Optics of Biological Particles
NATO Science Series A Series presenting the results of scientific meetings supported under the NATO Science Programme. The Series is published by IOS Press, Amsterdam, and Springer in conjunction with the NATO Public Diplomacy Division Sub-Series I. Life and Behavioural Sciences II. Mathematics, Physics and Chemistry III. Computer and Systems Science IV. Earth and Environmental Sciences
IOS Press Springer IOS Press Springer
The NATO Science Series continues the series of books published formerly as the NATO ASI Series. The NATO Science Programme offers support for collaboration in civil science between scientists of countries of the Euro-Atlantic Partnership Council. The types of scientific meeting generally supported are “Advanced Study Institutes” and “Advanced Research Workshops”, and the NATO Science Series collects together the results of these meetings. The meetings are co-organized bij scientists from NATO countries and scientists from NATO’s Partner countries – countries of the CIS and Central and Eastern Europe. Advanced Study Institutes are high-level tutorial courses offering in-depth study of latest advances in a field. Advanced Research Workshops are expert meetings aimed at critical assessment of a field, and identification of directions for future action. As a consequence of the restructuring of the NATO Science Programme in 1999, the NATO Science Series was re-organised to the four sub-series noted above. Please consult the following web sites for information on previous volumes published in the Series. http://www.nato.int/science http://www.springer.com http://www.iospress.nl
Series II: Mathematics, Physics and Chemistry – Vol. 238
Optics of Biological Particles edited by
Alfons Hoekstra University of Amsterdam, The Netherlands
Valeri Maltsev Institute of Chemical Kinetics and Combustion, Novosibirsk, Russia and
Gorden Videen Army Research Laboratory, Adelphi, MD, U.S.A., University of Amsterdam, The Netherlands
Published in cooperation with NATO Public Diplomacy Division
Proceedings of the NATO Advanced Research Workshop on Fluorescence and other Optical Properties of Biological Particles for Biological Warfare Agent Sensors Novosibirsk, Russia 3--6 October 2005 A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN-10 ISBN-13 ISBN-10 ISBN-13 ISBN-10 ISBN-13
1-4020-5501-3 (PB) 978-1-4020-5501-0 (PB) 1-4020-5500-5 (HB) 978-1-4020-5500-3 (HB) 1-4020-5502-1 (e-book) 978-1-4020-5502-7 (e-book)
Published by Springer, P.O. Box 17, 3300 AA Dordrecht, The Netherlands. www.springer.com
Printed on acid-free paper
All Rights Reserved © 2007 Springer No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
TABLE OF CONTENTS
Preface ..............................................................................................vii Organizational Structure..................................................................ix Light Scattering from a Cell Gorden Videen and Dat Ngo ............................................................... 1 Modeling of Light Scattering from Inhomogeneous Biological Cells Andrew K. Dunn................................................................................ 19 Angularly Resolved Elastic Scattering from Airborne Particles Paul H. Kaye, Kevin Aptowicz, Richard K. Chang, Virginia Foot, and Gorden Videen .................................................... 31 Bio-aerosol Fluorescence Yong-Le Pan, Jay D. Eversole, Paul H. Kaye, Virginia Foot, Ronald G. Pinnick, Steven C. Hill, Michael W. Mayo, Jerold R. Bottiger, Alan Huston, Vasanthi Sivaprakasam, Richard K. Chang .............................................................................. 63 Optical and Morphological Characterization of Selected Phytoplanktonic Communities Antonio Palucci ............................................................................... 165 Astro-biological Signatures T.M. Gledhill, W.B. Sparks, Z. Ulanowski, J.H. Hough, and S. DasSarma.............................................................................. 193
v
vi
Table of Contents
Modeling of Light Scattering by Single Red Blood Cells with the FDTD Method Jun Q. Lu, R. Scott Brock, Ping Yang and Xin-Hua Hu ................. 213 Optics of Erythrocytes Peter Tarasov, Maxim Yurkin, Pavel Avrorov, Konstantin Semyanov, Alfons Hoekstra, and Valeri Maltsev........................................................................... 243 Optics of Platelets Irina Kolesnikova, Sergey Potapov, Peter Tarasov, Konstantin Semyanov, and Valeri Maltsev ..................................... 261 Optics of Leucocytes Konstantin Semyanov, Alexey Zharinov, Peter Tarasov, Maxim Yurkin, Ilya Skribunov, Dirk van Bockstaele, and Valeri Maltsev........................................................................... 269 Index................................................................................................ 281
Folk singers in the three color forest, reproduced in vivid BW
PREFACE
The NATO Advanced Research Workshop on “Optics of Biological Particles” met in Novosibirsk, Russia 3 – 6 October, 2005. The focus of the meeting was the potential of light scattering in the detection and characterization of biological particles, on novel detection systems using polarized light scattering, imaging (microscopy), inelastic scattering, absorption, and emission over all EM spectral regions. The concept for the ARW came about at the culmination of a NATO Science for Peace Project, headed by Alfons and Valeri. It was thought this would be a convenient vehicle to advertise and demonstrate the apparatus resulting from this multi-year collaboration. Potential applications of the device for bio-aerosol characterization interested the third
Valeri Maltsev and Alfons Hoekstra.
member of the triumvirate. While the co-chairs were occupied by bureaucratic necessities, the logistics were worked out by the Local Organizing Committee, the seemingly endless number of students and colleagues of Valeri, and especially the fixer Dina Goloshchapova who dealt with all the difficult problems. Optics of biological Gorden Videen, Alfons Hoekstra and Valeri Maltsev at particles encompasses a conference dinner. vii
viii
Preface
great many fields, and our workshop and this volume can only scratch the surface. We were lucky to have a group of outstanding lecturers willing to invest the time to provide illuminating and entertaining lectures on the fundamental research in their fields. This book is a compilation of Valeri Maltsev demonstrates prototype of the Scanning significant contributions Flow Cytometer. taken primarily from these key lectures. We are grateful to those who were able to devote the significant time and effort necessary to document this work. While the lectures represent the key component of a NATO ARW, another critical component is providing opportunities for interactions. Not only is this important for lecturers to elucidate key points and to provide details on a more personal level, but it is also critical to provide the time for the communications that ultimately lead to advances and collaborations that will drive the field into the future. We are especially grateful for the members of the LOC who eliminated the problems and distractions that accompany extended travel and gave us opportunities for interaction. Alfons Hoekstra Valeri Maltsev Gorden Videen May, 2006
ORGANIZATIONAL STRUCTURE
CHAIRS Alfons Hoekstra
University of Amsterdam Amsterdam, The Netherlands
Valeri Maltsev
Institute of Chemical Kinetics and Combustion, Novosibirsk, Russia
Gorden Videen
Army Research Laboratory Adelphi, Maryland, USA
PRIMARY LECTURERS Valeri Maltsev - Quantitative biology - the first step in predictive biology. Role of physical methods. Valery Lopatin - Formation of light-scattering patterns of optically soft particles under variation of their parameters Gorden Videen - Aerosol elastic scattering Konstantin Semyanov - Scattering by Leucocytes, theory, simulations and experiments Alexander Priezzhev - Light scattering of erythrocytes and theirs aggregates Antonio Palucci - Optical and morphological characterization of natural phytoplanktonic communities Dirk van Bockstaele - Single cell light scattering in the biomedical field: Why would it provide more than just a trigger for fluorescence collection? Virginia Foot - Characterisation of Bioaerosols using spatial elastic scattering and fluorescence Richard K. Chang - Enriching Bio-aerosols for Species Identification by Optical or Biochemical Assay Techniques
ix
x
Organizational Structure
LOCAL ORGANIZING COMMITTEE Valeri Maltsev Peter Tarasov Dina Goloshchapova
Institute of Chemical Kinetics and Combustion, Novosibirsk, Russia Institute of Chemical Kinetics and Combustion, Novosibirsk, Russia Institute of Chemical Kinetics and Combustion, Novosibirsk, Russia
ACKNOWLEDGMENT OF SUPPORT: Primary support for the NATO ARW was provided by the NATO Science Committee. Additional funding was provided by Econova, The Russian Foundation for Basic Research, The European Research Office of the US Army, the Office of Naval Research International Field Office, the Siberian Branch of the Russian Academy of Sciences, Laboratory of Cytometry and Biokinetics, Institute of Chemical Kinetics and Combustion. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of NATO Econova, The Russian Foundation for Basic Research, the European Research Office of the US Army the Office of Naval Research International Field Office, The Siberian Branch of the Russian Academy of Sciences, the Laboratory of Cytometry and Biokinetics, Institute of Chemical Kinetics and Combustion, or the editors.
Pavel Avrorov, Maxim Yurkin, Alexey Zharinov, Dmitry Strokotov, Sergey Potpov, Irina Kolesnicova, Konstantin Semyanov, and Dina Goloschapova.
Organizational Structure
Conference attendees play in the leaves.
xi
LIGHT SCATTERING FROM A CELL A simple analytical case Gorden Videen1 and Dat Ngo2 1 US Army Research Laboratory
Adelphi Maryland 20783-1197 USA
[email protected] 2 Ngo Solutions, LLC
34 Marc Drive Howell, NJ 07731 USA
[email protected]
Abstract:
We present a semi-analytical scattering solution for a system resembling a simple cell. The model of the cell includes a spherical cytoplasm surrounded by a concentric cell membrane. Contained within the cytoplasm is a nonconcentric spherical nucleus. Because of the nature of the multi-pole expansion solution, numerical results can be acquired rapidly and accurately. Because of the spherical symmetry of the components, the light-scattering Mueller matrix elements include large amplitude oscillatory structure that may not be present in actual cells.
Keywords:
cell, light scattering, T-matrix
1.
INTRODUCTION
From a modeler’s point of view, biological systems are ugly. They are inhomogeneous and their constituent parts may vary significantly between samples. Morphological structures and internal chemistry, which defines their optical constants, are often ill-defined and also may vary. Unless they are placed in a liquid medium for which anomalous diffraction approximations can be used, their refractive indices are also significant. Even the sizes of the cells that depend on their growth conditions may vary by 100%. While numerical algorithms exist that can be used to calculate the scatter from virtually any system, this does us little good if we cannot describe the system. 1 A. Hoekstra et al. (eds.), Optics of Biological Particles, 1–17. © 2007 Springer.
2
G. Videen and D. Ngo
One school of thought teaches that if there is such extreme variation in samples that we cannot hope to describe the system of interest exactly, then we would be clever to approximate the scattering system by the simplest one, preferably one that we can solve analytically and easily. While there are obvious faults in this logic that will be discussed in more detail later, it does provide us with a starting point to consider modeling the light scattering from a common biological building block, a cell. In this chapter we describe a semi-analytical solution to the light scattered by a simple cell. This technique uses multi-pole expansions of the fields internal and external to the cell in terms of vector spherical harmonics. In this respect it may be considered a T-matrix solution. As such, the solution is amenable to all of its inherent features, like analytical expressions for system rotations and orientation averaging (Mishchenko et al., 2002). The system is shown in Figure 1. It is composed of a homogeneous spherical cytoplasm surrounded by a homogeneous spherical cell membrane. Contained within the cytoplasm is a homogeneous spherical nucleus. The nucleus can be placed at any location within the cytoplasm. Although this model is also severely limited, since the components of the cell are restricted to being spherical, it does have some advantages. First, calculations can be made rapidly. And second, the calculations are exact, even if the description of the cell is not. This model can be used to examine the effects of certain system parameters, such as the cell membrane composition or thickness, the nucleus size and position, or changes in the cytoplasm chemistry. This derivation is based on that of Videen and Ngo (1998).
2.
SOLUTION
We begin by describing the geometry of the cell shown in Figure 1. A spherical host cytoplasm of radius a1 and complex refractive index m1 is centered on the x1 , y1 , z1 coordinate system. An outer, concentric cell membrane having outer radius a3 and complex refractive index m3 is also centered on the x1 , y1 , z1 coordinate system such that a1 < a3 . A spherical inclusion nucleus of radius a2 and complex refractive index m2 is centered on the x2 , y2 , z2 coordinate system at a position x1 = 0, y1 = 0, z1 = d such that a1 − a2 > |d|. In order for the scattering geometry to be completely general, the wavevector of the incident radiation is oriented at angle α with respect to the z1 axis. The wavelength and wavevector of the plane wave in the non-absorbing, nonmagnetic incident medium are λ and k, respectively. The complex wavevector in media of refractive index mj is kj . To simplify the equations, we take the permeability of the spheres and the surrounding media to be the same. The scattering solution can be found by simultaneously satisfying the boundary conditions at each interface. We consider the fields incident on each cell
3
Light Scattering from a Cell m3 z2
a2 m1
y2 m2
z1
a3
y1
x2 x1
a1
Figure 1. Geometry of the cell: a thin cell membrane of refractive index m3 concentrically surrounds a spherical cytoplasm of refractive index m1 . A spherical nucleus of refractive index m2 is located eccentrically within the cytoplasm.
component separately. These fields are expanded in terms of the vector spherical harmonics that have the following form in this derivation:
(ρ) Mnm,j
(ρ) Nnm,j
im (ρ) m imϕj ˜ z (krj )Pn (cos θj )e − sin θj n d ˜m ϕˆj zn(ρ) (krj ) P (cos θj )eimϕj , dθj n
= θˆj
1 (ρ) m imϕj ˜ z (krj )n(n + 1)Pn (cos θj )e + = rˆj krj n d 1 d (ρ) m imϕ j θˆj rj zn (krj ) + P˜ (cos θj )e krj drj dθj n im 1 d (ρ) m imϕj ˜ P (cos θj )e rj zn (krj ) , ϕˆj krj drj sin θj n
(1)
(2)
where the index j corresponds to the coordinate system used (j = 1, 2) and (ρ) zn (krj ) are the spherical Bessel functions of the first, second, third, or fourth
4
G. Videen and D. Ngo
kind (ρ = 1, 2, 3, 4), and P˜nm (cos θj ) =
(2n + 1)(n − m)! m Pn (cos θj ), 2(n + m)!
(3)
where Pnm (cos θj ) are the associated Legendre polynomials. We assume a time dependence of exp(−iωt).
2.1
Outer Cell-membrane Interface
By expressing the field components as a vector spherical harmonic expansion centered at each spherical interface, the boundary conditions at each interface can be satisfied exactly, in a manner similar to that of Lorenz-Mie theory (cf. Bohren and Huffman, 1983). We first examine the fields that strike the outermost cell-membrane interface (r1 = a3 ). We consider an arbitrary field incident on the system that can be expanded using the spherical Bessel functions of the first kind, jn (kr1 ): E1inc =
∞ n
(1)
(1)
anm Mnm,1 + bnm Nnm,1 .
(4)
n=0 m=−n
Similarly, the scattered electric field may be expanded using the spherical (1) Bessel functions of the third kind, hn (kr1 ): E1sca =
∞ n
(3)
(3)
cnm Mnm,1 + dnm Nnm,1 .
(5)
n=0 m=−n
The fields inside the cell membrane may be expanded into incoming and outgo(2) ing spherical waves using spherical Bessel functions of the fourth kind hn (k1 r1 ) (1) and third kind hn (k1 r1 ): E1con
=
∞ n
(3)
(3)
(4)
(4)
inm Mnm,1 + jnm Nnm,1 + knm Mnm,1 + lnm Nnm,1 . (6)
n=0 m=−n
The application of boundary conditions at the outer cell-membrane interface for the above three equations yields two sets of equations:
anm
k3 k3 ψn (ka3 ) + cnm ξn(1) (ka3 ) = inm ξn(1) (k3 a3 ) + knm ξn(2) (k3 a3 ) (7) k k
anm ψn′ (ka3 ) + cnm ξn′(1) (ka3 ) = inm ξn′(1) (k3 a3 ) + knm ξn′(2) (k3 a3 )
(8)
bnm ψn (ka3 ) + dnm ξn(1) (ka3 ) = jnm ξn(1) (k3 a3 ) + lnm ξn(2) (k3 a3 )
(9)
5
Light Scattering from a Cell
bnm
k3 ′ k3 ψn (ka3 ) + dnm ξn′(1) (ka3 ) = jnm ξn′(1) (k3 a3 ) + lnm ξn′(2) (k3 a3 ) (10) k k (q)
where ψn (r) and ξn (r) (q = 1, 2) are the Riccati-Bessel functions defined by ψn (r) = rjn (r) and ξn(q) (r) = rh(q) n (r)
(11)
and the primes denote derivatives with respect to the argument.
2.2
Cytoplasm-cell-membrane Interface
In a similar fashion, we examine the fields that strike the cytoplasm-cellmembrane interface (r1 = a1 ). The fields in the region |d| < r1 < a1 may be expanded into incoming and outgoing spherical waves using spherical Bessel (2) (1) functions of the fourth kind, hn (kr1 ), and third kind, hn (kr1 ) : E1sph =
∞ n
(3)
(3)
(4)
(4)
enm Mnm,1 +fnm Nnm,1 +gnm Mnm,1 +hnm Nnm,1 . (12)
n=0 m=−n
The application of boundary conditions at the cytoplasm-cell-membrane interface yields two sets of equations: k1 k1 (1) ξn (k3 a1 ) + knm ξn(2) (k3 a1 ) = enm ξn(1) (k1 a1 ) + gnm ξn(2) (k1 a1 ), k3 k3 (13) inm ξn′(1) (k3 a1 ) + knm ξn′(2) (k3 a1 ) = enm ξn′(1) (k1 a1 ) + gnm ξn′(2) (k1 a1 ), (14) inm
jnm ξn(1) (k3 a1 ) + lnm ξn(2) (k3 a1 ) = fnm ξn(1) (k1 a1 ) + hnm ξn(2) (k1 a1 ),
(15)
k1 ′(1) k1 ξn (k3 a1 ) + lnm ξn′(2) (k3 a1 ) = fnm ξn′(1) (k1 a1 ) + hnm ξn′(2) (k1 a1 ). k3 k3 (16) Since our primary concern is with the scattered fields, we can write the scattered and internal field coefficients directly in terms of the cytoplasm internal field coefficients: jnm
(J) (J) (J) anm A(J) n + cnm Cn = enm En + gnm Gn ,
(17)
bnm Bn(J) + dnm Dn(J) = fnm Fn(J) + hnm Hn(J) ,
(18)
where J = 1 or 2. The coefficients can be found from Eqns. 7 - 10 and 13 - 16, and by applying the Wronskian formula for Riccati-Bessel functions (Abramowitz and Stegun, 1972): W ξn(1) (z), ξn(2) (z) = −2i, (19)
6
G. Videen and D. Ngo
the following expressions for these coefficients may be derived: ′(J) ′ (J) A(J) n = k1 k3 ψn (ka3 )ξn (k3 a3 ) − kk1 ψn (ka3 )ξn (k3 a3 ),
(20)
Bn(J) = kk1 ψn (ka3 )ξn′(J) (k3 a3 ) − k1 k3 ψn′ (ka3 )ξn(J) (k3 a3 ),
(21)
Cn(J) = k1 k3 ξn(1) (ka3 )ξn′(J) (k3 a3 ) − kk1 ξn′(1) (ka3 )ξn(J) (k3 a3 ),
(22)
Dn(J) = kk1 ξn(1) (ka3 )ξn′(J) (k3 a3 ) − k1 k3 ξn′(1) (ka3 )ξn(J) (k3 a3 ),
(23)
En(J) = kk3 ξn(1) (k1 a1 )ξn′(J) (k3 a1 ) − kk1 ξn′(1) (k1 a1 )ξn(J) (k3 a1 ),
(24)
Fn(J) = kk1 ξn(1) (k1 a1 )ξn′(J) (k3 a1 ) − kk3 ξn′(1) (k1 a1 )ξn(J) (k3 a1 ),
(25)
G(J) n
−
kk1 ξn′(2) (k1 a1 )ξn(J) (k3 a1 ),
(26)
−
kk3 ξn′(2) (k1 a1 )ξn(J) (k3 a1 ).
(27)
Hn(J)
2.3
=
kk3 ξn(2) (k1 a1 )ξn′(J) (k3 a1 )
=
kk1 ξn(2) (k1 a1 )ξn′(J) (k3 a1 )
Cytoplasm-nucleus Interface
Finally, we examine the fields that strike the cytoplasm-nucleus interface. We will examine these fields in the x2 , y2 , z2 coordinate system (j = 2). The fields inside the nucleus may be expressed by the spherical Bessel functions of the first kind jn (k2 r2 ): E2int =
n ∞
(1)
(1)
pnm Mnm,2 + qnm Nnm,2 .
(28)
n=0 m=−n
The fields in the cytoplasm may be expressed into incoming and outgoing (2) spherical waves using spherical Bessel functions of the fourth kind hn (k1 r2 ) (1) and third kind hn (k1 r2 ): E2ext
=
∞ n
(3)
(3)
(4)
(4)
rnm Mnm,2 +snm Nnm,2 +tnm Mnm,2 +unm Nnm,2 . (29)
n=0 m=−n
Applying boundary conditions at the inclusion sphere interface yields two sets of equations: pnm k1 ψn (k2 a2 ) = rnm k2 ξn(1) (k1 a2 ) + tnm k2 ξn(2) (k1 a2 ),
(30)
pnm ψn′ (k2 a2 ) = rnm ξn′(1) (k1 a2 ) + tnm ξn′(2) (k1 a2 ),
(31)
qnm ψn (k2 a2 ) = qnm k1 ψn′ (k2 a2 )
=
snm ξn(1) (k1 a2 )
+
snm k2 ξn′(1) (k1 a2 )
unm ξn(2) (k1 a2 ),
+
unm k2 ξn′(2) (k1 a2 ).
(32) (33)
We can eliminate the nucleus field coefficients (pnm and qnm ) to find relationships for the cytoplasm field coefficients. After a little bit of algebra, we have:
7
Light Scattering from a Cell
(2)
′(2)
rnm = tnm
snm =
k1 ξn (k1 a2 )ψn (k2 a2 ) − k2 ξn (k1 a2 )ψn′ (k2 a2 ) (1) k2 ξn (k1 a2 )ψn′ (k2 a2 )
−
′(1) k1 ξn (k1 a2 )ψn (k2 a2 )
= Qrn tnm , (34)
′(2) (2) k2 ξn (k1 a2 )ψn (k2 a2 ) − k1 ξn (k1 a2 )ψn′ (k2 a2 ) unm (1) ′(1) k1 ξn (k1 a2 )ψn′ (k2 a2 ) − k2 ξn (k1 a2 )ψn (k2 a2 )
= Qsn unm .
(35) The coefficients (Qrn and Qsn ) are similar to the Mie scattering coefficients (Bohren and Huffman, 1983).
2.4
Fields in the Host Cytoplasm
The fields interior to the cytoplasm are expressed by equations 7-10, while the fields exterior to the nucleus are expressed by equations 34 and 35. In actuality, these fields are identical: they are, however, expressed in terms of harmonic expansions about different coordinate systems. Through the translation addition theorem given in the appendix, it is possible to express one vector spherical harmonic centered about the x2 , y2 , z2 coordinate system in terms of a vector spherical harmonic expansion in the x1 , y1 , z1 coordinate system: (q)
Mnm,2 =
∞
An′
∞
Bn′
(n,m)
Mn′ m,1 + Bn′
(q)
(n,m)
(n,m)
Mn′ m,1 + An′
(q)
(n,m)
(q)
Nn′ m,1 ,
(36)
n′ =0 (q)
Nnm,2 =
(q)
Nn′ m,1 ,
(37)
n′ =0
where q denotes the order of the spherical Bessel functions (q = 3, 4). This relationship is valid in the region where r > |d| . We can equate the two sets of field expressions and express the coefficients enm , fnm , gnm , and hnm in terms of the coefficients rnm , snm , tnm , and unm . Substituting equations 36 and 37 into equation 29 yields enm =
∞
′
′
′
′
′
′
rn′ m An(n ,m) + sn′ m Bn(n ,m) ,
(38)
n′ =0
fnm =
∞
sn′ m An(n ,m) + rn′ m Bn(n ,m) ,
(39)
n′ =0
gnm =
∞
n′ =0
tn′ m An(n ,m) + un′ m Bn(n ,m) ,
(40)
8
G. Videen and D. Ngo
and hnm =
∞
′
′
un′ m An(n ,m) + tn′ m Bn(n ,m) .
(41)
n′ =0
Substituting equations 38 and 41 into equations 17 and 18 yields (J) anm A(J) n + cnm Cn =
∞
′ tn′ m An(n ,m) Gn(J) + Qrn′ En(J)
(42)
∞
′ tn′ m Bn(n ,m) Hn(J) + Qrn′ Fn(J)
(43)
n′ =0
′ s (J) + un′ m Bn(n ,m) G(J) n + Qn′ En
and
bnm Bn(J) + dnm Dn(J) =
n′ =0
′ + un′ m An(n ,m) Hn(J) + Qsn′ Fn(J) .
The exterior field coefficients of the nucleus, tnm and unm , may be calculated by eliminating the scattering coefficients, cnm and dnm , in equations 42 - 43: anm αn =
∞
tn′ m Tn′
∞
tn′ m Tn′
(n,m,1)
+ un′ m Un′
(n,m,2)
+ un′ m Un′
(n,m,1)
(44)
n′ =0
bnm βn =
(n,m,2)
(45)
n′ =0
where
(2) (2) (1) αn = A(1) n Cn − An Cn
(46)
(47) βn = Bn(1) Dn(2) − Bn(2) Dn(1)
′ (n,m,1) r (1) r (2) Tn′ = An(n ,m) Cn(2) G(1) − Cn(1) G(2) n + Qn′ En n + Qn′ En (48)
(n,m,2) (2) r (2) (1) (1) r (1) (2) (n′ ,m) Dn Hn + Qn′ Fn − Dn Hn + Qn′ Fn = Bn Tn′ (49)
(n,m,1) (2) s (2) (1) (1) s (1) (2) (n′ ,m) Cn Gn + Qn′ En − Cn Gn + Qn′ En = Bn Un′ (50)
(n,m,2) (2) s (2) (1) (1) s (1) (2) (n′ ,m) . Dn Hn + Qn′ Fn − Dn Hn + Qn′ Fn = An Un′ (51) Since the incident field coefficients are known, equations 44 and 45 represent two sets of simultaneous equations that can be solved through matrix inversion techniques for the two sets of field coefficients. Although the solution is
9
Light Scattering from a Cell
general for any incident field, we consider specifically the case of plane wave illumination whose wavevector is oriented at angle α with respect to the z1 axis as shown in Figure 1. Since we restrict the nucleus to be centered on the z1 axis, we remove the restriction on the incident plane wave for the derivation to be completely general. When the plane wave is polarized perpendicular to the x − z plane (TE), the coefficients are found to be in (n − m)(n + m + 1)P˜nm+1 (cos α) n(n + 1) − (n − m + 1)(n + m)P˜nm−1 (cos α)
anm = aTE nm =
2in+2 ∂ ˜ m P (cos α), n(n + 1) ∂α n
=
in+2 (2n + 1) = n(n + 1)
(52) (53)
(n − m + 1)(n − m + 2) ˜ m−1 Pn+1 (cos α) (2n + 1)(2n + 3)
(n + m + 1)(n + m + 2) ˜ m+1 Pn+1 (cos α) + (2n + 1)(2n + 3)
bnm = bTE nm
=
2in+2 mP˜nm (cos α) . n(n + 1) sin α
(54)
When the plane wave is polarized in the x-z plane (TM), the coefficients are found to be TE (55) anm = aTM nm = ibnm and TE bnm = bTM nm = ianm .
2.5
(56)
The Scattering Amplitudes and Efficiencies
We consider the scattering amplitudes in the far field, where kr1 ≫ ka. The scattered fields in this case are in the θˆ and ϕˆ directions. In this limit, the spherical Hankel functions reduce to spherical waves: h(1) n (kr) ∼
(−i)n ikr e . ikr
The scattering amplitudes can be expressed in the form of the matrix inc sca eikr1 Eϕ ETE S1 S4 . = inc S S Eθsca E −ikr1 3 2 TM
(57)
(58)
10
G. Videen and D. Ngo
The scattering amplitude matrix elements are solved by expanding the scattered electric fields (Eqn. 4) in terms of the vector wave functions and then expanding the vector wave functions (Eqn. 1) in terms of the polarization directions. In the far field, the rˆ component of the electric field becomes negligible compared with the θˆ and ϕˆ components. After some algebra, we have the following: ∞ n (−i)n eimϕ1 (59) S1 = n=0 m=−n
TE ∂ ˜ m TE m ˜ m P (cos θ1 ) + cnm P (cos θ1 ) , × dnm sin θ1 n ∂θ1 n
S2 = −i
∞ n
(−i)n eimϕ1
(60)
n=0 m=−n
m ˜m TM ∂ ˜ m (cos θ ) + d (cos θ ) , × cTM P P 1 1 nm nm sin θ1 n ∂θ1 n S3 = −i
∞ n
(−i)n eimϕ1
(61)
n=0 m=−n
TE ∂ ˜ m TE m ˜ m P (cos θ1 ) + dnm P (cos θ1 ) , × cnm sin θ1 n ∂θ1 n S4 =
∞ n
(−i)n eimϕ1
(62)
n=0 m=−n
TM m ˜ m TM ∂ ˜ m × dnm P (cos θ1 ) + cnm P (cos θ1 ) . sin θ1 n ∂θ1 n Using the following relationship for normalized, associated Legendre polynomials (63) P˜n−m (cos θ1 ) = (−1)m P˜nm (cos θ1 ) the following relationships between the scattering coefficients may be derived: m TE cTE nm ¯ = (−1) cnm , m+1 TM cTM cnm , nm ¯ = (−1) m+1 TE dnm , dTE nm ¯ = (−1)
where m ¯ = −m.
m TM dTM nm ¯ = (−1) dnm .
(64)
11
Light Scattering from a Cell
The scattering, extinction, and absorption efficiencies of the system are defined as the cross sections per projected area and may be expressed as 2 (65) (ka1 )2
∞ n 2 TE 2 TM 2 TM 2 TE cnm + dnm + cnm + dnm , n(n + 1) ×
Qsca =
n=1
Qext
m=−n
∞ −2 = n(n + 1) × Re 2 (ka1 ) n=1
×
n
m=−n
TE TE∗ TE TE∗ TM TM∗ TM TM∗ , cnm anm + dnm bnm + cnm anm + dnm bnm Qabs = Qext − Qsca ,
(66) (67)
where a∗nm and b∗nm are the complex conjugates of anm and bnm , respectively. The asymmetry parameter g is a measure of radiation pressure on the system and is a necessary parameter used in cell levitation. This quantity can be expressed as 4 TE∗ TM TM∗ mRe cTE nm dnm + cnm dnm 2 Qsca (ka) n,m (n − m + 1)(n + m + 1) + n(n + 2) × (2n + 3)(2n + 1) TE∗ TE TE∗ TM TM∗ TM TM∗ Re i(cTE nm cn+1m + dnm dn+1m + cnm cn+1m + dnm dn+1m .
g=
(68)
Detailed derivations for the asymmetry parameter and the efficiencies are given by Ngo (1994) and Videen et al. (1998).
3.
RESULTS AND DISCUSSION
Although Eqns. 59-62 describe the electromagnetic field scattered by the cell, the meat of the problem is solving the simultaneous equations described by Eqns. 44 and 45. To solve the infinite number of unknowns, some truncation of the coefficients is necessary. Since the dominant parameter affecting the number of modes necessary to describe the light scattering is the overall size of the system, we refer to previous discussions of convergence for a LorenzMie sphere by Wiscombe (1980) and Bohren and Huffman (1983). Perhaps the most commonly used cutoff number is N = x + 4x1/3 + 2 where x is
12
G. Videen and D. Ngo 106
S12 (percent polarization)
S11 (total intensity)
104 10
100
d = 0.0 d = 0.5 d = 1.0
105
3
102 101 1 10–1
0
-50
-100 0
30
60 90 120 150 Scattering angle ı (degrees)
180
100
0
30
60 90 120 150 Scattering angle ı (degrees)
180
0
30
60 90 120 150 Scattering angle ı (degrees)
180
100 S34 (percent polarization)
S33 (percent polarization)
50
50
0
-50
-100
50
0
-50
-100 0
30
60 90 120 150 Scattering angle ı (degrees)
180
Figure 2. Light scattering Mueller matrix elements as a function of scattering angle at three different cell nucleus locations. For this system, λ = 0.6328 µm, a1 = 2.4 µm, a2 = 1.0 µm, a3 = 2.5 µm, m1 = 1.05, m2 = 1.15, m3 = 1.20.
the size parameter of the sphere (in our case x = 2πm3 a3 /λ) and all higher order modes are set to zero. While this criterion usually provides accurate results, some care must be taken, since the placement of the nucleus near the cell membrane results in the presence of high order modes in the expansion of the internal fields. These modes are not reflected in the total number of modes necessary to describe the light scattering accurately, but their presence affects the value of the other modes. This criterion may also break down when the cytoplasm or nucleus refractive indices are excessive. We provide some sample calculations in Figure 2, which shows light-scattering Mueller matrix elements for cells having different nucleus positions. The structure of the total intensity element S11 is quite similar for the three systems shown. The forward-scatter region is determined predominantly from diffraction, which is determined by the external morphology of the system. This region shows only a small dependence on nucleus placement. This effect is reflected primarily in the backscatter region. The polarization Mueller matrix elements (matrix elements S12 , S33 , and S34 are shown) show similar sensitivities, with the backscatter region tending to show a greater dependence on nucleus position. Perhaps the most significant feature present in Figure 2 is the high-amplitude oscillatory structure. Polarization swings of nearly 200% and intensity changes
13
Light Scattering from a Cell 1 10
d= d=2 d=3
–1
10–2
/
10–3 10–4 10–5 10–6 10–7 10–8 0
30
60 90 Scattering Angle
sca
120 (degrees)
150
180
Figure 3. The average cross-polarized light-scattering intensities Icross / I for an a1 = 6.00λ and m1 = 1.05 spherical host containing an a2 = 3.00λ and m2 = 1.15 spherical inclusion. In this example, there is no cell membrane (a3 = a1 ).
of over an order of magnitude in a few degrees are not uncommon. This highamplitude oscillatory structure is a result of the use of the spherical shape approximation. Such amplitude swings are not as pronounced for particles of irregular shape and performing orientation averaging removes much more of the structure when the external morphology is non-spherical. Depending on the cell morphology, such structure may or may not be present. One other side-effect of assuming a spherical shape is that the calculated asymmetry parameters are significantly larger than for irregular particles of similar size and chemistry (Fu, 1996; Stephens et al., 1990; Takano and Liou, 1989). This severely limits the usefulness of such model calculations in performing radiative-transfer calculations in biological media like skin. Structural asymmetries can be a useful discriminating feature that is reflected in the light scattering. While differences in the light scattering can be seen in Figure 2 when the nucleus is displaced, a powerful tool for quantifying asymmetry is cross-polarization (Videen et al., 2001). The system is illuminated with one polarization state and the scattered light of the cross-polarized state is measured. For a perfectly symmetric system, e.g., when d = 0, the cross-polarized signal is zero. When the symmetry is broken (d = 0), the cross-polarized signal tends to be non-zero, and is expected to increase with the amount of asymmetry. Figure 3 shows the orientation-averaged crosspolarized light-scattering intensities of spherical host particles containing offset inclusion nuclei as a function of scattering angle. As the inclusion offset distance d increases, the cross-polarization intensity signal also increases. In this instance, the increase is almost the same at all scattering angles, making quantification of the displacement relatively straightforward.
14 (b)
1.0
0.0
-0.5
-1.0 -1.0
0.0
0.0
0.5
1.0
-1.0 -1.0
y (m)
0.0
-0.5
-0.5
-0.5
1.0
0.5
0.5
z (m)
z (m)
0.5
(c)
1.0
z (m)
(a)
G. Videen and D. Ngo
-0.5
0.0
y (m)
0.5
1.0
-1.0 -1.0
-0.5
0.0
0.5
1.0
y (m)
Figure 4. Two-dimensional angular optical scattering calculated from a spherical a1 = 6λ, m1 = 1.4599 spherical host containing a spherical m2 = 1.33 inclusion whose center is located a distance d = 3.0λ from the host center, and is illuminated at α = 90◦ . The inclusion radius is a) a2 = 1λ; b) a2 = 2λ; and c) a2 = 2.99λ. In this example, there is no cell membrane (a3 = a1 ).
Perhaps the most dramatic effect is the dependence of the spatial scattering on the position of the nucleus. Figure 4 shows spatial light-scattering intensities calculated from a spherical a1 = a3 = 6λ host containing a spherical inclusion offset at d = 3λ. The systems are illuminated at α = 90◦ , and the scattered light is shown as it would be projected on a screen one meter from the particle. The dominant patterns seen are concentric circles, the result of the Lorenz-Mie scattering of the host sphere. As the nucleus size is increased, the presence of interference ripples become apparent throughout the pattern. The most striking feature, however, is the asymmetric ring structure apparent in the bottom portion of these figures. These are diffraction rings of the inclusion, and their frequency increases with inclusion size. They are located on the side opposite to that of the inclusion, and are the result of the light refracted by the host surface and diffracted by the inclusion before exiting the host toward the side opposite the specular peak (Videen et al., 2000; Prabhu et al., 2001). These features have been measured experimentally from droplets containing inclusions (Bronk et al., 1993; Secker et al., 2000).
4.
CONCLUSION
We have provided an analytical solution for a cell-like system. The solution is such that results can be calculated rapidly and accurately. The nature of the solution also allows for rapid averaging and rotation through analytical T-matrix operations. Unfortunately, the components of the scattering system are perfectly spherical, and scattering from such regular shapes is accompanied by inherent light-scattering artifacts. Among these are high-amplitude oscillations in the light-scattering Mueller matrix elements and unrealistically large
15
Light Scattering from a Cell
asymmetry parameters. For these reasons care must be taken when applying results to real biological systems.
Appendix Stein (1961) and Cruzan (1962) derived translation addition theorems for vector spherical wave functions. These tools allow one to express a vector spherical harmonic in one coordinate system in terms of an expansion of vector spherical harmonics in another coordinate system. Since our coordinate systems are related only through a simple translation along the z axis and there is no coordinate rotation, the relations are relatively simple: (q)
Mnm,2 =
∞
(q)
(n,m,q)
(q)
(n,m,q)
(n,m,q)
Mn′ m,1 + Bn′
(n,m,q)
Mn′ m,1 + An′
An′
(q)
(A.1)
(q)
(A.2)
Nn′ m,1 ,
n′ =0 (q)
Nnm,2 =
∞
Bn′
Nn′ m,1 ,
n′ =0
where q denotes the order of the spherical Bessel functions (q = 3, 4). This relationship is (n,m,q) (n,m,q) and Bn′ can be valid in the region where r > |d| . The translation coefficients An′ (n,m,q) : calculated from the scalar translation coefficients, Cn′ (n′ − m + 1)(n′ + m + 1) (n,m,q) k1 d (n,m,q) (n,m,q) An′ = Cn′ − ′ Cn′ +1 n +1 (2n′ + 1)(2n′ + 3) k1 d (n′ − m)(n′ + m) (n,m,q) , C ′ − ′ n (2n′ + 1)(2n′ − 1) n −1 (n,m,q)
Bn′
=
−ik1 md (n,m,q) . C ′ n′ (n′ + 1) n
(A.3)
(n,m,q)
are scalar translation coefficients. These can be found via recursion relations The Cn′ given by Ngo (1994): √ (0,0,q) (A.4) Cn′ = 2n′ + 1jn′ (k1 d), √ (−1,0,q) Cn′ = − 2n′ + 1jn′ (k1 d), (A.5) (n+1,0,q) Cn′
(n,m,q)
Cn′
2n + 3 2n + 1 (n,0,q) 1 = + C ′ n′ (n + 1) 2n′ + 1 2n′ − 1 n −1 2n′ + 1 (n−1,0,q) 2n + 1 (n,0,q) − (n′ + 1) n Cn′ C , ′ 2n − 1 2n′ + 3 n +1
(n′ − m + 1)(n′ + m)(2n′ + 1) (n,m−1,q) Cn′ − = (n − m + 1)(n + m)(2n′ + 1) (n′ − m + 2)(n′ − m + 1) (n,m−1,q) k1 d − C ′ ′ (2n + 3)(n − m + 1)(n + m)(2n′ + 1) n +1 (n′ + m)(n′ + m − 1) (n,m−1,q) k1 d C ′ , (2n′ − 1)(n − m + 1)(n + m)(2n′ + 1) n −1
(A.6)
(A.7)
16
G. Videen and D. Ngo
and, (n,m,q)
Cn′
(n,−m,q)
= Cn′
.
(A.8)
From these equations, we see that (n,m,3)
= An′
(n,m,3)
= Bn′
An′ Bn′
(n,m,3) Cn′
=
(n,m,4)
= An′
(n,m,4)
= Bn′
(n,m,4) Cn′
=
(n,−m,3)
= An′
(n,−m,3)
= Bn′
(n,−m,3) Cn′
(n,m)
,
(n,m)
=
(A.9)
,
(A.10)
(n,m) Cn′ .
(A.11)
References Abramowitz M., and I.A. Stegun. (1972). Handbook of Mathematical Functions. New York, Dover. Bohren C.F., and D.R. Huffman. (1983). Absorption and Scattering of Light by Small Particles. New York, Wiley. Bronk B.V., M.J. Smith, and S. Arnold. (1993) “Photon-correlation spectroscopy for small spherical inclusions in a micrometer-sized electrodynamically levitated droplet,” Opt. Lett., 18, 93-95. Cruzan O.R. (1962). “Translational addition theorems for spherical vector wave functions,” Quart. Appl. Math. 20, 33-40. Fu Q. (1996). “An accurate parameterization of the solar radiative properties of cirrus clouds for climate models,” J. Climate 9, 2058-2082. Mishchenko M.I., L.D. Travis, and A.A. Lacis (2002). Scattering, Absorption, and Emission of Light by Small Particles. Cambridge, Cambridge University Press. Ngo D. (1994) Light Scattering from a Sphere with a Nonconcentric Spherical Inclusion, Ph.D. dissertation. Dept. of Physics, New Mexico State University, Las Cruces. Prabhu D., Melvin Davies and Gorden Videen (2001). “Light scattering calculations from oleicacid droplets with water inclusions.” Opt. Express 8, 308-313. Secker D.R., R. Greenaway, P.H. Kaye, E. Hirst, D. Bartley, G. Videen (2000). “Light scattering from deformed droplets and droplets with inclusions: I. Experimental Results.” Appl. Opt. 39, 5023-5030. Stein S. (1961). “Addition theorems for spherical wave functions,” Quart. Appl. Math. 19, 1524. Stephens G.L., S.C. Tsay, P.W. Stackhouse, and P.J. Flatau (1990). “The relevance of the microphysical and radiative properties of cirrus clouds to climate and climatic feedback,” J. Atmos. Sci. 47, 1742-1753. Takano Y., and K.-N. Liou (1989). “Solar radiative transfer in cirrus clouds. Part I: Singlescattering and optical properties of hexagonal ice crystals,” J. Atmos. Sci. 45, 3-19. Videen G., and D. Ngo (1998). “Light scattering multipole solution for a cell,” J. Biomed Opt. 3(2), 212-220. Videen, G., R.G. Pinnick, D. Ngo, Q. Fu, and P. Ch´ylek (1998). “The asymmetry parameter and aggregate particles,” Appl. Opt. 37, 1104-1109. Videen, G., W. Sun, Q. Fu, D.R. Secker, R. Greenaway, P.H. Kaye, E. Hirst, D. Bartley (2000). “Light scattering from deformed droplets and droplets with inclusions: II. Theoretical treatment,” Appl. Opt. 39, 5031-5039. Videen, G., D.R. Prabhu, M. Davies, F. Gonzlez, and F. Moreno (2001). “Light scattering fluctuations of a soft spherical particle containing an inclusion,” Appl. Opt. 40, 4054-4057. Wiscombe W.J. (1980). “Improved Mie scattering algorithms,” Appl. Opt. 19, 1505-1509.
Light Scattering from a Cell
Dirk van Bockstaele and Valery Loiko discuss conference flow.
Svetlana Khomicheva, Sergey Netesov, Alfons Hoekstra, and Gorden Videen
17
MODELING OF LIGHT SCATTERING FROM INHOMOGENEOUS BIOLOGICAL CELLS Andrew K. Dunn Biomedical Engineering Department University of Texas at Austin Austin, TX 78712 USA [email protected]
Abstract:
This chapter describes the results of a finite-difference time-domain model of light scattering from inhomogeneous biological cells. The FDTD approach enables realistic three-dimensional modeling of light scattering from cells. The effects of small cytoplasmic organelles and nuclear morphology on the angular distribution of scattered light are examined. The results suggest that the small-scale refractive-index variations found in small cytoplasmic organelles and within the nucleus largely determine the scattering properties of cells at larger scattering angles.
Keywords:
light scattering, finite-difference time-domain, cells
1.
INTRODUCTION
Optical diagnostics has seen rapid growth in the past decade. One of the primary goals of optical diagnostics is to obtain information about tissue morphology and function through the measurement of both reflected and fluorescent light. The ability to infer cellular information from these indirect measurements relies on an understanding of the interaction of light with tissue structures on the cellular and subcellular level. Due to the complicated physical composition of cells, it has been difficult to obtain such an understanding. In vivo imaging of tissue on the cellular level has become possible with confocal and two-photon microscopies, as well as optical coherence tomography (OCT). These methods rely on differences in the backscattering or fluorescence properties of small volumes within the tissue to provide contrast. The scattering differences arise from the inhomogeneities within cells and other tissue structures. Therefore, an understanding of the relationship between the scattering properties and physical structure of cells is important for interpretation 19 A. Hoekstra et al. (eds.), Optics of Biological Particles, 19–29. © 2007 Springer.
20
A.K. Dunn
of images. Each method is limited in both resolution and penetration depth and although the cause of these limitations is not well understood, scattering plays an important role. Biological cells are complex structures with components ranging in size from nanometers to tens of microns. Most cells consist of a cytoplasm (10 − 30µm) containing numerous organelles. The largest organelle is the nucleus, whose size ranges between 3 and 10 µm. The nucleus is filled with proteins, the most important one being chromatin. Mitochondria are small organelles comprised of a series of folded membranes with sizes ranging from 0.5-1.5 µm. Other cell components include endoplasmic reticulum (ER) (0.2-1 µm), lysomes (0.2-0.5 µm), and peroxisomes (0.2-0.5 µm). The total volume of each component relative to the entire cell volume varies, depending on tissue type, but typically the cytoplasm occupies 50-80% of the cell volume, the nucleus 5-10%, mitochondria 5-15%, and other organelles 1-10% (Alberts et al., 1989). The cell is surrounded by a thin (∼ 10 nm) plasma membrane and many organelles are bounded by similar membranes. The fraction of the total membrane area occupied by the plasma membrane is only 2-5%, while the membranes surrounding mitochondria (20-40%) and ER (40-60%) account for the largest fraction of membrane area (Alberts et al., 1989). In addition to the spatial structure of cell components, the dielectric properties of the cell determine its scattering properties. Although many studies have attempted to characterize the index of cell components, there are no definitive values for each component due to the difficulties inherent in these measurements, as well as the natural variations across cells and tissues. Table 1 summarizes some of the previously published refractive index values, which are used in our scattering model. Table 1.
Index of refraction values of cell components taken from previously published data.
Cell Component cytoplasm, rat liver cells mitochondria, rat liver cells lipid cytoplasm protein cytoplasm, hamster ovary cells mitochondria, rat liver melanin cytoplasm cortical cytoplasm dried protein
Index 1.38 1.40 1.48 1.35 1.50 1.37 1.42 1.7 1.358-1.374 1.353-1.368 1.58
Reference (Beuthan et al., 1996) (Beuthan et al., 1996) (Beuthan et al., 1996) (Kohl and Cope, 1994) (Kohl and Cope, 1994) (Brunsting and Mullaney, 1974) (Liu et al., 1996) (Vitkin et al., 1994) (Lanni et al., 1985) (Bereiter-Han et al., 1979) (Barer and Joseph, 1954)
Modeling of Light Scatteringfrom Inhomogeneous Biological Cells
21
The refractive index of cells and cell components is largely determined by the protein concentration within the cell component. Any cell component can be considered as a protein solution and its index can be written as (Barer, 1957) n = no + αC
(1)
where no is the index of the solvent, which is approximately equal to water for cells, α is the specific refraction increment, and C is the concentration of the solute (g/100 ml). For protein, α = 0.0018 (Barer and Joseph, 1954), and for other solutes found in cells such as sodium, α = 0.0016 (Barer and Joseph, 1954). While the specific refraction increments are similar for proteins and other solutes, proteins play the largest role in determining the index of refraction because their concentrations in terms of weight per volume are considerably greater than other solutes (Barer, 1957). In this chapter we summarize some of our work on modeling the scattering properties of inhomogeneous biological cells. The finite-difference timedomain (FDTD) method was chosen due to its relative simplicity and its ability to incorporate arbitrary three-dimensional spatial variation in dielectric properties. We find that the composition of organelles within the cell largely determines the higher angle scattering profile of the cells.
2.
FINITE-DIFFERENCE TIME-DOMAIN MODEL
The FDTD algorithm was developed in 1966 by Yee (Yee, 1966) and has been used extensively to solve a wide variety of electromagnetics and scattering problems (Taflove, 1995). The FDTD approach is a full vector solution of the electric and magnetic fields in a small region surrounding an object. The electric and magnetic fields are discretized on a spatial grid and are updated at alternate half steps in time. An example of the discretized equation for the x component of the electric field, Ex , is Exn+1 (i + 1/2, j, k) = Ai+1/2,j,k Exn (i + 1/2, j, k)
(2)
+Bi+1/2,j,k [Hzn+1/2 (i + 1/2, j + 1/2, k) − Hzn+1/2 (i + 1/2, j − 1/2, k) +Hyn+1/2 (i + 1/2, j, k − 1/2) − Hyn+1/2 (i + 1/2, j, k + 1/2)] where i, j, k are indices to the three-dimensional spatial grid and n is the time index. The spatially varying coefficients A and B are given by σ(i, j, k) ǫ(i, j, k) ∆t ǫ(i, j, k)δ
Ai,j,k = 1 −
(3)
Bi,j,k =
(4)
22
A.K. Dunn
Figure 1. Illustration of inhomogeneous cells generated for use in the FDTD program.
where ǫ and σ are the spatially varying permittivity and conductivity. All three Cartesian components of both the electric and magnetic fields are updated in a similar manner. To construct the three-dimensional cells in our model, each cell component is constructed as an ellipsoid and assigned a different refractive index (Table 1). The ellipsoids are created by assigning all grid points within the ellipse the corresponding ǫr for that component. The organelles are distributed throughout the cytoplasm using a series of random numbers for both placement and size. The center point of each organelle is chosen by generating sets of 3 uniformly distributed random numbers until the selected point fell within the cytoplasm. It is possible to preselect the location of each organelle, rather than place them randomly, but for a large number of organelles, random placement is considerably more simple and results in a better distribution within the cytoplasm. The size distribution of the small organelles is controlled by specifying an average diameter ravg , and a deviation from the mean, rd . Each of the three axes of each ellipsoid is determined from r = ravg + ξrd where ξ is uniformly distributed on [−1, 1]. An example of a cell used in the program is shown in Figure 1. The image on the left shows a volumetric view of the cell and the image on the right contains two slices through the cell. In this cell, the cytoplasm has a mean diameter of 13 µm, the nucleus is approximately 3.5 µm in diameter and the organelle volume fraction is 8%. The average organelle size is 0.5 µm with a deviation of 0.3 µm. The FDTD model was used to determine the far field angular scattering pattern, Fs (θ, φ) for a variety of cells. The far field pattern is calculated using a far field transform of the E and H fields on a closed surface surrounding the cell. All of the results are shown only as a function of θ, where Fs (θ, φ) has been averaged over azimuthal angle, φ,
Modeling of Light Scatteringfrom Inhomogeneous Biological Cells 10
23
-3
10-4 10-5
P(θ)
10-6 10-7 10-8 10
-9
10
-10
10
-11
10
-12
vf = 0.15 vf = 0.05 vf = 0
0
20
40
60 80 100 120 Scatter Angle (degrees)
140
160
180
Figure 2. Effect of small organelles, such as mitochondria on the scattering pattern of cells (cell diameter = 15 µm). Each curve represents a cell with different volume fractions of the organelle.
1 P (θ) = 2π
2π
Fs (θ, φ) dφ.
(5)
0
3.
ANGULAR SCATTERING PATTERNS OF CELLS
In this section, we summarize the results of several FDTD simulations of heterogeneous cells. In particular, we examine the effects of small cytoplasmic organelles and the morphological properties of the nucleus on the angular distribution of scattered light. Some of these results have been previously described as well (Dunn and Richards-Kortum, 1996).
3.1
Effects of Small Organelles
The effects of small organelles on the far-field scattering patterns of single cells are an important factor in determining the macroscopic, tissue-level scattering properties. To assess the effects of cytoplasmic organelles, such as mitochondria, on the scattering pattern, simulations were run on cells with varying amounts of organelles. Figure 2 shows the effect of small organelle volume fraction on the angular distribution of scattered light. The cell diameter in each case was 15 µm and the cell consists of a 5 µm nucleus (homogeneous) as well as small cytoplasmic organelles with a mean diameter of 0.5 µm and a volume fraction of either 0, 0.05 or 0.15. Note that in all of the results, the apparent increase in intensity at θ ∼ 180◦ is a numerical artifact due to the imperfect
24
A.K. Dunn 280
0.995
2
0.994
240 220
0.993
g
Scattering Cross-Section (µm )
260
200 g 180
0.992 Csca
160 140 0
0.05
0.1 Volume Fraction
0.991 0.2
0.15
Figure 3. Scattering cross section and anisotropy as a function of organelle volume fraction.
boundary conditions used in the FDTD model. We use a second-order Liao absorbing boundary condition (Liao et al., 1984) and note that much of this artifact could be eliminated by using a perfectly matched layer boundary condition (Berenger, 1994) instead. The presence of the small organelles results in an intensity increase at large angles, which begins at θ ∼ 20◦ for the cell in Figure 2. This pattern demonstrates the dependence of the scattered intensity on organelle volume fraction. To quantitatively compute the effects of organelle volume fraction, the scattering cross section and anisotropy were computed as a function of organelle volume fraction. The scattering cross section is computed by integrating the scattering pattern over all angles, C=
2π π 0
Fs (θ, φ) sin θdθ dφ.
(6)
0
The anisotropy, g, is computed from the normalized phase function, p(θ), g=
π
p(θ) cos θ sin θ dθ
(7)
0
where p(θ) =
π
P (θ)
.
(8)
P (θ) sin θ dθ
0
In Figure 3 the scattering cross section and anisotropy are plotted as a function of organelle volume fraction. The cross section increases with volume
Modeling of Light Scatteringfrom Inhomogeneous Biological Cells
25
fraction, indicating an increase in the total amount of scattered light, while the anisotropy decreases, reflecting the increase in high angle scatter from the organelles. Despite the decrease in g with volume fraction, the difference in g between a volume fraction of 0 and a volume fraction of about 18% is only about 0.004. The fact that g > 0.99 also indicates that the anisotropy may not be sufficient to completely characterize the scattering patterns of single cells, since the pattern is so highly forward peaked. Instead, higher order moments of the scattering pattern may be a more appropriate quantification of the angular spread of scattered light for single cells.
3.2
Effect of Nucleus Size and Homogeneity
The effect of nuclear morphology on scattering properties is important to understand because this is where many of the distinctions between normal and cancerous cells are made. In general, cancerous cells are characterized by an increased nuclear to cytoplasmic ratio. Since diseased cells are dividing more rapidly than normal cells, diseased cells contain greater amounts of nuclear proteins, such as chromatin (Alberts et al., 1989). The index of refraction of protein depends on its concentration (Barer and Joseph, 1954), but is greater in areas of the cell with large protein densities, such as the nucleus. To study the relationship between the physical structure of the nucleus and its scattering properties, the FDTD model was used to compute the scattering properties of cells with nuclei ranging in size from 2.5 to 9 µm for both homogeneous and inhomogeneous nuclei. The cytoplasm was assumed to be homogeneous and the variations in the refractive index of the nuclei were uniformly distributed between ∆n = 0.02 and 0.08 at an average spatial frequency of 2.5µm−1 . The scattering properties of cells with homogeneous nuclei (no refractiveindex variations) with diameters of 2.5, 5.0, and 9.0 µm are plotted in Figure 4. A slight increase in forward scatter exists for the cell with the largest nucleus (9 µm), but there is little difference in the patterns at angles greater than about 5◦ . A more realistic model of the nucleus takes into account the spatial variations in refractive index within the nucleus, caused by structures such as the nucleolus and nuclear proteins. The average index of the nucleus has been measured to be about 1.39 (Brunsting and Mullaney, 1974) and the variations in nuclear index described above were about this value. Figure 5 demonstrates the effect of the inhomogeneous nucleus for the same cell and nuclear diameters used in Figure 4. In contrast to the homogeneous nucleus, there are significant differences in the scattering patterns for different nucleus sizes. In general, the scattered intensity increases with nuclear diameter at angles greater than 5◦ , while the intensity change is less significant at angles less than 5◦ .
26
A.K. Dunn -3
10
d = 5.0 µm d = 2.5 µm d = 9.0 µm
-4
10
-5
10
-6
10 P(θ)
-7
10
10-8 -9
10
-10
10
10-11 -12
10
0
20
40
60 80 100 120 Scatter Angle (degrees)
140
160
180
Figure 4. Scattering patterns for cells with homogeneous nuclei of different size.
-3
10
d = 5.0 µm d = 2.5 µm d = 9.0 µm
-4
10
-5
10
-6
P(θ)
10
-7
10
-8
10
-9
10
-10
10
-11
10
0
20
40
60 80 100 120 Scatter Angle (degrees)
140
160
180
Figure 5. Scattering patterns for cells with inhomogeneous nuclei of different size.
Modeling of Light Scatteringfrom Inhomogeneous Biological Cells
27
The behavior of the inhomogeneous nucleus closely resembles the calculated behavior of cytoplasmic organelles (Figure 2), where an increase in the volume fraction of the organelles results in an increase in the total scattered intensity, particularly at high angles. Since the spatial scales of the organelles and nuclear inhomogeneities are approximately equal, the net effect on the scattering patterns is also similar. Since the spatial variations in index occur within the entire nucleus, a larger nucleus contains more spatial variations, analogous to an increase in cytoplasmic organelle volume fraction. One potential difference, however, is that the nuclear variations are confined to a smaller volume, depending on nuclear diameter, while the cytoplasmic organelles are distributed throughout the cytoplasm. Therefore, more multiple scattering may occur within the nucleus where the inhomogeneities are more dense.
3.3
Mie Theory Approximation to Cell Scattering
The major drawback of the FDTD method is its large demand on computational resources, and it may not be possible to access sufficient computing hardware in some cases. Since Mie theory is often used as an approximation to cell scattering, it is useful to know how accurate this approximation is. Since Mie theory assumes a homogeneous sphere with a fixed index of refraction, a suitable choice for n could be the volume averaged refractive index, navg , navg =
N
fi ni
(9)
i=1
where fi is the volume fraction and ni is the index of refraction of the i − th component, and N is the number of distinct cell components. Figure 6 shows a comparison of the scattering patterns of an inhomogeneous cell and a homogeneous sphere of the same size whose index is determined by Equation 9. The cell contains organelles at a volume fraction of 0.1 with an index of 1.46 and a nucleus with an index of 1.39. The scattered intensity of the homogeneous sphere is significantly less than that of the FDTD computed pattern for the cell at angles greater than about 30◦ . However, the scattering cross section of the homogeneous sphere is 414 µm2 , while the cross section of the inhomogeneous cell is 345 µm2 . Despite the higher cross section of the sphere, the scattered intensity is less at high angles, which is evident from the anisotropy values for the two curves: g = 0.9976 for the sphere and g = 0.9831 for the cell. While the approximation of the index of refraction with Equation 9 takes into account the volume fraction of the organelles, it does not consider the size distribution of the cell components. The small size of the organelles is responsible for the increase in high angle scatter, and this is the primary difference
28
A.K. Dunn 10
-2
10-3 10-4 10
-5
P(θ)
10-6 10
-7
10
-8
10
-9
10
FDTD cell
-10
10-11 Mie with average index
10-12 10
-13
0
20
40
60 80 100 120 Scatter Angle (degrees)
140
160
180
Figure 6. Comparison of FDTD and Mie theory scattering patterns for an inhomogeneous cell and a homogeneous sphere with volume averaged index of refraction.
between the two plots in Figure 6. Therefore, to accurately predict the scattering pattern at all angles, a more exact method, such as FDTD, must be used. However, because the cross section of the FDTD and average index Mie approximation are approximately equal, this method may be used to predict total tissue scattering. While this may not be useful for high resolution imaging, where the detailed scattering pattern is required, it could potentially be applied to light dosimetry problems that do not require detailed information about the scattering pattern.
4.
CONCLUSIONS
The FDTD approach can reveal detailed information about the effects of various morphological parameters on the scattering properties of cells. One limitation of the method however, is the large computational resources required to model even a single cell. As computing hardware advances, it may become feasible to model tissue level scattering with the FDTD method, enabling one to examine the influence of changes at the subcellular level on the macroscopic scattering properties of tissue.
References Alberts, B., Bray, D., Lewis, J., Raff, M., Roberts, K., and Watson, J. (1989). Molecular biology of the cell. Garland Publishing, New York. Barer, R. and Joseph, S. (1954). Refractometry of living cells. Quarterly Journal of Microscopical Science, 95:399–423. Barer, Robert (1957). Refractometry and interferometry of living cells. Journal of the Optical Society of America, 47:545–556.
Modeling of Light Scatteringfrom Inhomogeneous Biological Cells
29
Bereiter-Han, J., Fox, C., and Thorell, B. (1979). Quantitative reflection contrast microscopy of living cells. Journal of Cell Biology, 82:767–779. Berenger, Jean-Pierre (1994). A perfectly matched layer for the absorption of electromagnetic waves. Journal of Computational Physics, 114:185–200. Beuthan, J., Minet, O., Helfman, J., and Muller, G. (1996). The spatial variation of the refractive index in biological cells. Physics in Medicine and Biology, 41:369–382. Brunsting, A. and Mullaney, P. (1974). Differential light scattering from spherical mammalian cells. Biophysical Journal, 14:439–453. Dunn, A. and Richards-Kortum, R. (1996). Three-dimensional computation of light scattering from cells. IEEE Journal of Special Topics in Quantum Electronics, 2:898–905. Kohl, M. and Cope, M. (1994). Influence of glucose concentration on light scattering in tissue. Optics Letters, 17:2170–2172. Lanni, F., Waggoner, A., and Taylor, D. (1985). Internal reflection fluorescence microscopy. Journal of Cell Biology, 100:1091. Liao, Z.P., Wong, H.L., Yang, B.P., and Yuan, Y.F. (1984). A transmitting boundary for transient wave analysis. Sci. Sin., Ser. A, 27:1063–1076. Liu, H., Beauvoit, B., Kimura, M., and Chance, B. (1996). Dependence of tissue optical properties on solute-induced changes in refractive index and osmolarity. Journal of Biomedical Optics, 1:200–211. Taflove, Allen (1995). Computational electrodyamics: the finite-difference time-domain method. Artech House. Vitkin, I., Woolsey, J., Wilson, B., and Anderson, R. (1994). Optical and thermal characterization of natural (sepia oficinalis) melanin. Photochemistry and Photobiology, 59:455–462. Yee, K. (1966). Numerical solutions of initial boundary value problems involving maxwell’s equations in isotropic media. IEEE Transactions on Antennas and Propagation, AP-14:302– 307.
ANGULARLY RESOLVED ELASTIC SCATTERING FROM AIRBORNE PARTICLES Potential for characterizing, classifying, and identifying individual aerosol particles Paul H. Kaye,1 Kevin Aptowicz,2 Richard K. Chang,3 Virginia Foot,4 and Gorden Videen5 1
University of Hertfordshire, Hatfield AL10 9AB, UK West Chester University, West Chester, PA 19383, USA 3 Yale University, PO Box 208284 New Haven, CT 06520-8284, USA 4 Defence Science & Technology Laborator, Porton Down, Salisbury, Wilts SP4 0JQ, UK 5 University of Amsterdam, Kruislaan 403, 1098 SJ Amsterdam, The Netherlands and Army Research Laboratory, 2800 Powder Mill Road, Adelphi MD 20783, USA 2
Abstract:
Analysing the light scattering properties of individual airborne particles has become a powerful tool by which they may be characterized, classified, and in some cases, identified. The approach offers a non-invasive, nondestructive, and potentially real-time monitoring capability that has widespread application in environmental pollution and occupational fields as well as in the detection of possible deliberate releases of pathogens. In this chapter, we provide an overview of the historical development of the theoretical models and experimental techniques underpinning angularly resolved light scattering, address key methods of data analysis used to derive particle characteristics, and describe some of the very latest research results in the field.
