© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
IN THIS ISSUE
Focus on optical imaging This issue brings together a selection of articles on new techniques in optical imaging. Over the past decade, the move from single-photon confocal microscopy to multiphoton excitation has transformed optical imaging, enabling the capture of high-resolution three-dimensional images of living cells and tissues. Zipfel et al. discuss recent achievements in multiphoton microscopy and the most immediate challenges for this field, which include the optimization of beam power and characteristics to increase tissue penetration and the development of lower-energy lasers with which to expand the spectrum of fluorescent dyes [Reviews, p. 1369]. With some simple modifications, a multiphoton laser-scanning microscope can also image second-harmonic generation signals from samples, which enable high-resolution, nondestructive imaging of structures within cells and tissues without the need for external labeling. Because the technique does not involve excitation of fluorophores, it is less likely to induce phototoxic effects or photobleaching in tissue, which often is a problem of fluorescence microscopy [Perspective, p. 1356]. Fluorescence can be used to image molecular interactions through the phenomenon of resonance energy
transfer, which has given rise to a multitude of techniques for measuring intra- and intermolecular distances below at below ∼10 nm. Jares-Erijman and Jovin present an overview of these techniques and propose a few new variants of their own [Reviews, p. 1387]. Great strides are also being made in breaking the diffraction limit of light microscopy. Superresolution techniques have achieved axial resolutions in the tens of nanometers, promising to delineate three-dimensional structures inside the cell in unprecedented detail [Perspective, p. 1347]. Another way in which light can be exploited in imaging is optical coherence tomography (OCT), an approach that measures the echo time delay and magnitude of light (in a process analogous to ultrasound). OCT is already being used in the clinic as a type of ‘optical biopsy’ to probe tissue pathology at shallow depths [Perspective, p. 1361]. And in a final article, which provides a link between conventional optical imaging and scanned probe microscopy, Lewis et al. describe near-field scanning optical microscopy, which is enabling researchers to image the surfaces of biological materials with hitherto unachievable resolution. According to the authors, the true potential of this technology will only be realized once scanned probe microscopy devices are fully integrated into optical microscopes. [Reviews, p. 1378] AM, GTO, KA & MS
Superresolution imaging
Anthrax antidote
‘Stimulated emission depletion’ (STED) is a technique for surpassing the diffraction-limited resolution of light microscopy by sculpting the microscope’s focal spot. Previously, Hell and colleagues used STED-4Pi microscopy to image cells whose membranes had been labeled nonspecifically with a hydrophobic dye, attaining an axial resolution of ∼33 nm. Now they have demonstrated the method for standard immunofluorescence labeling of a cytosolic structure, achieving an axial resolution of ∼50 nm. Human embryonic kidney cells were fixed and stained with an anti-β-tubulin antibody and a secondary antibody coupled to a red dye. Individual microtubules could be clearly resolved (right), whereas the confocal reference showed an amorphous mass (left). The diameter of the decorated microtubules was found to be 60–70 nm, in agreement with electron microscopy results. [Brief Communications, p. 1303] KA
Bowdish and colleagues have identified high-affinity human antibodies to anthrax toxin by panning phage display libraries of Fabs, derived from immunized donors, against the protective antigen components of anthrax toxin. Two of the selected Fabs and their corresponding IgGs were effective at protecting rats from challenge with anthrax toxin. These antibodies may prove useful for treating individuals exposed to anthrax. [Brief Communications, p. 1305] KA
In This Issue was written by Kathy Aschheim, Aaron Bouchie, Michael Francisco, Andrew Marshall, Meeghan Sinclair and Gaspar Taroncher-Oldenburg.
Sequence of a killer The bacterium Photorhabdus luminescens has a fascinating and complex life cycle that includes a symbiotic phase in the gut of a nematode and an entomocidal (insect-killing) phase within a variety of insects attacked and infested by the nematodes carrying it. This life style is possible because P. luminescens is capable of unleashing a range of strategies to poison and control its hosts during the different phases of its life cycle. The full genome of P. luminescens provides a rich trove of information on the organism’s insect killing strategies, information that should be useful to researchers attempting to adapt some of these mechanisms in designing new methods to fight insect pests. [Articles, p. 1307; News and Views, p. 1294] GTO
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
vii
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
IN THIS ISSUE
Enzymatic tour de force
Hi-Fi yeast reconstruction
Heparan sulfate (HS) proteoglycans regulate many different biological processes such as blood coagulation, viral infection, tumor metastasis and cell growth. However, there are substantial limitations on their synthesis by chemical means, which has hindered investigation of proteoglycan function. Rosenberg and coworkers report the biosynthesis of HS oligosaccharides using an engineered set of cloned enzymes that mimics the Golgi apparatus in vitro. They assembled antithrombin III–binding heparan sulfate pentasaccharide in six steps, instead of the 60 required for chemical synthesis. Their yield was also twofold greater and 100 times faster than that achieved with chemical synthesis. The approach may also be applicable to the synthesis of other oligosaccharides. [Letters, p. 1343] MS
With the advent of full genome sequences, proteomes and ever–expanding data sets on protein–protein and protein-DNA interactions, the need for algorithms that can generate a high-fidelity image of a cell’s inner workings from the individual parts has grown. Gifford and colleagues describe GRAM (genetic regulatory motives), an algorithm that integrates gene expression and protein-DNA interaction data to reconstruct a ‘physically’ informed network of regulatory interactions in yeast. This approach contrasts with the more ‘functionally’ informed algorithms used so far, and allows the authors to piece together a regulatory network in yeast linking 655 genes with 68 transcription factors. [Articles, p. 1337; News and Views, p. 1295] GTO
Adenoviral tumor lysis and immunity
Parsing chemokine receptor function Thus far, around 20 receptors for chemokines (chemotactic cytokines) have been identified, but teasing out their functions has proven difficult. Borrowing a viral stratagem for attacking the immune system, Su and colleagues find an original way of inactivating chemokine receptors. With the knowledge that the HIV-1 protein Vpu downregulates CD4 through proteasomal degradation in the endoplasmic reticulum, the authors show that a fusion protein consisting of a chemokine and the C-terminal domain of Vpu localizes to the endoplasmic reticulum and downregulates the associated chemokine receptor. The approach is demonstrated in vitro for three different chemokine receptors. Experiments in lethally irradiated mice revealed that the chemokine receptor CXCR4 is required for homing of transplanted hematopoietic stem cells to the bone marrow. [Articles, p. 1321] KA
Patent roundup • A ruling issued on September 26 by the US Court of Appeals for the Federal Circuit (Washington, DC, USA) in Festo v. Shoketsu Kinsoku Kogyo Kabushiki declares that patentees can use the doctrine of equivalents in infringement litigation only if the equivalent invention would have been ‘unforeseeable’ at the time of the amendment. [News in brief, p. 1263] AB • A summary of recently published patent applications involving gene expression includes antisense oligonucleotides and synthetic ribozymes capable of inhibiting gene expression, and a method for monitoring the expression of a specific gene in vivo using nuclear magnetic resonance signal modification. [Patents, p. 1399] MF • The Federal Circuit recently ruled in Hoffmann-La Roche v. Promega that a patent may be invalid because an example was written in the past tense, when the experiments had not been conducted as written. Here, Potter and Talukder provide guidelines for writing examples based upon experimental work that has been done, in contrast to work that is planned. [Patents, p. 1397] MF
viii
Tumor replication–selective adenoviruses (oncolytic adenoviruses) have shown promising results in immunodeficient mouse-human tumor xenograft models. However, these results have not been replicated in clinical trials with cancer patients where rapid viral clearance has been observed and tumor regression was shortlived or not seen. Kirn and coworkers have now developed an immunocompetent tumor efficacy model in mice to study the role of adenoviral E3 immunoregulatory proteins’ interactions with the immune system and their effects on tumor killing efficacy. The majority of oncolytic adenoviruses in clinical trials have deletions in these proteins. The researchers found that deletion of the E3B gene region results in decreased viral gene expression, decreased replication, accelerated viral clearance and/or reduced antitumoral efficacy in their model. These results have important implications for the field of oncolytic viral therapy. [Articles, p. 1328] MS
Cytokines live long and prosper Cytokines, such as interferon-β (IFN-β), are key mediators of cellular communication. Their therapeutic use is hampered by rapid clearance from the bloodstream and toxic side effects when administered systemically. Chernajovsky and colleagues have engineered a long-lived, targeted cytokine by attaching the latency-associated protein (LAP) of transforming growth factor-β1 and a matrix metalloproteinase (MMP) cleavage site to the cytokine IFN-β. The LAP provides IFN-β with a protective cover, shielding it from degradation, and also preventing it from activating its receptors until it reaches sites of inflammation where MMPs are expressed, such as in the joints of patients with arthritis. At these sites, MMPs cleave off the LAP, releasing the active cytokine. The researchers show that the engineered cytokine has a 40-fold increase in half-life compared with normal IFN-β and has a greater therapeutic effect in a mouse model of arthritis. [Articles, p. 1314; News and Views, p. 1293] MS
Next month in • Targeted antibiotics • Engineering an antibody into GFP • Protein oligomerization method
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
EDITORIAL
Out of the shadows he accolades have been rolling in for the use of magnetic resonance in biological imaging. Richard Ernst received a Nobel for contributions to the development of high-resolution nuclear magnetic resonance spectroscopy, and last year Kurt Wüthrich got another for application of the phenomenon in determining the three-dimensional structure of biological macromolecules in solution. Just last month, Paul Lauterbur and Sir Peter Mansfield also received their summons to a fancy dinner in Stockholm for demonstrating the use of gradients in a magnetic field for creating two-dimensional images of internal structures and its refinement through mathematical analysis—findings that provided the foundation for magnetic resonance imaging, a diagnostic tool now in use in over 20,000 clinics and laboratories worldwide. But what of light microscopy, for so long the ubiquitous and routine imaging tool of the amateur and professional biological investigator? This issue of Nature Biotechnology presents a series of articles describing recent developments in the optical imaging field. And great things are clearly afoot. Since its initial description by Robert Hooke, the light microscope has been augmented with all manner of gadgets and innovations: phase contrast, differential interference contrast, laser confocal scanning systems, video, solid-state cameras, lasers and image analysis software to name a few. And yet, two seemingly insurmountable constraints on the technology have remained. The first of these is Abbe’s resolution limit (or the diffraction limit)—the smallest distance that can be resolved between two lines by optical instruments. The best that most confocal microscopes with single or even multiphoton excitation can achieve is a (spatial) resolution of 180 nm in the focal plane (x,y) and only 500–800 nm along the optic (depth) axis (z). For biologists, unfortunately, most macromolecular complexes and signaling domains have dimensions of ∼5–500 nm and the largest virus (pox virus) has a diameter of 250 nm. Thus, we have lacked the means to image, in real time and in live samples, biologically relevant molecules and entities at a resolution less than 200 nm. The good news is that several pioneering super-resolution technologies, including I5 microscopy, 4Pi microscopy and stimulated emission depletion microscopy (see p. 1347), are now taking the resolution of light microscopes beyond this limit. Lensless technology, such as scanning near-field optical microscopy (see p. 1378)— a technique that crosses the boundary between atomic force microscopy and optical microscopy and provides information about surfaces at spatial (x,y) resolutions down to 50 nm and to 10 nm in the axial (z) plane—is also breaking new ground.
T
Unfortunately, these techniques are also rather rough on their labeling agents, causing photo-bleaching (essentially light-mediated destruction of the label), which could potentially compromise attempts to improve resolution. The other major problem for optical imaging is that biological tissues are very good at absorbing and scattering light. This limits analysis of cellular events to just a few hundred micrometers below the surface. In this respect, microscopes that use near infrared, longer wavelength light, multiphoton absorption or optical coherence tomography (p. 1361) are now achieving greater tissue penetration (up to 2–3 mm) than traditionally thought possible, with the additional benefits of reduced photodamage of tissues and longer probe lifetimes. And while some are working to broaden the palette of reporters available (e.g., through mutagenesis of fluorescent proteins to extend excitation peaks and emission maxima to longer wavelengths), others are focusing on technologies that dispense with reporters altogether, attempting instead to visualize cellular structures through the measurement of intrinsic fluorescence. Looking ahead, the current renaissance in optical imaging technologies bodes well for biology and medicine. Until now, most light microscopy has focused on probes that report transcriptional activity. As it becomes increasingly clear that a large proportion of the signaling pathways and regulatory mechanisms in the cell act not at the level of transcription but rather at the level of protein-protein interactions and within specific cellular compartments, optical techniques for monitoring a protein’s local physico-chemical environment and the proteins in its immediate vicinity will become increasingly important. Microarrays and other global assays of gene expression activity that have dominated biotech in recent years will be increasingly complemented by imaging technologies for visualizing a much greater spectrum of cellular processes, including mRNA turnover, protein phosphorylation and glycosylation states, translation initiation and progress, and DNA structural and chemical modification. As the technology is both extended from molecular imaging to the visualization of cell, tissues, anatomy and physiology, and combined with other types of imaging (e.g., positron emission tomography, computed tomography and ultrasound), its promise for improving the speed and accuracy of disease diagnosis is quite real and definitely not the stuff of biotech entrepreneurial dreams. The latest $9.5 billion endorsement of this promise came in October when the world’s largest company by market value, General Electric, bought Amersham. Wondering what is one of Amersham’s core businesses? Contrast agents for enhancing the imaging of organs and tissue.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1251
NEWS
US NHGRI launches chemical attack on drug development gap The US National Institutes of Health NIH; Bethesda, MD, USA has launched a major push into an arena until now almost the sole dominion of drug and biotech companies: the large scale screening of small molecules, both to probe protein function and to develop drug leads. The Molecular Libraries initiative, which is part of the NIH’s Roadmap for Medical Research announced on September 30, is the biggest public effort to date in the relatively new field of chemical genomics— the genome-wide screening of libraries of small molecules to find new drug targets and leads. The frustrating failure to commercialize many publicly funded medical discoveries, as much as the desire to annotate the human genome, is driving the new initiative. “None of us are served by the current lack of success of developing novel therapeutics,” says Chris Austin, a senior advisor to Francis Collins, director of the National Human Genome Research Institute NHGRI; Bethesda, MD, USA, which will spearhead the new initiative. “That’s one of the reasons we’re doing what we’re doing.” Several universities, charitable foundations and regional initiatives have
Maggie Bartlett, NHGRI/NIH
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ALSO IN THIS SECTION Public–private partnerships boost research on neglected diseases p1254 Concerns raised over declining antiinfectives R&D p1255 Japan’s biotech sector shows signs of life p1256 GM confusion in Brazil p1257 News in brief p1261
NIH Director Elias Zerhouni and NHGRI Director Francis Collins are launching initiatives that may cloud the intellectual property landscape for drug development companies.
already taken steps to address this development gap (see Box and p. 1254), and any additional source of cash to address the problem is generally seen as welcome.
The NIH is quick to say that it’s not getting into the drug business. Although drug candidates will certainly emerge, the “NIH’s main purpose should not be to do what the phar-
Box 1 University strategies to fill the development gap Fed up with medical inventions that go unlicensed or ignored, more universities are trying to move them forward themselves. “Universities are starting to realize that the NIH isn’t putting all this money into their coffers to get professors promoted,” says Colleen Brophy, director of the Center for Protein and Peptide Pharmaceuticals at Arizona State University (AzBio; Tempe, AZ, USA). The Center is part of ASU’s new Arizona Biodesign Institute, which is headed by George Poste, former R&D chief of SmithKline Beecham (Philadelphia, PA, USA). AzBio’s goal is to get medical discoveries to market as soon as possible. In August, three major California research Universities joined with the nonprofit clinical research organization Strategic Research Institute (SRI; Palo Alto, CA, USA) to form PharmaSTART, a drug development consortium (Bioentrepreneur, 11 September 2003, doi:10.1038/bioent767). “Pharmaceutical companies frequently are not interested in small market products or niche drugs,” says
PharmaSTART steering committee member Jerrold Olesfsky, an endocrinologist at the University of California (San Diego). “It isn’t like there’s a hundred biotech companies out there just waiting to take on a new development project.” These efforts have ambitious goals, but money remains the sticking point. AzBio’s stated mission is to “generate protein-based pharmaceuticals that are clinically applicable for targeted disease processes such as vascular disease, cancer and wound healing.” AzBio plans to commercialize its discoveries conventionally, through a venture capital—funded startup company or biotech licensing deals. PharmaSTART investigators will receive free consulting services from SRI, but no money has been earmarked for actual drug development. The most advanced universitybased program to date is the Laboratory for Drug Discovery in Neurodegeneration at Harvard University (Cambridge, MA, USA), which won a $37.5 million grant from an anonymous donor. KG
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1253
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS maceutical companies are doing already,” says Austin. Instead, the initiative will “validate new targets, and even perhaps supply some small molecule compounds, which large or small companies may want to pick up and develop.” Noting “screening is the easy part,” Austin says the NIH will not do advanced drug development work, except perhaps for certain orphan diseases. Taking advantage of government and commercial sources, the NIH will build an initial library of 500,000 compounds for screening, all the results of which will enter a public database. The NIH is especially interested in phenotypic screening, in which small molecules are deployed without knowing their biological targets. Drug companies rarely do such screening, because conventional wisdom holds that the targets hit by the small molecules are too hard and time consuming to identify. But large-scale phenotypic screening is a new and largely untried idea. “Unlike the pharmaceutical industry, which takes [a target, such as] COX2 and only COX2, and then hits it over the head with 500,000 different compounds, we’re targeting a biological process,” says Craig Crews, professor of pharmacology and chemistry at Yale University (New Haven, CT, USA). Although biological effects of the screens will be unpredictable, they will result in more knowledge about biological mechanisms and structure than a simple ‘yes or no’ answer to the question, ‘did the drug inhibit a single target?’ Such screening “may very well generate a lot of very interesting biological activities, but then the bottleneck will be target identification and followup,” says Steve Adams, CSO for NeoGenesis Pharmaceuticals (Cambridge, MA, USA). Adams recommends a broader approach, including affinity-based and functional-based screening against known targets. Crews and Austin are confident that current methods, and new methods in development, will identify individual drug targets. Although chemical genomics is mainly a research tool, the initiative does put the NIH closer to the drug business, and intellectual property issues remain unresolved. With NIH taking drugs further down the pipeline, public ownership issues will inevitably arise when it comes time to license. Bitter disputes, like those over the anti-HIV treatment AZT and the cancer drug Taxol, where public outrage over exclusive licenses and pricing eventually led to protracted lawsuits involving state and federal governments, could become commonplace. “Problems in the past have arisen when the government has wanted to maintain some level of control over the downstream commercial aspects,” says Adams. “The
1254
pharma industry needs, indeed insists, on being able to have control over the development and commercialization process.” Austin says any policy decision concerning the licensing of resulting drugs and targets will follow consultation with industry. The field of chemical genomics only began in 1999, when chemist Stuart Schreiber’s group at Harvard University (Cambridge, MA, USA) used a phenotypic screen to identify a novel anticancer drug target, the mitotic kinesin Eg5. Schreiber thus demonstrated that chemical genomics screening could work in academia. “You can see the proliferation of academic small molecule screening facilities around the nation because of that,” says
Crews. Cytokinetics (S. San Francisco, CA, USA) has now taken an Eg5 inhibitor into a phase 1 clinical trial. The new initiative “is a very good idea, because there are certain areas of cell biology that just aren’t amenable to, say, traditional genetic analysis,” says Crews. “But it will require a large-scale effort, and that’s why I think this is quite an appropriate undertaking for NHGRI.” Austin acknowledges that a vast new infrastructure will be needed, but doesn’t yet know the price tag. “What we’re talking about is not cheap,” he says.“But there is commitment at the highest level of the NIH to do this. So it will happen.” Ken Garber, Ann Arbor, MI, USA
Public–private partnerships boost research on neglected diseases The Bill and Melinda Gates Foundation (Seattle, WA, USA) announced on September 21 the availability of $168 million in grants to fight malaria, including $100 million earmarked to develop vaccines through the public-private partnership Malaria Vaccine Initiative (Seattle, WA, USA). The charitable sector, by financing early-stage development, has lowered the risk in creating vaccines for diseases prevalent in developing countries. As a result, targeting neglected diseases has now become attractive to biotech companies that are willing to explore new markets.
Creating vaccines for developing countries has not traditionally attracted small biotech companies that typically require high-price markets in order to recoup R&D costs. Recent donations from charitable organizations, such as the Gates Foundation (endowed with $25 billion), have changed the landscape of drug development, and not only for malaria. “Charities have created a vaccine market specialized in diseases from developing countries such as tuberculosis and malaria. These diseases have, up until now, been overlooked by the industry as a potential
Table 1 Select biotech companies involved in public-private partnerships Company
Public-private partner
Disease
Apovia (Martinsried, Germany)
Medicine for Malaria Venture (MMV; Geneva)
Malaria
Bayer (Leverkusen, Germany)
MMV
Malaria
Celera (Rockville, MD, USA)
Institute for OneWorld Health (San Francisco)
Chagas disease
Chiron (Emeryville, CA, USA)
Global Alliance for Tuberculosis drug Development (New York)
Tuberculosis
Corixa (Seattle, WA, USA)
Infectious Disease Research Institute (Seattle, WA, USA)
Leishmaniasis
GlaxoSmithKline (London)
MMV
Malaria
Sequella (Rockville, MD, USA)
Aeres Foundation (Rockville, MD, USA)
Tuberculosis
Targeted Genetics (Seattle, WA, USA)
International AIDS Vaccine Initiative (IAVI; New York)
Human Immunodeficiency Virus (HIV)
Therion Biologics (Cambridge, MA)
IAVI
HIV
Berna Biotech (Berne, Switzerland)
IAVI
HIV
Sources: The Initiative on Public-Private Partnerships for Health (Geneva, Switzerland) and company web sites.
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
market,” says Jaap Goudsmit, CSO at vaccine company Crucell (Leiden, The Netherlands). Third-party funding has helped bridge the development gap between fundamental research in neglected diseases and its industrial application through public-private partnerships that associate nonprofit organizations, governments and industry (see Table 1). “Gates has taken a lot of the risk from product development by putting money in interim development,” says Carol Nacy, CEO of Sequella (Rockville, MD, USA), a company specializing in tuberculosis. Sequella works in parallel with the Aeres Foundation (Rockville, MD, USA), a private group that identifies potential tuberculosis drugs from basic research funded by the National Institutes of Health. It then performs proof of principle experiments in humans, going as far as phase 2 efficacy studies, in order to make the product appealing to potential biotech and pharma partners. As a result, biotech and pharmaceutical companies are more likely to take over development of such vaccines at these later stages, whereas they would not have during the early, riskiest stage of development. “We’re prepared to take greater risks because it is not our shareholder’s money,” says Jeffrey Almond, senior vice president of discovery research and external research and development at Aventis Pasteur (Lyons, France), which is developing a dengue vaccine with Acambis (Cambridge, UK) by adapting the biotech firm’s existing yellow fever technology. But a few partnerships have discovered some intellectual property (IP) roadblocks when attempting to fill the development gap for vaccines that target the developing world. Drug development for most indications requires some IP consolidation, and companies are generally happy to foot the bill in courtrooms when there is a large market at stake.“Because there is little market value for IP on malaria antigens, patenting them can actually hinder rather than promote innovation,” says Melinda Moree, director of MVI. Successes in developing a vaccine technology for neglected diseases can have further benefits, in addition to the positive publicity that such an achievement would generate.“Third-party funding helps us validate our technology,” says Vijay Samant, CEO of Vical (San Diego, CA, USA), which has used its plasmid DNA vaccine technology to do research for a malaria vaccine backed by US Navy funding. Once validated, the technology might be applied to vaccines for lucrative markets.
