezmes, J.; Vilanova, X.; Bouchikhi, B.; Gomez, R.; Carrasco, J. A.; Correig, X. Chemom. Intell. Lab. Syst. 2007, 85, 253. (257) Gardner, J. W.; Boilot, P.; Hines, E. L. Sens. Actuators, B 2005, 106, 114. (258) Ballabio, D.; Cosio, M. S.; Mannino, S.; Todeschini, R. Anal. Chim. Acta 2006, 578, 170. (259) Ryan, M. A.; Zhou, H. Y.; Buehler, M. G.; Manatt, K. S.; Mowrey, V. S.; Jackson, S. R.; Kisor, A. K.; Shevade, A. V.; Homer, M. L. IEEE Sens. J. 2004, 4, 337. (260) Shirokova, E.; Schmiedeberg, K.; Bedner, P.; Niessen, H.; Willecke, K.; Raguse, J. D.; Meyerhof, W.; Krautwurst, D. J. Biol. Chem. 2005, 280, 11807. (261) Vidic, J. M.; Grosclaude, J.; Persuy, M. A.; Aioun, J.; Salesse, R.; Pajot-Augy, E. Lab Chip 2006, 6, 1026. (262) Liu, Q. J.; Cai, H.; Xu, Y.; Li, Y.; Li, R.; Wang, P. Biosens. Bioelectron. 2006, 22, 318. (263) Schoning, M. J.; Poghossian, A. Electroanalysis 2006, 18, 1893.

CR068121Q

Chem. Rev. 2008, 108, 726−745

726

Hydrogen-Bond Acidic Polymers for Chemical Vapor Sensing Jay W. Grate* Pacific Northwest National Laboratory, P.O. Box 999, Richland Washington 99352 Received May 31, 2007

Contents 1. Introduction 2. Background 2.1. Sorption-Based Sensors 2.2. Partition Coefficients and Sensor Responses 2.3. Fluorinated Alcohols and Phenols 3. Fluoroalcohol-Containing Organic Polymers 4. Silicon-Based Fluoroalcohol and Fluorinated-Phenol Polymers 4.1. Linear Silicon-Containing Polymers 4.2. Hyperbranched and Polyhedral Architectures 5. Linear Solvation Energy Relationships 6. Acoustic Wave Sensors and Arrays 6.1. Chemical Agent Detection 6.2. Acoustic Wave Sensor Array Systems 6.3. Explosives Detection 7. Microcantilever Sensors 8. Sensors Responding to Electrical Properties 8.1. Chemiresistors 8.2. Chemicapacitors 9. Optical and Luminescent Sensors 10. Separations and Preconcentration 11. Discussion 12. Acknowledgments 13. References

726 727 727 727 728 729 731 731 733 735 738 738 739 740 740 741 741 741 742 742 743 743 743

1. Introduction Strongly hydrogen-bond acidic polymers for vapor sensors were introduced in the mid 1980s. Originally, these polymers were of interest for obtaining high sensitivity to organophosphorus nerve agents using acoustic wave sensors. The polymer served as the sorbent layer on the sensor and interacted with the strongly hydrogen-bond basic organophosphorus compounds by hydrogen bonding. These interactions promoted sorption of the vapors into the polymer film on the device surface, which increased the sensor response. The property of hydrogen-bond acidity was incorporated into the polymer by including fluorinated alcohol or fluorinated phenol functional groups in the polymer structure. These polymers have also been demonstrated in explosives sensing applications. From the time of the earliest sensor studies using hydrogen-bond acidic polymers, they were included in sensor arrays where they increase the chemical diversity and help to discriminate among vapors. * To whom correspondence should be addressed. Phone: 509-376-4242. Fax: 509-376-5106. E-mail: [email protected].

Jay W. Grate is a Laboratory Fellow at the Pacific Northwest National Laboratory (PNNL) and an Affiliated Professor with the Chemistry Department of the University of Washington. He received his B.A. degree in Chemistry at Rollins College and Ph.D. degree in Chemistry from the University of California, San Diego. After postdoctoral research at the University of California Irvine, he joined the Naval Research Laboratory in 1984, moving to PNNL in 1992. He spent a sabbatical at the Scripps Research Institute prior to joining PNNL. His research has focused on chemically interactive polymers and nanomaterials, chemical vapor sensors, radiochemical separations and sensing, and bioanalytical fluidics for biothreat detection. His work integrates aspects of the chemical sciences, material sciences, and measurement sciences into new microanalytical principles, methods, and systems. He is noted for developing a rational approach to understanding vapor sorption and polymer design as it relates to chemical detection with polymer-coated sensors. He has published over 100 papers in peer-reviewed journals and numerous book chapters and is author or coauthor on over a dozen patents, several of which have been licensed. He received an R&D 100 Award in 2004 for his work in developing a rationally designed polymer for chemical threat detection and an American Chemical Society Regional Industrial Innovation Award in 2007.

In addition to their use in important applications, this class of polymers has been significant because no hydrogen-bond acidic polymers with the desirable low glass to rubber transition temperatures (Tg) were commercially available. Furthermore, none of the conventional gas chromatographic stationary phases were strongly hydrogen-bond acidic. Those phases that were rather modest hydrogen-bond acids (e.g., docosanol, sorbitol, and diglycerol) were not polymers. Therefore, developing this class of polymers for chemical sensors required design and synthesis. A great variety of hydrogen-bond acidic polymers and architectures have now been developed, from linear organic polymers containing fluoroalcohols to silicon-containing polymers with phenolic or fluoroalcohol functionalities in linear and branched architectures. While they remain useful on acoustic wave sensors for chemical agent detection, these polymers have also been demonstrated in chemiresistor

10.1021/cr068109y CCC: $71.00 © 2008 American Chemical Society Published on Web 01/26/2008

Hydrogen-Bond Acidic Polymers

Figure 1. Sorption-based sensor.

sensors, chemicapacitor sensors, microcantilever sensors, and a variety of optical approaches. This review aims to describe the original motivation and principles behind the use of hydrogen-bond acidic polymers on chemical sensors and review the types of polymers developed. Accordingly, this review covers the period from 1986 to early 2007. The polymer materials designed and synthesized to be hydrogen-bond acidic sorbents will be presented in detail, followed by selected illustrative examples of their use in sensing. These polymers have been most widely used on acoustic wave sensors; their use on other types of sensors are less well known, and this review will seek to bring such examples forward. This review will also describe some examples where they have been used in separation or preconcentration approaches for vapor analysis. Vapor concentrations will be reported in either mg/m3 or parts by volume (ppm or ppb) according to the units used in the work being reviewed with conversions sometimes provided.1 Priority will be given to covering research published as original contributions in peer-reviewed journals, and dates noted will refer to journal publication rather than proceedings when possible. Proceedings papers will be cited in this review, at the author’s discretion, in order to include relevant findings likely to be of interest to the sensor community. It is unfortunate that some polymers and application studies have been published only in proceedings papers or in articles from journal issues that are, in actuality, conference proceedings. For some polymers, syntheses and characterization have never been provided in a peer-reviewed journal. In some peer-reviewed journals, including Sensors and Actuators, Talanta, and even Science, sensor results have been published where the structures of the sensing polymers and/or their synthesis and characterization are not given, and the references do not lead to published synthesis and characterization in peer-reviewed journals. These practices can undermine the scientific credibility of the published sensor results.

2. Background 2.1. Sorption-Based Sensors A sorption-based sensor in the context of this review comprises a sorbent material that collects and concentrates vapor molecules from the gas phase in combination with a means to transduce the reversible sorption process into an analytical signal. Typically, transduction arises from an acoustic wave, mechanical, optical, or electronic device to which the sorbent is applied as a thin film. This concept is shown in Figure 1. Acoustic wave devices such as the quartz crystal microbalance (QCM) or surface acoustic wave (SAW) device represent the most fundamental sorption-based sensors; when acting as pure gravimetric devices, they detect the amount

Chemical Reviews, 2008, Vol. 108, No. 2 727

of vapor sorbed as a mass and the response is not dependent on any other analyte property other than those that influence sorption. Nevertheless, several other types of devices provide platforms for sorption-based sensors including microcantilevers responding to mass uptake and/or bending forces, chemiresistor devices responding to changes in the conductivity of a sorbent thin film, chemicapacitors responding to dielectric changes, and optical fibers, waveguides, or other configurations responding to the change in refractive index, absorbance, or fluorescence of a sorbent material. While the sorbent material for these sensors could, in principle, be selected from a great variety of material types,2 polymers or polymer-based composites are most often used. As polymers are the subject of this review, we will typically refer to the sorbent material interchangeably as a polymer or sorbent.

2.2. Partition Coefficients and Sensor Responses The equilibrium distribution of a vapor between the gas phase and a sorbent polymer phase is characterized by the thermodynamic partition coefficient, K. Taking the concentration in the gas phase as Cv and the equilibrated concentration in the sorbent phase as Cs, the partition coefficient is the ratio defined according to eq 1. Higher values indicate a greater degree of collection and concentration of vapors in the polymer phase, where their presence can be transduced into an analytical signal. The relevance of the partition coefficient to vapor detection using coated acoustic wave sensors was noted in both quartz crystal microbalance (QCM) sensor studies3-5 and SAW sensor studies.6,7

K ) Cs/Cv

(1)

The partition coefficient is related to the Gibb’s free energy of solution of a gaseous solute, ∆GS°, according to eq 2. The Rconst in this equation is the gas constant as usual, and T is the temperature. (The subscript was added to Rconst here to clearly distinguish it from the R2 vapor solvation parameter to be described later in this review.)

∆GS° ) -RconstT ln K

(2)

Since the response of a sorption-based chemical sensor is based on the amount of vapor sorbed into the film, which is related to the amount of vapor in the gas phase by the partition coefficient (Cs ) CvK), the response is a function of the partition coefficient and the vapor-phase concentration according to eq 3. Hence, promoting interactions that increase the sorption of vapors will increase sensor responses. The partition coefficient may be a constant in the case of linear absorption isotherms but will be a function of vapor concentration in the case of nonlinear sorption isotherms.

response ) f(Cs) ) f(CvK)

(3)

This generalized response function can alternatively be written according to eqs 4 and 5, i.e., as a function of the amount of vapor in the film times a sensitivity factor, S. The sensitivity factor may, in turn, be a function of factors specific to the vapor molecule, i.e., analyte-specific factors, as well as specific factors related to the sensing device and its transduction mechanism.

response ) SCs ) SCvK

(4)

728 Chemical Reviews, 2008, Vol. 108, No. 2

Grate

S ) f(analyte-specific factors, transduction-specific factors) (5) For a purely gravimetric polymer-coated acoustic wave sensor the response, a frequency shift denoted by ∆fv, is related to the partition coefficient and vapor concentration according to eq 6.

∆ fv ) ∆ fsCvK/Fs

(6)

The parameters ∆fs and Fs are transduction-specific factors, where ∆fs is a measure of the amount of polymer on the sensor surface (specifically the frequency shift due to deposition of the film material onto the bare sensor) and Fs is the density of the sorbent polymer phase. For a purely mass-sensitive device, the analytical signal is related to the amount of vapor absorbed as a mass without dependence on other analyte-specific factors. The response of a sensor to the amount of vapor sorbed, where the transducer detects the volume rather than the mass of vapor, can be expressed as in eq 7.

R ) VvS′CvK

Table 1. Hydrogen-Bonding Solvation Parametersa for Selected Hydrogen-Bond Acidic Molecules

(7)

The parameter S′ is a sensitivity factor as in eq 4. Of particular note is that the volume sensitivity is related to the vapor-specific volume, Vv, which relates to the volume fraction of vapor in the polymer/vapor solution, φv ) VvCvK.8 The vapor-specific volume represents an analyte-specific sensitivity factor as expressed in eq 5. The quartz crystal microbalance (QCM) is a purely gravimetric sensor for sufficiently thin polymer films that move synchronously with the device surface.9-15 The SAW device can be gravimetric but often contains sensitivity to changes in the modulus of the polymer upon vapor sorption. The modulus change is related to the volume of the sorbed molecules through their influence on polymer film free volume.8-10,16-21 Accordingly, a SAW sensor acting as a mass-plus-volume-transducing device has a response that can be expressed as in eq 8. The first term has no analyte-specific sensitivity factors, while the second term has Vv, an analytespecific sensitivity factor which will vary from vapor to vapor.

∆ fv ) (∆ fsCvK/Fs) + VvS′CvK

Figure 2. Hydrogen-bonding interactions between fluorinated hydrogen-bond acidic molecules hexafluoroisopropanol or 3,5-bis(trifluoromethyl)phenol and DMMP, a hydrogen-bond basic organophosphorus compound.

(8)

Since rapid sensor responses are typically required or at least desired, use of polymers with Tg values below the operating temperature of the sensor is preferable. Vapor diffusion is much more rapid above this transition temperature due to greater polymer free volume and polymer segmental chain motion, leading to more rapid equilibration of the film material with the gas-phase vapor concentration.

2.3. Fluorinated Alcohols and Phenols The interactions that promote sorption of vapor molecules into a polymer film, where the vapor molecules act as solutes and the polymer acts as the solvent, are, by definition, solubility interactions.22-25 These interactions include van der Waals interactions and the hydrogen-bonding interactions of interest in this review. Figure 2 shows the concept of hydrogen bonding between a fluorinated alcohol or a fluorinated phenol with a hydrogen-bond base, represented in this case by the nerve agent simulant dimethyl methylphosphonate (DMMP).

a

molecule

ΣRH2

ΣβH2

ethanol 2,2,2-trifluoroethanol hexafluoroisopropanol phenol 3-fluorophenol 3,5-bis(trifluoromethyl)phenol acetic acid

0.37 0.57 0.77 0.60 0.68 0.82 0.62

0.48 0.25 0.10 0.30 0.17 0.44

Values are from refs 115 and 22.

The solvation parameters ΣRH2 and ΣβH2 have been developed to characterize the hydrogen-bond acidity and hydrogenbond basicity, respectively, of solute molecules. It is important to distinguish hydrogen-bond acidity and hydrogenbond basicity as they relate to hydrogen-bonding interactions from proton-transfer acidity and basicity as they relate to acid-base reactions. These are fundamentally different processes, and there is no general relationship between pKa values and hydrogen-bond acidity, for example.24-26 Resonance stabilization of a conjugate base can be quite significant in influencing proton dissociation but is not so relevant to hydrogen-bonding interactions. Though general correlations can sometimes be made within chemical families, no general relationship exists between proton-transfer acidities and basicities and hydrogen-bonding acidities and basicities. For example, phenol and acetic acid have similar hydrogenbond acidities as indicated by their ΣRH2 values of 0.60 and 0.62, respectively, but acetic acid is a stronger proton dissociation acid in water by 5 orders of magnitude. Similarly, although urea is a weaker proton-acceptor base than triethylamine, it is a stronger hydrogen-bond base (ΣβH2 ) 0.90 for urea and 0.79 for triethylamine). Use of solvation parameters in linear solvation energy relationships (LSERs) will be described below. Solvation parameters can also be used to relate the structural features in molecules to the resulting solubility properties. This can provide insight into the structures that one might incorporate into a polymer to obtain particular properties and potential interactions. The solvation parameters for the hydrogen-bond acidity and hydrogen-bond basicity of several hydrogen-bond acidic molecules are given in Table 1. These two parameters are scaled to free energies in the same way. The first three molecules illustrate increasing hydrogen-bond acidity of alcohols as the fluorination is increased, as indicated by the ΣRH2 values. Simultaneously, the hydrogen-bond basicity goes down, which can decrease self-association. Hexafluoroisopropanol, the most hydrogen-bond acidic of these alcohols, is shown in Figure 2. It is also apparent that the unfluorinated

Hydrogen-Bond Acidic Polymers

alcohol, ethanol, is only a weak to moderate hydrogen-bond acid while being even a better base. The next three molecules illustrate increasing hydrogen-bond acidity of phenols as the fluorination is increased. A phenol is intrinsically a better hydrogen-bond acid than an aliphatic alcohol and becomes even better if fluorinated. 3,5-Bis(trifluoromethyl)phenol, the most hydrogen-bond acidic of these phenols, is included in Figure 2. Accordingly, strong hydrogen-bond acidity can be obtained by incorporating fluorinated alcohol or fluorinated phenolic functional groups as substituents in a polymer structure.7,22,27 These groups maximize the hydrogen-bond acidity of the hydroxyl groups by the electron-withdrawing effect of the fluorine atoms while simultaneously minimizing the hydrogenbond basicity of the hydroxylic oxygen atoms, and they contain no other significantly basic heteroatoms. Minimizing hydrogen-bond basicity within the structure is important to minimize self-association. The energetic gain of the interaction of a basic analyte vapor with a formerly free hydroxyl group hydrogen, where there is no cost to breaking a pre-existing hydrogen bond, will be substantially greater than the gain from the basic vapor interacting with a formerly hydrogen-bonded hydroxyl group hydrogen and hence will provide a greater driving force for sorption. Finally, acetic acid is included in Table 1, since carboxylic acids are well known as acidic organic functional groups in the sense of proton dissociation acidity. Carboxylic acids are both hydrogen-bond acids and hydrogen-bond bases, and so can be expected to self-associate in condensed phases. In fact, even in the gas phase acetic acid exists primarily as a hydrogen-bonded dimer. The carbonyl oxygen provides a basic heteroatom that promotes self-association. For these reasons, carboxylic acids are not ideal functional groups for strongly hydrogen-bond acidic polymers.

3. Fluoroalcohol-Containing Organic Polymers This section on organic polymers (section 3) and the subsequent section on silicon-containing polymers (section 4) will focus on the structures and synthesis of the hydrogenbond acidic polymers being described as well as sorptive or spectroscopic studies that establish their interactions with basic solutes via hydrogen bonding. Use of these polymers on sensors and devices for vapor detection will be highlighted in later sections (sections 6-10) of the review. Use of hexafluoroisopropanol substituents in a polymer to promote the sorption of organophosphorus vapors was recognized in early studies published in 1987 by Barlow et al.28,29 These authors used both analog calorimetry and measurements of vapor uptake on a QCM to demonstrate and investigate sorption of DMMP by a modified polystyrene copolymer material, which interacted by hydrogen bonding to the pendant hexafluoroisopropanol groups. These studies followed those of Pearce, who first modified polystyrene with pendent hexafluoroisopropanol groups as part of a study on promoting the miscibility of polymer blends through polymerpolymer hydrogen bonding.30 At around the same time as Barlow’s initial report in 1984,31 scientists at the Naval Research Laboratory (NRL) were testing a fluoroepoxy prepolymer, dubbed fluoropolyol (FPOL), as a sorbent on SAW devices for organophosphorus vapor detection.7,27,32,33 This prepolymer was originally prepared as a random copolymer containing a mixture of isomers for the development of protective coatings for naval applications.34 The synthesis is shown in Scheme 1, produc-

Chemical Reviews, 2008, Vol. 108, No. 2 729 Scheme 1

ing FPOL(a).35 In most publications, the nominal repeat unit structures of these types of materials have been represented more simply by structures like that in FPOL(b), which illustrate the functional groups and dominant isomers.7,33,34,36 (In subsequent work to be described below, specific isomeric FPOL polymers were prepared using diglycidyl ether and diol reagents as indicated in the dashed box.) Hydroxyl groups along the polymer backbone, with electron-withdrawing fluorine and oxygen atoms, are responsible for the hydrogen-bond acidic properties. The FPOL sample used in early SAW sensor studies had a reported Tg of 10 °C. Thus, it had desirable fluoroalcohol interactive groups in a polymer with a low Tg. Additional studies of FPOL synthesis are described below.37-39 FPOL has been used in early SAW array studies, organophosphorus simulant and agent detection, comparisons of synthesized hydrogen-bond acidic polymers, studies of polymer solubility properties using LSERs, development of new chemometric approaches, sensor array systems, polymer film deposition methods, and development of other types of sensors including chemiresistors.7,8,21,22,27,32,33,37-62 It represented the de facto standard for hydrogen-bond acidic sensor materials for many years. Subsequent to the studies of Barlow et al. and the initial demonstration of FPOL as a sensor coating, Snow et al. published a detailed study of a variety of fluoroalcohol-containing polymers in 1991.27 Among these were a series of hexafluoroisopropanol-substituted polystyrenes as shown in Scheme 2. These were prepared by first making the corresponding hexafluoroisopropanol-substituted styrene monomers containing ortho-, meta-, or para-substituted phenyl groups and polymerizing them by a free-radical-initiated reaction to obtain the product homopolymers. The monomers were prepared from the reactions of bromostyrenes with magnesium to prepare Grignard reagents that were then reacted with hexafluoroacetone. The parent polymer in this group, polystyrene, has a Tg of 100 °C, while the para- and meta-substituted polymers had Tg values of 122 and 84 °C, respectively. No value was reported for the ortho-substituted polymer because the ortho-substituted monomer polymerized rather poorly, possibly due to a steric effect, and the yield was low. In the same paper, Snow et al. described the modification of polyisoprene with hexafluoroacetone to produce hexa-

730 Chemical Reviews, 2008, Vol. 108, No. 2 Scheme 2

Scheme 3

Scheme 4

fluoroisopropanol-substituted hydrocarbon polymers as shown in Scheme 3. This reaction of hexafluoroacetone with double bonds at elevated temperatures follows the chemistry described by Urry et al.63 The reaction proceeds most readily on terminal double bonds. In principle, up to two hexafluoroacetone groups could be incorporated per isoprene repeat unit. However, there was no evidence to indicate that more than one hexafluoroacetone was added to any single isoprene repeat unit. Two samples were prepared with different hexafluoroacetone to monomer ratios, yielding samples with 80% (PIPFA1) and 46% (PIPFA2) of repeat units functionalized as determined by elemental analysis. The PIPFA2 was a dense glassy material with a Tg of 32 °C, while the more substituted PIPFA1 reported to be a fibrous glassy material with a Tg of 60 °C. One additional fluoroalcohol polymer based on an acrylic ester backbone was prepared by Snow et al. Both the fluoroalcohol-substituted PAFA polymer shown in Scheme 4 and a control polymer without a free fluoroalcohol PA were prepared. The Tg values for PA and PAFA were 96 and 175 °C, respectively. All these polymers were examined as films on SAW devices to measure DMMP vapor uptake and in infrared (IR) spectroscopy studies to demonstrate hydrogen bonding to DMMP. SAW sensor responses were plotted against the vapor concentration as P/Psat, i.e., the test vapor partial pressure relative to the saturated vapor pressure. Since the SAW response is proportional to the amount of vapor sorbed,

Grate

these are related to sorption isotherms. (If the response were purely gravimetric, the calibration curve would be the same as the sorption isotherm, but modulus effects may occur.) For each type of fluoroalcohol polymer, the curves of the test polymer or polymers were compared with one or more controls. For example, the response of FPOL-coated sensor was compared with that of a sensor coated with a FPOL derivative that had been acetylated to convert the free hydroxyls to non-hydrogen-bond acidic esters. The sensor response to DMMP at the lowest concentrations, using the FPOL with free hydroxyl groups, was about 5 times greater than that for the sensor coated with the acetylated derivative of FPOL. The polystyrene polymers with pendant hexafluoroisopropanol groups were compared with unmodified polystyrene and an acetylated derivative of the parasubstituted polymer. The para- and meta-substituted polymers with free fluoroalcohols sorbed DMMP more than 5 times greater than the acetylated control and more than 10 times greater than the polystyrene control. Similarly, fluoroalcoholcontaining polymers in the polyisoprene and polyacrylate cases were about 10 times more sorbent than the controls. These experiments clearly confirmed the importance of the fluoroalcohol hydroxyl groups for promoting DMMP sorption, consistent with the hydrogen-bonding rationale for polymer design. IR spectroscopy studies were also carried out comparing the fluoroalcohol hydroxyl region before and after DMMP sorption. The absorption frequencies of the free hydroxyls were reported and compared with the shifted frequency in the presence of DMMP. Hydrogen-bond formation was clearly evident in these spectral comparisons. In carbon tetrachloride, hexafluoroisopropanol is known to hydrogen bond to DMMP with a spectral shift from 3600 to 3190 cm-1.28 The fluoroalcohol polymers showed similar shifts. It was noted that prior studies of hydrogen-bond formation between hexafluoroisopropanol and various Lewis bases had demonstrated that the size of the IR spectral shift of the hydroxyl region could be related to the enthalpy of hydrogen bonding.64 This prior study noted that the fluorinated alcohol was a better hydrogen-bond acid than an unfluorinated phenol. These studies by Snow et al., published in 1991, successfully demonstrated the approach of using fluoroalcohols incorporated into polymer structure to promote DMMP sorption by hydrogen-bonding interactions. However, all of the new polymers had Tg values above room temperature. Hence, they were not ideal in making a sensor coating yielding a very rapid response. In work published in 1997 and 1998, researchers from France investigated FPOL in detail, working with samples of the original prepolymer as well as samples that they synthesized themselves.37,38 These investigators noted that the original FPOL was a mixture of isomers and that the material contained low molecular weight oligomers. Both of these factors could have a plasticizing effect that lowers the Tg. They set out to synthesize FPOL polymer samples for each of the possible isomer combinations. Since there are meta- and para-isomers with regard to the phenyl group in the repeat structure and cis- and trans-isomers with regard to the double bond, there are four isomer combinations, each of which was prepared as the pure polymer. The original FPOL sample contained primarily the meta-aromatic/transethylenic units. The dashed box in Scheme 1 shows the reactants used to prepare one of the isomeric fluoropolyol

Hydrogen-Bond Acidic Polymers

Chemical Reviews, 2008, Vol. 108, No. 2 731

Scheme 5

materials, shown as FPOL(b). These authors found a Tg value for the original FPOL sample of 16 °C, whereas they reported Tg values from 35 to 45 °C for the pure isomeric polymers. In terms of vapor sensitivity in tests with chemical agent GB, these authors reported that the polymer with metaaromatic/trans-ethylenic units provided the best responses. Response times at temperatures below the Tg (30 °C) were extremely slow and much faster above the Tg (40 °C). These results are consistent with the expected slow diffusion of vapors in glassy polymers and the design rationale of selecting polymers with low Tg in order to obtain rapid sensor responses. Additional preparations of fluoropolyol-type materials were reported in 2004.39 These authors used reactions of epichlorohydrin and fluorinated diols to produce their epoxy polymers and tested them on 10 MHz QCM sensors. The resins obtained were pale yellow to light brown liquids at room temperature with molecular weights on the order of 1600-1800, confirming their oligomeric nature. Swager et al. synthesized a hexafluoroisopropanolsubstituted organic polymer with a conjugated backbone structure, which they reported in 2005.65 Three poly(phenylene-ethynylene) polymers were prepared, one of which is shown in Scheme 5. The pendant hexafluoroalcohol groups did not appear to influence the polymerization process by the palladium-catalyzed Sonogashira-Hagihara crosscoupling reaction. The rigid iptycene groups on the polymer provide several benefits, including reduced solid-state aggregation, increased solubility, and increased free volume that facilitates vapor diffusion within the solid polymer material for binding and fluorescence response. These materials were investigated as fluorescent sensors (see below) in experiments with nitroaromatic and pyridine vapors.

4. Silicon-Based Fluoroalcohol and Fluorinated-Phenol Polymers 4.1. Linear Silicon-Containing Polymers To address the issue of obtaining a fluoroalcohol-containing polymer with intrinsically low Tg, Grate turned to the use of a siloxane polymer backbone.44,66,67 Polydimethylsiloxane has one of the lowest Tg values known among polymers. Unless the silicon atoms in a polysiloxane have

Scheme 6

large substituents, the siloxane (Si-O-Si) linkage can move and rotate with very little hindrance, resulting in low Tg values. Polysiloxanes are also the most popular sorbent stationary phases used in gas-liquid chromatography due to their chemical and thermal stability, favorable wetting characteristics for coating columns or supports, low Tg values, and fast vapor diffusion. Synthetically, their selectivity can be readily tailored by variation of the substituent organic groups, and there are a variety of cross-linking and immobilization approaches available. Grate designed and synthesized a polysiloxane with a pendant hexafluoroisopropanol group on each repeat unit.66,67 The synthesis of this polymer, dubbed SXFA for siloxane fluoroalcohol, was published in 1995 as part of a study of sorbent sensing polymers using LSERs.44 Reaction of hexafluoroacetone with a terminal alkene was used to generate the pendant group, after first preparing a polysiloxane polymer containing pendant allyl groups, as shown in Scheme 6. Thus, this synthesis uses the reaction of hexafluoroacetone with terminal alkenes on a pre-existing polymer much like the reaction with polyisoprene introduced by Snow (see Scheme 3). The product was a viscous liquid. SXFA has been used by several investigators on a variety of sensor platforms, including SAW devices, microcantilevers, flexural plate wave devices, and chemicapacitors for applications including chemical agent detection and explosives detection, and it has been used in studies of vapor sorption and coating deposition methods.8,44,46,47,51,52,57,61,68-75 In subsequent work reported in 1997, Grate and Kaganove created hybrid organic/inorganic polymers incorporating oligosiloxanes as the inorganic segment in the polymer backbone.49 The functional organic segment was derived from a diallyl-substituted fluorinated bisphenol (2,2-bis(3allyl-4-hydroxyphenyl)hexafluoropropane; F-BSP), as shown

732 Chemical Reviews, 2008, Vol. 108, No. 2 Scheme 7

in Scheme 7. These polycarbosiloxane polymers, assembled using Pt-catalyzed hydrosilylation chemistry, incorporated both siloxane and carbosilane linkages. Careful characterization of the products confirmed that the reaction occurred by hydrosilylation to produce Si-C bonds rather than potentially competing dehydrocondensation reactions that could have produced O-Si bonds and consumed the desired phenolic hydroxyl groups.49 Materials were made with three, six, or tens of silicon atoms in the inorganic segment. The first of these, now commonly known as BSP3, is shown in Scheme 7. Use of the silicon-containing inorganic segment succeeded in yielding polymers with low Tg values, reported to be 6 °C for BSP3. BSP3 was isolated as a very viscous gum phase. Hydrosilylation polymerization to produce a carbosiloxane polymer containing a nonfluorinated bisphenol (2,2-bis(3allyl-4-hydroxyphenyl)propane; H-BSP) had been previously demonstrated by Mathias in a brief communication.76 Similar polymers with unfluorinated bisphenols have been investigated for surface modification applications by Boileau.77,78 Selection of fluorinated phenolic functional groups in BSP3 was motivated, in part, by prior work by Abraham and Rose, where a variety of low volatility liquid phenolic materials were compared using gas-liquid chromatographic measurements and LSERs.79 The F-BSP monomer shown in Scheme 7 along with a propyl-substituted variant were the most strongly hydrogen-bond acidic phenols in the study. Moreover, they were much more acidic than nonfluorinated bisphenols (H-BSP and a propyl analog), confirming the role of electron-withdrawing fluorine atoms to yield more strongly hydrogen-bond acidic hydroxyl groups. For example, the observed partition coefficient for ethylamine at 25 °C was 10 800 in F-BSP, whereas it was only 56 in the nonfluorinated H-BSP. In experiments with SAW devices, BSP3 polymer proved to be an excellent sorbent for DMMP, yielding sensors that Scheme 8

Grate

were twice as sensitive at trace concentrations compared to FPOL-coated sensors.49 BSP3 has been used in SAW sensor studies, SAW array systems with preconcentrators and gas chromatographic separation systems, comparison of polymercoated SAWs with nanoparticle chemiresistor sensors, fluorescence chemical agent sensors, solid-phase microextraction, and studies on the use of nanoparticles to suppress polymer dewetting on surfaces.49,51,55,57,59,80-90 The synthesis of BSP3 introduced hydrosilylation chemistry to the development of hydrogen-bond acidic polymers for sensors and was the first to examine a polymer with a fluorinated phenol rather than a fluorinated alcohol. In addition, these authors set out hydrosilylation polymerization as a versatile method to produce low Tg polymers with a variety of chemical selectivities, including hydrogen-bond basic, dipolar, polarizable, and nonpolar,55,59 as required to obtain chemical diversity in sorptive sensor arrays.2,22,26 Hydrosilylation chemistry was also shown to be a viable route to cross-link such hybrid polymers. In addition, using a photoactivated hydrosilylation catalyst, Pt(acac)2, sorptive polymers could be photopatterned.55,91 Illuminated regions of the film that are crosslinked by the photoactivated catalyst are retained, while regions that are not illuminated or crosslinked are removed in the wash step to develop the pattern. The ability to photopattern a sorptive polymer may be desirable for some sensor types in order to localize the polymer onto a specific region of the transducer. In principle, photopatterning could also be used to prepare large numbers of coated sensors in parallel at the wafer level. To obtain hydrogen-bond acidic photopatterned films, a method was developed that combined photoactivated Pt-catalyzed polymerization and cross-linking in a single step. The monomers and cross-linker used to prepare photopatterned film materials related to BSP3 are shown in Scheme 8. An image showing photopatterned domains of this polymer as curved lines on a silicon wafer is shown in Figure 3. Poole and Abraham developed a new fluoroalcoholsubstituted siloxane polymer dubbed PSF6, as shown in Scheme 9.92 Platinum-catalyzed hydrosilylation chemistry was used to add a trimethylsilyl-protected hexafluoroisopropanol-functionalized alkene to a preexisting methylhydrosiloxane-dimethylsiloxane copolymer material. Residual unreacted silicon hydrides were then capped by addition of octene, and finally, the alcohol was deprotected. Approximately 26% of the repeat units were functionalized with the fluoroalcohol, and the molecular weight was estimated at 4700. This polymer was developed as a high-temperature gas chromatographic phase. A control polymer with the corresponding unfluorinated alcohol was also prepared for comparison. Using the LSER

Hydrogen-Bond Acidic Polymers

Chemical Reviews, 2008, Vol. 108, No. 2 733 Scheme 10

Scheme 11 Figure 3. Image of a small silicon wafer on which a film of the materials in Scheme 8 was photopatterned to produce three lines of polymer material containing fluorinated bisphenol repeat units. Reprinted with permission from ref 55. Copyright 2000 American Chemical Society. Scheme 9

Scheme 12

method to evaluate the solubility properties of these polymers (see below), it was found that the fluorinated material was a strong hydrogen-bond acid with virtually no hydrogenbond basicity. By contrast, the unfluorinated control was a significantly weaker hydrogen-bond acid and actually a better hydrogen-bond base than acid. These studies further confirmed the principles used to select fluoroalcohols as one of the best hydrogen-bond acidic groups for obtaining selective sorbent materials for hydrogen-bond bases. Fluorination increases hydrogen-bond acidity by two effects: by the inductive electron-withdrawing effect increasing the intrinsic hydrogen-bond acidity of the unassociated hydroxyl group and by reducing self-association as a result of the lower hydrogen-bond basicity of the alcohol, resulting in more free hydroxyls available for hydrogen-bonding interactions. This paper also reviewed a variety of hydrogen-bond acidic phases that had been prepared and evaluated by LSERs up to that date, including FPOL, PSpFA, SXFA, and some nonpolymeric bisphenol molecules. In a 2001 journal paper describing explosives detection with coated SAW devices,61 McGill and Houser reported a polysiloxane and two linear carbosilane polymers, all with pendant hexafluoroisopropanol-substituted phenyl groups. Two of these, SXPHFA and CS3P2, are shown in Scheme 10. This paper introduced hydrogen-bond acidic polycarbosilanes; like polysiloxanes, many carbosilane polymers have low Tg values. However, neither this paper nor

preceding proceedings articles93-96 that described the use of SXPHFA provided the synthetic procedures or characterization. A series of subsequent proceedings papers, such as Polymer Preprints or PMSE Preprints, provided synthetic methods and characterization for various hydrogen-bond polymers prepared by this group97-102 and described polycarbosilanes with hexafluoroalcohol groups derived from the reaction of hexafluoroacetone with alkenyl-substituted linear polycarbosilanes (isolated as yellow oils).99,102 One of these, PMSFA, is also shown in Scheme 10. (The designation PMSFA is by this review author, not the inventors.) Another group reported the linear polysiloxane dubbed PLF (shown in Scheme 11), which was prepared with pendant hexafluoroisopropanol substituents added to an existing polymethylhydrosiloxane (PMHS) polymer by Ptcatalyzed hydrosilylation chemistry, in 2000 with additional papers appearing in 2001 and 2004.103-105 Synthesis and characterization details were not given. A linear polysiloxane with pendant fluorinated phenol groups has been prepared by Wheeler at Sandia using Ptcatalyzed hydrosilylation chemistry to add an allyl-substituted 3,5-bis(trifluoromethyl)phenol to an existing PMHS polymer.84,106 This polymer, dubbed DKAP, has the repeat unit shown in Scheme 12. It has been used on SAW sensors as part of Sandia’s microChemLab SAW array system84 and mentioned in connection with studying the effects of nanoparticles for suppressing the dewetting of polymer films from substrate surfaces.107 The synthesis and characterization remains unpublished.

4.2. Hyperbranched and Polyhedral Architectures Hyperbranched polysiloxane and polycarbosilane structures have also attracted interest. A hyperbranched material

734 Chemical Reviews, 2008, Vol. 108, No. 2

Grate

Scheme 13

Scheme 14

dubbed PBF, containing both siloxane and carbosilane linkages, was claimed in 2000.103 A proposed structure corresponding to a branched version of PLF (Scheme 11) was depicted, but no synthesis or characterization data were reported. A thorough full journal paper on hyperbranched polymers with strongly hydrogen-bonded acidic groups was published in 2004 by Dvornic and co-workers at the Michigan Molecular Institute.108 These authors generated hyperbranched silicon-based polymers using hydrosilylation as shown in Scheme 13, which were subsequently functionalized as shown in Scheme 14. The functional groups include fluorinated alcohols and phenols similar to those in previous linear polymers, added using either hydrosilylation chemistry or the reaction of hexafluoroacetone with an allyl group (as in SXFA, Scheme 6). The resulting polymers were all yellow oils. The hyperbranched backbones had reported molecular weights of 2913 and 6322 for HB-PCSOX, a hyperbranched polycarbosiloxane, and HB-PCS, a hyperbranched polycarbosilane, respectively. The molecular weights were not

degraded by functionalization. The molecular weight gains due to functionalization and the resulting polydispersity were quite variable. (The names in Scheme 14 were created here by combining “HB” for hyperbranched and the number corresponding to the number in these authors manuscript.) These polymers were coated onto 500 MHz SAW devices and tested against DMMP. In a series of reports in PMSE Preprints and Polymer Preprints starting in 2003, Houser, Simonson, and McGill described hyperbranched polycarbosilanes with fluoroalcohol groups and indicated their synthetic approaches.99-102 The fluoroalcohol-substituted polymers are viscous oils.102 A hyperbranched structure dubbed HC, used in a number of subsequent sensor studies, was developed. The pendant functional groups on HC are shown in ref 74 and correspond to those in the linear PMSFA (Scheme 10).109 Linear and hyperbranched polycarbosilanes have been used in SAW devices for explosives and chemical agent detection, on microcantilever beams, with chemicapacitive sensors, and as sorbents on microfabricated preconcentrators.61,74,99,110-112

Hydrogen-Bond Acidic Polymers

Chemical Reviews, 2008, Vol. 108, No. 2 735

Scheme 15

Full publication of the synthesis and characterization of the linear and hyperbranched polycarbosilane polymers in a peerreviewed journal would be desirable and would help to support the published performance of HC-coated sensors and devices. Polyhedral oligosilsesquioxane (POSS) compounds represent yet another silicon-containing architecture that has been functionalized with hydrogen-bond acidic groups, as shown in Schemes 15 and 16.113 This report appeared in early 2007. The POSS starting materials used were of a precisely defined size of the general formula R8Si8O12 with the R groups on the corners of the nanoscopic cubic structure. Hydrosilylation chemistry was used to create octafunctionalized POSS compounds starting with the octa(oxidimethylsilane)POSS shown in Scheme 15. Functionalization was designed to match prior linear polymers, leading to octasubstituted BSP3-POSS, for example, which has bisphenol groups similar to BSP3 (Scheme 7). The octasubstituted material designated ‘FPOL’-POSS (quotation marks added by this author), however, has functionalization more similar to PSmFA (Scheme 2), PSF6 (Scheme 9), or SXPHFA (Scheme 10) than to the original FPOL (Scheme 1). An unfluorinated phenol was also used to prepare an octasubstituted phenol-POSS. A monosubstituted SXFA-POSS was prepared to mimic the linear SXFA (Scheme 6) polymer, as shown in Scheme 16. All the new POSS compounds were fully characterized. The BSP3-POSS and phenol-POSS were isolated as white solids. ‘FPOL’-POSS and SXFA-POSS were isolated as yellow gums. These compounds were investigated as functionalized “nanofillers” in nanofiller-polymer blends used as sorptive coatings on SAW sensors. The POSS compounds were blended with either the corresponding linear fluoropolymer or an unfunctionalized polycarbosilane. The vapors used to test the coatings were nitroaromatic compounds.

Scheme 16

5. Linear Solvation Energy Relationships Linear solvation energy relationships, or LSERs, are semiempirical models for solubility-dependent phenomena, expressing a measure of the phenomenon as a linear combination of terms related to fundamental interactions. The partition coefficient, for example, is a measure of the sorption of a vapor from the gas phase into a sorbent phase serving as the solvent. The vapor is the solute, and the interactions are by definition solubility interactions. LSERs have been successful in correlating a vast amount of solubilitydependent phenomena, often to the precision of the available data.23,25,114-116 For a sensor whose response is directly proportional to the amount of vapor sorbed in a polymer layer, the sensor response represents a measure of the solubility-dependent phenomenon. Application of LSERs to the study of polymercoated chemical sensors was introduced in 19887 and has been described in detail in a number of articles and reviews by Abraham, Grate, and McGill.2,11,22,26,57,117 LSERs have been discussed independently by Hierlemann et al.118 The basic form of the LSER developed by Abraham for vapor sorption is given in eq 9, where K is the partition coefficient as defined in eq 1.22,23,26,115

log K ) c + rR2 + sπH2 + aΣRH2 + bΣβH2 + l log L16 (9) A set of solvation parameters R2, πH2 , ΣRH2 , ΣβH2 , and log L16

736 Chemical Reviews, 2008, Vol. 108, No. 2

Grate

Table 2. LSER Coefficientsa for Selected Polymers and Phenolic Liquids Determined at 298 K materialb

LSER coefficients

abbreviation

type

polarizability, r

dipolarity/polarizability, s

basicity, a

acidity, b

dispersion/cavity, l

FPOL PSpFA SXFA PIB PECH SXPH F-BSP H-BSP

HBApolymerc HBApolymerc HBApolymerc LPpolymerc LPpolymerc LPpolymerc HBAliquidc HBAliquidc

-0.67 -1.54 -0.42 -0.08 0.10 0.18 -0.48 -0.92

1.45 2.49 0.60 0.37 1.63 1.29 1.04 2.24

1.49 1.51 0.70 0.18 1.45 0.56 0.89 2.79

4.09 5.88 4.25 0.00 0.71 0.44 4.56 2.41

0.81 0.90 0.72 1.02 0.83 0.89 0.863 0.975

a Data are from refs 44 and 79 for measurements at 298 K. b The polymers are fluoropolyol (FPOL, Scheme 1), poly(4-vinylhexafluorocumyl alcohol) (PspFA-Scheme 2), a hexafluoroisopropanol-substituted polysiloxane (SXFA, Scheme 6), poly(isobutylene) (PIB, Scheme 17), poly(epichlorohydrin) (PECH, Scheme 17), and a 75%-phenyl-25%-methylpolysiloxane (SXPH, Scheme 17). The phenols are 2,2-bis(3-allyl-4hydroxyphenyl)hexafluoropropane (F-BSP in the table, shown in Schemes 7 and 8) and the unfluorinated 2,2-bis(3-allyl-4-hydroxyphenyl)propane (H-BSP in the table). c HBApolymer ) hydrogen-bond acidic polymer, LPpolymer ) low-polarity polymer, and HBA liquid ) hydrogen-bond acidic bisphenolic liquid.

characterize the solubility properties of the monomeric vapor molecules.23,115,119 Their corresponding coefficients, r, s, a, b, and the letter l represent the complementary properties of the sorbent phase acting as the solvent. These coefficients, and the constant c, are obtained by regressing the measured partition coefficients of a series of diverse solute compounds against the known solvation parameters of those compounds by the method of multiple linear regression. Typically, the required partition coefficients are determined from the retention times of injected solutes on a gas-liquid chromatographic column using the sorbent phase of interest as the stationary phase. The coefficients that characterize sorbent phase properties are of particular interest in the context of this review. The solvation parameter R2 is a calculated excess molar refraction parameter that provides a quantitative measure of polarizable nonbonding and π electrons. The parameter πH2 measures a molecule’s ability to stabilize a neighboring charge or dipole through dipole-dipole or dipole-induced dipole interactions, for example. The hydrogen-bonding parameters ΣRH2 and ΣβH2 measure effective hydrogen-bond acidity and basicity, respectively. The log L16 parameter is the liquid/gas partition coefficient of the solute on hexadecane at 298 K (determined by gas-liquid chromatography). The log L16 parameter is a combined measure of exoergic dispersion interactions that increase log L16 and the endoergic cost of creating a cavity in hexadecane leading to a decrease in log L16. All of these parameters, except R2, were derived fromequilibriummeasurementsofcomplexationorpartitioning,23,120-124 and the LSER method is free-energy related. However, the parameters do not all scale with free energy equivalently. While the parameters for hydrogen-bond acidity and basicity are similarly scaled, for example, the log L16 parameter is quite different. Free energy contributions must be calculated for particular solvent/solute pairs for comparison as we shall illustrate below. The complicated notation for the solvation parameters arose from the process of developing the scales and revising them, each revision requiring a modification of the notation. As the descriptor scales are now well established, the notation was recently revised,23 such that solvation parameters R2, πH2 , ΣRH2 , ΣβH2 , and log L16 are now denoted by E, S, A, B, and L, respectively. The LSER eq 9 is then more easily written as eq 10. The coefficients remain the same except the r coefficient is now represented as an e coefficient. The development of the solute descriptors and the new notation were described in detail in a recent review.23 In the current

treatment we will retain the old notation in eq 9 for consistency with the papers being reviewed.

log K ) c + eE + sS + aA + bB + lL

(10)

In eq 9, the l coefficient to log L16 is related to dispersion interactions and the cost of cavity formation in the sorbent phase. The s coefficient is related to the sorbent phase dipolarity and polarizability. Similarly, the r coefficient is related to polarizability. The a and b coefficients, being complementary to the vapor hydrogen-bond acidity and basicity, represent the sorbent phase hydrogen-bond basicity and acidity, respectively. Thus, the b coefficient can be used to characterize hydrogen-bond acidic polymers. A preliminary equation for FPOL was provided in 1991,22 and LSER equations for four fluoroalcohol polymers were published in 1995 as part of a study of 14 sorbent phases at 25 °C.44 This set included FPOL (Scheme 1), PSpFA (Scheme 2), SXFA (Scheme 6), and a hydroxy-terminated Fomblin Z-dol liquid. Also published in 1991 was a paper examining a large set of phenolic liquids, such as F-BSP, which is the fluorinated bisphenol monomer shown in Schemes 7 and 8. Results of these studies were also summarized in a 1998 paper that introduced the fluoroalcohol-substituted siloxane PSF6, shown in Scheme 9. LSER coefficients for PSF6 were determined at temperatures from 81 to 171 °C. In 2000, Chehimi et al. reported gas chromatographic characterization and LSER coefficient determination for PLF (Scheme 11) and PBF at 35 °C. The LSER coefficients for selected polymers and bisphenols at 25 °C are summarized in Table 2. We are particularly interested in hydrogen-bonding properties, given by the a and b coefficients, which like their corresponding solvation parameters scale equivalently with free energy. (However, the a and b coefficients cannot be compared directly with the l coefficient, for example, which scales differently) Three low-polarity polymers, whose structures are shown in Scheme 17, are included for comparison. The fluoroalcohol polymers all have large b coefficients, indicating their hydrogen-bond acidity, while the a coefficients are low. By comparison, the low-polarity polymers without fluorinated alcohol or phenolic groups lack significant b coefficients. Like the fluorinated alcohol polymers, the fluorinated bisphenol F-BSP also has a large b coefficient and a low a coefficient. By contrast, the unfluorinated bisphenol H-BSP has a much smaller b coefficient than the fluorinated bisphenol and is actually a stronger hydrogen-bond base than

Hydrogen-Bond Acidic Polymers Scheme 17

acid, as noted previously. These studies confirm the importance of fluorination in achieving high hydrogen-bond acidity and simultaneously lowering hydrogen-bond basicity. The latter factor improves overall selectivity and simultaneously reduces self-association; reduced self-association then also contributes to stronger hydrogen-bond acidity. For a given vapor/polymer combination, the LSER expresses the log of the partition coefficient as a linear combination of terms. Given the respective solvation parameters and coefficients for a specific vapor-polymer pair, the magnitude of each term can be calculated and compared to determine which interactions make the largest contributions. While dispersion interactions are nearly always important, the hydrogen-bonding interaction between a strongly hydrogenbond acidic polymer and a moderately to strongly basic vapor can also be quite significant.26 Table 3 illustrates the determination of interactions using SXFA as a representative hydrogen-bond acidic polymer and four vapors. Polarizability and dipolarity interactions can be taken as the sum of (rR2 + sπH2 ), where the rR2 term is normally a small correction to the overall dipolarity/ polarizability interaction indicated by sπH2 . The hydrogenbonding terms aΣRH2 and bΣβH2 represent hydrogen-bonding interactions where the polymer is a base or an acid, respectively. It is difficult to separate the dispersion interactions favoring sorption from the cost of forming a cavity. Together, these can be best represented as the sum of the regression constant c and the l log L16 term.92 It is clear in Table 3 that when the polymer is an acid (SXFA) and the vapor is a base (e.g., triethylamine, DMF, or ethanol), the hydrogen-bonding term is a significant contributor to the magnitude of log K. Dispersion interactions are significant contributors for most vapors from compounds that are condensed liquids at room temperature. Sensor signals can also be used to develop LSER equations for sorbent polymers if the response is proportional to the amount sorbed as a mass, as shown above in eq 6, since responses are directly proportional to the partition coefficient without any analyte-specific sensitivity factors. Thus, the responses of a polymer-coated QCM or SAW could be used to develop an LSER equation for the polymer if the responses are purely gravimetric. For a SAW sensor with a modulus contribution to the response, this condition is not strictly true, and variations in vapor-specific volumes will cause the observed responses to vary in their proportionality to K values (see above, eq 8). In this regard, determining LSERs from SAW sensor responses is somewhat less rigorous than using true partition coefficients or responses of purely gravimetric sensors. Nevertheless, it has been shown empirically that good correlations can be obtained between SAW sensor responses and vapor solvation parameters, providing LSER coefficients that characterize the polymer solubility properties in a sensible way. This approach was first shown by Zellers in 1993 for four polymers, none of them hydrogen-bond

Chemical Reviews, 2008, Vol. 108, No. 2 737

acids.125 Hierlemann, Zellers, and Ricco further discussed the use of SAW sensor responses for determining LSER coefficients in 2001.118 Also in 2001, Grate et al. reported LSER equations determined from SAW sensor responses for 14 polymers that were “well-behaved” as sensor coatings and 6 less well-behaved materials.57 Among the well-behaved polymers were three hydrogen-bond acidic polymers, including SXFA (Scheme 6), BSP3 (Scheme 7), and a BSP3 analog with longer oligosiloxane segments. The b coefficients for these three polymers were significantly higher than any of the other well-behaved polymers, consistent with the hydrogenbond acidity expected from the fluorinated functional groups. The first use of QCM devices to obtain data for determination of LSER coefficients was reported in 2001 by Hierlemann et al. Results for six polysiloxanes were reported, none with fluorinated alcohol or phenol functionality.118 One polysiloxane with 10% carboxylic acid groups was listed as an acidic polymer; however, the LSER analysis showed little hydrogen-bond acidity and much greater hydrogen-bond basicity. These results reaffirm the principle that fluorinated alcohols or phenols are preferred for obtaining hydrogenbond acidity in a sorptive polymer for the purpose of sorbing hydrogen-bond basic vapors22 and provides experimental evidence that pendant carboxylic acids do not result in effective hydrogen-bond acidic polymers for vapor sensing. The latter result is consistent with the notion that carboxylic acids will self-associate. The conventional use of LSERs in sensor research has been to characterize sorptive polymers or develop LSER equations that can be used to predict partition coefficients and/or sensor responses based on the polymer parameters and vapor solvation parameters. Grate and Wise proposed instead that the responses of an array of sensors could be used to determine the solvation parameters of the sensed vapor.52 Given an array with a suitably diverse set of known coatings and gravimetric sensor responses, the array pattern vector could be transformed into a vector containing the solvation parameters as descriptors of the vapor. In this way, an array detecting a vapor that had not been included in a prior training set might be able to characterize the vapor in terms of these descriptors and suggest what vapor it might be based on comparison of the found descriptors with those for known vapors. In order to be a diverse array, hydrogenbond acidic polymers would be required. A method similar to classical least-squares (CLS) calibration (often used in spectroscopy) was derived for simultaneously obtaining the full set of descriptor values from the array response vector. The approach requires that the interactive properties of the sorbent sensing layers be known and quantified as LSER coefficients (polymer parameters). In addition, inverse least-squares (ILS) methods could be used to process the array response vectors, in which case individual models are developed for each vapor descriptor. The ILS approach does not require advance knowledge of polymer parameters, but it does require that an adequate calibration data set be available to derive the ILS models. Once the CLS or ILS models are developed, an array might be used to characterize an unknown vapor in terms of its descriptor values, even if the specific unknown vapor had not been in the training set. This approach stands in contrast to most conventional pattern recognition approaches that are based on matching patterns from unknowns to patterns from known compounds in the training set.

738 Chemical Reviews, 2008, Vol. 108, No. 2

Grate

Table 3. Calculated Interaction Termsa for Four Vapors Sorbed by SXFA, a Hydrogen-Bond Acidic Polymer vapor

dipolarity/polarizability (rR2 + sπH2 )

n-hexane triethylamine DMFc ethanol

0.05 0.64 0.15

hydrogen bonding, a, ΣRH2

hydrogen bonding, b, ΣβH2

dispersion/cavity (c + l log L16)

partition coefficient,b log K

0.26

3.36 3.15 2.04

1.84 2.10 2.20 0.99

1.84 (1.74) 5.51 5.99 3.44 (3.54)

a No data indicate terms calculated to be zero. b These are calculated values, each derived from the sum of the interaction terms plus the constant. Values in parentheses are measured values. c N,N-Dimethylformamide.

Figure 4. Calculated vs reference values for the ΣβH2 vapor solvation parameter for calibration from six-factor PCR models. Reprinted with permission from ref 57. Copyright 2001 American Chemical Society.

These approaches for converting array responses to chemical information were initially derived for mass-transducing sensors such as acoustic wave sensors,52 extended to volumetransducing sensors17 such as chemiresistors with carbon particle/polymer composite sensing layers,126,127 and then extended to mass-plus-volume-transducing sensors.8,16,57A polymer-coated SAW device can be regarded as a massplus-volume transducing sensor if the response includes a modulus contribution as well as a gravimetric contribution. Actual SAW array response vectors were used to derive ILS models that could correlate and predict vapor solvation parameters.57 Successful correlations can be developed by multiple linear regression (MLR), principal components regression (PCR), and partial least-squares (PLS) regression. The best fits to the training data were obtained using MLR; however, cross-validation indicated that prediction of vapor descriptors for vapors not in the training set was significantly more successful using PCR or PLS. The array data included responses from sensors coated with strongly hydrogen-bond acidic polymers such as SXFA (Scheme 6) and BSP3 (Scheme 7). Having at least one of these types of polymers in the array is essential for obtaining good information about a vapor’s hydrogen-bond basicity as represented by the ΣβH2 solvation parameter. Figure 4 shows the correlation between the ΣβH2 values found for vapors in the training set, using six-factor PCR models to process the array data, and the known reference ΣβH2 values for those vapors. The correlation is very good.

6. Acoustic Wave Sensors and Arrays A significant portion of the development of acoustic wave sensor arrays has been focused on surface acoustic wave (SAW) devices. Wohltjen introduced these devices as the basis for chemical vapor sensors,128,129 and they have since

been investigated by several groups.2,11,12,15,53,125,130-141 The SAW device is a member of a family of devices called acoustic wave devices that include the QCM (also referred to as a thickness shear mode or TSM device), surface transverse wave devices, Love wave devices, flexural plate wave devices, Leaky SAW devices, and shear horizontal acoustic plate mode devices. Acoustic wave devices have been reviewed in many prior treatments,9-11,20,22,142-154 and acoustic wave devices in arrays for chemical vapor sensing were specifically reviewed in this journal.2 Polymer-coated acoustic wave devices are sorption-based sensors as shown in Figure 1. They detect the mass loading and sometimes the modulus changes that occur upon vapor sorption in the polymer film. The signal measured is typically a resonant frequency. Hydrogen-bond acidic polymers are relevant in two primary senses. First, they afford sensitivity to hydrogen-bond basic analytes of interest such as chemical agents and explosives (see below), and second, they help to ensure the diversity of chemical selectivities in a sensor array.22 Diverse chemical selectivities help to ensure that the array collects as much chemical information as possible about the sample by probing all of the available interactions and solubility properties that can be used to distinguish one vapor from another. Hydrogen-bond acidic polymers ensure that the array responds to the hydrogen-bond basicity of sorbed vapors. Such arrays can be useful for a variety of volatile organic compound detection applications. While generally useful for this purpose, the importance of including a hydrogen-bond acidic polymer in an array depends on the analytical task the array is expected to perform.2,51

6.1. Chemical Agent Detection Hydrogen-bond acidic polymers emerged as key coatings for acoustic wave sensors to promote the sorption and detection of organophosphorus chemical agents and their simulants. The principle of hydrogen bonding in this application was shown in Figure 2 above. Organophosphorus compounds are particularly strong hydrogen-bond bases; the ΣβH2 value for DMMP, for example, is 1.05. For comparison, the ΣβH2 values for other basic volatile organic compounds such as nitromethane, acetone, and ethyl amine are 0.27, 0.49, and 0.61, respectively. Much of the synthetic work described in the preceding sections was motivated by this application need. Typically, DMMP or diisopropyl methylphosphonate (DIMP) was used as a simulant, although in a few cases responses to actual chemical agents such as GD (soman) or GB (sarin) have been reported.37,38,43,84,105 The benefit of including a polar hydrogen-bond acidic functional group for improving sensitivity to organophosphorus compounds can be seen in Figure 5. This plot shows the calibration curves for SAW sensors coated with two hydrogen-bond acidic polymers, FPOL and BSP3 plotted with solid lines, compared with sensors coated with nonacidic

Hydrogen-Bond Acidic Polymers

Figure 5. Calibration curves for organophosphorus compound DMMP as sensed by polymer-coated SAW sensors, comparing the hydrogen-bond acidic polymers BSP3 and FPOL with nonacidic polymers PDMS, SXPH, and PECH.

polymers PECH, SXPH, and polydimethylsiloxane (PDMS) shown with dashed lines. The PDMS polymer is a nonpolar polymer containing the same repeat units that are found in the oligosiloxane segments of BSP3 and hence illustrates DMMP sorption by a polymer if the polar organic groups are left out altogether. Promotion of sorption and response by including hydrogen-bond acidic functional groups is clearly evident and confirmed by including other nonacidic but somewhat more polar polymers, PECH and SXPH (see Scheme 17 for their structures). Most acoustic wave sensors for agent detection have been based on SAW sensors; however; there are reports using other acoustic wave devices such as Love wave sensors.104,105 A number of reports give an indication of the sensitivity of polymer-coated SAW sensors to organophosphorus compounds. Using FPOL (Scheme 1) on 158 MHz SAW devices, signals of over 1000 Hz were reported at 1 mg/m3 DMMP.43 Using BSP3 (Scheme 7) on 200 MHz SAW devices, signals of over 20 000 Hz were observed in response to 8 mg/m3. Taking the noise to be about 3 Hz, for a minimum detectable signal of about 10 Hz at a signal-to-noise ratio of 3, the detection limit would be 0.004 mg/m3 or about 1 ppb by volume.49 Using SXFA (Scheme 6) on a 250 MHz SAW device, responses of 65 000 Hz to10 mg/m3 DMMP were reported.68 SAW sensors (500 MHz) freshly coated with HB-4 (Scheme 14), which has phenolic functionality similar to BSP3, were reported to give responses of 3629 Hz to 0.5 mg/m3 DMMP.108 Signals and detection limits depend on several factors, including the sensor temperature, SAW device frequency, noise, and polymer coating thickness, which vary among different studies. These data therefore provide indications of the high sensitivity without necessarily providing rigorous quantitative comparisons. Value as a SAW coating also depends on other factors such as the quality of the thin film that can be prepared on sensor. The variety of polymers shown in the previous sections provides a wealth of choices for developing effective sensors. One group reported that SAW sensors coated with hydrogen-bond acidic polymers rapidly lose their sensitivity

Chemical Reviews, 2008, Vol. 108, No. 2 739

Figure 6. Responses of two electrospray BSP3-coated SAW sensors to DIMP, initially and over 2 months later.

to basic compounds after the initial coating application and testing.108,113 These authors spray-coated polymer films onto 500 MHz SAW devices and tested with DMMP or DNT. In one paper, it was reported that a BSP3-coated SAW device lost 90% of its sensitivity in just 10 days.113 Indeed, all their hydrogen-bond acidic polymer-coated sensors showed significant declines in sensitivity with time. By contrast, Grate and Rapp examined BSP3 on Rapp’s SAW sensor system.155,156 Coatings were applied by electrospray157 to 434 MHz SAW sensors. As shown in Figure 6, these sensors did not show a significant aging effect. Thus, it would appear to be unlikely that poor aging behavior represents an intrinsic property of the hydrogen-bond acidic polymers themselves, as opposed to an operational property of a polymer film on a surface.

6.2. Acoustic Wave Sensor Array Systems In sensor systems with preconcentrators, even greater operational sensitivities and lower detection limits can be achieved relative to those for direct sensing as just described. Indeed, much of the development of SAW sensors for agent detection has focused on sensor arrays and sensor arrays preceded by preconcentrators. More recent systems have also placed gas chromatographic separation columns between the preconcentrator and the SAW array. In the first investigations of SAW sensor arrays with pattern recognition analysis,33,40 FPOL was included in the polymer coatings and provided the highest sensitivity to simulant DMMP. On the basis of hierarchical cluster analysis, FPOL was one of the most distinctive polymers in the data set, and it was among the coatings selected for a subset of four polymers that could successfully discriminate between one class of vapors including chemical agent simulants and another class of vapors containing a diversity of potential interferences.33 This work eventually lead to a complete prototype SAW array system with automated sample preconcentration using FPOL as the agent-sensitive coating.43 This was the first example of an acoustic wave sensor array system with on-board preconcentration. The Sandia microChemLab system has integrated a microfabricated preconcentrator, a microfabricated chromatographic column, and an array of polymer-coated SAW

740 Chemical Reviews, 2008, Vol. 108, No. 2

sensors. Use of BSP3 and DKAP (Scheme 12) polymers has been essential to the success of this platform in detecting DMMP and chemical agents.84,136 Zellers developed portable gas chromatographic instruments for chemical vapor analysis using SAW array detectors and included BSP3-coated sensors in the array.83,85 Similarly, he included BSP3 in a personal vapor monitor based on SAW sensors.87 Flexural plate wave acoustic sensors and systems have also used hydrogen-bond acidic polymers.54,55,72

6.3 Explosives Detection Detection of nitroaromatic compounds as simulants for nitroaromatic explosives such as trinitrotoluene (TNT), using a sensor coated with a hydrogen-bond acidic polymer, was first described by McGill et al. These authors used SXPHFA (Scheme 10) as the polymer to promote sorption of the nitroaromatic compounds through hydrogen bonding with basic nitro groups.93,94 In tests using 250 MHz SAW sensors, the authors observed responses of 8500 Hz to 400 ppb of 2,4-dinitrotoluene (2,4-DNT), which was generated by passing nitrogen through a column of 2,4-DNT-coated sand. The authors extrapolated these results to a detection limit of 0.235 ppb at a signal-to-noise ratio of 3. Infrared spectroscopic studies of SXPHFA containing sorbed nitrobenzene (NB) were carried out to support hydrogen bonding between nitro groups and the polymer hydroxyl groups. Free hydroxyl groups declined (but did not disappear), and hydrogenbonded hydroxyl groups increased, as shown in difference spectra between the neat polymer and polymer containing sorbed NB. McGill’s group provided more results on explosives detection on SAW devices in 2001, including IR spectroscopic studies of polymers with and without sorbed NB or DMMP.61 Both of these basic sorbates reduced the free hydroxyl stretch. The OH stretch for the hydroxyl group hydrogen bonded to analyte appeared at lower wavenumbers, with DMMP shifting the stretch significantly farther as a result of its stronger hydrogen-bond basicity. Response behaviors of 250 MHz SAW devices coated with CS3P2 (Scheme 10) and CS6P2 (a similar polymer with a longer methylene chain between silicon atoms) were presented and discussed. These polymers had similar responses to DNT in tests at 31 ppb, and a detection limit of 0.095 ppb was extrapolated at a signal-to-noise ratio of 3. The nitroaromatic compound DNT has also been used in the evaluation of the POSS compounds shown in Schemes 15 and 16, which were blended with linear polymers to prepare composite SAW sensors coatings.113 When explosives are compared to organophosphorus compounds with regard to sorption to hydrogen-bond acidic polymers, the relative roles of hydrogen-bonding and dispersion interactions may differ. The effective solvation parameter ΣβH2 for NB, DNT, and TNT has been reported to be 0.28, 0.47, and 0.61, respectively.93 For comparison, nitromethane has a value of 0.27, similar to NB. The nitro group is not intrinsically a very strong hydrogen-bond base, whereas the phosphoryl group of an organophosphorus compound is quite a strong hydrogen-bond base (e.g., DMMP has a ΣβH2 value of 1.05). On the other hand, explosives have much higher log L16 values than DMMP. This parameter is related to promotion of sorption by dispersion interactions. For example, the log L16 values for TNT and 2,4-DNT are 7.85 and 6.26, respectively, whereas that for DMMP is only 3.75. In general, nitroaromatics are more prone to interact by

Grate

dispersion interactions and less prone to interact by hydrogen bonding than typical organophosphorus compounds related to chemical agents. It is also the case that nitroaromatics have large R2 and πH2 parameters, indicating a very good ability to interact by dipole-dipole and dipole-induced dipole interactions.

7. Microcantilever Sensors Micromachined cantilever structures represent a more recent transducer for chemical and biological sensors and arrays.158-160 The cantilever structure is supported at one end by the bulk chip material and extends either over an etch pit or over the end of the chip like a diving board. These structures can be made in various sizes with 200 µm long by 50 µm wide by 1 µm thick being a representative size. Several such devices can be fabricated as part of a single chip for array-on-a-chip configurations. Two transduction mechanisms are typically used, either the resonant frequency shift or a bending mode observation. The resonant frequency is shifted by mass loading on the surface or in a surfaceapplied film. Bending arises not from the weight of collected analyte but rather from the effect of the sorbed analyte on surface stresses, leading to bending. Microcantilever movements can be monitored with optical methods or operated using capacitive, piezoresistive, or piezoelectric methods. Thundat et al. reported detection of 2,4-DNT using a polymer-coated microcantilever in 2004.70 A commercial “V”-shaped cantilever was coated by SXFA (Scheme 6) using matrix-assisted pulsed laser evaporation (MAPLE)69,161 to deposit a uniform film of the polymer on one side of the cantilever. The other side of the cantilever was coated with gold, which facilitated reflection of a laser bean for optical detection of cantilever motions. The film thickness was reported to be roughly 600 nm. 2,4-DNT concentrations obtained by passing carrier gas over a temperature-controlled 2,4-DNT sample were determined by trapping a known volume on a Tenax trap and analyzing the amount of sorbate by gas chromatography with mass spectrometric detection. Reversible and repeatable responses at 45 ppb 2,4-DNT were observed in 5 s exposures, leading to an observed sensitivity of 4.5 nm/ppb. These results were extrapolated to a detection limit of 0.300 ppb at a signal-to-noise ratio of 3. Several advantages of using the SXFA polymer were noted. Bending mode responses were found to be critically dependent on the film thickness, with coatings of 150 nm or less yielding very small responses compared to those with the 600 nm film. This paper also presented mass-loading responses from resonant frequency shifts. Frequency decreased with added mass as expected. The signal-to-noise for this mass-transduction mechanism was clearly less than that obtained in the bending mode response for the type of cantilevers being used. This group has also described detection of nitroaromatic compounds using cantilevers coated with a monolayer of 4-mercaptobenzoic acid.162 A cantilever beam sensor with a hydrogen-bond acidic coating for chemical agent detection has also been described.111 A sophisticated device design for operating the cantilever in a resonant frequency mass detection mode was developed, incorporating an electrostatic actuation mechanism to drive the beam and a piezoresistive transduction method. The device also incorporated an on-beam heater. A functionalized polycarbosilane dubbed HCSA2163 was deposited from dilute chloroform solution using a piezoelectric

Hydrogen-Bond Acidic Polymers

Chemical Reviews, 2008, Vol. 108, No. 2 741

film are not necessarily known, it is clear that the amount sorbed in the film (see eqs 3-5) is a dominant influence on the sensitivity and selectivity among the tested vapors. The Ni PCCP sensor was very sensitive to DMMP, with responses at 17 mg/m3 providing a signal-to-noise of 684, suggesting an extrapolated detection limit on the order of 0.1 mg/m3 at a signal-to-noise of 3. The strategy of using the sorptive polymer to influence the sensitivity and selectivity of a conducting particle/insulating polymer composite on a chemiresistor sensor has also been used in the development of arrays of carbon black/polymer composite chemiresistors.126,165 FPOL and SXFA have been used as overcoats on conductive polymer fabric sensors in an effort to promote sensitivity and selectivity for organophosphorus compounds.47 However, the fluoroalcohol-containing polymers did not increase the responses to DMMP.

8.2. Chemicapacitors

Figure 7. Correlation of the relative sensitivities of a NiPCCP/ FPOL chemiresistor to a range of vapors with the tendency of those vapors to be sorbed by FPOL as indicated from measurements on SAW devices. Figure plotted from data in ref 41.

inkjet dispensing head. Response to dilute DMMP at 0.1 mg/m3 or 20 ppb was reported to be 30 Hz at a 10 Hz noise level. Hence, this concentration represents the detection limit at a signal-to-noise ratio of 3.

8. Sensors Responding to Electrical Properties In this section, we consider sensors that respond to conductivity changes or capacitance changes and include a hydrogen-bond acidic polymer as a component.

8.1. Chemiresistors In 1990, Grate and co-workers at NRL described the use of FPOL-phthalocyanine composite films on interdigitated electrodes at room temperature as chemiresistor sensors for organic vapors.41 At a constant test voltage, these sensors respond with a change in current as sorbed vapors alter the resistance of the sensing film. Chemiresistor sensors with phthalocyanine films were already well known. This paper sought to use the FPOL in the film to promote the sorption of organophosphorus vapors and hence influence the sensitivity. The phthalocyanines were tetrakis(cumylphenoxy)phthalocyanines (PCCP) with various metal ions; Ni PCCP provided the most sensitive sensors in response to DMMP. The films were applied by the Langmuir-Blodgett (LB) technique, resulting in contiguous nanometer-scale discshaped crystallites of the semiconducting phthalocyanine in a matrix of FPOL.164 Thus, this sensor represents dispersed conducting particles as a first component in sorbent insulating polymer matrix as a second component. Indeed, it was found that response sensitivity was strongly correlated with the sorption of organic vapors in the FPOL matrix material. Figure 7 shows the log of the relative sensitivity of the Ni PCCP/FPOL chemiresistor for a range of vapors against a measure of the sorption of those vapors by FPOL as determined on a SAW device,7 taking the data from a table in ref 41. While all the analyte-specific factors that may influence the response of a composite Ni-PCCP-containing

Chemicapacitors166,167 represent a relatively new platform for the use of hydrogen-bond acidic polymers.71,74,75 Patel and Mlsna described microfabricated devices in parallel plate and elevated interdigitated electrode configurations. The latter design provides excellent access of vapors to the polymer coated on the electrodes. The parallel plates are designed with openings in the “top” plate for polymer deposition into the capacitor gap and vapor access to the polymer. In chemical agent tests it was observed that the interdigitated design was faster to respond but slightly less sensitive than the parallel plate design. Chemicapacitors respond to changes in the permittivity within the sensed volume of the capacitor that contains the sorptive polymer. Changes in the polymer due to vapor sorption (e.g., swelling) and analyte-specific dielectric properties influence the observed response. Polymers are typically applied from solution using a piezoceramic inkjet head. Using a variety of polymers, including some hydrogen-bond acidic polymers, these investigators reported detection limits for many volatile organic compounds, warfare agent simulants, and nitroaromatic compounds. In one study using parallel plate capacitors,71 SXFA-coated devices were reported to respond to DMMP at 0.18 ppm (ca. 1 mg/m3) with a signal-to-noise of 300, leading to an extrapolated detection limit of 2 ppb (ca. 0.01 mg/m3) at a signal-to-noise of 3. A detection limit of 0.1 ppb was indicated for nitrotoluene. SXFA was also shown to give low ppm detection limits for many volatile industrial solvent vapors. In subsequent work, these authors reported an extrapolated limit of detection of 2 ppb for 2,6-DNT using SXFA.74 In actual chemical agent tests, limits of detection of 0.047 mg/m3 for GD using SXFA and 0.4 mg/m3 for GB using HC were reported. Although hydrogen-bond acidic polymers are useful for nitroaromatic compound detection on this platform, this class of polymers is not the only choice that provides high sensitivity to nitroaromatics, at least as far as chemicapacitive sensing is concerned. Polymers such as OV-225 and OV-275 have been shown to provide sensitive sensors.75 OV-275 is a polysiloxane that contains dipolar nitrile groups, while OV-225 contains nitrile groups and phenyl substituents. These dipolar and polarizable groups apparently promote sensitivity through dipole-dipole and dipole-induced dipole as well as dispersive interactions. Chemicapacitors have also been reported by Snow and Houser that are based on semiconducting single-walled

742 Chemical Reviews, 2008, Vol. 108, No. 2

carbon nanotubes prepared as an electrically continuous network acting as one plate of the capacitor. The other electrode was a heavily doped silicon substrate, and the two were separated by a thermal oxide layer.112 Fringing electric fields radiate outward from the ca. 1 nm diameter nanotubes when a bias is applied. The fields are strongest at the nanotube surfaces where adsorbates can be detected as a change in the capacitance. The hydrogen-bond acidic polymer HC was applied to the sensor in a ca. 100 nm thick layer to promote responses to organophosphorus compounds. Compared to an uncoated sensor, the polymer-coated sensor gave responses to DMMP that were 500 times greater. The detection limit was estimated to be 0.5 ppb (0.0025 mg/m3) with a response time of 370 s to reach 90% of response. A monolayer with a pendant hexafluoroisopropanol functionality was fabricated on nanotube surfaces to provide a thinner chemoselective coating; while the responses were faster, it was 100 times less sensitive. These authors previously described chemiresistors and transistors using carbon nanotube networks for chemical vapor sensing.168 While HC was mentioned in this work, it was used as a component of an upstream filter to remove response to DMMP rather than as a chemoselective sensor layer.

9. Optical and Luminescent Sensors Polymers are important components to many optical sensors, either as a sorbent layer or as a matrix for other functional components. Polymers related to BSP3 were coated onto optical fibers for potential sensing applications.49 A polymer formulation including cross-linker and hydrosilylation catalyst was delivered to a cladding cup. A glass fiber freshly pulled from the melt passed through the cladding cup, a furnace, and finally was collected onto a roller. The polymer formulation coated onto the fiber was cured in the furnace. The best sections of coated fiber had 25 µm thick polymer coatings on 180 µm diameter cores and guided light as well as similar fibers coated with PDMS, a standard cladding material. A number of polymers, including FPOL, were examined as sorbent matrixes for Reichardt’s betaine, a solvatochromic dye whose absorbance peak shifts significantly in response to polarity changes in the local solvent environment.62 The spectra due to inclusion in the polymer environment were examined as well as spectral shifts arising from the sorption of vapor in the polymers. FPOL was a rather unique polymer among those tested because of its strong interaction with the dye, and as a result, the responses of the dye in FPOL were also distinctive. FPOL has also been investigated as a matrix for chemiluminescent reagents (luminol, KOH, and ferric ion catalyst) used in developing sensors for detection of oxygen or nitrogen dioxide.45 Use of hydrogen-bond acidic polymers as components of fluorescence sensors appears to be particularly promising. Nile Red, another solvatochromic dye well known for use in chemical vapor sensors,169-173 was incorporated into films of BSP3 and PSmFA.82 Additional experiments were carried out using poly(methyl methacrylate) (PMMA) as a control non-hydrogen-bond acidic matrix polymer. Several fluorescent dyes in each of these polymers were evaluated for responses to DMMP vapor. Of all the combinations, Nile Red in BSP3 gave the strongest responses with a large blue shift and strong fluorescence enhancement. Large (relative to the noise) fluorescence signals were observed at DMMP

Grate

vapor diluted to 100-200 ppb (0.5-1 mg/m3) concentrations. Extrapolated detection limits were not reported. The response behavior was interpreted in terms of a competition between Nile Red and DMMP for hydrogen-bonding sites on the polymer. The interaction of the Nile Red by hydrogen bonding with hydroxyl groups in BSP3 greatly reduced the dye fluorescence. Displacement of basic dye functional groups by DMMP, a stronger hydrogen-bond base, freed the dye from the quenching hydrogen bond and the fluorescence light output increased, providing the analytical signal. In a subsequent study, Nile Red and a second dye were incorporated into BSP3 to set up a fluorescence signal enhancement system.81 The phenylene-ethynylene polymer shown in Scheme 5 is a fluorescent polymer with hydrogen-bond acidic groups pendant to the conjugated polymer chain.65 These polymers with and without hexafluoroisopropanol substituents were coated on the inside of a glass capillary, and their fluorescence was monitored in response to vapors. Response consists of a fluorescence quenching effect. For the nitroaromatic compound 2,4-DNT, similar responses were observed regardless of whether pendant hexafluoroalcohol groups were present or not. The strong electrostatic interaction between the nitroaromatic compound and the phenylene-ethynylene polymer apparently overwhelms any influence the hydrogenbond acidic groups may have. On the other hand, pyridine induced strong poorly reversible decreases in fluorescence only in the polymers that contained the fluoroalcohol substituents. The authors suggested that the electron-deficient, strongly hydrogen-bonded pyridinium species in the film could undergo photoinduced charge-transfer reactions, a mechanism supported by additional experiments on more electron-rich substituted pyridines that did not lead to a large fluorescence response.

10. Separations and Preconcentration As noted above, hydrogen-bond acidic polymers developed for chemical vapor sensors have been used as gas chromatographic stationary phases in the investigation of the polymer solubility properties and development of LSERs. However, in the case of PSF6 (Scheme 9), the phase was developed primarily for gas chromatography.92 These authors noted the lack of hydrogen-bond acidic polymers for chromatographic purposes and set out to prepare a stationary phase with the desired solubility properties and high thermal stability. The material was routinely used at temperatures from 50 to 200 °C. It was an efficient material providing 1500-1800 theoretical plates per meter of packed column, which was noted to be similar to other conventional poly(siloxane) stationary phases on the same support material. The desired hydrogen-bond acidity was achieved, and although hydrogenbond acidity decreases with increasing temperature, the phase still retained significant and useful hydrogen-bond acidity at 200 °C. It had essentially no hydrogen-bond basicity, as is also desirable. The polymer BSP3, originally prepared as a sensing polymer, has been used as the sorbent phase in solid-phase microextraction (SPME) studies for chemical agent analysis.80 This technique involves a polymer-coated fiber that is equilibrated with a sample to extract the analyte or analytes of interest. It is then retracted into a syringe needle and inserted into a gas chromatographic injection port, where the fiber is pushed out of the needle and analytes are released into the instrument by thermal desorption. The vast majority

Hydrogen-Bond Acidic Polymers

of the initial SPME studies used PDMS as the polymer. Packed column chromatographic studies demonstrated that BSP3 polymer had much greater affinity for GB than PDMS. To demonstrate SPME with this polymer, BSP3-coated fibers were used to sample GB in the headspace above a hexane solution and then desorbed into the gas chromatograph at 150 °C. These fibers could be used repeatedly, and they exhibited 10-20-fold higher affinity for sarin than PDMS fibers. A hydrogen-bond acidic polymer, dubbed HCSA2, has been applied as a sorbent to a microfabricated thermally desorbed vapor preconcentrator.110 The device was fabricated as a flow-through microhotplate array with the polymer applied to the hot plate using an inkjet device. The polymer was a hyperbranched polycarbosilane with pendant hexafluoroisopropanol groups.163 Vapors collected by sorption in the polymer could be released by heating the preconcentrator to 180 °C in just 40 ms. The device was interfaced as a sampling front end to commercial ion mobility spectrometer instruments. In DMMP analyses with 60 s collection times, uncoated devices provided no measurable sample enrichment while the polymer-coated device increased the signal by six times. In explosives detection tests using TNT, a signal increase of three times was observed in preliminary tests. The sample enrichment factors appear to be based on detector peak heights rather than areas, so these numbers may underestimate the actual enrichment that was achieved.

11. Discussion Fluorinated hydrogen-bond acidic polymers for chemical sensing began with organic polymers for acoustic wave sensors. The polymers were not commercially available, and scientists wishing to use them needed to make or borrow samples. With the introduction of silicon-containing polymers with hydrogen-bond acidic groups, there has been a significant expansion in the numbers of polymers and polymer architectures that have been developed. Macromolecules in this class now range from linear polymers to hyperbranched materials and even POSS nanoarchitectures. Many of these are easier to make than the early polymers such as FPOL, fluorinated reagents and monomers are now more readily available, and some of the polymers can be obtained commercially. With increasing recognition and increasing availability, these polymers have been applied to many other sensing approaches. Sensitive sensors for explosives, chemical agents or simulants, and/or volatile organic compounds have been developed using these polymers on various acoustic wave devices, chemiresistors, chemicapacitors, and microcantilevers and in fluorescence sensing methods. These materials are also finding their way into separation and preconcentration applications. Creation of functionalized nanomaterials, other than polymers, with hydrogen-bond acidic groups is also beginning. The POSS materials113 have just been mentioned, and carbon nanotubes modified with hexafluoroisopropanol groups112 have also been cited. Gold nanoparticles have been prepared with monolayers that have terminal phenolic groups or fluoroalcohol groups.174,175 Undoubtedly, more such nanomaterials with fluorinated hydrogen-bond acidic groups will be developed in the future.