Key words:
Elastic light scattering; angular-resolved scattering; TAOS; aerosols; particle size; particle shape.
1.
INTRODUCTION
Elastic scattering by small particles has the largest intensity signal of all optical diagnostic techniques. Angularly resolved elastic scattering 31 A. Hoekstra et al. (eds.), Optics of Biological Particles, 31–61. © 2007 Springer.
32
P.H. Kaye et al.
intensity contains information rich in particle morphology and chemistry (size, shape and index of refraction). The information is so rich that it often requires some type of pattern recognition algorithms to unravel some of the basic information. In fact, the inversion of angular scattering characteristics to obtain particle morphology is an extremely difficult technical problem and may not have a unique solution. Even for the simplest case of homogeneous spherical particles, it is still a challenge to extract the complex index of refraction and the particle size from the measured data (intensity versus polar and azimuthal scattering angles and φ, respectively). The spatial alignment of the collection optics, the dynamic range of the detector, and signal-to-noise ratio all affect the accuracy of the parameters that we wish to extract. The situation is exacerbated for non-spherical particles and clusters of particles. Any inversion technique (i.e. interpreting the spatial light scattering pattern to determine the characteristics of the particle that produce it) is complicated by the fact that orientation of the non-spherical particle or the particle cluster may vary relative to the axis of illumination. Compounding this difficulty is the fact that the calculations are extremely computer intensive and usually place restrictions on the particle size parameter (x = 2 a/ where 2a is the particle extent and the wavelength of the illuminating radiation) and the cluster number, typically to be < 30. Nevertheless, such computationally intensive calculations are being performed to compare with experimental results. To bring the subject of elastic light scattering up to date and even projecting future advances being driven by various research interests, we have produced this chapter, starting from its inception, working through its many stages, till the present and beyond. What results is an inter-woven chapter giving the reader a good sense of the timeline of important discoveries, technical advancements, and future directions for elastic light scattering research and applications. Prof. Richard K. Chang Yale University
2.
COMPUTATIONAL SIMULATIONS OF LIGHT SCATTERING
The first theoretical derivation of the light scattered by a finite-sized particle was given by Lorenz (1890) and Mie (1908) for a homogeneous sphere. This theory for the most fundamental of particles has had a great influence on how we view the scattering from aerosol particles. For nearly a century, until other theories were developed and computers made it
Angularly Resolved Elastic Scattering from Airborne Particles
33
practical to use more numerically intense algorithms, it was almost the only method used to calculate the light scattered by aerosol particles. Its widespread use colored many of our perceptions of how these particles scatter light. Spheres are a very special class of particles. Most obviously, they are completely symmetric, and the total intensity of their scattering can be completely described using only the polar angle. It was many years before light-scattering measurements were routinely taken in both scattering angles and analysis to obtain information using the azimuthal scattering angle φ is, at this writing, still in its infancy. This symmetry also contributes to the very sharp angular resonances that are not present for more irregular particles. Such differences were not appreciated until other techniques capable of calculating scattering properties from other types of particles became available. For many years, and even up to today, Lorenz-Mie theory has been used to represent the scatter from irregularly shaped aerosol particles. In the minds of many, it is better to use an exact theory of the wrong particle than to use an approximate theory of the right particle. The consequences of this assumption are largely unknown, but the differences between the scattering from a sphere and from most naturally occurring irregularly shaped aerosols are vast. As light-scattering studies became more widespread, computer codes became available to calculate light scattering. Bohren and Huffman (1983) popularized the subject with an easy-to-read volume complete with a FORTRAN code of the Lorenz-Mie theory in the appendix. Barber and Hill (1990) provided a number of well documented codes to calculate scattering from several particle systems with emphasis on the T-matrix method. The field exploded with the development of the worldwide web. Mishchenko and Travis (1998) developed a state-of-the-art T-matrix code to calculate the scattering from ellipsoids and Chebyshev particles and distributed it freely on the web. During this timeframe, others began distributing their codes and web libraries of computer codes were founded. The oldest and currently most comprehensive was set up by Wriedt at the University of Bremen (www.t-matrix.de). Currently, there are three well developed sets of algorithms that are widely available for calculating the scattering from irregularly shaped particles. An early approach to calculate scattering by spheroidal particles was to use separation of variables in spheroidal coordinates, in a direct generalization of Lorenz-Mie theory, and to match the fields at the boundaries (Asano and Yamamoto, 1975). The approach works and is exact. The main drawbacks are that it only works for spheroids, and it requires calculating spheroidal wave functions which can be problematic. The T-matrix theory of Waterman (1971) was one of the first tools developed to treat more irregularly shaped particles. It can be thought of as
34
P.H. Kaye et al.
a generalization of Lorenz-Mie theory. The electromagnetic fields are expanded in terms of the vector spherical harmonics (VSH). These functions are assumed to be able to represent the fields at all interfaces, although it is not clear what the limitations of this assumption are. The Tmatrix relates the scattered field vector to the incident field vector as do the scattering coefficients in Lorenz-Mie theory, but for non-spherical particles there is mode mixing between the elements. The boundary conditions on the electric and magnetic fields are enforced on the surfaces of the scatterer. The primary advantage of using the T-matrix over other theories is that it is semi-analytical, so results can be obtained relatively rapidly. This is extremely beneficial, since the number of calculations for most algorithms typically scale with the cube of the particle dimension. In addition, the major computational investment is finding the T-matrix for a particle and once this is found, the scattering properties are completely defined. This means that the scatter from the particle in any orientation can be found at little extra cost. In addition, analytic formulae exist to calculate orientationaveraged scattering properties of the particles. The primary disadvantages for the T-matrix are convergence for highly elongated particles and the difficulty in dealing with particle heterogeneities, although some specific algorithms have been developed to address these issues. The discrete dipole approximation (DDA) developed by Purcell and Pennypacker (1973) addresses many of the limitations of the T-matrix. The approach can be seen as an example of the generalized method of moments (MoM) as described by Harrington (1968), which is used extensively for both forward and inverse problems. There are, however, multiple formulations of each of these approaches, and multiple ways to compare their mathematical interrelations (Lakhtakia, 1993; Lakhtakia and Mullholland, 1993), so that the similarities may at times be overlooked. In the DDA approximation the particle is represented by an array of dipoles each having a polarizability corresponding to that of the refractive index of the particle at that location. The scattering properties of the particle can be found from the dipole moments induced at each location by the incident field and the interactions of all the other dipoles in the particle system. From a little different perspective, the DDA (like the comparable MoM) is a solution to a volume integral equation for the electromagnetic fields in which the scatterers are mathematically discretized. The advantage of the DDA and other MoM approaches is that in theory these can be used to calculate the light scattered from any particle system, even heterogeneous ones. The disadvantage is that the algorithms are computationally intensive in both time and memory, and convergence problems may occur for large refractive indices. As the refractive index increases, the resolution of the dipole mesh must increase, and with it the costs. Currently, with modest
Angularly Resolved Elastic Scattering from Airborne Particles
35
use of parallel processing there is no problem calculating the light scattered by water particles whose spatial dimension is 10 times the wavelength. This is by no means the current upper limit. Like the T-matrix method, the DDA came into widespread use when Draine and Flatau (1994) provided a free version of their algorithm on the web (http://astro.princeton.edu/~draine/). Like the DDA, the finite-difference time-domain (FDTD) algorithm developed by Yee (1966) is a numerical technique in which the scattering particle is discretized onto a matrix in which each element has a refractive index corresponding to that of the particle. Unlike other methods, and as its name implies, the FDTD is solved not at a specific wavelength in the frequency domain, but in the time domain: the boundary conditions at each element are satisfied as the incident field marches through the particle space in time. The near field in the time domain is transformed to the frequency domain using a FFT and finally the far field is found from the near field using the volume integral equation. The major problem with initial FDTD algorithms was spurious reflections off the boundary of the array. These have been solved with the application of so-called perfectly absorbing boundary conditions. What remains is a very stable, accurate, albeit slow algorithm for calculating scattering from particles. Like the DDA, the resolution of the discretization space must increase with refractive index. Also, the modest use of parallel processing allows us to calculate light scattered by water particles whose spatial dimension is 10 times the wavelength without difficulty. Although there have been no direct comparisons of the DDA with FDTD, it appears that the DDA may be faster, but the FDTD may be more stable for large refractive indices. Of course this depends greatly on the algorithm used. The FDTD has not enjoyed nearly the widespread use as the DDA or T-matrix, probably the result of it not having early support on the worldwide web. We have touched on the principal techniques for calculating light scattering from aerosol particles. There are many variations on these techniques depending on the individual applications. Some of these are detailed in the bible compiled by Mishchenko, Hovenier and Travis (2000).
3.
BRIEF HISTORICAL REVIEW OF INSTRUMENT DESIGN INNOVATIONS
Optical systems to detect angularly resolved elastic light scattering from single flowing airborne particles have been evolving for almost half a century, as shown in Table 1. Initially, the systems used a single detector element with a broadband source and interference filters. Today,
36
P.H. Kaye et al.
instruments are built using laser sources and multi-element detector arrays. In the section that follows, a brief history of this evolution is presented with emphasis on major innovations on instrument design, rather than a complete survey of all the different variations of instruments that were published. Recommended review articles discussing these design implementations have been written by Kerker (1997) and Kaye (1998). Table 1. Timeline of design innovations Year Design Innovation 1960 Using electro-dynamic levitation to suspend the aerosol particle in mid-air, detected a span of polar angles (40° to 110°) in a fixed scattering plane. 1971 Commercialization of Differential II by Science Spectrum Inc. 1973 Use of annular strip of an ellipsoidal reflector, detected light scattering from single particles in laminar airflow over a polar angle span of almost 360° (7° to 173° and 187° to 353°); fixed scattering plane. 1980 Improved instrument based on ellipsoidal reflector by detecting the scattered light in parallel with a circular array of 60 photodiodes. 1988 DAWN-A introduced by Wyatt Technology Corp. that detects 16 scattering angles with varying polar angles and azimuthal angles using optical fibers and photomultiplier tubes. 1992 Ellipsoidal mirror for light collection, detected a scattering profile (high angular-resolution scattering pattern over a large solid angle) using an intensified CCD detector. The detected polar angle span from 30° to 141° and the azimuthal angles spanned 360°. 1997 Utilizing a custom 33-element photodiode detector array with information efficient layout, increased particle analysis rates in excess of 103/s. 2000 Measured simultaneous light scattering and intrinsic fluorescence for improved particle characterization. 2000 Using a camera lens in the Abbé sine condition, measured scattering patterns from spherical clusters that could be compared with theoretical calculations. 2004 Captured simultaneous light scattering patterns over identical angular ranges from two laser sources.
Reference Gucker and Egan, 1961
Phillips and Wyatt, 1972 Gucker et al., 1973
Bartholdi et al., 1980
Wyatt et al., 1988
Kaye et al., 1992
Kaye et al., 1997
Kaye et al., 2000 Holler et al., 2000
Aptowicz et al., 2004
3.1
Detection of Light Scattered in a Single Plane
3.1.1
Polar scattering in a single azimuthal plane
In the late 1940s and early 1950s, Frank T. Gucker led the development of optical particle sizers (Gucker et al., 1947a; Gucker et al., 1947b). In
Angularly Resolved Elastic Scattering from Airborne Particles
37
these devices, a fine stream of aerosol particles traverses an intense beam of light, then a photomultiplier tube (PMT) detects the flashes of light caused by elastic scattering. The resulting electrical pulses are sorted based upon the pulse height, which loosely correlates to particle size. In 1961, realizing that there was more information in resolving the scattered light over different angles, Gucker designed and constructed a device to detect multiangle light scattering that utilized electro-dynamic levitation and a single photo multiplier tube on a motor-driven rotation stage (Gucker and Egan, 1961; Gucker and Rowell, 1960). The device could detect light scattered in a single fixed scattering plane over a polar angle θ range from 40° to 110°, with an angular resolution of 5.2°. A little under ten years later, Phillip J. Wyatt led a design team that constructed an instrument (the Differential II) that had the same basic design, but improved the components (e.g. a laser source and solid state electronics) and detected the scattered light over a larger angle range from 10° to 170° with a 2° angular resolution (Phillips et al., 1970). This improved design was capable of extracting information about a particle’s morphology (Wyatt, 1972) with success mainly limited to the optical properties of homogenous spheres or coated spheres (e.g. certain bacterial spores).
Figure 1. Ellipsoidal mirror used to detect almost 180 degrees of scattered light from an aerosol particle in laminar airflow. Adapted from (Marshall et al., 1976).
Recognizing the limited utility for instruments based upon electrodynamic levitation due to extremely low thru-put capabilities, Gucker set out to develop another instrument, one that could detect angularly resolved
38
P.H. Kaye et al.
scattered light from aerosols in a laminar flow. The result was an extremely clever design (Gucker et al., 1973). The key element is a narrow strip of mirror in the form of a prolate ellipsoid as shown in Fig. 1. A laser beam, which enters and exits through holes in the mirror, is scattered by individual airborne particles carried in a laminar airflow at the focal point of the mirror. A portion of the scattered light is reflected by the mirror to the second focal point where it is intercepted by a rapidly rotating disc fitted with an aperture. The aperture only allows a small solid angle of the ring of scattered light to propagate to a photomultiplier tube (located at the second focal point) that records a time varying signal corresponding to a ~360° polar sweep of the scattering light. Using this device, the size and refractive index of latex spheres (diameter ≅ 1.2 µm) were determined independently with a fractional uncertainty of about 0.5% and 0.7% respectively (Marshall et al., 1976). This setup was improved upon by Bartholdi et al. (1980) by intercepting the circle of light reflected by the strip of ellipsoidal mirror with a circular photodiode array made up of 32 elements allowing for higher data collection rates. 3.1.2
Azimuthal scattering at a single polar angle
The instruments discussed above detect light elastically scattered in a single scattering plane, in particular polar scattering in an azimuthal plane. For spherical particles illuminated by circular polarized light, the scattering intensity over polar angle is invariant in different azimuthal planes due to symmetry. However, the presence of non-spherical particles breaks this symmetry, and thus the scattering intensity in a particular azimuthal plane is dependent upon the particle’s orientation during the scattering event. One can imagine designing an instrument that measures this variation in intensity over azimuthal angle to determine whether or not a particle is spherical. Indeed, this approach was taken by (Ludlow, 1982) and further developed by (Kaye et al., 1991; Kaye et al., 1992). In Ludlow’s original 1982 system, three photodiode detectors were arranged symmetrically at a 40° scattering angle in a ring around the axis of an incident circularly polarized laser beam (z-axis), as in Fig. 2. The detectors captured light scattered by individual aerosol particles carried in a laminar airflow through the beam. The signals recorded by these azimuthal detectors were then used to differentiate between spherical particles, for which the signals were equal to within signal-to-noise limits, or nonspherical particles, where differences in the signals were observed. The purpose of this was to attempt to differentiate in real-time between spherical droplets (potentially agent droplets) and other solid airborne particle in the field.
Angularly Resolved Elastic Scattering from Airborne Particles
39
Figure 2. Triple azimuthal detector arrangement for assessing particle sphericity (Ludlow, 1982).
This system was suitable only for particles greater than ~5µm in size. The sensitivity was subsequently improved down to ~1µm particles by arranging for the scattering particle to lie at the focus of a parabolic reflector that captured the light over a large solid angle (~5sr) and directed it to three miniature photomultiplier detectors placed symmetrically about the laser beam. By 1987, this design had been further refined by replacement of the parabolic reflector with an ellipsoidal reflector that increased the solid angle of capture by a further 60%, taking a form as shown in Fig. 3.
Figure 3. Measurement of the variation in azimuthal scattering using three miniature photomultiplier tubes (Kaye, et al., 1991).
40
3.2
P.H. Kaye et al.
Detection of Intensity Variations Over Both Azimuthal and Polar Angles
Thus far, instrument designs have been limited to the detection of light in a single plane. By detecting light at a fixed azimuthal angle, one is essentially limited to analyzing spherical particles. For these particles, one can determine both the size and refractive index. Conversely, by detecting light at a fixed polar angle, it is possible to analyze the shape of an aerosol particle, in particular whether or not it is spherical. In this section, instruments that detect variation in intensity over both azimuthal and polar angles are described. 3.2.1
Detection at a limited number of discrete angles
The first device that could detect angularly resolved light elastically scattered from single particles in a flowing gas stream over different azimuthal angles φ as well as different polar angles θ was the DAWN-A (Wyatt Technology Corp.) developed by Wyatt et al. (1988). The geometry of the set-up is quite simple. Fiber bundles coupled to photodiodes are placed at sixteen scattering angles over the surface of a spherical chamber, shown in Fig. 4. The peak current from each PMT is then recorded every time a particle flows through the laser beam at the focal volume. The DAWN-A was used in a study to determine the fraction of spherical particles in ambient aerosols on the south-western edge of the Great Smoky Mountains National Park near Townsend, Tennessee (Dick et al., 1998; Sachweh et al., 1995). In this study, more than 90% of ambient particles in the 0.2 – 0.8 µm size range sampled from Jul. 15-Aug. 25, 1995 were determined to be spherical. 3.2.2
Detection of a high-resolution scattering profile
In the early nineties, Kaye et al. (1992) and Hirst et al. (1994) utilized Ludlow’s prolate ellipsoidal reflector design (Fig. 3) to collect light over a large solid angle spanning a polar angular range of 30° to 141° and an azimuthal angular range 0° to 360°. However, they replaced the three miniature PMT detectors with an intensified CCD camera (format size 385x288) to capture very high-resolution scattering profiles. By adding a second ICCD camera to the opposite end of the same instrument, (i.e., replacing PMT Detector E4 in Fig. 3), scattering profiles could be recorded simultaneously from an individual particle at both low (forward) and high angular ranges.
Angularly Resolved Elastic Scattering from Airborne Particles
41
Figure 4. Scattering chamber of the DAWN-A that detects light scattered over different azimuthal as well as polar angles. Adapted from (Wyatt et al., 1988).
Figures 5 and 6 show experimental scattering patterns from a variety of particle types recorded by this instrument at high and low (forward) scattering angle ranges, respectively. In the former case, the outer circumference of the image corresponds to light scattered to the edge of the ellipsoidal reflector (141° scattering angle ), and the innermost dark circle results from the aperture in the rear of the ellipsoidal reflector corresponding to a scattering angle of 28°. In the latter case, the scattering profile images cover a scattering angle range of approximately 4° to 27°. The dark circle at the centre of each image in this case is the shadow of a beam stop that prevents detector illumination by the unscattered laser beam.
Figure 5. High-angle spatial light-scattering patterns recorded using an ICCD camera in place of the triple PMT detectors in Fig. 3. Top row: 10 µm water droplet; ellipsoidal particle, copper flake (~5 µm); Bottom row: salt crystal, fiber tilted from vertical, doublet of 3 µm PSL spheres (Hirst et al.,1994).
42
P.H. Kaye et al.
Figure 6. Low-angle (forward) spatial light-scattering patterns recorded using an ICCD camera in place of the PMT E4 in Fig. 3. Top-row: water microdroplet, straight asbestos fiber (crocidolite), salt crystal (corner-on), 2 µm ellipsoidal particle. Bottom row: salt crystal (edge-on), irregular silica particle, curved asbestos fiber (chrysotile), water droplet with oleic acid droplet inclusion.
The high degree of variability in both the high-angle and low-angle scattering patterns illustrates the extent to which they contain information relating to particle morphology. However, a critical factor, especially in applications involving real-time monitoring of particles, is the efficiency with which such information may be extracted automatically and rapidly by machine. Various methods for determining particle characteristics from these scattering patterns and attempts to deduce particle morphology by inversion of the scattering data are given later in this chapter. Following Kaye’s advancements, Pan et al., (2003) designed a similar system with some notable improvements. In particular, Pan’s design utilized a tightly focused CW laser beam as a trigger beam to minimize the size of the scattering volume. When a particle traversed the trigger beam, the scattering event detected by a PMT would trigger a pulsed laser source to fire. These improvements resulted in an enhanced signal-to-noise ratio as well as minimizing the effects of aberrations on the scattering profiles. In the paper, they refer to the scattering profiles as LATAOS (large-anglerange two-dimensional angular optical scattering) patterns.
3.3
Miscellaneous Design Innovations
Since the realization of an instrument that could detect high angular resolution scattering profiles over a large angular range, other instruments that are variations on this design, have been developed. For instance, Kaye et al. (1997) produced a simplified forward-scattering instrument in which the spatial scattering profile data were captured using a 33-channel
Angularly Resolved Elastic Scattering from Airborne Particles
43
photodiode detector array rather than an ICCD camera. This allowed for much higher scattering data acquisition rates (>103 /s) that are required in real-time particle monitoring applications such as airborne asbestos fiber detection. The photodiode array consisted of two 16-segment concentric rings surrounding a single continuous annular ring. This was an optimal geometry determined by the authors through simulations to maximize particle discrimination efficiency. In later work (Kaye et al., 2000), a multi-element hybrid photodiode (HPD) detector of 3 segmented annular rings was used to measure spatial scattering at high particle throughput rates. The outer ring contained 24 elements; the middle had 6 elements, while the inner ring was continuous around a central beamstop. This detector was tested on a chamber similar to that of Fig. 3, replacing both the forward-scatter and backscatter PMTs. Figure 7 shows the outer 24 elements plotted to give polar diagram images of the forward scatter and backward scatter for spherical, fibrous and cubic particles. This spatial light scattering analysis was then integrated with intrinsic particle fluorescence to help classify biological particles within an ambient aerosol. This topic is discussed further in the next chapter.
Figure 7. Polar diagrams of high-angle scatter for spheres, fibers and cubes (top row), and low angle scatter for the same particle shapes (lower row), recorded using a multi-element hybrid photodiode.
To measure single particle absorption, Aptowicz et al. (2004) developed a system that could detect light scattering profiles simultaneously from two laser sources, as shown in Fig. 8. Using dichroic beamsplitters, two laser sources are combined pre-scattering and then separated post-scattering so that light scattering profiles are simultaneously captured from two laser sources. Note that the orientation of the aerosol particle is the same for
44
P.H. Kaye et al.
both scattering events allowing one to compare directly the two scattering profiles and decipher if a strong absorption band lies at either wavelength.
Figure 8. Experimental setup for collecting simultaneous LATAOS patterns at two wavelengths.
4.
OPTICAL SCATTERING PATTERNS FROM URBAN AEROSOL PARTICLES
In a recent study by Aptowicz et al. (2006), light scattering patterns from approximately 6000 ambient aerosol particles were collected at the U.S. Army Research Laboratory Harry Diamond Building in Adelphi, Maryland over the course of 18 hours in October 2004. The patterns suggest that background aerosol particles have diverse morphologies ranging from single spheres to complex structures. These patterns are presented in Fig. 9 as a visual introduction to light scattering patterns. Twenty scattering patterns of atmospheric aerosols captured sequentially are shown in Fig. 9. The polar angles span from 75° to 130° and the azimuthal angles span from 0° to 360°. There are several experimental artifacts embedded in the LATAOS patterns. Holes drilled in the ellipsoidal mirror for one of the trigger beams as well as the particle-laden airstream appear as off-centered ovals in the patterns. A mounting post used to hold a beam-steering mirror appears as a black bar on the bottom of the image. The intensity of the patterns is adjusted for ease of viewing. As is evident, there is a high level of particle-to-particle variability in the patterns. Some of the patterns are recognizable; for example, image 10 was likely generated from a sphere. Some patterns appear to arise from perturbed spheres (like image 14), as suggested by their similarity to
Angularly Resolved Elastic Scattering from Airborne Particles
45
sphere-type patterns. The patterns were visually sorted to give a sense of the variability of the patterns, as shown in Fig. 10. In addition, possible particle shapes are listed to the left, although these are speculative. Particle size was roughly determined from the integrated scattering intensity.
Figure 9. Twenty consecutive scattering patterns of ambient particles captured during October 2004.
46
P.H. Kaye et al.
Figure 10. A qualitative classification of scattering patterns that depicts the variability of atmospheric particle morphology.
Angularly Resolved Elastic Scattering from Airborne Particles
47
Given that elastic scattering patterns are dependent on particle size, internal refractive index distribution, and morphology, and given that there may not be a unique relation between scattering patterns and these particle parameters, what particle characteristics can be determined from scattering patterns? The technical challenges required to answer these questions are difficult. However, researchers have derived some simple metrics from light scattering patterns and have successfully used these metrics to discriminate between different particle types. In the following section, an overview of some of these different procedures is presented.
5.
DETERMINATION OF PARTICLE CHARACTERISTICS FROM SCATTERING DATA
5.1
Deduction of Asymmetry Factor Af
One of the simplest methods of reducing light scattering pattern data to a single metric relating to particle shape is the Asymmetry Factor Af (Kaye et al. 1991, 1996), a measure of the azimuthal variation of light scattered by a particle. Note that this is quite different from the asymmetry parameter g, defined as , commonly used in radiative transfer calculations. The original implementation of Af measurement was based on an instrument similar to that shown in Fig. 3 in which the three azimuthal detectors E1, E2, and E3 provided data relating to particle shape, whilst the sole fourth detector E4 provided simultaneous data relating to particle size. The principle of the derivation of Af is illustrated in Fig. 11. A spherical particle, such as a liquid droplet, scatters light equally to the three azimuthal detectors as indicated left in Fig. 11. In contrast, an elongated particle, such as a high-aspect ratio fiber, tends to be aligned parallel to the direction of the sample airflow (i.e.: vertical in this case) and therefore scatters light predominantly horizontally to just one detector, E2, as shown at right in Fig. 11. Other particle morphologies result in distributions of scattered light to the three detectors that are somewhere between these two extremes. Af is determined from
where E-bar is the mean of E1 + E2 + E3, n is the number of discrete detectors used, and k is a constant selected to ensure the maximum value of
48
P.H. Kaye et al.
Af is 100 (i.e., when one detector is a positive value and the other two are zero). Thus, for a spherical particle, Af is zero, whilst for a vertically orientated fiber, Af approaches 100.
Figure 11. Principle of the Asymmetry factor derivation using three azimuthal detectors.
Despite the simplicity of the detection methodology, the evaluation of Af can be effective in discriminating particles of differing morphology in realtime. Fig. 12 shows the data display from the B1010 Aerosol Shape Analysis instrument (Kaye et al., 1996) developed through the UK Defence Science and Technology Laboratory and employing a triple-detector arrangement as above. The left-hand triangular display represents each particle dataset as a single point whose position on the triangle denotes the centroid of the three detector outputs E1, E2, and E3. Thus spherical particles result in dots at the centre of the triangle and non-spherical particles result in dots away from the centre. The color of the dot represents the frequency of particles at that location. In terms of Asymmetry Factor Af, the centre of the Centroid plot corresponds to Af = 0, and each apex Af = 100 with a linear radially symmetric function between these two extremes. The aerosol being investigated in this example comprised three different particle types, corn starch flour, 2 µm ellipsoidal haematite particles (a model for bacterial cells), and 1 µm polystyrene latex spheres, as indicated in the photomicrographs at the foot of the Fig. 12. It can be seen that differentiation between these particle types cannot be achieved using the Centroid Plot alone. However, in the right-hand Contour Plot graph where values of particle Af are plotted against the signal measured by the fourth detector that corresponds roughly with particle size, clear differentiation between the particle types is achieved. The Af approach also was applied to subsequent instrument systems, referred to in section 3.3, that used larger numbers of discrete azimuthal detectors or multi-pixel detectors comprising segmented radial arrays to record the azimuthal variations in more detail than could be achieved in the triple-detector systems.
Angularly Resolved Elastic Scattering from Airborne Particles
49
Figure 12. Data display screen from a B1010 instrument illustrating discrimination of particle types within a mixed aerosol using Af and Size data (Kaye et al., 1996).
However, whilst the computation of Af data from individual particles may be achieved at high throughput rates, an advantage in real-time monitoring applications, there is information in the azimuthal variation of scattered light from a particle that the simple Af calculation is unable to elucidate. Because of this, more sophisticated data inversion and pattern classification methods have been exploited to maximize the utility of the azimuthal light-scattering data. These are discussed later in this chapter.
50
P.H. Kaye et al.
5.2
Deduction of Degree of Symmetry, Dsym
As an extension of Kaye’s Af as well as the Sphericity Index developed by Dick et al., (1998), a Degree of Symmetry (Dsym) factor that quantifies the degree to which a scattering pattern has rotational and mirror symmetries was introduced recently by Aptowicz et al. (2006). The patterns analyzed with this rubric were those of ambient aerosols depicted in Section 4. Before determining Dsym, a mask is applied to each image to remove unwanted regions where experimental artifacts appear. The masked example pattern, shown in Fig. 13A, is labeled matrix A. The rotational and mirror symmetries of this matrix are assessed using the following expression:
Dsym
=
1 3 ∑ 3 i =1
1 −
∑ ( A − Bi ) 2 pixel subset 2 × RMS (A )
1/ 2
where B1 is 180° rotation of matrix A (Fig. 13B), and B2 and B3 are mirror images of matrix A (Fig. 13C and 13D). Note the mirror symmetry planes were chosen to be parallel and orthogonal to the laser polarization. The root-mean-square (RMS) of matrix A is a normalization factor and the factor of 2 accounts for double counting. With this definition, the pattern of a spherical particle, which has a high degree of symmetry, results in a Dsym value of 1 if there are no sources of optical aberrations or noise in the system. The Dsym calculated from the LATAOS patterns first displayed in Fig. 10 are shown in Fig. 14. As expected, patterns 3 and 10 have high Dsym values. Note that pattern 5 also has a high Dsym value, although this is due to the intensity saturation of the picture. The pattern with the lowest Dsym value is number 16 where the upper right corner of the pattern is particularly bright. This asymmetric feature leads to the low Dsym value. Note also that although pattern 14 has rings it does not have a high Dsym value, since the rings are not symmetric.
Angularly Resolved Elastic Scattering from Airborne Particles
51
Figure 13. Pattern manipulations to determine Dsym value.
Dsym was calculated for the entire data set of ambient particles (5993 patterns). In addition, data sets were also collected of single PSL spheres (39 patterns), single or small clusters of Bacillus subtilis spores (97 patterns), droplets of dioctyl phthalate (200 patterns), as well as particulate matter in diesel exhaust (288 patterns). For all the patterns, the mean number of photoelectron events per pixel (MPE) was determined for each scattering pattern. This should roughly correlate to the aerosol size, since the scattering cross-section scales approximately as the aerosol crosssectional area. A plot of Dsym versus MPE is shown in Fig. 15 for the different data sets. Apparent in the plot is a ceiling of the Dsym value of 0.9. This ceiling is an experimental artifact and is due to slight distortions in the mirror surface. The Dsym of the Polystrene Latex (PSL) spheres (radius ~ 0.5 µm) range from 0.85 to 0.87 with one exception. Dioctyl phthalate are droplets of various diameters but the same refractive index of 1.485. As expected the Dsym values are all above 0.75, ignoring the low intensity patterns, which are most likely dust. The particulate matter in diesel exhaust appears to show slight symmetry with a majority of the Dsym values lying above 0.70 with a maximum value of 0.85. The Dsym values of BG spores and clusters of BG spores range from 0.15 to 0.83, which follows from the spores’ diverse morphologies.
52
P.H. Kaye et al.
Figure 14. Dsym and corresponding light-scattering patterns.
Angularly Resolved Elastic Scattering from Airborne Particles
53
Figure 15. Degree of Symmetry ( Dsym ) versus the mean number of photoelectron events per pixel for various aerosols.
Finally, the LATAOS patterns of ambient aerosols are distributed over Dsym values ranging from 0.4 to 0.88. A bulk of the values are centered around a Dsym value of 0.60, however there appears to be another clustering of LATAOS patterns in the 0.8 to 0.9 range, which most likely represents spherical particles. To see this clustering, a histogram of the LATAOS patterns that fall in different Dsym ranges is displayed in Fig. 16. This result suggests that there is a subpopulation of highly spherical aerosol particles in the ambient atmosphere. These observations are consistent with the commonly accepted notion that most micron-sized particles (in the accumulation mode) appear to be nearly spherical, and are probably formed in the atmosphere through gas-particle reactions; whereas, most supermicron particles appear to be non-spherical and are likely directly injected into the atmosphere.
54
P.H. Kaye et al.
Figure 16. Histogram of the fraction of particles in different D sym ranges for different particle types.
5.3
Data inversion – Non-Spherical Particles
Whilst the properties (size, refractive index) of spherical particles may be determined with some certainty by comparison with theoretical predictions, carrying out this exercise for non-spherical particles becomes far more difficult and in many cases intractable. Hirst et al. (1994) recorded experimental scattering images of the type shown in Fig. 5 and subsequently determined particle parameters by achieving a best-fit to theoretical patterns generated using a simple Rayleigh-Gans approach. Although the matches between experiment and theory were surprisingly good (Fig. 17) given the limitations of the model, the process was extremely time-consuming and was limited to particle geometries of regular shape (fibers, cubes, cuboids, etc.).
Angularly Resolved Elastic Scattering from Airborne Particles
55
Figure 17. Matching experimental scattering profiles to ‘best-fit’ theoretical models. (Hirst et al., 1994).
6.
PATTERN MATCHING /PROBABILISTIC TECHNIQUES
An alternative to data inversion techniques and one that is applicable where known particle types are being sought in a monitoring or detection scenario is through the use of pattern matching or probabilistic techniques. These are characterized by the attempted matching of the unknown experimental scattering data to that from known particles recorded in the same way. Datasets from known particle types are recorded experimentally and are used to train the matching algorithm. Unknown particles are then attributed to one of the known particle classes based on the closest match between the particle’s scattering data and the training set data. For example, Kaye et al. (1997) recorded low-angle ( = 4° to 30°) azimuthal light scattering data from four particle types (droplets, cubes, flakes, and fibers) using a 32-pixel radial photodiode array. These data were used as training data for a variety of standard classification algorithms: Linear Discriminant, K-Nearest Neighbours, Fuzzy-K Nearest Neighbours, Normal Distribution, and Radial Basis Function (RBF) neural network. Each trained algorithm was then challenged with scattering data from
56
P.H. Kaye et al.
similar particle classes to establish the efficiency with which correct classification occurred. Data were presented in the form of square probability maps (Fig. 18) in which each corner represented the ideal classification for one of the particle classes. Not surprisingly, droplets (spheres) usually were classified accurately whilst the cubes, flakes, and fibers, whose populations all varied in size and morphology, exhibited more dispersion from their ideal corner.
Figure 18. Radial Basis Function probability maps for four aerosol types. Ideally, cubes should be classified into top-left corner, flakes into the top-right, droplets into the bottomleft, and fibers into the bottom-right.
7.
DUAL WAVELENGTH SCATTERING FOR MEASURING SINGLE-PARTICLE ABSORPTION
Infrared (IR) absorption spectra can be thought of as a fingerprint for chemical agents. Thus, by measuring the IR spectra from aerosol particles, one can classify different types of aerosols and discriminate between those that pose a health threat and those that are harmless constituents of the ambient atmosphere. However, measuring absorption at the single particle level is a trying task. Typically, one measures the extinction of an incident beam to determine the absorption in aerosols. However, the scattering
Angularly Resolved Elastic Scattering from Airborne Particles
57
cross-section can be equal to if not larger than the absorption cross-section, thus it is difficult to differentiate between the two when making this measurement. Of course, the scattering and absorption from a single particle are not mutually exclusive events, but are linked. In particular, absorption affects the angularly resolved scattering pattern through the imaginary component of the complex refractive index. Therefore, experiments were conducted to explore whether single particle absorption information could be extracted from scattering patterns (Aptowicz et al., 2004) The scheme to collect scattering patterns at two wavelengths simultaneously was shown in Fig. 8. In the actual experiment, the two laser sources were optically pumped GaSb-based semiconductor lasers with typeII InAs/InGaSb quantum well gain regions emitting at 3.9 µm and 5.1 µm with a peak power of ~0.4 watts and pulse duration of 100 µs. Droplets composed of a mixture of H2O and D2O were utilized to demonstrate the technique. At both wavelengths H2O is relatively transparent; whereas D2O has relatively high absorption at 3.9 µm while being transparent at 5.1 µm. For D2O, the imaginary component of the refractive index (κ) is 0.260 at 3.9 µm in contrast to 0.002 at 5.1 µm. Beam-shaping optics (spherical and cylindrical lenses) as well as an F/3.75 CaF2 focusing lens were utilized to achieve a desired spot size of 50 µm x 500 µm where the major axis is perpendicular to the propagation direction of the droplet. The polarizations of both laser beams were perpendicular to the propagation direction of the droplets. An ellipsoidal mirror collected the backward hemisphere of scattered light (0° ≤ φ ≤ 360°, 90° ≤ θ ≤ 163°) and focused it through a spatial filter located at the ellipsoid’s second focal point. To reduce aberration effects, an F/1 ZnSe aspheric lens collimated the scattered light, after which the two wavelengths were separated via a dichroic mirror. Finally, a bi-convex F/1 CaF2 lens coupled the light onto the InSb detectors. Droplets from three different mixtures of H2O-D2O were analyzed: 100%-0%, 75%-25%, and 50%-50%. The experimentally collected data, as well as numerical simulations based on Lorenz-Mie theory, are shown in Fig. 19. There are several experimental artifacts embedded in the TAOS patterns that include the following: (1) the shadow of the droplet generator nozzle protruding into the ellipsoidal mirror; (2) the shadow of the beam block mount; and (3) the diffraction of the illuminating laser beam around the beam block. Note that the patterns were processed to be linear in θ by taking into account the optical arrangement of the system.
58
P.H. Kaye et al.
Figure 19. (Row I) TAOS pattern at λ = 3.9 µm detected from a single droplet (diameter ~ 55 µm) composed of (a) H2O, (b) 75%-25% H2O-D2O, and (c) 50%-50% H2O-D2O. (Row II) Corresponding numerical simulations of row I based upon Lorenz-Mie theory. (Row III) TAOS pattern collected simultaneously at 5.1 µm from the same single droplets as Row I. (Row IV) Corresponding numerical simulations of row III based upon Lorenz-Mie theory. (Aptowicz et al., 2004).
The TAOS patterns shown in Fig. 19 were collected from single droplets illuminated simultaneously by the two collinear 3.9 µm and 5.1 µm laser beams. Each column from left to right indicates a different H2O-D2O droplet composition, as labeled. TAOS patterns (row I) collected at 3.9 µm show the effects of increasing the concentration of D2O that leads to an increase in absorption. These patterns qualitatively match the numerical simulations based on Lorenz-Mie theory (row II), although there appears to be some discrepancy for the 75%-25% case (comparison between (b) and (e)) believed to be due to oversimplification in estimating the absorption by taking a linear interpolation of known absorption values for neat fluids and ignoring effects of isotopes (Max and Chapados, 2002). TAOS patterns (row III) are collected simultaneously at 5.1 µm from the same single droplets. These patterns also agree with predictions from Lorenz-Mie theory in which there was a slight increase in scattering intensity of the
Angularly Resolved Elastic Scattering from Airborne Particles
59
central region (θ > 135°) with increasing concentrations of D2O. For the numerical simulations, the droplet diameter was estimated to be 55 µm and the refractive index was gathered from the literature. (Bertie et al., 1989) By comparing the data in row I with row III, it is clear that one could distinguish between droplets of pure H2O and droplets that contain a considerable amount of D2O by comparing the angular scattering patterns.
REFERENCES Aptowicz, K.B., Pan, Y.L., Chang, R.K., Pinnick, R.G., Hill, S.C., Tober, R.L., Goyal, A., Leys, T., and Bronk, B.V., 2004, Two-dimensional angular optical scattering patterns of microdroplets in the mid infrared with strong and weak absorption, Opt. Lett. 29(17):1965-1967. Aptowicz, K.B., Pinnick, R.G., Hill, S.C., Pan, Y.L., and Chang, R.K., 2006, Optical scattering patterns from single urban aerosol particles at Adelphi, Maryland, USA: A classification relating to particle morphologies, J. Geophys. Res., 111, D12212. Asano, S., and Yamamoto, G., Light scattering by a spheroidal particle, 1975, Appl. Optics 14(1): 29-49. Barber, P.W., Hill, S.C., 1990, Light Scattering by Particles: Computational Methods, World Scientific Publishing. Bartholdi, M., Salzman, G.C., Hiebert, R.D., and Kerker, M., 1980, Differential lightscattering photometer for rapid analysis of single particles in flow, Appl. Optics, 19(10):1573-1581. Bertie, J.E., Ahmed, M.K., and Eysel, H.H., 1989, Infrared intensities of liquids, 5. Optical and dielectric-constants, integrated-intensities, and dipole-moment derivatives of H2O and D2O at 22-degrees-c, J. Phys. Chem-Us., 93(6):2210-2218. Bohren, C.F., and Huffman, D.R., 1983, Absorption and scattering of light by small particles, Wiley, New York. Dick, W.D., Ziemann, P.J., Huang, P.F., and McMurry, P.H., 1998, Optical shape fraction measurements of submicrometre laboratory and atmospheric aerosols, Meas. Sci. Technol., 9(2):183-196. Draine, B.T. and Flatau, P.J., 1994, Discrete-dipole approximation for scattering calculations. J. Opt. Soc. Am. A 11:1491-1499. Gucker, F.T., and Egan, J.J., 1961, Measurement of angular variation of light scattered from single aerosol droplets, J. Coll. Sci. Imp U Tok, 16(1):68-&. Gucker, F.T., and Rowell, R.L., 1960, The angular variation of light scattered by single dioctyl phthalate aerosol droplets, Discuss. Faraday Soc., 30(185). Gucker, F.T., Okonski, C.T., Pickard, H.B., and Pitts, J.N., 1947a, A photoelectronic counter for colloidal particles, J. Am. Chem. Soc., 69(10):2422-2431. Gucker, F.T., Pickard, H.B., and Okonski, C.T., 1947b, A photoelectric instrument for comparing the concentrations of very dilute aerosols, and measuring low light intensities, J. Am. Chem. Soc., 69(2):429-438. Gucker, F.T., Tuma, J., Lin, H.M., Huang, C.M., Ems, S.C., and Marshall, T.R., 1973, Rapid measurement of light-scattering diagrams from single aerosol particles in an aerosol stream and determination of the latex particle size, Aerosol Science, 4:389-404.
60
P.H. Kaye et al.
Harrington, R.F., 1968, Field Computation by Moment Methods, Krieger Publishing Co., Inc. Hirst, E., and Kaye P.H., 1996, Experimental and theoretical light scattering profiles from spherical and non-spherical particles, J. Geophys. Res-Atmos. 101 (D14): 19231-19235. Hirst, E., Kaye, P.H., Guppy, J.R., 1994, Light scattering from non-spherical airborne particles: experimental and theoretical comparisons, Appl. Optics 33 (30):7180-7186. Holler, S., Auger, J.C., Stout, B., Pan, Y., Bottiger, J.R., Chang, R.K., and Videen, G., 2000, Observations and calculations of light scattering from clusters of spheres, Appl. Optics, 39(36):6873-6887. Kaye, P.H., 1998, Spatial light-scattering analysis as a means of characterizing and classifying non-spherical particles, Meas. Sci. Technol., 9:141-149. Kaye, P.H., Alexander-Buckley, K., Hirst, E., and Saunders S., 1996, A real-time monitoring system for airborne particle shape and size analysis, J. Geophys. Res-Atmos. 101 (D14): 19215-19221. Kaye, P.H., Barton, J.E., and Hirst, E., 2000, Simultaneous light scattering and intrinsic fluorescence measurement for the classification of airborne particles, Appl. Optics, 39:3738-3745. Kaye, P.H., Eyles, N.A., Ludlow, I.K., and Clark, J.M., 1991, An instrument for the classification of airborne particles on the basis of size, shape, and count frequency, Atmos. Environ. A-Gen, 25(3-4): 645-654. Kaye, P.H., Hirst, E., Clark, J.M., and Micheli, F., 1992, Airborne particle-shape and size classification from spatial light-scattering profiles, J. Aerosol. Sci., 23(6):597-611. Kaye, P., Hirst, E., and Wang-Thomas, Z., 1997, Neural-network-based spatial lightscattering instrument for hazardous airborne fiber detection, Appl. Optics, 36(24):61496156. Kaye, P.H., Eyles, N.A., Ludlow, I.K. and Clark, J.M., 1991, An instrument for the classification of airborne particles on the basis of size, shape and count frequency. Atmospheric Environment 25A, 3/4, 645-654. Kerker, M., 1997, Light scattering instrumentation for aerosol studies: An historical overview, Aerosol. Sci. Tech., 27:522-540. Lakhtakia, A, and Mulholland, G.W., 1993, On two numerical techniques for light scattering by dielectric agglomerated structures, J. Res. Nat. Inst. Stand. Technol., 98:699-716. Lakhtakia, A., 1992, General theory of the Purcell-Pennypacker scattering approach and its extension to bianisotropic scatterers, Astrophys. J., 394:494-499. Lorenz, L., 1890, Lysbevaegelsen I og uden for en haf plane lysbolger belyst kluge, Vidensk. Selsk. Skr. T. 6, 1-62. Ludlow, I.K., 1982, Particle analysis by light scattering, UK Ministry of Defense contract report, ER1A/9/4/2161/CDE. Marshall, T.R., Parmenter, C.S., and Seaver, M., 1976, Characterization of polymer latex aerosols by rapid measurement of 360 degree light-scattering patterns from individual particles, J Colloid Interf Sci, 55(3):624-636. Max, J.J., and Chapados, C., 2002, Isotope effects in liquid water by infrared spectroscopy, J. Chem. Phys., 116(11):4626-4642. Mie, G., 1908, Considerations on the optics of turbid media, especially colloidal metal sols, Ann. d. Physik, 25:377-442. Mishchenko, M.I., and Travis, L.D., 1998, Capabilities and limitations of a current FORTRAN implementation of the T-matrix method for randomly oriented rotationally symmetric scatterers, J. Quant. Spectrosc. Radiat. Transfer 60:309-324.
Angularly Resolved Elastic Scattering from Airborne Particles
61
Mishchenko, M., Hovenier, J.W. and Travis, L.D., 2000, Light scattering by nonspherical particles. Academic Press Inc. Pan, Y.L., Aptowicz, K.B., Chang, R.K., Hart, M., and Eversole, J.D., 2003, Characterizing and monitoring respiratory aerosols by light scattering, Opt. Lett., 28(8):589-591. Phillips, D.T., and Wyatt, P.J., 1972, Single-particle light-scattering measurement photochemical aerosols and atmospheric particulates, Appl. Optics, 11(9):2082-&. Phillips, D.T., Wyatt, P.J., and Berkman, R.M., 1970, Measurement of Lorenz-mie scattering of a single particle - polystyrene latex, J Colloid Interf Sci, 34(1):159-&. Purcell, E.M., and Pennypacker, C.R., 1973, Scattering and absorption of light by nonspherical dielectric grains, Astrophys. J. 186:705-714. Sachweh, B.A., Dick, W.D., and McMurray, P.H., 1995, Distinguishing between spherical and nonspherical particles by measuring the variability in azimuthal light-scattering Aerosol Sci. Tech., 23:373-391. Taflove, A., Ed., 1988, Advances in computational electrodynamics: The finite-difference time-domain method, Artech House, Boston. Waterman, P.C., 1971, Symmetry, unitarity, and geometry in electromagnetic scattering. Phys. Rev. D, 3:825-839. Wyatt, P.J., 1972, Light-scattering in microbial world, J. Colloid Interf. Sci., 39(3):479-&. Wyatt, P.J., Schehrer, K.L., Phillips, S.D., Jackson, C., Chang, Y.J., Parker, R.G., Phillips, D.T., and Bottiger, J.R., 1988, Aerosol-particle analyzer, Appl. Optics, 27(2):217-221. Yee, S.K., 1966, Numerical solution of initial boundary value problems involving Maxwell’s equations in isotropic media. IEEE Trans. Antennas Propagat. 14:302-307.
Paul Kaye, Richard Chang, Gorden Videen, Virginia Foot, and Kevin Aptowicz
BIO-AEROSOL FLUORESCENCE Detecting and characterising bio-aerosols via UV light-induced fluorescence spectroscopy Yong-Le Pan,1 Jay D. Eversole,2 Paul H. Kaye,3 Virginia Foot,4 Ronald G. Pinnick,5 Steven C. Hill,5 Michael W. Mayo,6 Jerold R. Bottiger,7 Alan Huston,2 Vasanthi Sivaprakasam,2 Richard K. Chang1 1
Department of Applied Physics and Center for Laser Diagnostics, Yale University New Haven, Connecticut 06520, USA 2 US Naval Research Laboratory, Code 5611, 4555 Overlook Ave SW Washington, DC 20375, USA 3 Science & Technology Research Institute, University of Hertfordshire Hatfield, Herts. AL10 9AB, U.K. 4 Defense Science & Technology Laboratory, Porton Down, Salisbury, Wilts. SP4 0JQ, U.K 5 U. S. Army Research Laboratory, Adelphi, Maryland 20783-1197, USA 6 Nanohmics Inc., 6201 East Oltorf Street, Austin, TX 78741, USA 7 U. S. Army Edgewood Chemical, Biological Center, Aberdeen Proving Ground, Maryland 21010, USA Abstract:
All aspects of fluorescence from a single or a cluster of bio-aerosol particles are treated in this chapter. The following topics are covered: light sources to induce fluorescence, detectors to sense fluorescence, excitation and emission spectra from simulants of known threat bio-aerosols, standards for monitoring aerosols, performance metrics, generation of dry and wet simulant bioaerosols, examples of commercially available bio-sensors, techniques for enriching aerosol samples by aerodynamic puffing, and the need for an identifier following the enrichment of aerosols sorted according to the aerosols’ fluorescence spectra. There is general agreement that fluorescence is a discriminator capable of separating bio- from non-bio-aerosols; however, an appropriate particle identifier has yet to be fully adapted to the monitoring of threat aerosols in the highly variable atmospheric environment.
Key words:
Fluorescence, Bio-aerosol, Detection, Classification.
63 A. Hoekstra et al. (eds.), Optics of Biological Particles, 63–164. © 2007 Springer.
64
1.
Y.-L. Pan et al.
INTRODUCTION
Fluorescence is an inelastic optical process whose intensity signal, for particles containing organic carbon, is about 3-6 orders of magnitude less than elastic scattering. Among all the inelastic processes, fluorescence generally has the strongest signals. The fluorescence process involves three steps: 1) transition from the vibronic states of the ground electronic manifold to some vibronic states in an upper electronic manifold caused by the absorption of an incident photon; 2) a non-radiative transition among vibronic states within the upper electronic manifold; and 3) lastly, a photon emission from the radiative transition of the lowest vibronic states in the upper electronic manifold to some vibronic states in the ground electronic manifold. Biological molecules are typically very large, and consequently the vibronic states and the electronic manifolds are dense and result in broad, diffuse emission bands. Furthermore, the many similar biomolecules are ubiquitous among living organisms. Consequently, the fluorescence signal can only be used as a classifier, and not as a specieslevel identifier. For example, among the 20 amino acids, only 3 amino acids (tryptophan, phenyalanine, and tyrosine) can give rise to laser-induced fluorescence (LIF). Among the three amino acids, tryptophan is the only one that emits (at 350 nm) without being absorbed by the other two. The excitation reaches a secondary maximum at 280 nm, but continues to increase for wavelengths shorter than 240 nm. By illumination around the 280 nm region, one also can observe fluorescence in a broad band centered roughly at 450 nm. In many cases a likely candidate source for this emission is the reduced form of nicotinamide adenine dinucleotide (NADH) a co-enzyme found in many living cells, and indicative of metabolic activity of bacteria. Independent measurements on purified samples of NADH show a similar emission spectral shape. Although other fluorophore biomolecules may either contribute to, or in some cases be solely responsible for, the actual observed emission in this spectral range, any fluorescence in this band has come to be referred to as the “NADH peak” as a kind of spectroscopic short-hand. By shifting the excitation wavelength from 266 nm to 350 nm, the NADH peak grows while the tryptophan peak diminishes. Hence we can perceive the power of dual wavelengths excitation and the advantage of 2-band monitoring. From analyzing the difference and ratio of the tryptophan and NADH fluorescence peaks, one can decide that some molecules in the particle are biological. Currently the most advanced fluorescence-based aerosol diagnostic techniques use a) dualwavelength excitation for the tryptophan and NADH absorption bands and
Bio-aerosol Fluorescence
65
two-wavelenth (UV and visible) detection bands, and b) single UVwavelength excitation near the tryptophan absorption band and the entire UV-visible detection band using 32-channel multi-anode photomultiplier tube (PMT) technology. There is general agreement in the bio-aerosol detection community that fluorescence can be used as a classifier of bio- and non-bio aerosols. Using current techniques, with fluorescence spectra as a cue, aerosols can be interrogated at rates up to 90,000 per second. The suspect bio-aerosols can be sorted (aerodynamically deflected) from the dominant population of innocuous background particles at rates up to 1200 per second. The bioenriched aerosols can be deposited onto the surface of a substrate or into a liquid in a microfluidic reservoir for further analyses. The bio-enriched particles on a substrate can be identified by optical techniques such as FTIR or Raman scattering. Bio-enriched particles in the microfluidic reservoir can be identified by specific binding between antigens and fluorescent-tagged antibodies, or possibly by DNA/RNA sequencing. The minimum detectable background of threat bio-warfare (BW) agent particles depends on the highly variable and site-dependent atmospheric background of organic carbon aerosols that act as interferents to detection. These interferent particles may be polycyclic aromatic hydrocarbons, humic and fulvic acids, humic-like substances, or bacteria that have fluorescence emission similar to BW agents, and can cause false alarms. This chapter begins with the quantum mechanical description of fluorescence, spontaneous Raman scattering, and resonance Raman scattering (Section 2). In Section 3 Steve Hill and Yong-Le Pan present an overview of the fluorescence of biological molecules. Then Steve Hill and Michael Mayo present their unpublished work on the excitation and emission spectra (EEM) of biological spores and cells. This EEM work is a roadmap to help the reader best select the excitation wavelengths of tryptophan and NADH in order to get the optimum signal-to-noise ratio for bio-aerosol detection. Although Hill and Mayo’s work, performed some ten years ago, has never been published, their EEM plots are frequently found pinned on the laboratory walls of bio-aerosol research labs. In Section 4, Jay Eversole discusses the use of confusion matrices and the receiver operating characteristic or ROC curve. In Section 5, Yong-Le Pan gives an illuminating presentation on the types of light sources that could be used for exciting the intrinsic fluorescence of bio-aerosols starting with traditional UV lamps to LEDs to the harmonics of YAG lasers and ultimately to UV laser diodes. He includes a description of the ingenious usage of 2 xenon lamps at different UV wavelengths (Paul Kaye et al.). In Section 6, Yong-Le Pan discusses the latest advances in detector technology, starting out with CCD and
66
Y.-L. Pan et al.
ICCD, then APD linear array, and ending with a 32-anode photomultiplier (PMT). Before any bio-aerosol sensor enters into a field trial, one needs to create aerosols for testing in the laboratory. These test aerosols are simulants for spores, bacteria, toxins, and viruses. Test aerosols are generated by a variety of techniques. In Section 7, Ron Pinnick, Jerold Bottiger, and Yong-Le Pan give a short review of some commercial nebulizers and spray atomizers. To produce dry aggregates, the Ink Jet Aerosol Generator (IJAG from U.S. Army Edgewood Chemical, Biological Center) developed by Bottiger et al. has been widely used. There are several bio-sensor instruments based on UV-LIF and elastic scattering. In Section 8, Paul Kaye, Virginia Foot, and Jay Eversole provide a detailed review of four current instruments that are considered to be the most reliable because they give the lowest number of false alarms. These three co-authors describe these instruments firsthand. The instruments were designed and developed by researchers at the U.S. Naval Research Lab and the University of Hertfordshire and the U.K. Defense Science and Technology Laboratory. In Section 9, Ron Pinnick succinctly summarizes some ambient aerosol fluorescence measurements. These measurements were made at Adelphi, Maryland, using a Fluorescence Particle Spectrometer (FPS), developed by the Yale/ARL team. A smart and fast spectral data analysis and comparison technique (hierarchical cluster analysis by Steve Hill) was used to analyze the vast amount (15,000 spectra) of data that was provided by the FPS. In Section 10, Yong-Le Pan provides a different facet for spectral data analysis and comparison technique: spectral comparison in quasi real-time. This allows various instruments to classify bio-aerosol particles based on fluorescence and elastic scattering. Together, the fluorescence spectra and the elastic scattering act as a cue for particle sorting by a puff of air from a puffer. This technique is used to deflect particles with a pre-chosen spectral line-shape. In Section 11, Yong-Le Pan discusses high discrimination methodology for bio-aerosols. After fluorescence detection, there are several techniques that are potentially able to identify a bio-aerosol. They are all optical or biochemical techniques. The output from the fluorescence-based bio-sensor can be a first stage in an identifier system for bio-aerosols. The future for bio-sensors that combine fluorescence combined with elastic scattering is truly exciting. We are at the threshold of developing something phenomenal. Currently, the objectives are not only to decrease the false alarm rate, but to identify the one threat aerosol in the ambient atmospheric background. This can be best accomplished with continuous
Bio-aerosol Fluorescence
67
monitoring and providing data in quasi-real time. With the highest quantum efficiency, elastic scattering and fluorescence provide the first choice for the instrumentation of bio-aerosol detection and characterization. Besides satisfying the detection requirements, the instrument should be inexpensive, maintenance free, sensitive to one Agent Containing Particles per Liter of Air (ACPLA), low power consumptive, and producing low false alarm rates. That will be a real challenge! Prof. Richard K. Chang and Dr. Ronald G. Pinnick
2.
A BRIEF LOOK AT FLUORESCENCE SPECTROSCOPY
2.1 Fluorescence and Resonance Raman Effect When an aerosol particle is illuminated by a light source, most of the photons that are incident on the particle are elastically scattered and a small fraction of the photons are inelastically scattered. The elastically scattered photons have the same energy (frequency) and, therefore, the same wavelength as the incident photons. Inelastic scattering includes fluorescence, phosphorescence, and Raman scattering (see Figure 1). The inelastically scattered photons from the fluorescence process result from a series of stepwise transitions. The first step is the absorption of the incident photon that causes a transition from the ground electronic manifold to the upper electronic manifold. The second step is a non-radiative decay from the upper rotational vibrational states to the lower rotational vibrational states in the same electronic manifold. The third and final step is a radiative transition down to the ground electronic manifold. Therefore, the wavelength re-emitted is red-shifted with respect to the elastic scattering photons. Among the elastic-inelastic processes, fluorescence generally has the largest signal and a lifetime on the order of ns, and the photo-induced transitions occur between real vibro-rotational energy levels of different electronic manifolds. Therefore, the frequency and lineshape of the fluorescence spectrum is determined by the vibro-rotational energy levels of the involved upper and ground electronic manifold of the specific structure of the scattering molecule. In most cases there exists a big frequency shift between the absorption and fluorescence bands. For most molecules, the lineshape of the fluorescence band is a mirror image of the lineshape of the absorption band. The energy shift is related to the details
68
Y.-L. Pan et al.
of the electronic band structure. For simple diatomic molecules, this energy shift is called the Franck-Condon shift. While phosphorescence involves forbidden transitions between electronic states with different spin multiplicity (e.g. T1 and S0), the intensity of these transitions is several orders of magnitude weaker than that of fluorescence.
Figure 1. Potential energy diagram of the inelastic scattering process. Photons incident on the molecule are absorbed, then the energy is released by radiatively emitting fluorescence, phosphorescence, Raman scattering (solid arrows), and non-radiative transitions, such as intersystem crossing and internal conversion (wave arrows) between the ground state (S0), excited singlet (S1) and triplet states (T1), and even virtual states for Raman scattering.
The Raman scattering process involves virtual transitions from rotational and vibrational states of the ground electronic manifold to the rotational and vibrational states in the upper electronic manifold, Eupper. The energy difference between the Eupper and the incident photon energy (Eupper - Eincident ) can be considered as the inverse time the electron lives in the upper state, in accordance with Heisenberg Uncertainty Principle. During this short time the electron must interact with the rotational and vibrational modes. The following example will give the reader a sense of why the Raman effect is so weak and why the enhancement provided by the resonance Raman effect is important. Assume that the energy difference (Eupper - Eincident ) is 1 eV and thus lives in the excited state for about 10-15s. During this short lifetime in the excited electronic manifold, the probability of interacting with a rotational and vibrational mode is very unlikely and hence one gets the weak Raman scattered signal (see the curve on the right in Fig. 1). When the incident photon energy is tuned to
Bio-aerosol Fluorescence
69
the absorption linewidth of the upper manifold, the lifetime is increased, and hence one gets an enhancement of the resonance Raman effect (RRE). The distinction between the RRE and fluorescence can be subtle and is an interesting topic all by itself. Suffice to say that the fluorescence process has a lifetime decay factor while the RRE remains an instantaneous process with no de-phasing effects during the transitions to and from the upper rotational and vibrational states in the excited electronic manifold.
2.2 Energy Levels and Fluorescence Spectrum For a molecular system, we begin with the Schrödinger equation (Atkins and Friedman, 2005; Demtroder, 1982),
i¥
∂Ψ = HΨ ∂t
(1)
where, is the wavefunction, and H is the Hamiltonian operator for the molecule system. As we cannot solve this equation analytically, several approximations are generally taken into consideration. First, when the potential is time independent, the total wavefunction can be separated into time-dependent and space-variation parts. In addition, the nuclear motion is much slower than the electronic motion, and the total wavefunction is a product of the electronic, vibrational, and rotational wavefuctions. This is called the Born-Oppenheimer approximation:
f Ψ( r , t ) = ψ el ( r− el Rn )ψ vib ( Rn )ψ rot (θ , φ )ξ (t )
(2)
where, Rn is the nuclear coordinate, and r-el is the electronic coordinate. Based on these, the eigenfunctions and eigen energy levels (as shown in Figure 1) can be solved analytically or numerically according to the complexity of the molecule system. Therefore, the matrix element for a transition between two energy levels i and k of different electronic states can be expressed as
' " ' " Rik = ∫ ∫ψ e' * M −eψ e " dτ el ψ vib *ψ vib dτ vib ∫ ψ rot *ψ rot dτ rot θ φ R r , ≈ M −el ( Re ) Fvib H rot
(3)
where, M − el ( Re ), Fvib , and H rot are the electronic transition momentum, Franck-Condon factor, and Hönl-London factor, which mainly represent the transition strength, overlap integral of the vibrational wave functions, and
70
Y.-L. Pan et al.
transition selection rules, respectively. Therefore, the line strength S(i, k) of (nk, vk, Jk) from energy level i to level k by the a transition (ni, vi, Ji) notation of electronic, vibrational, and rotational quantum numbers n, v, J can be written as (Atkins and Friedman, 2005; Demtroder, 1982), 2
2
S (i, k ) = M − el (i, k ) Fvib ( vi , ni , v k , nk ) H rot ( J a , J b )
2
(4)
Thus, the intensity I( ) of fluorescence spectrum at a certain frequency is the sum of all the possible transitions from all energy levels i to k between two electronic states with all the photon emissions at the same frequency and population Ni at energy level i by excitation,
I (γ ) = ∑ N i S (i, k )
(5)
i,k
and the whole fluorescence spectrum is formed with different intensity I( ) at different frequency . Generally, for large molecules such as biological molecules, the allowed transitions between levels are so dense that it is difficult to observe the discrete fluorescence spectral characteristics of small molecules; therefore, for bio-aerosol detection and characterization little consideration is given towards high-resolution fluorescence excitation and emission. The focus is mainly on the fluorescence spectral profile, peak position, intensity, and quantum efficiency.
2.3 Fluorescence Quantum Yields and Cross Section In preparing for bio-aerosol detection during field trials, knowledge of the fluorescence quantum yields and absorption cross-sections are critical. This is particularly important for those materials that have almost identical fluorescence lineshapes but different fluorescence quantum efficiencies. The light loss or extinction in the forward direction (Iloss) is the result of the scattering cross section scatt and absorption cross-section absorp. When the particle interacts with light of flux I0 (photons/sec cm2)
I loss = I 0 (σ scatt + σ absorp )
(6)
The absorbed photons I0 absorp are re-emitted via fluorescence Iflu and other nonradiative decay Inr from the excited states of the molecules. Therefore, the fluorescence quantum yields Q can be written as (Demtroder, 1982; Lakowicz, 1999)
Q = I flu / I 0σ absorp = I flu /( I flu + I nr ) = Γ /( Γ + κ nr )
(7)
Bio-aerosol Fluorescence
71
where Γ is the Einstein A coefficient and κnr are depopulate rates of the excited state through fluorescence emission and nonradiative decay. When the molecules are in a low concentration with linear absorption, the quantum yield Q also can be obtained directly by experimental measurement via the relationship (Mycek and Pogue, 2003)
Iflu(measured) = kI0Qε bc = kI0Qσabsorp
(8)
where is the molar absorptivity of the molecule, b is the pathlength of the absorption cell, c is the fluorophore concentration, and k is the instrumental parameter. Therefore, the minimal detectable fluorescence needs to be at least one photon and the number of photons incident on the particle needs to be ≥ 1 / kQσ absorp .