For companies targeting niche markets, there are a number of opportunities in the developing world.“The SARS epidemic has been a catalyst in generating interest for the prevention of other diseases such as influenza, or flu,” says Gurinder Shahi, CEO of life science consultancy Bioentreprise Asia (Singapore). Indeed,“a flu vaccine market is emerging because of SARS,” agrees Goudsmit. He believes that only by vaccinating people against flu will it be possible to distinguish people affected by SARS, because both conditions have similar symptoms at an early stage of the disease. Although governments may foot the bill for a flu vaccine as a preventative measure, middle-income private markets also constitute an untapped opportunity in Asia, says Shahi. But companies can also make profits by adapting vaccines to the needs of developed
countries. Indeed, the emergence in developed countries of diseases from developing countries has generated dual markets that, until now, existed only for travelers’ vaccines, such as diarrhea and yellow fever vaccines. For example, companies such as Crucell, Bavarian Nordic (Copenhagen) and Acambis are currently focusing on vaccines against the West Nile virus, and SARS has attracted a number of vaccine players (Nat. Biotechnol. 21, 720, 2003) as such diseases become an economic burden to countries affected. In addition, the US BioShield proposal now pending before Congress would, if enacted, pony up $6 billion in public funds to address infectious diseases, such as Ebola, that are now considered security issues (Nat. Biotechnol. 21, 216, 2003). Sabine Louët, London
Concerns raised over declining antiinfectives R&D In mid-September, officials at the US Food and Drug Administration (FDA; Rockville, MD, USA) approved daptomycin (Cubicin, previously Cidecin) for clinical treatment of skin infections—a seeming renaissance because it is the first time in decades that a product belonging to a new class of antibiotics gained approval. Yet, FDA officials point to a distinct trend toward fewer antiinfective agents being approved per year over the past 20-year period. Thus, even while Cubist Pharmaceuticals (Lexington, MA, USA) is celebrating success with Cubicin, a cyclic lipopeptide, many other would-be developers of antiinfective products at biotech and pharma companies are sounding gloomy about near-term prospects within this market sector. Moreover, despite resurgent federal interest and support for developing such products for biodefense purposes (Nat. Biotechnol. 21, 469, 2003), continuing uncertainties about the BioShield proposal and more recent concerns about the regulation of federal Small Business Innovation Research grants are adding to the gloomy mood (Bioentrepeneur; 4 September 2003, doi:10.1038/bioent765). With venture capital sources scarce, interest in this product sector among big pharma, particularly in antibacterial agents, is “evaporating and the sense of bleakness is pervasive,” says Deborah Nosca of Nereus Pharmaceuticals (San Diego, CA, USA), which is seeking to develop antiinfective products.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
Cubist Pharmaceuticals
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS
Gram-positive skin infections can now be treated with Cubicin (daptomycin), which is the first of a new class of antibiotics to be approved by the FDA in over two decades.
“A large number of pharmaceutical companies—and many biotech companies—have exited the field or reduced their efforts, especially for antibacterials,” says Steven Projan of Wyeth Research (Cambridge, MA, USA), one of several industry representatives who addressed these issues during the Interscience Conference on Antimicrobial Agents and Chemotherapy, convened during midSeptember in Chicago.“And when big pharma sneeze, biotech companies get pneumonia and
1255
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS drop dead.” Many larger companies have indicated that drug development attention is being shifted to those products that will be used for treating chronic diseases, rather than acute infections for which treatment courses typically are brief. This sort of behavior among major pharmaceutical corporations not so long ago was often what spurred upstart biotech companies to seize on smaller market sectors that offered them adequately sized product-development opportunities. But this strategy has become increasingly difficult for biotech companies to exploit, according to David Shlaes of Idenix Pharmaceuticals (Cambridge, MA, USA), another biotech seeking to develop products to treat infectious diseases. “It is a myth that biotech companies will take over this field,” he says. One big problem revolves around capital, particularly the sums that are needed when any company plans a large-scale clinical trial, Shlaes explains. “Cubist was an exceptional case. Most small companies can’t take on the cost of trials, and the departure of big pharma removes a major source of funding for small companies.” Antbiotics represent a “complex market,” counters Ralph Christoffersen at Morgenthaler Ventures (Boulder, CO, USA). He acknowledges that some big pharmaceutical companies are leaving the field because they think it is too small, but he says that it nonetheless remains “superb for some small biotech companies.” Moreover, he adds, “the notion that big pharma must fund clinical trials is over-generalizing. It’s much more feasible for venture capital to fund clinical trials for [drugs such as antibiotics] to treat acute rather than chronic diseases—and because animal models are more highly predictive, it costs much less money, and there is a higher probability of success.” Other discouraging forces include costly unresolved issues over intellectual property (IP) rights and continuing regulatory difficulties having to do with clinical trials, according to both Shlaes and Projan. Reforms being considered as part of the BioShield legislative proposal pending in Congress could provide new incentives, but remain unsettled. For example, one proposal being floated—to provide a given company with ‘wildcard’ extended market exclusivity for any drug within its portfolio in exchange for development of valuable but less profitable drugs— could prove a useful incentive to attract companies back into the antiinfectives field, Shlaes says. Of course, not all pharma are sniffling over the value of antiinfective agents, and not every small biotech company seeking to
1256
develop treatments for infectious diseases has developed fatal corporate pneumonia. For example, Johnson & Johnson (Raritan, NJ, USA) is continuing to seek such drugs, according to Karen Bush of the company’s Pharmaceutical R&D division. She suggests that medical needs for new antibiotics outweigh any IP issues that may deter a pharmaceutical company from entering the field. “A major reason to stay in the field is drug resistance,” she notes. Moreover, there are still solid commercial justifications for pursuing new antimicrobial treatments because this field is considered the
third largest in terms of annual sales, surpassed only by drugs for treating central nervous system disorders and cardiovascular disease. Antibacterial agents represent about a $26 billion worldwide annual market, according to Projan. Even a small piece of that action can be attractive for small companies, says Zhengyu Yuan of Vicuron Pharmaceuticals (Fremont, CA, USA). A product with annual sales of “more than $150 million is very suitable for biotech companies,” making the search for antimicrobial products still “very attractive” for some companies. Jeffrey L. Fox, Washington
Japan’s biotech sector shows signs of life Cell therapy firm MEDINET (Yokohama, Japan) completed an initial public offering (IPO) on October 8 in Japan, joining four other biotech companies that have gone public in the country since September 2002. Industry observers expect around six more will have IPOs before the end of 2004 (see Table 1). The spate of offerings is seen as a harbinger of a Japanese economic upturn and contrasts sharply with the biotech industry’s lack of activity on public markets elsewhere in the world. There has not been a biotech IPO in Europe or the United States since June 2002. “This is virtually unknown outside Japan, but the nation’s biotech sector has been exploding in recent years,” says Christopher Savoie, CEO of Gene Networks (Tokyo), a pharmaceutical development company that hopes to go public within the next few years. Indeed, share prices of four of the five biotech firms, which all spun out from universities, have increased since their IPO completions. The IPO boom reflects the recent recovery of the Japanese economy. The country’s GDP grew by 1.0% in the second quarter, 0.4% higher growth than was expected by the Cabinet Office, and the benchmark NIKKEI 225 index rose above 10,000 in August for the first time since August 2002. Insiders say this positive economic climate in Japan gives an advantage to biotech firms there. Japan’s biotech sector has also benefitted from changes in economic infrastructure and government deregulation of industry– academic collaboration that is favorable to entrepreneurs (Nat. Biotechnol. 18, 256, 2000). Part of this system fell into place when Japan’s markets for emerging stocks were restructured to make it easier and faster for companies to conduct IPOs, reducing the application approval time from years to one or two months, say financial market experts.
Legislation enacted in 1998 granting investors limited liabilities, in the hope that it would boost investment by venture capitalists, has also helped entrepreneurs, says Takeo Matsumoto, CEO of Biotech-Healthcare Partners (Tokyo). It appears to have worked. In fiscal year 2002, overall investment in biotech firms stood at ¥8.8 billion ($77 million), an increase of 33% over the previous year, according to statistics from the Japan Economic Journal (19 August 2003, p.19). Japanese biotech startups are also drawing increasingly on the expertise of the country’s universities. Faculty members—once confined to their ivory towers—have been allowed to run R&D companies and technology licensing organizations, since 2000. The universities have thus become more active players in the biotech industry, says Robert Kneller, professor of intellectual property at the University of Tokyo. And Japan’s Ministry of Economy, Trade and Industry (METI; Tokyo) aims to increase university-backed venture firms to 1,000 by the fiscal year 2006. As of March 2003, 131 such companies were involved in activities classified by METI as biomedical; of these, six plan to go public by 2009, according to METI. “Japan has traditionally possessed a myriad of biotech business seeds. It was just that there was no system to bridge them with venture opportunities,” says Savoie. Among those seeds, the fusion of biotech and other technologies is particularly hot, says Katsuya Tamai, a professor of intellectual property at the University of Tokyo.“Japan has been strong in nanotechnology, so the combination of nanotech and biotech is the future.” Tamai hopes there will be more leeway for university professors to conduct unique research and run biotech ventures when national universities become
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
NEWS
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Date of IPO
Company
Business
September 25, 2002
AnGes MG (Osaka)
Gene therapy
December 10, 2002
TRANS GENIC (Kumamoto)
Mouse and antibody production methods
July 30, 2003
A&T Corporation (Kanagawa)
Clinical trial reagents
September 18, 2003
MediBic (Tokyo)
Consulting, informatics
October 8, 2003
MEDINET (Yokohama)
Cell therapy
December 2003
Soiken (Osaka)
Human clinical research
December 2003
OncoTherapy Science (Tokyo)
Gene therapy research
Spring 2004
Shin Nippon Biomedical Laboratories (Tokyo)
Clinical pharmacology and safety research
Spring 2004
SOSEI (Tokyo)
Drug development
Spring 2004
HuBit Genomix (Tokyo)
SNPs
The second half of 2004
DNA Chip Research (Yokohama)
DNA chip development
Sources: Tokyo Stock Exchange Mothers, JASDAQ, Japan Economic Journal (12 August 2003, p.1).
independent organizations in April 2004. Although Japan’s penchant for biotech ventures is increasing, there are obstacles that must be overcome, says Kneller. One of the biggest challenges in Japan is a reluctance of large pharmaceutical companies to partner with small biotechnology companies that are focused on biomedical research, which consists of almost half of all Japanese biotech firms, according to statistics by the Japan Bioindustry Association (Tokyo).“Drug discovery in Japanese large pharmaceutical companies occurs predominantly in-house. In contrast, European and US pharmaceutical companies rely more on alliances with university based startups and other biotechnology companies for drug
discovery,” says Kneller. He further suggests that Japan’s inflexible labor market, which is partly sustained by a pension system and retirement money program that favors lifetime employment at a single job, discourages changing career paths into entrepreneurship. In the meantime, Japan’s biotech sector still has some catching up to do.“There are already 400 biotech [public] ventures in the US, whereas Japan has only a handful,” points out Steven Burrill, CEO of Burrill & Co. (San Francisco), a life sciences merchant bank. “We will see thirteen companies that go public in the fourth quarter this year in the US,” he predicts. Keiko Kandachi, Tokyo
GM confusion in Brazil The Brazilian government temporarily lifted the country’s ban on planting genetically modified (GM) soy on September 26, but long-awaited legislation that would allow such crops to be made commercially available in Brazil has yet to be voted on by Congress. The confused legal situation has set back both agbiotech firms and local academic scientists developing new GM organisms and could therefore hamper the overall competitiveness of the country’s agbiotech sector. Although the new measure still forbids the sale of GM seeds, the government’s decision brings Monsanto (St. Louis, MO, USA) a step nearer to selling its GM soy in Brazil. “This is another time that Roundup Ready soy’s safety has been attested [to] by the federal government,” says Lúcio Mocsányi, director of communications for Monsanto, referring to a
decision made in 1998 by the Comissão Técnica Nacional de Biossegurança (National Technical Biosafety Committee, CTNBio; Brasília, Brazil) that the GM soy could be safely commercialized (see Box 1). As a result of CTNBio’s decision, Greenpeace filed a lawsuit against Monsanto that resulted in a court ruling that banned GM soy until the government’s recent removal of the ban (Nat. Biotechnol. 17, 848, 1999). Agbiotech industry insiders hope this announcement will eventually lead to the end of restrictions on GM crops. On March 26, another presidential provisional measure (later converted to Law 10688/03) established rules for the commercialization of the 2002/2003 soy crop, including transgenic soy. This measure gives amnesty to soy farmers who illegally planted GM soy when the ban
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
©Ernesto Paterniani
Table 1 Completed and proposed initial public offerings of Japanese biotech firms since September 2002.
Roundup ready soy in a farm in Rio Grande do Sul.
was in place; there are no official statistics, but farmers believe around 70% to 80% of soy in Brazil’s southernmost state of Rio Grande do Sul is now transgenic (see Fig. 1). Even so, the most recent provisional measure implemented by the government is only a temporary solution; it has yet to be confirmed by Congress and converted into new legislation on GM crops. This draft law, which was initially expected to be ready before October, the soy-planting season, has been delayed, partly due to disagreement within the government. The minister of agriculture, Roberto Rodrigues, defends the planting of GM crops, whereas the minister of the environment, Marina Silva, is against the commercial planting, saying that more studies are necessary. “The worst thing that could happen is not to have a clear regulation pattern,” says Fernando de Castro Reinach, CEO of Alellyx, a genomics company created by researchers involved in the sequencing of plant pathogens, such as Xylella fastidiosa. “We are losing the concept that we must judge GMOs in a case-by-case manner.” Although the lack of clear legislation affects multinational companies such as Monsanto, Brazil is only one country (albeit an important one) within their market. However, for smaller companies like Alellyx, which work on crops of major economic significance to Brazil, such as sugarcane, eucalyptus and orange, the scenario may be more dire. “Right now we’re not suffering any damage, because we’re still in a developing phase regarding transgenics, but in one or two years we would need field testing, and then we may be put in jeopardy,” says Reinach. He also warns that a GM-unfriendly scenario “with neither clear rules nor clear prohibitions” will cause companies to close and scientists to leave the country. Not only are biotech companies affected, but Brazil’s research community is also feel-
1257
NEWS
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
• January 5, 1995. Brazil passes Biosafety Law that legalizes the screening of GM organisms by CTNBio, which is established to oversee all issues related to GM organisms on a case-by-case basis. • 1998. CTNBio approves the planting of Monsanto’s Roundup Ready soy. • 1998. The Brazilian office of Greenpeace and IDEC, a São Paulo–based consumers defense institute, file a lawsuit against Monsanto and CTNBio to forbid the planting of Monsanto’s GM soy. Brazil’s own Ministry of the Environment later joined the legal action. A judge sides with Greenpeace, implementing a ban against the planting and commercialization of GM crops in Brazil. • March 26, 2003. Provisional Measure number 113/03 (later converted to Law 10688/03 on June 16) provides amnesty to soy farmers who have illegally planted GM soy during the ban. • September 26, 2003. Brazilian vice president, José Alencar, acting on behalf of President Luiz Inácio Lula da Silva, who was abroad, enacts a Provisional Measure (Medida Provisória 131/03) that lifts the ban on the planting of GM soy. RBN
ing the impact. “We hear people saying they favor research with transgenics, but not their commercial planting. So what are we doing research for?” asks Francisco José Lima Aragão, project leader at Embrapa Genetic Resources and Biotechnology (Brasília), one of the key recipients of agriculture ministry funds for applied research. “Brazil could have been the second developing country to plant its own GM crops, after China. Now we are three years late in our research.” Leila Macedo Oda, who is president of the National Biosafety Association (ANBio; Rio de Janeiro), a nongovernment organization, draws a parallel between the objection to GM organisms and ‘market reserve’ in computing back in the 1980s. By forbidding the import of computers, the Brazilian government hoped to favor a local computer industry, but the lack of competition only made Brazil lag behind other countries in the use of computers. Many people then smuggled computers the way soy farmers now smuggle GM seeds. Ricardo Bonalume Neto, São Paulo, Brazil
1258
RR
Folha de S. Paulo and Veja
Box 1 Time line of a ban AP
AM
MA
PA
CE PI
AC
AL SE
TO
RO
DA
MT Brazilian soy producing areas in 1970 Brazilian soy producing areas in 2003
GO MG
ES
MS
Total planted area: 18.5 million hectares Estimated transgenic soy: 350,000 hectares in the state of Mato Grasso (MT) 354,000 hectares in the state of Mato Grasso do Sul (MS) 450,000 hectares in the state of Paraná (PR) 2,250,000 hectares in the state of Rio Grande do Sul (RS)
RN PB PE
SP
RJ
PR SC RS
Figure 1 Repartition of soy in Brazil. Source: Brazilian Institute of Geography and Statistics, Embrapa Soy and Brazilian Ministry of Agriculture.
Benefits of biotech clusters questioned The official opening on October 29 of Biopolis, Singapore’s new biotechnology R&D hub, further ratchets up the competition among governments that view biotechnology clusters as a key component of national economic development. But a recent UK parliamentary report on biotechnology questions the extent to which clustering concepts derived from leading locations can be applied elsewhere. Biopolis will be a ready-made cluster: a campus to house publicly funded R&D labs, biotech and pharmaceutical companies, venture capital firms, law firms and other components deemed necessary to facilitate the commercialization of innovative biotechnology inventions. And Singapore is not alone in this effort—dozens of national administrations, regional authorities and economic development agencies are also striving to emulate the success of locations such as Boston, San Francisco and Cambridge (UK) in developing mature, self-sustaining biotechnology clusters. But can they all succeed? A report published on September 3 by the UK House
of Commons Committee on Trade and Industry suggests not. The emergence of Cambridge, Boston and San Francisco as successful biotechnology locations, it argues, is the result of a coincidence of factors—the presence of centers of academic excellence and investors rather than any deliberate design or public policy interventions. Indeed, Jeroen Bart Carrin, assistant director of the Japan External Trade Organization (Geneva) who studied biotech clusters in Switzerland, says authorities in the Zurich area, for example, are reluctant to actively support the idea of clustering in biotech, simply because they do not need to. “Startup companies come to, or stay in, Zurich by themselves, conveniently located within the gravitational pull of The Federal Institute of Technology Zurich, the University of Zurich and other centers of excellence, as well as other biotech companies,” says Carrin. The cluster concept was originally developed in 1990 by Harvard Business School (Cambridge, MA, USA) professor Michael Porter in a book called The Competitive Advantage of Nations, which
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
describes clusters as the geographic concentrations of interconnected companies, specialized suppliers, service providers, firms in related industries and associated institutions in particular fields that compete but also cooperate. This theory “provides a good description of the factors involved in the success of biotechnology in certain regions,” the report states.“It does not, however, necessarily provide a blueprint for establishing biotechnology elsewhere.” But many business developers argue that budding locations can learn from the experiences of existing centers. Gurinder Shahi, CEO of the life sciences consultancy BioEnterprise Asia (Singapore), says companies that emerge later than the pioneers can mature more quickly if they manage to learn from business models and technology pathways that worked in other regions. Kjell Carlsson, research associate at the Institute for Strategy and Competitiveness at Harvard Business School (Cambridge, MA, USA), says that although emerging regions can learn from both successful and unsuccessful clusters, he warns that each cluster has an individual dynamic that cannot be mapped precisely onto another.“There aren’t many—if any—world-leading clusters that specialize in exactly the same thing or [that] have the same makeup.” The parliamentary committee also argues that UK policy should focus primarily on reinforcing the success of the country’s most competitive clusters, instead of supporting the emergence of biotechnology in multiple locations.“Not only may considerable sums of public money be wasted in trying to force into existence local biotechnology companies, but also rivalry between regions may adversely affect those with existing strengths in the sector, thus undermining the success of biotechnology in the UK as a whole.” Glenn Crocker, former head of Ernst & Young’s (Cambridge, UK) life sciences practice, and CEO of the recently established incubator BioCity Nottingham (Nottingham, UK), disagrees with the committee’s views. “I can’t believe they actually meant that biotech clusters shouldn’t develop wherever there is a demand,” he says.“It would be incredibly small-minded to suggest that the few little pockets of activity we have at the moment are all we can sustain in the UK. Silicon Valley stretches for a hundred miles; Cambridge covers about ten at most.” One of the main theoretical benefits of clustering is that the critical mass it engenders acts as a magnet for investors, entrepreneurs and scientific talent. However, according to
JTC
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS
Singapore’s Biopolis includes five buildings for national institutes and two for biotechs, all connected by walkways and an atrium.
Jason Rushton, an associate with life sciences venture capital fund Merlin Biosciences (Cambridge, UK), location is not a major factor in driving individual investment decisions, notwithstanding the obvious networking advantages that clusters offer. Although being based in a high-profile area
can confer advantages to a company and make it easier to attract talent, he says, human factors and serendipity often weigh heavily on a company’s original choice of location. “Many companies are where they are because that’s where they are,” he says. Cormac Sheridan, Dublin
Biotech parks proliferate, despite concerns over sustainability The rush to develop biopharmaceutical research and industry parks is heating up around the world, with China’s island province of Hainan announcing its intention on September 14 to develop yet another biomedicine valley—joining ten other existing Chinese biopharmaceutical parks. But industry analysts warn that many of these ambitious development plans may be thwarted because some entrepreneurs are attracted to biotech parks to take advantage of tax incentives or real estate bargains. Critics also say most companies that occupy these parks rely on the commercialization of traditional knowledge, whereas innovative R&D is required to develop and sustain a biotechnology industry. Many regions hope to boost their economic development by creating R&D parks (see Table 1) that provide infrastructure to host public and private research that is focused on
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
biotechnology in hopes of attracting ancillary players such as intellectual property specialists or investors to create a cluster effect (see p.1258). But that is not always a realistic option. “Except [for] research hubs like Beijing or Shanghai, it is quite difficult for other provinces and cities to develop [a biotech drug] industry due to their lack of talent and technological [know-how],” says Zhou Yongchun, a senior research fellow with China Science and Technology Promotion Center (Beijing). A key issue that threatens the sustainability of these parks is that companies do not produce the R&D-intensive products that characterize biotechnology in the United States or Europe where the industry is more mature. Instead, some traditional Chinese medicine makers in China try to expand their market shares or push up their stock prices by falsely
1259
declaring that they have developed some biotech drugs or developed new biotechnological processes to produce their drugs, says Mei Xiaodan, founder of recombinant human protein and biochip company Bio Integration Tech (Dalian, China). According to China Datong Securities (Dalian, China), of 200 Chinese biopharmaceutical companies surveyed in 2002, only 14 specialized in producing biotech drugs, and no more than four of them had annual sales above Yuan 100 ($12) million. But Yongchun says the Chinese biopharmaceutical industry has great potential given the large sums of public and private funding available to develop the industry. The government is putting forth Yuan 0.6 billion ($72 million) into biotechnology research for the 2001–2005 period, and Chinese companies have invested a total of Yuan 13 ($1.7) billion up to the end of 2002. India, which has a strong tradition of producing generic drugs (Nat. Biotechnol. 21, 1115—1116 (2003)), faces similar problems due to lack of innovative R&D outside established biotech centers, such as Bangalore (Karnataka) and Hyderabad (Andhra Pradesh), according to Narayanan Suresh, editor of BioSpectrum, a national business magazine on biotechnology. And in Latin America, several countries, such as Argentina, Chile, Brazil, Mexico, Cuba and Uruguay, have vowed to develop biotechnology sectors. Other than Cuba, most of these countries have biotechnology industries that rely mainly on technologies that have already been established elsewhere, such as diagnostic kits for salmon in Chile and cattle
Hepeng Jia
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS
Entrance of the Beijing Bioengineer-ing and Pharmaceutical Industrial Base.
in Argentina, according to José Luis Ramírez, director of the United Nations University Program for Biotechnology in Latin America and the Caribbean (Caracas, Venezuela). Another problem arises from the lack of controls when allocating government incentives, such as tax rebates and inexpensive land prices, designed to boost the creation of biotech parks. In China, for example, once a park has been ratified by a central government, local park developers—often affiliated with a local government—can obtain massive bank loans with low interest. As a result, some of these parks might crumble when real estate developers have made their profit. Indeed, in India where similar issues exist, Devinder Sharma, president of the nongovernmental organization Forum for Biotechnology and Food Policy (New Delhi),
warns that biotech parks could have the same fate as that of the government-backed industrial parks created a few years ago. “The entrepreneurs came, bought the land at throw-away prices, and within a couple of years, most of them vanished after selling off the plots,” says Sharma. Not only are biotech parks becoming real estate businesses, but they also drain government funding into inadequate projects. A Mexican industry observer who wishes to remain anonymous denounces the allocation of government funding to “pseudo-biotechnology” projects led by people who are taking advantage of the current political and economic momentum that is favorable to biotechnology. Ana María Sandino, virology researcher at the Santiago de Chile University (Santiago) and founder of Diagnotec (Santiago), says Chilean companies only get government support if they are associated with academics—a policy that makes scientists turn to biotech just to get funding. But some industry observers believe that biotech parks will play an important part in the future of developing nations. Kondapuram Raghavan, director of the Indian Institute of Chemical Technology (Hyderabad), which is setting up an incubation center at the biotechnology park to be created in Hyderabad, says, “The investor community and the government realize that biotech is a long term play arena and cannot yield quick returns. Therefore, they are unlikely to pull back.” Hepeng Jia, Beijing, with additional reporting from K.S. Jayaraman, Hyderabad, India, and Claudia Orellana, Brecon, UK.
Table 1 Biotechnology parks created recently in China, India and Latin America. Park
Date announced
Specialization
Money invested
Hainan Biomedicine Valley (Hainan province, China)
September 2003
Biomedicine
Yuan 7.2 billion ($870 million) promised by government
A TCM Town (Bozhou, Anhui province, China)
September 2003
Drugs derived from traditional Chinese medicines (TCM)
Yuan 413 million ($50 million) in deals signed
Beijing Bioengineering and Pharmaceutical Industrial Base (Beijing)
June 2003 (launch date)
Bioengineering and Pharmaceutical
Yuan 2 billion ($241.5 million)
Biotech Park (Vishakhapatnam, Andhra Pradesh, India)
August 2003
Marine biotechnology
*
International Biotech Park (Hinjewadi, Maharashtra, India)
January 2003
Biopharmaceuticals
$140 million
Jalna Park (Jalna, Maharashtra, India)
January 2003
Agbiotechnology
$140 million
Whockhardt’s biotech park (Aurangabad, Maharashtra, India)
August 2003
Biopharmaceuticals
$33 million
Bangalore Park (Bangalore, Karnataka, India)
March 2002
Industrial biotech; biopharmaceuticals
$120 million
Biotec Plaza (Montevideo, Uruguay)
October 2002
Recombinant vaccines
$3 million
* Data not available.