12. Acknowledgments The author would especially like to acknowledge former co-workers and collaborators who played key roles in the

Chemical Reviews, 2008, Vol. 108, No. 2 743

development of polymer-coated acoustic wave sensors, hydrogen-bond acidic polymers, and LSERs, including Hank Wohltjen, Arthur Snow, Michael Abraham, and Steven Kaganove. The author thanks Arthur Snow for additional discussions on polymer structures during the preparation of this manuscript. Michael Rapp is acknowledged for providing independent sensor testing data for BSP3. The author gratefully acknowledges funding from the U.S. Department of Energy, National Nuclear Security Administration, Office of Nonproliferation Research and Development (NA-22), for past funding and Laboratory Directed Research and Development funds of the U.S. DOE, administered by the Pacific Northwest National Laboratory, for current funding. The William R. Wiley Environmental Molecular Sciences Laboratory, a U.S. DOE scientific user facility operated for the DOE by PNNL, is also acknowledged. The Pacific Northwest National Laboratory is a multiprogram national laboratory operated for the U.S. Department of Energy by Battelle Memorial Institute.

13. References (1) Chemical agents and simulant concentrations are typically reported in mg/m3, the international standard units for gas-phase concentrations. These units will be used here, and conversions for concentrations originally reported in ppb will be provided. Explosives, however, have been typically reported in ppb, and these units will be retained for this class of analytes. (2) Grate, J. W. Chem. ReV. 2000, 100, 2627. (3) Janghorbani, M.; Freund, H. Anal. Chem. 1973, 45, 325. (4) Edmunds, T. E.; West, T. S. Anal. Chim. Acta 1980, 117, 147. (5) McCallum, J. J.; Fielden, P. R.; Volkan, M.; Alder, J. F. Anal. Chim. Acta 1984, 162, 75. (6) Snow, A.; Wohltjen, H. Anal. Chem. 1984, 56, 1411. (7) Grate, J. W.; Snow, A.; Ballantine, D. S.; Wohltjen, H.; Abraham, M. H.; McGill, R. A.; Sasson, P. Anal. Chem. 1988, 60, 869. (8) Grate, J. W.; Wise, B. M.; Gallagher, N. B. Anal. Chim. Acta 2003, 490, 169. (9) Grate, J. W.; Martin, S. J.; White, R. M. Anal. Chem. 1993, 65, 987A. (10) Grate, J. W.; Martin, S. J.; White, R. M. Anal. Chem. 1993, 65, 940A. (11) Grate, J. W.; Frye, G. C. In Sensors Update; Baltes, H., Goepel, W., Hesse, J., Eds.; VSH: Weinheim, 1996; Vol. 2, p 37. (12) Martin, S. J.; Frye, G. C.; Senturia, S. D. Anal. Chem. 1994, 66, 2201. (13) Martin, S. J.; Frye, G. C. Technical Digest of the 1992 IEEE Solid State Sensor and Actuator Workshop, IEEE, Hilton Head Island, South Carolina, June 22-25, 1992, pp. 27-31. (14) Martin, S. J.; Frye, G. C. Proc. IEEE Ultrason. Symp. 1991, 393. (15) Bodenhofer, K.; Hierlemann, A.; Noetzel, G.; Weimar, U.; Goepel, W. Anal. Chem. 1996, 68, 2210. (16) Wise, B. M.; Gallagher, N. B.; Grate, J. W. J. Chemometrics 2002, 17, 463. (17) Grate, J. W.; Wise, B. M. Anal. Chem. 2001, 73, 2239. (18) Grate, J. W.; Zellers, E. T. Anal. Chem. 2000, 72, 2861. (19) Grate, J. W.; Kaganove, S. N.; Bhethanabotla, V. R. Faraday Discuss. 1997, 107, 259. (20) Grate, J. W. In Assignment of the Glass Transition; Seylor, R. J., Ed.; ASTM STP 1249; ASTM: Philadelphia, PA, 1994; p 153. (21) Grate, J. W.; Klusty, M.; McGill, R. A.; Abraham, M. H.; Whiting, G.; Andonian-Haftvan, J. Anal. Chem. 1992, 64, 610. (22) Grate, J. W.; Abraham, M. H. Sens. Actuators B 1991, 3, 85. (23) Abraham, M. H.; Ibrahim, A.; Zissimos, A. M. J. Chromatogr. A 2004, 1037, 29. (24) Abraham, M. H.; Doherty, R. M.; Kamlet, M. J.; Taft, R. W. Chem. Br. 1986, 22, 551. (25) Kamlet, M. J.; Doherty, R. M.; Abboud, J.-L. M.; Abraham, M. H.; Taft, R. W. CHEMTECH 1986, 16, 566. (26) Grate, J. W.; Abraham, M. H.; McGill, R. A. In Handbook of Biosensors: Medicine, Food, and the EnVironment; Kress-Rogers, E., Nicklin, S., Eds.; CRC Press: Boca Raton, FL, 1996; p 593. (27) Snow, A. W.; Sprague, L. G.; Soulen, R. L.; Grate, J. W.; Wohltjen, H. J. Appl. Polym. Sci. 1991, 43, 1659. (28) Barlow, J. W.; Cassidy, P. E.; Lloyd, D. R.; You, C. J.; Chang, Y.; Wong, P. C.; Noriyan, J. Polym. Eng. Sci. 1987, 27, 703. (29) Chang, Y.; Noriyan, J.; Lloyd, D. R.; Barlow, J. W. Polym. Eng. Sci. 1987, 27, 693.

744 Chemical Reviews, 2008, Vol. 108, No. 2 (30) Ting, S. P.; Pearce, E. M.; Kwei, T. K. J. Polym. Sci., Part C: Polym. Lett. 1980, 18, 201. (31) Barlow, J. W. 1984 Scientific Conference on Chemical Defense Research, Aberdeen Proving Ground, MD, Nov 1984. (32) Wohltjen, H.; Snow, A. W.; Barger, W. R.; Ballantine, D. S. IEEE Trans. Ultrason., Ferroelect., Freq. Contr. 1987, UFFC-34, 172. (33) Ballantine, D. S.; Rose, S. L.; Grate, J. W.; Wohltjen, H. Anal. Chem. 1986, 58, 3058. (34) Field, D. E. J. Coatings Technol. 1976, 48, 43. (35) Field, D. E.; Griffith, J. R. Fluorinated Polyether Network Polymers, U.S. Patent 4132681, Jan 2, 1979. (36) Field, D. E.; Griffith, J. R. Ind. Eng. Chem. Prod. Res. DeV. 1975, 14, 52. (37) Rebiere, D.; Dejous, C.; Pistre, J.; Lipskier, J.-F.; Planade, R. Sens. Actuators B 1998, B49, 139. (38) Rebiere, D.; Dejous, C.; Pistre, J.; Planade, R.; Lipskier, J.-F.; Robin, P. Sens. Actuators B 1997, B43, 34. (39) Gupta, D. C.; Gutch, P. K.; Lal, G.; Vyas, K. D.; Jaiswal, D. K. Def. Sci. J. 2004, 54, 229. (40) Rose-Pehrsson, S. L.; Grate, J. W.; Ballantine, D. S.; Jurs, P. C. Anal. Chem. 1988, 60, 2801. (41) Grate, J. W.; Klusty, M.; Barger, W. R.; Snow, A. W. Anal. Chem. 1990, 62, 1927. (42) Fox, C. G.; Alder, J. F. Anal. Chim. Acta 1991, 248, 337. (43) Grate, J. W.; Rose-Pehrsson, S. L.; Venezky, D. L.; Klusty, M.; Wohltjen, H. Anal. Chem. 1993, 65, 1868. (44) Abraham, M. H.; Andonian-Haftvan, J.; Du, C. M.; Diart, V.; Whiting, G.; Grate, J. W.; McGill, R. A. J. Chem. Soc., Perkin Trans. 1995, 2, 369. (45) Collins, G. E.; Rose-Pehrsson, S. L. Anal. Chem. 1995, 67, 2224. (46) Grate, J. W.; Patrash, S. J.; Abraham, M. H. Anal. Chem. 1995, 67, 2162. (47) Collins, G. E.; Buckley, L. J. Synth. Met. 1996, 78, 93. (48) Grate, J. W.; Kaganove, S. N. Polym. Prepr. 1997, 38, 955. (49) Grate, J. W.; Kaganove, S. N.; Patrash, S. J.; Craig, R.; Bliss, M. Chem. Mater. 1997, 9, 1201. (50) Zellers, E. T.; Park, J.; Tsu, T.; Groves, W. J. Anal. Chem. 1998, 70, 4191. (51) Grate, J. W.; Kaganove, S. N.; Patrash, S. J. Anal. Chem. 1999, 71, 1033. (52) Grate, J. W.; Wise, B. M.; Abraham, M. H. Anal. Chem. 1999, 71, 4544. (53) Park, J.; Groves, W. A.; Zellers, E. T. Anal. Chem. 1999, 71, 3877. (54) Cai, Q.-Y.; Park, J.; Heldsinger, D.; Hsieh, M.-D.; Zellers, E. T. Sens. Actuators B 2000, B62, 121. (55) Grate, J. W.; Kaganove, S. N.; Nelson, D. A. Chem. InnoVations 2000, 30 (11), 29. (56) Bubb, D. M.; Horwitz, J. S.; McGill, R. A.; Chrisey, D. B.; Papantonakis, M. R.; Haglund, R. F., Jr.; Toftmann, B. Appl. Phys. Lett. 2001, 79, 2847. (57) Grate, J. W.; Patrash, S. J.; Kaganove, S. N.; Abraham, M. H.; Wise, B. M.; Gallagher, N. B. Anal. Chem. 2001, 73, 5247. (58) Shen, C.-Y.; Shen, Y.-T.; Wu, L. J. Mater. Sci. 2002, 37, 295. (59) Grate, J. W.; Nelson, D. A. Proc. IEEE 2003, 91, 881. (60) Shen, Y.-T.; Shen, C.-Y.; Wu, L.; Huang, C.-L. Jpn. J. Appl. Phys. Part 1 2003, 42, 1358. (61) Houser, E. J.; Mlsna, T. E.; Nguyen, V. K.; Chung, R.; Mowery, R. L.; McGill, A. R. Talanta 2001, 54, 469. (62) Krech, J. H.; Rose-Pehrsson, S. L. Anal. Chim. Acta 1997, 341, 53. (63) Urry, W. H.; Niu, J. H. Y.; Lundsted, L. G. J. Org. Chem. 1968, 33, 2302. (64) Purcell, K. F.; Stikeleather, J. A.; Brunk, S. D. J. Am. Chem. Soc. 1969, 91, 4019. (65) Amara, J. P.; Swager, T. M. Macromolecules 2005, 38, 9091. (66) Grate, J. W. Siloxanes with Strongly Hydrogen Bond Donating Functionalities, U.S. Patent 5,756,631, May 26, 1998. (67) SXFA was originally invented and synthesized by Grate at the Naval Research Laboratory. He supplied a sample to Abraham and McGill in England for LSER characterization. SXFA has since been used by several investigators, including Grate, who moved to the Pacific Northwest National Laboratory, McGill, after he moved to the Naval Research Laboratory, and their various collaborators. (68) McGill, R. A.; Chung, R.; Chrisey, D. B.; Dorsey, P. C.; Matthews, P.; Pique, A.; Mlsna, T. E.; Stepnowski, J. L. IEEE Trans. Ultrason., Ferroelect., Freq. Contr. 1998, 45, 1370. (69) Pique, A.; McGill, R. A.; Chrisey, D. B.; Leonhardt, D.; Mlsna, R. E.; Spargo, B. J.; Callahan, J. H.; Vachet, R. W.; Chung, R.; Bicaro, M. A. Thin Solid Films 1999, 536, 355-356. (70) Pinnaduwage, L. A.; Thundat, T.; Hawk, J. E.; Hedden, D. L.; Britt, R.; Houser, E. J.; Stepnowski, S.; McGill, R. A.; Bubb, D. Sens. Actuators B 2004, 99, 223. (71) Patel, S. V.; Mlsna, T. E.; Fruhberger, B.; Klaassen, E.; Cemalovic, S.; Baselt, D. R. Sens. Actuators B 2003, B96, 541.

Grate (72) Cunningham, B. T.; Kant, R.; Daly, C.; Weinberg, M. S.; Pepper, J. W.; Clapp, C.; Bousquet, R.; Hugh, B. Proc. SPIE-Int. Soc. Opt. Eng. 2000, 4036, 151. (73) Chrisey, D. B.; Pique, A.; McGill, R. A.; Horwitz, J. S.; Ringeisen, B. R.; Bubb, D.; Wu, P. K. Chem. ReV. 2003, 103, 553. (74) Mlsna, T. E.; Cemalovic, S.; Warburton, M.; Hobson, S. T.; Mlsna, D. A.; Patel, S. V. Sens. Actuators B 2006, 116, 192. (75) Patel, S. V.; Hobson, S. T.; Cemalovic, S.; Mlsna, T. E. Proc. SPIEInt. Soc. Opt. Eng. 2005, 5986, 59860M/1. (76) Mathias, L. J.; Lewis, C. M. Macromolecules 1993, 26, 4070. (77) Tronc, F.; Lestel, L.; Boileau, S. Polymer 2000, 41, 5039. (78) Tronc, F.; Lestel, L.; Amiel, C.; Boileau, S. Langmuir 1999, 15, 7080. (79) Abraham, M. H.; Hamerton, I.; Rose, J. B.; Grate, J. W. J. Chem. Soc., Perkin Trans. 1991, 2, 1417. (80) Harvey, S. D.; Nelson, D. A.; Wright, B. W.; Grate, J. W. J. Chromatogr., A 2002, 954, 217. (81) Levitsky, I. A.; Krivoshlykov, S. G.; Grate, J. W. J. Phys. Chem. B 2001, 105, 8468. (82) Levitsky, I.; Krivoshlykov, S. G.; Grate, J. W. Anal. Chem. 2001, 73, 3441. (83) Lu, C. J.; Zellers, E. T. Analyst 2002, 127, 1061. (84) Lewis, P. R.; Manginell, R. P.; Adkins, D. R.; Kottenstette, R. J.; Wheeler, D. R.; Sokolowski, S. S.; Trudell, D. E.; Byrnes, J. E.; Okandan, M.; Bauer, J. M.; Manley, R. G.; Frye-Mason, G. C. IEEE Sens. J. 2006, 6, 784. (85) Lu, C. J.; Whiting, J.; Sacks, R. D.; Zellers, E. T. Anal. Chem. 2003, 75, 1400. (86) Bhadury, P. S.; Dubey, V.; Singh, S.; Saxena, C. J. Fluorine Chem. 2005, 126, 1252. (87) Hsieh, M. D.; Zellers, E. T. J. Occup. EnViron. Hyg. 2004, 1, 149. (88) Holmes, M. A.; Mackay, M. E.; Krishnan, R. S.; Giunta, R. K.; Hawker, C. J.; Malkoch, M. PMSE Prepr. 2004, 90, 692. (89) Holmes, M. A.; Mackay, M. E.; Giunta, R. K. PMSE Prepr. 2004, 91, 1049. (90) Cai, Q.-Y.; Zellers, E. T. Anal. Chem. 2002, 74, 3533. (91) Grate, J. W.; McGill, R. A.; Abraham, M. H. Proc. IEEE Ultrason. Symp. 1992, 275. (92) Martin, S. D.; Poole, C. F.; Abraham, M. H. J. Chromatogr. A 1998, 805, 217. (93) McGill, R. A.; Mlsna, T. E.; Chung, R.; Nguyen, V. K.; Stepnowski, J. Sens. Actuators B 2000, B65, 5. (94) Mlsna, T. E.; Mowery, R.; McGill, R. A. Technical Digest of the 7th International Meeting on Chemical Sensors, International Academic Publishers: Beijing, China, July 27-30, 1998; p 65. (95) McGill, R. A.; Mlsna, T. E.; Mowery, R. Proceedings of the IEEE 1998 International Frequency Control Symposium, IEEE: 1998; p 630. (96) McGill, R. A.; Mlsna, T. E.; Chung, R.; Nguyen, V. K.; Stepnowski, J.; Abraham, M. H.; Kobrin, P. Proc. SPIE-Int. Soc. Opt. Eng. 1998, 3392, 384. (97) Simonson, D. L.; Houser, E. J.; Stepnowski, J. L.; Stepnowski, S. V.; McGill, R. A. Polym. Prepr. 2005, 46, 1186. (98) Simonson, D. L.; Houser, E. J.; Stepnowski, J. L.; McGill, R. A. Polym. Prepr. 2004, 45, 558. (99) Houser, E. J.; Simonson, D. L.; Stepnowski, J. L.; Ross, S. K.; Stepnowski, S. V., III.; McGill, R. A. Polym. Prepr. 2004, 45, 541. (100) Simonson, D. L.; Houser, E. J.; Stepnowski, J. L.; Pu, L.; McGill, R. A. PMSE Preprints 2003, 89, 866. (101) Simonson, D. L.; Houser, E. J.; Stepnowski, J. L.; McGill, R. A. PMSE Prepr. 2003, 88, 546. (102) Houser, E. J.; Simonson, D. L.; Stepnowski, J. L.; McGill, R. A. PMSE Prepr. 2003, 88, 548. (103) Demathieu, C.; Chehimi, M. M.; Lipskier, J. F. Sens. Actuators B 2000, B62, 1. (104) Zimmermann, C.; Rebiere, D.; Dejous, C.; Pistre, J.; Chastaing, E.; Planade, R. Sens. Actuators B 2001, B76, 86. (105) Zimmermann, C.; Mazein, P.; Rebiere, D.; Dejous, C.; Pistre, J.; Planade, R. IEEE Sens. J. 2004, 4, 479. (106) Wheeler, D. (Sandia National Labs) Personal communication, May 2007. (107) Holmes, M. A.; Mackay, M. E.; Giunta, R. K. J. Nanoparticle Res. 2006, DOI:10.1007/s11051-006-9118-1. (108) Hartmann-Thompson, C.; Hu, J.; Kaganove, S. N.; Keinath, S. E.; Keeley, D. L.; Dvornic, P. R. Chem. Mater. 2004, 16, 5357. (109) A chemical structure diagram for HC has been given in a poster: http://www.nrl.navy.mil/nanoscience/files/nanochemistry_poster. pdf. (110) Voiculescu, I.; McGill, R. A.; Zaghloul, M. E.; Mott, D.; Stepnowski, J.; Stepnowski, S.; Summers, H.; Nguyen, V.; Ross, S.; Walsh, K.; Martin, M. IEEE Sens. J. 2006, 6, 1094. (111) Voiculescu, I.; Zaghloul, M. E.; McGill, R. A.; Houser, E. J.; Fedder, G. K. IEEE Sens. J. 2005, 5, 641.

Hydrogen-Bond Acidic Polymers (112) Snow, E. S.; Perkins, F. K.; Houser, E. J.; Badescu, S. C.; Reinecke, T. L. Science 2005, 307, 1942. (113) Hartmann-Thompson, C.; Keeley, D. L.; Dvornic, P. R.; Keinath, S. E.; McCrea, K. R. J. Appl. Polym. Sci. 2007, 104, 3171. (114) Kamlet, M. J.; Taft, R. W. Acta Chem. Scand. B 1985, 39, 611. (115) Abraham, M. H. Chem. Soc. ReV. 1993, 22, 73. (116) Abraham, M. H.; Poole, C. F.; Poole, S. K. J. Chromatogr. A 1999, 842, 79. (117) McGill, R. A.; Abraham, M. H.; Grate, J. W. CHEMTECH 1994, 24 (9), 27. (118) Hierlemann, A.; Zellers, E. T.; Ricco, A. J. Anal. Chem. 2001, 73, 3458. (119) Abraham, M. H.; Andonian-Haftvan, J.; Whiting, G.; Leo, A.; Taft, R. W. J. Chem. Soc., Perkin Trans. 1994, 2, 1777. (120) Abraham, M. H.; Grellier, P. L.; Prior, D. V.; Duce, P. P.; Morris, J. J.; Taylor, P. J. J. Chem. Soc., Perkin Trans. II 1989, 699. (121) Abraham, M. H.; Grellier, P. L.; Prior, D. V.; Morris, J. J.; Taylor, P. J. J. Chem. Soc., Perkin Trans. 1990, 2, 521. (122) Abraham, M. H.; Whiting, G. S.; Doherty, R. M.; Shuely, W. J. J. Chromatogr. 1991, 587, 213. (123) Abraham, M. H.; Fuchs, R. J. Chem. Soc., Perkin Trans. II 1988, 523. (124) Abraham, M. H.; Grellier, P. L.; McGill, R. A. J. Chem. Soc., Perkin Trans. II 1987, 797. (125) Patrash, S. J.; Zellers, E. T. Anal. Chem. 1993, 65, 2055. (126) Severin, E. J.; Lewis, N. S. Anal. Chem. 2000, 72, 2008. (127) Lonergan, M. C.; Severin, E. J.; Doleman, B. J.; Beaber, S. A.; Grubbs, R. H.; Lewis, N. S. Chem. Mater. 1996, 8, 2298. (128) Wohltjen, H.; Dessy, R. E. Anal. Chem 1979, 51, 1465. (129) Wohltjen, H. Sens. Actuators 1984, 5, 307. (130) Rapp, M.; Boss, B.; Voigt, A.; Bemmeke, H.; Ache, H. J. Fresenius J. Anal. Chem. 1995, 352, 699. (131) Zellers, E. T.; Batterman, S. A.; Han, M.; Patrash, S. J. Anal. Chem. 1995, 67, 1092. (132) Bodenhoefer, K.; Hierlemann, A.; Seemann, J.; Gauglitz, G.; Christian, B.; Koppenhoefer, B.; Goepel, W. Anal. Chem. 1997, 69, 3058. (133) Hierlemann, A.; Ricco, A. J.; Bodenhoefer, K.; Goepel, W. Anal. Chem. 1999, 71, 3022. (134) Ricco, A. J.; Crooks, R. M.; Osbourn, G. C. Acc. Chem. Res. 1998, 31, 289. (135) Ricco, A. J.; Crooks, R. M.; Xu, C.; Allred, R. E. In Interfacial Design and Chemical Sensing; Mallouk, T. E.; Harrison, D. J., Ed.; ACS Symposium Series 561; American Chemical Society: Washington, DC, 1994; p 264. (136) Heller, E. J.; Hietala, V. M.; Kottenstette, R. J.; Manginell, R. P.; Matzke, C. M.; Lewis, P. R.; Casalnuovo, S. A.; Frye-Mason, G. C. Proc. Electrochem. Soc., Chem. Sens. IV 1999, 138, 99-23. (137) Groves, W. A.; Zellers, E. T.; Frye, G. C. Anal. Chim. Acta 1998, 371, 131. (138) Frye, G. C.; Martin, S. J.; Ricco, A. J. Sens. Mater. 1989, 1, 335. (139) Liron, Z.; Kaushansky, N.; Frishman, G.; Kaplan, D.; Greenblatt, J. Anal. Chem. 1997, 69, 2848. (140) Liron, Z.; Greenblatt, J.; Frishman, G.; Gratziani, N. Sens. Actuators B 1993, 12, 115. (141) Shaffer, R. E.; Rose-Pehrsson, S. L.; McGill, R. A. Field Anal. Chem. Technol. 1998, 2, 179. (142) Alder, J. F.; McCallum, J. J. Analyst 1983, 108, 1169. (143) Guilbault, G. G.; Jordan, J. M. CRC Crit.l ReV. Anal. Chem. 1988, 19, 1. (144) Fox, C. G.; Alder, J. F. Analyst 1989, 114, 997. (145) D’Amico, A.; Verona, E. Sens. Actuators 1989, 17, 55. (146) Nieuwenhuizen, M. S.; Venema, A. Sens. Mater. 1989, 5, 261. (147) Ballantine, D. S.; Wohltjen, H. Anal. Chem. 1989, 61, 704A. (148) McCallum, J. J. Analyst 1989, 114, 1173. (149) Frye, G. C.; Martin, S. J. Appl. Spectrosc. ReV. 1991, 26, 73.

Chemical Reviews, 2008, Vol. 108, No. 2 745 (150) Thompson, M.; Kipling, A. L.; Duncan-Hewitt, W. C.; Rajakovic, L. V.; Cavic-Vlasak, B. A. Analyst 1991, 116, 881. (151) Nieuwenhuizen, M. S.; Venema, A. In Sensors A ComprehensiVe SurVey; Goepel, W., Jones, T. A., Kleitz, M., Lundstrom, J., Seiyama, T., Eds.; VCH: Weinheim, 1991; Vol. 2, p 647. (152) Grate, J. W.; Baer, R. L. In Current Topics in Biophysics: Biosensors; Frangepol, P. T., Nilolelis, D. P., Eds.; Iasi University Press: Iasi, Romania, 1996; p 51. (153) Ballantine, D. S.; White, R. M.; Martin, S. J.; Ricco, A. J.; Zellers, E. T.; Frye, G. C.; Wohltjen, H. Acoustic WaVe Sensors. Theory, Design, and Physico-Chemical Applications; Academic Press: New York, 1997. (154) Thompson, M. H.; Stone, D. C. Surface-Launched Acoustic WaVe Sensors: Chemical Sensing and Thin-Film Characterization; John Wiley and Sons: New York, 1997. (155) Unpublished results, with permission. Grate supplied BSP3 polymer to Michael Rapp, whose group independently coated and tested sensors in his laboratories. (156) Bender, F.; Barie, N.; Romoudis, G.; Voigt, A.; Rapp, M. Sens. Actuators B 2003, B93, 135. (157) Bender, F.; Waechter, L.; Voigt, A.; Rapp, M. Proceedings of IEEE Sensors 2003. IEEE International Conference on Sensors, 2nd, Toronto, ON, Canada, Oct 22-24, 2003; IEEE: 2003; Vol. 1, p 115. (158) Sepaniak, M. J.; Datskos, P. G.; Lavrik, N.; Tipple, C. Anal. Chem. 2002, 74, 569A. (159) Thundat, T.; Chen, G. Y.; Warmack, R. J.; Allison, D. P.; Wachter, E. A. Anal. Chem. 1995, 67, 519. (160) Pinnaduwage, L. A.; Ji, H.-F.; Thundat, T. IEEE Sens. J. 2005, 5, 774. (161) Houser, E. J.; Chrisey, D. B.; Bercu, M.; Scarisoreanu, N. D.; Purice, A.; Colceag, D.; Constantinescu, C.; Moldovan, A.; Dinescu, M. Appl. Surf. Sci. 2006, 252, 4871. (162) Pinnaduwage, L. A.; Hedden, D. L.; Gehl, A.; Boiadjiev, V. I.; Hawk, J. E.; Farahi, R. H.; Thundat, T.; Houser, E. J.; Stepnowski, S.; McGill, R. A.; Deel, L.; Lareau, R. T. ReV. Sci. Instrum. 2004, 75, 4554. (163) The polymer designated by HCSA2 is apparently the hyperbranched HC, a polycarbosilane with hexafluoroalcohol substituents. (164) Barger, W. R.; Dote, J.; Klusty, M.; Mowery, R.; Price, R.; Snow, A. Thin Solid Films 1988, 159, 369. (165) Albert, K. J.; Lewis, N. S.; Schauer, C. L.; Sotzing, G. A.; Stitzel, S. E.; Vaid, T. P.; Walt, D. R. Chem. ReV. 2000, 100, 2595. (166) Kummer, A. M.; Hierlemann, A.; Baltes, H. Anal. Chem. 2004, 76, 2470. (167) Cornila, C.; Hierlemann, A.; Lenggenhager, R.; Malcovati, P.; Baltes, H.; Noetzel, G.; Weimar, U.; Gopel, W. Sens. Actuators B 1995, 25, 357. (168) Novak, J. P.; Snow, E. S.; Houser, E. J.; Park, D.; Stepnowski, J. L.; McGill, R. A. Appl. Phys. Lett. 2003, 83, 4026. (169) Johnson, S. R.; Sutter, J. M.; Engelhardt, H. L.; Jurs, P. C.; White, J.; Kauer, J. S.; Dickinson, T. A.; Walt, D. R. Anal. Chem. 1997, 69, 4641. (170) Albert, K. J.; Gill, S. D.; Pearce, T. C.; Walt, D. R. Anal. Bioanal. Chem. 2002, 373, 792. (171) Albert, K. J.; Myrick, M. L.; Brown, S. B.; James, D. L.; Milanovich, F. P.; Walt, D. R. EnViron. Sci. Technol. 2001, 35, 3193. (172) White, J.; Kauer, J. S.; Dickinson, T. A.; Walt, D. R. Anal. Chem. 1996, 68, 2191. (173) Barnard, S. M.; Walt, D. R. EnViron. Sci. Technol. 1991, 25, 1301. (174) Grate, J. W.; Nelson, D. A.; Skaggs, R. Anal. Chem. 2003, 75, 1868. (175) Snow, A. W.; Wohltjen, H.; Jarvis, N. L. 2002 NRL Review, pp 45-55, 2002. [Online]. Available: http://www.nrl.navy.mil/content. php?P)02REVIEW45.

CR068109Y

Chem. Rev. 2008, 108, 746−769

746

Composites of Intrinsically Conducting Polymers as Sensing Nanomaterials David W. Hatchett† and Mira Josowicz*,‡ Department of Chemistry, University of Nevada, Las Vegas, Nevada 89154-4003 and Department of Chemistry and Biochemistry, Georgia Institute of Technology, Atlanta, Georgia 30332-0400 Received August 10, 2007

Contents 1. Introduction 2. Properties and Limitations of Composite Components 2.1. Intrinsically Conducting Polymers (ICP) 2.2. Carbon-Structured Materials 2.3. Metals and Metal Oxides 2.4. Phthalocyanines and Related Complexes 2.5. Biologically Active Materials 3. Advances in Composites of Inherently Conducting Polymers 3.1. Composites with Nonconducting Polymers 3.1.1. Composites with Hydrophilic Polymers 3.1.2. Composites with Hydrophobic Polymers 3.1.3. Multicomponent Composites 3.2. Composites with Carbon Black and Carbon Nanotubes 3.3. Composites with Metallic or Metal-Containing Components 3.3.1. Gold 3.3.2. Platinum and Palladium Metal 3.3.3. Copper and Nickel Metal 3.3.4. Metal Oxides 3.3.5. Metal Phthalocyanines and Porphyrins 3.4. Composites with Biological Materials 4. Conclusions and Outlook 5. Acknowledgments 6. References

746 747 747 749 750 752 752 753 753 754 754 755 756 758 758 760 761 762 762 763 764 765 765

1. Introduction Composites of intrinsically conducting polymers (ICPs) are materials that utilize conjugated polymers and at least one secondary component that can be inorganic or organic materials or biologically active species. The goal is to produce a new composite material that has distinct properties that were not observed in the individual components. This may include either new or improved chemical properties that can be exploited for chemical or biological sensing. For example, adding carbon nanotubes tends to drastically influence the electrical and thermal conductivity of ICPs. A secondary aspect concerns the stabilization of the secondary component in the polymer matrix. Enhanced optical, electri* To whom the correspondence should be addressed. Phone: 404-894-4032. Fax: 404-894-7452. E-mail: [email protected]. † University of Nevada. ‡ Georgia Institute of Technology.

cal, or mechanical properties such as stiffness and strength are common. In some cases, the physical and chemical properties of the secondary component are much different after composite formation. For the purpose of this review we will primarily focus on the ICPs such as polyaniline, polypyrrole, and polythiophene and their derivatives. The resonance-stabilized structure of ICPs allows, for example, incorporation of ions, nanoparticles, or nanowires of metals, metal oxides, carbon, or molecular species such as metallophthalocyanines or biologically active components such as enzymes, antibodies, and antigens.1 In some cases, the ICP will simply act as a template for the incorporation of the secondary component. In that case, the secondary component will impart the chemical properties required for chemical sensing. In other cases the materials are linked through electrostatic interactions which influence the electronic and physical properties of the materials used to prepare the composite. The combined physical/chemical properties of the components are exploited for chemical sensing when the materials are joined. Differences in the properties of composites containing similar components are often tied to the method of preparation. Therefore, preparation methods will be discussed in more detail with respect to the “how” the secondary component is incorporated in the ICPs. The major advantage of ICP composite materials over the ICP alone is based on the increase in active surface area and ability to form good electronic contact between the composite components and the transducer. The parent polymer provides high dispersion and high surface area for the secondary components to be integrated and creates templates for chemical reactions and interactions. The inherent stability and symbiosis between the two components used to create the composite material is often superior to the bulk components alone. This review is organized as follows. First, we briefly summarize the properties and limitations of the most frequently used materials in chemical and biological sensors. Then different approaches to combine them with ICPs are discussed. The methods have been broadly classified as polymerization, dispersion, redox reactions, and electrostatic interactions. When applicable the strategies used to control the size, shape, and distribution of the secondary component in the composite are emphasized. When relevant the catalytic applications of the composite materials are examined. The advantages of using these composites as sensing material can be expressed with respect to increased surface area, higher numbers of analytical recognition sites, lower detection limits, low resistivity or faster response times, and improved environmental stability. The mechanical properties of the material are often improved which leads to more robust sensors.

10.1021/cr068112h CCC: $71.00 © 2008 American Chemical Society Published on Web 01/03/2008

Composites of Intrinsically Conducting Polymers

David W. Hatchett received his B.Sc. degree from California State University, Stanislaus, in 1992. He received his Ph.D. degree from the University of Utah, Salt Lake City, in 1997 for work on the electrochemical formation of sulfide and thiolate monolayers on Ag. Dr. Henry S. White supervised this work. After graduation he accepted a postdoctoral associate position in the Department of Chemistry in Professor Jiri Janata’s Group at the Georgia Institute of Technology. His postdoctoral research focused on the sensing properties of polyaniline (PANI) and the acid doping and control of species in conductive polymer membranes. His current research interests are diverse including the chemical and electrochemical synthesis of PANI/metal composites of PANI/Au, PANI/Pt, and PANI/Pd. The primary goal of this research is to produce composite materials with unique sorption, chemical, sensing, and catalytic properties. In addition, composite ion- and gas-selective membranes for chemical sensing applications are examined. His research focuses on the development and testing of gasand ion-sensing materials that can be used to manufacture chemical sensors. Finally, electrochemical characterization of lanthanide and actinide species in room-temperature ionic liquids (RTILs) has been explored. This work examines the potential-mediated separation of species using the large potential window provided by nonaqueous RTIL systems. The range of materials produced and characterization of the materials are examined with respect to defined and diverse applications. His research provides insight into the rational design of materials for chemical sensing, catalysis, ion transport, and sorption processes.

In summary, the aim of the review is to explore a fundamental and technological incentive for ICP composite sensors, the sensing properties, and how molecular ICP composites are made. The behavior of the secondary component relative to the primary component and how the combined properties of the composite improve sensing applications is also examined. The references selected in the text do not reflect the chronology of the advances in the research field within the last 7 years, and we do not claim to present every possible reference for completeness. Rather the references are meant to highlight specific aspects and applications that are important in the synthesis and application of ICP composite to chemical and biosensors.

2. Properties and Limitations of Composite Components 2.1. Intrinsically Conducting Polymers (ICP) In general, intrinsically conductive polymers (ICPs), also referred to as organic semiconductors, are polymers with a delocalized π-electron system with an “intrinsic” wide band gap that defines their affinity for electrons (work function).2 They are good candidates for developing chemical or electrochemical sensors for two reasons: (1) The electronic conductivity, related to the redox state (doping level) of a conducting polymer, is modulated by the interaction with various analytes resulting in changes in parameters such as

Chemical Reviews, 2008, Vol. 108, No. 2 747

Mira Josowicz studied chemistry at the Politechnika Slaska, Gliwice, Poland, where she graduated in 1966 in Physical Chemistry. She received her Ph.D. degree from the Technical University in Munich, Germany, in the field of Inorganic Chemistry with Professors Ch. Quentin and H.-P. Fritz. From 1983 to 1986, she was a Feodor-Lynnen-Humboldt postdoctoral research fellow at the University of Utah with Professor J. Janata. After her return to the University of Bundeswehr in Munich, she finished her habilitation in 1992 in the field of chemical sensors. From 1992 to 1997 she was a staff scientist in the Materials and Chemical Sciences Center at the Pacific Northwest National Laboratory, Richland, Washington. Since 1997 she has been Principal Research Scientist in the School of Chemistry and Biochemistry, Georgia Institute of Technology. Her main research interests include chemical sensors and properties of organic semiconductors.

resistance, current, or electrochemical potential/work function. Since the ICP acts as an electronic transducer, the magnitude and rise of the electrical signal correlates with the type of physical-chemical interaction involved in the signal generation. The interaction with analytes introduces changes in the physicochemical properties resulting from structural reorganization of the polymer chains. These changes are facilitated by an increase or a decrease of the density of charged species through polymer/analyte interaction or hydrogen bonding between the species.3 (2) They are attractive as sensing layers because they retain processing properties of polymers, lowering manufacturing costs of films, powders, or wires. The processing of ICPs, that are of poly(heterocyclic) types, such as polypyrrole (PPy), polyfuran, polythiophene (PT), poly(p-phenylene vinylene), aromatic poly(azomethine)s, and polyaniline (PANI), from solution or using heat treatment has been reviewed with emphasis on the design of dopant, special catalytic properties, and special optical or spectroscopic properties.4 Consequently, specific functionality can be added to the ICP matrix by entrapment during the polymerization process. Charge carriers are introduced into the polymer by doping, which can be accomplished chemically, electrochemically, or optically.5 For example, polyaniline can be protonated by triflic acid generated by UV photolysis of triphenylsulfonium triflate salt.6 The strong acid, such as triflic acid or tetrafluoroboric acid, lowers both the work function and increases the conductivity of polyaniline (PANI) base according to the known proton doping mechanism. In contrast, addition of triflate salts to the emeraldine base form of PANI in the solid state leads to formation of localized states that do not contribute to the conductivity but lower the work function of the polymer.7 We call this form of doping charge-transfer doping as opposed to oxidative and proton doping. This distinction has important implications for use of PANI in various types of sensing device structures. When PANI is used as the replacement for the metal in the insulated gate

748 Chemical Reviews, 2008, Vol. 108, No. 2

field-effect transistor the value of the work function is critically important because it controls the threshold voltage. Since there is no current passing through the PANI layer in such a device, its conductivity is irrelevant. However, when PANI itself is used as a conductor of electronic current, the type of doping that leads to conductivity changes is important. Moreover, in some devices, such as thin-film transistors, both charge transfer and proton doping play a role in the operational characteristics of the device. Furthermore, the conductivity of ICPs depends on its ability both to transport charge carriers along the polymer backbone and to hop between polymer chains.8 The work function of the ICP is defined by its redox state and the basis for modulation of optical transitions and ionization potential.9 The environmental stability of the material is crucial for analytical calibration of the sensor. It has been demonstrated that the functional groups of camphorsulfonic acid used as a dopant for PANI can actually help to fulfill such requirements.10 Intrinsically conductive polymers have advantageous morphological and structural properties that are critical to the sensitivity and selectivity of gas or ion sensors. These properties are sensitive to polymerization parameters such as the type of the counterion, supporting electrolyte, and solvent monomer/counterion concentrations. Introduction of subtly different dopant groups into ICPs including variations in doping level results in different conductivities of the material. Upon exposure to gases or analytes the ICPs show significant differences in conductivity responses due to changes in charge mobility and the amount of charge doping in the films. Selectivity of ICPs has been demonstrated through addition of a functional group to the precursor monomer or/and copolymerization to produce different ICP precursors. For example, the sensitivity to alkali-metal ions has been introduced by addition of polyalkyl ether or crown ether functionalized groups to the parent polymer chain. Furthermore, mixing pyridyl-based ligands with the parent polymer produces coordination sites. Introduction of metal ions can introduce conformational changes in the ICP that govern optical or electrochemical signals upon exposure to chemical or biological analytes. When the ICP is polymerized in the presence of a chiral compound, molecular-size specificity can be incorporated in the matrix. All of these possibilities were discussed in much more detail in a previous review by Swager et al.11 Different synthetic routes and their effect on the gas sensing properties of ICPs have also been the subject of several reviews.12 The sensing mechanism, configurations, and factors that affect the performance of gas sensors fabricated using conducting polymers such as polyaniline (PANI), polypyrrole (PPy), and poly(3,4-ethylenedioxythiophene) (PEDOT) as the active layers have been reviewed by others within this issue of Chemical ReViews. In brief, the interactions of ICPs with ambient gases can be monitored using the membrane as a chemical resistor (conductivity changes), diode or field effect transistor (work function changes), optical waveguide (absorption changes), or surface acoustic devices (mass change). Since changes in work function are based on formation of a charge-transfer complex between the gaseous species and the conducting polymer,13 methods for a stable and adjustable primary doping of the ICPs are necessary.14 Electron donation or withdrawal by the analyte vapors leads to conductivity changes in the sensor films. A systematic study of the steadystate electrical resistance responses of alkyl-substituted polypyrroles (PPy) and a polyalkylterthiophene (PAT) doped

Hatchett and Josowicz

with a variety of counter anions to volatile organic chemicals (VOC) comprising alcohols, esters, aromatics, and alkanes revealed a linear correlation between vapor concentration producing a fixed amplitude of sensor response with respect to analyte-saturated vapor pressures.15 It has also been observed that the sensor sensitivity toward an analyte can be correlated with the differences in analyte polarity. Changing the dopant ion in the conducting polymer has a large effect on absolute sensor sensitivity but does not markedly affect the relative sensitivity toward different analyte functional groups. Humidified carrier gas depresses sensor sensitivity, suggesting competition between water and gas molecules. Polymer swelling based on humidity also implies that the pathways become saturated with water, which would then change the diffusion rate of the gases into the polymer matrix. Activity coefficients for solutions of organic vapors dissolved in conducting polymers are not available in the literature. Overall, a nonspecific partition model of conducting polymer interaction with organic vapors is proposed in which, for a given analyte functional group, the electrical resistance transduction mechanism is insensitive to large variations in analyte size and shape. Furthermore, conducting polymers are useful as both ionto-electron transducers and sensing membranes in solid-state ion-selective electrodes. However, the influence of ambient media such as oxygen, acids, bases, redox reactants, water, and organic vapors on changes of the polymer structure, composition, and degradation should not be overlooked. The benefits and disadvantages of such processes for analytical characteristic of polypyrrole-based electrochemical sensors have been reviewed previously.16 The physical properties of ICPs can also be a hindrance for certain applications and must be considered when choosing which polymer will be used. For example, in bioelectrochemical applications it is important that the polymer film remains conductive at pH values higher than 4. The conductivity of PANI is reduced as the pH is increased, leading to a decrease in proton doping. However, the conductivity of ICPs such as polythiophene or polypyrrole demonstrates very little pH dependence. In addition, the effect of counterion incorporation into the polymer as a dopant during polymer synthesis or testing on the pH stability of the ICP can also be important.17 In some cases the lack of proton or anion doping can be used to create sensor membranes. Specifically, the physical properties of undoped conjugated polymers where the band gap depends on chemical constitution of the conjugated backbone and the nature of the substituents attached to the main chain have been exploited as electroluminescent (diode) sensors. The luminescence generation is caused by excitation that occurs during interaction of the polymer with the analyte. The wide variety of chemical sensors produced from ICPs demonstrates the diversity of the material. The progress in development of potentiometric sensors based on the use of ICPs within the last 5 years has been reviewed.18 Optical chemical sensors based on the electrochromic properties of ICPs have also been reviewed in this issue and by others.19 Although these reviews focus on the sensing properties of ICPs only, incorporation of secondary materials with ICPs to produce composite materials provides another approach for producing diverse sensing materials with higher chemical stability and improved analytical selectivity.

Composites of Intrinsically Conducting Polymers

2.2. Carbon-Structured Materials The chemistry and physics of carbon nanotube (CNT) have been the focus of a large number of studies. The enormous development of the carbon structures in recent years can be attributed to the high electrical conductivity, high chemical stability, and extremely high mechanical strength and modulus. All of these properties can be exploited to develop a new generation of chemical sensors.20 One significant drawback to the use of CNTs in sensors remains their lack of chemical selectivity. Selectivity is minimal due to the fact that the response mechanism is based on the electron affinity and density ratio of the target analyte relative to the CNT transducer. Therefore, formation of ICP/CNT composites is important and functionalization is critical for use in chemical sensors. CNTs are used in composites because they can be treated as inorganic or organic reagents based on their inherent “graphene” structure. They have electron-transport properties consistent with π-conjugated systems with metallic or semiconducting properties depending on their diameter.21 The rolled graphene structure forms single-walled nanotubes (SWNT) that can be then treated as one-dimensional conductors.22 They have great potential for conductance- and capacitance-based sensors because they are sensitive to changes in dipole moment and polarizability of the adsorbate when charge-transfer complexes are formed between adsorbates and the SWNT.23 The difficulty associated with the assembly of SWNTs has led to the use of SWNT networks that contain a large number of randomly positioned nanotubes (bundles). Consequently, the properties of SWNT devices depend on the position, number, and electronic type of its constituent nanotubes. A wide number of synthetic conditions are currently used to produce SWNTs, but none of the parameters discussed can currently be controlled with a high degree of reproducibility. The key to the electronic quality of random networks of SWNTs is the electrical contact that is formed between intersecting structures. SWNTs adhere to surfaces such as SiO2 via van der Waals forces that deform the SWNTs at the point of intersection and pull them closer together. The proximity of the two structures increases the probability of internanotube tunneling. Multiwalled nanotubes (MWNT) are based on a concentric tube within a tube and can have either metallic or semiconducting characteristics.24 These electrical hybrid tubes can be expected to behave as metallic nanowires due to the average current density across the characteristically different walls. In addition, wall-to-wall conduction is possible within the MWNT, allowing the properties of the metallic MWNTs tubes to dominate charge transport. Some researchers have reported that MWNTs behave as diffusive conductors.25 The principles for CNT gas sensors for the detection and quantitation of gases are based on changes in electrical properties induced by charge transfer with the gas molecules or changes in the adsorption.26 The ability of the adsorbate to transfer charge depends on the electron affinity of the adsorbate with respect to the CNT. For example, the conductivity of the semiconducting SWCNT upon exposure to 200 ppm of NO2 (electron-withdrawing gas) changed within 10 s, up to 3 orders of magnitude. In contrast, the conductivity decreases by 2 orders of magnitude upon exposure to 1% NH3 (electron-acceptor gas) within 2 min. In principle, the response mechanism can be applied to many other electron-accepting or -donating gases (NO, CO, CO2, etc.). However, a major obstacle in manipulating the electron

Chemical Reviews, 2008, Vol. 108, No. 2 749

affinity of CNTs is their limited solubility in solvent. They have a tendency to “rope up” or aggregate in solutions rather than dispersing uniformly.27 Aggregation of CNTs is typically minimized through chemical modification. For example, introduced carboxyl groups enable covalent coupling of molecules through the creation of amide or ester bonds. High solubility in organic solvents can be obtained by covalent or ionic attachment of long-chain aliphatic amines onto the carboxyl groups. The presence of the “modified” carboxyl groups leads to reduction of the van der Waals interactions between the CNTs, allowing separation and further chemical functionalization in aqueous or organic solutions. Functionalization also provides an effective tool to broaden the spectrum of electroanalytical chemical sensing and electrochemical biosensing applications of CNTs through subsequent solution-based chemistry.28 These modifications also allow secondary aspects to be tailored including electronic and mechanical properties.29 High solubility in water solution was achieved through carboxyl-based coupling of hydrophilic polymers such as poly(ethylene glycol) (PEG). Use of CNTs as transducers in chemical and physical processes after functionalization is also possible. Functionalization of the CNTs can be conducted in a controlled and selective manner electrochemically, where an active radical is generated from a target reagent present in solution, resulting in polymer formation on the CNT. Attention has been directed also to electron-donor-acceptor molecules such as ferrocene and phthalocyanine because they can act as secondary chemical transducers in electrochemical processes. Finally, functionalization of SWNT with zincphthalocyanine (ZnPc-SWNT) results in a photoinduced electron-transfer effect that has potential to be exploited in optical sensors.30 Significant progress in interfacing carbon nanotubes with biomolecular materials was made in key areas such as aqueous solubility, chemical and biological functionalization, biocompatibility, and specificity with respect to the electronic sensing properties of proteins.31 Development of hybrid nanostructures using MWNTs has been realized via adsorption of protonated amino groups that are able to anchor gold nanoparticles using electrostatic interactions. Functional species can then be affixed using the well-known mechanism for thiol sorption at gold surfaces.32 Recent advances in the asymmetric end functionalization of CNTs, with two different building blocks, allow a variety of materials to be produced. The end groups can be used to control hydrophobicity and hydrophilicity, ion transport through the CNT channels, and photoinduced electron transfer between donors and acceptors. Both approaches show great promise in the rational design and realization of a host of new chemical sensing materials.33 Despite the fast response of CNTs to low-vapor-pressure analytes, their long recovery time, extremely high affinity to adsorb oxygen, and complex fabrication process still remain a major drawback to their use as sensing materials. The homogeneity of the materials suffers due to inadequate separation procedures for fractionation based on conductivity, length, and diameter. Therefore, the low reproducibility of the physical properties has a major impact on the quality of the response and understanding of the mechanisms. Furthermore, electrical properties of the contacts often dependent on the geometry and external factors such as ambient light are not well understood. In contrast, the conductivity of most ICPs is lower than that of the CNTs and is well defined.

750 Chemical Reviews, 2008, Vol. 108, No. 2

2.3. Metals and Metal Oxides Metals in “nano” form exhibit unique properties relative to bulk metals. For example, gold nanoparticles change color and are very good catalysts relative to the bulk material. In addition, nanoparticles act as semiconductors, and their melting temperature decreases.34 Material clusters of nanoscale diameter often exhibit structural, electronic, spectral, magnetic, and chemical characteristics that are unique to the size regime. These properties cannot be easily extrapolated or deduced through scaling arguments based on the knowledge of these properties at the bulk limit.35 A number of approaches have been demonstrated for the synthesis and assembly of metal nanoparticles. Examples include two-phase synthesis of monolayer-protected nanoparticles, stepwise layer-by-layer construction, DNA complimentary binding, polymer- or dendrimer-mediated assembly, and mediator-template assembly.36 Supported metal particles were also grown in the liquid phase and then deposited on a substrate or formed directly on a surface from a solid-liquid interfacial layer.37 The particle size can be controlled most efficiently using sol deposition methods or zeolites, where the metal component is introduced and retained within the pores. However, problems can arise with regard to the stabilizing agent that maintains the particle in solution. Stabilization of the particles inside the pores through complexation can also occur. Use of metal carbonyl clusters to produce precursors of the nanoparticles seems to be least efficient regarding the difficulties in size regulation. Understanding the basis of the fundamental variations in the physical-chemical properties of materials ranging from the molecular level to solid-state chemistry has a great impact on organic, coordination, and solid-state chemistry, catalysis, physics, and materials science. Extensive studies are currently being performed that attempt to correlate the material properties with cluster size. In addition, factors that influence the surface properties with respect to catalysis and optical properties of nanomaterials are of general importance.38 The published knowledge of the frontiers of cluster science extends rapidly and is a topic of several books and several hundred publications.39 Topics related to metal clusters at surfaces have included the synthesis and assembly of nanocrystals, theory and spectroscopy of inter- and intraband optical transitions, single-nanocrystal optical and tunneling spectroscopies, electrical transport in nanocrystal assemblies, and the physical and engineering aspects of nanocrystal-based devices.40 An overview of the structure and resulting electronic and optical properties of metal nanoparticles embedded in insulating polymer matrices with emphasis on preparation of such materials have been reviewed with a focus on thin film deposition techniques.41 A general introduction to nanoscience based on the synthesis of colloidal nanocomposite particles and their use in making metal-nanoparticles such as silver, iron, ligands shell, inorganic core, oxide shell, semiconductor nanocrystals, nanorods, nanotubes, nanobelts, etc., has also been presented in a review recently.42 The long-term stability of these superlattices depends on the method used for their stabilization. It was reported that thiol-derivatized Au clusters are stable in toluene for over a 3-year period. When the particles are allowed to crystallize under ambient conditions, the protective alkanethiol coating slowly degrades. Under these conditions it becomes energetically favorable for the Au particles to minimize their surface

Hatchett and Josowicz

energy through aggregation.43 The larger particles (∼8 nm diameter) destabilize at a much slower rate than the smaller ones (∼4 nm). The reason for the decreased stability of the smaller particles as compared to the larger ones is possibly related to their geometrical structure. It was observed that heating of nanoparticle arrays of different sizes at different rates introduces instability into their array structures. That instability is initiated by the smaller nanoparticles that have a tendency to melt first. By melting a string of nanoparticles it is possible to produce a wire that is 10 times thinner than any wire made using the standard microelectronic process of electron beam lithography. The conductivity can be controlled by altering the size and separation between nanostructures that make the supercrystal. The influence of size, shape, and surface chemistry of small metal clusters that are dispersed on a support is crucial to understanding their electronic structure.44 The intrinsic properties of the cluster are nonscalable from bulk analogues for the reduction of the particle size to only a few dozen atoms. For example, gold reduced to clusters of eight atoms on an appropriate support transformed into an efficient, active center for oxidation of CO. The book Nanocatalysis recently published by Landmann and Heiz introduces the properties of nanosized materials and focuses on the influence and effect of dimensionality and size in the new and unique chemical properties.45 Use of gold nanoparticles in the field of sensing and recognition demonstrates that these nanoparticles are an ideal starting material for this purpose.46 One of the most attractive features of monolayer-protected clusters (MPC) for sensor application is the ease of introduction of functional groups coupled to biomolecules. For example, thiolated oligonucleotide are attractive for the voltammetric detection of hybridized DNA targets, for encapsulation with conjugated carbohydrates for labeling of specific proteins, or coupled to antibody conjugates for detection of antigens. Furthermore, these MPC systems are good candidates for modeling recognition processes and development of new biomimetic catalysts because they have globular shape and size comparable to many biological molecules including proteins, nucleic acids, enzymes, and many cellular substructures. Some interesting examples of these systems have been recently reviewed.47 Functionalization of gold nanoparticles is also of general importance. For example, a soft shell of organic chains surrounding the Au core can easily be exchanged with different thiols present in solution. The exchange allows introduction of the desired (sensitive) functionalized thiols using a reaction that has been termed “place-exchange reactions”.48 The “place-exchange reaction” is the macroscopic consequence of the dynamic self-assembled nature of the protective monolayer. Studies of the mechanism of the “place-exchange reaction” are still not conclusive, but the reported data suggest that an associative-dissociative process takes place. These processes suggest that the core Au particles may be converted to functional species with designed chemical structure and reactivity. More importantly, they provide the ability to tailor selectivity of the surface through thiolate bonding and exchange. Bimetallic nanoparticles (bi-MNPs) containing two metals are of great interest since they can exhibit catalytic, electronic, and optical properties distinct from the corresponding monometallic nanoparticles. A number of ap-

Composites of Intrinsically Conducting Polymers

proaches have been demonstrated for the synthesis and assembly of bimetallic or trimetallic nanoparticles.49 More recently, the optical,50 magnetic,51 and catalytic properties of multicomponent metal particles have been examined.52 For example, the composition of Au/Ag alloy nanoparticles can be varied by changing the salt ratio used during the bimetallic particle synthesis. The different mole ratios of metals both in and on the bimetallic cores has been utilized to manipulate colors and surface plasmon resonance bands.53 The interaction of the bimetal with target analytes changes surface plasmon resonance and can be used in developing sensors. Stable magnetization reversal transitions at room temperature of Pt/Fe alloy particles as small as 4 nm has also been achieved.54 The materials can act as electrocatalyst for methanol oxidation55 and oxygen reduction.56 Whereas the synthesis of the nanoparticles assembly has been studied in detail, the interparticle distance of the multicomponent nanocrystal surfaces is not fully understood. It has been reported that manipulation of the core composition and linker chain length could provide new opportunities for exploiting binary nanoparticle-structured sensing and catalytic properties.57 Metal oxides exhibit electrical behavior that can vary from electrically insulating (MgO, and Al2O3), wide-band semiconductor (TiO2, SnO2, ZnO, Ti2O3), to metal-like (V2O3, ReO3, RuO2) behavior. Some oxides have several stable oxidation states that are very important in surface chemistry. The oxidation state can control the types of defect that may be formed and the chemisorption that takes place at their surface when they are used as a sensing layer. Usually the process of electron exchange between the conductance band of metal oxides and the adsorbed species is fast. However, the chemisorption process of gases is governed by both surface and bulk properties, which can be slow for sorption and desorption. Reduction of the bulk surface to nanometer or micrometer size is an approach for influencing both the electronic and sensing properties of the metal oxides. The identification and control of structural features of metal oxide cluster size are prerequisite for preservation of its electronic, optical, or magnetic properties. The formulation conditions, time, and temperature can be used to form nanocrystals, nanorods, nanowires, or even nanoparticles with controllable sizes. Classical sol-gel synthesis can be used to control parameters including the diameter and aspect ratio based on the hydrolysis and condensation of metal-halide or -alkoxide precursors in aqueous solution.58 The reported CO gas sensing of single-crystalline one-dimensional (1-D) SnO2 nanocrystals (rod, wire) with a controllable size indicates that the sensitivity can be correlated with the specific surface area of the nanocrystals.59 In order to improve the crystalline arrangement of very small particles, the nonaqueous solution, halide, and surfactant-free synthesis routes were used to produce SnO2, In2O3, Nb2O5, and perovskite nanopowders.60 The oxides are in the form of nanostructured powders that are composed of single crystallites with minimal aggregation and an average size of 2-20 nm.61 Growth of 1-D oxide nanomaterials such as ZnO, SnO2, In2O3, Ga2O3, SiOx, MgO, and Al2O3 with controlled microstructured and size using chemical vapor transport and condensation systems has been reviewed.62 Porous networks of metal oxides provide high surface area relative to bulk structures. In many cases the void geometry afforded by the structure of the material provides enhanced uptake of species relative to the two-

Chemical Reviews, 2008, Vol. 108, No. 2 751

dimensional bulk surface.63 The sensitivity of any device is enhanced due to the increased number of interaction sites associated with the three-dimensional structure. Finally, the chemical and physical properties of nanostructured materials can be much different than those observed for the bulk materials. The conductivity of arrays of TiO2 nanotubes was reported to be approximately 9 orders higher that of bulk TiO2 surfaces.64 The enhanced conductivity was explained in terms of increased surface area, defined stoichiometry, and a greater level of crystallinity than in multigranular oxides. These materials showed potential for reducing the instability associated with percolation or hopping conduction. It is possible to envision that chemical sensitivity will be greatly influenced by the chemistry taking place at the surface since for most semiconductor oxides nanowires the Debye length (measure of field penetration into the bulk) is comparable to the radius of the wire. Furthermore, the increased electron and hole diffusion rate to the surface of a nanostructure allows the analyte to be rapidly photodesorbed from the surface even at room temperature. This implies that surface photoinduced redox reactions occur rapidly and interfacial charge-transfer rates are not limiting. Use of photoinduced desorption of a sorbed analyte for rapid detection is conceivable. Photoassisted desorption will increase the conductance of the nanowire, providing signal transduction for the sensor. Therefore, nanostructures of metal oxides show promise in improving the efficiency of electrochemical and optical sensing devices, encompassing a broad range of applications. Studies have shown that TiO2 nanotubes fabricated by electrochemical anodic oxidation exhibit a highly ordered structural morphology that enhances the photocurrent and photocatalytic behavior relative to TiO2 nanoparticle films prepared by the regular sol-gel methods.65 In addition, incorporation of SnO2 (2.5 nm particles) with TiO2 (25 nm) showed an enhancement (1.5 times) in the photocatalytic activity compared to the pure TiO2 for decomposition of gaseous 2-propanol and evolution of CO2.66 Although the role of dimension, morphology, and composition is clearly identified in the study, the overall recombination rate between electrons and holes is considered to be retarded by trapping the photoexcited electrons from the conduction band of TiO2 by the dopant species (e.g., SnO2). In recent years it has become clear that the reactivities and thermodynamics of small transition-metal oxides clusters exhibit many fascinating and unique properties. Reactions and thermochemistry of small transition-metal cluster ions demonstrate that quantitative thermodynamic information concerning the stability and reactivity of small transitionmetal clusters in the gas-phase provides an understanding of the their function as catalysts at the molecular level, which is applicable to the bulk-phase limit.67 The role of the catalyst in chemical sensor applications is to lower the reaction barrier in metal oxides based on weakening the O-O bond through charge transfer.68 A second possibility is to use systems that contain O atoms without the O-O bonds. The reactivity of transition-metal oxides with oxygen atoms bound only to the transition metal remains largely unstudied.69 The molecular understanding of environmental catalysis of metal oxide clusters and their reactions has been the subject of a review.70 Newer investigations show how particular size-selected clusters of Fe2O3 can become a potential candidate for promoting NO reduction and CO oxidation. The systems may even become dual-task self-

752 Chemical Reviews, 2008, Vol. 108, No. 2

regulating systems for conversion of CO to CO2.71 Unlike the bulk metal oxide system, occupation of different MO orbitals is possible with small clusters, resulting in different ionic charge states. The combined theoretical and experimental study of FeO3( with CO reveals that cation clusters are more efficient in lowering reaction barriers and facilitate catalytic conversion of CO to CO2.72 The importance of d-orbital aromaticity has recently become evident in metal oxide clusters, [M3O9]- and [M3O9]2- (M ) Mo, W). They are unique in that they involve a single fully delocalized metal-metal bond and may exhibit novel chemical, electrochemical, and catalytic properties.73 More recently it has become clear that the cluster activity depends on the number of atoms, geometrical structure, and sorption site at which adsorption of the gas participates.74 Quite apart from any practical application, studies of the reaction kinetics of metal clusters with small molecules have demonstrated dramatic cluster size dependence of reaction rate constants.75 An understanding of the properties of metal oxides as a function of size, geometry, and reactivity is important if these materials are to be incorporated into polymers for sensor applications. The composites formed from these materials may be much different that the individual materials. Growth of a crystal also coincides with advances in understanding intermolecular interactions and supramolecular chemistry.76 Several aspects of solid-state chemistry are of increasing relevance to chemical sensors such as the application of cooperative binding to the differentiation of similar analytes.77 Formation of metal oxide clusters with surface-adsorbed reactive organic groups has been achieved. Preparation of an interesting new type of inorganic-organic hybrid polymer with unique properties is also possible.78 In summary, metal oxides provide both catalysts and catalytic supports that if used with ICPs may produce composite materials with unique sensing properties. However, the catalytic and reactive properties must be understood and exploited to develop selective chemical sensors which are targeted to specific analyte detection.

2.4. Phthalocyanines and Related Complexes Pyrrole-based macrocycles such as metalloporphyrins have a similar structure to phthalocyanines (Pc). Porphyrins, which are less stable to degradation than phthalocyanines, have been used to mimic naturally occurring metal porphyrin complexes (e.g., cytochrome c, hemoglobin, and myoglobin). Specifically, metallophthalocyanines have been of interest for their high thermal stability and well-defined redox activity in sensing applications.79 These compounds are organic p-type semiconductors formed from three distinct components: the macrocycle, the peripheral substituent, and the metal ion coordinated in the core of the molecule.80 Each of these components contributes to the total selectivity of a sensing layer via manipulation of the metal center and substitution of functional groups on the organic ring.81 Their ability to coordinate with almost all the metals and some nonmetals present in the Periodic Table results in a huge number of sensors that have been prepared using the same macrocycle. In solid-state gas sensors the conduction mechanism and molecular interaction contribute to the signal through either binding of the analyte to metal coordination sites or competition with O2 for occupied metal surface sites. There is also a possibility of weak binding (physisorption) to the organic region of the phthalocyanine molecule for noncoordinating analytes, which may be governed by weak hydrophobic or

Hatchett and Josowicz

charge-transfer interactions. The later interaction is related to formation of charge-transfer complexes through coordination of the target, O2 (π-electron acceptor), to metallophthalocyanine (MPc) metal centers at the air/phthalocyanine interface. The interaction leads to formation of oxidized MPc+ complex where the positive charge is delocalized over the phthalocyanine ring and reduces O2- species with injection of hole charge carriers into the bulk solid.82 Resistive sensing concentrates mostly on p-type metallophthalocyanine thin films interacting with oxidizing gases, such as ozone and NOx. The sensing mechanism is based on the gas interacting with the MPc, resulting in injection of holes and an increase in the current. Interaction of phthalocyanines with reducing gases, such as NH3, has the opposite effect.83 The advantage of these materials is that phthalocyanine films can be utilized for gas sensing at temperatures much lower than comparable metal oxide sensors.84 An alternative method for detection involves use of the materials as electrocatalysts. For example, detection of nitric oxide in biological systems on gold electrode and determination of nitrite in real samples using polymeric nickel tetraaminophthalocyanine (p-NiTAPc) film coated glassy carbon electrode has been achieved using the materials as electrocatalysts.85 In addition, detection of NO in biological samples in the presence of interfering ions such as NO2 and neurotransmitters has been achieved by immobilization of the porphyrin on the electrodes using thiolate coordination chemistry.86 For good catalytic activity the catalyst must lower the oxidation and/or reduction potentials of the species. Therefore, an important factor in the catalytic activity of immobilized MPc complexes is identification of the oxidation or reduction site. MPc complexes with metal-based oxidation processes are expected to show better catalytic activity toward oxidation or reduction of analytes than ring-based processes. However, this fact has not been proven to be a general rule. It was found that in some cases the monomeric species of MPc are better catalysts than the aggregated complex. The catalytic activity depends on the thickness, conductivity, and method of immobilization of the film. In order to improve the selectivity of the sensor, Nafion has been included to help eliminate interference from other ions. However, Nafion tends to reduce the sensitivity of the sensor. In summary, MPc should act as a good catalyst by enhancing sensitivity, be stable for prolonged use at different temperatures and pH values, and offer good selectivity for analyte over its oxidation products and other interfering species. One can therefore envision that incorporation of MPc into an ICP may optimize the sensing properties due to the nanostructure of the polymer. The improved nanostructured architecture of thin composite films may enhance interactions between the components and the conducting substrate and maximize diffusion of the guest molecules into the film.

2.5. Biologically Active Materials The field of bioanalytical chemistry continues to expand encompassing biologically active materials, which utilize enzymes, proteins (amino acids), DNA, antibodies, and antigens. A large portion of the biosensors developed thus far have utilized enzymatic reactions. These systems use enzymes that are affixed, immobilized, or encapsulated in a variety of matrix materials. More recently more novel materials have been used, including carbon nanotubes,87 gold particles,88 proteins (amino acids),89 and DNA,90 to provide interaction sites for biological recognition components or

Composites of Intrinsically Conducting Polymers

Chemical Reviews, 2008, Vol. 108, No. 2 753

analytes. The matrix must be chemically inert and not change the properties of the biological component so that the reactivity, potency, or interaction is not lost. In addition, the matrix must have open geometry so that the target analyte can access the biological component. Use of polymer matrices for biosensors has been reviewed previously.91 Detection of biologically active species is typically achieved using either optical or electrochemical methods. For electrochemical systems the biological transducer is incorporated in such a way that signal transduction results in a measurable current. Optical methods require that the biologically active component causes changes in the optical properties such as UV-vis absorbance and fluorescence intensity. In most cases this is achieved by incorporation of a biologically selective enzyme through electrostatic interactions or covalent bonding.92 These systems have been reviewed previously and will not be extensively discussed in this review.93 Polymers are one common host matrix that are suitable for incorporation and trapping of enzymes.94 Electropolymerization of nonconducting films like polyphenol, poly(1,2- diaminobenzene), poly(1,3-diaminobenzene), and poly(o-aminophenol) in the presence of an enzyme has been achieved with controlled enzyme loading. The enzymes remain active during polymerization and encapsulation because these films are generated from water-soluble monomers. Thin uniform films that filter electroactive interferents are possible, minimizing poisoning of the enzyme. Enzyme layers prepared by electropolymerization of p-aminophenol and glucose oxidase (GOx) have shown excellent characteristics in glucose sensor performance.95 Similarly, use of high-volume sol-gels96 and zeolytes97 as immobilization matrices has been realized. The large open volumes of sol-gels and zeolites provide attractive voids for incorporation of biologically active materials. In addition, the large open volumes allow rapid diffusion of analytes into the structure. Finally, two-dimensional arrays of self-assembled biologically active materials have been achieved, eliminating the need for secondary matrices. In these systems the biologically active component is typically covalently attached to the surface using well-known assembly chemistry such as the metal/thiolate interaction.98

transducer matrix to convert the ionic charge into an electronic one, as a mediator and sometimes even as a catalyst. In the following section we will try to describe the role of the various components in the ICP composite materials with respect to sensitivity, response, and recovery time of the generated sensor signal that is a complex combination of transport properties within the host matrix and chemical interaction with the analyte. For application of composites in sensor devices, two important factors need to be addressed: functionality and processability. These properties are related to the aspect ratio of the two components in the matrix, the interfacial adhesion between the additives, electrical conductivity, and structure. All of the properties are greatly influenced by the method used to obtain them, which essentially defines the fundamental challenge for applied ICP/CNT composite research. How does one effectively manipulate nanoscale building blocks to assemble useful nanoscale materials? Besides these essential factors, the properties of the composite can also be affected by the geometric structure of the material itself since it can be processed in the form of films, pellets, and wires. The differences in the volumetric factors between these structures cannot be neglected because they dramatically influence the diffusion of species into the composite and can affect the magnitude of the sensor response. More compact structures or thicker films may slow diffusion and be responsible for slow sensor response. On the other hand, thin films or fibers will be limited in the number of interacting sites, which may become the limiting factor for the dynamic range of the sensor response. Furthermore, processing parameters used for fabrication of structured composites affect, e.g., crystallinity and porosity in these materials. Although the method of ICP/polymer composite sensors preparation is important, selection of the detection method for an analyte is equally important. For example, electronic properties that are strongly influenced by the preparation methods of ICP/polymer composites will have an impact on conductivity changes in conductive sensors. In contrast, in potentiometric sensors the measured change in anion or cation concentration in solution or in work function will depend strongly on the starting redox composition of the polymer.

3. Advances in Composites of Inherently Conducting Polymers

3.1. Composites with Nonconducting Polymers

The general sensing mechanism is based on changes in the physical, chemical, or mechanical properties of the polymer when exposed to analyte. For example, when a conducting composite is exposed to a vapor it permeates into the polymer causing it to expand. The vapor-induced expansion of the polymer reduces the number of electrically conducting pathways for charge carriers. As a result, the electrical resistance of the composite rises. The high segment chain mobility of the ICP in the composite is an important requirement for good absorption of vapors or gases. This property can be improved through addition of nonconducting polymer, carbon, metal, or metal oxide particulates, or biological materials as they create amorphous, flexible regions around the ICP polymer chains in the composite. Consequently, they can be used to increase the number of interacting sites with the analyte, increase the intra- and interchain mobility of charge in ICP polymer chains or even change the affinity of the composite for the electron-donor or -acceptor gas. Furthermore, they can be used as a

A useful approach to overcome some of the physical and chemical limitations, including the processability and mechanical and thermochemical stability of ICPs, involves combining these materials with other well-known nonconducting polymers, such as poly(vinylalcohol) (PVA), polystyrene (PS), poly(methyl methacrylate) (PMMA), and poly(vinylacetate) (PVAc).99 The resulting composite combines the advantageous electrical, redox, or optical properties of ICP with the mechanical properties of the host polymer through the ratio of ICP verses the insulating polymer.100 The intrinsic conductivity of the ICP is not the key factor governing the electrical conductivity of the ICP/polymer composite films. Rather the processability of the two components in solution and the extent of dispersion of the conducting polymer inside the insulating polymer matrix are key.101 In addition, the extent of dispersion of the ICP particles based on the solubility of the insulating matrix in the common solvent and the solvent’s ability to swell and diminish heterogeneous domains inside the insulating polymer matrix strongly influences the electrical properties.

754 Chemical Reviews, 2008, Vol. 108, No. 2

The solvent contributes to changes in the morphology of the film, which changes the response magnitude and reversibility of the sensor. Therefore, it is possible to vary sensitivity toward alcohols and amines and relative humidity by choosing the appropriate polymer matrix and solvent system for casting. However, it is not possible to modify the sensor’s selectivity within the same polymer matrix. Higher solubility for the ICP in common solvent has been achieved using an anionic surfactant such as dodecyl benzenesulfonic acid (DBSA) that is also a dopant for the ICP precursor. Chemical sensors based on coupling of the polymer swelling to optical and magneto- and electrochemical transduction technologies have been recently reviewed.102 These composites have been categorized in terms of the production process as either homogeneous or heterogeneous systems. In homogeneous systems polymerization of ICP monomer is conducted in the presence of the nonconducting host of the polymeric matrix in solution. In heterogeneous systems polymerization is carried out via emulsion, colloidal dispersion of ICP particles, or thermal processing.103 The thermally processed composites have percolation thresholds that are at least 1 order of magnitude higher than the lowest values obtained from solution-processed blends. Thermal processing methods provide poor structural control of the morphology. The solution and thermal processability of doped conjugated polymers allows production of conducting fibers with more open geometry. Properties of these composites such as degradation mechanism, electrochemical behavior in nonaqueous media, resistivity changes in relation to dispersion concentration, ion-exchange properties, conductivity, and photoconductivity studies have been examined previously.104 It can be concluded that the structure and mechanical properties of the electrically insulating polymer in the composite govern the sorption of the analyte molecule into the blend matrix and thus facilitate a change in the effective volume fraction of the conductive ICP phase. The ability of adjacent charge carriers in the ICP to hop or tunnel is controlled by potential barrier between the hopping sites and the host polymer, which can be manipulated by temperature.105 In general, temperature-dependent sensor sensitivity to a particular gas is governed by the ratio of the two components in the composite and by the sensor-gas equilibria.

3.1.1. Composites with Hydrophilic Polymers The solubility of various nonconducting polymers in water is well known and can be utilized to prepare ICP/polymer composites. However, water solubility implies that the materials will be sensitive to water vapor and that relative humidity must be monitored and evaluated before use as sensor materials. In some cases the sensitivity to water vapor can be used to produce highly reproducible humidity sensors. Various ICP polymers have been prepared in the presence of “water-soluble” polymers including PVA and PMMA, polymer (poly-N-vinyl pyrrolidin-2-one (PVP)), and ethylene vinyl acetate (EVA). Composites of polypyrrole with poly(vinyl acetate) with ∼27% PPy (PPy/PVAc) were cast as smooth, strong, and flexible films with electrical conductivity comparable to that of PPy and mechanical properties similar to those of the host polymer. The composite material demonstrates the importance of combining two materials and the benefit of mixing physical and chemical properties of the two dissimilar materials. The threshold conductivity for

Hatchett and Josowicz

the composite was reached at 5% of PPy in the host matrix. The environmental stability of the PPy/polymer composite is improved relative to PPy films that can loose up to 5065% of their conductivity after 45 days of exposure to the atmosphere. PANI/PVA composites have also been prepared with variable amounts of PANI in the composite. Composites with more than 20% by weight of PANI in PVA are immiscible. However, when less than 10% of PANI is present, the composite is miscible. In general, processed PANI/PVA composite films (150 µm thick) do not show any change with respect to relative humidity between 10% and 60%. Also, PVA alone does not show significant impedance changes over the same relative humidity range. When the miscible PANI/PVA composite is exposed to relative humidity between 60% and 90%, the resistance decreases from the high MOhm range to lower values. The resistance of the immiscible PANI/PVA composite increases from the lower kOhm range to higher values for relative humidity between 60% and 90%.106 It is important to note that a relatively small increase in the minority component PANI has a dramatic influence on the overall properties of the composite. The changes have been attributed to swelling of the PVA, which leads to separation of the conducting ICP/ polymer domains for the composites. The overall resistance is based on competitive processes where the resistivity of PANI and PVA decreases and increases, respectively, at high humidity. PANI/PVP composites provide another example of composite materials with high sensitivity to relative humidity. These composites are prepared using poly-N-vinyl pyrrolidin-2-one (PVP), which has high affinity for the water vapor.102 However, this film is less sensitive toward NH3 than pure PANI. This discrepancy was attributed to the low affinity of the PVP for NH3 sorption. In this example the properties of the PANI composite do not necessarily produce a better sensing material. Rather, it demonstrates the influence of each material in the composite on the overall sensing characteristics.

3.1.2. Composites with Hydrophobic Polymers An attractive alternative to the PANI/PVA, water-soluble composite, is the PANI with styrene-butyl acrylate (SBA) copolymer films doped with weak acid.107 The composite shows a linear resistive response for relative humidity between 20% and 95%, fast response time (4-5 s) measured in stepwise increments (of 10% relative humidity), and fast recovery times (10 s). The films were found to be stable for more than 1 year and at temperatures below 50 °C. Wallace et al. reported reliable and predictable linear responses with over a 100% resistance change over a 95% change in relative humidity for PANI and poly(butyl acrylate-co-vinyl acetate) (P(BuA-VAc)) copolymer (40% w/w). The composite with 15% w/w of PANI showed an exponential response with increased sensitivities with an 800% resistance change over a 95% change in relative humidity. The only disadvantage to PANI/P(BuA-VAc) composite is that the response showed a slow but consistent drift which was attributed to an overoxidation of PANI in air. However, the high sensitivity of the composite far outweighs the drift associated with decomposition. Despite the use of relatively low hydrophilic matrix polymers such as poly(methylmethacrylate) or more hydrophobic polymers such as polystyrene, the influence of humidity on the ammonia response of PANI/PMMA and

Composites of Intrinsically Conducting Polymers

PANI/PS sensor films was consistent with expected results. PANI doped with bis(2-ethyl hexyl) hydrogen phosphate (DiOHP) was mixed with poly(methyl methacrylate) (PMMA) and polystyrene (PS) and dissolved in toluene.108 The resulting solution of polymers in toluene was cast to produce free-standing films. Use of toluene as casting solvent introduces porosity in the PANI-PS and PANI-PMMA blend favorable to high diffusion of gases into the membranes. Furthermore, PANI aggregates formed that were homogeneously distributed and completely covered by the host polymer in the matrix in the PANI/DiOHP/PS film. In contrast, PANI aggregates appear to form inside the porous matrix of the composite for PANI/DiOHP/PMMA composites. These aggregates introduce porosity into the PANIPMMA blend and enhance the sensor response and reversibility. The composition of the protonated PANI in the PANI/ DiOHP/PMMA and PANI/DiOHP/PS blends was 65.3%. The conductivity of the films decreased monotonously with increasing relative humidity from 15% to 85%. It is likely that the water molecules interact with the doped PANI, forming hydrogen bonds between the water molecules and the PANI backbone or the donor molecules, influencing the electronic properties of the composites. The PANI/DiOHP/ PMMA composite showed larger decreases in conductivity than PANI/DiOHP/PS due to the higher porosity of PMMA host matrix. It was shown that the PANI/DiOHP/PS and PANI/DiOHP/PMMA blends have much better environmental stability than pure PANI films. In contrast to many of the composites discussed thus far, polypyrrole films chemically polymerized on poly(etheretherketone) (PEEK) and poly(methylmethacrylate) (PMMA) show high sensitivity to ammonia and amines at 100% RH with detection limits of 1 ppm. In addition, the composites can be used in biochemical applications because they are stable in aqueous environment.109 Synthesis of 250 nm thick layer of PPy/PEEK composite has been achieved. In contrast, PPy/PMMA formed only a monolayer of ∼50 nm spheres. The PPy polymerized on PEEK was more sensitive to ammonia and showed little sensitivity to ethanol or water in contrast to PPy deposited on PMMA or to a neat PPy. In addition, the PPy/PEEK composite was approximately three times more stable over time than PPy/PMMA. Composites of PMMA or PS processed with PPy doped with naphthalenesulfonate were also successfully used as acetone/toluene and acetone/ acetic acid sensors.110 The gas-sensing properties of PANI/nylon-6 composite have been evaluated with respect to NH3, CO, and C3H8 detection and quantitation. These studies have focused on the influence of the acid dopant on the sensing properties of the materials. The composites with monocarboxylic acid doped PANI including formic acid, acrylic acid, and dodecyl benzenesulfonic acid respond rapidly to NH3 gas. In contrast, CO and C3H8 gases have no significant effect on changes in the electrical resistance of the composite or neat PANI material.111 It was also shown that pellets produced from PANI doped with HNO3 or camphorsulfonic acid (CSA) have slightly higher electrical conductivity when compared to pellets of PANI/polyimide (PI) doped with either HNO3 or CSI. The electrical conductivity values of PANI-HNO3/PI and PANICSA/PI composites are comparable, in contrast to pellets made from homopolymer alone. The crystallinity of the PANI/PI composites doped with different acids is the key similar conductivity of the materials. Upon contact of the

Chemical Reviews, 2008, Vol. 108, No. 2 755

PANI-CSA in a form of a pristine material or composite with CO, known as oxidizing gas, an electron is transferred from the π-electron network of PANI to the gas, making the polymer positively charged. The charge carriers give rise to increased electrical conductivity of the film. The presence of polyimide in the PANI-CSA pellet improves the sensitivity toward CO interaction due to the fact that the conductivity is smaller by an order of magnitude when compared to the pristine polymer. On the other hand, the electrical conductivity responses are comparable.112 Addition of polyimide to PANI not only reduces brittleness of the film but also improves electrical sensitivity toward CO by increasing the range of exposure temperatures accessible. On the basis of this observation it is expected that the temperature dependence may be used to improve selectivity of PANI/PI composites to gases or vapors by influencing the permeability or the displacement of water from the surface. Generally, the gas sensitivity of the ICP/nonconducting polymers is considerably larger (1 or 2 order of magnitude) than that of the ICPs for the hydrophobic vapor analytes (xylene, benzene, acetone) at high vapor concentrations. Selective conductivity responses have been reported toward acetone and toluene or acetone and acetic acid for polypyrrole doped with naphthalene sulfonic acid (NS) using polystyrene (PS) or polymethacrylate (PMMA) as a nonconducting matrix.113 The selectivity was attributed to solubility parameters of the nonconducting polymers that are comparable to acetone.114 The selective conductivity response was controlled by temperature. It originated from two effects: the swelling of the nonconducting matrix that separated the conductive PPy particles and the PPy-NS interactions with the vapor itself. Moreover, the sensitivity to water was successfully eliminated. However, the long induction period and poorly reproducible responses remains a challenge in the use of these materials as sensors. In summary, preparation of ICP/nonconducting polymer as a composite is simple. However, use of ICP/nonconducting polymers as sensitive, selective sensors can only be accomplished if the properties of the combined materials are adequately understood. For example, the ratio of the components in the matrix must be strictly controlled since it affects the reproducibility of the film sheet resistance and sensing device performance.