2.4 Power/energy Requirements for Detectable Fluorescence To maximize the detectable fluorescence emission, the key is to have highly efficient fluorescence-collection instruments, represented by k, as well as to have the largest possible fluorescence quantum yield Q, and the largest absorp. The last two parameters are related to the electronic structure and depend on which state is excited or are dependent on the excitation wavelength and fluorescence emission wavelength. What follows is a typical estimation for the power or energy requirements for detectable fluorescence from a single biological spore of 1-10 µm in diameter. We assume that the quantum efficiency for the detector is generally around 0.5. The efficiency of the fluorescence dispersed grating is less than 0.5 and for the fluorescence signal that the optics can collect from the 4π sr is less than 20%. Other reflection and scattering losses of the instrument could be 20%, then k ≤ 0.5 × 0.5 × 0.2 × (1 - 0.20) = 0.04. In most molecules Γ >> κ nr , so Q ≈ 1. For a single Bacillus subtilis (BG) spore illuminated near the absorption peak at 280 nm (Lakowicz, 1999), the absorption cross section is approximately =3.8×10-14cm2 absorp /(nm·sr·spore) (Faris et al., 1997; Kaye et al., 2005; Weichert et al., 2002). Another UV absorption peak at around 229nm has even stronger absorption. However, not all the photons emitted by a light source can interact with the particles. For example, a laser with 1 nm wavelength width is focused to a 500 µm diameter spot to match the size of the laminar aerosol stream (Pan et al., 2003b). If it is a continuous wave (CW) light source, it
72
Y.-L. Pan et al.
illuminates the particle for 50 µs if the particle speed is around 10 m/s. If it is a Q-switched laser, it illuminates the particle for 10 ns. Consequently, we have to account for the interaction volume where the light source is focused to a spot (area A). Therefore, the detectable photons from fluorescence emission are given by Nphoton=kQ ( or
absorp/A)
Nphoton=kQ (
(Pτ/hγ)
absorp/A)
(E/hγ)
(9) (10)
for a CW laser with power P, or a pulsed laser with pulse energy E, respectively. Thus, for detecting a single photo-electron generated by fluorescence from a single BG spore, the power incident on the spore must be at least
Ahγ kQσ absorpτ
=
π (500/2[ µm]) 2 6.62 ⋅ 10 −34 [Js]3 ⋅ 10 8 [m/s]/280 [nm] 0.04 × 1 × 3.8 × 10 -14 [cm 2 ] × 4π ⋅ 50 [ µs]
≈ 1.5 [mW], or the energy must be higher than 1.5 [mW] ×50 [µs] ≈ 75 nJ. In order to achieve reasonable signal-to-noise ratio (S/N) for fluorescence-band detection, 4 times the power or energy for the above limited detection requirement is recommended; i.e., a 6 mW CW or 0.3 µJ/pulse light source at 280 nm. As the above parameters are related to the excitation wavelength, especially for the absorption cross section absorp, which is about 7 times weaker at 340 nm, and 60 times weaker at 400 nm (Faris et al., 1997; Kaye et al., 2005; Weichert et al., 2002), a higher power UV light source is required for excitation of the amino acid tryptophan in the biological spores. The commercially available 4th harmonic and 3rd harmonic of the Nd:YAG (266 nm and 355 nm) or Nd:YLF (263 nm and 351 nm) lasers have been widely used for bio-aerosol fluorescence excitation. The absorption cross section absorp at these wavelengths are several to ten times weaker than that at 280 nm; therefore, 60 mW or 3 µJ/pulse light source, more precisely with power flux 24 W/cm2 or with pulse energy flux 1.2 mJ/cm2, are required. For spectrally dispersed fluorescence spectrum detection, even stronger light sources are needed.
Bio-aerosol Fluorescence
73
2.5 Fluorescence Dependence on Aerosol Particle Size, Concentration of Fluorophores, Detection Direction, and Illumination Intensity Generally, fluorescence from molecules in the gas and liquid conditions is considered to be uniformly distributed over all directions. However, for aerosol particles, the fluorescence is angularly dependent. For example, plane-wave illumination on a spherical particle distorts the internal electromagnetic field in a known way: the excitation profile is not uniform and the emission profile, which depends on the location of the excited molecules, is also inhomogeneous. Lorenz-Mie theory has shown that the internal field has a focal point in the equatorial plane within the shadow fore. Hill et al. (2000) has shown that the reemission of an ensemble of molecules distributed uniformly is not uniform, but enhanced in the 180o direction. The fluorescence via three-, two-, and one-photon excitation from dye molecules in spherical microcavities has an asymmetrical angular distribution and has an enhancement ratio of intensities at 180º to 90º of 9, 5, and 1.8, respectively (Hill et al., 2000). Meanwhile, experimental results show that the dependence of fluorescence on size or concentration, from both droplets and dry-particle aggregates of bacteria, is less than predicted. This probably is the result of concentration quenching (Hill et al., 2002). Within a certain laser power range, the fluorescence intensity is proportional to the illumination intensity. If the power exceeds a certain threshold, the fluorescence spectral profile changes, breakdown (Pan et al., 1999) or the generation of new materials (Pan et al., 2001a) may occur.
3.
FLUORESCENCE OF BIOLOGICAL MATERIALS
3.1 Fluorescence Emission of Some Biological Fluorophors The molecules responsible for most of the fluorescence in most biological cells (Lakowicz, 1999), are listed as follows: 1) The amino acids tryptophan, tyrosine, and phenylalanine. Each of these constitutes 1% to 5% of the dry weight of some typical bacteria. The excitation and emission maxima of these fluorophors in solution are about 255 nm and 282 nm for phenylalanine; 275 nm and 303 nm for tyrosine, and 280 nm and 348 nm for tryptophan. In proteins that contain tryptophan as well as one of these other fluorescent amino acids, it is common for the
74
Y.-L. Pan et al.
energy absorbed by phenylalanine and tyrosine to be transferred to tryptophan, and to appear around 350 nm. 2) The reduced nicotinamide adenine dinucleotides, NADH and NADPH, which have excitation emission maxima around 340 nm and 450 nm. 3) The flavin compounds such as riboflavin, the flavin adenine dinucleotides (FAD and FADH), flavin mononucleotides (FMN), and a varitety of flavoproteins. In aqueous solutions flavins have excitation and emission maxima around 450 and 520 nm, respectively. Spectra of some of these molecules are illustrated in the literature (Lakowicz, 1999; Teale and Weber, 1957). The emission from biological materials can be more complex than suggested by the relatively small numbers of primary fluorophors listed above. First, the fluorescence of the primary biological fluorophors can depend upon their local environment, e.g. the pH or concentrations of certain ions. Many biological molecules are polymers or oligomers of smaller building blocks. For example, (a) RNA is a polymer of ribonucleic acids; (b) glycogen, starch, and cellulose are polymers of sugars; (c) proteins are polymers of amino acids, where the number of common amino acids is in the low 20s. Because many biological molecules are polymers or oligomers and the numbers of building blocks for many of these polymers is large, and because and other biological molecules are nonpolymeric combinations of simpler molecules, the number of different biological molecules is larger than any number that would make any difference to us. In a protein, different tyrosine or tryptophan molecules can exhibit different fluorescence depending upon their proximity to other amino acids in the same or other protein molecules. That proximity is determined by the 3D structure of the protein and that structure can vary with pH, ion concentrations, temperature, etc. Because of this, the amount of variation in fluorescence of mixtures can be much larger than would be expected by looking only at the individual fluorophors. One example of the effect of the local environment is that the excitation and emission maxima from different biological materials are not necessarily the same in an aerosol as in solution, because the aerosols are often dry or somewhat dry. We have measured the fluorescence spectra of some of the primary fluorophors in biological aerosols (tyrosine, tryptophan, NADH, and riboflavin). To measure these spectra, we dissolve the materials in water and aerosolize the solutions with an IJAG (Bottiger et al., 1998). This generator has a built-in drying column. The single-shot, fluorescence spectra from individual (nominal 5-µm-diameter) particles of these fluorophors, measured “on-the-fly,” are presented in Figure 2. The
Bio-aerosol Fluorescence
75
excitation is at 266 nm. Tyrosine and tryptophan have emission peaks at about 310 nm and 340 nm. The NADH emission peaks around 450 nm. The riboflavin emission from dry particles peaks near 560 nm; its emission is strongly red shifted from its emission when in an aqueous solution where it peaks near 520 nm.
Figure 2. Single-particle 266-nm-excited fluorescence spectra of common fluorophors found in biological particles. Each spectrum is for a nominal 5-µm diameter (Hill et al., 1999).
Second, there are many other fluorescent molecules in various biological systems, e.g., the chlorophylls in plants and blue-green algae, the fluorescent compound in the bacterium Pseudomonas fluorescens, the green fluorescent protein. These are less relevant to aerosol characterization. For example, there is an enormous amount of chlorophyll in the environment, but we have not noticed evidence for it in the fluorescence of aerosols. Perhaps this is the result of not looking hard enough or in regions where it might be more common, e.g., where blue-green algae may be aerosolized on a windy day. In our experience with freshly picked leaves (as measured and described in Section 3.3 below) the chlorophyll emission appears to decay rapidly; it may be that by the time a leaf degrades enough to have parts of it blown up as an aerosol particle the chlorophylls have been modified enough that they do not fluoresce.
76
Y.-L. Pan et al.
3.2 Biomolecules Exhibiting UV-excited Fluorescence Have Conjugated Double Bonds For our interest in biological and non-biological molecules that fluoresce when excited with visible or ultraviolet light, it is useful to consider some simple concepts to help estimate which of the large numbers of building blocks and more complex molecules fluoresce and at what wavelengths. First, biomolecules exhibiting strong UV-excited fluorescence have conjugated double bonds. We cannot think of counter examples. Second, the absorption and, maybe to a lesser extent, the emission from organic fluorophors can be approximated by treating the π electrons in a double bond or in a series of conjugated double bonds using a particle-in-abox model (Kuhn, 1949; O'Neill et al., 2005);. The length of this box is approximately that of the spacing between the atoms in the bond, or in the case of conjugated double bonds, the relevant length is the overall distance the electron might move through the set of conjugated double bonds. The energy levels of an electron in a box of length L are
En = ( nh ) 2 / 8mL2
(11)
where n is the quantum number, h is Plank’s constant, and m is the mass of the electron. For benzene, phenylalanine, and other ring structures, simple estimates based on a particle in a ring can be an improvement. The energy levels of an electron in a ring of perimeter l (Castanho, 2002) are
En = ( nh ) 2 / 2ml 2
( n = 0, 1, 2, 3....)
(12)
For electromagnetic energy to be absorbed, the differences between the energy levels should be less than the energy of a photon of the excitation wavelength. The length of a single carbon-carbon bond is approximately 0.14 nm, and is so short that excitation of an electron to the next electronic energy level would require photons with wavelengths below 200 nm. Molecules that absorb light in the 200 nm to 500 nm range and emit at 280 nm or longer wavelengths have conjugated electrons. If the number of conjugated π electrons is N, then the wavelength of the lowest transition band in its simplest expression is (Kuhn, 1949)
λ = 8mcL2 / h( N + 1) .
(13)
More complex expressions based on box models (O’Neill, 2005) are available, but the simplest may be best for illustrating the concepts.
Bio-aerosol Fluorescence
77
With these concepts in mind, we can re-examine some of the fluorophors mentioned above. Phenylalanine is an amino acid where the side group is a benzene ring. Its fluorescence is similar to benzene, i.e., weak at wavelengths below about 215 nm. Tyrosine’s absorption and emission occurs at longer wavelengths than phenylalanine’s because the side group in tyrosine is a phenol, i.e., a benzene group with an OH moiety. If the H of the OH group leaves, the electrons in the -C=O can be conjugated to the rest of the benzene ring, thereby increasing the overall length of the box in which the conjugated electrons can move. Tryptophan includes a combination of a 6-member ring and a 5-member ring (overall, an indole ring) and the conjugated electrons can move over these combined rings. Therefore, the absorption and emission wavelength maxima for tryptophan are longer than they are in tyrosine or phenylalanine. In NADH the conjugated electrons can move throughout a larger box than tryptophan, and in the flavins the conjugated electrons can move over an even larger distance. The excitation and emission wavelength maxima of these molecules are longer accordingly. One reason to think about these box and ring sizes and excitation and emission wavelengths, even with these simple models, is that these ideas help clarify that the majority of biological molecules should have relatively weak fluorescence. For example, some toxic biological compounds which have been a concern for their biowarfare or bioterrorist potential have no conjugated π electrons, and so would not be expected to fluoresce. Examples are the neurotoxic agents tetrodotoxin from the puffer fish and palytoxin from bacteria that live in filefish, and saxitoxin from shellfish that eat dinoflagellates. Other examples are the tricothene mycotoxins, and a toxin from blue-green algae, anatoxin A. The list could be much longer. The simple box models are most useful when they can be applied simply. However, for many non-chemists, it is not simple to estimate and understand the degree of conjugation and the expected fluorescence. For example, the bases adenine and guanine which occur in DNA and RNA are similar to tryptophan in that each has a six-member ring bound to a fivemember ring and each has one double bond in the 5-member ring. These molecules absorb in the 240 to 280 nm range, while tryptophan absorbs at a slightly longer wavelengths, 270 nm to 290 nm. However, these bases have very weak fluorescence. The conjugation is different, e.g., there are four nitrogen atoms in each of these bases, while the indole ring of tryptophan has only one nitrogen atom, but unfortunately we cannot present simple rules to explain these effects.
78
Y.-L. Pan et al.
3.3 Excitation/Emission Spectra of Some Example Biological Materials For our goal of discriminating between harmful biological aerosols and other aerosols, we would like to use excitation wavelengths that: a) excite a relatively strong fluorescence so that very small aerosol particles can be examined, and b) are relatively useful for discrimination. Excitation wavelengths that excite the strongest fluorescence need not be the ones most useful for discrimination. Also, in designing instruments, some excitation wavelengths are easier to obtain than others (see the discussion in Section 5 on light sources). Also, we would like to know the optimal emission wavelengths to measure for sensitivity and discrimination. To obtain the data needed to make the above determinations, we and others have measured excitation/emission spectra (EEM spectra) of some biological materials. EEM spectra are the fluorescence emission spectra measured for a series of excitation wavelengths. Here we describe the instrument we have used to measure EEM spectra (see Fig. 3), and then present in Figs. 4 to 15 the EEM spectra of a range of biological and non-biological particles. These are example EEM spectra taken from a larger set of EEM spectra of typical bacteria cells, spores, pollens, vegetation, and other materials (both dry and suspended in solution) measured by Mayo and Hill in 1995-1996. Other groups have measured EEM spectra of similar and additional biological and nonbiological materials. However, as with our work shown here, very little of this is published. Duarte et al. (2004) have published EEM spectra from the water-soluble components of aerosol samples gathered from the atmospheric environment. However it is not clear how to relate these measurements of complex, mixed samples, to single-particle fluorescence measurements of the type we hope to use for discrimination. Figure 3 shows the typical measurement configuration for measuring EEMs by utilizing a Fluorolog-2 Model FL2T2 Spectrofluorometer System (Spex Industries, Inc.). This continuous wave spectrofluorometer consists of a T-box sampling module with a 90o/22.5o selection mirror to reflect the fluorescence from either the front-face or angle of the sample (either in 22.5o reflection mode for dry samples, or 90o collection mode for wet samples), double-grating (1200 grooves/mm) spectrometers for both the excitation and emission in order to increase the stray- and scattered-light rejection and spatial resolution. The T-box sampling module includes a rhodamine-B reference quantum counter chamber to provide a reference signal (by using a Hamamatsu R508 photomultiplier to monitor the output. By monitoring the fluorescence signal to the reference signal, Mayo and
Bio-aerosol Fluorescence
79
Hill were able to compensate for wavelength dependence associated with the light output of the 450 W Xenon lamp and wavelength variations of the UV-blazed gratings of the excitation spectrometer.
Figure 3. The T-cell spectrofluorometer for measuring fluorescence excitation and emission spectra.
The excitation and emission spectrometer slits were set for 1 nm bandpass for all measurements in order to reduce the effects of scattered light on the spectra. EEM fluorescence spectra were acquired by varying the excitation wavelength from 250 nm to 530 nm at 20 nm bandpass increments. The emission was measured starting at 20 nm above the excitation wavelength. The ending emission wavelength was 20 nm before the second order wavelength line, or 800 nm, whichever was smaller. The integration time was 0.1 second for all samples. No background correction or data smoothing was used. The excitation and emission wavelengths were varied automatically. The sample suspensions were placed in quartz cuvettes (1 cm) in a Tbox sampling module. The quartz cuvettes were excited at normal incidence and the fluorescence emission measured at 90o. Solid samples were measured as follows: Bacteria, pollen, and the biological compounds were pressed into a solid sample holder (SPEX Model 1933) in the T-box sampling module without a cover slip. The solid sample holder does not exhibit any significant fluorescence. The dry sample is then illuminated
80
Y.-L. Pan et al.
with light from the 450 W xenon lamp through the double-grating monochromator at an angle of about 4o from normal incidence. Fluorescence emission was measured through the front face port of the Tbox sampling module which is at 22.5o. Figures 4–11 show the fluorescence EEM spectra of several typical bacterial cells, Bacillus anthracis (Vollum) vegetative cells as a suspension in water, Bacillus anthracis (Vollum) vegetative spores as a suspension in water, Bacillus subtilis vegetative cells in dry form, Bacillus subtilis vegetative cells as a suspension in water, Pesudomonas fluorescens in dry form, Yersinia pestis in dry form, Micrococcus luteus in dry form, Staphylococcus aureus in dry form, respectively. Figure 12 -- 15 show the fluorescence EEM spectra of mulberry pollen, corn pollen, pecan pollen, and kaolin particles, respectively. The emission from dry materials appear red-shifted from that of the corresponding wet materials, an effect that was mentioned above for flavins.
3.4 The Distinction between Biological and Non-biological is Fuzzy in Some Cases One objective for the measurement of laser-induced fluorescence of individual airborne particles is to determine which particles in an air sample emit fluorescence consistent with biological particles or possibly harmful biological particles. If the LIF can be used to determine subsets of these categories, that is even better. If there is an increase in the fraction of particles estimated to be biological and possibly harmful, then this estimation may be used to decide if there is a need to try to determine what those possibly harmful particles are. Alternatively, an individual particle having fluorescence consistent with biological particles may be separated for further analysis. Because of these and related objectives, and because we often use the categories of biological and non-biological, and because it is not always clear where to draw the line between these, we discuss here briefly one way to define the difference between these categories, and limitations of distinguishing these. We use the term “biological molecules” to indicate molecules that occur in or are made by living things or viruses, but are relatively uncommon outside of living things. For example, water, sodium, magnesium, chloride and bicarbonate ions, are all common in living things, but also are common elsewhere, so we do not classify these as biological. Carbonate anions, carbon dioxide, and oxygen are made
Bio-aerosol Fluorescence
Figure 4. Fluorescence excitation and emission (EEM) spectra of Bacillus anthracis (Vollum) vegetative cells in water.
81
Figure 5. Fluorescence excitation and emission (EEM) spectra of Bacillus subtilis vegetative cells in water.
Figure 6. Fluorescence excitation and Figure 7. Fluorescence excitation and emission (EEM) spectra of Pseudomonas emission (EEM) spectra of Bacillus anthracis fluorescens in dry form. (Vollum) vegetative spores, in water.
Figure 8. Fluorescence excitation and Figure 9. Fluorescence excitation and emission (EEM) spectra of Bacillus subtilis emission (EEM) spectra of Yersinia pestis in vegetative cells in dry form. dry form.
82
Y.-L. Pan et al.
Figure 10. Fluorescence excitation and Figure 11. Fluorescence excitation and emission (EEM) spectra of M. Lysodeikticus emission (EEM) spectra of Saureus vegetative cells in dry form. vegetative cells in dry form.
Figure 12. Fluorescence excitation and Figure 13. Fluorescence excitation emission (EEM) spectra of Mulberry pollen and emission (EEM) spectra of Pecan in dry form. pollen in dry form.
Figure 14. Fluorescence excitation and Figure 15. Fluorescence excitation and emission (EEM) spectra of corn pollen in dry emission (EEM) spectra of kaolin particles in form. dry form.
Bio-aerosol Fluorescence
83
primarily by biological systems. We do not classify these as biological because once these have been formed by biological systems they may have remained in the environment and been transformed, in some cases independently of biological action, for many millions of years. We do not classify as biological coal, oil extracted from the earth, or black carbon generated by the burning of wood, because these materials have been modified extensively by non-biological processes. Some of the simpler molecules that we call biological are very small, e.g., urea, the acetate anion, ethanol, glycerol, glucose, and formic acid (made by ants to make a painful bite). In discussing how we use laser-induced fluorescence to distinguish between biological and non-biological materials in airborne particles, it is useful to remember that the distinction between biological and nonbiological molecules may be fuzzy. The fuzziness of the division, in our definition, is illustrated by molecules that are initially clearly biological, e.g., lignins in a living tree, which are slowly oxidized and transformed after the tree dies. A decaying tree may be oxidized and otherwise metabolized by bacteria and or fungi, largely to carbon dioxide and water that we consider non-biological, but also to a host of partially oxidized materials, some of which may be, or may become, humic and fulvic acids, which are common in the environment. We are hesitant to classify these materials as non-biological, partly because some have structures similar to compounds in plants and may result directly from biological degradation. The fuzziness of the division is further illustrated with a related question: When has biological plant material decayed enough or been transformed enough in the direction of coal or oil to be considered non-biological? In using LIF to distinguish between biological and non-biological materials in airborne particles, it is useful to remember that many nonbiological molecules have fluorescence similar to biological molecules. For example, naphthalene has a fluorescence emission somewhat similar to that of the amino acid tryptophan. We think of naphthalene as a non-biological molecule that results from complex transformations of molecules that were at one time biological. For another example, in our measurements of single particle UV-LIF at Adelphi, MD (Pinnick et al., 2004), a large fraction of particles had fluorescence that appeared consistent with humic acids, which could be considered to be biological, or of somewhat recent biological origin. However, similar substances (for example, humic-like substances or HULIS) having similar fluorescence can be obtained by partial oxidation of other hydrocarbons that come from oil or coal and are non-biological in our terminology.
84
Y.-L. Pan et al.
3.5 Fluorescence Spectra Are Not Expected Typically to Discriminate Between Species The differences between fluorescence spectra of different, well-washed bacteria may be very minor (Hill et al., 1999). The same species of bacteria when prepared in different ways display more significant differences in their fluorescence spectra. The differences occur because other fluorescent materials can be mixed with the bacteria when they are prepared and washed in different ways; e.g., the remaining growth media may fluoresce. Several facts strongly suggest that UV-excited fluorescence spectral measurements are not sufficient for characterizing particles at the biological-species level, except possibly in some special cases. Some are as follows: 1) there are relatively few primary fluorophors in the biological cells; 2) the fluorescence emission from each of these primary fluorophors is broad; 3) there are many thousands of different bacterial species, including hundreds of types of spore-forming bacteria; 4) each of these bacterial species contains many thousands of different proteins, many of them similar from species to species and so are not very likely to be useful for discrimination; 5) the fluorescence can depend upon the metabolic state of the cell, e.g., NADH, the reduced form of NAD, is far more fluorescent than NAD, and it is more prevalent in cells that have sufficient energy and are actively metabolizing than it is in cells that are starving; and, 6) there are many different materials in atmospheric particles, including bacteria, proteins, pollens, humic acids, atmospheric hydrocarbons, soil minerals, cement particles, etc.; all combinations of these can occur in atmospheric particles. Also, the signal to noise ratios for the single-shot, single-particle, fluorescence spectra are not ideal, and the problem cannot be circumvented by increasing the laser intensity because nonlinear effects play a significant role with increased intensity. Some additional discrimination may be obtained because the primary fluorophors can fluoresce differently depending upon their local environment. For example, each tryptophan molecule in a 3-D protein structure may have a different local environment, interacting differently with the solvent and closest other amino acids at each location. And so there can be differences between the fluorescence from different proteins. Also, the time dependence of the fluorescence emission can be used to discriminate between different fluorophors that have similar emission spectra. However, measurements of time-dependent, single-particle, fluorescence spectra of individual particles have not, to our knowledge, been described in the literature, and it does not appear likely that there
Bio-aerosol Fluorescence
85
would be sufficient signal for such measurements to be made on 1micrometer-diameter particles at even a few emission wavelengths. Even if these local-environment-dependent differences were large, there are many thousands of different proteins in any bacteria, and many of these proteins are similar or identical in different species, and so these localenvironment-dependent effects may be similar from species to species. For an example of across-species similarities, penicillin inhibits the growth of a broad spectrum of bacteria. It does so because bacteria make the same peptidoglycan cell wall. Differences in the efficacy of penicillin depend more on how the bacteria resist the penicillin than they do on differences in the proteins that make that cell wall. Other antibiotics inhibit growth of a broad spectrum of bacteria because they bind to the same 30S ribosomal subunit. Many more examples of conserved proteins and groups of proteins could be given. In the natural environment bacteria may appear in the same airborne particle with cellulose, lignins, humic and/or fulvic acids, chlorophylls, etc. These other materials may suggest something relating to sources of such samples, and how they were prepared. For example, if the fluorescence from some known types of growth media is measured in particles, and if the background aerosol had been measured over some time previously and nothing that fluoresced similar to growth media had appeared in that background aerosol, then the measurement of the fluorescence spectra of growth media may suggest that manmade particles had been introduced into the atmosphere. That information may be sufficient to set off alarms. Although in this subsection we hammered home the point that LIF-spectral measurements are not sufficient for characterizing particles at the biological-species level, most of the particles in the atmosphere do not exhibit LIF that looks like bacteria, and so LIF can provide a strong discrimination against these other particles that do not look like threats, but this is getting ahead of the story.
4.
STANDARDS FOR BIO-AEROSOL MONITORS
4.1 Concentration of Aerosols BW agents are classified into three main groups by the US Department of Defense: bacterial, viral and toxin. There are no known biological agents that are molecular vapors, i.e. have measurable vapor pressure at STP, so when considering airborne BW detection, it is implicit that the agent has been dispersed as aerosols in some manner. The essence of BW aerosol
86
Y.-L. Pan et al.
detection is to differentiate and detect certain aerosols on the basis of their composition, independent of particle size. Fortunately the size range is somewhat restricted. For many BW agents there exists a minimum unit size. As an example, Bacillus anthracis spores are typically about a micrometer in length, by about 0.5 to 0.75 micrometers in diameter. These dimensions are representative for most bacterial agents, vegetative cells as well as endospores, and represent a minimum unit size, since fragments of cells or spores are ineffective as agents. For viral agents the indivisible units, virions, are smaller, typically of dimension 100 nanometers or less. For toxin agents, the basic units are even smaller: a single biochemical molecule. Botulinum toxin, for example, has a molecular weight of about 50,000 for the active, light chain part. However, as a practical matter, there has not been any documented method by which aerosols in the size range of 0.1 micrometer or less are dispersed as single aerosols without significant agglomeration taking place, so as part of the aerosolization process the dispersed mass distribution will likely reach a maximum at sizes larger than a micron. Consequently, most discussions of BW agent aerosol detection consider particle sizes in the range from 1 to 10 µm corresponding to the optimum retention range for lung tissue. Therefore, it is possible to conceptually approach the topic of airborneBW-agent detection in terms of aerosol accounting. In the following discussion, we assume that the detection method involves drawing an air sample into the sensor or detection instrument, and that the aerosols are interrogated individually. With these two caveats, let us designate the sensor inlet flow rate as F liters/minute, so that within the sample time τS of the instrument, a volume V of air is interrogated where V = FτS. Let N0 be the total number of particles contained in V in the appropriate size range, so that N0 = NAB + NT, where NAB is the number of ambient background aerosols and NT is the number of threat particles. Clearly, in normal situations NT is zero. Now let us designate the detection threshold concentration of the sensor as CT threat particles per liter. This means that the sensor should be able to respond with an alarm when a total number of threat particles corresponding to a concentration of CT are present in the sample volume, or NT = CTV. However, this last relation is not quite correct; the detection threshold is meaningful as an ensemble average over a large number of equivalent volumes V. The actual number of particles in a given volume V fluctuates; therefore, the correct expression is given in terms of the average number of threat particles: = CTV. If we now assume that the statistics for aerosol number in a fixed volume are described by a Poisson distribution, then in order to achieve the detection concentration threshold of CT, the actual alarm threshold must be set lower than depending on the confidence level, or probability of detection PD
Bio-aerosol Fluorescence
87
that the sensor is required to operate. A quantitative example maybe helpful to illustrate these points: Let us assume that the threshold concentration CT is 5 threat particles per liter, that the sensor has an inlet flow rate F of 10 liters/minute, and that the required sample time τS is 30 seconds for which the desired confidence level should be 95% or better. From the relations above, we have: V = 10 × 0.5 = 5 liters, = 5 × 5 = 25 threat particles, and for Poisson statistics with a mean of 25 (standard deviation of 5), one minus the cumulative probability can be calculated to reach 0.96 for an expected particle number of 16, so that for this system, NT = 16. The purpose of this example is to illustrate that there are constraints and physical limitations carried in the sensor design and performance requirements. In this example, one can see that if the sensor had a flow rate of 1 liter/min instead of 10 liter/min, would be 2.5 particles, and that the actual threshold, NT, would have to be set to less than 1 aerosol to achieve the 95% confidence level, or PD. Since the aerosols are indivisible units, this is not a physically realizable criterion, and consequently one would have to relax either the PD requirement or the response time to develop a realistic design.
4.2 Performance Metrics This simple discussion highlights three of the four fundamental performance metrics (FPMs) that characterize all detection systems in general: (1) response time, (2) detection threshold (CT) and (3) probability of detection PD. The fourth FPM is probability of false alarm PFA, which will be discussed in later paragraphs. These performance metrics are highlighted in Table 1. The word “sensitivity” is frequently used in discussions on this topic instead of detection threshold, but in the author’s experience, using “sensitivity” here for characterizing a detection system frequently leads to confusion due to the multiple usages and meanings of the word, and therefore has been deliberately avoided. In an effort to be more precise, the term “threat concentration threshold” is used. The quantity that is really of interest here is the minimum amount of material that can be confidently measured and identified. However, this amount of material is connected to a level of confidence, or the probability of detection PD. The phrase “limit of detection” might be considered in this context, but this term is defined specifically for a PD of 0.99. Depending on the user and the requirements that the detection system is intended for, the PD could be 0.90 or 0.95, etc., so a more generalized term is needed. From the example in the preceding paragraph, it should not be surprising that these metrics are interrelated. It should also be noted that for airborne BW
88
Y.-L. Pan et al.
detection the units of measure for CT is a concentration of aerosols per unit volume, and are frequently given as agent-containing particles per liter of air (ACPLA). While these units are convenient for aerosol accounting concepts as just discussed, they are not well suited to consequence prediction that depends on dosage exposure. Since the number of organisms per aerosol can be assumed to scale with the particle volume or mass, then the nominal size range of 1 to 10 micrometers translates to a dose variability range of 103 per particle. In practice, the issue of particle size distribution also directly impacts the sensor performance as well. In summary, an isolated statement such as, “the instrument should be capable of detecting 25 ACPLA” is simply incomplete, and only acquires meaning when combined with specifications of the remaining performance metrics as a complete set. Table 1. Fundamental Performance Metrics (FPMS) for Sensors τ Response Time CT Threat Concentration Threshold PD Probability of Detection PFA Probability of False Positive 4.2.1 A simple model for a generic detection system To facilitate a more detailed discussion of detection-system characterization, a simple model of a generic detection system is shown schematically in Figure 16. This generalized detection system consists of three sequential modules or components: (a) sample acquisition, (b) measurement (detection), and (c) data processing and classification. As before, it is assumed that a BW aerosol detector samples the air and individually interrogates all the particles in a volume compatible with the desired concentration threshold, as described above. So the first stage, sample acquisition, could take from seconds to minutes, depending on the specific operating parameters of the sensor. For the second stage, measurement, this model is generic, but since this chapter has focused on optical phenomenologies (especially elastic scattering and laser-induced fluorescence) to provide the aerosol data for detection and discrimination, these methods will be highlighted. Therefore, in the measurement stage a set of optical signatures are acquired by photodetectors for each particle and converted to electrical signals or data. This step can occur relatively quickly as the scattering and fluorescence signals are created and propagate in nanoseconds or less. The greatest time-consuming aspect of the measurement step will likely be signal digitization and recording (data
Bio-aerosol Fluorescence
89
acquisition). However, the focal volume for optical measurement must be smaller than the reciprocal total aerosol concentration to avoid simultaneous interrogation of multiple particles. Also, the inlet flow rate may be constrained by the maximum frequencies of the laser excitation and/or electronic data acquisition systems. Therefore, the speed of the sample acquisition stage is somewhat interdependent on the measurement (data acquisition) stage. The third and final stage, data processing, extracts the desired signal features, and classifies the aerosol data in some way to determine whether or not a threat is present in the ambient background aerosol (clutter).
Figure 16. A general schematic diagram of the components of a detection system illustrating the origin of the basic metrics: Sensitivity - noise equivalent signal, Response time - sum of sample acquisition, detection, and data processing times, and a decision boundary as a user defined input determining the tradeoff between PD and PFA.
Using this basic model, the first FPM, total response time of the detection system, is easily seen to be the sum of the sample acquisition time τS, measurement time and data processing time. Ideally, one would optimize the system to work at near maximum data acquisition rate so that the air inlet flow rate would vary depending on the total aerosol loading. In practice this is relatively difficult to achieve, and development efforts have so far focused on more fundamental measurement and discrimination issues. The statistical level of confidence of the detection system and the minimum threat concentration threshold dictates the volume of air or number of particles that must be interrogated, and therefore the sample acquisition time. In general, a trade-off exists where higher confidence and/or lower concentration thresholds require longer response times. In terms of the BW-detection system utility, one of the motivations for using
90
Y.-L. Pan et al.
optical phenomenologies is to take advantage of their inherent speed. The more quickly the detection system can alarm, the greater the benefit in terms of minimizing or avoiding exposure. A realistic but challenging goal is to achieve a time response of a minute or less while satisfying the remaining FPM requirements.
4.3 Minimum Threat Concentration In considering the second FPM, the minimum threat concentration CT, the sources of noise and variability in the sample measurement can be examined in the context of this generic model. In general, there are three distinct noise sources: (a) the noise floor of the detector in the absence of any sample, (b) the shot noise of the signal from an aerosol, and (c) the ambient aerosol background in the absence of any target. Sources (a) and (b) are conceptually very distinct from (c) to the extent that they can both be addressed directly in terms of physical parameters involving the optical design and the optical properties of the aerosol materials, such as the fluorescence cross section discussed in Section 2 of this chapter. The scattered light signal from the optical system in the absence of an aerosol, and the inherent noise level of the photodetectors (e.g., PMT dark current) determine the level of the signal in the absence of a particle, and therefore the smallest particle size that can be detected. In this regard, the practical limit for single aerosol detection is probably in the range of 0.1 to 0.5 micrometers in size based on the mature technology of commercial aerosol counters. Since the minimum size of interest is 1 micrometer, this implies that effectively all the particles of interest in a given volume of air can be detected. However, the much more significant limitation in terms of minimum threat concentration threshold is the last source of noise, the misclassification of non-threat aerosols as threats. To reinforce a point regarding terminology here, optical measurements of aerosol elastic scattering and/or laser-induced fluorescence can have sufficient sensitivity to detect a single one-micrometer-sized aerosol, yet a single particle is far from sufficient to declare an alarm. Fluorescence signals from a 1 micrometer particle can have poor signal-to-shot noise ratios, so that the classification that occurs based on these data are more prone to error than from larger particle, so we see also that size does matter.
4.4 Data Processing for Aerosol Classification The next stage of the detection system, data processing, is where decisions regarding the classification of the measured aerosols are made
Bio-aerosol Fluorescence
91
based on the acquired data. As an example, consider the intensity of fluorescence in the spectral range of 320-350 nm excited by a laser pulse at some wavelength shorter than 290 nm. If the fluorescence intensity is normalized to the incident laser pulse intensity, and to the particle size to reduce the variability in the measured data due to those external parameters, then the measured intensity indicates something about the composition of the aerosol.
Figure 17. shows a drawing of hypothetical histograms for two sets of measured aerosols for a single channel of fluorescence intensity normalized to particle size. The threshold is an adjustable intensity level for separating individual aerosols into two groups, but always has some misclassification since the two distributions overlap.
Figure 17 is a hypothetical illustration of such data collected for two groups of aerosols: ambient background and threat aerosols. Data on the threat aerosols would be collected in a closed chamber or flow system of some type. The plot shows the relative frequency of occurrence for each of the two classes of aerosols, and one can see that the threat aerosols have a significantly higher mean and median fluorescence intensity than the ambient aerosols. Given this type of data, a classification capability could be proposed based on a constant threshold: aerosols with fluorescence intensities higher than this number would be classified as threat particles, and otherwise as ambient. Assuming that these distributions are stationary, then the probability of detection PD can be estimated by simply calculating the fraction of the total threat aerosol population that exceeds the threshold. By moving the threshold up or down (right or left on the plot in Figure 17) one can respectively either decrease or increase this fraction from 0 to 1. However, as shown, the two populations substantially overlap. The probability of producing a false positive, that is of misclassifying an ambient aerosol as a threat, is estimated by the fraction of ambient aerosols that exceed the threshold value. As the threshold is decreased (moved to
92
Y.-L. Pan et al.
the left) to increase the PD toward 1, the fraction of included ambient aerosols increases nonlinearly. This simple example illustrates the intrinsic trade-off relationship between PD and PFA. It also illustrates the necessity for having an a priori, or user-defined threshold or classification criterion. In general, the data used to classify is referred to as a feature. In this case, the normalized fluorescence intensity provides discrimination between two classes of samples, and is exploited as a classification feature.
4.5 Multiple Measurements for Statistical Characterization: Confusion Matrices Making multiple measurements, for example, a spectral profile of emission, rather than a single integrated total intensity frequently provides greater discrimination capability because multiple features may be used simultaneously to provide classification information. In this context the word “feature” means some prescribed aspect of the measured signal. As examples, one could consider a spectrum as a measured signal, and a feature could be a peak position, a peak width, the slope or curvature of a peak as extractable features of the spectral data. In the case of an image signal, the ratio of particle circumference to area, the aspect ratio, or the absence of any concave part of its shape could all be examples of features. In general, a feature is any aspect of a measured signal that can be used to provide separation among different classes of samples, and any measured signal such as an image or a spectrum could possess multiple features. An established method of statistical characterization of the performance of a measurement technique or its associated features for classification is the use of confusion matrices. This approach is especially useful in situations with multiple features and/or multiple classes of objects or samples. For the application considered here both multiple features and multiple sample classes or targets are important. Figure 18 illustrates two generic features of a hypothetical measurement signal that together show separation among three sample classes represented as triangles, circles and squares. In this context, the exact nature of the features themselves is not relevant, but together they define a data space, in this case a twodimensional plane, that exhibits the measured data. The individual points in Figure 18 represent independent measurements, and the spread in these data indicates either noise or a variation of some other unknown aspect within each sample class. If enough data measurements are collected on these samples, we assume the frequency of these points would define a probability distribution for the sample materials specifying the probability for a measurement occurring in any particular area element (dx. dy) of the
Bio-aerosol Fluorescence
93
data space. We assume this distribution is intrinsic to these sample classes. One can see from this plot that using either feature alone would increase the overlap between the classes, but that by using both features, separation of the classes is improved, and in particular, class 3 appears to separate completely from classes 1 and 2. However, classes1 and 2 remain slightly overlapped.
A B C
Figure 18. Schematic representation of a measurement that has two features (1 & 2) that together provide some separation among three different samples. Lines A, B and C illustrate alternative decision boundaries in the data space.
4.5.1 An Example To quantify an example, consider the line labeled A to be the decision boundary for separating sample 1 from sample 2. The performance of these features over that set of samples with the designated boundary can be described in terms of the confusion matrix shown in Figure 19. The (1,1) element of this matrix would read as follows: the probability of correctly identifying Sample 1 is 19/22, since 19 of the 22 data points for that sample occur to the upper left of the boundary line A. The (1,2) element would read as the probability of (mis)identifying (confusing) Sample 1 for Sample 2 is 3/22, and in general, the element (i, j) would be the probability of identifying the ith sample as the jth sample. Therefore, in this representation, the diagonal elements are the probabilities of correctly identifying each sample, while the off-diagonal elements are the probabilities of confusing one sample with another. In this particular example, the matrix reflects the fact that the data shows Sample 3 as
94
Y.-L. Pan et al.
being completely separable from Samples 1 and 2, by having 0s in the off-diagonal elements for column 3 and row 3 (decision boundary for sample 3 not shown). Likewise, the fact that the data points for samples 1 and 2 are overlapped is reflected in the off-diagonal elements (1,2) with a value of 3/22, and (2,1) with a value of 3/17, and the fact that the diagonal elements (1,1) and (2,2) are less than unity. Notice that the probabilities of confusing Sample 1 with Sample 2, and Sample 2 with Sample 1 do not have to be equal. Only the sums of the individual rows should equal unity since a given class must be distributed among a fixed set of possible categories and the sum of those probabilities is conserved.
1
2
3
Sample 1
19/22
3/22
0
Sample 2
3/17
14/17
0
Sample 3
0
0
1
Figure 19. Confusion Matrix.
With this approach, it is important to emphasize that specific values of the confusion matrix elements are not unique. In separating the locus of data points between Samples 1 and 2, the user must identify a decision boundary to specify areas of the data space and compute the probabilities that a data point occurring in an area of the data space is correctly or incorrectly counted as one sample or another. In general, there is no fixed or prescribed method for generating the confusion matrix, and in the example above, the values in the matrix were achieved by using the straight line labeled A as a decision boundary for separating Samples 1 and 2. A different decision boundary, such as the line labeled B would provide a zero probability of confusing Sample 1 with Sample 2 at the expense of a slightly lower probability of correctly identifying Sample 1 and a slightly higher probability of confusing sample 1 with sample 2. Alternatively, Sample 1 could have a detection probability of 1 with a decision boundary such as line C, but at the expense of an increased probability of misclassifying Sample 2 as Sample 1. Frequently the decision boundary is implicit, and usually one does not directly control it. However, we would like to emphasize that the focus is not on the boundary per se, but on the probability distributions themselves as represented by the data. Overlap of the probability distributions may be intrinsic to the classes themselves and
Bio-aerosol Fluorescence
95
the measurements, which defines the features or data space. In this example, if Sample 1 is a target of primary interest, and the other samples are regarded as background, then picking a decision boundary of line B minimizes the probability of false positives PFA for detecting Sample 1 at the expense of its probability of detection PD.
4.6 Receiver Operating Characteristic or ROC Curve For a single target sample, and a one-dimensional data space (i.e., having one feature) as illustrated in Figure 17, the decision boundary simply becomes a threshold number, and one can easily represent the trade-off between PD and PFA in a graphical manner as a function of this threshold parameter. This kind of plot is known as the receiver operating
Figure 20. Confusion Matrix The classic ROC curve compares probability of detection PD versus probability of false positive PFA on linear scales from 0 to 1. Three performance curves are plotted.
characteristic or ROC curve. However, in a multi-dimensional feature space, and/or with multiple target samples, such a simple decision boundary parameterization is not usually available. In terms of confusion matrices, the diagonal elements are easily recognized as the probabilities of correct identification or PDs for each of the samples, while the off-diagonal elements affect the probability of false positive. Figure 20 shows the classic representation of a ROC curve in which the probabilities of
96
Y.-L. Pan et al.
detection and false positive are plotted linearly from 0 to 1. The baseline for this representation is the straight diagonal for which the probabilities of correct and incorrect classification are equal. In reference to Figure 17, this condition would be represented if the distributions of the two sample classes would be identical so that the computed fractions for PD and PFA would be equal for any value of the threshold. In this sense the ROC curve is a parameterized plot like the trajectory of a projectile, where the parameter not shown is time. In the ROC case the parameter is threshold, or in the multi-dimensional feature space, a complex decision boundary. In the classic plot of Figure 20 the quality of the detector performance is gauged by the deviation from the diagonal baseline into the upper left quadrant, and the performance is sometimes further simplified by quantifying the area between the performance curve and the diagonal baseline. So, the detection system performance can vary between 0 (diagonal) and 0.5 (filling the upper left half of the square). In Figure 20, two hypothetical performance curves were generated, one labeled “good” (course dashed curve, squares) and another labeled “excellent” (fine dashed curve, diamonds). The good curve is a segment of a circle, while the excellent curve is indistinguishable from the axes of the plot frame in this scale. There are a host of difficulties with this representation and ROCs in general. First, the ROC curve is a distillation of information contained in the actual probability distribution function as represented by the data. In the typical case where there are more than two features, some type of summary is desirable since it is not possible to visualize more than 3 dimensions. However, the reduction of performance data to a single ROC is frequently so great that distinction between different systems is barely perceptible. The second issue that is hidden by the use of ROC curves is that the decision boundary provides the curve. In the simple case for a single feature and scalar threshold the curve is unambiguous. However, for multi-dimensional feature spaces there may be many different ways in which the decision boundary can be varied, each giving rise to a different ROC curve on the same data base. Finally, the issue of performance in terms of a variable background, which is the whole point of autonomous sensors, means that a given physical instrument with a given detection algorithm (classifier) will have a different ROC curve for different environments and different threat concentration thresholds, so as a practical matter it is not clear that the summary representation of the ROC is necessarily the most efficient representation for conveying the essential characterization of the system. Since BW aerosol detection systems are attempting to move to response times of a minute, the probability of false alarm must move to the values of 10-5 (once per week). So at a minimum, it seems essential to at least agree on rescaling the usual representation of the
Bio-aerosol Fluorescence
97
ROC curve to a logarithmic scale for PFA as shown in Figure 21. Here, the same values that were plotted in Figure 20 are simply replotted on the new coordinate scale. This at least allows the differences between values of PFA that are of interest (10-5 versus 10-6 ) to be seen.
Figure 21. For bio-detection, it is desirable for PFA to be small (e.g. < 10-5). The linear plot is not adequate to show these small values, and the ROC curve should be rescaled as a loglinear plot. Here the same performance curves from Fig. 20 are plotted again on a logarithmic horizontal axis for PFA.
5.
LIGHT SOURCES FOR UV EXCITED FLUORESCENCE
5.1 Selection of Appropriate Excitation Wavelengths The choice of an appropriate excitation light source for UV-excited fluorescence depends on the wavelength and power requirements. For the wavelength, it is critical that the resulting fluorescence spectra at the excitation wavelength contain enough information for discrimination, but have the least interference fluorescence from other bio- or non-bio-aerosols. This excitation wavelength should be generated by the highest efficiency available, the most inexpensive, and the lightest light source, in addition to
98
Y.-L. Pan et al.
producing a measurable fluorescence signal from bio-aerosols in the size range (1-10 µm). For the best discrimination among biological materials composed of bacterial spores, pollens, algae etc., laser wavelengths that excite all four of the main types of fluorescent compounds (amino acids, NADH, flavins, and chlorophylls) are expected to be most useful. These correspond to wavelengths shorter than 310 nm. Wavelengths near 280 nm are efficient for tyrosine, tryptophan, and phenylalanine excitation, but even higher efficiencies can be obtained at 240 nm. Efficient wavelengths to excite NADH and flavin compounds are near 340 nm; however, once the wavelength gets shorter, there are more interferents that could get excited. In order to get better discrimination, multiple wavelengths excitations are currently being implemented.
5.2 Excitation Wavelength for UV-LIF and Raman Spectroscopy Raman spectroscopy conducted in the UV, visible, or near IR, is affected by the choice of incident wavelength. The dependence of the incident wavelength is distinctively different for the Raman process and for fluorescence, as discussed in Section 2. The Raman peak is always at a fixed energy difference between the incident photon and the Raman photon. The energy difference is usually measured in wave numbers where 1 eV corresponds to about 8067 cm-1. The Raman peaks are measured in energy shifts from the incident photon energy with the energy shift equal to zero at the incident photon energy. The Raman peak is always slaved to the incident photon in units of cm-1 which corresponds to the vibrational frequency. The fluorescence peak, however, is not slaved to the incident photon energy. The incident photon needs to be resonant with an absorption band. The efficiency for Raman scattering is proportional to 1/λ4 if there is no absorption band nearby. An example of the ratio of efficiency between generating Raman scattering with λ = 1.06 µm (fundamental) to 0.266 µm (4th harmonic) is 44 or 256. However, if the incident photons are able to reach an absorption band, then the Raman scattering efficiency increases faster than 1/ λ4. This additional increase is called resonance Raman effect (RRE). Therefore, RRE and fluorescence can both be observed on the red side of the incident wavelength. However, the RRE signal is much weaker than the fluorescence signal and it also occurs much closer to the incident photon frequency. RRE applied to biological samples can be found in the literature (Chadha et al., 1993; Manoharan et al., 1990; Manoharan et al., 1993;
Bio-aerosol Fluorescence
99
Nelson et al., 1992; Nelson et al., 2004; Storrie-Lombardi et al., 2001; Wu et al., 2001). Non-resonance Raman measurements with visible light excitation also have been used in various bio-aerosol detection systems and some fluorescence interference were found (Dalterio et al., 1987; Dalterio et al., 1986; Sengupta et al., 2005). Therefore, it is considered by some researchers that deep UV (200–280 nm) excitation may not only give rise to strong RRE signals but also to the fluorescence of amino acid around 330 nm. The RRE and the fluorescence emission are greatly separated in wavelengths and therefore, avoid being mis-identified. Various types of UV light sources have been used in bio-aerosol detection and characterization. These sources can be divided into three groups: laser, lamp, and light emitting diode (LED). The advantages and disadvantages of these light sources for purposes of bio-aerosol discrimination are discussed below.
5.3 Pulsed and CW Lasers The first (solid state) Ruby laser was invented in 1960. On the heels of this invention, many kinds of pulsed and CW lasers, including gas lasers, metal-vapor lasers, dye lasers, semiconductor lasers, free-electron lasers, Xray lasers, and Raman lasers were developed. These lasers span the wavelength range from a few tens of nm to a few tens of µm. There are two key laser-source characteristics essential for fluorescence excitation of bioaerosols: (1) the power flux (W/cm2) and spectral power density (W/δλ) is many orders of magnitude higher than other light sources; and (2) the much lower divergence of a collimated laser beam makes the plane wave approximation valid. Due to the energy requirements necessary to detect the fluorescence from a single bacterial spore, a certain incident energy is required of the light source. This requirement restricts the available lasers that could provide a suitable wavelength with enough signal-to-noise ratio. Most fluorescence-based bio-aerosol detection systems use lasers as their excitation light source. Wavelengths suitable for bio-aerosol fluorescence excitation include the following: 263 nm and 266 nm from the 4th harmonic of Nd:YLF and Nd:YAG laser; 351 nm and 355 nm from the 3rd harmonic of Nd:YLF and Nd:YAG laser; 248 nm (KrF), 308 nm (XeCl), 353 nm (XeF) from excimer lasers; 325 nm from HeCd laser; 337 nm from nitrogen gas laser; 363 nm and 488 nm from Argon ion laser; and 280 to 316 nm from Ce:LiSAF, Ce:LiCAF lasers. In order to achieve a fairly good S/N fluorescence spectrum from bioaerosols, a light source with power flux 24 W/cm2 or pulse energy flux 1.2
100
Y.-L. Pan et al.
mJ/cm2 is recommended as discussed in section 2. For a laser focused to a 500 µm diameter spot the equivalent laser power for a flux of 24 W/cm2 is 60 mW. For most lasers, 60 mW can be achieved, even for the semiconductor laser. A pulsed laser must be able to produce a single pulse of light whenever the targeted particle traverses the sample volume. However, for CW or continuously running pulse lasers, there is no need to trigger the laser, but triggering the detector may be necessary to achieve optimal S/N. In sensors that measure single-particle fluorescence, particle velocities generally range between 1 to 10 m/s.
Figure 22. Image of a real-time, in-situ single-shot, single-particle fluorescence spectrum based bio-aerosol discriminating, sorting, and collecting system, which is employed in the ventilation room of a major international airport.
The Yale University and Army Research Laboratory (ARL) research team has been developing a rapid point-detection system that is suitable for bio-aerosol particle discrimination, sorting, and collecting based on dispersed UV-LIF spectra. The fluorescence spectra are excited by the 4th
Bio-aerosol Fluorescence
101
harmonic of a Nd:YLF (263 nm) or a Nd:YAG (266 nm) lasers. The 4th harmonic of a Nd:YLF (263 nm, model DC50-263, Photonics Industries, Inc.) laser can be triggered externally on demand from a single shot to 10 KHz with a maximum pulse energy 50 µJ, pulse duration 15 ns upon arrival of an individual aerosol particle. Once focused to a 1 mm2 spot, this provides a pulse energy flux of 5 mJ/cm2 (about 4 times stronger than the recommended flux 1.2 mJ/cm2), and highly resolved fluorescence spectra (250 nm to 700 nm with 1024 or 32 pixels) from laboratory-generated particles or ambient aerosols from outdoor testbeds (size within 1-10 µm in diameter). Very good S/N spectra have been observed from single aerosol particles on-the-fly, excited by such a laser system. Sandia National Laboratory has been working with the Yale/ARL team together to reproduce a bio-aerosol sensor that is similar to the Yale/ARL sensor to be employed in a major international airport as part of the Department of Home Security Prospective and Responsive Options for Airport Counter-Terrorism testbed to diagnose false alarms of current commercial bio-aerosol sensors (shown in Figure 22). Fluorescence particles with potential bio-threat spectral characteristics are sorted and collected for further off-line analysis by scanning electron microscopy and dispersive Xray spectroscopy. Details of the system are discussed in sections 9, 10, and 11.
5.4 Traditional Lamp Light Source The most common light sources for spectro-fluorometers and absorption spectrophotometers are traditional lamps, especially the Xenon lamp that provides a quasi-continuous light output from 250 to 700 nm with some small sharp lines around 450 nm and beyond 800 nm. The wavelength dependent output of these lamps is a major reason for distortion of fluorescence excitation spectrum, which generally needs careful calibration. Mercury lamps can supply even higher intensity output than Xenon lamps with high power concentrated at a few sharp lines, the most powerful at 253.7 nm and a few around 300 nm; all suitable for bio-aerosol excitation. The Mercury-Xenon lamp has higher UV output than the Xenon lamp. Other lamps such as Quartz-Tungsten halogen lamps have continuous output from visible to IR, close to that of blackbody radiation, which are not suitable for fluorescence excitation of bio-aerosols, but are good for the calibration of optical spectral systems. Lamp light sources are generally collected and collimated by a reflector and a group of lenses to deliver maximum light output to the objective. In most case, a monochromater or narrow band filter is used for narrow-band output selection. Compared with a laser source, lamps are much harder and
102
Y.-L. Pan et al.
more cumbersome for delivering light output to a small spot. Assuming 20% of the light output can be focused to a 1 mm spot, the conversion efficiency of the corresponding lamp is about 2%. If a 50 nm bandwidth is needed for fluorescence excitation with power flux 24 W/cm2, then the lamp power must be higher than (24W/cm2)×(1mm2)/{20%×2%×[50 nm/(800nm-250nm)]}= 660 W. Generally, traditional lamps can provide this level of power, but 98% of the electronic power is converted into heat. If cost is an issue, pulsed lamps can supply more energy in a smaller and less expensive package than lasers for fluorescence excitation. Until a very inexpensive, compact, high efficiency light source is commercially available, the traditional lamp is still an affordable alternative for fluorescence-based bio-aerosol sensors for both military and civilian applications.
Figure 23. Image of a bio-aerosol sensor using two xenon lamps at different UV wavelengths as the excitation sources, approximately (26x22x28) cm in size (Kaye et al., 2005).
A low-cost multi-channel aerosol fluorescence sensor using two pulsed xenon flash lamps with different wavelengths as the excitation light sources have been developed by Paul Kaye et al. (shown in Figure 23). The two xenon lamps (Hamamatsu) have a small physical size (10 x 4 x 3.5cm) and can be externally-triggered with a maximum discharge energy of 40 mJ, flash duration ~1 µs, and a 126 Hz maximum repetition rate, set by the thermal dissipation limit of 5 W. By using such a setup, the two xenon lamps are sequentially triggered upon the detection of the arriving aerosol particle. The two triggered UV pulses at ~280 nm and ~370 nm have fluences in excess of ~300 µJ/cm2, which is optimal for excitation of
Bio-aerosol Fluorescence
103
bio-fluorophores tryptophan and NADH. For each excitation wavelength, the fluorescence in two bands (320 – 600 nm and 420 – 600 nm) are detected for each particle. Bio-aerosol particles are discriminated based on the relative intensity of the two fluorescence bands and by the elastic scattering intensity from the individual particles (Kaye et al., 2005).
5.5 Light Emitting Diode (LED) and Laser Diode The LED is a special type of semiconductor diode that emits incoherent narrow-spectrum light when electrically pumped. The wavelength of the light emitted depends on the bandgap energy of the materials forming the pn junction. There are LEDs with emission spanning from the deep UV to IR. Recently, a deep UV LED (260-365 nm) based on the materials aluminium gallium nitride (AlGaN) has been produced by Sensor Electronic Technology (SET) through the support of the DARPA Semiconductor Ultraviolet Optical Sources (SUVOS) program. This deep UV-LED provides an optical output to 0.5 mw at 20 mA, which introduces a new compact and robust semiconductor source for bio-aerosol sensing. In the visible wavelength region, typical LEDs are designed to operate at about 10 mW of electrical power with 30% conversion efficiency, which is much higher than most lasers and conventional lamps. Unlike most lasers and lamps, the LED has several advantages: inexpensive, very small size, highly robust, highly efficient, low heat generation, long life time, and fast response in a pulsed mode. The most severe disadvantage of LEDs is the power limitation imposed by heat dissipation. Hopefully, deep UV LEDs with much higher optical output will be realized in the near future. The laser diode is a special kind of laser using semiconductor materials as the working media, which is also similar to the LED formed from a p-n junction and powered by injected electrical current. It has the some advantages as a LED, but with a much smaller angle divergence of the emission, that greatly brings a number of experimental conveniences to the spectroscopist. Therefore, deep UV LEDs and ultimately UV laser diodes offer the best prospects for miniaturizing the fluorescence excitation from bio-aerosols in the near future. 5.5.1 Linear array of synchronously pulsed LEDs Under the support of the SUVOS program, several research groups have tried to observe the fluorescence from single bio-aerosol particles by using LEDs as excitation light sources (e.g. Yale/Brown, LL, MIT). Joint efforts
104
Y.-L. Pan et al.
by the Brown/Yale group demonstrated a compact system (Davitt et al., 2005; Pan et al., 2003c), incorporating a 32-element linear array of UV (290 nm and 340 nm) LEDs and a multi-anode photomultiplier tube, to obtain the fluorescence spectra from single aerosol particles on-the-fly. The 32-element linear array of UV LEDs, composed of individual elements 200 µm wide and 50 µm high, is illustrated in Figure 24. The emission from this LED linear array is focused onto a 3.2 mm long, 200 µm diameter column that overlaps the laminar aerosol stream. In order to expose each particle with maximum LED radiation, the LEDs are fired sequentially one-by-one and timed to follow the trajectories of aerosol particles in the stream. By
Figure 24. A typical spectral output of two UV-LEDs. Note that optical filters are needed to block the tail emission at the long wavelength side in order not to overwhelm the tryptophan and NADH emissions (Davitt et al., 2005).
accumulating the fluorescence photons of a single aerosol particle from 32 pulsed LEDs, a 100-times stronger signal was obtained compared to single LED illumination operating in CW mode. The emission of LEDs in the long-wavelength tail, as shown in the electroluminescence spectra of Figure 24 (a), is a potential problem for bio-aerosol detection systems. This tail can overlap the fluorescence bands of interest, particularly that of tryptophan. The successful demonstration of LED array-based fluorescence spectral detection from single bio-aerosol particles offers a practical stop-gap alternative to current large and expensive solid-state laser-based systems. Ultimately, UV laser diodes will be the silver bullet light sources.
Bio-aerosol Fluorescence
6.
105
DETECTORS FOR BIOFLUORESCENCE
The detector for fluorescence-spectra-based real-time, in-situ instruments must satisfy the following conditions: (1) high enough sensitivity to detect the dispersed fluorescence spectra from a single cell or spore; (2) sufficient spatial resolution in one-dimension for recording dispersed spectra; (3) linearity of dynamic range of the order of a thousand in order to cover the signal intensity variation for aerosols from 1 to 10 µm; (4) sufficiently fast frame capture rate to record and analyze thousands of particles per second; and (5) lower cost, more compact, lighter, and less power consumption compared to present systems. Presently the most used detectors are Charge Coupled Devices (CCDs), Intensified CCDs (ICCDs), linear arrays of avalanche photodiodes, and multi-anode photomultiplier tubes (PMTs). Linear arrays of photodiodes do not have the needed detectivity for bio-aerosol detection applications.
6.1 CCD and ICCD A Charge-Coupled Device (CCD) is a spatially resolved photoncounting device that can take an image formed by the different light intensities incident on each pixel of the CCD chip. The chip generally consists of 1024 by 512 pixels forming a rectangular active area. When light is incident on the chip, it passes through a polysilicon layer, then a silicon dioxide layer, and eventually into a silicon substrate. Electrons are excited by the penetration of photons into the substrate, move into the conduction band then rapidly to the nearest electrode, and are collected by the potential well generated by the electrode. Pixels are read out by computer, and electrodes are cleared to allow another cycle accumulation of fresh electrons periodically after a certain time of exposure. The electrons from each pixel are proportional to the photons incident on the pixel. The CCD is an exceptional device for imaging applications because a) the CCD has a very high and broad quantum efficiency that can reach 80% with a spectral response from 200 nm to 1100 nm, b) excellent spatial resolution limited only by the pixel size, c) virtually no cross talk between adjacent pixels, d) wide linear dynamic range (generally 12 bits or better), e) low dark current, f) the ability to accumulate single photoelectrons over periods of hours, and g) with recent advances of InSb and HgCdTe technologies extended sensitivity into the IR. The long-wavelength sensitivity of CCDs may have particular applications in elastic scattering. The main weakness of the CCD for pulsed fluorescence detection is the large readout noise of a few photoelectrons, equivalent to a few
106
Y.-L. Pan et al.
photoelectrons per pixel. A second weakness is the slow readout rate requiring about 1 sec to read out the entire image. With an image intensifier in the front of a CCD detector, the resulting ICCD camera is a very sensitive gateable detector suitable for single-pulse detection. Photons incident on a photocathode generate photo-electrons with about 40% quantum efficiency. These photo-electrons are accelerated within the tube of a Micro Channel Plate (MCP). Secondary electron emission from the hollow tubes of the MCP multiplies the number of electrons with gain exceeding 1 million. The electrons are accelerated from the MCP tubes to a phosphorus screen that emits green photons. The CCD detector placed in back of the MCP intensifier detects these photons. Since the MCP can be gated with a high voltage pulse, the ICCD assembly becomes a gated detector, capable of being synchronous with the arrival of an aerosol particle as well as a laser pulse. The ICCD has a number of characteristics: a) single-photon sensitivity for pulsed events, b) higher cross-talk (lower spatial resolution) and lower dynamic range compared to CCD detectors, c) particular utility for sensing fluorescence signals with lifetimes longer than the laser pulses that are typically a few nanoseconds (Carranza et al., 2003; Denvir and Coates, 2002; Pinnick et al., 2004).
6.2 APD Linear Array An avalanche photodiode (APD) is a relatively new solid-state detector formed by a p-n junction. It behaves likes the photodiode PIN but with high internal gain and single-photon sensitivity. A typical APD consists of an absorption region and a multiplication region. The incident photons generate holes and electrons that are separated in the absorption region. One carrier is swept towards the multiplication region where it gets amplified by a high electric field by impact ionization. Gains of at least 100 are realized for silicon APDs, 10-40 for germanium and InGaAs APDs. Generally, the APD can be biased to reach a high gain above 104 for singlephoton detection and is capable of achieving very short (20-50 ps) resolving times (Campbell et al., 2004; Dautet et al., 1993). Current silicon technology allows for fabrication of avalanche photodiode (APD) arrays with high resolution and large numbers of pixels at affordable cost. Due to this fact, the CCD and PMT detectors may be obsolete in the future. The APD is considered to be the ideal detector in fluorescence-based bio-aerosol detection instruments, providing a combination of high speed on the order of ns, high sensitivity on the order
Bio-aerosol Fluorescence
107
of a single photon, good QE around 40%, high gain that can reach 106, and even a broad spectral response from 300 nm to 1700 nm with different materials and structures. The nonlinearity of magnification for different light incidence at a certain bias voltage (gain), dark current and multiplication noise hold back some uses of APDs, especially for the fluorescence spectral detection of bio-aerosols that need a wide linear dynamic range covering a signal intensity range over 1000.