1260
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
NEWS IN BRIEF
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Personalized drug orphans Countries may currently lack adequate legislation to regulate drugs that are personalized to fit the genetic profiles of patients, and corresponding genetic tests to identify those patients, according to the Nuffield Council on Bioethics (London) in a report ‘Pharmacogenetics: Ethical Issues,’ released on September 23. Because grouping patients according to their genetic features could result in the fragmentation of the market for medicines, the council suggests orphan drug laws could be extended to personalized medicines in order to encourage companies to tackle such markets. (Orphan drug laws, which exist in Europe, the United States, Australia and Japan, provide tax incentives and several years of market exclusivity for sponsors to develop products for rare diseases.) Nuffield recommends that European and US drug regulatory agencies provide guidance on how best to incorporate pharmacogenetics tests into the new drug license conditions. Sara Radcliffe, director of Science Policy and Bioethics at the Biotechnology Industry Organization (Washington, DC, USA), says that regulators will have to provide more guidance in “close coordination with the diagnostics developers and with experts in pharmaceutical sciences to make sure the guidance keeps pace with rapidly advancing knowledge in this area.” ED
Japan calls in scientists Japan’s government will introduce the ‘Special Advisor System’ in April 2004 to enable 100 experts in science and technology to lend their advice to judges dealing with mounting intellectual property cases in civil courts. Japan’s legal system is notoriously lacking in judges specializing in technology-related issues. Out of 2,000 judges nationwide, only about ten hold bachelor of science degrees, according to the Japan Economic Journal (7 June 2003). Hiroshi Konno, a professor of financial engineering at Chuo University in Tokyo, says,“It is virtually impossible to make [judges] understand the highly technical material if they don’t have any scientific background.” Under the new system, university professors and other researchers will take on advisory roles on a part-time basis to help judges better understand the technical issues involved. But Konno
News in Brief written by Paroma Basu, Aaron Bouchie, Laura DeFrancesco, Emma Dorey, Jeffrey L. Fox, Keiko Kandachi, Sabine Louët and Pete Mitchell.
GM Nation unites GM opponents A UK government-backed public consultation has found that the majority of the UK public are opposed to genetically modified (GM) crops and distrust both the agbiotech industry and the government’s ability to regulate such products. The consultation, launched in June as ‘GM Nation,’ received feedback from over 35,000 people, at least 80% of whom are clearly anti-GM (see Nat. Friends of the Earth Biotechnol. 21, 957, 2003). A parallel series of over 600 public meetings and ten focus-group sessions found that the more information people were given about GM technology, the more they were opposed to it. “You can argue about the details, but when you add this result to earlier data like Eurobarometer [a 2002 survey that showed only about half the British public to be against GM food] it does show there is a consistent level of concern about food biotechnology,” said Professor Nick Pidgeon, an expert in public opinion surveys and director of the Center for Environmental Risk at the University of East Anglia (Norwich, UK). A spokesperson for the pro-industry Agricultural Biotechnology Council (London) dismisses the consultation as unrepresentative, saying the responses had been “orchestrated by campaigning groups.” Leading environmentalist groups, Greenpeace (London) and Friends of the Earth (London), deny any organized campaign. PM
says the annual budget for the advisory fee per expert—¥200,000 ($1,800) including transportation and accommodations—is too small for such responsibility, and Japan ultimately needs to produce more judges with scientific expertise. However, few of the 72 law schools that are set to open next year under the new legal education system in Japan offer programs in science or technology. KK
NAS advises biodefense board To guard against bioterrorism-related misuses of legitimate biotechnology research in the public and private sectors, members of a panel convened by the National Academy of Sciences (NAS; Washington, DC, USA) recommend the US Department of Health and Human Services (Washington, DC, USA) create an independent National Science Advisory Board for Biodefense (NSABB) to provide advice not only on publishing matters but also more generally on the relative risks of new technologies. The NSABB, which is to include top scientists and national-security experts, would oversee a series of local committees that would operate as a “tiered system of review” to identify experiments that raise concern because of their “high potential for misuse,” according to NAS panel members. This proposal for a review system emphasizes “self-governance” by the scientific community and would extend to decisions over whether to publish findings that are considered “sensitive but not classified,” according to the panel. Several federal agencies are deeply involved in discussing these issues, making it
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
questionable “whether we need someone else to enter the fray,” says Michael Warner of the Biotechnology Industry Organization (BIO; Washington, DC). JLF
EBI supports SMEs The European Bioinformatics Institute (EBI; Heidelberg, Germany), a nonprofit academic organization, launched on October 13 a support bioinformatics forum open to members of small biotechnology firms. Industry insiders say the initiative could eventually increase operating efficiencies within the industry. The new forum aims to create a community for bioinformaticians working at small to medium-sized enterprises (SME), defined as those firms that employ fewer than 250 staff members and earn less than €50 ($58.2) million yearly, which represent 83% of all European biotechnology firms. The EBI will give forum participants access to its industry support services—such as technical expertise, hands-on training workshops and participation in annual meetings—an advantage previously extended only to large pharmaceutical firms. The European Commission’s Framework 6 research fund set aside for SMEs more than 15%—€1.8 ($2.1) billion—of its budget until 2006. Clive Brown, head of bioinformatics at gene sequencing firm Solexa (Cambridge, UK), says, “this [forum] should lead, via neutral mechanisms like the SME program, to an industry-wide benefit in terms of reduced costs, better training, shared standards and recruitment and retention of key staff.” PB
1261
NEWS IN BRIEF
Table 1 US biotechnology fundraising ($ million)
120
115
3Q03 IPO
110
$0
$0
$0
$130
$183
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Debt/other Venture capital
100
95 April
May
3Q02
Secondary public $1,374 PIPEsa
105
2Q03
$676
$522
$136
$2,779
$2,499
$214
$670
$628
$560
Partneringb
$2,556
$2,256
$1,242
Total
$8,055
$6,035
$2,335
June aPIPE,
Nasdaq
Amex
Amex biotech
Nasdaq biotech
private investment in public equity. bPartnering figures based on total deal value disclosed for transactions worldwide.
Source: Burrill & Co.
The base value is 100 as of July 1, 2003. Source: Yahoo.
3Q financial roundup The US biotechnology industry raised over $8 billion in the third quarter of 2003 (3Q03), the most since the fourth quarter of 2000 when the industry raised $11 billion. The quarter’s most significant trend is the industry’s preparations for an opening of the initial public offering (IPO) window (Nat. Biotechnol. 21, 1116–1117, 2003). As Nature Biotechnology went to press, 13 firms had filed for an IPO on US stock markets, which appear to be gearing up to receive them—the amount raised from secondary public offerings ($1,374 million) increased over tenfold from the previous quarter (see Table 1). In parallel, Japan expects six IPOs before the end of 2004 (See p. 1256). And investment from venture capitalists in biotech ($670 million) was higher than it has been since the fourth quarter of 2001. Although there was good regulatory news in the form of key product approvals in the US, such as Cubist Pharmaceuticals’ (Lexington, MA, USA) antibiotic Cubicin (see p. 1255), there is increasing concern among investors over a proposed rule that affects the reimbursement of injectable biotech drugs when administered in the physician’s office (Nat. Biotechnol. 21, 1119–1120, 2003). Outside the US, Indian
firms received marketing clearance for two generic biotechnology products, and a phase 3 clinical trial clearance for another, in India in 3Q03, signaling an emergence of a biogenerics industry there (Nat. Biotechnol. 21, 1115–1116, 2003). AB
Hurdles for Canadian CEOs The recruitment firm Egon Zehnder International (Zurich) released on September 16 the results of a survey of 35 CEOs in Canada’s biotechnology healthcare sector. Through telephone interviews with executives, the study unveiled key factors that are blocking the Canadian biotech sector from reaching its potential. For example, 30% of those interviewed stated that raising capital is an immediate concern because of the lack of biotech-savvy venture capitalists in Canada. In addition, 70% reported difficulties navigating the regulatory environment in the United States, Canada and Europe. Other problems highlighted in the survey include the recruitment and retention of skilled employees who have a combination of scientific and business skills, and the adequate compensation of executives. Graeme McRae, CEO of drug discovery and
veterinary company BioNiche (Belleville, ON, Canada) points out that his company has been able to recruit workers by modeling its compensation packages on those of the pharmaceutical industry. BioNiche has also cleared regulatory hurdles—eight products were approved in the last two years—by receiving guidance from consulting groups. PB
EU tables markets Directive On October 7, the Council of Europe (Brussels) agreed on the terms of a new Investment Services Directive that is designed to protect noninstitutional investors and enable companies to raise money outside the stock exchange. The directive is a step towards the completion of a harmonized financial market in Europe allowing investment firms to operate across the European Union, once firms are accredited in their country. As a result, “[the Directive] will increase competition [among investment firms],” says a European Commission spokesperson. The main complaint of opposing countries (the United Kingdom, Luxembourg, Sweden and Finland) was the introduction of ‘pretrade transparency,’ which requires banks to advertise the price at which investors are ready to buy or sell a security that enables other firms to fulfill that order if they can. This rule is designed to protect noninstitutional investors from unfair price fluctuations but does not affect investors dealing with larger-than-average market volumes. The directive also opens the door for banks in countries like France, Spain, Italy, Portugal, Austria, Greece, Belgium and Germany to trade off-market financial products, which was previously disallowed in those countries. “Any measure that goes in the direction of harmonizing the market for financial ventures is much welcome by the biotech community,” says Hugo Schepens, secretary general at the European bioindustry association Europabio (Brussels). SL
New product approvals Product
Companies
Details
Cubicin
Cubist Pharmaceuticals
The US FDA granted market approval on September 12 on to Cubicin to treat skin-related infections in hospitalized
(daptomycin)
(Lexington, MA, USA)
patients. Cubicin is the first approved product in a new class of antibiotics called cyclic lipopeptides, which cause the fatal depolarization of gram positive bacterial membranes. Analysts predict Cubicin may reach sales of $350 million by 2008 for a global market of 600,000.
Ganite
Genta
On September 18, the US FDA granted marketing approval to Ganite for the treatment of cancer-related hyper-
(gallium
(Berkeley Heights,
calcemia that is resistant to hydration. Hypercalcemia, a life-threatening elevation of calcium levels in the blood,
nitrate)
NJ, USA)
occurs in up to 50% of patients with advanced cancer, most commonly in patients with cancers of the lung, breast, head and neck, kidney, and multiple myeloma. Ganite is also in phase 2 trials to treat non-Hodgkin’s lymphnoma. PB&AB
1262
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
NEWS IN BRIEF
On September 26, The US Court of Appeals for the Federal Circuit (CAFC; Washington, DC, USA), on remand from the US Supreme Court, issued a ruling in Festo v. Shoketsu Kinsoku Kogyo Kabushiki limiting patent holders’ use of the doctrine of equivalents. In a November 2000 decision, the CAFC ruled that the doctrine of equivalents—which extends the literal meaning of a patent claim to include a range of similar inventions—could not be used in an infringement lawsuit if a patentee amended an original claim to gain patent approval. Many biotechnology firms worried that their inventions would be easily copied, for example, by replacing a single amino acid in an antibody product. But in May 2002, the US Supreme Court imposed a “flexible bar rule” whereby the doctrine of equivalents could still be allowed under certain circumstances (Nat. Biotechnol. 20, 639, 2002). The September CAFC ruling clarifies that patentees can only use the doctrine of equivalents if the equivalent would have been “unforeseeable” at the time of the amendment. In their dissension, Judges Newman and Meyer say the decision, in that it does not allow for the use of the doctrine of equivalents in enough situations, “places new and costly burdens on inventors, and reduces the incentive value of patents.” AB
Regeneron, Aventis strike deal Regeneron Pharmaceuticals (Tarrytown, NY, USA) has attracted an $80 million upfront fee from Aventis Pharmaceuticals
Ultimate DNA microarray competition heats up Several companies are trying to make a DNA microarray chip that contains the entire complement of human genes, but Affymetrix (Santa Clara, CA, USA) is the first to bring one to market. With its latest offering, the Human Plus 2.0 Array, launched on October 2, Affymetrix puts 1.3 million probes on a 1.3 cm square wafer—enough to probe all 30,000 human genes plus 17,000 splice variants. According to CEO Steve Fodor, Affymetrix is moving toward more cost-effective and faster high-throughput applications for use in drug discovery and toxicology by miniaturizing their chips so that they fit into microwell plates. Although JP Morgan (New York) analyst Tycho Peterson sees Affymetrix as the undisputed market leader in DNA microarray chips, competitors Agilent Technologies (Palo Alto, CA, USA) and Applied Biosystems (Foster City, CA, USA) plan to launch similar products by the end of 2003. Those not wanting to buy into Affymetrix’s system might turn to Agilent’s, which does not require specialized scanning equipment; Applied Biosystems is pushing the data validation and annotation that it receives from its association with sister company Celera Genomics (Rockville, MD, USA), according to Applied Biosystems spokesperson Lori Murray. LD Affymetrix
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
CFAC rules on Festo
(Strasbourg, France) for its fusion protein vascular endothelial growth factor (VEGF) inhibitor, developed for oncology and ophthalmology. The deal is the highest valuation ever made for a phase 1 drug candidate. “It is a remarkable deal in size given the stage of the product,” comments Douglas Fambrough, principal at venture capital firm Oxford Bioscience Partners (Boston). He says that pharma pipelines are thin enough for companies like Aventis to take a risk to buy a product based on the combination of a validated target—demonstrated by Genentech’s (S. San Francisco,
CA, USA) VEGF competitor Avastin, which is in phase 3—and a validated mechanism of action, based on data from Amgen’s (Thousand Oaks, CA, USA) tumor necrosis factor inhibitor Enbrel, which is already on the market. In addition to the up-front payment, Aventis will pay $45 million in equity and possibly $25 million for the first clinical milestones on top of development costs. Should the drug receive market approval, profits will be shared but Regeneron would have to reimburse 50% of the estimated $700 million development costs. SL
Selected research collaborations Company 1
Company 2
Acambis
Cangene Corporation
(Cambridge, UK)
$ (millions) *
Details A collaboration to develop purified antibodies from human plasma to bolster human immune
(Winnipeg, Manitoba,
systems against the West Nile virus (WNV). The companies will use Acambis’ WNV vaccine
Canada)
to vaccinate Cangene’s plasma donors. Cangene’s manufacturing facility will be used for joint development at shared costs.
Siga (New York, NY, USA)
Innate
*
(Umeå, Sweden)
A partnership to discover antiinfective therapeutics against bacterial biowarfare agents, with a preliminary focus on the plague, Yersinia pestis. Siga will search for bacterial surface proteins that could lead to organism-specific therapeutics. The companies will use Innate’s screening platform for discovering compounds that impede bacterial protein expression.
ParAllele BioScience (S. San Francisco, CA,
Roche
*
(Basel, Switzerland)
An agreement to explore the genetic basis of diabetes for the development of drugs and diagnostics. ParAllele will use its genetic screening technology to discover the genetic
USA)
variations present in Roche’s patient samples. Roche will fund the research at an undisclosed level with plans to expand the test to various patient populations.
Evotec OAI (Hamburg, Germany)
DeveloGen (Goettingen, Germany)
*
A partnership to discover and develop compounds to treat obesity and diabetes. The deal combines Evotec’s screening and medicinal chemistry technologies with more than 200 potential and 30 validated targets in DeveloGen’s pipeline. Program costs and profits will be shared, but development and commercialization rights will be out-licensed.
*Financial details not disclosed
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
PB
1263
E R R AT A
Erratum: 3Q financial roundup Aaron Bouchie Nat. Biotechnol. 21, 1262 (2003) The y-axis of the graph was labeled April, May and June. It should have been labeled July, August and September.
Erratum: Cancer trials get set for biomarkers Aaron Bouchie Nat. Biotechnol. 22, 6–7 (2004) The figure on p.7 shows the receptor being blocked by the constant region of the Herceptin antibody. The receptor should be shown binding to the variable and not the constant region of the antibody.
Erratum: Make or break for costimulatory blockers Ken Garber Nat. Biotechnol. 22, 145–147 (2004) On p.145, CTLA4-Ig is described as having achieved “outstanding phase 3 results.” In fact, the data were from a phase 2 trial.
NATURE BIOTECHNOLOGY VOLUME 22 NUMBER 3 MARCH 2004
341
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
CORRESPONDENCE
Equivocal role of micelles in Eprex adverse events To the editor: The news article “Lessons from Eprex for biogeneric firms” in the September issue (Nat. Biotechnol. 21, 956–957) cites an unpublished study conducted by Huub Schellekens and associates that proposes the hypothesis that the stabilizer polysorbate 80 led to the formation of micelles in the formulation of Eprex and that the micelles are a possible cause of pure red cell aplasia associated with Eprex treatment of patients with chronic renal failure. Unfortunately, the article treats Schellekens’ hypothesis as if it were conclusive and exclusive. Schellekens himself does not claim this (see below); he has presented this hypothesis at scientific meetings and readily acknowledges that there are other factors that may also contribute to pure red cell aplasia. Furthermore, he has stated that his hypothesis needs additional work, including testing in animal models. None of this information was included in the news article. We have consulted with Schellekens and have been supporting his efforts in finding new factors that may underlie the increased incidence of pure red cell aplasia in patients who have been treated with our drug Eprex. As a matter of public record, it is also well known that cases of pure red cell aplasia have been reported with other recombinant erythropoietins that contain stabilizers other than polysorbate 80, which would rule out the possibility that micelles are the sole cause, as the article implies. Since 2001, our company has been conducting a thorough investigation into the causes of erythropoietin-associated pure red cell aplasia and of the factors that have led to more cases in patients treated with Eprex. Our investigation, and the database of knowledge on pure red cell aplasia, clearly point to the causes of the syndrome as multifactorial. Route of administration for patients with chronic renal failure, storage and handling, and changes in the stabilizer used in Eprex have all been identified as potential factors. We are working closely with health authorities and health-care professionals around the world to better understand and
reduce the incidence of pure red cell aplasia. Our efforts to shift the route of administration from subcutaneous to intravenous have borne fruit. In the first seven months of 2003, we have seen a marked decline in reported cases of the condition. Worldwide, the incidence has dropped from 3.16 cases per 10,000 patient years in the first half of 2002 to 0.43 cases per 10,000 patient years in the first half of 2003. In addition, we are working on several hypotheses regarding the formulation of Eprex, including the micelle hypothesis. We welcome all responsible scientific inquiry into this issue. However, we also expect a scientifically accurate rendering of all the facts when news of a study appears in a wellrespected journal such as Nature Biotechnology. Janice M Smiell Johnson & Johnson Pharmaceutical Research & Development, Senior Director/Clinical Leader, Global Development, 920 Route 202, P.O. Box 300, Raritan, NJ 08869, USA. e-mail:
[email protected]
To the editor: As the senior authors referenced in the news article “Lessons from Eprex for biogeneric firms” (Nat. Biotechnol. 21, 956–957), we would like to point out that the news article draws conclusions that are not supported in the paper entitled “Micelle-associated protein in epoetin formulations” that is in press at the journal Pharmaceutical Research. The opening line of the news article states: “Dutch scientists have found that aggregates of small molecules (micelles) in the formulation of erythropoietin alpha (EPO), sold as Eprex in Europe, were responsible for an immunogenic reaction that triggered severe side effects”. In the third paragraph, the article further states: “But the high concentration of the sorbitol [polysorbate 80] led to the formation of micelles, which were found to cluster with EPO in a form that triggers an immunogenic reaction. The resulting antibodies not only abrogated the effects of Eprex, but also neutralized naturally occurring EPO.”
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
The above statements mistakenly imply that the unpublished research paper has provided definitive proof for the cause of erythropoietinassociated pure red cell aplasia in patients with chronic renal failure. In the Pharmaceutical Research paper, we and our coauthors clearly present the development of an important hypothesis that requires further study. For example, we state that “this micellar form of epoetin may be an important risk factor for the development of antibodies in patients,” adding “we recognize that the presence of micelle-associated epoetin as a risk factor for immunogenicity in patients at the moment remains hypothetical” and “Follow-up studies in which the micelle-associated epoetin is tested in animal models may shed light on the correlation between the formulation of epoetin and its immunogenicity.” We make no claim to a direct cause-effect relationship between the micelle-associated epoetin and immunogenicity. We only propose that such an interaction could be possible and deserves serious further investigation. Although the work published in the Pharmaceutical Research article was not supported by the manufacturer of Eprex, we are now collaborating with the company (Johnson & Johnson, Raritan, NJ, USA) to further investigate our initial results. We have joint analytical efforts ongoing to confirm and better characterize the nature of the micelleassociated erythropoietin, and we will be conducting animal studies to explore the immunogenicity of the complex. This collaboration will allow us to obtain verification of results in separate and independent laboratories. Huub Schellekens Director, Central Animal Institute, Utrecht University, P.O. Box 80190, 3508 TD Utrecht, The Netherlands Daan J A Crommelin Department of Pharmaceutics, Utrecht Institute for Pharmaceutical Sciences, Utrecht University, P.O. Box 80082, 3508 TB Utrecht, The Netherlands e-mail:
[email protected]
1265
Sabine Louët responds: In researching the news story, I had several interviews with Huub Schellekens, who explained to me the key findings of his laboratory’s research on Eprex. At no point during these interviews did he strongly underline the fact that there was such a level of uncertainty regarding the findings of his study. However, it is clear that the activity of a therapeutic protein is likely to depend on many factors; indeed, the news article pointed out this fact: “Not only could the immunogenic
reaction be triggered by a change in formulation—as in the Eprex case—but also by variations in amino acid sequence, glycosylation or even by impurities cropping up during manufacturing or administration of the drug.” The adverse events associated with the manufacture, formulation and administration of Ortho Biotech’s (a Johnson & Johnson affiliate) erythropoietin alpha (Eprex) exemplify the difficulties faced by companies that seek to manufacture and formulate generic biopharmaceuticals.
Chaperonins govern growth of Escherichia coli at low temperatures
1266
Yakimov, M.M., Golyshin, P.N. & Timmis, K.N., unpublished data; Swiss-Prot accession numbers Q8KM30 and Q8KM31, respectively). Both chaperonins show high protein refolding activities in vitro at temperatures of 4–12 °C (16-fold higher than at 30 °C; Fig. 1b). We reasoned that if the cold-sensitive GroEL and GroES chaperonins of E. coli determine its lower growth temperature, and if the coldadapted Cpn60 and Cpn10 chaperonins of O. antarctica can assume the roles of GroEL and GroES in E. coli, then introduction of the corresponding genes into, and their expression in, E. coli should extend its temperature range of growth by decreasing its lower temperature limit. We therefore cloned and expressed the O. antarctica genes cpn60 and cpn10, encoding the two chaperonins, under the – cpn 60/10 + cpn 60/10
a
control the Plac promoter in E. coli strain XLOLR and examined the growth characteristics of the transgene after induction of expression with isopropyl-Dgalactopyranoside (IPTG; Fig. 1a). The strain bearing the construct grows much faster than the parental strain at low temperatures: 3-fold faster than the parental strain at 15 °C, 36-fold faster at 10 °C and 141-fold faster at 8 °C (growth rate of parental E. coli ∼0.002 h–1; that of the transgenic strain ∼0.282 h–1). No growth of the parental E. coli was detected below 8 °C, whereas the transgenic strain grew at temperatures below 4 °C As determined using the square-root growth model of Ratkowsky et al.6, the theoretical minimum temperatures for the parental and transgenic E. coli would be 7.5 °C and –13.7 °C, respectively (see Supplementary Methods online). To rule out the possibility that hyperexpression of chaperones per se lowers the growth limit of E. coli, we also expressed the GroEL and GroES chaperonins to similar cellular levels— 160 µg GroEL/ES per milligram of protein versus 120 µg Cpn60/10 per milligram of protein, using plasmids pBB528 and pBB541 (kindly provided by E. Betiku and U. Rinas (GBF)), in which the chaperonins are expressed from the same Plac promoter (for details, see Supplementary Fig. 1 online). The growth characteristics of E. coli at temperatures below 15 °C were not influenced by hyperexpression of the homologous chaperonins (data not shown). This demonstrates that the depression of the lower limit of growth of E. coli by Cpn60 and Cpn10 is due to a O. antarctica Cpn60/10 E. coli GroEL/ES
b 90
)
Refolding activity (%)
1 -1
To the editor: Growth and multiplication of specific cells and organisms occurs within narrow physico-chemical conditions. Despite the fundamental importance of one (or at most two) cellular functions that determine the growth range of a cell or an organism, in most cases we have little idea of their identity. Here, we report the finding that chaperonins determine growth at lower temperatures of the bacterium Escherichia coli K-12. The finding has implications for the use of bacteria in environmental biotechnology, biochemical engineering and recombinant protein production. E. coli is a mesophilic bacterium able to grow well in the temperature range from 21 °C to 49 °C, with an optimum at about 37 °C. The growth rate of E. coli strain XLOLR drops rapidly as incubation temperatures decrease from 20 °C, and the minimum for measurable growth is around 7.5 °C (ref. 1; Fig. 1a). Interestingly, the ability of the E. coli chaperonins GroEL and GroES to fold denatured proteins also rapidly decreases below 15 °C (ref. 2; Fig. 1b). These chaperones promote the folding and/or assembly of over 30% of cellular proteins, are required for bacteriophage morphogenesis and have a role in protein secretion3,4. The question thus arises of whether the vital role of chaperonins is the function that determines the lower temperature limit of E. coli growth. We recently isolated a new psychrophilic bacterium, Oleispira antarctica strain RB-8 T (DSMZ14852 T), from Antarctic seawater5 and characterized its chaperonin Cpn60 and co-chaperonin Cpn10 (Ferrer, M., Lünsdorf, H., Chernikova, T.N.,
Growth rate (h
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
CORRESPONDENCE
0.1
0.01
0.001
75 60 45 30 15 0
0
6 12 18 24 30 36 42 48
Temperature (°C)
0 6 12 18 24 30 36 42 48 54
Temperature (°C)
Figure 1 In vivo and in vitro properties of the chaperonins of Oleispira antarctica. (a) Effect of expression of the O. antarctica chaperonins on the growth of E. coli at different temperatures. (b) In vitro refolding activities of O. antarctica Cpn60/10 and E. coli GroEL/ES chaperonins at different temperatures. Data are not fitted to any model. For details see Supplementary Methods online.