3.1.3. Multicomponent Composites In ternary blend systems a conducting polymer is soluble in a uniform mixed nonconducting polymer matrix and does not interfere with the miscibility between the components of the nonconducting matrix. It has been reported that mixed matrices of copolyamide (CoPA)/ethylene vinyl acetate) (EVA) prepared with various PANI concentrations can be utilized as sensors for a homologous series of alcohols. The highest decrease in resistance correlated well with the polarity of the analytes.115 The solubility difference of ICP in CoPA and EVA creates a continuous morphology for containing a highly conjugated path in the nonconjugated matrix. Therefore, the properties of the ternary polymer blend system are frequently better than conductive composites containing nonpolymeric conductive phases like carbon black or metal particles. The most prominent difference between these systems is that carbon black (CB) or metal particles do not produce as high an electronic conduction path in the matrix. Generally speaking, the higher electrical conductivity of the composite films provides higher signal-to-noise and reproducibility.

756 Chemical Reviews, 2008, Vol. 108, No. 2

Favorable miscibility of ICP with the polymer components of the host matrix can easily be investigated using UV-vis, FTIR spectroscopy, optical microscopy, and differential scanning calorimetry. For example, the blue shift of the π-π* transition of ICP observed in the UV-vis spectrum signals the change in the conjugation length of the ICP and it splitting the miscibility with the polymer blend, vibrational shifts in the FTIR, and flattening of bands signals possible molecular interactions in the system, melting point depression, the miscibility and visual examination of the changes in the morphology of the material. In summary, the ternary type of polymer system offers expanded flexibility for realizing high-performance semiconducting architectures which can eliminate the need to design all of the performance requirements into the active semiconducting polymer. In general, sensors using these blends have low power consumption, operate at ambient temperature, and show reversible adsorption/desorption kinetics which are favorable in sensor applications.

3.2. Composites with Carbon Black and Carbon Nanotubes Electrically conductive composites with carbon black (CB) or carbon nanotubes (CNT) and ICP have been produced and their advances in chemical sensing application reviewed.116 These composites are obtained using two methods. The first method consists of the direct mixing of carbon black (CB), dispersion of CNT with or without ICP, or polymerization of ICP on the dried CNT networks. In the chemical or electrochemical oxidative polymerization of the ICP monomer CNTs act a supplementary reactive reagent. All the preparation methods lead to similar composites with improved electrical and mechanical properties of the ICP. Carbon black particulates are one of the simplest carbon materials that have been particularly useful in the form of ICP/CB composite for the electrochemical detection of metal cations using stripping voltammetry.117 The environmentally inert PANI/CB composite, prepared in porous glass as an electrode, showed high chemical and physical stability and allowed high reproducibility for the electrochemical accumulation and sensing of metal cations. These composite electrodes allow more hazardous Hg electrodes to be eliminated. Addition of carbon structures using carbon nanotubes produces composites with enhanced properties relative to ICP/CB composites due primarily to the higher surface area. Most of the sensor applications of ICP/CNT composite have focused primarily on polyaniline, polypyrrole, polythiophene, and poly(3,4-ethylenedioxythiophene) (PEDOT) and characterization of their synergistic electronic and sensing properties. Different interface reactions between the ICP polymer and CNTs can be created which depend on the synthesis process. The interaction of aniline with CNT during polymerization can form charge-transfer complexes associated with disorder in the graphite lattice and defects on the carbon nanotubes. Alternatively, CNTs can create binding sites with the precursor ICP. The interaction between the dispersed ICP and carboxylated CNTs in a solvent can form covalent bonds that lead to polymer-functionalized CNTs and CNT doped polymer. In the case of chemical polymerization of ICP monomer in the presence of CNTs, the radical anion of the CNTs interacts with the positive charges on the ICP, resulting in polymer-functionalized CNTs.118 The electrical conductivity and structure of the composite is greatly influenced by the method used prepare the com-

Hatchett and Josowicz

posite and the relative composition of CNT versus polymer. Preparation of the PANI/CNT composites from solutions of PANI (EB) in N,N -dimethyl propylene urea (DMPU) containing different amounts of the CNTs (1-5%)119 have been compared with poly(3-hexylthiophene) (P3HT) with nonfunctionalized and functionalized solubilized SWCNTs in chloroform.120 A rise in conductivity was observed only in the case of functionalized SWCNTs-P3HT composite having 1-5% weight content of nanotubes. In contrast, the nonfunctionalized SWCNTs-P3HT composite requires almost 20% by weight of nanotubes in the matrix to obtain the same level of conductivity as the functionalized SWCNTs. Although conductivity alone is not the dominant factor for ICP/ CNT sensor applications it demonstrates the influence of carbon composition on the electronic properties of the composites. Since conductivity changes can be utilized as a detection method the functionalized composites may provide a more suitable starting material for development of chemical sensors. Finally, functionalization of the CNT surface not only leads to better dispersibility of the CNTs in various organic solvents121 but also allows grafting of the ICP on to the wall of the nanotube, (CNT-f-ICP).122 In addition, it increases the strength of the interface between the CNT and the ICP matrix and contributes to its environmental stability. Fabrication of PANI/CNT composites through the electrostatic adsorption of the anilinium cation to MWNTs functionalized with COO--functionalized MWNT requires only a very small amount of CNTs to form a conducting network chemiresistor.123 The thickness of the films is an issue because the films are typically much thinner using this synthetic technique. Therefore, the thin film response is faster, but the overall analytical sensitivity is limited.124 For thicker films chemisorption and physisorption can result in long recovery times. In fact, pure ICP does not show a high degree of chemisorption or physisorption in the absence of the MWNT. Operation of the conductometric sensor at higher temperature may enhance the desorption process of the chemisorbed gas and shorten the recovery time but can also decrease the overall sensor response. The mechanism for the interaction of ICP/CNT composite with gas or vapor and the molecular selectivity have not yet been detailed. It has been suggested that it is based on chargetransfer interactions between the composite and the analyte. Furthermore, it has been also reported that cast membranes of chemically polymerized ICP/CNT material may yield more uniform coatings with much higher surface area than electrochemically polymerized material.125 Multipulse chronoamperometric electropolymerization (MCE) has been used to fabricate composites with larger specific area, enhanced electronic conductivity, and ionic transport capacity critical for electroanalytical sensors.126 In general, higher surface areas provide an increased number of interaction sites within the composite that are available for sensing. The selective responses of chemoresistor to polar and nonpolar vapors can be predicted based on the ionization potential difference of the vapors. The greater the ionization potential, the weaker the polarization should be. This was demonstrated using a chemiresistor fabricated with a paste made from chemically polymerized 3-methylthiophene in the presence of COOH-functionalized MWCNTs, mixed with polyethylene oxide (used as a binder) deposited between two palladium electrodes. This chemiresistor was sensitive to chloromethanes (CH2Cl2, CHCl3, CCl4, CH4) with fast response times. However, it was not sensitive to acetone,

Composites of Intrinsically Conducting Polymers

acetaldehyde, benzaldehyde, tetrahydrofuran, methanol, and ethanol vapors providing high selectivity.127 The high selectivity of this composite was based on the analyte adsorptivities during analyte interaction. The suggested mechanism involves lowering of the free energy of the analyte due to the polarized structure of the composite, reducing electron flow in the sensor circuit. Similar interactions have been exploited using electron donors such of ammonia, trimethylamine, and triethylamine vapors with CNTs-COOH/PANI composites.128 The changes in electron affinity associated with functional groups such as methyl, ethyl, and hydrogen leads to variation in gas sensitivity and response rate of the chemiresistor. It was shown that addition of CNT to PANI actually reduced the sensitivity to triethylamine, but the current baseline increased greatly which is favorable for the stability of the sensor. The existence of a strong interaction between CNT and PANI was shown to be an efficient way to adjust the baseline current and properties of gas sensors. In contrast, strong interactions between the analyte and the specific binding sites in the composite contribute to a high, irreversible response. The effect of adsorption of gas molecules on functionalized field-effect-transistor have also been tested. However, the mechanism for the enhanced sensitivity has not been determined. The results are not fully understood and somewhat controversial due to possible modulation of Schotky barriers.129 For example, two distinct responses were observed for the polyethyleneimine (PEI)/CNT-functionalized FET upon exposure to the electron-withdrawing NO2 molecules. A steep and nonlinear conductance change at low gas concentration (up to 20 ppm) and linear conductance change at high gas concentration were observed. It has been suggested that the response at high concentrations may account for the charge transfer between the adsorbed gas molecules and the CNT/composite. A Langmuir-type isotherm can be used to describe adsorption at low concentrations of NO2 consistent with a steep and nonlinear change in conductivity. This observation was independent of the type of the coating (CNT with PEI or without), suggesting that the PEI does not alter the binding energy of adsorbed molecules. The response at high concentration has to be taken into account when the vapor is at ppt concentration.130 Recent studies have also demonstrated that CNTs enhance the electrocatalytic properties of phosphomolybdic acid (H3PMo12O40) in polypyrrole films when used as an amperometric sensor. The studies are predicated on the enhanced catalytic activity associated with the high electron density of the polymer and surface reactivity of the composite. Use of galvanostatically deposited film of PPy doped with the heteropolyanion on CNT-modified gold electrode has been particularly useful in the detection and quantitation of nitrite.131 An enhanced electrocatalytic reduction current for nitrite was observed for PPy-(H3PMo12O40)/CNT/Au as compared to the electrode without CNT. The reduced response of PPy-(H3PMo12O40)/Au demonstrated that CNTs promote electron transfer between the heteropolyanion and the electrode surface. The amperometric response to nitrite was about four times higher than that of the electrode without CNT reaching a detection limit of 1 × 10-6 M. A similar method for grafting electrochemically deposited PANI films on CNT and incorporation of nanoclusters of nickel hexacyanoferrates has been applied for detection of cesium ions.132 New electroanalytical applications have also been examined with ICP/CNT composites. The oxidation state of the

Chemical Reviews, 2008, Vol. 108, No. 2 757

ICP is particularly relevant to these applications. Specifically the equilibrium potential of the two partially oxidized redox couples of the ICPs are critical. In polyaniline the equilibrium potential is defined by the ratio of the deprotonated base (emeraldine base, EB) and the protonated salt (emeraldine salt, ES).133 Composites of PANI/CNT networks (150-200 nm thick) deposited from sodium dodecylsulfate (SDS) and coated electrochemically with PPy or PANI showed reproducible, linear, fast, and stable pH responses in buffer solution of pH 1-13.134 The dynamic range and stability of the response was significantly improved due to the higher thickness of the film and different types of binding sites present in the film when compared to pure CNT network. In addition, excellent electrocatalytic properties have been reported for overoxidized PPy/SWNT composite deposited on glassy carbon electrode (GCE) from sodium dodecyl sulfate. The oxidized composite has been used for amperometric sensing of nitrite, ascorbic acid, dopamine, and uric acid.135 The useful concentration ranges for dopamine, uric acid, and ascorbic acid were between 1.0 × 10-3 and 1.0 × 10-6 M with detection limits of 3.8 × 10-7, 7.4 × 10-7, and 4.6 × 10-6, respectively. The electrodes were stable, selective, and sensitive to the biological species. The field of biosensing has also benefited from ICP/carbon composite materials. For example, amperometric glucose sensors have been made by modifying the platinum surface with CNTs followed by electrochemical deposition of 3,3′diaminobenzidine (DAB). The DAB film introduces NH2 functional groups into the composite that are use to immobilize glucose oxidase.136 Although there are numerous glucose biosensors discussed in the literature, this material is more selective and has a higher response to glucose at a much lower potential than other materials. The lower detection potential allows suppressed interference signals from ascorbic acid or uric acid. The enhanced properties are the direct result of the symbiotic properties of the composite. Preparation of poly(1,2-diaminobenzene)/CNTs nanoporous composite on a glassy carbon electrode via a multipulse chronoamperometric electropolymerization process has been demonstrated. The composite provides an excellent platform for fast determination (5 s) of NADH at a lower concentration range (from 2.0 µM to 4 mM) with detection limits of 0.5 µM in phosphate buffer. These values are significantly lower than the response of the polymer alone. That performance is a direct result of the enhanced electronic structure of the composite and high surface area afforded by the CNTs coated with polymer.126 ICP/carbon composites have been applied to a number of common biosensor systems. For example, an effective molecular template for the electrostatic rejection of dopamine has been prepared from a DNA-dispersed CNT in self-doped poly(anilineboronic acid). The sensor has improved conductivity, redox activity, and stability of the polymer in neutral solutions during dopamine detection at very low concentration (1 nM).137 Similarly, electrochemical DNA biosensors based on nucleic acid recognition processes exploit the properties of ICP/CNT composites. For this system, PPy/ MWCNTs-COOH-modified glassy carbon electrodes are incubated in the presence of 1-ethyl-3(3-dimethylaminopropyl) carbodiimide (EDAC), and NH2-terminated ssDNA probes were used as the DNA hybridization transducer.138 Finally, ICP/CNT composites have been used to detect complementary DNA sequences as low as 5.0 × 10-12 mol/L using electrochemical impedance without a hybridization

758 Chemical Reviews, 2008, Vol. 108, No. 2

marker or intercalator. Use of hybridized DNA probes provides a high degree of selectivity at a high cost. Therefore, when possible nonhybridized systems are preferred provided they are selective and sensitive for the target DNA structure. In summary, it can be said that ICP/CNT composites are very useful for facilitating electron transfer and increasing efficiency of binding sites for specific sensing interactions without the need of redox mediators. However, they require functionalization because the carbon structures alone do not provide selective interfaces within the sensor composite. In addition, difficulties in incorporation of the carbon species must be overcome for proper accommodation within the sensor device structure. Provided these issues are addressed, the enhanced electronic conductivity of the carbon materials is promising to provide high sensitivity responses to a variety of analytes.

3.3. Composites with Metallic or Metal-Containing Components Incorporation of metals, metal oxides, and organometallic species such as metallocenes in intrinsically conductive polymers can enhance electron transfer through a direct or mediated mechanism with improved conductivity and enhanced stability.139 For these systems the electron-rich polymer often acts as a chemical receptor or scaffold for the secondary component. The electron density of the conductive polymer is important, providing stability to species that may be electron deficient. However, it has also been suggested that the polymer provides stability through exclusion of surface contaminants that may interfere with the surface chemistry of the deposit. The structure of the conductive polymer plays a role in the dispersion of the species, their aggregation, and formation of the composite material. The polymer is used to provide high surface, protection against the fouling of the metal catalyst, and a scaffold for high dispersion and anchoring of the metal particles. The conductivity of the composite allows the electron density of the polymer to be controlled through the applied thermodynamic potential which influences chemical reactions at the metal surface. The ability to finely disperse the secondary species in the polymer ensures high surface area and possible enhancement of the unique characteristics of the composite. The synthetic method employed is an important factor to consider with regard to the homogeneous distribution of the secondary component. There are multiple methods used to prepare the ICP composite materials with metallic species distributed at the polymer surface in the vicinity of the electrode/polymer interface or throughout the bulk of the polymer. The approach used to introduce the metallic species typically determines the degree of dispersion or deposition. For example, dispersion of the polymer using pulsed deposition and relaxation (spontaneous reduction of the species) or co-deposition of the ICP with the active species using cyclic voltammetry has been utilized to introduce metal species.140 One common approach involves the chemical reaction and uptake of a species by the conductive polymer. In this case the species are introduced by controlled reduction of metallic anions such as AuCl4-, PtCl42-, and PtCl62- by manipulating the redox sites within the ICP.141 The uptake of the negatively charged anion of interest is based on the oxidation of the polymer and the need to maintain charge neutrality in the system. Reduction of the anion occurs as the potential is switched to negative values. In contrast, the oxidized form of the metal

Hatchett and Josowicz

species can be introduced to the polymer using metal salts followed by the electrochemical, chemical, or spontaneous reduction of the metallic species. The most common metals incorporated into the ICP include gold, platinum, and palladium as nano- or/and microparticles. In many cases formation of the metal nanoparticles occurs prior to incorporation into conductive polymer matrix. In such cases, the preformed clusters of metal, metal oxides, or metallocenes are dispersed in solution containing the ICP precursor. Chemical oxidation of the precursor results in encapsulation of the metal species to form the ICP composite.142 The resulting materials can then be cast onto substrates for subsequent applications. Recently, the direct chemical synthesis of composite materials using polymer precursors and oxidizing anions such as AuCl4- and PtCl62- has been reported.143 Each of the methods discussed has its advantages and disadvantages that must be weighed before they are used to prepare composite materials. The properties and application of composite materials containing metals, metal oxides, and metallocenes are quite diverse. For example, polyaniline, polythiophene, and polypyrrole have all been utilized as scaffolds for the incorporation of metal species in catalytic applications.144 Catalysis at metal surfaces embedded in conducting polymers is enhanced due to the conductive environment. The electrooxidation of CO adsorbed on Pt particles incorporated into the PANI film surface occurs at a lower positive potential than on bulk Pt alone. The oxidation potential depends not only on the thickness of PANI film but also on the amount of Pt particles incorporated into the PANI matrix.145 Likewise, metal oxides such as SnO2 and TiO2 have received much attention due to their unique properties and use in applications such as fuel cells and batteries.146 Metallocenes can be used for a host of applications based on their optical and electronic properties.147 When combined with ICPs each of these materials provides unique properties that are consistent with the physical and chemical properties of the species. The properties of the composite materials are exploited when developing chemical sensing applications. It is well known that nanoparticles demonstrate special size- and shape-dependent properties.148 For example, the nanoparticle size and shape have significant effects on the localized surface plasmon resonance spectrum of Ag and Au nanoparticles.149 The sensing application typically dictates the critical size of the metal species in the composite. It has been observed that aggregation and size distribution of the metal nanoparticles vary with the reaction time. It is difficult to control both the size distribution and aggregation of the metal particles embedded in the polymer matrix regardless of the technique used to create the composite. Therefore, the size of the metal particle is often inhomogeneous or variable. Within this section we would like to review some approaches for the synthesis and use of ICP/metallic species composites for chemical sensing.

3.3.1. Gold Chemical sensors based on gold nanostructured particles are of interest for sensing gases and vapors and as binding sites for enzymes, proteins, and nucleotides. While the size of the gold cluster used to affix bioactive species is not critical, detection of gases at gold surfaces is dependent on this parameter. Therefore, the size distribution of gold in ICPs must be monitored based on the application. The dependence

Composites of Intrinsically Conducting Polymers

is governed by the shift of the binding energy of Au clusters in the PANI matrix associated with the cluster size. The binding energy of the smallest Au particle size (0.2 nm) has been determined to be 84.49 eV. In contrast, the binding energy for the largest (30 nm) Au particle is 84.09 eV, a difference of 0.40 eV. The properties of gold clusters, ∼5 nm, approaches bulk metal with respect to binding energies. In contrast, clusters of 1-2 nm diameter have approximately 30-250 atoms with higher binding energies.150 In addition, the nanoscopic gold catalysts are active at much lower temperatures in comparison to platinum-group catalysts for CO oxidation.151 There is a debate regarding the underlying basis for the activity of these nanoscale materials contributing to the complexity of Au-based catalysts. The optical properties of nanoscale metal particles and composites are often unique. For example, femtosecond transient absorption dynamics of the gold and goldpolypyrrole nanoparticles by photoexcitation at wavelengths of 580 and 400 nm are different.152 In contrast to gold nanoparticles, PPy/Au nanoparticles exhibit negligible changes in the time constants for electron-phonon interactions for increasing beam intensity. In addition, the amplitudes for phonon-phonon interactions of the gold polypyrrole nanoparticles are enhanced significantly with increasing beam intensity in comparison to bare gold nanoparticles. These features illustrate that polypyrrole provides an efficient channel for thermal energy relaxation of the gold nanoparticles in the PPy/Au composite. Not all optical properties are enhanced by formation of the PPy/Au composite. Photoexcitation of gold particles and the PPy/Au composite at 400 nm is nearly the same regardless of changes in the excitation intensity. Chemical oxidation-reduction reactions that result in the uptake of gold into the conducting polymer can also be spontaneous. When the reaction conditions become thermodynamically unfavorable the process ceases. It was demonstrated that the spontaneous reduction of Au from halogenaurate salt in acidic solution into previously formed PANI films containing either HBF4 or CH3COOH acid dopant changes the work function of the PANI/Au composite. It needs to be pointed out that the nucleation and growth of the gold particles in the PANI is a continuous interplay between nucleation and diffusion of the aurate ions in the solution. The film morphology depends on the thickness and acid used for its polymerization, which affect the availability of nucleation sites (imine to amine ratio). The size of the Au clusters and their distribution within the polymer matrix is a function of the immersion time in solution.153 More recent studies have used the aniline monomer and metal anion, AuCl4-, to produce PANI/Au composites.154 The one-step process has also been used to form PANI/Au structures in the presence of soft templates such as D-camphor10-sulfonic acid (CSA), which acts as both a dopant and a surfactant in formation of one-dimensional PANI/Au coaxial nanocables with an average diameter of 50-60 nm and lengths of more than 1-2 µm.155 It was found that the probability of formation and the size of the PANI/Au nanofiber depends on the molar ratio of aniline to HAuCl4 and the concentration of CSA, respectively. The conductivity of a single gold/polyaniline nanocable was high, 77.2 S cm-1. Hollow PANI nanotubes, with an average diameter of 50-60 nm, were also obtained successfully by dissolving the Au from the composite.

Chemical Reviews, 2008, Vol. 108, No. 2 759

In contrast to the chemical synthesis of PANI/Au composite the thermodynamics of the synthesis can be controlled electrochemically with the applied electrochemical potential. In these processes the potential dependent uptake and reduction of AuCl4- ions is achieved in the preformed polymer film of either PANI or PPy.156 In addition, studies of the uptake of HCl in PANI films have identified a similar method for creating PANI/Au composites termed “moving electrochemical front” synthesis. The technique is based on the potential-dependent generation of chloroaurate ion at the sacrificial Au electrode surfaces during normal acid doping of the polymer. The electrochemical generation of AuCl4was reduced in the PANI film to form the PANI/Au composite.148 Comparative experiments between the electrochemical and chemical method for Au formation in the PANI matrix have identified the synthetic differences between the composites.139 In the case of electrochemically prepared PANI/Au composites the reduction of Au has been shown to reduce proton doping at the nitrogen heteroatom. Acid doping for the chemically prepared PANI/Au composite is relatively unchanged in comparison. The results indicate that electrochemically reduced gold nanoparticles interact more strongly with the polymer when they are formed electrochemically. A unique approach for fabrication of H-bonded multilayer thin films with gold nanoparticles has been reported.157 This system is unique because the composite is produced without the use of an additional reducing agent. The system reveals interesting nanostructural and spectroscopic properties that require further investigation. The impact of the synthetic method remains to be determined. The number of sensors that utilize ICP/Au composite materials is quite small. However, the optical and catalytic properties of the composites are well studied and suggest that the materials can be utilized for chemical sensing. For example, conductive polymer/gold composite materials have been used in the amperometric sensing of NO2. The analytical sensitivity of PANI/Au/Nafion composite is on the order of 2.54-1.40 mA/ppm for NO2. The material does not specifically utilize gold clusters embedded in the polymer matrix. Rather, the gold is deposited on Nafion and the polyaniline is electrochemically grown on the Nafion/Au substrate to create the Nafion/Au/PANI sandwich structure. The membrane is then used to monitor the cathodic reduction of NO2. The conductive polymer is an important component that improves the properties of the composite material. The Nafion/Au/PANI coating increases the sensitivity of the sensor relative to Au/Nafion alone.158 Aqueous-based ion sensing has been achieved using stripping analysis at gold surfaces. The motivation was primarily based on eliminating the need for mercury working electrodes in stripping analysis. Gold-modified carbon paint screen-printed electrodes were developed as an alternative to mercury-based electrodes for determination of metals such as lead, copper, and mercury.159 More recently, conductive polymers have been utilized as a template for gold stripping electrode and detection of arsenic.160 The composite material is prepared through the electrochemical reduction of gold in a PANI membrane formed on a glassy carbon electrode. For comparison, additional electrodes were prepared in the same manner on a pristine glassy carbon electrode without a thin layer of PANI. The work demonstrates that PANI provides a much higher surface area than glassy carbon alone that translates into higher sensitivity of the analytical device. The

760 Chemical Reviews, 2008, Vol. 108, No. 2

limit of detection for the sensor was calculated to be 0.4 ppb, well below concentrations that are considered safe. More extensive work exists with respect to the application of conductive polymer/gold composite materials as biosensors. In many cases the gold is chemically modified to impart unique chemical characteristics that can be utilized for sensing. For example, use of gold and silver nanoparticles for glucose sensing has been recently reviewed.161 This work utilized dextran rather than conductive polymers to prevent nanoparticles in solution from aggregating. Physical changes in the aggregation of gold particles in the presence of glucose were examined using the fluorescence activity of the gold at 650 nm. Use of conducting polymer/gold nanoparticles for glucose sensing has been achieved using the absorbance of the gold nanoparticles in the presence of glucose. The absorbance spectrum of polyaniline boronic acid/gold nanoparticle composite is extremely sensitive to the dielectric environment and shows a decrease in intensity of the band located at 543 nm as glucose concentration increases. The key to these sensors is the dispersion of nanoscopic gold into the polymer. In addition to the high surface area provided by the dispersion of nanoparticles of gold, the chemical properties of the nanodeposits are different from the bulk metal.162 These two works were preliminary studies and did not provide any information regarding the selectivity or sensitivity of the composite systems with respect to the target analyte, glucose. These studies are examples of how ICP/Au nanoparticle composites can be utilized as glucose sensing materials. They also provide insight in the preparation and incorporation of the materials in electrochemical systems. A more thorough study of glucose biosensing using conductive polymer/gold nanoparticle composites has been conducted.163 In this study polyaniline/gold nanocomposites were produced in two steps. First, the HAuCl4 was reacted with H2O2 to produce the nanoparticles of gold. This solution was then reacted with aniline to produce the composite material, PANI/Au. Electrodes were prepared by casting PANI/Au composite onto an electrode’s surface, reacted with glucose oxidase, and capped with Nafion. Detection utilizes the H2O2 produced by the PANI/Au/enzyme/Nafion when glucose reacts with glucose oxidase. The sensor produced has a dynamic range between 1.0 × 10-6 and 8.0 × 10-4 mol/L and a detection limit of 5.0 × 10-7 mol/L. The amperometric sensitivity was calculated to be 2.3 mA/M. The influence of structured composites on glucose sensing has also been examined. PANI/polystyrene/Au nanoparticle rods were prepared on electrode surfaces with glucose oxidase incorporated into the membrane.164 Oxidation of glucose was examined for membranes with and without Au nanoparticles, and calibration plots were obtained for both. The membranes with Au were shown to be approximately two times more sensitive than membranes without the metal species. In addition, the structured rod membranes were 25 times more sensitive than simple membranes of the composite for equal gram weights of material used. The increase is attributed to higher surface area and a more open geometry for the rod membranes. Although there are few examples of PANI/Au composite biological sensors, the glucose studies suggest that other biosensing systems can be produced using similar enzymatic processes. Plasma resonance sensors are quite common and have been used for a variety of surface-based sensing. In these sensors the gold nanoparticles are modified with secondary compo-

Hatchett and Josowicz

nents such as antibodies, antigens, and self-assembled monolayers (SAMs) that act as signal transducers.165 Interactions at the modified surface influence the localized surface plasmon resonance of the modified gold particles through changes in UV-vis absorption. The secondary components attached to the gold provide a selective interface for the target analyte to interact.166 The high molar absorptivity associated with surface plasmon resonance ensures high analytical sensitivity for the devices.167 A recent article has suggested that electroactive plasmonic sensors constructed from conductive polymer/metal composites may become a reality.168 In this work, gold nanoparticles are sandwiched between PANI membranes and potential-dependent modulation of the surface plasmon resonance of the gold is observed. Reduction of the surface plasmon resonance is observed as PANI becomes more oxidized. This work suggested that the oxidation state of PANI might be utilized in sensing regimes based on surface plasmon resonance of the gold nanoparticles. The sensing mechanism highlights a symbiosis that exists between the polymer matrix and the gold particles being exploited in chemical sensing. In addition, modification of gold nanocomposites can also be employed, opening other sensing pathways based on the immunoassay work discussed previously. Modification of gold surfaces is well known.169 Monolayers of n-alkanethiol and functionalized alkanethiols have been self-assembled onto a gold surface, providing stable metal/organic surfaces that permit introduction of a variety of functional groups onto surfaces.170 However, use of alkanethiol derivatives in conductive polymer/gold composites has not been thoroughly explored. One explanation could be their poor oxidative stability. The recently synthesized poly(propylenesulfide)based copolymers that chemisorbed on gold surfaces and is much more robust to oxidation as compared to alkanethiolates may lead to a novel approach for surface modification.171 Aside from that, a similar approach using thiol moieties seems logical to be transferred to conductive polymer/gold composites for gas-based sensing. The large surface area of the polymer and high dispersion of gold in the matrix should provide enhanced signal for sensor devices through the increase in surface area. More importantly, use of derivatized thiols imparts chemical selectivity to an otherwise nonselective surface. A recent review of selfassembled monolayers provides examples of thiol surfaces that have been utilized to develop selective sensors.172

3.3.2. Platinum and Palladium Metal Inherently conducting polymer/platinum or palladium composites can be formed in much the same manner as the conductive ICP/gold composite materials discussed previously. The most common method to prepare PANI/Pt composites involves direct reduction of anions such as PtCl42- and PtCl62- into the polymer173 or chemical synthesis of PANI in the presence of PtCl42- and PtCl62-.141 Similarly, synthesis of Nafion-PANI-Pd or Nafion-PANI-Pd composites on a Nafion-modified glassy carbon electrode has been accomplished.174 Synthesis of ICP-metal nanoparticle composite materials having a nanofiber-like morphology has been carried out by Mallick et al.175 The nanofibers composite was synthesized using 3,5-dimethyl aniline as a polymer precursor and Pd-acetate. During the reaction 3,5-dimethyl aniline undergoes oxidation and forms poly(3,5-dimethyl aniline), whereas reduction of Pd-acetate results in formation of Pd nanoparticles. The major advantage of this procedure

Composites of Intrinsically Conducting Polymers

is that both the polymer and the nanosized metal particulates are formed simultaneously yielding a good internal electrical contact between them. The dispersed Pd nanoparticulates were all of the same size (∼2 nm). Polyaniline-Pd nanoparticle composite was also prepared by mixing previously formed polymer with Pd nanoparticles (Pd-NPs). Oyama et al. reported that the amount of Pd-NPs dispersed within the PANI film affects it redox chemistry.176 The Pd-NPs are stable as metallic state Pd(0) in the PANI-Pd composite film, whereas it exists as ionic state Pd(II) by electrochemical reaction with PANI. Conductive polymer/Pt and conductive polymer/Pd composites have been primarily utilized in fuel cell applications and catalysis. The catalytic oxidation reactions of formic acid, methanol, ethanol, and hydrazine at platinum composites have been evaluated for fuel cell applications.177 Polyaniline/ Nafion membranes with Pd have been used to examine the electrocatalytic oxidation of formic acid for fuel cell applications.178 Other studies have examined the catalytic hydrogenation and oxidative coupling reactions at PANI/Pd composite materials.179 Use of conductive polymer/Pt and Pd composites for chemical sensing has been limited in comparison to development of materials for catalysis. However, there are a few studies that have identified sensing regimes that can utilize conductive/Pt and Pd composites. For example, amperometric sensing of hydrogen has been achieved at Nafion membranes containing platinum.180 In this study the influence of Pt loading was examined with respect to the analytical sensitivity of the device. The results indicated that the highest sensitivity was on the order of 0.716 µA/ppm in the concentration range of 1260-5250 ppm. Finally, hydrogen sensing using Nafion/Pt composites has been examined.181 The selectivity and sensitivity of the conductive polymer/Pt composite was maximized while minimizing Pt loading in this study. The hydrogen sensor produced using the Nafion/ Pt composite had long-term stability at room temperature with an analytical sensitivity of 0.01 µA cm-2 ppm-1. Fewer examples of conductive polymer/Pd sensors exist in the literature. The chemical sensing of methanol using PANI/Pd nanocomposites has been achieved.182 The material is produced through oxidative polymerization of aniline in the presence of Pd nanoparticles. The resistance of the composite was examined as a function of methanol vapor concentration over a concentration range of 1-2000 ppm. However, saturation of the sensor was observed at concentrations higher than 10 ppm, indicating that sensor has a very small dynamic range. The response of mixtures of other alcohols with methanol did not change the response of the sensor appreciably, indicating the composite selectivity responds to methanol in the presence of possible interfering species. Finally, palladium-modified multiwalled nanotubes have been utilized in the chemical sensing of benzene at room temperature.183 Although there are few examples of conductive polymer/ Pd composite sensors in the literature, there are many examples of the use of Pd with other materials to produce sensors. Use of polystyrene/Pd nanocomposite for hydrogen sensing has been achieved.184 Electrochemical detection of hydrazine has been achieved using multiwalled nanotubes “decorated” with Pd nanoparticles.185 These works suggest that ICP/Pd composites may be utilized for similar sensing applications that build on the principals developed previously in the study of nonconducting polymer and CNT composites.

Chemical Reviews, 2008, Vol. 108, No. 2 761

Biosensing applications have utilized the catalytic properties of the metal component of the ICP/Pt composites. Highly dispersed Pt particles in superfine-fibrous polyaniline (PANI) synthesized on stainless steel electrode have excellent electrocatalytic activity for H2O2 oxidation. When glucose oxidase is combined with the PANI/Pt composite the material functions as a glucose sensor. Detection of hydrogen peroxide generated in the presence of glucose oxidase is utilized to detect the sugar.186 The enzyme electrode has excellent performance with a large steady-state current response. In addition, the amperometric response time to glucose is fast with minimal interference from other biological components encountered including uric and ascorbic acid. It was observed that with addition of Pt the interference from uric and ascorbic acid was greatly diminished and minimized. In summary, ICPs can be utilized for dispersion of metal species in place of other materials to produce composites with similar or new sensing properties. Specifically, catalytic reactions and sensing regimes that utilize Pd and Pt particles and conductive polymers should be examined more thoroughly. When possible, sensing applications that utilize nonconductive/metal composites should be reevaluated with ICPs. The advantages of high electron density and potentialdependent control of the oxidation state of the conductive polymer should be utilized and when possible evaluated with respect to known sensing mechanisms.

3.3.3. Copper and Nickel Metal Although Au, Pt, and Pd are the most common metallic species that have been used to produce ICP/metal composites, Cu and Ni have also received some attention. Use of PANI/ Cu for detection of chloroform vapor at ppm levels has been tested.187 The composite was prepared by adding copper colloid during chemical polymerization of aniline. It was used as a pressed pellet for sensing applications where the composite functioned as a chemoresistor. On the basis of FTIR spectra and a comparative study of a neat PANI it was concluded that the unique surface activity of Cu nanoparticles plays a major role in the reproducible and reversible response of the composite to chloroform vapor. Free-standing PANI/Ni film with a Ni content as high as 4.12% was prepared by a simple one-step electrochemical method.188 It was shown that the size and distribution of the Ni deposits are influenced by the electrolytic bath composition and potential sweep rate. In addition, the PANI matrix exhibited excellent ferromagnetism after incorporation of Ni. It was determined that the electrical conductance of PANI was also modified by the presence of Ni in the polymer matrix.189 For example, the conductivity of the composite was lower than that of neat PANI. The decrease in conductivity was attributed to the partial blockage of the conductive path by the Ni particles embedded in the polymer matrix. The synthesis of PANI/Ni/PANI composite materials has also been achieved using layer-by-layer deposition.190 The composite was characterized electrochemically. It was shown that the nickel-nanoclusters induce changes in the structural morphology of PANI matrix, influencing the electrooxidation of PANI chains. In addition, the clusters cause bridging between the PANI chains/fibrils, promoting delocalization of electrons within the polymer. A more novel application can be envisioned that utilizes the electrocatalytic activity of Ni/PANI and NiMo/PANI layers.191 The materials show promise as catalytic materials for chemical and biological sensors. In addition, alloying

762 Chemical Reviews, 2008, Vol. 108, No. 2

nickel with transition metals (W, Mo, Fe) should increase the intrinsic electrocatalytic activity. Furthermore, the magnetic alloys can act as functional components for the separation of biorecognition complexes and amplified electrochemical sensing of DNA or antigen/antibody complexes. Preliminary studies on sensing properties of nickel incorporated into a Nafion have been conducted.192 Potentialdependent switching of a magnetic responsive composite of 2.5 µM Nafion with 300 nm Ni particles obtained through a standard ion-exchange method as a function of applied current has been observed. Within a proper operational range of (0.7 V, Ni-doped multifield responsive, ionic polymermetal composite shows good paramagnetic characteristics and excellent electric responsive properties without oxidationinduced performance degradation.

3.3.4. Metal Oxides Composites containing conductive polymers with metal oxides, TiO2, ZrO2, and WO3 to name a few, have been produced. These composites minimize the need to operate at high temperature (300-500 °C), typical of bulk metal oxides. Composite sensors produced using ICP/metal oxides have enhanced stability and respond to both oxidizing (NO2) and reducing gases (CO2, CO, H2). It is documented that the method of preparation of TiO2 or SiO2 influences the operating temperature. More recently it has been shown that sensors prepared from sol-gel microstructures of TiO2 and SiO2 can be operated at temperatures as low as 200 °C with high stability in the presence of CO and NO2 gas.193 TiO2 and SiO2 have been added to solutions containing aniline monomer and the chemical oxidant ammonium persulfate to form the sol-gel composite.194 These materials show very high thermal stability with 75% retention of mass at temperatures in excess of 800 °C. Although these materials were not used as chemical sensors, the study demonstrates the high stability of these materials. Use of ICP/TiO2, SiO2 nanocomposites for CO gas sensing has been examined, but the study is not detailed and does not provide either the dynamic concentration range for CO gas or the analytical sensitivity of the device prepared from the composite.195 In addition, there is no attempt to examine possible interference from other gases. However, the study does demonstrate the response of the composite material to CO gas at room temperature. This study is important because it demonstrated that elevated temperatures are not required to observed changes in conductivity/resistance of the PANI/ TiO2, SiO2 composite. It suggests that the ICP and nanostructure of the TiO2 and SiO2 deposits are important for lowering the operating temperatures of gas sensors when compared to bulk films. Polypyrrole-, polyaniline-, or polyhexylthiophene-coated SnO2 or TiO2 nanoparticles were also prepared using the selfassembly (SLB) procedure.196 The process was repeated multiple times to obtain the desired thickness of the nanocomposite films. The ultrathin films were used for CO, aromatic hydrocarbon, and NO2 sensing applications. The change in resistance at room temperature upon exposure to NO2 was recorded, decreasing exponentially with concentration. The results suggest that the gas may act as a dopant for the conducting polymer. The material demonstrated high sensitivity with large changes in resistance at all concentrations. It is worth noting that the response showed thickness dependence with instantaneous and highly reproducible response as the film thickness decreased. In addition, the

Hatchett and Josowicz

thin films are better for low gas (ppb) detection levels. The composites demonstrated higher sensitivity, stability, and reversibility when compared to neat ICP films. In addition, elevated operating temperatures were not required, with polyhexylthiophene/SnO2 nanocomposite films exhibiting good detection characteristics at room temperature. Additional studies of metal oxides doped with platinum and palladium have been examined. For example, surface-doped tin nanoparticles show a large increase of sensitivity in humid air and a surprising inversion of sensitivity toward carbon monoxide in dry atmosphere (relative humidity ≈ 5%) when going from undoped to Pd- or Pt-doped sensors.197 These studies are preliminary in nature and demonstrate the influence of metal doping on metal oxides. The impact of these systems remains to be evaluated. In contrast, the ICP/SnO2 or TiO2 composite systems have been shown to be one of the most interesting nanostructured materials for gas-sensing application. These materials show an excellent change in conductivity at room temperature and possess excellent optical and catalytic properties that can be utilized to develop chemical sensors.

3.3.5. Metal Phthalocyanines and Porphyrins The composite materials produced from ICP and phthalocyanines demonstrate high electrical conductivity consistent with the properties of the polymer. For example, polyaniline/ phthalocyanine (PANI/Pc) and polypyrrole/phthalocyanine (PPy/Pc) films have been prepared electrochemically on platinum electrodes from aqueous solutions containing pyrrole or aniline and a salt of the oligomeric nickel or cobalt phthalocyanine.198 Incorporation of large anions based on oligomeric metal/phthalocyanines and polypyrrole or polyaniline favors cation insertion during their normal redox process. In contrast to PPy/NiPPc or PPy/CoPc, PANI/NiPPc is depositing only in very thin films. The proton transport was found to be responsible for maintaining the electroneutrality inside the film during cycling.199 The conductivity for PANI/CoPC composites decreases from that of pure PANI, PANI/PVC, or PANI/PMMA.200 Few examples exist for the use of ICP/phthalocyanine composites in chemical sensors. Polyaniline/Cu-phthalocyanine (PANI/CuPc), polypyrrole/Cu-phthalocyanine (PPy/ CuPc), and polythiophene/Cu-phthalocyanine (PT/CuPc) have all been examined as gas sensors for vapors such as methanol, ammonia, and nitrogen dioxide. These composites have limited sensitivity to methanol or ammonia gas. However, all polymer composites showed enhanced sensitivity to gas vapors relative to the conductive polymer alone. Increases in analytical sensitivities with respect to nitrogen dioxide gas were on the order of 103 depending on the phthalocyanine concentration in the composite. The study neglects discussing or demonstrating the enhanced properties of the composite relative to phthalocyanine films without the polymer. PANI/phthalocyanines (PANI/MePc, Me ) Fe, Ni, or Cu), prepared by a layer-by-layer (LBL) method, have been used to detect dopamine by cyclic voltammetry.201 Composites produced using PANI/PcFe demonstrated superior sensing properties for dopamine in comparison to polyvinyl chloride/ phthalocyanine or PANI/NiPc or PANI/CuPc. The concentration range accessible for PANI/FePc composite was from 0.25 to 8 mM, and the detection limit was 1.0 × 10-4 M. Differences in the voltammetric response were observed as the metal center of the phthalocyanine composites was

Composites of Intrinsically Conducting Polymers

changed. However, all materials were able to distinguish the dopamine from ascorbic acid based on electrochemical oxidation of the species. The membranes are extremely stable, showing little chemical degradation of the composite after multiple sensing experiments. The use of metallophthalocyanine as a component of polymeric membrane ion-selective electrodes has the advantages that the active component is not leaching out and does not readily decompose. Modified carbon electrode with hybrid films of poly(o-aminophenol) and nickel sulfonated phthalocyanine produced by electrochemical polymerization of o-aminophenol in the presence of the metal complex has been used in electrochemical sensors for nitric oxide detection.202 The oxidation peak current is linear with concentrations up to 200 µM of NO. Overall, phthalocyanines show promise as chemical sensor materials because they are stable with reactive centers that allow the sensing properties to be manipulated, and they can easily be incorporated into ICPs. However, further studies are required to fully adapt these macrocyclic materials to composite materials for chemical sensing.

3.4. Composites with Biological Materials The emerging trends in biosensors based on ICPs, covering approximately the last 5 years, have been described in the literature.203 Various methods have been investigated for biosensing applications using ICP and biomaterials composites. In contrast to environmental-based sensing, the materials used to create the composite must be biocompatible to be used in vivo. These materials should be inert and provide high loading of the biologically specific component with minimal leaching. A recent review has been presented regarding the advantage of ICP membrane-based biosensors.204 Although biosensing continues to expand encompassing enzymatic systems, proteins (amino acids), DNA, antibodies, and antigens, the vast majority of systems incorporate enzymes and the biologically active material. These systems use enzymes that are affixed, immobilized, or encapsulated in a variety of matrix materials. The primary advantages of ICPs over other possible matrices include the enhanced speed, stability, and sensitivity of enzyme biosensors. Incorporation of single enzymes or groups of enzymes provides the biosensor selectivity to either single or multiple biological components. The ICP acts as the transducer, converting the chemical response into electric current. Synthesis of the material usually involves chemical or electrochemical polymerization in the presence of an enzyme. However, in some cases the polymers functional groups can be used to covalently attach the enzyme to the matrix. Common analytes include species such as glucose, cholesterol, urea, and uric acid due to their biological relevance with respect to the health of an individual. A few examples are provided to demonstrate how conductive polymer/enzyme composites have been utilized for biosensing applications. Use of ICPs in the design of bioanalytical sensors has primarily focused on polypyrrole (PPy) and polyaniline (PANI) because they are biologically compatible and do not dissolve in water. Immobilization of biologically active molecules within PPy during electrochemical polymerization has been examined.205 The polymers have also gained recognition in the field of mimicking natural sensing organs by combining electromechanical actuators with the ability to control biological processes in drug delivery systems.206

Chemical Reviews, 2008, Vol. 108, No. 2 763

Daily glucose monitoring is an important factor in the management of diabetes. The enzyme, glucose oxidase, has been successfully incorporated into chitosan-PANI membranes.207 Inclusion of chitosan provides chemical-binding sites to affix the enzyme, increases the stability, and prevents leaching of the enzyme. A near linear increase in the current associated with 5 mM glucose is observed for the layer-bylayer growth of chitosan/polyaniline/glucose oxidase sensor. The current associated with glucose oxidation is not saturated over the first 12 layers. The sensor produced using 12 layers shows linearity in the concentration range from 0.5 to 16 mM. In comparison to the chitosan/glucose oxidase sensors, inclusion of PANI increases the dynamic concentration range and decreases the time required to reach the steady-state response by a factor of 2. Amperometric biosensors have been prepared using conducive polymers such as PANI and PPy with the enzyme tyrosinase.208 Tyrosinase is used to monitor the oxidation of phenolic residues that can be found in air, soil, and water. The release of phenol compounds is typically associated with decomposition of plastics and plasticides, pesticides, and paper processing. Phenolic residues are easily absorbed in humans through ingestion, transdermal uptake, and inhalation. The enzymatic response of a PANI/tyrosinase composite membrane is demonstrated for phenol. Although the authors demonstrate that the composite can be used in the oxidative detection of phenolic species, they fail to fully evaluate the figures of merit for the sensor material. Specifically, the authors do not evaluate the dynamic concentration range, limit of detection, and analytical sensitivity. In contrast the PPy/tyrosinase composite was fully characterized with respect to phenol, catechol, and p-chlorophenol. The PPy/ tyrosinase sensor was sensitive to phenol, catechol, and p-chlorophenol over concentration ranges from 9.9 to 84.7, 6.7 to 72.6, and 3.9 to 48.8 mM, respectively. Sensitivities for phenol, catechol, and p-chlorophenol of 0.14, 0.21, and 0.36 A M-1 cm-2 were obtained. Detection limits for the three components were between 1 and 2 mM, much lower than previous examples of tyrosinase sensors in the literature. Finally, the long-term stability of the PPy/tyrosinase was found to be superior to conventional biosensors used for detection of phenolic residues. The link between high cholesterol and heart disease has provided the impetus required for development of cholesterol biosensors. Composites of conductive polymer with the enzymes cholesterol esterase and cholesterol oxidase have been utilized as biosensors.209 The studies provide a simple method for producing cholesterol sensors that are thermally and chemically stable with good analytical sensitivity. In both studies the enzyme is covalently linked to prevent enzymatic leaching in solution. In addition, the covalent linkages improve the stability of the sensor over a 5 week period. The optimum pH range for detection of cholesterol oleate was determined to be between 7.0 and 7.5 in these studies. Temperature also influences the sensor response with optimum conditions observed at 45 °C. These studies provided a dynamic concentration range of 1-500 mg/dL, a detection limit of 25 mg/dL, and analytical sensitivity of 0.042 mA (mg/dL)-1 for the conductive polymer/enzyme composite sensors. Incorporation of uricase into conducting polymers for evaluation of uric acid biological samples has been examined. In the first study, polyaniline was synthesized in the ionic liquid 1-ethyl-3-methylimidazolium-ethyl sulfate (EMIES).210

764 Chemical Reviews, 2008, Vol. 108, No. 2

The conductive polymer membranes prepared in the ionic solution show high electric conductivity up to pH values of 12. In addition, the polymer has been shown to have high flexibility and good mechanical properties. Acetaminophen, glutathione, L-cysteine, and ascorbic acid were combined with solutions containing uric acid (0.5 mmol dm-3) and compared to the response of uric acid alone. The biosensor composites produced from PANI/uricase prepared in ionic liquid are resistant to interference and impurities in comparison to PANI/uricase prepared in acid solutions. The sensor displays a linear dynamic concentration range of uric acid between 1.0 × 10-3 and 1.0 mmol dm-3. The authors do not provide a measure of the analytical sensitivity of the composite prepared in ionic liquid. However, the response is significantly enhanced relative to uric acid sensing using PANI prepared in acid. Although the authors state that uricase is immobilized in the polymer matrix, the enzyme actually acts as a dopant, with no covalent contact with polymer. Immobilization of uricase into PANI membranes has also been achieved, and the biosensing of uric acid has been evaluated using the composite.211 Covalent bonding of uricase is achieved in PANI using glutaraldehyde as a cross-linker. The covalent bond keeps the enzyme immobilized in the polymer matrix and enhances the stability of the sensor membrane, remaining 95% active after 18 weeks. The authors also demonstrate that the composite with immobilized uricase is superior to PANI/uricase composite produced through simple doping of the polymer. The response of the composite to uric acid displays two linear ranges with respect to concentration between 0.01 and 0.05 mM and 0.1 and 0.6 mM. The analytical sensitivity obtained from the calibration is 47.2 mA nM-1 at low concentrations and 10.66 mA nM-1 at higher concentrations. Optimum results are obtained at temperatures between 30 and 35 °C and pH values of 6.5 with limited activity up to pH 8. The influence of possible interference from other biologically relevant species was also examined using the PANI/uricase composite. The sensor showed little activity with respect to urea, glucose, ascorbic acid, cholesterol, and lactic acid. The authors also examined serum samples using the composite and compared the results to spectroscopic measurement of uric acid. The results are very similar, indicating the composite can measure uric acid in real biological samples.

4. Conclusions and Outlook This review of the trends in the synthesis and application of ICP composites in chemical sensing demonstrates the versatility of these materials. It highlights the concept that rational design of composites can be utilized to develop novel sensing materials with improved properties such as enhanced resistance to humidity, lower detection limits, increased sensitivity, lower sensing temperatures, and enhanced stability. The simple application of combinatorial chemistry techniques can be used to produce large volumes of sensor material. However, this approach does not represent the rational design of composite sensors. Rational design requires that a more fundamental understanding of the properties of the ICP and secondary component be evaluated independently and as a combined entity. Knowledge of the individual materials and the changes associated with the combined species provides the necessary guidance for enhancing favorable and reducing unfavorable attributes. The design of a sensor can only proceed with a fundamental understand-

Hatchett and Josowicz

ing of the chemical, physical, and mechanical properties of the composite materials. When these parameters are understood they can be utilized to develop sensors with designed reactivity, sensitivity, and selectivity. It is apparent that not all of the unique, tunable properties that are observed in composite materials have been utilized for chemical sensor applications. However, the enhanced properties of the composites, relative to each of the individual material, have been used extensively in other applications including batteries, fuel cells, and catalyst. In addition, the extensive knowledge of functional components including organized mixed monolayers on gold nanoparticles is just starting to be applied to the design of diverse sensors. Furthermore, a complete understanding of the properties of the individual components influence the sensing properties of the composite has not been extensively studied. For example, charge propagation in the composite, the limits of charge mobility through the composite, the true effect of water vapor, oxygen, or light on the interaction of analyte with the composite, and how the applied processing method of the composite influences the properties are all factors that must be addressed. The rational development of composite materials that provide enhanced sensing capabilities should use standardized methods for evaluation of ICP and ICP composites, respectively. This approach would allow comparison of sensing results achieved with different layers and also could contribute to the rapid progression of the design and implementation of novel materials in sensing applications. Further developments in metal nanomaterials and their electronic properties, chemical functionalization, and catalytic properties will continue to impact the sensor field. Development of methodologies to control the shape, size, and range of distribution of CNTs and metal clusters within ICP matrices is required. In addition, characterization of the chemical and physical properties of ICP composite is required to fully understand the influence of secondary components on the sensing properties. It is impossible to control the properties of the secondary component if the properties of the ICP matrix are not fully understood. Identification of new properties of nanomaterials requires new methods for the preparation or assembly of usable composite structures. The work of Penner and co-workers is an example that demonstrates how new methods can be utilized in the preparation of naked (palladium) and bimetallic beaded nanowires ranging in diameter from tens of nanometers to 1 µm. These materials are used in the detection of hydrogen gas.212 Further work is required to develop applications for structured nanowire materials. In contrast to CNTs and in spite of the enormous progress in synthesizing and structural characterization of metal oxide nanowires such as SnO2, ZnO, and TiO2, few of these structured materials have been utilized in gas sensors for detection of common gases such as CO,213 O2,214 or water vapor.215 The advantages of metal oxides have been shown with respect to lowering the required operation temperature of the analytical sensor. Implementation of new building blocks that can be used to assemble 1-D nanowires216 or nanoparticles to produce 2-D nanofilms217 have been reviewed but not yet implemented in the design of novel ICP composites. There has been progress in making supercrystals that are produced from rare-earth oxides M2O3 (M ) Pr, Nd, Sm, Eu, Gd, Tb, Dy) that are uniformly and spontaneously assembled to form nanorods, nanowires, and nanoplates.218 Ultranarrow semi-

Composites of Intrinsically Conducting Polymers

conductors (ZnS, ZnSe, CdS, CdSe) with tunable subnanometer increments have been produced using microwave irradiation.219 Nanorods with a controllable interparticle spacing have been achieved using capping agents of varying alkyl chain lengths.220 In addition, the recently developed theoretical description of the evolution of an ensemble of nanoparticles in colloidal solution used to control the nanocrystal monodispersity of semiconductors and magnetic alloy nanocrystals as building blocks of 2-D and 3-D arrays and colloidal supercrystals should find application in developing advanced sensing layers.221 These studies highlight the advancement of synthetic methods to produce new, unique materials that can be utilized to produce new composite materials. New electronic, optical, and mechanical properties that are interesting to solid-state electrochemical and optical sensors may be obtained from inorganic fullerene-like WS2 and MoS2 polyhedra and nanotubes which are capable of modulating the energy gap with shrinking diameter. In this respect they are prevalent as compared to many of the bulk 2D materials because they exhibit a direct gap rather than an indirect gap transition.222 Their thermodynamic stability and the ability to fine tune the Fermi level by doping and intercalation of metal atoms makes them novel materials for sensor applications because their catalytic properties can be controlled.223 In summary, the next generation of ICP materials for chemical sensors should be able to balance their processability with their sensing performance with regard to faster response times, sensitivity, selectivity, lower detection limits, and larger dynamic concentration ranges. It should be feasible to control the molecular structure and morphology when the material is processed in order to address the issue of analyte sorption and diffusion. Additionally, controlling or tailoring the properties of the interface can be used to minimize charge trapping and contact resistance within the sensor device structure. Any sensor should incorporate an understanding of the properties of each component independently while evaluating the new properties of the composite. Rational sensor design that utilizes the combined chemical, physical, and mechanical properties of the composite materials is crucial in the development of selective, sensitive sensing mechanisms.

5. Acknowledgments This work was supported by the NSF Grant CHE 0452045 (M.J.) and DOE Grant DE-FG36-05GO85028 (D.W.H.).

6. References (1) Wang, H.-L.; Li, W.; Jia, Q. X.; Akhadov, E. Chem. Mater. 2007, 19, 520. (2) Heeger Alan, J. Angew. Chem., Int. Ed. 2001, 40 (14), 2591. (3) Persaud, P.; Krishna, C. Mater. Today 2005, 8 (4), 44. (4) Pron, A.; Rannou, P. Prog. Polym. Sci. 2001, 27 (1), 135. JagurGrodzinski J. Polym. AdV. Technol. 2002, 13 (9), 615. Malinauskas, A. Polymer 2001, 42 (9), 3972. (5) Handbook of ConductiVe Molecules and Polymers; Nalva, H. S., Ed.; John Wiley & Sons Ltd.: Chichester, 1997; Vols. 1-4. (6) Potje-Kamloth, K; Polk, B. J.; Josowicz, M.; Janata, J. Chem. Mater. 2002, 14 (6), 2782. (7) Polk, B. J.; Potje-Kamloth, K.; Josowicz, M.; Janata, J. J. Phys. Chem. B 2002, 106 (44), 11457. (8) Kaiser A. B. Rep. Prog. Phys. 2001, 64 (1), 1. (9) Guernion, N. J. L.; Hayes, W. Curr. Org. Chem. 2004, 8 (7), 637. (10) Sasaki, I.; Janata, J.; Josowicz, M. Polym. Degrad. Stab. 2007, 92 (7), 1408. (11) McQuade, D. T.; Pullen, A. E.; Swager, T. M. Chem. ReV. 2000, 100, 2537.

Chemical Reviews, 2008, Vol. 108, No. 2 765 (12) Hua, B.; Shi, G. Sens.-Basel 2007, 7 (3), 267. Nicolas-Debarnot, D.; Poncin-Epaillard, F. Anal. Chim. Acta 2003, 475, 1. Potje-Kamloth, K. Crit. ReV. Anal. Chem. 2002, 32 (2), 121. Adhikari, B.; Majumdar, S. Prog. Polym. Sci. 2004, 29 (7), 699. (13) Josowicz, M. Chem. Anal. 2001, 46 (2), 161. Bailey, R. A.; Persaud, K. C. Polym. Sens. Actuators 2000, 149. (14) Janata, J.; Josowicz, M. Nat. Mater. 2003, 2 (1), 19. Petr, A.; Zhang, F.; Peisert, H.; Knupfer, M.; Dunsch, L. Chem. Phys. Lett. 2004, 385 (1, 2), 140. (15) Bissell, R. A.; Persuade, K. C.; Travers, P. Phys. Chem. Chem. Phys. 2002, 4, 3482. (16) Maksymiuk, K. Electroanalysis 2006, 18, 1537. (17) Bret, C. M. A.; Thiemann C. J. Electroanal. Chem. 2002, 58, 215. Bobacka, J. Electroanalysis 2006, 18 (1), 7. Michalska, A. Pol. Anal. Bioanal. Chem. 2006, 384 (2), 391. (18) Trojanowicz, M. Microchim. Acta 2003, 143 (2-3), 75. Bobacka, J.; Ivaska, A.; Lewenstam, A. Electroanalysis 2003, 15 (5-6), 366. Maksymiuk, K. Electroanalysis 2006, 18, 1537. (19) Lin, J. Trends Anal. Chem. 2000, 19 (9), 541. Gazotti, W. A.; Nogueira, A. F.; Girotto, E. M.; Micaroni, L.; Martini, M.; das Neves, S.; De Paoli, M.-A. Handbook AdV. Electron. Photon. Mater. DeVices 2001, 10, 53. (20) Sinha, N; Ma, J.; Yeow, J. T. W. J. Nanosci. Nanotechnol. 2006, 6 (3), 573. Snow, E. S.; Perkins, F. K.; Robinson, J. A. Chem. Soc. ReV. 2006, 35, 790. Stetter, J. R.; Maclay, G. J. AdV. Micro. Nanosys. 2004, 1, 357. Merkoci, A.; Pumera, M.; Llopis, X.; Perez, B.; del Valle, M.; Alegret, S. Trends Anal. Chem. 2005, 24 (9), 826. Moulton, S. E.; Minett, A. I. Sens. Lett. 2005, 3 (3), 183, Sherigara, B. S.; Kutner, W.; D’Souza, F. Electroanalysis 2003, 15 (9), 753. (21) Scho¨onenberger, C. Semicond. Sci. Technol. 2006, 21, 51. Dresselhaus, M.; Dresselhaus, G.; Avouris, P. H. Carbon Nanotubes: Synthesis, Structure, Properties and Applications; Springer-Verlag: Berlin, 2001. Tanaka, K.; Okada, M.; Huang, Y. The Science and Technology of Carbon Nanotubes; Tanaka, K., Yamabe, T., Fukui, K., Eds.; Elsevier: Oxford, 1999. (22) McEuen, P. L. Phys. World 2000, 31. Fuhrer, M. S.; Nygard, J.; Shih, L.; Forero, M.; Yoon, Y. G.; Mazzoni, M. S. C.; Choi, H. J.; Ihm, J.; Louie, S. G.; Zettl, A.; McEuen, P. L. Science 2000, 288, 494. (23) Snow, E. S.; Perkins, F. K.; Robinson, J. A. Chem. Soc. ReV. 2006, 35 (9), 790. (24) Rice, P.; Wallis, T. M.; Russek, S. E.; Kabos, P. Nano Lett. 2007, 4, 1086. (25) Scho¨nenberger, C.; Bachtold, A.; Strunk, C.; Salvetat, J.-P.; Forro, L. Appl. Phys. A 1999, 69, 283. Bachtold, A.; Fuhrer, M. S.; Plyasunov, S.; Forero, M.; Anderson, E. H.; Zettl, A.; McEuen, P. L. Phys. ReV. Lett. 2000, 84, 6082. (26) Dai, L.; Soundarrajan, P.; Kim, T. Pure Appl. Chem. 2002, 74 (9), 1753. Migone, A. D.; Talapatra, S. Encyclopedia Nanosci. Nanotechnol. 2004, 3, 749. Li, J.; Ng, H. T. Encyclopedia of Nanosci. Nanotechnol. 2004, 1, 591. Parikh, K.; Cattanach, K.; Rao, R.; Suh, D. S.; Wu, A. M.; Manohar, S. K. Sens. Actuators B 2006, 113 (1), 55. Li, J.; Lu, Y.; Ye, Q.; Cinke, M.; Han, J.; Meyyappan, M. Nano Lett. 2003, 3 (7), 929. (27) Luong, J. H. T.; Hrapovic, S.; Wang, D.; Benseba, F.; Simard, B. Electroanalysis 2004, 16 (1-2), 132. (28) Ye, J. S.; Sheu, F. S. Curr. Nanosci. 2006, 2 (4), 319. Trojanowicz, M. Trends Anal. Chem. 2006, 25 (5), 480. Gong, K.; Yan, Y.; Zhang, M.; Su, L.; Xiong, S.; Mao, L. Anal. Sci. 2005, 21 (12), 1383. Merkoci, A. Microchim. Acta 2006, 152 (3-4), 157. He, P.; Dai, L. BioMEMS Biomed. Nanotechnol. 2006, 1, 171. Portney, N. G.; Ozkan, M. Anal. Bioanal. Chem. 2006, 384 (3), 620. Burghard, M. Small 2005, 1 (12), 1148, Lin, Y. H.; Wang, J. Electroanalysis 2005, 17 (1), 7. Yantasee, W.; Wang, J. Front. Biosci. 2005, 10, 492. Fernando, K. A. S.; Lin, Y.; Sun, Y. P. Langmuir 2004, 20, 4777. Bahr, J. L.; Tour, J. M. J. Mater. Chem. 2002, 12, 1952. Sun, Y.-P.; Fu, K.; Lin, Y.; Huang, W. Acc. Chem. Res. 2002, 35, 1096. Chen, J.; Rao, A. M.; Lyuksyutov, S.; Itkis, M. E.; Hamon, M. A; Hu, H. R.; Cohn, W.; Eklund, P. C.; Colbert, D. T.; Smalley, R. E.; Haddon, R. C. J. Phys. Chem. B 2001, 105, 2525. (29) Banerjee, S.; Hemraj-Benny, T.; Wong, S. S. AdV. Mater. 2005, 17 (1), 17. Velasco-Santos, C.; Martinez-Hernandez, A. L.; Castano, V. M. Composite Interfaces 2005, 11 (8-9), 567. Lin, Y.; Hill, D. E.; Bentley, J.; Allard, L. F.; Sun, Y. P. J. Phys. Chem. B 2003, 107 (38), 10453. (30) Ballesteros, B.; de la Torre, G; Ehli, C.; Rahman, G. M. A.; AgulloRueda, F.; Guldi, D. M.; Torres, T. J. Am. Chem. Soc. 2007, 129 (16), 5061. (31) Teker, K.; Sirdeshmukh, Ra.; Sivakumar, K.; Lu, S.; Wickstrom, E.; Wang, H.-N.; Vo-Dinh, T.; Panchapakesan, B. NanoBiotechnol. 2005, 1 (2), 171. (32) Wang, T.; Hu, X. G.; Qu, X. H.; Dong, S. J. J. Phys. Chem. B 2006, 110 (13), 6631.

766 Chemical Reviews, 2008, Vol. 108, No. 2 (33) Tasis, D.; Tagmatarchis, N.; Orato, M. Chem. ReV. 2006, 106, 1105, Burghard, M. Surf. Sci. Rep. 2005, 58 (1-4), 1. (34) Li, T. X.; Ji, Y. L.; Yu, S. W.; Wang, G. H. Solid State Commun. 2000, 116, 547. (35) Landman, U. Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 6671. (36) Gittins, D. I.; Bethell, D.; Schiffrin, D. J.; Nichols, R. J. Nature 2000, 408, 67. Lahtinen, R. M.; Mertens, S. F. L.; East, E.; Kiely, C. J.; Schiffrin, D. J. Langmuir 2004, 20, 3289. Kariuki, N. N.; Luo, J.; Maye, M. M.; Hassan, A.; Menard, T.; Naslund, H. R.; Lin, Y.; Wang, C.; Engelhard, M. H.; Zhong, C. J. Langmuir 2004, 20, 11240. Mallin, M. P.; Murphy, C. J. Nano Lett. 2002, 2, 1235. RodriguezGonzalez, B.; Sanchez-Iglesias, A.; Giersig, M.; Liz-Marzan, L. M. Discuss. Faraday Soc. 2004, 125, 133. Zamborini, F. P.; Hicks, J. F.; Murray, R. W. J. Am. Chem. Soc. 2000, 122, 4514. Templeton, A. C.; Zamborini, F. P.; Wuelfing, W. P.; Murray, R. W. Langmuir 2000, 16, 6682. Wuelfing, W. P.; Zamborini, F. P.; Templeton, A. C.; Wen, X. G.; Yoon, H.; Murray, R. W. Chem. Mater. 2001, 13, 87. Taton, T. A.; Mucic, R. C.; Mirkin, C. A.; Letsinger, R. L. J. Am. Chem. Soc. 2000, 122, 6305. Storhoff, J. J.; Lazarides, A. A.; Mucic, R. C.; Mirkin, C. A.; Letsinger, R. L.; Schatz, G. C. J. Am. Chem. Soc. 2000, 122, 4640. Mbindyo, J. K. N.; Reiss, B. D.; Martin, B. R.; Keating, C. D.; Natan, M. J.; Mallouk, T. E. AdV. Mater. 2001, 13, 249. Kariuki, N. N.; Han, L.; Ly, N. K.; Patterson, M. J.; Maye, M. M.; Liu, G. J.; Zhong, C. J. Langmuir 2002, 18, 8255. Boal, A. K.; Rotello, V. M. Langmuir 2000, 16, 9527. Chen, S. W.; Pei, R. J. J. Am. Chem. Soc. 2001, 123, 10607. Kim, B.; Tripp, S. L.; Wei, A. J. Am. Chem. Soc. 2001, 123, 7955. Shon, Y.-S.; Choo, H. Chem. Commun. 2002, 2560. Guldi, D. M.; Zilbermann, I.; Anderson, G.; Kotov, N. A.; Tagmatarchis, N.; Prato, M. J. Am. Chem. Soc. 2004, 126, 14340. Boal, A. K.; Ilhan, F.; DeRouchey, J. E.; Thurn-Albrecht, T.; Russell, T. P.; Rotello, V. M. Nature 2000, 404, 746. Srivastava, S.; Frankamp, B. L.; Rotello, V. M. Chem. Mater. 2005, 17, 487. Zeng, W. X.; Maye, M. M.; Leibowitz, F. L.; Zhong, C. J. Anal. Chem. 2000, 72, 2190. Han, L.; Maye, M. M.; Leibowitz, F. L.; Ly, N. K.; Zhong, C. J. J. Mater. Chem. 2001, 11, 1258. Han, L.; Daniel, D. R.; Maye, M. M.; Zhong, C. J. Anal. Chem. 2001, 73, 4441. Lim, I.-I. S.; Maye, M. M.; Luo, J.; Zhong, C. J. J. Phys. Chem. B 2005, 109, 2578. (37) Guczi, L.; Beck, A.; Horvath A.; Horvath, D. Top. Catal. 2002, 19, 157. (38) Kondow, T.; Mafune, F. Progress in Experimental and Theoretical Studies of Clusters (AdVanced Series in Physical Chemistry); World Scientific Publishing Co.: London, 2003. Metal Clusters in Chemistry; Braunstein, P., Oro, L. A., Raithby, P. R., Eds.; Wiley-VCH: Weinheim, 2000. (39) Meiwes-Broer, K.-H. Metal Clusters at Surfaces: Structure, Quantum Properties, Physical Chemistry; Springer: New York, 2000. Kawazoe, Y.; Kondow, T.; Ohno, K. Clusters & Nanomaterials; Springer: New York, 2001. (40) Klimov, V. I. Semiconductor and Metal Nanocrystals: Synthesis and Electronic and Optical Properties (Optical Engineering); CRS: Enfield, 2003. (41) Heilmann, A. Polymer Films with Embedded Metal Nanoparticles; Springer: New York, 2002. (42) Capek, I. Nanocomposite structures and dispersions; Elsevier Science: New York, 2006. (43) Kiely, C. J.; Fink, J.; Brust, M.; Bethell, D.; Schiffrin, D. J. Nature 1998, 396, 444. (44) Heiz, U.; Bullock, E. L. J. Mater. Chem. 2004, 14, 564. (45) Landmann, U.; Heiz, U. Nanocatalysis; Springer: New York, 2007. (46) Daniel, M.-C.; Astruc, D. Chem. ReV. 2004, 104 (1), 293. (47) Pasquato, L.; Pengo, P.; Scrimin, P. J. Mater. Chem. 2004, 14 (24), 3481. (48) Song, Y.; Murray, R. W. J. Am. Chem. Soc. 2002, 124, 7096. (49) Lim, I.-I. S.; Maye, M. M.; Luo, J.; Zhong, C. J. J. Phys. Chem. B 2005, 109, 2578. (50) Stroyuk, A. L.; Shvalagin, V. V.; Kuchmii, S. Y. J. Photochem. Photobiol. A: Chem. 2005, 173 185. (51) Sun, S. H.; Murray, C. B.; Weller, D.; Folks, L.; Moser, A. Science 2000, 287, 1989. Sun, S. H.; Fullerton, E. E.; Weller, D.; Murray, C. B. IEEE Trans. Magn. 2001, 37, 1239. (52) (a) Lu, G. Q.; Wieckowski, A. Curr. Opin. Colloid Interface Sci. 2000, 5, 95. Roucoux, A.; Schulz, J.; Patin, H. Chem. ReV. 2002, 102, 3757. Long, J. W.; Stroud, R. M.; Swider-Lyons, K. E.; Rolison, D. R. J. Phys. Chem. B 2000, 104, 9772. Schadt, M.; Lin, Y.; Naslund, H. R.; Zhong, C. J. Catal. Today 2005, 99, 291. Maye, M. M.; Kariuki, N. N.; Luo, J.; Han, L.; Njoki, P.; Wang, L.; Lin, Y.; Naslund, H. R.; Zhong, C. J. Gold Bull. 2004, 37, 217. Zhong, C. J.; Luo, J.; Maye, M. M.; Han, L.; Kariuki, N. N. Nanostructured Gold and Alloy Electrocatalysts. In Nanotechnology in Catalysis; Zhou, B., Hermans, S., Somorjai, G. A., Eds.; Kluwer Academic/Plenum Publishers: New York, 2004; Vol. 1, Chapter 11, pp 222.

Hatchett and Josowicz (53) Kariuki, N. N.; Luo, J.; Maye, M. M.; Hassan, A.; Menard, T.; Naslund, H. R.; Lin, Y.; Wang, C.; Engelhard, M. H.; Zhong, C. J. Langmuir 2004, 20, 11240. Mallin, M. P.; Murphy, C. J. Nano Lett. 2002, 2, 1235. (54) Sun, S. H.; Fullerton, E. E.; Weller, D.; Murray, C. B. IEEE Trans. Magn. 2001, 37, 1239. (55) Luo, J.; Maye, M. M.; Kariuki, N. N.; Wang, L.; Njoki, P.; Han, L.; Schadt, M.; Lin, Y.; Naslund, H. R.; Zhong, C. J. Catal. Today 2005, 99, 291. (56) Maye, M. M.; Kariuki, N. N.; Luo, J.; Han, L.; Njoki, P.; Wang, L.; Lin, Y.; Naslund, H. R.; Zhong, C. J. Gold Bull. 2004, 37, 217. (57) Kariuki, N. N.; Luo, J.; Hassan, S. A.; Lim, I.-I. S.; Wang, L.; Zhong, C. J. Chem. Mater. 2006, 18, 123. (58) Dahl, J. A.; Maddux, B. L. S.; Hutchison, J. E. Chem. ReV. 2007, 107 (6), 2228. Khan, M. I. ChemInform 2003, 34, 255. (59) Zhang, D.-F.; Sun, L.-D.; Xu, G.; Yan, C.-H. Phys. Chem. Chem. Phys. 2006, 8 (42), 4874. Jitianu, A.; Altindag, Y.; Zaharescu, M.; Wark, M. J. Sol-Gel Sci. Technol. 2003, 26, 483. (60) Pinna, N.; Neri, G.; Antonietti, M.; Niederberger, M. Angew. Chem., Int. Ed. 2004, 43, 4345. Niederberger, M.; Garnweitner, G. Chem. Eur. J. 2006, 12 (28), 7282. Niederberger, M.; Garnweitner, G.; Buha, J.; Polleux, J.; Ba, J.; Pinna, N. J. Sol-Gel Sci. Technol. 2006, 40 (2-3), 259. (61) Neri, G.; Bonavita, A.; Rizzo, G.; Galvagno, S.; Pinna, N.; Niederberger, M.; Capone, S.; Siciliano, P. Sens. Actuators B 2007, 122, 564. (62) Fang, X.; Zhang, L. J. Mater. Sci. Technol. 2006, 22 (1), 1. (63) Jitianu, A.; Altindag, Y.; Zaharescu, M.; Wark, M. J. Sol-Gel Sci. Technol. 2003, 26, 483. (64) Grimes, C. A. J. Mater. Chem. 2007, 17, 1451. (65) Zhuang, H.-F.; Lin, C.-J.; Lai, Y.-K.; Sun, L.; Li, J. EnViron. Sci. Technol. 2007, 41, 4735. (66) Chai, S. Y.; Kim, Y. S.; Lee, W. I. J. Electroceram. 2006, 17, 323. Liu, X. M.; Wu, S. L.; Chu, P. K.; Zheng, J.; Li, S. L. Mater. Sci. Eng. A 2006, 426, 274. (67) Armentrout, P. B. Ann. ReV. Phys. Chem. 2001, 52, 423. Zemski, K. A.; Justes, D. R.; Castleman, A. W., Jr. J. Phys. Chem. B 2002, 106 (24), 6136. (68) Reddy, B. V.; Rasouli, F.; Hajaligol, M. R.; Khanna, S. N. Chem. Phys. Lett. 2004, 384 (4-6), 242. (69) Reddy, B. V.; Rasouli, F.; Hajaligol, M. R.; Khanna, S. N. Fuel 2004, 83 (11-12), 1537. Khan, M. I. J. Solid State Chem. 2000, 152 (1), 105. (70) Bernstein, E. R.; Matsuda, Y. Toward a Molecular Understanding of EnVironmental Catalysis: Studies of Metal Oxide Clusters and their Reactions, EnVironmental Catalysis; Grassian, V. H., Ed.; CRC Press: Boca Raton, 2005; pp 421-469. (71) Reddy, B. V.; Khanna, S. N. Phys. ReV. Lett. 2004, 93 (6), 068301/ 1-068301/4. Reilly, N. M.; Reveles, J. U.; Johnson, G. E.; Khanna, S. N.; Castleman, A. W., Jr. J. Phys. Chem. A 2007, 111 (20), 4158. (72) Reilly, N. M.; Reveles, J. U.; Johnson, G. E.; Khanna, S. N.; Castleman, A. W. Chem. Phys. Lett. 2007, 435 (4-6), 295. (73) Huang, X.; Zhai, H.-J.; Kiran, B; Wang, L.-S. Angew. Chem., Int. Ed. 2005, 44 (44), 7251. (74) Li, S.; Dixon, D. A. J. Phys. Chem. A 2006, 110 (19), 6231. Sun, Q.; Sakurai, M.; Wang, Q.; Yu, J. Z.; Wang, G. H.; Sumiyama, K.; Kawazoe, Y. Phys. ReV. B: Condens. Matter Mater. Phys. 2000, 62 (12), 8500. (75) Rainer, D. R.; Goodman, D. W. J. Mol. Catal. A 1998, 131 (1-3), 259. Vardhan, D.; Liyanage, R.; Armentrout, P. B. J. Chem. Phys. 2003, 119 (8), 4166. Griffin, J. B.; Armentrout, P. B. J. Chem. Phys. 1997, 107 (14), 5345. (76) Moulton, B.; Zaworotko, M. J. Chem. ReV. 2001, 101 (6), 1629. (77) Robertson, A.; Shinkai, S. Coord. Chem. ReV. 2000, 205 157. (78) Schubert, U. Macromol. Containing Met. Met.-Like Elem. 2006, 7, 55. Schubert, U. J. Sol-Gel Sci. Technol. 2004, 31 (1-3), 24. (79) Malinski, T. In Porphyrin Handbook; Kadish, K. M., Smith, K. M., Guilard, R., Eds.; Academic Press: New York, 2000; Vol. 6, Chapter 44. (80) Kadish, K.; Smith, K. M.; Guilar, R. The Porphyrin Handbook; Academic Press: New York, 2003; Vols. 15-20. (81) Walt, D. R. Chem. ReV. 2000, 100, 2595. Di Natale, C.; Paolesse, R.; Macagnano, A.; Mantini, A.; Goletti, C.; D’Amico, A. Sens. Actuators B 1998, 52, 162. (82) Kerp, H. R.; Westerdui, K. T.; van Veen, A. T.; van Faassen, E. E. J. Mater. Res. 2001, 16, 503. Ho, K. C.; Tsou, Y. H. Sens. Actuators B 2001, 77, 253. (83) Snow, A. W.; Barger, W. R. In Phthalocyanines: Properties and Applications; Lever, A. B. P., Ed.; John Wiley and Sons: New York, 1989; Vol. 1, p 341. (84) Lee, Y.-L.; Sheu, C.-Y.; Hsiao, R.-H. Sens. Actuators B 2004, 99, 281.

Composites of Intrinsically Conducting Polymers (85) Nyokong, T.; Vilakazi, S. Talanta 2003, 61, 27. Wen, Z.-H.; Kang, T.-F. Talanta 2004, 62, 351. (86) Praneeth, V. K. K.; Haupt, E.; Lehnert, N. J. Inorg. Biochem. 2005, 99 (4), 940. Paulat, F.; Lehnert, N. Inorg. Chem. 2007, 46 (5), 1547. (87) Gooding, J. J. Nanotechnology in Biology and Medicine; Tuan, V.D., Ed.; CRC Press LLC: Boca Raton, 2007; Vol. 21, pp 1-21. (88) Vastarella, W.; Livia, D. S.; Masci, A.; Maly, J.; De Leo, M.; Moretto, L. M.; Pilloton, R. Int. J. EnViron. Anal. Chem. 2007, 87 (10-11), 701. (89) Deore, B.; Nagaoka, T. Curr. Top. Anal. Chem. 2001, 2, 135. Shiigi, H.; Kijima, D.; Ikenaga, Y.; Hori, K.; Fukazawa, S.; Nagaoka, T. J. Electrochem. Soc. 2005, 152 (8), H129-H134. Kriz, D.; Ansell, R. J. Technol. Instrum. Anal. Chem. 2001, 23, 417. (90) Wong, E. L. S. Biophys. ReV. Lett. 2007, 2 (2), 167. Odenthal, K. J.; Gooding, J. J. Analyst 2007, 132, 603. (91) Pal, P. S.; Sarkar, P. J. Indian Chem. Soc. 2002, 79 (3), 211. Goddard, J. M.; Hotchkiss, J. H. Prog. Polym. Sci. 2007, 32, 698. (92) Heller, A. Acc. Chem. Res. 1990, 23 (5), 128. Lin, C.-L.; Shih, C.L.; Chau, L.-K. Anal. Chem. 2007, 79 (10), 3757. Barsan, M. M.; Klincar, J.; Batic, M.; Brett, C. M. A. Talanta 2007, 71, 1893. Olea, D.; Moreau, P.; Faure, C. J. Electroanal. Chem. 2007, 605, 125. (93) Nagl, S.; Wolfbeis, O. S. Analyst 2007, 132, 507. Ramanathan, M.; Simonian, A. L. Biosens. Bioelectron. 2007, 22, 3001. Bosch, M. E.; Sanchez, A. J. R.; Rojas, F. S.; Ojeda, C. B. Sensors 2007, 7, 797. Passaro, V. M. N.; Dell’Olio, F.; Casamassima, B.; De Leonardis, F. Sensors 2007, 7, 508. (94) Yuqing, M.; Jianrong, C.; Xiaohua, W. Trends Biotechnol. 2004, 22, 227. Kueng, A.; Kranz, C.; Mizaikoff, B. Biosens. Bioelectron. 2004, 19, 1301. Zhang, Y.; Lim, C. T.; Ramakrishna, S.; Huang, Z.-M. J. Mater. Sci.: Mater. Med. 2005, 16, 933. (95) Han, J. H.; Taylor, J. D.; Kim, D. S.; Kim, Y. S.; Kim, Y. T.; Cha, G. S.; Nam, H. Sens. Actuators B 2007, 123, 384. (96) Kim, M. A.; Lee, W.-Y. Anal. Chim. Acta 2003, 479, 143. Wang, G.; Xu, J.-J.; Chen, H.-Y.; Lu, Z.-H. Biosens. Bioelectron. 2003, 18, 335. Shin, J. H.; Marxer, S. M.; Schoenfisch, M. H. Anal. Chem. 2004, 76, 4543. Sotiropoulou, S.; Chaniotakis, N. A. Biomaterials 2005, 26, 6771. (97) Yu, T.; Zhang, Y.; You, C.; Zhuang, J.; Wang, B.; Liu, B.; Kang, Y.; Tang, Y. Chem. Eur. J. 2006, 12, 1137. Shan, W.; Yu, Tao, Wang, B.; Hu, J.; Zhang, X.; Tang, Y. Chem. Mater. 2006, 18, 3169. Dai, Z.; Fang, M.; Bao, J.; Wang, H.; Lu, T. Anal. Chim. Acta 2007, 591, 195. (98) Calvo, E. J.; Wolosiuk, A. Chem. Phys. Chem. 2004, 5, 235. Boozer, C.; Ladd, J.; Chen, S.; Yu, Q.; Homola, J.; Jiang, S. Anal. Chem. 2004, 76, 6967. Campuzano, S.; Pedrero, M.; Montemayor, C.; Fatas, E.; Pingarron, J. M. Electrochem. Commun. 2006, 8, 299. (99) De Paoli, M. A. In Handbook of Organic ConductiVe Molecules and Polymers; Nalwa, H. S., Ed.; Wiley: New York, 1997; Vol. 2. (100) Omastova, M.; Pionteck, J.; Kosina, S. Electronic and Optical Properties of Conjugated Molecular Systems in Condensed Phases; Hotta, S., Ed.; Research Signpost: Kerala, India, 2003; pp 153186. (101) Barra, G. M. O.; Leyva, M. E.; Soares, B. G.; Synth. Met. 2002, 130, 239. (102) Lavine, B. K.; Kaval, N.; Westover, D. J.; Oxenford, L. Anal. Lett. 2006, 39 (9), 1773. (103) Mirmohseni, A.; Wallace, G. G. Polymer 2003, 44 (12), 3523. (104) Aguilar-Hernandez, J.; Skarda, J.; Potje-Kamloth, K. Synth Met. 1998, 95 (3), 197. Bhat, N. V.; Gadre, A. P.; Bambole, V. A. J. Appl. Polym. Sci. 2001, 80 (13), 2511. (105) Segal, E.; Tchoudakov, R.; Mironi-Harpaz, I.; Narkis, M.; Siegmann, A. Polym. Int. 2005, 54 (7), 1065. (106) McGovern, S. T.; Spinks, G. M.; Wallace, G. G. Sens. Actuators B 2005, B107 (2), 657. (107) Jain, S.; Chakane, S.; Samui, A. B.; Krishnamurthy, V. N.; Bhoraskar, S. V. Sens. Actuators, B 2003, 96 (1-2), 124. (108) Matsuguchi, M.; Okamoto, A.; Sakai, Y. Sens. Actuators B 2003, 94, 46. (109) Guernion, N.; Ewen R. J.; Pihlainen, K.; Ratcliffe N. M.; Teare, G. C. Synth. Met. 2002, 126 (2-3), 301. (110) Ruangchuay, L.; Sirivat, A.; Schwank, J. React. Funct. Polym. 2004, 61 (1), 11. Ruangchuay, L.; Sirivat, A.; Schwank, J. Talanta 2003, 60 (1), 25. (111) Hong, K. H.; Oh, K. W.; Kang, T. J. J. Appl. Polym. Sci. 2004, 92, 37. (112) Watcharaphalakorn, S.; Ruangchuay, L.; Chotpattananont, D.; Sirivat, A.; Schwank, J. Polym. Int. 2005, 4 (8), 1126. (113) Ruangchuay, L.; Sirivat, A.; Schwank, J. React. Funct. Polym. 2004, 61 (1), 11. Ruangchuay, L.; Sirivat, A.; Schwank, J. Talanta 2003, 60 (1), 25-30. (114) Grulke, E. A. Solubility Parameter Values. In Polymer Handbook, 3rd ed.; Brandrup, J., Immergut, E. H., Eds.; Wiley: New York, 1989; pp VII/519.