6.3 PMT Linear Array and on-Board Processor The photomultiplier tube (PMT) was developed several decades earlier than the semiconductor-based detectors. A PMT is a vacuum tube that consists of an input window, photocathode, electron multiplier of dynode stages, and anode. Photons incident on the photocathode generate photoelectrons that are emitted into the vacuum and focused onto the first dynode. These electrons are amplified repeatedly by the second and successive dynodes. The amplified electrons from the last dynode eventually are collected by the anode for signal output that is proportional to the incident light intensity. It has a very fast response time, on the order of sub-ns, single-photon sensitivity, high QE of around 30%, high gains that can be 106 or higher, a broad spectral response from 110 nm to 1200 nm, a wide linear dynamic range greater than 1000, low dark current and noise, a fairly affordable price, no water cooling needed, and is compact (Kume, 1994), so it has been used widely in bio-aerosol detection systems, especially for applications with few fluorescence bands associated with their scattering intensity. Recently, Hamamatsu developed a new PMT detector with multiple anodes like a linear array. This multi-anode PMT behaves like multiple PMTs within a single housing as shown in Figure 25. Each anode has its own output connected by a pin from the base of the PMT housing. The special design of the electron-multiplication section for the multi-anode makes this a unique detection unit. It consists of nine dynode stages for electron multiplication. The spatial integrity of photoelectrons that come from a particular location of the photocathode in the region around each anode makes the output from each anode for the corresponding pin. The inset of Figure 25 illustrates the schematic of this design. The photons detected on a particular photocathode are multiplied, and a gain of nearly 106 can be reached when 800V is applied, preserved in the right location, and read out of each of the corresponding pins of the 32 anodes. Experiments verify that such a multi-anode PMT assembly (Hamamatsu, Corp., H7260) has single-photon sensitivity, a linear dynamic range
108
Y.-L. Pan et al.
spanning more than three orders of magnitude, and less than 3% cross talk between adjacent anodes. Thus, the 32-anode PMT is ideally suited for fluorescent spectra detection. For bio-aerosol detection systems that are based on fluorescence spectra, the ICCDs are generally selected in laboratory instruments for its high spatial resolution with single-photon sensitivity. However, the slow datatransfer rate of couple tens of spectra frame per second limits their use within a real-time fluorescence detection system.
Figure 25. Illustration of the 32-anode PMT: a rectangular photocathode (1×3.5 cm), and 32output pins. Electron multiplication is achieved via a 9-stage-dynode unit that preserves the spatial integrity of the photoelectrons emitted along the x-axis of the photocathode.
Thus, this recently available multi-anode PMT assembly (Model H7260 with 32-anode) from Hamamatsu, Corp. holds promise for building on these recent advances for detection of bio-aerosol particles. Replacing the cooled ICCD with this 32-anode PMT as a key detector has dramatically increased
Bio-aerosol Fluorescence
109
the spectral recording speeds and reduced the physical size and price of the detection system, but still preserves enough resolution for the fluorescence spectra from bio-materials (32 bands in total). This detector has been the main detector in several instruments that are a clone of the Yale/ARL instrument. Based on fluorescence, the Yale/ARL systems can detect, discriminate, sort, and collect suspect bio-aerosol particles, one-by-one, onthe-fly, with single UV laser shots (Pan et al., 2004; Pan et al., 2001b). This system has been run continuously and autonomously for long periods at several field sites such as Applied Physics Laboratories, John Hopkins University and US Navy Research Laboratory. It has the ability to monitor up to 90,000 particles/sec using an on-board processor (developed by Vtech Engineering Corporation, see detail in section 10) and an optimized algorithm to discriminate threat-like particles based on UV LIF. This highspeed discriminator allows only the threat-like particles, which comprise only a very small fraction of all the particles queried via UV LIF, to continue into the second-stage identifier, minimizing background clutter and the possibility of contamination of the second-stage identification.
7.
TEST AEROSOLS FOR BIO-AEROSOL SENSORS
A number of simulants have been used as test aerosols for bio-aerosol sensors: Bacillus globigii (BG) for spores, Erwinia herbicola (Eh) for bacteria, ovalbumin (Ov) for toxin, and MS-2 for virus. Test aerosols of these simulants have been generated by a variety of techniques including a) spray atomization of liquid suspensions with commercial Collison, Sonotek, and Royco nebulizers, and the Ink Jet Aerosol Generator (IJAG) developed by Bottiger et al. (1998), and b) dry dispersal.
7.1 Collison and Sonotek Nebulizers Typical size distributions of test aerosols for BG generated by spray atomization with both Collison and Sonotek nebulizers at the DARPA Spectral Sensing of Bio-Aerosols (SSBA) trials conducted at the John’s Hopkins University Applied Physics Laboratory BSL2 test chamber are shown in Figure 26.
110
Y.-L. Pan et al.
Figure 26. Typical size distributions of a) BG (generated by Sonotek), b) BG (generated by Collison) measured with the Thermal Systems Inc, Aerodynamic Particle Sizer at the DARPA SSBA Round 3 trials conducted at the John’s Hopkins University Applied Physics Laboratory (17.5 cubic meter) test chamber Jan-Feb 2005. Both distributions contain what are believed to be small concentrations of unknown background aerosols. As can be seen from the distributions, the Sonotek produces larger particles (BG agglomerates) compared to the Collison nebulizer.
7.2 ECBC Dry Disperser for Aerosol Generation Similar simulant aerosol size distributions can be obtained with dry dissemination. A technique for dry dissemination has been developed by scientists at ECBC (Jerold R. Bottiger, Edward W. Stuebing, and Paul J. Deluca). Their handheld dry disperser is the ultimate in portable convenient aerosol sources, and ejects a small reproducible cloud of simulant aerosol particles at the push of a button. It can be used as a source of particles in the laboratory and as a confidence checker for equipment deployed in the
Bio-aerosol Fluorescence
111
field. The dry disperser is a metered dose inhaler that, when activated, releases a 60-mg spray of its contents into the air. A typical formulation is 0.1% to 1.0% by weight of simulant material in 1,1,1,2-tetrafluoroethane (HFA-134a) as the propellant. HFA-134a is an environmentally friendly replacement for R-12 Freon in refrigeration applications. The inhaler body is filled under pressure with liquified HFA-134a propellant and crimped shut with a cap that holds the metering valve. When the dry disperser is activated a measured volume of the contents is released to the atmosphere where it expands and evaporates instantly, leaving a small cloud of the loaded particles. The most popular dry disperser are loaded with BG spores or ovalbumin powder. Figure 27 shows a typical size distribution (measured with the TSI Model 3320 Aerodynamic Particle Sizer) of BG spores generated by the dry disperser. These data were obtained by dispersing five shots into a 2-foot diameter (118 liter) aluminum aerosol chamber.
Figure 27. (a) Number and (b) mass particle size distribution of BG aerosol particles generated by the ECBC dry disperser aerosol generator.
112
Y.-L. Pan et al.
As is evident from the distributions, a significant fraction of spores are found in clusters up to 3-4 micrometers aerodynamic diameter. The mass peak of the clusters is just over 2 micrometers diameter, corresponding to 10-15 spores per cluster. The number distribution shows a mode of submicron size particles not present in the original powder which is an oillike contaminant in either the propellant or the dry disperser canister body.
7.3 ECBC Ink Jet Aerosol Generator (IJAG) More monodisperse size distributions of aerosol simulants can be obtained with the IJAG developed at ECBC by Bottiger and collaborators (Bottiger et al., 1998). The IJAG adapts ink-jet-printer technology for the production of well controlled and well characterized aerosol samples. The IJAG comprises three components: the dispenser, the controller, and a computer, as shown in Figure 28. The cartridge is the 12-nozzle model used in old Hewlett Packard Thinkjet and Quietjet printers. Cartridges are purchased empty and loaded with a slurry (or solution) of water and the material of which particles are to be made, at a concentration appropriate for the final particle size desired. The droplets emitted by this cartridge are about 50 micrometers in diameter, so, for example, to obtain a dried residue particle 5 micrometers in diameter the concentration should be 0.1%. The loaded cartridge is mounted on the top of the dispenser, or head, and fires droplets downward through an oven that is the bulk of the dispenser. The oven is simply a brass tube wrapped with heat tape and held at 250 degrees Fahrenheit. It is contained within a white delrin tube, with the head mounted on top. As the droplets of slurry travel down the oven the water evaporates off and the non-volatile contents coagulate into compact and roughly spherical aggregates before reaching the bottom. Air pumped into the bottom of the dispenser is heated as it rises, and divides into two flows: most of the hot air leaves through the oven pipe, carrying the particles with it, while a smaller amount flows through an aperture into the head region and is returned to the pump. The counterflow of air is used to winnow out the small satellite particles that are formed when an ink-jet cartridge nozzle fires, resulting in a more uniform size distribution of particles at the outlet.
Bio-aerosol Fluorescence
113
Figure 28. Computer-controlled IJAG comprising dispenser with cartridge mounted on top, controller, and laptop computer.
Particles generated with a properly working IJAG have an approximate 30% standard deviation dispersion in particle size as suggested in the micrograph of the clusters formed by BG spores and 2.29 µm PLS spheres shown in Figures 29 (a) and (b), respectively. The uniformity of the droplets or particle size generated by the IJAG can be estimated by counting the number of primary particles included in a sampling of droplets. In Figure 29 (c) we show the results of counting the number Nsph of 2.29-µm psl spheres in each of 227 randomly selected splats of wet droplets. The fraction of the 227 clusters that contains Nsph spheres is plotted versus Nsph. The most probable number of spheres was found to be 18. Also plotted is the Poisson distribution probability of finding Nsph spheres in a cluster, given that the average Nsph is 18 (Holler et al., 2000). The IJAG produces relatively monodispersed particles compared to other aerosol generators.
114
Y.-L. Pan et al.
Figure 29. A typical sample of nominal 5-micrometer (a) BG clusters (formed by BG spores), (b) Polystyrene sphere clusters (formed by 2.29 µm PLS spheres) generated with the ECBC IJAG. (c) shows the uniformity of a typical distribution of particle size generated by the IJAG generator. It indicates the fraction of 227 clusters that contain the number of 2.29 µm PLS spheres within one cluster (Nsph) plotted versus Nsph. The solid curve represents a Poisson distribution, given that the most probable number of spheres is 18 (Holler et al., 2000).
8.
REVIEW OF INSTRUMENTATION
There are several biosensor systems based on UV light-induced fluorescence and elastic scattering. The most popular one might be the Biological Aerosol Warning System (BAWS) developed by MIT, Lincoln Laboratory. Several similar systems based on one or a few fluorescence bands with or without elastic scattering also have been developed such as the BioLert system by Pacific Scientific Instruments, Bio-Vigilant by BioVigilant Systems, Inc., Canadian Integrated Bio/Chemical Agent Detection System (CIBADS), the Fluorescence Aerodynamic Particle Sizer (FLAPS) by Defence R&D Canada Suffield, the Fluorescence Aerosol
Bio-aerosol Fluorescence
115
Shape Analysis System (FLASAS) by UK company BIRAL, University of Hertfordshire and Defence Science Technology Laboratories (Dstl), and the SPFA by the US Navy Research Laboratory (NRL). Other advanced prototype instruments include (WIBS) by University of Hertfordshire and Dstl, the NRL multi-wavelength excitation biosensor system, and the Single-Particle Fluorescence Spectrometer (SPFS) system by Yale University and US Army Research Laboratory that detects the dispersed fluorescence spectrum over an appreciable wavelength range. Typical systems are discussed in detail in the following sections.
8.1 NRL Biosensor via Multi-wavelength Excitation Both 355 and 266 nm laser excitation wavelengths have been developed to interrogate individual aerosols on-the-fly as a method of biological particle characterization (Agranovski et al., 2003; Eversole et al., 1999; Eversole et al., 2001; Hill et al., 1999; Pan et al., 2003b), but the relative capability of one wavelength or the other was not clear. The Naval Research Laboratory initiated a study in 2002 to evaluate the relative utility of the resulting fluorescence signatures from these laser sources by sequentially exciting single aerosols with both wavelengths. This was accomplished by aligning the output from two pulsed Nd:YAG lasers and a CW 780 nm diode laser to form collinear beams. Elastically scattered light at 780 nm from an aerosol transiting the focal volume was detected and used to trigger the Q-switch of the first pulsed laser, while a delayed trigger pulse was sent to the second laser Q-switch. The time difference between the two pulsed lasers was on the order of one microsecond, so that from the time of initiation by the particle entering the 780 nm beam, less than 2 µs elapse before both pulsed lasers fired. For typical aerosol velocities of a few meters/sec, the entire interrogation is completed in a travel distance of a several micrometers for the aerosol. The resulting laser induced fluorescence (LIF) from the aerosol was collected with an elliptical reflector and directed through a series of dichroic beam splitters to photomultiplier (PMT) detectors in roughly three spectral bands: 315-380 nm, 415-495 nm, and 495-605 nm. The incident intensities of all three lasers also were monitored for each particle. Initial studies were conducted using a wide range of materials generated as aerosols in the laboratory and fed into this table-top optical system. The details of this approach and results have been published (Sivaprakasam et al., 2004), and it was concluded that by utilizing information from both fluorescence excitation sources (266 nm and 355 nm), additional discrimination ability could be obtained among a set of biological aerosol classes.
Y.-L. Pan et al.
Fluorescence from 355 nm Excitation
116
Fluorescence from 266 nm Excitation
Figure 30. Shows scatter plots of individual aerosol fluorescence resulting from 266 nm versus 355 nm excitation. Four classes of aerosols are shown: ambient outdoor, and labgenerated kaolin, ovalbumin, and Bacillus atrophaeus spores (BG).
One of the technical challenges in achieving this capability was the design and fabrication of the electronic data acquisition system for the pairs of PMT output pulses that result from the fluorescence pulses of the two excitation lasers fired with a fixed temporal delay. Since the emission lifetime is on the order of a ns or less, each of the emission signals essentially follows the excitation pulse shape that is nominally 10 ns in width. To accurately characterize emission for a wide set of aerosol materials that can vary in size from one to ten microns, demanded high precision digital data acquisition capability that could achieve a 105 dynamic range in signal strength. Customized charge integrators were designed and fabricated (Vtech corporation) to meet these specifications using the 780 nm elastic scattering onset as a trigger. The minimum spacing that could be achieved in recording these PMT pulses was about 150 ns with integration windows of about 100 ns. After investigating laboratory-generated aerosols with this system, attention was shifted to characterizing ambient aerosols and air was brought into the system through a duct vented to the outside. Figure 30 shows a scatter-plot of the emission intensities of individual aerosols. For this plot the intensities in the various spectral bands were summed to provide a composite fluorescence intensity from 266 nm excitation versus 355 nm excitation. The darkest points are of ambient outdoor aerosols that cluster in the lower left quadrant of the plot defined by thresholds c1 and c2. These thresholds are arbitrary numbers, but are useful in conveying the relative frequency of occurrence of 85% of the ambient aerosols having
Bio-aerosol Fluorescence
117
Fluorescence from 355 nm Excitation
fluorescence intensities below both these values. Roughly 7% of the ambient aerosols had fluorescence higher than c1, but less than c2., and roughly 8% had fluorescence that exceeded c2 at this particular location and time. Aerosols consisting of Bacillus atrophaeus (BG) spores (dark grey), kaolin (light grey), and ovalbumin (medium grey) were generated in the laboratory using a Collison nebulizer that creates particles with a size distribution mode and median near 1 micron. One can see by inspection that if the only discriminant was 266 nm excited emission, then the overlap between the ambient and BG aerosols would be significantly increased: Imagine all the points collapsed to a horizontal line. Having an additional data dimension along 355 nm improves the ability to differentiate the population of BG aerosols from the ambient. The kaolin particles exhibit the lowest fluorescence values due to either excitation source, and are completely contained in the lower left quadrant defined by c1 and c2. Ovalbumin shows significantly higher 266 nm excited emission than all other categories, but with less 355 nm emission that BG particles on average.
Yersinia Outdoor Air
Diesel Exhaust BG Ovalbumin
Kaolin
Fluorescence from 266 nm Excitation
Figure 31. Shows scatter plots of individual aerosol fluorescence resulting from 266 nm versus 355 nm excitation. As in Fig. 30, the ambient outdoor aerosols with the diesel soot particles present are shown with the darkest shade of dots. Laboratory-generated aerosols of other materials are shown in lighter shades of grey for contrast.
118
Y.-L. Pan et al.
A similar plot shows another set of ambient outdoor data in Figure 31. As with the aerosols shown in Fig. 30, the ambient aerosols (darkest spots) form a cluster with lower values of fluorescence with both excitation wavelengths. However, in this case, a diesel-powered truck was parked in proximity to the duct inlet, so that the aerosols pulled into the instrument on this day were a combination of diesel engine exhaust and ambient air. One can see that there is a population labeled diesel exhaust that has significantly higher 266 nm emission values. This population of particles disappeared when the truck engine was switched off. Also, as before, several subpopulations of laboratory sample aerosols have been superimposed on the ambient aerosol plots as lighter shades of grey. The class of kaolin particles remains in the lowest region of fluorescent intensities, while the ovalbumin aerosols again show the highest 266 nm emission intensities. BG particles and Yersinia rohdei (vegetative bacterial cells in media) show up on the boundary between the non-diesel and diesel ambient particle clusters. The ability to discriminate among the classes of BG, Yersinia, and ambient diesel is highly dependent on the presence of both excitation dimensions. All of the aerosol clusters tend to show a roughly ellipsoidal shape with a major to minor axis aspect ratio of 2 to 3, which is tilted upward at an angle. This is likely largely due to the increase of fluorescence cross section with increasing size. Laboratory data on polystyrene latex spheres (with and without fluorescent dyes) shows a power dependence of fluorescence emission cross section with aerosol size in the range of 2.5 to 2.7.
8.2 WIBS1 Researchers at the University of Hertfordshire and Defence Science and Technology Laboratory in the UK examined the use of xenon discharge tubes in a prototype bio-aerosol monitoring system, WIBS1 (Kaye et al., 2004). WIBS1 was developed with the self-imposed constraint of using only low-cost COTS (commercial off-the-shelf) components. In addition, it was required to utilize two fluorescence excitation wavebands and two fluorescence emission detection wavebands tuned to the tryptophan and NADH emission spectra to facilitate bio-aerosol discrimination. Miniature xenon discharges tubes, similar to those found in the ubiquitous disposable camera, are readily commercially available. As optical sources, they suffer the limitations of pulsed output, low pulse repetition rate relative to pulsed lasers, and essentially isotropic emission. Nevertheless, compared to solidstate harmonic lasers they are very cheap and relatively efficient producers of UV radiation in the required 260-290 nm and 340-380 nm wavebands.
Bio-aerosol Fluorescence
119
WIBS1 used two Perkin Elmer RSL3100 series xenon modules because of their low cost, low power requirement (2W), small size, (8 x 4 x 2.5 cm approx.), and built-in quasi-collimation optics. They operated with a maximum flash energy (electrical) of 40 mJ and a flash duration of ~1µs. Although their maximum repetition rate was 50 Hz, operation at 2 - 3 Hz would achieve a useable xenon lifetime (quoted at 108 flashes) of at least 12 months continuous operation. 8.2.1 Sensor Configuration Figure 32 shows the schematic of the WIBS1 in which the two xenon sources were mounted on either side of a central scattering chamber. Orthogonal to the axis of the xenon sources were two fluorescence detector channels, FL1 and FL2, incorporating apertured spherical glass mirrors (Edmund Optics T43-470) to reflect fluorescence light through appropriate optical filters and onto the photocathodes of two miniature photomultiplier (PMT) detector modules (Hamamatsu H6779).
Figure 32. Plan view of WIBS1 sensor layout.
In operation, the aerosol flow of 10 l/min is drawn from the ambient environment through the central chamber of the sensor by a small electrical fan (not shown). No filtering or ensheathing of the aerosol flow takes place in order to minimise power requirements. The xenon sources fire alternately at 2 Hz frequency, each optically filtered to the required spectral band (centred on 280 nm and 370 nm respectively). Each is configured to
120
Y.-L. Pan et al.
provide a near-collimated beam of circular cross-section ~10mm diameter at the chamber mid-point, thus illuminating approximately 4 ml of aerosol with a UV fluence of ~ 15 – 20 µJ/cm2. Fluorescent light emanating from the particles contained within this volume passes through optical filters such that the fluorescence detector channels 1 and 2 record light in bands 320600 nm and 420-600 nm respectively. The higher band covers the majority of the NADH emission spectrum, whilst the difference between the two bands represents the peak of the tryptophan emission. Despite its simplicity, WIBS1 proved capable of reasonably good levels of aerosol discrimination. For example, Figure 33 shows the value of the difference in fluorescence detector channels (FL1-FL2) for Xenon 1 excitation (~280nm) as a function of the FL2 response under Xenon 2 excitation (~370nm) for four different aerosol types. For these, albeit pure, aerosol types, differentiation is achieved by this simple comparison of FL1 and FL2 responses. Clearly, environmental aerosols would present a greater challenge. Nevertheless, the low unit cost of this type of sensor would allow its deployment in potentially very large networked arrays, and the limited discrimination capabilities of each individual sensor would be offset by the ability to collate aerosol fluorescence data over a spatially wide area.
(FL1-FL2) @ 280nm excitation (arb. units)
700 600
Ovalbumen
500
Paper Mulberry
400
NADH 300
BG dry
200 100 0 -200
0
200
400
600
800
1000
1200
1400
1600
-100 -200
FL2 @ 370nm excitation (arb. units)
Figure 33. Aerosol discrimination performance of WIBS1 sensor.
1800
Bio-aerosol Fluorescence
121
8.3 WIBS2 Sensor The most significant limitation of the WIBS1 sensor arises from the fact that it records intrinsic fluorescence data from an ensemble of particles simultaneously rather than from a succession of particles individually. This reduces its ability to detect potentially low concentrations of particles of interest where these form part of a much larger aerosol population. The WIBS2 sensor (Kaye et al., 2005) was designed to overcome this by providing single-particle fluorescence detection whilst retaining the lowcost benefits of the xenon sources used in WIBS1. WIBS2, shown in Figure 34, employs a central optical chamber around which are arranged a continuous-wave 660 nm diode laser used in the detection and sizing of particles, two pulsed xenon UV sources emitting at different wavebands, and two fluorescence detection channels, FL1 and FL2, detecting intrinsic particle fluorescence across two wavebands. Thus, for each particle, a 2x2 excitation-emission matrix is recorded along with an estimate of particle size. Aerosol flow Scattering volume Xenon 1 source
cw laser beam
Fluorescence emission capture
Xenon 2 source
Figure 34. Schematic layout of WIBS2 sensor head.
Aerosol is drawn from the ambient atmosphere via a laminar-flow delivery system which presents suspended particles in essentially single file as they traverse the focused beam from the diode laser at the centre of the chamber. Total aerosol flow is 4.0 l min-1, of which ~ 3.6 l min-1 is filtered
122
Y.-L. Pan et al.
before being re-introduced to form a sheath flow confining the remaining sample flow. Each particle entering the scattering volume produces a scattered light signal from whose magnitude an estimate of particle (spherical equivalent) size may be determined using Mie theory. Particles deemed greater than ~1 µm in size initiate the sequential firing of the two xenon UV sources Xe1 and Xe2 to excite particle fluorescence. The mean value of UV fluence experienced by the particle is approximately 310 µJ/cm from each xenon source. WIBS2 uses two Hamamatsu L9455 xenon modules (Hamamatsu Photonics K.K., Japan). These are compact (10 x 4 x 3.5 cm), externally triggerable modules which incorporate a precision xenon discharge tube with an arc length of 1.5 mm. The maximum repetition rate for the xenon modules is 126 Hz, set by their thermal dissipation limit of 5 W. With its current sample flow-rate of 0.4 l min-1, WIBS2 is therefore capable of exciting fluorescence in all particles for particle concentrations up to ~2 x 104 particles/l. Beyond this concentration, the additional particles are counted and sized only. The xenon module outputs are optically filtered to the primary absorption bands of tryptophan and NADH (ie: ~280 nm and 370 nm respectively). The intrinsic fluorescence emission from the particle is collected, as in WIBS1, via detection channels FL1 and FL2 that collect fluorescent light over the wavebands ~310-600 nm and ~420-600 nm respectively. When the particle enters the cw laser beam, the FL2 channel detects the elastically scattered light signal. This ‘Particle Detect’ pulse initiates the firing of the two xenon sources. Xe1 (280 nm) fires some 10 µs after particle detection, producing fluorescence signals at channels FL1 and FL2 respectively. Twenty microseconds after this, Xe2 (370 nm) fires and channel FL2 records the resulting fluorescence emission from the particle. The total particle measurement time is approximately 30 µs, after which the system is ready to detect a further particle. However, since the xenon sources require approximately 5 ms to re-charge, any particles passing through the sensing volume during this 5 ms period are sized and recorded, but no fluorescence data is acquired. Thus, for each particle (up to 126 particles per second), a 2-dimensional fluorescence excitation-emission matrix is recorded together with an estimate of particle size. Figure 35 shows typical WIBS2 data recorded from a variety of test biological and non-biological aerosols. The x-axis ‘Scatter’ is in arbitrary units and corresponds to a particle size range from ~0.5 to 10 µm. The yaxis shows the ‘FL2_370’ fluorescence signal, again in arbitrary units, whilst the z-axis ‘FL1_280 / FL2_280’ is a dimensionless quantity that is primarily a function of the particle material alone, reflecting as it does the
Bio-aerosol Fluorescence
123
spectral distribution of the fluorescence emission rather than its absolute magnitude. Figure 35 shows results from aerosols of the following: washed BG spores (Bacillus atrophaeus, a simulant for B. anthracis spores); dry dispersed non-viable BG spores; washed and unwashed E. coli (Escherichia coli) vegetative cells; 0.1 mmol solutions of tryptophan and NADH (both in 1% sucrose solution); a 1% solution of ovalbumen in water; 3 µm latex spheres; Tonic Water (1% in distilled water); and 1.7 µm fluorescent latex spheres. With the exception of the dry BG spores which were aerosolised using filtered compressed air, each material was nebulised from liquid suspension.
FL1 280 / FL2 280
10
10
10
3µm PSL 1.7µm fluorescent PSL BG spores (dry) BG spores (washed) NADH (0.1mmol) Tryptophan (0.1mmol) E. coli veg. (unwashed) E. coli veg. (washed) Ovalbumen (1%) Tonic water (1%)
1
0
-1
-2
10 -2 10
10 10
-1
-1
10
0
10
10
1
10
0
2
10
3
10
1
Size (µm)
FL2 370
Figure 35. WIBS2 data recorded from both biological and non-biological aerosols.
These results illustrate the ability of the WIBS2 sensor to differentiate between various biological and non-biological airborne particles down to ~1 µm in size, even to the point of discriminating between washed and unwashed cells (a result of residual growth media on the unwashed cells that tends to fluoresce at longer wavelengths), and thus demonstrate the potential of xenon sources for low-cost field-based LIF bio-aerosol detection.
124
Y.-L. Pan et al.
8.4 FLASAS FLASAS, or Fluorescence Aerosol Shape Analysis System, was an instrument developed by the Defence Science and Technology Laboratory in the UK. It exploited previous work (chapter 3) that illustrated the potential of classifying particles on the basis of the variation in azimuthal scattering, measured by just three symmetrically placed detectors. FLASAS uses a combination of shape measurement with laser induced fluorescence to achieve real-time detection of biological particles, as shown in Fig. 36.
Figure 36. FLASAS bio-aerosol monitor.
Individual particles are detected and sized by scattering light from the red diode laser beam (635 nm) to the forward scattering PMT detector. This signal is then used to trigger the pulsed firing of a frequency doubled Compass 501QM-VD 532nm that irradiates the particle with UV at 266 nm. An ellipsoidal mirror reflects both the elastically scattered and fluorescent light from the particle to a dichroic filter that separates the two. The elastically scattered light then falls onto a multi-pixel hybrid photodiode detector, HPD (having pixels arranged in annular rings of 24, 6, and 1 respectively; see Ch.3), whilst the fluorescence magnitude in the 300-500 nm band is recorded by a photomultiplier tube detector. The former provides an ‘image’ of the spatial scattering from which the particle shape classification is derived. The total intensity of the collected light, in
Bio-aerosol Fluorescence
125
combination with the forward-scattered light, is used to estimate particle size. The acquisition electronics process the 31 outputs from the HPD device, the elastic scatter and fluorescence data. The UV laser power is monitored by a photodiode at the end of its path through the chamber to allow subsequent correction of pulse power variations. The time of flight of each particle through the red laser beam also is measured and used to help reject occurrences of particle coincidence in the beam. The collected data can be displayed in a variety of ways to assist in real-time monitoring. The software facilities enable archiving of data for subsequent analysis, and allow data to be analysed in real time using artificial intelligence techniques. Figure 37 shows electron-micrographs of a) spherical and b) fibrous sample particles and their corresponding polar plots of the scattered intensity signals received by the outer 24 pixel-annulus of the HPD detector. Clear differences can be seen in the cylindrical symmetry of the scatter for spheres and fibres.
Figure 37. SEM of a) 5um PSL spheres and polar intensity plots of the signal output of the outer 24 HPD pixels, and b) caffeine fibers and corresponding polar plots.
Tests with commercial polystyrene latex beads (PSL) highlighted that different manufacturers’ materials fluoresce at different levels, and that variation in fluorescence occurs even between batches from the same supplier (as they are supplied in suspension with surfactants). Since the instrument response is dependent on the material as well as particle size, droplets of oleic acid, generated using a TSI model 3450 vibrating orifice aerosol generator, were instead used as a standard material for instrument characterisation (Figure 38).
126
Y.-L. Pan et al.
Figure 38. FLASAS fluorescence from oleic acid droplets of increasing size.
The FLASAS instrument has been tested during a number of outdoor field trials, collecting data from the ambient background as well as from a large number of deliberately generated aerosols. While a trained observer watching the displays of the instrument could detect changes in the measured aerosol against a clean background, an automated recognition of unusual events is required for most potential applications of the instrument. Neural networks are well known for their pattern recognition capabilities and have therefore been extensively tested with the instrument data. A supervised Learning Vector Quantisation (LVQ) method has been implemented on the FLASAS system. The LVQ consists of 3 layers of processing elements or nodes with weighted links between nodes. The nodes in the input layer correspond to selected values of data from each particle, including raw fluorescence photomultiplier and HPD element data, time of flight, and processed size, asymmetry and normalised fluorescence data. The nodes in the output layer represent the different particle classes upon which the network was trained. The network was trained on laboratory data from many well-controlled particle samples split into 5 classes, covering bacteria, pollen, smokes, spheres and non-spheres. Data from a training set was presented to the input of the network, while the correct answer was given to the output. The weights of the hidden, middle layer were then adjusted until similar input data gave the correct class in the output. This back-propagation method was repeated for all the training data, and once complete, the network weights were frozen. The network was then tested on separate test data files. For each input particle of known class, the network output assigned a class and the known and generated classes then could be compared to give performance statistics for each network trained.
Bio-aerosol Fluorescence
127
The network was optimised on laboratory data and then tested on data collected during field trials. Figure 39 shows the particle classifications resulting from a release of dilute Erwinia herbicola (EH) in a salt buffer. It shows that the neural network can correctly classify the EH as bacteria, underneath the majority of dried salt particles that were classified as nonspheres and smoke (due to their small size). The instrument classification system also has been used to process data collected during blind trials, where the instrument sampled for 96 hours while aerosols were generated at unknown times, to identify when the instrument was being challenged with biological aerosols or other interferent aerosols. The resulting processed graphs (Figure 40) show clearly that although biological challenges were often generated at the same time as potential interferents, the classification process is able to place individual particles into the appropriate class. 8
Classification of Particles 50 per. Mov. Avg. (Pollen) 50 per. Mov. Avg. (Bacteria)
6
50 per. Mov. Avg. (Smokes) 50 per. Mov. Avg. (NonSphere) 50 per. Mov. Avg. (Sphere)
4
2
0
13:55:00
14:05:00
14:15:00
Figure 39. Neural network output classification from an EH and salt release.
While the early work concentrated on using Neural network methods, a number of alternative automatic classification techniques could be used (see section 10). For example, Support Vector Machines (SVM) has shown even better classification success than Neural Networks when using the same data.
128
Y.-L. Pan et al.
8.5 AFS, Aerosol Fluorescence Sensor The Aerosol Fluorescence Spectrometer (Hirst et al., 2004) was a development of FLASAS (section 8.4) and sought to combine spatial light scattering analysis with particle fluorescence spectrometry to help maximise the degree of confidence with which biological aerosols could be discriminated within an ambient environment. Several groups have demonstrated successful measurement of fluorescent spectra from single particles and have used this technique, usually in combination with particle size measurement, to aid particle discrimination on the basis of fluorescence (for example, Nachman et al., 1996; Chen et al., 1996; Pan et al., 1999 and 2003). The AFS was the first instrument to simultaneously record full spectral fluorescence data with spatial elastic-scatter data. It was also unique in using a novel quasicontinuous wave UV fiber-laser (Stratophase Ltd., Romsey, UK) operating at 266nm to elicit both data types, as illustrated in Figure 40.
Figure 40. Neural network output classification from a blind trial, showing 3 bacterial challenges against 2 interferent releases.
In operation, ambient aerosol is drawn into the instrument at a rate of ~5 l min-1 and is configured in the conventional way using filtered sheath air to produce a sample flow diameter to approximately 0.5 mm as it passes through the laser beam. The laser itself is, strictly speaking, a pulsed laser, though the pulse frequency (~80 MHz) means that it may be considered quasi-continuous in respect of the particle illumination, each particle being
Bio-aerosol Fluorescence
129
illuminated by up to 400 pulses. The average beam power at approximately 266 nm wavelength is variable up to ~100 mW maximum, and the beam is circularly polarised by virtue of a quarter-wave plate positioned appropriately in the beam path. Being almost as compact as a conventional frequency-quadrupled Nd-YAG laser of similar output power, the fiber laser is small enough to be suitable for use in field equipment. Each individual particle passing through the scattering volume results in radiation scattered in the forward direction up to 30° from the beam axis, being imaged onto a hybrid photodiode (HPD) detector as used by the FLASAS instrument.
Figure 41. Schematic diagram of the AFS instrument showing spatial scatter and fluorescence data paths.
Intrinsic fluorescence from the illuminated particle is reflected by an angled ellipsoidal mirror, as shown in Figure 41, via a UV blocking filter at 295 nm (WG-295, CVI Laser Corp., Albuquerque, NM, USA) into a UV transmitting optical fiber bundle of circular cross-section and 5 mm diameter. The exit end of this fiber bundle is rectangular in cross-section so as to maximize light delivery into the 8 mm high entrance slit of a concave
130
Y.-L. Pan et al.
grating spectrometer (Jovin Yvon model CP140-1602; 285-715nm grating). At the focal plane of the spectrometer, the resulting spectrum falls onto a 16-anode linear array photomultiplier tube (Hamamatsu R5900U-L16) which covers the 300 – 570 nm part of the spectrum. The use of such a photomultiplier for fluorescence spectroscopy has been successfully demonstrated by Pan et al. (2001) who employed a 32-channel device in the acquisition of transient spectra from particles in flow.
Figure 42. AFS data display screen corresponding to 1 µm PSL microsphere.
Figure 42 shows the real-time AFS display screen corresponding to the passage of a 1 µm polystyrene latex microsphere through the UV laser beam. The upper HPD ‘Scattering’ graph shows left-most the inner-annulus signal magnitude (first column) proportional to particle size, followed by three values of the middle-annulus detectors (grouped as three pairs), and the 24 outer-annulus values. For a perfect sphere, these 24 values should be equal; the slight variations seen in the Figure are due to both signal noise and minor variations in individual detector element gains within the HPD. The latter are removed in subsequent data processing. The lower ‘Fluorescence’ histogram shows the characteristic intrinsic fluorescence spectrum recorded from the PSL particle, with peak emission at ~350 nm.
Bio-aerosol Fluorescence
131
Figure 43. AFS data display screen corresponding to a vertically orientated gypsum fiber.
This PSL microsphere signature may be contrasted with that shown in Figure 43 which shows the AFS output for a single gypsum (calcium sulphate) fiber. In this case, the upper HPD spatial scattering graph exhibits two distinct peaks in the outer-ring detectors resulting from the classic horizontal scattering produced by the vertically-orientated fiber. Gypsum is essentially non-fluorescent, and therefore the lower histogram shows only minor fluctuations, again resulting from signal noise. The development and testing of the AFS is continuing, though early results confirm that the potential benefits of combining spatial scatter and intrinsic fluorescence measurements will be realised in improved particle discrimination performance.
9.
AMBIENT AEROSOL FLUORESCENCE MEASUREMENTS
During the last decade a number of research groups have constructed detector prototypes to measure the LIF of individual aerosol particles in one or two broadband wavelength channels (Eversole et al., 1999; Eversole et al., 2001; Hairston et al., 1997; Ho, 2002; Kaye et al., 2000; Pinnick et al., 2004; Reyes et al., 1999; Seaver et al., 1999). More advanced prototypes to excite fluorescence at two wavelengths (Kaye et al., 2005; Sivaprakasam
132
Y.-L. Pan et al.
et al., 2004), and to detect the dispersed fluorescence spectrum over an appreciable wavelength range (Hill et al., 1999; Pan et al., 2003b) also have been constructed. None of these efforts focused on measurement of natural background aerosol. Recently Pinnick et al. (2004) have described and operated a Fluorescence Particle Spectrometer (FPS) for sampling atmospheric particles that occur at far lower concentrations than in laboratory air or prepared test aerosols in field environments. This latter work appears to be the first focused effort to apply the UV-LIF technique to ambient aerosols in the atmosphere. A schematic diagram of the FPS for atmospheric aerosols is illustrated in Figure 44.
Figure 44. Schematic of the Fluorescence Particle Spectrometer. Aerosol is sampled through a virtual impactor-focusing nozzle inlet, which concentrates supermicron particles into a flowing laminar jet. Single particles within the aerosol jet are probed on-the-fly with a pulsed UV laser, exciting fluorescence within the particles. Fluorescence emission is collected by a reflective objective, focused onto a spectrograph slit, and detected with a gated CCD camera. This arrangement permits rapid measurement of single particle fluorescence spectra, with aerosol sample rates of a few liter per min for supermicron particles.
Bio-aerosol Fluorescence
133
Since the concentration of super micrometer-size particles in the ambient atmosphere falls off rapidly with increasing particle size, a virtual impactor is used to concentrate the aerosol into a laminar jet so that an effective sample rate of a few liter/min over a large dynamic range of particle size can be achieved. Triggering of the UV probe laser is achieved by detecting elastically scattered light from two crossed diode lasers (635 and 670 nm) when a particle is in the desired sampling volume. Single particles within the aerosol jet are probed on-the-fly with a pulsed UV laser (266nm), exciting fluorescence within the particles. In addition, the ICCD is gated open only when the UV laser fires. Fluorescence emission is collected by a reflective objective, focused onto a spectrograph slit, and detected with a gated ICCD camera (or multi-anode PMT as depicted). With this approach, expensive UV laser energy is used efficiently, as the laser is only pulsed when a targeted particle is in the sample volume, permitting rapid measurement of single particle fluorescence spectra (from about 295 to 605nm), undiluted and uncontaminated by background. A number of other smaller but nevertheless important details also have been addressed by Pinnick et al. (2004) such as uniformity of sample illumination, accurate determination of effective sample rate, net sampling efficiency as a function of particle size, and demonstration with test laboratory aerosols.
9.1 Previous FPS Results Data was taken with the FPS in 16 data runs lasting 15-40 minutes each over a two month period (March-April) at Adelphi Maryland, USA. About 15,000 useable spectra were obtained, which were analyzed using a defined hierarchical cluster grouping analysis. The result showed that almost all of the spectra could be fit into 8 relatively distinct groups. An example of this grouping analysis is shown in Figure 45, which illustrates spectra for a 26 minute period as well as the “template” spectra (derived through formal methods) for that group as determined from the complete data set of all 16 collection periods. As can be seen, the spectra appear to fit surprisingly well into the 8 groups. We view this as quite a remarkable result and if confirmed under more general temporal and spatial circumstances, it would represent a significant advance in the capabilities of FPS for the study of organic carbon aerosols in the ambient atmosphere. As discussed by Pinnick et al. (2004) the cluster classes themselves suggest possible composition of the particles (see labels in Figure 45) but at this time, with limited experience and lacking supplemental information such identifications, are not definitive.
134
Y.-L. Pan et al. 0.8 0.6
0.6
6 spectra cluster 1
0.4
183 spectra cluster 2
0.4 0.2
0.2
0
0 300
400
500
300
600
400
500
600
0.5
0.6
0.4
78 spectra cluster 3
0.4
60 spectra cluster 4
0.3 0.2
0.2
0.1 0
0 300
400
500
600
0.6
300
400
500
600
0.8 243 spectra cluster 5
0.4
0.6
2 spectra cluster 6
0.4 0.2
0.2 0
0 300
400
500
300
600 0.5
0.5 10 spectra cluster 7
0.4 0.3
0.4
400
500
600
482 spectra, cluster 8
0.3
0.2
0.2
0.1
0.1 0
0 300
500 400 wavelength (nm)
600
300
500 400 wavelength (nm)
600
Figure 45. Cluster analysis of atmospheric single particle spectra measured March 28, 2003 from 12:52-13:18 at Adelphi, Maryland, USA. Of the 10,000 total spectra measured, 1127 have fluorescence signals exceeding a noise threshold and are used in the analysis. Shown are template spectra (dotted lines) and normalized fluorescence spectra (solid lines) that are similar to the template spectra. The number of spectra in each cluster type is shown with each cluster. The possible materials for each cluster from 1 to 8 are (1) single-ring aromatics, (2) single, double -ring aromatics, PAHs, (3) bacteria plus HULIS, (4) bacteria, double -ring aromatics, (5) marine aerosol, bacteria, (6) unkown, (7) cellulose, NADH, and (8) Humic acids, HULIS.
10. SPECTRAL DATA ANALYSIS AND COMPARISONS TECHNIQUES The fluorescence spectra based bio-aerosol sensors need not only detect the BW agent particles with highest sensitivity and probability, but also need an optimized algorithm to distinguish the BW agent particles from
Bio-aerosol Fluorescence
135
other interferent and background aerosol particles based on UV-LIF rapidly, easily, and clearly. This requires smart and fast spectral data analysis and comparison techniques to do this job. Current spectral analysis algorithms can be summarized into three methodology groups: (1) BW agent particle is determined by its signal location in two-dimensional or three-dimensional coordinates where the coordinates are the fluorescence intensity, ratio of fluorescence intensity from UV and visible bands, or the size and shape information from elastic scattering of the detected aerosol particles; (2) BW-agent particle is determined by its fast fluorescence spectral profile match through the relative intensity comparison between different wavelengths (or multiple bands) associated with its fluorescence quantum efficiency; and (3) BW agent particle is determined by principle component analysis (PCA), hierarchical cluster analysis, discriminant function analysis (DFA), or least-square fit etc., numerically multivariate analysis and classification.
10.1 Special Location of Biothreats in a Coordinate Most of the real-time bio-aerosol detection instruments, especially those with one or two fluorescence spectral bands with or without one elastic scattering band, use their signal ratio to discriminate BW agent particles from other aerosol particles. Typical works are those done by Pinnick and Hill et al. at Army Research Laboratory (Pinnick et al., 1995), Eversole et al. at Navy Research Laboratory (Eversole et al., 1999; Eversole et al., 2001; Seaver et al., 1999; Sivaprakasam et al., 2004), Hairston et al. at Canadian Department of Defense (Hairston et al., 1997), Jeys et al. at Lincoln Laboratory, Mass. Institute of Technology (LL, MIT, (Reyes et al., 1999), and Kaye et al. at University of Hertfordshire, UK (Kaye et al., 2000; Kaye et al., 2005). Figure 46 shows spatial dimensions of specific particle classes, including BW agents (Cox et al., 1995). Therefore, the location of BW agent particles in a two-dimensional coordinate of size and fluorescence intensity is used to discriminate between particle classes. Generally, aerosol size is obtained by aerodynamic sizers or scattering sizers, and the fluorescence intensity is with or without size normalization to include the physical properties of fluorescence quantum yields and cross section (Faris et al., 1997; Kunnil et al., 2005a; Weichert et al., 2002). Kaye et al., 2005 use particle shape information including an asymmetry factor (Af) and normalized fluorescence intensity to get even better classification.
136
Y.-L. Pan et al.
Figure 46. The size range of the most related aerosol particles with bio-agents.
Figure 47 shows the fluorescence normalized by scatter intensity as a function of particle asymmetrical factor. The scattered intensity here is taken as the mean around the whole 360° of an azimuthal ring of detectors, and this yields a nonlinear function related to the scattering cross section of the particle which can itself be correlated with a spherical-equivalent particle diameter by Mie theory. The Af is a measure of the variation of the scattered light intensity around the azimuthal ring that relates to the particle asymmetrical degree of the shape (see Ch.3). Looking at the ratio of fluorescence intensity to scattering intensity is a simplistic approach to normalizing the fluorescence signal for particle volume but nevertheless it provides an effective means of data separation, which has been used in the middle 1900s (Pinnick et al., 1995). As can be seen from Fig. 11.2 (Kaye et al., 2000), the fluorescent 1.7-µm PSL is, as would be expected, clearly differentiated from the non-fluorescent PSL varieties, having an order of magnitude greater fluorescence to scattered intensity ratio. Importantly, the BG spores and gypsum have clearly distinguished themselves from the other limited types of aerosol particles.
Bio-aerosol Fluorescence
137
Figure 47. The fluorescence intensity normalized by scattering intensity versus asymmetry factor of several kinds of aerosol particles (Kaye et al., 2000).
The BAWS system developed by LL, MIT is presently considered as the standard system for a fast fluorescence-based bio-aerosol trigger, it has been used in various laboratories and field test as a reference system. BAWs uses a similar principle as the above algorithm for bio-aerosol discrimination. It is based on the ratio of two fluorescence bands that improve the accuracy of assignment for bio-aerosols, although it still struggles with a high false alarm rate. The early BAWS system detects the entire fluorescence emission by two PMTs for two fluorescence bands, the UV (300–400 nm) and visible (400–600 nm) respectively, when excited by a 266 nm laser (Reyes et al., 1999). The UV band is sensitive to biofluorescence from the amino acid tryptophan, while the visible band is sensitive to bio-fluorescence from NADH and flavin compounds. With the use of an aerosol chamber in the laboratory, the sensor was exposed to many different kinds of aerosols for the purpose of establishing aerosol fluorescence spectral signatures. The normalized difference [(UVVis)/(UV+vis)] of the coincident signals in the UV PMT and the visible PMT, on a particle-by-particle basis, has proven that the threat simulants have fairly good distinguishable signatures from typical natural background aerosols. Even so, there still are many interferent aerosol particles whose signals are located at the same coordinate as the bio-agents to cause false alarm. Now, LL, MIT is developing the Super BAWS system with
138
Y.-L. Pan et al.
fluorescence bands extended to 16 channels similar to that of the Yale University/ARL team (described in the next section) to improve the discriminating ability for bio-threats. Currently there are several systems available that incorporate similar ideas for data analysis as the pioneer works discussed above. The BioLert system from Pacific Scientific Instruments is based on the UV-LIF around 450 nm (the NADH fluorescence from bio-aerosols) and elastic scattering with the laser diode at 370 nm. The new system is currently configured to excite aerosols at two wavelengths 370 nm and 410 nm, with plans to use 280 nm and 340 nm diode lasers for the future. BioVigilant has developed a bio-viability verification detector also like the Biological Aerosol Warning System (BAWS), and Canadian Fluorescence Aerodynamic Particle Sizer (FLAPS) system by combining the information from UV fluorescence and scattering particle sizer. The British FLASAS, system by which the airborne particles are optically analyzed in real-time using laser scattering and fluorescence, characterizes particles in terms of size, shape, fluorescence, and concentration. It is believed that this system has better classification ability than the other fluorescence band systems such as BioLert.
10.2 Spectral Profile Match through Multiple Fluorescence Bands The systems described in the previous section that utilize just one or two fluorescence bands for detection with or without a scattering channel can cause high false alarm rate. Researchers try to get higher accuracy discrimination using fluorescence-based instruments either through higher resolution fluorescence spectroscopy (Chen et al., 1996; Cheng et al., 1999; Gray et al., 1998; Hill et al., 1996; Hill et al., 1995; Hill et al., 1999; Nachman et al., 1996; Pan et al., 2003b; Pan et al., 1999; Pan et al., 2001b; Schroder et al., 1999) or through multiple excitations (Kaye et al., 2005; Sivaprakasam et al., 2004). Discrimination ability can be improved greatly by detecting the dispersed spectra. Figure 48 shows the average single-shot fluorescence spectra excited by a 263 nm laser from several typical 5 µm single-particle, on-the-fly. A fast and smart data analysis algorithm is required to extract the information from the spectral data. The Yale/ARL team developed such an algorithm by matching the fluorescence spectra through the relative intensity comparison between different wavelengths (divided into multiple bands) associated with its fluorescence quantum efficiency.
Bio-aerosol Fluorescence
139
Figure 48. Single-shot fluorescence spectra from 5 µm single particles excited by a 263 nm laser
As mentioned previously (Frain et al., 2006; Pan et al., 2003b; Pan et al., 2004; Pan et al., 2001b), this real-time, in-situ bio-aerosol detection and discrimination system based on single-shot UV-LIF dispersed spectra from single individual aerosol particles on-the-fly consists of the following key elements (see schematic diagram in Figure 49): (1) A concentrator that draws air at about 300 liters/min and is based on the virtual impact principle. The concentrator directs most of the particles in the 1 to 10 µm size range toward a slower speed pump end that exit about 1 liter/min. (2) By using a specially designed nozzle, the particles are forced to flow in a straight trajectory, localized within a cylindrical area of 600 µm in diameter for over a distance of at least 1 cm. (3) The exiting particles are aligned to flow through the intersecting volume (referred to as the trigger volume) of two diode laser beams with different wavelengths. Only if the particles travel through the intersection of these beams will the elastic scattered signal be present at the two PMTs and an AND gate output be issued. (4)
140
Y.-L. Pan et al.
This output serves as a trigger to the UV laser (263 nm, the 4th-harmonic of YLF), which is synchronized to illuminate the detected particle from the trigger volume flying through the sample volume defined as the intersection of the UV-laser beam and the focal point of the fluorescence collection lens. When this particle is irradiated with UV radiation, the UV-LIF spectrum is dispersed by a compact spectrograph covering a wavelength span of 250 nm to 700 nm. This wavelength span of 450 nm is aligned with the detector elements from a 32 anode-PMT. (5) Every aerosol particle that transits through the trigger volume and subsequently transits though the sample volume is irradiated. The resulting fluorescence spectrum for each of these particles is captured and analyzed by the readout and processing electronics (PhotoniQ-QEM), designed and built by Vtech Engineering Corporation. (6) The on-board processor determines whether or not a particle has the characteristic spectrum of known bio-threat aerosols. If a particle matches pre-determined signature criteria, the electronics trigger the air puffer to blast out a puff of air, which then deflects that particle for further analysis. A brief description on the operation of the readout and processing electronics, analysis and comparison of the spectral data is as follows. Figure 50 (a) shows the electronics are actually composed of two boards. The smaller board (3" x 2.5") interfaces to the multi-channel detector, the 32-anode PMT. The larger board (5" x 7") contains the analog processing electronics, the analog-to-digital converters, the input and output triggering circuitry, various instrument control I/O circuitry, and the digital signal processing (DSP) circuits. The two boards are connected with a shielded cable. A daughter-card (not shown) containing a high voltage power supply for biasing the PMT can be mounted to the larger board. The HV bias on this daughter-card is controlled by the DSP within the PhotoniQ. All of these electronics are powered from a single +5V, 9 Watt power supply. The PhotoniQ electronics receive an incoming digital trigger signal for each particle in the trigger volume. The internal triggering circuitry then adds a variable delay (about 2 µs) from the trigger that corresponds with the time necessary for the laser output to reach the particle sample volume as well as the time for the particle to travel to this spot. The trigger circuitry initiates a corresponding boxcar integration cycle to collect the fluorescence spectral signal while the particle is within the sample volume. The 32 integrated signals are then further processed in the analog domain before being digitized by onboard analog-to-digital converters (ADCs). This resulting digital “vector” sample represents the particle’s spectral information. The vector is further processed digitally by the DSP. Finally, a spectral match algorithm is performed within the PhotoniQ’s DSP to determine if the particle matches a stored particle signature and whether it should be discriminated.
Bio-aerosol Fluorescence
141
Figure 49. Schematic diagram of the real-time, in-situ single-shot, single-particle fluorescence spectrum based bio-aerosol discriminating, sorting, and collecting system developed by Yale/ARL.a
The PhotoniQ operates as a stand-alone readout and processing system with the 32-anode PMT; however, the digital vector samples from each particle can be output to a PC using a high speed parallel interface. In this mode of operation, the data from particles that match the signature criteria are marked in the log file for continued analysis and algorithm verification. In addition to data logging, a LabView interface allows the user to control the various operational parameters of the PhotoniQ such as trigger settings, integration times, and particle algorithm/discrimination settings.
142
Y.-L. Pan et al.
Figure 50. (a) PhotoniQ-OEM readout and processing electronics system for a multiplechannel detector (Vtech Engineering Corporations); (b) The averaged fluorescence spectra from 100 aerosol particles of Bacillus subtilis vegetative cells (solid) and humic-like ambient particles (dashed) associated with eight defined fluorescence bands for sorting aerosol particles that is characterized as Bacillus subtilis vegetative cells.
Figure 50 (b) shows how eight fluorescence bands (B1, B2, …B8) are defined to optimize the differentiation ability and speed for sorting aerosol particles of Bacillus subtilis (BG) vegetative cells. The two averaged fluorescence spectra are from 100 aerosol particles of BG vegetative cells (solid) and humic-like ambient particles (dashed) used as a database for characterizing BG vegetative cells and humic-like ambient particles respectively. The sharp peak at 685 nm is due to elastic scattering of the particle with a diode laser. Its intensity is proportional to the particle volume that can be used to normalize the fluorescence signal, and is related to the fluorescence quantum efficiency and provides an important vector for discrimination, especially for these materials that have similar spectral profiles, but different quantum efficiencies. Table 2 shows the intensity ratios between the different spectral bands. To sort the fluorescence spectra similar to BG with 1.5 times standard deviations, 8 flags are defined in Table 3. Table 2. Intensity ratios of the different fluorescence bands for BG.
Bands Ratio
B1/B8 0.1
B2/B8 2.3
B3/B8 4.2
B4/B8 2.5
B5/B8 0.9
B6/B8 0.3
B7/B8 0.4
Bio-aerosol Fluorescence
143 Table 3. Flags for BG.
#
1
2
3
4
5
6
7
8
Flag
B3 > 3 .8 B8
B3 < 4 .6 B8
B2 > 20 B1
B3 > 1 .6 B2
B3 > 1 .4 B4
B4 > 2 .3 B5
B5 > 2.4 B6
B5 <2 B7
The particle will be considered to be BG vegetative cells and sorted only if only the detected UV-LIF spectrum satisfy all the criteria in Table 3. If the particle meets the algorithm’s criteria, a digital signal is output from the PhotoniQ to actuate the puffer. Meanwhile, all the spectra are saved and the concentration of matched particles is counted. Once it reaches the preset threshold, a warning alarm for a matched potential bio-threat is given. As all the processes are finished within the on-board PhotoniQ chip, and no talk with the computer is needed, the system can detect fluorescence spectra and discriminate particles with throughput speed as high as 90,000 particles per second. Because of the re-programmable nature of the DSP, many different algorithmic detection methods can be supported using the same hardware, according to the specialty of different bio-agents; e.g., to sort the humic-like aerosol particles, different spectral bands should be defined to optimize the analysis algorithm.
Figure 51. 500 successive single-shot fluorescence spectra measured from ambient aerosol particles (NRL on Aug. 18, 2004) illuminated by the 263 nm Q-switched Nd:YLF laser (50 µJ/pulse). Any detected ambient aerosol particles that have bio-threat-like fluorescence spectra are deflected and collected for further analysis.
144
Y.-L. Pan et al.
Based on this data analysis and comparison algorithm, the subsequent device has been successfully tested in various indoor and outdoor field trial situations. The technological advancements have been transitioned to the Defense Advanced Research Project Agency (DARPA), and one of the systems is currently operating in the San Francisco International Airport facility by Sandia-Livermore National Laboratories for calibrating other fluorescence based biosensors. Figure 51 shows 500 successive fluorescence spectra from ambient aerosol particles illuminated by the 263 nm Q-switched Nd:YLF laser. Among them, 19 particles have been assigned to be bio-threat-like aerosols and were deflected and collected, all of them have a strong fluorescence peak around 330 nm. And the other 481 particles with their corresponding fluorescence spectra did not meet the criteria conditions and have been considered as non-suspect aerosols and traveled in straight trajectories without deflection. The corresponding collections of the deflected aerosols with different criteria definitions of fluorescence spectra are under analysis by Fourier Transform InfraRed (FTIR) spectroscopy. In the blind test at Applied Physics Laboratories, Johns Hopkins University for the DARPA Spectral Sensor for Biological Agent program, this system has been used as the front-end to prescreen aerosol particles. This system collected the suspect particles on a substrate for further analysis by visible Raman, UV Raman, and fluorescence imaging. Results reveal that the biological simulants were recognized and identified with much higher detection probability and less false alarm rate with the use of the preselection of biothreat-like particles than the direct deposit of all aerosol particles without any sorting process. We select two typical datasets to illustrate our procedures used to quantify how well the algorithm that is based on spectral band profile matching can reduce the background overlap with that of BW agents. The first dataset (named BGS) was collected from a 5-minute release of pure BG aerosol particles that was generated by a Sonotek nebulizer at middle concentration level (~25 CFU/L) in SSBA test bed and consists of 27,354 fluorescence spectra. The second dataset (named MDAIR) was from a 30 minute direct sampling of the ambient air on June 17, 04:47 am - 05:17 am, 2004 at APL, Johns Hopkins University in Baltimore, Maryland and consists of 48,505 fluorescence spectra. We reduced the dimensional spectral space into either two bands or 8 bands. The 8 bands were defined as above for the spectral band profile-matching algorithm. The two bands cover the wavelength range spanned by the 32-anode PMT where channels 5-9 cover the 300 – 400 nm (UV) range and channels 10-19 cover the 400 – 600 nm (VIS) range.
Bio-aerosol Fluorescence
145
By analyzing the BGS dataset, we find that 81% of the particles (22,217 out of 27,354 spectra) have a fluorescence spectral peak in Band 3, channel 7 of the PMT. Note that the tryptophan peak at 350 nm lies within this band. This spectral profile is distinguishable if the fluorescence intensity at Band 3 is above 0.010 pC (~5 photons). Based on these BG spectra, the UV fluorescence band to VIS fluorescence band has a fixed ratio of 1.970 ± 0.354 standard deviation (STD). The scatter plot in Figure 52 (a) shows 92.4% of the data within the 22,217 particles (black dots) are located within the range of UV/VIS ratio equal to 1.970 ±1.5 times STD. The remaining data (gray dots) are either out of this ratio range or the spectral intensity in Band 3 is below 0.010 pC (around the left-bottom corner). The same BGS dataset is analyzed to provide the mean and standard deviation of the ratios between the 8 spectral bands as those defined in Figure 51. The results are presented in a two-dimensional scatter plot as shown in Figure 52 (b). This plot indicates that 90.6% of the spectra (in black color symbols) satisfy the 8 flags conditions of Table 3 with 1.5 times the STD. The two-band and 8-flag algorithms show little difference. The latter approach is a little wider than the UV/VIS ratio distribution.
Figure 52. Scatter plot of fluorescence spectra from Bacillus Globigii (BG) in a reduced two-dimensional space. (a) Spectra discriminated by the ratio of UV band to VIS band (1.970 ± 0.354×1.5). (b) Spectra discriminated by spectral profile match 8 flags through 8 fluorescence bands.
When we use the discriminating standard deduced from the BGS dataset to classify the fluorescence spectra from the ambient-aerosol-particle dataset (MDAIR) a large difference is found. As shown in Figure 53(a) 30.8 % of the particles have a UV/VIS ratio of BG (1.970 ±1.5 × SDV), which is one of the main reasons biosensors based on one or two
146
Y.-L. Pan et al.
fluorescence spectral bands have a high false alarm rate. The gray dots in Figure 53(a) represent the UV/VIS ratio deviations outside the 1.5 times STD due to the complex compositions of the ambient aerosol particles. The spectral band profile-matching algorithm shows a much better discrimination ability. The scatter plot in Figure 53(b) indicates that only 2.0 % of the particles in this MDAIR dataset satisfy the criteria when using the 8-flags approach. There are many fluorescence spectra from the ambient particles located within the correct UV/VIS ratio range of BG particles but many fewer particles that also match the spectral profile of BG. The 8-flags approach with the 32-anode PMT analysis is also comparable to the data analysis of the fluorescence spectra taken at Adelphi, MD by a 1024channel ICCD detector (Pinnick et al., 2004). The 1024 channels are grouped into 40 binned bands and treated by a hierachical clustering method. About 1% of the particles meet the criteria for look alike BG particles. The output of this 32-anode PMT provides fluorescence profile information to enable high-speed analysis in quasi real-time for good discrimination among particles.
Figure 53. Scatter plot of fluorescence spectra from ambient aerosols in a reduced twodimensional space. The aerosols were obtained during a 30 minute direct sampling of the ambient air on June 17, 04:47 am - 05:17 am, 2004 at APL, Johns Hopkins University in Baltimore, Maryland, USA. (a) Spectra discriminated by the ratio of UV band to VIS band (1.970 ± 0.354×1.5). (b) Spectra discriminated by spectral profile match (8 flags) through 8 fluorescence bands.
10.3 Numerically Multivariate Analysis and Classification In order to achieve a good balance between speed and accuracy, the above method that matches spectral profiles through multiple fluorescence bands and quantum efficiency is, in practice, one of the best algorithms for
Bio-aerosol Fluorescence
147
bio-aerosol discrimination. Theoretically, the principal component analysis (PCA) of (Giana et al., 2003; Holler et al., 2004; Tjarnhage et al., 2001; Ye and Auner, 2004), hierarchical cluster analysis (Pinnick et al., 2004), discriminant function analysis (DFA) of Jarvis et al. (2006), Fourier transfer analysis (Fisher et al., 1998), and least-square fit analysis will exhibit better bio-aerosol classification ability if one can ignore the speed and suitability for real-time instrumentation. Principal components analysis (PCA) is a multivariate technique (Flury, 1988; Giana et al., 2003; Kunnil et al., 2005b; Tjarnhage et al., 2001; Ye and Auner, 2004) used to extract information from large sets of data by minimizing data redundancy, which reduces data dimensionality by keeping the most important data signatures. In PCA, a set of reduced variables obtained from a large matrix of correlated data are generated. This new orthogonal principal component data, which are eigenvectors of a covariance matrix of the original data set, is in fact a linear combination of the original variables. In general, only the most important principal components are kept, which reduces the dimensionality of the data by ignoring those components that do not contribute significantly to the information content. Even, just with the first two principal components, it represents 85% of the variation in the original data (Tjarnhage et al., 2001). The application of PCA turns out to work much better for recognizing large sets of high dimensional data in this new coordinate projection formed by orthogonal principal components, which indicates the data in terms of the differences and similarities between them. Eventually, each data set (e.g., a fluorescence spectrum) is represented as a point in this new multidimensional coordinate. Points with similar properties tend to cluster creating groups of high similarity. Consider a fluorescence spectra dataset from m aerosol particles with n data points (determined by the spectral resolution, it can be easily up to 1000 by a CCD or ICCD detector). This data will be represented by an m × n data points matrix. Before the application of PCA, the fluorescence intensity spectra from each particle should be transformed using a Standard Normal Variate (SNV) transformation to center the spectra with zero mean and normalize the spectra with standard deviation scale of one. n
xi ' k = ( xik − xi ) /(
∑ ( xi
k
− xi ) 2
k =1
( n − 1)
)
(14)
where, xi′k is the standard normal variate corresponding to the kth original fluorescence intensity xik (k = 1, 2, …, n) at a certain wavelength noticed
148
Y.-L. Pan et al.
numerically at k, and xi (i = 1, 2,…, m) is the ith fluorescence intensity data vector from the ith particle. xi is the mean value of fluorescence spectrum from the ith particle with n sample data points. Therefore, the n-dimensional AdjustedData matrix can be written as
x11 ' , x12 ' ,......, x1n ' x 21 ' , x 2 2 ' ,......, x 2 n ' AdjustedData(X) = ...... xm1 ' , xm2 ' ,......, xmn '
(15)
Based on the data set, an n × n covariance matrix can be generated as
cov( x1 , x1 ), cov( x1 , x 2 ),......, cov( x1 , x n ) cov( x 2 , x1 ), cov( x 2 , x 2 ),......, cov( x2 , xn ) cov(X ) = ...... cov( , ), cov( , ),......, cov( , ) x x x x x x n 1 n 2 n n
(16)
where each covariance element is determined by m
cov( xi , x j ) =
∑ ( xk
i
− xi )( xk j − x j )
k =1
( m − 1)
(17)
xi and x j are the mean values of fluorescence intensity at a particular wavelength observed at i and j from the mth particle. Thus, a set of n orthogonal eigenvectors can be found from the covariance matrix having the same dimensionality as the variables. The eigenvectors of the covariance matrix form the principal components, and the corresponding eigenvalues are used to rank them. The eigenvector with the highest eigenvalue is the first principal component of the data set. By ordering the eigenvectors according to their corresponding eigenvalues, highest to lowest, yields the principal components in order of significance. Eigenvector (X) = (eig1, eig2, ……, eign)
(18)
The principal components of less significance may be disregarded without significant loss of information. The reduced components form a new set of eigenvectors (components), called the FeatureVector set having less dimensionality than the original data set. FeatureVector (X) = (eig1, eig2, ……, eigl), l ≤ n
(19)
Bio-aerosol Fluorescence
149
Transpose the FeatureVector and AdjustedData matrix to get the corresponding transposed matrices like
eig1 eig 2 T FeatureVector (X) = and . eigl
x11 ' , x 2 1 ' ,......, xm1 ' x12 ' , x 2 2 ' ,......, xm 2 ' T AdjustedData (X) = ...... x1n ' , x 2 n ' ,......, xm n ' Finally, a new reduced data set is generated by the last step of PCA treatment, Newdata (X) = FeatureVectorT(X)×AdjustedDataT(X)
(20)
This new data is an m (row) × l (column) matrix that has reduced the dimensionality from n to l with only the terms of the eigenvectors kept having the most importance. The data in each column are the corresponding coordinates of the compressed fluorescence spectral information from each of the m different particles in the l-dimensional orthogonal space, generally l=3 is large enough to include greater than 99% of the information. This analysis method has been successfully used to classify fluorescence spectra from different bio-aerosols into different cluster groups. In addition to PCA, other multivariate analysis techniques have been employed for bio-aerosol classification. Researchers even have tried to use Fourier transformations (Fisher et al., 1998), time-resolved (Brahma et al., 1985), and two-wavelength pump-probe depletion fluorescence spectroscopy (Courvoisier et al., 2005) to get better discrimination of the biological materials with similar fluorescence spectra.