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
CORRESPONDENCE qualitative change in cellular function effected by the psychrophilic chaperonins and not to a quantitative change in chaperonin level. We have thus established causality between increased growth rates of E. coli at lower temperatures and depression of its minimum temperature for growth, on one hand, and recruitment of the O. antarctica cold-adapted chaperones, on the other. This, in turn, demonstrates that the chaperones of E. coli are the rate-limiting cellular determinant of growth at lower temperatures. As the principal function of chaperones is protein folding, and we have shown a correlation between protein folding ability and growth at lower temperatures, it is highly probable that the cellular function determining growth of E. coli at lower temperatures is protein folding. Nevertheless, we cannot presently exclude the possibility that another chaperoninmediated cellular function is responsible for the altered growth characteristics. Two related questions arise from this finding: how widespread is chaperonedetermined growth among other organisms and under different environmental conditions, and will it be possible to extend the temperature ranges of growth of other cells by recruiting chaperones that have the required properties (e.g., can the temperature ranges of growth of psychrophiles and mesophiles be extended by recruitment of chaperones from mesophiles or thermophiles)? Whatever the case, the finding presented here has implications for biotechnology. One important strategy for developing new biocatalytic processes is to mine biodiversity by creating genomic libraries of DNA resources in E. coli and screening them for desired activities7. Enzymes from psychrophiles are particularly interesting for certain enzymatic bioconversions8, but some cannot be produced in an active form in E. coli because they are denatured in vivo at the temperatures used in cultivating this bacterium9. Thus, screens for psychrophilic enzyme activities would clearly benefit from growth of such E. coli libraries at low temperatures, and subsequent production of identified enzymes will also require low-temperature growth of the host organism. Use of an E. coli host producing the O. antarctica chaperonins, or other cold-tolerant chaperones, will permit both lower growth temperatures and efficient
folding of the psychrophilic proteins produced. It is noteworthy that, if the temperature ranges of growth of organisms generally prove to be modifiable by recruitment of heterologous chaperones, this could become a generic means of altering their biogeography and of making them more robust either for a wide range of environmental applications that are subject to climate-related fluctuations in temperature (waste treatment, bioremediation, microbially mediated plant growth promotion and protection, retting, biomining, etc.) or for biotechnological processes at temperatures that are stressful for the organisms used. It may also have other interesting implications for agriculture if the cold adaptation strategy we have developed for E. coli is also applicable to plants and can be used to increase their robustness to weather conditions and extend their growth windows in time (length of growing season) and space (latitude growth range). Note: Supplementary information is available on the Nature Biotechnology website. ACKNOWLEDGMENTS M.F. thanks the European Commission for a Marie Curie postdoctoral fellowship, and K.T. thanks the Fonds der Chemischen Industrie for generous support.
Manuel Ferrer1, Tatyana N Chernikova1, Michail M Yakimov2, Peter N Golyshin1 & Kenneth N Timmis1 1Division of Microbiology, German Research
Centre for Biotechnology (GBF), Braunschweig 38124, Germany. 2Istituto Sperimentale Talassografico, CNR, Messina, Italy. e-mail:
[email protected] 1. Ingraham, J.L. & Marr, A.G. Escherichia coli and Salmonella: Cellular and Molecular Biology edn. 2 (American Society for Microbiology, Washington, DC, USA, 1996). 2. Mendoza, J.A., Dulin, P. & Warren, T. Cryobiology 41, 319–323 (2000). 3. Gething, M.-J. & Sambrook, J. Nature 355, 33–45 (1992). 4. Walter, S. & Buchner, J. Angew. Chem. Int. Ed. Eng. 41, 1098–1113 (2002). 5. Yakimov, M.M. et al. J. Syst. Evol. Microbiol. 53, 779–785 (2003). 6. Ratkowsky, D.A., Lowry, R.K., McMeekin, T.A., Stokes, A.N. & Chandler, R.E. J. Bacteriol. 154, 1222–1226 (1983). 7. Olsen, M.J. et al. Nat. Biotechnol. 18, 1071–1074 (2000). 8. Cavicchioli, R., Siddiqui, K.S., Andrews, D. & Sowers, K.R. Curr. Opin. Biotechnol. 13, 253–261 (2002). 9. Feller, G., Le Bussy, O. & Gerday, C. Appl. Environ. Microbiol. 64, 1163–1165 (1998).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1267
C O M M E N TA R Y
Invention and commercialization in optical bioimaging Daniel L Farkas “It would be madness and inconsistency to suppose that things not yet done can be done, except by means not yet tried.” -Francis Bacon
Dimensions and complexity Imaging Discipline
Throughout history, light has been closely associated with inquiry and knowledge. In Novum Organum1, Francis Bacon wrote that “experiments of use are not enough”; there must also be “experiments of light”— meaning lumen siccum, the light of understanding. Optical imaging does make us understand better by seeing better. It is about as old as the living world, and one of the earliest methods of scientific inquiry. Yet in many respects, the development of novel imaging applications outside of basic research has been slow, and in areas of clinical practice, such as pathology, optical technology remains the same as it was a century ago. Light fantastic Light is the richest, most versatile imaging radiation. It is noninvasive and able to create contrast not only by intensity (e.g., like x-rays, positron-emission, or sound waves), but also through several other properties, such as wavelength, polarization, coherence, lifetime and nonlinear effects. Imaging methods that take advantage of one or more of these attributes can be combined for complementarity or even synergy. The light microscope, an icon of the sciences, is a highly versatile instrument with applications in fields as diverse as molecular biology, neuroscience, forensics, surgical pathology, silicon wafer inspection or art Daniel L. Farkas is at the Department of Surgery, Cedars-Sinai Medical Center, 8700 Beverly Blvd., Los Angeles California 90048, USA. e-mail:
[email protected]
Hierarchical level Information Spending (super-log) Commercialization
Microscopic Molecular biology
DNA
Cell biology
Protein
Genomics
Mesoscopic Histopathology
Cell
Proteomics
$$$
$$
Strongly emerging
Clearly emerging
Tissue Cytomics
Embryology
Embryo
Macroscopic In vivo
Medicine
Animal
Human
Pathology and phenomics
$
$
Somewhat emerging
Almost nonexistent
Clinical
$$$$$$ Established, but traditional
Erin Boyle
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
FOCUS ON OPTICAL IMAGING
Figure 1 Researching the living world—steps toward more informed intervention. The very small size of subcellular entities (left) requires special technologies for their study, as does the increased complexity of more clinically relevant organizational levels (right). The point of current ‘gold standard’ diagnostics, representing pathology, is shown by the red arrow separating these two domains. On the left of it, silicon and DNA are converging, providing a strong market. On the right, new technologies and clinical medicine are not yet converging convincingly in spite of huge demand, owing to obstacles more likely removable by strategic investment and smart legislation than market forces.
conservation. It extends visual perception to objects well below the limits of normal sight, providing the human eye with access to the ultramicroscopic world. The first modern microscopes were produced in the 1870s as a result of collaboration between German instrument maker Carl Friedrich Zeiss, mathematician and physicist Ernst Abbe and glass chemist Otto Schott (http://micro.magnet.fsu.edu/optics/ timeline/people/zeiss.html). By 1886, Zeiss’s company had successfully marketed over 10,000 instruments. Today, a handful of established companies produce light microscopes to high industrial and scientific standards, and there are several tens of thousands of research-grade microscopes in the world, with about ten times as many
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
simpler ones mostly in the hands of students and educators. The convergence of advances in hardware (e.g., video, solid-state cameras, micropositioning, lasers and computers) and our ability to manipulate living systems (e.g., to produce genetically encoded probes) has transformed what was previously a static, two-dimensional visualization instrument into a dynamic, four-dimensional research tool. Better light sources and detectors have enabled the introduction of such elegant methods as confocal scanning and multiphoton microscopy, and digital data handling has allowed deconvolution and enhancement of complex images. In tandem, the ability of antibody-antigen recognition specificity and recombinant
1269
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
C O M M E N TA RY technology to effect highly specific labeling of biological structures has enabled microscopists to focus on single entities at different locations in cells and tissues and at different times. Studies of molecular structures at atomic resolution have also enabled site-specific mutagenesis of natural probes to enhance their properties as reporters. Light microscopy is now capable of monitoring cellular and subcellular activities, all the way down to single molecular events2, digitally and quantitatively. Indeed, the sheer range of intensities (about 12 orders of magnitude), of time (femtoseconds to years) and of space (nanometers to centimeters), as well as the resolution within these ranges of optical imaging is orders of magnitude better than any other method. Certain instruments now also recruit light as an effector in experiments, enabling the manipulation (with high spatio-temporal resolution) of the very cellular structures that are being visualized. In basic research, fluorescence recovery after photobleaching3,4, photoactivation5, and laser tweezing6 have all been widely adopted, with new variations, such as laser-tracking microrheology and the optical stretcher, also emerging7. In medicine, photodynamic therapy (in which light is used to activate a prodrug) and laser surgery (in which the intensity of light is used as a photonic scalpel, or soldering iron) are fast gaining ground. An enabling technology? The expansion of applications for optical bioimaging is creating new market opportunities in the life and biomedical sciences; however, many of these are currently unevenly developed (see Fig. 1). Optical imaging has been enthusiastically adopted by the biology community, mostly as an analytical technology in scientific investigation. More recently, light microscopy has been applied in confocally scanned microarrays, optically scanned gels and fast cellular imaging for high-throughput screening. In research, the six-figure price tag of an advanced microscopy workstation does not seem to significantly inhibit sales (and, in fairness, compares well with the tenfold higher price tag of a magnetic resonance imaging machine). In 2000, the annual market for light microscopes was $520 million, with confocal microscopy and scanning probe microscopy comprising $55 million and $78 million, respectively (http://www.the-infoshop.com/study/ tk12574_microscopy.html). Innovations have started building on existing optical instrumentation platforms, and commer-
1270
cialization of the resulting products in the research community has been slowed only by some rather unfortunate patent disputes. In confocal microscopy, litigation between competing vendors has bankrupted several smaller companies and in multiphoton microscopy the first warning shots were fired this year between Zeiss (Jena, Germany) and Bio-Rad (Hercules, CA, USA)8. The adoption of optical bioimaging in medicine has been slow for several reasons. First, imaging structures deep within the body presents a significant challenge for visualization by optical methods. In the body, which is highly opaque and scatters light, one cannot use UV light for excitation, or high laser intensities or any other extreme conditions because of poor penetration and potential damage from the molecular all the way up to the gross tissue level. There is also a marked paucity of proper contrast agents, particularly for in vivo applications. Most of the agents currently used were introduced two to five decades ago, and fall short of today’s standards in both safety and efficiency, not having been developed with current technological capabilities in mind, or optimized for machine vision. The cost of properly testing a new agent is almost prohibitive, as regulatory hoops require their safety and efficacy verification in a very broad set of circumstances, to newly raised standards. However, one cannot overemphasize the importance of labeling, as biological objects lack intrinsic contrast. To quote Floyd Bloom, a former editor of Science, “The gain in brain lies mainly in the stain.”9 Second, there is a need for regulatory oversight and approval in point-of-care instruments. Laser eye surgery achieved fast success because of the quality of the instruments, the ‘optical nature’ of the target application (ophthalmology) and the relatively minor modifications needed for adoption. In gastrointestinal (specifically, small bowel) imaging, untethered optical capsule endoscopy (http://www.givenimaging.com/ Cultures/en-US/given/english) is an unexpected, inventive and quantifiable technology that has been approved by the FDA in record time because of its clear advantages over established procedures. And third, the medical community has traditionally been much more conservative than the research community in adopting new technology; physicians still largely rely on their own senses and intuition in diagnosis and treatment. Egregious examples
include clinical pathology, where the entire surgical procedure and its time line (including subsequent additional surgery because of false positives that can run as high as 25%) is set by the subjective call of a pathologist, and melanoma diagnosis, where the presurgical examination is about as advanced technologically as it was a millennium ago. Quantitatively speaking, some of the most important applications of optical bioimaging (e.g., as an analytical tool in pathology) are still done on fixed specimens, as they were a century ago. In an era when we trust satellite imaging to predict global meteorology, optical fibers to beam communications around the world and lasers to manufacture goods in factories and guide missiles with ‘reduced collateral damage,’ we somehow cannot bring ourselves to use advanced optical imaging to help diagnose disease10. Competition for eyeballs Several factors will be important in realizing the commercial potential of optical imaging in biology and medicine. In research, imaging technologies will continue to be essential for readout of biological assays that marry DNA and silicon. Gene sequencing machines and microarrays could hardly exist without optical technologies, and these platforms will likely continue to develop in the hands of engineers in the emerging areas of proteomics and cytomics. The increasing exploitation of imaging technology in high-throughput platforms (driven by the purchasing power of big pharma, biotechnology companies and large university centers that focus on applied chemical and genomic discovery) in drug discovery, teratology and clinical studies promises to expand potential markets for imaging instrumentation in biomedical and biological sciences. To realize the potential of clinical medicine as a market, strategic investment may be needed in mesoscopic imaging and advanced endoscopy, yielding much-increased relevance of the optical products. A second challenge for imaging is that human capital needs proper preparation. There are currently precious few training programs in imaging anywhere and this needs to be addressed. Notable exceptions include the US National Science Foundation Integrative Graduate Education and Research Training Program at the University of Texas (Austin, TX, USA; http://www.ece.utexas.edu/igert/optical_imag ing.html) and the US National Institutes of
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
C O M M E N TA RY Health (Bethesda, MD, USA)-supported Course on Fluorescence Spectroscopy at the University of Maryland (Baltimore, MD, USA; http://cfs.umbi.umd. edu/course/). One possible reason for the lack of education is that optical imaging’s markedly interdisciplinary nature makes it difficult to teach. Third, there is an opportunity for biomedical researchers exploiting optical imaging technology to collaborate with nontraditional partners. For instance, most imaging is now digital, and image processing is key. The entertainment industry has currently more processing power, creativity and savvy than any research laboratory. For the younger generation’s education, this science-entertainment collaboration might be one way of making the educational material exciting and visually stunning (www.ptei.org/educational_programs/planetarium). Additionally, methods developed for space exploration and satellite reconaissance, such as hyperspectral imaging, can also be highly useful.
One is struck by how limited exploitation of advanced optical imaging has been in the medical arena. There is clearly an acute need for high-resolution optical technologies for imaging within the body (that is, mesoscopic imaging) to facilitate diagnosis and treatment. As Francis Bacon, astutely observed centuries ago1: “The inadequacy of these microscopes, for the observation of any but the most minute bodies, and even those if part of a larger body, destroys their utility; for if the invention could be extended to greater bodies, or the minute part of greater bodies, so that...the latent minutiae and irregularities of liquids, urine, blood, wounds, and many other things could be rendered visible, the greatest advantage would, without doubt, be derived.” It is likely that, as we wait for molecular medicine and nanobiotechnology to mature, we will require a combination of optical methods (e.g., intensity-based, lifetime, coherence-domain, hyperspectral and photon diffusion) to address this challenge.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
Some of our brightest hopes are represented by the methods described in this issue, and with their help, one hopes that Bacon’s ‘light of understanding’ can truly be shifted from bench to bedside for the first time. 1. Bacon, F. Novum Organum (published 1620) (eds. Urbach, P. & Gibson, J.) (Open Court, Chicago, 1993). 2. Ishijima, A. & Yanagida, T. Trends Biochem. Sci. 26, 438–444 (2001). 3. Reits, E.A. & Neefjes, J.J. Nat. Cell. Biol. 3, E145–E147 (2001). 4. Braeckmans, K., Peeters, L., Sanders, N.N., De Smedt, S.C. & Demeester, J. Biophys J. 85, 2240–2252 (2003). 5. Callaway, E.M. & Yuste, R. Curr. Opin. Neurobiol. 12, 587–592 (2002). 6. Greulich, K.O. in Micromanipulation by Light in Biology and Medicine: The Laser Microbeam and Optical Tweezers, Enabling Techniques in Bioimaging (ed. Farkas, D.L.) (Birkhäuser Publishing Company, Basel, 1999). 7. Kuo, S.C. Traffic 2, 757–763 (2001). 8. Carl Zeiss Jena GmbH and Carl Zeiss Inc. v. Bio-Rad Laboratories, Inc. and Cornell Research Foundation, Inc., US District Court, Southern District of New York (CV-98-8012 RCC). 9. Appel, N.M. Ann. NY Acad. Sci. 820, 14–28 (1997). 10. Taylor, D.L. et al. Toxicol. Pathol. 22, 145–159 (1994). 11. Farkas, D.L. in Methods in Cellular Imaging (ed. Periasamy, A.) 345–361 (Oxford University Press, 2001).
1271
E R R AT A A N D C O R R I G E N D A
Erratum: Why biotech don’t pay dividends–yet Tom Jacobs Nat. Biotechnol. 21, 1283 (2003)
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
The title of this article contained a typographical error. The title should have read: “Why biotechs don’t pay dividends–yet”. Nature Biotechnology regrets the error.
Erratum: New biotech hubs may emerge as industry matures Paroma Basu Nat. Biotechnol. 21, 1123, 2003 The title of Table 1 incorrectly indicates the presence of data for 48 North American cities or counties. The original article, which appears in the News section of the Bioentrepreneur web portal (http://www.nature.com/bioent), does indeed contain these data. But the version reprinted here displays a truncated version of the table with ten data points: North American cities or counties that rank 1–5 and 43–48 in total annual operating costs for a biomedical research and development facility. Nature Biotechnology regrets the error.
Corrigendum: Invention and commercialization in optical bioimaging Daniel L. Farkas Nat. Biotechnol. 21, 1269–1271, 2003 The URL that appeared on p. 1271 was incorrect. The correct URL is http://www.ptei.org/educational_programs/Planetarium/ planetarium_project.html.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 12 DECEMBER 2003
1513
C O M M E N TA RY
Crystal gazing in optical microscopy
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Scott E Fraser For most of the three centuries after Hooke introduced optical microscopy, refinement of the instrumentation and approach made microscopes more convenient than anything else. A modern reader of Hooke’s classic treatise Micrographia1 (ca. 1655) has no problem in visualizing the instrument employed; as in modern microscopes, a light source, an objective lens and an eyepiece were used to project an image magnified a 100-fold to a 1,000-fold into a human eye. This range of magnifications and resolution (∼200 nm) has brought cellular morphology and tissue structure into view and made optical microscopy the perfect partner for biological investigation. With the maturation of the theory of optical design in the past century, however, significant enhancements in the resolution and the imaging power were achieved. Instruments offering reliable performance to the diffraction limit became routinely available. Contrast techniques were established that maximized the in-plane and depth resolution of intrinsic contrast (e.g., Nomarsky or differential interference contrast offers depth resolution of ∼300 nm). And fluorescence microscopes were refined that offered dramatically improved contrast on stained materials. The broader use of classic optical microscopy continued to be restricted, however, by two major limitations. First, most of the techniques required the fixation, sectioning and staining of the tissues to make the specimen sufficiently transparent and thin. Second, the human eye, and even the film that sometimes replaced it, restricted the field to imaging within established spatial and temporal resolution limits. Enabling steps The dramatic advances in resolution and information from modern microscopy have been made possible by replacing the eye of the observer with an area detector, such as a video camera, or a point detector, such as a photomultiplier tube. The use of a computer to control, acquire and process the image per-
Scott E. Fraser is at the Biological Imaging Center and Division of Biology, Beckman Institute 139-74, California Institute of Technology, Pasadena, California 91125, USA. e-mail:
[email protected].
1272
mits objects much smaller than the theoretical limit of resolution of the conventional light microscope to be detected and even followed within living cells. Strategies that exploit the quantitative nature of such digital microscopes permit subtle variations in intensity to be followed and interpreted in ways that traditional microscopists could never have imagined. In one of many such examples, computer-enhanced differential interference contrast optics have tracked the ∼8 nm steps of single molecular motors as they ‘walk’ along a microtubule2. Thus, the emergence of digital microscopies promised that optical imaging could be used to obtain molecular insights. The useful depth of optical imaging has also been redefined by such advances as optical coherence tomography3 and multiphoton microscopy4. High-resolution images can now be obtained over depths of hundreds of micrometers to as deep as millimeters. Second harmonic imaging microscopy captures signals from macromolecular assemblages, offering molecular insights into living tissues, even without the addition of stains. The resolution limits have been broken by a family of far-field techniques, such as 4 Pi microscopy5 or I5 microscopy6, in thin, transparent specimens, and now by advanced laser-scanning approaches that should be applicable to even optically unforgiving specimens7. Molecular insights offered by fluorescence resonance energy transmission imaging8 and direct molecular imaging is now made possible by the growing family of near-field scanning optical microscopies9. These impressive advances do not yet represent, however, the best the field can do. To really enable optical imaging with exquisite resolution, deep imaging potential and precision, several challenges lie ahead. Challenges The ultimate goal for microscopy is to image single molecules and their interactions with other molecules in complex biological structures, such as cells and tissues. But attainment of that goal will require new types of instruments that can overcome current technical challenges associated with resolution at the molecular level. As in other forms of imaging, the resolution of optical imaging is restricted by a
trade-off between the number of pixels in an image, the signal-to-noise ratio of the image data contained in those pixels and the image acquisition time. If all other things were kept constant, the time required to acquire an image climbs dramatically as the pixel size is decreased. Dye molecules are limited by the maximal rate at which they can give off fluorescence by the length of their excited-state lifetime (∼5 ns) and are present at relatively low concentrations inside of a cell (at a concentration of 1 µM, a volume element in a laser-scanning image contains approximately a dozen dye molecules). If these dyes were 100% efficiently excited, each dye molecule would give off ∼200 photons in the ∼1 µs that the signal from a single volume element (voxel) is collected, resulting in a total yield of about 2,500 photons per voxel. As encouraging as this might sound, two significant limitations are hidden in these numbers. The first of these is that a microscope collects and conveys to the detector only a small fraction of the emitted photons because of light that does not enter the objective is lost en route to the detector or fails to excite the detector; this results in an image that is noisy as a result of the stochastic fluctuations inherent in the small number of photons collected per voxel. The expected fluctuations are given by counting error (number counted ± square root of the number). In the above example, ∼25 photons might reach and excite the detector, resulting in an unavoidable fluctuation of ±5 (note that 25 ± 5 is a 20% standard deviation). Given that the dye was already completely excited in our example, there is no gain in signal made possible by turning up the exciting laser power; therefore, the only way to improve upon this unacceptable variation would be either to slow down the scan rate (so each voxel is interrogated for longer) or to average repeated scans. The second issue is that each dye molecule has a finite lifetime, giving off perhaps 10,000–100,000 photons before it is lost by bleaching. At the excitation rate given above, dyes would survive only a minute or two of continuous scanning. Thus, improving the signal-to-noise ratio by slowing down the scan rate of the laser beam or averaging frames is not without cost. The reality of fluorescence imaging is that there never seems to
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
C O M M E N TA RY be enough dye molecules/unit volume to offer the highest resolution images. Although this is probably not limiting for optical coherence tomography, the photon budget of modern imaging tools is one of the critical concerns for second harmonic imaging microscopy (due to low excitation probability), fluorescence resonance energy transfer, near-field scanning optical miscroscopy imaging and especially stimulated emission depletion imaging because of its increased bleaching rate. Fortunately, there are approaches to extract data from the small numbers of light-emitting molecules in each voxel, such as fluorescence correlation spectroscopy (FCS)10 and image correlation spectroscopy (ICS)11. In FCS, a single focal volume is excited and detected resulting in a noisy trace as labeled molecules diffuse into and out of the active volume. The autocorrelation of these signal fluctuations yields a measure of the diffusion constant of the labeled molecules; the crosscorrelation between two differentially labeled molecules can yield insights into the dissociation constant of the pair. ICS scans an active volume through an entire scene, and might be thought of as a form of spatially resolved FCS.