Chemical Reviews, 2008, Vol. 108, No. 2 767 (115) Cooper, H.; Segal, E.; Srebnik, S.; Tchoudakov, R.; Narkis, M.; Siegmann, A. Polym. AdV. Technol. 2004, 15 (10), 573. Huang, Y.-J.; Hsieh, T.-Y.; Wang, Y.-Z.; Chuang, Ch.-N.; Chen, Ya.-P.; Huang, P.-T.; Ho, K.-S. J. Polym. Sci. 2007, 104 (2) 773. (116) Dai, L; Soundarrajan, P.; Kim, T. Pure Appl. Chem. 2002 74 (9), 1753, Lukaszewicz, J. P. Sens. Lett. 2006, 4, 53. (117) Zimer, A. M.; Bertholdo, R.; Grassi, M. T.; Zarbin, A. J. G.; Mascaro, L. H. Electrochem. Commun. 2003, 5, 983. (118) Baibarac, M.; Baltog, I.; Lefrant, S.; Mevellec, J. V.; Chauvet, O. Chem. Mater. 2003, 15, 4149. Baibarac, M.; Gomez-Romero, P. J. Nanosci. Nanotechnol. 2006, 6 (2), 289. (119) Ramamurthy, P. C.; Harrell, W. R.; Gregory, R. V.; Rao, A. M. J. Electrochem. Soc. 2007, 154 (6), H495. (120) Kumar, J.; Singh, R. K.; Kumar, V.; Rastogi, R. C.; Singh, R. Diamond Relat. Mater. 2007, 16 (3), 446. (121) Viswanathan, G.; Chakrapani, N.; Yang, H.; Wei, B.; Chung, H.; Cho, K.; Ryu, C. Y.; Ajayan, P. M. J. Am. Chem. Soc. 2003, 125, 9258. (122) Philip, B.; Xie, J. N.; Chandrasekhar, A.; Abraham, J.; Varadan, V. K. Smart Mater. Struct. 2004, 13 (2), 295. (123) Yan, X. B.; Han, Z. J.; Yang, Y.; Tay, B. K. J. Phys. Chem. C 2007, 111 (11), 4125. (124) Wanna, Y.; Srisukhumbowornchai, N.; Tuantranont, A.; Wisitsoraat, A.; Thavarungkul, N.; Singjaic, P. J. Nanosci. Nanotechnol. 2006, 6 (12), 3893 (125) An, K. H.; Jeong, S. Y.; Hwang, H. R.; Lee, Y. H. AdV. Mater. 2004, 16, 1005. (126) Zeng, J; Gao, X; Wei, W.; Zhai, X.; Yin, J.; Wu, L.; Liu, X.; Liu, K.; Gong, S. Sens. Actuators B 2007, 120, 595. (127) Santhanam, K. S. V.; Sangoi, R.; Fuller, L. Sens. Actuators B 2005, 106 (2), 766. (128) Ma, X.; Zhang, X.; Li, Y.; Wang, M.; Chen, H.; Mi, Y. Macromol. Mater. Eng. 2006, 291, 75. (129) Kim, S.; Lee, H. R. Appl. Phys. Lett. 2007, 91, 093126. (130) Qi, P.; Vermesh, O.; Grecu, M; Wang, Q.; Dai, H.; Peng, S.; Cho, K. J. Nano Lett. 2003, 3 (3), 347. Snow, E. S.; Perkins, F. K.; Robinson, J. A. Chem. Soc. ReV. 2006, 35, 790. (131) Tang, Q; Luo, X.; Wen, R. Anal. Lett. 2005, 38, 1445. (132) Lin, Y.; Cui, X. J. Mater. Chem. 2006, 585. (133) Ge, C. H.; Armstrong, N. R.; Saavedra, S. S. Anal. Chem. 2007, 79, 1401. (134) Ferrer-Anglada, N.; Kaempgen, M.; Roth, S. Phys. Status Solidi B 2006, 243 (13), 3519. (135) Li, Y.; Wang, P.; Wang, L.; Lin X. Biosens. Bioelectron. 2007, 22, 3120. (136) Kian Ping, L.; Sheng Liang, Z; Wei De, Z. Diamond Relat. Mater. 2004, 13 (4-8), 1075. (137) Ma, Y.; Ali, S. R.; Dodoo, A. S.; He, H. J. Phys. Chem. B 2006, 110, 16359. Ali, S. R.; Ma, Y.; Parajuli, R. R.; Balogun, Y.; Lai, W. Y.-C.; He, H. Anal. Chem. 2007, 79, 2583. (138) Xu, Y.; Ye, X.; Yang, L.; He, P.; Fang, Y. Electroanalysis 2006, 18 (15), 1471. (139) Muraviev, D. N.; Macanas, J.; Farre, M.; Munoz, M.; Alegret, S. Sens. Actuators B 2006, 118, 408. Muraviev, D. N.; Pividory, M. I.; Soto, J. L. M.; Alegret, S. SolV. Extract. Ion Exchange 2006, 24, 731. (140) Li, H. S.; Josowicz, M.; Baer, D. R.; Engelhard, M. H.; Janata, J. J. Electrochem. Soc. 1995, 142 (3), 798. (141) Kinyanjui, J. M.; Hanks, J.; Hatchett D. W.; Smith, A. J.; Josowicz, M. J. Electrochem. Soc. 2004, 151, D113. O’Mullane, A. P.; Dale, S. E.; Day, T. M.; Wilson, N. R.; Macpherson, J. V.; Unwin, P. R. J. Solid State Electrochem. 2006, 10, 792. Niu, L.; Li, Q.; Wei, F.; Wu, S.; Liu, P.; Cao, X. J. Electroanal. Chem. 2005, 578, 331. (142) Grace, A. N.; Pandian, K. Electrochem. Commun. 2006, 8, 1340. Amaya, T.; Saio, D.; Hirao, T. Tetrahedron Lett. 2007, 48, 2729. Baranton, S.; Coutanceau, C.; Leger, J.-M. Roux, C.; Capron, P. Electrochim. Acta 2005, 51, 517. Bian, C.; Yu, Y.; Xue, G. J. Appl. Polym. Sci. 2007, 104, 21. (143) Kinyanjui, J. M.; Hanks, J.; Hatchett, D. W. Macromolecules 2004, 37, 8745. Selvakannan, P. R.; Mandal, S.; Pasricha, R.; Sastry, M. J. Colloid Interface Sci. 2004, 279, 124. Guo, Z.; Zhang, Y.; Huang, L.; Wang, M.; Wang, J.; Sun, J.; Xu, L.; Gu, N. J. Colloid Interface Sci. 2007, 309, 518. (144) Selaraj, V.; Alagar, M.; Hamerton, I. Appl. Catal. B 2007, 73, 172. Selvaraj, V.; Alagar, M. Electrochem. Commun. 2007, 9, 1145. Choudary, B. M.; Roy, M.; Roy, S.; Kantam, M. L.; Sreedhar, B.; Kumar, K. V. AdV. Synth. Catal. 2006, 348, 1734. (145) Mascaro, L. H.; Gonc¸ alves, D.; Bulho˜es, L. O. S. Thin Solid Films 2004, 461, 243.

768 Chemical Reviews, 2008, Vol. 108, No. 2 (146) He, Z.-Q; Liu, W.-P.; Xiong, W.-P.; Shu, L.-Z.; Shu, H.; Wu, X.-M.; Chen, S.; Huang, K.-L. Wuji Huanxue Xuebao 2007, 23, 813. Luo, J.; Huang, H.; Lin, Z.; Hepal, M. ACS Symp. Ser. 2003, 832, 113. (147) Varela, H.; Bruno, R. L.; Torresi, R. M. Polymer 2003, 44, 5369. Lin, C.-L.; Lee, C.-C.; Ho, K.-C. J. Electroanal. Chem. 2002, 81, 524 - 525. (148) Dawson, A.; Kamat, P. V. J. Phys. Chem. B 2001, 105, 960. (149) Englebienne, P.; Hoonacker, A. V. J. Colloid Interface Sci. 2005, 292, 445. (150) Smith, A. J.; Josowicz, M.; Engelhard, M.; Baer, D. R.; Janata, J. Phys. Chem. Chem. Phys. 2005, 7 (20), 3619. (151) Min, B. K.; Friend, C. M. Chem. ReV. 2007, 107 (6), 2709. (152) Shin, H. J.; Hwang, I. W.; Hwang, Y. N.; Kim, D.; Han, S. H.; Lee, J. S.; Cho, G. J. Phys. Chem. B 2003, 107, 4699. (153) Smith, A. J.; Josowicz, M.; Janata, J. J. Electrochem. Soc. 2003, 150, E384. (154) Kinyanjui, J. M.; Hatchett, D. W.; Smith, A. J.; Josowicz, M. Chem. Mater. 2004, 16, 3390. Mallick, K.; Witcomb, M. J.; Dinsmore, A.; Scurrell, M. S. Macromol. Rapid Commun. 2005, 26, 232. Guo, Z.; Zhang, Y.; Huang, L.; Wang, M.; Wang, J.; Sun, J.; Xu, L.; Gu, N. J. Colloid Interface Sci. 2007, 309, 518. Mallick, K.; Witcomb, M. J.; Scurrell, M. S. J. Phys.: Condens. Matter 2007, 19, 1. (155) Huang, K.; Zhang, Y.; Long, Y.; Yuan, J.; Han, D.; Wang, Z.; Niu, L.; Chen, Z. Chem. Eur. J. 2006, 12 (20), 5314. (156) Hatchett, D. W.; Josowicz, M.; Baer, D.; Janata, J. Chem. Mater. 1999, 11, 2989. Chen, W.; Li, C. M.; Chen, P.; Sun, C. Q. Electrochim. Acta 2007, 52, 2845. (157) Patton, D.; Locklin, J.; Meredith, M.; Xin, Y.; Advincula, R. Chem. Mater. 2004, 16 (24), 5063. (158) Do, J.-S.; Chang, W.-B. Sens. Actuators B 2001, 72, 101. Do, J.-S.; Chang, W.-B. Sens. Acuators B 2004, 101, 97. (159) Bonfil, Y.; Brand, M.; Kirowa-Eisner, E. Anal. Chim. Acta 1999, 395, 85. Wang, J.; Tian, B. Anal. Chem. 1993, 65, 1529. Wang, J.; Tian, B. Anal. Chim. Acta 1993, 274, 1. (160) Chowdhury, A.-N.; Ferdousi, S.; Islam, M. M.; Okajima, T. J. Appl. Polym. Sci. 2007, 104, 1306. (161) Aslan, K.; Zhang, J.; Lakowicz, J.; Geddes, C. D. J. Fluoresc. 2004, 14, 391. (162) Ma, Y.; Li, N.; Yang, C.; Yang, X. Colloids Surf. A 2005, 269, 1. Grace, A.; Pandian, K. Colloids Surf. A 2007, 302, 113. (163) Xian, Y.; Liu, F.; Xian, Y.; Wang, H.; Jin, L. Biosens. Bioelectron. 2006, 21, 1996. (164) Granot, E.; Katz, E.; Basnar, B.; Willner, I. Chem. Mater. 2005, 17, 4600. (165) Haes, A. J.; Stuart, D. A.; Nie, S.; Van Duyne, R. P. J. Fluoresc. 2004, 14, 355. (166) Andersen, G. P.; Merrick, E. C.; Trammell, S. A.; Chinowsky, T. M.; Shenoy, D. K. Sens. Lett. 2005, 3, 151. Dutra, R. F.; Mendes, R. K.; da Silva, V. L.; Kubota, L. T. J. Pharm. Biomed. Anal. 2007, 43, 1744. (167) Farre´, M.; Martinez, E.; Ramo´n, J.; Navarro, A.; Radjenovic, J.; Mauriz, E.; Lechuga, L.; Marco, M. P.; Barcelo´, D. Anal. Bioanal. Chem. 2007, 388, 207. (168) Leroux, Y.; Eang, E.; Fave, C.; Trippe, G.; Lacroix, J. C. Electrochem. Commun. 2007, 9, 1258. (169) Ford, M. J.; Masens, C.; Cortie, M. B. Surf. ReV. Lett. 2006, 13 (2-3), 297. (170) Dubrovsky, T. B. In Protein Architecture: Interfacing Molecular Assemblies and Immobilization Biotechnology; Lvov, Y., Mohwald, H., Eds.; CRC Press: Boca Raton, 2000; pp 25-54. (171) Bearinger, J. P.; Tterretaz, S.; Michel, R.; Tirelli, N.; Vogel, H.; Textor, M.; Hubbell, J. A. Nat. Mater. 2003, 2, 259. (172) Gooding, J. J.; Mearns, F.; Yang, W.; Liu, J. Electroanalysis 2003, 15, 81. (173) Hatchett, D. W.; Wijeratne, R.; Kinyanjui, J. M. J. Electroanal. Chem. 2006, 593, 203. (174) Andreev, V. N.; Matveev, V. V.; Pisarev, S. A. Russ. J. Electrochem. 2006, 42 (9), 974. (175) Mallick, K.; Witcomb, M. J.; Dinsmore, A.; Scurrell, M. S. Langmuir 2005, 21, 7964. (176) Park, J.-E.; Park, S.-G.; Koukitu, A.; Hatozaki, O.; Oyama, N. Synth. Met. 2004, 141, 265. (177) O’Mullane, A. P.; Dale, S. E.; Day, T. M.; Wilson, N. R.; MaICPherson, J. V.; Unwin, P. R. J. Solid State Electrochem. 2006, 10, 792. Venancio, E. C.; Napporn, W. T.; Motheo, A. J. Electrochim. Acta 2002, 47, 1495. Niu, L.; Li, Q.; Wei, F.; Wu, S.; Liu, P.; Cao, X. J. Electroanal. Chem. 2005, 578, 331. (178) Andreev, V. N. Russ. J. Electrochem. 2006, 42, 1936. (179) Drelinkiewicz, A.; Hasik, M.; Kloc, M. Catal. Lett. 2000, 64, 41. Mourato, A.; Wong, S. M.; Siegenthaler, H.; Abrantes, L. M. J. Solid Sstate Electrochem. 2006, 10, 140. Amaya, T.; Saio, D.; Hirao, T. Tetrahedron Lett. 2007, 48, 2729.

Hatchett and Josowicz (180) Liu, Y.-C.; Hwang, B.-J.; Tzeng, I.-J. J. Electrochem. Soc. 2002, 149, H173. (181) Sakthivel, M.; Weppner, W. Ionics 2005, 11, 177. (182) Athawale, A. A.; Bhagwat, S. V.; Katre, P. P. Sens. Actuators B 2006, 114, 263. (183) Guo, M.; Pan, M.; Chen, J.; Mi, Y.; Zhang, X.; Chen, Y. Fenxi Huaxue 2006, 34, 1755. (184) Atashbar, M. Z.; Banerji, D.; Singamaneni, S.; Bliznyuk, V. Cryst. Liq. Cryst. 2005, 427, 217. (185) Ji, X.; Banks, C. E.; Holloway, A. F.; Jurkschat, K.; Thorogood, C. A.; Wildgoose, G. G.; Compton, R. G. Electroanalysis 2006, 18, 2481. (186) Zhou, H.; Chen, H.; Luo, S.; Chen, J.; Wei, W.; Kuang, Y. Biosens. Bioelectron. 2005, 20, 1305. Hong, N. X.; Luo, S.; Wei, W.; Kuang, Y. Fenxi Huaxue 2004, 32, 1445. (187) Sharma, S.; Nirkhe, C.; Pethkar, S.; Athawale, A. Sens. Actuators B 2002, 85, 131. (188) Chowdhury, A.-N.; Islam, M. S.; Azam, M. S. J. Appl. Polym. Sci. 2007, 103, 321. (189) Chowdhury, A. N.; Islam, M. S.; Azam, M. S. J. Appl. Polym. Sci. 2007, 103, 321. (190) Trung, T.; Trung, T. H.; Ha, C.-S. Electrochim. Acta 2005, 51 (5), 984. (191) Damian, A.; Omanovic, S. J. Power Sources 2006, 158 (1), 464. (192) Park, I.-S.; Kim, K.-J. Sens. Actuators A 2007, 135, 220. (193) Mohammadi, M. R.; Fray, D. J.; Cordero-Cabrera, M. C. Sens. Actuators B 2007, 124, 74. (194) Schnitzler, D. C.; Meruvia, M. S.; Humelgen, I. A.; Zarbin, A. J. G. Chem. Mater. 2003, 15, 4658. (195) Ram, M. K.; Yavuz, O ˜ .; Lahsangah, V.; Aldissi, M. Sens. Actuators B 2005, 106, 750. (196) Ram, M. K.; Yavuz, O.; Aldissi, Mt. Synth. Met. 2005, 151, 77. (197) Erades, L.; Grandjean, D.; Nayral, C.; Soulantica, K.; Chaudret, B.; Menini, P.; Parret, F.; Maisonnat, A. New J. Chem. 2006, 30, 1026. (198) Vijayanathan, V.; Venkatachalam, S.; Krishnamurthy, V. N. Synth. Met. 2000, 114, 273. (199) Varela, H.; Bruno, R. L.; Torresi, R. M. Polymer 2003, 44, 5369. (200) Saravanan, S.; Anantharaman, M. R.; Venkatachalam, S. Mater. Sci. Eng. B 2006, 135, 113. (201) Zucolotto, V.; Ferreira, M.; Cordeiro, M. R.; Constantino, C. J. L.; Moreira, W. C.; Oliveira, O. N., Jr. Sens. Actuators B 2006, 113, 809. (202) Miras, M. C.; Badano, A.; Bruno, M. M.; Barbero, C. Port. Electrochim. Acta 2003, 21 (3), 235. (203) Malhotra, B. D.; Chaubey, A.; Singh, S. P. Anal. Chim. Acta 2006, 578 (1), 59. Wei, D.; Ivaska, A. Chem. Anal. 2006, 51 (6), 839. Ramanaviciene, A.; Ramanavicius, A. A Crit. ReV. Anal. Chem. 2002, 32 (3), 245. (204) Malhortra, B. D.; Chaubey, A.; Singh, S. P. Anal. Chim. Acta 2006, 578, 59. (205) Ramanavicius, A.; Ramanaviciene, A.; Malinauskas, A. Electrochim. Acta 2006, 51 (27), 6025. (206) Low, L. M.; Seetharaman S.; He, K. Q.; Madou, M. J. Sens. Actuators B 2006, 67 (1-2), 149. Entezami, A. A.; Massoumi, B. Iranian Polym. J. 2006, 15 (1), 13. Otero, T. Proc. SPIE 2005, 5839, 286. (207) Xu, X. H.; Ren, G. L.; Cheng, J.; Liu, Q.; Li, D. G. J. Mater. Sci 2006, 41, 3147. (208) Pandy, R. S. S.; Takashima, W.; Kaneto, K. J. Appl. Polym. Sci. 2004, 93, 927. Li, X.; Sun, C. Anal. Chem. 2005, 60, 1073. (209) Singh, S.; Solanki, P. R.; Pandey, M. K.; Malhotra, B. D. Anal. Chim. Acta 2006, 568, 126. Singh, S.; Solanki, P. R.; Pandey, M. K.; Malhotra, B. D. Sens. Actuators B 2006, 115, 534. (210) Jiang, Y.; Wang, A.; Kan, J. Sens. Actuators B 2007, 124, 529. (211) Arora, K.; Sumana, G.; Saxena, V.; Gupta, R. K.; Gupta, S. K.; Yakhmi, J. V.; Pandey, M. K.; Chand, S.; Malhotra, B. D. Anal. Chim. Acta 2007, 594, 17. (212) Walter, E. C.; Penner, R. M.; Liu, H.; Ng, K. H.; Zach, M. P.; Favier, F. Surf. Interface Anal. 2002, 34, 409. (213) Kolmakov, A.; Zhang, Y. X.; Cheng, G. S.; Moskovits, M. AdV. Mater. 2003, 15, 997. Kolmakov, A.; Moskovits, M. Annu. ReV. Mater. Res. 2004, 34, 151. Zhang, Y.; Kolmakov, A.; Chretien, S.; Metiu, H.; Moskovits, M. Nano Lett. 2004, 4, 403. Zhang, Y.; Kolmakov, A.; Lilach, Y.; Moskovits, M. J. Phys. Chem. B 2005, 109, 1923. (214) Fan, Z. Y.; Lu, J. G. Appl. Phys. Lett. 2005, 86, 123510. Li, Q. H.; Liang, Y. X.; Wan, Q.; Wang, T. H. Appl. Phys. Lett. 2004, 85, 6389. (215) Zhang, Y. S.; Yu, K.; Jiang, D. S.; Zhu, Z. Q.; Geng, H. R.; Luo, L. Q. Appl. Surf. Sci. 2005, 242, 212. (216) Ozin, G.; Arsenault, A. Nanochemistry: A Chemical Approach to Nanomaterials; RSC Publishing: Cambridge, 2005.

Composites of Intrinsically Conducting Polymers (217) Kiely, C. J.; Fink, J.; Hutchinson, T.; Zheng, J. G.; Brust, M.; Bethell, D.; Schiffrin, D. J. Elect. Micro. Anal. 1999, 161, 515. (218) Panda, A. B.; Glaspell, G.; El-Shall, M. S. J. Phys. Chem. C 2007, 111 (5), 1861. (219) Panda, A. B.; Glaspell, G.; El-Shall, M. S. J. Am. Chem. Soc. 2006, 128 (9), 2790. (220) Pradhan, N.; Efrima, S. J. Phys. Chem. B 2004, 108 (32), 11964.

Chemical Reviews, 2008, Vol. 108, No. 2 769 (221) Rogach, A. L.; Talapin, D. V.; Shevchenko, E. V.; Kornowski, A.; Haase, M.; Weller, H. AdV. Funct. Mater. 2002, 12 (10), 653. (222) Tenne, R. J. Mater. Res. 2006, 21 (11), 2726. Tenne, R. Chem. Nanostruct. Mater. 2003, 147. (223) Tenne, R. Angew. Chem., Int. Ed. 2003, 42 (42), 5124.

CR068112H

Chem. Rev. 2008, 108, 770−813

770

Combinatorial and High-Throughput Development of Sensing Materials: The First 10 Years Radislav A. Potyrailo*,† and Vladimir M. Mirsky‡ Chemical and Biological Sensing Laboratory, Materials Analysis and Chemical Sciences, General Electric Global Research, Niskayuna, New York 12309, and Institute of Analytical Chemistry, Chemo- and Biosensors, University of Regensburg, 93040 Regensburg, Germany Received July 30, 2007

Contents 1. Introduction 1.1. Requirements for an Ideal Sensor 1.2. Challenges in Rational Design of Sensing Materials 2. Combinatorial and High-Throughput Materials Screening 2.1. General Principles 2.2. Opportunities for Sensing Materials 2.3. Gradient and Discrete Sensing Material Libraries 2.4. Dynamic Combinatorial Libraries 3. Inorganic Sensing Materials 3.1. Catalytic Metals for Field-Effect Devices 3.2. Metal Oxides 3.2.1. Conductometric Metal Oxide Sensors 3.2.2. Cataluminescent Metal Oxide Sensors 3.3. Plasmonic Nanomaterials 3.3.1. Nanoscale Materials for Surface-Enhanced Raman 3.3.2. Nanoscale Materials for Plasmon Resonance 3.4. Semiconductor Nanocrystals 4. Organic Sensing Materials 4.1. Indicator Dyes 4.2. Polymeric Compositions 4.3. Homo- and Copolymers 4.4. Conjugated Polymers 4.5. Molecularly Imprinted Polymers 5. Summary and Outlook 6. Acknowledgments 7. References

770 770 771 772 772 773 774 775 775 775 777 777 783 785 785 786 788 790 790 792 794 797 801 806 807 807

1. Introduction Chemical and biological sensors have found their niche among modern analytical instruments when real-time determination of the concentration of specific sample constituents is required. Development of sensors with new capabilities is driven by the ever-expanding monitoring needs of a wide variety of species in gases and liquids. On the basis of a variety of definitions of sensors,1-3 here we will accept that * Corresponding author (e-mail [email protected]). † General Electric Global Research. ‡ University of Regensburg.

a chemical or biological sensor is an analytical device that utilizes a chemically or biologically responsive sensing layer to recognize a change in a single or multiple chemical or biological parameters of a measured environment and to convert this information into an analytically useful signal. In a sensor device, a sensing material is applied onto a suitable physical transducer to convert a change in a property of a sensing material into a suitable physical signal. The obtained signal from a single transducer or an array of transducers is further processed to provide useful information about the identity and concentration of species in the sample. The energy transduction principles that have been employed for chemical and biological sensing involve radiant, electrical, mechanical, and thermal types of energy.4,5 Specific sensing concepts are further implemented with each energy transduction. Sensors based on radiant energy of transduction can employ intensity, wavelength, polarization, phase, or time resolution detection. Sensors based on electrical energy of transduction can employ conductometric, potentiometric, or amperometric detection. Sensors based on mechanical energy of transduction can employ gravimetric or viscoelastic detection. Sensors based on thermal energy of transduction can employ calorimetric or pyroelectric detection. Hyphenated techniques in sensing are of significant importance and combine several transduction techniques in one sensor.6,7 In addition to a sensing material layer and a transducer, a modern sensor system often incorporates other important components such as sample introduction and data processing components. In contrast to sensing based on intrinsic analyte properties (e.g., spectroscopic, dielectric, thermal), indirect sensing utilizes a responsive sensing material.2,5,8-13 This approach dramatically expands the range of detected species, improves sensor performance (e.g., analyte detection limits), and is more straightforwardly adaptable for miniaturization.6,14-33 These attractive features can be offset by some limitations of indirect sensors, for example, insufficient selectivity, poisoning, poor long-term stability, and slow response and recovery times. Nevertheless, indirect sensors constitute the most active research area in developing sensing approaches that cannot be addressed with direct sensing.

1.1. Requirements for an Ideal Sensor The exact configuration of a sensor for a particular application may be dictated by the nature and requirements of that application. Nevertheless, it is useful to set the features that one would wish in an ideal sensor for chemical and biochemical species. Requirements for an ideal sensor system

10.1021/cr068127f CCC: $71.00 © 2008 American Chemical Society Published on Web 01/23/2008

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 771 Table 1. Sensor Requirementsa system requirements and performance

Radislav Potyrailo is a Principal Scientist and Project Leader with General Electric Global Research Center in Niskayuna, NY, and an Adjunct Industrial Professor in the Department of Chemistry, Indiana University, Bloomington, IN. He received an engineering degree in optoelectronic instrumentation from Kiev Polytechnic Institute, Ukraine, in 1985. After holding a Visiting Scientist position at the Department of Electrical and Computer Engineering, University of Toronto, he received his Ph.D. in Analytical Chemistry from Indiana University, Bloomington, IN, in 1998. His research is focused on the development of new sensing platforms, sensing materials, and microanalytical instrumentation. He has over 40 issued U.S. patents; he has coauthored/coedited 7 books, published over 60 peer-reviewed papers, and given over 40 invited and keynote lectures at national and international technical meetings. He is the initiator and a co-organizer of the First Gordon Research Conference on Combinatorial and High Throughput Materials Science and other conferences and symposia on combinatorial methods in materials science and nanostructures for plasmonic sensing. Dr. Potyrailo is Editor of the Springer series Modern Microanalytical Systems.

Vladimir M. Mirsky received a M.D. in Biophysics from Moscow Medical University in 1981. Then he studied physical chemistry and electrochemistry, completed a postgraduate course in electrochemistry and physical chemistry at the Frumkin Institute of Electrochemistry of the USSR Academy of Sciences and obtained a Ph.D. in electrochemistry in 1986. In 1991 he was awarded the Alexander-von-Humboldt Research Fellowship and moved to Germany. In 1993−1994 he worked for the CNRS Centre of Molecular Biology in France. Since 1995, Dr. Mirsky has been working at the Institute of Analytical Chemistry, Chemo- and Biosensors, University of Regensburg (Germany), currently as professor. His research interests comprise applied biophysics and physical chemistry of interfaces and development of new chemical and biological sensors. He is/was a supervisor of 23 Master and Ph.D. works, a member of the editorial boards of three scientific journals, and an editor of a number of topic issues including the recent book Ultrathin Electrochemical Chemo- and Biosensors in the Springer Series on Chemical Sensors and Biosensors. The results of his work are presented in 18 patents or patent applications and in about 100 peer-reviewed scientific papers.

can be analyzed from the viewpoint of requirements for individual subsystems that include (1) sample introduction,

dynamic range ergonomic design false-positive rate initial cost long-term stability maintenance simplicity multicomponent detection multiple operation modes operation cost operation simplicity power consumption probability of detection response speed response reversibility robustness selectivity self-calibration sensitivity shelf life size sterilizability a

sample sensing transduction data introduction material principle analysis X X X X X X X X X X X X X X X X X

X

X

X

X X X X X X

X X X X X X X

X X X X X X

X X X X X X X

X X

X X X X X X X X

X X X X X X

Analytical definitions are provided in refs 5 and 34.

(2) sensing material, (3) transduction principle and implementation, and (4) data analysis. As shown in Table 1, the proper design of each individual subsystem has a significant impact on the overall system requirements and performance. For example, a desired sensor dynamic range can be obtained with an appropriate design of sample introduction, type of sensing material, transduction principle selected for the generation of the sensor signal, and data analysis. In real-world applications, the qualities of an ideal sensor are often weighted differently according to application. High reliability, adequate long-term stability, and resolution top the priority list for industrial sensor users, whereas often the size and maturity of the technology are the least important factors.2,3,6 A low false-positive rate and an ergonomic design are very critical for first responders.34 In contrast, medical users focus on cost for disposable sensors. Specific requirements for medical in vivo sensors include blood compatibility and minute size.35 Resistance to gamma radiation during sterilization, drift-free performance, and cost are the most critical specific requirements for sensors in disposable bioprocess components.36 The importance of continuous monitoring also differs from application to application. For instance, glucose sensing should be performed two to four times a day using home blood glucose biosensors, whereas blood-gas sensors for use in intensive care should be capable of continuous monitoring with sub-second time resolution.37,38

1.2. Challenges in Rational Design of Sensing Materials As illustrated in Table 1, all subsystems, including sensing materials, are equally important for a successful sensor. Rational design of sensing materials based on prior knowledge is a very attractive approach because it avoids timeconsuming synthesis and testing or numerous candidates. However, to be quantitatively successful, rational design39-43 requires detailed knowledge regarding how the intrinsic properties of sensing materials relate to their performance properties (e.g., magnitude of response to analyte and interferences, long-term stability, shelf life, resistance to poisoning, response and recovery times, best operation temperature, adhesion stability to substrate).

772 Chemical Reviews, 2008, Vol. 108, No. 2

Unfortunately, the multidimensional nature of the interactions between the function and the composition, preparation method, and end-use conditions of sensing materials often makes difficult the rational design of sensing materials for real-world applications.28,44-48 Thus, in addition to rational design, a variety of sensing materials have been developed using detailed experimental observations or discovered by chance.49-57 Such an approach in the development of sensing materials reflects a more general situation in materials design that is “still too dependent on serendipity” with only limited capability for rational materials design as recently noted by Eberhart and Clougherty.58 Examples of significant discoveries that have been recognized with the Nobel Prize in chemistry and that are currently being explored as the foundation of new sensing technologies include discovery and development of conductive polymers by Heeger, MacDiarmid, and Shirakawa (2000), discovery of fullerenes by Curl, Jr., Kroto, and Smalley (1996), development and use of molecules with structure-specific selective interactions (crown ethers) by Cram, Lehn, and Pedersen (1987), and investigations in surface chemistry by Langmuir (1932). Conventionally, detailed experimentation with sensing materials candidates for their screening and optimization consumes tremendous amounts of time and project cost without adding to “intellectual satisfaction”. Fortunately, new synthetic and measurement principles and instrumentation significantly accelerate the development of new materials. Several of these developments that are currently routinely used in creation and characterization of sensing and other materials have been recognized with Nobel Prizes in chemistry and physics. These include design of the first electron microscope by Ruska (1986), design of the scanning tunneling microscope by Binnig and Rohrer (1986), development of methodology for chemical synthesis on a solid matrix by Merrifield (1984), construction of maser and laser systems by Townes, Basov, and Prokhorov (1964), invention of the phase contrast microscope by Zernike (1953), invention of partition chromatography by Martin and Synge (1952), discovery of new light scattering effect by Raman (1930), and invention of the method of microanalysis of organic substances by Pregl (1923). The practical challenges in rational sensor material design also provide tremendous prospects for combinatorial and high-throughput research, which is the applied use of technologies and automation for the rapid synthesis and performance screening of relatively large numbers of compounds.59-67 Sensing materials can be categorized into three general groups that include inorganic, organic, and biological materials. In this review, inorganic sensing materials are defined as materials that have inorganic signal generation components (e.g., metals, metal oxides, semiconductor nanocrystals) that may or may not be further incorporated into a matrix. Organic sensing materials comprise indicator dyes, polymer/ reagent compositions, conjugated polymers, and molecularly imprinted polymers. The emphasis of this review is to comprehensively cover combinatorial and high-throughput development of inorganic and organic sensing materials. Development of biological receptors such as aptamers, peptides, and antibodies using combinatorial approaches has been extensively reviewed elsewhere68-73 and is beyond the scope of this review. Other earlier reviews were focused on applications of sensors47,74 and MEMS devices63,75 for highthroughput materials characterization and on high-throughput

Potyrailo and Mirsky

development of polymer- and biopolymer-based sensing materials.13 This review is organized in several sections. First, it sets a stage with the general principles of combinatorial and highthroughput materials screening technologies followed by the analysis of opportunities that are provided for sensing materials from these new technologies. A section on combinatorial inorganic sensing materials provides a critical analysis of developments in catalytic metals for field-effect devices and metal oxides for conductometric and cataluminescent sensors, plasmonic, and semiconductor nanocrystal materials. A section on combinatorial organic sensing materials provides a critical analysis of developments in indicator dyes, polymeric compositions, homo- and copolymers, conjugated polymers, and molecularly imprinted polymers. Because the widespread applications of combinatorial techniques for sensing materials are quite recent, this review also serves as an introduction to this field. We demonstrate that new parallel synthesis and advanced analytical instruments and data mining tools accelerate the discovery and optimization of sensing materials and provide more fundamental knowledge on the material fabrication with tailored initial and long-term stability properties.

2. Combinatorial and High-Throughput Materials Screening 2.1. General Principles In pharmaceutical and biotechnology industries, combinatorial synthesis and high-throughput analysis methods have found their applications in systematic searching of large parameter spaces for new candidate therapeutic agent molecules.76 Combinatorial chemistry originated in several laboratories around the world when Frank in Germany,77 Geysen in Australia,78 and Houghten in the United States79 developed methods to make more compounds in a shorter period of time.80 In materials science, the materials properties depend not only on composition but also on morphology, microstructure, and other parameters related to the material preparation conditions. As a result of this complexity, a true combinatorial experimentation is rarely performed in materials science with a complete set of materials and process variables explored. Instead, carefully selected subsets of the parameters are often explored in an automated parallel or rapid sequential fashion using highthroughput experimentation (HTE). The terms “combinatorial chemistry” and “combinatorial materials science” are often applied for all types of automated parallel and rapid sequential materials evaluation processes. Thus, an adequate definition of combinatorial and high-throughput materials science is a process that couples the capability for parallel production of large arrays of diverse materials together with different high-throughput measurement techniques for various intrinsic and performance properties followed by the subsequent navigation in the collected data for identifying “lead” materials.61-64,81 Individual aspects of accelerated materials development have been known for decades. These include combinatorial and factorial experimental designs,82 parallel synthesis of materials on a single substrate,83,84 screening of materials for performance properties,85 and computer data processing.86,87 However, it took the innovative scientific vision of Joseph Hanak to suggest in 1970 an integrated materials

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 773

Figure 1. Combinatorial and high-throughput approach for materials development: (A) typical combinatorial materials development cycle; (B) development phases of new sensing materials in sensors and opportunities for combinatorial and high-throughput experimentation.

development workflow.88 Its key aspects included (1) complete compositional mapping of a multicomponent system in one experiment, (2) simple, rapid, and nondestructive allinclusive chemical analysis, (3) testing of properties by a scanning device, and (4) computer data processing. Hanak was truly ahead of his time and “it took 25 years for the world to realize his idea”.60 In 1995, Xiang, Schultz, and co-workers initiated the avalanche of applications of combinatorial methodologies in materials science with the publication of their paper “A combinatorial approach to materials discovery”.89 Since then, combinatorial materials science has enjoyed much success, rapid progress for over a decade, and tremendous diversification into a wide variety of types of materials. Besides sensing materials, discussed in this review, examples of materials reported in conjunction with combinatorial and high-throughput screening techniques include superconductor,89 ferroelectric,90 magnetoresistive,91 luminescent,92 agricultural,93 structural,94 hydrogen storage,95 and organic light-emitting96 materials; ferromagnetic97 and thermoelastic98 shape-memory alloys; heterogeneous,99 homogeneous,100 polymerization,101 electrochemical,102 and hydrogen evolution103 catalysts; polymers,104 zeolites,105 and metal alloys;106 materials for methanol fuel cells,107 solid oxide fuel cells108 and solar cells;109 automotive,110 waterborne,111 vapor-barrier,112 marine,113 and fouling-release114 coatings, and others. A typical modern combinatorial materials development cycle is outlined in Figure 1A. Compared to an initial idea, the modern workflow has several new important aspects such as planning of experiments, data mining, and scaleup. In combinatorial screening of materials, concepts originally thought of as highly automated have been recently refined to have more human input, with only an appropriate level of automation. For the throughput of 50-100 materials formulations per day, it is acceptable to perform certain aspects of the process manually.115,116 To address numerous materials-specific properties, a variety of high-throughput characterization tools are required. Characterization tools are used for the rapid and automated assessment of single or multiple properties of the large number of samples fabricated together as a combinatorial array or “library”.62,117,118 Typical library layouts can be discrete82,89 and gradient.83,88,119-121 A specific type of library layout will depend on the required density of space to be explored, available library fabrication capabilities, and capabilities of high-throughput characterization tools.

In addition to the parallel synthesis and high-throughput characterization instrumentation that significantly differs from conventional equipment, the data management approaches also differ from conventional data evaluation.66 In an ideal combinatorial workflow, one should “analyze in a day what is made in a day”,122 which requires significant computational assistance. In an exemplary combinatorial workflow (Figure 1A), design and synthesis protocols for materials libraries are computer assisted, materials synthesis and library preparation are carried out with computer-controlled manipulators, and property screening and materials characterization are also software controlled. Furthermore, materials synthesis data as well as property and characterization data are collected into a materials database. Data in such a database are not just stored but also processed with the proper statistical analysis, visualization, modeling, and data mining tools. Table 2 illustrates important current capabilities and remaining needs for the data management system.123 Production of combinatorial leads on a larger scale reveals how reliable and realistic are the data obtained on the combinatorial scale. Materials developed at the combinatorial scale and validated on scale-up versions or in practical applications include catalysts, polymers, phosphors, formulated organic coatings, and sensing materials.13,67,116,124-130

2.2. Opportunities for Sensing Materials Development of a sensor system with a new sensing material includes several phases such as discovery with initial observations, feasibility experimentation, and laboratory-scale detailed evaluation, followed by the transition to the pilot scale and to commercial manufacturing (see Figure 1B). At the initial stage, performance of the sensing material is matched with the appropriate transducer for the signal generation. The stage of the laboratory-scale evaluation is very labor-intensive because it involves a detailed testing of sensor performance. Some of the aspects of this evaluation include optimization of the sensing material composition and morphology, its deposition method, detailed evaluation of response accuracy, stability, precision, selectivity, shelf life, long-term stability of the response, effects of potential poisons, etc. The pilot-scale manufacturing focuses on the identification and elimination of manufacturing issues that affect the reproducible, high-yield manufacturing of the sensors. During this phase, alpha and beta tests are typically performed. The alpha tests are typically performed by the

774 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Table 2. Functions of Data Management System123 function

current capabilities

remaining needs

experimental planning

composition parameters process parameters library design

iterative intelligent experimental planning based on results from virtual or experimental libraries

database

entry/search of composition/process variables operation with heterogeneous data unification of data between different instruments

storage and manipulation (search) of large amounts (terabytes and more) of data

instrument control

operation of diverse instruments

interinstrument calibration full instrument diagnostics plug’n’play multiple instrument configurations

data analysis

visualization of composition/process conditions and measured parameters of library elements univariate/multivariate processing of steady-state and kinetic data quantitative analysis outlier detection

advanced data compression processing of large amounts (terabytes and more) of data, cloud computing when required

data mining

prediction of properties of new materials virtual libraries cluster analysis molecular modeling QSAR

identification of appropriate descriptors on different levels (atomic, molecular, process, etc.)

researchers on an advanced sensor device prototype to identify issues related to sensor operation and functionality. Beta tests are typically performed on the further improved version of the sensor system by the identified group of endusers (“early adopters”) to seek their feedback on sensor performance, ease of use, failure modes, etc. The pioneering work by Xiang, Schultz, and co-workers published in 199589 inspired applications of combinatorial methodologies for sensing materials. In 1996, Lundstro¨m proposed to make arrays of sensing films with multiple types of metals.131 In 1996, Natan and co-workers applied a solution-based combinatorial assembly of metal nanoparticles to create an efficient substrate for surface-enhanced Raman measurements.132 In 1997, Walt and co-workers performed combinatorial polymerizations of sensing films and fabricated discrete and gradient film arrays.44 Combinatorial and highthroughput experimentation provides an attractive opportunity to accelerate the development and optimization of sensing materials. These development aspects of sensing materials are highlighted in Figure 1B and include initial screening, detailed performance screening, accelerated shelf life, and long-term stability testing. In this review, we will provide critical analysis of these developments. It is quite remarkable that most of the types of sensing materials have been explored with combinatorial technologies, which demonstrates the desire of the sensing community for the accelerated development of sensing materials using newly introduced research tools. Go¨pel showed that a theoretical

dimensionality of the hyperspace of independent chemical sensor features is ∼1021 (see Figure 2) and includes the permutations of various sensor materials, transducer principles, and modes of operation for each sensor/transducer combination.133-135

2.3. Gradient and Discrete Sensing Material Libraries In the past, spatial gradients in functional materials were generated by varying composite and structural characteristics.119 Gradients in polymeric materials were produced by changing the chemical nature of monomers, the molecular constitution of polymers, and their supramolecular structure or morphology.120 Several methods for fabrication of surface molecular chemical gradients have been reported.136-139 In sensing materials, additional parameters of gradients are also required that can include concentrations of formulation additives, thickness, temperature, extent of cross-linking, and some others. Gradient sensor libraries (or arrays) can be produced using solvent-assisted polymerization,44 fiber drawing,140,141 draw coating,48,142,143 or ink jet printing.144 Once a gradient sensor array is formed, it is important to estimate the possibilities to adequately measure the variation of properties along the gradient. These can be intrinsic (thickness, chemical composition, morphology, etc.) and performance (response magnitude, selectivity, stability, immunity to poisoning, etc.) properties. Discrete sensor arrays can be produced using ink jet printers,145-147 liquid-dispensing

Figure 2. Hyperspace of chemical sensor features with about 1021 independent features. Reprinted with permission from ref 133. Copyright 1998 Elsevier.

Combinatorial Development of Sensing Materials

robots,45,148 microarrayers,149 and automated dip-coating150 or during in situ polymerizations.151,152 Ink jet printing typically produces dispense volumes of around several picoliters, whereas the liquid-dispensing robots can dispense several nanoliters and up to several microliters.149 Once the gradient or discrete sensor film array is fabricated, it is exposed to an environment of interest and steadystate or dynamic measurements are acquired. Scanning (e.g., complex impedance, UV-vis, excitation-emission fluorescence, X-ray diffraction) systems often provide more detailed information compared to imaging systems. When a dynamic process (e.g., a response time) of sensor materials arranged in an array is monitored with a scanning system, the maximum number of elements in a sensor library that can be measured with the required temporal resolution can be limited by the data acquisition ability of the scanning system.143

2.4. Dynamic Combinatorial Libraries An emerging field in combinatorial chemistry is dynamic combinatorial chemistry,153-155 which is very attractive for the development of synthetic receptors and indicator dyes.156 Traditional combinatorial chemistry involves the use of irreversible reactions to generate static libraries of related compounds. Dynamic combinatorial chemistry involves the use of reversible reactions to generate equilibrating mixtures of molecules, known as dynamic combinatorial libraries (DCLs). The composition of a DCL is able to respond to molecular recognition events resulting from the addition of a target of interest. The preferential binding of one member of the DCL to the target induces a shift in the equilibrium toward the formation of that particular compound. Thus, whereas in combinatorial chemistry library synthesis and screening are two separate processes that are performed sequentially, dynamic combinatorial chemistry offers in situ screening of the combinatorial library simply by comparing its composition in the absence or presence of the target and identifying library members with a high affinity for the respective target.157

3. Inorganic Sensing Materials 3.1. Catalytic Metals for Field-Effect Devices In 1973, in the laboratory of Prof. Lundstro¨m, a phenomenon of hydrogen response of a thin Pd metal film that was tried as a gate of a field-effect transistor was discovered.158,159 Later observations in the same laboratory of sensing response on defective films with cracks and holes led to the discovery of effects of discontinuous metal gates and to the development of sensors for ammonia gas.160 Further developments of these sensors with gates of a variety of catalytic metals (Pd, Pt, Rh, Ir) demonstrated their sensitivity to numerous other gases (e.g., hydrogen sulfide, ethylene, ethanol, different amines). The mechanism of the response involves a change in the work function of the catalytic metal gate due to chemical reactions on the metal surface. Currently, it is understood that chemical reaction mechanisms in these sensors depend on the specific gas molecules as shown in Figure 3.161 For example, the response to hydrogen gas is due to a dipole layer, which is induced by trapping of hydrogen atoms at the metal-insulator interface of the device (see Figure 3A). The hydrogen atoms are first formed by dissociation of hydrogen molecules on the

Chemical Reviews, 2008, Vol. 108, No. 2 775

Figure 3. Mechanisms of sensing response to H2 and NH3 with the catalytic Pt metal gate: (A) in hydrogen gas, hydrogen atoms are mostly formed on the metal surface; (B) in ammonia gas, hydrogen atoms are produced mostly at the three phase boundaries. Reprinted with permission from ref 161. Copyright 2007 Elsevier.

Figure 4. Diversity of optimization parameters during the preparation of catalytic metals for field-effect devices.

catalytic metal surface, and then they diffuse through the metal layer until a steady state is established between the concentrations of hydrogen atoms on the surface and at the metal-insulator interface. Some other hydrogen-containing molecules, such as hydrocarbons and alcohols, give rise to a response in a similar manner as hydrogen. However, the response mechanism to ammonia gas differs from the hydrogen response because of the fundamental differences in the dissociation sites. The ammonia sensitivity is related to the pores (exposed oxide) in the thin metal film (see Figure 3B).161,162 Optimization approaches of materials for these sensors involve several degrees of freedom163 as illustrated in Figure 4. To simplify screening of the desired material compositions and to reduce a common problem of batch-to-batch differences of hundreds of individually made sensors for materials development, the scanning light pulse technique (SLPT) has been developed by Lundstro¨m and co-workers.161,164,165 In these measurements, a focused light beam is scanned over a large-area semitransparent catalytic metal-insulator-semiconductor structure, and the photocurrent generated in the semiconductor depletion region is measured and creates a 2-D response pattern of the sensing film (aka “a chemical

Potyrailo and Mirsky

image”). An alternative method to scan such gradients utilizes a vibrating capacitor method developed by Mizsei.166-169 These chemical images were used to optimize properties such as chemical sensitivity, selectivity, and stability.170 When combined with surface characterization methods, this information also has led to the increased knowledge of gas response phenomena. Early in these investigations,164,171 it was shown how to take advantage from the ability of forming a temperature gradient along the test structure to study the temperature-dependent catalytic activity of metals. The catalytic activity of metals is related to the reactive sticking coefficient s, which. in turn, relates to the probability that an adsorbate is formed when a molecule hits an unoccupied part of the sensor surface. The sticking coefficient s and the rate constants for reactions r1 and r2, which remove an adsorbate from the surface, are thermally activated with activation energies Es, E1, and E2, respectively. The variables s, r1, and r2 are the functions of position, x, along the surface due to the temperature gradient. Thus, the temperaturegradient-induced response can be described as171

(

)

s(x) d ∼ {(Es - E1) exp(-E1/kT) + dx r1(x) + r2(x) r21(x) (Es - E2) exp(-E2/kT)}

dT (1) dx

where dT/dx is determined by the variation of the operating temperature T along the surface, k is Boltzmann’s constant, and r21 ) c2/c1 is the ratio of the frequency factors of the two reaction rates. The use of combinatorial gradient techniques facilitated the increase of the understanding of the properties of catalytic metal gates and their influence on the selectivity and sensitivity of gas-sensitive field-effect devices. Temperature gradient experiments provided the required knowledge for selection of specific operating temperatures for detection of different gases.161,164 By using the 1-D thickness gradients to study catalytic films, the effects of the variation of the film thickness that influence the gas response sensitivity, selectivity, and stability were discovered.172-174 It was suggested that a 2-D gradient made from two types of metal films as a double-layer structure should provide new capabilities for sensor materials optimization, unavailable from thickness gradients of single-metal films.175 To make a 2-D gradient, the first metal film was evaporated on the insulator with the linear thickness variation in one dimension by moving a shutter with a constant speed in front of the substrate during evaporation. On top of the first gradient thickness film, a second metal film was evaporated with a linear thickness variation perpendicular to the first film. As the validation of the 2-D array deposition, the response of devices with 1-D thickness gradients of Pd, Pt, and Ir films to several gases has been studied with SLPT, with results similar to those of corresponding discrete components.170 The 2-D gradients have been used for studies and optimization of the two-metal structures170,175 and for determination of the effects of the insulator surface properties on the magnitude of sensing response.176 Two-dimensional gradients of Pd/Rh film compositions were also studied to identify materials compositions for the most stable performance.170 The Pd/Rh film compositions were tested for their response stability to 1000 ppm of hydrogen upon aging for 24 h at 400 °C while exposed to 250 ppm of hydrogen (see Figure 5A,B). This accelerated aging experiment of the 2-D

Figure 5. Results of the accelerated aging of 2-D combinatorial library of Rh/Pd film: chemical response images to 1000 ppm of hydrogen (A) before and (B) after the accelerated aging; (C) differential response after and before the accelerated aging (the most stable regions have the darkest color). Aging conditions: 24 h of exposure to 250 ppm of hydrogen at 400 °C. Reprinted with permission from ref 170. Copyright 2005 The Institute of Electrical and Electronics Engineers, Inc.

776 Chemical Reviews, 2008, Vol. 108, No. 2

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 777

Figure 7. Key strategies for development of semiconducting metal oxide materials for conductometric sensors: (A) polycrystalline SnO2, with two discontinuities; (B) single-crystal straight nanowire. Solid color and lines on the left correspond to vacuum or reducing conditions. Lighter areas on the right show the formation of depletion layers that reduce the effective diameter of the conducting channel (dashed arrows). A- is adsorbed electron-acceptor species, Vs is adsorbate-induced Schottky barrier, r is the initial radius of the conducting channel in a straight nanowire, W is radius of the conducting channel upon adsorption. Reprinted with permission from ref 184. Copyright 2007 Institute of Physics Publishing.

different gate metallizations (see Figure 6). These four evaporation steps (n ) 4) through two masks will produce 2n ) 16 different materials compositions. Such an approach should be very attractive in the future to study layered and alloyed multicomponent film compositions that are difficult or impossible to make using gradient deposition techniques.

3.2. Metal Oxides 3.2.1. Conductometric Metal Oxide Sensors Figure 6. Use of binary masks for synthesis of a combinatorial library of catalytic metals for a gas-sensing field-effect chip to produce different selectivity patterns at different areas on the chip: (A) 16 different areas are obtained with variable metal compositions and 4 evaporation steps (1, 500 Å of Pd; 2, 100 Å of Pt; 3, 100 Å of Ir; 4, 100 Å of Ru); (B) example of the principal structure of the gate in area 15. The exact composition of the gates (alloys, mixtures, layered structures) depends on the metals used and the details of the fabrication methods. Reprinted with permission from ref 131. Copyright 1996 Elsevier.

gradient film surface demonstrated the existence of two of the most stable local regions. One region was a “valley” of a stable response shown as a dark color in Figure 5C. Another region was a thicker part of the two-component film with a ∼20 nm thick Rh film and a ∼23 nm thick Pd film. This new knowledge inspired new questions of position stability of the valley and the possibility to improve sensor stability by an initial annealing process. In the developed SLPT methodology, the gate layers were typically electrically conducting metal films. To expand the applicability of this high-throughput screening technique to other functional materials beyond catalytic metals, a new grid gate structure has been developed.177 This advancement made possible investigations of semiconductor or insulating materials with the possibility of studying nanometer-sized clusters of functional deposits in atmospheric conditions. To make sensing films with multiple types of metals, Lundstro¨m131 proposed to adapt a sensing film deposition strategy from combinatorial chemistry178 and combinatorial materials synthesis89 applications. In this approach, the desired pattern of sensing films can be produced using a set of binary masks and a mask that defines the gate areas on the chip. The use in sequence of only two shadow masks and four metal evaporation steps can give 16 areas with

The origin of conductometric gas sensors that utilize semiconducting materials goes back to the 1950s to the discoveries of gas reaction effects with germanium by Brattein and Bardeen179 and with semiconducting metal oxides by Heiland180 and Bielanski and co-workers.181 In the early 1960s, Seiyama182 and Taguchi183 fabricated the first such sensors. At present, in conductometric sensors, semiconducting metal oxides are typically used as gas-sensing materials that change their electrical resistance upon exposure to oxidizing or reducing gases. While over the years significant technological advances have been made that resulted in practical and commercially available sensors, new materials are being developed that further improve the sensing performance of these sensors. Current strategies for materials development in this type of sensors are illustrated in Figure 7.184 The first strategy (Figure 7A) typically employs polycrystalline 2-D films in which electron transport is determined by dimensionally small interconnections between metal oxide grains. Assuming (1) a narrow size distribution of the electrically active grains in the sensing layer and hence homogeneous electrical properties, (2) percolation paths of the conductivity independent of the work function changes, and (3) constant mobility of the charge carriers in the nanoparticles, the sensor conductance G can be described as185

G ) G0 exp{[(EC - EF)b - eVS]/kT}

(2)

where (EC - EF)b is the energy difference between the Fermi level and the conduction band in the bulk, e is the electron charge, and VS is the band bending. In general, both (EC EF)b and eVS may change upon gas exposure. The second strategy (Figure 7B) is based on adsorbateinduced change in the effective cross section of the conduct-

778 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 8. Double-gradient sensor microarray for selective gas detection. (A) Sensor schematic illustrating a single metal oxide thin film segmented by electrodes (SE) and arranged on a temperature gradient heater. The sensing film is further covered with a gradient thckness ceramic membrane. Used with kind permission from Goschnick. (B) False color thermal infrared image of the heated gradient sensor array with a temperature gradient of 6.7 °C/mm. The white arrow depicts the airflow direction. Reprinted with permission from ref 198. Copyright 2004 Molecular Diversity Preservation International. (C) Radar plot of resistance of 38 sensor segments operated with a temperature gradient from 310 to 360 °C (6.7 °C/mm). Resistance values (Mohm): acetone (red), ethanol (blue), ammonia (green), and propanol vapors (light blue). Reprinted with permission from ref 198. Copyright 2004 Molecular Diversity Preservation International. (D) Results of the linear discrimination analysis of the signal patterns in practical tests to detect gaseous precursors of smoldering fires induced by overheated cable insulation (ETFE: ethylene tetrafluorine ethylene). Reprinted with permission from ref 195. Copyright 2002 The Institute of Electrical and Electronics Engineers, Inc.

ing channel of a single-crystal nanowire from its initial radius r to the changed radius of the conducting channel W upon adsorption184

W ∼ LD(eVS/kT)1/2

(3)

where LD is the Debye length for the semiconductor metal oxide, which is a measure of the field penetration into the bulk, LD ) (0kT/e2n0)1/2,  being the dielectric constant, 0 the permittivity of free space, and n0 the carrier concentration. Details on these response mechanisms are available in several reviews.12,186-191 Realizing the opportunities that arise with the temperature dependence of the sensor response described by eqs 2 and 3, temperature-gradient-based sensors that utilize a single metal oxide thin film segmented by electrodes have been developed by Goschnick and co-workers.192-198 In addition to the spatial temperature gradient heater, one of the designs of the sensor chip also had a SiO2 or Al2O3 membrane with a gradient thickness from 2 to 50 nm (see Figure 8A).199 Such a ceramic membrane provided an additional response selectivity200 through the thickness-dependent gas transport. To fabricate such a temperature and membrane gradient sensor, a gas-sensitive SnO2:Pt film (Pt content of 0.8 atom %) was deposited onto a thermally oxidized Si wafer by RF magnetron sputtering using a shadow mask. Next, Pt strip electrodes and two meander-shaped thermoresistors were

sputtered on the same side of the substrate as the SnO2 film, under a shadow mask for structuring the films. The arrangement of the electrodes subdivided the monolithic SnO2 film into 38 sensor segments on an area of 4 × 8 mm2. Finally, Pt heaters were deposited onto the backside of the substrate to operate the chip with the 50 °C temperature gradient from 310 to 360 °C (see Figure 8B).198 Operation of the sensor with the SnO2:Pt film for the detection of acetone, ethanol, ammonia, and propanol vapors is illustrated in Figure 8C. The application of a temperature gradient increased the gas discrimination power of the sensor by 35%. The sensor with a SiO2 gradient thickness membrane was employed for detection of gaseous precursors of smoldering fires induced by overheated cable insulation (see Figure 8D).195 The microstructure of the metal oxides deposited onto sensors depends on the deposition method (e.g., evaporation, sputtering, sol-gel techniques, aerosol methods, screenprinting) and material processing conditions.12,201 For example, in chemical vapor deposition (CVD), one must choose suitable precursor chemistry, reagent gas concentrations, and precursor partial pressure to control the composition and microstructure of metal and metal oxide thin films.202 Sputter deposition processes are typically sensitive to target condition, sputtering power, reactant concentrations, and substrate orientation. Nearly all deposition processes are strongly influenced by substrate temperature during and after deposition.

Combinatorial Development of Sensing Materials

Figure 9. Results of CVD experiments with TiO2 films performed using a single 16-element array and incremental deposition temperatures: (A) optical micrograph of a 16-element microhotplate array; (B-E) SEM microstructure images for four elements within the 16-element microarray. Dependence of microstructure on deposition temperature (oC): (B) 120; (C) 210; (D) 296; (E) 398. Reprinted from ref 202.

Semancik and co-workers have applied arrays of micromachined silicon microhotplates as a platform to tailor metal oxide thin film properties for microsensors for gas detection in air203-207 and for volatile organic compounds in water.208 To screen materials properties, microarrays of 4, 16, 36, and 48 individually addressable elements, each element with its own independent heating and electrical probe contacts, were employed. The nominal size and mass of the suspended structure were 100 × 100 µm and 0.2 µg, respectively. The low thermal mass and embedded heater enabled heat rates of 105-106 °C/s and operation temperatures of >500 °C. Figure 9 depicts representative results of CVD experiments with TiO2 films performed using a single 16-element array and incremental deposition temperatures to illustrate the microstructure’s temperature dependence when Ti(NO3)4 was used as the precursor.202 At the lowest deposition tempera-

Chemical Reviews, 2008, Vol. 108, No. 2 779

ture, formation of large grains was observed. With the increase in deposition temperature to 250 °C, the large grains subdivided along a particular crystallographic direction and formed a plate-like microstructure. Increasing the deposition temperature further to 400 °C caused these plates to subdivide into smaller grains. In the studies by Semancik and co-workers, the TiO2 was in the anatase form as shown by X-ray diffraction data.209 The anatase and rutile crystalline structures of TiO2 have different growth kinetics, thermodynamic stabilities, surface energies, and physicochemical properties. Thus, to facilitate the understanding of the mechanisms underlying gas response, Mazza et al. developed a method for the fabrication of libraries of nanostructured TiO2 films with a controlled gradient of the rutile/anatase ratio, film thickness, morphology, and crystalline dimensions.210 This was obtained by using rutile <10 nm diameter clusters as seeds for crystal growth in competition with anatase nucleation from the amorphous phase. With a simple one-step postdeposition thermal treatment, a 14 × 15 array with individual 1 × 1 mm2 chemoresistive sensing elements was formed and characterized for vapor response at 300-500 °C. Nanostructured TiO2 films with variable anatase/rutile ratio were controlled independently from the size of the nanoparticles that allowed detailed exploration of chemical and process conditions parameters. Using a micro-Raman mapping, the relative abundance of the rutile/anatase phases in annealed samples was determined as shown in Figure 10A. These determinations were done by examining the Raman Eg mode of anatase (144 cm-1) and Eg and A1g modes of rutile (447 and 612 cm-1); those intensities depend on abundances. Measurements of vapor response were performed using a volt amperometric technique at a constant 10 V potential and measuring the resultant current.211 The variation of conductance ∆G can be described as ∆G/G ) ACgB, where A and B are empirically determined constants related to the temperature, grain size, film porosity, and gas adsorption.211 Figure 10B shows responses to 200 ppm of methanol from three representative locations in the array. The sensitivity of the sensors was found to be correlated to the rutile/anatase ratio of the sensing layers. To enhance the response selectivity and stability, an accepted approach is to formulate multicomponent materials that contain additives in metal oxides. Introduction of additives into base metal oxides can change a variety of materials properties including concentration of charge carriers, energetic spectra of surface states, energy of adsorption and desorption, surface potential and intercrystallite barriers, phase composition, sizes of crystallites, catalytic activity of the base oxide, stabilization of a particular valence state, formation of active phases, stabilization of the catalyst against reduction, the electron exchange rate, etc.12,189 Dopants can be added at the preparation stage (bulk dopants) that will affect the morphology, the electronic properties of the base material, and its catalytic activity. However, the fundamental effects of volume dopants on base materials are not yet predictable.212 Addition of dopants to the preformed base material (surface dopants) can lead to different dispersion and segregation effects depending on the mutual solubility12 and influence of the overall oxidation state of the metal oxide surface.12,188,189,212 The diverse optimization parameters during the preparation of metal oxide sensing films are summarized in Figure 11.

780 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 10. Screening of vapor response of cluster-assembled nanostructured TiO2 films with a gradient in the rutile/anatase ratio: (A) quantitation of rutile/anatase ratio from ratiometric Raman analysis; (B) dynamical response of three sensors exposed to 200 ppm of methanol. (Inset) Map of the deposited sensor 14 × 15 element array; gradient in film thickness is shown as gray scale. The central white column indicates no deposition of TiO2. Green, blue, and red dashed vertical lines in (A) indicate corresponding sensor responses in (B). Reprinted with permission from ref 210. Copyright 2005 American Institute of Physics.

Figure 11. Diversity of optimization parameters during the preparation of metal oxide sensing materials.

To improve the productivity of materials evaluation by using combinatorial screening, Semancik and co-workers

Potyrailo and Mirsky

employed 36-element sensor arrays to evaluate various surface-dispersed catalytic additives on equivalent CVD SnO2 films.213,214 Catalysts were deposited by evaporation to nominal thicknesses of 3 nm, and then the microhotplates were heated to effect the formation of a discontinuous layer of catalyst particles on the SnO2 surfaces. The layout of the fabricated 36-element library is shown in Figure 12A. The response characteristics of SnO2 with different surfacedispersed catalytic additives are presented in Figure 12B. These radar plots show sensitivity results to benzene, hydrogen, methanol, and ethanol for operation at three temperatures. Fabrication of gas sensor materials was also demonstrated using a combinatorial pulsed-laser deposition (PLD) approach that provides a spatially selective deposition of compositionally varying discrete or gradient samples with arbitrary layout designs.215 Takeuchi and co-workers fabricated thinfilm combinatorial gas sensor libraries based on doped semiconducting SnO2 thin films.216 Deposition of 50 nm thick sensor films of different compositions was performed at 550 °C on 2 × 2 mm Au electrode patterns (see Figure 13A). The electrode patterns were fabricated on Al2O3 substrates using a photolithographic lift-off process prior to the sensor film deposition. Each sensor array consisted of 16 different compositions of the SnO2 host material and ZnO, WO3, In2O3, Pt, and Pd dopants, added to modify the selectivity pattern of the sensors.186,217 The ability of the developed materials to distinguish different gas species is depicted in Figure 13B for responses to 100 ppm of chloroform, formaldehyde, and benzene. Each gas produced a different response pattern with the five sensing materials. The response patterns for different gases were distinct with gas concentrations down to 12.5 ppm. It was suggested that this combinatorial PLD technique can be applied for not only screening of different dopants in SnO2 and other semiconducting films but also for manufacturing of compact sensor arrays. To expand the capabilities of screening systems, it is attractive to characterize not only the conductance of the sensing materials with DC measurements but also their complex impedance spectra.218 The use of complex impedance spectroscopy provides the capability to test both ionand electron-conducting materials and to study electrical properties of sensing materials that are determined by the material microstructure, such as grain boundary conductance, interfacial polarization, and polarization of the electrodes.219,220 Simon and co-workers designed and built a 64 multielectrode array for high-throughput impedance spectroscopy (10-107 Hz) of sensing materials (see Figure 14A).219 In this system, an array of interdigital capacitors was screen-printed onto a high-temperature-resistant Al2O3 substrate. To ensure the high quality of determinations, parasitic effects caused by the leads and contacts have been compensated by a software-aided calibration.219 After the system validation with doped In2O3 and automation of the data evaluation,220 the system was implemented for screening of a variety of additives and matrices with the long-term goal to develop materials with improved selectivity and longterm stability. Sensing films were applied using robotic liquid-phase deposition based on optimized sol-gel synthesis procedures. Surface doping was achieved by the addition of appropriate salt solutions followed by library calcination. Screening results at 350 °C of thick films of WO3 and In2O3 base oxides surface doped with various metals are presented as bar diagrams in Figure 14B,C, respectively.221,222 The

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 781

Figure 12. Combinatorial study of effects of surface dispersion of metals into CVD-deposited SnO2 films: (A) layout of a 36-element library for study of the sensing characteristics of SnO2 films with 3 nm of surface-dispersed Pt, Au, Fe, Ni, or Pd (Con. ) control; each sample was made with six replicates); (B) radar plots of sensitivity results to benzene, hydrogen, methanol, and ethanol for operation at 150, 250, and 350 °C. Reprinted from ref 213.

effects of various surface doping elements on the gas-sensing properties of In2O3 thick films sensors at multiple temperatures over 250-400 °C are compared in Figure 14D.222 It was found that some doping elements lead to changes in both the conductivity in air and gas-sensing properties toward oxidizing (NO2, NO) and reducing (H2, CO, propene) gases. Correlations between the sensing and the electrical properties in reference atmosphere indicated that the effect of the doping elements was due to an influence on the oxidation state of the metal oxide surface rather that to an interaction with the respective testing gases. This accelerated approach for generating reliable systematic data was further coupled to the data mining statistical techniques that resulted in the development of (1) a model associating the sensing properties and the oxidation state of the surface layer of the metal oxide based on oxygen spillover from doping element particles to the metal oxide surface and

(2) an analytical relationship for the temperature-dependent conductivity in air and nitrogen that described the oxidation state of the metal oxide surface, taking into account sorption of oxygen.222 Simon and co-workers further employed this highthroughput complex impedance screening system for the reliable screening of a wide variety of less explored material formulations. Polyol-mediated synthesis has been known as an attractive method for the preparation of nanoscaled metal oxide nanoparticles.223 It requires only low annealing temperatures and provides the opportunity to tune the composition of the materials by mixing the initial components on the molecular level.224,225 To explore previously unknown combinations of p-type semiconducting nanocrystalline CoTiO3 with different volume dopants as sensing materials, the polyol-mediated synthesis method was used to synthesize nanometer-sized CoTiO3 followed by the volume-doping

782 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 13. Application of a microfabricated electrode sensor array for evaluation of gas responses of doped semiconducting PLD-deposited SnO2 thin films: (A) microfabricated electrode sensor array; (B) radar plot of the normalized drift-corrected resistance change of five sensors to different gases at 400 °C. Reprinted with permission from ref 216. Copyright 2003 American Institute of Physics.

with Gd, Ho, K, La, Li, Na, Pb, Sb, and Sm (all at 2 atom %). The SEM-estimated primary particle size of the volume doped CoTiO3 materials was in the range from 30 to 140 nm, with the smallest particle size for CoTiO3:La and the largest for CoTiO3:K. From the temperature-dependent responses to propylene and ethanol, it was discovered that the CoTiO3:La [with a precursor of 0.730 mmol of Ti[OCH(CH3)2]4 and 0.715 mmol of Co(CH3COO)2‚4H2O] had an outstanding ethanol response compared to all other materials and insensitivity to air humidity [0-90 % relative humidity (RH)]. Evaluation of response selectivity was tested for all materials in the library. The selectivity of the volume-doped materials toward ethanol is shown in Figure 15A, demonstrating that highest selectivity (at 425 °C) was found for CoTiO3:Pb and CoTiO3:K. Because of the highest ethanol sensitivity, the new promising CoTiO3:La material was further used for the long-term stability testing in 45% RH air for 200 h to examine the baseline stability and reproducibility of the signal change to a given analyte concentration. The stability test for CoTiO3: La was done by periodic measurements of the sample resistance in air and the response and recovery behavior toward propylene (20 ppm) every 12 h as shown in Figure 15B. Doping strategy for this CoTiO3:La material has been also expanded with its surface doping with various metals (Au, Ce, Ir, Pd, Pt, Rh, Ru) in different concentrations (0.2-0.6 atom %).224 The enormous productivity of this highquality, high-throughput experimentation approach has been further demonstrated in the fabrication and testing of p-type semiconducting perovskite-type LnMO3 (Ln ) La, Pr, Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, Lu; M ) Cr, Fe) oxides. It was found that these materials exhibited a correlation between the (Ln-O) binding energy and the gassensing properties.226 It is known that there are some similarities between chemical sensing and heterogeneous catalysis, making catalytic activity of materials an important (but not a determining) parameter for sensing materials selection.189,190 Because of the possibility of similarity between the sensing and catalytic activity of materials, Maier and co-workers prepared libraries of sensing materials with different base oxides (WO3, SnO2, TiO2, ZrO2, In2O3, Bi2O3, Ce2O3) and bulk and/or surface dopants (Ag, Au, Ce, Co, Cu, Er, Fe, Gd, Lu, Mn, Pr, Ru, Sc, Sm, Tb, Th, Ti, V, Y, Yb)227 and evaluated their catalytic activity by the emissivity corrected IR-thermography, as

previously applied to heterogeneous catalysis.99 The motivation for the use of the IR-thermography was to use it as a rapid primary screening tool, capable of discovering new compositions with sensor properties. As a control, response of these materials at 250-400 °C was evaluated concurrently by resistance measurements. It was found that emissivitycorrected IR-thermography, although providing rapid imaging capability, ranked tested materials with deviations from the ranking obtained using automated resistive measurements.227 In particular, it was found that most materials (except the noble metal dopants) did not show the thermography/ resistivity correlation (see Figure 16).228 The thermography/ resistivity correlation for doped WO3 was slightly better than that for doped In2O3.228 Possibly, the materials-ranking ability of the IR-thermography could be improved when applied for screening of more diverse types of sensing materials including those operating at relatively low, even room, temperatures.190 The enormous amount of data collected during these experiments facilitated the successful efforts of Maier, Simon, and co-workers to develop data mining techniques229,230 and a database system.231 The developed data mining tools (see Figure 17) have been successfully applied to identify from resistance measurements several promising materials candidates such as In99.5Co0.5Ox, W99Co0.5Y0.5Ox, W98.3Ta0.2Y1Mg0.5Ox, W99.5Ta0.5Ox, and W99.5Rh0.5Ox with different gas-selectivity patterns.232 The formation of mixed oxides has an enormous potential for sensing materials originating from the opportunities for tailoring of chemical composition, microstructure, porosity, and surface properties.232-234 In contrast to crystalline materials, these amorphous mixed oxides are prepared under mild reaction conditions in ambient atmosphere, making available a variety of precursors, additives, modifiers, solvents, catalysts, and post-treatment conditions to provide numerous fine-tuning options. The functional properties of such solids are largely unexplored234 and provide a tremendous opportunity for the development of new sensing materials. The facile preparation and accessibility of these materials make them ideally suited for the application of high-throughput technologies to allow investigators for the first time to access and optimize mixed oxides on a realistic time scale.234 Initial results include the works by Maier and co-workers232 on In2O3/WO3 libraries and by Lee and co-workers233 on SnO2/ZnO libraries.

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 783

Figure 14. Screening of sensor metal oxide materials using complex impedance spectroscopy and a multielectrode 64-sensor array. (A) Layout of 64-sensor array. Reprinted with permission from ref 219. Copyright 2002 American Chemical Society. (B) Relative gas sensitivities at 350 °C of the WO3 base oxide materials library surface-doped with multiple salt solutions, all at 0.5 atom % (bulk dopant, 0.5 atom % Ta). Sequence of test gases and their concentrations (with air between) was H2 (50 ppm), CO (50 ppm), NO (5 ppm), NO2 (5 ppm), and propene (50 ppm). Reprinted with permission from ref 221. Copyright 2004 Wiley-VCH Publishers. (C) Relative gas sensitivities at 350 °C of the In2O3 base oxide materials library surface-doped with multiple salt solutions; concentration 0.1 atom % if not denoted otherwise; ND ) undoped. Sequence of test gases and their concentrations (with air between) was H2 (25 ppm), CO (50 ppm), NO (5 ppm), NO2 (5 ppm), and propene (25 ppm). Reprinted with permission from ref 222. Copyright 2007 American Chemical Society. (D) Dependence of the relative gas sensitivities on the concentration of doping elements of a library type shown in (C). Reprinted with permission from ref 222. Copyright 2007 American Chemical Society.

In the area of metal oxide sensing materials, combinatorial techniques will tackle very challenging and rewarding directions in the development of conceptually new materials that will provide new applications for practical metal oxide sensors. One significant application area could be to develop metal oxide materials that will be sensitive, selective, stable and free from water interferences when operating at low (80-150 °C) and room temperatures.235-240 Such materials will become a foundation for a new generation of low-power sensors. The second significant application area is sensors for exhaust monitoring in diesel engines and other harsh environment applications.241,242 Although diesel engines are one of the greener technologies, the operation temperatures of these engines are too high for conventional metal oxide

sensors. Thus, development of new sensing materials that will in situ selectively detect pollutants at high temperatures (500-700 °C) could be another focus area for combinatorial experimentation.

3.2.2. Cataluminescent Metal Oxide Sensors In 1976, Breysse and co-workers243 observed a chemiluminescence emission during a catalytic oxidation of carbon monoxide on thoria. Because this emission was due to the catalytic effect, it was named cataluminescence (CTL). Oxidation of gas molecules at the surface of the solid catalyst is the heterogeneous catalytic reaction. The mechanism of the CTL emission involves recombinant radiation and radiation from excited species.244 The reaction process is

784 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 17. Hierarchical clustering map of 2112 responses of diverse sensing materials to H2, CO, NO, and propene (Prop.) at four temperatures established from the high-throughput constant current measurements and processed with Spotfire data-mining software (clustering algorithm was “complete linkage” of the Euclidean distances). Reprinted with permission from ref 232. Copyright 2006 Molecular Diversity Preservation International.

Figure 15. Performance of nanometer-sized CoTiO3 synthesized via the polyol method with different volume dopants: (A) results of high-throughput screening of response selectivity to ethanol; (B) evaluation of long-term stability of CoTiO3:La at 400 °C. Reprinted with permission from ref 225. Copyright 2007 Elsevier.

Figure 18. Five stages of the heterogeneous catalytic reaction process that produce cataluminescence emission through the recombinant radiation and radiation from excited species. See text for details. Reprinted with permission from ref 244. Copyright 2005 Springer.