11. HIGH-DISCRIMINATION METHODOLOGIES As we discussed in the above sections, detecting threat bacteria among the wide variety of background aerosols is analogous to looking for a needle in the haystack. The present available optical systems have been developed to distinguish bio-aerosols from non-bio-aerosols based on fluorescence
150
Y.-L. Pan et al.
(Chen et al., 1996; Cheng et al., 1999; Eversole et al., 1999; Eversole et al., 2001; Hill et al., 1996; Hill et al., 1995; Hill et al., 1999; Leblanc and Dufour, 2002; Nachman et al., 1996; Pan et al., 2003b; Pan et al., 1999; Pinnick et al., 1998; Schroder et al., 1999; Sivaprakasam et al., 2004) and/or elastic scattering (Holler et al., 1998; Kaye et al., 1991; Kaye et al., 2000; Pan et al., 2003a; Wyatt et al., 1988) from individual single aerosol particles. These instruments can run continuously, in situ and in real time to provide rapid warning/alarm for the existence of a few potentially lifethreatening BW-agent aerosol particles in the midst of a vast number of non-biological particles. The current instruments are capable of analyzing a few hundred liters of sample air per minute at different environmental conditions and locations (e.g., subway stops, mines, and airports). Since the early stages of field trials, bio-arosols were released and the potential harmful aerosols were generally the minority entrained within a majority of particles that are innocuous background aerosol (an example of this situation can be 1 bio-aerosol particle in 104 background aerosol particles per liter of air). However, the major flaw inherent in these instruments is the high frequency of false alarms, because the UV-visible fluorescence systems are incapable of distinguishing different molecules (such as tryptophan and diesel vapor or cigarette smoke), Furthermore, the fluorescence based instruments are not capable of identifying different kinds of harmless or harmful pathogenic micro-organisms. The currently available biochemical techniques, such as polymerase chain reaction (PCR) (Belgrader et al., 1999; Ho, 2002; Makino et al., 2001; Murray et al., 2003), fluorescence in situ hybridization (FISH) (Ho, 2002; Murray et al., 2003), antibiotic resistance determination (Pourahmadi et al., 2000), or chip matrix of biochemical microsensors (Hagleitner et al., 2001), can identify the genus and species of the bacteria or virus. However, after the alarm has sounded, these biochemical assay procedures are time consuming and require complicated sample preparation. Even the recently developed ultrafast B cell adaptive immune-based sensor (Rider et al., 2003) and realtime PCR (Belgrader et al., 1999) take at least several minutes, and the total assay requires more than half an hour (Rider et al., 2003). The optical methods are potentially faster and include Raman spectroscopy, mass spectroscopy, laser-induced-breakdown spectroscopy (LIBS), Fourier Transfer Infrared spectroscopy (FTIR). Some of these optical methodologies require pre-selection of bio-aerosols by the fluorescencecued first stage. What follows is a brief review of these high-discrimination techniques.
Bio-aerosol Fluorescence
151
11.1 High-discrimination or Identification Techniques for Bio-aerosol Particles The currently available high-discrimination or identification techniques for bio-aerosol particles mostly are biochemical assay-based or opticsbased, and the best identification method might be based on biochemical assay associated techniques. The Polymerase Chain Reaction (PCR) is one that works by producing a number of copies of a specific region of DNA in order that it can be tested. PCR has become one of the most widely used techniques in molecular biology. It is a rapid and simple means of producing relatively large numbers of copies of DNA molecules from minute quantities of source DNA material. It generally takes denaturization, hybridization, and DNA synthesis or replication, three steps to finish the whole process (Murray et al., 2003). Fluorescence in situ hybridization (FISH) is a method for visualizing the location of nucleotide sequences on chromosomes, nuclei, and even in the tissues. Generally, it requires the preparation of short sequences of singlestranded DNA, called probes, which are complementary to the DNA sequences to be examined. These probes hybridize, or bind, to the complementary DNA. They are labeled with fluorescent tags, which are bright enough to observe the location of those sequences of DNA. This specific match of the complementary DNA sequence ensures the highdiscriminating identification (Murray et al., 2003). The other chemical assay based method also uses fluorescent-tagged bio-recognition molecules such as antibodies. They selectively bind to a targeted threat particle for identification. This method can reach a rapid decision, without wash or incubation steps by the trade of specificity. However, the fluorescent-tagged antibody solution generally is very expensive, and the large quantity of background aerosol particles can clog the microfluidics channel and contaminate the reactions, if the system is not prepared for a large influx of particles. The preprocessing of the sample particles is always a key problem towards reducing the cost and getting accurate identification. For spectroscopic techniques, LIBS, which is also called laser-inducedplasma spectroscopy (LIPS) or laser-assisted spark spectroscopy (LASS), based on the emission spectroscopy of the investigating aerosol particle ablated into its atomic constituents uses a high-power pulsed laser. The particle is identified by looking at the specific atomic emission wavelength and relative intensity of the fingerprint atomic elements of the specific biological materials. The advantages of LIBS are the simplicity of experimental arrangement, and possible multi-element analyses that can be
152
Y.-L. Pan et al.
performed simultaneously without the need for sample preparation. LIBS has become renowned as a high-discrimination method for its in situ and real-time bio-aerosol detection. However, this technique has difficulty in discriminating between simple molecular compounds. It is also hard to classify individual complex biological aerosols at very low concentrations, as well as determine absolute concentrations. Self-absorption of emission, line broadening, and continuum background also affect the usage of LIBS (Beddows and Telle, 2005; Cheng et al., 1999; Hahn et al., 2003; Hettinger et al., 2006; Hybl et al., 2003; Martin et al., 1999). Laser-based bio-aerosol mass spectroscopy (Beddows and Telle, 2005; Ben-David and Ren, 2003; Chadha et al., 1993; Gittins et al., 1999; Jarvis et al., 2006; Manoharan et al., 1990; Manoharan et al., 1993; Nelson et al., 1992; Nelson et al., 2004; Sengupta et al., 2005; Storrie-Lombardi et al., 2001; Wu et al., 2001) is similar to LIBS in that it also uses the high power of a pulse laser to break the investigating aerosol particle into small negative or positive fragments via laser-induced desorption/ionization (not high enough to break into plasma form), then determine the contained compositions of the aerosol particle by detecting the spectral intensity via the mass/charge ratio of these ion fragments. Several laser-based singleparticle mass spectroscopy systems such as BAMS, RSMS, PALMS, RTAMS, ATOFMS, LAMPAS, and MALDI-MS have been used for realtime detection and characterization of individual bio-aerosol and atmospheric particles. Among the laser-based desorption/ionization methods, two approaches exist. One utilizes the inherent advantages of matrix-assisted laser desorption/ionization mass spectrometry as the MALDI-MS system to provide a soft ionization method for large biomolecules with the least amount of fragmentation for the detection of biomarkers with large molecular mass. Another like the BAMS system uses matrix-free laser desorption/ionization to eliminate the need for reagents and sample preparation, which can impede in situ real-time particle analysis. Mass spectroscopy also can produce enough signal intensity with a single laser-shot from single aerosol particle; however, the time it takes for collecting and treating large datasets with sufficiently high resolution currently limits both LIBS and mass spectroscopy from becoming a realtime, in-situ bio-aerosol discrimination tool. The pre-selection of suspect bio-aerosol becomes important for the real-field application. Systems like BAMS that use on-line pre-selection by comparing the ratio of the two fluorescence bands from UV and VIS greatly increase the system applicability. Raman and FTIR spectroscopy (Beddows and Telle, 2005; Harris et al., 2005; Russell et al., 2005; Steiner et al., 2005) are able to discriminate or identify aerosol particles by looking at the specific frequency of vibro-
Bio-aerosol Fluorescence
153
rotational structures of the ground state from specific molecules via scattering-frequency shifts or absorption of the excitation frequency. However, the signal from a single aerosol particle is so weak that it requires significant integrate time. The integration time can be a short pulse if there is a large enough sample of pure bio-aerosols on a suitable substrate; however, this greatly limits the applicability for using Raman and FTIR spectroscopy for real-time bio-aerosol detection. Because of the signal weakness of Raman and FTIR spectroscopic techniques as well as the specificity of biochemical assays, particle accumulation is used for getting sufficient S/N, but the large quantity of background aerosol particles could contaminate the reactions and disable the biochemical assays within minutes, or the few target biological particles may be overwhelmed by the presence of non-biological particles, in which case the intensity fluctuations in the spectral background of the vast number of non-bio-aerosol particles can mask the Raman and FTIR peaks from the few bio-aerosol particles. For various identification techniques to work rapidly and precisely, high purity and concentration of suspect bio-threat particles are preferred. Thus, it is essential to enrich the suspect particle concentration relative to the ambient aerosol concentration. Most particle-sorting instruments, such as the widely used flow cytometer-cell sorter, are based on electrostatic deflection cued by the particle’s electric resistance, absorption, fluorescent color, or scattering size (Asano et al., 1997; Melamed et al., 1990). Because the ambient aerosol particles have different dielectric constant, size, shape, and velocity, they enter the electrostatic sorting area with different time delay, charge/mass ratio, and momentum, and these factors increase the difficulty in sorting out the wanted particles with high efficiency and purity. The Yale/ARL team developed a rapid classification scheme based on fluorescence spectrum of the aerosol that provides the cue as to whether to deflect that particular aerosol particle by a puff of air (hence the name Puffer) emerging from a pulsed valve to sort and regroup by a particle aerodynamic localizer (PAL), and collect the potential bio-threat particles on different substrates or inside a small volume of liquid for further analysis or identification (Frain et al., 2006; Pan et al., 2004).
11.2 Preselection of Potential Biothreat Particles by Puffer and PAL The concept of the rapid aerodynamic sorting system cued on single particle UV LIF fluorescence spectrum is shown in Figure 54 (Frain et al., 2006; Pan et al., 2004). The Puffer, an electromagnetic actuated pulsed valve, can generate a very short (60 µs) supersonic air-packet (18 psi)
154
Y.-L. Pan et al.
within 20 µs. An aerodynamic nozzle, mounted on the exit of the valve, tightly focuses the air-packet down to 500 µm in diameter and allows it to propagate over distances exceeding several mm. Such a well localized airpacket only causes a very short disturbance of the main aerosol stream and therefore deflects only very few neighboring non-biological particles along with the sorted biological particle. Assuming that there is one biological particle entrained in 103 non-biological particles per liter of air, and as the system is capable of detecting and analyzing these particles one by one within 1 minute, then the pulsed air-packet deflects a fraction of 1 liter sample air (∆V/V) by the air packet, and
∆V /V = πR2 v (∆t) / V
(21)
here R=300 µm and v = 10 m/s is the radius and speed of the air stream, respectively, ∆t = 60 µs is the duration of the pulsed air-packet. Therefore, the total non-biological particles in the stream within one liter is C = 103, and the segment of puffed stream contains C × (∆V /V) = 1.7 x 10-4 nonbiological particles, which means within a 60 µs puff, the probability of deflecting a non-biological particles with the suspect biological particle is 1.7 x 10-4. Therefore, the concentration ratio of biological to non-biological particles within the deflected particles will be 1/1.7x10-4 = 6000. Comparing to the initial concentration ratio 1/103, the enrichment factor could be 6000/(1/103) = 6x106 under such conditions! Therefore, such a sorting and enriching process could greatly benefit biochemical identification. By monitoring trajectories of the particles disturbed by the puffer, this sample illuminates that the puffer can selectively deflect either of two neighboring particles only 500 µm apart. A mixture of two kinds of particles having equal volumes of ~8±2 µm diameter equivalent spheres, are obtained by combining the outputs of two IJAGs to demonstrate the sorting capability of the Puffer. Bacillus subtilis (BG) or tryptophan particles are chosen to be the target bio-aerosol in a minority concentration (15%) of the mixture. Arizona road dust (ARD), riboflavin (RBF), or sodium chlorine (NaCl) is used as the majority constituent containing 85% of the mixture. ARD is one of the standard background aerosols, RBF is a biological material but has a shape and fluorescence spectrum different from BG particles, and NaCl has a shape similar to the BG or tryptophan particles when they are generated by the IJAG. Figure 54 shows how the system performs for a wide variety of mixtures that combine different shapes of fluorescent particles (RBF and BG), different shapes of fluorescent and non-fluorescent particles (ARD and BG), or similar shaped and sized fluorescent and non-fluorescent particles (NaCl and Tryptophan). The fluorescent particles (BG, RBF, or tryptophan) are observed by their fluorescence images induced by UV lamp illumination, while ARD and NaCl particles are observed with additional
Bio-aerosol Fluorescence
155
white-light illumination. The microscopic images clearly indicate that the center of the undeflected aerosol pile (on left side of the Figure) is dominated by ARD, RBF, or NaCl aerosols, and 2 mm from the center along the direction of the puffed air-packet are mainly occupied by the deflected particles (BG or tryptophan, see right side of the Figure). These aerosols have been sorted based on their similar UV-LIF fingerprint, despite background aerosol fluorescence, or being similarly shaped as the BG particles. The counting results show that all enrichment factors are higher than 103.
Figure 54. Suspect bio-aerosols, bacillus subtilis (BG) or tryptophan (see the images at the right side) are sorted from different mixed groups of aerosols: riboflavin (RBF) and BG, Arizona road dust (ARD) and BG, or sodium chlorine (NaCl) and tryptophan. RBF, ARD, or NaCl (see the images at the left side) represent background aerosols that remain mainly undeflected.
156
Y.-L. Pan et al.
However, for rapid and sensitive identification, the collected particle number density per unit area on the substrate must get high enough to achieve a certain signal to noise ratio to permit the rapid analysis of the particles within a short signal integration time. For this reason, it is necessary to localize the deflected aerosol onto a smaller spot on the substrate, on the order of a few hundred micrometers in diameter. Unfortunately, the inertia of the gas-packet emitted by the Puffer is so strong (300 m/s from the exit of the puffer nozzle) that, although the pressure driving the puff ends abruptly, the gas-packet entrains surrounding gas and particles for milliseconds after the puff is triggered. The particles are launched on different trajectories depending on exactly when the particle interacted with the puffed air (Frain et al., 2006). Experiments have demonstrated that shorter gas pulses (60 µs) and lower gas pressure in the puffer nozzle (15 psi) help reduce the dispersion of the deflected particles. Even so, the deposits of 4 µm deflected particles on a horizontally oriented substrate are still highly scattered. In order to achieve a high density of purified and enriched potential bio-threat particles, a Particle Aerodynamic Localizer (PAL) apparatus, in the form of a focusing funnel, was introduced to the system and the substrate was oriented vertically to collect the deflected particles. In this configuration, a high puffer gas pressure and long gas pulse duration are required to ensure that the gas-packet or puff has enough residual momentum to push the particles through the PAL and deposit them onto the substrate, but not so much as to cause the particles to bounce off the substrate after impact. In addition, it is desirable to maintain a short duration to avoid deflecting unwanted non-biological particles in the main particle stream. The optimal conditions for this configuration are 18 psi puffer gas pressure and 100 µs pulse duration. A CFD simulation and a series of experiments have been performed to try to acquire the best conditions to achieve the highest delivering efficiency, largest enrichment factor, and smallest area in deposition. The rejection of particles from the funnel and bounced particles from the deposit substrate are the main losses for collection. Figure 55 shows the pattern of deflected particles deposited on the substrate, theoretically and experimentally. The pattern indicates most of the deposited particles are localized within a region less than 1 mm in diameter.
Bio-aerosol Fluorescence
157
Figure 55. CFD prediction and experimental test of the deposition pattern of deflected particles on a vertical substrate.
ACKNOWLEDGEMENT One of the authors (Y. L. Pan) particularly likes to acknowledge the editorial help of Dr. Sung-Wen Chang who transforms streams of words into smoother sentences. Her patience, expertise, and willingness to go over many versions of this chapter are greatly appreciated. In addition, Y. L. Pan wishes to thank Dr. Stephen Holler for his scientific and editorial helps in the writing of section 10.3.
REFERENCES Agranovski, V., Ristovski, K., Hargreaves, M., Blackall, P.J., and Morawska, L., 2003, Real-time measurement of bacterial aerosols with the UVAPS: Performance evaluation, J Aerosol Sci, 34: 301-317. Asano, K., Miyatake, Y., Yatsuzuka, K., and Higashyama, Y., 1997, The Eeffect of Particle Velocity on Electrostatic Particle Sorting, Journal of Electrostatic, 42: 17-23. Atkins, P., and Friedman, E. 2005. Molecular Quantum Mechanics (4th ed.). Oxford University press, Oxford New York. Beddows, D.C.S., and Telle, H.H., 2005, Prospects of real-time single-particle biological aerosol analysis: A comparison between laser-induced breakdown spectroscopy and aerosol time-of-flight mass spectrometry, Spectrochim Acta B, 60(7-8): 1040-1059. Belgrader, P., Benett, W., Hadley, D., Richards, J., Stratton, P., Mariella, R., and Milanovich, F., 1999, Infectious disease - PCR detection of bacteria in seven minutes, Science, 284(5413): 449-450.
158
Y.-L. Pan et al.
Ben-David, A., and Ren, H., 2003, Detection, identification, and estimation of biological aerosols and vapors with a Fourier-transform infrared spectrometer, Appl Optics, 42(24): 4887-4900. Bottiger, J.R., Deluca, P.J., Stuebing, E.W., and VanReenen, D.R., 1998, An Ink-Jet Aerosol Generator, J Aerosol Sci, 29(suppl. 1): s965-s966. Brahma, S.K., Baek, M.P., Gaskill, D., Force, R.K., Nelson, W.H., and Sperry, J., 1985, The Rapid Identification of Bacteria Using Time-Resolved Fluorescence and Fluorescence Excitation Spectral Methods, Appl Spectrosc, 39(5): 869-872. Campbell, J.C., Demiguel, S., Ma, F., Beck, A., Guo, X.Y., Wang, S.L., Zheng, X.G., X. w. Li, J.D.B., Kinch, M.A., Huntington, A., Coldren, L.A., Decober, J., and Tscherptner, N., 2004, Recent Advances in Avalanche Photodiodes, IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, 10(4): 777-787. Carranza, J.E., Gibb, E., Smith, B.W., Hahn, D.W., and Winefordner, J.D., 2003, Comparison of nonintensified and intensified CCD detectors for laser-induced breakdown spectroscopy, Appl Optics, 42(30): 6017-6021. Castanho, M.A.R.B., 2002, Teaching molecular applications of the particle-in-a-ring model using azulene, J Chem Educ, 79(9): 1092-1093. Chadha, S., Nelson, W.H., and Sperry, J.F., 1993, Ultraviolet Micro-Raman Spectrograph for the Detection of Small Numbers of Bacterial-Cells, Rev Sci Instrum, 64(11): 30883093. Chen, G., Nachman, P., Pinnick, R.G., Hill, S.C., and Chang, R.K., 1996, Conditional-firing aerosol-fluorescence spectrum analyzer for individual airborne particles with pulsed 266-nm laser excitation, Opt Lett, 21(16): 1307-1309. Cheng, Y.S., Barr, E.B., Fan, B.J., Hargis, P.J., Rader, D.J., O'Hern, T.J., Torczynski, J.R., Tisone, G.C., Preppernau, B.L., Young, S.A., and Radloff, R.J., 1999, Detection of bioaerosols using multiwavelength UV fluorescence spectroscopy, Aerosol Sci Tech, 30(2): 186-201. Courvoisier, F., Boutou, V., Wood, V., Bartelt, A., Roth, M., Rabitz, H., and Wolf, J.P., 2005, Femtosecond laser pulses distinguish bacteria from background urban aerosols, Appl Phys Lett, 87(6): -. Cox, C., Wathes, C., and edited 1995. Bio-aerosols handbook. Oxford University press, Oxford New York. Dalterio, R.A., Nelson, W.H., Britt, D., Sperry, J.F., Tanguay, J.F., and Suib, S.L., 1987, The Steady-State and Decay Characteristics of Primary Fluorescence from Live Bacteria, Appl Spectrosc, 41(2): 234-241. Dalterio, R.A., Nelson, W.H., Britt, D., Sperry, J., Psaras, D., Tanguay, J.F., and Suib, S.L., 1986, Steady-State and Decay Characteristics of Protein Tryptophan Fluorescence from Bacteria, Appl Spectrosc, 40(1): 86-90. Dautet, H., Deschamps, P., Dion, B., MacGregor, A.D., MacSween, D., McIntyre, R.J., Trottier, C., and Webb, P.P., 1993, Photon counting techniques with silicon avalanche photodiodes, Appl Optics, 32(21): 3895-3900. Davitt, K., Song, Y.K., Patterson, W.R., Nurmikko, A.V., Gherasimova, M., Han, J., Pan, Y.L., and Chang, R.K., 2005, 290 and 340 nm UV LED arrays for fluorescence detection from single airborne particles, Optics Express, 13(23): 9548-9555. Demtroder, W. 1982. Laser spectroscopy: Basic concepts and instrumentation. SpringerVerlag, Berlin Heidelberg New York. Denvir, D.J., and Coates, C.G., 2002, Electron-multiplying CCD technology: application to ultrasensitive detection of biomolecules, Proc. SPIE, Biomedical Nanotechnology Architectures and Applications, 4626: 502-512.
Bio-aerosol Fluorescence
159
Duarte, R.M.B.O., Pio, C.A., and Duarte, A.C., 2004, Synchronous scan and excitationemission matrix fluorescence spectroscopy of water-soluble organic compounds in atmospheric aerosols, J Atmos Chem, 48(2): 157-171. Eversole, J.D., Hardgrove, J.J., Cary, W.K., Choulas, D.P., and Seaver, M., 1999, Continuous, rapid biological aerosol detection with the use of UV fluorescence: Outdoor test results, Field Anal Chem Tech, 3(4-5): 249-259. Eversole, J.D., Cary, W.K., Scotto, C.S., Pierson, R., Spence, M., and Campillo, A.J., 2001, Continuous bio-aerosol monitoring using UV excitation fluorescence: Outdoor test results, Field Anal Chem Tech, 5(4): 205-212. Faris, G.W., Copeland, R.A., Mortelmans, K., and Bronk, B.V., 1997, Spectrally resolved absolute fluorescence cross sections for Bacillus spores, Appl Optics, 36(4): 958-967. Fisher, M., Bulatov, V., Hasson, S., and Schechter, I., 1998, Fast aerosol analysis by Fourier transform imaging fluorescence microscopy, Analytical Chemistry, 70(11): 2409-2414. Flury, B. 1988. Common principal components and related multivariate models. John Wiley & Sons, New York, Chichester, Brisbane, Toronto, Singapore. Frain, M., Schmidt, D.P., Pan, Y.L., and Chang, R.K., 2006, Selective deflection and localization of flowing aerosols onto a substrate, Aerosol Sci Tech, 40(3): 218-225. Giana, H.E., Silveira, L., Zangaro, R.A., and Pacheco, M.T.T., 2003, Rapid identification of bacterial species by fluorescence spectroscopy and classification through principal components analysis, Journal of Fluorescence, 13(6): 489-493. Gittins, C.M., Piper, L.G., Rawlins, W.T., Marinelli, W.J., Jensen, J.O., and Akinyemi, A.N., 1999, Passive and active standoff infrared detection of bio-aerosols, Field Anal Chem Tech, 3(4-5): 274-282. Gray, P.C., Shokair, I.R., Rosenthal, S.E., Tisone, G.C., Wagner, J.S., Rigdon, L.D., Siragusa, G.R., and Heinen, R.J., 1998, Distinguishability of biological material by use of ultraviolet multispectral fluorescence, Appl Optics, 37(25): 6037-6041. Hagleitner, C., Hierlemann, A., Lange, D., Kummer, A., Kerness, N., Brand, O., and Baltes, H., 2001, Smart single-chip gas sensor microsystem, Nature, 414(6861): 293-296. Hahn, D.W., Miziolek, A.W., and Palleschi, V., 2003, Laser-induced breakdown spectroscopy: an introduction to the feature issue, Appl Optics, 42(30): 5937-5937. Hairston, P.P., Ho, J., and Quant, F.R., 1997, Design of an instrument for real-time detection of bio-aerosols using simultaneous measurement of particle aerodynamic size and intrinsic fluorescence, J Aerosol Sci, 28(3): 471-482. Harris, W.A., Reilly, P.T.A., and Whitten, W.B., 2005, MALDI of individual biomoleculecontaining airborne particles in an ion trap mass spectrometer, Analytical Chemistry, 77(13): 4042-4050. Hettinger, B., Hohreiter, V., Swingle, M., and Hahn, D.W., 2006, Laser-induced breakdown spectroscopy for ambient air particulate monitoring: Correlation of total and speciated aerosol particle counts, Appl Spectrosc, 60(3): 237-245. Hill, S.C., Pinnick, R.G., and Nachman, P., 1996, Conditional-sampling spectrograph detection system for fluorescence measurements of individual airborne biological particles, Appl Optics, 35(7): 1069-1076. Hill, S.C., Pinnick, R.G., Nachman, P., Chen, G., Chang, R.K., Mayo, M.W., and Fernandez, G.L., 1995, Aerosol-Fluorescence Spectrum Analyzer - Real-Time Measurement of Emission-Spectra of Airborne Biological Particles, Appl Optics, 34(30): 7149-7155. Hill, S.C., Boutou, V., Yu, J., Ramstein, S., Wolf, J.P., Pan, Y.L., Holler, S., and Chang, R.K., 2000, Enhanced backward-directed multiphoton-excited fluorescence from dielectric microcavities, Phys Rev Lett, 85(1): 54-57.
160
Y.-L. Pan et al.
Hill, S.C., Pinnick, R.G., Niles, S., Fell, N.F., Pan, Y.L., Bottiger, J., Bronk, B.V., Holler, S., and Chang, R.K., 2002, Fluorescence from airborne microparticles: dependence on size, concentration of fluorophores, and illumination intensity (vol 40, pg 3005, 2001), Appl Optics, 41(21): 4432-4432. Hill, S.C., Pinnick, R.G., Niles, S., Pan, Y.L., Holler, S., Chang, R.K., Bottiger, J., Chen, B.T., Orr, C.S., and Feather, G., 1999, Real-time measurement of fluorescence spectra from single airborne biological particles, Field Anal Chem Tech, 3(4-5): 221-239. Ho, J., 2002, Future of biological aerosol detection, Anal Chim Acta, 457(1): 125-148. Holler, S., Pan, Y.L., Chang, R.K., Bottiger, J.R., Hill, S.C., and Hillis, D.B., 1998, Twodimensional angular optical scattering for the characterization of airborne microparticles, Opt Lett, 23(18): 1489-1491. Holler, S., Zomer, S., Crosta, G.F., Pan, Y.L., Chang, R.K., and Bottiger, J.R., 2004, Multivariate analysis and classification of two-dimensional angular optical scattering patterns from aggregates, Appl Optics, 43(33): 6198-6206. Holler, S., Auger, J.C., Stout, B., Pan, Y., Bottiger, J.R., Chang, R.K., and Videen, G., 2000, Observations and calculations of light scattering from clusters of spheres, Appl Optics, 39(36): 6873-6887. Hybl, J.D., Lithgow, G.A., and Buckley, S.G., 2003, Laser-induced breakdown spectroscopy detection and classification of biological aerosols, Appl Spectrosc, 57(10): 1207-1215. Jarvis, R.M., Brooker, A., and Goodacre, R., 2006, Surface-enhanced Raman scattering for the rapid discrimination of bacteria, Faraday Discuss, 132: 281-292. Kaye, P.H., Eyles, N.A., Ludlow, I.K., and Clark, J.M., 1991, An Instrument for the Classification of Airborne Particles on the Basis of Size, Shape, and Count Frequency, Atmos Environ a-Gen, 25(3-4): 645-654. Kaye, P.H., Barton, J.E., Hirst, E., and Clark, J.M., 2000, Simultaneous light scattering and intrinsic fluorescence measurement for the classification of airborne particles, Appl Optics, 39(21): 3738-3745. Kaye, P.H., Hirst, E., Foot, V.E., Clark, J.M., and Baxter, K., 2004, A low-cost multichannel aerosol fluorescence sensor for networked deployment., Proc. SPIE European Symposium Optics/Photonics in Security and Defence, London, ,, 5617: 388-398. Kaye, P.H., Stanley, W.R., Hirst, E., Foot, E.V., Baxter, K.L., and Barrington, S.J., 2005, Single particle multichannel bio-aerosol fluorescence sensor, Optics Express, 13(10): 3583-3593. Kuhn, H., 1949, A Quantum-Mechanical Theory of Light Absorption of Organic Dyes and Similar Compounds, J Chem Phys, 17(12): 1198-1212. Kume, H. 1994. Photomultiplier tube. Hamamatsu Photonics K.K., Japan. Kunnil, J., Sarasanandarajah, S., Chacko, E., and Reinisch, L., 2005a, Fluorescence quantum efficiency of dry Bacillus globigii spores, Optics Express, 13(22): 8969-8979. Kunnil, J., Sarasanandarajah, S., Chacko, E., Swartz, B., and Reinisch, L., 2005b, Identification of Bacillus spores using clustering of principal components of fluorescence data, Aerosol Sci Tech, 39(9): 842-848. Lakowicz, J. 1999. Principles of fluorescence spectroscopy (2nd ed.). Kluwer Acadimic/Plenum Publisher, New York Boston Dordrecht London Moscow. Leblanc, L., and Dufour, E., 2002, Monitoring the identity of bacteria using their intrinsic fluorescence, FEMS Microbiology Letters, 211: 147-153. Makino, S.I., Cheun, H.I., Watarai, M., Uchida, I., and Takeshi, K., 2001, Detection of anthrax spores from the air by real-time PCR, Lett Appl Microbiol, 33(3): 237-240.
Bio-aerosol Fluorescence
161
Manoharan, R., Ghiamati, E., Sperry, J.F., and Nelson, W.H., 1990, Rapid Bacterial Id Via Resonance Raman-Spectroscopy, Abstr Pap Am Chem S, 200: 138-Biot. Manoharan, R., Ghiamati, E., Chadha, S., Nelson, W.H., and Sperry, J.F., 1993, Effect of Cultural Conditions on Deep Uv Resonance Raman-Spectra of Bacteria, Appl Spectrosc, 47(12): 2145-2150. Martin, M.Z., Cheng, M.D., and Martin, R.C., 1999, Aerosol measurement by laser-induced plasma technique: A review, Aerosol Sci Tech, 31(6): 409-421. Melamed, M.R., Lindmo, T., and Mendelsohn, M.L. 1990. Flow Cytometry and Sorting. A John Wiley & Sons, Inc., New York. Murray, P.R., Baron, E.J., and Pfaller, M.A. 2003. Manual of Clinical Microbiology. Ameronican Society for Microbiology, New York. Mycek, M.A., and Pogue, B. 2003. Handbook of Biomedical Fluoresecnce. Marcel Dekker, inc. New York Basel. Nachman, P., Chen, G., Pinnick, R.G., Hill, S.C., Chang, R.K., Mayo, M.W., and Fernandez, G.L., 1996, Conditional-sampling spectrograph detection system for fluorescence measurements of individual airborne biological particles, Appl Optics, 35(7): 10691076. Nelson, W.H., Manoharan, R., and Sperry, J.F., 1992, Uv Resonance Raman Studies of Bacteria, Appl Spectrosc Rev, 27(1): 67-124. Nelson, W.H., Dasari, R., Feld, M., and Sperry, J.F., 2004, Intensities of calcium dipicolinate and Bacillus subtilis spore Raman spectra excited with 244 nm light, Appl Spectrosc, 58(12): 1408-1412. O'Neill, L., Lynch, P., McNamara, M., and Byrne, H.J., 2005, Structure property relationships in conjugated organic systems, Synthetic Met, 153(1-3): 289-292. Pan, Y.L., Aptowicz, K.B., Chang, R.K., Hart, M., and Eversole, J.D., 2003a, Characterizing and monitoring respiratory aerosols by light scattering, Opt Lett, 28(8): 589-591. Pan, Y.L., Hartings, J., Pinnick, R.G., Hill, S.C., Halverson, J., and Chang, R.K., 2003b, Single-particle fluorescence spectrometer for ambient aerosols, Aerosol Sci Tech, 37(8): 628-639. Pan, Y.L., Boutou, V., Chang, R.K., Ozden, I., Davitt, K., and Nurmikko, A.V., 2003c, Application of light-emitting diodes for aerosol fluorescence detection, Opt Lett, 28(18): 1707-1709. Pan, Y.L., Boutou, V.E., Bottiger, J.R., Zhang, S.S., Wolf, J.P., and Chang, R.K., 2004, A puff of air sorts bio-aerosols for pathogen identification, Aerosol Sci Tech, 38(6): 598602. Pan, Y.L., Holler, S., Chang, R.K., Hill, S.C., Pinnick, R.G., Niles, S., and Bottiger, J.R., 1999, Single-shot fluorescence spectra of individual micrometer-sized bio-aerosols illuminated by a 351- or a 266-nm ultraviolet laser, Opt Lett, 24(2): 116-118. Pan, Y.L., Pinnick, R.G., Hill, S.C., Niles, S., Holler, S., Bottiger, J.R., Wolf, J.P., and Chang, R.K., 2001a, Dynamics of photon-induced degradation and fluorescence in riboflavin microparticles, Appl Phys B-Lasers O, 72(4): 449-454. Pan, Y.L., Cobler, P., Rhodes, S., Potter, A., Chou, T., Holler, S., Chang, R.K., Pinnick, R.G., and Wolf, J.P., 2001b, High-speed, high-sensitivity aerosol fluorescence spectrum detection using a 32-anode photomultiplier tube detector, Rev Sci Instrum, 72(3): 18311836. Pinnick, R.G., Hill, S.C., Pan, Y.L., and Chang, R.K., 2004, Fluorescence spectra of atmospheric aerosol at Adelphi, Maryland, USA: measurement and classification of single particles containing organic carbon, Atmos Environ, 38(11): 1657-1672.
162
Y.-L. Pan et al.
Pinnick, R.G., Hill, S.C., Nachman, P., Videen, G., Chen, G., and Chang, R.K., 1998, Aerosol fluorescence spectrum analyzer for rapid measurement of single micrometersized airborne biological particles, Aerosol Sci Tech, 28(2): 95-104. Pinnick, R.G., Hill, S.C., Nachman, P., Pendleton, J.D., Fernandez, G.L., Mayo, M.W., and Bruno, J.G., 1995, Fluorescence Particle Counter for Detecting Airborne Bacteria and Other Biological Particles, Aerosol Sci Tech, 23(4): 653-664. Pourahmadi, F., Taylor, M., Kovacs, G., Lloyd, K., Sakai, S., Schafer, T., Helton, B., Western, L., Zaner, S., Ching, J., McMillan, B., Belgrader, P., and Northrup, M.A., 2000, Toward a rapid, integrated, and fully automated DNA diagnostic assay for Chlamydia trachomatis and Neisseria gonorrhoeae, Clin Chem, 46(9): 1511-1513. Reyes, F.L., Jeys, T.H., Newbury, N.R., Primmerman, C.A., Rowe, G.S., and Sanchez, A., 1999, Bio-aerosol fluorescence sensor, Field Anal Chem Tech, 3(4-5): 240-248. Rider, T.H., Petrovick, M.S., Nargi, F.E., Harper, J.D., Schwoebel, E.D., Mathews, R.H., Blanchard, D.J., Bortolin, L.T., Young, A.M., Chen, J.Z., and Hollis, M.A., 2003, A B cell-based sensor for rapid identification of pathogens, Science, 301(5630): 213-215. Russell, S.C., Czerwieniec, G., Lebrilla, C., Steele, P., Riot, V., Coffee, K., Frank, M., and Gard, E.E., 2005, Achieving high detection sensitivity (14 zmol) of biomolecular ions in bio-aerosol mass spectrometry, Analytical Chemistry, 77(15): 4734-4741. Schroder, K.L., HargisJr, P.J., Schmitt, R.L., Rader, D.J., and Shokair, I.R., 1999, Development of an unattended ground sensor for ultraviolet laser induced fluorescence detection of biological agent aerosols, Conference on Air Monitoring and Detection of Chemical and Biological Agents II, SPIE, 3855: 82-91. Seaver, M., Eversole, J.D., Hardgrove, J.J., Cary, W.K., and Roselle, D.C., 1999, Size and fluorescence measurements for field detection of biological aerosols, Aerosol Sci Tech, 30(2): 174-185. Sengupta, A., Laucks, M.L., Dildine, N., Drapala, E., and Davis, E.J., 2005, Bio-aerosol characterization by surface-enhanced Raman spectroscopy (SERS), J Aerosol Sci, 36(56): 651-664. Sivaprakasam, V., Huston, A.L., Scotto, C., and Eversole, J.D., 2004, Multiple UV wavelength excitation and fluorescence of bio-aerosols, Optics Express, 12(19): 44574466. Steiner, W.E., Klopsch, S.J., English, W.A., Clowers, B.H., and Hill, H.H., 2005, Detection of a chemical warfare agent simulant in various aerosol matrixes by ion mobility timeof-flight mass spectrometry, Analytical Chemistry, 77(15): 4792-4799. Storrie-Lombardi, M.C., Hug, W.F., McDonald, G.D., Tsapin, A.I., and Nealson, K.H., 2001, Hollow cathode ion lasers for deep ultraviolet Raman spectroscopy and fluorescence imaging, Rev Sci Instrum, 72(12): 4452-4459. Teale, F.W.J., and Weber, G., 1957, Ultraviolet Fluorescence of the Aromatic Amino Acids, Biochem J, 65(3): 476-482. Tjarnhage, T., Stromqvist, M., Olofsson, G., Squirrell, D., Burke, J., Ho, J., and Spence, M., 2001, Multivariate data analysis of fluorescence signals from biological aerosols, Field Anal Chem Tech, 5(4): 171-176. Weichert, R., Klemm, W., Legenhausen, K., and Pawellek, C., 2002, Determination of fluorescence cross-sections of biological aerosols, Particle & Particle Systems Characterization, 19(3): 216-222. Wu, Q., Hamilton, T., Nelson, W.H., Elliott, S., Sperry, J.F., and Wu, M., 2001, UV Raman spectral intensities of E. coli and other bacteria excited at 228.9, 244.0, and 248.2 nm, Analytical Chemistry, 73(14): 3432-3440.
Bio-aerosol Fluorescence
163
Wyatt, P.J., Schehrer, K.L., Phillips, S.D., Jackson, C., Chang, Y.J., Parker, R.G., Phillips, D.T., and Bottiger, J.R., 1988, Aerosol-Particle Analyzer, Appl Optics, 27(2): 217-221. Ye, Z.M., and Auner, G., 2004, Principal Component Analysis Approach for Biomedical Sample Identification, 2004 IEEE lntemational Conference on Systems, Man and Cybernetics: 1384-1354.
Kevin Aptowicz and Sunny Chang enjoy the warm Siberian hospitality.
Suping, Kyle, and Yong-Le Pan
164
Y.-L. Pan et al.
Xin-Hua Hu, Richard Chang, Bailey Hu, Jun Lu and Belinda Hu.
OPTICAL AND MORPHOLOGICAL CHARACTERIZATION OF SELECTED PHYTOPLANKTONIC COMMUNITIES Applications for remote sensing Antonio Palucci Laser Applications Section, ENEA, Via Fermi 45, 00044 Frascati, Italy Tel.: +39 (06) 9400-5299, fax: +39 (06) 9400-5312, e-mail: [email protected]
Abstract:
A systematic laboratory characterization of different phytoplankton cultures has been performed in combination with LIF (Laser Induced Fluorescence) measurements aimed to investigate the possibility of their remote monitoring by means of lidar fluorosensor systems. Cultures of microalgae characterized by different pigment content have been analyzed in the visible region upon UV laser excitation. High-resolution laboratory spectra have been measured in order to obtain the fingerprint of each species. Emission wavelengths related to the main pigments’ contribution have been identified. Detection limits of the emitted red chlorophyll signal have been evaluated for the different species after dilution in the culture medium and in real seawater. Prior to the LIF excitation aimed at remote characterization, the algal cultures were morphologically analyzed by fluorescence and scanning electron microscopy. In addition, the relevant cell number was counted for biomass estimation.
Key words:
algae, fluorescence, morphology, laser remote sensing, phytoplankton.
1.
INTRODUCTION
In the early stage of running a lidar fluorosensor apparatus, the ENEA laser remote-sensing laboratory in Frascati initiated a detailed parallel investigation on the organisms accountable for the well known red fluorescence emission.1 The participation in national and international oceanographic campaigns2,3 led us to explore this aspect, due to the possible 165 A. Hoekstra et al. (eds.), Optics of Biological Particles, 165–192. © 2007 Springer.
166
A. Palucci
capability of distinguishing dominant species dispersed in seawaters. In fact, aside from the chlorophyll-a (Chl-a) pigment, many other pigments are present in these living organisms and several of them show the ability to fluoresce if stimulated by proper light emission. Therefore, a systematic laboratory characterization of different phytoplankton cultures has been performed in combination with LIF (Laser Induced Fluorescence) measurements aimed at investigating the possibility of remote monitoring by means of lidar fluorosensor systems. Cultures of microalgae characterized by different pigment contents have been analyzed in the visible region upon UV laser excitation. High resolution laboratory spectra have been measured in order to obtain the fingerprint of each species. Emission wavelengths related to the main pigments have been identified, and detection limits of the emitted red chlorophyll signal have been evaluated for the different species after dilution in the culture medium and in real seawater.4 Prior to the LIF excitation, the algal cultures were morphologically analyzed by fluorescence and scanning electron microscopy. In addition, the relevant cell number was counted for biomass estimation, and the chlorophyll content was determined by different chemical methods. The laboratory study was preparatory to the design of a complete fluorosensor mobile laboratory equipped with a multichannel fluorosensor apparatus successfully applied in monitoring seawater quality and phytoplankton during four Antarctic oceanographic campaigns.5,6 The instrument includes a double-pulse transmitter, which allows us to evaluate phytoplankton photosynthetic activity from differential measurements of Chl-a fluorescence combined with PAR determinations at the sea surface.5 Successively, a submersible laser fluorometer payload was developed adopting a similar LIF technique in order to study phytoplankton stratification or pollutant releases in deeper layers. The final optical fluorescence database and the foremost morphological characterization are the base to implement the promising scanning flow cytometry technique7 in a new submersible stand-alone instrument8 whose future application in buoy or submersible devices are foreseen. In the following sections we discuss the primary characteristics of phytoplankton, followed by the application of the LIF technique to marine characterization with the goal of recognizing families usually held responsible for noxious blooms. Various phytoplankton species have been examined: diatoms, green algae, cyanobacteria and dinoflagellates, either in their culture solution or in natural seawater at different concentrations. All the samples were grown under optimal environmental conditions and characterized by several bio-chemical techniques. In the experimental section, the laboratory version of our marine lidar fluorosensor apparatus,
Characterization of Phytoplankton
167
which is in fact a laser fluorometer equipped with an optical multichannel detector, is described together with the data analysis procedures. Results, relating the measured LIF intensities to cell numbers, chlorophyll concentration and characteristic pigment ratios, have been obtained for various species of the considered phytoplankton groups and are here discussed in detail. Remarks on the design of a compact multichannel lidar system and of a scanning flow cytometry dedicated to phytoplankton monitoring are given at the end, together with conclusions.
2.
THE GRASS OF THE SEA: PHYTOPLANKTON
Oceans sustain a key role in the climate equilibrium through the energy transport from poles to equator, thus strongly affecting the global biogeochemical cycles. The waters represent 99% of the available living space. The oceans produce more than half of the atmospheric oxygen and enclose in their sediments more than 99% of the CO2 (Carbon Dioxide). Phytoplankton are microscopic plants that live in the oceans as well as in different aquatic environments and can be considered the best indicator of the biogeochemical oceanic cycles because their growth depends on their surrounding conditions. It is the base of the trophic sea chain, it produces almost all the oxygen released by the oceans in the atmosphere, and it is responsible for over 40% of the global carbon fixation. Its biomass can be estimated by the Chl-a concentration. Marine ecosystems support more than 35% of the total (marine and terrestrial) primary production with a very fast turnover, weeks and more than ten years, respectively.9 It can be considered as the grass of the sea,10 and as a consequence of its internal composition, it radically influences the color of the oceans and affects the depth of light penetration through the water column.11 Many phytoplanktonic species are able to control their vertical position in the water column in small changes by floating, following marine currents, or by flagellar mobility. Their migration from surface to sublayers is mainly due to variations in nutrients and radiance availability along the water column. Seven main taxonomic groups of phytoplankton can be recognized:12,13,14 Chlorophyta (Green Algae) - large and diverse group. Chl-a, -b, carotenes, and some xanthophylls pigments are present. Cell walls consist of inner cellulose and outer pectinaceous layers (e.g., Ulothrix, Chlamydomonas, Scenedesmus, Pediastrum, Eudorina, Ankistrodesmus) Examples are shown in Figure 1.
168
A. Palucci
Figure 1. Picture of Eudorina and Volvox Chlorophyta algae, from collection15.
Cyanophyta (Cyanobacteria or Blue-green Algae) - only alga with prokaryotic cells (undifferentiated) and only group able to fix atmospheric N. Cells lack mitochondria, chloroplasts, internal membranes. Most are filamentous. Pigments consist of Chl-a, β-carotene, some xanthophylls, and phycobilins, particularly phycocyanin-C (gives them blue-green colour). Cell wall is mucopeptide and many species are surrounded by a gelatinous sheath. Vegetative reproduction may be more rapid than in most other phytoplankton (e.g., Anabaena, Oscillatoria, Microcystis, Gleotrichia, Aphanizomenon). Examples are shown in Figure 2.
Figure 2. Pictures of Anabena and Oscillatoria Cyanophyta algae from collection 15.
Bacillariophyta (Diatoms) – wider spread and most important group of algae. They include both unicellular and colonial forms. Pigments consist of Chl-a, -c, carotene, and xanthophylls. Energy is stored as fat and oil in large globules. The cell wall is silica embedded in a pectinaceous matrix and construction consists of overlapping valves, giving a pillbox-like structure. They are divided into 2 groups: Centrales have radial symmetry and Pennate diatoms have bilateral symmetry (Pennales). Vegetative reproduction at night is their most common mode (e.g., Asterionella, Cyclotella, Melosira, Fragillaria, Tabellaria, Stephanodiscus). Examples are shown in Figure 3.
Characterization of Phytoplankton
169
15
Figure 3. Pictures of diatoms algae from collection .
Chrysophyta (Golden-brown Algae) - colour is due to dominance of βcarotene and some xanthophyll carotenoids in addition to Chl-a. Most species are unicellular; a few are colonial. Many lack cell walls and are bound only by cytoplasmic membranes. Some colonial forms such as Synura, Uroglena, and Dinobryon are widely distributed and are often major components of phytoplankton. Cryptophyta (Cryptomonads) - most species are naked, unicellular, and motile. They are usually flattened dorso-ventrally, with anterior invagination with 2 equal-length flagella. Reproduction is due to longitudinal division. Sexual reproduction is unknown as are many facets of their life cycle. (e.g., Cryptomonas, Rhodomonas, Chroomonas) Pyrrhophyta (Dinoflagellates) - unicellular flagellated algae, many of which are motile. Most have a conspicuous cell wall elaborated with sculptured patterns and the formation of large spines and cell wall processes. Transverse and longitudinal furrows in their cell surface connect and contain flagella, which allow weak movement (e.g. Ceratium, Gymnodinium, Peridinium). An example is shown in Figure 4.
170
A. Palucci
Figure 4. Picture of Ceratium dinoflagellates from collection 15.
Euglenophyta (Euglenoids) - relatively large and diverse group but few species are truly planktonic. Most are unicellular but lack a cell wall and possess 1-3 flagella arising from an invagination of the external membrane. Euglenoids are found most often in seasons, depth strata, or lake systems in which ammonia is high. (e.g., Euglena, Phacus). Size is one of the main characteristics of these unicellular compounds that can be included in between the very small viruses and larger zooplankton species. In Table 1 is reported the main size classification of algae and their relative abundance in a case study of the Antarctic Ross Sea.16 Micro-plankton are the main components of the natural algal community while the pico-plankton contribution is not negligible. Table 1. Percentage contribution of the pico-, nano-, micro-plankton in the Antarctic Ross sea polynya. Dimension Fractions Pico-plankton < 2 micron 15 – 30% Nano-plankton 10-2 micron 20 – 30% Micro-plankton > 10 micron 35 – 50%
Seawater grass is a complex biosphere where different species can contribute to the so-called natural phytoplanktonic community. Algae are dispersed along the water column in equilibrium with the nutrient gradient, in the thermocline, due to their capability to float in the water either by changing their buoyancy or by flagellar motility. In Table 2 are reported the algae concentrations at different seawater depths in the same case study of the Antarctic Ross Sea17 as an example of the composite of changes of the natural phytoplanktonic community along the water column.
Characterization of Phytoplankton
171
Table 2. Distribution of algal species at superficial, 5m, 50m and 350m seawater depth in a station of the Antarctic Ross Sea. Depth
Sup.
5m
50m
350m
Species (cell/l) Bacillariophyceae Actinocyclus actinochilus Chaetoceros app. Cylindrotheca closterium
120
120
1000
160
240
120
Eucampia sp.
40
Flagilariopsis curta
4680
4520
6160
Flagilariopsis kergulensis
2600
160
4800
1200
1480
Pleurosigma spp.
20
Pseudo-nitzschia pseudodelicatissima Pseudo-nitzschia spp.
800
Stellarima microtrias Thalassiosira Antarctica
23200
40
160
8640
560
Thalassiosira spp.
40
Dinophyceae Amphidinium hadai
20
Gyrodinium fusiforme
20
40
Prymniosiophyceae Phaeocystis cf. Antarctica
68800
642000
1126000
1000
2560
2000
2600
102380
659400
1142520
2600
Other flagellate species Undeterminated flagellates (<10µm) Total
The photosynthetic apparatus enclosed in these microscopic plants is composed of different pigments. Among them, Chl-a is primarily responsible for the capture of solar energy and of energy transfer down a chain of electron-transfer components that assist in the manufacture of energy in the form of ATP that can be used to synthesize cellular components from CO2. The global carbon cycle, which regulates the temperature of our planet and affects life-sustaining oxygen essential to the metabolism of all aerobic organisms, are controlled by the actions of phytoplankton. Perhaps no other group of organisms plays such an important role in the maintenance of life on Earth.
172
A. Palucci
Therefore, the importance of monitoring Chl-a concentration in the worldwide oceanic environment has emerged as an important undertaking. Among the various possible techniques, passive and active remote sensing can be a major contributor to this task. The constellation of different ocean color passive remote sensing satellites can supply valuable information on the oceanic water optic characteristics, which are affected by the presence of different phytoplanktonic species; thus, increasing the radiance emitted in the green and the absorbance in the blue and red spectral regions. In active remote sensing, the fluorescence of Chl-a is observed, e.g. at 680 nm, induced by an artificial source, e.g. a laser emitting light at 355 nm.
3.
MATERIALS AND METHODS
3.1 LIF Techniques in Monitoring Phytoplankton Species Since visible and near UV lasers are suitable to excite chromophores, especially chlorophyll and carotenoids, in the living vegetative tissues of algae and plants,18 LIF techniques are commonly used to investigate vegetation targets both in local and remote applications. In order to ascertain the possibility of remotely classifying algae from their pigment contents, previous laboratory measurements were performed on different phytoplankton cultures in saline solutions simulating real brackish water in natural environments, including coastal lakes and open sea. The emission spectra collected from a water column excited by UV radiation contain as a dominant feature the water Raman peak red-shifted from 3330 cm-1, which is currently used for normalization of remotely sensed data.19,20 In the absence of chemical pollution due to industrial releases, the visible LIF emission is due mostly to Dissolved Organic Matter (DOM), which is always present in natural waters,21 and to different phytoplankton pigments. The visible algal emissions show peculiar structures, extending from the blue to the red region, which occur in spectral windows suitable for remote determination.22 The most important red phytoplankton fluorescing pigments,23 which can be traced either upon UV or visible excitation, include phycoerythrin (580 nm), phycocyanin (660 nm), and chlorophylls (peaked around 685 nm with a shoulder at 730 nm). By using an UV excitation source it might be possible to detect also the blue pigments bands overlapping the DOM emission, which contain contributions due to spectral absorption and re-emissions from xanthophyll, NaPDH, fucoxantin24 and minor light-harvesting carotenoids.25 In the
Characterization of Phytoplankton
173
absorption blue band some structures have been related recently to the Peridinin-Chlorophyll-a Protein (PCP) characteristics of dinoflagellates.26 However most of the transitions occurring in the blue spectral region overlap the broad DOM emission. This fact reduces the importance of blue pigments in field classification of different algal specie.
3.2 Phytoplankton Cultures Different algal cultures were grown during parallel auto-ecological laboratory studies of various species that might originate noxious and toxic blooms. A deeper knowledge of each single species and its cellular characterization, including the theca composition, is very important to applied interdisciplinary studies and to evaluate the possibility of their remote detection. After identification of the dominant species by optical microscopy, the cultures were grown in their respective enriched media, Guillard (1975) medium (F/2) for dinoflagellates species and "ETRSW" medium in case of green algae and diatomeae. Details on laboratory growing conditions and biochemical characterizations are well documented.4 Classification and general characteristics of all the algae cultures analyzed are summarized in Table 3. Most of the algae listed in Table 1 have been grown from samples fished through a vertical phytoplankton net (25 µm diameter) along the Italian coasts either in the north-west Adriatic (all the Prorocentraceae) or in the northern Tyrrenian Sea (Chlorophyceae and Bacillarophyceae). Dinoflagellates belonging to the Alexandrium genera were taken from the Atlantic.
3.3 LIF Apparatus A laser fluorometer has been used to record the complete visible LIF emission spectra of algal cultures in the laboratory (Figure 5). After proper dilution, algal cultures have been analyzed in a small volume quartz cuvette during a non-turbulent fast flow obtained by gravity from a large volume reservoir. UV radiation emitted from the laser source at 355 nm excites the fluorescence of the algal chromophores contained in the solution flowing through the cuvette. The fast flow allows researchers to analyze a fresh sample at each subsequent laser shot, thus avoiding degradation effects due to possible laser-induced damage of the cell membrane. For the same reason no pump has been used to sustain the flow.
174
A. Palucci
Table 3. Classification of the algal cultures here examined, main pigments present in each class are listed between brackets as well as special characteristics of some families. (*) Different Chlorophyll c are indicated for single species. Common name Green algae
Diatoms
Class (pigments)
Order
Family (remarks)
Species
Chlorophyceae Chl-a Chl-b; α, β, γ carotenoids
Volvocales
Clamydomonaceae (with cell wall)
Clamydomonas sp.
Prasinophyceae Polyblepharidaceae (without cell wall) Phaeodactylinaceae
Platymonas suecica Dunaliella terctiolecta.
Bacillarophyceae Chl-a and chl-c*; β-carotenoids, xantophylls (fucoxantin, diatoxantin, diadinoxantin)
Pennates
Nitzschiaceae “ Fragilariaceae Dinoflagellates
Dinophyceae Chl-a and chl-c; β-carotenoids, xantophylls (peridin, neoperidin, dinoxantin, diadinoxantin)
Prorocentrales
Gonyaulacales Cyanobacteria
Cyanophyceae Chl-a and chl-c; β-carotenoids, phycocyanin, phycoerythrin
Chroocoecales
Prorocentraceae
“ Goniodomataceae “ Chroococcacaceae
“
Phaeodactylum tricornutum (Chl-c1 and c2)
Nitzschia closterium (Chl-c2 and c3) Nitzschia delicatissima Fragilaria sp. (Chl-c1 and c2) Prorocentrum minimum
Prorocentrum micans Alexandrium tamarense Alexandrium lusitanicum Synechococcus leopoliensis (Racib) Komarek
Synechococcus 625
The fluorometer collection optics, mounted nearby the cuvette at a right angle to the laser excitation beam, include an appropriate low-pass filter to detect fluorescence while cutting off the scattered laser radiation, a 10 cm lens and a quartz fiber optic 10 m long, all installed on a single optical mount for easy alignment.
Characterization of Phytoplankton
175
RESERVOIR LASER Nd:YAG λ = 355 nm
CELL
F L Monochromator
DISCHARGE
F.O.
OMA
PC
GATE
Figure 5. Laboratory Laser fluorometer for liquid samples analysis: F = filter; L = lens; FO = fiber optic.
In order to record single-shot spectra in a broad visible range, a timegated Optical Multichannel Analyzer (OMA) has been chosen as a fluorescence detector for the laboratory application. The OMA head can be mounted on a small monochromator, which in turn can be equipped either with low- or high-resolution gratings. Proper optical matching between the glass fiber and each monochromator is ensured by small lenses focusing the signal on the respective entrance slit. The primary components of the apparatus with details of the relevant characteristics have been published previously.4 The laser fluorometer was always operated at 10 Hz during this laboratory classification. Prior to any data processing, averages of 100 spectra were stored on the PC controlling the experiment. During all the LIF measurements, concentrated phytoplankton species solutions have been carefully diluted either in their growth solution or in fresh seawater samples collected in the middle Tyrrenian Sea near Anzio harbor. In the latter case it has been carefully checked that the seawater used for dilution originally contained a negligible concentration of phytoplankton, i.e. a concentration below the detection limits of the instrument.
176
A. Palucci
3.4 Fluorescence Data Analysis Procedure In order to correlate the LIF intensity measured on different spectral channels to the respective pigment content, the spectra measured upon UV excitation at 355 nm were successively processed, integrating over a 10 nm width on the water Raman peak (402 nm) and on the different pigment bands listed in Table 4, for chl-a the integration has been performed on the sharp peak related to PS-II and always found near 680 nm. Table 4. Spectral features in laser excitation of phytoplankton. λexc [nm] 355
H2O Raman [nm] 402
LIF emission [nm] 420-480 450-600 580 660 680 730
Chromophores DOM Carotenoids (Fucoxanthin, Xanthophyll), NaPDH, PCP Phycoerythrin Phycocyanin Chl-a (PSII) Chl-a (PSI)
A multiple Gaussian deconvolution has been applied, in the case of overlapping spectral features in order to take into account the contribution of nearby structures on each integral. An example of this deconvolution is given in Figure 6 for an algal species containing all the chromophores listed in Table 4 and the typical integration band is indicated for each deconvoluted peak. 2000 Water Carotenoids Raman
LIF intensity [rel. un.]
1500
1000
Phycoeritrin
Phycocyanin 500 Chlorophylls
0 400
450
500
550
600
650
700
750
Wavelength [nm]
Figure 6. Gaussian deconvolution on the LIF spectrum of a concentrated solution of Synecochoccus leopoliensis. The assignment of the pigments is indicated and different integration bands are marked.
Characterization of Phytoplankton
177
Successively LIF intensities have either been expressed in Raman units for calibration with the different pigment content or reduced to spectral ratios for various species recognition.
3.5 Fluorescence Calibration For absolute calibration with pigment content, the integrated LIF intensities measured at the corresponding emission wavelengths have been expressed in Raman units, through normalization to the Raman water peak. In the latter procedure a subtraction of the blue tail due to DOM and other pigment contributions on the water Raman peak at 402 nm was performed after the appropriate 4 Gaussian deconvolutions of the blue portion of the spectrum (Figure 6). Non-negligible contributions of the DOM and blue pigments tail were found only for concentrated cultures and in the cases of dilution in natural seawater. LIF measurements of different pigment concentrations were performed after dilution of concentrated samples from the original cultures. The LIF spectra of natural or synthetic seawaters used for dilution are shown in Figure 7a. The residual DOM emission at 402 nm has been subtracted from all the data here reported before their final reduction. Other contributions to the blue region of the emission spectra can be due to the culture growth medium. As shown in Figure 7b, the ETRSW terrain containing soil extract, which was used for growing green algae and diatomeae, shows a non-negligible broad blue emission. Due to the contemporary presence of DOM from degradation of biological material during the algae growth, the terrain spectrum cannot simply be subtracted from the LIF spectra of cultures, however its presence must be kept in mind for the ETRSW case. In the absence of proper chemical determinations of different dissolved impurities, the LIF data taken from aqueous solutions are expressed in Raman units20 with a procedure utilizing different spectral channel intensities measured on the same laser-induced spectrum. The procedure, described in the following, makes it possible to compare concentration data taken with different laser systems on different waters. Upon laser excitation of an aqueous medium, fluorescence and Raman emission signals are generated, which in the linear regime after time integration and spectral analysis, can be related to the chromophore concentrations and to the cross section efficiency as follows:27
178
A. Palucci 50000 Anzio harbour Synthetic
40000 30000 20000 10000
(a)
0 50000
400
450
500
550
600
650
700
750
ETRSW F/2
40000 30000 20000 10000
(b)
0 400
450
500
550
600
650
700
750
Wavelenght (nm)
Figure 7. LIF spectra of water solutions used to dilute algal cultures: a) Anzio and synthetic seawater; b) F/2 and ETRSW.
Fλem =
σ (λem , λex ) A E ex m2 kT
(1)
where F is the total signal emitted at wavelength λem, m is the refraction index of water, A is a system dependent constant, Eex is the excitation laser energy, σ is the quantum efficiency of the process and kT is the total extinction coefficient kT = kex.+ kem given by the sum of the terms at the excitation and emission wavelength. The ratio of the chromophore fluorescence signals on the water Raman scattering intensity can be expressed as
F σ F k ex + k R = R σ R k ex + k F
(2)
where σs and the ks refer to the respective wavelengths. Normalizing this to the Raman signal eliminates the dependence on the system parameters, on the laser energy, and on the water optical properties. In Eq. (2) the ratio of the extinction coefficients can be set approximately to a constant and neglected, provided that a careful choice of excitation and emission wavelengths is performed in order to avoid errors due to differential absorption. The different chromophore concentrations expressed in Raman units are thus independent of the experimental parameters.
Characterization of Phytoplankton
179
The LIF intensity at selected wavelengths (hereafter labeled Fλ), once expressed in Raman units, are directly related to cell number, chlorophyll and carotenoid content. Results provide the calibration of the laser fluorometer in the absolute determination of pigment content for the examined algal cultures.