It can yield insights into the number of labeled molecules and their regional distribution, as well as their motions and diffusion. Thus, both approaches offer molecular insights from optical tools that otherwise would not even approach molecular scale spatial resolution. These insights do not come without some cost, however. The primary limitation of FCS and ICS is that they extract information from statistical tests made against a model of the label behavior; this limits the rate of data collection, both because of the time involved in processing of the data and because of the time needed to sample the signal from the active voxel(s). Again the photon budget limits the spatial and temporal resolution to events much longer than a nerve action potential. The vision Based on these musings, what vision emerges of the hurdles and promise of optical imaging tools? First, imaging technology is moving rapidly from a technology development phase to a phase in which the resulting tools are being actively applied to research problems in a number of fields. The venturesome technology developments of yesterday and today are
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
destined to become the routine laboratory instruments of tomorrow. Second, the photon budget is likely to remain the major limitation to the broader application of these technologies. As such, attention must be focused on the optimization of the light path of the instrumentation, the efficiency of the detectors and the robustness of the labels. With a combined improvement of only tenfold, today’s impossible project can become tomorrow’s routine research topic. 1. Hooke, R. Micrographia (Warnock Library, London, 1665). 2. Schnitzer, M.J & Block, S.M. Nature 388, 386–390 (1997). 3. Fujimoto, J.A. Nat. Biotechnol. 21, 1361–1367 (2003). 4. Zipfel, W.R., Williams, R.M. & Webb, W.W. Nat. Biotechnol. 21, 1369–1377 (2003). 5. Hell, S.W. & Stelzer, E.H.K. J. Opt. Soc. Am. A 9, 2159–2166 (1992). 6. Gustafsson, M.G.L., Agard, D.A. & Sedat, J.W. Proc. Soc. Photoopt. Instrument. Eng. 2412, 147–156 (1995). 7. Hell, S.W. Nat. Biotechnol. 21, 1347–1355 (2003). 8. Jares-Erijman, E.A. & Jovin, T.M. Nat. Biotechnol. 21, 1387–1395 (2003). 9. Lewis, A. et al. Nat. Biotechnol. 21, 1378–1387 (2003). 10. Schwille, R., Haupts, U., Maiti, S. & Webb, W.W. Biophys. J. 77, 2251–2265 (1999). 11. Wiseman P.W., Squier J.A., Ellisman M.H., & Wilson K.R. J. Microsc. 200, 14–25 (2000).
1273
C O M M E N TA RY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
A framework for designing transgenic crops—science, safety and citizen’s concerns Ariane König European legislation requires the phasing out of transgenic crops containing selectable markers conferring resistance to clinically used antibiotics by 2004. The interpretation of the law will affect research scientists, policy makers and stakeholders in the agro-food chain worldwide. In this commentary, I develop a framework for the assessment of methods used for the selection of transgenic cells according to criteria relating to product development, regulatory safety assessment and market acceptance. The assessment indicates that publicly funded research on plant selection methods should be encouraged, whereas the use of the few available, cost-effective, widely accessible and safe methods should not be restricted. Broader use of similar integrated assessment frameworks is recommended to inform research and product development strategies, evolving regulations and guidelines, and the public debate on transgenic crops. Rationale The polarized and value-laden debate on transgenic crops over the past decade has highlighted the need for governments and industry alike to address concerns voiced by nongovernmental organizations and the media, and to restore the public’s trust in the technology1. Improved approaches to technology assessment addressing the concerns of citizens, scientific experts, regulators and stakeholders are required to inform deliberation on applications of biotechnology. The low efficacy of currently routinely used DNA-delivery technologies results in only a very small proportion of targeted plant cells actually integrating the recombinant DNA with the trait-conferring genes stably in the nucleus. Means allowing identi-
Ariane König is at Harvard University, Harvard Center for Risk Analysis, 718 Huntington Avenue, Boston, Masschusetts 02115, USA. e-mail:
[email protected]
1274
fication of the transformed plant cells are therefore required. Selectable marker genes, which are linked to the trait-conferring gene before transformation, usually confer to all cells that have integrated the recombinant DNA the ability to grow on a specific nutrient medium that nontransformed cells can’t grow on. These markers have, however, no function in the commercial transgenic crop. The technology framework consists of the qualitative comparison of selectable markers conferring resistance to antibiotics, other selectable markers that might be used instead of antibiotic resistance markers and methods that avoid the presence of any selectable marker in or their removal from the transgenic crop before it is commercialized. (The assessment’s focus on the selection of tools for the transformation process precludes consideration of the novel traits conferred to transgenic crops, or the enduse of the new product.) The issue merits systematic assessment for two main reasons: first, transgenic crops selected using genes that confer resistance to antibiotics have been a major point of contention in the public debate; second, transgenic crops containing markers conferring resistance to clinically used antibiotics will have to be phased out by 2004, according to European legislation implemented in October 2002 (ref. 2). In fact, a European Commission (Brussels) Working Group charged with providing guidance on the interpretation of this law met in April 2003 for the first time. I developed the framework for two main purposes: to guide choices between alternative methods for plant cell transformation and selection, and to inform the development of recommendations on best practices in the design of transgenic crops and the phasing out of transgenic crops containing antibiotic resistance marker genes. The assessment finds that some of the new methods required for phasing out antibiotic resistance markers are less cost effective and not as widely accessible and applicable as previously established practices; others may raise new acceptance-related concerns.
A framework for comparative technology assessment The proposed comparative assessment framework uses three criteria to assess transformation methods: the suitability for research and development, likelihood of regulatory approval and societal/market acceptance. Suitability for research and development. Three factors influence the suitability of a method for the genetic modification of a crop for routine use in research and development: maturity, resource intensity and technology accessibility. The maturity of a method is best assessed by establishing whether efficient and reproducible laboratory protocols exist for the method’s use in a wide range of crops. The development of a new method for genetic modification and its adaptation to specific crops can take years; success is far from guaranteed. The lack of mature alternatives to an established practice is, however, a strong signal that the established practice should not be phased out in the near future, provided the established practice is recognized as safe. Methods that are not as yet mature but are promising in terms of safety and acceptance-related criteria (discussed in the next section) are suggested as a research priority; public funding of such research is recommended, in particular where a method has not already been patented by a private sector entity. The second factor, R&D resource intensity, is a function of the efficiency of the genetic modification process (transformation efficiency) that determines to a large extent the overall resource requirements for obtaining a modified plant line suitable for commercialization. The overall transformation efficiency, in turn, is largely a function of the efficiencies of the DNA transfer into plant cell genomes and of the selection of successfully modified cells. The subsequent screening process to select a plant line with commercial potential comprises three steps: first, transgenic inserts are characterized and plants that have only a single insert (single copies of the recombinant
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
C O M M E N TA RY genes are generally preferred, as this both simplifies the risk assessment and reduces the likelihood of gene silencing effects); second, plant lines that reliably express the introduced trait are selected; and third, the resultant transgenic plant lines are assessed to determine whether the genetic modification resulted in unintended effects that might affect human health or the environment. This involves analysis of morphology, plant development, agronomic performance, and the nutritional and antinutritional composition of the plant. Any significant deviations from these characteristics in the modified crop from nonmodified counterparts that serve as controls are assumed to be indicative of unintended effects. If further investigation confirms that changes in crop composition or other properties may have implications for human health, such plant lines are eliminated. The process from conception to production of a commercially viable plant line that fulfills all of the above criteria can take up to ten years3. The resource intensity of the transformation process can then be established through assessment of the numbers of plant lines that need to be analyzed at each stage of the product development process; this allows direct extrapolation of material and labor costs. Some selectable markers significantly affect transformation efficiency, and some methods for the elimination of selectable markers require additional screening or breeding steps. The relative resource intensity of different transformation technologies can then be estimated by comparing the number of plant lines that have to be analyzed with the number of plant lines handled if a well-established method like the neomycin phosphotransferase (nptII) marker is used in the same crop/trait combination. High relative resource efficiency reduces not only the monetary cost of product development, but also the opportunity costs within the firm in terms of how many different research and development projects can be carried out in parallel as space and laboratory equipment is saved. This has direct implications for the innovation capacity of groups engaged in research and development. If most alternatives to an established practice are more resource intensive, regulatory policy recommendations to phase out this practice will likely have significant negative impacts on the industry and innovation. The last factor, ease and cost of access to the technology for research and development in the public and private sector (in both developed and developing world), is largely determined by the patent situation.
Geographical marketing restrictions, or the need to obtain technology licenses are relevant to assessment of accessibility. Information on the patent holder may help to predict the likelihood and cost of obtaining a license. License fees for patented selection or elimination technologies contribute to the overall costs of the use of a particular transformation method in product development. Policies on best practices that restrict the use of established tools whose alternatives are patented by large multinationals will likely affect the industry structure and place smaller, resource-poor public laboratories and laboratories in the developing world at a disadvantage. The maturity, resource intensity and accessibility of a particular method should, however, be periodically reassessed to account for continued improvement of technologies over time. Regulatory and market acceptance. Three factors have been defined to assess the likelihood of regulatory and market acceptance of a technology: first, regulatory familiarity of a technology helps to highlight whether there are uncertainties on how to regulate the technology and what data should be required for its risk assessment; second, regulatory complexity highlights uncertainties of the magnitude of potential risks and limitations of methods used to assess risks; and third, market acceptance may be distinct from regulatory acceptance where the framing of questions on risk may differ substantially between lay persons and experts. The weighting of risks and benefits and associated uncertainties may play a role in this. The regulatory familiarity of a particular method to genetically modify crops is assessed through determination of the number and geographical spread of countries that have approved a transgenic crop developed with this or similar methods. The existence of regulatory guidelines for data requirements for the risk assessment of transgenic crops developed with particular methods is an additional indicator of regulatory familiarity. Like maturity, this factor should not deter investment of public funds into research and development of such methods—to the contrary. This is, however, a warning flag in product development, as regulatory acceptance will be more uncertain and will likely take longer to obtain. Regulatory complexity depends on the nature of the trait introduced through the genetic modification, recombinant gene sequences and proteins present in the transgenic crop, and the safety of the recombinant material and associated uncertainties. In both the United States and the European
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
Union the environmental and food safety assessment of transgenic crops starts with the comparison of the novel product to a closely related nonmodified counterpart that has an accepted standard of safety4. This allows identification of any changes resulting from the genetic modification that might represent a threat to human health or the environment. These changes are then the focus of the subsequent safety assessment. Assessment of the introduced DNA and proteins takes into consideration existing knowledge on the donor organism, including previous human exposure and known adverse effects to the donor organism. Introduced proteins have to be of known function and specificity, and are characterized as to their stability and possible structural or functional similarities with known allergens or protein toxins. Accepted and validated methods must exist to characterize potential hazards and to assess risks. For instance, a protein’s stability is often used as an indicator of a protein’s potential to induce allergic responses in humans. The introduction of a stable recombinant protein into a transgenic crop is likely to jeopardize regulatory acceptance of the product, because no validated and widely accepted methods exist to prove that stable proteins are not allergens5. The choice of transformation method might add to the regulatory complexity of a crop if this method results in the presence of recombinant material that has no role in the commercially cultivated transgenic crop. This holds true in particular where additional recombinant material adds uncertainty to the risk assessment; examples of this include DNA sequences that are prone to genetic rearrangements or enzymes with broad substrate specificity that might interfere with the crop’s own metabolism in unanticipated ways. Regulatory complexity should be seen as a warning flag for investment of research funds, adoption for product development and for recommendations as a future best practice for the design of transgenic crops. Uncertainties in the risk assessment of transgenic crops and regulatory complexity are often also the basis for scare-mongering stories on the technology in the media, which in turn may affect acceptance of the technology by regulators, politicians and others in the agro-food chain. Regulatory and market acceptance of a transgenic crop largely depends on whether an application for a similar transgenic crop developed with this or a similar method has already been approved by regulatory authorities and on whether there are significant uncertainties in assessment of potential risks
1275
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
C O M M E N TA RY to human health or the environment. These help anticipate whether a specific method used in the transformation process might increase time and resources required for completion of the regulatory approval procedure or even reduce the likelihood of regulatory and market acceptance of the final product. Researchers should consider early consultation with regulators from several countries to establish a safety assessment approach and data requirements for methods that have not been considered previously in a regulatory system. Moreover, to anticipate broader concerns that might ultimately affect regulatory and market acceptance, as in the case of the nptII marker, it is recommended to foster broader participation involving stakeholders and other interested parties in product development decisions as well as in the formulation of policies on best practices. Procedural approaches for soliciting structured input from a range of interested parties into deliberations on technological choices have been reviewed in detail elsewhere6. Such participatory technology assessment approaches may need periodic review as the understanding and development of the technologies proceed and values and norms affecting the perception of a technology and any uncertainties that might be associated with it may also evolve with time. The assessment of a method’s suitability for research and development, and its likely regulatory and market acceptance should be iterative and recursive. Iteration helps to integrate information on broader societal concerns as understood through participatory methods into product development decisions and safety assessments for regulatory purposes. Recursion over time is required, as both the efficacy of a method still in development and societal norms and values relating to new technologies may change rapidly over time. Assessing selectable markers and elimination methods Past academic reviews are quite upbeat about the diverse and expanding array of alternative selectable markers and marker elimination methods7,8. Policy makers in the United Kingdom9, the European Commission10 and the United Nations Food and Agricultural Organisation/World Health Organisation (Geneva, Switzerland)5 have provided recommendations on best practices in the design of transgenic crops emphasizing the need to keep recombinant DNA inserted into transgenic crops to a minimum. Deployment of methods for marker elimination is considered desirable. The implications for research
1276
and development of the resulting recommendations have, however, not been systematically considered. By applying the above-described assessment framework to published information on alternative selectable markers and methods for their elimination, I provide below a summary of the transformation technologies available today for the development of transgenic crops intended for commercialization. I do not evaluate systems that may in the longer term help to obviate the need for marker-based selection in some crops altogether because they are only in the early stages of development and will likely in the medium term only be applicable in a very few cases (e.g., oligonucleotide-mediated mutagenesis11, microinjection12). Some methods referred to in reviews7,8 and policy documents9,10 are not as yet sufficiently mature, cost-effective or broadly accessible to allow routine deployment across the industry; others raise new uncertainties on risk or stability of product performance. Selectable markers. Two categories of selectable markers are found in plants: negative selectable markers conferring resistance to cytotoxic agents that kill or inhibit growth of plant cells, and positive metabolic markers that allow plant cells to grow on unusual nutrient media that do not foster growth of nonmodified plant cells13. Currently used selectable markers are primarily genes that confer to plant cells resistance to cytotoxic agents, such as antibiotics and herbicides. The most widely used plant selectable marker gene in the developed and developing world is nptII, which confers resistance to the antibiotics kanamycin and neomycin. The technology is mature and cost-effective: reproducible transformation protocols using nptII as a selection marker yielding relatively high transformation efficiencies have been established for many crops of agronomic importance14. The patent on the nptII coding sequence combined with a particular regulatory sequence that allows expression of the bacterial gene in plants will expire soon. Moreover, the patent was not enforced when the gene was used for the commercial development of transgenic crops with traits of particular value to developing countries, such as the virus resistant papaya. Regulatory familiarity for the nptII selectable marker gene is a given: most regulators and scientific experts involved in risk assessment of transgenic crops are familiar with the nptII marker; several transgenic crops containing this gene, like Calgene’s (Davis, CA, USA) Flavr Savr tomatoes, virus-resistant papaya, and insect-protected potatoes/
maize/cotton have been approved in a number of countries, including Argentina, Canada, the European Union, Mexico, South Africa, the United Kingdom and the United States15,16. The presence of the nptII gene in the final product adds to the regulatory complexity, although the safety assessment of the recombinant material, the nptII gene and protein product, is not associated with significant scientific uncertainties. Because of the frequent occurrence of nptII in microbes in the human gut, humans have had significant exposure to the gene and its protein product. The protein’s substrate specificity is high. It does not inactivate structurally similar but more important antibiotics like amikacin, and it does not interfere with plant metabolism. It is readily digestible in vitro, and acute toxicity studies in mice using over 5,000 times the worst case exposure scenario through transgenic crops confirmed the absence of acute adverse effects17. The risk assessment of specific antibiotic resistance markers also has to address whether they might undermine the therapeutic effectiveness of antibiotics to which they confer resistance. This involves assessing the relative occurrence of the gene in microbial populations and taking into account the clinical value of the antibiotics. The nptII gene was derived from a common microbe in the human gut18. Cultivation and consumption of crops containing the nptII gene is unlikely to affect its frequency of occurrence in microbes, as the trait is already widespread in microbial communities and efficiently transferred between microbes. Transfer of a frequently occurring resistance gene like nptII between microbes is several orders of magnitude more likely to occur than microbial uptake of a transgenic crop-derived nptII gene in the environment or in the human gut19,20. Moreover, kanamycin and neomycin are not of clinical importance because of severe side effects and widespread bacterial resistance that limits their effectiveness21. However, the scientific community at large agrees that the closely related antibiotic resistance marker nptIII, which confers tolerance to the clinically important antibiotic amikacin, should not be used22. European policy makers have requested a phase out of the use of selectable markers conferring resistance to clinically used antibiotics. Because of difficulties in drawing distinctions between distinct antibiotic resistance markers in risk communication this might also affect the use of the nptII gene. The method is mature, cost-effective and widely accessible. However, in spite of given regula-
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
C O M M E N TA RY tory familiarity and limited complexity, the market acceptance is greatly reduced. An alternative category of cytotoxic marker is herbicide-tolerance conferring genes. The efficiency of herbicide-tolerance as a selectable marker depends on the herbicide-tolerance trait and the crop that is being modified. Transformation efficiencies using glyphosate tolerance in maize are lower than those using nptII in biolistic gun transformation. Agrochemical companies that produce the respective herbicide are usually the owners of the herbicide tolerance genes—access to herbicides for product development purposes is thus often restricted or comes at a cost. Furthermore, the presence of some herbicide-resistance-conferring selectable markers may not always be desirable, as it may raise regulatory and market acceptance concerns. For instance, glyphosate-tolerance is not desirable in potatoes that are cultivated in western Europe as glyphosate is the agent of choice to control potato volunteers. A herbicide-tolerance marker in a transgenic crop that is not registered for the herbicide-tolerance trait may tempt farmer’s misuse of the herbicides that the crop is tolerant to, but uses of which on this crop are not authorized. Herbicide tolerance is recommended as a selectable marker in those crops in which it is also the desired trait, provided the genes confer sufficient tolerance to allow establishment of a sufficiently effective transformation protocol. In summary, the accessibility and applicability of herbicide tolerance genes is limited in most cases, and regulatory complexities may be associated with them on a case-by-case basis. Glyphosate-tolerance as a marker is mature, its applicability is limited, however, as the efficacy of selection depends on crop species, the gene is patented by Monsanto (St. Louis, MO, USA), and for above-mentioned reasons. Regulatory familiarity exists, the safety assessment is straight forward and market acceptance is not a problem. One more recently developed alternative to antibiotic resistance markers are metabolic markers that confer to plant cells the ability to grow on unusual carbon sources or precursors for growth hormones that will only support growth of plant cells that have taken up the recombinant DNA19,23. Closest to commercial application is the marker gene phosphomannose isomerase, which enables plants to derive carbon from the sugar mannose. It is not applicable in plants that naturally contain the enzyme. Preliminary results indicate that transformation frequencies in corn using mannose selection can exceed frequencies using kanamycin selection up
to tenfold24. To date, however, the technology cannot be considered ‘mature.’ Transformation protocols have been developed only for maize and sugar beet25. The patent is owned by a private sector entity, and access to the technology by competitors is likely to be restricted. There is no regulatory familiarity as no transgenic crop developed with a metabolic selectable marker has been considered by regulators for commercialization; no guidelines on data requirements for their risk assessment exist. Regulatory complexities might arise from the need to assess potential unintended effects through interference with other related plant metabolic pathways. Furthermore, potential ecological impacts of plants with new growth advantages will need to be assessed on a case-bycase basis26. Investment of research funds in this area will certainly be beneficial, but the lack of maturity, accessibility and regulatory familiarity do not allow it to be considered as a suitable alternative to present uses of the nptII marker in product development or in formulation on policies on best practices in the design of transgenic crops. Marker elimination methods Promising methods that are being developed for the removal of marker genes from transgenic crops include cotransformation, homologous recombination, and recombinase-mediated excision7,8,27,28. All allow, in theory, phasing out the use of antibiotic resistance markers and would minimize the amount of introduced recombinant DNA sequences. Potential benefits from such research include the facilitation of the risk assessment of transgenic crops developed with this technology through reduction of recombinant sequences present in the final product, and increased regulatory and market acceptance of transgenic crops without antibiotic resistance markers. Other advantages of marker elimination in product development include the possibility of retransforming transgenic crops using the same marker gene to insert additional traits. Moreover, if multiple new traits are combined in one transgenic crop line using conventional breeding methods the removal of marker genes and other superfluous DNA sequences reduces the risk of gene silencing, which can occur when several identical gene regulatory or coding sequences are present. Homologous recombination relies on the occurrence of base pairing between identical sequences that are in close proximity during the DNA replication process. This can lead to the excision of DNA sequences that are located between the two repeated DNA
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
sequences29–31. At present, recombination frequencies are, however, very low. The method is, therefore, not as yet sufficiently mature or resource efficient for routine use in product development31. Use of this approach will also likely add to the regulatory complexity of resulting transgenic crops. Careful investigation of the stability of the remaining insert after recombination will be required in safety assessments of transgenic crops developed using such approaches, as high recombination frequencies may be indicative of transgene insertion in recombination hotspots or other less stable areas of chromosomes28,32,33. The incentives to invest in further development of this approach are high as improved control over homologous recombination in plants promises in the long term to help the development of gene targeting techniques, allowing greater control over the insertion locus, and expression of recombinant genes, which in turn would reduce uncertainties related to risk assessment and product performance. The second method for marker elimination, cotransformation, introduces changes into routinely used Agrobacterium-mediated DNA delivery protocols to obtain plant cells in which the gene of interest and the marker gene integrate into separate genomic locations, preferably on separate chromosomes, to allow segregation of the two insertions by breeding. The method is considered mature, as it has been developed over the past decade to work in a range of crop species34,35. In the most successful experiments, up to 25% of the cotransformed cell lines contained single copies of both marker and trait-conferring genes that could be segregated through breeding35–37. A fourfold increase in the number of transformed cell lines that have to be handled has, however, significant repercussions on laboratory space, cost and time needed for the development of crops with new advantageous traits. This can be prohibitive for use of the method in smaller laboratories. Furthermore, use of the array of related methods for those with commercial intentions is restricted through patents. Even so, a significant advantage of this method, at least for large corporations, is that no regulatory or market acceptance concerns are likely to result from it; it allows clean separation of marker and trait in the transgenic crop for commercial cultivation. The third strategy for marker elimination is the use of site-specific recombinase enzyme-mediated marker excision systems. A recombinase enzyme specifically cuts DNA at the two short parallel DNA recognition sites and then reseals the two DNA strands after
1277
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
C O M M E N TA RY intervening DNA between the two sites has been removed. In transgenic crops in which markers are to be removed using recombinase-mediated excision, the selection marker gene is flanked by the specific recombination target sequences. A subsequently introduced recombinase protein precisely excises the marker sequence from the genome. Recombination then results in clean marker excision. Several different sitespecific recombination systems derived from yeasts38–40 and from viruses41 have been shown to function in plant cells. Three distinct strategies for introduction of the recombinase protein are being investigated, mainly using the bacteriophage P1derived Cre/lox system. In the first strategy, autoexcision, the DNA introduced into the transgenic crop contains the marker gene with an adjacent cre gene flanked by a pair of lox sites. The expression of the cre gene is controlled by an inducible, developmental stage or tissue-specific regulatory element. This allows switching on the cell’s production of the recombinase enzyme at a specific point in time during the genetic modification process, after successful selection of modified cells. The promoter induction results in recombinase production and in the excision of the marker gene and the cre gene that are positioned in between the two Cre-recognition sites. In practice, the strategy is tricky, as any basal level of expression of the recombinase gene that occurs in the absence of promoter induction will result in premature elimination of the selection marker and cell death on the selective growth medium. Successful auto-excision of a marker gene and the adjacent cre recombinase gene flanked with the lox-sites that mediate Crebinding and recombination has now been achieved42. The transformation efficiency, that is the number of cells surviving on the selectable medium is, however, still too low to allow routine use of this method in product development. An alternative method that allows transient direct introduction of the Cre protein into plant cells that contain lox-flanked markers is by micro-injection or transformation43. To date, neither of the two approaches for the transient introduction of the Cre protein into cells has yielded transformation efficiencies that would allow their routine use for commercial product development. Furthermore, a license is required for such uses. Auto-excision and transient protein introduction are, however, methods that promise marker elimination in a very controlled and precise manner, adding no additional recombinant material to the plant
1278
genome other than the 35 base pair lox site. Experimentally, the most successful strategy for Cre introduction into a plant line that contains the trait-conferring gene and the lox-site-flanked marker involves a sexual cross with a plant line containing the cre recombinase gene. According to mendelian inheritance patterns, if both the Cre-containing and trait-containing plant lines are homozygous for their respective traits, all of the plant lines in the F1 generation contain the Cre and marker genes; Cre-mediated marker excision will occur in these plant lines. The cre gene and the trait-conferring gene can then be segregated through breeding in plants of the F2 generation44. The final product will contain only the gene of interest with one lox site44,45. Marker-free plants can be obtained in the following generation. The method is mature; protocols have been developed for a range of crops with commercial interest. It is, however, more resource intensive than the use of nptII as two additional breeding steps are required to segregate marker from trait. In summary, the most promising variation of the Cre/lox system, auto-excision, is not as yet sufficiently mature. Moreover, use of the method requires licensing. Regulatory familiarity is not given, as no regulatory applications for crops developed with this method have been submitted to date and no guidelines exist on what data are required for the risk assessment of such crops. The regulatory complexity of crops developed with this method is increased, as the likelihood of occurrence of unintended recombination events will need to be assessed through comprehensive study of the binding site-specificity of recombinases and the occurrence of potential recombinase-recognition sites in plant genomes. Such unintended rearrangements will likely only occur at low frequency compared with natural mechanisms for rearrangements of plant genomes46. Such uncertainties may, nevertheless, represent a welcome target by opponents to genetic engineering; market acceptance is therefore difficult to predict. Conclusions Only a limited number of methods for the genetic modification of crops is currently available. The above assessment finds that some of the new methods that are developed to replace or eliminate antibiotic resistance markers are less cost effective than previously established practices; others may raise new acceptance-related concerns. Some of the methods, like homologous recombination and recombinase-mediated excision, are still
too immature and inefficient to allow their widespread use in product development. It can take years to develop and optimize selection methods for use within specific crops and with specific transformation methods; success is not always granted. Furthermore, most of the few new selectable markers and elimination methods that are sufficiently mature for use in product development are already patented; access to them is restricted. Not all other selection methods are suitable for use in all crops, or in conjunction with all transformation methods. Consideration of the resource-intensity criterion clearly demonstrates that a method like cotransformation, one of the few currently available methods that allow avoidance of selectable markers in the final marketed product is significantly more resource intensive than nptII and may place small laboratories at a disadvantage. The use of metabolic markers that may in future be used instead of antibiotic resistance markers and recombinase-mediated excision for clean excision of selectable markers open regulatory new questions that are to be addressed in risk assessments. The preliminary application of the framework outlined here clearly demonstrates that aggressive enforcement of European Community law requiring the phasing out of transgenic crops developed with the nptII gene may well turn research on improvement of transformation methods into a bottleneck for innovation in plant biotechnology. If barriers to innovation are to be minimized, this indicates that policymakers should not only encourage the development of an increasing array of technologies, but also not restrict the use of the few available, cost effective, widely accessible and safe methods. One limitation of the type of technology assessment outlined here is that rapid scientific progress and changes in societal norms and values may mean that the results only apply in a narrow window of time. The assessment of suitability for research and development and regulatory and market acceptance should therefore be iterative and recursive. Iteration helps to integrate information on broader societal concerns as understood through participatory methods to feed back into safety assessments for product development decisions and regulatory purposes. Recursion over time is required, as both the efficacy of a method still in development and societal norms and values relating to new technologies may change rapidly over time. Another consideration is that risk and uncertainty are concepts grounded in culturally embedded norms and values that can differ across societies. The design of parti-
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
C O M M E N TA RY cipatory methods for broader engagement to better understand possible concerns surrounding a method from a wide range of perspectives should take this into account. It can also be challenging to attribute different weights to opinions on risk and uncertainties from experts and lay members of panels in cases where they diverge. The use of a more formal decision framework for technology assessment enables more comprehensive analysis and facilitates the comparison of similar assessments for alternative methods. A more systematic approach also can help to identify and prioritize knowledge gaps that need to be filled. Structured approaches can also facilitate the communication of decisions. Deployment of comprehensive technology assessment frameworks is recommended for setting research priorities, guiding product development strategies and informing evolving regulations and guidelines on transgenic crops. Technology assessment frameworks similar to the one presented here might, for example, be developed for the evaluation of new traits for transfer to transgenic crops, and/or for whole new transgenic crops with specific novel characteristics that result in new uses. The iterative and recursive approach to, and greater engagement in technology assessments, may help producers, regulators, policymakers and the interested public to develop a better understanding of the potential impacts on society of individual methods and whole new technologies for producing transgenic crops. ACKNOWLEDGEMENTS I am grateful for helpful comments from Mark Bailey, Michael Boutros, Alan Gray, John P. Holdren and Bruno Tinland. The research was funded by the German Bundesministerium für Bildung und Forschung through a contract from the Biologische Bundesanstalt. 1. Moore, J.A. Issues Sci. Technol. XVII, 31–36 (2001). 2. European Council Directive 2001/18/EC. Official J. L 106, 0001–0039 (2001). http://www.europa.eu.int/ comm/food/fs/sc/scp/out31_en.html 3. Armstrong, C. et al. Crop Sci. 35, 550–557 (1995). 4. FAO/WHO. Safety Aspects of Genetically Modified Foods of Plant Origin. Report of a Joint FAO/WHO Expert Consultation (FAO/WHO, Geneva, 2000). http://www.fao.org/es/esn/food/gmreport.pdf 5. Kuiper, H.A., Kleter, G.A., Noteborn, H.P.J.M. & Kok, E.J. Plant J. 27, 503–528 (2001). 6. Renn, O., Webler, T. & Wiedemann, P. Fairness and Competence in Citizen Participation: Evaluating Models for Environmental Discourse (Kluwer Academic Publishers, Dordrecht, 1995). 7. Hohn, B., Levy, A.A. & Puchta, H. Curr. Opin. Biotechnol. 12, 139–143 (2001). 8. Ebinuma, H. et al. Plant Cell Reports 20, 383–392 (2001). 9. UK Advisory Committee on Releases into the Environment (ACRE). Subgroup on Best Practices in GM Crop Design. Guidance Principles of Best Practice in the Design of Genetically Modified Plants. (ACRE, London, UK, 2001). http://www.environment.defr.gov.