Figure 16. Correlation between temperature change and relative gas sensitivity of a library of In2O3 sensing films doped with a wide variety of surface dopants including different concentrations of noble metal (Au, Ag, Pd) dopants upon exposure to H2 at 350 °C. Used with kind permission of Maier and Simon.228

schematically depicted in Figure 18 and involves five stages. Initially, gas molecules R and O diffuse from the outer gas phase and reach the proximity of the catalyst surface. Next, gas molecules are chemisorbed to form Rad and Oad on the

catalyst surface with a part of the adsorbate desorbed to the gas phase. Furthermore, chemisorbed Rad and Oad react to produce chemisorbed ROad on the surface. Next, the reaction product RO is desorbed from the surface. Finally, RO diffuses off to the gas phase. Upon exposure to different organic vapors, the CTL emission effect has been observed on a variety of nanosized materials with different particle sizes such as MgO (∼28 nm), TiO2 (∼20 nm), Al2O3 (∼18 nm), Y2O3 (∼90 nm), LaCoO3:Sr2+ (∼50 nm), and SrCO3 (∼25 nm).245 ZnO nanoparticles were shown to have the CTL upon exposure to ethanol.246 The rate of CTL emission strongly depends on temperature through the rate of formation of chemisorption surface state

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 785

science. After numerous rejections, the eventual publication of the discovery of the surface enhancement Raman scattering (SERS) cross section250 is a classical example of slow acceptance by fellow scientists of new phenomena that are difficult or impossible to predict using rational approaches. At present, two commonly considered mechanisms for SERS include an electromagnetic enhancement and a chemical enhancement.251 For plasmonic structures and appropriate excitation conditions, the electromagnetic enhancement mechanism dominates. In this mechanism, the SERS enhancement factor EF at each molecule is a result of enhancing the incident excitation electromagnetic field Eout(ω) and the resulting Stokes-shifted Raman electromagnetic field Eout(ω - ων). The SERS electromagnetic enhancement factor is given by252

EFSERS(ων) )

|Eout(ω)2| |Eout(ω - ων)2| E04

(4)

where E0 is the magnitude of the applied field. On the basis of experimental measurements, the enhancement factor can be calculated from the SERS-enhanced Raman intensity ISERS(ων) normalized by the number of molecules bound to the enhancing metallic substrate Nsurf by dividing the normal Raman intensity INRS(ων) normalized by the number of molecules in the excitation volume Nvol given by252 Figure 19. Spectroscopic images on the cataluminescence emission-based sensor under the gradient temperature distribution from 440 to 530 °C over the 12 mm of heater length: (A) response to 800 ppm of ethanol in air; (B) response to 250 ppm of acetone in air. Reprinted with permission from ref 247. Copyright 1998 Elsevier.

or production of excited species. Nakagawa and co-workers247 developed a temperature gradient technique coupled with spectroscopic imaging to evaluate the sensing capabilities of different candidate materials during catalytic oxidation of organic vapors as a function of temperature and wavelength. A 12 mm long gradient heater was operated to produce a 90 °C temperature gradient from 440 to 530 °C. Figure 19 illustrates typical results obtained from the temperature gradient experiments. Several degrees of freedom were suggested for the discrimination of different vapors. These included the vapordependent ratio of CTL emission peaks, the activation energy of the CTL, and the temperature at the turning point of the CTL intensity peak in temperature dependence. In addition, it was possible to construct a sensor array that contained nanosized SrCO3, γ-Al2O3, and BaCO3 for quantitative analysis of explosive gases such as propane, n-butane, and isobutane in tertiary mixtures with detection limits in the 5-80 ppm range.248 Future possible improvements for the CTL sensor materials could be in the areas of discovery of catalysts with better selectivity, more active catalysts for lower temperature of operation, activators of the catalysts that emit at shorter wavelengths to avoid incandescent radiation, and higher efficiency emission catalysts.244

3.3. Plasmonic Nanomaterials 3.3.1. Nanoscale Materials for Surface-Enhanced Raman In a recent article,249 Van Duyne provided a brief historical perspective on difficulties to accept unpredicted concepts in

EFSERS(ων) )

{ISERS(ων)/Nsurf} {INRS(ων)/Nvol}

(5)

Reports on the enormously large SERS enhancement factors of 1014-1015 have facilitated the single-molecule detection253,254 and have inspired the development of new sensing materials. The SERS enhancements of 108-1010 are required to detect a single molecule.255 The maximum electromagnetic enhancement factor at the single-particle level was calculated256 to be ∼1011 and can be obtained at interstitial sites between particles and at locations outside sharp surface protrusions. It was suggested that the rest of the enhancement has to be contributed from the chemical enhancement. There are three types of the charge-transfer process that contribute to the chemical enhancement:257 (1) the change of molecular polarizability when the molecule interacts with the surface or other surface species; (2) the change of molecular polarizability when the molecule forms a surface complex with a metal ion or electrolyte ion; and (3) the photon-driven charge-transfer process that occurs when the incident laser energy matches the energy difference between the surface molecules’ HOMO or LUMO and Fermi level or surface state of the metal substrate. A chemical enhancement of 107 from Au bow-tie nanoantennas has been reported,258 which is much larger than previously believed (10-104). Detection of SERS has been accomplished on chemical systems with small or no plasmon resonance contributions, including small Ag clusters such as Ag8 and Ag20 (105 enhancement)259,260 and semiconductor nanocrystals (104 enhancement).261,262 Combinatorial approaches have been applied to optimize metallic SERS substrates. These approaches include 2-D variation of nanoparticle density to discover the existence of an optimal surface coverage for the most effective SERS enhancement132 and combinatorial exploration of the roughness effects of metallic substrates.263 Natan and co-workers132 proposed that the SERS enhancement of optimized periodic

786 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 20. Combinatorial discovery of the most active SERS region from a 2-D gradient of variable surface coverage of 12 nm diameter Au nanoparticles and the variable thickness of Ag shell: (left) background-subtracted SERS intensity map for the 1168 cm-1 phenylnitroso stretch of p-nitrosodimethylaniline (each shaded box represents one SERS spectrum collected in a 1 × 1 mm2 area); (right) AFM images of nanoparticles (500 × 500 nm2) from regions A-D. Reprinted with permission from ref 132. Copyright 1996 American Chemical Society.

structures can be much larger than that of simple island films. They demonstrated the power of solution-based combinatorial approaches for synthesis of surfaces exhibiting nanometer-scale variation in mixed-metal composition and architecture. The SERS response with a variable surface coverage of colloidal Au and the amount of Ag coating on Au nanoparticles (Ag staining) was studied in detail. A gradient in particle coverage was produced along the direction of immersion of a glass slide by a fixed rate immersion of the (3-mercaptopropyl)trimethoxysilane-coated glass slide into an aqueous solution of 12 nm diameter colloidal Au particles. The slide was further rotated by 90°, and fixed-rate immersion was performed into an Ag-ioncontaining solution for Ag staining of Au. This second immersion step produced a gradient in particle size over the new immersion direction. The resulting surface exhibited a continuous variation in nanometer-scale morphology as was defined by particle coverage and particle sizes. The SERS signal for adsorbed p-nitrosodimethylaniline was measured over a 2 × 2 cm sample exhibiting continuous gradients in Au coverage, and Ag cladding thickness showed a spatial map of the background-corrected SERS intensity for the phenylnitroso stretch at 1168 cm-1. A detailed interrogation revealed a region that was >103-fold more SERS-enhancing than the least active sites (see Figure 20). The nanometerscale morphology at positions of interest was determined by atomic force microscopy. These results showed the significant changes in SERS enhancement factor over only small alterations in surface morphology. Reproducible surface enhancement and surface treatment conditions that extend the shelf life of SERS surfaces have been recently reported (for several examples, see refs 264 and 265). The combinatorial computational approaches will play an increasingly important role for the identification of the best morphologies for “hot spots” and their gaps,266 for the best surface treatment conditions to extend the shelf life of SERS surfaces, and for deeper understanding of the effect of chemical enhancement.267,268

3.3.2. Nanoscale Materials for Plasmon Resonance First explanations of colored effects from colloidal Au were reported by Faraday in his Bakerian lecture.269,270 At present, for sensing applications, plasmon resonance on metal nanoparticles and surface plasmon resonance on thin metallic

films are widely used for chemical and biological sensing as summarized in recent reviews and books.252,271-275 In SPR with a uniform film, the sensor response R can be defined either as the shift in the wavelength ∆λ or angle ∆θ of the SPR minimum in reflected light intensity associated with analyte adsorption. The SPR sensor response R is related to the change in the refractive index ∆n due to the presence of adsorbed species as276

R ) m∆n {1 - exp(-2d/Ld)}

(6)

where m is the refractive index sensitivity, d is the effective thickness of a layer that experiences the change in the refractive index, and Ld is the characteristic electromagnetic field decay length. The localized surface plasmon resonance (LSPR) response is measured as the change in the extinction (or scattering) wavelength maximum, which is related to ∆n also through eq 6 but with a shorter Ld.252 The nanoparticle LSPR approach was established for biosensing with the pioneering work of Mirkin and co-workers, who demonstrated the selective colorimetric detection of polynucleotides with the use of the distance-dependent optical properties of Au nanoparticles.22 The approach that utilizes the aggregation-induced, red-to-blue color change associated with Au nanoparticles is attractive for myriad applications ranging from screening of combinatorial libraries of DNA-binding molecules with DNA-functionalized Au nanoparticles277 to chemical sensing.278-281 Analysis of hybridization events in combinatorial DNA arrays using oligonucleotide-modified and Ag-stained Au nanoparticle probes on a conventional flatbed scanner had a detection sensitivity exceeding that of the analogous fluorophore system by 2 orders of magnitude.282 Although plasmonic nanoparticles potentially offer properties that are unavailable in molecular or mesoscopic systems, in order to benefit from these properties in new functional components such as sensors, it is critical to develop methods for placing particles into chemically and structurally welldefined environments.283 To generate 2-D and 3-D assemblies of plasmonic nanoparticles, Genzer and co-workers developed an approach to use polymer brushes that offer environments for controlled organization of nanoparticles within polymer matrices.283-286 This desired arrangement of nanoparticles in polymers requires control of several parameters

Combinatorial Development of Sensing Materials

Figure 21. Diversity of optimization parameters during the preparation of polymeric sensing materials with plasmonic nanoparticles.

shown in Figure 21. The large number of parameters made probing the behavior of brush/nanoparticle hybrids very difficult without a combinatorial approach.285,287 Systematic studies of the effects of the loading of nanoparticles and polymer structure were performed by fabricating orthogonal polymer gradient substrates, in which the polymer molecular weight (MW) and grafting density σ were changed independently in two orthogonal directions (see Figure 22). The citrate-covered Au nanoparticles were attached to a poly(dimethylaminoethyl methacrylate) (PDMAEMA) polymer brush grafted on a silica substrate as shown in Figure 22A. The electrostatic interaction between the positively charged DMAEMA groups and citrate ions attached to the surface of Au nanoparticles at pH 6.5 was responsible for binding the nanoparticles to the underlying brush. Thus, the extinction wavelength maximum of plasmonic nanoparticles caused the color of the slide in Figure 22B to change from light red (region of low MW and low σ) to dark violet blue (region of high MW and high σ), indicating interparticle plasmon coupling associated with an increase in uptake of particles in the brush upon an increase of MW and σ. The increase in intensity of the plasmon absorbance peak in the direction of increasing MW or σ (see Figure 22C) was due to the increasing number of particles attached to the polymer chains. The concomitant red shift of the plasmon peak position suggested intensified interparticle plasmon coupling accompanying the nanoparticle crowding on the substrate. Upon variation of σ, the color variation was more pronounced for low MW (shorter) chains relative to longer chains, suggesting that the number of particles bound to the brush was dependent on the number of favorable sites that particles have access to.284 Initially employed Au nanoparticles were relatively large (16 nm) to penetrate the brush; however, the use of smaller ones (3.5 nm) made it possible to load the nanoparticles inside the polymer.285 In subsequent studies, Bhat and Genzer also developed an approach to control the number density of citrate-stabilized Au nanoparticles on flat substrates by varying the concentration of the grafted amino groups on the surfaces and their degree of ionization. The concentration of grafted amino groups was controlled by decorating silicabased substrates with a molecular gradient of (3-aminopropyl)triethoxysilane (APTES). The degree of ionization of the

Chemical Reviews, 2008, Vol. 108, No. 2 787

-NH2 groups in APTES was controlled by varying the pH of the Au colloid.286 To visualize and quantify the gradient distribution of plasmonic nanoparticles in sensing films, Dovidenko, Potyrailo, and co-workers developed an approach based on focused ion beam (FIB) cross-sectioning of a sensing film followed by the 3-D reconstruction of the spatial distribution of nanoparticles (see Figure 23).280 These new methods for controlled organization of nanoparticles within polymer matrices283-286 coupled with the visualization and quantitation of nanoparticles in the polymer280 are important for the development of chemical and biological sensors with the tailored dynamic range of sensor response. Plasmonic nanoparticles are also attractive for deposition onto solid sensor substrates in desired discrete arrangements. Numerous lithographic and other nanofabrication techniques are readily applicable for the formation of nanoparticle arrays.288-293 Among the lithographic techniques, dip-pen nanolithography294-296 is one of the most versatile for the fabrication of combinatorial materials libraries for their screening of ligand-binding events. Dip-pen nanolithography, when operated in nanoplotter mode, is a powerful tool for combinatorial nanotechnology297 because it can be programmed to generate a series of patterns that vary with respect to composition, feature size, and feature spacing and can be deposited onto different substrates. These patterns subsequently can be used to study chemical and biochemical recognition in the combinatorial fashion as was recently demonstrated for several combinatorial applications.298-300 The use of dip-pen nanolithography with conventional cantilevers offers a 5 nm spatial resolution,301 making it a straightforward tool to arrange plasmonic nanoparticles in a nanoarray format.302 The addressable combinatorial arrays were successfully used for light-directed, spatially addressable parallel chemical synthesis,178 combinatorial synthesis of materials arrays,89 light-directed assembly of metallic nanoparticles,303 and combinatorial patterning of nanocrystals304 and were proposed for catalytic metal film sensors.131 Koenderink and co-workers305 have theoretically demonstrated an approach for addressable combinatorial arrays using the principles of surface plasmon resonant nanolithography.289,290 The distinct emission patterns of hot spots from 1-D or 2-D arrays of plasmonic subwavelength nanoparticles were created even though all particles in the array were irradiated.305 Illumination with unfocused light of all particles in the array allowed optical addressing of particles by varying the wavelength, incidence angle, and polarization of the incident wave. The coherent coupling of all fields in closely spaced particles in an array was tuned through the relative phases with which particles were excited to controllably create hot/cold spots of constructive/destructive interference on a single or several dipoles depending on array geometry and illumination conditions.306 Figure 24 illustrates all five symmetry-distinct patterns that were created by illuminating a 2 × 2 square array of Ag particles (radius ) 25 nm, spacing ) 75 nm) with linearly polarized light under various angles. When symmetry was taken into account, combinatorial 24 binary combinations of four particles that can be exposed or unexposed was reduced to five unique patterns with at least one particle exposed. Such on-demand combinatorial excitation of particles in the array upon an illumination of the whole array with an unfocused light brings a host of opportunities for sensing,

788 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 22. Combinatorial approach for systematic studies of the simultaneous effects of the loading of Au nanoparticles and polymer structure: (A) schematic of the attachment of citrate-covered gold nanoparticles to a PDMAEMA polymer brush grafted on a silica substrate; (B) photograph of a glass slide showing a reflected light color originated from 16 nm diameter Au nanoparticles bound to an orthogonal σ-MW gradient of surface-grafted PDMAEMA; (C) visible light absorbance spectra taken along red circles (path A, constant MW, σ gradient) and green squares (constant σ, gradient MW) shown in (B). Reprinted with permission from ref 284. Copyright 2004 Wiley-VCH Publishers.

Figure 23. Visualization and quantitation of the gradient distribution of plasmonic nanoparticles in sensing films utilizing FIB crosssectioning followed by the 3-D reconstruction: (A) 3-D representation of the set; (B) demonstration of ability to visualize data along X, Y, and Z axes by moving one slice at a time along Z and continuously in X and Y directions. Data sets of ∼25 nm thick FIB serial slices are sectioned normal to Z direction in the picture (lateral dimensions, X ) Y ) 3.4 µm). Reprinted with permission from ref 280. Copyright 2006 Materials Research Society.

ranging from sequential excitation of differently functionalized particles in the array that have the same optical response to sequential metal-enhanced fluorescence readout to avoid photobleaching of the whole array at once. Further implementation of local plasmonic effects in sensors requires the development and optimization of highly reproducible yet cost-effective surfaces containing plasmonic nanostructures.251,307 High-thoughput materials development techniques for optimization of these structures as well as computational and combinatorial development of plasmonic negative refractive index metamaterials308,309 for sensors310,311 will be critical for the timely introduction of these new physical phenomena in practical bioanalytical applications.

3.4. Semiconductor Nanocrystals Semiconducting nanocrystals were independently introduced by Alivisatos and co-workers312 and Nie and Chan313

as labels for biodiagnostic applications and biotechnology. Although at present organic fluorophores dominate sensing applications because of the diversity of their functionality and well-understood methods of their synthesis, new semiconducting nanocrystal labels have several advantages (photostability, relatively narrow emission spectra, and broad excitation spectra30,314) over organic fluorophores. Thus, finding a solution to complement the existing organic fluorescent reagents with more photostable, yet chemically or biologically responsive, nanocrystals is very attractive. It is known that a variety of photoluminescent materials are sensitive to the local environment.315 In particular, polished or etched bulk CdSe semiconductor crystals316,317 and nanocrystals318,319 were shown to be sensitive to environmental changes. To better understand the environmental sensitivity of semiconductor nanocrystals upon their incorporation into polymer films, Potyrailo and Leach incorporated mixtures

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 789

Figure 24. Individual addressing of all combinations of Ag nanoparticles in a 2 × 2 array. The filled-in dots show the relative amount of excitation on each sphere; the contour plots show the resulting electric field intensity in a plane just 5 nm above the array; the location of the spheres is indicated by open circles. Reprinted with permission from ref 305. Copyright 2007 American Chemical Society. Table 3. Polymer Matrices for Incorporation of Different Sizes of CdSe Nanocrystals322 polymer

polymer type

1

poly(trimethylsilyl)propyne

2 3 4 5 6 7 8 9

poly(methyl methacrylate) silicone block polyimide polycaprolactone polycarbonate polyisobutylene poly(dimethylaminoethyl) methacrylate polyvinylpyrrolidone styrene-butadiene ABA block copolymer

of multisize CdSe nanocrystals into numerous rationally selected polymeric matrices (see Table 3) and screened these films for their photoluminescence (PL) response to vapors of different polarities upon excitation with a 407 nm laser.320-322 It was discovered that CdSe nanocrystals of different sizes (2.8 and 5.6 nm diameter) and passivated with tri-noctylphosphine oxide using known methods323,324 had dramatically different PL response patterns upon exposure to methanol and toluene after incorporation into polymeric matrices (see Figure 25A). As an example, Figure 25B shows response patterns of gas-dependent PL of the two sizes of CdSe nanocrystals in poly(methyl methacrylate) (PMMA) sensor film. The difference in the response patterns of the nanocrystals was attributed to the combined effects of the dielectric medium surrounding the nanocrystals, their size, and surface oxidation state. The sensing films were tested for 16 h under a continuous laser excitation and exhibited a

rationale for selection as sensor matrix polymer with largest known solubility of oxygen, candidate for efficient oxidation of CdSe nanocrystals polymer for solvatochromic dyes polymer with very high partition coefficient for sorbing organic vapors polymer for solvatochromic dyes polymer with high Tg for sorbing of organic vapors polymer with low Tg for sorbing of organic vapors polymer for surface passivation of semiconductor nanocrystals polymer for sorption of polar vapors polymer for sorption of nonpolar vapors

high stability of PL intensity.325 Results of cluster analysis of PL response patterns upon exposure to methanol and toluene after incorporation into polymeric matrices are demonstrated in the dendrogram in Figure 25C. The dendrogram was constructed by performing principal component analysis (PCA) on the data from Figure 25A and further using Mahalanobis distance on three principal components (PCs). From this dendrogram, it is clear that polymers 6 and 7 were the most similar in their vapor response with studied CdSe nanocrystals as demonstrated by a very small distance to K-nearest neighbor between them. Polymer 4 was the most different from the rest of polymers as indicated by the largest diversity distance to K-nearest neighbor. Such data mining tools provide a means to quantitatively evaluate polymer matrices. When coupled with quantitative structure-property relationship simulation tools that will incorporate molecular descriptors, new knowledge generated from high-throughput experiments may provide additional insights for the rational

Figure 25. Diversity of steady-state PL response of two-size (2.8 and 5.6 nm) mixtures of CdSe nanocrystals to polar (methanol) and nonpolar (toluene) vapors: (A) magnitude of PL change in nine polymer matrices listed in Table 3 (reprinted with permission from ref 322; copyright 2006 Materials Research Society); (B) gas-dependent PL of the two-size CdSe nanocrystals sensor film (polymer 2) with emission of 2.8 nm nanocrystals at 511 nm and emission of 5.6 nm nanocrystals at 617 nm (reprinted with permission from ref 321; Copyright 2006 American Institute of Physics); (C) results of cluster analysis of PL response patterns upon exposure to methanol and toluene after incorporation into nine polymer matrices. Numbers 1 and 2 in (B) are replicate exposures of sensor film to methanol (6% vol) and toluene (1.5% vol), respectively.

790 Chemical Reviews, 2008, Vol. 108, No. 2

design of gas sensors based on incorporated semiconductor nanocrystals. In the future, such work in vapor sensors promises to complement existing solvatochromic organic dye sensors with more photostable and reliable sensor materials.319 In sensors for ionic species,326 ligand screening will be also attractive to perform using combinatorial techniques.

4. Organic Sensing Materials 4.1. Indicator Dyes Although photoluminescent nanocrystals are becoming attractive for sensing applications as discussed in section 3.4, development of new colorimetric and fluorescence dyes as analyte-responsive reagents remains important in the design of new sensors with improved selectivity and sensitivity. Using existing knowledge, it is possible to computationally predict general spectral features of dyes.327,328 However, quantitatiVe computations of the selectivity of analyte recognition, extinction coefficient, and quantum yield of emission for new reagents are very challenging. Additional practical challenges may occur from difficulties in the synthesis of those best reagent structures predicted from computations. Thus, combinatorial synthetic approaches have been applied for the development of new fluorescent329-335 and colorimetric336-339 reagents. Recent reviews are available on the fluorescent labels and probes.340,341 Recent developments resulted in the discovery of a wide variety of new useful reagents for the detection of metal ions,329,330,333,336 ATP,334 GTP,335 dipeptides,338 tripeptides,339 and reagents that selectively bind to amyloid332 and different cell components.342,343 To navigate in the vast amount of data, Rosania, Chang, and co-workers342,343 demonstrated a cheminformatic strategy for a multiparameter analysis of combinatorial libraries of dyes to predict not only their spectral properties but also their analyte-binding abilities. This approach has been demonstrated with a library of styryl dyes that were fluorescent lipophilic cations and that were selectively accumulated in mitochondria and other regions in living cells. Furthermore, it was suggested that in addition to organellespecific binding, some dyes may possess DNA, RNA, or protein-specific binding features.342 Dye synthesis (see Figure 26A) was done by the condensation of 41 aldehydes (building block A) that were of various sizes, conjugation lengths, and electron-donating or -withdrawing capabilities and 14 pyridinium salts (building block B) that were 2- or 4-methylpyridine derivatives condensed with each other with a secondary amine catalyst. Due to the structural diversity, the emission colors of the library of compounds covered a broad range from 470 to 730 nm. The binding ability of the dyes library in the cells was assessed by the incubation of dyes with live human melanoma cells. From the detailed analysis of 119 of 276 fluorescent compounds that localized to specific subcellular compartments (i.e., mitochondria, endoplasmic reticulum, vesicles, nucleoli, chromatin, cytoplasm, or granules), the structure-binding relationships have been developed as presented in Figure 26B. A model was developed that related the spectral and subcellular localization characteristics of styryl compounds to the two chemical building blocks A and B that were used to synthesize the molecules. The model predicted the subcellular localization and spectral properties of the styryl compounds from numerical scores that were independently associated with the individual building blocks of the molecule. It was found that

Potyrailo and Mirsky

more complex, nonadditive interactions between the two building blocks also played a role in determining the molecule’s optical and biological properties. Dynamic combinatorial libraries were utilized by Buryak and Severin as highly selective colorimetric sensors for dipeptides338 and tripeptides.339 An analyte-induced adaptation of a DCL was used to identify the respective analyte when the library was composed of compounds of different colors. The compounds in the libraries were chelating dyemetal salt complexes that were able to undergo ligandexchange reactions. For such a sensor, the information about the sample was distributed over the entire UV-vis spectrum, which was a “fingerprint” for the analyte. In conventional sensor arrays, each sensor is independent, and the sample is identified by analysis of several nonspecific sensors with a construction of a characteristic fingerprint.24 In contrast, a DCL sensor consisted of compounds that were connected by exchange reactions.338 The addition of a target molecule that selectively interacted with some members of the library caused a re-equilibration of the whole library, and this adaptation was used to identify library members with a high affinity for the respective target molecule. Crego-Calama and co-workers explored the surfaceconfinement effects in sensing self-assembled monolayers (SAMs) that were developed for determinations of cations and anions.344-348 The key advantages of SAMs for surfaceconfined sensing are in the possibility of the introduction of additional chelating effects from the preorganization of the surface-immobilized reagent. The binding groups and fluorophore molecules in the SAM are in close enough proximity, thus the binding group-analyte interaction is communicated to the fluorophore, resulting in a modulation of the fluorescence intensity (see Figure 27A). Sensing SAMs were made on glass tailored with two building blocks. One block included small molecules that supplied different functionalities acting as binding groups (e.g., ureas, amides, thioureas, sulfonamides). Another block included fluorescent dye molecules for reporting the recognition event. The properties of the layer were a result of the combination of the nature of the different binding groups, the fluorescent dye molecule, and surface functionalization. To make combinatorial libraries of sensing SAMs, glass slides were functionalized with a SAM of N-[3-(trimethoxysilyl)propyl]ethylenediamine, followed by the sequential covalent attachment of fluorophore molecules and small binding groups. These libraries were produced by two methods that were (1) a solutionbased procedure with a sequential dipping and (2) a microcontact printing. For determination of anions, fluorophores in one of these combinatorial libraries344 were lissamine rhodamine B sulfonyl chloride and tetramethylrhodamine5-(and 6)-isothiocyanate, whereas the anion binding functional groups were amino, amide, sulfonamide, urea, and thiourea (see Figure 27B). From the fluorescence response of this combinatorial library to 10-4 M acetonitrile solutions of tetrabutylammonium salts of HSO4-, NO3-, H2PO4-, and AcO- anions (see Figure 27C), a general trend in the response magnitude between the library elements based on different fluorophores was discovered. This trend was attributed to the differences in the attachment point functionality of two fluorophores. From these experiments, the variables of fluorophores, binding groups, and their substituents that affected the sensitivity and selectivity of the sensing SAM were applied to develop more understanding

Combinatorial Development of Sensing Materials Chemical Reviews, 2008, Vol. 108, No. 2 791

Figure 26. Chemoinformatic strategy for a multiparameter analysis of combinatorial libraries of styryl dyes to predict their spectral properties and analyte-binding ability: (A) schematic of dye synthesis by the condensation of 41 aldehydes (building block A) and 14 pyridinium salts (building block B); (B) localization distribution of the organelle specific styryl dyes [(#) nuclear, (/) nucleolar, ([) mitochondria, (b) cytosolic, (×) endoplasmic reticular, (9) vesicular, and (2) granular]. Reprinted with permission from ref 342. Copyright 2003 American Chemical Society.

792 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 27. Combinatorial approach for sensing SAMs: (A) schematic representation of an analyte (purple ovals, top schematic) interaction with the SAM due to its coordinating properties followed by reporting of the interaction by the fluorophore [binding groups are depicted by octahedrons, and fluorescent groups are depicted by orange spheres (bottom schematic)]; (B) combinatorial SAM compositions of the anion-sensing library; (C) normalized fluorescence intensity of sensing SAMs in the presence of 10-4 M solutions of HSO4-, NO3-, H2PO4-, and AcO- as tetrabutylammonium salts in acetonitrile. Reprinted with permission from ref 344. Copyright 2004 American Chemical Society.

on the origin of selectivity and response magnitude for ion sensing.344 In the future, combinatorial techniques promise to provide experimental data to generate descriptors for further computational design of diverse indicator molecules to predict not only their optical properties but also their analyte-binding selectivity in the presence of interferences.

4.2. Polymeric Compositions Applications of formulated compositions of analyte responsive reagents immobilized onto a solid support go back to ancient times when the Romans used papyrus impregnated with an extract of acorns for selective colorimetric determinations of iron sulfate and copper sulfate.349 In those and many other more recent applications (for example, in the detection of acids and alkalis with a litmus paper by Lewis in 1767) only qualitative determinations were performed, yet they provided critical analytical information about the presence or absence of analytes of interest in a sample. Modern sensing based on immobilized reagents is the most widely used sensing approach because of the diversity of analytes that can be detected and the diversity of transduction principles involved in detection that include radiant, electrical, mechanical, and thermal energies. The diversity of requirements for such sensors can be quite broad and can range from the need for the long-term autonomous monitoring in remote locations350 without sensor film degradation

to extra-stable and accurate performance of sensors in bioprocess control351,352 over only several weeks of operation, to nonleaching of reagents from an immobilization matrix in the in vivo sensors for analysis of blood parameters,353 to selective determination of a wide variety of ionic species in environmental and industrial water,354-356 to extremely sensitive sensing of contaminants in drinking water.357 For these and many other applications, the generally accepted strategy is to employ formulated sensing compositions. The key components of formulated sensing films include analyte-responsive reagents, polymer matrices, functional additives, and common solvents. An extensive optimization is required to identify sensor formulations with best sensor performance (e.g., largest sensor sensitivity, smallest response to interferences, shortest response time, enhanced stability). There can be easily five or six functional additives in formulated sensing films for optical358,359 and potentiometric360 sensors, not taking into an account a solvent, which in turn can be binary or even ternary to ensure the solubility of all components. Figure 28 depicts needed types of formulation components to tailor dynamic range, selectivity, accuracy, sensitivity, long-term stability, spectral response, and response time of formulated materials. Often, optimization of formulated sensor materials requires evaluation of numerous polymeric matrices or multiple additives at their multiple concentrations and ratios. Of course, general knowledge exists for the design of formulated sensing films.

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 793

Figure 28. Diversity of optimization parameters during the preparation of formulated polymeric sensing materials.

However, rational design of formulated sensing materials is often limited by poor solubility and compatibility of formulation components, photobleaching, and other practical issues.45,361-363 These significant knowledge gaps prevent a more efficient application of rational approaches of development of formulated sensor materials. Walt and co-workers44 pioneered in situ polymerization of combinations of starting monomers and an analyteresponsive indicator and employed this approach to fabricate discrete and gradient arrays of sensing films with responses that were not simply related to the proportion of the starting materials. Bright and co-workers364,365 developed an approach to use a high-speed printer to rapidly produce and screen biodegradable polymer and xerogel-based formulations for biosensors. Robotic-based approaches for fabrication and testing of libraries of solvent-cast formulated sensing film compositions for gas and water analysis have been developed by Wolfbeis and co-workers45,46 and Potyrailo and coworkers.48,143,148,366 Optimization of concentrations of formulation components can require significant effort because of the nonlinear relationship between additive concentration and sensor response.44,367-373 For detailed optimization of formulated sensor materials, Potyrailo and co-workers374 used concentration- and thickness-gradient sensor material libraries. The one-, two-, and three-component composition gradients were made by flow-coating individual liquid formulations onto a flat substrate and allowing them to merge under diffusion control when still containing solvents.375 These gradient films were applied for optimization of sensor material formulations for the analysis of ionic and gaseous species.143,375 A very low reagent concentration in the film is expected to produce only a small signal change. The small signal change is also expected when the reagent concentration is too high. Thus, the optimal reagent concentration will depend on the analyte concentration and activity of the immobilized reagent. Because the activity of reagents upon immobilization is too difficult to quantitatively predict, an optimization of reagent concentration is typically performed. To illustrate this approach, concentration optimization of a colorimetric reagent was performed in a polymer film for the detection of trace concentrations of chlorine in water. A

Figure 29. Colorimetric formulated poly(2-hydroxyethyl methacrylate) hydrogel sensing films for detection of ions in water: (A) concentration optimization of a colorimetric chlorine-responsive reagent in a formulated polymeric gradient sensing film [exposure, 1 ppm of chlorine; (inset) spectrum of the optimal dye concentration in the film]; (B) conventional optical disks with printed sensing regions with optimized concentration of the reagent immobilized into sensing films for detection of chlorine in water [(inset) closeup of the screen-printed sensing film].

concentration gradient of a near-infrared cyanine dye was formed in a poly(2-hydroxyethyl methacrylate) hydrogel sensing film. The optical absorption profile A0(x) was obtained before analyte exposure to map the reagent concentration gradient in the film. A subsequent scanning across the gradient after the analyte exposure (1 ppm of chlorine) resulted in the determination of the optical response profile AE(x). The difference in responses, ∆A(x) ) A0(x) - AE(x), revealed the spatial location of the optimal concentration of the reagent that produced the largest signal change (see Figure 29A). Unlike traditional concentration optimization approaches,372,373 the new method provided opportunities for time-affordable optimization of the concentration of multiple formulation components using concentration gradients. Sensing films with the optimized concentration of the cyanine dye for chlorine determinations in industrial water were further screen-printed as a part of sensing arrays355 onto conventional optical disks as shown in Figure 29B. The quantitative readout of changes in film absorbance was performed in a conventional optical disk drive in a recently developed laboratory-on-a-disk system.354-356 With this system, it was possible to quantify signal changes from sensing films with dimensions down to several tens of micrometers, limited only by the size of the laser beam on the disk surface.354,356 The effect of the thickness of sensor films on the stability of the response in water to ionic species has been also evaluated using gradient-thickness-sensing films.143 Sensor reagent stability in a polymer matrix upon water exposure is one of the key requirements. In the gradient sensor arrays,

794 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 30. Application of gradient-thickness sensor film arrays for evaluation of reagent leaching kinetics: (A) variable film thickness; (B) reagent-leaching kinetics at pH 10. Reprinted with permission from ref 143. Copyright 2005 American Institute of Physics.

the thickness of sensor films was determined from the absorbance of the film-incorporated bromothymol blue reagent (Figure 30A). When these arrays were further exposed to a pH 10 buffer (Figure 30B), an “activation” period was observed before leaching of the reagent from the polymer matrix became detected as the absorbance decrease. This apparent activation period was roughly proportional to the film thickness. However, the leaching rate was independent of the film thickness as indicated by the same slopes of the response curves at 3-9.5 h exposure time periods. In sensing materials based on solid polymer electrolytes [e.g., polyethylene oxide,376 poly(dimethyldiallylammonium chloride),377 Nafion,378 etc.], the conductivity depends on ionic mobility rather than electron mobility. Modifications of selectivity patterns of this type of sensing materials in response to different analytes have been achieved by formulating them with different functional additives. For example, Nafion films have been formulated with hydrogels,379 ionic liquids,380 salts,381 surfactants,382 and many other additives. Potyrailo and Morris378,383 recently demonstrated an approach for multianalyte sensing using a single conventional radio frequency identification (RFID) tag that has been adapted for chemical sensing. Unlike other approaches of using RFID sensors, where a special tag is designed at a much higher cost, conventional RFID tags (<$1) were utilized and simply coated with chemically sensitive films (see Figure 31A). In such RFID chemical sensors, both the digital tag ID (see Figure 31B) and the complex impedance of the resonant circuit of the RFID antenna were measured.378,384 The measured digital ID provided information about the sensor and the object onto which the sensor was attached. By measuring simultaneously several parameters of the complex impedance from a Nafion-coated resonant LC circuit of the RFID sensor and applying multivariate statistical analysis methods, the identification and quantitation of several vapors of interest with a single RFID sensor were demonstrated with parts per billion vapor detection limits.378 To induce an additional selectivity in vapor response and to study long-term stability, Potyrailo and co-workers385

Potyrailo and Mirsky

employed several different phthalic acid derivatives as additives in Nafion. Phthalate plasticizers in different polymeric films were used previously to induce the diversity in vapor response.386-389 Potyrailo and co-workers formulated five different phthalate plasticizers in Nafion at 10% vol, and the films were deposited onto RFID sensors. An array of 48 RFID sensors was formed390 as shown in Figure 31C and further arranged on a gradient temperature heater to generate a linear temperature gradient from 40 to 140 °C. The interrogation of RFID sensors in the array was done with a single transmitter (pick-up antenna) coil positioned on an X-Y translation stage and connected to a network analyzer. After temperature annealing in air, the differential impedance response after and before the annealing from each sensor was calculated (see Figure 31D). The smallest response was associated with the most stable film composition after annealing. This demonstrated approach provided the capability of using conventional passive RFID tags as high-performance transducers for rapid aging studies of sensing materials. As pointed out earlier by Potyrailo and co-workers,140 in general, the increase of the level of environmental stress may be problematic if the correlation with traditional test methods is lost. To avoid this situation, it will be critical to plan the detailed accelerated-aging highthroughput experiments with positive and negative controls.

4.3. Homo- and Copolymers At Pittcon of 1963, King reported a new gas-sensing technique with thickness shear mode (TSM) resonators (aka quartz crystal microbalances, QCMs) coated with a variety of sorbing materials including several types of hydrophilic polymers.391 This study has inspired many generations of scientists to explore the use of different films on acousticwave resonant and cantilever devices to develop sensors for practical gas- and liquid-phase analysis. Numerous reviews and books provide a comprehensive coverage of the current state of the art in such transducers, (see, for example, refs 392 and 393). In gas sensing with polymeric materials, polymer-analyte interaction mechanisms include dispersion, dipole induction, dipole orientation, and hydrogen bonding.394,395 The response of polymeric matrices was shown to be stable over several years.56,396,397 Although there have been several models developed to calculate polymer responses,398-402 the most widely employed model is based on the linear solvation energy relationships (LSER).398,399 The LSER method systematically explores the role of vapor-solubility properties and fundamental interactions in selectivity and diversity of sensing polymers. The LSER modeling was initially performed using ∼2000 compounds with the goal of understanding the development principles of more effective stationary phases for gas chromatography.399,403 The results were further expanded into sensor applications.398,399 The LSER method calculates polymer/gas partition coefficients as a linear combination of terms that represent several molecular types of interactions,398,399

log K ) c + rR2 + sπ2H + aΣR2H + bΣβ2H + l log L16 (7) where R2, π2H, R2H, β2H, and log L16 are parameters that characterize the solubility properties of the vapor in a sorbent polymer, coefficients r, s, a, b, and l are the corresponding sorbent polymer parameters, and c is the regression constant.

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 795

Figure 31. Combinatorial screening of stability of sensing film compositions using passive RFID sensors: (A) strategy for adaptation of conventional passive RFID tags for chemical sensing by deposition of a sensing film onto the resonant circuit of the RFID antenna [(inset) analyte-induced changes in the film material affect film resistance (RF) and capacitance (CF) between the antenna turns]; (B) photo of a typical employed RFID sensor (memory chip type ) I*CODE1, memory chip ID ) 0900 000 457D 5E12); (C) photo of an array of 48 RFID sensors prepared for temperature-gradient evaluations of response of Nafion/phthalates compositions; (D) results of temperature annealing of 48-film library in air plotted as the differential impedance response Zp after and before the annealing from each sensor as a function of annealing temperature and material composition. Nafion sensing film compositions: 1, control without plasticizer; 2, dimethyl phthalate; 3, butyl benzyl phthalate; 4, di-(2-ethylhexyl) phthalate; 5, dicapryl phthalate; 6, ditridecyl phthalate.

Figure 32. Approach for high-throughput evaluation of sensing materials for field applications: (A) setup schematic of a 24-channel TSM sensor array for gas-sorption evaluation of sorbing polymeric films (reprinted with permission from ref 412; copyright 2004 American Institute of Physics); (B) photo of 24 sensor crystals (including 2 reference sealed crystals) in a gas flow cell; (C) multilevel high-throughput materials screening strategy of sensing materials.

Experimentally, the polymer/gas partition coefficients are measured as the ratio of the analyte concentration in the polymer sensor film to the analyte concentration outside the film.404,405 The LSER method has proven to be very effective at correlating the polymer-vapor sorption properties with R > 0.95 correlation between the predicted and experimentally obtained partition coefficients for single vapors.406,407 Potyrailo and co-workers applied the LSER method as a guide to select a combination of available polymers to construct a TSM sensor array for the determination of organic solvent vapors in the headspace above groundwater.408 Field testing of the sensor system409 demonstrated that its detection limit with available polymers was too high (several parts per million) to meet the requirements for the detection of groundwater contaminants. Potyrailo and Sivavec have found a new polymer for sensing (silicone block polyimide) that had the partition coefficient >200,000 to parts per billion concentrations of trichloroethylene (TCE) and provided at

least a 100 times more sensitive response for the detection of chlorinated organic solvent vapors than other known polymers.56,410 To provide discrimination between analytes and interfering species, polymer modifications have been introduced to the base silicone block polyimide polymer. For screening of sensing materials candidates, a 24-channel TSM sensor system was built that matched a 6 × 4 microtiter wellplate format (Figure 32A,B). A comprehensive materials screening was performed with three levels411,412 as shown in Figure 32C. In the primary (discovery) screen, materials were exposed to a single analyte concentration. In the secondary (focused) screen, the best materials subset was exposed to analytes and interferences. Finally, in the tertiary screen, remaining materials were tested under conditions mimicking the long-term application. Although all of the screens were valuable, the tertiary screen provided the most intriguing data because aging of base polymers and copolymers is difficult or impossible to model.41 From the tertiary screening, the decrease in materials response to the nonpolar analyte vapors

796 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 33. Application of the 24-channel TSM sensor array system for mapping of solvent resistance of polycarbonate copolymers: (A) general view of the screening system with a 6 × 4 microtiter wellplate positioned below the sensor array (reprinted with permission from ref 150; copyright 2004 American Chemical Society); (B) example of property/composition mapping of solvent resistance of polycarbonate copolymers in tetrahydrofuran (reprinted with permission from ref 414; copyright 2006 American Chemical Society). Numbers in the contour lines are normalized sensor frequency shift values (hertz per milligram of polymer in a well of the microtiter wellplate).

and the increase in response to a polar interference vapor were quantified. This 24-channel TSM sensor array system was further applied for the high-throughput screening of solvent resistance of a family of polycarbonate copolymers prepared from the reaction of bisphenol A (BPA), hydroquinone (HQ), and resorcinol (RS) with the goal to use these copolymers as solvent-resistant supports for deposition of solvent-containing sensing formulations.413 During the periodic exposure of the TSM crystals to polymer/solvent combinations (Figure 33A150), the mass increase of the crystal was determined, which was proportional to the amount of polymer dissolved and deposited onto the sensor from a polymer solution. The high mass sensitivity of the resonant TSM sensors (10 ng), use of only a minute volume of a solvent (2 mL), and parallel operation (matching a layout of the available 24 microtiter wellplates) made this system a good fit with available polymer combinatorial synthesis equipment. These parallel determinations of polymer-solvent interactions also eliminated errors associated with serial determinations. The data were further mined to construct detailed solvent-resistance maps of polycarbonate copolymers and to determine quantitative structure-property relationships (see Figure 33B414). The application of this sensor-based polymer-screening system provided a lot of stimulating data that were difficult to obtain using conventional one-sample-at-a-time approaches. To eliminate the direct wiring of individual TSM sensors and to permit materials evaluation in environments where wiring is not desirable or adds a prohibitively complex design, Potyrailo and Morris415 developed a wireless TSM sensor array system in which each sensor resonator was coupled to a receiver antenna coil and an array of these coils was scanned with a transmitter coil (Figure 34A). Using this sensor wireless system, sensing materials can be screened for their gas sorption properties, analyte binding in liquids, and changes in chemical and physical properties upon weathering and aging tests. The applicability of the wireless sensor materials screening approach has been demonstrated for the rapid evaluation of the effects of conditioning of

polymeric sensing films at different temperatures on the vapor-response patterns. In one set of high-throughput screening experiments, Nafion film-aging effects on the selectivity pattern were studied. Evaluation of this and many other polymeric sensing materials lacks the detailed studies on the change of the chemical selectivity patterns as a function of temperature conditioning and aging. Conditioning of Nafion-coated resonators was performed at 22, 90, and 125 °C for 12 h. Temperature-conditioned sensing films were exposed to water (H2O), ethanol (EtOH), and acetonitrile (ACN) vapors, all at concentrations (partial pressures) ranging from 0 to 0.1 of the saturated vapor pressure P0, and their responses were measured using a network analyzer. The collected data were processed using PCA as shown in Figure 34B-D. It was found that conditioning of sensing films at 125 °C compared to room temperature conditioning provided (1) an improvement in the linearity in response to EtOH and ACN vapors, (2) an increase in relative response to ACN, and (3) a 10-fold increase of the contribution to principal component 2. The latter point signifies an improvement in the discrimination ability between different vapors upon conditioning of the sensing material at 125 °C. This new knowledge will be critical in designing sensors for practical applications when a need exists to preserve sensor response selectivity over long exploitation times or when there is a temperature cycling for an accelerated sensor-film recovery after vapor exposure. In the area of homo- and copolymers, combinatorial technologies have been also employed for the development of sensing materials for chemical analysis in liquids. Using biocatalytic polymer synthesis, Dordick and co-workers416,417 created a 15-member library of polyphenol polymers. Although the general knowledge exists on complexation of polyphenol polymers with metal ions,416 the origin of the selective response to different metal ions is less known, and when a more selective polymer response is desired, it is more difficult to predict such responses quantitatively. Thus, this only general knowledge inspired a detailed exploration of combinations of a variety of starting monomers with an

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 797

Figure 34. Concept for wireless high-throughput screening of materials properties using thickness shear mode resonators: (A) configuration of a wireless proximity resonant sensor array system for high-throughput screening of sensing materials with a single transmitter coil that scans across an array of receiver coils attached to resonant sensors; (B-D) evaluation of selectivity of Nafion sensing films to several vapors after conditioning at different temperatures [(B) 22, (C) 90, and (D) 125 °C]. Vapors: H2O (water), EtOH (ethanol), and ACN (acetonitrile). Concentrations of vapors are 0, 0.02, 0.04, 0.07, and 0.10 P/P0. Arrows indicate the increase of concentrations of each vapor. Reprinted with permission from ref 415. Copyright 2007 American Institute of Physics.

expectation to find polymeric materials for high-selectivity determination of metal ions of environmental importance. An array of 15 phenolic homopolymers and copolymers was combinatorially generated from 5 phenolic monomers by peroxidase-catalyzed oxidative polymerization. These polymers were screened for their intrinsic UV fluorescence (322 nm excitation) response to Fe3+, Cu2+, Co2+, and Ni2+ ions at 0.2-1 mM concentrations.416,417 As shown in Figure 35, the combinatorial approach generated a diverse metal ion response of new polymers from a limited number of phenolic monomers. New polymers demonstrated preferential selectivity to certain metal ions yet with some cross-sensitivity.

4.4. Conjugated Polymers Conjugated polymers are organic polymers with conjugated bonds between single monomers that have electrical, electronic, magnetic, and optical properties similar to the properties of metals and semiconductors while preserving the processability and other properties of conventional polymers.418 Examples of conjugated polymers include polyparaphenylene, polyphenylenevinylene, polypyrrole, polyacetylene, polythiophene, polyfuran, polyheteroaromatic vinylenes, polyaniline, and numerous derivatives of these compounds. Conjugated polymers have found their wide applicability as unique sensing materials28,419-423 because recognition and transduction can be performed within the same chemical moiety. This feature is complementary to immobilization of individual recognition and transducing additives into sensing films based on nonconjugated polymeric compositions. Current efforts in the development of conjugated polymers as sensing materials include introduction of diverse receptors into polymers, copolymerization with monomers that

have desired receptor groups, incorporation of biological receptors, and molecularly imprinted polymerization. Several roles of conjugated polymers in sensing films include catalytic layers, redox mediators, molecular recognition receptors, analyte preconcentrators, and electrical and optical transducers.424-429 Properties of conjugated polymers for sensing applications depend on many factors summarized in Figure 36. Synthesis of conjugated polymers is typically realized either by the addition of oxidizing agents or by electrochemical oxidation.430,431 The type and physical conditions of polymerization, choice of solvent, counterions, and the presence and type of additional dopants affect final properties.432 Upon careful selection of the factors shown in Figure 36, these polymers can recognize, transduce, and, sometimes, amplify chemical or biological information into an optical or electrical signal.433,434 Specifics of immobilization of receptor biomolecules include adsorption onto electropolymerized films, entrapment during the electropolymerization process, covalent binding on electrogenerated polymers, and anchoring by affinity interactions between biomolecules and conjugated polymers.435 To effectively evaluate the complexity of the composition and process parameters space, combinatorial approaches have been applied for screening of conjugated polymers for numerous applications.436-438 Manipulation of reaction components during combinatorial synthesis of conjugated polymers include microfluidic and liquid-dispensing approaches. Microfluidic and microflow systems are well established in combinatorial chemistry439-442 because of several important features they provide that include high efficiency, short time scales, safe conditions, and low amounts of waste. Microfluidic systems for combinatorial materials science have been

798 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 35. Results of combinatorial synthesis of a 15-member array of metal-ion sensitive homo- and copolymers from five phenolic monomers 1-5; fluorescence responses of 15 materials to 4 metal ions: (A) 0.2 mM Fe3+; (B) 1.0 mM Cu2+; (C) 1.0 mM Co2+; (D) 1.0 mM Ni2+. Phenolic monomers: 1, p-cresol; 2, p-phenylphenol; 3, p-methoxyphenol; 4, p-hydroxyphenylacetic acid; 5, p-hydroxybenzoic acid. Results were calculated from data reported by Dordick and co-workers.417

previously applied for organic phase synthetic reactions, formation of surface composition gradients, and many other applications.366,443-445 Combinatorial synthesis of polypyrrole-based polymers reported by Xiang and LaVan446 involved a microfluidic system for gradient mixing of reagents and subsequent parallel electropolymerization in multichannels (see Figure 37). The microfluidic electrochemical polymer reactor generated stable concentration gradients of solutions of pyrrole and the sodium salt of polystyrenesulfonate (PSS) acid and other additives that resulted in controlled deposition of polypyrrole films with different thicknesses (see Figure 38). Doping of conjugated polymers [e.g., polypyrrole, polyaniline, poly(3,4-ethylenedioxythiophene)] with PSS is known to provide an improvement in chemical sensing ability via self-doping.447-449 Multiple polypyrrole compositions were synthesized in 11 parallel fluidic channels over 4 electrodes per channel. However, with the gradient mixer it was possible to generate multiple but fixed compositions, and such a device did not allow any arbitrary ratio of reactants when multiple constituents were used. By using immiscible liquid plugs that form a droplet-based microfluidic system,445 Xiang and LaVan demonstrated that it was possible to produce arbitrary compositions from more than two starting constitu-

Figure 36. Diversity of optimization parameters during the preparation of conjugated polymer sensing materials.

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 799

Figure 37. Design of a microfluidic system for combinatorial electrochemical synthesis of conjugated polymers: (A) general schematic of the system (the microfluidic channels are shown in blue and electrodes in red); (B) gradient generator [two different species A (red) and B (blue) were injected into the gradient generator from the top]; (C) parallel electrochemical synthesis reactor. Reprinted with permission from ref 446. Copyright 2006 The Institute of Electrical and Electronics Engineers, Inc.

Figure 38. Combinatorial polymer deposition from the mixture of pyrrole and polystyrene sulfonic acid on platinum anodes placed at distances of 500, 100, 1000, and 2000 µm from the platinum cathode. Reprinted with permission from ref 446. Copyright 2006 The Institute of Electrical and Electronics Engineers, Inc.

ents.446 Immiscible liquid plugs can provide rapid mixing of components, control of the timing of reactions, control of interfacial properties, and the ability to synthesize and transport solid reagents and products.445 A developed prototype was composed of a plug microgenerator and a multiwell polymer microreactor. A microarray of as many as 100 plugs was generated, and redox reactions of monomers and dopants were conducted by properly activating microelectrodes installed in each microwell. By varying the ratio of input constituents, such a device was able to form droplets with any arbitrary composition and to deliver them within the immiscible carrier fluid to the prescribed locations.

Mirsky and co-workers developed an approach based on electrically addressed polymerization that did not require a controlled liquid flow or delivery to each individual polymerization region.151,152,450,451 Instead, the addressable electrochemical polymer synthesis on the defined electrodes was performed by controlling the electrical potential of the given electrode group of the electrode array while all other electrode groups were kept below the required polymerization potential. To deposit different materials onto sensing regions in the electrode array, polymerization solutions in the electropolymerization microcell were simply changed between the polymerizations. For combinatorial electrochemical polymerizations, Mirsky and co-workers developed a screening system that utilized a chip with 96 groups of electrodes. Each electrode group had a size of 400 × 400 µm2 and consisted of four electrodes designed for two- and four-point measurements. The 96-element chip operated in an electropolymerization microcell through an electronic multiplexer. The delivery of polymerization solutions and rinsing between individual polymerizations was performed with an automated dosing station. To provide an automated high-throughput screening capability of chemosensitive properties of synthesized materials, the system was integrated into the combinatorial information workflow with the automated data analysis of analyte exposures of sensing films (see Figure 39). The key aspects of the workflow are summarized in Figure 39A.452 The developed automated procedure to test polymerized conjugated sensing films included two analyte-pulsed exposures at one concentration and a sequence of analytepulsed exposures at increasing concentrations. An automated data analysis of materials responses included calculated absolute and relative analytical sensitivity, response and recovery rate, recovery efficiency, reversibility, reproduc-

800 Chemical Reviews, 2008, Vol. 108, No. 2

Figure 39. Combinatorial electrochemical polymerizations of conjugated sensing polymers: (A) combinatorial information workflow with the automated data analysis of analyte exposures of the sensing films (in the photo of 96-element combinatorial library of sensing films, dark regions in the center of the combinatorial library are formed by synthesized polymers) (reprinted with permission from ref 452; copyright 2005 American Chemical Society); (B) example of user interface demonstrating automatically computed relative sensitivity of response of combinatorial library of sensing films to 3.5 ppm of HCl gas as a function of polymerization charge. The leftmost region of the interface panel is the menu for selection of analyzed parameters for display. See text for details.

ibility, binding constant (for sensor materials that obey Langmuir’s adsorption isotherm), and response linearity (for sensor materials that obey Henry’s adsorption isotherm).152 This system was employed for screening a wide variety of sensing materials. In screening of chemosensitive properties of polyanilines, copolymers of anilines, and aniline derivatives, combinatorial libraries were tested for their response to HCl gas.151,450 The measured electrical resistance was detected simultaneously by two- and four-point techniques.451,453 Figure 39B illustrates an example of data analysis interface with a user-selected display of calculated

Potyrailo and Mirsky

parameters. As seen in Figure 39B, the relative sensitivity of response (normalized to initial conductance) of polymerized sensing materials had its optimal (maximum) value at the polymerization charge of ∼0.5 mC, further employed for more detailed studies. A summary of exemplary screening results for different binary copolymers is presented in Figure 40. An introduction of nonconductive monomers into polymer decreased the polymer conductance and therefore decreased the difference between conductive and insulating polymer states. This caused a decrease of the absolute sensitivity (Figure 40A). Normalization to the polymer conductance without analyte exposure compensated this effect and demonstrated that the polymer synthesized from the mixture of anthranilic acid and aniline possessed the highest relative sensitivity (Figure 40B). This effect may be explained by the strong dependence of polymer conductance on the defect number in polymer chains. In comparison with pure polyaniline, this copolymer had better recovery efficiency but a slower response time (Figure 40C,D). The developed high-throughput screening system was capable of reliable ranking of sensing materials and required only ∼20 min of manual interactions with the system and ∼14 h of computer-controlled combinatorial screening compared to ∼2 weeks of laboratory work using traditional electrochemical polymer synthesis and materials evaluation.151 Mirsky and co-workers used the developed system also for optimization of enzymatic biosensors for glucose with electrocatalytic transduction.151 Developed sensing films for glucose were multilayer systems that contained an electrocatalyst, an enzyme, and a conjugated polymer. To form glucose-sensitive films, a layer of variable thickness of Prussian Blue (an electrocatalyst for decomposition of hydrogen peroxide) was electrochemically synthesized on the electrode groups in the array. Then the films were exposed to hydrogen peroxide, and their electrochemical response was compared to optimize the thickness of the catalyst layer. Next, a solution of pyrrole-containing glucose oxidase was electropolymerized with variable thickness on the electrode groups in the array, and the electrocatalytic currents observed upon glucose additions were analyzed to determine the best sensing film candidates. This convenient layout of combinatorial libraries in the form of small amounts of polymers of a few micrograms on the surface of the electrode array allowed not only characterization but also further storage and organization of banks of sensing materials for further investigations. Schuhmann and co-workers developed a general purpose setup for combinatorial electrochemistry by combining a liquid-dispensing mechanical robot and electrochemical system.454 The setup operated with one electrode set (consisting of classical three-electrode configuration), moving it between different cells or with an eight-electrode set providing simultaneous eight-channel measurements compatible with a microtiter wellplate layout. Such a system was applied successfully for high-throughput investigation of porphyrins455 and would be certainly very helpful for screening of other types of combinatorial libraries, for example, of metallic nanoparticles.456,457 This system was used by Ba¨uerle and co-workers458 for high-throughput electrochemical characterization of combinatorial libraries of π-conjugated oligothiophenes. Such materials are widely used for optical and electrical detection of gases and metal ions.459-462 A regioregular head-to-tail coupled quater(3-arylthiophene) was

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 801

Figure 40. Selected results of screening of sensing materials for their response to HCl gas: (A) best absolute sensitivity; (B) best relative sensitivity; (C) best response rate; (D) best recovery efficiency, performed by heating. Sensor materials: ANI indicates polyaniline; 4ABA, 3ABSA, 3ABA, and AA indicate polymers synthesized from aniline and 4-aminobenzoic acid, 3-aminobenzenesulfonic acid, 3-aminobenzoic acid, and anthranilic acid, respectively. Gray and black bars are the results obtained by two- and four-point techniques, respectively.

selected as the lead structure (Figure 41A) for combinatorial synthesis to systematically investigate the substitutent influence on the energy levels of the molecular orbitals and to develop structure-property relationships. This oligothiophene was substituted at the 3-position of each thiophene subunit with phenyl groups and at the para positions with one of four groups (CF3, H, CH3, OCH3). These four groups have different electronic nature expected to affect the electronic structure of the quaterthiophenes without greatly changing the overall geometry of the molecule. The synthesized 256-oligomer library included all possible permutations of the four diversity elements. The electronic consequences of the oligomer substitution on the relative energy levels of the frontier orbitals were further investigated by automated cyclic voltammetry to obtain the data on the substituent effects on redox potentials. A systematic shift to the higher first oxidation potentials E10 occurred when the substituents had a more pronounced acceptor character as shown in Figure 41B. A similar trend was also found for the second oxidation potentials E20 of the quaterthiophenes. The correlation of the collected redox potentials was further established with the substituent descriptor Σσp+. This descriptor was defined as the sum of the Hammett constants σp+ of the individual substituents and numerically reflected the overall contributions of the individual substituents. The relationship between the first E10 and second E20 oxidation potentials and substituent descriptor Σσp+ is illustrated in Figure 41C. Thus, combinatorial methodologies allowed the synthesis and evaluation of carefully planned libraries of conjugated polymers to deduce structure-property relationships. In the future, combinatorial screening of electrical properties of conjugated polymers can be combined with automated analysis of optical properties to gain more insight into the function of differently doped conjugated polymers. Elec-

tropolymerization can be combined with electrochemical deposition of metals or other types of conductive or nonconductive layers with different functions. Postsynthetic chemical modifications or grafting photopolymerization can be also additional ways for formation of more advanced structures.

4.5. Molecularly Imprinted Polymers The key feature of molecularly imprinted materials is their engineered selectivity of binding to an analyte of interest. These synthetic materials mimic the hypothesis of generation of antibodies by the immune system and trace their origin to the works of Mudd463 and Pauling.464 The early work in molecular imprinting involved silica gels and was done in the 1930 and 1940s by Polyakov in the Soviet Union465 and by Dickey in the United States.466 Wulff and Sarhan were the first to synthesize and characterize molecularly imprinted organic polymers.467 At present, the area of research on molecularly imprinted materials is not only tremendously active but also one of the most advanced in the combinatorial and rational aspects of materials design. A comprehensive database of literature on molecularly imprinted materials is accessible at the homepage of the Society for Molecular Imprinting (www.molecular-imprinting.org). Recent reviews are available in refs 468-477. Synthesis of molecularly imprinted polymers (MIPs) includes several essential steps (Figure 42) such as (1) preparation of a non-covalent complex or covalent conjugate between polymerizable functional monomers and analyte or its analogue, (2) polymerization of these functional monomers, and (3) removal of the analyte or its analogue from the polymer. Thus, MIPs are synthesized using several functional monomers and an analyte molecule (known as a

802 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 42. Application of molecularly imprinted polymerization for development of chemical and biological sensing materials.

Figure 41. Deduction of structure-property relationships using combinatorial synthesis and high-throughput evaluation of quater(3-arylthiophene)s library: (A) structure of quater(3-arylthiophene)s; (B) results from automated cyclic voltammetry determinations of the first oxidation potentials E10 (in the color bar, [a] is not determined because of sample impurity); (C) relationship between the substituent descriptor Σσp+ and the first E10 and second E20 oxidation potential of the combinatorial library of quaterthiophenes. Reprinted with permission from ref 458. Copyright 2001 WileyVCH Publishers.

“template”). During the MIP synthesis, monomers selfassemble and cross-link around the template. The presence of template during the polymerization process of the MIP facilitates the formation of a cavity that closely matches the template. Once the template is removed from the fabricated MIP, the material can be used as a polymeric receptor for

sensing applications with different types of transducers. In a liquid-phase detection, replacement of water by analyte molecules leads to changes of electrical (e.g., capacitance) and/or optical (e.g., refractive index) properties of the polymer layer, which may occur due to analyte binding or as a result of a subsequent osmotic swelling. These changes can be detected by impedance478-481 or surface plasmon resonance279,482,483 measurements. A “gate effect”484,485 accompanying analyte binding to a MIP film can be detected by conductivity measurements. Mechanical changes of the polymer layer (including its mass and acoustic thickness) lead to broad applications of TSM transducers486-488 for sensing with MIP films. Fluorescent sensing can be used for detection of fluorescent analytes in competitive assays489,490 or in fluorescent MIPs.491,492 Electrochemical techniques can be applied for direct detection of electrochemically active analytes or for replacement of electrochemically active markers in competitive assays.493-495 Other detection approaches include colorimetric,496,497 calorimetric,498 and ISFET.499,500 In addition to the use of various sensing platforms, detailed investigations of affinity of MIPs are performed by HPLC. There are two specific aspects for MIP-based receptors for application in chemical sensors. The first aspect is related to the polymer morphology and thickness and is driven by the relatively slow analyte diffusion in MIPs. In conventional 3-D imprinting, the thickness of the employed polymer layers is from several nanometers to about a micrometer. A deposition of thin polymer layers can be performed by spincoating, electropolymerization, and grafting photopolymerization.478-480,501 In many detection techniques the sensor signal can be increased by increasing the sensor surface; this is a motivation for development of MIPs based on immobilized dendrimers or nanoparticles.279,502 The second aspect of MIPs is related to the heterogeneity of their binding sites with only scarce reports on quasi-homogeneous binding sites. This heterogeneity causes broad distribution of the binding constants of different binding sites of the same polymer. This leads to a broader range of measured analyte concentrations, but this advantage is compromised by a less predictable sensor behavior. After analyte removal, some of the analyte molecules remain in the stronger binding sites,

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 803

Figure 44. Workflow for automated analyte-binding evaluation of combinatorial libraries of MIPs.

Figure 43. Diversity of optimization parameters during the preparation of molecularly imprinted polymers.

and only weak binding sites can be occupied during the next analyte exposure. Thus, the binding constant and binding kinetics of such a sensing material will be defined mainly by the weak binding sites. This results in the dependence of apparent sensor affinity on the desorption efficiency. Therefore, the critical requirement for analyte quantitation using MIP materials is the reproducible desorption of analyte. The complex between analyte and polymerizable monomers can be stabilized by cleavable covalent bonds and/or by non-covalent interactions. Advantages and disadvantages of covalent and non-covalent types of complexes were summarized earlier.470 Covalent imprinting was primarily developed by Wulff and co-workers.467,503-505 An example of the functional monomer that is often used for covalent imprinting is phenylboronic acid as a receptor for cis-1,2and cis-1,3 diols, including saccharides, nucleotides, AMP, and NAD(P).506-508 A more sophisticated covalent imprinting typically requires a synthesis of cleavable polymerizable derivative of an analyte and results in the formation of more homogeneous binding sites. Non-covalent imprinting, introduced by Mosbach and coworkers,509 is more flexible but leads to higher heterogeneity of binding sites.510 The non-covalent interactions are provided by hydrogen bonds, ionic bonds, van der Waals forces, and hydrophobic interactions. Higher flexibility and simplicity of the non-covalent approach are, of course, very attractive. However, at the same time this flexibility makes an optimization of sensing materials extremely resource- and time-consuming. Figure 43 illustrates the diversity of parameters involved in the fabrication of MIPs with desired selectivity and capacity. Molecular organization around the template is provided by the non-covalent interactions between the template and the functional and cross-linking monomers. The morphology and selectivity of the resulting MIP are affected by the stoichiometry and concentration of the template and monomers.511 The timing of the phase separation during polymerization that influences the binding properties of the resulting MIP is determined by the polymerization porogen (solvent) and by the physical conditions (temperature, pressure) during polymerization.511-514 A number of successful imprintings of different analytes by

exactly the same mixtures of polymerization monomers were reported,479-481 indicating the possibility of finding a mixture of monomers that can be used for more than one application for molecular imprinting. In 1999, groups of Takeuchi515 and Sellergren516 independently introduced combinatorial technologies into molecularly imprinted polymerization. In these first combinatorial studies, both groups selected herbicides as model analyte substances and validated the high-throughput combinatorial molecularly imprinting technique as a working method for finding optimal conditions of MIP preparation. Following these pioneering studies, the combinatorial screening technology has been improved by decreasing the sample volumes in combinatorial libraries to ∼50 mg, increasing the size of combinatorial libraries up to 60 polymer samples, and application of additional initiation approaches such as thermoinitiation.517 The MIP synthesis was further performed by liquid-handling robots in 96-well microtiter plates with the automated binding detection analyzed by analytical instrumentation designed for combinatorial screening518-523 as shown in Figure 44. Multiple parameters shown in Figure 43 that could be optimized to find an ideal MIP material may lead to a costly and complicated workflow even when high-throughput automated systems are employed. Thus, it may be attractive to optimize not the final MIP product but rather each step of the product preparation. A complex formation between functional monomers and template shown in Figure 42 is the crucial step in the synthesis of effective MIPs. Therefore, recent optimization efforts have been focused on the template-functional monomer complexes. The complex stoichiometry has been evaluated with microcalorimetry.524 Also, 1H NMR was used for the fast evaluation of functional groups of the template that can form bonds with functional monomers.525,526 The Job’s plot of chemical shifts measured for binary mixtures of the template-functional monomer complex demonstrated the 1:1 stoichiometry of the complex.525,526 Both methods belong to simple prescreening techniques that can be used for further design of combinatorial libraries for functional screening. Besides the growing range of combinatorially developed artificial receptors,527 the data mining technologies also have been significantly improved. Simple comparisons of trends and maxima of the most important analytical characteristics such as binding and selectivity have been complemented with the multivariate statistical data analysis.528-531 An application of this approach for development of MIP for sulfonamides is shown in Figure 45. The optimal molar ratio of template,

804 Chemical Reviews, 2008, Vol. 108, No. 2

Potyrailo and Mirsky

Figure 45. Optimization of the template/monomer/cross-linker (T:M:X) ratio of the MIP using a three-level full-factorial design. Reprinted with permission from ref 530. Copyright 2004 Elsevier.

Figure 46. Rational design of MIPs: (A) virtual library of 20 monomers used for in silico optimization of the complex with ephedrine; computed structures of the ephedrine complex with (B) methacrylic acid and (C) hydroxyethyl methacrylate. Reprinted with permission from ref 534. Copyright 2001 The Royal Society of Chemistry.

monomer, and cross-linker (T:M:X) was predicted from statistical analysis and confirmed by a direct experiment.530 Independent multivariate analysis was applied for optimization of MIP for bisphenol A. In addition to the T:M:X ratio, a concentration of initiator, a polymerization porogen (tetrahydrofuran, chloroform, toluene, or acetonitrile) and an initiation procedure (photochemical or thermal) were optimized.529 The goal to find approaches for rational design of MIPs was formulated in 1998.532,533 However, the task was considered to be too complicated.492 The first study to replace experimental screening of molecularly imprinted polymers by screening in silico was performed by Piletsky and coworkers.534 To develop a MIP specific to ephedrine, a virtual

library of 20 functional monomers was developed and computationally evaluated (see Figure 46). Molecular modeling was used to calculate binding energy between the monomers and the template. Although effects of cross-linker and porogen were not considered during the calculations, a good correlation between calculated results and a further HPLC study was obtained. Later, this approach was used by Piletsky and co-workers to design MIPs for creatinine,535 microcystin,536 biotin,537 and carbaryl.538 Table 4 demonstrates a comparison of properties of MIPs optimized by molecular modeling with properties of mono- and polyclonal antibodies.536 Molecular modeling was also implemented by other scientific groups, and MIPs for other analytes were developed.487,539-543

Combinatorial Development of Sensing Materials

Chemical Reviews, 2008, Vol. 108, No. 2 805

Table 4. Comparison of MIPs Based on Methacrylic Acid, MIPs Optimized by Computational Design, and Polyclonal and Monoclonal Antibodies (Adapted from Reference 536)

properties of sensitive materials

MIP optimized by computational design

MIP based on methacrylic acid

monoclonal antibodies

polyclonal antibodies

binding properties sensitivity range, µg/L

reciprocal binding constant, nM 0.1-100

0.3 ( 0.1 0.8-100

0.9 ( 0.1 0.025-5

0.030 ( 0.004 0.05-10

0.50 ( 0.07

cross-reactivity, % Microcystin-YR Nodularin

Microcystin-RR 27 ( 2 22 ( 2

21 ( 1 30 ( 3 36 ( 1

19 ( 1 44 ( 2 18 ( 1

106 ( 1 142 ( 1 73 ( 1

92 ( 2

stability under harsh conditions (remaining affinity of receptors) pH 2 pH 11 10 mM CuSO4

80 °C 80% DMF 100 ( 3 102 ( 3 98 ( 1

89 ( 1 102 ( 2 102 ( 2 118 ( 5 56 ( 2

52 ( 2 97 ( 2 16 ( 1 18 ( 5 24 ( 1

17 ( 5 19 ( 4 94 ( 2 44 ( 7 17 ( 1

9.9 ( 0.2 18 ( 1

Artificial receptors synthesized by molecularly imprinted polymerization were also formed as 2-D structures on solid supports. The development of 2-D molecular imprinting was initially independent from 3-D imprinting. One of the first observations of this effect was reported by Sagiv,544 who described a memory effect and molecular shape recognition for the monolayers of octadecyltrichlorosilane. This effect was further observed for non-cross-linked self-assembled monolayers of alkylthiols on gold.545 Further attempts to stabilize the system resulted in the development of the spreader-bar approach, in which the template molecules were not removed.545-547 The principle of the spreader-bar approach is illustrated in Figure 47A.547 This approach was

evaluated to prepare large amounts of different receptors that can distinguish not only molecules of different compounds547 but also the molecules with different chiralities.548 However, a number of factors complicate an affinity prediction of these spreader-bar structures. First, a design of preparation conditions of mixed monolayers in equilibrium conditions requires information on the affinity of corresponding compounds to the electrode surface,549 which is currently available only for a few compounds. Additionally, mixed monolayers from compounds with strong intermolecular interaction leading to the formation of domains should be excluded. Detailed structural investigation of such systems is time-consuming.293,550 Even for well-characterized mixed monolayers it

Figure 47. Approach for high-throughput development of 2-D imprinting with stabilization by the molecular spreader-bar technique: (A) principle of the spreader-bar technique (reprinted with permission from ref 547; copyright 2003 The Royal Society of Chemistry); (B) schematic of an electrochemical system for parallel investigation of biosensing materials formed by the spreader-bar technique; (C) responses of 2-D imprinted materials (as a scores plot of the first and second principal components) on additions of caffeine (1), uracil (2), adenine (3), cytosine (4), thymine (5), and uric acid (6) (reprinted with permission from ref 547; copyright 2003 The Royal Society of Chemistry).

806 Chemical Reviews, 2008, Vol. 108, No. 2

is not easy to predict their affinity properties. This was the motivation for the development of a system for high-throughput impedance screening of affinity properties. The system for simultaneous capacitance measurements was based on eight lock-in amplifiers with current input as shown in Figure 47B. This approach was used for the development of a sensor array for the determination of purines and pyrimidines using PCA (Figure 47C).547 The true 2-D molecularly imprinted polymers, with polymerization and removal of the template, have been developed mainly for large analytes (e.g., proteins and bacteria) that are difficult to extract from the bulk polymer.488,551-554 Typical technologies of their preparation are very similar to that for 3-D MIPs. For example, Tappura et al. reported 2-D imprinting of morphine555 when the structure of binding sites was optimized by molecular dynamic simulation. Most of the approaches developed for high-throughput screening of 3-D polymers can be adapted for 2-D imprinted materials. In the future, promising approaches for the design of MIPs may be based on a two-step procedure including an initial optimization of the complex content and stoichiometry (by either experimental combinatorial screening of binding or/and theoretical modeling) and a final experimental functional optimization of the imprinted polymer.

5. Summary and Outlook Combinatorial and high-throughput technologies in materials science have been successfully accepted by research groups in academia and governmental laboratories that have overcome the entry barrier of dealing with new emerging aspects in materials research such as automation and robotics, computer programming, informatics, and materials data mining. The main driving forces for combinatorial materials science in industry include broader and more detailed explored materials and process parameters space and faster time to market. Industrial research laboratories working on new catalysts and inorganic luminescent materials were among the first adopters of combinatorial methodologies in industry. The classical example of an effort by Mittasch, who has spent 10 years (over 1900-1909) to conduct 6500 screening experiments with 2500 catalyst candidates to find a catalyst for industrial ammonia synthesis,556 will never happen again because of the availability and affordability of modern tools for high-throughput synthesis and characterization. In the area of sensing materials, reported examples of significant screening efforts are less dramatic, yet also breathtaking. For example, a decade ago, Cammann, Shulga, and co-workers386 reported an “extensive systematic study” of more that 500 compositions to optimize vapor-sensing polymeric materials. Walt and co-workers57 reported screening of over 100 polymer candidates in a search for “their ability to serve as sensing matrices” for solvatochromic reagents. Seitz and co-workers557 investigated the influence of multicomponent compositions on the properties of pHswellable polymers by designing 3 × 3 × 3 × 2 factorial experiments. Clearly, combinatorial technologies have been introduced at the right time to make the search for new materials more intellectually rewarding. Naturally, numerous academic groups that were involved in the development of new sensing materials turned to combinatorial methodologies to speed knowledge discovery.44,45,152,161,219,221,365 From numerous results achieved using combinatorial and high-throughput methods, the most successful have been in

Potyrailo and Mirsky

the areas of molecular imprinting, polymeric compositions, catalytic metals for field-effect devices, and metal oxides for conductometric sensors. In those materials, the desired selectivity and sensitivity have been achieved by the exploration of multidimensional chemical composition and process parameters space at a previously unavailable level of detail at a fraction of the time required for conventional one-at-atime experiments. These new tools provided the opportunity for the more challenging, yet more rewarding, explorations that previously were too time-consuming to pursue. Future advances in combinatorial development of sensing materials will be related to several key remaining unmet needs that prevent researchers from having a complete combinatorial workflow and to “analyze in a day what is made in a day”.122 First, new fabrication methods of combinatorial libraries of sensing materials will be implemented ranging from those adapted from other materials synthesis and fabrication approaches558,559 to those developed specifically for sensing applications.356 Second, although the evaluation of performance properties of sensing material has been automated and numerous sensing systems have been developed to collect reliable response data from sensing materials, the remaining need is to develop screening tools for high-throughput characterization of intrinsic materials properties to keep up with the rates of performance screening of sensing materials candidates. For example, in the area of conductometric metal oxide sensors, a variety of employed techniques (e.g., Hall, catalytic conversion, and work function measurements, DRIFT spectroscopy) are at different stages in their high-throughput screening capabilities. Third, certain portions of the data management aspects of the combinatorial workflow are still under development as summarized in Table 2. However, over the past several years, there have been a growing number of reports on data mining in sensing materials.232,414,528,531 “Searching for a needle in the haystack” was popular in the early days of combinatorial materials science.59,560,561 At present, it has been realized that screening of the whole materials and process parameters space is still too costly and time prohibitive even with the availability of existing tools. Instead, designing the highthroughput experiments to discover relevant descriptors will become more attractive.562 Fourth, predictive models of behavior of sensing materials under realistic conditions over long periods of time are needed. These modeling efforts will require inputs not only from screening of the performance and intrinsic properties of sensing materials but also from screening of the effects of interfaces between sensing materials and transducers. This review has attempted to critically analyze the benefits of combinatorial technologies from the standpoint of practitioners of these tools. Perhaps the best response to one’s possible skeptical arguments that “this is not intellectually satisfying”, “this is not science”, and “this is too Edisonian” is two observations. The first observation is a quote from a book chapter by Go¨pel and Reinhardt563 published in 1996 before the broad acceptance of combinatorial technologies into materials science. Go¨pel and Reinhardt mentioned “...it is surprising that no sensor group has so far screened systematically the many well-established metal oxide based catalysts for their potential use as sensor materials. On the other hand, it is surprising that only a few catalysis groups make use of the possibility of characterizing their catalysts

Combinatorial Development of Sensing Materials

by complementary monitoring their sensor properties.” The second observation is that 10 years later, the multidisciplinary essence of combinatorial technologies has brought together sensor and catalysis groups221,230-232, 564 and many other diverse research groups and has affected researchers as well. At present, an effective combinatorial scientist acquires skills as diverse as experimental planning, automated synthesis, basics of high-throughput materials characterization, chemometrics, and data mining. These new skills can be now obtained through the growing network of practitioners and through the new generation of scientists educated across the world in combinatorial methodologies. Combinatorial and high-throughput experimentation was able to bring together several previously disjointed disciplines and to combine valuable complementary attributes from each of them into a new scientific approach.

6. Acknowledgments We gratefully acknowledge GE components for support of our combinatorial sensor research. We thank J. Cui, W. Morris, A. Pris, N. V. Roznyatovskaya, C. Surman, and O. S. Wolfbeis for fruitful discussions; R. Oudt for the help with figures; and T. Leib, G. Chambers, W. Flanagan, and A. Linsebigler for support and encouragement.

7. References (1) Hulanicki, A.; Glab, S.; Ingman, F. Pure Appl. Chem. 1991, 63, 1247. (2) Wolfbeis, O. S., Ed. Fiber Optic Chemical Sensors and Biosensors; CRC Press: Boca Raton, FL, 1991. (3) Taylor, R. F., Schultz, J. S., Eds. Handbook of Chemical and Biological Sensors; IOP Publishing: Bristol, U.K., 1996. (4) Middelhoek, S.; Noorlag, J. W. Sens. Actuators 1981/1982, 2, 29. (5) Janata, J. Principles of Chemical Sensors; Plenum Press: New York, 1989. (6) Hirschfeld, T.; Callis, J. B.; Kowalski, B. R. Science 1984, 226, 312. (7) Hirschfeld, T. Science 1985, 230, 286. (8) Bakker, E.; Bu¨hlmann, P.; Pretsch, E. Chem. ReV. 1997, 97, 3083. (9) Potyrailo, R. A.; Hobbs, S. E.; Hieftje, G. M. Fresenius’ J. Anal. Chem. 1998, 362, 349. (10) Janata, J.; Josowicz, M.; Vanysek, P.; DeVaney, D. M. Anal. Chem. 1998, 70, 179R. (11) Wolfbeis, O. S. Anal. Chem. 2006, 78, 3859. (12) Franke, M. E.; Koplin, T. J.; Simon, U. Small 2006, 2, 36. (13) Potyrailo, R. A. Angew. Chem., Int. Ed. 2006, 45, 702. (14) Bergman, I. Nature 1968, 218, 396. (15) Hardy, E. E.; David, D. J.; Kapany, N. S.; Unterleitner, F. C. Nature 1975, 257, 666. (16) Peterson, J. I.; Vurek, G. G. Science 1984, 224, 123. (17) Barnard, S. M.; Walt, D. R. Nature 1991, 353, 338. (18) Tan, W.; Shi, Z.-Y.; Smith, S.; Birnbaum, D.; Kopelman, R. Science 1992, 258, 778. (19) Charych, D. H.; Nagy, J. O.; Spevak, W.; Bednarski, M. D. Science 1993, 261, 585. (20) Dickinson, T. A.; White, J.; Kauer, J. S.; Walt, D. R. Nature 1996, 382, 697. (21) Holtz, J. H.; Asher, S. A. Nature 1997, 389, 829. (22) Elghanian, R.; Storhoff, J. J.; Mucic, R. C.; Letsinger, R. L.; Mirkin, C. A. Science 1997, 277, 1078. (23) Lin, V. S.-Y.; Motesharei, K.; Dancil, K.-P. S.; Sailor, M. J.; Ghadiri, M. R. Science 1997, 278, 840. (24) Rakow, N. A.; Suslick, K. S. Nature 2000, 406, 710. (25) Kong, J.; Franklin, N. R.; Zhou, C.; Chapline, M. G.; Peng, S.; Cho, K.; Dai, H. Science 2000, 287, 622. (26) Hagleitner, C.; Hierlemann, A.; Lange, D.; Kummer, A.; Kerness, N.; Brand, O.; Baltes, H. Nature 2001, 414, 293. (27) Ivanisevic, A.; Yeh, J.-Y.; Mawst, L.; Kuech, T. F.; Ellis, A. B. Nature 2001, 409, 476. (28) Janata, J.; Josowicz, M. Nat. Mater. 2002, 2, 19. (29) Li, Y. Y.; Cunin, F.; Link, J. R.; Gao, T.; Betts, R. E.; Reiver, S. H.; Chin, V.; Bhatia, S. N.; Sailor, M. J. Science 2003, 299, 2045. (30) Alivisatos, A. P. Nat. Biotechnol. 2004, 22, 47. (31) Rose, A.; Zhu, Z.; Madigan, C. F.; Swager, T. M.; Bulovic, V. Nature 2005, 434, 876.