3.6 Species Recognition In field surveys, as well as in cytometry, identifying a large number of different phytoplankton species in real marine environments may require the use of several different indicators, according to the quality and the color of the examined waters. As a first diagnostic, the presence of characteristic red emissions, peaked at the wavelengths listed in Table 4 for the main algal pigments is suitable to recognize the presence of several algal families. Further information can be derived from accurate measurements of the peak position for each resolved red band. The use of pigment ratios can be appropriate in remote recognition problems. In fact red emissions can be normalized to the broad blue feature centered at 450nm containing both the carotenoid and the DOM contribution in real seawater. Alternatively LIF intensities related to different pigments can be normalized to the total visible emission (Ftot) recorded between 375 and 800 nm. This choice of normalizing different Fλ to the Ftot has the advantage that it removes the spectral dependence of the sensitivity, which is due to the combination of optics transmission and detector efficiency at each wavelength. However, the latter experimental parameter must be taken into account before comparing the absolute value of ratios measured with different fluorosensor systems. For very complex situations, e.g., when several algal families are present, it may be necessary to use a large set of ratios covering the whole visible spectral range from the blue to the red. If n is the minimum number of spectral ratios Rsj (j = 1,n) required for identifying species s, laboratorymeasured averages must be made available to form a reference database of Species Specific Ratios (SSP), determined with the appropriate spectral resolution and extended to all phytoplankton of importance for blooms. Due to the overlap of DOM with blue pigment emission, the building up of this reference database is a crucial task. In fact, a database suitable for reliable field recognition should include only laboratory determinations performed on algal cultures diluted in seawaters with a DOM content as close as possible to that of the area kept under control during the blooms.
180
A. Palucci
Remotely collected spectral ratios Rij (i = 1, N; j = 1, n), where N is the total number of measurements taken over the same area and n is the number of spectral ratios considered, can then be mapped to the database SSPs to disentangle each species s. As an indicator of the presence of a certain species, we suggest using the average distance D in the n-dimensional space of spectral ratios, i.e.
D=
1 N
∑ ∑ (R N
i
n j
ij
− 〈 Rsj 〉 )
2
(3)
which should exhibit a minimum only in one case, thus identifying the presence of the corresponding algal family or even species in the field data population. In the latter case the confidence level for the recognition can be estimated by evaluating the typical variances of the SSPs .
4.
RESULTS
4.1 Morphology The morphology of the examined dinoflagellates species has been carefully investigated by SEM microscopy, results relevant to the images of the four algae are reported. The different cell dimensions and shapes can be observed in these images together with the characteristic flagellum, hanging on the outer cell membrane, especially evident in the case of the large cell of Pr. Micans. (Figure 8b). Fluorescence images also allow us to estimate the fluorescing pigment content.
4.2 LIF Spectra Low-resolution LIF spectra measured from different algal cultures upon UV excitation are shown in Figure 9 – Figure 12 for concentrated aqueous solutions of different species belonging to Dinophyceae, Chlorophyceae, Bacillarophyceae and Cyanophyceae, which are listed in Table 5. Spectral features characteristic of each algal species and of different families within the same class are evident in these spectra.
Characterization of Phytoplankton
181
A
B
C
D
Figure 8. SEM (left) and fluorescence (right) microscope images of Dinophyceae: a) Prorocentrum minimum (size: length 35-70 µm; width 20-50 µm); b) Prorocentrum micans (size: length 14-22 µm width 10-15 µm); c) Alexandrium lusitanicum (size: length 16-30 µm); d) Alexandrium tamarense (size: length 24-50 µm).
The LIF spectra of four dinoflagellates measured on the concentrated growth solutions whose characterization is given in Table 5 are shown in Figure 9. In common to all the spectra is a characteristic structured emission in the blue region overlapping the broad DOM emission. Three peaks centered around 435, 460 and 490 nm respectively can be distinguished. The spacing between these unresolved structures suggests their assignment to carotenoids absorption and reemission, whose vibrational progression in the S2 state has an energy gap of about 1400 cm1 25 . The progression peaked at 457, 497 and 539 nm has been tentatively assigned to peridinin in toxic dinoflagellates.26 Table 5. Cell numbers and pigment content of dinoflagellates concentrated solutions. Algae
Cells conc. [cell/ml]
Chl-a g/l]
Prorocentrum minimum
6 104
296
Prorocentrum micans
3 104
440
Alexandrium lusitanicum
5 10
4
360
Alexandrium tamarense
3 104
220
182
A. Palucci
The spectra reported in Figure 9 also show that the intensity of red chlorophyll emission, always peaked around 683 nm, differs strongly within the considered group of dinoflagellates. The characteristic F450/F683 ratio could be an indication of the presence of a certain family or species in a natural community of dinoflagellates. 1400
Prorocentrum micans Prorocentrum minimum Alexandrium lusitanicum Alexandrium tamarense
LIF intensity [rel. un.]
1200
1000 800
600
400
200 0 400
450
500
550
600
650
700
750
800
Wavelength [nm]
Figure 9. LIF spectra of different dinoflagellates cultures given in Table 5.
The LIF spectra of four algal species classified as belonging to chlorophyceae and bacillarophyceae are reported in Figure 10 as measured from their concentrated growth solutions, which are characterized from the data listed in Table 6. Note that the overall highest intensity measured on the spectra in Figure 10 in comparison with the dinoflagellates spectra shown in Figure 9 is due to the higher cell concentration examined in the latter case. Furthermore the spectra in Figure 10 have an overall less intense red chlorophyll emission, whose central wavelength position between 680 and 690 nm depends on the algal species. Table 6. Cell numbers and pigments content of chlorophyceae and bacillarophyceae cultures. Algae
Cell conc. [cell/ml]
Chl-a g/l]
Platimonas suecica
84⋅104
6333
Dunaliella terctiolecta
72⋅10
4
7044
Clamydomonas sp.
70⋅104
6860
4
8110
Pheodactylum tricornutum
82⋅10
Nitzschia closterium
<104
2757
Nitzschia delicatissima
60⋅104
3530
Fragilaria sp.
<104
4524
Characterization of Phytoplankton
183
Figure 10. LIF spectra of different chlorophyceae and bacillarophyceae cultures, the relevant cell numbers and pigments content are given in Table 5.
The LIF spectra of three other Bacillarophyceae are reported in Figure 11, as measured on very concentrated growth solutions (Table 6). This group of algae, formed by Nitzschia and Fragilaria families, is characterized by an extremely weak red chlorophyll emission (see the expanded inset). Note that in case of Nitzschia closterium and Nitzschia delicatissima almost no red peak could be detected, although in the latter case measurements were performed on a solution with a cell number of 60 104 which is comparable to the one used to obtain the other spectra of chlorophyceae and bacillarophyceae shown in Figure 11. The LIF spectra of two Cyanobacteria are reported in Figure 12 as measured on diluted growth solutions. In spite of the fact that both the algae belong to the Synechococcus family, their pigment content is quite different, thus resulting in different features in the detected LIF spectra. The cell number for the Synechococcus 625 was 343 104, while it was not determined for the other species. Both the species show the red phycocyanin peak at 660 nm, partially overlapping the weak chlorophyll band, which is peculiar to all Cianobacteria. Furthermore the Synecochoccus leopoliensis, being characterized by a high abundance of phycoerythrin, the main light-harvesting pigment for PS-II23 instead of Chla, shows a dominant feature peaked at 580 nm ascribed to this pigment (see Table 1).
184
A. Palucci
Figure 11. LIF spectra of different bacillarophyceae cultures.
In order to investigate the possibility of using the red chlorophyll peak position as an indicator for the identification of different diatoms families, high resolution spectra have been measured in this region by using both the UV and green laser excitation. Results obtained from the concentrated cultures (relevant data in Table 5) using an excitation wavelength of 532 nm, thus avoiding spectral distortions in the absence of a strong DOM tail extending toward the red, are shown in Figure 13. This figure shows that the chl-a emission in the chlorophyceae P. suecica and D. terctiolecta is peaked at 683 nm, as in the case of all the considered dinophyceae (Figure 9). A red shift to 685 nm is observed for the other chlorophycea investigated (Clamydomonas sp.), while all the bacillarophyceae shown in Figure 13 (P. tricornutum and N. clostrerium) as well those reported in Figure 11, show the red peak maximum at 680 nm.
Characterization of Phytoplankton
185
1600
LIF intensity [rel. un.]
1400
Synechococcus 625 Synechococcus aureulum
1200 1000 800 600 400 200 0 400
450
500
550
600
650
700
750
800
Wavelength [nm]
Figure 12. LIF spectra of different Cianobacteria cultures.
Platimonas suecica Dunaliella terticolata Clamydomonas sp. Pheodactilum triconutum Nizschia Closterium
LIF intensity [rel. un.]
500
400
300
200
100
0 6500
6600
6700
6800
6900
7000
7100
7200
Wavelength [Å]
Figure 13. High resolution LIF spectra of different chlorophyceae and bacillarophyceae cultures. The relevant cell numbers and pigment content are given in Table 6.
186
A. Palucci
4.3 Pigment Concentration After normalization to the water Raman peak at 402 nm, the correlation of red LIF signals with cell number and chl-a concentration has been investigated on diluted solutions. Results are shown in Figure 14 (top) for dinophyceae and (bottom) for most of the examined chlorophyceae and one bacillarophycea, all diluted in synthetic seawater. Analogous results, having
0,5
F683 [Raman un.]
0,4 Prorocentrum minimum Prorocentrum micans Alexandrium lusitanicum Alexandrium tamarense
0,3
0,2
0,1
(a) 0,0 0
100
200
300
400
500
Chl-a conc. [µg/l]
F683 [Raman un.]
0,20 Platimonas suecica Phaeodactilum tricornutum Clamydomonas sp. Dunaliella terctiolecta
0,15
0,10
0,05 (b) 0,00 0
2500
5000
7500
Chl-a conc. [µg/l]
Figure 14. LIF bands intensity at 683 nm vs relevant Chl-a pigment content in: (top) dinophyceae; (bottom) chlorophyceae and bacillarophyceae.
Characterization of Phytoplankton
187
slightly worse statistics, have been obtained after dilution in natural seawater. For the dinophyceae, the red chlorophyll integral has been calculated at 683 nm, while in the case of the other classes this integral (FChl) is centered at the corresponding position of the maximum (680, 683 or 685 nm depending on the family). Both figures show a linear dependence of Chl-a LIF intensity with concentration in all cases. The data also show clearly that Chl-a concentration as low as 10 µg/l can be detected even in a small volume cuvette without optimizing either the optical path or the angular acceptance of the collection optics. Solutions with known dilution factors also have been prepared in synthetic seawater from a highly concentrated sample of Synechococcus leopoliensis. These solutions have been utilized to check the linearity of the LIF intensity measured on the peak corresponding to each pigment (i.e. F580, F660, F680) with its concentration in Raman units. For this Cianobacterium, results similar to those reported in Figure 14 for chlorophyll, have been obtained also in the case of phycoerythrin and phycocyanin contents.
4.4 Pigment Ratios To demonstrate the possibility of local and remote recognition of different algae families, the spectral ratios obtained from the Fλ measurements taken at four different wavelengths over the total visible intensity Ftot, are shown in Figure 15. These LIF ratios might be directly related to pigment ratios once the spectral response of the receiving subsystem has been calibrated as a function of the wavelength. As expected, due to the unresolvable overlap of carotenoids with DOM emission and to the non-negligible terrain contribution, no clear information can be deduced from the blue channel F460, while the red channels’ intensity correspond to the relevant pigment content. In fact, the chlorophyll relative intensity (labeled as F680) appears to provide an unambiguous classification in the main families. Dinophyceae (samples identified as N = 1 to 4 are Pr. micans, Pr. minimum, Al. tamarense and Al. lusitanicum, respectively) have the highest chlorophyll red emission, followed by Chlorophyceae (samples identified as N = 5 to 7 are Pl. suecica, Dun. terctiolecta and Clamydomonas sp., respectively). Extremely low chlorophyll emission is found for all bacillarophyceae, especially for Nitzschiaceae (samples identified as N = 8 to 11 are Ph. tricornutum, N. closterium, N. delicatissima, and Fragilaria sp., respectively). Cianobacteria species belonging to the Chroococcacaceae family (samples identified as N = 12 to 13 are Syn. leopoliensis and Syn. 625, respectively)
188
A. Palucci
show also a relatively low chlorophyll content, however they are unambiguously recognized by the presence of the intense phycoerythrin and phycocyanin emissions (F580 and F660, respectively), which do not occur in all the other classes. F460
F580
F660
F680
0,08
Fλ/Ftot
0,06
0,04
0,02
0,00 1
2
3
4
5
6
7
8
9
10
11
12
13
Samples N. Figure 15. Spectral ratios at different emission wavelengths for 14 different algae species: 14 Dinophyceae, 5-7 Chlorophyceae, 8-11 Bacillarophyceae, 12-13 Cianobacteria.
5.
DISCUSSION
The red LIF intensity of chlorophyll peaks around 680 nm, varies significantly by about two orders of magnitude among the different algal species, and correlates well with pigment concentration. A linear relationship has been found between F680 and chlorophyll concentration within a family. However, the calibration constant does depend on the considered algal species, namely some dinophyceae (the Alexandrium families) give rise to the highest LIF emission, F683 per chlorophyll unit. High intensities also occur for all the chlorophyceae considered, while bacillarophyceae emit much less in the red, with a minimum emission in the case of the Nitzschiaceae family. Data on cyanobacteria, also characterized by a weak chorophyll emission, were not reported since their absolute calibration in chl-a content was not available. Data confirm that it is not possible to ascertain the absolute value of chla concentration from only LIF data without a calibration on the same algal species, since the excitation and emission efficiency depend on the target biological system.
Characterization of Phytoplankton
189
Chl-a is the major fluorescence emitter near 680 nm. The overall chemical composition of photosynthetic pigments is eventually responsible also for re-absorption phenomena. In fact the lowest chlorophyll emission observed for chlorophyceae and bacillarophyceae with respect to the dinophyceae might be related to the abundance in the former classes of chlorophylls different from the fluorescing chl-a (Table 3). While dinophyceae, apart from chl-a contain only chl-c2, chlorophyceae contain also the chl-b and bacillarophyceae the chl-c1. The red absorption of the latter pigments could be responsible both for the low red chlorophyll emission and for characteristic wavelength shifts observed in chlorophyceae and bacillarophyceae. Furthermore, morphologic parameters, such as the presence of thick membranes or even of a semi-transparent theca, may affect strongly the LIF signal generated by a defined concentration of chl-a contained inside certain unicellular organisms. The comparison of the red pigments ratios with the LIF spectral ratios measured for the same species at the characteristic emission wavelengths reported in Figure 15 show that a straightforward assignment is not possible without the suggested construction of a laboratory library (SSP) of LIF data relevant to different algae families even when absorption characteristics are considered.28
6.
CONCLUSIONS
Present data taken on phytoplankton cultures show that their visible emission spectra measured upon near-UV excitation of seawater contains information relevant to the remote identification of various algal families characterized by different fluorescing pigments. This identification requires a careful laboratory characterization, taking into account the fluorescence of the growth terrain, for many different species. However, fluorescence information, together with scattering and shape data, can be valuable techniques to provide remote (lidar fluorosensors) and local (flow cytometry) mappings of the distribution of various algal species in their natural environment.
ACKNOWLEDGEMENTS The contribution of C. Micheli is kindly acknowledged for her precious help in biological laboratory support in growing, characterizing and analyzing the different algae species and their pigment content. A special
190
A. Palucci
thank is addressed to my colleagues R. Fantoni, L. Fiorani, A. Lai and C. Micheli for suggestions and encouragements. The author is deeply grateful to the book reviewers A. Hoekstra and G. Videen in their constant encouragement.
REFERENCES 1.
2. 3.
4.
5.
6.
7. 8.
9. 10. 11. 12. 13. 14. 15.
Barbini R., Fantoni R., Palucci A., Ribezzo S., Van der Steen H.J.L., 1992, Spectral and time resolved measurements of pollutants on water surface by a XeCl laser fluorosensor. EARSel Adv. in Remote Sensing. 1: 46-51. Barbini R., Fantoni R., Colao F., Palucci A., Ribezzo S., 1999, Shipborne laser remote sensing of the Venice lagoon. Int. Jou. Remote Sensing, 20, N° 12, 2405-2421. Barbini R., Colao F., Fantoni R., Palucci A., Ribezzo S., Micheli C., 1998, Design and application of a lidar fluorosensor system for remote monitoring of phytoplankton. ICES Journal of Marine Science, 55: 793-802. Barbini R., Colao F., Fantoni R., Palucci A., Ribezzo S., Micheli C., 1997, Laboratory LIF characterization of different phytoplankton species originating harmful blooms. ENEA RT/INN/97/21. In English available upon request. Barbini R., Fantoni R., Colao F., Palucci A., Ribezzo S., 2001, Differential lidar fluorosensor system used for phytoplankton Bloom and sea water quality monitoring in Antarctica. Int. Jou. of Remote Sensing, 22, N°. 2 & 3, 369-384. Barbini R., Colao F., Fantoni R., Fiorani L., Galli M., Palucci A., Artamonov. E.S., 2003, Remotely sensed primary production in the western Ross Sea: results of in situ tuned models. Antarctic Science, 15, Issue 01. March 2003. pp 77-84. Maltsev V.P., 2000, Scanning flow cytometry for individual particle analysis. Review of Scientific Instruments.. 71, 243-255. Barnaba, F. Fiorani L., Palucci A., Tarasov P., 2005, ENEA laser scanning flow cytometer (CLASS) for optical characterization of marine particles. NATO Advanced Research Workshop on Optics of Biological Particles, Novosibirsk, Russian Federation, Oct. 3-6. Falkowski P.G., Raven, J.A., 1997, Aquatic photosynthesis. Blackwell Science ISBN 086542-387-3. Cerullo M.M., Curtsinger B., 1999, Sea Soup: Phytoplankton. Tilbury House Publishers ISBN: 0884482081. Kirk, J.T.O., 1994, Light and Photosynthesis in Aquatic Ecosystems. Cambridge: Cambridge University Press, ISBN 0-521-45966-4. Sournia, 1979, Phytoplankton Manual (Monographs on oceanographic methodology) Unipub eds. ISBN: 9231015729. Raymont, 1980, Plankton and Productivity in the Ocean, Pergamon International Library of Science, Technology, ISBN: 0080215521. Dring M.J., 1982, Biology of Marine Plants. Edward Arnold Publisher, ISBN: 0713128607. Phytoplankton pictures collection. Courtesy of Biology Department of Raymond Walters College http://biology.rwc.uc.edu/HomePage/BWS/planktonkey/bluegeen.html
Characterization of Phytoplankton
191
16. Saggiomo V., Carrada G.C., Mangoni O., Ribera d’Alcala M., Russo A., 1998, Spatial and temporal variability of size-fractionated biomass and primary production in the Ross Sea Antarctica during austral spring and summer. Journal of Marine Systems 17, 115-127. 17. Umani S.F., Accornero A., Budillon G., Capello M., Tucci S., Cabrini M., Negro P.D., Monti M., De Vittor C., 2002, Particulate matter and plankton dynamics in the Ross Sea Polynya of Terra Nova Bay during the Austral Summer 1997/98. Journal of Marine Systems 36, 29-49. 18. Lichtenthaler H.K. and Rinderle U., 1988, The role of chlorophyll fluorescence in the detection of stress conditions in plants. CRC in Analytical Chemistry, 19, S25-S85. 19. Bristow M, Nielsen D., Bundy D., Furtek R., 1981, Use of water Raman emission to correct airborne laser fluorosensor data for effect of water optical attenuation. Appl. Optics 20, 2889-2906. 20. Reuter R., Diebel D., Hengstermann T., 1993, Oceanographic laser remote sensing: measurements of hydrographic fronts in the German Bight and in the northern Adriatic sea. Int. J. Remote Sensing 14, 823-848. 21. Patsayeva S.V., 1995, New methodological aspect of the old problem. Laser diagnostic of dissolved organic matter. EARSel Adv. in Remote Sensing, 3, 66-70. 22. Hoge F.E. and Swift R.N., 1981, Absolute tracer dye concentration using airborne laser-induced water Raman backscatter. Appl. Optics 20, 1191-1202. 23. Ikeya T., Ohki K., Takahashi M., Fujita Y., 1994, Photosynthetic characteristics of marine Synechococcus spp. with special reference to light environments near the bottom of the euphotic zone of the open ocean. Marine Biology 118, 215-221. 24. Shreve A.P., Trautman J.K., Owens T.G. and Albrecht A.C., 1991, A femtosecond study of electronic state dynamics of fucoxanthin and implications for photosynthetic carotenoid-to-chlorophyll energy transfer mechanisms. Chem. Phys. 154, 171-178. 25. Mimuro M., Nagashima U., Nagaoka S., Takaichi S., Yamazaki I., Nishimura Y., Katoh T., 1993, Direct measurement of the low-lying singlet excited (21Ag)state of a linear carotenoid, neurosporene, in solution. Chem. Phys. Lett. 204, 101-105. 26. Ogata T., Komada M., Nomura S., Kobayashi M., Nozawa T., Katoh T., Mimuro M., 1994, A novel peridin-chlorophyll a protein (PCP) from the marine dinoflagellate Alexandrium cohorticula: a high pigment content and plural spectral forms of piridinin and chlorophyll a. FEBS Letters 356, 367-371. 27. Measures R.M., 1984, Laser Remote Sensing J. Wiley & Sons New York. 28. Lazzara L, Bricaud A, Claustre H., 1996. Spectral absorption and fluorescence excitation properties of phytoplanktonic populations at a mesotrophic and an oligotrophic site in the tropical North Atlantic (EUMELI program). Deep-Sea Research Part I-Oceanographic Research Papers 43 (8): 1215-1240.
192
A. Palucci
Antonio Palucci (right) toasting with comrades.
ASTRO-BIOLOGICAL SIGNATURES Prospects for the detection of non-terrestrial biological material T.M. Gledhill1, W.B. Sparks2, Z. Ulanowski1, J.H. Hough1 and S. DasSarma3 1
Science and Technology Research Institute, University of Hertfordshire, College Lane, Hatfield, AL10 9AB, UK 2 Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA 3 University of Maryland Biotechnology Institute, Center of Marine Biotechnology, 701 East Pratt Street, Baltimore, MD 21202
Abstract:
The possibilities for detecting non-terrestrial biological material are briefly reviewed in light of the flurry of extrasolar planet detections over the last ten years and the forthcoming space missions designed to search for signs of extraterrestrial life. We summarize the evidence for non-terrestrial organic material and the challenges involved in the development of robust and reliable biosignatures, i.e. features whose presence requires a biological origin. It is necessary to distinguish between prebiotic material, which is likely to be a necessary precursor to life but which is not produced by a biological system, and the detection of either biological material itself or its effect on the environment. We discuss the use of chirality and spectrally resolved circular polarization as a remote-sensing technique for distinguishing between biological and non-biological substances.
Key words:
Astro-biology, chirality, circular polarization
1.
INTRODUCTION
The detection of non-terrestrial biological material is complicated by the lack of a strict definition of what is meant by “biological.” However, we do know that life on Earth has a carbon-based organic chemistry, requiring liquid water as a solvent to mediate reactions, and so it is reasonable to search for evidence of similar systems elsewhere. Although other 193 A. Hoekstra et al. (eds.), Optics of Biological Particles, 193–211. © 2007 Springer.
194
T.M. Gledhill et al.
biochemistries have been suggested, there are good reasons to suppose that a carbon/water-based biochemistry may be universal (Chyba & Hand 2005). Carbon is a by-product of stellar evolution and is returned to the Galaxy in huge quantities in the winds of giant stars. To date, almost 100 species of carbon-based organic molecules have been detected in a range of astrophysical environments, from nearby star-forming clouds to distant galaxies (Ehrenfreund et al., 2002), suggesting that the building blocks of life are widely available. Studies of chondritic meteorites, dating from the earliest phases of the solar system, show evidence for complex organics including amino-acids (e.g. Cronin et al., 1981). This material will have been delivered to the early Earth, as well as to other planets, potentially kick-starting the biological process at more than one site in the solar system. Although Mars appears largely desiccated today (however see Murray et al., 2005 for recent evidence of a frozen sea close to Mars' equator), there is abundant evidence that surface water has existed there in the geologically recent past (Baker 2001), raising the possibility that microbial life may have existed at some point. Another site of considerable interest from a biological point of view is Europa, a moon of Jupiter, which is thought to possess an ocean of liquid water below its icy surface (Carr et al., 1998; Kivelson et al., 2000). The tenacity of life and its ability to flourish in extreme environments, often without access to solar radiation, is demonstrated by communities of extremophile microorganisms on Earth (Rothschild & Mancinelli 2001). Many of these extremophiles have been found to belong to an ancient branch of life, dubbed “Archaea,” which have fundamental similarities to higher organisms (Woese 1987). Extremophilic microorganisms have been found in some of the harshest environments, such as in hydrothermal vents above 100 oC, frozen ice cores, permanently saturated sodium chloride lakes, and highly acidic and alkaline waters, greatly expanding our definition of the habitable zone on our planet. Notable examples include the large extremophile populations discovered in deep anoxic basins (van der Wielen et al., 2005) and in sub-seafloor sediments (Schippers et al., 2005). Techniques for detecting biological material are especially important at a time when space missions are currently investigating the most likely sites for life elsewhere in the solar system. The STARDUST mission, to capture cometary material for analysis, returned safely to Earth in January 2006 after an encounter with comet Wild 2. It is also only ten years ago that the first detection was made of an extrasolar planet orbiting a main sequence star (Mayor & Queloz 1995), opening up the field of extrasolar planetary research. Within the next few years, with the launch of new planet-finding space missions, such as KEPLER (Borucki et al., 1996) and COROT (Deleuil et al., 1996), we will have, for the first time in our history, a good
Astro-biological Signatures
195
idea of the number of Earth-like planets (ELPs) in our Galaxy. These ELPs will then be investigated by the Terrestrial Planet Finder (NASA) and Darwin (ESA) space missions, due for launch in the next decade, which are intended to search directly for evidence of habitability and the presence of life.
2.
PREBIOTIC MATERIAL
Spectral-line studies of circumstellar environments have identified many complex organic molecules (Friedel et al., 2005, Ehrenfreund et al., 2002, White 2004). The detection of glycolaldehyde (Hollis et al., 2000; Halfen et al., 2006) is of particular biological significance, since it is the simplest monosaccharide sugar and may be a key stage in the formation of ribose. Acetic acid, which may be involved in the synthesis of glycine, the simplest amino acid, has been detected towards Sgr B2 (Mehringer et al., 1997). Searches for interstellar glycine in the gas phase have so far been inconclusive. Kuan et al., (2003a) reported a tentative detection of glycine towards three hot molecular cloud cores, but this has since been disputed by Snyder et al., (2005). Attempts to detect other prebiotic materials, such as pyrimidine (a component of nucleic acids) in the interstellar medium (ISM) have also so far proved to be unsuccessful (Kuan et al., 2003b). The largest carbon molecules in the ISM are likely to be polycylic aromatic hydrocarbons (PAHs), w+hich are thought to be responsible for a range of infrared interstellar emission features (Léger & Puget 1984). The various emission features (principally at 3.3, 6.2, 7.7, 8.6 and 11.2 µm) are due to stretching and bending modes of the C-C and C-H bonds, although there has not been, as yet, a conclusive identification of a specific PAH with an astronomical spectral feature. In an attempt to fit the 6.2 µm interstellar emission band, Hudgins et al., (2005) conclude that PAHs with a nitrogen atom substituted into the structure are required, forming endoskeletal PANHs (polycyclic aromatic nitrogen heterocycles). Although PAHs do not play a role in terrestrial biology, PANHs are possible precursors of a prebiotic chemistry and bear some similarities with biologically important molecules, such as chlorophyll. Most complex organic molecules are expected to form more readily on the surface of dust grains, rather than in the gas-phase ISM. UV photolysis of interstellar ice analogues in the laboratory, under conditions that are thought to be typical of the molecular cloud from which the solar system condensed, has been shown to produce amino acids such as glycine, alanine and serine, after hydrolysis (Bernstein et al., 2002). Muñoz Caro et al., (2002) identify 16 amino acids in their UV-irradiated ice residue. In both
196
T.M. Gledhill et al.
cases, the chiral amino acids appear to be racemic (equal numbers of the left and right handed enantiomers are produced). Prebiotic molecules produced in the solar nebula will have been incorporated into primitive solar system bodies such as comets and asteroids and may have been delivered to the early Earth during the heavy bombardment phase (Chyba et al., 1990). Direct evidence for this process is provided by the detection of indigenous amino acids in the Murchison meteorite, a carbonaceous chondrite that fell to Earth in 1969 (Kvenvolden et al., 1970). More than 70 amino acids have been identified in Murchison, over 50 of which are not involved in biology on Earth. Although the initial analysis suggested a small excess of L-enantiomers, which was attributed to terrestrial contamination, Engel & Nagy (1982) reported a larger L-enantiomer excess for several amino acids, for which they claimed an extraterrestrial origin. This has been disputed by Bada et al., (1983), who attributed the result to contamination, as in the original study. To avoid the contamination issue, Cronin & Pizzarello (1997) analyzed 4 of the non-terrestrial amino acids and again detected an L-enantiomer excess, of between 7.0 and 9.1 per cent, concluding that, at least in the case of some of the Murchison amino acids, the L-excess must be indigenous to the meteorite. The issue is an important one, since it is unclear whether the origin of homochirality in terrestrial biology (with nearly all amino acids being L-enantiomers and sugars Denantiomers) predates life, or is a consequence of it.
Figure 1. Circular polarization image of the OMC-1 star-forming region in Orion at 2.2 µm, from Bailey et al. (1998). Total intensity is shown on the left with the bright BecklinNeugebauer (BN) object at coordinate (0,0). Percentage circular polarization is shown on the right, ranging from -5% (black) to +17% (white). A region of high (up to 17%) circular polarization is seen to the west of the BN object at coordinates (-20, -7).
Astro-biological Signatures
197
It has been shown in the laboratory that irradiation of racemic substances with circularly polarized light can result in enantiomeric excesses (Bonner 1991). A mechanism for UV photolysis by circularly polarized light in star-forming regions was proposed by Bailey et al. (1998), following their discovery of high degrees of circular polarization (up to 17% at 2.2 µm) in Orion (see Fig. 1). They argued that this could account for the chiral asymmetry of the amino acids in the Murchison meteorite, as well as explaining the origin of homochirality in terrestrial biological molecules. If enantiomeric excess can be produced abiotically in this way, then the usefulness of chirality as a biosignature may be complicated (Section 5).
3.
EXTRASOLAR PLANETS
The prime motivation for the development of remote-sensing techniques, designed to identify biosignatures, is provided by the discovery of more than 170 extrasolar planets during the last 10 years (see Marcy et al., 2005 for a review). To date, nearly all of these detections have been made indirectly using the radial velocity technique, which measures the Doppler shift in the spectral lines of the star as it orbits about the gravitational centre of the star-planet system. The magnitude of the effect depends upon the orientation of the planet's orbit relative to our line of sight, and so the apparent mass, Msini, is measured, where i is the inclination angle and M the actual planetary mass. Most of the discovered planets are gas giants, with Msini in the range 0.1 to 13 Jupiter masses (1 Jupiter mass is equivalent to 318 Earth masses), orbiting close to their parent stars. As the precision of the Doppler radial velocity technique has improved, this mass limit has been pushed down, and a number of Neptune-mass planets (Msini in the range 14 to 21 Earth masses) are now known (Santos et al., 2004; Butler et al., 2004; McArthur et al., 2004). More recently, the first likely “super-Earth” has been detected, orbiting a nearby (4.7 pc) star (Rivera et al., 2005). This planet is part of a multiple system already known to contain other Jupiter-mass planets, and by making assumptions regarding the coplanarity of the orbits, Rivera et al., determine a mass of 7.5 Earth-masses. The mass distribution of the planets detected so far can be approximated by an inverse power law, dN/dM ∝ M-1.05 (Marcy et al., 2005), which rises steeply toward the lower mass limit, suggesting that there may be a potentially large number of Earth-mass planets awaiting discovery. This possibility is further supported by the very recent report of a 5.5 Earth-mass planet orbiting a cool M-dwarf star, discovered through gravitational microlensing (Beaulieu et al., 2006).
198
T.M. Gledhill et al.
The radial velocity technique is unlikely to be able to detect Earth-mass planets around solar-type stars. The Doppler precision required to detect an Earth orbiting at 1 au around a solar-mass star is 3 cm s-1 (Tarter 2001); whereas, limitations due to line broadening in the stellar spectrum are likely to restrict accuracy to not much below 1 m s-1. Gravitational microlensing surveys, on the other hand, are well-suited to detecting Earth-mass planets with orbital radii between 1 and 5 au. Statistically, however, planets discovered using this technique are likely to be much too distant to follow up with direct detections (the recent discovery by Beaulieu et al. (2006) lies at a distance of 6.6 kpc). An alternative technique is to use precision astrometry to measure the stellar wobble due to an unseen orbiting planet. The ESA GAIA mission, due for launch in 2011, should be capable of discovering thousands of planets around nearby main sequence stars (Perryman et al., 2001). The SIM Planetquest mission, scheduled for launch next decade, is expected to be capable of detecting planets with mass greater than 3 Earth-masses orbiting between 0.1 and 2 au around nearby (<20 pc) stars (Marcy et al., 2005). This range of orbital radii is important for the future detection of ELPs, since it encompasses the habitable zone (HZ) for solar-type stars, the region around the star in which life-supporting planets might exist. The HZ is usually defined by the requirement that liquid water should exist on the planet's surface and, for our own solar system, a conservative estimate of its width is from 0.95 au to 1.37 au (Kasting et al., 1993). For cooler K and M spectral-type stars, the HZ occurs further in and is narrower. For hotter F type stars, it occurs further out and is wider. In a study of the development of photosynthesis, Wolstencroft & Raven (2002a) conclude that an F0V star provides the best environment for an ELP. Stars hotter than F0 will have main sequence lifetimes less than 2 Gy and so these systems may be less favourable to the evolution of life. For planets orbiting in a plane close to our line of sight to the star, there is a possibility that the planet will transit the star, causing an occultation. The first transiting extrasolar planet was discovered in 2000 (Henry et al., 2000, Charbonneau et al., 2000, Mazeh et al., 2000) around the star HD209458, and a total of 7 are known so far, 5 of which were discovered using their transits. As well as providing information on the orbit, mass and radius of the planet, transits can also allow the planetary atmosphere to be studied, as the background stellar light is transmitted through it. In the case of the planet around HD209458, several transits have been observed and absorption lines attributable to atomic sodium and hydrogen have been detected, probing both the lower and upper atmosphere (Charbonneau et al., 2002; Vidal-Madjar et al., 2003). Most recently both carbon and oxygen have been detected in the upper atmosphere (Vidal-Madjar et al., 2004), the
Astro-biological Signatures
199
first detection of these biologically important elements in an extrasolar planet atmosphere. However, HD209458b is not an ELP but rather a gas giant, orbiting extremely close to its star (within 8.5 stellar radii). The hot, extended atmosphere is in the process of being evaporated. Recently, Deming et al. (2005), using Spitzer observations, report the detection of mid-infrared radiation from HD209458b, by observing the reduction in flux during secondary eclipse, when the planet passes behind the star.
4.
POTENTIAL BIOSIGNATURES
A major goal of the TPF (Beichman et al., 1999) and Darwin (Léger et al., 1996) missions is to search for habitable planets around a large number of nearby solar-type stars and to detect spectroscopic evidence for life. The first objective involves determination of the orbital and physical properties of the planet to assess habitability. If the planet is located within the star's HZ, with a modest orbital eccentricity, then liquid water can exist at the surface. In order to maintain a constant surface temperature over long periods, however, it is likely that a climate stabilization mechanism similar to the carbon-silicate weathering cycle on Earth will be necessary. This negative feedback loop is capable of reversing periods of global glaciation as well as preventing a runaway H2O greenhouse (e.g. Kasting & Catling 2003) and would have been particularly important in early Earth history when the Sun was less luminous than today. The process requires long-term tectonic activity so that it is likely that any ELP will be volcanically active, meaning that planetary mass is important. The lack of long-term volcanic activity on Mars, due to its small mass (one ninth that of Earth) and short cooling timescale, probably contributed to the loss of its atmosphere. The second objective, for the spectroscopic detection of life, necessitates the establishment of reliable biosignatures.
4.1 Atmospheric Biosignatures The detection of an atmospheric biosignature would provide indirect evidence for life, as long as non-biological processes can be ruled out. The best candidate is usually considered to be O2, and its photolytic product O3. Although O2 can be produced non-biologically by the photodissociation of water, this is not likely to be a long-term source unless the planet is in a runaway greenhouse phase, when all of the water may be evaporated into the atmosphere. This situation is thought to have led to complete water loss on Venus and defines the inner edge of the HZ. However, large amounts of
200
T.M. Gledhill et al.
atmospheric O2 on an ELP with a reducing crust and surface temperature compatible with liquid water, may be a strong indicator of continuous production and, hence, biological (photosynthetic) activity. Léger et al. (1993) calculate that if photosynthetic production of O2 were to cease on Earth, then O2 would be removed from the atmosphere within 2·107 yrs. Although O2 has an absorption band at 760 nm, this would be difficult to observe unless the planet eclipsed the star. The contrast ratio between a solar-type star and an ELP varies from about 109 in the visible to 106 in the mid-infrared (Angel et al., 1986) so that observations of the O3 absorption feature at 9.7 µm are likely to be more feasible. The photochemical production and destruction of O3 are considered by Léger et al. (1993) and the process is non-linear. The relatively large O3 column density at low O2 partial pressures makes O3 a potentially sensitive tracer of O2; however, the non-linearity in the relationship means that it is difficult to determine the actual amount of O2 present. The presence of O3 in the atmosphere is in itself a necessity if surface dwelling organisms are to exist, since it provides an essential screen against the damaging effects of UV radiation. In marine environments on Earth, phytoplankton are significantly inhibited by solar radiation in the top 5-10 m or more of the water column (Herndl et al., 1993). In order to use light as an energy source, organisms need to be on the surface or in less than 300 m depth of water (Wolstencroft & Raven 2002a). A potential difficulty in using O3 as a biosignature is that its detection may be complicated by the existence of other molecules with spectral features around 9.7 µm, such as NH3, which absorbs in the 9-11 µm region, as seen in the spectra of Jupiter and Saturn (Léger et al., 1993). Although the detection of large amounts of O2 in the atmosphere of an ELP would be a strong indicator of oxygenic photosynthesis, and hence life, a non-detection would not prove the absence of life. Photosynthesis on Earth is thought to have evolved by at least 2.7 Gya, based on evidence for cyanobacteria in rocks from the Pilbarra (Brocks et al., 1999), but global oxygenation did not occur until 2.3 Gya, representing a 400 million year delay. Life may have formed long before the advent of photosynthesis, perhaps in the early Archean, approximately 3-3.5 Gya (Chyba & Hand 2005), although this is still controversial (e.g. Moorbath 2005). If so, then the Earth would not have had an O2 biosignature for the first 0.7-1.2 Gy of its biological history. This has led some authors to consider biosignatures based on microbial ecosystems, thought to be characteristic of Earth's earliest biosphere (e.g. Pilcher 2003). Other atmospheric gases, such as H2O and CO2, although not biosignatures, are important to biological systems. The presence of liquid water on a planet is considered to be a requirement for life; whereas, CO2 is required for photosynthesis, and both have been designated as high priority
Astro-biological Signatures
201
observations for TPF/Darwin (des Marais et al., 2002). CO2 has an absorption band at 15 µm and may be a good indicator of a terrestrial planet. A detection of methane (CH4) on an ELP would be interesting, since it may have been the dominant greenhouse gas in the Archean atmosphere, when CO2 concentrations are thought to have been low (Kasting & Catling 2003). However, it can be produced both biologically, by methanogenic bacteria, and non-biologically in mid-ocean ridge systems, and so is ambiguous as a biosignature.
4.2 Surface Biosignatures Whereas atmospheric biosignatures are indirect indicators of the presence of life, remote sensing of surface features may provide direct confirmation of extraterrestrial organisms. Such a feature could be provided by an abrupt change in reflectance of the planet's surface as a function of wavelength. The primary photosynthetic molecule, chlorophyll, has absorption peaks at 430 nm and 662 nm (chlorophyll A), giving vegetation its characteristic green colour. Between 700 and 1400 nm, green vegetation shows a large and abrupt increase in reflectance (from ~5% at 670 nm to ~70% at 800 nm) giving rise to a "red edge" in the reflectance spectra around 700 nm (e.g. Arnold et al., 2002). This red edge, due to vegetation, has been detected in observations of the Earthshine spectrum (see Fig. 2), which use the moon as a diffuse reflector by which to observe the integrated spectrum of the illuminated Earth (Arnold et al., 2002; Woolf et al., 2002; Seager et al., 2005). These observations detect the red edge feature at the 4% to 10% level and seem to confirm its potential use as a biosignature. The red edge feature has also been detected in low-resolution spectra of the Earth taken from the Galileo spacecraft (Sagan et al., 1993). Absorption features due to ozone, oxygen and water vapour, as well as the rise in the blue part of the spectrum due to Rayleigh scattering in the atmosphere are also seen. Although in a single plant leaf the red edge is very prominent, in the integrated Earthshine spectrum it is only present at the few-percent level and may be very difficult to detect in the spectrum of an extrasolar planet, especially when unknown atmospheric features are present. The strength of the red edge is expected to be variable, however, since it will depend on the degree of cloud cover and the area of vegetation being illuminated, and this may help matters. Diurnal variations as an ELP rotates and vegetated areas come into view, may be easier to detect, as well as seasonal changes occurring during the planet's orbit (Seager et al., 2005). The use of the red edge as an extrasolar biosignature also assumes that similar photosynthetic structures and pigments to those on Earth will have evolved elsewhere. The
202
T.M. Gledhill et al.
increase in reflectance at 700 nm, giving rise to the red edge, is largely due to the cell structure in the leaves of terrestrial vegetation, which scatter light longward of this wavelength very efficiently. This may be an evolutionary mechanism to harvest light at shorter wavelengths whilst simultaneously keeping the leaf structure cool by not absorbing at longer wavelengths. If so, then this could be an essential strategy for any land-based vegetation which might, therefore, readily evolve elsewhere (Seager et al., 2005).
Figure 2. Earthshine observations (adapted from Seager et al., 2005) corresponding to vegetated (top) and non-vegetated regions of the Earth. A tentative detection of the red-edge feature is seen in the top spectrum when a vegetated area of the Earth (South America) is illuminated.
Approximately half of all photosynthesis on Earth is carried out by oceanic phytoplankton (Field et al., 1998), which does not, in general, exhibit a red-edge feature. Neither do all photosynthetic pigments have absorption bands in the same wavelength range as the chlorophyll A and B used by land plants. For example, purple bacteria use bacteriochlorophyll A and B, which have absorption peaks longward of 700 nm. A different lightharvesting system is used for phototrophic growth by halophilic Archaea, which are salt-loving extremophilic microorganisms, robust against UV radiation and tolerant of desiccation. They use a light-harvesting system featuring the retinal-based chemical bacteriorhodopsin, which has a broad absoption peak centred around the solar maximum of 550 nm. Interestingly,
Astro-biological Signatures
203
chlorophyll absorbs light either side of this peak, suggesting the possibility of evolutionary competition between the two, and that both systems may have been present in quantity throughout time (DasSarma 2004). The tolerance shown by halophilic Archaea for harsh cold, dry, and salty conditions makes this species a prime candidate for life on Mars, should it exist at the present time (Landis 2001). Recently, retinal proteins similar to bacteriorhodopsin have been discovered in common planktonic bacteria in the oceans, which are spectrally tuned depending on their depth in the water column (Man et al., 2003). The rapid appearance of microbial life on Earth, and the lengthy period over which it dominated, renders it statistically likely that if extrasolar terrestrial counterparts are found, then they too may well be inhabited and even dominated by microbial life. However, the spectral detection of microbial life is likely to prove even more difficult than red-edge detection of vegetation (Knacke 2003), and will be a challenge for the post TPF/Darwin era.
5.
CHIRAL SIGNATURES
A characteristic of life is that it is predominantly homochiral. The vast majority of organisms use only left-handed L-amino acids in proteins and right-handed D-sugars in nucleic acids. This is likely to be a necessity for self-replication in which case it will characterize all biochemical life (Popa 2004). As mentioned in Section 2, analysis of the Murchison meteorite has shown L-excesses of up to 9% for a number of α-methyl amino acids (Cronin & Pizzarello 1997), with slightly smaller excesses found in the Murray meteorite (Pizzarello & Cronin 2000). This suggests that homochirality may result from the delivery to Earth of pre-biotic matter with an in-built enantiomeric excess (Bailey et al., 1998); alternatively, it may arise from a surface interaction that favours one handedness over the other. Nevertheless, chirality is a property of terrestrial life, and studies of chirality have been proposed as a robust means of searching for life that may even differ from terrestrial life. Organic material displays optical activity, arising from differential absorption (circular dichroism) or scattering of left and right circularly polarized light by its component chiral molecules (Pospergelis 1969, Wolstencroft 1974, Wolstencroft et al., 2002b). Circular dichroism spectroscopy is a standard analysis technique for studying protein structure and function (Purdie & Brittain 1993, Kelly & Price 2000). Chlorophyll, for example, induces 0.1–1% circular dichroism in its absorption bands (Houssier & Sauer 1970), as shown in Fig. 3. The spectra of intact chloroplasts from plant leaves show additional peaks, probably due to the
204
T.M. Gledhill et al.
presence of other chromophores such as carotenoids (Faludi-Daniel et al., 1973). Circular Intensity Differential Scattering (CIDS) considers the circular polarization induced by scattering from chiral samples and has been studied by Bustamante et al. (1985). Circular polarization also can be caused by optical interaction associated with the chirality of subcellular structures, such as membranes and macromolecules (Bustamante et al., 1983; Finzi et al., 1985). Light emitted through fluorescent processes, commonly associated with molecules of biological significance, also exhibits circular polarization (Steinberg 1978; Riehl & Richardson 1986). Microbes typically produce fractional polarizations ~10-2 to 10-3 (Salzman & Gregg 1984). It is clear that the interaction between light and living organisms can produce, in a wide variety of ways, a potentially detectable signal ~0.1% or more, in the form of circular polarization. While ocean water containing phytoplankton has been reported to produce fractional polarizations of the order of 10-4 at 488 nm (Voss et al., 1984), values near absorption bands may well be considerably higher. Circular dichroism and related phenomena in photosynthesis can be induced by the intrinsic chirality of the molecules, by excitonic coupling between chromophores in molecular complexes and by macroscopic organization of the system (Garab 1996). The spectra of chlorophyll and bacteriorhodopsin show a substantial circular dichroism signal as well as a distinctive sign-flip through the absorption bands (Fig. 3).
Figure 3. (a) Absorption spectrum of chlorophyll (solid) and circular dichroism spectrum (dashed) adapted from Houssier & Sauer (1970). (b) Absorption spectrum (solid) and circular dichroism spectrum (dashed) of bacteriorhodopsin, adapted from Becher & Cassim (1977).
Non-biological contributions to circular polarization can arise from atmospheric and aerosol scattering and from mineralogical scattering processes. While some minerals are also optically active, integrating over a
Astro-biological Signatures
205
naturally produced sample is expected to give equal fractions of enantiomorphs, which average to zero in their optical activity. Scattering can be a source of circular polarization if the scattering object itself has a chiral geometry. Likewise, chiral asymmetry in the scattering system geometry, for example associated with preferential alignment of particles, can lead to artifacts, as can birefringence in the detection optics (Francke et al., 1994, Riehl & Richardson 1986, Rosen & Pendleton 1995). Multiple reflection and phase effects can also introduce circular polarization, but would produce a smooth distribution recognizably related to the geometry of the scattering in the case of spatially resolved solar system observations (Bandermann et al., 1972, Degtjarev & Kolokolova 1992). Empirically, solar system measurements of circular polarization consistently find very low fractional circular polarization levels, ~10-4 to 10–5 (Kemp et al., 1971, Swedlund et al., 1973, Meierhenrich et al., 2002), some two or three orders of magnitude smaller than likely organic effects due to pure biological samples. Imaging circular polarimetry of the Martian surface at 210 km resolution (Sparks, Hough & Bergeron 2005) found upper limits to the polarization of ~10-3, consistent with globally averaged values. Even if local enantiomeric excesses of optically active minerals were to exist, it is expected that they would be distinguishable from circular polarization having an organic origin by their spectral properties (Pospergelis 1969, Kelly & Price 2000). The unique homochirality of biology can, therefore, potentially yield a macroscopic circular polarization signal especially in photosynthetic and phototrophic systems. From an astronomical perspective, these activities are extremely attractive, since they are surface phenomena and are likely to be tuned to the wavelength of maximum spectral flux of the host star (Wolstencroft & Raven 2002a). For the same reason they are also tuned to the wavelength of maximum transmission of the hypothetical planetary atmosphere. Both of these combine to maximize our ability to observe the source directly. Circular-polarization spectroscopy has the potential to distinguish between an organic-based absorption feature and a mineralogical one, thus helping to eliminate features that can mimic the chlorophyll edge, yielding potential false positives. However, very few laboratory studies of the circular polarization signatures of intact biological materials (i.e. in macroscopic or whole cell form) have been undertaken to date (Wolstencroft et al., 2002b). Due to the potential for false positives, carefully designed studies are necessary to ensure that robust techniques are developed. In addition to biological targets, it is essential to investigate inorganic samples, such as pure minerals and mineral mixes (such as sand
206
T.M. Gledhill et al.
and rock), to establish whether circular polarization allows discrimination between strong edge features in organic and inorganic substances. Nevertheless, it is encouraging that no false positives were found by Sparks et al. (2005) in their imaging circular polarimetry of the Martian surface. Finally, it is essential to carry out terrestrial field studies to provide the necessary validation of these concepts in practical remote sensing applications.
REFERENCES Angel J.R.P., Cheng A.Y.S., Woolf N.J., A space telescope for infrared spectroscopy of Earth-like planets, 1986, Nature, 322, 341-343. Arnold L., Gillet S., Lardière O., Riaud P., Schneider J., A test for the search for life on extrasolar planets: Looking for the terrestrial vegetation signature in the Earthshine spectrum, 2002, Astron. & Astrophys., 392, 231-237. Bader J.L., Cronin J.R., Ho M.-S., Kvenvolden K.A., Lawless J.G., Miller S.L., Oro J., Steinberg S., On the reported optical activity of amino acids in the Murchison meteorite, 1983, Nature, 301, 494-496 . Bailey J.A., Chrysostomou A., Hough J.H., Gledhill T.M., McCall A., Clark S., Ménard F., Tamura M., Circular polarization in star-formation regions: implications for biomolecular homochirality, 1998, Science, 281, 672-674. Baker V.R., Water and the Martian landscape, 2001, Nature, 412, 228-236. Bandermann L.W., Kemp J.C., Wolstencroft R.D., Circular polarization of light scattered from rough surfaces, 1972, Mon. Not. R. Astron. Soc., 158, 291-304. Beaulieu J.-P. et al., Discovery of a cool planet of 5.5 Earth masses through gravitational microlensing, 2006, Nature, 439, 437-440. Becher B., Cassim J.Y., Effects of bleaching and regeneration on the purple membrane structure of Halobacterium halobium, 1977, Biophys. J., 19, 285-297. Beichman C.A., Woolf N.J., Lindensmith C.A., The terrestrial planet finder: A NASA origins program to search for habitable planets, 1999, JPL publication 99-3. Bernstein M.P., Dworkin J.P., Sandford S.A., Cooper G.W., Allamandola L.J., Racemic amino acids from the ultraviolet photolysis of interstellar ice analogues, 2002, Nature, 416, 401-403. Bonner W.A., The origin and amplification of biomolecular chirality, 1991, Orig. Life, 21, 59-111. Borucki W.J., Kock D.G., Dunham E.W., Jenkins J.M., The Kepler Mission: A mission to determine the frequency of inner planets near the habitable zone of a wide range of stars, 1996, in Planets beyond our solar system and next generation space missions, STScI, Baltimore. Brocks J.J., Logan G.A., Buik R., Summons R.E., Archean molecular fossils and the early rise of eukaryotes, 1999, Science, 285, 1033-1036. Bustamante C., Tinoco I., Maestre M.F., Circular differential scattering can be an important part of the circular dichroism of macromolecules, 1983, Proc. Nat. Acad. Sci. 80, 35683572.
Astro-biological Signatures
207
Bustamante C., Maestre M.F., Keller D., Expressions for the interpretation of circular intensity differential scattering of chiral aggregates, 1985, Biopolymers, 24 (8), 15951612. Butler R.P. Vogt S.S., Marcy G.W., Fischer D.A., Wright J.T., Henry G.W., Laughlin G.L., Lissauer J.J., A Neptune-mass planet orbiting the nearby M dwarf GJ 436, 2004, ApJ., 617, 580-588. Carr M.H., et al., Evidence for a subsurface ocean on Europa, 1998, Nature, 391, 363-365. Charbonneau D., Brown T.M., Latham D.W., Mayor M., Detection of planetary transits across a Sun-like star, 2000, ApJ, 529, L45-L48. Charbonneau D., Brown T.M., Noyes R.W., Gilliland R.L., Detection of an extrasolar planet atmosphere, 2002, ApJ, 568, 377-384. Chyba C.F., Hand K.P., Astrobiology: The study of the living universe, 2005, Ann. Rev. Astron. Astrophys., 43, 31-74. Chyba C.F., Thomas P.J., Brookshaw L., Sagan C., Cometary delivery of organic molecules to the early earth, 1990, Science, 249, 366-373. Cronin J.R., Gandy W.E., Pizzarello S., Amino acids of the Murchison meteorite: I., 1981, Journal of Molecular Evolution, 17, 265-272. Cronin J.R., Pizzarello S., Enantiomeric excesses in meteoritic amino acids, 1997, Science, 275, 951-955. DasSarma, S., Genome sequence of an extremely halophilic archaeon, 2004, Microbial Genomes, ed. Fraser, C.M., Read, T., Nelson, K.E., Humana Press, Totowa, NJ, 383399. Degtjarev V.S., Kolokolova L.O., Possible application of circular polarization for remote sensing of cosmic bodies, 1992, Earth, Moon & Planets, 57, 213-223. Deleuil M., Barge P., Léger A., Schneider J, 1996, Detection of Earth-like planets with the COROT space mission, in Planets beyond the solar system and the next generation of space missions, STScI, Baltimore. Deming D., Seager S., Richardson L.J., Harrington J., Infrared radiation from an extrasolar planet, 2005, Nature, 434, 740-743. Des Marais D.J. et al., Remote sensing of planetary properties and biosignatures on extrasolar terrestrial planets, 2002, Astrobiology, Vol. 2, No. 2, 153-181. Ehrenfreund P. et al., Astrophysical and astrochemical insights into the origin of life, 2002, Rep. Prog. Phys. Vol. 65, 1427-1487. Engel M.H., Nagy B., Distribution and enantiomeric composition of amino acids in the Murchison meteorite, 1982, Nature, 296, 837-840. Faludi-Daniel A., Demeter S., Garay A.S., Circular dichroism spectra of granal and agranal chloroplasts of maize, 1973, Plant Physiol. 52, 54-56. Field C.B., Behrenfeld M.J., Randerson J.T., Falkowski P., Primary production of the biosphere: Integrating terrestrial and oceanic components, 1998, Science, 281, 237-240. Finzi L., Bustamante C., Garab G., Juang C.B., Direct observation of large chiral domains in chloroplast thylakoid membranes by differential polarization microscopy, 1989, Proc. Nat. Acad. Sci. 86, 8748-8752. Francke C., Otte S.C.M., Van Der Heiden J.C., Amesz J., Spurious circular dichroism signals with intact cells of heliobacteria, 1994, Biochim. Biophys. Acta Bioen., 1186, 7580. Friedel D.N., Snyder L.E., Remijan A.J., Turner B.E., Detection of interstellar acetone towards the Orion-KL hot core, 2005, Astrophysical Journal, 632, L95-L98. Garab G., Linear and circular dichroism, 1996, Biophysical Techniques in Photosynthesis, Advances in Photosynthesis, Vol. 3, eds J. Amesz & A.J. Hoff, 11-40.
208
T.M. Gledhill et al.
Halfen D.T., Apponi A.J., Woolf N., Polt R., Ziurys L.M., A systematic study of glycolaldehyde in Sagittarius B2(N) at 2 and 3 µm: criteria for detecting large interstellar molecules, 2006, Astrophysical Journal, 639, 237-245. Henry G.W., Marcy G.W., Butler R.P., Vogt S.S., A transiting 51 Peg-like planet, 2000, ApJ, 529, L41-L44. Herndl G.J., Muller-Niklas G., Frick J., Major role of ultraviolet-B in controlling bacterioplankton growth in the surface layer of the ocean, 1993, Nature, 361, 717-719. Hollis J.M., Lovas F.J., Jewell P.R., Interstellar glycolaldehyde: The first sugar, 2000, Astrophysical Journal, 540, L107-L110. Houssier C., Sauer K., Circular dichroism and magnetic circular dichroism of the chlorophyll and protochlorophyll pigments, 1970, J. Amer. Chem. Soc., 92:4, 779-791. Hudgins D.M., Bauschlicher C.W., Allamandola L.J., Variations in the peak position of the 6.2 µm interstellar emission feature: A tracer of N in the interstellar polycyclic aromatic hydrocarbon population, 2005, Astrophysical Journal, 632, 316-332. Kasting J.F., Whitmire D.P., Reynolds R.T., Habitable zones around main sequence stars, 1993, Icarus, 101, 108-128. Kasting J.F., Catling D., Evolution of a habitable planet, 2003, Annu. Rev. Astron. Astrophys., 41, 429-463. Kelly S.M., Price N.C., The use of circular dichroism in the investigation of protein structure and function, 2000, Current Protein and Peptide Science, Vol. 1, 349-384. Kemp J.C., Wolstencroft R.D., Swedlund J.B., Circular polarization: Jupiter and other planets, 1971, Nature, 232, 165-168. Kivelson M.G., Khurana K.K., Russell C.T., Volwerk M., Walker R.J., Zimmer C., Galileo magnetometer measurements: A stronger case for a subsurface ocean at Europa, 2000, Science, 289, 1340-1343. Knacke R.F., Methodologies and techniques for detecting extraterrestrial (microbial) life, 2003, Astrobiology, Vol. 3, No. 3, 531- 541. Kuan Y.-J., Charnley S.B., Huang H.-C., Tseng W.-L., Kisiel Z., Interstellar glycine, 2003a, Astrophysical Journal, 593, 848-867. Kuan Y.-J., Yan C.-H., Charnley S.B., Kisiel Z., Ehrenfreund P., Huang H.-C., A search for interstellar pyrimidine, 2003b, Mon. Not. R. Astron. Soc., 345, 650-656. Kvenvolden K. et al., Evidence for extra-terrestrial amino acids and hydrocarbons in the Murchison meteorite, 1970, Nature, 228, 923-926. Landis G.A., Martian Water: Are there extant halobacteria on Mars?, 2001, Astrobiology, Vol. 1, No. 2, 161-164. Léger A., Puget J.L., Identification of the unidentified IR emission features of interstellar dust?, 1984, Astron. Astrophys., 137, L5-L8. Léger A., Pirre M., Marceau F.J., Search for primitive life on a distant planet: relevance of O2 and O3 detections, 1993, Astron. Astrophys., 277, 309-313. Léger A., Mariotti J.M., Mennesson B., Ollivier M., Puget J.L., Rouan D., Schneider J., Could we search for primitive life on extrasolar planets in the near future?, 1996, Icarus, 123, 249-255. Man D., Wang, W., Sabehi, G., Aravind, L., Post, A.F., Massana, R., Spudich, E.N., Spudich, J.L., Beja, O., Diversification and spectral tuning in marine proteorhodopsins, 2003, EMBO J., 22, 1725-1731. Marcy G., Butler R.P., Fischer D., Vogt S., Wright J.T., Tinney C.G., Jones H.R.A., Observed properties of exoplanets: Masses, orbits and metallicities, 2005, Progress of Theoretical Physics Supplement, No. 158, 24-42.
Astro-biological Signatures
209
Mayor M., Queloz D., A Jupiter-mass companion to a solar-type star, 1995, Nature, 378, 355-357. Mazeh T. et al., The spectroscopic orbit of the planetary companion transiting HD209458, 2000, Astrophysical Journal, 532, L55-L58. McArthur B.E. et al., Detection of a Neptune-mass planet in the ρ Cancri system using the Hobby-Eberly telescope, 2004, Astrophysical Journal, 614, L81-L84. Mehringer D.M., Snyder L.E., Miao Y., Lovas F.J., Detection and confirmation of interstellar acetic acid, 1997, Astrophysical Journal, 480, L71-L74. Meierhenrich U.J., Thiemann W.H.-P., Barbier B., Brack A., Alcaraz C., Nahon J., Wolstencroft R.D., Circular polarization of light by planet Mercury and enantiomorphism of its surface minerals, 2002, Origins of Life and Evolution of the Biosphere, 32, 181190. Moorbath S., Dating earliest life, 2005, Nature, 434, 155. Muñoz Caro G.M., Melerhenrich U.J., Schutte W.A., Barbier B., Arcones Segovia A., Rosenbauers H., Thlemann W.H.-P., Brack A., Greenberg J.M., Amino acids from ultraviolet irradiation of interstellar ice analogues, 2002, Nature, 416, 403-406. Murray J.B. et al., Evidence from the Mars Express high resolution stereo camera for a frozen sea close to Mars' equator, 2005, Nature, 434, 352-356. Perryman M.A.C. et al., Gaia: Composition, formation and evolution of the Galaxy, 2001, Astron. & Astrophys., 369, 339-363. Pilcher C.B., Biosignatures of early Earths, 2003, Astrobiology, Vol. 3, No. 3, 471-486. Pizzarello S., Cronin J.R., Non-racemic amino acids in the Murray and Murchison meteorites, 2000, Geochim. Cosmochim. Acta, 64, 329-338. Pospergelis M.M., Spectroscopic measurements of the four Stokes parameters for light scattered by natural objects, 1969, Soviet Physics - Astronomy, 12, 973-977. Popa, R., Between Necessity and Probability: Searching for the Definition and Origin of Life, 2004, Springer. Purdie N., Brittain H.G., Analytical applications of circular dichroism, 1993, Techniques & instrumentation in analytical chemistry, 14, Elsevier. Riehl J.P., Richardson F.S., Circularly Polarized Luminescence Spectroscopy, 1986, Chemical Reviews 86, 1-16. Rivera E.J. et al., A ~7.5 Earth mass planet orbiting the nearby star, GJ 876, 2005, Astrophysical Journal, 634, 625-640. Rosen D.L., Pendleton J.D., Detection of biological particles by the use of circular dichroism measurements improved by scattering theory, 1995, Appl. Opt. 34, 5875-5884. Rothschild L.J., Mancinelli R.L., Life in extreme environments, 2001, Nature, 409, 10921101. Sagan C., Thompson W.R., Carlson R., Gurnett D., Hord C., A search for life from the Galileo spacecraft, 1993, Nature, 365, 715. Salzman G.C., Gregg C.T., Current and experimental methods of rapid microbial identification, 1984, Biotechnology, Vol. 2, 243-248. Santos N.C. et al., The HARPS survey for southern extrasolar planets II: A 14 Earth-masses exoplanet around µ Arae, 2004, Astron. & Astrophys., 426, L19-L23. Schippers A., Neretin L.N., Kallmeyer J., Ferdelman T.G., Cragg B.A., Parkes R.J., Jergensen B.B., Prokaryotic cells of the deep sub-seafloor biosphere identified as living bacteria, 2005, Nature, 433, 861-864. Seager S., Turner E.L., Schaffer J., Ford E.B., Vegetation's red edge: A possible spectroscopic biosignature of extraterrestrial plants, 2005, Astrobiology, Vol. 5, No. 3, 372-390.
210
T.M. Gledhill et al.