uk/environment/acre/bestprac/guidance/index.htm 10. European Commission. Guidance Document for the Risk Assessment of Genetically Modified Plants and Derived Food and Feed (6-7 March 2003 - Prepared for the (Scientific Steering Committee by The Joint Working Group on Novel Foods and GMOs composed of members of the Scientific Committees on Plants, Food and Animal Nutrition). Scientific Steering Committee, European Commission, Brussels, 2003). http://europa.eu.int/comm/food/fs/sc/ssc/out327_en.pdf 11. Rice, M.C., Czymmek, K. & Kmiec, E.B. Nat. Biotechnol. 19, 321–326 (2001). 12. Knoblauch, M., Hibberd, J.M., Gray, J.C. & Van Bel, A.J. Nat. Biotechnol. 17, 906–909 (1999). 13. Joersbo, M. & Okkels, F.T. Plant Cell Reports 16, 219–221 (1996). 14. Gasser, C.S. & Fraley, R. Science 244, 1293–1299 (1989). 15. United States Food and Drug Administration. Fed. Reg. 59, 26700–26711 (1994). 16. James, C. Global Review of Transgenic Crops: 2001. ISAAA Brief 23 (International Service for the Acquisition of Agribiotech Applications, New York, 2001). http://www.isaaa.org/Publications/Downloads/ Briefs%2024.pdf 17. Fuchs, R. et al. Biotechnology 11, 1543–1547 (1993). 18. Horsch, R.B. Science 227, 1229–1231 (1985). 19. Nap, J.-P. Transgenic Res. 1, 239–249 (1992). 20. Nielsen, K.M., Bones, A.M., Smalla, K. & van Elsas, J.D. FEMS Microbiol. Rev. 22, 79–103 (1998). 21. Siegenthaler, W.E. Am. J. Med. 80, 2–14 (1986). 22. Anonymous. Scientists warn of GM crops link to meningitis. Daily Mail 26 April (1999), p. 10. 23. Haldrup, A., Petersen, S.G. & Okkels, F.T. Plant Mol. Biol. 37, 287–296 (1998). 24. Negrotto, D., Jolley, M., Beer, S.R.W.A. & Hansen, G. Plant Cell Reports 19, 798–803 (2000). 25. Joersbo, M. et al. Mol. Breeding 4, 111–117 (1998). 26. Kuiper, H.A., Kleter, G.A., Noteborn, H.P.J.M. & Kok, E.J. Plant J. 27, 503–528 (2001). 27. Puchta, H. Trends Plant Sci. 5, 273–274 (2000). 28. Hare, P.D. & Chua, N.-H. Nat. Biotechnol. 20, 575–580 (2001). 29. Peterhans, A., Schlupmann, H., Basse, C. & Paszkowski, J. EMBO J. 9, 3437–3445 (1990). 30. Reiss, B., Klemm, M., Kosak, H. & Schell, J. Proc. Natl. Acad. Sci. USA 93, 3094–3098 (1996). 31. Zubko, E., Scutt, C. & Meyer, P. Nat. Biotechnol. 18, 442–445 (2000). 32. Yoder, J.I. & Goldsborough, A.P. Biotechnology 12, 263–267 (1994). 33. Puchta, H. & Barbara, H. Trends Plant Sci. 1, 340–345 (1996). 34. Depicker, A., Herman, L., Jacobs, A., Schell, J. & van Montagu, M. Mol. Gen. Genet. 201, 477–484 (1985). 35. De Framond, A.J., Back, E.W., Chilton, W.S., Kayes, L. & Chilton, M.-D. Mol. Gen. Genet. 202, 125–131 (1986). 36. Komari, T., Hiei, Y., Saito, Y., Murai, N. & Kumashiro, T. Plant J. 10, 165–174 (1996). 37. McKnight, T.D., Lillis, M.T. & Simpson, R.B. Plant Mol. Biol. 8, 439–445 (1987). 38. Kilby, N., Davies, G.J., Snaith, M.R. & Murray, A.H. Plant J. 8, 637–652 (1995). 39. Lyznik, L., Mitchell, J.C., Hirayama, L. & Hodges, T.K. Nucleic Acids Res. 21, 969–975 (1993). 40. Onouchi, H. Nucleic Acids Res. 19, 6373–6378 (1991). 41. Dale, E.C. & Ow, D.W. Proc. Natl. Acad. Sci. USA 88, 10558–10562 (1991). 42. Zuo, J., Nui, Q.W., Geir Moeller, S. & Chua, N.-H. Nat. Biotechnol. 19, 157–161 (2001). 43. Gleave, A., Mitra, D., Mudge, S. & Morris, B. Plant Mol. Biol. 40, 223–235 (1999). 44. Russell, S.H., Hoopes, J.L. & Odell, J.T. Mol. Gen. Genet. 234, 49–59 (1992). 45. Ow, D.W. Curr. Opin. Biotechnol. 7, 181–186 (1996). 46. König, A. in Proceedings of the 6th International Symposium on The Biosafety of Genetically Modified Organisms. (eds. Fairbairn, C., Scoles, G. & McHughen, A.) 171–179 (University Extension Press, University of Saskatchewan, Saskatoon, Canada, 2000).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1279
C O M M E N TA RY
Gene medication or genetic modification? The devil is in the details © 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Grethe S Foss and Sissel Rogne Medication based on the transfer of genes, like gene therapy and DNA vaccines, holds the promise of combating diseases both in humans and animals. Similar methods of gene transfer are used when making genetically modified (GM) animals. When exactly is a ‘gene-medicated’ animal also a genetically modified organism (GMO)? In Europe and elsewhere, GMOs are subject to limited release and to rules of labeling; therefore, the answer could have implications for pharmaceutical companies, veterinarians, food producers, consumers and even pet owners. The overlap of the fields of gene medication and genetic modification is a challenging area where two distinct cultures and regulatory systems hold sway: medicine and biosafety. The first focuses mainly on intended effects in target animals, the second on unintended effects on ecosystems; the fields often represent conflicting perspectives. Central to the territorial battle are the regulatory definitions of ‘medicinal product’, ‘genetic modification’ and, as a result of exemptions already laid down in European regulations, the issue of what constitutes an ‘organism.’
eliminate the boar taint from pork—a process called ‘immunocastration’—and the vaccines being developed to reduce the fertility of pest animals like the wild Australian rabbit. Similarly, genetic modification is not limited to the addition of heritable properties. The definition of genetic modification of organisms is based on the technology used and not on the intention. In the EU directive 2001/18/EC on deliberate release of GMOs, the term GMO is defined as ‘an organism, with the exception of human beings, in which the genetic material has been altered in a way that does not occur naturally by mating and/or natural recombination.’ In the Cartagena Protocol on Biosafety to the Convention on Biological Diversity, the focus is likewise on the technology used and not on the properties added. Here, the equivalent of a GMO, a ‘living modified organism’ (LMO), is defined as ‘any living organism that possesses a novel combination of genetic material obtained through the use of modern biotechnology.’ Medication using an integrative gene therapy construct could, for instance, be seen to lead to a ‘novel combination of genetic material.’
Definitions People intuitively understand the term ‘medicinal product’ to mean preventing, diagnosing or treating disease. The definition is in fact broader, in that a substance that may be administered to animals with a view to modifying physiological functions is likewise considered a veterinary medicinal product, as stated in the European Union (EU; Brussels) Directive 2001/82/EC. Illustrative examples are CSL Animal Health’s (Parkville, Victoria, Australia) gonadotropin-releasing factor vaccine (Improvac) developed for male pigs to
When is medication also modification? The definition of genetic modification in the EU directive 2001/18/EC is made deliberately vague to cover new methods developed: “genetic modification occurs at least through the use of techniques listed in Annex I A, part 1.” Injection of genetic material into testes is now being explored as a new method for genetically modifying animals1. In the annex list, this method is covered under “techniques involving the direct introduction into an organism of heritable material prepared outside the organism including micro-injection, macro-injection or micro-encapsulation.” Therefore, genetic modification can occur by injecting genes into whole animals, rather than manipulating cells in the laboratory. It is also genetic modification if, instead of direct injection, the foreign nucleic acid molecules are inserted ‘into any virus, bacterial plasmid or other vector system’ and incorporated ‘into a host organism in which they do not occur naturally but in which they are capable of continued propagation.’ Thus, gene medication applying heritable material or recombinant nucleic acid
Grethe S. Foss and Sissel Rogne are at the Norwegian Biotechnology Advisory Board, an independent body appointed by the government. S.R. is also in the Department of Public Health, University of Bergen, Norway, and the Department of Nature Conservation at the Agricultural University of Norway. An English version of the Board’s report and statement on the issue can be found on http://www.bion.no. e-mail:
[email protected]
1280
molecules capable of continued propagation could be seen as genetic modification. According to the Cartagena Protocol as well, methods of gene medication can be covered under the LMO definition in the protocol. Here ‘modern biotechnology’ means “the application of a) in vitro nucleic acid techniques, including recombinant DNA and direct injection of nucleic acid into cells or organelles, or b) fusion of cells beyond the taxonomic family that overcome natural physiological reproductive or recombination barriers and that are not techniques used in traditional breeding and selection.” When the fate of the added DNA is uncertain, it is not clear how the definitions will be interpreted. Is testing of the offspring needed to decide whether the added gene was indeed ‘heritable material’? Different interpretations Regulatory bodies can interpret the definition of genetic modification differently. In a report last year, The British Agriculture and Environment Biotechnology Commission (London) took a stand on DNA-vaccinated animals2: “Importantly, the foreign DNA is not expected to integrate into the host’s genome and so the vaccinated animal is not genetically modified.” A more precautionary view is held by the Norwegian Directorate for Nature Management (Trondheim, Norway). In response to specific enquiries regarding gene medication of farmed salmon, they stated that a DNA-vaccinated fish is to be considered genetically modified for as long as the added DNA is present in the fish. Interestingly, in the United States, where there is no specific regulation of GMOs, genetic modification of animals is to be regulated as medication. The US Food and Drug Administration (FDA; Rockville, MD, USA) has asserted that the genetic constructs used to create transgenic fish (and other animals) fall under the legal definition of a drug as a substance “...intended to affect the structure or function of the body of man or other animals.” In a report issued in January and prepared for the Pew Initiative on Food and Biotechnology (Cambridge, MA, USA)3, concern is expressed over the FDA’s legal authority and its limited ability to consider the ecological risks of
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
C O M M E N TA RY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
GMOs. No federal agency seems to have clearcut legal authority to regulate or ban transgenic fish on environmental grounds. The report also points to the lack of transparency and public participation associated with the assessments. Unintended risks and product authorization The risk of an event is defined as the probability of the event times its consequences. Therefore, one could argue that where the consequences are particularly adverse, the risk could still be high, despite a low probability of the event. To address the risk of unintended as well as intended consequences, a thorough risk assessment is required when applying for deliberate release of GMOs into the environment. The assessment is evaluated by the relevant national GMO authority. In the case of gene medication, unintended spreading, uptake and integration of the foreign DNA are likely to happen at a certain low frequency. The questions then are, how adverse are the possible consequences, which institution(s) should evaluate these potential GMOs, and with whom should the responsibility of assessing the risks lie? In Europe, pharmaceuticals based on biotechnology are authorized through a centralized procedure by the European Agency for the Evaluation of Medicinal Products (EMEA; London). For medicinal products containing or consisting of GMOs, a compromise has been reached; they are exempted from the EU directive on deliberate release of GMOs when placed on the market, provided that the authorization procedure includes an environmental risk assessment equivalent to that provided for by the directive. As part of the procedure, the national GMO authorities are involved in evaluating the environmental risks of the medicinal products and the animals receiving them. Noticeably, the applications concerning GMO medicinal products lack the public openness central to other GMO applications and the GMO authorities do not have access to all the information needed for a cost-benefit analysis of the GMO. Moreover, the decision on market authorization is taken by the medicinal agency and not by the GMO authority as it is for other GMOs. The Cartagena Protocol, which entered into force on September 11 this year, does not have a similar exemption for veterinary medicinal products, only for products for human use. Therefore, countries that have ratified the protocol, including the EU, have to disclose detailed information about these products if a receiving country wishes to do a cost-benefit
analysis before importation. To date, few veterinary gene medication products have been granted market authorization. The first application of an experimental veterinary DNA vaccine was seen in the US in January this year. Californian condors in zoos and in the wild were vaccinated in an attempt to protect this endangered species from the West Nile virus4. Labeling issues in the EU When administering gene medication, food producers may risk having their living animals termed ‘GMO’ for a shorter or longer period. What may have more influence, however, is the fear of having to label the products of the animals as ‘GM food.’ Interestingly, the new EU regulation on GM food and feed (2001/0173(COD)) specifies that products of animals treated with GM medicinal products are not to be labeled as GM food. Considering the overlap between gene medication and genetic modification, does this exemption pave the way for the use of gene medication as a means of genetic modification through the backdoor? Gene medication products that are not GMOs in themselves are neither evaluated by the GMO authority as part of the authorization procedure, nor covered by the Cartagena Protocol. The philosophical question of what constitutes an ‘organism’ becomes a highly practical one the moment these regulatory differences influence the choice of vector used for gene medication. In the EU directive on deliberate release of GMOs, ‘organism’ means any biological entity capable of replication or of transferring genetic material. Similarly, in the Cartagena Protocol, ‘living organism’ means “any biological entity capable of transferring or replicating genetic material, including sterile organisms, viruses and viroids.” Plasmids are not seen as organisms in themselves. Although standard viral and plasmid vectors are easily categorized, the difference between genetically engineered viral genomes and plasmid-based viral genes packed into viral particles can be far from obvious. To a large extent, vectors can be engineered to suit the category preferred. It is not obvious whether the labeling exemption is limited to the medicinal products authorized as GMOs, or whether it also covers medicinal products based on naked recombinant DNA. Should the exemption cover GMOs only, organisms might be favored over nonorganisms as vectors for gene medication, regardless of the fact that plasmids might represent a lower risk both for the animal and the environment compared with viral vectors.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
A possible compromise? The Norwegian Biotechnology Advisory Board (Oslo) has made suggestions for how to regulate the overlap between gene medication and genetic modification. In theory, genemedicated animals could have the same kind of adverse effects as transgenic animals, although generally with a substantially lower probability. The Board therefore recommends that all gene transfer to animals outside the laboratory, whether the intent is medicinal or not, should be assessed on a case-by-case basis by the authority assessing deliberate releases of GMOs and evaluated according to the same principles. However, the Board sees it as important not to dilute the concept of GMO and recommends that animals treated with gene medication products in general should not be termed GMO. To avoid creating a new category of ‘gene-medicated organisms’ that has no foundation in international regulations, the risk evaluation should be part of a process of considering whether the animal ought to be termed genetically modified or not. The Norwegian Biotechnology Advisory Board suggests that the gene-medicated animal could be termed GMO if any of the following scenarios can be shown probable: first, that the added genetic material will be inherited by the offspring; second, that the genetic material will pose a risk to health or the environment if it is inherited; third, that the genetic material, through recombination, can result in organisms with new, unwanted properties; or fourth, that the genetic material will give the organism properties that will lead to a public outcry. Detailed guidelines need to be worked out for this system to become functional. In the present overlap between the fields of gene medication and genetic modification, the devil is in the details, both in the biological and regulatory sense. To avoid bizarre interpretations, creative loopholes and unforeseen environmental effects, a more subtle system is needed. We have presented a possible compromise where the GMO authority is formally involved in assessing the environmental risk of all gene transfers to animals outside the laboratory. This allows better regulation where less depends on the definitions, the most appropriate vector can be chosen and the development of low-risk medicinal products is encouraged. 1. Sato, M. et al. Mol. Reprod. Dev. 61, 49–56 (2002). 2. The British Agriculture and Environment Biotechnology Commission. Animals and Biotechnology (BAEBC, London, September 2002). 3. The Pew Initiative on Food and Biotechnology. Future Fish, Issues in Science and Regulation of Transgenic Fish (Pew Initiative, Cambridge, MA, January 2003). 4. Bouchie, A. Nat. Biotechnol. 21, 11 (2003).
1281
C O M M E N TA RY
The tyranny of ‘genethics’
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Leigh Turner Whether you work in bioethics, medical anthropology, science and technology studies or health law, if you have an entrepreneurial frame of mind, now is an excellent time to cultivate a ‘genethics’ research agenda. Around the world, ethics and policy centers related to genetics are proliferating. In Australia, Canada, the United Kingdom, the United States and many other regions around the globe, research teams are pursuing various projects on ethical and social issues related to genetic testing, genetic screening, genetics databases, germ line gene therapy, ‘cloning,’ xenotransplantation and embryonic stem cell research. Clearly, important ethical, legal, social and psychological issues are related to genetics. Scholars, policy makers and legislators rightly worry about how access to genetic information might lead to discriminatory practices in the workplace and in the provision of health insurance. Genetics research raises important questions concerning privacy, confidentiality, stigmatization, individual consent, community consent, resource allocation, access to genetic testing and screening, and intellectual property. There is a need to develop analyses, policies and practice guidelines that might potentially influence physicians, researchers, regulatory bodies and legislators. Still, we need to consider whether placing such an emphasis on ‘genethics’ themes ignores other important ethical, legal and social issues. Between 1990 and 1999, the Ethical Legal and Social Issues (ELSI) Research Program of the National Human Genome Research Institute (Bethesda, MD, USA) spent $58.3 million on ELSI funding. Over that same period, the ELSI Branch of the US Department of Energy spent $18.5 million.