Chemical Reviews, 2008, Vol. 108, No. 2 807 (32) Potyrailo, R. A.; Ghiradella, H.; Vertiatchikh, A.; Dovidenko, K.; Cournoyer, J. R.; Olson, E. Nat. Photonics 2007, 1, 123. (33) Armani, A. M.; Kulkarni, R. P.; Fraser, S. E.; Flagan, R. C.; Vahala, K. J. Science 2007, 317, 783. (34) Carrano, J. C.; Jeys, T.; Cousins, D.; Eversole, J.; Gillespie, J.; Healy, D.; Licata, N.; Loerop, W.; O’Keefe, M.; Samuels, A.; Schultz, J.; Walter, M.; Wong, N.; Billotte, B.; Munley, M.; Reich, E.; Roos, J. In Optically Based Biological and Chemical Sensing for Defence; Carrano, J. C., Zukauskas, A., Eds.; SPIEsThe International Society for Optical Engineering: Bellingham, WA, 2004; Vol. 5617, p xi. (35) Meyerhoff, M. E. Trends Anal. Chem. 1993, 12, 257. (36) Clark, K. J. R.; Furey, J. BioProcess Int. 2006, 4(6), S16. (37) Newman, J. D.; Turner, A. P. F. Biosens. Bioelectron. 2005, 20, 2435. (38) Pickup, J. C.; Alcock, S. Biosens. Bioelectron. 1991, 6, 639. (39) Newnham, R. E. Crystallogr. ReV. 1988, 1, 253. (40) Akporiaye, D. E. Angew. Chem., Int. Ed. 1998, 37, 2456. (41) Ulmer, II, C. W.; Smith, D. A.; Sumpter, B. G.; Noid, D. I. Comput. Theor. Polym. Sci. 1998, 8, 311. (42) Suman, M.; Freddi, M.; Massera, C.; Ugozzoli, F.; Dalcanale, E. J. Am. Chem. Soc. 2003, 125, 12068. (43) Lavigne, J. J.; Anslyn, E. V. Angew. Chem., Int. Ed. 2001, 40, 3119. (44) Dickinson, T. A.; Walt, D. R.; White, J.; Kauer, J. S. Anal. Chem. 1997, 69, 3413. (45) Apostolidis, A.; Klimant, I.; Andrzejewski, D.; Wolfbeis, O. S. J. Comb. Chem. 2004, 6, 325. (46) Chojnacki, P.; Werner, T.; Wolfbeis, O. S. Microchim. Acta 2004, 147, 87. (47) Potyrailo, R. A. Macromol. Rapid Commun. 2004, 25, 77. (48) Potyrailo, R. A. Polym. Mater: Sci. Eng. (Prepr.) 2004, 90, 797. (49) McKusick, B. C.; Heckert, R. E.; Cairns, T. L.; Coffman, D. D.; Mower, H. F. J. Am. Chem. Soc. 1958, 80, 2806. (50) Bu¨hlmann, P.; Pretsch, E.; Bakker, E. Chem. ReV. 1998, 98, 1593. (51) Steinle, E. D.; Amemiya, S.; Bu¨hlmann, P.; Meyerhoff, M. E. Anal. Chem. 2000, 72, 5766. (52) Pedersen, C. J. J. Am. Chem. Soc. 1967, 89, 7017. (53) Hu, Y.; Tan, O. K.; Pan, J. S.; Yao, X. J. Phys. Chem. B 2004, 108, 11214. (54) Svetlicic, V.; Schmidt, A. J.; Miller, L. L. Chem. Mater. 1998, 10, 3305. (55) Martin, P. D.; Wilson, T. D.; Wilson, I. D.; Jones, G. R. Analyst 2001, 126, 757. (56) Potyrailo, R. A.; Sivavec, T. M. Anal. Chem. 2004, 76, 7023. (57) Walt, D. R.; Dickinson, T.; White, J.; Kauer, J.; Johnson, S.; Engelhardt, H.; Sutter, J.; Jurs, P. Biosens. Bioelectron. 1998, 13, 697. (58) Eberhart, M. E.; Clougherty, D. P. Nat. Mater. 2004, 3, 659. (59) Jandeleit, B.; Schaefer, D. J.; Powers, T. S.; Turner, H. W.; Weinberg, W. H. Angew. Chem., Int. Ed. 1999, 38, 2494. (60) Maier, W.; Kirsten, G.; Orschel, M.; Weiss, P.-A.; Holzwarth, A.; Klein, J. In Combinatorial Approaches to Materials DeVelopment; Malhotra, R., Ed.; American Chemical Society: Washington, DC, 2002; Vol. 814, p 1. (61) Takeuchi, I., Newsam, J. M., Wille, L. T., Koinuma, H., Amis, E. J., Eds. Combinatorial and Artificial Intelligence Methods in Materials Science; Materials Research Society: Warrendale, PA, 2002: Vol. 700. (62) Potyrailo, R. A.. Amis, E. J., Eds. High Throughput Analysis: A Tool for Combinatorial Materials Science; Kluwer Academic/Plenum Publishers: New York, 2003. (63) Koinuma, H.; Takeuchi, I. Nat. Mater. 2004, 3, 429. (64) Potyrailo, R. A., Karim, A., Wang, Q., Chikyow, T., Eds. Combinatorial and Artificial Intelligence Methods in Materials Science II; Materials Research Society: Warrendale, PA, 2004; Vol. 804. (65) Potyrailo, R. A., Takeuchi, I., Eds. Special Feature on Combinatorial and High-Throughput Materials Research; Measurement Science and Technology; AIP: Melville, NY, 2005; Vol. 16. (66) Potyrailo, R. A., Maier, W. F., Eds. Combinatorial and HighThroughput DiscoVery and Optimization of Catalysts and Materials; CRC Press: Boca Raton, FL, 2006. (67) Maier, W. F.; Sto¨we, K.; Sieg, S. Angew. Chem., Int. Ed. 2007, 46, 6016. (68) Burton, D. R. Acc. Chem. Res. 1993, 26, 405. (69) Osborne, S. E.; Ellington, A. D. Chem. ReV. 1997, 97, 349. (70) Clark, S. L.; Remcho, V. T. Electrophoresis 2002, 23, 1335. (71) Srinivasan, N.; Kilburn, J. D. Curr. Opin. Chem. Biol. 2004, 8, 305. (72) Schrader, T., Hamilton, A. D., Eds. Functional Synthetic Receptors; Wiley-VCH: Weinheim, Germany, 2005. (73) Klussmann, S., Ed. The Aptamer Handbook: Functional Oligonucleotides and Their Applications; Wiley-VCH: Weinheim, Germany, 2006. (74) Takeuchi, I.; Lauterbach, J.; Fasolka, M. J. Mater. Today 2005, 8 (10), 18.

808 Chemical Reviews, 2008, Vol. 108, No. 2 (75) Ludwig, A.; Cao, J.; Brugger, J.; Takeuchi, I. Meas. Sci. Technol. 2005, 16, 111. (76) Czarnik, A. W., DeWitt, S. H., Eds. A Practical Guide to Combinatorial Chemistry; American Chemical Society: Washington, DC, 1997. (77) Frank, R.; Heikens, W.; Heisterberg-Moutsis, G.; Blocker, H. Nucleic Acids Res. 1983, 11, 4365. (78) Geysen, H. M.; Meloen, R. H.; Barteling, S. J. Proc. Natl. Acad. Sci. U.S.A. 1984, 81, 3998. (79) Houghten, R. A. Proc. Natl. Acad. Sci. U.S.A. 1985, 82, 5131. (80) Lebl, M. J. Comb. Chem. 1999, 1, 3. (81) Xiang, X.-D., Takeuchi, I., Eds. Combinatorial Materials Synthesis; Dekker: New York, 2003. (82) Birina, G. A.; Boitsov, K. A. ZaVod. Lab. 1974, 40, 855 (in Russian). (83) Kennedy, K.; Stefansky, T.; Davy, G.; Zackay, V. F.; Parker, E. R. J. Appl. Phys. 1965, 36, 3808. (84) Hoffmann, R. Angew. Chem., Int. Ed. 2001, 40, 3337. (85) Hoogenboom, R.; Meier, M. A. R.; Schubert, U. S. Macromol. Rapid Commun. 2003, 24, 15. (86) Anderson, F. W.; Moser, J. H. Anal. Chem. 1958, 30, 879. (87) Eash, M. A.; Gohlke, R. S. Anal. Chem. 1962, 34, 713. (88) Hanak, J. J. J. Mater. Sci. 1970, 5, 964. (89) Xiang, X.-D.; Sun, X.; Bricen˜o, G.; Lou, Y.; Wang, K.-A.; Chang, H.; Wallace-Freedman, W. G.; Chen, S.-W.; Schultz, P. G. Science 1995, 268, 1738. (90) Chang, H.; Gao, C.; Takeuchi, I.; Yoo, Y.; Wang, J.; Schultz, P. G.; Xiang, X.-D.; Sharma, R. P.; Downes, M.; Venkatesan, T. Appl. Phys. Lett. 1998, 72, 2185. (91) Bricen˜o, G.; Chang, H.; Sun, X.; Schultz, P. G.; Xiang, X.-D. Science 1995, 270, 273. (92) Danielson, E.; Devenney, M.; Giaquinta, D. M.; Golden, J. H.; Haushalter, R. C.; McFarland, E. W.; Poojary, D. M.; Reaves, C. M.; Weinberg, W. H.; Wu, X. D. Science 1998, 279, 837. (93) Wong, D. W.; Robertson, G. H. AdV. Exp. Med. Biol. 1999, 464, 91. (94) Zhao, J.-C. AdV. Eng. Mater. 2001, 3, 143. (95) Olk, C. H. Meas. Sci. Technol. 2005, 16, 14. (96) Zou, L.; Savvate’ev, V.; Booher, J.; Kim, C.-H.; Shinar, J. Appl. Phys. Lett. 2001, 79, 2282. (97) Takeuchi, I.; Famodu, O. O.; Read, J. C.; Aronova, M. A.; Chang, K.-S.; Craciunescu, C.; Lofland, S. E.; Wuttig, M.; Wellstood, F. C.; Knauss, L.; Orozco, A. Nat. Mater. 2003, 2, 180. (98) Cui, J.; Chu, Y. S.; Famodu, O. O.; Furuya, Y.; Hattrick-Simpers, J.; James, R. D.; Ludwig, A.; Thienhaus, S.; Wuttig, M.; Zhang, Z.; Takeuchi, I. Nat. Mater. 2006, 5, 286. (99) Holzwarth, A.; Schmidt, H.-W.; Maier, W. Angew. Chem., Int. Ed. 1998, 37, 2644. (100) Cooper, A. C.; McAlexander, L. H.; Lee, D.-H.; Torres, M. T.; Crabtree, R. H. J. Am. Chem. Soc. 1998, 120, 9971. (101) Lemmon, J. P.; Wroczynski, R. J.; Whisenhunt, Jr., D. W.; Flanagan, W. P. Polym. Mater: Sci. Eng. (Prepr.) 2001, 42, 630. (102) Reddington, E.; Sapienza, A.; Gurau, B.; Viswanathan, R.; Sarangapani, S.; Smotkin, E. S.; Mallouk, T. E. Science 1998, 280, 1735. (103) Greeley, J.; Jaramillo, T. F.; Bonde, J.; Chorkendorff, I.; Nørskov, J. K. Nat. Mater. 2006, 5, 909. (104) Brocchini, S.; James, K.; Tangpasuthadol, V.; Kohn, J. J. Am. Chem. Soc. 1997, 119, 4553. (105) Lai, R.; Kang, B. S.; Gavalas, G. R. Angew. Chem., Int. Ed. 2001, 40, 408. (106) Ramirez, A. G.; Saha, R. Appl. Phys. Lett. 2004, 85, 5215. (107) Jiang, R.; Rong, C.; Chu, D. J. Comb. Chem. 2005, 7, 272. (108) Lemmon, J. P.; Manivannan, V.; Jordan, T.; Hassib, L.; Siclovan, O.; Othon, M.; Pilliod, M. In Combinatorial and Artificial Intelligence Methods in Materials Science II. MRS Symposium Proceedings; Potyrailo, R. A., Karim, A., Wang, Q., Chikyow, T., Eds.; Materials Research Society: Warrendale, PA, 2004; Vol. 804, p 27. (109) Ha¨nsel, H.; Zettl, H.; Krausch, G.; Schmitz, C.; Kisselev, R.; Thelakkat, M.; Schmidt, H.-W. Appl. Phys. Lett. 2002, 81, 2106. (110) Chisholm, B. J.; Potyrailo, R. A.; Cawse, J. N.; Shaffer, R. E.; Brennan, M. J.; Moison, C.; Whisenhunt, D. W.; Flanagan, W. P.; Olson, D. R.; Akhave, J. R.; Saunders, D. L.; Mehrabi, A.; Licon, M. Prog. Org. Coat. 2002, 45, 313. (111) Wicks, D. A.; Bach, H. Proceedings of The 29th International Waterborne, High-Solids, and Powder Coatings Symposium; University of Southern Mississippi: Hattiesburg, MS, 2002; Vol. 29, p 1. (112) Grunlan, J. C.; Mehrabi, A. R.; Chavira, A. T.; Nugent, A. B.; Saunders, D. L. J. Comb. Chem. 2003, 5, 362. (113) Stafslien, S. J.; Bahr, J. A.; Feser, J. M.; Weisz, J. C.; Chisholm, B. J.; Ready, T. E.; Boudjouk, P. J. Comb. Chem. 2006, 8, 156. (114) Ekin, A.; Webster, D. C. J. Comb. Chem. 2007, 9, 178. (115) Potyrailo, R. A.; Chisholm, B. J.; Olson, D. R.; Brennan, M. J.; Molaison, C. A. Anal. Chem. 2002, 74, 5105.

Potyrailo and Mirsky (116) Potyrailo, R. A.; Chisholm, B. J.; Morris, W. G.; Cawse, J. N.; Flanagan, W. P.; Hassib, L.; Molaison, C. A.; Ezbiansky, K.; Medford, G.; Reitz, H. J. Comb. Chem. 2003, 5, 472. (117) MacLean, D.; Baldwin, J. J.; Ivanov, V. T.; Kato, Y.; Shaw, A.; Schneider, P.; Gordon, E. M. J. Comb. Chem. 2000, 2, 562. (118) Potyrailo, R. A.; Takeuchi, I. Meas. Sci. Technol. 2005, 16, 1. (119) Bever, M. B.; Duwez, P. E. Mater. Sci. Eng. 1972, 10, 1. (120) Shen, M.; Bever, M. B. J. Mater. Sci. 1972, 7, 741. (121) Pompe, W.; Worch, H.; Epple, M.; Friess, W.; Gelinsky, M.; Greil, P.; Hempel, U.; Scharnweber, D.; Schulte, K. Mater. Sci. Eng. A 2003, 362, 40. (122) Cohan, P. E. In 2002 COMBIsThe 4th Annual International Symposium on Combinatorial Approaches for New Materials DiscoVery; Knowledge Foundation: Arlington, VA, 2002. (123) Potyrailo, R. A.; Maier, W. F. In Combinatorial and High-Throughput DiscoVery and Optimization of Catalysts and Materials; Potyrailo, R. A., Maier, W. F., Eds.; CRC Press: Boca Raton, FL, 2006; p 3. (124) Li, G. Y. COMBI 2002sThe 4th Annual International Symposium on Combinatorial Approaches for New Materials DiscoVery, Jan 235, San Diego, CA, The Knowledge Foundation, Inc.: Brookline, MA, 2002. (125) Bricker, M. L.; Gillespie, R. D.; Holmgren, J. S.; Sachtler, J. W. A.; Willis, R. R. In High Throughput Analysis: A Tool for Combinatorial Materials Science; Potyrailo, R. A., Amis, E. J., Eds.; Kluwer Academic/Plenum Publishers: New York, 2003; Chapter 26. (126) Spivack, J. L.; Webb, J.; Flanagan, W. P.; Sabourin, C.; May, R.; Hassib, L. In Combinatorial and Artificial Intelligence Methods in Materials Science II. MRS Symposium Proceedings; Potyrailo, R. A., Karim, A., Wang, Q., Chikyow, T., Eds.; Materials Research Society: Warrendale, PA, 2004; Vol. 804, p 199. (127) Spivack, J. L.; Cawse, J. N.; Whisenhunt, Jr., D. W.; Johnson, B. F.; Shalyaev, K. V.; Male, J.; Pressman, E. J.; Ofori, J.; Soloveichik, G.; Patel, B. P.; Chuck, T. L.; Smith, D. J.; Jordan, T. M.; Brennan, M. R.; Kilmer, R. J.; Williams, E. D. Appl. Catal., A 2003, 254, 5. (128) Boussie, T. R.; Diamond, G. M.; Goh, C.; Hall, K. A.; LaPointe, A. M.; Leclerc, M.; Lund, C.; Murphy, V.; Shoemaker, J. A. W.; Tracht, U.; Turner, H.; Zhang, J.; Uno, T.; Rosen, R. K.; Stevens, J. C. J. Am. Chem. Soc. 2003, 125, 4306. (129) Archibald, B.; Bru¨mmer, O.; Devenney, M.; Giaquinta, D. M.; Jandeleit, B.; Weinberg, W. H.; Weskamp, T. In Handbook of Combinatorial Chemistry. Drugs, Catalysts, Materials; Nicolaou, K. C., Hanko, R., Hartwig, W., Eds.; Wiley-VCH: Weinheim, Germany, 2002; Vol. 2, Chapter 34. (130) Potyrailo, R. A.; Ezbiansky, K.; Chisholm, B. J.; Morris, W. G.; Cawse, J. N.; Hassib, L.; Medford, G.; Reitz, H. J. Comb. Chem. 2005, 7, 190. (131) Lundstro¨m, I. Sens. Actuators A 1996, 56, 75. (132) Baker, B. E.; Kline, N. J.; Treado, P. J.; Natan, M. J. J. Am. Chem. Soc. 1996, 118, 8721. (133) Go¨pel, W. Sens. Actuators B 1998, 52, 125. (134) Weimar, U.; Go¨pel, W. Sens. Actuators B 1998, 52, 143-161. (135) Mitrovics, J.; Ulmer, H.; Weimar, U.; Go¨pel, W. Acc. Chem. Res. 1998, 31, 307. (136) Meredith, J. C.; Karim, A.; Amis, E. J. Macromolecules 2000, 33, 5760. (137) Efimenko, K.; Genzer, J. AdV. Mater. 2001, 1560. (138) Morgenthaler, S.; Lee, S.; Zurcher, S.; Spencer, N. D. Langmuir 2003, 19, 10459. (139) Xu, C.; Wu, T.; Drain, C. M.; Batteas, J. D.; Fasolka, M. J.; Beers, K. L. Macromolecules 2006, 39, 3359. (140) Potyrailo, R. A.; Wroczynski, R. J.; Pickett, J. E.; Rubinsztajn, M. Macromol. Rapid Commun. 2003, 24, 123. (141) Potyrailo, R. A.; Szumlas, A. W.; Danielson, T. L.; Johnson, M.; Hieftje, G. M. Meas. Sci. Technol. 2005, 16, 235. (142) Potyrailo, R. A.; Olson, D. R.; Brennan, M. J.; Akhave, J. R.; Licon, M. A.; Mehrabi, A. R.; Saunders, D. L.; Chisholm, B. J. Systems and methods for the deposition and curing of coating compositions; U.S. Patent 6,544,334 B1, 2003. (143) Potyrailo, R. A.; Hassib, L. ReV. Sci. Instrum. 2005, 76, 062225. (144) Turcu, F.; Hartwich, G.; Scha¨fer, D.; Schuhmann, W. Macromol. Rapid Commun. 2005, 26, 325. (145) Lemmo, A. V.; Fisher, J. T.; Geysen, H. M.; Rose, D. J. Anal. Chem. 1997, 69, 543. (146) Calvert, P. Chem. Mater. 2001, 13, 3299. (147) de Gans, B.-J.; Schubert, U. S. Langmuir 2004, 20, 7789. (148) Hassib, L.; Potyrailo, R. A. Polym. Mater: Sci. Eng. (Prepr.) 2004, 45, 211. (149) Schena, M. Microarray Analysis; Wiley: Hoboken, NJ, 2003. (150) Potyrailo, R. A.; Morris, W. G.; Wroczynski, R. J.; McCloskey, P. J. J. Comb. Chem. 2004, 6, 869.

Combinatorial Development of Sensing Materials (151) Mirsky, V. M.; Kulikov, V. In High Throughput Analysis: A Tool for Combinatorial Materials Science; Potyrailo, R. A., Amis, E. J., Eds.; Kluwer Academic/Plenum Publishers: New York, 2003; Chapter 20. (152) Mirsky, V. M.; Kulikov, V.; Hao, Q.; Wolfbeis, O. S. Macromol. Rapid Commun. 2004, 25, 253. (153) Huc, I.; Lehn, J.-M. Proc. Natl. Acad. Sci. U.S.A. 1997, 94, 21062110. (154) Reinhoudt, D. N.; Crego-Calama, M. Science 2002, 295, 2403. (155) Corbett, P. T.; Leclaire, J.; Vial, L.; West, K. R.; Wietor, J.-L.; Sanders, J. K. M.; Otto, S. Chem. ReV. 2006, 106, 3652. (156) Otto, S.; Severin, K. Top. Curr. Chem. 2007, 277, 267. (157) Bugaut, A.; Toulm, J.-J.; Rayner, B. Angew. Chem., Int. Ed. 2004, 43, 3144. (158) Lundstro¨m, I.; Shivaraman, S.; Svensson, C.; Lundkvist, L. Appl. Phys. Lett. 1975, 26, 55. (159) Lundstro¨m, I.; Svensson, C.; Spetz, A.; Sundgren, H.; Winquist, F. Sens. Actuators B 1993, 13-14, 16. (160) Winquist, F.; Spetz, A.; Armgarth, M.; Nylander, C.; Lundstro¨m, I. Appl. Phys. Lett. 1983, 43, 839. (161) Lundstro¨m, I.; Sundgren, H.; Winquist, F.; Eriksson, M.; KrantzRu¨lcker, C.; Lloyd-Spetz, A. Sens. Actuators B 2007, 121, 247. (162) Spetz, A.; Armgarth, M.; Lundstro¨m, I. J. Appl. Phys. 1988, 64, 1274. (163) Eriksson, M.; Klingvall, R.; Lundstro¨m, I. In Combinatorial and HighThroughput DiscoVery and Optimization of Catalysts and Materials; Potyrailo, R. A., Maier, W. F., Eds.; CRC Press: Boca Raton, FL, 2006; p 85. (164) Lundstro¨m, I.; Erlandsson, R.; Frykman, U.; Hedborg, E.; Spetz, A.; Sundgren, H.; Welin, S.; Winquist, F. Nature 1991, 352, 47. (165) Lo¨fdahl, M.; Eriksson, M.; Lundstro¨m, I. Sens. Actuators B 2000, 70, 77. (166) Mizsei, J. Solid-State Electron. 2000, 44, 509. (167) Mizsei, J.; Ress, S. Sens. Actuators B 2002, 83, 164. (168) Mizsei, J. Thin Solid Films 2003, 436, 25. (169) Mizsei, J. Thin Solid Films 2005, 490, 17. (170) Klingvall, R.; Lundstro¨m, I.; Lo¨fdahl, M.; Eriksson, M. IEEE Sensors J. 2005, 5, 995. (171) Lundstro¨m, I.; Sundgren, H.; Winquist, F. J. Appl. Phys. 1993, 74, 6953. (172) Lo¨fdahl, M.; Utaiwasin, C.; Carlsson, A.; Lundstro¨m, I.; Eriksson, M. Sens. Actuators B 2001, 80, 183. (173) Lo¨fdahl, M.; Eriksson, M.; Johansson, M.; Lundstro¨m, I. J. Appl. Phys. 2002, 91, 4275. (174) Spetz, A.; Helmersson, U.; Enquist, F.; Armgarth, M.; Lundstro¨m, I. Thin Solid Films 1989, 177, 77. (175) Klingvall, R.; Lundstrom, I.; Lofdahl, M.; Eriksson, M. Proc. IEEE Sensors 2003, 2, 1114. (176) Eriksson, M.; Salomonsson, A.; Lundstro¨m, I.; Briand, D.; A° bom, A. E. J. Appl. Phys. 2005, 98, 034903. (177) Filippini, D.; Gunnarsson, J.; Lundstro¨m, I. J. Appl. Phys. 2004, 96, 7583. (178) Fodor, S. P. A.; Read, J. L.; Pirrung, M. C.; Stryer, L.; Lu, A. T.; Solas, D. Science 1991, 251, 767. (179) Brattein, W. H.; Bardeen, J. Bell Syst. Tech. J. 1953, 32, 1. (180) Heiland, G. Z. Phys. 1954, 138, 459. (181) Bielanski, A.; Deren, J.; Haber, J. Nature 1957, 179, 668. (182) Seiyama, T.; Kato, A.; Fujiishi, K.; Nagatani, M. Anal. Chem. 1962, 34, 1502. (183) Taguchi, N. Jpn. Patent 45-38200, 1962. (184) Dmitriev, S.; Lilach, Y.; Button, B.; Moskovits, M.; Kolmakov, A. Nanotechnology 2007, 18, 055707. (185) Barsan, N.; Weimar, U. J. Phys.: Condens. Matter 2003, 15, R813. (186) Go¨pel, W.; Reinhardt, G. In Sensors Update; Baltes, H., Go¨pel, W., Hesse, J., Eds.; VCH: Weinheim, 1996; Vol. 1, p 49. (187) Kolmakov, A.; Moskovits, M. Annu. ReV. Mater. Res. 2004, 151. (188) Barsan, N.; Koziej, D.; Weimar, U. Sens. Actuators B 2007, 121, 18. (189) Korotcenkov, G. Sens. Actuators B 2005, 107, 209. (190) Korotcenkov, G. Mater. Sci. Eng. B 2007, 139, 1. (191) Korotcenkov, G. Sens. Actuators B 2007, 121, 664. (192) Goschnick, J.; Frietsch, M.; Schneider, T. Surf. Coat. Technol. 1998, 108-109, 292. (193) Koronczi, I.; Ziegler, K.; Kruger, U.; Goschnick, J. IEEE Sensors J. 2002, 2, 254. (194) Arnold, C.; Andlauer, W.; Ha¨ringer, D.; Ko¨rber, R.; Goschnick, J. Proc. IEEE Sensors 2002, 1, 426. (195) Arnold, C.; Harms, M.; Goschnick, J. IEEE Sensors J. 2002, 2, 179. (196) Goschnick, J.; Koronczi, I.; Frietsch, M.; Kiselev, I. Sens. Actuators B 2005, 106, 182. (197) Schneider, T.; Betsarkis, K.; Trouillet, V.; Goschnick, J. Proc. IEEE Sensors 2004, 1, 196. (198) Sysoev, V. V.; Kiselev, I.; Frietsch, M.; Goschnick, J. Sensors 2004, 4, 37.

Chemical Reviews, 2008, Vol. 108, No. 2 809 (199) Frietsch, M.; Koronczi, I.; Goschnick, J. Environmental Odour Management, Nov 17-19, Cologne Germany, 2004; Lecture 26. (200) Goschnick, J. Microelectron. Eng. 2001, 57-58, 693. (201) Graf, M.; Gurlo, A.; Barsan, N.; Weimar, U.; Hierlemann, A. J. Nanoparticle Res. 2006, 8, 823. (202) Taylor, C. J.; Semancik, S. Chem. Mater. 2002, 14, 1671. (203) Suehle, J. S.; Cavicchi, R. E.; Gaitan, M.; Semancik, S. IEEE Electron DeVice Lett. 1993, 14, 118. (204) Semancik, S.; Cavicchi, R. Acc. Chem. Res. 1998, 31, 279. (205) Panchapakesan, B.; DeVoe, D. L.; Widmaier, M. R.; Cavicchi, R.; Semancik, S. Nanotechnology 2001, 12, 336. (206) Semancik, S.; Cavicchi, R. E.; Wheeler, M. C.; Tiffany, J. E.; Poirier, G. E.; Walton, R. M.; Suehle, J. S.; Panchapakesan, B.; DeVoe, D. L. Sens. Actuators B 2001, 77, 579. (207) Meier, D. C.; Evju, J. K.; Boger, Z.; Raman, B.; Benkstein, K. D.; Martinez, C. J.; Montgomery, C. B.; Semancik, S. Sens. Actuators B 2007, 121, 282. (208) Zhu, L.; Meier, D.; Boger, Z.; Montgomery, C.; Semancik, S.; DeVoe, D. L. Sens. Actuators B 2007, 121, 679. (209) Taylor, C. J.; Gilmer, D. C.; Colombo, D. G.; Wilk, G. D.; Campbell, S. A.; Roberts, J.; Gladfelter, W. L. J. Am. Chem. Soc. 1999, 121, 5220. (210) Mazza, T.; Barborini, E.; Kholmanov, I. N.; Piseri, P.; Bongiorno, G.; Vinati, S.; Milania, P.; Ducati, C.; Cattaneo, D.; Li, Bassi, A.; Bottani, C. E.; Taurino, A. M.; Siciliano, P. Appl. Phys. Lett. 2005, 87, 103108. (211) Barborini, E.; Bongiorno, G.; Forleo, A.; Francioso, L.; Milani, P.; Kholmanov, I. N.; Piseri, P.; Siciliano, P.; Taurino, A. M.; Vinati, S. Sens. Actuators B 2005, 111-112, 22. (212) Siemons, M.; Koplin, T. J.; Simon, U. Appl. Surf. Sci. 2007, doi: 10.1016/j.apsusc.2007.06.073. (213) Semancik, S. Correlation of Chemisorption and Electronic Effects for Metal Oxide Interfaces: Transducing Principles for Temperature Programmed Gas Microsensors; Final Technical Report, Project EMSP 65421, Grant 07-98ER62709; U.S. Department of Energy Information Bridge, 2002. (214) Semancik, S. In Combinatorial Materials Synthesis; Xiang, X.-D., Takeuchi, I., Eds.; Dekker: New York, 2003; p 263. (215) Chang, K.-S.; Aronova, M. A.; Takeuchi, I. Appl. Surf. Sci. 2004, 223, 224. (216) Aronova, M. A.; Chang, K. S.; Takeuchi, I.; Jabs, H.; Westerheim, D.; Gonzalez-Martin, A.; Kim, J.; Lewis, B. Appl. Phys. Lett. 2003, 83, 1255. (217) Lampe, U.; Fleischer, M.; Reitmeier, N.; Meixner, H.; McMonagle, J. B.; Marsh, A. In Sensors Update; Baltes, H., Go¨pel, W., Hesse, J., Eds.; VCH: Weinheim, Germany, 1996; Vol. 2, p. 1. (218) Barsoukov, E., Macdonald, J. R., Eds. Impedance Spectroscopy: Theory, Experiment, and Applications, 2nd ed.; Wiley: Hoboken, NJ, 2005. (219) Simon, U.; Sanders, D.; Jockel, J.; Heppel, C.; Brinz, T. J. Comb. Chem. 2002, 4, 511. (220) Simon, U.; Sanders, D.; Jockel, J.; Brinz, T. J. Comb. Chem. 2005, 7, 682. (221) Frantzen, A.; Scheidtmann, J.; Frenzer, G.; Maier, W. F.; Jockel, J.; Brinz, T.; Sanders, D.; Simon, U. Angew. Chem., Int. Ed. 2004, 43, 752. (222) Sanders, D.; Simon, U. J. Comb. Chem. 2007, 9, 53. (223) Feldmann, C. Scr. Mater. 2001, 44, 2193. (224) Siemons, M.; Simon, U. Sens. Actuators B 2006, 120, 110. (225) Siemons, M.; Simon, U. Sens. Actuators B 2007, doi:10.1016/ j.snb.2007.04.009. (226) Siemons, M.; Simon, U. Sens. Actuators B 2007, 126, 181. (227) Scheidtmann, J.; Frantzen, A.; Frenzer, G.; Maier, W. F. Meas. Sci. Technol. 2005, 16, 119. (228) Sanders, D.; Simon, U.; Frentzer, G.; Maier, W. F. Personal communication, 2007. (229) Sieg, S.; Stutz, B.; Schmidt, T.; Hamprecht, F.; Maier, W. F. J. Mol. Mod. 2006, 12, 611. (230) Sieg, S. C.; Suh, C.; Schmidt, T.; Stukowski, M.; Rajan, K.; Maier, W. F. QSAR Comb. Sci. 2007, 26, 528. (231) Frantzen, A.; Sanders, D.; Scheidtmann, J.; Simon, U.; Maier, W. F. QSAR Comb. Sci. 2005, 24, 22. (232) Frenzer, G.; Frantzen, A.; Sanders, D.; Simon, U.; Maier, W. F. Sensors 2006, 6, 1568. (233) Kim, K.-W.; Cho, P.-S.; Kim, S.-J.; Lee, J.-H.; Kang, C.-Y.; Kim, J.-S.; Yoon, S.-J. Sens. Actuators B 2007, 123, 318. (234) Frenzer, G.; Maier, W. F. Annu. ReV. Mater. Res. 2006, 36, 281. (235) Srivastava, R.; Dwivedi, R.; Srivastava, S. K. Sens. Actuators B 1998, 50, 175. (236) Wang, Y.; Jiang, X.; Xia, Y. J. Am. Chem. Soc. 2003, 125, 16176. (237) Wei, B.-Y.; Hsu, M.-C.; Su, P.-G.; Lin, H.-M.; Wu, R.-J.; Lai, H.-J. Sens. Actuators B 2004, 101, 81.

810 Chemical Reviews, 2008, Vol. 108, No. 2 (238) Kim, Y. S.; Ha, S.-C.; Kim, K.; Yang, H.; Choi, S.-Y.; Kim, Y. T.; Park, J. T.; Lee, C. H.; Choi, J.; Paek, J.; Lee, K. Appl. Phys. Lett. 2005, 86, 213105. (239) Wang, H. C.; Li, Y.; Yang, M. J. Sens. Actuators B 2006, 119, 380. (240) Patil, D. R.; Patil, L. A. Sens. Actuators B 2007, 123, 546. (241) Miura, N.; Nakatou, M.; Zhuiykov, S. Ceram. Int. 2004, 30, 1135. (242) Fergus, J. W. Sens. Actuators B 2007, 121, 652. (243) Breysse, M.; Claudel, B.; Faure, L.; Guenin, M.; Williams, R. J. J.; Wolkenstein, T. J. Catal. 1976, 45, 137-144. (244) Nakagawa, M.; Yamashita, N. Springer Ser. Chem. Sens. Biosens. 2005, 3, 93. (245) Zhu, Y.; Shi, J.; Zhang, Z.; Zhang, C.; Zhang, X. Anal. Chem. 2002, 74, 120. (246) Huarong Tang; Yaming Li; Chengbin Zheng; Jun Ye; Xiandeng Hou; Yi Lv. Talanta 2007, 72, 1593. (247) Nakagawa, M.; Okabayashi, T.; Fujimoto, T.; Utsunomiya, K.; Yamamoto, I.; Wada, T.; Yamashita, Y.; Yamashita, N. Sens. Actuators B 1998, 51, 159. (248) Cao, X.; Zhang, X. Luminescence 2005, 20, 243. (249) Haynes, C. L.; Yonzon, C. R.; Zhang, X.; Van, Duyne, R. P. J. Raman Spectrosc. 2005, 36, 471. (250) Jeanmaire, D. L.; Van, Duyne, R. P. J. Electroanal. Chem. 1977, 84, 1. (251) Vo-Dinh, T. Trends Anal. Chem. 1998, 17, 557. (252) Willets, K. A.; Van Duyne, R. P. Annu. ReV. Phys. Chem. 2007, 58, 267-297. (253) Nie, S.; Emory, S. R. Science 1997, 275, 1102. (254) Kneipp, K.; Wang, Y.; Kneipp, H.; Perelman, L. T.; Itzkan, I.; Dasari, R.; Feld, M. S. Phys. ReV. Lett. 1997, 78, 1667. (255) Le Ru, E. C.; Etchegoin, P. G.; Meyer, M. J. Chem. Phys. 2006, 125, 204701. (256) Xu, H.; Aizpurua, J.; Ka¨ll, M.; Apell, P. Phys. ReV. E 2000, 62, 4318. (257) Ren, B.; Liu, G.-K.; Lian, X.-B.; Yang, Z.-L.; Tian, Z.-Q. Anal. Bioanal. Chem. 2007, 388, 29. (258) Fromm, D. P.; Sundaramurthy, A.; Anika, K.; Schuck, P. J.; Kino, G. S.; Moerner, W. E. J. Chem. Phys. 2006, 124, 061101. (259) Peyser-Capadona, L.; Zheng, J.; Gonzales, J. I.; Lee, T.-H.; Patel, S. A.; Dickson, R. M. Phys. ReV. Lett. 2005, 94, 058301. (260) Zhao, L. L.; Jensen, L.; Schatz, G. C. Nano Lett. 2006, 6, 1229. (261) Quagliano, L. G. J. Am. Chem. Soc. 2004, 126, 7393. (262) Lombardi, J. R.; Birke, R. L. J. Chem. Phys. 2007, 126, 244709. (263) Kahl, M.; Voges, E.; Kostrewa, S.; Viets, C.; Hill, W. Sens. Actuators B 1998, 51, 285. (264) Schierhorn, M.; Lee, S. J.; Boettcher, S. W.; Stucky, G. D.; Moskovits, M. AdV. Mater. 2006, 18, 2829. (265) Young, M. A.; Dieringer, J. A.; van Duyne, R. P. In Tip Enhancement; Kawata, S., Shalaev, V. M., Eds.; Elsevier: Amsterdam, The Netherlands, 2007; p 1. (266) Qin, L.; Zou, S.; Xue, C.; Atkinson, A.; Schatz, G. C.; Mirkin, C. A. Proc. Natl. Acad. Sci. U.S.A. 2006, 103, 13300. (267) Schatz, G. C. Proc. Natl. Acad. Sci. U.S.A. 2007, 104, 6885. (268) Zhou, Q.; Chao, Y.; Li, Y.; Xu, W.; Wu, Y.; Zheng, J. ChemPhysChem 2007, 8, 921. (269) Faraday, M. Philos. Trans. R. Soc. London 1857, 147, 145. (270) Edwards, P. P.; Thomas, J. M. Angew. Chem., Int. Ed. 2007, 46, 5480. (271) Shipway, A. N.; Katz, E.; Willner, I. ChemPhysChem 2000, 1, 18. (272) Katz, E.; Willner, I. Angew. Chem., Int. Ed. 2004, 43, 6042. (273) Hutter, E.; Fendler, J. H. AdV. Mater. 2004, 16, 1685. (274) Zhao, J.; Zhang, X.; Yonzon, C.; Haes, A. J.; Van Duyne, R. P. Nanomedicine 2006, 1, 219. (275) Homola, J., Ed. Surface Plasmon Resonance Based Sensors; Springer: Berlin, Germany, 2006. (276) Jung, L. S.; Campbell, C. T.; Chinowsky, T. M.; Mar, M. N.; Yee, S. S. Langmuir 1998, 14, 5636. (277) Han, M. S.; Lytton-Jean, A. K. R.; Oh, B.-K.; Heo, J.; Mirkin, C. A. Angew. Chem., Int. Ed. 2006, 45, 1807. (278) Tokareva, I.; Minko, S.; Fendler, J. H.; Hutter, E. J. Am. Chem. Soc. 2004, 126, 15950. (279) Matsui, J.; Akamatsu, K.; Hara, N.; Miyoshi, D.; Nawafune, H.; Tamaki, K.; Sugimoto, N. Anal. Chem. 2005, 77, 4282. (280) Dovidenko, K.; Potyrailo, R. A.; Grande, J. In Combinatorial Methods and Informatics in Materials Science; MRS Symposium Proceedings; Fasolka, M., Wang, Q., Potyrailo, R. A., Chikyow, T., Schubert, U. S., Korkin, A., Eds.; Materials Research Society: Warrendale, PA, 2006; Vol. 894, p 231. (281) Nuopponen, M.; Tenhu, H. Langmuir 2007, 23, 5352. (282) Taton, T. A.; Mirkin, C. A.; Letsinger, R. L. Science 2000, 289, 1757. (283) Bhat, R. R.; Tomlinson, M. R.; Wu, T.; Genzer, J. AdV. Polym. Sci. 2006, 198, 51. (284) Bhat, R. R.; Tomlinson, M. R.; Genzer, J. Macromol. Rapid Commun. 2004, 25, 270.

Potyrailo and Mirsky (285) Bhat, R. R.; Genzer, J. Appl. Surf. Sci. 2005, 252, 2549. (286) Bhat, R. R.; Genzer, J. Nanotechnology 2007, 18, 025301. (287) Bhat, R. R.; Tomlinson, M. R.; Genzer, J. J. Polym. Sci. B: Polym. Phys. 2005, 43, 3384. (288) Xia, Y.; Rogers, J. A.; Paul, K. E.; Whitesides, G. M. Chem. ReV. 1999, 99, 1823. (289) Martin, O. J. F. Microelectron. Eng. 2003, 67-68, 24. (290) Luo, X.; Ishihara, T. Appl. Phys. Lett. 2004, 84, 4780. (291) Gates, B. D.; Xu, Q.; Stewart, M.; Ryan, D.; Willson, C. G.; Whitesides, G. M. Chem. ReV. 2005, 105, 1171. (292) Zhang, X.; Whitney, A. V.; Zhao, J.; Hicks, E. M.; Van Duyne, R. P. J. Nanosci. Nanotechnol. 2006, 6, 1920. (293) Hirsch, T.; Zharnikov, M.; Shaporenko, A.; Stahl, J.; Weiss, D.; Wolfbeis, O. S.; Mirsky, V. M. Angew. Chem., Int. Ed. 2005, 44, 6775. (294) Piner, R. D.; Zhu, J.; Xu, F.; Hong, S.; Mirkin, C. A. Science 1999, 283, 661. (295) Ginger, D. S.; Zhang, H.; Mirkin, C. A. Angew. Chem., Int. Ed. 2004, 43, 30. (296) Salaita, K.; Wang, Y.; Mirkin, C. A. Nat. Nanotechnol. 2007, 2, 145. (297) Mirkin, C. A. MRS Bull. 2001, July, 535. (298) Weinberger, D. A.; Hong, S.; Mirkin, C. A.; Wessels, B. W.; Higgins, T. B. AdV. Mater. 2000, 12, 1600. (299) Demers, L. M.; Mirkin, C. A. Angew. Chem., Int. Ed. 2001, 40, 3069. (300) Ivanisevic, A.; McCumber, K. V.; Mirkin, C. A. J. Am. Chem. Soc. 2002, 124, 11997. (301) Mirkin, C. A.; Hong, S.; Demers, L. ChemPhysChem 2001, 2, 37. (302) Lee, K.-B.; Kim, E.-Y.; Mirkin, C. A.; Wolinsky, S. M. Nano Lett. 2004, 4, 1869. (303) Vossmeyer, T.; DeIonno, E.; Heath, J. R. Angew. Chem., Int. Ed. 1997, 36, 1080. (304) Vossmeyer, T.; Jia, S.; DeIonno, E.; Diehl, M. R.; Kim, S.-H.; Peng, X.; Alivisatos, A. P.; Heath, J. R. J. Appl. Phys. 1998, 84, 3664. (305) Koenderink, A. F.; Hernandez, J. V.; Robicheaux, F.; Noordam, L. D.; Polman, A. Nano Lett. 2007, 7, 745. (306) Hernandez, J. V.; Noordam, L. D.; Robicheaux, F. J. Phys. Chem. B 2005, 109, 15808. (307) Henzie, J.; Lee, M. H.; Odom, T. W. Nat. Nanotechnol. 2007, 2, 549. (308) Smith, D. R.; Pendry, J. B.; Wiltshire, M. C. K. Science 2004, 305, 788. (309) Soukoulis, C. M.; Linden, S.; Wegener, M. Science 2007, 315, 47. (310) Potyrailo, R. A.; Barash, E.; Dovidenko, K.; Le Tarte, L. In DeVelopments in Protein Interaction Analysis DiPIA 2007, May 6-9; GE Healthcare Biacore: Phoenix, AZ, 2007. (311) Andreev, G. O.; Driscoll, T.; Palit, S.; Cho, S. Y.; Jokerst, N. M.; Smith, D. R.; di Ventra, M.; Basov, D. N. In Book of Abstracts, Nanoelectronic DeVices for Defense & Security (NANO-DDS) Conference, June 18-21; Arlington, VA, 2007; p 98. (312) Bruchez, M., Jr.; Moronne, M.; Gin, P.; Weiss, S.; Alivisatos, A. P. Science 1998, 281, 2013. (313) Chan, W. C. W.; Nie, S. Science 1998, 281, 2016. (314) Medintz, I. L.; Uyeda, H. T.; Goldman, E. R.; Mattoussi, H. Nat. Mater. 2005, 4, 435. (315) Ko, M. C.; Meyer, G. J. In Optoelectronic Properties of Inorganic Compounds; Roundhill, D. M., Fackler, Jr., J. P., Eds.; Plenum Press: New York, 1999; p 269. (316) Lisensky, G. C.; Meyer, G. J.; Ellis, A. B. Anal. Chem. 1988, 60, 2531. (317) Seker, F.; Meeker, K.; Kuech, T. F.; Ellis, A. B. Chem. ReV. 2000, 100, 2505. (318) Nazzal, A. Y.; Qu, L.; Peng, X.; Xiao, M. Nano Lett. 2003, 3, 819. (319) Vassiltsova, O. V.; Zhao, Z.; Petrukhina, M. A.; Carpenter, M. A. Sens. Actuators B 2007, 123, 522. (320) Potyrailo, R. A.; Leach, A. M. In Proceedings of TRANSDUCERS’05, The 13th International Conference on Solid-State Sensors, Actuators and Microsystems, Seoul, Korea, June 5-9, 2005; The Institute of Electrical and Electronics Engineers, Inc.: Piscataway, NJ, 2005; p 1292. (321) Potyrailo, R. A.; Leach, A. M. Appl. Phys. Lett. 2006, 88, 134110. (322) Leach, A. M.; Potyrailo, R. A. In Combinatorial Methods and Informatics in Materials Science, MRS Symposium Proceedings; Wang, Q., Potyrailo, R. A., Fasolka, M., Chikyow, T., Schubert, U. S., Korkin, A., Eds.; Materials Research Society: Warrendale, PA, 2006; Vol. 894, p 237. (323) Murray, C. B.; Norris, D. J.; Bawendi, M. G. J. Am. Chem. Soc. 1993, 115, 8706. (324) Kovalevskij, V.; Gulbinas, V.; Piskarskas, A.; Hines, M. A.; Scholes, G. D. Phys. Stat. Sol. B 2004, 241, 1986. (325) Potyrailo, R. A.; Leach, A. M. In Abstracts Book: Asia-Pacific Conference of Transducer and Micro-Nano Technology APCOT 2006, June 25-28, Singapore; World Scientific Publishing Co. Pte. Ltd.: Singapore, 2006; p 121.

Combinatorial Development of Sensing Materials (326) Chen, Y.; Rosenzweig, Z. Anal. Chem. 2002, 74, 5132. (327) Wang, B.-C.; Chang, J.-C.; Pan, J.-H.; Xue, C.; Luo, F.-T. J. Mol. Struct. THEOCHEM 2003, 636, 81. (328) Strohmeier, G. A.; Fabian, W. M. F.; Uray, G. HelV. Chim. Acta 2004, 87, 215. (329) Singh, A.; Yao, Q.; Tong, L.; Still, W. C.; Sames, D. Tetrahedron Lett. 2000, 41, 9601. (330) Castillo, M.; Rivero, I. A. ARKIVOC 2003, 11, 193. (331) Gao, J.; Watanabe, S.; Kool, E. T. J. Am. Chem. Soc. 2004, 126, 12748. (332) Li, Q.; Lee, J.-S.; Ha, C.; Park, C. B.; Yang, G.; Gan, W. B.; Chang, Y.-T. Angew. Chem., Int. Ed. 2004, 46, 6331. (333) Mello, J. V.; Finney, N. S. J. Am. Chem. Soc. 2005, 127, 10124. (334) Hagihara, M.; Fukuda, M.; Hasegawa, T.; Morii, T. J. Am. Chem. Soc. 2006, 128, 12932. (335) Wang, S.; Chang, Y.-T. J. Am. Chem. Soc. 2006, 128, 10380. (336) Szurdoki, F.; Ren, D.; Walt, D. R. Anal. Chem. 2000, 72, 5250. (337) Regnier, T.; Garbacia, S.; Laurent, P.; des Abbayes, H.; Lavastre, O. QSAR Comb. Sci. 2005, 24, 1164. (338) Buryak, A.; Severin, K. Angew. Chem., Int. Ed. 2005, 44, 7935. (339) Buryak, A.; Severin, K. J. Comb. Chem. 2006, 8, 540. (340) Finney, N. S. Curr. Opin. Chem. Biol. 2006, 10, 238. (341) Altschuh, D.; Oncul, S.; Demchenko, A. P. J. Mol. Recognit. 2006, 19, 459. (342) Rosania, G. R.; Lee, J. W.; Ding, L.; Yoon, H.-S.; Chang, Y.-T. J. Am. Chem. Soc. 2003, 125, 1130. (343) Shedden, K.; Brumer, J.; Chang, Y. T.; Rosania, G. R. J. Chem. Inf. Comput. Sci. 2003, 43, 2068. (344) Basabe-Desmonts, L.; Beld, J.; Zimmerman, R. S.; Hernando, J.; Mela, P.; Garcia, Parajo, M. F.; van Hulst, N. F.; van den Berg, A.; Reinhoudt, D. N.; Crego-Calama, M. J. Am. Chem. Soc. 2004, 126, 7293. (345) Basabe-Desmonts, L.; Zimmerman, R. S.; Reinhoudt, D. N.; CregoCalama, M. Springer Ser. Chem. Sens. Biosens. 2005, 3, 169. (346) Basabe-Desmonts, L.; Reinhoudt, D. N.; Crego-Calama, M. AdV. Mater. 2006, 18, 1028. (347) Basabe-Desmonts, L.; Reinhoudt, D. N.; Crego-Calama, M. Chem. Soc. ReV. 2007, 36, 993. (348) Basabe-Desmonts, L.; Van Baan, F. D.; Zimmerman, R. S.; Reinhoudt, D. N.; Crego-Calama, M. Sensors 2007, 7, 1731. (349) Zolotov, Y. A.; Ivanov, V. M.; Amelin, V. G. In Wilson & Wilson’s ComprehensiVe Analytical Chemistry; Barcelo, D., Ed.; Elsevier: Amsterdam, The Netherlands, 2002. (350) Tabacco, M. B.; Uttamlal, M.; McAllister, M.; Walt, D. R. Anal. Chem. 1999, 71, 154. (351) Agayn, V. I.; Walt, D. R. Bio/Technology 1993, 11, 726. (352) Hanson, M. A.; Ge, X.; Kostov, Y.; Brorson, K. A.; Moreira, A. R.; Rao, G. Biotechnol. Bioeng. 2007, 97, 833. (353) Miller, W. W.; Yafuso, M.; Yan, C. F.; Hui, H. K.; Arick, S. Clin. Chem. 1987, 33, 1538. (354) Potyrailo, R. A.; Morris, W. G.; Leach, A. M.; Sivavec, T. M.; Wisnudel, M. B.; Boyette, S. Anal. Chem. 2006, 78, 5893. (355) Potyrailo, R. A.; Morris, W. G.; Leach, A. M.; Sivavec, T. M.; Wisnudel, M. B.; Krishnan, K.; Surman, C.; Hassib, L.; Wroczynski, R.; Boyette, S.; Xiao, C.; Agree, A.; Cecconie, T. Am. Lab. 2007, May, 32. (356) Potyrailo, R. A.; Morris, W. G.; Leach, A. M.; Hassib, L.; Krishnan, K.; Surman, C.; Wroczynski, R.; Boyette, S.; Xiao, C.; Shrikhande, P.; Agree, A.; Cecconie, T. Appl. Opt. 2007, 46, 2007. (357) Forzani, E. S.; Zhang, H.; Chen, W.; Tao, N. EnViron. Sci. Technol. 2005, 39, 1257. (358) Potyrailo, R. A.; Sivavec, T. M.; Xiao, C.; Cecconie, T. J.; Hassib, L.; Leach, A. M.; Engel, D. B. Material compositions for sensors for determination of chemical species at trace concentrations and method of using sensors; U.S. Patent Appl. 20070092973, 2007 (available from www.uspto.gov). (359) Xiao, C.; Sivavec, T. M.; Engel, D. B.; Leach, A. M. Self-contained phosphate sensors and method for using same; U.S. Patent Appl. 20070092972, 2007 (available from www.uspto.gov). (360) Liu, D.; Meruva, R. K.; Brown, R. B.; Meyerhoff, M. E. Anal. Chim. Acta 1996, 321, 173. (361) Mills, A.; Lepre, A.; Wild, L. Anal. Chim. Acta 1998, 362, 193. (362) Bedlek-Anslow, J. M.; Hubner, J. P.; Carroll, B. F.; Schanze, K. S. Langmuir 2000, 16, 9137. (363) Wang, J.; Musameh, M.; Lin, Y. J. Am. Chem. Soc. 2003, 125, 2408. (364) Cho, E. J.; Bright, F. V. Anal. Chem. 2002, 74, 1462. (365) Cho, E. J.; Tao, Z.; Tang, Y.; Tehan, E. C.; Bright, F. V.; Hicks, W. L., Jr.; Gardella, J. A., Jr.; Hard, R. Appl. Spectrosc. 2002, 56, 1385. (366) Amis, E. J. Nat. Mater. 2004, 3, 83. (367) Collaudin, A. B.; Blum, L. J. Sens. Actuators B 1997, 38-39, 189. (368) Mills, A. Sens. Actuators B 1998, 51, 60-68. (369) Eaton, K. Sens. Actuators B 2002, 85, 42.

Chemical Reviews, 2008, Vol. 108, No. 2 811 (370) Papkovsky, D. B.; Ponomarev, G. V.; Trettnak, W.; O’Leary, P. Anal. Chem. 1995, 67, 4112. (371) Levitsky, I.; Krivoshlykov, S. G.; Grate, J. W. Anal. Chem. 2001, 73, 3441. (372) Florescu, M.; Katerkamp, A. Sens. Actuators B 2004, 97, 39. (373) Basu, B. J.; Thirumurugan, A.; Dinesh, A. R.; Anandan, C.; Rajam, K. S. Sens. Actuators B 2005, 104, 15. (374) Potyrailo, R. A.; Hassib, L. In MACRO 2004sWorld Polymer Congress, the 40th IUPAC International Symposium on Macromolecules, July 4-9; IUPAC: Paris, France, 2004. (375) Potyrailo, R. A.; Hassib, L. In Proceedings of TRANSDUCERS’05, The 13th International Conference on Solid-State Sensors, Actuators and Microsystems, Seoul, Korea, June 5-9, 2005; The Institute of Electrical and Electronics Engineers, Inc.: Piscataway, 2005; p 2099. (376) Hughes, R. C.; Yelton, W. G.; Pfeifer, K. B.; Patel, S. V. J. Electrochem. Soc. 2001, 148 (4), H37. (377) Rauen, K. L.; Smith, D. A.; Heineman, W. R.; Johnson, J.; Seguin, R.; Stoughton, P. Sens. Actuators B 1993, 17, 61. (378) Potyrailo, R. A.; Morris, W. G. Anal. Chem. 2007, 79, 45. (379) Madaras, M. B.; Buck, R. P. Anal. Chem. 1996, 68, 3832. (380) Bennett, M. D.; Leo, D. J. Sens. Actuators A 2004, 115, 79. (381) DeLongchamp, D. M.; Hammond, P. T. Chem. Mater. 2003, 15, 1165. (382) Singh, K.; Shahi, V. K. J. Memb. Sci. 1998, 140, 51. (383) Potyrailo, R. A.; Mouquin, H.; Morris, W. G. Talanta 2007, doi: 10.1016/j.talanta.2007.06.023. (384) Potyrailo, R. A.; Morris, W. G.; Welles, K. B. I.; Leach, A. M.; Pris, A. D. Chemical and biological sensors, systems and methods based on radio frequency identification; U.S. Patent Appl. 20070090927, 2007 (available from www.uspto.gov). (385) Potyrailo, R. A.; Morris, W. G.; Surman, C. Unpublished results, 2007. (386) Buhlmann, K.; Schlatt, B.; Cammann, K.; Shulga, A. Sens. Actuators B 1998, 49, 156. (387) Matzger, A. J.; Lawrence, C. E.; Grubbs, R. H.; Lewis, N. S. J. Comb. Chem. 2000, 2, 301. (388) Burl, M. C.; Sisk, B. C.; Vaid, T. P.; Lewis, N. S. Sens. Actuators B 2002, 87, 130. (389) Kim, Y. S.; Ha, S.-C.; Yang, Y.; Kim, Y. J.; Cho, S. M.; Yang, H.; Kim, Y. T. Sens. Actuators B 2005, 108, 285. (390) Potyrailo, R. A.; Morris, W. G. In Abstracts of Papers; 232nd National Meeting of the American Chemical Society, Sept 10-14, 2006, San Francisco, CA; American Chemical Society: Washington, DC, 2006. (391) King, W. H., Jr. Anal. Chem. 1964, 36, 1735. (392) Lavrik, N. V.; Sepaniak, M. J.; Datskos, P. G. ReV. Sci. Instrum. 2004, 75, 2229. (393) Steinem, C., Janshoff, A., Eds. Piezoelectric Sensors; Springer: Berlin, Germany, 2007. (394) Grate, J. W.; Abraham, H.; McGill, R. A. In Handbook of Biosensors and Electronic Noses. Medicine, Food, and the EnVironment; KressRogers, E., Ed.; CRC Press: Boca Raton, FL, 1997; p 593. (395) Grate, J. W. Chem. ReV. 2000, 100, 2627. (396) Hierlemann, A.; Weimar, U.; Kraus, G.; Schweizer-Berberich, M.; Go¨pel, W. Sens. Actuators B 1995, 26, 126. (397) Wohltjen, H. Plenary Talk at the 11th International Meeting on Chemical Sensors, UniVersity of Brescia, Italy, July 16-19, 2006; Elsevier Science: Amsterdam, The Netherlands, 2006. (398) Grate, J. W.; Abraham, M. H. Sens. Actuators B 1991, 3, 85. (399) Abraham, M. H. Chem. Soc. ReV. 1993, 22, 73. (400) Maranas, C. D. Ind. Eng. Chem. Res. 1996, 35, 3403. (401) Wise, B. M.; Gallagher, N. B.; Grate, J. W. J. Chemom. 2003, 17, 463. (402) Belmares, M.; Blanco, M.; Goddard, III, W. A.; Ross, R. B.; Caldwell, G.; Chou, S.-H.; Pham, J.; Olofson, P. M.; Thomas, C. J. Comput. Chem. 2004, 25, 1814. (403) Abraham, M. H.; Andonian-Haftvan, J.; Whiting, G. S.; Leo, A.; Taft, R. S. J. Chem. Soc., Perkin Trans. 2 1994, 1777. (404) Grate, J. W.; Snow, A.; Ballantine, D. S., Jr.; Wohltjen, H.; Abraham, M. H.; McGill, R. A.; Sasson, P. Anal. Chem. 1988, 60, 869. (405) Grate, J. W.; Patrash, S. J.; Abraham, M. H. Anal. Chem. 1995, 67, 2162. (406) Grate, J. W.; Patrash, S. J.; Abraham, M. H.; Du, C. M. Anal. Chem. 1996, 68, 913. (407) Hierlemann, A.; Zellers, E. T.; Ricco, A. J. Anal. Chem. 2001, 73, 3458. (408) Potyrailo, R. A.; May, R. J.; Sivavec, T. M. Sensor Lett. 2004, 2, 31. (409) Potyrailo, R. A.; Sivavec, T. M.; Bracco, A. A. Proc. SPIE-Int. Soc. Opt. Eng. 1999, 3856, 140. (410) Sivavec, T. M.; Potyrailo, R. A. Polymer coatings for chemical sensors; U.S. Patent 6,357,278 B1, 2002. (411) Potyrailo, R. A.; Morris, W. G.; Wroczynski, R. J. In High Throughput Analysis: A Tool for Combinatorial Materials Science;

812 Chemical Reviews, 2008, Vol. 108, No. 2

(412) (413)

(414) (415) (416) (417) (418) (419) (420) (421) (422) (423) (424) (425) (426) (427) (428) (429) (430) (431) (432) (433) (434) (435) (436) (437) (438) (439) (440) (441) (442) (443) (444) (445) (446)

(447) (448) (449) (450) (451) (452) (453) (454)

Potyrailo, R. A., Amis, E. J., Eds.; Kluwer Academic/Plenum Publishers: New York, 2003; Chapter 11. Potyrailo, R. A.; Morris, W. G.; Wroczynski, R. J. ReV. Sci. Instrum. 2004, 75, 2177. Potyrailo, R. A.; McCloskey, P. J.; Ramesh, N.; Surman, C. M. Sensor devices containing co-polymer substrates for analysis of chemical and biological species in water and air; U.S. Patent Appl. 2005133697, 2005 (available from www.uspto.gov). Potyrailo, R. A.; McCloskey, P. J.; Wroczynski, R. J.; Morris, W. G. Anal. Chem. 2006, 78, 3090. Potyrailo, R. A.; Morris, W. G. ReV. Sci. Instrum. 2007, 78, 072214. Wu, X.; Kim, J.; Dordick, J. S. Biotechnol. Prog. 2000, 16, 513. Kim, D.-Y.; Wu, X.; Dordick, J. S. In Biocatalysis in Polymer Science; American Chemical Society: Washington, DC, 2003; Vol. 840; p 34. MacDiarmid, A. G. ReV. Mod. Phys. 2001, 73, 701. Bai, H.; Shi, G. Sensors 2007, 7, 267. Malhotra, B. D.; Chaubey, A.; Singh, S. P. Anal. Chim. Acta 2006, 578, 59. Ramanavicius, A.; Ramanaviciene, A.; Malinauskas, A. Electrochim. Acta 2006, 51, 6025. Borole, D. D.; Kapadi, U. R.; Mahulikar, P. P.; Hundiwale, D. G. Des. Monomers Polym. 2006, 9, 1. Mabeck, J. T.; Malliaras, G. G. Anal. Bioanal. Chem. 2006, 384, 343. Guiseppi-Elie, A.; Wallace, G. G.; Matsue, T. In Handbook of Conducting Polymers; Skotheim, T. A., Elsenbaumer, R. L., Reynolds, J. R., Eds.; Dekker: New York, 1998; p 963. Luo, X.; Killard, A. J.; Smyth, M. R. Chem. Eur. J. 2007, 13, 2138. Li, X.; Ta, N.; Sun, C. Bull. Electrochem. 2005, 21, 173. Ivanov, S.; Tsakova, V.; Mirsky, V. M. Electrochem. Commun. 2006, 8, 643. Samoylov, A. V.; Mirsky, V. M.; Hao, Q.; Swart, C.; Shirshov, Y. M.; Wolfbeis, O. S. Sens. Actuators B 2005, 106, 369. Hao, Q.; Wang, X.; Lu, L.; Yang, X.; Mirsky, V. M. Macromol. Rapid Commun. 2005, 26, 1099. Yano, J. Curr. Trends Polym. Sci. 1998, 3, 131. Sabouraud, G.; Sadki, S.; Brodie, N. Chem. Soc. ReV. 2000, 29, 283. Hao, Q.; Rahm, M.; Weiss, D.; Mirsky, V. M. Microchim. Acta 2003, 143, 147. Leclerc, M. AdV. Mater. 1999, 11, 1491. Thomas III, S. W.; Joly, G. D.; Swager, T. M. Chem. ReV. 2007, 107, 1339. Cosnier, S. Anal. Lett. 2007, 40. Gross, M.; Mu¨ller, D. C.; Nothofer, H.-G.; Scherf, U.; Neher, D.; Bra¨uchle, C.; Meerholz, K. Nature 2000, 405, 661. Lavastre, O.; Illitchev, I.; Jegou, G.; Dixneuf, P. H. J. Am. Chem. Soc. 2002, 124, 5278. Fukumoto, H.; Muramatsu, Y.; Yamamoto, T.; Yamaguchi, J.; Itaka, K.; Koinuma, H. Macromol. Rapid Commun. 2004, 25, 196. Watts, P.; Haswell, S. J. Curr. Opin. Chem. Biol. 2003, 7, 380387. Cullen, C. J.; Wootton, R. C. R.; De Mello, A. J. Curr. Opin. Drug DiscoVery DeV. 2004, 7, 798. Suga, S.; Okajima, M.; Fujiwara, K.; Yoshida, J. QSAR Comb. Sci. 2005, 24, 728. Wong Hawkes, S. Y. F. W.; Chapela, M. J. V.; Montembault, M. QSAR Comb. Sci. 2005, 24, 712. Cygan, Z. T.; Cabral, J. T.; Beers, K. L.; Amis, E. J. Langmuir 2005, 21, 3629. Xu, C.; Barnes, S. E.; Wu, T.; Fischer, D. A.; DeLongchamp, D. M.; Batteas, J. D.; Beers, K. L. AdV. Mater. 2006, 18, 1427. Song, H.; Chen, D. L.; Ismagilov, R. F. Angew. Chem., Int. Ed. 2006, 45, 7336. Xiang, Y.; LaVan, D. Proceedings of the 2nd IEEE/ASME International Conference on Mechatronic and Embedded Systems and Applications; The Institute of Electrical and Electronics Engineers, Inc.: Piscataway, 2006; p 1. Caglar, P.; Wnek, G. E. J. Macromol. Sci. A 1995, 32, 349. Mousavi, Z.; Bobacka, J.; Ivaska, A. Electroanalysis 2005, 17, 1609. Kanungo, M.; Kumar, A.; Contractor, A. Q. Anal. Chem. 2003, 75, 5673. Kulikov, V.; Mirsky, V. M. Meas. Sci. Technol. 2004, 15, 49. Kulikov, V.; Mirsky, V. M.; Delaney, T. L.; Donoval, D.; Koch, A. W.; Wolfbeis, O. S. Meas. Sci. Technol. 2005, 16, 95. Mirsky, V. M.; Kulikov, V.; Wolfbeis, O. S. Polym. Mater.: Sci. Eng. (Prepr.) 2005, 93, 1053. Hao, Q.; Kulikov, V.; Mirsky, V. M. Sens. Actuators B 2003, 94, 352. Erichsen, T.; Reiter, S.; Ryabova, V.; Bonsen, E. M.; Schuhmann, W.; Markle, W.; Tittel, C.; Jung, G.; Speiser, B. ReV. Sci. Instrum. 2005, 76, 062204.

Potyrailo and Mirsky (455) Ryabova, V.; Schulte, A.; Erichsen, T.; Schuhmann, W. Analyst 2005, 130, 1245. (456) Sljukic, B.; Baron, R.; Salter, C.; Crossley, A.; Compton, R. G. Anal. Chim. Acta 2007, 590, 67. (457) Baron, R.; Sljukic, B.; Salter, C.; Crossley, A.; Compton, R. G. Electroanalysis 2007, 19, 1062. (458) Briehn, C. A.; Schiedel, M.-S.; Bonsen, E. M.; Schuhmann, W.; Ba¨uerle, P. Angew. Chem., Int. Ed. 2001, 40, 4680. (459) Smith, R. C.; Tennyson, A. G.; Won, A. C.; Lippard, S. J. Inorg. Chem. 2006, 45, 9367. (460) Maynor, M. S.; Nelson, T. L.; O’Sullivan, C.; Lavigne, J. J. Org. Lett. 2007, 9, 3217. (461) Tang, Y.; He, F.; Yu, M.; Feng, F.; An, L.; Sun, H.; Wang, S.; Li, Y.; Zhu, D. Macromol. Rapid Comm. 2006, 27, 389. (462) Li, B.; Sauve, G.; Iovu, M. C.; Jeffries-EL, M.; Zhang, R.; Cooper, J.; Santhanam, S.; Schultz, L.; Revelli, J. C.; Kusne, A. G.; Kowalewski, T.; Snyder, J. L.; Weiss, L. E.; Fedder, G. K.; McCullough, R. D.; Lambeth, D. N. Nano Lett. 2006, 6, 1598. (463) Mudd, S. J. Immunol. 1932, 23, 423. (464) Pauling, L. J. Am. Chem. Soc. 1940, 62, 2643. (465) Polyakov, M. V.; Stadnik, P. M.; Paritzkii, M. V.; Malkin, I. M.; Dukhina, F. S. J. Phys. Chem. (U.S.S.R.) 1933, 4, 454. (466) Dickey, F. H. Proc. Natl. Acad. Sci. U.S.A. 1949, 35, 227. (467) Wulff, G.; Sarhan, A. Angew. Chem., Int. Ed. 1972, 11, 341. (468) Sellergren, B., Ed. Molecularly Imprinted Polymers; Elsevier: Amsterdam, The Netherlands, 2000; Vol. 23. (469) Haupt, K.; Mosbach, K. Chem. ReV. 2000, 100, 2495. (470) Komiyama, M.; Takeuchi, T.; Mukawa, T.; Asanuma, H. Molecular Imprinting: From Fundamentals to Applications; Wiley-VCH: Weinheim, Germany, 2003. (471) Batra, D.; Shea, K. J. Curr. Opin. Chem. Biol. 2003, 7, 434. (472) Haupt, K. In Molecularly Imprinted Materials: Science and Technology; Yan, M., Ramstro¨m, O., Eds.; CRC Press: New York, 2004; p 685. (473) Hall, A. J.; Emgenbroich, M.; Sellergren, B. In Templates in Chemistry II. Topics in Current Chemistry; Schalley, C. A., Do¨tz, K.-H., Vo¨gtle, F., Eds.; Springer-Verlag: Heidelberg, Germany, 2005; Vol. 249, p 317. (474) Piletsky, S. A.; Turner, N. W.; Laitenberger, P. Med. Eng. Phys. 2006, 28, 971. (475) Zhang, H.; Ye, L.; Mosbach, K. J. Mol. Recognit. 2006, 19, 248. (476) Diaz-Garcia, M. E.; Pina-Luis, G.; Rivero, I. A. Trends Anal. Chem. 2006, 25, 112. (477) Bossi, A.; Bonini, F.; Turner, A. P. F.; Piletsky, S. A. Biosens. Bioelectron. 2007, 22, 1131. (478) Panasyuk, T. L.; Mirsky, V. M.; Piletsky, S. A.; Wolfbeis, O. S. Anal. Chem. 1999, 71, 4609. (479) Panasyuk-Delaney, T.; Mirsky, V. M.; Ulbricht, M.; Wolfbeis, O. S. Anal. Chim. Acta 2001, 435, 157. (480) Panasyuk-Delaney, T.; Mirsky, V. M.; Wolfbeis, O. S. Electroanalysis 2002, 14, 221. (481) Delaney, T. L.; Zimin, D.; Rahm, M.; Weiss, D.; Wolfbeis, O. S.; Mirsky, V. M. Anal. Chem. 2007, 79, 3220. (482) Lavine, B. K.; Westover, D. J.; Kaval, N.; Mirjankar, N.; Oxenford, L.; Mwangi, G. K. Talanta 2007, 72, 1042. (483) Matsunaga, T.; Hishiya, T.; Takeuchi, T. Anal. Chim. Acta 2007, 591, 63. (484) Yoshimi, Y.; Ohdaira, R.; Iiyama, C.; Sakai, K. Sens. Actuators B 2001, 73, 49. (485) Hattori, K.; Hiwatari, M.; Iiyama, C.; Yoshimi, Y.; Kohori, F.; Sakai, K.; Piletsky, S. A. J. Memb. Sci. 2004, 233, 169. (486) Yan, S.; Fang, Y.; Gao, Z. Anal. Lett. 2007, 40, 1013. (487) Yan, S.; Fang, Y.; Gao, Z. Biosens. Bioelectron. 2007, 22, 1087. (488) Dickert, F. L.; Lieberzeit, P. A. In Piezoelectric Sensors; Steinem, C., Janshoff, A., Eds.; Springer: Berlin, Germany, 2007; p 173. (489) Navarro-Villoslada, F.; Urraca, J. L.; Moreno-Bondi, M. C.; Orellana, G. Sens. Actuators B 2007, 121, 67. (490) Benito-Pena, E.; Moreno-Bondi, M. C.; Aparicio, S.; Orellana, G.; Cederfur, J.; Kempe, M. Anal. Chem. 2006, 78, 2019. (491) Cywinski, P.; Sadowska, M.; Danel, A.; Buma, W. J.; Brouwer, A. M.; Wandelt, B. J. Appl. Polym. Sci. 2007, 105, 229. (492) Kubo, H.; Yoshioka, N.; Takeuchi, T. Org. Lett. 2005, 7, 359. (493) Piletsky, S. A.; Turner, A. P. F. Electroanalysis 2002, 14, 317. (494) Ulyanova, Y. V.; Blackwell, A. E.; Minteer, S. D. Analyst 2006, 131, 257. (495) Warsinke, A.; Nagel, B. Anal. Lett. 2006, 39, 2507. (496) Greene, N. T.; Shimizu, K. D. J. Am. Chem. Soc. 2005, 127, 5695. (497) Jakusch, M.; Janotta, M.; Mizaikoff, B.; Mosbach, K.; Haupt, K. Anal. Chem. 1999, 71, 4786. (498) Lettau, K.; Warsinke, A.; Katterle, M.; Danielsson, B.; Scheller, F. W. Angew. Chem., Int. Ed. 2006, 45, 6986. (499) Zayats, M.; Lahav, M.; Kharitonov, A. B.; Willner, I. Tetrahedron 2002, 58, 815.

Combinatorial Development of Sensing Materials (500) Pogorelova, S. P.; Kharitonov, A. B.; Willner, I.; Sukenik, C. N.; Pizem, H.; Bayer, T. Anal. Chim. Acta 2004, 504, 113. (501) Panasyuk, T.; Dall’Orto, V. C.; Marrazza, G.; El’skaya, A.; Piletsky, S.; Rezzano, I.; Mascini, M. Anal. Lett. 1998, 31, 1809. (502) Zimmerman, S. C.; Wendland, M. S.; Rakow, N. A.; Zharov, I.; Suslick, K. S. Nature 2002, 418, 399. (503) Wulff, G.; Vesper, W.; Grobe-Einsler, R.; Sarhan, A. Makromol. Chem. 1977, 178, 2799. (504) Wulff, G. Angew. Chem., Int. Ed. 1995, 34, 1812. (505) Wulff, G. In Molecularly Imprinted Materials: Science and Technology; Yan, M., Ramstro¨m, O., Eds.; Dekker: New York, 2005; p 59. (506) Deore, B.; Freund, M. S. Analyst 2003, 128, 803. (507) Sallacan, N.; Zayats, M.; Bourenko, T.; Kharitonov, A. B.; Willner, I. Anal. Chem. 2002, 74, 702. (508) Raitman, O. A.; Chegel, V. I.; Kharitonov, A. B.; Zayats, M.; Katz, E.; Willner, I. Anal. Chim. Acta 2004, 504, 101. (509) Mosbach, K. Trends Biochem. Sci. 1994, 19, 9. (510) Haupt, K. In Ultrathin Electrochemical Chemo- and Biosensors; Mirsky, V. M., Ed.; Springer: Berlin, Germany, 2004; p 23. (511) Andersson, H. S.; Karlsson, J. G.; Piletsky, S. A.; Koch-Schmidt, A. C.; Mosbach, K.; Nicholls, I. A. J. Chromatogr., A 1999, 848, 39. (512) Piletsky, S. A.; Mijangos, I.; Guerreiro, A.; Piletska, E. V.; Chianella, I.; Karim, K.; Turner, A. P. F. Macromolecules 2005, 38, 1410. (513) Piletsky, S. A.; Guerreiro, A.; Piletska, E. V.; Chianella, I.; Karim, K.; Turner, A. P. F. Macromolecules 2004, 37, 5018. (514) Lu, Y.; Li, C.; Wang, X.; Sun, P.; Xing, X. J. Chromatogr., B 2004, 804, 53. (515) Takeuchi, T.; Fukuma, D.; Matsui, J. Anal. Chem. 1999, 71, 285. (516) Lanza, F.; Sellergren, B. Anal. Chem. 1999, 71, 2092. (517) Lanza, F.; Hall, A. J.; Sellergren, B.; Bereczki, A.; Horvai, G.; Bayoudh, S.; Cormack, P. A. G.; Sherrington, D. C. Anal. Chim. Acta 2001, 435, 91. (518) Piletsky, S. A.; Piletska, E. V.; Chen, B.; Karim, K.; Weston, D.; Barrett, G.; Lowe, P.; Turner, A. P. F. Anal. Chem. 2000, 72, 4381. (519) Takeuchi, T.; Seko, A.; Matsui, J.; Mukawa, T. Instrum. Sci. Technol. 2001, 29, 1. (520) Dirion, B.; Cobb, Z.; Schillinger, E.; Andersson, L. I.; Sellergren, B. J. Am. Chem. Soc. 2003, 125, 15101. (521) Dirion, B.; Schillinger, E.; Sellergren, B. In Molecularly Imprinted Materialss2003. MRS Symposium Proceedings; Materials Research Society: Warrendale, PA, 2004; Vol. 787, p 53. (522) Chassaing, C.; Stokes, J.; Venn, R. F.; Lanza, F.; Sellergren, B.; Holmberg, A.; Berggren, C. J. Chromatogr., B 2004, 804, 71. (523) Lanza, F.; Sellergren, B. Macromol. Rapid Commun. 2004, 25, 59. (524) Fish, W. P.; Ferreira, J.; Sheardy, R. D.; Snow, N. H.; O’Brien, T. P. J. Liq. Chromatogr. Relat. Technol. 2005, 28, 1. (525) Salvador, J. P.; Estevez, M. C.; Marco, M. P.; Sanchez-Baeza, F. Anal. Lett. 2007, 40, 1294. (526) O’Mahony, J.; Molinelli, A.; Nolan, K.; Smyth, M. R.; Mizaikoff, B. Biosens. Bioelectron. 2006, 21, 1383. (527) Zhu, Q.-Z.; Haupt, K.; Knopp, D.; Niessner, R. Anal. Chim. Acta 2002, 468, 217. (528) Villoslada, F. N.; Takeuchi, T. Bull. Chem. Soc. Jpn. 2005, 78, 1354. (529) Navarro-Villoslada, F.; San Vicente, B.; Moreno-Bondi, M. C. Anal. Chim. Acta 2004, 504, 149. (530) Davies, M. P.; De Biasi, V.; Perrett, D. Anal. Chim. Acta 2004, 504, 7. (531) Mijangos, I.; Navarro-Villoslada, F.; Guerreiro, A.; Piletska, E.; Chianella, I.; Karim, K.; Turner, A.; Piletsky, S. Biosens. Bioelectron. 2006, 22, 381. (532) Nicholls, I. A. J. Mol. Recognit. 1998, 11, 79. (533) Nicholls, I. A.; Adbo, K.; Andersson, H. S.; Andersson, P. O.; Ankarloo, J.; Hedin-Dahlstro¨m, J.; Jokela, P.; Karlsson, J. G.; Olofsson, L.; Rosengren, J.; Shoravi, S.; Svenson, J.; Wikman, S. Anal. Chim. Acta 2001, 435, 9.

Chemical Reviews, 2008, Vol. 108, No. 2 813 (534) Piletsky, S. A.; Karim, K.; Piletska, E. V.; Day, C. J.; Freebairn, K. W.; Legge, C.; Turner, A. P. F. Analyst 2001, 126, 1826. (535) Subrahmanyam, S.; Piletsky, S. A.; Piletska, E. V.; Chen, B.; Karim, K.; Turner, A. P. F. Biosens. Bioelectron. 2001, 16, 631. (536) Chianella, I.; Lotierzo, M.; Piletsky, S. A.; Tothill, I. E.; Chen, B.; Karim, K.; Turner, A. P. F. Anal. Chem. 2002, 74, 1288. (537) Piletska, E.; Piletsky, S.; Karim, K.; Terpetschnig, E.; Turner, A. Anal. Chim. Acta 2004, 504, 179. (538) Sanchez-Barragan, I.; Karim, K.; Costa-Fernandez, J. M.; Piletsky, S. A.; Sanz-Medel, A. Sens. Actuators B 2007, 123, 798. (539) Wei, S.; Jakusch, M.; Mizaikoff, B. Anal. Chim. Acta 2006, 578, 50. (540) Farrington, K.; Regan, F. Biosens. Bioelectron. 2007, 22, 1138. (541) Dineiro, Y.; Menendez, M. I.; Blanco-Lopez, M. C.; Lobo-Castanon, M. J.; Miranda-Ordieres, A. J.; Tunon-Blanco, P. Anal. Chem. 2005, 77, 6741. (542) Dineiro, Y.; Menendez, M. I.; Blanco-Lopez, M. C.; Lobo-Castanon, M. J.; Miranda-Ordieres, A. J.; Tunon-Blanco, P. Biosens. Bioelectron. 2006, 22, 364. (543) Liu, Y.; Wang, F.; Tan, T.; Lei, M. Anal. Chim. Acta 2007, 581, 137. (544) Sagiv, J. Israel J. Chem. 1980, 18, 346. (545) Mirsky, V. M.; Hirsch, T.; Piletsky, S. A.; Wolfbeis, O. S. Angew. Chem., Int. Ed. 1999, 38, 1108. (546) Lahav, M.; Katz, E.; Willner, I. Langmuir 2001, 17, 7387. (547) Hirsch, T.; Kettenberger, H.; Wolfbeis, O. S.; Mirsky, V. M. Chem. Commun. 2003, 432. (548) Prodromidis, M. I.; Hirsch, T.; Mirsky, V. M.; Wolfbeis, O. S. Electroanalysis 2003, 15, 1795. (549) Mirsky, V. M. In Metal Nanoclusters in Catalysis and Materials Science: the Issue of Size-Control; Corain, B., Schmid, G., Toshima, N., Eds.; Elsevier: Amsterdam, The Netherlands, 2007; p 321. (550) Hirsch, T.; Shaporenko, A.; Mirsky, V. M.; Zharnikov, M. Langmuir 2007, 23, 4373. (551) Turner, N. W.; Jeans, C. W.; Brain, K. R.; Allender, C. J.; Hlady, V.; Britt, D. W. Biotechnol. Prog. 2006, 22, 1474. (552) Hayden, O.; Mann, K. J.; Krassnig, S.; Dickert, F. L. Angew. Chem., Int. Ed. 2006, 45, 2626. (553) Hayden, O.; Haderspoeck, C.; Krassnig, S.; Chen, X.; Dickert, F. L. Analyst 2006, 131, 1044. (554) Voicu, R.; Faid, K.; Farah, A. A.; Bensebaa, F.; Barjovanu, R.; Py, C.; Tao, Y. Langmuir 2007, 23, 5452. (555) Tappura, K.; IVikholm-Lundin, I.; Albers, W. M. Biosens. Bioelectron. 2007, 22, 912. (556) Ertl, G. Angew. Chem., Int. Ed. 1990, 29, 1219. (557) Conway, V. L.; Hassen, K. P.; Zhang, L.; Seitz, W. R.; Gross, T. S. Sens. Actuators B 1997, 45, 1. (558) de Gans, B.-J.; Wijnans, S.; Woutes, D.; Schubert, U. S. J. Comb. Chem. 2005, 7, 952. (559) Egger, S.; Higuchi, S.; Nakayama, T. J. Comb. Chem. 2006, 8, 275. (560) Potyrailo, R. A.; Olson, D. R.; Chisholm, B. J.; Brennan, M. J.; Lemmon, J. P.; Cawse, J. N.; Flanagan, W. P.; Shaffer, R. E.; Leib, T. K. Presented at the Invited Symposium “Analytical Tools For High Throughput Chemical Analysis And Combinatorial Materials Science”, Pittsburgh Conference on Analytical Chemistry and Applied Spectroscopy, March 4-9, New Orleans, LA, 2001. (561) Jansen, M. Angew. Chem., Int. Ed. 2002, 41, 3746. (562) Potyrailo, R. A. Presented at the 6th DPI Workshop on Combinatorial and High-Throughput Approaches in Polymer Science, Sept 10-11, 2007, Darmstadt, Germany, 2007. (563) Go¨pel, W.; Reinhardt, G. In Sensors Update; Baltes, H., Go¨pel, W., Hesse, J., Eds.; VCH: Weinheim, Germany, 1996; Vol. 1, p 47. (564) Frantzen, A.; Sanders, D.; Scheidtmann, J.; Simon, U.; Maier, W. F. QSAR Comb. Sci. 2005, 24, 22.