Snyder L.E., Lovas F.J., Hollis J.M., Friedel D.N., Jewell P.R., Remijan A., Ilyushin V.V., Alekseev E.A., Dyubko S.F., A rigorous attempt to verify interstellar glycine, 2005, Astrophysical Journal, 619, 914-930. Sparks W.B., Hough J.H., Bergeron L.E., A search for chiral signatures on Mars, 2005, Astrobiology, 5, 737-748. Steinberg I.Z., Circular polarization of luminescence: Biochemical and biophysical applications, 1978, Ann. Rev. Biophys. Bioeng. 7, 113-137. Swedlund J.B., Kemp J.C., Wolstencroft R.D., Circular polarization of Saturn, 1973, Astrophysical Journal, 178, 257-266. Tarter J., The search for extraterrestrial intelligence (SETI), 2001, Annu. Rev. Astron. Astrophys., 39, 511-48. van der Wielen, P.W., Bolhuis, H., Borin, S., Daffonchio, D., Corselli, C., Giuliano, L., D'Auria, G., de Lange, G.J., Huebner, A., Varnavas, S.P., Thomson, J., Tamburini, C., Marty, D., McGenity, T.J., Timmis, K.N., 2005, The enigma of prokaryotic life in deep hypersaline anoxic basins, Science, 307,121-123. Vidal-Madjar A., Lecavelier des Etangs A., Désert J.-M., Ballester G.E., Ferlet R., Hébrard G., Mayor M., An extended upper atmosphere around extrasolar planet HD209458b, 2003, Nature, 422, 143-146. Vidal-Madjar A. et al., Detection of oxygen and carbon in the hydrodynamically escaping atmosphere of the extrasolar planet HD209458b, 2004, Astrophysical Journal, 604, L69L72. Voss K.J., Fry E.S., Measurement of the Mueller matrix for ocean water, 1984, Appl. Opt. 23, 4427-4439. White G.J., Araki M., Greaves J.S., Ohishi M., A spectral survey of molecules in the Orion Nebula from 455-507 GHz: An inventory of prebiotic chemistry, 2004, IAU Symposium 213, 169-172. Woese, C.R., 1987, Bacterial evolution, Microbiol Rev., 51, 221-271. Wolstencroft R.D., The circular polarization of light reflected from certain optically active surfaces, 1974, Planets, Stars, and Nebulae: Studies with photopolarimetry, ed. Gehrels, University of Arizona Press, 495-499. Wolstencrof R.D., Raven J.A., Photosynthesis: Likelihood of occurrence and possibility of detection on Earth-like planets, 2002a, Icarus, 157, 535-548. Wolstencroft R.D., Tranter G., Le Pevelen D.D., Diffuse reflectance circular dichroism for the detection of molecular chirality: an application in remote sensing of flora, 2002b, Bioastronomy 2002: Life among the stars, IAU Symp. 213, eds. R. Norris & F. Stootman, ASP, 149-153. Woolf N.J., Smith P.S., Traub W.A., Jucks K.W., The spectrum of Earthshine: A pale blue dot observed from the ground, 2002, Astrophysical Journal, 574, 430-433.
Astro-biological Signatures
Bill Sparks (right) discusses the merits of Russian vodka with Valeri Maltsev and Virginia Foot.
211
Tim Gledhill demonstrates his proficiency in Russian folk dancing.
Valery Loiko, Joseph Ulanowski, and Tim Gledhill enjoy the Siberian wilderness.
MODELING OF LIGHT SCATTERING BY SINGLE RED BLOOD CELLS WITH THE FDTD METHOD
Jun Q. Lu1, R. Scott Brock1, Ping Yang2 and Xin-Hua Hu1 1
Department of Physics, East Carolina University, Greenville, NC 27858, U.S.A. Department of Atmospheric Sciences, Texas A&M University, College Station, TX 77843, U.S.A. 2
Abstract:
Scattering of light is one of major pathways that biological cells interact with optical radiation in a wide spectral region from ultraviolet to near-infrared. In this region the size parameters of most human cells vary from about 10 to 100, and accurate modeling of scattering requires treatment of light as electromagnetic waves. In this chapter, we review the generic aspects of the finite-difference-time-domain (FDTD) method for solving the problems of light scattering by biological cells and a parallel FDTD algorithm for numerical computation. Furthermore, we review a specific application of FDTD to the simulation of the scattering properties of single human red blood cells with biconcave and deformed shapes.
Key words:
light scattering; FDTD method; parallel computing; red blood cells.
1.
INTRODUCTION
Despite of the ubiquitous presence of light scattering, accurate modeling remains a difficult problem and becomes more challenging for biological cells as the scatterers due to their complex morphologies. In the spectral region from the ultraviolet to near-infrared, the dependence of the spatial distribution of scattered light on scatterer morphology provides a fast and non-invasive method to probe cell structures and changes.1-5 Since the size parameters of most biological cells span from 10 to 100, accurate modeling 213 A. Hoekstra et al. (eds.), Optics of Biological Particles, 213–241. © 2007 Springer.
214
J.Q. Lu et al.
of the scattering process requires consideration of light as electromagnetic waves. Analytical approaches are not applicable to light scattering problems involving biological cells due to their complex morphologies. Thus, modeling of light scattering in these cases was beyond the reach of the conventional methods, with exact or approximate solutions in closed form, to which many mathematicians and physicists had become accustomed to.6 For this reason, it was common for many researchers to treat light scattering as a source of noise, or imperfection, rather than signals. Before the 1980s, the Lorenz-Mie theory of sphere and its extensions 7-9 were applied to the simulation of light scattering by biological cells. Even to this date, the sphere models still serve many researchers in their attempts to understand light scattering by cells in “simple terms”. With rapid progress in computing technology and various highperformance computing techniques over the last two decades, it is practical nowadays to numerically solve the light scattering problems, or electromagnetic radiation problems in general for systems with moderate to large size parameters. Several numerical methods have been developed to solve the Maxwell equations or the corresponding wave equations for modeling light scattering by biological cells. These include the T-matrix method, the discrete dipole method, the multipole method and the finitedifference-time-domain (FDTD) method.10,11, 12 Compared to others, the FDTD method is free of the possible instability due to matrix inversion. More importantly, its relative simple algorithm suits the modeling of biological cells with complex and inhomogeneous morphologies. As a result, the FDTD method has attracted increasing efforts for modeling the complex and wide-ranged configurations of electromagnetic radiation problems that are the hallmarks of diverse biomedical applications.13-22 The FDTD method requires the entire computational domain to be meshed. This mesh is usually uniform so that the mesh density or the stepping size of computation is determined by the smallest detail of the configuration. Fortunately, the FDTD calculations proceeds through the mesh grid points locally and its total computing time can thus be reduced with parallel computing technique for scatterers of large size parameters by cutting the computational domain into multiple sections. The computational efficiency of parallel FDTD codes provides a significant incentive for adopting this technique in solving the complex light scattering problems. The performance-to-cost ratio of computers with single or multiple Central Processing Units (CPUs) continues to decrease rapidly even as the CPU speed remains relatively stagnant over the last few years. Therefore, parallelization of the FDTD method provides a practical approach to model accurately the light scattering problems by biological cells on low-cost computer clusters. In this chapter, we briefly review the FDTD algorithm
Modeling of Light Scattering by Single Red Blood Cells
215
and its parallel implementation. The Mueller matrix is used as a theoretical framework for characterizing light scattering by biological cells. The calculated Mueller matrix elements of human red blood cells (RBCs) with different shapes are then presented as the examples of the results acquired using serial and parallel FDTD simulations20, 21 and the implications of these simulation results on experimental measurements are discussed. No originality is claimed in this review although we made an effort to systematically summarize our previous studies of light scattering by single RBCs using the FDTD method.
2.
THE SCATTERING CONFIGURATION AND MUELLER MATRIX
The problem of light scattering by a single biological cell consists of a wave incident on the scatterer in an open space occupied by a host medium. To account for the polarization configurations associated with light scattering, we adopted a framework on the basis of the Stokes vectors for the incident and scattered light fields and the Mueller matrix for the scatterer.7, 23 Here we first describe the scattering problem in general and then introduce the Mueller matrix.
2.1 Setting up the Problem Figure 1 illustrates an incident wave (E0, H0) along the direction of k0 scattered by a red blood cell with a rotational symmetry, leading to a scattered wave (Es, Hs) in the direction of k. The host medium is described by a dielectric constant εh while the cell is characterized by a spatially variant dielectric constant, ε(r), for its intracellular structure. The dielectric constant of the cell can be also written as ε(r, ω), for ε often varies with the frequency ω of the electromagnetic fields. For a source-free, nonmagnetic, linear dielectric system (µ=1), typical for a system consisting of a cell embedded within a medium, modeling of the scattered fields starts by solving the Maxwell equations, 24
∂B ∂D = 0 , ∇×H = (1) , ∇ i D = 0 , ∇ iB = 0 , ∂t ∂t where D = ε0εE, H = B/µ0, ε0 and µ0 are the permittivity and permeability of ∇×E +
vacuum. In the frequency domain, we can obtain the wave equations satisfied by E(r, ω) in regions inside and outside of the cell, respectively, as follows:
216
J.Q. Lu et al.
(∇ 2 + k 2 )E(r, ω ) = −4π f (r, ω ), r ∈ V ,
(2)
(∇ + k )E(r, ω ) = 0,
(3)
2
2
otherwise ,
where k = ( µ0ε 0ε hω )1/ 2 = ω v is the wave number in the host medium, V is the cell volume and the vector function f(r, ω) on the right-hand side of Eq. (2) is given by
f (r, ω ) =
ε (r, ω ) 1 2 − 1)E(r, ω ) . (k + ∇∇⋅)( 4π εh
(4)
Figure 1. The scattering configuration with the A-axis as the axis of rotational symmetry to mark the orientation of a red blood cell.
The term of −4πf(r, ω) in Eq. (2) describes the optical heterogeneity, in the background of the homogeneous host medium, introduced by the cellular structure that is responsible for light scattering. The general solution for E(r, ω) outside of the cell volume V can be obtained by applying the Green’s second identity in the region 24
E(r, ω ) = E0 (r, ω ) + ∫ f (r ', ω )G (r, r ', ω )d 3r ' , V
(5)
where G (r, r ', ω ) = eik |r −r '| 4π | r − r ' | is the free space Green function and V in the volume integral refers to the volume of the cell. The second term on the right side of Eq. (5) provides the field of scattered light Es. In the far-field zone satisfying kr→∞, we can replace k | r − r ' | by kr − k ⋅ r ' and the scattered field becomes
E s (r, ω ) = F(k , k 0 )
eikr , r
(6)
where the vector function F(k, k0), with |k0|=|k|=k for elastic scattering, is given by
Modeling of Light Scattering by Single Red Blood Cells
F (k , k 0 ) =
1 4π
∫
V
(k 2 + ∇∇⋅)(
ε (r ', ω ) − 1)E(r ', ω )e − ik ⋅r ' d.3 r ' εh
217
(7)
Equation (6) represents an outgoing spherical wave modified by a factor F(k, k0), denoted as the scattering amplitude. It is evident from Eq. (7) that the angular distribution of scattered light represented by F(k, k0) and the cell optical structure represented by ε(r, ω) are related via a volume integral. Because of this feature, it is difficult to determine cellular structure from the measured light distribution, and could lead to ill-posted inverse problems if the measured signals do not provide sufficient information. Furthermore, to evaluate F(k, k0), it is necessary to first solve the near-field field, E(r’,ω), within a volume V containing the scatterer. One approach is to apply the FDTD method to the two curl equations in Eq. (1) for solving E(r’,ω).
2.2 The Stokes Parameters and Muller Matrix Formulation Due to the vector nature of the electromagnetic fields, the scatterer not only redirects the energy flux of the incident light but also modifies its polarization state along the way. Therefore, light scattering problems are in general polarization-dependent. Before we introduce the Mueller matrix, let’s first cast the Eqs. (6) and (7) with an amplitude matrix on the basis of the linearity of the Maxwell equations by resolving the incident and scattering fields into components parallel and perpendicular to the scattering plane defined by k and k0
EEs e−ikr +ikz = −ikr E⊥ s
S2 S3 EE0 . S4 S1 E⊥0
(8)
The elements of the 2x2 amplitude matrix in Eq. (8) are complex functions of the scattering polar angle θs and the scattering azimuthal angle φs, measured from the incident direction of k0 to the scattering direction of k. The amplitude matrix elements can be calculated from the scattering amplitude F(k, k0) using two independent polarization configurations for the incident wave as the input data to the FDTD calculations.25 The electromagnetic fields of light cannot be directly measured due to their high frequencies. Rather, optical signals are proportional to the squares of the field amplitudes. This prompts the use of the real valued Stokes vectors and Mueller matrix for relating the scattered light signals to those of the incident, up to a multiplicative constant, as follows:7, 23
218
J.Q. Lu et al.
Is S11 Qs = 1 S 21 U s kr 2 S31 Vs S41
S12
S13
S 22 S32 S42
S 23 S33 S43
S14 I 0 S 24 Q0 , S34 U 0 S44 V0
(9)
where I, Q, U, V are the Stokes parameters of the incident light beam (with subscript 0) and the scattered (with subscript s), and Sij (with i, j = 1,2,3,4) are the elements of the Mueller matrix characterizing the optical properties of the scatterer and are functions of θs and φs. The element S11 yields the probability of an unpolarized incident light being scattered into the direction of θs and φs while other elements provide information on scattering of light with different polarization states. When normalized over the solid angle of 4π steradian, S11 is also called the scattering phase function. For a scatterer of a single biological cell, there are at most 7 independent elements in the Mueller matrix and each element can be derived from the amplitude matrix of 4 complex elements.7 For a homogeneous sphere, the Mueller matrix contains only 4 independent elements among the 8 non-zero elements, i.e., S11=S22, S12=S21, S33=S44, S34=S43. Under the single-scattering condition, it has been shown that a Mueller matrix can be similarly defined for an ensemble of single scatterers at random locations by adding the matrices of individual scatterers as a result of incoherent superposition of the scattered Stokes parameters.26, 27 In the case of spheres with the same size and refractive index, the Muller matrices are identical for single or sphere ensemble as the scatterer because of the spherical symmetry. For biological cells, however, the shape and optical structure posses no symmetry in general and vary even among the cells of same phonotype due to their different growth patterns and stages in the cell cycle. Thus, all 16 Mueller matrix elements for a scatterer consisting of multiple cells are expected to be independent in general.
3.
THE FDTD METHOD
The FDTD computational codes used for the numerical results recaptured in this review from the authors’ previous studies are based on a code written in Fortran 77, which was designed for light scattering by ice crystals.25, 28 The serial and parallel FDTD codes have been developed in
Modeling of Light Scattering by Single Red Blood Cells
219
Fortran 90/95 with enhanced structural design for parallel implementation 20-22
and are constantly improved for performance enhancement, 3D structural handling capability and portability.
3.1 The FDTD Algorithm The FDTD method that was first proposed by Yee in 1966 12 is to numerically solve the two curl equations in the Maxwell equations in the time domain. In this method, the interested region is divided into rectangular grid cells, and the scatterer is defined by properly assigning ε(r, ω) over the grid points denoted by integer indices (i, j, k) with ∆x, ∆y, and ∆z as the sizes of a grid cell along the x, y, and z-directions, respectively. Additionally, time is discretized with a step size of ∆t. The FDTD algorithm uses a central-difference scheme to approximate the spatial and temporal derivatives of a field component u as 11, 28 u x +∆x / 2,t − u x −∆x / 2,t ∂u n i n (10) = i + O ( ∆x) 2 , ∂x ∆x xi ,tn
∂u ∂t
= xi ,tn
u x ,t
i n +∆t / 2
− u x ,t
i n −∆t / 2
∆t
+ O (∆t ) 2 .
(11)
Ey(i-1/2,j,k+1/2) Ex(i,j-1/2,k+1/2) Hz(i,j,k+1/2) Ey(i+1/2,j,k+1/2)
Ex(i,j+1/2,k+1/2) Ez(i-1/2,j+1/2,k)
Ez(i+1/2,j-1/2,k) Hy(i,j+1/2,k) Hx(i+1/2,j,k) Ez(i+1/2,j+1/2,k) Ex(i,j+1/2,k-1/2) Ey(i+1/2,j,k-1/2) Figure 2. The E and H field components on a Yee grid cell.
In order to satisfy the constraints imposed by the central difference scheme, the six electromagnetic field components are specified on a grid cell at staggered locations, as shown in Fig. 2. The electric field components, (Ex, Ey, Ez), are sampled on the edges of the grid cell and computed at multiples of time steps t=n∆t. The magnetic field components, (Hx, Hy, Hz), are sampled at the face centers of the unit grid cell and calculated at multiples
220
J.Q. Lu et al.
of half time steps t=(n+1/2)∆t. After the spatial and temporal derivatives in the curl equations are converted into finite difference forms, the resultant finite difference equations are then solved in a time marching sequence by alternatively computing the electric and magnetic fields on a dual, staggered, orthogonal grid within an appropriately selected space (white space) containing the scatterer. Examples of the finite-difference equations for the field components Ex and Hx can be found as n +1
Ex i , j , k = Ex i , j , k
n + 12
n
n + 12 ∆t H z i , j + 12 ,k − H z + ε i , j ,k ∆y
n − 12
H x i , j ,k = H x i , j ,k
∆t E y + µ0
n i , j , k + 12
− Ey
n + 12 i , j − 12 , k
n+ n+ H y i , j ,2k + 1 − H y i , j ,2k − 1 2 2 , − ∆z 1
n i , j , k − 12
∆z
−
Ez
n i , j + 12 , k
1
− Ez ∆y
n i , j − 12 , k
.
(12)
(13)
Equations for the remaining electric and magnetic components are similar. This time and space leapfrog scheme yields an efficient, secondorder accurate algorithm and enforces automatically the divergence equations in Eq. (1) and boundary conditions. Furthermore, the field calculations are local, i.e., only the values of field in neighboring grid points are needed. This makes the FDTD algorithm scalable for efficient parallel computation. To ensure numerical stability of the algorithm, the spatial and temporal increments must satisfy the Courant-Friedrichs-Levy (CFL) condition given by 28
∆t ≤
1 v 1/ ∆x 2 + 1/ ∆y 2 + 1/ ∆z 2
,
(14)
where v is the light speed in the host medium. Considering an incident wave with λ = 1µm, the time step ∆t must be less than approximately 0.1fs for a grid cell size of ∆x =∆y =∆z =λ/20.
3.2 Implementation Details 3.2.1
Absorbing boundary condition (ABC)
Light scattering from a single biological cell is often treated as an open region problem, i.e., the spatial domain is unbounded. For efficient computation, the domain must be limited to just enclose the scatterer, and a suitable boundary condition on the outer perimeter of the computational domain must be applied to “extend” the finite domain to infinity in
Modeling of Light Scattering by Single Red Blood Cells
221
simulations. For this purpose, a special ABC should be explicitly enforced to minimize the artificial reflections of outgoing waves. Several schemes have been developed, such as the widely used Mur’s second-order ABC 29, outgoing-wave annihilators 30, transmitting boundary conditions 31, the Higdon method,32 and the perfectly matched layers (PML) ABC introduced by Berenger.33 Among these choices, the PML ABC has been proven to be very efficient in reducing the artificial reflection better than other ABCs by orders of magnitudes.34 The PML ABC method is based upon a splitting of electric and magnetic field components at the absorbing boundary with the possibility of assigning losses to the split components. It creates a fictitious anisotropic absorbing medium adjacent to the FDTD domain boundary backed by a perfectly conducting surface that has a wave impedance independent of the angle of incidence and frequency of outgoing scattered waves. We further found that PML is very suitable for parallel FDTD simulations since no additional inter-processor communication is required. Moreover, they can be located generally very close to the scatterer, which is advantageous over other ABCs for its relatively small memory requirement because of reduced domain size or the “white space”, as shown in Fig. 3.
Figure 3. The sectional view of a 3-d FDTD grid with the PML ABC and light incident from the left. The domain can be divided into three regions: a total field region enclosing the scatterer; a scattered field region and the PML region.
In applying PML ABC, the computational domain is now surrounded by the PML region as shown in Fig. 3, enclosed by the perfectly conducting boundary. In the PML region, each field component is split into two subcomponents to achieve a high absorption rate of incident power that are updated according to PML ABC 33. For example, Hx is split into Hxy and Hxz with Hxy updated according the following relation
222
J.Q. Lu et al.
H
1 2 1 xy i + , j ,k 2 n+
=e
−σ *y ∆t / µ
H
1 2 1 xy i + , j ,k 2 n+
− Ezx i + 1 , j + 1 ,k + Ezy n
2
2
(1 − e
+
−σ *y ∆t / µ
σ ∆y * y
n
− Ezy
1 1 i + , j − ,k 2 2
)
n
[ Ezx i + 1 , j − 1 ,k
n 1 1 i + , j + ,k 2 2
2
2
(15)
],
where σ *y is the fictitious magnetic conductivity in the y-direction in the PML region. 3.2.2
Fields in different regions and matrix elements
To increase the speed of FDTD simulation, the incident field will only be accounted for in the region very close to the scatterer. This allows the separation of the computational domain into two parts: a near-field region close to the scatterer where the total (incident and scattered) fields are considered as the near-fields, and a far-field region including the PML region where only the scattered field or the far-fields are calculated. This approach is ensured by the linearity of the Maxwell equations with the relations given by
E = E0 + E s
and
H = H0 + Hs .
(16) 35
It can be realized by using the equivalence theorem to represent the effect of the incident field in the near-field region by the equivalent electric and magnetic currents on a virtual surface enclosing this region. This scheme of zoning provides several benefits for optimization of the FDTD algorithm. It allows arbitrary forms of incident wave for relatively simple coding of the interaction structures, and a large computational dynamic range. It also minimizes residue reflections from the ABC at the outmost surface of the FDTD domain and allows easy calculation of the far-fields. According to Eqs. (6) and (7), it is necessary to calculate the near-fields in frequency domain, E(r, ω) and H(r, ω), for obtaining the scattered farfields and the angle-resolved light scattering distributions in terms of the Mueller matrix elements as defined in Eq. (9). The FDTD calculations provide the near-fields in the time domain which can be transformed into the frequency domain with a discrete Fourier transform
E(r, ω ) ≈
∆t 2π
N
∑ E(r, t ) e ω
i tn
n =1
n
,
(17)
and H (r, ω ) ≈
∆t 2π
N
∑ H(r, t ) e ω
i tn
n =1
n
.
(18)
Modeling of Light Scattering by Single Red Blood Cells
223
The scattering amplitude F(k, k0) is obtained from Eq. (7) with numerical integrations for different scattering direction k with a given k0. Using two incident waves with crossed polarizations, e.g., in the x and y directions as shown in Fig. 1, the amplitude matrix defined in Eq. (8) can be related to F(k, k0) in the following form 25, 36 FE x S= F⊥ x
FE y cos θ s F⊥ y sin θ s
FE, x −ik 3 = F⊥ , x 4π
∫
V
(
EE, x (r ', ω ) − ik ⋅r ' 3 ε (r ', ω ) − 1) d r', e εh E⊥ , x (r ', ω )
where FE, y −ik 3 = F⊥ , y 4π
sin θ s , − cos θ s
∫
V
(
EE, y (r ', ω ) − ik ⋅r ' 3 ε (r ', ω ) − 1) d r'. e εh E⊥ , y (r ', ω )
(19) (20)
(21)
To obtain the angle-resolved amplitude matrix, and subsequently the Mueller matrix elements,7 the four volume integrals in Eqs. (20) and (21) must be calculated for each scattering direction k. This step is time consuming for large scatterers. It should be pointed out that the amplitude matrix elements can be obtained using a surface integral method that could reduce computing time and memory requirement.11, 36, 37 This method, however, can result in errors for large scattering angles and a modification has been introduced recently to address this problem.38 We are currently investigating the possibility of improving our parallel code with this scheme. The FDTD results allow the calculation of other important parameters that are often used to characterize the scatterer such as the scattering cross section σs, the absorption cross section σa, and the extinction cross section σt. The scattering cross section σs is defined as the ratio of the total power scattered by the cell from the incident beam to the incident power per unit area
σs =
1 v | E0 |2
∫
S1
Re(E s × B*s ) ⋅ da'
(22)
while total cross section σt is defined by the total power taken due to scattering and absorption from the incident beam to the incident power per unit area
σt = −
1 v | E0 |2
∫
S1
Re(E s × B*0 + E*0 × B s ) ⋅ da'
The absorption cross section σa can be obtained as σa = σt − σs.
(23)
224
J.Q. Lu et al.
3.3 Parallel Code Development The computing complexity of the FDTD algorithm to model scatterers of large size parameters often exceeds the capabilities of single-CPU computers, even the supercomputers. Fortunately, the FDTD method is well suited for parallel processing and such techniques have already been used for the analysis of microwave circuits and antennas 39, 40, of optical integrated devices 41, and for scattering 42 and wave interaction problems 43. As the performance-to-cost ratio of the personal computers (PCs) and workstations rapidly decreases, the benefits of developing parallel FDTD code outweigh significantly the cost of code development, making FDTD parallelization an attractive option. The most computationally intensive portion of the FDTD algorithm is the field update during time marching in the near-field region. A typical PC cluster consists of multiple nodes and each node with one or more CPUs or processing elements (PEs). The cluster nodes are connected by a network switch and thus belong to a loosely coupled multiple-instruction multipledata architecture. The FDTD algorithm has demonstrated excellent speedups on various architectures of multi-PE computers11 and workstation clusters. 40-42 The common element of the parallel FDTD algorithms is that the field update tasks are split among all PEs using a spatial decomposition as shown in Fig. 4. (1)
section 1
Ez
Hx
Hy Ey Ex
Hz
section 2
Ez
Hx
Hy
Ey Ex
Hz
(2) Figure 4. Data exchange between neighboring sections. Note that only tangential components on the boundary surfaces are exchanged.
Modeling of Light Scattering by Single Red Blood Cells
225
To perform FDTD simulations on a parallel computing cluster, we divide evenly the computational grid into slab sections along the planes that are perpendicular to one of the coordinate axes with the spatial decomposition performed along the grid edges. The sections are nonoverlapping and of equal size, sharing only common surfaces, or interfaces, chosen by taking slices along the edges of the primary grid. Parallelization in the time domain requires consideration of load balancing and inter-PE communication. Each PE of the cluster executes the field updates on one section and communicates boundary values to and from the PEs of its neighboring sections, thus enabling load balances and minimizing processor idle time. A close examination of the arrangement of the electromagnetic field components in a grid cell (see Fig. 2) reveals that the field components are located either on a cell surface plane or on a middle plane (the plane that divides a grid cell into two halves along an axis), and they are either tangential or normal to the planes. It can be seen from Eqs. (12) and (13) and Fig. 2 that the four nearest-neighboring field components are needed to update each field component in any grid cell. For a field component normal to its plane, the four nearest neighboring field components are located in the same plane. As for a field component tangential to its plane, two of the nearest neighboring field components are located in the same plane while the other two are located in the two parallel neighboring planes, one on each side, and both are tangential to their own planes. This means that when we divide the computing domain into sections along the grid cell surface planes or the middle planes, no information is needed from its neighboring section to update a normal field component on the boundary plane because all the nearest neighboring field components are available on the same plane. But to update a tangential field component on the boundary plane, one tangential field component is needed from the boundary plane of its neighboring section. An example of the section division is illustrated in Fig. 4 where the boundary plane is perpendicular to the y-axis. It can be seen that to update a tangential component Ex on the boundary plane of section 2, a tangential component Hz on the boundary of the neighboring section 1 is needed. Conversely, calculation of a tangential component Hz in section 1 needs Ex from section 2. Based on the above considerations and the fact that the electric and magnetic field updates are separated in time, a simple scheme is used to manage communications between PEs on neighboring sections. An extra layer of one half grid cell is added to each section at the boundary so that among the two neighboring sections one has a copy of the tangential components of E on the boundary of its neighbor and the other has a copy of the tangential components of H on the boundary plane of its neighbor, as
226
J.Q. Lu et al.
marked by the unlabelled planes in Fig. 4. In this scheme, the field components everywhere within each section, except in the copied plane of the added layer, can be updated independently without inter-PE communication. At the beginning of a time marching step, the H components everywhere except the copied plane are updated without interPE communication. The updated tangential components of H at the original boundary surface are then copied to its neighboring PE (communication (1) in Fig. 4) followed by the updating of E components everywhere except the copied plane. The updated tangential components of E at the original boundary are then copied to its neighboring PE (communication (2)) to prepare the system for next cycle of updating. As a result, two tangential components in each grid cell are communicated to a neighboring PE in one direction at each half time step. This approach is slightly different from previous strategies 11, 39 where only the H components at the shared boundary plane are exchanged at each time step at the expense of redundant updating of the tangential components of E and the normal components of H at the boundary of each section. The above discussed communication leads to a communication overhead that increases with the PE number. For PC clusters where most PEs are connected through network switches the overhead can be significantly larger than that of parallel networks with PEs of shared memory. In comparison, calculations in the frequency domain are performed mostly on calculating the four volume integrals shown in Eqs. (20) and (21). The volume integrals are divided among the PEs and the end results are obtained by adding all data from each PE. Thus, communications are necessary only at the beginning and end of the integral calculation. Efficient parallel FDTD simulations also require load balancing. It is possible that one PE has to perform more computations than its neighbors. Consequently, the neighboring PEs idle during each time step due to the wait for the boundary values from the calculating PE. A load balancing procedure has been developed in our code to minimize idling time by optimization of the section division.
3.4 Code Validation and Performance Evaluation The parallel FDTD code has been evaluated on three different computer clusters with inter-PE communications managed by the Message Passing Interface (MPI). The Fortran 90 compilers and MPI, however, are from different sources. We validated the FDTD code on the cases of homogeneous spheres of different radius by comparing to the Mie theory
227
Modeling of Light Scattering by Single Red Blood Cells
and evaluated the code performance on the three clusters, DSC, LSC, and BLLC, with technical information listed in Table 1. Table 1. Three computer clusters Cluster name Cluster node # of nodes (memory) Communication Software Institution URL
DSC eight 1.5GHz IBM Power4 CPUs 176 (16GB)
LSC two 3.06GHz Intel xeon CPUs 525 (2GB)
BLLC two 3.06GHz Intel xeon CPUs 8 (1 GB)
2Gbps switch IBM versions of F90 compiler and MPI San Diego Supercomputer Center http://www.sdsc.edu/u ser_services/datastar/
2Gbps switch Intel F90 compiler and MPI by Argonne* Texas Advanced Computing Center http://www.tacc.utexas .edu/hpc/
1Gbps switch Intel F90 compiler and MPI by Argonne* Biomedical Laser Laboratory, ECU http://bmlaser.physics. ecu.edu/computing/ind ex.htm
Mie Theory FDTD
S11 (normalized)
101 100 10-1 10-2 10-3 10-4 10-5 0
30
60
90
120
Scattering Angle (degree)
(a)
150
180
0.15 15 0.00 0 -0.15 -15
S11
-S12/S11
-0.30
-30 0.05
0.1
0.00
Absolute Error
102
Absolute Error Relative Error (%)
* freely available at http://www-unix.mcs.anl.gov/mpi/mpich/
0.0 -0.05
S33/S11
-0.1 0
30
60
S43/S11 90 120 150
0
30
60
-0.10 90 120 150 180
Scattering Angle (degree)
(b)
Figure 5. Comparison of the Mueller matrix elements between FDTD and Mie theory for a sphere of r=2.5µm: (a) angle-resolved S11; (b) errors of the other elements. (Adapted from Fig. 2 of ref. 20 with permission from OSA.)
Among the 8 non-zero Mueller matrix elements for a homogeneous sphere, only 4 are independent. We calculated S11, S12, S33 and S34 for spheres of radii r =1.60, 2.5, 3.75, and 5 µm for code validation. With an incident wave of wavelength λ0 = 1 µm in vacuum, the sphere was assumed to have a complex refractive index of nsp = 1.4019 + i1.6805x10-5 in a host
228
J.Q. Lu et al.
medium with refractive index of nh = 1.35. This corresponds to a wavelength of λ=0.74µm in the host medium, and the size parameters for the tested spheres thus are 13.5, 21.2, 31.8, and 42.4, respectively. A Gaussian pulse with a duration of 30 time steps and a time step of ∆t=0.98∆x/c was assumed for the incident light. The grid cell size ∆x was set as λ/30 and the total number of time marching steps for the FDTD simulations was set as 8N, where N is the number of time steps needed for the incident pulse to travel across the total field region. 1.00 g: FDTD g: Mie 200
0.98
100
0.96
Anisotropy Factor
Extinction Cross Section (µm2)
300
σe: FDTD σe: Mie 0
0.94 0
1
2
3
4
5
6
Sphere Radius (µm)
Figure 6. Comparison of the extinction efficiency σe and the anisotropy factor g between FDTD and Mie theory versus the radius of the spheres. The lines are provided as the visual aids. (Adapted from Fig. 3 of ref. 20 with permission from OSA.)
The angle-resolved Mueller matrix elements calculated from the parallel FDTD code show excellent agreement with the Mie theory for all sphere cases. An example of r =2.50 µm is shown Fig. 5 with normalized S11 in (a) and the relative error of S11 and the absolute error of the other elements in (b). The extinction efficiencies and anisotropy factors of different spheres are compared in Fig. 6. The code portability was tested on the three clusters and found to be satisfactory with identical results produced. We evaluated the performance of the parallel FDTD code by compiling the CPU times for execution of the two main portions of the code: one for the FDTD calculations in the time domain and the other in the frequency domain. A parameter of speedup S has been used for evaluation of the parallel code performance, which is defined as the total CPU time of a simulation by executing the parallel code over a cluster of PE divided by the CPU time of executing the identical simulation with a serial code on one PE. The ideal value of speedup, therefore, is given by the number of PE, p, and can be achieved when communication overhead takes negligible time. We first tested the code with a moderate grid of 159x159x159 in size. This is the case in Fig. 7(a), (b) and (c) for a sphere with a radius of r
Modeling of Light Scattering by Single Red Blood Cells
229
=1.6µm. The total number of time marching steps is 2397. For the sphere case of radius r =2.5µm and λ=0.74µm in the host medium, the corresponding size parameter is about 21 with results shown in Fig. 7 (d) to (f).
10000
60000 80
80 40000
60
Ideal speedup Speedup
(a) 0 20000
20
40
60
80
(b)
0
80 60
10000
40
(b)
5000 0 20
40
60
20 0 100
80
(d)
0
0 100 100
15000
0
40 20000
20
Speedup
0
60
40
Time per PE (s)
5000
Time per PE (s)
100
100
Total time Frequency domain Time domain
20
40
60
0 100 100
80
60000
80 60
40000
40 20000
(e)
0 0
20
40
60
20 0 100
80
15
15
20000
20
Speedup
15000
60000 15000
10
10 40000
10000
(c)
5000 0 0
4
8
Number of PEs
12
5
0 16
5
20000
(f)
0 0
4
8
12
0 16
Number of PEs
Figure 7. FDTD simulation times per PE and the speedup versus the PE number for a sphere of r=1.60µm with grid size 159x159x159 and 2397 time marching steps: (a) on DSC; (b) on LSC; (c) on BLLC and for the case of r = 2.5µm with grid size 231x231x231 and 3421 time marching steps: (d) on DSC; (e) on LSC; (f) on BLLC. (Adapted from Fig. 4 of ref. 20 with permission from OSA.)
The data in Fig. 7 demonstrated the high performance of the parallel FDTD on clusters with p up to about 100. The large difference in the performance results between DSC and LSC/BLLC indicated the significant dependence of code performance on cluster architecture designs and handling of data communications by different versions of F90 compiler and MPI. Even with the reduced speedup values for the LSC/BLLC clusters, however, the benefits of parallel computing are apparent for these low-cost PC clusters. In the case of r=2.5µm sphere, with a size parameter of 21, the
230
J.Q. Lu et al.
total CPU time was reduced from about 335 hours with the serial code on one PE to about 33.8 hours with the parallel code on the 16-PE cluster of BLLC. Comparison of the data in Fig. 7(a) to (d) also indicates that the communication overhead for large number of PEs starts to become noticeable in the case of small sphere more quickly than the cases of large sphere.
4.
NUMERICAL SIMULATIONS OF LIGHT SCATTERING BY HUMAN RED BLOOD CELLS
Red blood cells (RBCs) constitute about 45% to 50% of human blood in volume as the carriers of oxygen, carbon dioxides and other molecules in the circulation system. Optics of red blood cells (RBCs) attracted wide attention for their intriguing shapes and clinical importance as the most abundant cells in the blood. Light scattering signals from RBCs can be used for cell counting, sorting, size and shape determination.5, 44-48 It has also been reported that the Mueller matrix element elements S14 can be correlated to the hemoglobin polymerization in sickle cells.49 Consequently, light scattering by single RBC and aggregates have been studied with various approximate and numerical methods 5, 50-52. In this section, we present our recent FDTD results.20, 21
4.1 RBC with Biconcave Shapes A mature human RBC is a non-nucleated cell filled with suspensions of hemoglobin proteins within the membrane and can be modeled as a homogenous body. The hemoglobin proteins have different derivatives according to their binding molecules and we consider here only the two major components of oxyhemoglobin (HbO2) and deoxyhemoglobin (Hb). For a homogeneous RBC, the imaginary part of the refractive index n = nr + ini of the protein suspension is related to the molar extinction coefficients γ1 and γ2 of HbO2 and Hb, respectively, by
ni =
ln10 f λ0 (h1γ 1 + h2γ 2 ) , 4π nr
(24)
where f is the hemoglobin concentration, h1, h2 (=1−h1) are the volume ratios of HbO2 and Hb, respectively. The spectroscopy data of γ1 and γ2 are available for λ0 ranging from 200 to 1000nm 53. For a normal RBC, We assumed a hemoglobin oxygen saturation of h1=97% and nr=1.400.46 In addition, we adopted f=335 g/liter or 5.04×10-3 mole/liter for the
Modeling of Light Scattering by Single Red Blood Cells
231
hemoglobin concentration.54 With these data, the imaginary refractive index of a normal RBC was obtained as a function of λ0 between 400 and 1000nm 20, which is about one order of magnitude smaller than the values used in previous publications.50 In calculating the scattered fields, we assumed that the RBC is suspended in a host medium of blood plasma with no absorption and a refractive index of nh=1.350. The viscoelastic structure of a human RBC enables its shape change in response to different shear or physiologic conditions, which can lead to significant pathologic consequences.55 In modeling the biconcave shape and related symmetric changes, we adopted a surface shape function 50 to describe the radial coordinate of a surface location as a simple function of the polar and azimuthal angles of θA and φA, respectively r (θ A , φ A ) = a sin q θ A + b .
(25)
relative to the normal of the RBC disc or the A-axis shown in Fig. 1. The radius of the biconcave disc is determined by R = a + b, center thickness by 2b, and the shape by q with q=0 corresponding to a sphere, as shown in Fig 8(a). The shape parameters of q=5, a=3.0µm and b=0.75µm were used as the undeformed biconcave shape 50 in the following results. A typically sized RBC was found to have a volume of about 94.1µm3 and a surface area of 134.1µm2,56 equivalent to the volume of a sphere with a radius of R = 2.82µm. Two additional biconcave shapes with values of q = 2 and 9 were also used to study the effect of shape changes of a biconcave RBC. The volume of a biconcave RBC was kept as a constant by adjusting a and b for the two cases: q=2, a=2.1µm, b=1.12µm and q=9, a=3.8µm, b=0.41µm. The cross-sections of these cases are shown in Fig. 8(a). Although the shape function given by Eq. (25) does not fit precisely to the shape of RBC determined experimentally,57 it was easy to be implemented in the code and sufficiently accurate for our purpose of investigating light scattering dependence on shape and orientation. From the FDTD results for biconcave RBC, we found that the differences between two pairs of the elements, S22 and S11, S44 and S33, are quite small. Thus, we present here only 4 matrix elements, which are plotted in Fig. 9 for the different shape of the biconcave RBC under the condition of equal volume and a “face-on” orientation (θ0=0°). It is clear from these data that the shape change correlates with the scattering pattern change, and the elements S11 and S43 exhibit the highest sensitivity to the shape. As the shape deviates from the spherical shape by isovolumetrically “flattening” along the z-axis, by increasing q, the phase function S11 shows large enhanced backscattering near θs=180° by more than 10-fold while the side scattering near θs=90° demonstrates large pattern shifting.
232
J.Q. Lu et al.
0
A
q=9 q=2
q=5
q=0 ∆P=26 10 3
1.2
10
0
(c)
(b)
(a)
Figure 8. Schematic representation of RBC shapes: (a) cross-sections of biconcave shapes, (b) cross-sections of deformed shapes with rotational symmetry around the A-axis, (c) projection views of biconcave and deformed shapes with ∆P=0 and 10.
103
1.0 q=2 q=5 q=9
100
0.5
10-1
0.0
10-2 10-3
-0.5
10-4 10-5
S33 / S11
-S12 / S11
101
1.0
-1.0 1.0
0.5
0.5
0.0
0.0
-0.5
-0.5
S43 / S11
S11 (normalized)
102
-1.0
-1.0 0
30
60
90
120
θs (degree)
150
180
30
60
90
120
150
180
θs (degree)
Figure 9. The angle-resolved Muller matrix elements at different values of shape parameter q for a biconcave RBC with θ0=0° and λ0=1 µm.
Modeling of Light Scattering by Single Red Blood Cells
233
The geometrical cross-section σg toward the incident light was calculated numerically for each shape and orientation in the efficiency factor calculations. The optical parameters of scattering and absorption efficiencies, defined as the respective cross-section divided by the geometric cross-section, are listed in Table 2 together with those of the sphere case (q=0). The optical parameters listed in Table 2 show large drops in scattering and absorption power for increasing q, but the correlation between g and q is relatively weak. q 0 (sphere) 2 5 9
Table 2. The shape dependence of efficiencies (θ0 = 0°, λ0=1 µm) Qa g Qs 1.287 1.09×10-3 0.9913 (Mie: 1.311) (Mie: 1.08×10-3) (Mie: 0.9912) 0.6084 7.20×10-4 0.9916 0.3441 5.27×10-4 0.9930 0.1965 3.90×10-4 0.9904
To study the effect of cell orientation on light scattering patterns, we obtained the Muller matrix elements as a function of the scattering angle θs for different θ0 of orientation angle, as shown in Fig. 10. When θ0≠0, these elements were averaged over the azimuthal angle φs and the angular oscillations were smoothed out in comparison with the case of θ0=0°. For biconcave RBC suspended in blood plasma, the cells orientate arbitrarily. Experimental measurements of single scattering light distributions from a cell suspension correspond in general to the calculated values averaged over φs and θ0. For this reason, we averaged the elements of the Muller matrix over eight orientation angles between 0° and 90° which resulted in much smoother curves that are also plotted in Fig. 10. The anisotropy factor corresponding to the averaged phase function, i.e. the normalized S11, is g=0.9888. The averaged efficiencies were also computed by obtaining the ratio of averaged cross-sections to the averaged geometric cross-section, which yielded 0.8091 for Qs and 8.36x10-4 for Qa. In the study of multiple light scattering in turbid tissues, a singleparameter function of Henyey-Greenstein (HG) is widely used as a single scattering phase function 58
PHG (θ s ) =
1 1− g2 . 2 (1 − 2 g cos θ s + g 2 )3/ 2
(26)
We plotted in Fig. 10 the normalized HG phase function using the g obtained above for comparison with the averaged S11 from our FDTD calculation. It is evident that the two phase functions differ significantly
234
J.Q. Lu et al.
θ0=0o θ0=45
1.0 o
0.5
θ0=90o averaged HG
0.0
-S12/S11
104 103 102 101 100 10-1 10-2 10-3 10-4 10-5
-0.5
1.0
-1.0 1.0
0.5
0.5
0.0
0.0
-0.5
-0.5
-1.0
S43/S11
S33/S11
S11 (normalized)
between 60° and 170° of scattering angle. More importantly, the averaged S11 increases as θs approaches to 180° while PHG(θs) is a monotonically decreasing function.
-1.0 0
30
60
90
120
θs (degree)
150
180
30
60
90
120
150
180
θs (degree)
Figure 10. The angle-resolved Muller matrix elements at different orientation angles of θ0 and the averaged values for a RBC with q=5 and λ0=1 µm. All elements shown were averaged over the azimuthal angle φs for different θ0. The HG phase function was obtained with the g averaged over θ0. (Adapted from Fig. 7 of ref. 19 with permission from SPIE.)
By comparing the results shown in Figs. 9 and 10, we note that the shape and orientation changes affect the Mueller matrix elements in different manners: shape changes lead to often translated angular distribution patterns without much effect on the fine structures while orientation difference affects the fine structures. These results confirms the expectation that angle-resolved measurement of scattered light should yield more useful data for detecting cellular morphology than the current practice with only averaged forward and side scatter measurements.
Modeling of Light Scattering by Single Red Blood Cells
235
4.2 RBC with Deformed Shapes A RBC assumes a biconcave shape when free of constraint and deforms from the biconcave shapes in the blood flow through microvessels in response to the upstream hydrostatic pressure drop across the cell and the opposing drag of the capillary wall. The deformable membrane of RBC has viscoelastic properties, permitting isochoric shape changes without an accumulation of elastic energy 59 and preserving volume and surface area.60 Elastic deformations of RBCs have been analyzed previously for the effect on the blood flow in which the shapes of deformed RBCs are assumed to possess rotational symmetry around their center axis. 61-65 The 3D structure of a deformed RBC is generated by rotating its crosssection around the center axis. We considered 5 cases of deformed RBCs proposed by Zarda et al.,62 corresponding to 5 different values of a dimensionless pressure drop ∆P in a blood flow, as shown in Fig. 8 with the cross section views in (b) and the projection views of the 3D structures in (c). For the five shapes shown in Fig. 8(b), the cell volume varies within 2% from the target value of 94.1 µm3, but the related surface areas differ from the target value of 134 µm2 by a margin of 7% ~ 12%. In the FDTD calculations, the real and imaginary part of the complex refractive index of RBC was chosen to be nr=1.400 and ni =1.6804x10-5, respectively, at the wavelength of λ0 = 1µm for an RBC with an oxygen saturation of 97%.20 The host medium of blood plasma was assumed to be transparent with a real refractive index of nh=1.350. The FDTD grid cell size was set at ∆x=λ/25 with λ as the wavelength of the incident light in the host medium. Tests shown that sufficient accuracy can be achieved at this resolution. The shape change of the deformed RBCs can cause significant variation in the angular dependence of phase function or normalized S11. One example is shown in Fig. 11 in which 2D plots of S11 are presented in the plane of the scattering angle θs and azimuthal scattering angle φs for 3 deformed shapes of RBC at an orientation angle of θ0 = 30o. These results clearly demonstrate that the 2D angular distribution of the phase function from a single cell should yield a very sensitive probe to detect morphology changes in the cell. Combination of the FDTD method as an accurate modeling tool with experimental data 66 could lead to robust solutions of the inverse problems to determine cellular morphology from light scattering signals.
236
J.Q. Lu et al.
(a)
(b)
(c)
Figure 11. The gray-scale plot of the normalized S11 as a function of scattering angles for deformed RBCs at θ0=30o at 3 values of ∆P: (a) 0; (b) 3; (c) 26.
The Mueller matrix elements have been calculated as the functions of scattering angle θs at selected values of orientation angle with θ0 = 0o, 30o, 60o, 90o, 120o, 150o and 180o. The elements were averaged over θ0 with equal weights for ∆P = 0, 3 and 26. Only the cases of θ0 = 0o and averaged one are presented here (more detailed presentation of these results can be seen in ref. 21). It can be seen from these two figures that the sensitivity of elements S11, S12, and S33 on the degree of deformation (∆P) is weaker than that of S22 and S34. Further analysis of S34 /S11 suggests that circular polarization related light scattering signals may provide a sensitive approach to detect shape changes of deformed RBCs. Most light scattering measurements of cells signals, however, are often integrated over the full range of φs of a single cell or from suspensions of multiple cells. Therefore, in the following results we investigated the averaged Mueller matrix elements over φs, and the averaging may be carried over θ0 if desired. The rotational symmetry of the deformed RBCs allows at most 6 independent elements in the Mueller matrix 7, namely S11, S12, S22, S33, S34, and S44. Among them, the elements S33 and S44 are nearly identical in our cases because the cells can be classified as soft scatterers for their small refractive index contrast with the host medium, i.e., (nr−nh)/nh << 1.67 For this reason, only the results on S11, S12, S22, S33 and S34 are plotted in Figs. 12 and 13. The averaged element S33/S11, shown in Fig. 12, exhibits a good agreement with the Rayleigh-Gans (R-G) theory in its dependence on θs, obtained by averaging results from soft scatterers with random orientations 67, which provided additional validation of our parallel FDTD code.
237
S11 (normalized)
Modeling of Light Scattering by Single Red Blood Cells
103 102 101 100 10-1 10-2 10-3 10-4 10-5
θ0=0o
averaged
-S12/S11
1.0
averaged
0.5 0.0 -0.5
θ0=0o
-1.0
S33 / S11
1.0
θ0=0o
0.5
averaged
∆P=0 ∆P=3 ∆P=26
0.0 -0.5 -1.0 0
30
60
90
R-G theory
120
150
0
30
60
90
120
150
180
θs (degree) Figure 12. The angle-resolved Mueller matrix element of deformed RBCs at θ0=0o and averaged over 7 values of θ0 for 3 values of ∆P. In the lower-right panel an extra dash-dotdot line is added to represent S33/S11 calculated from the Rayleigh-Gans (R-G) theory.
S22/S11
1.000 0.999
averaged
0.992
∆P=0 ∆P=3 ∆P=26
0.998 1.0
S34/S11
1.000
o θi=0
0.984
o θi=0
0.5
0.976 0.2
averaged
0.0
0.0
-0.5 -1.0 0
30
60
90
120
150
0
30
60
90
120
150
-0.2 180
θs (degree) Figure 13. The angle-resolved Mueller matrix element S22 and S34 of deformed RBCs at θ0=0o and averaged over 7 values of θ0 for 3 values of ∆P.
238
5.
J.Q. Lu et al.
SUMMARY
In this chapter, we review the FDTD algorithm and development of a parallel code to numerically simulate light scattering by single RBC in terms of the Mueller matrix elements.20,21 The performance of the parallel code shows that, despite its reputation as a brute force method, the computing complexity of the accurate FDTD method can be considerably reduced with the parallel computing technique. Moreover, the modeling results of light scattering by RBC demonstrate the adaptability of the FDTD method for modeling of light interaction with biological cells of varied morphologies. During the last two decades, the FDTD method has been applied to various problems. As the performance-to-cost ratio of computer clusters continues to increase, we expect the FDTD method to maintain its significant expansion into the optical region from ultraviolet to nearinfrared and to provide a powerful tool for accurately modeling irregular cellular systems.
ACKNOWLEDGEMENTS This work was supported in part by a NIH grant (1R15GM70798-01) and by NPACI through supercomputer time allocations. Ping Yang also acknowledges support from the National Science Foundation (ATM0239605).
REFERENCES 1. 2.
3.
4.
5.
P. J. Wyatt, D. T. Phillips, Structure of single bacteria from light scattering, J. Theor. Biol., 37(3), 493-501, (1972). G. C. Salzman, J. M. Crowell, C. A. Goad, K. M. Hansen, R. D. Hiebert, P. M. LaBauve, J. C. Martin, M. L. Ingram, P. F. Mullaney, A flow-system multiangle lightscattering instrument for cell characterization, Clin Chem, 21(9), 1297-304, (1975). W. S. Bickel, J. F. Davidson, D. R. Huffman, R. Kilkson, Application of polarization effects in light scattering: A new biophysical tool, Proc. Nat. Acad. Sci. USA, 73, 486490, (1976). L. T. Perelman, V. Backman, M. Wallace, G. Zonios, R. Manoharan, A. Nusrat, S. Shields, M. Seiler, C. Lima, T. Hamano, I. Itzkan, J. Van Dam, J. M. Crawford,M. S. Feld, Observation of periodic fine structure in reflectance from biological tissue: a new technique for measuring nuclear size distribution, Phys. Rev. Lett., 80(3), 627-630, (1998). M. A. Yurkin, K. A. Semyanov, P. A. Tarasov, A. V. Chernyshev, A. G. Hoekstra, V. P. Maltsev, Experimental and theoretical study of light scattering by individual mature
Modeling of Light Scattering by Single Red Blood Cells
6.
7. 8. 9. 10. 11. 12. 13.
14.
15.
16. 17. 18.
19.
20. 21.
22.
23. 24.
239
red blood cells by use of scanning flow cytometry and a discrete dipole approximation, Appl Opt, 44(25), 5249-56, (2005). J. J. Bowman, T. B. A. Senior, P. L. E. Uslenghi, J. S. Asvestas, Electromagnetic and acoustic scattering by simple shapes, Rev. print. ed. (Hemisphere Pub. Corp., New York, 1987). H. C. van de Hulst, Light scattering by small particles. (Wiley, New York, 1957). A. L. Aden, M. Kerker, Scattering of Electromagnetic Waves from Two Concentric Spheres, J. Appl. Phys., 22(10), 1242-1246, (1951). O. B. Toon, T. P. Ackerman, Algorithms for the calculation of scattering by stratified spheres, Appl. Opt., 20(20), 3657-3660, (1981). T. Wriedt, A Review of Elastic Light Scattering Theories, Part. Part. Syst. Charact., 15, 67-74, (1998). A. Taflove, S. C. Hagness, Computational Electrodynamics: The Finite-Difference Time-Domain Method, 2nd ed. (Artech House, Boston, Mass., 2000). S. K. Yee, Numerical solutions of initial boundary problems involving Maxwell's equations in isotropic materials, IEEE Trans. Antennas. Propg., 14, 302-307, (1966). R. W. Lau, R. J. Sheppard, G. Howard, N. M. Bleehen, The modelling of biological systems in three dimensions using the time domain finite-difference method: II. The application and experimental evaluation of the method in hyperthermia applicator design, Phys Med Biol, 31(11), 1257-66, (1986). P. J. Dimbylow, O. P. Gandhi, Finite-difference time-domain calculations of SAR in a realistic heterogeneous model of the head for plane-wave exposure from 600 MHz to 3 GHz, Phys Med Biol, 36(8), 1075-89, (1991). G. B. Gentili, M. Leoncini, B. S. Trembly, S. E. Schweizer, FDTD electromagnetic and thermal analysis of interstitial hyperthermic applicators. Finite-difference time-domain, IEEE Trans Biomed Eng, 42(10), 973-80, (1995). A. Dunn, R. Richard-Kortum, Three-dimensional computation of light scattering from cells, IEEE J. Selected Topics Quantum. Electron., 2, 898-890, (1996). W. G. Scanlon, N. E. Evans, J. B. Burns, FDTD analysis of close-coupled 418 MHz radiating devices for human biotelemetry, Phys Med Biol, 44(2), 335-45, (1999). R. Drezek, A. Dunn, R. Richards-Kortum, A Pulsed Finite-Difference Time-Domain (FDTD) Method for Calculating Light Scattering from Cells Over Broad Wavelength Ranges, Opt. Express, 6(7), 147-157, (2000). D. Arifler, M. Guillaud, A. Carraro, A. Malpica, M. Follen, R. Richards-Kortum, Light scattering from normal and dysplastic cervical cells at different epithelial depths: finitedifference time-domain modeling with a perfectly matched layer boundary condition, J. Biomed. Opt., 8(3), 484-94, (2003). J. Q. Lu, P. Yang, X. H. Hu, Simulations of Light Scattering from a Biconcave Red Blood Cell Using the FDTD method, J. Biomed. Opt., 10(2), 024022, (2005). R. S. Brock, X. H. Hu, P. Yang, J. Q. Lu, Evaluation of a parallel FDTD code and application to modeling of light scattering by deformed red blood cells, Opt. Express, 13(14), 5279-5292, (2005). R. S. Brock, X. H. Hu, D. A. Weidner, J. R. Mourant, J. Q. Lu, Parallel FDTD Modeling of Light Scattering by a B-cell with 3D Structure Reconstructed from Confocal Images, submitted to J. Quant. Spectrosc. R.A., (2005). C. F. Bohren, D. R. Huffman, Absorption and scattering of light by small particles. (Wiley, New York, 1983). J. D. Jackson, "Classical electrodynamics," 3rd ed. (Wiley, New York: 1999) pp. Ch. 10.
240
J.Q. Lu et al.
25. P. Yang, K. N. Liou, Finite-difference time domain method for light scattering by small ice crystals in three-dimensional space, J. Opt. Soc. Amer., A, 13, 2072-2085, (1996). 26. E. S. Fry, G. W. Kattawar, Relationships between elements of the Stokes matrix, Appl. Opt., 20(16), 2811-2814, (1981). 27. M. I. Mishchenko, J. W. Hovenier, D. W. Mackowski, Single scattering by a small volume element, J. Opt. Soc. Am. A, 21(1), 71-87, (2004). 28. P. Yang, K. N. Liou, "Finite difference time domain method for light scattering by nonspherical and inhomogeneous particles," in Light Scattering by Nonspherical Particles, M. I. Mishchenko, J. W. Hovenier, and L. D. Travis, Eds. (Academic Press, San Diego: 1999) pp. Ch. 7. 29. G. Mur, Absorbing boundary conditions for the finite difference approximation of the time-domain electromagnetic-field equations, IEEE Trans. Electromagn. Compat., 23, 377-382, (1981). 30. A. Bayliss, E. Turkel, Radiation boundary conditions for wave like equations, Commun. Pure Appl. Math., 33, 707-725, (1980). 31. Z. Liao, H. L. Wong, B. Yang, Y. Yuan, A transmitting boundary for transient wave analyses, Scie. Sin., 27, 1063-1076, (1984). 32. R. L. Higdon, Absorbing boundary conditions for difference approximations to the multidimensional wave equation, Math. Comput., 47, 437-459, (1986). 33. J. P. Berenger, A perfectly matched layer for the absorption of electromagnetic waves, J. Comput. Phys., 114, 185-200, (1994). 34. D. S. Katz, E. T. Thiele, A. Taflove, Validation and extension to three dimensions of the Berenger PML absorbing boundary condition for FD-TD meshes, IEEE Microwave Guided Wave Lett., 4, 268-270, (1994). 35. S. A. Schelkunoff, Electromagnetic waves. (D. Van Nostrand Company inc., New York,, 1943). 36. P. W. Zhai, Y. K. Lee, G. W. Kattawar, P. Yang, Implemting the near- to far-field transformation in the finite-difference time-domain method, Appl. Opt., 43, 3738-3746, (2004). 37. W. Sun, Q. Fu, Z. Chen, Finite-difference time-domain solution of light scattering by dielectric particles with a perfectly matched layer absorbing boundary condition, Appl Opt, 38, 3141-3151, (1999). 38. X. Li, A. Taflove, V. Backman, Modified FDTD near-to-far-field transformation for improved backscattering calculation for strongly forward-scattering objects, IEEE Antennas and Wireless Propagation Letter, 4, 35-38, (2005). 39. S. Gedney, Finite-difference time-domain analysis of microwave circuit devices on high performance vector/parallel computers, IEEE Trans. on Microwave Theory and Techniques, 43, 2510-2514, (1995). 40. H. Hoteit, R. Sauleau, B. Philippe, P. Coquet, J. P. Daniel, Vector and parallel implementations for the FDTD analysis of milimeter wave planar antennas, International J. of High Speed Computing, 10, 209-234, (1999). 41. K. C. Chew, V. F. Fusco, A parallel implementaiton of the finite-difference timedomain algorithm, Int. J. Numerical Modeling,, 8, 293-299, (1995). 42. P. S. Excell, A. D. Tinniswood, K. Haigh-Hutchinson, Parallel computation of largescale electromagnetic field distributions, Appl. Comput. Electromagn. Soc. J., 13, 179187, (1998). 43. V. Varadarajan, R. Mittra, Finite-difference time-domain analysis using distributed computing, IEEE Microwave Gauided Wave Lett., 4, 144-145, (1994).
Modeling of Light Scattering by Single Red Blood Cells
241
44. J. W. M. Visser, "Analysis and sorting of blood and bone marrow cells," in Flow Cytometry and Sorting, M. R. Melamed, T. Lindmo, and M. L. Mendelsohn, Eds., 2nd ed. (Wiley, New York: 1990) pp. Ch. 33. 45. W. Groner, N. Mohandas, M. Bessis, New optical technique for measuring erythrocyte deformability with the ektacytomter, Clin. Chem., 26, 1435-1442, (1980). 46. D. H. Tycko, M. H. Metz, E. A. Epstein, A. Grinbaum, Flow-cytometric light scattering measurement of red blood cell volume and hemoglobin concentration, Appl. Opt., 24, 1355-1365, (1985). 47. P. Mazeron, S. Muller, H. El Azouzi, On intensity reinforcements in small-angle light scattering patterns of erythrocytes under shear, Eur. Biophys. J., 26, 247-252, (1997). 48. A. V. Priezzhev, O. M. Ryaboshapka, N. N. Firsov, I. V. Sirko, Aggregation and disaggregation of erythrocytes in whole blood: study by backscattering technique, J. Biomed. Opt., 4(1), 76-84, (1999). 49. C. T. Gross, H. Salamon, A. J. Hunt, R. I. Macey, F. Orme, A. T. Quintanilha, Hemoglobin polymerization in sickle cells studied by circular polarized light scattering, Biochim Biophys Acta, 1079(2), 152-60, (1991). 50. A. G. Borovoi, E. I. Naats, U. G. Oppel, Scattering of light by a red blood cell, J. Biomed. Opt., 3(3), 364-372, (1998). 51. S. V. Tsinopoulos, D. Polyzos, Scattering of He Ne laser light by an average-sized red blood cell, Appl. Opt., 38(7), 5499-5510, (1999). 52. S. V. Tsinopoulos, E. J. Sellountos, D. Polyzos, Light scattering by aggregated red blood cells, Appl Opt, 41(7), 1408-17, (2002). 53. O. W. Van Assendelft, Spectrophotometry of Haemoglobin Derivatives. (C.C. Thomas, Springfield, Ill., 1970). 54. R. B. Pennell, "Composition of normal human red cells," in The Red Blood Cell, C. Bishop and D. M. Surgenors, Eds. (Academic Press, New York: 1964. 55. G. Cicco, A. Pirrelli, Red blood cell (RBC) deformability, RBC aggregability and tissue oxygenation in hypertension, Clin Hemorheol Microcirc, 21(3-4), 169-77, (1999). 56. E. Evans, Y. C. Fung, Improved measurements of the erythrocyte geometry, Microvasc Res, 4(4), 335-47, (1972). 57. Y. C. Fung, W. C. O. Tsang, P. Patitucci, High-resolution data on the geometry of red blood cells, Biorheology, 18, 369-385, (1981). 58. L. G. Henyey, J. L. Greenstein, Diffuse radiation in the galaxy, Astrophys J, 93, 70-83, (1941). 59. S. Chien, R. G. King, R. Skalak, S. Usami, A. L. Copley, Viscoelastic properties of human blood and red cell suspensions, Biorheology, 12(6), 341-6, (1975). 60. A. J. Grimes, "Human Red Cell Metabolism." (Blackwell Scientific Pub, Oxford: 1980) pp. 57. 61. R. Skalak, P. I. Branemark, Deformation of red blood cells in capillaries, Science, 164(880), 717-9, (1969). 62. P. R. Zardar, S. Chien, R. Skalak, "Interaction of viscous incompressible fluid with an elastic body," in Computational Methods for Fluid-Solid Interaction Problems, T. L. Geers, Ed. (American Society of Mechanical Engineers, New York: 1977) pp. 65-82. 63. P. R. Zarda, S. Chien, R. Skalak, Elastic deformations of red blood cells, J Biomech, 10(4), 211-21, (1977). 64. T. W. Secomb, R. Skalak, N. Ozkaya, J. F. Gross, Flow of axisymmetric red blood cells in narrow capillaries, J. Fluid Mech., 163, 405-423, (1986).
242
J.Q. Lu et al.
65. T. W. Secomb, R. Hsu, A. R. Pries, Motion of red blood cells in a capillary with an endothelial surface layer: effect of flow velocity, Am. J. Physiol. Heart Circ. Physiol., H629-H636, (2001). 66. K. B. Aptowicz, Y. L. Pan, R. K. Chang, R. G. Pinnick, S. C. Hill, R. L. Tober, A. Goyal, T. Jeys, B. V. Bronk, Two-dimensional angular optical scattering patterns of microdroplets in the mid infrared with strong and weak absorption, Opt Lett, 29(17), 1965-7, (2004). 67. L. Turner, Rayleigh-Gans-Born Light Scattering by Ensembles of Randomly Oriented Anisotropic Particles, Appl. Opt., 12(5), 1085-1090, (1973).
OPTICS OF ERYTHROCYTES
Peter Tarasov,1 Maxim Yurkin,1,2 Pavel Avrorov,1,3 Konstantin Semyanov,1 Alfons Hoekstra,2 and Valeri Maltsev1,3 1
Institute of Chemical Kinetics and Combustion, Institutskaya 3, Novosibirsk,630090, Russia; Faculty of Science, Section Computational Science, of the University of Amsterdam, Kruislaan 403, 1098 SJ, Amsterdam, The Netherlands; 3 Novosibirsk State University Pirogova 3, Novosibirsk,630090, Russia; 2
Abstract:
We present optical methods to study erythrocytes and consider selected models to compute their light scattering.
Key words:
erythrocyte, cassini, DDA
1.
INTRODUCTION
Light-scattering properties of blood depend on the optical properties of red blood cells (RBC) because these cells form the disperse phase of blood. Moreover, the importance of RBC volume and hemoglobin concentration in clinical hematological analysis also emphasizes their importance. RBCs also play an important role in verification of solutions of the direct lightscattering problem for non-spherical particles because of their simple internal structure and stable biconcave discoid shape. Nevertheless the large size to wavelength ratio for an individual RBC (approximately 40) encourages the use of approximations in light-scattering theory and/or simplifying the shape in order to simulate light scattering from individual RBCs. For instance, RBCs have been assumed to be volume equivalent spherical,1,2 spheroidal,3 or ellipsoidal,4,5 dielectric particles. Such assumptions can only be valid under special experimental conditions, such as isovolumetric sphering, osmotic swelling or deformation in shear flow. Approximate theories, such as Fraunhofer and anomalous diffraction theories,4 Wentzel-Kramers-Brillouin6 or physical optics5 approximations 243 A. Hoekstra et al. (eds.), Optics of Biological Particles, 243–259. © 2007 Springer.