Leigh Turner is in the Biomedical Ethics Unit, Department of Social Studies of Medicine, Faculty of Medicine, McGill University, 3647 Peel Street, Montreal, Quebec H3A 1X1, Canada, and at Montreal General Hospital, 1650 Avenue Cedar Montréal, Québec, H3G 1A4, Canada. In 2003–2004, he is a member at the School of Social Science, Institute for Advanced Study, Einstein Drive, Princeton, New Jersey 08540, USA. e-mail:
[email protected]
1282
The yearly expenditures for ‘ELSI’ research might just look like chump change to you if you work outside the humanities and social sciences. However, if you are an anthropologist or bioethicist, government-funded genethics research programs are the wealthiest ‘sugar daddies’ you are likely to find. Genethics is hot. Genethics is sexy. Genethics is generating a lot of poorly designed, repetitive, marginally useful research. Genethics is obscuring the careful consideration of other equally or more pressing social issues. The availability of funding for genethics research is attracting researchers who—in the absence of this cornucopia of financial resources—would likely never consider pursuing scholarship in this area. It is hard to believe that pure intellectual fascination or a sense of civic duty are the only factors prompting so many scholars from anthropology, sociology, media studies and bioethics to pursue genethics-related research. The problem with this focus is that a vast number of important social issues are neglected by scholars who might otherwise dedicate their careers to more pressing social concerns. For example, although many homeless individuals have psychiatric disorders, are meaningful solutions to the homeless problem plaguing many cities in North America really going to be provided by molecular biology and psychiatric genetics? Are famines and malnutrition in developing nations going to be solved by the introduction of transgenic crops rather than by addressing broader issues concerning the global distribution of basic human resources? Do we need developments in genetics to reduce global mortality and morbidity from the use of heavily advertised tobacco products? Can we expect geneticists to provide meaningful social responses to the widening gap between wealthy, resourcerich, developed nations and poor, developing nations? Can any properly informed individual think that AIDS has a solution rooted solely in genetics rather than in a multi-pronged effort involving the provision of contraceptives and preventive measures, better education, better job opportunities and greater equality between men and women? Many of the great social problems facing the world today are not going to be
solved by breakthroughs in genetics. Similarly, the focus on genethics obscures the extent to which most pressing social issues have rather little to do with genetics. Genetics research is tremendously important. Similarly, ethical, legal and social issues related to genetics are worthy of careful investigation and deliberation. Still, I am concerned that the abundance of funding for genethics scholarship is skewing research agendas and luring scholars away from the study of other topics that are as important, or more significant, than the careful consideration of genethics. Should governments and funding agencies dedicate less funding to ‘ELSI’ and ‘genetics, ethics law and society (GELS)’ research? Should far more resources be directed toward the study of profoundly important social issues, such as homelessness and international inequalities in health? There is little point in attempting to provide a general response to priority-setting exercises that need to be attuned to local needs and circumstances. That said, it is a cause for great concern that so many philosophers, lawyers, bioethicists, media studies researchers, medical anthropologists, medical sociologists and other ‘social critics’ are throwing their hats into the genethics arena and filling their caps with abundant research funds. My guess is that if the money were not there many of these scholars would turn their attention to social concerns, ethical issues and legal matters that are far more significant than the topics for which they are now preparing grants, writing reports, ‘building capacity’ and ‘teaching-theteachers.’ As someone who works within the medical school of a research-intensive university, I understand as well as anyone the pressure to obtain funding for scholarly research. Nonetheless, I am concerned that the financial carrots offered by government agencies, companies and philanthropic foundations are being consumed by scholars who know there are more important topics deserving their attention. Ten years ago, the big money in bioethics was in the study of ethical, legal and social issues at the end-of-life. We will see how many ELSI and GELS researchers continue to ply their trade when the next big thing comes along. Neuroethics, anyone?
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
F O O L’ S C O R N E R
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Why biotech don’t pay dividends — yet Tom Jacobs, of the Internet site Motley Fool (http://www.fool.com/), provides his angle on biotechnology investments. Read on and become “Foolishly” informed*. He can be contacted about biotechnology and investing at
[email protected]. Jacobs cannot give individual investment advice but welcomes any. Tom Jacobs owns shares of Meridian Bioscience.
The US financial news of the year is the Senate’s narrow vote in May to cut the tax on dividends. With dividends now taxable to individuals at a lower 15% rate, corporations that already pay taxes on their profits have a much weaker case for withholding cash. Investors in Japan, the United Kingdom and Sweden, for example, with higher dividend taxation, can only dream. And the same goes for biotech investors; of 366 companies that touch on the biotech and drugs universe and trade on US exchanges, only 23 pay a dividend. We wonder, where’s our piece of the dividend pie? Dividends 101 A dividend is essentially a cash return on your investment, commonly expressed in terms of yield or a percentage of the stock price. A company distributing $2 annually per share and trading at $50 has a 4% dividend yield. To pay the dividend, management gives up some percentage of the money generated by product sales, but left over after selling costs; operating expenses, like salaries and advertising; taxes; spending for property, plant and equipment; and any other extraordinary bills are paid. The more a company pays out, the less it has to invest in its business. Thus, to offer a dividend, management must believe that the money cannot be more profitably invested in a company’s own or other research and development—into chimeric monoclonal antibodies, for example—or in buying back shares (see Nat. Biotechnol. 21, 746, 2003). Dividends: good or bad? You might fault managers for their inability to find better investments for corporate profits, but academic research shows dividends account for half of the annual long-term gains for the S&P 500 (Standard & Poor’s Index that includes ∼70% of all publicly traded US companies)—8% to 11% depending on the source. And after a stock market boom and bust, some investors prefer the dependable dividends of stodgy businesses to the potential growth of speculative businesses, thank you very much. They invest in the likes of tobacco and food giant Altria Group (New York, NY, USA; NYSE:MO; 6.2% dividend) or automaker General Motors (Detroit, MI, USA; NYSE:GM; 4.9% dividend) that generate a lot of cash but don’t have great growth opportunities. Or in real estate investment trusts like Annaly Mortgage (New York, NY, USA; NYSE:NLY; 6.9% dividend) or business development compa-
nies like Allied Capital (Washington, DC, USA; NYSE:ALD; 9.3% dividend) that receive tax exempt status on the condition they pay out a high percentage of profits as dividends. Many giddily receive a steady 9.3%, rather than suffer the gyrations of the market. But nothing is sacred about a dividend. The more financially savvy pigs of George Orwell’s Animal Farm might observe that some dividends are ‘more equal than others.’ Not every dividend-paying company can always produce enough cash to meet its basic expenses, fund research and development for growth and save for a rainy day. Business conditions fluctuate, so a dividend may not only grow over time, but may be ruthlessly cut, harming the stock price. Take ailing big pharma Schering-Plough (Madison, NJ, USA; NYSE:SGP). In August, the company slashed its dividend 68% to conserve cash. This, and the concomitant earnings warning, lopped another 9% off the stock price. Most other big pharmas aren’t hurting and take the middle road. They return some profits to shareholders in small dividends and still invest for future growth (Table 1). Why no dividend? The obvious reason that few biotechs pay dividends is simple: few profits. But even where there are profits, younger growing companies must plow it all back into research and development. Older big pharmas have cash-eating research and development too, but they also sport stodgier consumer products that generate steady cash, but offer fewer opportunities to invest for growth. While cash machines Amgen and Genentech may today redeploy every penny into profitable recombinant DNA and monoclonal antibody research, would you bet against either offering over-the-counter products in 20 years—and then choosing to return some profits to shareholders as dividends? Biotechs want to grow up to be big pharmas, but as parents never tire of telling teenagers, “The privileges of adulthood bring responsibilities and burdens.” Some day, Amgen and Genentech will pay a dividend, however small, and others will follow suit when more profitable and mature. It’s not why we may choose to invest in them today, but investors won’t refuse a nice bonus when it comes.
* Nature Biotechnology does not guarantee the veracity, reliability, or completeness of any information provided on this page; it is not responsible for any errors or omissions or for any results obtained from the use of such information; it will not be liable for any loss, damage, or investment decision arising from a reader’s reliance on the information provided.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
Table 1 Top 5 yields for big pharma and biotech Company
Annual dividend dividend
Yield
Big pharma Bayer (Leverkusen, Germany; NYSE:BAY)
$1.01
Bristol-Myers Squibb (Princeton, NJ, USA: NYSE:BMY)
$1.12
4.7% 4.4%
Merck (Whitehouse Station, NJ, USA: NYSE:MRK)
$1.48
2.9%
GlaxoSmithKline (Brentford, UK, USA; NYSE:GSK)
$1.15
2.7%
Eli Lilly (Indianapolis, IN, USA; NYSE:LLY)
$1.34
2.3%
Biotech Psychemedics (Cambridge, MA, USA: AMEX: PMD)
$0.32
4.0%
Meridian Bioscience (Cincinnati, OH, USA: NASDAQ:VIVO)
$0.36
3.6%
ICN Pharm (Costa Mesa, CA, USA; NYSE:ICN)
$0.31
1.8%
Sanofi-Synthelabo (Paris, France; NYSE:ADR)
$0.49
1.6%
Altana (Bad Homburg, Germany; NYSE:AAA)
$0.86
1.4%
Sources: Yahoo! Finance, AAII Stock Investor Professional.
1283
E R R AT A A N D C O R R I G E N D A
Erratum: Why biotech don’t pay dividends–yet Tom Jacobs Nat. Biotechnol. 21, 1283 (2003)
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
The title of this article contained a typographical error. The title should have read: “Why biotechs don’t pay dividends–yet”. Nature Biotechnology regrets the error.
Erratum: New biotech hubs may emerge as industry matures Paroma Basu Nat. Biotechnol. 21, 1123, 2003 The title of Table 1 incorrectly indicates the presence of data for 48 North American cities or counties. The original article, which appears in the News section of the Bioentrepreneur web portal (http://www.nature.com/bioent), does indeed contain these data. But the version reprinted here displays a truncated version of the table with ten data points: North American cities or counties that rank 1–5 and 43–48 in total annual operating costs for a biomedical research and development facility. Nature Biotechnology regrets the error.
Corrigendum: Invention and commercialization in optical bioimaging Daniel L. Farkas Nat. Biotechnol. 21, 1269–1271, 2003 The URL that appeared on p. 1271 was incorrect. The correct URL is http://www.ptei.org/educational_programs/Planetarium/ planetarium_project.html.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 12 DECEMBER 2003
1513
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
BOOK REVIEW
How to sell an industry Building Global Biobrands: Taking Biotechnology to Market by Françoise Simon and Philip Kotler Free Press, 2003 400 pp. hardcover, $35 ISBN 0-7432-2244-X
Reviewed by Donna Murray
Biotechnology and marketing are both enormously complex areas, and a book that attempts to combine both topics is never going to be an easy read. However, authors Françoise Simon and Philip Kotler have used their extensive experience to produce a book on biotech marketing that is both interesting and informative. Although the text does not shy away from topics such as nanotechnology and genetic engineering, the reader does not necessarily have to be familiar with the subjects to understand the issues discussed. A comprehensive glossary is included, which is invaluable for unfamiliar terms used in the text. Simon and Kotler state in their preface that they have focused primarily on the biopharmaceutical sector, and the book is indeed heavily biased toward biotechnology in the health-care industry. However, they argue that their findings are relevant to all industries involved with biotech, and it would seem possible to take their findings and apply them to marketing in other areas, such as genetically modified crops. The text covers a wide range of topics and marketing techniques, from alliances to consumer communications, and uses industry case studies to illustrate how companies responded—with varying degrees of success—to particular situations. The authors have obviously had a high level of access to the companies involved, and the case studies provide valuable insight into how companies behave and manage issues. The book is divided into three parts. Part one looks at the current state of biotechnology and at innovations, including bioinformatics, that are driving this sector forward. Part two looks at how market identity for biotech products can be built and maintained. Finally, part three looks at some of the challenges facing the biotech industry. Simon and Kotler state that biotechnology will be driven by the need to feed, clothe and shelter an ever-increasing world population. However, on reading the book, the overwhelming message seems to be one of profit, and it is perhaps naive to overlook the fact that the vast majority of the world’s population cannot pay for the things they need. The book raises the interesting question of whether biotech companies are so altruistic that they will undertake programs to feed, clothe and
Donna Murray is at the Scottish Institute for Enterprise, The University of Edinburgh, The King’s Buildings, Edinburgh EH9 3JL, UK. e-mail:
[email protected]
provide medicine for billions of people for little or no profit. From the evidence presented, there appears little to justify such a claim with regard to health issues. For example, although some companies offer their treatments free to low-income patients, most do not. The authors also indicate that companies are resistant to measures that would allow patients in poorer countries access to life-saving drugs and therapies. The recent debate over genetically modified (GM) food in the United Kingdom indicates that there is a high level of mistrust among the public about whether multinationals have the will to deliver the benefits of biotechnology to developing countries. As well as a thorough analysis of biopharma, the authors offer a very succinct analysis of why the first wave of GM organisms were widely rejected by consumers. They indicate that these products were seen as being ‘all risk, no benefit’ and that the companies involved simply did not give enough consideration to the public acceptance problem caused by this perception. Having worked in biotech community education projects, I saw this problem with the first generation, and it has caused second-generation crops to start out with an extremely poor public image. Initial GM traits in crops were seen as being purely for the benefit of the producers, yet the companies involved seemed to expect consumers to accept them with no problems. In addition, the beneficial aspects of the crops (for example, the reduction in the use of herbicides with herbicide-resistant crops) were never made clear to the public. When consumers see a clear benefit, there is evidence that their attitudes are less hostile. The first GM food in the UK was a tomato puree made from GM tomatoes that ripen more slowly. This resulted in a lower price, due to reduced wastage, and increased flavor. Sales of the product were good; however, it was pulled from supermarket shelves in 1999 as a result of the adverse publicity around other GM crops. Overcoming the public mistrust that has built up as a result of the mishandling of the first generation of GM crops will not be an easy task. Simon and Kotler obviously know that they are covering a huge topic, and that many readers will find it difficult to follow some of the findings they mention. Therefore, key points are repeated where relevant in different chapters. Given that the book focuses so much on biopharma, this is important if your interests are in other areas of biotech, as you will have to apply the findings to your own area of interest. This would seem to be the book’s most obvious failing. It would have been interesting to read more about the authors’ views on how their findings apply to nonpharma biotech. I was frequently intrigued by some of the technologies the authors claim are already being researched or are on their way to market. The technologies discussed range from genetically engineered spiders producing nonsticky silk—up to ten times as strong as steel—to the latest health-care developments. Such information makes the book interesting and relevant for nonspecialists. At other points, the book focuses so heavily on the US medical system that one’s interest level can wane, making those chapters drag. Finally, the book opens with a clichéd paragraph about family life in the future. This is in no way an indication of the quality of the work. I will definitely be recommending this book to young biotech companies as a reference manual. Many MBA students who have an undergraduate science degree will also benefit from such a specialist marketing book.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1285
The rise of European venture capital for biotechnology Michael Howell, Melanie Trull & Mark D Dibner Changes in economic institutions and the entrepreneurial climate, together with a growth in venture capital funds, have greatly increased European biotechnology venture formation in recent years.
Michael Howell is at the Curriculum in Neurobiology, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina 27599, USA. Melanie Trull and Mark Dibner are at BioAbility, LLC, PO Box 14569, Research Triangle Park, NC 27709, USA. e-mail:
[email protected]
100 90
US 30%
80 US 55%
70
Canada 9%
60 50 Canada 14%
40
Europe 48%
30 Europe 17%
20
Asia/Pacific 15%
Asia/Pacific 12%
Global public companies
Global private companies
10 0
Bob Crimi
The European biotechnology industry is now ranked as the largest outside the United States. Europe boasts some 1,800 biotechnology companies, which compares with roughly the same number in the United States (ref. 1; http://www.bioability.com/us_biotech_companies.htm). The industry has shown steady growth in Europe, with the number of biotechnology companies doubling since 1997, and now employs over 82,000 workers, compared to 162,000 employed by US biotechnology firms1. At the heart of Europe’s astronomical growth in biotechnology over the past five years is the availability of venture capital (VC). This article is based on surveys of US and European VC firms carried out in mid to late 2002 by the biotechnology consulting company BioAbility (Research Triangle Park, NC, USA). For the purposes of this study, VC funds were examined in four distinct stages of their life cycles: fundraising, currently investing, fully invested and liquidating (see Box 1 for definition of terms). Furthermore, VC investment in biotechnology was examined in six distinct fundraising stages: seed funding, rounds A/B, rounds C/D, mezzanine, bridge financing and buyout (see Box 1). Our data show that European VCs are filling the gap in funding for laterstage, pre–initial public offering (IPO) companies, but start-ups may still have trouble finding funding at the seed and early rounds of funding.
Percentage
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
F E AT U R E
Figure 1 Geographical distribution of public and private biotechnology companies around the world. (Adapted from ref. 9.)
Opportunities in Europe Fifteen years ago, US biotechnology companies were flooded with money, while European companies struggled to find financing. This was largely due to a lack of European stock markets that would list these young, entrepreneurial firms. With a lack of ‘exit strategy’ possibilities for investors in European biotechnology firms, venture capitalists were hesitant, if not unwilling, to invest in European firms. For many years, the number of European biotechnology firms remained low, at 20–25% of the number of firms in the United States. However, restrictions for listing on the London Stock Exchange eased in 1995, and 1996 brought the development of the Nouveau Marché in Paris, France and of the EASDAQ (the European equivalent of the NASDAQ), today called
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
NASDAQ Europe (London, UK and Brussels, Belgium), the majority of which is now owned by NASDAQ (New York, NY, USA). The rate of new company formation in Europe has soared and VC investment in European firms has grown tremendously, to such an extent that there are now more private biotechnologycompanies in Europe than in the United States (see Fig. 1). Many groups around the world see investment in European biotechnology as an attractive opportunity. The attractiveness of the European biotechnology market comes from the knowledge that the drivers of the industry are sound and long lasting2. Like the US population, the population of western Europe is steadily aging, and demand for age-related pharmaceuticals and therapeutics will undoubtedly be on the rise. Adding to the demand is the
1287
F E AT U R E
Box 1 Stages of financing
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Stages of a VC fund At any given time, VC funds fall into one of the following four categories: Fundraising. The fund has been set up and is raising money from investors of many sorts, including qualified individuals, companies, pension funds and others. Currently investing. The VC fund is investing in companies. At this stage, the fundraising has typically closed, or certain minimum amounts may be met and the fund may be raising the final, additional funds. Fully invested. The fund is no longer investing in additional companies. Some funds may still be available for follow-on or bridge financing of existing investments. Liquidating. The VC fund is returning the investment, equity and profit to its various partners, according to the original partnership agreement. Once liquidated, the fund no longer exists.
Stages of a company Likewise, the maturity of a company can be described in terms of the stage of financing that it is receiving: Seed stage. A small initial financing, typically under $1 million, to validate a concept, get the company started and complete the initial business plan. This financing round could be in the form of a straight equity investment, convertible preferred equity, convertible debt or a combination. Warrants to purchase additional shares of stock at a later time and under certain conditions would usually be included. Investors could be individual qualified investors (called ‘angels’), organized groups of angels, or venture capitalists. However, at this stage angel investors are involved far more frequently than venture capitalists.
European Union’s annexation of the former Eastern Bloc countries, causing the potential market size for biotechnology products in Europe to expand by another 125 million people. Unmet needs in disease treatment and prevention, combined with constant new technology and scientific advancement, provide a continual stream of newly emerging markets and prospects for industry growth. The 6th Framework Programme (FP6) of Research for the European Community places great emphasis on the growth of new technology and commercialization, especially related to the biosciences. Of the €17.5 billion budgeted for the years 2002–2006, €2.2 billion are earmarked for biotechnology in human health, and another 20% or more of the funds could be biotechnology related, at least in part. Further, there is an understanding that small and medium enterprises (SMEs) play a crucial role in European competitiveness and job creation that is considered highly important by the European Commission, with 15% of FP6 dedicated to the support of SMEs. (http://www.efbweb. org/FP6/fp6a.htm).
1288
Series A/B. One or two early rounds for $1–5 million (Series A) and $6–10 million (Series B). This is typically VC money, but could come from private or other investors such as pension funds. With the first (Series A) round, the founders’ shares are typically diluted out by about a half. Often, venture capitalists join together to form a syndicate, each putting in a piece of this round, and one venture capitalist acts as the lead investor. Typically this investor negotiates the terms of the round and is afforded a board of directors seat (or can designate a seat). Each subsequent round generally requires at least one new bona fide investor to lead the round and value the enterprise. Often, one or more representative of the lead investor group in a round will also be placed on the board of directors. Series C/D. To take the company through product development and on to an IPO, another $15–50 million may be needed in one or two rounds of financing. Often at this stage smaller funds cannot particip-ate and are significantly diluted because larger amounts are required. Mezzanine financing. After some validation such as a partnership to provide credibility, this should be the last financing before an anticipated IPO or some other liquidity event. Mezzanine financing also serves to help justify IPO valuation and gives another benchmark to the share price before the IPO. Bridge financing. Not desired by investors after previous rounds, this short-term and very costly (to the biotechnology company) VC funding is provided to a company that is in dire need for cash. Buyout. The purchase of a company by a VC firm or investor group, after which the incumbent and/or incoming management will be given or acquire a large stake in the business. MDD, MT and MH
Also bolstering biotechnology industry growth in Europe is the plight of large pharmaceutical companies. These companies are less effective innovators than biotechnology firms and their in-house drug discovery programs are consistently failing to provide a sufficient supply of new drug products that can counteract the expiration of patents on blockbusters and address the expanding overheads and sustainable growth of the giant companies formed through consolidation in the past few years3. As large pharmaceutical companies realize that they can’t do everything in house—they are spending more money on R&D, yet putting fewer drugs into the pipeline—biotechnology companies help fill the need for innovation. Biotechnology venture capital financing The growth of the European biotechnology industry is due in large part to the parallel growth of the European VC industry. Government-sponsored small business initiatives, as exemplified by the German BioRegio initiative (see Box 2), also account for many start-ups. This expansion is reflected by the
increasing amount of funds being raised by venture capitalists for investment in the life sciences overall. According to the National Venture Capital Association (Arlington, VA, USA), total VC investment in the life sciences in Europe rose from $200 million in 1996 to $2.6 billion in 2000. This 13-fold change greatly overshadows the 2.5-fold growth in US VC investment over the same period, from $3.1 billion to $7.5 billion4. The primary origin of the funds raised by European VC firms is domestic, with the proportion of funds raised from sources outside of Europe decreasing. In 2001, 34.6% of the $46.3 billion raised for VC in Europe came from non-European sources, including 25.1% from the United States. In 2002, with only $31.8 billion raised, 28.9% came from outside of Europe, with 18.5% from the United States (http://www.evca.com/images/attachments/tm pl_13_art_35_att_305.pdf). Despite overall optimism for the industry, biotechnology companies were not immune to the global economic downturn, and the public markets reflected a loss of investors. In 2001, the European biotechnology sector
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
European VC profile In the second half of 2002, BioAbility surveyed 231 European VC firms about their investment practices in biotech. Of those firms, 117 responded, providing information on 157 funds. The majority of the venture capitalists responding to our survey remain bullish on the sector, despite the difficult market conditions. Eighty percent of our
Fundraising current
$204 $102 $165
Investing current
$230 $85
Investing complete
$122
Europe US
$40
Liquidating
$108 $0
$50
$100
$150
$200
Average fund size (Millions)
$250
Bob Crimi
began the year with an $80 billion market cap (stock price times the total number of shares outstanding); at the end of 2002, the market cap was down to $22 billion. The IPO window in Europe slammed shut in 2001, just as it did for US high-tech companies; 39 European biotechnology companies went public in 2000 versus just 5 in 2001 and 2 in 2002. European IPOs raised $3 billion in 2000 but only a meager $24 million in 2002 (ref. 1). The rapid rise in the number of European biotechnology companies in the late 1990s followed by the dramatic downturn in the public markets in 2000 have left many companies in precarious financial positions. In the absence of public funds, the role of VC financing gets elevated from important to critical. Last year, European biotechnology companies continued to rely on VC as their primary source of financing. In 2000, biotechnology firms raised $1.2 billion; in 2001, as the markets were bottoming, they raised a record $1.4 billion from venture capital1. Total money flowing into biotechnology companies from VC for 2002 was $1.2 billion5. Consider the difference between public and private financing for biotechnology companies in two short years. In 2000, VC investments were less than half (40%) the amount of public financing; in 2002, VC investment was 45 times as large as public investment! The commitment of venture capitalists to sustain biotechnology companies through the downturn of the economy is reflected in the size of deals. Deal size has been increasing steadily over the past six years; in 1997, the average amount of a deal was $7.6 million per European company; in 2000 that number averaged $17 million6. More companies are landing larger deals: in 1998, only three companies raised more than $20 million, whereas in 2001, 23 companies managed to do so1. Venture capitalists are willing to finance companies that have put their IPO plans on hold. They know that biotechnology companies are cash intensive and that the time frame for product development could be as long as ten years. And, unlike the downturn in the early 1990s, venture capitalists are confident they can raise the money necessary to sustain their businesses6.
Fund life cycle
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
F E AT U R E
Figure 2 Average size of funds in the four stages of the fund life cycle—European versus US VC funds. Source: BioAbility Survey, 2002.