CR068127F

Chem. Rev. 2008, 108, 814−825

814

Electrochemical Glucose Biosensors Joseph Wang* Biodesign Institute, Center for Bioelectronics and Biosensors, Departments of Chemical Engineering and Chemistry and Biochemistry, Box 875801, Arizona State University, Tempe, Arizona 85287-5801 Received March 29, 2007

Contents 1. Introduction 2. Brief History of Electrochemical Glucose Biosensors 3. First-Generation Glucose Biosensors 3.1. Electroactive Interferences 3.2. Oxygen Dependence 4. Second-Generation Glucose Biosensors 4.1. Electron Transfer between GOx and Electrode Surfaces 4.2. Use of Nonphysiological Electron Acceptors 4.3. Wired Enzyme Electrodes 4.4. Modification of GOx with Electron Relays 4.5. Nanomaterial Electrical Connectors 5. Toward Third-Generation Glucose Biosensors 6. Solid-State Glucose Sensing Devices 7. Home Testing of Blood Glucose 8. Continuous Real Time in-Vivo Monitoring 8.1. Requirements 8.2. Subcutaneous Monitoring 8.3. Toward Noninvasive Glucose Monitoring 8.4. Microdialysis Sampling 8.5. Dual-Analyte Detection 9. Conclusions: Future Prospects and Challenges 10. Acknowledgments 11. References

814 815 815 815 816 817 817 817 817 818 818 818 819 819 820 820 822 822 822 823 823 824 824

1. Introduction Diabetes mellitus is a worldwide public health problem. This metabolic disorder results from insulin deficiency and hyperglycemia and is reflected by blood glucose concentrations higher or lower than the normal range of 80-120 mg/ dL (4.4-6.6 mM). The disease is one of the leading causes of death and disability in the world. The complications of battling diabetes are numerous, including higher risks of heart disease, kidney failure, or blindness. Such complications can be greatly reduced through stringent personal control of blood glucose. The diagnosis and management of diabetes mellitus thus requires a tight monitoring of blood glucose levels. Accordingly, millions of diabetics test their blood glucose levels daily, making glucose the most commonly tested analyte. Indeed, glucose biosensors account for about 85% of the entire biosensor market. Such huge market size makes diabetes a model disease for developing new biosensing * To whom correspondence should be addressed. E-mail: joseph.wang@ asu.edu.

Joseph Wang has been the Director of the Center for Bioelectronics and Biosensors (Biodesign Institute) and Full Professor of Chemical Engineering and Chemistry and Biochemistry at Arizona State University (ASU) since 2004. He has also served as the Chief Editor of Electroanalysis since 1988. He obtained his higher education at the Israel Institute of Technology and was awarded his D.Sc. degree in 1978. He joined New Mexico State University (NMSU) in 1980. From 2001−2004, he held a Regents Professorship and a Manasse Chair position at NMSU. His research interests include nanobiotechnology, bioelectronics, biosensors, and microfluidic devices. He has authored over 725 research papers, 9 books, 15 patents, and 25 chapters. He was the recipient of the 1994 Heyrovsky Memorial Medal (of the Czech Republic) for his major contributions to voltammetry, the 1999 American Chemical Society Award for Analytical Instrumentation, the 2006 American Chemical Society Award for Electrochemistry, and the ISI ‘Citation Laureate’ Award for being the Most Cited Scientist in Engineering in the World (during 1991−2001).

concepts. The tremendous economic prospects associated with the management of diabetes along with the challenge of providing such reliable and tight glycemic control have thus led to a considerable amount of fascinating research and innovative detection strategies.1,2 Amperometric enzyme electrodes, based on glucose oxidase (GOx), have played a leading role in the move to simple easy-to-use blood sugar testing and are expected to play a similar role in the move toward continuous glucose monitoring. Since Clark and Lyons proposed in 1962 the initial concept of glucose enzyme electrodes,3 we have witnessed tremendous effort directed toward the development of reliable devices for diabetes control. Different approaches have been explored in the operation of glucose enzyme electrodes. In addition to diabetes control, such devices offer great promise for other important applications, ranging from bioprocess monitoring to food analysis. The great importance of glucose has generated an enormous number of publications, the flow of which shows no sign of diminishing. Yet, in spite of the many impressive advances in the design and use of glucose biosensors, the promise of tight diabetes management has

10.1021/cr068123a CCC: $71.00 © 2008 American Chemical Society Published on Web 12/23/2007

Electrochemical Glucose Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 815

not been fulfilled. There are still major challenges in achieving clinically accurate continuous glycemic monitoring in connection to closed-loop systems aimed at optimal insulin delivery. Such feedback response to changes in the body chemistry has broader implications upon the management of different diseases. The management of diabetes thus represents the first example of individualized (personalized) medicine. This review discusses the principles of operation of electrochemical glucose biosensors, examines their history, discusses recent developments and current status, surveys major strategies for enhancing their performance, and outlines key challenges and opportunities in their further development and use. Emphasis is given to fundamental advances of glucose sensing principles and related materials. It is not a comprehensive review but rather discusses key developments and applications. Given the very broad field and long history of electrochemical glucose biosensors, the author apologizes for possible oversights of important contributions.

2. Brief History of Electrochemical Glucose Biosensors The history of glucose enzyme electrodes began in 1962 with the development of the first device by Clark and Lyons of the Cincinnati Children’s Hospital.3 Their first glucose enzyme electrode relied on a thin layer of GOx entrapped over an oxygen electrode via a semipermeable dialysis membrane. Measurements were made based on the monitoring of the oxygen consumed by the enzyme-catalyzed reaction glusoce oxidase

glucose + O2 98 gluconic acid + H2O2 (1) A negative potential was applied to the platinum cathode for a reductive detection of the oxygen consumption +

-

O2 + 4H + 4 e f 2H2O

(2)

The entire field of biosensors can trace its origin to this original glucose enzyme electrode. Clark’s original patent4 covers the use of one or more enzymes for converting electroinactive substrates to electroactive products. The effect of interference was corrected by using two electrodes, one of which was covered with the enzyme, and measuring the differential current. Clark’s technology was subsequently transferred to Yellow Spring Instrument (YSI) Company, which launched in 1975 the first dedicated glucose analyzer (the Model 23 YSI analyzer) for direct measurement of glucose in 25 µL whole blood samples. Updike and Hicks5 further developed this principle by using two oxygen working electrodes (one covered with the enzyme) and measuring the differential current in order to correct for the oxygen background variation in samples. In 1973, Guilbault and Lubrano6 described an enzyme electrode for the measurement of blood glucose based on amperometric (anodic) monitoring of the hydrogen peroxide product

H2O2 f O2 + 2H+ + 2e-

(3)

The resulting biosensor offered good accuracy and precision in connection with 100 µL blood samples. A wide range of amperometric enzyme electrodes, differing in electrode design or material, immobilization approach, or membrane composition, has since been described. Use of electron

acceptors for replacing oxygen in GOx-based blood glucose measurements was demonstrated in 1974.7 Continuous exvivo monitoring of blood glucose was also proposed in 1974,8 while in-vivo glucose monitoring was demonstrated by Shichiri et al. in 1982.9 During the 1980s biosensors became a ‘hot’ topic, reflecting a growing emphasis on biotechnology. Considerable efforts during this decade focused on the development of mediator-based ‘second-generation’ glucose biosensors,10-12 introduction of commercial screen-printed strips for selfmonitoring of blood glucose,13,14 and use of modified electrodes and tailored membranes/coatings for enhancing sensor performance.15 In the 1990s, we witnessed extensive activity directed toward the establishment of electrical communication between the redox center of GOx and the electrode surface.16-20 Of particular note is the work of Heller, who introduced the use of flexible polymer with osmium redox sites.16,17 During this period, we also witnessed the development of minimally invasive subcutaneously implantable devices.1,21-24 It is possible also to use glucose dehydrogenase (GDH) instead of GOx for amperometric biosensing of glucose. However, the construction of glucose biosensors based on GDH requires a source of NAD+ and a redox mediator to lower the overvoltage for oxidation of the NADH product. Quinoprotein GDH can also be used in connection to a pyrroloquinoline quinone (PQQ) cofactor

glucose + PQQ(ox) f gluconolactone + PQQ(red) (4) While eliminating the need for a NAD+ cofactor, such PQQ enzymes have not been widely used owing to their limited stability.

3. First-Generation Glucose Biosensors First-generation glucose biosensors rely on the use of the natural oxygen cosubstrate and generation and detection of hydrogen peroxide (eqs 1 and 3). The biocatalytic reaction involves reduction of the flavin group (FAD) in the enzyme by reaction with glucose to give the reduced form of the enzyme (FADH2)

GOx(FAD) + glucose f GOx(FADH2) + gluconolactone (5) followed by reoxidation of the flavin by molecular oxygen to regenerate the oxidized form of the enzyme GOx(FAD)

GOx(FADH2) + O2 f GOx(FAD) + H2O2

(6)

Measurements of peroxide formation have the advantage of being simpler, especially when miniaturized devices are concerned. Such measurements are commonly carried out on a platinum electrode at a moderate anodic potential of around +0.6 V (vs Ag/AgCl). A very common configuration is the YSI probe, which involves the entrapment of GOx between an inner anti-interference cellulose acetate membrane and an outer diffusion-limiting/biocompatible one.

3.1. Electroactive Interferences The amperometric (anodic) measurement of hydrogen peroxide at common working electrodes requires application of a relatively high potential at which endogenous reducing species, such as ascorbic and uric acids and some drugs (e.g.,

816 Chemical Reviews, 2008, Vol. 108, No. 2

acetaminophen), are also electroactive. The current contributions of these and other oxidizable constituents of biological fluids can compromise the selectivity and hence the overall accuracy of measurement. Considerable efforts during the late 1980s were devoted to minimizing the interference of coexisting electroactive compounds. One useful avenue in diminishing electroactive interferences is to employ a permselective coating that minimizes the access of these constituents toward the electrode surface. Different polymers, multilayers, and mixed layers with transport properties based on charge, size, or polarity have thus been used for blocking coexisting electroactive compounds.25-31 Such films also exclude surface-active macromolecules, hence protecting the surface and imparting higher stability. Electropolymerized films, particularly poly(phenylendiamine), polyphenol, and overoxidized polypyrrole, have been shown to be extremely useful in imparting high selectivity (by rejecting interferences based on size exclusion) while confining GOx onto the surface.25,27,28 The electropolymerization process makes it possible to generate coatings on extremely small surfaces of complex geometries, although the resulting films often have limited stability for in-vivo work. Other commonly used coatings include sizeexclusion cellulose acetate films,29 the negatively charged (sulfonated) Nafion or Kodak AQ ionomers,30 and hydrophobic alkanethiol or lipid layers.31 Use of overlaid multilayers, which combines the properties of different films, offers additional advantages. For example, alternate deposition of Nafion and cellulose acetate has been used to eliminate the interference of the neutral acetaminophen and negatively charged ascorbic and uric acids, respectively.32 Another avenue for achieving high selectivity involves the preferential electrocatalytic detection of the generated hydrogen peroxide.33-41 Such detection relies on tuning the operating potential to the optimal region (+0.0 to -0.20 V vs Ag/AgCl) where contributions from easily oxidizable interfering substances are eliminated. Remarkably high selectivity coupled with a fast and sensitive response has thus been obtained. For example, a substantial lowering of the overvoltage for the hydrogen peroxide redox process, and hence a highly selective glucose sensing, can be achieved using metal-hexacyanoferrate-based transducers.36-41 In particular, Prussian-Blue (PB; ferric-ferrocyanide) modified electrodes have received considerable attention owing to their very strong and stable electrocatalytic activity. Karyakin et al. showed the catalytic rate constant for H2O2 reduction at PB film to be 3 × 103 M-1 s-1.38 Prussian-Blue offers a substantial lowering of the overvoltage for the hydrogen peroxide redox process and hence permits highly selective biosensing of glucose at a very low potential (-0.1 V vs Ag/AgCl). The high catalytic activity of PB leads also to a very high sensitivity toward hydrogen peroxide. Further improvements in the stability and selectivity of PB-based hydrogen peroxide transducers can be obtained by electropolymerizing a nonconducting poly(1,2-diaminobenzne) permselective coating on top of the PB layer.39 A glucose nanosensor, based on the co-deposition of PB and GOx on a carbon-fiber nanoelectrode, has also been reported.40 PB-based carbon inks were developed for fabricating electrocatalytic screen-printed glucose biosensors.41 Similarly, metallized carbons such as rhodium or ruthenium on carbon33-35 have been shown to be extremely useful for highly selective biosensing of glucose. The high selectivity of metallized carbon transducers (such as Rh-C or

Wang

Ru-C) reflects their strong preferential electrocatalytic detection of hydrogen peroxide at an optimal potential range around 0.0 V, where most unwanted background reactions are negligible. Such catalytic oxidation of the peroxide product relies on the presence of a metal oxide film. The hydrogen peroxide reduces the surface metal oxide film to the metal, which is then reoxidized electrochemically, generating the anodic current signal. Miniaturized or disposable glucose microsensors have thus been prepared by electrochemical co-deposition of ruthenium and glucose oxidase onto carbon fiber microelectrodes35 or dispersing metal microparticles or metallized carbon particles within screen-printable inks.33,34 Additional improvements can be achieved by combining this preferential catalytic activity with a discriminative layer, e.g., by dispersing rhodium particles within a Nafion film.42 Low-potential selective detection of the GOx-generated hydrogen peroxide is possible also by coupling with another enzyme horseradish peroxidase (HRP) that catalyzes the peroxide oxidation.45 The marked reduction in the overvoltage for hydrogen peroxide at carbon-nanotube (CNT)-modified electrodes offers highly selective lowpotential biosensing of glucose.43,44 Yet, some controversy exists on whether the improved electrochemical behavior of hydrogen peroxide at CNT electrodes reflects the intrinsic CNT electrocatalysis or associated with metal impurities. Low-potential selective detection of the GOx-generated hydrogen peroxide is possible also by coupling with another enzyme such as horseradish peroxidase (HRP) that catalyzes the peroxide oxidation.45 The coupling of CNT with platinum nanoparticles has been shown to be extremely useful for enhancing the sensitivity and speed of GOx-based glucose biosensors (down to 0.5 µM within 3 s).46 Use of CNT molecular wires, connecting the electrode and the redox center of GOx, will be discussed in section 4.5.

3.2. Oxygen Dependence Since oxidase-based devices rely on the use of oxygen as the physiological electron acceptor, they are subject to errors resulting from fluctuations in oxygen tension and the stoichiometric limitation of oxygen. These errors include changes in sensor response and a reduced upper limit of linearity. This limitation (known as the “oxygen deficit”) reflects the fact that normal oxygen concentrations are about 1 order of magnitude lower than the physiological level of glucose. Several avenues have been proposed for addressing this oxygen limitation. One approach relies on the use of masstransport-limiting films (such as polyurethane or polycarbonate) for tailoring the flux of glucose and oxygen, i.e., increasing the oxygen/glucose permeability ratio.1,47,48 A twodimensional cylindrical electrode, designed by Gough’s group,47,48 has been particularly attractive for addressing the oxygen deficit by allowing oxygen to diffuse into the enzyme region of the sensor from both directions while glucose diffuses only from one direction (of the exposed end). This was accomplished by using a two-dimensional sensor design with a cylindrical gel containing GOx and an outside silicone rubber tube which is impermeable to glucose but highly permeable to oxygen. We addressed the oxygen limitation of glucose biosensors by designing oxygen-rich carbon paste enzyme electrodes.49,50 This biosensor is based on a fluorocarbon (Kel-F oil) pasting liquid, which has very high oxygen solubility, allowing it to act as an internal source of oxygen. The internal flux of oxygen can thus support the enzymatic

Electrochemical Glucose Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 817

Figure 1. Sequence of events that occur in ‘second-generation’ (mediator-based) glucose biosensors-mediated system.

reaction, even in oxygen-free glucose solutions. It is possible also to circumvent the oxygen demand issue by replacing the GOx with glucose dehydrogenase (GDH), which does not require an oxygen cofactor.51

4. Second-Generation Glucose Biosensors

Figure 2. Use of a redox polymer for wiring GOx: efficient electrical communication between the redox center of the enzyme and electrode surfaces.

4.1. Electron Transfer between GOx and Electrode Surfaces Further improvements (and solutions to the above errors) can be obtained by replacing the oxygen with a nonphysiological (synthetic) electron acceptor capable of shuttling electrons from the redox center of the enzyme to the surface of the electrode. The transfer of electrons between the GOx active site and the electrode surface is the limiting factor in the operation of amperometric glucose biosensors. Glucose oxidase does not directly transfer electrons to conventional electrodes because of a thick protein layer surrounding its flavin adenine dinucleotide (FAD) redox center and introducing an intrinsic barrier to direct electron transfer. Accordingly, different innovative strategies have been suggested for establishing and tailoring the electrical contact between the redox center of GOx and electrode surfaces.52-54

4.2. Use of Nonphysiological Electron Acceptors Particularly useful in developing glucose biosensors has been the use of artificial mediators that shuttle (carry) electrons between the FAD center and the electrode surface by the following scheme

glucose + GOx(ox) f gluconic acid + GOx(red) (7) GOx(red) + 2M(ox) f GOx(ox) + 2M(red) + 2H+ (8) 2M(red) f 2M(ox) + 2e-

(9)

where M(ox) and M(red) are the oxidized and reduced forms of the mediator. The reduced form is reoxidized at the electrode, giving a current signal (proportional to the glucose concentration) while regenerating the oxidized form of the mediator (eq 9). Such mediation cycle is displayed in Figure 1. Diffusional electron mediators, such as ferrocene derivatives, ferricyanide, conducting organic salts (particularly tetrathiafulvalene-tetracyanoquinodimethane, TTF-TCNQ), quinone compounds, transition-metal complexes, and phenothiazine and phenoxazine compounds, have been particularly useful to electrically contact GOx.9-12 The former received considerable attention owing to their low (pH-independent) redox potentials and larger number of derivatives. As a result of using these electron-carrying mediators, measurements become largely independent of oxygen partial pressure and can be carried out at lower potentials that do not provoke interfering reactions from coexisting electroactive species. In order to function effectively, the mediator should react

rapidly with the reduced enzyme (to minimize competition with oxygen), possess good electrochemical properties (such as a low pH-independent redox potential), and have low solubility in aqueous medium. The mediator must also be insoluble, nontoxic, and chemically stable (in both reduced and oxidized forms). The oxygen competition can be minimized if the rate of electron transfer via the mediator is high compared to the rate of the enzyme reaction with oxygen. In most cases, however, oxidation of the reduced GOx by oxygen can occur even in the presence of mediator (particularly as oxygen is freely diffusing), hence limiting the accuracy (especially at low glucose levels). In addition, the low potential of most mediators minimizes but does not eliminate the oxidation of endogenous species (particularly ascorbate). Such endogenous electroactive compounds can also consume the mediator, leading to additional errors. Commercial blood glucose self-testing meters, described in section 7, commonly rely on the use of ferricyanide or ferrocene mediators. Most in-vivo devices, however, are mediatorless due to potential leaching and toxicity of the mediator. Mediated systems also display low stability upon an extended continuous operation.

4.3. Wired Enzyme Electrodes Enzyme wiring with a redox polymer offers additional improvements in the electrical contact between the redox center of GOx and electrode surfaces (Figure 2). An elegant nondiffusional route for establishing a communication link between GOx and electrodes was developed by Heller’s group.16,55 This was accomplished by ‘wiring’ the enzyme to the surface with a long flexible hydrophilic polymer backbone [poly(vinylpyridine) or poly(vinylimidazole)] having a dense array of covalently linked osmium-complex electron relays. The redox polymer penetrates and binds the enzyme (through multiple lysine amines) to form a threedimensional network that adheres to the surface. Such folding along the GOx dramatically reduces the distance between the redox centers of the polymer and the FAD center of the enzyme. The resulting film conducts electrons and is permeable to the substrate and product of the enzymatic reaction. Electrons originating from the redox site of GOx are thus transferred through the gel’s polymer network to the electrode. The resulting three-dimensional redox-polymer/ enzyme networks thus offer high current outputs and fast response and stabilize the mediator to electrode surfaces. Current densities as high as mA/cm2 were reached upon wiring multiple enzyme layers. Such huge current densities

818 Chemical Reviews, 2008, Vol. 108, No. 2

Wang

facilitate the use of ultrasmall enzyme electrodes. The remarkable sensitivity is coupled with very high selectivity (e.g., negligible interferences from ascorbic and uric acids, acetaminophen, and cysteine at +0.20V vs SCE).56 Such wired enzyme electrodes are particularly attractive for invivo applications where leaching of diffusional mediators is to be avoided and when small size is important.

4.4. Modification of GOx with Electron Relays Chemical modification of GOx with electron-relay groups represents another attractive route for facilitating the electron transfer between the GOx redox center and the electrode surface. In 1984 Hill described the covalent attachment of ferrocene-monocarboxylic acid to the lysine residues of GOx using isobutyl choloformate,11 while Heller16 used carbodimide coupling for attaching the same mediator to GOx. Such covalent attachment of ferrocene groups led to direct oxidation of the flavin center of GOx at unmodified electrodes with the bound ferrocenes allowing electron tunneling in a number of consecutive steps. Bartlett described the carbodimide-based covalent attachment of TTF to the peptide backbone of GOx.20 Direct oxidation of the FAD centers of the enzyme was demonstrated without the need for soluble species. Glucose biosensors with extremely efficient electrical communication with the electrode can be generated by the enzyme reconstitution process. Willner’s group57 reported on an elegant approach for modifying GOx with electron relays and obtaining efficient electrical contact. For this purpose, the FAD active center of the enzyme was removed to allow positioning of an electron-mediating ferrocene unit prior to the reconstitution of the apoenzyme with the modified FAD. The attachment of electron-transfer relays at the enzyme periphery has also been considered by the same group for yielding short electron-transfer distances.52,54 While clearly illustrating a direct coupling, demonstration of a stable response would be required prior to practical applications of this elegant approach.

4.5. Nanomaterial Electrical Connectors The emergence of nanotechnology has opened new horizons for the application of nanomaterials in bioanalytical chemistry. Recent advances in nanotechnology offer exciting prospects in the field of bioelectronics. Owing to the similar dimensions of nanoparticles and redox proteins such nanomaterials can be used for effective electrical wiring of redox enzymes. Various nanomaterials, including gold nanoparticles or carbon nanotubes (CNT), have thus been used as electrical connectors between the electrode and the redox center of GOx. For example, apo-glucose oxidase can be reconstituted on a 1.4 nm gold nanocrystal functionalized with the FAD cofactor.58 The gold nanoparticle, immobilized onto the gold electrode by means of a dithiol linker, thus acts as an “electrical nanoplug” (relay unit) for the electrical wiring of its redox-active center. This leads to a high electron-transfer turnover rate of ∼5000 per second. Carbon nanotubes (CNT) represent additional nanomaterials that can be coupled to enzymes to provide a favorable surface orientation and act as an electrical connector between their redox center and the electrode surface. Particularly useful for this task have been vertically aligned CNTs that act as molecular wires (‘nanoconnectors’) between the underlying electrode and a redox enzyme.59-61 Willner’s group59 dem-

Figure 3. Carbon nanotube (CNT) connectors with long-range electrical contacting. Assembly of the CNT electrically contacted glucose oxidase electrode. (Reprinted with permission from ref 59. Copyright 2004 Wiley-VCH.)

onstrated that aligned reconstituted glucose oxidase (GOx) on the edge of single-wall carbon nanotubes (SWCNT) can be linked to an electrode surface (Figure 3). Such enzyme reconstitution on the end of CNT represents an extremely efficient approach for ‘plugging’ an electrode into GOx. Electrons were thus transported along distances higher than 150 nm with the length of the SWCNT controlling the rate of electron transport. An interfacial electron-transfer rate constant of 42 s-1 was estimated for 50 nm long SWCNT. Efficient direct electrical connection to GOx was reported also by Gooding’s group in connection to aligned SWCNT arrays.60 At present, activation of the bioelectrocatalytic functions of GOx by nanoparticles or CNT requires electrical overpotentials (beyond the thermodynamic redox potential of the enzyme redox center). Improving the contact between the nanomaterial and the electrode might decrease this overpotential.

5. Toward Third-Generation Glucose Biosensors Ultimately, one would like to eliminate the mediator and develop a reagentless glucose biosensor with a low operating potential, close to that of the redox potential of the enzyme. In this case, the electron is transferred directly from glucose to the electrode via the active site of the enzyme. The absence of mediators is the main advantage of such third-generation biosensors, leading to a very high selectivity (owing to the very low operating potential). However, as discussed earlier, critical challenges must be overcome for the successful realization of this direct electron-transfer route owing to the spatial separation of the donor-acceptor pair. Efficient direct

Electrochemical Glucose Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 819

Figure 5. Carbon nanotube (CNT)-based transistor for biosensing of glucose. Schematic picture of two electrodes connecting a semiconducting CNT with GOx enzymes immobilized on its surface. (Reprinted with permission from ref 69. Copyright 2003 American Chemical Society.) Figure 4. Three generations of amperometric enzyme electrodes for glucose based on the use of natural oxygen cofactor (A), artificial redox mediators (B), or direct electron transfer between GOx and the electrode (C).

electron transfer at conventional electrodes has been reported only for few redox enzymes. There are mixed reports in the literature regarding the direct (mediatorless) electron transfer catalyzed by GOx.53 Although several papers claim such direct electron transfer between GOx and the electrode, only few provide the level of proof for such mediatorless detection. Unsuccessful efforts to obtain direct electron transfer of GOx at conventional electrodes led to exploration of new electrode materials. The optimally designed electrode configuration has to ensure that the electron-transfer distance between the immobilized protein and the surface is made as short as possible. One route for creating third-generation amperometric glucose biosensors is to use conducting organic salt electrodes based on charge-transfer complexes such as tetrathiafulvalene-tetracyanoquinodimethane (TTF-TCNQ).62-64 Different electron-transfer mechanisms at TTF-TCNQ electrodes have been proposed by different authors, and the precise mechanism of GOx catalysis remains controversial. Khan et al.63 described a third-generation amperometric glucose sensor based on a stable charge-transfer complex electrode. The device relied on the growing tree-shaped crystal structure of TTF-TCNQ. The authors claimed that the close proximity and favorable orientation of the enzyme at the crystal surface allowed direct oxidation of the enzyme and selective glucose measurements at 0.1 V (vs Ag/AgCl), although they did not provide a convincing evidence for such direct electron transfer. Palmisano et al.64 described a disposable third-generation amperometric glucose sensor based on growing TTF-TCNQ tree-like crystals through an anti-interference layer of a nonconducting polypyrrole film. Arguments against direct electron transfer were presented by Cenas and Kulys.65 These authors suggested that the electron transfer of GOx at TTF-TCNQ electrodes is mediated and involves corrosion of the TTF-TCNQ to produce dissolved components of these organic salts that mediate the electron transfer of the enzyme. Mediatorless third-generation glucose biosensors based on the GOx/polypyrrole system were suggested by Aizawa66 and Koopal.67 However, the relatively high anodic potential of this system (vs the redox potential of FAD/FADH2, -0.44 V) suggests the possibility of electron transfer mediated by oligomeric pyrroles present on the surface. Oxidized boron-doped diamond electrodes also indicated recently some promise for mediator-free glucose detection based on direct electron transfer.68 Figure 4 summarizes various generations of amperometric glucose biosensors based on different mechanisms of electron transfer, including the use of natural secondary substrates, artificial redox mediators, or direct electron transfer. Although substantial progress has been made on the electronic

coupling of GOx, further improvements in the charge transport between its FAD redox center and electrodes are desired.

6. Solid-State Glucose Sensing Devices The unique electrical properties of 1-dimensional nanomaterials, such as carbon nanotubes, have been shown to be useful for developing conductivity based nanosensors for glucose.69 Dekker’s group demonstrated that GOx-coated semiconducting SWNTs act as sensitive pH sensors and that the conductance of GOx-coated semiconducting SWNTs changes upon addition of glucose substrate (Figure 5). A conductivity-based glucose nanobiosensor based on conducting-polymer-based nanogap has been developed by Tao and co-workers.70 Such nanojunction-based sensor was formed by using polyaniline/glucose oxidase for bridging a pair of nanoelectrodes separated with a small gap (ca. 20-60 nm). Solid-state transistor-like switchable glucose sensing devices were reported earlier by Bartlett’s group.71 Such ‘enzyme-transistor’ responsive to glucose was prepared by connecting two carbon band microelectrodes with poly(aniline) (PANI) film covered with a GOx/poly(1,2-diaminobenzene) layer. Addition of glucose, in the presence of TTF+, resulted in a conductivity change associated with the reduction of poly(aniline) by the enzyme mediated by TTF+

TTF + PANI(ox) f TTF+ + PANI(red)

(10)

7. Home Testing of Blood Glucose Electrochemical biosensors are well suited for addressing the needs of personal (home) glucose testing and have played a key role in the move to simple one-step blood sugar testing. Since blood glucose home testing devices are used daily to diagnose potentially life-threatening events they must be of extremely high quality. The majority of personal blood glucose monitors rely on disposable screen-printed enzyme electrode test strips.72,73 Such single-use electrode strips are mass produced by the rapid and simple thick-film (screenprinting) microfabrication or vapor deposition process.34,74 The screen-printing technology involves printing patterns of conductors and insulators onto the surface of planar solid (plastic or ceramic) substrates based on pressing the corresponding inks through a patterned mask. Each strip contains the printed working and reference electrodes, with the working one coated with the necessary reagents (i.e., enzyme, mediator, stabilizer, surfactant, linking, and binding agents) and membranes (Figure 6). The reagents are commonly dispensed by an ink-jet printing technology and deposited in the dry form. A counter electrode and an additional (‘baseline’) working electrode may also be included. Various membranes (mesh, filter) are often incorporated into the test strips and along with surfactants are used to provide a

820 Chemical Reviews, 2008, Vol. 108, No. 2

Wang

tracking of changes in the glucose level, are expected in the near future.73

8. Continuous Real Time in-Vivo Monitoring

Figure 6. Cross section of a commercial strip for self-testing of blood glucose (based on the Precision biosensor manufactured by Abbott Inc.): (A) electrode system; (B) hydrophobic layer (drawing the blood).

uniform sample coverage and separate the blood cells. Such single-use devices eliminate problems of carry over, cross contamination, or drift. Overall, despite their low cost and mass production such sensor strips are based on a high degree of sophistication essential for ensuring high clinical accuracy. The control meter is typically small (pocket-size), light, and battery operated. It relies on a potential-step (chronoamperometric) operation in connection with a short incubation (reaction) step. Such devices offer considerable promise for obtaining the desired clinical information in a simpler (“user-friendly”), faster, and cheaper manner compared to traditional assays. In 1987 Medisense Inc. (in the United Kingdom) launched the first product of this type, the penstyle Exactech device, based on the use of a ferrocenederivative mediator. Since then, over 40 different commercial strips and pocket-sized monitors have been introduced for self-testing of blood glucose.73,75 However, over 90% of the market consists of products manufactured by four major companies, including Life Scan, Roche Diagnostics, Abbott, and Bayer. Most of these meters rely on a ferricyanide mediator, except for the Abbott devices that employ a ferrocene derivative or an osmium-based redox polymer. In all cases, the diabetic patient pricks the finger, places the small blood droplet on the sensor strip, and obtains the blood glucose concentration (on a LC display) within 5-30 s. Some of the new meters allow sampling of submicroliter blood samples from the forearm, thus reducing the pain and discomfort associated with piercing the skin. For example, the FreeStyle monitor of Abbott relies on coulometric strips with a 50 µm gap thin-layer cell for assays of 300 nL blood samples. Widespread use of such alternative sampling sites requires that the collected samples properly reflect the blood glucose values (especially when these levels change rapidly). In addition to fast response and small size, modern personal glucose meters have features such as extended memory capacity and computer downloading capabilities. Overall, the attractive performance of modern blood glucose monitors reflects significant technological advances along with major fundamental developments (described in previous sections). Despite these remarkable technological advances, home testing of blood glucose often suffers from low and irregular testing frequency (related to the inconvenience and discomfort), inadequate interpretation of the results by the patient, or liability issues and requires compliance by patients. More integrated devices, offering multifunctional capability, enhanced interface with the physician’s work, and convenient

Although self-testing is considered a major advance in glucose monitoring, it is limited by the number of tests per day it permits. The inconvenience associated with standard finger-stick sampling deters patients from frequent monitoring. Such testing neglects the monitoring of nighttime variations. This means that measurements do not reflect the overall direction, trends, and patterns of the blood glucose level. Hence, they may result in poor approximation of blood glucose variations. Tighter glycemic control, through more frequent measurements or continuous monitoring, is desired for detecting sharp changes in the glucose level and triggering proper alarm in cases of hypo- and hyperglycemia. Continuous glucose monitoring provides maximal information about changing blood glucose levels throughout the day, including the direction, magnitude, duration, and frequency of such fluctuations. Continuous glucose monitoring thus addresses the deficiencies of test-strip-based meters and provides the opportunity of making fast and optimal therapeutic interventions (i.e., insulin delivery).76 This would minimize short-term crises and long-term complications of diabetes and lead to improved quality and length of life for people with diabetes. Glucose biosensors are thus key components of closed-loop glycemic control systems for regulating a person’s blood glucose. The concept of closed-loop (sense/release) systems is expected to have a major impact upon the treatment and management of other diseases and revolutionize patient monitoring.77,78 Such a ‘sense and act’ route for diabetes management system represents the first example of an individualized drug administration system for optimal therapeutic intervention. Legal and liability issues may impede the practical implementation of the ‘sense and act’ approach since a potential false high response from the in-vivo sensor may lead to an insulin overdose. A wide range of possible in-vivo glucose electrochemical biosensors, based on different needle designs, materials, and membrane coatings, has been studied over the past 25 years. The first application of such devices for in-vivo glucose monitoring was demonstrated by Shichiri et al. in 1982.9 His group’s needle-type glucose sensor relied on a platinum anode held at +0.6 V (vs a silver cathode), which was used to monitor the enzymatically produced hydrogen peroxide. The enzyme (GOx) entrapment was accomplished in connection with a cellulose-diacetate/heparin/polyurethane coating. The majority of glucose sensors used for in-vivo applications are based on the GOx-catalyzed oxidation of glucose by oxygen owing to concerns about potential leaching of mediators.

8.1. Requirements The major requirements of clinically accurate in-vivo glucose sensors have been discussed in several review articles.1,23,76,79 The ideal sensor would be one that provides a reliable real-time continuous monitoring of all blood glucose variations throughout the day with high selectivity and speed over extended periods under harsh conditions. The challenges for meeting these demands include rejection of the sensor by the body, miniaturization, long-term stability of the enzyme and transducer, oxygen deficit, in-vivo

Electrochemical Glucose Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 821

Figure 7. Inflammatory response of implantable sensor in the subcutaneous tissue. Sequence of events that leads to formation of fibrous capsules around chemical sensors. (Reprinted with permission from ref 80. Copyright 2006 American Chemical Society.)

calibration, short stabilization times, baseline drift, safety, and convenience. The sensor must be of a very tiny size and proper shape that allows for easy implantation and results in minimal discomfort. Last but not least is the powering of an autonomous sensor-transmitter system. Reducing the size of the power source (e.g., biofuel cell, battery) remains a major challenge. Undesirable interactions between the surface of the implanted device and biological medium cause deterioration of the sensor performance upon implantation and proved to be the major barrier to the development of reliable in-vivo glucose probes. Such adverse effects include the effect of the sensor upon the host environment as well as the environment effect upon the sensor performance. In blood, the major source of complication arises from surface fouling of the electrode by proteins and coagulation composites and the risk of thromboembolism. Due to this severe bloodinduced biofouling (that suppresses the glucose response), most glucose biosensors lack the biocompatibility necessary for reliable prolonged operation in whole blood. The danger of thrombus formation is another major concern (health risk) hindering the implementation of sensors implanted in the blood. Accordingly, the majority of the sensors being developed for continuous glucose monitoring do not measure blood glucose directly. Alternative sensing sites, particularly the subcutaneous tissue, have thus received growing attention. The subcutaneous tissue is minimally invasive, and its glucose level reflects the blood glucose concentration. However, such subcutaneous implantation generates a wound site that experiences an intense local inflammatory reaction. This inflammatory response associated with the wound formation is characterized with problems such as scar tissue formation accompanied by adhesion of bacteria and macrophage and distortion of the glucose concentration in the immediate vicinity of the sensor (Figure 7). The extent of this inflammatory response depends upon various factors, including the shape, size, and rigidity of the sensor as well as its physical and chemical character.1 Recent approaches for designing more biocompatible invivo glucose sensors focused on preparing interfaces that resist biofouling. These include a controlled release of nitric oxide (NO),80-83 protecting the outer surface with polymeric coatings (such as polyethylene glycols, polyethylene oxides, or the perfluorinated ionomer Nafion) that exhibit low protein adsorption84-86 or co-immobilization of the anticoagulant heparin.87 The former is attributed to the discovery that NO is an effective inhibitor of platelet and bacterial adhesion. Such NO-release glucose sensors were prepared by doping the outer polymeric membrane coating of previously reported

Figure 8. Nitric oxide-releasing coating for improved biocompatibility of glucose biosensors. Schematic of the hybrid xerogel/ polyurethane glucose biosensor employing NO-donor-modified sol-gel particles supported in a polyurethane matrix. (Reprinted with permission from ref 81. Copyright 2004 American Chemical Society.)

needle-type electrochemical sensors with suitable lipophilic diazeniumdiolate species82 or diazeniumdiolate-modified sol-gel particles (Figure 8).81 Histological examination of the implant site demonstrated a significant decrease in the inflammatory response. Similarly, poly(ethylene glycol) (PEG) containing polymers are among the least protein absorbing. Quinn et al.85 reported on a glucose permeable hydrogel based on cross-linking an 8-armed amine-terminated PEG derivative with a di-succinimidyl ester of a dipropionic acid derivative of PEG. The gel was evaluated as a biocompatible interface between an amperometric glucose microsensor and the subcutaneous tissue of a rat. Very few adherent cells were observed after 7 days. Calibration, i.e., the transformation of the time-dependent current signal i(t) into an estimation of glucose concentration at time t, CG(t), represents another major challenge to the development of sensors for continuous monitoring of glucose. This can be accomplished using one-point88 or two-point89 calibration procedures. In the one-point calibration procedure, the sensor sensitivity S is determined from a single blood glucose determination as the ratio between the current signal i and the blood glucose concentration CG. Such “one-point” in-vivo calibration can be used for highly selective sensors having a zero output current at zero glucose concentration.88 A single withdrawn blood sample can thus provide the one calibration point. If the intercept io is not negligible, a twopoint calibration procedure is essential.89 The two-point calibration involves an estimate of two parameters: the sensor sensitivity S and the intercept io (the sensor output observed in the absence of glucose). The glucose concentration at any time can thus be estimated from the current i

CG(t) ) (i - io)/S

(11)

Proper calibration thus ensures that the measured tissue glucose concentration accurately reflects the blood glucose level. A key issue is still maintaining the calibration over a period of several days. The calibration process should be repeated during implantation to account for variations in sensitivity. A calibration-free operation is the ultimate goal, but this would require detailed understanding of the sensitivity changes along with highly reproducible devices.

822 Chemical Reviews, 2008, Vol. 108, No. 2

Wang

ized glucose-oxygen biofuel cell, based on an implantable 7 µm carbon fiber anode and cathode (coated with GOx and laccase, respectively), for powering the autonomous sensortransmitter system.93 Additional devices based on patch-like sensors, nanoneedles, and microdialysis sampling are currently being developed by different organizations. The later are discussed below.

8.3. Toward Noninvasive Glucose Monitoring

Figure 9. Design of an implantable four-layered glucose biosensor for subcutaneous monitoring. (Reprinted with permission from ref 88. Copyright 1994 American Chemical Society.)

Although major advances have been made and several short-term in-vivo glucose sensors are approaching the commercial stage, major efforts are required before a reliable long-term minimally invasive or noninvasive sensing becomes a reality.

8.2. Subcutaneous Monitoring Most of the recent attention regarding real-time in-vivo monitoring has been given to the development of subcutaneously implantable needle-type electrodes.21-24 Such devices track blood glucose levels by measuring the glucose concentration in the interstitial fluid of the subcutaneous tissue (assuming the ratio of the blood/tissue levels is constant). Subcutaneously implantable devices are commonly designed to operate for a few days, after which they are replaced by the patient. They are commonly inserted into the subcutaneous tissue in the abdomen or upper arm. Success in this direction has reached the level of short-term human implantation; continuously functioning devices, possessing adequate (>1 week) stability, are expected in the very near future. Such devices would enable a swift and appropriate corrective action through use of a closed-loop insulin delivery system, i.e., an artificial pancreas. Computer algorithms correcting for the transient difference (short time lag) between blood and tissue glucose concentrations have been developed.24 These algorithms will be used in future closed-loop feedback systems to calculate the right amount of dispensed insulin. Subcutaneously implantable glucose sensors have moved from the purely experimental stage to commercially available products.90,91 The CGMS unit of Medtronic Minimed Inc. (Sylmar, CA) offers a 72 h period of such subcutaneous monitoring with measurement of tissue glucose every 5 min (nearly 300 readings per day) and data storage in the monitor’s memory.90 After 72 h, the sensor is removed and the information is transferred to a computer that identifies patterns of glucose variations. It was recommended that the management decision should rely on trends in the sensor recording and not upon a single-point reading.92A similar system is currently being developed by Abbott Inc.91 This system is based on the wired enzyme technology of Heller’s group (Figure 9), which involves insertion of a short needle into the skin and yields a reading every minute. The implanted element, designed to function for about 4 days between replacements, is small enough to be painlessly replaced by the user. Both the Abbott and Minimed devices include a limited range transmitter that relays the sensor data to a pager-like device that provides the necessary warnings and stores the data. Heller’s team has engineered a miniatur-

Noninvasive glucose sensing is the ultimate goal of glucose monitoring. This noninvasive route for continuous glucose monitoring is expected to eliminate the challenges, pain, and discomfort associated with implantable devices. Noninvasive methods are thus preferable to invasive techniques, provided that they do not compromise the clinical accuracy. Such ability to measure glucose noninvasively will thus revolutionize the treatment of diabetes. This approach has been directed toward glucose measurements in saliva, tears, or sweat. In particular, Cygnus Inc. has developed a watchlike glucose monitor based on the coupling of reverse iontophoretic collection of glucose across the skin with the biosensor function.94 The wearable GlucoWatch device (available now from Animas Technologies Inc.) contains both the extraction and the sensing functions along with the operating and data-storage circuitry. It provides up to three glucose readings per hour for up to 12 h (i.e., 36 readings within a 12 h period). The system has been shown to be capable of measuring the electroosmotically extracted glucose with a clinically acceptable level of accuracy. An alarm capability is included to alert the individual of very low or high glucose levels. However, the unit requires a long warm up and calibration against fingerstick blood measurement and is subject to difficulties due to skin rash with irritation under the device, long warm up times, sweating, or change in the skin temperature. Other routes for “collecting” the glucose through the skin and for noninvasive glucose testing are currently being examined by various groups and companies. Most of these approaches rely on optical detection, which is beyond the scope of this review. Despite these extensive efforts, no reliable method is presently available for continuous noninvasive glucose monitoring and it is still uncertain if such reliable monitoring will become available in the near future.

8.4. Microdialysis Sampling Another alternative to implanted needle glucose biosensors is to use microdialysis as an interface between the body and the biosensor. Here a hollow dialysis fiber is commonly implanted in the subcutaneous tissue and perfused with isotonic fluid. Glucose, diffusing from the tissue into the fiber, is thus pumped toward to the enzyme electrode. Various groups developed portable systems for continuous tissue glucose monitoring based on such combination of microdialysis and enzymatic amperometric glucose measurement.95-100 For example, Vering et al.96 described a microdialysis-based wearable system for continuous in-vivo monitoring of glucose. Sampling was performed by means of a biocompatible microdialysis needle probe inserted into the subcutaneous tissue. A microfabricated enzyme electrode was used in connection to a stopped-flow procedure. Langerman et al. applied a microdialysis system for determining glucose and lactate in the brain tissue of injured critical care patients.95 Several companies, such as Menarini Diagnostics

Electrochemical Glucose Biosensors

Chemical Reviews, 2008, Vol. 108, No. 2 823

9. Conclusions: Future Prospects and Challenges

Figure 10. Integrated needle-type glucose/insulin microsensor based on electrocatalytic (RuOx) insulin detection and biocatalytic (GOx) glucose sensing. (Reprinted with permission from ref 103. Copyright 2001 American Chemical Society.)

or Roche, are currently exploring commercial microdialysis glucose monitoring probes. The GlucoDay microdialysis system of Menarini Diagnostics displayed good correlation with venous blood glucose measurements of 70 diabetes patients.100 The Roche (Disentronic) system is non-enzymatic and relies on monitoring glucose-induced changes in the viscosity associated with binding to the lectin concanavalin A.97

8.5. Dual-Analyte Detection Various clinical situations require the simultaneous monitoring of glucose and of other clinically important analytes, such as lactate or insulin. Such coupling of two sensing elements requires both analytes to be monitored independently at different levels and without cross talk. For example, the simultaneous monitoring of lactate and glucose is of considerable interest for patient monitoring during intensivecare and surgical operations. Wilkins’s group described an integrated needle-type biosensor for intravascular glucose and lactate monitoring.101 In order to miniaturize the whole sensor and incorporate it into a hypodermic needle, the working electrode of the glucose sensor was made by electrodeposition of platinum on the needle surface, while the lactate sensor was made from platinum wire which was fixed in the needle hollow body. Palmisano et al. reported on a dual (side-by-side) Pt electrode amperometric biosensor for the simultaneous monitoring of glucose and lactate.102 The surface coating (based on electropolymerized overoxidized polypyrrole film) resulted in excellent selectivity and no cross talk. Wang and Zhang developed a needle-type sensor for the simultaneous continuous monitoring of glucose and insulin.103 The integrated microsensor consisted of dual electrocatalytic (RuOx) and biocatalytic (GOx) modified carbon electrodes inserted into a needle (Figure 10) and responded independently to nanomolar and millimolar concentrations of insulin and glucose, respectively.

The enormous activity in the field of glucose biosensors is a reflection of the major clinical importance of the topic. Such huge demands for effective management of diabetes have made the disease a model in developing novel approaches for biosensors. Accordingly, for nearly 50 years we have witnessed tremendous progress in the development of electrochemical glucose biosensors. Elegant research on new sensing concepts, coupled with numerous technological innovations, has thus opened the door to widespread applications of electrochemical glucose biosensors. Such devices account for nearly 85% of the world market of biosensors. Major fundamental and technological advances have been made for enhancing the capabilities and improving the reliability of glucose measuring devices. Such intensive activity has been attributed to the tremendous economic prospects and fascinating research opportunities associated with glucose monitoring. The success of glucose blood meters has stimulated considerable interest in in-vitro and in-vivo devices for monitoring other physiologically important compounds. Similarly, new materials (membranes, mediators, electrocatalysts, etc.) and concepts, developed originally for enhancing glucose biosensors, now benefit a wide range of sensing applications. Despite the impressive progress in the development of glucose biosensors, the promise of tight diabetes management has not been fulfilled, and there are still many challenges and obstacles related to the achievement of a highly stable and reliable continuous glycemic monitoring. Such monitoring of moment-to-moment changes in blood glucose concentrations is expected to lead to a substantial improvement in the management of diabetes. The motivation of providing such tight diabetes control thus remains the primary focus of many researchers. Clearly, success in this direction demands a detailed understanding of the underlying biochemistry, physiology, surface chemistry, electrochemistry, and material chemistry. Yet, the ultimate implementation of the new devices may rely on commercial and legal considerations rather than scientific ones. As this field enters its fifth decade of intense research, we expect significant efforts that couple the fundamental sciences with technological advances. This stretching of the ingenuity of researchers will result in advances including the use of nanomaterials for improved electrical contact between the redox center of GOx and electrode supports, enhanced “genetically engineered” GOx, new “painless” invitro testing, artificial (biomimetic) receptors for glucose, advanced biocompatible membrane materials, the coupling of minimally invasive monitoring with compact insulin delivery system, new innovative approaches for noninvasive monitoring, and miniaturized long-term implants. In addition to minimally invasive and noninvasive glucose monitoring, efforts continue toward the development of chronically implanted devices (aimed at functioning reliably for periods of 6-12 months). These and similar developments will greatly improve the control and management of diabetes. The concept of a feedback loop (sensing-delivery) system goes beyond diabetes monitoring. Such ability to deliver an optimal therapeutic dose in response to distinct changes in the body chemistry of each person offers a unique opportunity to deliver personalized medical care and dramatically change the treatment of other diseases through tailored administration of drugs.77,78

824 Chemical Reviews, 2008, Vol. 108, No. 2

10. Acknowledgments I am grateful to all my students, post-docs, visiting scholars, and collaborators for their wonderful contributions to our electrochemical biosensor program. This research was supported by grants from the NSF and NIH.

11. References (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) (29) (30) (31) (32) (33) (34) (35) (36) (37) (38) (39) (40) (41) (42) (43) (44) (45) (46) (47)

Reach, G.; Wilson, G. S. Anal. Chem. 1992, 64, 381A. Wang, J. Electroanalysis 2001, 13, 983. Clark, L., Jr.; Lyons, C. Ann. NY Acad. Sci. 1962, 102, 29. Clark, L., Jr. U.S. Patent 33,539,455, 1970. Updike, S.; Hicks, G. Nature 1967, 214, 986. Guilbault, G.; Lubrano, G. Anal. Chim. Acta 1973, 64, 439. Schlapfer, P.; Mindt, W.; Racine, P. Clin. Chim. Acta 1974, 57, 283. Albisser, A.; Lebel, B.; Ewart, G.; Davidovac, Z.; Botz, C.; Zingg, W. Diabetes 1974, 23, 397. Shichiri, M.; Yamasaki, Y.; Hakui, N.; Abe, H. Lancet 1982, 2, 1129. Cass, A.; Davis, G.; Francis, G.; Hill, H. A.; Aston, W.; Higgins, I. J.; Plotkin, E.; Scott, L.; Turner, A. P. Anal. Chem. 1984, 56, 667. Hill, H. A. O. Eur. Pat. Appl. EPO 125,139 A2, 14, 45-46, 1984. Frew, J.; Hill, H. A. Anal. Chem. 1987, 59, 933A. Hilditch, P.; M.; Green, M. Analyst 1991, 116, 1217. Matthews, D.; Holman, R.; Brown, E.; Streemson, J.; Watson, A.; Hughes, S. Lancet 1987, 2, 778. Murray, R. W.; Ewing, A.; Durst, R. Anal. Chem. 1987, 59, 379A. Degani, Y.; Heller, A. J. Phys. Chem. 1987, 91, 1285. Ohara, T.; Rajogopalan, R.; Heller, A. Anal. Chem. 1994, 66, 2451. Willner, I.; Heleg-Shabtai, V.; Blonder, R.; Katz, E.; Tao, G. J. Am. Chem. Soc. 1996, 118, 10321. Xiao, Y.; Patolsky, F.; Katz, E.; Hainfield, J.; Willner, I. Science 2003, 299, 1877. Bartlett, P. N.; Booth, S.; Carauana, D. J.; Kilburn, J. D.; Santamaria, C. Anal. Chem. 1997, 69, 734. Bindra, D.; Zhang, Y.; Wilson, G. S.; Sternberg, R.; Thevenot, D. R.; Reach, G.; Moatti, D. Anal. Chem. 1991, 63, 1692. Csoregi, E.; Schmidtke, D. W.; Heller, A. Anal. Chem. 1995, 67, 1240. Henry, C. Anal. Chem. 1998, 70, 594A. Schmidtke, D.; Freeland, A.; Heller, A.; Bonnecaze, R. Proc. Natl. Acad. Sci. U.S.A. 1998, 95, 294. Sasso, S.; Pierce, R.; Walla, R.; Yacynych, A. Anal. Chem. 1990, 62, 1111. Emr, S.; Yacynych, A. Electroanalysis 1995, 7, 913. Malitesta, C.; Palmisano, F.; Torsi, L.; Zambonin, P. Anal. Chem. 1990, 62, 2735. Palmisano, F.; Centonze, D.; Guerrieri, A.; Zambonin, P. G. Biosens. Bioelectron. 1993, 8, 393. Sternberg, R.; Bindra, D.; Wilson, G. S.; Thevenot, D. R. Anal. Chem. 1988, 60, 2781. Moussy, F.; Jakeways, S.; Harrison, D. J.; Rajotte, R. V. Anal. Chem. 1994, 66, 3882. Wang, J.; Wu, H. Anal. Chim. Acta 1993, 283, 683. Zhang, Y.; Hu, Y.; Wilson, G. S.; Moatti-Sirat, D.; Poitout, V.; Reach, G. Anal. Chem. 1994, 66, 1183. Wang, J.; Liu, J.; Chen, L.; Lu, F. Anal. Chem. 1994, 66, 3600. Newman, J.; White, S.; Tothill, I.; Turner, A. P. Anal. Chem. 1995, 67, 4594. Sakslund, H.; Wang, J.; Lu, F.; Hammerich, O. J. Electroanal. Chem. 1995, 397, 149. Karaykin, A.; Gitelmacher, O.; Karaykina, E. Anal. Chem. 1995, 67, 2419. Karaykin, A. Electroanalysis 2001, 13, 813. Karyakin, A. A.; Karyakina, E. E.; Gorton, L. J. Electroanal. Chem. 1998, 456, 97. Lukachova, L. V.; Kotelnikova, E. A.; D’Ottavi, D.; Shkerin, E. A.; Karyakina, E. E.; Moscone, D.; Palleschi, G.; Currulli, A.; Karyakin, A. A. Bioelectrochemistry 2002, 55, 145. Zhang, X.; Wang, J.; Ogorevc, B.; Spichiger, U. Electroanalysis 1999, 11, 945. O’Halloran, M. P.; Pravda, M.; Guilbault, G. G. Talanta 2001, 55, 605. Wang, J.; Wu, H. J. Electroanal. Chem. 1995, 395, 287. Wang, J. Electroanalysis 2005, 17, 7. Wang, J.; Musameh, M. Analyst 2004, 129, 1. Ruzgas, T.; Csoregi, E.; Emneus, J.; Gorton, L.; Marko Varga, G. Anal. Chim. Acta 1996, 330, 123. Hrapovic, S.; Liu, Y. L.; Male, K. B.; Luong, J. H. T. Anal. Chem. 2004, 76, 1083. Gough, D.; Lucisano, J.; Tse, P. Anal. Chem. 1985, 57, 2351.

Wang (48) Armour, J.; Lucisano, J.; Gough, D. Diabetes 1990, 39, 1519. (49) Wang, J.; Lu, F. J. Am. Chem. Soc. 1998, 120, 1048. (50) Wang, J.; Mo, J. W.; Li, S. F.; Porter, J. Anal. Chim. Acta 2001, 441, 183. (51) D’Costa, E.; Higgins, I.; Turner, A. P. Biosensors 1986, 2, 71. (52) Willner, I.; Katz, E. Angew. Chem., Int. Ed. 2000, 39, 1180. (53) Guindilis, A.; Atanasov, P.; Wilkins, E. Electroanalysis 1997, 9, 661. (54) Willner, I.; Katz, E.; Willner, B. Electroanalysis 1997, 9, 965. (55) Pishko, M. V.; Katakis, I.; Lindquist, S. E.; Ye, L.; Gregg, B. A.; Heller, A. Angew. Chem., Int. Ed. 1990, 29, 82. (56) Ohara, T. J.; Rajagopalan, R.; Heller, A. Electroanalysis 1994, 66, 2451. (57) Riklin, A.; Katz, E.; Willner, I.; Stocker, A.; Buckmann, A. F. Nature 1995, 376, 672. (58) Xiao, Y.; Patolsky, F.; Katz, E.; Hainfeld, J. F.; Willner, I. Science 2003, 299, 1877. (59) Patolsky F.; Weizmann Y.; Willner I. Angew. Chem., Int. Ed. 2004, 43, 2113. (60) Liu, J. Q.; Chou, A.; Rahmat, W.; Paddon-Row, M. N.; Gooding, J. J. Electroanalysis 2005, 17, 38. (61) Gooding, J. J.; Wibowo, R.; Liu, J. S.; Yang, W.; Losic, D.; Orbons, S.; Mearns, F. J.; Shapter, J. G.; Hibbert, D. B. J. Am. Chem. Soc. 2003, 125, 9006. (62) Albery, W.; Bartlett, P. N.; Craston, D. H. J. Electroanal. Chem. 1985, 194, 223. (63) Khan, G. F.; Ouwa, M.; Wernet, W. Anal. Chem. 1996, 68, 2939. (64) Palmisano, F.; Zambonin, P. G.; Centonze, D.; Quinto, M. Anal. Chem. 2002, 74, 5913. (65) Cenas, N. K.; Kulys, J. Bioelectrochem. Bioenerg. 1981, 8, 103. (66) Yabuki, S.; Shinohara, H.; Aizawa, M. J. Chem. Soc., Chem. Commun. 1989, 945. (67) Koopal, C. G. J.; de Ruitter, B.; Notle, R. J. M. J. Chem. Soc., Chem. Commun. 1991, 1691. (68) Jing, W.; Yang, Q. Anal. Bioanal. Chem. 2006, 385, 1330. (69) Besteman, K.; Lee, F.; Wiertz, F.; Heering, H.; Dekker, C. Nano Lett. 2003, 3, 727. (70) Forzani, E. S.; Zhang, H. Q.; Nagahara, L.; Amlani, I.; Tsue, R.; Tao, N. J. Nano Lett. 2004, 4, 1785. (71) Bartlett, P. N.; Birkin, P. R. Anal. Chem. 1994, 66, 1552. (72) Turner, A. P. F.; Chen, B.; Piletsky, S. A. Clin. Chem. 1999, 45, 1596. (73) Newman, J. D.; Turner, A. P. F. Biosens. Bioelectron. 2005, 20, 2388. (74) Wring, S.; Hart, J. Analyst 1992, 117, 1281. (75) Kirk, J.; Rheney, C. J. Am. Pharm. Assoc. 1998, 38, 210. (76) Wilson, G. S.; Gifford, R. Biosens. Bioelectron. 2005, 20, 2388. (77) Heller, A. AIChE J. 2005, 51, 1054. (78) Deo, S.; Moschou, E.; Peteu, S.; Daunert, S.; Eisenhardt, P.; Madou, M. Anal. Chem. 2003, 75, 207A. (79) Vadgama, P.; Desai, M.; Crump, P. Electroanalysis 1991, 3, 597. (80) Frost, M.; Meyerhoff, M. E. Anal. Chem. 2006, 78, 7371. (81) Shin, J. H.; Marxer, S. M.; Schoenfisch, M. H. Anal. Chem. 2004, 76, 4543. (82) Gifford, R.; Batchelor, M. M.; Lee, Y.; Gokulrangan, G.; Meyerhoff, M. E.; Wilson, G. S. J. Biomed. Mater. Res., Part A 2005, 75A, 755. (83) Oh, B. K.; Robbines, M. E.; Nablo, B. J.; Schoenfisch, M. H. Biosen. Bioelectron. 2005, 21, 749. (84) Chapman, R. G.; Ostuni, E.; Liang, M. N.; Meluleni, G.; Kim, E.; Yan, L.; Pier, G.; Warren, H. S.; Whitesides, G. M. Langmuir 2001, 17, 1225. (85) Quinn, C. A.; Connor, R. E.; Heller, A. Biomaterials 1997, 18, 1665. (86) Moussy, F.; Harrison, D. J.; O’Brien, D. W.; Rajotte R. V. Anal. Chem. 1993, 65, 2072. (87) Wang, J.; Chen, L.; Hocevar, S. B.; Ogorevc, B. Analyst 2000, 125, 1431. (88) Csoregi, E.; Quinn, C. P.; Schmidtke, D. W.; Lindquist, S. E.; Pishko, M. V.; Katakis, I.; Hubbel, J. A.; Heller, A. Anal. Chem. 1994, 66, 3131. (89) Choleau, C.; Klein, J. C.; Reach, G.; Aussedat, B.; Demaria-Pesce, V.; Wilson, G. S.; Gifford, R.; Ward, W. K. Biosens. Bioelectron. 2002, 17, 641. (90) Gross, T. M. Diabetes Technol. Ther. 2000, 2, S19. (91) Feldman, B.; Brazg, R.; Schwartz, S.; Weinstein, R. Diabetes Technol. Ther. 2003, 5, 769. (92) Pickup, J. C.; Hussain, F.; Evans, N. D.; Sachedina, N. Biosens. Bioelectron. 2005, 20, 1897. (93) Chen, T.; Calabrese Barton, S.; Binyamin, G.; Gao, Z.; Zhang, Y.; Kim, H.; Heller, A. J. Am. Chem. Soc. 2001, 123, 8630. (94) Tierney, M.; Kim, H.; Tamada, J.; Potts, R. Electroanalysis 2000, 12, 666. (95) Langermann, M.; Mendlelowitsch, A.; Landolt, H.; Alessandri, B.; Gratzl, O. Clin. Neurolog. Neurosurg. 1995, 97, 149.

Electrochemical Glucose Biosensors (96) Vering, T.; Adam, S.; Drewer, H.; Dumschat, C.; Steinkuhl, R.; Schulze, A.; Siegel, E. G.; Knoll, M. Analyst 1998, 123, 1605. (97) Bayer, U.; Schafer, D.; Thomas, A.; Aulich, H.; Hauter, U.; Reihl, B.; Ehwald, R. Diabetologia 2001, 44, 416. (98) Freckmann, G.; Kalatz, B.; Pfeiffer, B.; Hoss, U.; Haug, C. Exp. Clin. Endocrinol. Diabetes 2001, 109, S347. (99) Rhemrev-Boom, R. M.; Tiessen, R. G.; Jonker, A. A.; Venema, K.; Vadgama, P.; Korf, J. Clin. Chim. Acta 2002, 316, 1.

Chemical Reviews, 2008, Vol. 108, No. 2 825 (100) Maran, A.; Crepaldi, C.; Tiengo, A., et al. Diabetes Care 2002, 25, 347. (101) Yang, Q. L.; Atanasov, P.; Wilkins, E. Electroanalysis 1998, 10, 752. (102) Palmisano, F.; Rizzi, R.; Centonze, D.; Zambonin, P. G. Biosens. Bioelectron. 2000, 15, 531. (103) Wang, J.; Zhang, X. Anal. Chem. 2001, 73, 844.

CR068123A

Chem. Rev. 2008, 108, 826−844

826

ISFET and Fiber Optic Sensor Technologies: In Vivo Experience for Critical Care Monitoring Bruce A. McKinley* Department of Surgery, The Methodist Hospital, Houston, Texas 77030 Received June 14, 2007

Contents 1. Introduction 2. ISFET Sensor Technology 2.1. Early Prototypes and Concept Realization 2.2. Summary of the First Decade of ISFET Sensor Technology Development 2.3. Summary of ISFET Sensor Development Since 1980 3. Fiber Optic Sensor Technology 3.1. Development and Commercialization 3.2. Clinical Trials and Assessments 3.3. Issues Identified with Continuous Intraarterial Blood Gas Monitoring 3.4. Interstitial Fluid Monitoring Using Fiber Optic Sensor Systems 3.5. Summary of Fiber Optic Sensor System Development for in vivo Monitoring 4. Factors Affecting the Future of Discrete Sensor Development for in vivo Monitoring 4.1. Continuous IF Monitoring 4.2. Technology Problems 5. Summary 6. Conclusion 7. References

826 826 827 831 831 832 834 834 836 836 837 837 839 840 841 842 842

1. Introduction Technologies to provide measurements of (human) biochemistry and physiology developed with analytical chemistry techniques, and with understanding of physiology. Sensors to monitor a microenvironment within a living organism have fascinated physicians and physiologists.1-18 The capability to monitor some aspect of physiologic process began to appear with in vitro electrochemical sensors for the clinical laboratory. Discrete sensors useful for in vivo monitoring emerged with the earliest in vitro electrochemical sensors.19 A discrete sensor, configured as a tethered probe, placed in or near a local environment of interest to provide continuous measurement of the sensed variable, has been an enduring concept.1-13,20-29 In addition to solid-state temperature and pressure, electrochemistry-based sensors for PO2, pH, and other specific ions, e.g., K+ and Ca++, were early examples. Medical applications of sensor technologies have been seen as major market opportunities, and have had * To whom correspondence should be addressed. E-mail: BAMcKinley@ tmhs.org.

Bruce A. McKinley is a faculty member of The Methodist Hospital, Department of Surgery, Houston, TX, where he is Director of Translational Research and Bioengineering, Division of Surgical Critical Care and Acute Care Surgery. He received his doctoral degree in Bioengineering from the University of Utah, Salt Lake City, UT, in 1980. He received his M.S. degree in Biomedical Engineering from Iowa State University, Ames, IA, and undergraduate training in Chemical Engineering from Lehigh University, Bethlehem, PA. His work in medical applications of sensors began with his Ph.D. program at the University of Utah, which involved collaboration with electrochemists, bioengineers, and surgeons, under the direction of Drs. Janata, Westenskow, Houtchens, and Moody, to demonstrate the capability of monitoring interstitial electrolytes during shock and resuscitation using early prototype ISFET sensor technology. As a faculty member at the University of Texas, this work continued using then recently developed fiber optic sensor technology with early demonstration of monitoring of brain parenchyma and skeletal muscle pH, PCO2, and PO2 in studies of trauma, shock, and resuscitation and monitoring of the interstitium and microperfusion during shock and resuscitation using infrared spectrometry systems to monitor tissue hemoglobin O2 saturation. Currently, he is involved with computerized clinical decision support as a technique to enable effective clinical research and improve patient care.

varied development efforts. Development and market success, however, has been elusive. In this paper the author reviews these efforts for two innovative sensor technologies that appeared in recent decades: ISFET sensors and fiber optic (‘optode’) sensors. By realizing the concepts of discrete sensors with micrometer dimensions, these developments effectively revolutionized sensor technology. These developments are primary examples of medical sensor technology that continues to evolve. The markets foreseen for these discrete sensors remain as possibilities.

2. ISFET Sensor Technology The ISFET innovation is a generally applicable discrete sensor technology for different electrode sensors of micrometer dimensions. The innovation of the ISFET was

10.1021/cr068120y CCC: $71.00 © 2008 American Chemical Society Published on Web 01/08/2008

ISFET and Fiber Optic Sensor Technologies

integrated circuit (IC) based (i.e., solid state) electrode sensors with micrometer dimensions, physical ruggedness, and possibility of manufacture with extreme precision and reproducibility previously unavailable with other electrode technologies. On the basis of emerging IC technology with matrices of independent transistor devices occupying a few square millimeters, the ISFET sensor concept offered the potential of a ‘clinical laboratory on a chip’ able to be placed safely in a patient. The possible application of continuous monitoring of multiple physiologically important variables was quickly identified. The theory of ISFET function has been described elsewhere.30-37 The ISFET concept was based on planar metal oxide semiconductor (MOS) field effect transistor (FET) technology that was being commercialized in the 1970s, primarily as computer memory IC devices. The three material ‘metal oxide semiconductor’ structure refers to the physical structure of early (and, interestingly, now the very latest) FETs having a metal (aluminum) gate electrode overlying an oxide insulator overlying a semiconductor material. FET device architecture, fabrication methods, and chip surface wire connections were based on the then rapidly developing IC industry technology. The MOSFET structure, metal surface gate overlying insulator and semiconductor with drain and source connections, had been adequately characterized to permit reliable device fabrication. Dimensions of MOSFETs could be specified and a process developed to reliably produce matrices of hundreds of identical devices on a single silicon wafer. The simple MOSFET device modification to create the ISFET was to remove the gate surface metallization and connection and interpose an aqueous solution and reference electrode. Effectively, the reference electrode specified a gate potential to cause the ISFET to function in the nonsaturated region of semiconductor current flow, wherein drain to source current varies directly with gate voltage. Those variables detected using potentiometric electrode sensor systems would, in concept, use the same reference potential and a single reference electrode. Specific variables would be sensed using ion-selective ‘membranes’ specifically selective for ions of interest deposited on a planar integrated circuit electrode surface. Such membranes were the basis of ion-selective electrode technology.38,39 (Nonionic species would be sensed using other selective mechanisms.) The individual sensors would have precise micrometer dimensions and be reproducibly manufacturable.

2.1. Early Prototypes and Concept Realization Several efforts began in the 1970s that demonstrated sensor function and the possibility of a new sensor technology based on IC devices and technology.30,32-37,40-43 ISFET chips were designed and fabricated using state of the art materials and methods.31,34 The multidisciplinary nature of the ISFET brought together several disciplines that had rarely collaborated. The ISFET innovation involved then new electronics engineering with focus on integrated circuit (IC) and semiconductor device design and fabrication, and attracted the attention of classically trained electrochemists.35,36 The concept of multiple potentiometric sensors of micrometer dimensions as an IC ‘chip’ recalled the idea of in vivo monitoring of multiple variables simultaneously. This idea attracted the attention of surgical intensivists. Bioengineering, a new collaborative discipline, provided the science and engineering expertise to identify and solve medical instru-

Chemical Reviews, 2008, Vol. 108, No. 2 827

mentation, materials, and monitoring problems that this new sensor concept engendered. Important advances were made and continued throughout the 1970s and early 1980s in demonstration of ISFET sensors and useable probe designs at several laboratories.30-32,43-50 Uniquely at the University of Utah (Salt Lake City, UT), science, engineering, and medical expertise combined in the mid 1970s to focus on the ISFET concept as it might be applied to intensive care. As a result of multidisciplinary collaboration, concepts were rapidly realized and early prototype sensors demonstrated. Original concepts appeared and were proven in the 1970s as large-scale IC fabrication facilities were developing to support the emerging digital computer mass market. The first ISFET sensor to be demonstrated in an electrochemistry/ engineering laboratory setting was hydronium ion (pH).31 Analogous to conventional glass electrodes with SiO2 H+-selective ‘membranes’, the pH ISFET had similar selectivity and sensitivity. In addition to pH, ISFET sensor prototypes demonstrated the capability of other then available ion-selective electrodes. ISFET sensor prototypes were demonstrated for potassium (K+), calcium (Ca++), sodium (Na+), ammonium (NH4+), and fluoride (F-) ions. 34,36,37 These potentiometric electrode sensors borrowed ion-selective polymeric membranes in extremely small volumes, which were incorporated in the ISFET structure overlying the thin gate region. Focus on those variables of probable broadest clinical application led to development and demonstration of probes and techniques for sensor placement in anatomic sites likely to provide an indication of physiologic function. Studies to investigate possible applications of these probes were typically done in animal models of shock and resuscitation. A probe design that was used for early studies at the University of Utah is shown in Figure 1. To the author’s knowledge, the first clinically relevant demonstration of polymeric membrane K+ ISFET function involved intravenous placement of a K+ ISFET probe and monitoring during IV infusion of KCl, as is commonly done for ICU patients with hypokalemia as part of standard electrolyte replacement therapy.41 This first demonstration showed stable function of the sensor system placed in the region of the superior vena cava for more than 3 h. Data obtained from in vitro clinical laboratory analysis of serial blood samples confirmed the accuracy of the ISFET sensor system. This and other experiments demonstrated the feasibility and desirability of continuous monitoring of electrolyte variables in physiologic preparations that were very representative of intensive care experiences. These situations are still common and representative of intensive care today. Of early studies investigating possible applications of K+ ISFET sensors, those involving sensor placement and monitoring in the interstitial fluid (IF) space of certain tissues (i.e., extravascular, extracellular tissue space) had perhaps the greatest foresight. Animal models of hemorrhagic and burn shock showed clear indication of severe derangement of basic cellular physiology, as monitored by K+ ISFET sensors placed in skeletal muscle tissue.41,51 A few years prior to these studies, other basic aspects of cellular physiology during shock and resuscitation had been reported by Shires et al. 2,13,52-54 In these studies, derangements of skeletal muscle cellular membrane potentials and K+ transport during hemorrhagic shock in primates were measured and reported to be proportional to duration and severity of shock. During this same decade, these processes were able to be monitored with new proximity to cellular physiology using prototype

828 Chemical Reviews, 2008, Vol. 108, No. 2

McKinley

Figure 1. Early ISFET probe design used for studies of medical applications in laboratory animal models of hemorrhagic shock and calcium ion manipulation. Chip with two ISFET devices is shown with wire bonds to prototype copper wire tip connections, prior to encapsulation with epoxy compound. A reference electrode (Ag/AgCl) is depicted in the upper lumen of the catheter. The exposed Ag/ AgCl electrode was isolated from protein exposure using a distal hydrogel plug over the catheter port. (Reprinted with permission from ref 51. Copyright 1981 Lippincott Williams & Wilkins.)

ISFET sensors placed adjacent to viable cells in an intact tissue bed. ISFET K+ probes were also configured to study intracellular dynamics of K+.55 The ability to monitor the IF space had been investigated by others using then conventional glass pH and polymeric membrane ion-selective electrode technology, but investigators were typically limited to muscle surface probe placement.3,9,14,56 The ISFET probe, incorporating a reference electrode in the same device, provided the capability for minimally invasive, clinically practical placement in viable skeletal muscle beds. A study to investigate the use of ISFET K+ sensor probes for skeletal muscle IF K+ activity during hemorrhagic shock and resuscitation found a reproducible, near linear increase of IF K+ activity during the period of hemorrhage-induced hypotension.41,51 This finding was consistent with other reports of studies of hemorrhagic shock, including seminal studies of Shires et al.,13,52,54 and indicated progressive leakage of intracellular K+ to the interstitial space and failure of extracellular Na+-intracellular K+ gradient. After resuscitation, incomplete recovery of IF K+ was observed, an effect perhaps indicative of inadequate resuscitation. Figure 2a shows results of an animal experiment involving hemorrhagic shock and resuscitation. Two ISFET K+ probes, one with central venous placement and one with medial thigh skeletal muscle placement, monitored K+ activity in blood and IF simultaneously before, during, and after the shock insult (controlled hemorrhage to maintain hypotension with MAP ≈ 40 mmHg for 1 h). Interstitial and blood K+ activities differed significantly in this study. Changes in blood K+ activity were insignificant during the 1 h of hemorrhagic hypotension. Figure 2b shows composite results from a series of 10 experiments performed using the same protocol. During shock, deterioration of cellular oxidative metabolism, indicated by IF K+ activity increase, compared to other systemic variables that were monitored, including blood K+ activity, blood pressure, and heart rate, was especially noteworthy for the following reasons: (1) Blood pressure, a primary variable used to judge (remaining) vascular volume, has

compensatory mechanisms that can mask severity of shock or inadequacy of resuscitation. (2) Heart rate and increased extraction of O2 from blood also compensate for decreased total vascular volume. (3) Skeletal muscle seems to be less sensitive than other tissues to an acute decrease in O2 delivery, and trans membrane potential is maintained despite a decrease of muscle surface pH, possibly due to glycogen stores and anaerobic metabolism. ISFET sensors for Ca++ activity was a capability demonstrated with a polymeric Ca++-selective membrane placed over the thin gate region of the ISFET device36,37 (see Figure 1). A study was designed to investigate possible transients in systemic and IF Ca++ activity.57 In dose rate-dependent response, myocardial depression was observed with IV infusion of sodium citrate. Blood Ca++ was monitored continuously with a Ca++ ISFET placed in the superior vena cava and compared with serial blood sample Ca++ activity analyses using a clinical laboratory ion-selective electrode analyzer. Figure 3 depicts results from a series of eight experiments. Blood Ca++ activity was observed to decrease to ∼1 mEq/L with a dose of citrate administered equivalent to that contained in four units of citrate-preserved blood cells. Myocardial depression, indicated by severely decreased cardiac output, blood pressure, and Ca++ activity, and increased pulmonary capillary wedge pressure, recovered but remained depressed for >30 min following bolus infusion. It is still common for multiple units of citrate-preserved packed red blood cells (PRBC) to be given rapidly to severely injured patients who are in shock due to hemorrhage upon hospital arrival, during emergency surgery, and during subsequent resuscitation and likely contributes to myocardial depression that can be sustained depending on the dose of citrate. At Level 1 trauma centers in the United States, replacement of lost hemoglobin to regain and sustain systemic O2 delivery and hemodynamic function often involves volumes of PRBC that exceed 10 units in the first hours of hospitalization.58 Further demonstration of the ability to place a small ISFET sensor in an IF compartment involved direct monitoring of

ISFET and Fiber Optic Sensor Technologies

Chemical Reviews, 2008, Vol. 108, No. 2 829

Figure 2. (a) Single hemorrhagic shock experiment results showing continuously monitored interstitial fluid (IF) K+ activity in skeletal muscle of medial thigh using ISFET K+ sensor (in probe design depicted Figure 1), continuously monitored blood (serum, S) K+ activity in superior vena cava using ISFET K+ sensor (in probe design depicted Figure 1), and independent serial clinical laboratory analyses of blood (serum, S, off line) K+ activity. Mean arterial pressure and cardiac output measurements depict hemodynamic performance in response to hemorrhage (B), maintained hypotension (shock), and reinfusion of shed blood (R). IF K+ was observed to increase steadily during shock and decrease with reinfusion of shed blood, while S K+ changed insignificantly. Agreement of ISFET and off-line S K+ measurements was also noteworthy. (Reprinted with permission from ref 51. Copyright 1981 Lippincott Williams & Wilkins.) (b) Composite data from 10 hemorrhagic shock experiments showing (mean ( SD) continuously monitored interstitial fluid (IF) K+ activity in skeletal muscle of medial thigh using ISFET K+ sensor (15 min intervals), continuously monitored blood (serum, S) K+ activity in superior vena cava using ISFET K+ sensor (15 min intervals), and independent serial clinical laboratory analyses of blood (serum, S, off line) K+ activity. With hemorrhage to produce hypotension (shock), a monotonic increase in IF K+ activity and an insignificant increase in S K+ activity were observed; return to near baseline was observed upon reinfusion of shed blood. (Reprinted with permission from ref 51. Copyright 1981 Lippincott Williams & Wilkins.)

myocardial IF Ca++ activity.59 With appropriate surgical exposure of the heart, a Ca++ ISFET sensor was placed under the epicardial membrane of the left ventricle of an anesthetized dog without disturbing cardiac function. Ca++ activity was manipulated by IV infusion of, first, sodium citrate, to cause hypocalcemia, and then calcium chloride, to cause hypercalemia. Serial femoral artery blood samples were analyzed using a clinical laboratory ion-selective electrode

analyzer. The expected responses in the myocardium and blood were recorded in real time. Blood Ca++ activity was less than IF Ca++ activity during citrate infusion, and the opposite comparison was observed during calcium chloride infusion. Externally imposed changes in blood concentrations were in these experiments, the ‘forcing function’, with IF Ca++ responding with possible moderating effect of phosphorylation-related mechanisms in myocardium to maintain

830 Chemical Reviews, 2008, Vol. 108, No. 2

McKinley

Figure 3. Composite data from eight experiments to study cardiovascular depression associated with sodium citrate (blood preservative agent) administration to simulate rapid blood transfusion (mean ( SEM): off-line (clinical laboratory) analyses of Ca++ activity (solid dots, - SEM); continuous central venous blood ISFET Ca++ activity (open dots, + SEM); cardiac output, mean arterial pressure, pulmonary artery pressure; pulmonary artery wedge pressure (all - SEM). Cardiovascular depression was observed to be directly proportional to sodium citrate dose rate and inversely proportional to Ca++ activity. Rapid return of Ca++ activity was observed upon cessation of sodium citrate infusion, likely due to citrate metabolism and release of Ca++ ion, correlating with recovery of normal hemodynamic performance. Agreement of ISFET and off-line S Ca++ measurements was also noteworthy. (Reprinted with permission from ref 57. Copyright 1981 Lippincott Williams & Wilkins.)

stable Ca++ activity in perimyofibrillar IF, despite perfusion of the muscle tissue bed with hypo- or hyper-calcemic blood. The concept of multiple ISFET sensors on a single IC chip was also realized by the late 1970s.40,59 IC designs for two ISFET thin gate devices and two conventional MOSFET devices were fabricated (see Figure 1). Individual device dimensions and spacing permitted ISFET sensor encapsulation using previously described techniques. ISFET probes were fabricated with two independently functioning ISFET sensors: one for H+ and the other for K+. Analog instrumentation was designed to operate the two sensors simultaneously with biasing achieved using a single reference electrode, also incorporated in the probe. Calibration of both sensors used two solutions of known H+ and K+ activities and at two different temperatures to permit temperature compensation using a third device of the same IC, a conventional MOSFET. With this ISFET probe placed in a surgically exposed gastrocnemius muscle, IF pH and K+ activity in this skeletal muscle of an anesthetized, mechanically ventilated rabbit were monitored during and after ischemia, induced by temporarily occluding the aorta ∼1 cm proximal to the aortic

bifurcation to avoid collateral blood supply.59 Arterial blood samples were obtained for comparison with clinical laboratory ion-selective electrode analyses. During and after ∼1 h of ischemia, mirror image symmetry in fluctuations of IF pH and K+ activity was observed. Skeletal muscle IF pH decreased and K+ activity increased during ischemia. With release of aortic clamp and reperfusion, recovery toward preischemia baseline occurred. As with studies of IF K+ activity during hemorrhagic shock, recovery to pre-ischemia IF pH or K+ did not occur 2 h after the insult. IF pH and K+ activity changed with ischemia and reperfusion with a time course that was similar to the canine IF K+ activity during and after hemorrhagic shock. Reperfusion effects of pH, showing IF pH to differ more from blood (systemic) pH than K+ activity, demonstrated what is now more commonly understood to be ‘washout’ of acidemic products of anaerobic metabolism from a large vascular bed after ischemia. Close agreement of IF and systemic K+ throughout a 5 h experiment is remarkable. These changes had been observed previouslysbut separatelysby other investigators using similar laboratory preparations and conventional ion-selective (and pH) elec-

ISFET and Fiber Optic Sensor Technologies

trode technology. Simultaneous recordings of skeletal muscle IF pH and K+ activity at the same site had not been reported. This demonstration of multisensor capability further proved the original ISFET concept of multiple sensors on a single IC and configured in a single, small rugged probe.

2.2. Summary of the First Decade of ISFET Sensor Technology Development The ISFET as a sensor technology with potential versatility and broad application was recognized in the mid 1970s. In addition to pH, ISFET sensor prototypes demonstrated the capability of other then available ion-selective electrodes. ISFET sensor prototypes were demonstrated for potassium, calcium, sodium, ammonium, and fluoride ions. Prototype sensors were able to be incorporated in probes with dimensions practical for use in direct, continuous monitoring of critically ill patients and in other applications that included dentistry. The ISFET concept progressed rapidly during the late 1970s. Development of theory, technology, and practical applications advanced simultaneously. The ISFET innovation of a generally applicable discrete sensor technology for different electrochemical sensors of micrometer dimensions was realized by 1980. IC-based ISFET electrode sensors with micrometer dimensions and physical ruggedness previously unavailable with other electrode technologies had been demonstrated. Those variables detected using potentiometric ion-selective electrode sensor systems had been fabricated and demonstrated as ISFET sensors. Multisensor probes had been designed and fabricated. Technical progress occurred at several centers in the United States, Europe, and Japan. In certain settings, the multidisciplinary nature of this innovative sensor technology benefited from electrochemistry, bioengineering, and medical (surgical intensivist) collaboration. Designs of practical ISFET sensor systems were developed, fabricated, and tested for clinically relevant applications. Early development of a new sensor technology had integrated electrochemistry and electronics engineering disciplines to produce prototype ISFET ICs. Design and development of a practical ISFET probe to optimally use the specific features of small size and ruggedness required electrochemistry, electronics, bioengineering, and medical collaboration. Fabrication of ISFET probes was done manually using materials and techniques in trial and error, typical of early engineering development. The culmination of these efforts was a very small discrete sensor system able to be placed in specific tissue beds of interest to monitor physiologic processes more directly and reliably than had been possible. The results of these early studies to demonstrate feasibility of this new sensor technology confirmed studies of tissue cellular physiology during hemorrhagic shock and resuscitation that had been reported in the early 1970s. In the United States, this early work was accomplished using scientific grant funding. The concept of reproducibly manufacturable ISFET sensors and probes commercially available for this purpose, however, was not proven. A new sensor technology seemed ready to be commercialized. Commercialization of ISFET sensor technology proved more elusive than may have been expected by those involved in early conceptualization, development, and demonstration of prototypes. In the United States and elsewhere, funding from commercial entities would be required.