244
P. Tarasov et al.
can only be satisfactorily used for studies of the formation of structure peculiarities in the light-scattering patterns of individual RBCs. These approximations could not be applied in the solution of the inverse lightscattering problem to determine the RBC characterisitics from lightscattering data. Tsinopoulos and Polyzos7 were the first to rigorously simulate light scattering by realistic non-deformed RBCs. They used a boundary element method appropriately combined with fast Fourier transform (FFT) algorithms. This method also was applied to aggregated red blood cells.8 Another method that employs rotational symmetry of RBCs is the Discrete Sources Method (DSM), which was applied to RBC light scattering by Eremina et al.,9 who compared a realistic shape with spheroids and discspheres. Comparing the light-scattering profiles of biconcave disc-shaped RBC and approximating shapes, this study provided a determination of boundaries of usage of the approximating shapes in simulation of light scattering from mature RBCs. Most important conclusions are as follows: 1. The approximation of a RBC with a volume-equivalent sphere is not accurate. 2. The approximation of a RBC with a size–volume-equivalent oblate spheroid can be considered satisfactory for only the case of face-on incidence and quite successful only when one is interested in collecting data close to the forward direction, ≤ ~ 4°. 3. The scattering properties of a RBC are not essentially affected by the presence of the surrounded membrane. 4. The disk-sphere model of a RBC gives better agreement with the biconcave disk model comparedd with the spheroid model. While methods employing the axisymmetric properties of RBCs are the fastest, more general and widespread methods suitable for arbitrary-shaped scatterers, such as the Discrete Dipole Approximation (DDA) and the Finite Difference Time Domain method (FDTD), have recently been applied to RBC scattering. Karlsson et al.10 studied light scattering by RBCs with FDTD (λ from 600 to 1000 nm in vacuum), DDA (used only for validation of FDTD) and different approximate methods (Rytov approximation, superposition method). These approximate methods agree well with FDTD for small scattering angles. A similar analysis was done for two touching RBCs.11 Another FDTD study of RBC light scattering (λ = 633nm in vacuum) was performed by Lu et al.12 (see also their chapter in this volume). They used a simple geometrical parameterization of the RBC shape13 and studied effects of varying RBC shape, orientation, and the wavelength of the incident beam. Moreover, they also studied light scattering by RBCs deformed in blood flow,14 using a more advance parallel FDTD code.
Optics of Erythrocytes
245
Numerical results for the angular distributions of the Mueller scattering matrix elements of an RBC and their dependence on shape, orientation, and wavelength are presented. Yurkin et al.15 applied DDA to simulate light scattering by a RBC and compared this with experimental light-scattering patterns measured by a Scanning Flow Cytometer (SFC). A good agreement could be demonstrated, thus validating the biconcave disk model16 for a mature RBC. Moreover, this allowed the authors to solve the inverse lightscattering problem, resulting in the determination of a RBC’s diameter, volume and orientation in the flow from the measured SFC signal. Finally, scattering from the biconcave disk shape was compared with that of the spheroid and disc-sphere models. The conclusion of this15 and other9,11 papers was that simplified geometrical shapes models of RBCs are generally unsatisfactory. A detailed comparison of simulations by different authors is hindered by the fact that slightly different shape models are used for a native nondeformed RBC. Most papers7,8,10,11,15 describe a model based on the microscopic measurements by Fung et al.16 where an empirical polynomial fit for hypothetical intersample–averaged human cells is reported. Other shape models are due to Skalak et al.17 (used by Shvalov et al.6 and Eremina et al.9), a parametric model by Borovoi et al.13 (used by Lu et al.12), and a surface of rotation of a Cassini curve.5 In many particular cases all these approximate shape models may lead to the same satisfactory results. However, the validity of such shape models for the calculation of scattering by RBCs should be further studied for different sizes, and within the context of the specific application in which such scattering results are to be used. Until recently, the characterization of an erythrocyte population (in terms of volume distributions and hemoglobin concentrations) was achieved by isovolumetric sphering of the cells, followed by a 2-angular scattering measurement. This allows an efficient measurement of the mutual distribution of RBC volume and hemoglobin concentration.18,19 In the consequent sections new approaches to characterize erythrocytes from light scattering, with emphasis on application of scanning flow cytometry, are considered.20
2.
STATIC CHARACTERIZATION OF INDIVIDUAL ERYTHROCYTES
Classical flow cytometry of red cells uses low-angular and 90-degree scattering in combination with fluorescence.21 The low-angular scattering is
246
P. Tarasov et al.
thought to correlate with cell size. The main purpose of fluorescence is to discriminate between red cells and other cell types. A qualitative analysis of the resulting light scattering histograms, in combination with fluorescence results in important information about the mechanism of erythrocyte aging in vivo.22,23,24
2.1 Scanning Flow Cytometer A scanning flow cytometer20,25 (SFC) sufficiently extends the capabilities of flow cytometry in the morphological characterization of RBCs by measuring angularly resolved light-scattering profiles (LSP) of individual cells. The current set-up of the SFC provides measurement of the following combination of Mueller matrix elements: 2π
I (θ) =
∫ [S (θ, ϕ) + S (θ, ϕ)]dϕ 11
14
(1)
0
where I(θ) is the output signal of the SFC, θ and ϕ are the polar and azimuthal angles, with respect to incident radiation. A typical range of θ is from 50 to 800. The second term in the integral is zero for particles having a symmetry plane that contains the propagation vector of the incident beam.15 For arbitrary shaped particles this term is negligible. Absorption of hemoglobin is very small for the wavelength of incident radiation used in this study (0.6328 µm). Variations in the real part of the refractive index between erythrocytes can be attributed solely to variations in hemoglobin concentration.2 Absorption of hemoglobin is caused by an uncontrolled level of oxygenation variations. At the given wavelength, absorption can be neglected (~10-5).
2.2 RBC Sampling Whole blood was taken from a healthy volunteer by venopuncture with EDTA as anticoagulant and resuspended in buffered saline (0,01M HEPES buffer (Sigma), pH 7.4 with 0.15M NaCl). For the acid hemolysis erythrocytes were resuspended in 0.15M NaCl without buffer. Then the cells were stored at room temperature (230C) and used within three hours. Longer storage significantly affects kinetics of hemolysis and the volume of the cells. Experiments were also carried out at room temperature. All blood samples were taken from the same individual to allow for comparison. Murine cells were obtained from anaesthetized mice of a 57Bl6 line by cardiac puncture and were processed in the same manner. Murine cells were used to demonstrate the applicability of the analysis over a wide range of sizes.
Optics of Erythrocytes
247
2.3 Isovolumetric Sphering Isovolumetric sphering in the presence of SDS (sodium dodecyl sulfate) was applied to attain a spherical shape of the cells.18 The sphering close to ideal is achieved by balancing the weight ratio of SDS/protein at approximately 1/50 or less. The use of endogenous serum protein is possible: the dilution of the whole blood 1/50 in phosphate buffer saline with 0.003% SDS for a minute and consequent dilution 1/25 by isotonic saline of 0.1% glutaraldehyde and 0.001% SDS. The modification of the method has been chosen where 0.1% human serum albumin is present in the surrounding isotonic medium as a protecting agent and SDS concentration is 0.004%.
2.4 Lysing Procedures A volume containing 4µl of solution with erythrocytes was placed into a test-tube with 1ml of lysing solution to achieve a final concentration in the range of 5×105 – 2×107 cells/ml. To ensure homogeneity of cell concentrations the samples have been swirled for 2-3 seconds with a vortex mixer or stirred up by pipetting. The overall delay between introduction of a sample to lysing solution and start of measurements constituted about 1520 seconds. Isotonic ammonium chloride is a widespread erythrocyte lysing solution.26 The process of lysis is mediated by Jacobs Stewart’s cycle27 and based on transport properties of band 3 protein, which is present on mammalian RBCs. At present there is no complete mathematical model for this type of lysis. A typical composition of commercial lysing solution is as follows: NH4Cl (0.15M), NaHCO3 (~10 mM), disodium EDTA (~1 mM), pH=7.4. Ammonium chloride acts as equilibrating agent increasing intracellular osmolyte content.28 Dissolved CO2 - HCO3- pair performs a “catalysing” role with saturating concentrations of about 5 mM.29 At these concentrations a typical time of hemolysis is less than a minute, rather fast in comparison to the delay required to prepare a sample. Usually in “handmade” lysing solutions the addition of HCO3- ions is omitted, taking advantage of only atmospheric CO2 dissolved in water. At these conditions the whole process lasts about 5 minutes. We have chosen the following modification of lysing solution: addition of controlled intermediate concentrations of HCO3- ions (1 mM) in the form of sodium salt to degassed water (boiled in a microwave for 10 minutes), isotonic concentration of NH4Cl with 1.5mM buffer HEPES (Sigma). The resulting solution has buffer properties with pH ~ 7.2. Solutions prepared in this manner have been used for an hour.
248
P. Tarasov et al.
2.5 Osmotic Fragility Test The osmotic fragility was performed in a classical manner placing an aliquot of a stock solution of blood into a series of saline solutions of decreasing ionic strength with consequent calculation of a percentage of hemolysis. The lysis percentage was determined by measurement of the optical density of hemoglobin in supernatants. Measurements were performed using a microculture plate reader (Behring ELISA Processor II, Behringwerke, AG Diagnostica, Germany) at wavelengths of 570, 492, 450 nm to verify absence of variation between samples caused by oxygenation degree. Triplicate samples were tested at each concentration of saline. Hemolysis was expressed as a percentage of hemoglobin released in distilled water. Optical corrections for differences in concentration of saline were taken into account.
2.6 Method of Spectral Decomposition In order to provide an effective comparison of experimental and theoretical light-scattering data, the LSPs are modified30, 31 by multiplication with the weighting function f(θ) defined as
θ − θl F (θ) = f (θ) ⋅ I (θ) = sin 2 π θ h − θl
⋅ I (θ)
(2)
where θ l and θ k are the minimal and maximal significant scattering angles of the measured LSP. Typical values are θ l = 10° and θ h = 70°. The multiplication corresponds to a standard Hanning window procedure that strongly reduces the effects of discontinuities at the beginning and end of the sampling period of the SFC. Moreover, this function resembles the SFC instrument function25 and improves visual inspection of LSPs as the logarithmic scale can be substituted with a linear one. Peaks emerge in the frequency domain of the modified LSP, corresponding to light-scattering resonance on characteristic dimensions of the particle (Figure 1). The following empirical equation was derived to connect the location of the peak in the spectrum of modified light scattering LSP and the size parameter of a homogeneous spherical particle:31 α = C1 ⋅ Pf (3) where the coefficient C1 is ~ 189.12 for the given wavelength, α=m0πd/λ, d is the particle diameter, m0 is the refractive index of the medium, λ is the wavelength of incident light in vacuum, Pf is the position of the spectral peak. This equation gives a precision of 3% sizing a homogeneous spherical particle with refractive indices in the region from 1.37 to 1.70. This region of refractive indices covers the refractive indices of biological cells.
Optics of Erythrocytes
249 1.0
a)
b) "zero frequency peak" 0.8
15
Pf= 0.2 Af= 0.34
10
0.6
"main frequency peak" 0.4
5
Amplitude, arb. units
Modified indicatrix, F, arb. units
20
0.2
0
20
40
Angle, degrees
60
0.0
0.2
0.4
0.6
0.0 1.0
0.8
Frequency, (degree)
-1
Figure 1. Modified LSP F(θ) (a) of a spherical particle (diameter 5.7 µm, refractive index 1.39) and its normalized Fourier spectrum (b), where Pf is the location and Af is the amplitude of the main frequency peak.
2.6.1
Characterization of spherical RBC
The parametric solution of the inverse light-scattering problem for sphered erythrocytes is based on two parameters32: Pf and the integral of the modified LSP, denoted by J.
J=
θh
∫ f (θ) ⋅ I (θ)dθ
θl
V
1
3
∝ rsphere ∝ Pf
(4)
n R − n0 ∝ J const Pf ; const ≈ 0.43 β Here V and HBC are the volume and the hemoglobin concentration of a spherical particle, respectively, rsphere is the radius of the particle, β is the coefficient expressed in dL/g, nR is the real part of refractive index of a cell and n0 is the refractive index of the surrounding medium, β has a typical value of 0.0019 dL/g.2 HBC =
250
P. Tarasov et al.
The inversion algorithm is applicable to cells with spherical shape and volumes ranging from 30 µm3 to 250 µm3 and hemoglobin concentrations from 5 to 50 g/dL.32 With this method we analyzed murine and human RBCs to determine the distributions of cell volume and hemoglobin concentration. Two-dimensional contour plot shown in Figure 2 demonstrates the results of the analysis.
45
HBC, g/dl
40 human 35 30
mouse
25 40
60
80
100
120
3
V, µm
Figure 2. Two hematological indices of RBCs of human and mouse measured with SFC after isovolumetric sphering.
2.6.2
Recognition of spherical particles
Common practice to determine RBC size and hemoglobin concentration, as implemented in commercial analyzers, utilizes measurement over two spatial angles in the forward directions.2 This method allows characterization of spherized RBCs only. On the other hand this method does not distinguish between spherical and non-spherical cells, resulting in an uncertainty in determination of RBC characteristics. The LSPs of individual RBCs, as measured with the SFC, provide the possibility to recognize spherical cells before their characterization. We calculate the LSPs of oblate spheroids using the T-matrix method33 for spheroid
Optics of Erythrocytes
251
characteristics congruent with typical RBCs. We find that the amplitude of the peak Af of the normalized Fourier spectrum (i.e. the spectrum divided by the amplitude of the zero frequency peak) correlates with the aspect ratio of the spheroid. To model the conversion of an erythrocyte to a sphere, we assume an increase in volume of the spheroid while keeping the surface area constant (i.e. sphering in late stages of hemolysis). For our computations we use the available T-matrix code.34 Sphering of three spheroids with initial characteristics equal to a typical average and two extreme values of V0, and HBC0 have been simulated. The results shown in Figure 3 allow us to conclude that erythrocyte shape could be classified as a sphere if Af >0.25. This algorithm recognizes spheroids with aspect ratio ε ≥ 0.93. This algorithm allows us to estimate the error caused by non-sphericity ε ≥ 0.93 of the RBC in the determination of RBC characteristics with the spectral decomposition approach. We define the error in computation of the volume of a spheroid as
0.45
HBC0=33.7 g/dl, V0=88 fl
0.40
HBC0=31.2 g/dl, V0=74 fl
Af
0.35
HBC0=36.2 g/dl, V0=102 fl
0.30 0.25 0.20 0.15 0.10 0.6
0.7
0.8
0.9
1.0
aspect ratio of spheroid, ε
Figure 3. Amplitude of the main frequency peak Af as a function of aspect ratio ε of a spheroid. Graphs for mean and bounding values of RBC characteristics (V0, HBC0) are shown.
252
P. Tarasov et al.
ErrorV = Vcalc − Vtheor Vtheor
(5)
where Vtheor is the volume of spheroid used in the simulations, and Vcalc the volume obtained from the solution of the inverse problem, using model LSPs. The error was estimated for 200 theoretical LSPs of spheroids with ε ranging from 0.6 to 1, hemoglobin concentrations were 18.2, 28.9, 39.3 g/dL. The range of volumes was varied from 70 to 180 femtoliters. The resulting ErrorV as function of the aspect ratio is shown in Figure 4(a).
0.5
ErrorV
0.4
b)
a)
0.3
0.2
0.1
0.0 0.6
0.7 0.8 0.9 aspect ratio of spheroid, ε
0.1
0.2 0.3 0.4 amplitude of peak, Af
Figure 4. Error of RBC volume as a function of a) spheroid aspect ration and b) amplitude of the main frequency peak.
The results shown in Figure 4 allow us to fix the threshold in the amplitude of the peak Af such that the RBC volume is determined with an error smaller than 5%. The amplitude of the peak must exceed 0.25. Our approach allows us to estimate the fraction of non-spherical RBCs remaining during spherization. The amplitude of the spectral peak on spheroid aspect ratio calculated for typical RBC characteristics with the Tmatrix method is shown in Figure 5a). The distribution of the amplitudes of the spectral peak for isovolumetrically sphered RBCs measured with the SFC is shown in Figure 5b). The data shown in Figure 5 demonstrate that at least a part of the erythrocytes is not spherical. Nevertheless, the value of Af is larger than 0.2 for all cells, and the population can be characterized by our algorithm with precision in volume and concentration of hemoglobin of ~5%. Parameters obtained from this isovolumetric sphering experiments are
Optics of Erythrocytes
253
V 0 =87.1, wV0 =10.2, HBC 0 =33.4, wHBC0 =2.11, where wV0 and wHBC0 are the width of the distribution of RBC volume and hemoglobin concentration, respectively.
0.50
a)
b)
0.45
amplitude of peak, Af
0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00
0.80
0.85
0.90
0.95
1.00
aspect ratio of spheroid, ε
200
400
600
800
frequency count, a.u.
Figure 5. The amplitude of spectral peak as a function of spheroid aspect ratio calculated for typical RBC characteristics with T-matrix method a) and distribution of amplitude of spectral peak for isovolumetric sphered RBC measured with SFC.
2.6.3
Diameters of mature erythrocytes
The LSP structure of mature RBCs is sensitive to the RBC characteristics that may allow us to solve the inverse light-scattering problem to determine the RBC diameters from measured LSPs. We applied the spectral decomposition approach to relate the RBC diameter with a proper LSP parameter. This relation was discovered from simulations of light-scattering from biconcave disks using the Discrete Dipole Approximation (DDA).15 The LSPs and spectra were calculated for a cell volume of 100 fl and for different cell diameters ranging from 6.75 µm to 8.28 µm. The results of three representative simulations are shown in Figure 6. The symmetry axes of the disks were orthogonal to the direction of the incident laser beam. There are a few peaks in the spectra. We found that the location of the “last” essential peak (in the scale of degrees-1) corresponds to the diameter of the red blood cell.
254
P. Tarasov et al.
Intensity, arb. units
14
V = 100 µm 0 β = 90
12
3
10 8
d = 8.28 µm, ε = 0.294 d = 7.60 µm, ε = 0.380 d = 6.75 µm, ε = 0.542
6 4 2 0 10
15
20
25
30
35
40
45
50
Angle, degrees
Amplitude
0.2
0.1
0.0 0.0
0.1
0.2
0.3
0.4
0.5
-1
Frequency (degrees ) Figure 6. Modified theoretical light scattering LSPs (top) and their spectra (bottom) for biconcave disks of the three different shapes. V is the volume and d the diameter of the erythrocyte, ε is the ratio of the erythrocyte's thickness to diameter d. Cells are elongated along the direction of the incident laser beam.
In order to define in an unambiguous way this last essential peak we have to exclude the effect of background noise. The noise is due to laser noise of 0.5 % of the total power under current experimental conditions. A
Optics of Erythrocytes
255
3.0
0.06
2.5
0.05
2.0
0.04
1.5
0.03
Pf 1.0
0.02
0.5
0.0 10
L=0.003
20
30
40
50
Angle, degrees
0.0
0.2
0.4
0.6
Amplitude
Intensity, arb. units
considerable contribution to the noise for weak signals is caused by quantization noise of the analog-to-digital converter. This part was easily measured in the absence of a light-scattering signal and was 0.0006 in arbitrary units. The total noise level can be calculated by L=0.005×M + 0.0006, where M is the mean value of the measured signal. The essential peaks are defined as those peaks exceeding the noise level.
0.01
0.00
-1
Frequency, degrees
Figure 7. A typical experimental LSP (left) of a mature RBC with its spectrum (right). Calculated noise level L and the position of the last essential peak Pf are shown.
A typical LSP of an erythrocyte and its associated spectrum are shown in Figure 7. The location of the last essential peak has been used in equation (3) to calculate diameter of erythrocytes, d. Four thousand LSPs of mature RBCs were proceesed with this method and the resulting distribution of RBC diameters is shown in Figure 8. The mean diameter and width of the distribution are slightly different from the distribution parameters introduced by Fung et al.16 but the maximal and minimal diameters of both distributions are in good agreement. These results demonstrate the performance of the SFC and the spectral decomposition method to be a basis of a new hematological index – RBC diameter that can be determined in automatic regime.
256
P. Tarasov et al.
500
400
dmean
6.7 µm
σ
1.1 µm
Counts
300
200
100
0
2
4
6
8
Size, µm
10
12
14
Figure 8. The diameter distribution of mature RBCs measured with SFC.
3.
CONCLUSION
In this work we have illustrated how the SFC, in combination with a proposed inversion algorithm and common in vitro procedures can help to expand characterization of individual RBCs. For isovolumetric sphering, which is widespread in hematology analyzers, SFC provides the ability to monitor aspect ratios of swollen cells.
REFERENCES 1. J. M. Steinke and A. P. Shepherd, Comparison of mie theory and the light-scattering of red blood-cells, Appl. Opt. 27, 4027-4033 (1988). 2. D. H. Tycko, M. H. Metz, E. A. Epstein, and A. Grinbaum, Flow-cytometric light scattering measurement of red blood cell volume and hemoglobin concentration, Appl. Optics. 24, 1355-1365 (1985).
Optics of Erythrocytes
257
3. A. M. K. Nilsson, P. Alsholm, A. Karlsson, and S. Andersson-Engels, T-matrix computations of light scattering by red blood cells, Appl. Optics. 37, 2735-2748 (1998). 4. G. J. Streekstra, A. G. Hoekstra, E. J. Nijhof, and R. M. Heethaar, Light scattering by red blood cells in ektacytometry: Fraunhofer versus anomalous diffraction, Appl. Opt. 32, 2266-2272 (1993). 5. P. Mazeron and S. Muller, Light scattering by ellipsoids in a physical optics approximation, Appl. Opt. 35, (1996). 6. A. N. Shvalov, J. T. Soini, A. V. Chernyshev, P. A. Tarasov, E. Soini, and V. P. Maltsev, Light-scattering properties of individual erythrocytes, Appl. Opt. 38, 230-235 (1999). 7. S. V. Tsinopoulos and D. Polyzos, Scattering of He-Ne laser light by an average-sized red blood cell, Appl. Opt. 38, 5499-5510 (1999). 8. S. V. Tsinopoulos, E. J. Sellountos, and D. Polyzos, Light scattering by aggregated red blood cells, Appl. Opt. 41, 1408-1417 (2002). 9. E. Eremina, Y. Eremin, and T. Wriedt, Analysis of light scattering by erythrocyte based on discrete sources method, Opt. Comm. 244, 15-23 (2005). 10. A. Karlsson, J. P. He, J. Swartling, and S. Andersson-Engels, Numerical simulations of light scattering by red blood cells, IEEE Trans. Biomed. Engin. 52, 13-18 (2005). 11. J. He, A. Karlsson, J. Swartling, and S. Andersson-Engels, Light scattering by multiple red blood cells, J. Opt. Soc. Am. A 21, 1953-1961 (2004). 12. J. Q. Lu, P. Yang, and X. Hu, Simulations of light scattering from a biconcave red blood cell using the finite-difference time-domain method, J. Biomed. Opt. 10, 024022 (2005). 13. A. G. Borovoi, E. I. Naats, and U. G. Oppel, Scattering of light by a red blood cell, J. Biomed. Opt. 3, 364-372 (1998). 14. R. S. Brock, X. Hu, P. Yang, and J. Q. Lu, Evaluation of a parallel FDTD code and application to modeling of light scattering by deformed red blood cells, Opt. Expr. 13, 5279-5292 (2005). 15. M. A. Yurkin, K. A. Semyanov, P. A. Tarasov, A. V. Chernyshev, A. G. Hoekstra, and V. P. Maltsev, Experimental and theoretical study of light scattering by individual mature red blood cells with scanning flow cytometry and discrete dipole approximation, Appl. Opt. 44, 5249-5256 (2005). 16. Y. C. Fung, W. C. Tsang, and P. Patitucci, High-resolution data on the geometry of red blood cells, Biorheology 18, 369-385 (1981). 17. R. Skalak, A. Tozeren, R. P. Zarda, and S. Chien, Strain energy function of red blood cell membranes, Biophys. J. 13, 245-264 (1973). 18. Y. R. Kim and L. Ornstein, Isovolumetric sphering of erythrocytes for more accurate and precise cell volume measurement by flow cytometry, Cytometry 3, 419-427 (1983). 19. N. Mohandas, Y. R. Kim, D. H. Tycko, J. Orlik, J. Wyatt, and W. Groner, Accurate and independent measurement of volume and hemoglobin concentration of individual red cells by laser light scattering, Blood 68, 506-513 (1986). 20. V. P. Maltsev and K. A. Semyanov, Characterization of bioparticles from light scattering (Vista Science Press, Netherlands, 2004). 21. B. H. Davis, Diagnostic utility of red cell flow cytometric analysis, Clin. Lab Med. 21, 829-840 (2001). 22. L. K. Jennings, L. K. Brown, and M. E. Dockter, Quantitation of protein 3 content of circulating erythrocytes at the single-cell level, Blood 65, 1256-1262 (1985).
258
P. Tarasov et al.
23. S. J. Nance, Flow cytometry related to red cells, Transfus. Sci. 16, 343-352 (1995). 24. N. J. Nusbaum, Red cell age by flow cytometry, Med. Hypotheses 48, 469-472 (1997). 25. V. P. Maltsev, Scanning flow cytometry for individual particle analysis, Rev. Sci. Instruments 71, 243-255 (2000). 26. M. I. Tiirikainen, Evaluation of red blood cell lysing solutions for the detection of intracellular antigens by flow cytometry [see comments], Cytometry 20, 341-348 (1995). 27. V. L. Lew and R. M. Bookchin, Volume, pH, and ion-content regulation in human red cells: analysis of transient behavior with an integrated model, J. Membr. Biol. 92, 57-74 (1986). 28. M. D. Sass, Effect of ammonium chloride on osmotic behavior of red cells in nonelectrolytes, Am. J. Physiol 236, C238-C243 (1979). 29. R. J. Labotka, W. Galanter, and V. M. Misiewicz, Erythrocyte bisulfite transport, Biochim. Biophys. Acta 981, 358-362 (1989). 30. K. A. Semyanov K. A. and P. A. Tarasov, Measurement of mammalian erythrocyte indices from light scattering with scanning flow cytometer, in Diagnostic Optical Spectroscopy in Biomedicine II. Edited by Wagnieres, Georges A. Proceedings of the SPIE, 5141, 106-113 (2003). 31. K. A. Semyanov, P. A. Tarasov, A. E. Zharinov, A. V. Chernyshev, A. G. Hoekstra, and V. P. Maltsev, Single-particle sizing from light scattering by spectral decomposition, Appl. Opt. 43, 5110-5115 (2004). 32. P. A. Tarasov, Study of characteristic parameters of population of human erythrocytes using dynamic flow cytometry, [in Russian], PhD thesis, Krasnoyarsk, 2005. 33. M. I. Mishchenko, L. D. Travis, and D. W. Mackowski, T-matrix computations of light scattering by nonspherical particles: A review, J. Quant. Spectrosc. Radiat. Transf. 55, 535-575 (1996). 34. http://www.giss.nasa.gov/~crmim/
Optics of Erythrocytes
259
From left to right, front: Pavel Avrorov, Alexander Shvalov, Daria Orlova, Valeri Mun, Irina Kolesnikova, Valeri Maltsev, Dina Goloshchapova, Alexey Zharinov, Sergey Potapov; Back: Konstantin Gilev, Konstantin Semyanov, Vyacheslav Nekrasov, Peter Tarasov, Ilya Skribunov, Dmitry Strokotov, Andrei Chernyshev, Maxim Yurkin.
Konstantin Semyanov demonstrates the SFC to ARW participants.
OPTICS OF PLATELETS
Irina Kolesnikova,1,2 Sergey Potapov,1 Semyanov,1 and Valeri Maltsev1,2
Peter
Tarasov,1
Konstantin
1
Institute of Chemical Kinetics and Combustion, Institutskaya 3, Novosibirsk,630090, Russia; Novosibirsk State University Pirogova 3, Novosibirsk,630090, Russia;
2
Abstract:
Optical methods to study platelets with recent application to scanning flow cytometry are presented.
Key words:
platelet, spheroid, T-matrix, DDA
1.
INTRODUCTION
Platelets are a small subcellular fragment of blood that have an important role in maintaining haemostatic and arterial thrombosis, particularly in fibrillation. Activated platelets adhere to an injury and aggregate to form a haemostatic plug or thrombus. Blood platelet volume, size and shape are markers of their activation and function that can be used to distinguish activated and non-activated platelets. Characterization of human platelets is frequently used as a tool in investigation of disease, such as ischaemic heart disease, cerebrovascular disease, renovascular disease, etc.1 Just one characteristic of platelets, platelet volume, is available in instrumental clinical analysis. Unfortunately this instrumental analysis does not help us to clarify optical properties of these cells because these instruments utilize aperture-impedance techniques (e.g. Coulter S Plus).2 This approach does not take into account the shape variability of platelets. We believe that an enhanced characterization of platelets may have clinical relevance. With this work we have analyzed results obtained for blood platelets3 using a Scanning Flow Cytometer (SFC)4,5 that allows measurement of the 261 A. Hoekstra et al. (eds.), Optics of Biological Particles, 261–267. © 2007 Springer.
262
I. Kolesnikova et al.
angular dependence of the light-scattering intensity (the light-scattering profiles, LSP) of individual cells in the region ranging from 50 to 1200. The SFC output signal is proportional to Mueller matrix element S11 integrated over the azimuthal angle. In order to retrieve morphological characteristics from a platelet’s LSP we have to solve the inverse light-scattering problem. The obvious and only existing approach to determine characteristics of an arbitrary particle from LSP is the direct comparison of the experimentally measured LSP with the theoretically calculated LSP based on an optical model of the particle. Previously we have demonstrated the validity of this approach for characterization of red blood cells from light scattering.6
2.
CHARACTERIZATION OF INDIVIDUAL BLOOD PLATELETS FROM LIGHT SCATTERING
2.1 Sample Preparation Blood was withdrawn from the antecubital vein of normal volunteers with syringe and then put into a tube filled with EDTA as an anticoagulant. Care was taken to ensure a ratio of 1.5 mg EDTA to 1 ml blood in sample. All surfaces in contact with the blood were plastic. The sample was stored at room temperature. The measurement was conducted about 30 minutes after venesection.
2.2 Choice of Shape for Model Platelet: T-Matrix Simulation A platelet that is a discoid cell with diameter of 2–4 µm and thickness of 0.5–2 µm in the nonactivated state, and it becomes a spicular spheroid in the activated state. Activation of platelets starts when oxygen appears in the blood plasma. The process of platelet activation includes transformation of the cell to aggregate into a thrombus – pseudopodia emerge from a discoid platelet that becomes spicular.7 We applied the T-matrix method to simulate light scattering of a platelet modeled as an oblate spheroid, because pseudopodia does not have a substantial effect on the LSP in the current configuration of the SFC.3 A recent review of the -matrix approach has been performed by Mishchenko et al.8, and we applied the public-domain T-matrix code from Mishchenko (http://www.giss.nasa.gov/~crmim/) in simulations of light scattering of individual platelets.9
Optics of Platelets
263
2.3 Results We continuously measured 2000 LSPs of platelets with the SFC and few of them are shown in Figure 1.
T-matrix simulation SFC experiment
rv=0.9 µm
10
ε=3
rv=0.8 µm
0
β=50
Intensity, arb. units
1
ε=3 0
β=40 0.1
10
rv=0.8 µm
rv=0.8 µm
ε=3
ε=3 0
0
β=20
1
β=10
0.1
10
20
30
Angle, degrees
40
10
20
30
Angle, degrees
40
50
Figure 1. Typical light-scattering profiles of blood platelets measured with SFC (points), and best-fit LSPs of oblate spheroids calculated with T-matrix method (solid lines). The spheroid characteristics, volume-equivalent-sphere-radius rV, axis ratio ε, polar orientation angle β were used to calculate the best-fit light-scattering profile.
The characteristics of platelets were determined from a fit of the experimental and theoretical LSPs. We varied volume-equivalent-sphereradius rV, polar orientation angle β of oblate spheroid to find the best-fit LSP for LSP experimentally measured with the SFC. The polar orientation angle is the angle between the direction of the incident radiation and the axis of the oblate spheroid. The axis ratio ε and refractive index n of oblate spheroids were fixed at 3 and 1.41, respectively. We were unable to vary all spheroid characteristics because the iteration procedure requires an incredible amount of time for calculations.
264
I. Kolesnikova et al.
0.06
0.04
Amplitude
Intensity, arb. units
0.2
Pf
0.1
0.02
0.0 10
20
30
40
50
Angel, degrees
60 0.0
0.1
0.2
0.00 0.4
0.3 -1
Frequency (degree )
Figure 2 . Typical experimental light-scattering profile transformed by Hanning window (left) with spectrum (right) for a single platelet.
In order to determine the volume of measured blood platelets we apply the spectral approach introduced in our previous work.10 The spectral approach assumes transformation of the LSP with a standard Hanning window procedure and FFT. The typical transformed LSP and spectrum are shown in Figure 2. The location of the maximum peak Pf was used in equation (1) for calculation of the volume-equivalent-sphere-diameter d of platelets: α = 189.12 ⋅Pf ,
(1)
where α=m0πd/λ, m0 is the refractive index of the medium, λ is the wavelength of incident light in vacuum. The spectral approach allows us to form the volume distribution of blood platelets (solid line in Figure 3).
Optics of Platelets
265
SFC
Coulter (Vmean=9.4 fL)
150
Counts
(Vmean=10.3 fL)
100
50
0
0
5
10
15
20
25
30
Volume, fL
Figure 3. The volume distributions of platelets for the same blood sample obtained with SFC (solid line) and with a Coulter analyzer (dash line).
The volume distribution of platelets of the same blood sample was measured on the hematology analyzer Coulter MAXM (dash line in Figure 3). We found a good agreement between both volume distributions except for small volumes. Our results demonstrate that the maximum peak in the LSP spectrum for individual platelets corresponds to the diameter of the volume-equal sphere and SFC can be used in characterization of blood platelets from light scattering.
3.
CONCLUSION
A new method to determine the volume of individual blood platelets has been introduced. The method is based on Scanning Flow Cytometry and the measurement of multi-angle light scattering and a solution of the inverse light-scattering problem with the spectral approach. The parameters of the platelet volume distribution are in good agreement with independent measurements from commercial instruments and literature values.
266
I. Kolesnikova et al.
This work gives access to important hematological parameters and platelet volume, which can be measured using the optical technique. A clinical protocol should be developed for the SFC with the aim of measuring the mean and the width of the distribution of blood platelet volume.
REFERENCES 1. Bath P.M.V. and Butterworth R.J. Platelet size: measurement, physiology and vascular disease. Blood Coagulation and Fibrinilysis 1996; 7: 157-161. 2. Rowan R.M., Fraser C., Gray J.H. and McDonald G.A. The Coulter counter model S Plus - the shape of things to come. Clin Lab Haematol 1979; 1: 29-40. 3. I.V. Kolesnikova, S.V. Potapov, M.A. Yurkin,c, A.G. Hoekstra, V.P. Maltsev and K.A. Semyanov, Determination of volume, shape and refractive index of individual blood platelets, J. Quant. Spectrosc. Radiat. Transf. (accepted for publication). 4. V.P. Maltsev, Scanning flow cytometry for individual particle analysis, Rev. Sci. Instruments 71, 243-255 (2000). 5. V.P. Maltsev and K.A. Semyanov, Characterisation of Bio-Particles from Light Scattering Inverse and Ill-Posed Problems Series (VSP, Utrecht, 2004). 6. M.A. Yurkin, K.A. Semyanov, P.A. Tarasov, A.V. Chernyshev, A.G. Hoekstra, and V. P. Maltsev, Experimental and theoretical study of light scattering by individual mature red blood cells with scanning flow cytometry and discrete dipole approximation, Appl. Opt. 44, 5249-5256 (2005). 7. Michelson A.D. Platelets. New York: Academic Press/Elsevier Science, 2002. 8. M.I. Mishchenko, L.D. Travis, and D.W. Mackowski, T-matrix computations of light scattering by nonspherical particles: A review, J. Quant. Spectrosc. Radiat. Transf. 55, 535-575 (1996). 9. V.P. Maltsev, Light scattering of biological particles: instrumental solution, inverse light-scattering problem, experimental results, in Proceedings of 7th Conference on Electromagnetic and Light Scattering by Nonspherical Particles: Theory, Measurements, and Applications, Bremen, Germany, September 8-12, 2003, p. 145146. 10. K.A. Semyanov, P.A. Tarasov, A.E. Zharinov, A.V. Chernyshev, A.G. Hoekstra, and V.P. Maltsev, Single-particle sizing from light scattering by spectral decomposition, Appl. Opt. 43, 5110-5115 (2004).
Optics of Platelets
267
ARW Participants enjoy a nature hike.
OPTICS OF LEUCOCYTES
Konstantin Semyanov,1 Alexey Zharinov,1 Peter Tarasov,1 Maxim Yurkin,1,2 Ilya Skribunov,1 Dirk van Bockstaele,4 and Valeri Maltsev1,3 1
Institute of Chemical Kinetics and Combustion, Institutskaya 3, Novosibirsk,630090, Russia; 2 Faculty of Science, Section Computational Science, of the University of Amsterdam, Kruislaan 403, 1098 SJ, Amsterdam, The Netherlands; 3 Novosibirsk State University Pirogova 3, Novosibirsk,630090, Russia; 4 Antwerp University & Hospital, Wilrijkstraat 10, B-2650 Edegem, Belgium
Abstract:
Optical methods to study neutrophils, eosinophils, basophils, lymphocytes, and monocytes are reviewed. Recent applications of scanning flow cytometry to characterize lymphocytes and monocytes is presented.
Key words:
morphology, erythrocyte, platelet, lymphocyte, monocyte, neutrophil, basophil, mononuclear, granulocyte
1.
DIFFERENTIATION OF LEUCOCYTES FROM LIGHT SCATTERING
Ordinary flow cytometry has been applied to discriminate white blood cells. Granulocytes, monocytes, lymphocytes occupy unique positions in the two-dimensional space created by the forward- and side-scattered light intensities. De Grooth et al.1 proposed measurement of the depolarization of side-scattered light to discriminate within the granulocyte subpopulation. Measurement of the depolarized side-scattering enables one to discriminate human eosinophilic granulocytes from neutrophilic granulocytes.2 Finally they measured forward light scattering, orthogonal (side) light scattering, and the fluorescence intensities of unlysed peripheral blood cells labeled with CD45-phycoerythrin (CD# is cluster designation number) and the nucleic acid dyes LDS-751 and thiazole orange utilizing an ordinary flow 269 A. Hoekstra et al. (eds.), Optics of Biological Particles, 269–280. © 2007 Springer.
270
K. Semyanov et al.
cytometer. Erythrocytes, reticulocytes, platelets, neutrophils, eosinophils, basophils, monocytes, lymphocytes, nucleated erythrocytes, and immature nucleated cells occupy unique positions in the five-dimensional space created by the listmode storage of the five independent parameters.3 Such a technique is implemented in cytometrical protocol, employing higher depolarized orthogonal light scatter of eosinophils.4,5,6 Unfortunately light scattering measured in a fixed solid angle does not provide characterization of blood cells, i.e. determination of cell characteristics from light-scattering data. In order to systematically study optical properties of nucleated blood cells we must increase the quantity and improve the quality of lightscattering information read from individual cells. granulocyte monocyte lymphocyte
Intensity, arb.units
6
4
2
0
0
100
200
300
400
time, µs
Figure 1. Examples of LSTs of three basic types of leucocytes: granulocyte, monocyte, and lymphocyte. The zero point in time scale corresponds to about 900, 100 µs – 550, 150 µs – 230, 225 µs – 100, 400 µs – 50.
In the technique of scanning flow cytometry, the angular scattering profile of a moving particle is measured as a function time as the regions of the scattered field are reflected by a spherical mirror to a detector. The light-scattering trace (LST) contains both the forward-scattering (FSC) and side-scattering (SSC) signals of individual particles, i.e. the temporal dependence of the scattering intensity of the moving particles. The LST can be transformed into a light-scattering profile (LSP), i.e. the scattering phase function or scattering intensity as a function of angle. This transformation and detailed description of the scanning flow cytometry technique are described elsewhere.7,8,9,10 Examples of LSTs of leukocytes are shown in Figure 1. The integrals of the LST over different angular ranges are a superior alternative to the biparametric FSC/SSC plots, since they provide much more data. Three populations of leukocytes can be gated out (Figure 2) using the two integrals of the time-resolved LST: the integral from 350 µs to 400 µs corresponds to the FSC of ordinary cytometers and the integral
Optics of Leucocytes
271
from 0 µs to 200 µs corresponds to SSC of ordinary cytometers. The LST provides an opportunity to select a larger quantity of parameters, up to the number of measuring points, for differentiation of leukocyte or others cells.
Integral of LST (0-200), arb. u.
700
granulocytes
600 500 400 300 200 100 0
monocytes lymphocytes 0
10
20
30
40
50
60
Integral of LST (350-400), arb. u.
Figure 2. A leukocytes biparametric diagram of a light-scattering trace integral over 350-400 µs versus the light-scattering trace integral over 0-200 µs.
2.
LEUCOCYTE SIZING
Since the Scanning Flow Cytometer (SFC) increases the amount of light-scattering data read from individual cells we are able to test the applicability of the spectral approach11 in sizing leucocytes. The spectral approach assumes transformation of the LSP with a standard Hanning window procedure and FFT to produce a modified LSP. The typical modified LSP of leucocytes and their spectra are shown in Figure 3. The location of the peak Pf is used for calculating the volume-equivalentsphere-diameter d of a cell: (1) α = 189.12 ⋅Pf , where α = m0πd/λ, m0 is the refractive index of the medium, λ is the wavelength of the incident light in vacuum. The LSPs of individual blood cells are measured with a SFC and the fast Fourier transform (FFT) procedure was applied to the modified LSPs. The resulting spectrum is formed by a number of peaks. In order to determine a
272
K. Semyanov et al.
size of the cell the location of the maximal frequency peak in the spectrum must be identified.9 The threshold amplitude Ath that cuts noisy peaks from the LSP structure peaks can be estimated for each LSP with the following formulae: Ath = 0.005M + 0.0006,
(2)
where the first term is due to ~5% noise of the laser and the second term is due to noise of the A/D converter and background. Coefficient M is the mean value of the measured signal. The typical light-scattering profiles of three types of leucocytes: granulocytes, monocytes and lymphocytes and their spectra are shown in Figure 3. The horizontal line in the spectrum plot corresponds to the threshold amplitude Ath. The location of the maximal frequency peak Pf of the spectrum is used in equation (1) for the calculation of the effective size of leucocytes. We identify leukocyte types accordingly to the integral×size map shown in Figure 4 (a), where the integral is an integral of the LSP from 10 to 70 degrees. Each point on this map corresponds to a single leukocyte. The resulting size distributions for different types of leucocytes are shown in Figure 4 (b). The mean size d and standard deviation s.d. are as follows: d = 8.0 µm and s.d. = 2.0 µm for lymphocytes; d = 10.8 µm and s.d. = 1.5 µm for monocytes; d = 12.4 µm and s.d. = 2.3 µm for granulocytes. The measured mean sizes of leucocytes are in agreement with literature data.12,13 Similarity of plots in Figure 2 and in Figure 4 (a) confirms the wellknown fact that forward light scattering strongly correlates with cell size. However the dots in Figure 4 (a) demonstrate a correlation between cell size and the integral of the LSP; whereas, the analogous correlation is not observed in Figure 2. Assuming a correlation between the LSP integral and the size of the cellular nucleus we are able to conclude that large cells have large nuclei. This statement should be investigated with independent technique such as a light microscopy.
3.
OPTICS OF MONONUCLEAR CELLS
White blood cells, which include granular and agranular, cells are an important part of the body's immune system, helping to destroy invading microorganisms. The lymphocytes and monocytes are agranular cells with very clear cytoplasm. The cytoplasm is transparent. Monocytes and lymphocytes are distinguished by having a nucleus that may be eccentric in location, and a relatively small amount of cytoplasm. The small ring of cytoplasm contains numerous ribosomes. The nucleus is round and large in comparison to the cell and it occupies most of it. In any case, some of the cytoplasm remains visible, generally in a lateral position.
Optics of Leucocytes
273
0.10
Granulocyte
2.0
0.08 0.06
1.5
Pf
1.0
0.04
0.5
L=0.005
0.8
0.032
0.6
0.024
Pf
0.016
0.4
L=0.002
Amplitude
Intensity, arb. unit
Monocyte
0.2
0.008 0.000
0.0
Lymphocyte
0.20
0.008
0.15
0.006
Pf 0.10
0.004
L=0.001
0.05
20
30
40
Angel, degrees
50
Amplitude
Intensity, arb. unit
0.02 0.00
0.0
0.00 10
Amplitude
Intensity, arb. unit
2.5
0.002 0.000
0.0
0.2
0.4
0.6 -1
Frequency, degree
Figure 3. Typical experimental light-scattering profiles (left) and their spectra (right) for a single granulocyte (top), monocyte (middle) and lymphocyte (bottom)
In this study we analyze the light scattering of the most important subpopulation of white blood cells – lymphocytes. According to the quantity of cytoplasm, lymphocytes are divided into small, medium and large. These cells play an important role in the human immune response. The T-lymphocytes act against virus infected cells and tumor cells. The Blymphocytes produce antibodies. Formation of the lymphocyte LSPs was analyzed with the aim of development of an appropriate optical model of a lymphocyte to solve the inverse light-scattering (ILS) problem for their characterization. The light scattering of lymphocytes was simulated by means of an algorithm14 that allows calculations of the scattering matrix of two concentric spheres with the following characteristics: d and nc are the diameter and refractive index of the inner sphere; D and ns are the diameter and refractive index of the outer sphere, respectively. Additionally we were able to simulate lymphocyte light scattering with a multi-layered model15 defined by the layer diameter di and layer refractive index ni where i is the number of layer. The choice of the models is based on an analysis of experimentally measured LSPs of individual lymphocytes.10
274
K. Semyanov et al.
Integral, arb. units
(a) granulocytes
60
40
lymphocytes
20
monocytes
0
Counts
(b)
monocytes lymphocytes granulocytes
120
80
40
0
0
4
8
12
Size, µm
16
20
Figure 4. The (a) integral×size map and (b) the size distribution of leucocytes obtained using a SFC.
A sample that contains 5 × 105 lymphocytes was analyzed with the Scanning flow Cytometer. The spectral decomposition approach11 was applied to the measured LPSs. The modified LSPs of two single lymphocytes with spectral decomposition are shown in Figure 5 by points. To retrieve the lymphocyte characteristics from the measured LPS we fitted
Optics of Leucocytes
275
the experimental LSPs to theoretical LSPs while varying the parameters of the two concentric spheres as follows: D and d – cell and nucleus diameters, respectively; nc and ns – refractive index of nucleus and cytoplasm, respectively. The results shown in Figure 5 are as follows: (a) lymphocyte characterized by d = 6.66 µm, nc = 1.440, D = 9.17 µm, ns = 1.358; (c) lymphocyte characterized by d = 6.38 µm, nc = 1.456, D = 8.74 µm, ns = 1.355. The lymphocyte characteristics retrieved were used to calculate the modified LSP and spectral decomposition. These modified LSPs with spectral decomposition are shown in Figure 5 by grey lines. There is substantial disagreement between theory and experiment in the figure although the location of the maximal peak in the spectrum correlates with the diameter of the inner sphere. This disagreement is probably caused by the insufficient two-layer model of a single lymphocyte. 30 25 20
(a) experiment theory (two layers model) theory (five layers model)
(b)
(c)
(d)
15 Intensity (A.U.)
10 5 0
12.5 10.0 7.5 5.0 2.5 0.0 10
20 30 40 50 60 Scattering angle (deg.)
70 0
20 40 60 80 100 120 140 -1 Frequency index (deg. )
Figure 5. (a), (c) The modified light-scattering profiles and (b), (d) spectral decomposition of two single lymphocytes measured with the SFC. The experimental data is marked by points; whereas, the theoretical fits are marked by grey and black lines for two and five layers models, respectively.
In order to clarify the reason for disagreement of the theoretical and experimental results, we modeled a lymphocyte using a five-layered sphere. The experimental LSPs shown in Figure 5 were fitted by particles formed by five layers. The resulting particles can be characterized by the following parameters: (a) lymphocyte characterized by d1 = 1.468 µm, n1 = 1.402, d2 = 6.264 µm, n2 = 1.478, d3 = 7.895 µm, n3 = 1.368, d4 = 8.650 µm, n4 =
276
K. Semyanov et al.
1.360, d5 = 9.518 µm, n5 = 1.342; (c) lymphocyte characterized by d1 = 1.418 µm, n1 = 1.404, d2 = 5.975 µm, n2 = 1.465, d3 = 6.606 µm, n3 = 1.425, d4 = 8.918 µm, n1 = 1.366, d5 = 9.708 µm, n5 = 1.342. The corresponding LSPs with spectral decomposition are shown in Figure 5 by black lines. The five-layered model evidently has given better agreement between experiment and theory. On the other hand we have found that the central part of the lymphocyte has smaller refractive index compared with the next layer. We are assuming that the first and second layers are affected mostly by internal properties of the nucleus such as nucleolus, plaques of chromatin. The third layer is probably related with external properties defining the shape of the nucleus: indentations and over-all deviation from spherical shape. The last two layers could be related to the cytoplasm surrounding the nucleus which sometimes displays internal structures such as mitochondria and granules. We used the outer edge of the third layer as estimation for the diameter of the nucleus and the edge of the fifth layer to represent the diameter of the cell. The refractive index of the cytoplasm is considered to be equal the refractive index of the fourth layer. The refractive index of the nucleus was estimated as the volumetric average of refractive indices of the first, the second and the third layer.16,17 The theoretical study of scattering of multi-layered spheres and experimental measurement of LSPs of individual lymphocytes has allowed us to conclude that a lymphocyte could not be modelled by two concentric spheres. The multi-layered sphere or eccentric spheres models must be applied to simulate light scattering of individual lymphocytes.
4.
OPTICS OF GRANULOCYTES
Light scattering from a suspension of granulocytes was employed in dynamic studies. Light scattering intensity was used as a measure of neutrophil aggregation18 and shape change.19 Neutrophil degranulation was shown to correlate with the orthogonal light-scattering intensity20 and granulocyte aggregation with light extinction.21 Flow cytometrical studies of neutrophil biology were reviewed by Carulli.22 Granulocytes can be discriminated from other leukocytes in flow cytometers by their higher forward and side scattering (except basophils that are found in the light-scattering region of lymphocytes).23 To perform further discrimination of granulocytes into subclasses, additional measured parameters are needed. One characterization method23 employs measuring the CD45 antigen presence on the cell surface using labeled monoclonal antibodies. Eosinophils, lymphocytes, and monocytes express higher densities of CD45 than neutrophils and basophils.
Optics of Leucocytes
277
Another parameter of interest is autofluorescence. Its high values for eosinophils can be used to discriminate them from all other leucocytes.4,24 However, this parameter is not reliable, since it can be altered by chemical treatment of the cells or when using fluorescent dyes. It was combined with measurements of CD16 (specific for neutrophils) or CD49d (specific for eosinophils) to discriminate eosinophils from neutrophils.25 Eosinophils express several CD2 subfamily receptors on their surface: NTB-A, CD224 (2B4), CD84, CD58, and CD48. Neither basophils nor neutrophils express NTB-A. Basophils also express CD224 but not neutrophils. CD84, CD58, and CD48 are expressed by all granulocytes. Therefore, NTB-A presence can be used to identify eosinophils and CD224 absence – neutrophils.26 One of the first reliable methods to evaluate basophils in flow cytometry was based both on CD45 expression (low density) and on the presence of IgE on the cell surface, which can be combined with detection of low orthogonal light scattering compared to other granulocytes.27 Improvement of this method based on the expression of CRTH2 (chemoattractant receptor-homologous molecule expressed on Th2 cells) / DP2 (second receptor of prostaglandin D2), which is also expressed by Th2 cells and eosinophils, was proposed.28 CRTH2 detection is combined with CD3 negative (exclude Th2) and low side-scattering (exclude eosinophils) to identify basophils. The angular dependence of the light-scattering intensity was measured for granulocytes and lymphocytes. Orientation effects on the intensity of side scattering in different experimental setups were discussed.29
REFERENCES 1.
2.
3.
4.
5.
B. G. De Grooth, L. W. Terstappen, G. J. Puppels, and J. Greve, Light-scattering polarization measurements as a new parameter in flow cytometry, Cytometry 8, 539544 (1987). L. W. M. M. Terstappen, B. G. de Grooth, K. Visscher, F. A. van Kouterik, and J. Greve, Four-parameter white blood cell differential counting based on light scattering measurements, Cytometry 9, 39-43 (1987). L. W. M. M. Terstappen, D. Johnson, R. A. Mickaels, J. Chen, G. Olds, J. T. Hawkins, M. R. Loken, and J. Levin, Multidimensional flow cytometric blood cell differentiation without erythrocyte lysis, Blood Cells 17, 585-602 (1991). S. Lavigne, M. Bosse, L. P. Boulet, and M. Laviolette, Identification and analysis of eosinophils by flow cytometry using the depolarized side scatter-saponin method, Cytometry 29, 197-203 (1997). L. W. M. M. Terstappen, B. G. De Grooth, K. Visscher, F. A. van Kouterik, and J. Greve, Four-parameter white blood cell differential counting based on light scattering measurements, Cytometry 9, 39-43 (1988).
278 6.
7.
8. 9. 10. 11.
12.
13.
14. 15. 16.
17.
18.
19.
20.
K. Semyanov et al. A. G. Hoekstra and P. M. A. Sloot: Biophysical and Biomedical Applications of NonSpherical Scattering, in M.I. Mishchenko; J.W. Hovenier and L.D. Travis, editors, Light Scattering by Nonspherical Particles, Theory, Measurements, and Applications, pp. 585-602. Academic Press, San Diego, San Francisco, New York, Boston, London, Sydney, Tokyo, 2000. ISBN: 0-12-498660-9. V. P. Maltsev, A. V. Chernyshev, Method and device for determination of parameters of individual microparticles, US Patent Number: 5,650,847. Date of patent: Jul. 22, 1997. J. T. Soini, A. V. Chernyshev, P.E. Hanninen, E. Soini, and V. P. Maltsev, A new design of the flow cuvette and optical set-up for the scanning flow cytometer, Cytometry, 31, 78-84 (1998). V. P. Maltsev, Scanning flow cytometry for individual particle analysis, Review of Scientific Instruments, 71, 243-255 (2000). V. P. Maltsev and K. A. Semyanov, Characterization of bioparticles from light scattering (Vista Science Press, Netherlands, 2004). K. A. Semyanov, P. A. Tarasov, A. E. Zharinov, A. V. Chernyshev, A. G. Hoekstra, and V. P. Maltsev, Single-particle sizing from light scattering by spectral decomposition, Appl. Opt. 43, 5110-5115 (2004). R. J. Cave, R. L. Holder, T. K. Moms, J. Taylor, D. Smith, and N. K. Shinton, An evaluation of the Technicon H6000 haematology system. Clin Lab Haematol 5, 203-14 (1983). B. Weinberg, Mononuclear phagocytes, in: Wintrobe's Clinical Hematology, 11th ed., J. P. Greer, J. Foerster, and J. N. Lukens, eds. (Lippincott Williams & Wilkins Publishers, Baltimore, USA, 2003), v.1, pp. 349-386. C. F. Bohren and D. R. Huffman, Absorption and Scattering of Light by Small Particles (Wiley, New York, 1983). W. Yang, Improved algorithm for light scattering by a multilayered sphere, Appl Opt 42, 1710–1720 (2003). A. Zharinov, D. van Bockstaele, M. Lenjou, and K. Semyanov, Characterization of mononuclear blood cells from light scattering, In Abstracts of the NATO Advanced Research Workshop On Optics of Biological Particles Akademgorodok (Academic Town), Novosibirsk, Russian Federation, October 3 — October 6, 2005, Eds. V. Maltsev, A. Hoekstra, G.Videen, pp. 49-50. A. Zharinov, P. Tarasov, A. Shvalov, K. Semyanov, D. R. van Bockstaele, and V. Maltsev, A Study of Light Scattering of Mononuclear Blood Cells with Scanning Flow Cytometry, J. Quant. Spectrosc. Radiat. Transf. (accepted for publication). M. Niwa, Y. Kanamori, K. Kohno, H. Matsuno, O. Kozawa, M. Kanamura, and T. Uematsu, Usefulness of grading of neutrophil aggregate size by laser-light scattering technique for characterizing stimulatory and inhibitory effects of agents on aggregation, Life Sci. 67, 1525-1534 (2000). M. U. Ehrengruber, D. A. Deranleau, and T. D. Coates, Shape oscillations of human neutrophil leukocytes: characterization and relationship to cell motility, J. Exp. Biol. 199, 741-747 (1996). L. A. Sklar, Z. G. Oades, and D. A. Finney, Neutrophil degranulation detected by right angle light scattering: spectroscopic methods suitable for simultaneous analyses of degranulation or shape change, elastase release, and cell aggregation, J. Immunol. 133, 1483-1487 (1984).
Optics of Leucocytes
279
21. I. Yuli and R. Snyderman, Light scattering by polymorphonuclear leukocytes stimulated to aggregate under various pharmacologic conditions, Blood 64, 649-655 (1984). 22. G. Carulli, Applications of flow cytometry in the study of human neutrophil biology and pathology, Haemapathol. Mol. Hematol. 10, 39-61 (1996). 23. L. W. M. M. Terstappen, D. Johnson, R. A. Mickaels, J. Chen, G. Olds, J. T. Hawkins, M. R. Loken, and J. Levin, Multidimensional flow cytometric blood-cell differentiation without erythrocyte lysis, Blood Cells 17, 585-602 (1991). 24. G. J. Weil and T. M. Chused, Eosinophil autofluorescence and its use in isolation and analysis of human eosinophils using flow microfluorometry, Blood 57, 1099-1104 (1981). 25. A. M. Thurau, U. Schylz, V. Wolf, N. Krug, and U. Schauer, Identification of eosinophils by flow cytometry, Cytometry 23, 150-158 (1996). 26. A. Munitz, I. Bachelet, S. Fraenkel, G. Katz, O. Mandelboim, H. U. Simon, L. Moretta, M. Colonna, and F. Levi-Schaffer, 2B4 (CD244) is expressed and functional on human eosinophils, J. Immunol. 174, 110-118 (2005). 27. P. Gane, C. Pecquet, P. Lambin, N. Abuaf, F. Leynadier, and P. Rouger, Flow cytometric evaluation of human basophils, Cytometry 14, 344-348 (1993). 28. R. Boumiza, A. L. Debard, and G. Monneret, The basophil activation test by flow cytometry: recent developments in clinical studies, standartization and emerging perspectives, Clinical and Molecular Allergy 3, 9 (2005). 29. D. Watson, N. Hagen, J. Diver, P. Marchand, and M. Chachisvilis, Elastic light scattering from single cells: orientational dynamics in optical trap, Biophys. J. 87, 1298-1306 (2004).
Dirk van Bockstaele and Valeri Maltsev entertain the entertainers.
280
K. Semyanov et al.
Conference participants in the midst of scientific discussion
INDEX
Absorbing boundary condition (ABC), 210 absorption cross section, 213 ACPLA, 64 aerosol classification, 88 ambient aerosols, 138 ammonium nitrate, 134 ammonium sulfate, 134 angular scattering, 23 Arizona road dust, 149, 150 astrobiology, astrobiological signatures, 185 Asymmetry factor Af, 47, 131 Bacillus globigii, 106-114, 116, 118, 119, 132, 137, 140, 141, 150 Bacillus subtilis, also see Bacillus globigii, 134 backscatter, 43 bacteria, 31 BAWS, 110, 132, 133 biconcave shape of RBC, 220, 221 biological/non-biological distinction, 77 BioLert, 110 biosignatures, 191 Bio-Vigilant, 110 Born-Oppenheimer approximation, 67
carbon black, 134 cell scattering, 1 cell size, 20 cellular organelle scattering, 23 Central Processing Unit (CPU), 204 chirality, 193 chlorophyll, 194, 195 chondritic meteorites, 186, 188 CIBADS, 110 circular dichroism, circular polarization, 187, 195, 196, 197 circular dichroism spectroscopy, 197 confusion matrix 89-92 Courant-Friedrichs-Levy (CFL) condition, 210 cross polarization, 14 cytometry, 233 cytoplasm, 5 deformed RBC, 220, 224 degree of symmetry, 49 detection threshold, 84 detectors, 101-106 detectors, CCD and ICCD, 102-103 detectors, APD, 103 detectors, PMT 104-106 DFA, 130, 142 diesel exhaust, 113-114
281
282 discrete dipole approximation, DDA, 34, 232 discrete Fourier transform, 212 elastic scattering, 31 energy levels, 67 Erwina herbicola, 106, 122, 134 erythrocytes, size, 241 Escherichia coli, 118, 134 Europa, 186 excitation/emission spectra, 75-79 excitation wavelengths, 94 extinction cross section, 213 extrasolar planets, 188-190 extremophiles, 186 false positive, 84 finite-difference time-domain, FDTD, 21, 35, 204, 232 FLAPS, 110, 133 FLASAS, 110, 119-121, 124, 125, 133 fluorescence, 61 fluorescence, calibration, 171 fluorescence cross section, 68 fluorescence, data analysis, 170 fluorescence emission, fluorophors, 71 fluorescence, energy requirements, 69 fluorescence, particle size, 70 fluorescence Quantum yields, 68 fluorescence spectrum, 67 Fourier transform, 249, 255 Franck-Condon shift, 65 FTIR, 145, 147, 148 glycine, 187 glycolaldehyde, 187 granulocyte, 254, 255, 261 gypsum, 126, 132 habitable zone, 190 homochirality, 188, 195 Hönl-London factor, 67 HULIS, 129
Index IJAG, 106-110 inelastic scattering, 65 LATAOS, 42 LASS, see LIBS leucocytes, 253 leucocyte, sizing, 255 LIBS, 146 LIF for phytoplankton, 167, 176 LIF apparatus for phytoplankton, 169 LIPS, see LIBS Lorenz-Mie theory, 32 lymphocyte, 254, 255 lysing, 235 Mars, 186 Message Passing Interface (MPI), 216 methane, 192 Mie approximation, 27 Minimum threat concentration, 87 monocyte, 254, 255 mononuclear cells, 259 MS-2, 106 Mueller matrix, 207, 208 Murchison meteorite, 188 NADH, 62, 114, 116-118, 129, 132, 133 non-spherical particles, 54 nuclear heterogeneity, 27 nucleus scattering, 25 optical activity, 195, 196 ovalbumin, 106, 113, 114 oxygen, ozone, 191, 192 ozone, 191, 192 PAHS, 129, 187 PCA, 130, 141-144 PCR 145, 146 Perfectly Matched Layers (PML) ABC, 210, 211 performance metrics, 84 phenyalanine, 62 phytoplankton, 161
Index phytoplankton cultures, 168 phytoplankton morphology, 175 phytoplankton, species recognition, 173 pigment concentration, 180 pigment ratios, 181 platelets, 247 prebiotic material, 187 processing element (PE), 214 Puffer, 148-150 pyrimidine, 187 radial basis functions, 55 Raman spectroscopy, 147 red blood cell (RBC), 205, 219, 231 red edge, 193 refractive index of RBC, 220 resonance Raman effect, 65, 66, 95 response time, 84 refractive index, cell, 21 riboflavin, 149 ROC, 92-94 scanning flow cytometry, 234, 248 scattering amplitudes, 12, 207, 212, 213 scattering anisotropy, 24 scattering, azimuthal, 38 scattering coefficients, 12 scattering cross-section, 24 scattering, dual wavelength, 56
283 scattering efficiencies, 11 scattering fluctuations, 15 scattering phase function, 24, 208 scattering, polar, 36 scattering, two-dimensional, 40 sensitivity, 84 size parameters of most biological cells, 203 sources, 96-101 sources, lamps, 98-99 sources, LED, 100-101 space missions, 186, 187, 190 spectral decomposition, 236 SPFA, 111 SPFS, 111 spherical particles, 238 sphericity index, 49 spheroidal particles, 33 T-matrix, 33, 248 translation coefficients, 16 tryptophan, 62, 118 two-wavelength excitation, 62 tyrosine, 62 urban aerosols, 44 UV excitation, 62 vector spherical harmonics, 3, 4 white space, 210, 211 WIBS, 111, 114-119 Yersinia rohdei, 113, 114
284
Index
At the coffee break