Chemical Reviews, 2008, Vol. 108, No. 2 831

2.3. Summary of ISFET Sensor Development Since 1980 Developments have continued throughout recent decades that have demonstrated sustained interest in ISFET sensor technology.31 The early 1980s saw several exploratory efforts by major medical product corporations (Critikon (a Johnson & Johnson company), Corning Glass, U.S.; Cordis Europa, The Netherlands; Kuraray, Nihon Kohden, Japan) to develop ISFET sensors toward reproducible, easily manufactured prototypes.60-67 No U.S. commercial products for medical applications resulted from these efforts to the knowledge of the author, although ISFET pH-based PCO2 sensors were commercialized for medical applications in Japan (Nihon Kohden), and ISFET pH sensors were developed to manufactured prototypes in The Netherlands (Cordis Europa). ISFET pH sensors are manufactured and used for water, food, brewery, and pharmaceutical process monitoring (Mettler Toledo Inc., Columbus, OH; Honeywell International Inc., Morristown, NJ; Endress+Hauser Group, Reinach, CH). These current ISFET sensors exist as commercial products due to advantages of ruggedness, lack of breakable glass, and sterilizability compared to glass membrane electrodes, and the ability to manufacture a probe that meets the requirements of process monitoring. Similar advantages exist for continuous in vivo monitoring using ISFET sensors in critically ill patients but with probe dimensions that are much smaller than these commercially available process monitoring probes. Of note is the ability to manufacture ISFET pH sensors, the dimensions of which are similar to those of early ISFET prototypes. This capability involves encapsulation of ISFET IC devices without occlusion of ion-sensitive thin gate regions.45 A key to the problem of manufacture was encapsulation of parts of planar IC surfaces with ion-sensitive thin gate regions not occluded and able to be exposed to the aqueous solution of interest.31,45 Eventual hydration and leak between integrated circuit chip SiO2-exposed gate surface and encapsulation materials was a probable source of instability of the sensors. Long-term stability, though demonstrated over days and likely adequate for medical applications involving relatively short duration monitoring, was unlikely to sustain if packaging in aqueous solution was required. Many designs have appeared that addressed the problem of encapsulation.31,32,44,45,60-62,64,68-79 An early attempt to address manufacture adapted IC industry tape automated bonding, which provided printed (copper) electrical (‘beam’) leads in a pattern congruent with ISFET IC bonding pads. The ISFET IC (rectangular cross section; see Figure 1) was positioned between additional tape layers with appropriate adhesive or perforations to anchor and align and, with subsequent application of (thermoset) encapsulant, seal the IC, with the exception of thin gate regions left exposed as a pH sensor or for application of ion-selective membrane.64 Another approach used conventional thermoset epoxy encapsulant around an ISFET IC placed in a cast form with focused gas jet stream impinging over thin gate regions during curing to prevent flow of epoxy over thin gate regions that would remain exposed.61 Other reports involve photolithographic IC processing techniques to provide encapsulation as part of ISFET IC fabrication.68,80 For ISFET pH probe designs for laboratory and food industry process monitoring, manual encapsulation using thermoset epoxy has been devised and micromold techniques incorporating printed circuit electrical leads and elastomeric seals have been used.45,73

832 Chemical Reviews, 2008, Vol. 108, No. 2

A related problem encountered early with the Utah in vivo probe design was mechanical stability of the polymeric gel liquid ion-selective membrane placed over the thin gate region of the ISFET device. Convenient polyvinyl chloride and plasticizer formulations of K+ and Ca++ ionophore membrane solution permitted creation of a water-tight seal with an epoxy encapsulant well into which microliter solution volume was deposited using a micropipette and stereomicroscope, but such techniques were difficult to automate. Other methods were devised, e.g., suspended mesh of chemically inert, nonconductive, and mechanically stabilizing polymer,60 which was, in concept, generalizable to other partial encapsulation techniques. An alternative approach to ion-selective membranes involving functionalized polysiloxanes, able to be covalently linked to IC surfaces and combined with photolithographic techniques, was reported to provide K+, Ca++, and NO3- ISFET sensors.81 A key problem for a discrete ISFET sensor system for in vivo monitoring was stable reference potential necessarily provided by a reference electrode. Initial efforts to address this problem were undertaken by researchers at the University of Utah.40,82,83 A small liquid junction electrode was devised that used saline as internal reference electrolyte and excluded (blood) proteins from the Ag/AgCl equilibrium using a hydrogel barrier. Studies cited above were undertaken with this simple design, which proved adequate for relatively short-term (several hour) experiments. Although systematic study of the performance of reference electrodes for in vivo monitoring as a component of a discrete sensor system was not undertaken for the early studies cited, the problem was addressed as a component of commercial product developments.75,84-90 The problem of reference electrode design has had continued study as ion-selective electrodes became established as a part of clinical chemistry.91-97 Alternative methods have been devised to provide proper biasing of the ISFET device and theoretical (Nernstian) electrode response.31,82 A reference electrode that is integral to a probe is conceptually attractive. ‘On chip’ reference electrode designs were devised and tested. Various designs provided a liquid junction with micrometer dimensions but otherwise conventional metal halide electrode, and may have offered an adequate compromise between volume of internal reference electrolyte solution (adequate for reference potential stability and duration of probe use; minimal potential for toxic effect due to diffusion or accidental in situ disintegration), exclusion of proteins from the reference potential generating equilibrium reaction, and manufacturable design compatible with ISFET (IC) fabrication technique and medical product sterilization and storage/shelf life requirements.84-87,89,90,98 Stable reference potential is intended and described, but nonpatent literature reports of performance of such designs are unavailable. The versatility of the ISFET and chemically sensitive field effect transistor (ChemFET) concept became more apparent with innovations of the early 1980s. Reports continued to reflect the basic nature of the sensor technology.99-104 Recent reports indicate ongoing development in a wide variety of areas, including cell and tissue culture,105-108 transcutaneous blood glucose concentration monitoring,109 enhanced sensitivity glucose sensor,110 ISFET surface modification for immune function detection,111 urea sensor designs based on urease enzyme activity,112,113 organic thin film transistor biosensors,114 bacterial metabolism and bacteriology for food processing,115 single nucleotide polymorphism (SNP) detec-

McKinley

tion,116 protein detection using streaming or zeta potential for immunosensor or biocompatibility applications,117 and acetylcholinesterase-based ISFET for insecticide detection.118 Since 1976, nearly 400 U.S. patents have been issued that address ISFET technology.

3. Fiber Optic Sensor Technology A second discrete sensor technology had its inception during the 1970s. Fiber optic sensors for pH, PCO2, and PO2, initially reported in the mid 1970s,24,26,119,120 appeared to offer many of the same innovation advantages that had been envisioned for ISFET sensors, including extremely small size, reliability, and ease of manufacture. Sensors based on absorbance, fluorescence, and other optical mechanisms are fundamental to much of analytical chemistry. The innovation for discrete fiber optic sensors was incorporation of indicator chemistry at or near the tip of one or a pair of optical fibers. The technology was new and did not seem to involve the complexity of electrode sensors, most specifically a reference electrode. The technology was based on optical signals transmitted via submicrometer diameter optical fibers. Perceived advantages of optical sensors, compared with electrode sensors (including ISFET), included the following: 121 (1) No need for a reference electrode potential to obtain electrolyte (e.g., pH) measurement. (2) No susceptibility to electrostatic or electromagnetic interference. (3) Potential for better long-term stability in blood or other proteinaceous fluids, probably relating to reference electrode stability. (4) Possibility of multifunction probes of single optical fiber construction based on multiwavelength analysis of mixed indicator sensors and principal of noninteraction of photons of different wavelength sources. (5) Early success of prototype discrete sensor development. The discrete fiber optic sensor technology had its basis in fluorescence and absorbance chemistry.122 In concept, absorbance indicators react with the analyte of interest, e.g., H+, in a reversible equilibrium reaction and change the color of the indicator solution. Indicators, typically weak acids or bases, have different forms (ionized, tautomers) that coexist. The ionized form absorbs certain wavelengths of light. Absorbance of light transmitted through the indicator solution shifts when alterations in concentration or activity of the analyte cause a change in concentration of the indicator forms. Reflected or transmitted light is of the same wavelength or spectral range as incident light. The dye phenol red (phenolsulfonphthalein) is an example of an absorbance indicator that has been developed and used in an optical fiber pH sensor. Another type of optical indicator used fluorescence properties of certain compounds that fluoresce light (emit radiation) upon absorption of light (radiation) from another source, with the wavelength of fluoresced light being greater than that of the absorbed light. Attenuation of fluorescence (‘quenching’) occurs in proportion to analyte concentration, thus providing a sensor mechanism. Quenching occurs due to radiationless deactivation of excited states of the indicator by electromagnetic interaction of indicator and sensed molecules in a reversible, collisional process that does not consume the analyte. The intensity of fluorescent radiation of a specific wavelength is measured. The dye solvent green 5 (perylene dibutyrate) is a fluorescence indicator that was used in early development of fiber optic PO2 sensors. The indicator, therefore, transduces changes in the concentration of the chemical (in the body’s fluids) by altering

ISFET and Fiber Optic Sensor Technologies

Chemical Reviews, 2008, Vol. 108, No. 2 833

Figure 4. Indicator placement and schematic photon path for different fiber optic sensor designs. (Reprinted with permission from ref 122. Copyright 1994 University HealthSystem Consortium.)

absorbance of source light or changing fluorescence caused by source light. The optical fiber or fibers conduct source light to, and return light from, a small volume of indicator contained within the light path for measurement by remote instrumentation. The term ‘optode’, analogous to ‘electrode’, was popularized in the early 1980s. Figure 4 depicts indicator placement in different fiber optic sensor designs and schematic light paths for these designs. The optode sensors that were developed for medical applications were pH, PCO2, and PO2. Theory to explain the function of observed sensors was also developed and has been reported elsewhere.122-127 Not all reports of fiber optic sensor theory have confirmed an apparent basis of simplicity compared to electrode sensors. Thermodynamic principles applied to fiber optic pH sensors indicate that interphase effects and differences between complex bulk solution and submicroliter volumes of colorimetric indicator isolated in the light path of fiber optic sensors precludes measurement of pH (i.e., hydronium ion activity).128 Absorbance theory, related to concentrations of indicator conjugate base and conjugate acid, was commonly reported to explain fiber optic pH sensor function.122,127 For absorbance of incident light by the base form of an indicator dye, A-, the relative amount (concentration) of A- varies

with pH and is related to transmitted light intensity at an absorption wavelength, assuming uniform properties of the indicator solution. This simplistic approach ignored solutesolvent and solute-solute interactions and interphase surface adsorption effects, which largely determine the intended measured variable, hydrated hydrogen ion activity (not concentration) in the microenvironment of the optode sensor window. This analysis points out the fundamental difference between optical and electrochemical pH sensors. Electrode measurement is a potential difference between the bulk of solution and the bulk of an ion-selective membrane (e.g., hydrated glass for pH), and the potential of the ion species is equal in those two phases. None of the measured potentials is due to adsorbed species, e.g., proteins. Optode measurement of a charged species, e.g., hydronium ion (pH), originates with bulk solution-indicator surface window interaction. Here, the surface activity of the species of interest is related to its corresponding bulk value through its adsorption isotherm, competing adsorption, and surface equilibration effects, including ionic strength, and polyelectrolyte and solvent effects in bulk solution-indicator and indicator-optical fiber surface layers. Effectively, electrode and fiber optic sensor pH measurements are not the same,

834 Chemical Reviews, 2008, Vol. 108, No. 2

McKinley

and a fiber optic pH sensor does not measure hydronium ion activity but infers pH from optical characteristics of its color indicator. For electrically neutral species PCO2 and PO2, effects of activity coefficient variations are much less significant than in ionic sensors, and theory of fiber optic sensor technology based on concentration, and not activity, is consistent with thermodynamic analysis. These analyses do not explain the relatively close agreement of in vivo fiber optic sensor and in vitro pH measurements compared with PO2 sensors in subsequent clinical trials (see below), but do indicate overly simplistic theoretical justification of early optode sensors.

3.1. Development and Commercialization In contrast to ISFET sensor technology, development of fiber optic sensors began with commercial product focus in the early 1980s and proceeded for 10 years. The product focus was continuous intraarterial blood gas monitoring and became the objective of many development efforts in the United States and Europe. Immediate product focus was motivated by several factors: (1) Sensors for blood gas variables pH, PCO2, and PO2 had been demonstrated by early investigators. (2) Fiber optics was seen to be an attractive new technology, and sensor theory, perhaps deceptively simple, was readily understood. (3) Emerging successes of ‘high-tech’ (i.e., high growth and profit) computer companies motivated new venture funding by corporations, typically medical device and drug companies, and by then new venture capital firms. (4) A market was perceived for blood gas monitoring, a high value clinical laboratory analysis, that seemed clearly defined, to exceed threshold $100M annual revenue needed to sustain interest of investors, and able to be captured with the first product to be developed. The elegant simplicity of the fiber optic sensor concept apparently involved no principles of electrodes and electrochemistry, and sensor prototypes were demonstrated. With a certain product in mind, a technology rush to market was underway by the mid 1980s. The product concept, a continuous intraarterial blood gas monitor system comprising an indwelling single-use probe and a bedside instrument, was able to be defined. The essential part of the commercial product that would produce profit was a multisensor, discrete sensor probe that could be manufactured in large numbers at low cost, sold to hospitals at a multiple of manufacture cost, used once and thrown away. Sensor performance was assumed to be adequately stable and rapid in response to consider continuous monitoring, perceived to offer advance in patient care. Access to the patient’s blood was required, and was determined to be via a cannula already invading the patient’s vasculature for purposes of continuous monitoring of blood pressure. The probe diameter, therefore, needed to be sufficiently small to fit through a conventional disposable peripheral arterial cannula without perturbing the ex vivo blood pressure signal. This requirement could be met with optical fiber technology. Many development efforts were ongoing throughout the late 1980s. These development efforts led to three competitive continuous intraarterial blood gas monitor systems by the early 1990s: PB3300 Intra-Arterial Blood Gas Monitoring System (Puritan Bennett Inc., Carlsbad, CA), commercially available 1993, production discontinued 1994; Paratrend 7 (Pfizer/Biomedical Sensors Ltd., Highwycombe, U.K.), commercially available 1994-2002; Biosentry System (Optex Biomedical Inc., The Woodlands, TX), commercially available 1994, production discontinued 1995.

Figure 5. Multisensor indwelling probes with fiber optic sensors for pH, PCO2, and PO2, and a thermocouple for temperature measurement and compensation. Probe diameters were ∼0.2 mm and designed for sensors to be directly exposed to arterial blood. Connectors were designed to couple with previously placed intraarterial cannula for blood pressure monitoring. (Reprinted with permission from ref 122. Copyright 1994 University HealthSystem Consortium.)

Figure 5 shows photographs of indwelling probes developed for each of these fiber optic sensor systems. Figure 6 shows probes including cable extensions to connect the indwelling probes to bedside instruments, also possibly including a further extension cable ∼2 m in length. Within a decade of the start of focused product development, 18 years after initial demonstration of the optode, the discrete optode sensor was transformed from prototype to commercial product with FDA clearance to market. Many factors contributed to the limited market success of these systems.

3.2. Clinical Trials and Assessments Included in the development efforts were clinical trials of prototype and near commercial systems, and their introduction to clinicians who would use them. Operating rooms and intensive care units were the likely environments for the technology. Typically, the clinical trials were observational and compared continuous intraarterial blood gas monitoring

ISFET and Fiber Optic Sensor Technologies

Figure 6. Fiber optic probes including cable extensions to connect the indwelling probes to bedside instruments for operation of multisensor probes for intraarterial blood gas monitoring and display of continuous in vivo data. (Reprinted with permission from ref 122. Copyright 1994 University HealthSystem Consortium.)

and standard clinical laboratory in vitro analyses with the intent of demonstrating equivalence. In vivo monitor data were not used for clinical decision making. As of 1994, when systems had been commercialized, all human use trials reported were limited to groups of <15 patients per hospital. Although there were no reports of complications or patient morbidity due to experimental use of these systems, neither were there reports of cost-benefit analysis. Tests of four commercially developed systems were reported. The first human use study to be reported included some analysis of clinical use criteria.129 This study used a system that had been developed extensively but was not commercialized (CDI System 1000, Cardiovascular Devices Inc., Irvine, CA). Prior animal studies were conducted to

Chemical Reviews, 2008, Vol. 108, No. 2 835

assess performance and blood compatibility. Several hundred in vitro measurements were shown to agree well with in vivo measurements at times of blood sampling over common clinical ranges of pH, PCO2, and PO2. Human studies involved 12 patients and 79 in vitro analyses for comparison (duration of monitoring 3-26 h). A second trial in 14 operating room patients found similarly good in vivo monitor-in vitro analysis agreement (duration of monitoring 2-8 h). Studies of systems that were commercialized also maintained focus on the continuous intraarterial blood gas monitoring application for which they had been designed. Demonstration of the PB3300 system (Puritan Bennett Inc.) in mechanically ventilated ICU patients at a German university hospital emphasized use of the system in ‘routine clinical conditions’ (i.e., as a standard clinical monitor subject to no special maintenance or special conditions for proper function than other standard monitors).130 Probe use for >72 h was reported (range, 8-170 h). Placement via standard 20 gauge cannulae showed no degradation of arterial pressure waveforms. No probe failures occurred. Arterial hypotension, use of vasopressor agents, and core-peripheral temperature differences did not affect agreement of in vivo monitorstandard in vitro analyses. A second trial of the PB3300 system conducted in three U.S. hospitals showed similarly good performance in OR (mean time of monitoring ) 6 h) and ICU patients (mean time of monitoring ) 46 h) with several hundred in vivo monitor-in vitro sample analyses compared over ranges that included extremes encountered in critically ill patients.131 In vivo-in vitro comparison data was subjected to ‘proficiency testing’ used by clinical laboratory regulatory organizations (Health Care Financing Administration Clinical Laboratory Improvement Act) to assess reliability and measure agreement of an analyzer system with results from a peer group of similar analyzers measuring a standard test sample. The data obtained in this study were within the specifications of test sample targetmeasurement variation, although marginally for PO2. These studies demonstrated a discrete fiber optic sensor system that was clinically viable. The Paratrend 7 system (Pfizer/Biomedical Sensors Ltd.) was demonstrated in similar trials with similar success. In ICU patients, a mean duration of monitoring of 43 h (range 10-118 h) showed the ability to monitor pH, PCO2, and PO2 for time periods consistent with duration of clinical crises that require intensive care. With this system, ‘in vivo recalibration’ (adjustment of the system readout) to agree with an in vitro clinical laboratory sample analysis was done at 12 h intervals (but not during periods of blood gas instability, e.g., adjustment of fraction inspired O2) to correct for presumed sensor drift. A second similar trial in OR patients undergoing cardiopulmonary bypass surgery demonstrated performance during hypothermia (30 °C). In vivoin vitro agreement was poor for PO2 but acceptable for pH and PCO2. Of note, the Paratrend 7 system used an amperometric electrode PO2 sensor in its first commercially available system. In work that preceded that of other commercialized systems, the Biosentry System (Optex Biomedical Inc.) had similar results in trials that used a precommercial prototype system. In ICU patients, in vivo monitor-in vitro analysis agreement was acceptable and similar to a comparison of in vitro analyzers (mean monitor duration ) 55 h, range 4368 h).132 In OR patients, performance was adequate but found

836 Chemical Reviews, 2008, Vol. 108, No. 2

poor in vivo-in vitro agreement for PO2 during emergence from general anesthesia when PO2 changes might be rapid and peripheral blood flow could change.133

3.3. Issues Identified with Continuous Intraarterial Blood Gas Monitoring Reports of these trials in the critical care literature demonstrated a new sensor technology that functioned remarkably well in clinical environments for critically ill patients. Continuous intraarterial blood gas monitoring based on fiber optic sensor technology presented a significant advance in physiologic monitoring. Many issues became apparent as a result.122 In these clinical trials, many conditions were encountered that would impede general system acceptance for routine clinical use, e.g., sensor calibration, response time and temperature effects, and disagreement of some in vivo-in vitro measurements. The latter issue was one of constant focus throughout efforts to effectively market and sell these continuous intraarterial blood gas monitor systems. In this regard, sensor response times, exact time of sampling, flow variation in the radial artery (typical arterial pressure cannula and probe placement site), and clinical laboratory sample processing systems were examples of factors that could affect in vivo-in vitro analysis agreement. In the eyes of prospective clinician customers, the problem of imprecise agreement of a continuous monitor and the standard clinical laboratory was a persistent issue. Blood gas variables pH, PCO2, and PO2 had been marketed as critical physiologic information that require continuous monitoring to better manage critically ill patients, and any disagreement could therefore present a problem. Clinical (nonideal) situations that were encountered included hemodynamic instability, arterial hypotension, vasoconstriction, and cold extremities (probe temperature up to 7 °C less than body core temperature), all possibly contributing to low blood flow through the peripheral artery of probe placement, and disagreement of in vivo-in vitro analyses. Although limited in size, the initial observational clinical trials did demonstrate remarkably good agreement of in vivo-in vitro analyses, especially pH and PCO2. PO2 consistently had the poorest agreement, but arterial PO2 may be the most ‘volatile’, subject to the greatest and most rapid changes, and therefore most likely to have the poorest agreement with in vitro analyses of random blood samples. The cost of any new monitors or equipment was another rapidly growing issue in the early mid 1990s when these continuous intraarterial blood gas monitoring systems became commercially available. None of the clinical trials addressed this issue. Underlying this issue, the benefits of either intermittent, discrete sample analysis or continuous blood gas monitoring had ever been demonstrated or compared, making estimation of the value of one over the other difficult. The cost of the fiber optic sensor systems included the probe and bedside monitor instrument. The latter was especially expensive, having necessarily incorporated expensive optical components to select light of specific wavelength ranges from a broad spectrum source. Bedside monitor instrument prices of $20 000 or greater were problematic. Probe prices of $300 or greater were required due to the cost of manufacture, and were also problematic. To most potential customers, the cost of the continuous intraarterial blood gas monitor systems seemed much greater than that of standard clinical laboratory systems if the cost

McKinley

of one laboratory blood gas analysis was compared with the cost of one single-use probe. Many other common cost factors for laboratory blood gas analysis (e.g., nursing time required to obtain delayed or lost laboratory results and costs of other monitoring technologies that may provide indications or rationale for obtaining a laboratory blood gas analysis) were typically ignored. The value (benefit/cost ratio) of continuous blood gas data during an acute hypoxic crisis would be very high (could prompt life saving interventions), but probably very low during most of the time a patient is monitored. The cost of two continuous intraarterial blood gas monitor probes during 6 days in an ICU could still be considered a better value than, by crude analogy, the cost of seat belts and air bags in automobiles that are designed to save lives in a crisis.

3.4. Interstitial Fluid Monitoring Using Fiber Optic Sensor Systems With ongoing controversy over continuous arterial blood gas monitoring, the concept of interstitial fluid (IF) monitoring using this technology began to be investigated. The fiber optic sensor probes were very small in diameter (∼0.2 mm), and probe designs that incorporated sensors at or near the probe tip were potentially useful in this application. Monitoring brain parenchyma pH, PCO2, and PO2 was the first application to receive focused study.22,134-138 In studies of patients undergoing neurosurgical procedures, IF pH, PCO2, and PO2 were monitored and found to change dramatically.22,137 In an animal model of cerebral hypoxia and ischemia, differences in systemic and brain parenchyma baseline measurements and dramatic changes in brain parenchyma measurements were demonstrated.138 These realtime, simultaneous measurements had not been available previously and provided new insight to the effects of changes in brain perfusion on tissue cellular metabolism. Brain parenchyma norms and thresholds for intervention were considered and analyzed.135,138 As shown in Figure 7, remarkable long-term monitoring was described in one report and demonstrated the possibility of prolonged monitoring throughout typical ICU stay times, similar to blood pressure and, for brain trauma, intracranial pressure.139 This data confirmed the durability of the sensors and probe designs for IF monitoring. This data also demonstrated ongoing variability of continuous monitoring of basic cellular oxygenation and ventilation variables, and possibility of intervention to preempt and avert potentially disastrous effects of intracranial hypertension due to edema. A tissue bed of continued interest to many investigators was skeletal muscle, and the same fiber optic sensor technology developed and marketed for continuous intraarterial blood gas monitoring was applicable. Reports in animal models of hemorrhagic shock described the ability to simultaneously monitor IF pH, PCO2, and PO2 of skeletal muscle reliably and accurately.140,141 These reports also depicted clear differences of systemic and IF pH, PCO2, and PO2 norms and changes of these variables during and after shock. IF pH was shown to recover much more slowly than IF PCO2 and PO2, likely reflecting persistent cellular metabolic perturbation. Similar to studies of brain parenchyma, changes in skeletal muscle IF PCO2 appeared to reflect the adequacy of perfusion, and, together with IF pH, the presence of anaerobic, or perhaps defective metabolism. A case report of skeletal muscle IF monitored during ICU shock resuscitation using the same fiber optic sensor technol-

ISFET and Fiber Optic Sensor Technologies

Chemical Reviews, 2008, Vol. 108, No. 2 837

Figure 7. (a) Extended continuous monitor records of patient brain parenchyma interstitial fluid (IF) pH, PCO2, and PO2, and intracranial pressure (ICP; hourly) after severe closed head injury during eight successive ICU days. Cyclic oscillation of IF PCO2 and pH co-incident with hypothermia temperature controller; IF PO2 increases with start of pentobarbital therapy (ICU day 4); IF pH decrease and IF PCO2 increase 9 h prior to ICP increase to 50 mmHg (ICU day 4) and 3.5 h prior to second ICP increase (ICU day 5). (Reprinted with permission from ref 139. Copyright 1998 Springer.) (b) Extended continuous monitor records of patient brain parenchyma interstitial fluid (IF) pH, PCO2, and PO2, and intracranial pressure (ICP; hourly) after severe closed head injury during 10 successive ICU days: IF PO2 ) 4 mmHg, PCO2 ) 230 mmHg, pH ) 6.48 co-incident with arterial PCO2 ) 58 and pH ) 7.08, and norepinephrine infusion (ICU day 2); IF PO2 increase with start of pentobarbital therapy (ICU day 4); IF pH decrease and IF PCO2 increase 10.5 h prior to ICP increase to 50 mmHg (ICU day 4). (Reprinted with permission from ref 139. Copyright 1998 Springer.) For these studies, brain parenchyma interstitial fluid pH, PCO2, and PO2 data was obtained using fiber optic sensor system (Biosentry, Optex Biomedical Inc.) following an institutional review board approved protocol.

ogy and system designed for blood gas monitoring depicted derangements of skeletal muscle IF pH, PCO2, and PO2 due to severe shock following massive trauma.142 Continuous monitoring showed the return of these variables to normal with reestablished stable hemodynamic performance. In these reports, hemodynamic performance was monitored using standard highly invasive pulmonary artery and peripheral artery catheters. The skeletal muscle placement of the fiber optic probe with its multiple pH, PCO2, and PO2 and temperature sensors was via percutaneous puncture using a 20 gauge needle and catheter sheath, withdrawn to expose the sensors to IF in viable tissue. This placement technique was minimally invasive, easily accomplished, and of extremely low risk. The authors concluded that skeletal muscle IF (i.e., peripheral tissue) monitoring may provide a more complete picture of shock and resuscitation that do conventional systemic variables, with skeletal muscle IF pH, PCO2, and PO2 providing a direct indication of metabolism, respiration, and oxygenation, and an indirect indication of perfusion of peripheral tissue.142

3.5. Summary of Fiber Optic Sensor System Development for in vivo Monitoring By the late 1980s, development of fiber optic sensor systems for intraarterial blood gas monitoring was an international effort and in the United States had involved NIH, national laboratories, major pharmaceutical and device manufacturers, and numerous venture companies. By one estimate, $250 million of private industry and venture funding had been expended from the mid to late 1980s, all with the goal of intraarterial blood gas monitoring. By the early 1990s, three systems were developed, cleared to market

by the U.S. FDA, and, for a short time, sold competitively. By the late 1990s, only one of these systems was available, and its use was rarely encountered. Presently, one relatively new fiber optic sensor system designed for IF PO2 is commercially available and marketed primarily for laboratory animal research (Oxford Optronix Ltd., Oxford, U.K.; www.oxford-optronix.com). Reports of continued development of indicators and sensor systems continue to appear in the literature.143-145

4. Factors Affecting the Future of Discrete Sensor Development for in vivo Monitoring The concept of discrete sensors incorporated in a probe to monitor specific physiologic processes within a patient (or other living subject) has been demonstrated. ISFET and fiber optic sensor technology development in recent decades has provided a wealth of experience and rationale for continued development. The experience has included technology and performance requirements for medical applications of continuous monitoring using discrete sensors incorporated in a probe to provide remote access. Clinical trials and additional experience with fiber optic sensor continuous arterial blood gas monitor systems stimulated careful consideration of need for and implications of continuous blood gas monitoring. The value added by continuous (blood gas or electrolyte) monitoring over intermittent (clinical laboratory) measurement is not obvious. The need for continuous monitoring of variables is accepted for some variables in clinical critical care. This need has been less well established for those variables traditionally available from the clinical laboratory than for pressure, heart

838 Chemical Reviews, 2008, Vol. 108, No. 2

electrical activity, and certain oxygenation variables, especially arterial blood hemoglobin O2 saturation (SpO2; ‘pulse oximetry’). These variables, indicative of heart and lung function and of continuous concern in an ICU, may also be monitored because those sensor technologies (invasive blood pressure, Ag-AgCl EKG electrode, hemoglobin O2 saturation) provide reliable continuous inexpensive measurements. In general, conditions for continuous monitoring of a sensed variable should include the following: (1) likely to change rapidly and of a magnitude that adversely affects patient physiology; (2) an available, calibrated intervention that affects the measured variable (i.e., a variable that is able to be measured but for which there is no treatment or intervention of research interest would not be a candidate for clinical care); (3) not detectable by other existing techniques; (4) and total response time from decision to obtain a measurement to return an analytical result to be acted upon by a physician and therapy to have an effect should be decreased compared to existing measurement techniques For remote laboratory-based analyzers, the total response time might be 10 min (including decision to obtain measurement, acquisition of representative sample, transport of sample, processing and analysis, results quality assurance check, and transmit of result to bedside). For a bedside point of care analyzer, total response time might be 5 min. For a discrete sensor-based, continuous in vivo monitor system, total response time (defined using system response time constant to characterize a sensor, e.g., 63% of full response to a step increase in analyte) might also be 5 min, but this system would be more likely to detect important changes of the monitored variables. Clinical trials of fiber optic sensor-based blood gas monitor systems demonstrated reliable, safe function in ICU settings in patients who were at risk for or who were experiencing respiratory distress and other complications. Other technologies that could still presently be considered to compete with continuous intraarterial blood gas monitoring included the following: (1) physical examinationsability to identify signs or symptoms or likelihood of development of blood gas abnormalities; deemed inadequate to replace blood gas analysis but able to suggest need to obtain blood gas information; (2) pulse oximetrysby the late 1980s, a noninvasive continuous monitor that had become a clinical standard during intraoperative anesthesia and management of critically ill patients; insensitive to changes in PO2 >60 mmHg, thereby permitting large decreases in arterial PO2 indicative of deteriorating pulmonary function without affecting hemoglobin O2 saturation indicated by pulse oximetry; unreliable in patients with poor cutaneous perfusion, e.g., during shock or hypothermia, who could be unstable and likely to suffer sudden decreased oxygenation; (3) capnographysnoninvasive continuous monitor of airway end tidal PCO2 but not specifically indicative of arterial PCO2, i.e., decreasing end tidal PCO2, could indicate decreasing cardiac output or increasing alveolar ventilation and not indicative of arterial pH; (4) in vitro (clinical laboratory) arterial blood gas analysissreadily available in hospitals using highly automated analyzers with excellent quality control and rapid, accurate, repeatable analysis; major deficiency is typically prolonged total analysis time from time of sample acquisition to time of results report; (5) in vitro (bedside or point of care) arterial blood gas analysiss accurate, repeatable analysis, available in hospitals, typically in OR or ICU settings where immediate and frequent analysis

McKinley

may be required; self-contained electrode sensor-calibration solution-sample waste modules or single-use disposable electrode sensor modules;146-151 (6) ex vivo (bedside or point of care) analysisspreviously developed in systems using ISFET,48-50,102,152,153 electrode,154,155 and fiber optic sensors,154,155 incorporated in a patient-attached module that are periodically exposed to blood withdrawn from and reinfused to the patient via intravenous or intraarterial catheter and tubing. For discrete sensor-based continuous monitor systems other than blood gas, other technologies to supplant or cue need for electrolyte analysis include primarily EKG to detect heart rhythm abnormalities, ISFET K+, and Ca++ experience suggests rationale for continuous electrolyte monitoring in many intensive care situations. Bedside point of care analyzers are commercially available to provide random or interval analyses, but rapidity and magnitude of changes in circulating blood volume are common and meet conditions for continuous monitoring as described above. Decisions to obtain a measurement and with some frequency have been influenced by desire to contain overall patient care cost. The relative value of a continuously monitored variable available intermittently via clinical laboratory analysis must therefore be demonstrated to decrease cost, risk to the patient, and inconvenience of bedside caregivers. In clinical intensive care environments the frequency of essential clinical laboratory measurements tends to increase with perceived severity of a patient’s illness. Because risk to the patient is already established with placement of catheters for arterial and venous blood, urine, and upper gastrointestinal tract fluid, the decision for increased frequency of clinical laboratory analysis is unlikely to increase this risk. With increased severity of illness and physician judgment of possibility of survival with certain interventions, cost becomes a secondary factor affecting the frequency of clinical laboratory measurement.59 The concept of ex vivo analysis has been pursued, and systems were developed and commercialized. Systems were developed based on early ISFET sensors and measured K+, Na+, Ca++, and pH in patient blood with initial focus on intraoperative monitoring during cardiac surgery.48-50,153 Two systems were commercialized in the early 1990s. A system incorporating fiber optic pH, pCO2, and pO2 sensors was found to function reliably in clinical trials involving ICU patients.156,157 Another system incorporating electrode sensors for pH, pCO2, pO2, K+, Na+, and hematocrit also functioned reliably in ICU patients.154,155 Development and commercialization of these ex vivo systems occurred during and after that of intraarterial fiber optic blood gas monitor systems. The electrode system was recently recommercialized (VIA Medical, Austin, TX; www.viamedical.com). Rationale for development of these systems was in part based on the uncertainty of performance (primarily accuracy) and inability to check calibration of intravascular sensors that might be unpredictably compromised by effects of blood protein deposition and clot formation. Further rationale was associated with difficulty in design and manufacture of intravascular probes incorporating multiple sensors of microscopic dimension. Analysis of ex vivo vs in vivo system viability has been based on monitoring of systemic physiology in blood. Described above, focus on intravascular monitoring was a premise that was unsuccessful for fiber optic blood gas systems, whereas, described below, interstitial monitoring was found to be feasible using both ISFET and fiber optic

ISFET and Fiber Optic Sensor Technologies

sensor technologies, and to offer a new mode of physiologic monitoring. Rationale for continuous data for current clinical laboratory analyses has been described to relate to hospital personnel labor saving in the current environment of cost reimbursement for diseases treated and fixed annual budgets for care of defined patient populations.59 Instead of multiple discrete sample analyses and the attendant labor involving bedside nurse, clinical laboratory technicians, and sample courier, discrete sensor probes, calibrated and placed once by the bedside nurse, were demonstrated by fiber optic sensor-based continuous blood gas monitor systems described above. The cost to the patient of a multisensor probe, e.g., $300, which would be replaced every 2 or 3 days, would still compare favorably with individual blood gas analyses, e.g., $300 each. If a probe incorporated blood gas and electrolyte sensors, the cost equation would be more favorable for in vivo continuous monitoring. Quality of care, enhanced by continuous monitoring of critical clinical chemical variables, could be favorably affected. Discrete sensor-based continuous in vivo monitoring is now complemented with information technology in hospital critical care environments. Decision support algorithms could be developed to recognize abnormal trends of individual variables, compare patterns of concurrent variables, interpret abnormalities as possible differential diagnoses, and provide instructions for additional confirmatory analyses and interventions. With the reliability of monitoring systems established, closed loop control would be developed.158

4.1. Continuous IF Monitoring Of the ISFET and optode sensor system experiences, perhaps the greatest advance was demonstration and confirmed utility of IF monitoring as a new mode of monitoring in clinical critical care medicine. Reasons include the following: (1) IF presents compartments that are accessible by small, percutaneously placed probes. (2) Gas partial pressures and ion activities reflect adequacy of blood perfusion and cellular function. (3) Inadequate cellular utilization of delivered substrates is detectable early and adequacy of interventions for improved cellular utilization are able to be monitored. In large tissue beds, inadequate cellular utilization of delivered substrates is commonly termed ‘shock’. Shock is perhaps the most central and resource intense problem that confronts clinical critical care medicine.12,159 Recognition of shock, resuscitation, and subsequent maintenance of organ system and hemodynamic function are capabilities that have defined clinical critical care of severe trauma and sepsis patient populations. The current state of the art involves measurement of systemic variables to recognize and confirm shock, and to monitor the resuscitation process. Systemically measured variables, including blood pressure, pH, hemoglobin O2 saturation, and lactate concentration, and derived variables, including base deficit, systemic vascular resistance, and oxygen delivery, have been used clinically as indirect indicators of tissue perfusion, but they do not indicate onset and correction of cellular dysfunction in any particular tissue bed.160-164 From ISFET experience, IF K+ seems to be an indicator of cell function that is appropriate and able to be monitored. In most tissues, especially skeletal muscle which is perhaps most accessible, the large cellular transmembrane gradient between intracellular cytoplasm and extracellular fluid that

Chemical Reviews, 2008, Vol. 108, No. 2 839

is the physiologic norm is related to constant O2-dependent energy supply. An increase in IF K+, as demonstrated with early ISFET studies of hemorrhagic shock in animal models, is most likely indicative of disruption of O2-dependent energy supply.51 IF Ca++ monitoring, also demonstrated to be feasible in early ISFET studies, is another variable of potential relevance in monitoring cellular function during shock and resuscitation. Decreased blood Ca++ activity was reported in early studies of hemorrhagic and septic shock in animal models,15,16 and blood and IF Ca++ were shown to be decreased by IV citrate infusion comparable to that associated with blood transfusions.57,59 IF K+ and Ca++ may therefore be more direct indicators of cellular function and detectors of cellular dysfunction than traditional systemic variables, and these variables can be monitored using discrete sensors. From optode experience, as described previously, IF pH, PCO2, and PO2 were studied in animal models of hemorrhagic shock and resuscitation.140,141 Skeletal muscle was studied as a likely clinically available tissue bed affected by hemorrhage-induced perfusion abnormalities. Comparison of systemic arterial and venous measurements revealed IF pH < venous pH < arterial pH, IF PCO2 > venous PCO2 > arterial PCO2, and arterial PCO2 > IF PO2 > venous PO2, as expected with the tissue bed as the site of cellular respiration and source of acidic metabolic end products. Inverse time courses for IF pH and PCO2 occurred with hemorrhage, and pH was slowest to recover toward baseline measurements, possibly indicative of integrated metabolic status and its prolonged derangement. Rapidly decreased PO2 was observed, consistent with decreased perfusion of skeletal muscle (peripheral) tissues and preservation of perfusion of core organs and rapid recovery to measurements that temporarily exceeded baseline with resuscitation. After an hour of shock (MAP ≈ 45 mmHg), IF pH ≈ 6.9, IF PCO2 ≈ 130 mmHg, and IF PO2 ≈ 5 mmHg, compared with arterial pH ≈ 7.1, arterial PCO2 ≈ 35 mmHg, and arterial PO2 ≈ 100 mmHg, were recorded, demonstrating greater proximity of IF measurements to cellular processes than reflected by systemic measurements. In clinical studies of trauma shock resuscitation in which skeletal muscle was monitored during resuscitation, similar arterial-IF-venous gradients and extreme IF measurements were observed.165 Studies of IF pH, PCO2, and PO2 of brain parenchyma in an animal model of ischemia and hypoxia, and clinical studies monitoring effects of therapy after brain trauma, found more extreme measurements of IF pH, PCO2, and PO2.138,166 IF pH, PCO2, and PO2 therefore appear to be direct indicators of cellular function and detectors of dysfunction, and were also demonstrated to be reliably monitored using discrete sensors. The IF studies undertaken with ISFET and, later, optode discrete sensor systems show IF monitoring to be a clinically acceptable technique. Discrete sensor technologies, as described, could open a valuable window of continuous monitoring in critical care medicine. Requirements for IF monitoring of K+, Ca++, pH, PCO2, and PO2 seem to have been clearly met:59 (1) small probes were able to be placed without disruption or disturbance of the local tissue environment. (2) Sensed variables were able to be monitored in IF environments for many hours to days. (3) With systemic pathology (e.g., hemorrhagic shock, brain injury) and local tissue perfusion changes, sensed variables changed early and with greater sensitivity than conventional systemic param-

840 Chemical Reviews, 2008, Vol. 108, No. 2

eters. (4) The rate and magnitude of changes in sensed IF variables K+, Ca++, pH, PCO2, and PO2 are specific and striking in the presence of shock, or brain injury, to consistently diagnose inadequate perfusion and need for intervention. (5) These sensed IF variables provided timely and reliable responses to interventions to indicate adequacy of response at the tissue level. Importantly, in critical care settings of trauma and sepsis resuscitation and therapy for brain injury, outcomes are not able to be accurately predicted after resuscitation efforts based on conventional systemic measures. For this reason, the information obtained from continuous IF monitoring using discrete sensor systems could offer greatest utility in monitoring adequacy of resuscitation interventions. An alternative to direct tissue monitoring that evolved in the 1980s was CO2 tonometry of mucosa of hollow viscus organs, including stomach and upper gastrointestinal tract.140,167-175 Using a CO2-permeable membrane to surround a CO2 sensor, changes in mucosal interstitial PCO2 could be monitored by placing a flexible (balloon) tonometer in close proximity to metabolically active mucosal membranes lining the GI tract. Normally highly perfused, the body’s response to blood loss or increased skeletal muscle O2 demand is to shunt blood flow from the GI tract to tissues that require more constant blood flow, causing CO2 accumulation in the GI mucosa. CO2 diffusion from intra- to extracellular space and across the permeable membrane to the CO2 sensor is reliably detected and representative of mucosal interstitial PCO2. GI mucosal CO2 tonometry, available commercially beginning in the 1990s, was subjected to many studies and vigorous debate to establish its credibility as an indirect monitor of systemic and organ-specific O2 delivery and onset of anaerobic metabolism signaling O2 delivery and perfusion inadequacy. Oral mucosal CO2 tonometry systems were developed that incorporated an ISFET or fiber optic pH-based PCO2 sensor. A commercially available system incorporated a fiber optic pH-based PCO2 sensor, which provided single PCO2 measurements (CapnoProbe Sublingual Sensor, Nellcor (Tyco Healthcare/Mallinckrodt), Pleasanton, CA; no longer available). Recent reports indicate that continuous measurement of oral mucosal PCO2 identifies the severity of the volume deficit176 and may offer a more direct and readily obtainable assessment of tissue perfusion than blood pressure.177 Presently, there are two systems commercially available for continuous in vivo monitoring of oxygenation of tissue beds. A highly engineered electrode system was developed and tested in other applications and optimized to monitor brain parenchyma, i.e., brain IF, PO2 (Licox, Integra LifeSciences Inc., Plainsboro, NJ; www.integra-ls.com).10,20,28,178,179 The system incorporates a precalibrated electrode PO2 probe (0.8 mm diameter) and intraparenchyma placement system for use in monitoring brain oxygenation after injury during intensive care for time periods of days. Another technology, near-infrared spectrometry (NIRS), has been developed for noninvasively monitoring tissue hemoglobin O2 saturation for time periods of days (Inspectra, Hutchinson Technology Inc., Hutchinson, MN; www.htbiomeasurement.com). This technology does not use a discrete sensor that transduces a chemical analyte to electrical or optical signal but monitors IR absorbance by oxygenated hemoglobin (and tissue cytochromes) in skeletal muscle and subcutaneous tissue using a fiber optic probe that attaches to the overlying skin surface.180,181 In recent clinical trials, the monitored variable,

McKinley

StO2, has been shown to detect need for early massive transfusion and correlate with outcome in severely injured patients.182,183 Similar to ISFET and optode sensors, both systems have developed over a period of decades to their present form and function. Of note, these are among the latest monitor systems to become commercially available for intensive care use, and both are designed to monitor tissue and not systemic blood.

4.2. Technology Problems The requirements of an ideal monitor for use in clinical critical care environments are as follows:12 (1) a sensor for an essential variable(s) that is accurate and stable for indefinite duration of use; (2) a non- or minimally invasive sensor; (3) a continuous monitor function with the ability to display trend (record of recent measurements with time); (4) a system that is easily used and with measurement display that is easily understood; (5) small size and weight; and (6) ruggedness and transportability. This list of qualities is attained by very few monitors that are in currently in use. The pulse oximeter is perhaps the foremost example that embodies most of these requirements and is universally affordable throughout the health care systems of the United States and most nations. With new information that would be regarded as essential, e.g., to guide shock resuscitation to an endpoint, new ideals become obvious. Both ISFET and optode discrete sensor technologies were developed to the extent that features and functions of prototype and commercially available systems did address most of these ideal requirements. Probes were developed that were of 0.5 mm diameter (ISFET) and 0.2 mm diameter (optode). Optode probes were able to be placed percutaneously and without any bleeding in skeletal muscle IF space. Instrumentation systems, designed using early 1990s optical technology, were adequate in their designs. Much improved systems were developed by the late 1990s using new blue light (460 µm) emitting diode (LED) device technology. This single technical development led to electro-optical bedside instruments that were quite compact and transportable and incorporated displays with clearly viewable, easily understood graphical data presentation. Additionally, these latest electro-optical instrumentation systems to be developed, although for an ex vivo a fiber optic sensor system, were affordable (∼$2500).156,157 Issues encountered with invasive probes or implanted materials include biocompatibility. A device intended to be placed in direct, prolonged contact with blood in a peripheral artery must be non-thrombogenic, nontoxic, sterile and not injurious to the tissue or blood vessel endothelium it will contact. Probe designs for fiber optic sensor systems commercialized in the early to mid 1990s met these requirements. For IF monitoring, these blood compatibility requirements are not applicable because no blood contact occurs. A basic principle of intensive care, especially with evolving infectious agents resistant to antibiotic drugs, is to minimize penetration of skin, upper airway, and other anatomic protective barriers. Compared to intravascular placement, an advantage of percutaneous placement of a small diameter probe in the interstitium, e.g., anterior thigh, tricep, or other skeletal muscle bed, is the lack of blood stream invasion, ease of site preparation, and ability to visually monitor the placement site for development of erythema or presence of blood. Infection of bicep interstitial sites was not reported as a

ISFET and Fiber Optic Sensor Technologies

complication in extensive clinical trials using an electrode PO2 probe.20 Sensor response times have been addressed for ISFET and fiber optic discrete sensors. Optode systems, which balanced sensor response and life times to achieve adequate performance, offered sensor life time duration that was adequate for clinical critical care applications: 72 h was considered a performance requirement; intraarterial blood gas monitoring performance >150 h was reported in clinical trials.130 Sensor 90% response times for both ISFET (milliseconds) and fiber optic sensor (seconds to 1-3 min) systems were found to be much less than that of physiologic events monitored systemically or in IF compartments.40,55,122,184 Inherent to measurement provided by a sensor is accuracy. For continuous in vivo monitoring, accurate measurement provided by a precalibrated sensor is required. Precalibration methods for ISFET and fiber optic sensor systems described were in vitro and relied on characterizing sensor responses over physiologic ranges and specifying sensor measurements in an aqueous solution similar to that of the physiologic (micro) environment into which the sensor would be placed. Typically, two solutions of known gas partial pressures or ion activities bracketing the anticipated (patho)physiologic range were used. Upon placement, it was assumed that precalibration measurements continued to reflect the in vivo physiologic microenvironment. Drift of sensor responses, both baseline and magnitude, is a basic issue for continuous in vivo monitoring, with which sensor access may not be possible without disturbance of the placement site or the sensor. Ex vivo and in vitro systems provide sensor access to check calibration. Provision of in situ recalibration has been addressed using channels and chambers for introduction of calibration fluids but not incorporated in commercial systems.63 Recent ISFET developments, including options to Si3N4 gate regions, may have minimized (pH sensor) drift, and IC device or encapsulation degradation may be monitored by incorporation of other on-chip devices.45,185,186 Calibration of sensors prior to placement was a part of prototype ISFET and commercial optode systems operation. Neither ISFET or optode sensors have been fabricated with sufficient uniformity or package shelf life stability to preclude in vitro calibration just before use at patient bedside. Optode sensor system stabilities were characterized sufficiently to predict life time duration, or need for offset adjustment at some reasonable time interval based on representative sample clinical laboratory analysis, and included ports to permit (blood) sample acquisition at the placement site for independent clinical laboratory analysis.122 The question of accuracy has been addressed in studies of ISFET and fiber optic sensor systems in systemic blood of laboratory and clinical trials by comparing samples from the sensor microenvironment, as described above, and these studies have provided the best evidence that continuous in vivo monitoring is accurate and reliable in blood. IF, i.e., extravascular and extracellular, environments are arguably less complex and more favorable than intravascular blood for sensor placement and monitoring. Effects of protein or cell adsorption on in vivo sensor systems, e.g., ISFET ionselective membrane or reference electrode surfaces, in either blood or IF have not been studied systematically to the author’s knowledge. Related effects have been studied in commercial clinical laboratory (in vitro) systems, and reports confirm complexity of in vivo electrolyte and protein interaction.96,187,188 Empirical data, including blood and IF,

Chemical Reviews, 2008, Vol. 108, No. 2 841

provide convincing evidence of continuous in vivo monitoring capability for discrete blood gas and electrolyte sensors. The foremost problem, and the most obvious, for ISFET and optode discrete sensor systems is low-cost manufacture involving liquid and polymeric components with micrometer dimensions. To the knowledge of the author, these problems have not been adequately resolved, or if they had, a commercially viable product would be presently available. For ISFET sensors, insulation-encapsulation of part of an IC chip and exposure of the thin gate region or incorporation of liquid or gel ion-selective membrane material over the thin gate region has been the essential challenge. Recent developments that have been cited include:45 prefabricated microhousing assemblies; sensor fabrication using epoxy molds; photolithographic process to incorporate nonplanar structure; and elastomeric sealant materials. These developments have provided encouraging manufactured prototypes of single-sensor probes as small as ∼1-2 mm diameter with ruggedness much greater than that of conventional (glass membrane pH) electrodes. Discrete ISFET sensors manufacturable in large volume and suitable for in vivo monitoring are not yet reported. For ISFET sensor systems, a technical solution of this problem appears to preclude commercial endeavors for the high-price, high-volume medical market. For optode sensor systems, automation of processes was foreseeable with manufacturing methods developed by the early 1990s.

5. Summary ISFET and fiber optic sensors progressed rapidly from concept to reality in less than a decade, specifically for medical applications. Both technologies realized the concepts of discrete sensors with micrometer dimensions, and with potential for manufacture of a high-technology, low-cost monitor to dramatically improve medical care. Much was learned with these developments, and these technologies are principal examples of medical sensor technology that continues to evolve. Both ISFET and fiber optic sensors are basic technologies with broad applicability, including many sensors and sensor mechanisms and many potential applications. Development of these two sensor technologies progressed very differently. ISFET sensors progressed with continued demonstration of new sensors and mechanisms with engineering prototypes developed to study and confirm feasibility of applications. Based mostly on scientific grant and minimally on private funding, this progressed rapidly throughout the 1970s and early 1980s with the resulting demonstration of sensors for hydrogen, potassium, calcium, sodium, ammonium, and fluoride ions and, with broader ChemFET technology, developments of other families of sensors, and development of sensor theory and mechanism. Fiber optic sensors, from earliest demonstrations of pH and PO2 sensors, focused medical drug and device corporations on the blood gas variables, pH, PCO2, and PO2, and on a perceived market for continuous intraarterial blood gas monitoring. Developments of commercial products resulted and, as described, had limited market success for many reasons. ISFET and fiber optic sensor systems were found to be useful for continuous monitoring of interstitial fluid compartments of tissue beds. Skeletal muscle and brain parenchyma are tissue beds in which reliable monitoring was demonstrated and found to have probable clinical importance. With ISFET sensors in prototype probes, interstitial fluid monitor-

842 Chemical Reviews, 2008, Vol. 108, No. 2

ing of pH, K+, and Ca++ was investigated with early success. Fiber optic pH, PCO2, and PO2 sensor systems designed for intraarterial blood gas monitoring were used for interstitial fluid monitoring and further demonstrated capability and probable clinical utility of interstitial fluid monitoring. Interstitial fluid monitoring appears to be a mode of monitoring that should be pursued in clinical critical care medicine, and ISFET and fiber optics sensor systems offer feasibility. The information obtained from continuous interstitial fluid monitoring using ISFET or fiber optic sensor systems could offer greatest utility in monitoring adequacy of shock resuscitation interventions. Recognition, resuscitation, and prevention of shock is perhaps the most central and resource intense problem that continues to confront clinical critical care medicine.12,59,159 As demonstrated with continuous monitoring of ISFET or fiber optic K+, Ca++, pH, PCO2, and PO2 sensor systems, these variables may be more direct indicators of cellular function and detectors of cellular dysfunction than traditional systemic variables, e.g., blood lactate concentration or base deficit. The probe designs and sensor systems permitted bedside placement for continuous interstitial fluid monitoring of these variables with little restriction of patient access by clinicians. Experiences with these two sensor technologies have demonstrated capability and probable clinical importance in monitoring the interstitial fluid space of tissue beds. The developments of both sensor technologies progressed rapidly from basic scientific research and development of theoretical understanding to engineering prototypes and development of practical understanding. Technical problems have precluded a commercially viable medical product. ISFET pH sensors are commercially available for various applications in water, food, and drug process monitoring, but the foremost problem for both ISFET and fiber optic sensor systems, and that most obvious for medical applications, is low-cost manufacture involving liquid and polymeric components with micrometer dimensions.

6. Conclusion ISFET and fiber optic sensor developments since the 1970s have realized the concept of monitoring clinically important problems in critically ill patients. The ability to continuously monitor physiologic processes of viable cells in tissue beds has also been demonstrated. ISFET and fiber optic sensor technologies have had different development directions since their inceptions, but both have been and continue to be pursued with similar interest. Rapid early progression and developments that continue to show the enduring interest in the concept of discrete sensors to monitor otherwise invisible physiologic processes. It seems likely that these technologies will continue to provide insight into physiology as technical problems are incrementally resolved. These sensor technologies have both been used to demonstrate IF monitoring, a possible new window of continuous monitoring in critical care medicine. The breakthrough of a viable commercial product that provides a new mode of continuous monitoring for clinical critical care seems probable. With the future as the cradle of past developments,189 the medical markets foreseen for these sensor technologies remain as possibilities.

7. References (1) Collison, M. E.; Meyerhoff, M. E. Anal. Chem. 1990, 62, 425A.

McKinley (2) Cunningham, J. N., Jr.; Shires, G. T.; Wagner, Y. Am. J. Surg. 1971, 122, 650. (3) Dmochowski, J. R.; Deuvaert, F.; Rabelo, A., Jr.; Couch, N. P. Surg. Forum 1972, 23, 12. (4) Drake, H. F.; Treasure, T. IntensiVe Care Med. 1986, 12, 104. (5) Hopf, H. W.; Hunt, T. K.; Scheuenstuhl, H.; West, J. M.; Humphrey, L. M.; Rollins, M. D. Methods Enzymol. 2004, 381, 539. (6) Hunt, T. K.; Twomey, P.; Zederfeldt, B.; Dunphy, J. E. Am. J. Surg. 1967, 114, 302. (7) Ingvar, D. H.; Lubbers, D. W.; Siesjo, B. Acta Physiol. Scand. 1960, 48, 373. (8) Kost, G. J. J. Appl. Physiol 1982, 52, 85. (9) Ling, G. N. J. Physiol. 1978, 280, 105. (10) Maas, A. I.; Fleckenstein, W.; de Jong, D. A.; van Santbrink, H. Acta Neurochir. Suppl. (Wien) 1993, 59, 50. (11) Mindt, W.; Maurer, H.; Moller, W. Arch. Gynecol. 1978, 226, 9. (12) Moore, F. A.; McKinley, B. A.; Moore, E. E. Lancet 2004, 363, 1988. (13) Shires, G. T.; Cunningham, J. N.; Backer, C. R.; Reeder, S. F.; Illner, H.; Wagner, I. Y.; Maher, J. Ann. Surg. 1972, 176, 288. (14) Silver, I. A. MicroVasc. Res. 1977, 13, 409. (15) Trunkey, D.; Carpenter, M. A.; Holcroft, J. J. Trauma 1978, 18, 166. (16) Trunkey, D.; Holcroft, J.; Carpenter, M. A. J. Trauma 1976, 16, 633. (17) Gleichmann, U.; Ingvar, D. H.; Lubbers, D. W.; Siesjo, B. K.; Thews, G. Acta Physiol. Scand. 1962, 55, 127. (18) Lubbers, D. W. Int. Anesthesiol. Clin. 1966, 4, 103. (19) Astrup, P.; Severinghaus, J. W. The history of blood gases, acids and bases; Munksgaard: Copenhagen, 1986. (20) Boekstegers, P.; Weidenhofer, S.; Kapsner, T.; Werdan, K. Crit. Care Med. 1994, 22, 640. (21) Duroux, P.; Emde, C.; Bauerfeind, P.; Francis, C.; Grisel, A.; Thybaud, L.; Arstrong, D.; Depeursinge, C.; Blum, A. L. Gut 1991, 32, 240. (22) Hoffman, W. E.; Charbel, F. T.; Edelman, G. Anesth. Analg. 1996, 82, 582. (23) Lubbers, D. W.; Gehrich, J.; Opitz, N. Life Support Syst. 1986, 4 (Suppl. 1), 94. (24) Lubbers, D. W.; Opitz, N. Z. Naturforsch. [C] 1975, 30, 532. (25) Parker, D.; Key, A.; Davies, R. S. Lancet 1971, 1, 952. (26) Peterson, J. I.; Goldstein, S. R.; Fitzgerald, R. V.; Buckhold, D. K. Anal. Chem. 1980, 52, 864. (27) Shapiro, B. A.; Cane, R. D. Crit. Care Med. 1989, 17, 573. (28) van Santbrink, H.; Maas, A. I.; Avezaat, C. J. Neurosurgery 1996, 38, 21. (29) Zauner, A.; Bullock, R.; Di, X.; Young, H. F. Neurosurgery 1995, 37, 1168. (30) Bergveld, P. IEEE Trans. Biomed. Eng. 1970, 17, 70. (31) Bergveld, P. Sens. Actuators B 2003, 88, 1. (32) Esashi, M.; Matsuo, T. IEEE Trans. Biomed. Eng. 1978, 25, 184. (33) Janata, J. Workshop on theory, design and biomedical applications of solid state chemical sensors, West Palm Beach, FL, 1978, p 41. (34) Janata, J.; Huber, R. J. Ion SelectiVe Electrode ReV. 1979, 1, 31. (35) Janata, J.; Moss, S. D. Biomed. Eng. 1976, 11, 241. (36) McBride, P. T.; Janata, J.; Comte, P. A.; Moss, S. D.; Johnson, C. C. Anal. Chim. Acta 1978, 101, 239. (37) Moss, S. D.; Johnson, C. C.; Janata, J. IEEE Trans. Biomed. Eng. 1978, 25, 49. (38) Anker, P.; Ammann, D.; Meier, P. C.; Simon, W. Clin. Chem. 1984, 30, 454. (39) Lindner, E.; Toth, K.; Pungor, E.; Morf, W. E.; Simon, W. Anal. Chem. 1978, 50, 1627. (40) McKinley, B. A. Ph.D. dissertation, University of Utah, 1980. (41) McKinley, B. A.; Saffle, J.; Jordan, W. S.; Janata, J.; Moss, S. D.; Westenskow, D. R. Med. Instrum 1980, 14, 93. (42) Johnson, C. C.; Moss, S. D.; Janata, J. A. U.S. Patent 4,020,830, 1977, US PTO. (43) Shimada, K.; Yano, M.; Shibatani, K.; Komoto, Y.; Esashi, M.; Matsuo, T. Med. Biol. Eng. Comput. 1980, 18, 741. (44) Esashi, M.; Matsuo, T. Med. Prog. Technol. 1987, 12, 145. (45) Oelssner, W.; Zosel, J.; Guth, U.; Pechstein, T.; Babel, W.; Connery, J. G.; Demuthd, C.; Grote Gansey, M.; Verburg, J. B. Sens. Actuators B 2005, 105, 104. (46) Oeseburg, B.; Ligtenberg, H. C.; Schepel, S. J.; Zijlstra, W. G. Scand. J. Clin. Lab. InVest. Suppl. 1987, 188, 31. (47) Schepel, S. J.; de Rooij, N. F.; Koning, G.; Oeseburg, B.; Zijlstra, W. G. Med. Biol. Eng. Comput. 1984, 22, 6. (48) Sibbald, A.; Covington, A. K.; Carter, R. F. Clin. Chem. 1984, 30, 135. (49) Sibbald, A.; Covington, A. K.; Carter, R. F. Med. Biol. Eng. Comput. 1985, 23, 329. (50) Sibbald, A.; Covington, A. K.; Cooper, E. A.; Carter, R. F. Clin. Chem. 1983, 29, 405.

ISFET and Fiber Optic Sensor Technologies (51) McKinley, B. A.; Houtchens, B. A.; Janata, J. Crit. Care Med. 1981, 9, 845. (52) Cunningham, J. N., Jr.; Shires, G. T.; Wagner, Y. Surgery 1971, 70, 215. (53) Trunkey, D. D.; Illner, H.; Arango, A.; Holiday, R.; Shires, G. T. Surg. Forum 1974, 25, 1. (54) Trunkey, D. D.; Illner, H.; Wagner, I. Y.; Shires, G. T. Surgery 1973, 74, 241. (55) Haemmerli, A.; Janata, J.; Brown, H. M. Anal. Chem. 1980, 52, 1179. (56) Kost, G. J. Ph.D. dissertation, University of California, 1977. (57) McKinley, B. A.; Wong, K. C.; Janata, J.; Jordan, W. S.; Westenskow, D. R. Crit. Care Med. 1981, 9, 333. (58) Gonzalez, E. A.; Moore, F. A.; Holcomb, J. B.; Miller, C. C.; Kozar, R. A.; Todd, S. R.; Cocanour, C. S.; Balldin, B. C.; McKinley, B. A. J. Trauma 2007, 62, 112. (59) McKinley, B. A.; Houtchens, B. A.; Janata, J. Ion SelectiVe Electrode ReV. 1984, 6, 173. (60) Blackburn, G. F. U.S. Patent 4,486,292, 1984, US PTO. (61) Covington, A. K.; Sibbald, A. U.S. Patent 4,567,006, 1986, US PTO. (62) Gion, H.; Kubota, K.; Nakamura, M.; Yano, M. U.S. Patent 4,641,249, 1987, US PTO. (63) Koning, G.; Bergveld, P. U.S. Patent 4,478,222, 1984, US PTO. (64) Kratochvil, J.; Ho, N.; Janata, J. U.S. Patent 4,449,011, 1984, US PTO. (65) Ligtenberg, H. C. G.; Veld, A. U.S. Patent 4,589,970, 1986, US PTO. (66) Nakamura, M.; Yano, M. U.S. Patent 4,736,748, 1988, US PTO. (67) Nakamura, M.; Yano, M.; Ikeya, H.; Tsurumi, H.; Kusudo, O.; Hamamoto, Y. U.S. Patent 4,568,444, 1986, US PTO. (68) Munoz, J.; Jimenez, C.; Bratov, A.; Bartroli, J.; Alegret, S.; Dominguez, C. Biosens. Bioelectron. 1997, 12, 577. (69) Bare, R. O.; Scherer, A. U.S. Patent 5,665,653, 1997, US PTO. (70) Baxter, R. D. U.S. Patent 4,505,799, 1985, US PTO. (71) Benton, B. W. U.S. Patent 5,833,824, 1998, US PTO. (72) Covington, A. K.; Sibbald, A. U.S. Patent 4,437,969, 1984, US PTO. (73) Najafi, N.; Massoud-Ansari, S. U.S. Patent 6,140,144, 2000, US PTO. (74) Oda, S.; Seshimoto, O.; Sueyoshi, T.; Amano, H. U.S. Patent 4,877,582, 1989, US PTO. (75) Otten, J. M.; Boekema, B. U.S. Patent 4,706,678, 1987, US PTO. (76) Oyama, N.; Shimomura, T.; Yamaguchi, S. U.S. Patent 4,816,118, 1989, US PTO. (77) Sekiguchi, T.; Hamatani, T.; Ozawa, H.; Takahashi, M. U.S. Patent 4,903,099, 1990, US PTO. (78) Uhlig, A.; Lindner, E.; Teutloff, C.; Schnakenberg, U.; Hintsche, R. Anal. Chem. 1997, 69, 4032. (79) Yuqing, M.; Jianguo, G.; Jianrong, C. Biotechnol. AdV. 2003, 21, 527. (80) Bratov, A.; Munoz, J.; Dominguez, C.; Bartroli, J. Sens. Actuators B 1995, 823, 24-25. (81) van der Wal, P. D.; van den Berg, A.; de Rooij, N. F. Sens. Actuators B: Chem. 1994, 18, 200. (82) Comte, P. A.; Janata, J. Anal. Chim. Acta 1978, 101, 247. (83) Janata, J.; McBride, P. T. J. Bioeng. 1978, 2, 459. (84) Schwiegk, S.; Mathauer, K.; Wegner, G.; Hoffmann, B. F. W.; Vogel, A. U.S. Patent 5,182,005, 1993, US PTO. (85) Shimomura, T.; Yamaguchi, S.; Uchida, N.; Mori, T.; Katsube, T. U.S. Patent 5,200,053, 1993, US PTO. (86) Shirai, T.; Suzuki, K. U.S. Patent 4,645,583, 1987, US PTO. (87) Smith, R. L.; Collins, S. U.S. Patent 4,592,824, 1986, US PTO. (88) Van Der Velden, P. M. U.S. Patent 4,552,625, 1985, US PTO. (89) Yamaguchi, S.; Shimomura, T.; Uchida, N.; Katsube, T.; Oyama, N. U.S. Patent 5,213,675, 1993, US PTO. (90) Yano, M.; Shimada, K.; Shibatani, K.; Makimoto, T. U.S. Patent 4,269,682, 1981, US PTO. (91) Butler, S. J.; Payne, R. B.; Gunn, I. R.; Burns, J.; Paterson, C. R. Br. Med. J. (Clin. Res. Ed.) 1984, 289, 948. (92) Masters, P. W.; Payne, R. B. Clin. Chem. 1993, 39, 1082. (93) Payne, R. B. Ann. Clin. Biochem. 1982, 19 (Pt 4), 233. (94) Payne, R. B. Ann. Clin. Biochem. 1991, 28 (Pt 3), 235. (95) Payne, R. B.; Buckley, B. M.; Rawson, K. M. Ann. Clin. Biochem. 1991, 28 (Pt 1), 68. (96) Saris, N. E. J. Clin. Chem. Clin. Biochem. 1988, 26, 101. (97) Thode, J.; Fogh-Andersen, N.; Siggaard-Andersen, M.; SiggaardAndersen, O. Ann. Clin. Biochem. 1983, 20 (Pt 5), 271. (98) Huang, I.-Y.; Huang, R.-S.; Lo, L.-H. Sens. Actuators B 2003, 94, 53. (99) Caras, S. D.; Janata, J. Anal. Chem. 1985, 57, 1924. (100) Caras, S. D.; Janata, J.; Saupe, D.; Schmitt, K. Anal. Chem. 1985, 57, 1917. (101) Caras, S. D.; Petelenz, D.; Janata, J. Anal. Chem. 1985, 57, 1920. (102) Harrow, J.; Janata, J. Trans. Am. Soc. Artif. Intern. Organs 1985, 31, 416. (103) Janata, J. Diabetes Care 1982, 5, 271. (104) Janata, J.; Blackburn, G. F. Ann. N.Y. Acad. Sci. 1984, 428, 286.

Chemical Reviews, 2008, Vol. 108, No. 2 843 (105) Lehmann, M.; Baumann, W.; Brischwein, M.; Ehret, R.; Kraus, M.; Schwinde, A.; Bitzenhofer, M.; Freund, I.; Wolf, B. Biosens. Bioelectron. 2000, 15, 117. (106) Lehmann, M.; Baumann, W.; Brischwein, M.; Gahle, H.-J.; Freund, I.; Ehret, R.; Drechsler, S.; Palzer, H.; Kleintges, M.; Sieben, U.; Wolf, B. Biosens. Bioelectron. 2001, 16, 195. (107) Martinoia, S.; Rosso, N.; Grattarola, M.; Lorenzelli, L.; Margesin, B.; Zen, M. Biosens. Bioelectron. 2001, 16, 1043. (108) Mohri, S.; Shimizu, J.; Goda, N.; Miyasaka, T.; Fujita, A.; Nakamura, M.; Kajiya, F. Sens. Actuators B 2006, 115, 519. (109) Ito, N.; Saito, A.; Kayashima, S.; Kimura, J.; Kuriyama, T.; Nagata, N.; Arai, T.; Kikuchi, M. Front. Med. Biol. Eng. 1995, 6, 269. (110) Dam, T. V.; Pijanowska, D.; Olthuis, W.; Bergveld, P. Analyst 2003, 128, 1062. (111) Besselink, G. A.; Schasfoort, R. B.; Bergveld, P. Biosens. Bioelectron. 2003, 18, 1109. (112) de Melo, J. V.; Soldatkin, A. P.; Martelet, C.; Jaffrezic-Renault, N.; Cosnier, S. Biosens. Bioelectron. 2003, 18, 345. (113) Soldatkin, A. P.; Montoriol, J.; Sant, W.; Martelet, C.; JaffrezicRenault, N. Biosens. Bioelectron. 2003, 19, 131. (114) Bartic, C.; Borghs, G. Anal. Bioanal. Chem. 2006, 384, 354. (115) Castellarnau, M.; Zine, N.; Bausells, J.; Madrid, C.; Jua´rez, A.; Samitier, J.; Errachid, A. Sens. Actuators B 2007, 120, 615. (116) Purushothaman, S.; Toumazoua, C.; Oub, C.-P. Sens. Actuators B 2006, 114, 964. (117) Koch, S.; Woias, P.; Meixner, L. K.; Drost, S.; Wolf, H. Biosens. Bioelectron. 1999, 14, 413. (118) Flores, F.; Artigas, J.; Marty, J. L.; Valdes, F. Anal. Bioanal. Chem. 2003, 376, 476. (119) Goldstein, S. R.; Peterson, J. I.; Fitzgerald, R. V. J. Biomech. Eng. 1980, 102, 141. (120) Lubbers, D. W.; Opitz, N.; Speiser, P. P.; Bisson, H. J. Z. Naturforsch. [C] 1977, 32, 133. (121) Lubbers, D. W. Acta Anaesthesiol. Scand. Suppl. 1995, 104, 37. (122) McKinley, B. A.; Parmley, C. L.; Morris, W. P.; Allen, S. J.; Tonnesen, A. S. Technology report: Continuous in vivo arterial blood gas monitor systems, University Hospital Consortium Clinical Practice Advancement Center, 1994. (123) Costello, D. J.; Salter, J. R.; Schlain, L. A.; Kosa, N. Methodology and clinical applications of blood gases; International Federation for Clinical Chemistry: Copenhagen, Denmark, 1990; p 41. (124) Gehrich, J. L.; Lubbers, D. W.; Opitz, N.; Hansmann, D. R.; Miller, W. W.; Tusa, J. K.; Yafuso, M. IEEE Trans. Biomed. Eng. 1986, 33, 117. (125) Lippitsch, M. E.; Wolfbeis, O. S. Anal. Chim. Acta 1988, 205, 1. (126) Opitz, N.; Lubbers, D. W. Int. Anesthesiol. Clin. 1987, 25, 177. (127) Seitz, W. R. CRC Crit. ReV. Anal. Chem. 1988, 19, 135. (128) Janata, J. Anal. Chem. 1987, 59, 1351. (129) Shapiro, B. A.; Cane, R. D.; Chomka, C. M.; Bandala, L. E.; Peruzzi, W. T. Crit. Care Med. 1989, 17, 455. (130) Haller, M.; Kilger, E.; Briegel, J.; Forst, H.; Peter, K. Crit. Care Med. 1994, 22, 580. (131) Larson, C. P., Jr.; Vender, J.; Seiver, A. Anesthesiology 1994, 81, 543. (132) Zimmerman, J. L.; Dellinger, R. P. Crit. Care Med. 1993, 21, 495. (133) Smith, B. E.; King, P. H.; Schlain, L. Int. J. Clin. Monit. Comput. 1992, 9, 45. (134) Charbel, F. T.; Hoffman, W. E.; Misra, M.; Hannigan, K.; Ausman, J. I. Surg. Neurol. 1997, 48, 414. (135) Charbel, F. T.; Hoffman, W. E.; Misra, M.; Nabhani, T.; Ausman, J. I. Acta Neurochir. Suppl. 1999, 75, 41. (136) Doppenberg, E. M.; Zauner, A.; Bullock, R.; Ward, J. D.; Fatouros, P. P.; Young, H. F. Surg. Neurol. 1998, 49, 650. (137) Hoffman, W. E.; Charbel, F. T.; Edelman, G.; Ausman, J. I. Surg. Neurol. 1996, 46, 75. (138) McKinley, B. A.; Morris, W. P.; Parmley, C. L.; Butler, B. D. Crit. Care Med. 1996, 24, 1858. (139) McKinley, B. A.; Parmley, C. L. Acta Neurochir. Suppl. 1998, 71, 177. (140) McKinley, B. A.; Butler, B. D. Crit. Care Med. 1999, 27, 1869. (141) McKinley, B. A.; Parmley, C. L.; Butler, B. D. J. Trauma 1998, 44, 119. (142) McKinley, B. A.; Ware, D. N.; Marvin, R. G.; Moore, F. A. J. Trauma 1998, 45, 633. (143) He, H.; Mortellaro, M. A.; Leiner, M. J.; Fraatz, R. J.; Tusa, J. K. J. Am. Chem. Soc. 2003, 125, 1468. (144) He, H.; Mortellaro, M. A.; Leiner, M. J.; Young, S. T.; Fraatz, R. J.; Tusa, J. K. Anal. Chem. 2003, 75, 549. (145) Tusa, J. K.; Leiner, M. J. Ann. Biol. Clin. (Paris) 2003, 61, 183. (146) Murthy, J. N.; Hicks, J. M.; Soldin, S. J. Clin. Biochem. 1997, 30, 385.

844 Chemical Reviews, 2008, Vol. 108, No. 2 (147) Papadea, C.; Foster, J.; Grant, S.; Ballard, S. A.; Cate, J. C. t.; Southgate, W. M.; Purohit, D. M. Ann. Clin. Lab. Sci. 2002, 32, 231. (148) Schneider, J.; Dudziak, R.; Westphal, K.; Vettermann, J. Anaesthesist 1997, 46, 704. (149) Steinfelder-Visscher, J.; Weerwind, P. W.; Teerenstra, S.; Brouwer, M. H. Perfusion 2006, 21, 33. (150) Zaloga, G. P.; Dudas, L.; Roberts, P.; Bortenschlager, L.; Black, K.; Prielipp, R. J. Clin. Monit. 1993, 9, 341. (151) Zaloga, G. P.; Roberts, P. R.; Black, K.; Santamauro, J. T.; Klase, E.; Suleiman, M. Crit. Care Med. 1996, 24, 957. (152) Ruzicka, J.; Ramsing, A. Scand. J. Clin. Lab. InVest. Suppl. 1982, 161, 35. (153) Sibbald, A. Med. Instrum. 1985, 19, 164. (154) Bailey, P. L.; McJames, S. W.; Cluff, M. L.; Wells, D. T.; Orr, J. A.; Westenskow, D. R.; Kern, S. E. J. Clin. Monit. Comput. 1998, 14, 339. (155) Widness, J. A.; Kulhavy, J. C.; Johnson, K. J.; Cress, G. A.; Kromer, I. J.; Acarregui, M. J.; Feld, R. D. Pediatrics 2000, 106, 497. (156) McKinley, B. A.; Parmley, C. L. J. Crit. Care 1997, 12, 214. (157) McKinley, B. A.; Parmley, C. L. J. Crit. Care 1998, 13, 190. (158) Anderson, J. R.; East, T. D. Biomed. Sci. Instrum. 2002, 38, 289. (159) Dellinger, R. P.; Carlet, J. M.; Masur, H.; Gerlach, H.; Calandra, T.; Cohen, J.; Gea-Banacloche, J.; Keh, D.; Marshall, J. C.; Parker, M. M.; Ramsay, G.; Zimmerman, J. L.; Vincent, J. L.; Levy, M. M. Crit. Care Med. 2004, 32, 858. (160) Davis, J. W.; Mackersie, R. C.; Holbrook, T. L.; Hoyt, D. B. Ann. Emerg. Med. 1991, 20, 842. (161) Davis, J. W.; Parks, S. N.; Kaups, K. L.; Gladen, H. E.; O’DonnellNicol, S. J. Trauma 1996, 41, 769. (162) Davis, J. W.; Shackford, S. R.; Holbrook, T. L. Surg. Gynecol. Obstet. 1991, 173, 473. (163) Gore, D. C.; Jahoor, F.; Hibbert, J. M.; DeMaria, E. J. Ann. Surg. 1996, 224, 97. (164) James, J. H.; Luchette, F. A.; McCarter, F. D.; Fischer, J. E. Lancet 1999, 354, 505. (165) McKinley, B. A.; Ware, D. N.; Marvin, R. G.; Moore, F. A. J. Trauma 1998, 45, 271. (166) McKinley, B. A.; Parmley, C. L. Acta Neurochir. Suppl. (Wien) 1998, 71, 177. (167) Chang, M. C.; Cheatham, M. L.; Nelson, L. D.; Rutherford, E. J.; Morris, J. A. J. Trauma 1994, 37, 488. (168) Fiddian-Green, R. G.; Haglund, U.; Gutierrez, G.; Shoemaker, W. C. Crit. Care Med. 1993, 21, S25. (169) Gutierrez, G.; Bismar, H.; Dantzker, D. R.; Silva, N. Crit. Care Med. 1992, 20, 451.

McKinley (170) Gutierrez, G.; Palizas, F.; Doglio, G.; Wainsztein, N.; Gallesio, A.; Pacin, J.; Dubin, A.; Schiavi, E.; Jorge, M.; Pusajo, J.; Klein, F.; Roman, E. S.; Dorfman, B.; Shottlender, J.; Giniger, R. Lancet 1992, 339, 195. (171) Johnson, B. A.; Weil, M. H. Crit. Care Med. 1991, 19, 1432. (172) Weil, M. H.; Tang, W.; Bisera, J. U.S. Patent 5,579,763, 1996, US PTO. (173) Weil, M. H.; Tang, W.; Bisera, J. U.S. Patent 6,055,447, 2000, US PTO. (174) Weil, M. H.; Tang, W.; Bisera, J. U.S. Patent 6,216,024, 2001, US PTO. (175) Weil, M. H.; Nakagawa, Y.; Tang, W.; Sato, Y.; Ercoli, F.; Finegan, R.; Grayman, G.; Bisera, J. Crit. Care Med. 1999, 27, 1225. (176) Cammarata, G. A.; Weil, M. H.; Fries, M.; Tang, W.; Sun, S.; Castillo, C. J. J. Appl. Physiol 2006, 100, 304. (177) Pellis, T.; Weil, M. H.; Tang, W.; Sun, S.; Csapozi, P.; Castillo, C. J. Trauma 2005, 58, 817. (178) Boekstegers, P.; Fleckenstein, W.; Rosport, A.; Ruschewsky, W.; Braun, U. Anaesthesist 1988, 37, 287. (179) Boekstegers, P.; Heinrich, R.; Gunderoth-Palmowski, M.; Grauer, W.; Fleckenstein, W.; Schomerus, H. AdV. Exp. Med. Biol. 1986, 200, 429. (180) Cairns, C. B.; Moore, F. A.; Haenel, J. B.; Gallea, B. L.; Ortner, J. P.; Rose, S. J.; Moore, E. E. J. Trauma 1997, 42, 532. (181) McKinley, B. A.; Marvin, R. G.; Cocanour, C. S.; Moore, F. A. J. Trauma 2000, 48, 637. (182) Cohn, S. M.; Nathens, A. B.; Moore, F. A.; Rhee, P.; Puyana, J. C.; Moore, E. E.; Beilman, G. J. J. Trauma 2007, 62, 44. (183) Moore, F. A.; Nelson, T.; McKinley, B. A.; Moore, E. E.; Nathens, A. B.; Rhee, P.; Puyana, J. C.; Beilman, G. J.; Cohn, S. M. J. Trauma 2007, in press. (184) Opitz, N.; Lubbers, D. W. AdV. Exp. Med. Biol. 1987, 215, 45. (185) Gracia, I.; Cane, C.; Lora-Tamayo, E. Sens. Actuators B: Chem. 1995, 24, 206. (186) Gracia, I.; Cane, C.; Lozano, M.; Esteve, J. Sens. Actuators B: Chem. 1993, 15, 218. (187) Freaney, R.; Egan, T.; McKenna, M. J.; Doolin, M. C.; Muldowney, F. P. Clin. Chim. Acta 1986, 158, 129. (188) Payne, R. B.; Jones, D. P. Ann. Clin Biochem. 1987, 24 (Pt 4), 400. (189) Bell, M. In Mars Being Red; Copper Canyon Press: Port Townsend, WA, 2007.

CR068120Y