European respondents said they are likely to invest in biotechnology in the future; almost all expect to invest at equal or higher levels compared to the past. The funds range greatly in size, from the Apax Partners’ (London, UK) Apax Europe V fund at $4.4 billion, with $800 million earmarked for biotech, to the $300,000 PME Capital (Porto, Portugal) fund with only $48,000 invested in biotech. From our surveys of US and European VC firms, the sizes of European funds containing biotechnology concerns appear comparable to the sizes of US funds7. The funds that had current investments averaged $165 million in size, slightly smaller than the $230 million average for US funds in the same life cycle (Fig. 2). Funds where investing is complete were smaller on both continents. European funds averaged just $85 million, whereas US funds were an average of $122 million. Our data suggest that fundraising efforts for European biotechnology funds may be outpacing their US counterparts, at least as viewed in terms of fund size. The European funds with current fundraising activity averaged $200 million, double the size of US funds that are actively fundraising. One possible difference between US and European venture capitalists is that US venture capitalists appear much more likely to liquidate portfolio companies they feel are underperforming, rather than continuing financing8. Comparison between European and US funds from our surveys shows that US funds liquidated more than twice the amount of investments that European funds did, although investment lev-
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
els were comparable in their current funds (Fig. 2). Our data show that, overall, European venture capitalists who invest in biotechnology companies finance companies at stages almost identical to US firms (Fig. 3). Half the funds were invested in seed-stage companies and almost 70% have investments at the round A/B stage (for definitions, see Box 1). Despite similar averages across Europe, large differences exist in fund sizes and investment strategies between individual countries. The distribution of VC funds in Europe is illustrated in Figure 4. The UK, which has the largest and best-established biotechnology and VC industries, also had the largest fund sizes, averaging $153 million. Germany, on the rise in biotechnology since 1997, had fund sizes averaging just over $100 million (see Box 2). France had the third largest average, at just over $80 million per fund. The investment strategies for European venture capitalists also differ by region. Interestingly, respondents from two Nordic countries, Finland and Denmark, had all of their funds completely dedicated to biotechnology investments. Traditionally, Finland venture capitalists invest up to 50% in seed and early-stage companies. In the UK, biotechnology companies in their funds made up closer to half of the companies The UK also has very little seed-stage investment. UK venture capitalists invest almost 90% in expansion stages, with almost no investment at the seed stage. This trend toward late-stage investing has recently been reflected across the continent.
1289
F E AT U R E
Although the European biotechnology industry is large and growing larger, individual countries represent very different stages of biotechnology growth and VC progress. Germany provides a good example of how directed changes in governmental policy and commitment can lead to flourishing biotechnology and VC industries. Before the 1990s, Germany had an extensive research base but little interest in commercializing its technology. Public perception of genetic engineering was strongly negative, and scientists in general had a lack of entrepreneurial inclination. This began to change when the German government developed a favorable infrastructure for expansion in the area of biotechnology. In 1996, the German Minister for Science and Technology launched the BioRegio contest to promote the commercialization of biotechnology. BioRegio was a government initiative that promoted the development of biotechnology ‘clusters’, with the winning ‘model’ regions receiving $25 million over five years. At the same time, Germany set up a program called “Risk Capital for Small Technology Companies”— Beteiligungskapital fur kleine Technologieunternehmen (BTU)—which began subsidizing small high-tech ventures and guaranteeing private capital investors a portion of their investment for up to five years. The program matched VC funds, doubling their investment. This created a favorable investment environment, and VC industry grew exponentially. Our 2002 survey showed 75 VC firms in Germany that invest all or partly in biotechnology. Compare this rise to the United Kingdom, which has older and more established biotechnology and VC industries but only 43 sites that invest in biotechnology (Fig. 4). One reason for Germany’s growth is the change in public opinion. There is an increasing perception that biotechnology is a central component of the modern European economy. By 1999, Germany overtook the United Kingdom in total number of biotechnology firms. Europe’s largest economy wants to be the leader in biotechnology. MDD, MT and MH
The economic climate in the past two years has influenced the financing strategies of biotechnology venture capitalists away from the seed stage toward later stages—in 2001, $1 billion of the $1.4 billion invested went to later-stage companies9. More money is going into investments in pre-existing businesses
and less into new start-ups. The result has been that seed-stage companies are having difficulty getting funding in both the United States and Europe. However, in both regions life science companies fare better than those in most other sectors in garnering seed-stage investments10.
54% 47%
Seed
68% 75%
Series A/B
50% 44%
Series C/D
20% 18%
Mezzanine Bridge
Buyout 0%
19% 14%
Europe
9% 8% 10% 20%
US
30%
40%
50%
60%
70%
80%
Bob Crimi
Financing stage
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Box 2 How things have changed
Figure 3 Percentage of VC funds that invest in each financing stage. This is a snapshot of investment strategy for European and US VCs investing in biotechnology companies. This graph combines data from all fund stages: fundraising, investing, investing complete and liquidating. Note that funds may invest at multiple stages, so the data will add up to more than 100%. Source: BioAbility Survey 2002.
1290
The less-than-robust economy of 2002–2003 has also influenced the type of companies getting funding. There has been an increase in funding of later-stage companies that can market therapeutics, or are preparing to do so. Often these are spun out of large pharmaceutical houses. European venture capitalists are shying away from drug discovery start-ups, which need more money and take longer to develop their products4. Challenges The European VC industry is less well developed than its US equivalent, and only really picked up steam in the late 1990s. Europe’s fledgling biotechnology industry was financed by its fledgling VC industry and both were hurt in the severe economic downturn beginning in 2000. The European VC industry appears to be consolidating, with several of the newest and smallest firms disappearing. Some believe there are more VC firms and portfolio companies than the current economy can support, and a shakeout is expected9,11. Just as the creation of European stock markets helped biotechnology firms, the failure of public exchanges leaves public and private companies in limbo. After only five years, Germany closed its Neuer Markt in September 2002 after it had lost 96% of its value in two years. Switzerland also closed its SWX New Market after only three years, citing poor returns (http://www.swx.com/news/ media/media20020723a_en.html). Fragmentation of stock markets all over Europe creates a lack of exposure for biotechnology firms to sophisticated investors. Lack of a truly unified equity market, such as the London Stock Exchange, can cause some of them to move overseas to the United States’ NASDAQ1. Lack of an exit on public markets for VCs means that they must invest more money in existing companies to keep them alive. Although VC commitment to current portfolio companies is laudable, start-up and earlystage companies may not survive or even be funded due to a strain on available venture resources—not because the science is not viable or the market not attractive. Not surprisingly, investments in European private equity are smaller and are growing less quickly than that in the United States. Europe invested $23 billion in private equity in 2001, compared with the United States’ investment of $220 billion. A few years ago, European VC commitments were half the size of US investments; in 2000, they fell to one-fourth the US size9. Another reason for the disparity between US and European investments in private
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
F E AT U R E
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Iceland 4 Israel 3 Turkey 1
Figure 4 Where the money is: number of European VC sites with biotechnology companies in their portfolios. Within each country, VC sites are clustered around large urban centers such as London, Paris, Copenhagen and Helsinki. Perhaps because of the sheer number of sites and the BioRegio competition, Germany has many clusters of VC sites, including Munich, Heidelberg and Frankfurt-amMain. Source: BioAbility Survey 2002.
Sweden 11
Finland 5
Norway 7
Denmark 11
United Kingdom Netherlands 43 Poland 7 1 Germany Ireland 75 3 Belgium Czech Republic 8 Luxembourg 1 2 Austria Hungary 6 Switzerland 1 27 France Slovenia 16 1 Spain 1
Italy 3
Portugal 2
Greece 5
Cyprus 1
equity is tax laws relating to VC funds and portfolio companies. A report from the European Venture Capital Association (Brussels, Belgium) states that the fragmentation of the tax and legal environments in the European Union still hinders the development of entrepreneurial growth companies by venture capitalists in member countries (http://www.evca.com/images/attachments/ tmpl_13_art_30_att_270.pdf)12. The United Kingdom, in response to lobbying from the private equity industry, changed its tax codes, reducing capital gains taxes and ending a 20person limit on participants in a fund13. Venture capitalists still want more transparent tax laws and a lifting of limitations on cross-country investments1. Also affecting the VC industry is the general climate for supporting the biotechnology companies they create. For years, many Europeans had a cultural bias against entrepreneurship and the risks it represents, opting instead for more traditional careers. Many scientists cared little for commercializing their technology, and institutions lacked sup-
change tax laws for the benefit of private equity and encourage commercialization of technology developed at universities. Compared with the results of our study of the US biotechnology industry7 and the VC being raised to fund it, Europe has had a much higher growth rate over the past few years. The European biotechnology product pipeline continues to grow, with 50 products now in phase 3 trials. The number of biotechnology firms in Europe has skyrocketed. And European VC has followed suit. However, because they are newer, Europe’s biotechnology firms are small and are at the stages where they need additional funding to grow or even stay alive. Although the VC funds are out there for many of these companies, the firms would benefit more from a turnaround in the global economy. Europe, like the United States, has seen growth of biotechnology VC in 2001 and 2002. However, there appears to be a lessening of available VC in the United States in the first half of 2003 (ref. 7), and it is yet to be determined whether the same will hold true for Europe.
port organizations such as technology transfer offices to facilitate the development of commercially viable research discoveries. Tax burdens and bias against entrepreneurship caused many biotechnology companies to move to the United States, in turn, giving the European biotechnology industry a lagging start and allowing the US industry to dwarf Europe’s. Governments are now both responding to tax issues and fostering general entrepreneurial development by creating policies similar to Germany’s in the 1990s. France is budgeting $60 million for a contest to identify innovative companies that use academic research. In 2002, Ireland announced a $25 million fund for biotechnology companies through a government agency designed to promote cooperation between academia and industry. Conclusions There is cause for much optimism in the European biotechnology and VC industries. European governments are making concerted efforts to engender entrepreneurial spirit,
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1. Crocker, G. Endurance: Ernst and Young’s European Biotechnology Report, 10th Anniversary Edition (Ernst and Young, London, 2003) 2. Alt Assets. Delivering on Discovery: Private Equity Investing in Biotechnology. (AltAssets, London, UK; 2002) 3. Initiative Europe Ltd. Venture Capital Yearbook 2002 (Initiative Europe Ltd, Redhill, UK, 2003). 4. Subacchi, P. Eur. Venture Capital J. 1 November 2001. http://www.privateequityweek.com/evcj/protected/sectorreps/industry/ZZZ8E6NLITC.html 5. Birnbaum, M. European Life Sciences 2002: Deals, Sources, and Investors. (Windmill Reports, Amsterdam, 2003). 6. Paisner, G. Europe’s biotech VCs are waiting out the downturn—by making larger investments. Red Herring Magazine, 11 December 2001. http://www. redherring.com/Article.aspx?f=articles%2farchive%2fma g%2fissue108%2f973.xml 7. Dibner, M., Trull, M. & Howell, M. US venture capital for biotechnology. Nat. Biotechnol. 21, 613–617 (2003). 8. Essijck, K. European Venture Capitalists Kept Wallets Closed in 2001. Special to the Wall Street Journal. http://www.mysql.com/articles/inpress_wsj_01042002. html 9. Ernst & Young. Beyond Borders: The Global Biotechnology Report 2002 (Ernst & Young, New York, 2002). 10. Innovation in a harsh market. European Venture Capital Journal, 7 May 2003. http://www.ventureeconomics.com/evcj/protected/ctryreps/1047652019313. html 11. AltAssets. After the Goldrush: A Survey of European Venture Capital Firms. (AltAssets, London, 2002). 12. European Venture Capital Association. European Private Equity and Venture Capital Industry Hampered by the Fragmentation of EU Country’s Tax and Legal Environments. Press release, March 31, 2003. 13. Country Reports: UK-British Venture Capital Association, European Venture Capital Journal. March 3, 2003. (available at http://www.privateequityweek. com/evcj/protected/ctryreps/1045005237159.html)
1291
Engineering better cytokines Lawrence Steinman Designing enzymes with protective shells that are cleaved once they arrive at inflammatory sites can be used to enhance immune therapy. A drug that homes in on a diseased tissue and then becomes activated at that site, reversing the pathological process but causing minimal collateral damage, might fit the requirements for Paul Ehrlich’s ‘magic bullet.’ In this issue, Chernajovsky and colleagues1 report just such an approach with β-interferon, a cytokine with modest, though proven, therapeutic efficacy in some forms of human autoimmune disease, such as multiple sclerosis2. Their redesigned cytokine has an increased half-life, and more importantly, it becomes activated by inflammatory mediators at the site of disease, increasing its therapeutic efficacy in an animal model of rheumatoid arthritis. This work emphasizes the promise of using molecular techniques to optimize cytokines by activating them only where needed and reducing their systemic side effects. Cytokines—potent chemicals used in intercellular communication—are in some cases sheathed in protective covers, called latency peptides, that shield them from degradation and allow them to act where needed once the cover is removed. β-interferon does not normally have a protective cover, so Chernajovsky and colleagues borrowed a ‘latency-associated peptide’ from a different cytokine, transforming growth factor-β1, and spliced it to the gene encoding β-interferon (Fig. 1). The peptide was ingeniously engineered to contain cleavage sites for a metalloprotease (Fig. 1), an enzyme found at the site of inflammation in the joint lining in rheumatoid arthritis3 and at the site of inflammatory brain lesions in multiple sclerosis4. The engineered cytokine was shown to have a very long half-life, on the order of 55 hours,
Lawrence Steinman is in the Department of Neurological Sciences, Stanford University, Stanford, California 94305, USA. e-mail:
[email protected]
Protective shell Cleavage site
Cytokine
Address Cleavage site
Inflammatory enzyme cuts protective shell off the cytokine Inflammatory enzymes Cytokine receptor Mailbox for address
Disease site Erin Boyle
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS AND VIEWS
about 40 times longer than the normal half-life of injected β-interferon. In addition, fluids from the sites of inflammation, including synovial fluid from individuals with rheumatoid arthritis and cerebrospinal fluid from individuals with multiple sclerosis, activated the latent β-interferon by cleaving the sheathed cytokine and releasing the latency peptide. A longer half-life would permit the administration of lower doses of cytokine, and activating a cytokine only in the compartment where the disease is occurring would reduce the systemic side effects of the treatment. For β-interferon therapy in multiple sclerosis, a year’s cost is approximately $10,000 in the United States, and so a longer-acting drug might reduce expenses considerably. Flu-like symptoms accompanying dosing with β-interferon are a frequent complaint of individuals with multiple sclerosis who are taking this medication. If the drug is activated only at the site of disease, then such side effects might be reduced.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
Figure 1 A protecting shell, often called a latency peptide, is spliced onto a gene encoding a cytokine. The shell is attached by genetic engineering to the cytokine by bridges encoding sites that can be cleaved by enzymes. The sites are chosen so that the enzyme necessary for cleavage is located at the site of inflammation. At the disease site, the cytokine is unsheathed and activated. Other embellishments can be added to cytokines by bioengineering, including addresses, such as motifs that recognize adhesion molecules that are characteristically found at certain inflammatory sites.
Chernajovsky and colleagues also demonstrate the therapeutic efficacy of their engineered cytokine. Previously, they had developed techniques for delivering β-interferon genes with therapeutic effects in animal models of multiple sclerosis and rheumatoid arthritis5. Here, they delivered DNA plasmid constructs encoding the latent β-interferon and demonstrated reversal of paw swelling in already established arthritis after delivering a single intramuscular injection of DNA encoding the latent cytokine. This animal model has served as a good testing ground for new therapies in rheumatoid arthritis. In addition to engineering cytokines to have longer half-lives and to become activated at sites of inflammatory disease, other exciting approaches for improving these essential molecules are under investigation. Garren and colleagues6 successfully combined a DNA construct encoding the cytokine interleukin 4 and its signal sequence with another DNA con-
1293
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS AND VIEWS struct including a myelin gene to drive an immune response in the direction of suppressing autoimmunity to myelin. Others have used immune constructs to target cytokines to specific addresses, delivering the cytokine to the correct ‘mailbox’ where it will have maximal effect7. The use of ‘immunocytokines’ and other novel cytokine constructs should enhance the efficacy of these natural products in fighting disease. Other enzymatic cleavage sites on the cytokine engineered by Chernajovsky and colleagues include substrates of thrombospondin and transglutaminase. Enzymes like transglutaminase are more abundant at sites of neurodegeneration8. Thus, cytokine therapy might be enhanced for treatment of diseases like Alzheimer and Huntington disease associated with increased transglutaminase8. A motif in the engineered β-interferon construct encoding the peptide sequence arginine, glycine, aspartate (RGD) facilitates interaction with integrins. This provides additional capabilities to target adhesion molecules and extracellular matrix proteins9. One of the possible negative aspects of extensively engineering any natural molecule is the risk that the newly engineered structures will appear sufficiently different to the immune system to elicit limiting antibody responses or allergies. Even small modifica-
tions to accommodate an enzymatic cleavage site or to allow recognition by a crucial target could alter a self-molecule sufficiently to raise potential problems. Despite these caveats, optimizing enzymes provides exciting new opportunities in biotechnology. The use of protein engineering to create new receptor constructs for tumor necrosis factor-α or to engineer chimeric monoclonal antibodies against this cytokine has revolutionized the treatment of autoimmune diseases, such as rheumatoid arthritis. This feat was acknowledged in this year’s Lasker Prize for Medical Research to Feldmann and Maini10. Ever more spectacular uses of these natural molecules and their antagonists will continue to add to our arsenal against disease. 1. Adams, G., Vessillier, S., Dreja, H. & Chernajovsky, Y. Nat. Biotechnol. 21, 1314–1320 (2003). 2. Jacobs, L. et al. Arch. Neurol. 44, 589–595 (1987). 3. van Meurs, J. et al. Arth. Rheum. 42, 2074–2084 (1999). 4. Gijbels, K., Galardy, R. & Steinman, L. J. Clin. Invest. 94, 2177–2182 (1994). 5. Croxford, J.L. et al. J. Immunol. 160, 5181–5187 (1998). 6. Garren, H. et al. Immunity 15, 15–22 (2001). 7. Lode, H.N., Xiang, R., Becker, J.C., Gillies, S.D. & Risfeld, R.A. Pharmacol. Ther. 80, 277–292 (1998). 8. Karpuj, M.V. et al. Nat. Med. 8, 143–149 (2002). 9. von Adrian, U. & Engelhart, B. N. Engl. J. Med. 348, 68–73 (2003). 10. Feldmann, M. & Maini, R.N. Nat. Med. 9, 1245–1250 (2003).
Sequence of a symbiont Valerie M Williamson & Harry K Kaya The complete genome sequence of the insect pathogen and nematode symbiont Photorhabdus luminescens identifies a trove of antibiotic and toxin genes. The bacterium P. luminescens is pathogenic to a wide range of insects1. It is transmitted to the body cavity of an insect host by the nematode Heterorhabditis bacteriophora, an organism with which it has a complex symbiotic relationship2. The ability of P. luminescens to switch from symbiont in the nematode gut to virulent pathogen in the insect host—a process in which the microorganism unleashes a veritable trove of toxins to fight off microbial com-
Valerie M. Williamson and Harry K. Kaya are in the Department of Nematology, University of California, Davis, One Shields Ave., Davis, California 95616, USA. e-mail:
[email protected]
1294
petitors and invertebrate scavengers—has made it a compelling candidate for genome sequencing. In this issue, Kunst et al.3 report the genome sequence of P. luminescens, providing a first comprehensive look into how this specialized microbe controls its different life stages and identifying several genes that could potentially be exploited in new approaches to fight insect pests. The complex lifecycle of P. luminescens is summarized in Figure 1. Once introduced into an insect by infectious H. bacteriophora juveniles (Fig. 1), the bacterium produces a battery of toxins that kill the host within 48 hours. Hydrolysis of the insect’s body by P. luminescens produces a rich food source for the bacterium, which in turn becomes a food
source for the nematode. After one to three generations, depending on insect host size, nematodes in the infective juvenile stage leave the depleted insect carcass in search of a new host, carrying P. luminescens in their gut. Like Bacillus thuringiensis, P. luminescens and its carrier nematode H. bacteriophora have been exploited for their insect-killing properties as a biocontrol agent against insect pests4. The nematode-bacterium complex is most effective against a number of soil-inhabiting insects, but its high production costs limit its application to high-value niche markets. Moreover, poor shelf life, lack of persistence in the soil and the necessity of high application rates have constrained large-scale use in agricultural settings. The DNA sequence of the P. luminescens genome has exceeded expectations as more putative toxin-encoding genes have been found than in any other bacterial genome so far examined. A large number of these genes have been identified as encoding putative toxins based on similarities to other known bacterial toxins. P. luminescens had previously been shown to carry a number of toxin complex genes encoding high-molecular-weight secreted proteins that are toxic to Manduca sexta, the tobacco hornworm5, and additional toxin complex genes have been discovered in the genome sequence of P. luminescens. Another Photorhabdus protein encoded by the gene mcf (makes caterpillars floppy) triggers apoptosis in insect hemocytes and midgut epithelium6. The P. luminescens genome sequence has further identified two homologs of insect juvenile hormone esterases3. Such proteins would be predicted to cause inappropriate development of the insect host and thus may represent a novel insecticidal strategy. Indeed, the products of these genes, when expresssed in Escherichia coli and fed to mosquitoes and caterpillars cause death within 48 hours. Thus, the presence of multiple types of proteins with diverse modes of toxicity indicates that P. luminescens has evolved or acquired a wide range of strategies leading to the rapid death of a range of insects after infection1. The P. luminescens genome sequence has identified several categories of genes that help explain its success in leading such a specialized lifestyle. Sequences encoding polyketide and nonribosomal peptide synthases probably synthesize antibiotics to protect against microbial competitors. Genes encoding antimicrobial colicin-like factors and immunity proteins have also been identified. Genes encoding secreted enzymes predicted to have a role in bioconversion and metabolism of the insect cadaver have been identified, as well as transcriptional regulators of their expression. The
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
NEWS AND VIEWS
Nematode infective juveniles penetrate into the insect host and release bacteria. Bacteria produce toxins that kill the host, and secrete enzymes that hydrolize the cadaver.
INSECT PAT H N GE O
BI
ON
T
Nematode infective juveniles with bacteria in their guts emerge from depleted cadaver.
M NE D O FO
Nematodes reproduce inside insect cadaver, feeding on bacteria and degraded host tissues.
Erin Boyle
M
P. luminescens LIFE CYCLE
A SO T O D UR E CE
Y M AT O D E S
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NE
Infective juveniles carrying bacteria search for and find new hosts
Bacteria reproduce, feeding on rich resources of cadaver.
Figure 1 Symbiont or pathogen? During its complex life cycle, P. luminescens rotates between being a nematode symbiont, an insect pathogen and a nematode food source.
bacterium forms an intimate association with the gut of the nematode, and a large number of genes with potential roles in this symbiotic interaction have been identified, including those encoding proteins that probably function in surface adhesion. The availability of the complete sequence of the P. luminescens genome thus represents an important milestone toward exploiting the potential of this bacterium as a biocontrol agent. Access to new gene sequences encoding potential protein toxins has further implications for bioengineering insect resistance in plants. Transgenic plants expressing B. thuringiensis genes are among the most successful and widely applied biotechnological products7. In some instances, transgenic B. thuringiensis crops have had a large impact on yield and have resulted in less pesticide use. But there is concern that insect resistance to B. thuringiensis toxins in transgenic plants, arising from changes in insect populations, will reduce the effectiveness of this toxin and its transgenic products. In addition, B. thuringiensis toxin proteins are generally effective against a narrow range of insects, and toxins have not been identified or developed against some insect pests. Several of the predicted toxin proteins in P. luminescens have been shown to have oral toxicity, but toxicity upon expression in transgenic plants has not yet been reported. The best-characterized toxic proteins from P. luminescens are large, and their expression in plants may be problematic as was the case initially with B. thuringiensis peptides8.
Perhaps the most fascinating story yet to be told from the analysis of the P. luminescence genome is how this organism came to acquire the genes that allow it to fill its specialized niche so successfully. Comparison with the
genomes of related bacteria indicates that extensive horizontal gene transfer has occurred. For example, Yersinia pestis, a flea-colonizing bacterium and the causal agent of plague, is a close relative. Other clues to the evolution of the P. luminescens genome are provided by the multitude of pathogenicity islands, phage remains and abundant transposable elements found in the P. luminescens genome. Further genomic analyses will provide answers to such questions as how and why homologs of an insect juvenile hormone esterase gene were incorporated into the genome of P. luminescens and how P. luminescens implements and regulates all its different insecticidal capabilities. The answers will provide the tools to exploit this organism’s capabilities to fight insect pests in new, untested ways. 1. ffrench-Constant, R. et al. FEMS Microbiol. Rev. 26, 433–456 (2003). 2. Forst, S. & Clarke, D. in Entomopathogenic Nematology (ed. Gaugler, R.) 35–56 (CABI Publishing, Oxon, UK, 2002). 3. Duchaud, E. et al. Nat. Biotechnol. 21, 1307–1313 (2003). 4. Kaya, H.K. & Gaugler, R. Annu. Rev. Entomol. 38,181–206 (1993). 5. Bowen, D. et al. Science 280, 2129–2132 (1998). 6. Daborn, P.J. et al. Proc. Natl. Acad. Sci. USA 99, 10742–10747 (2002). 7. Sheldon, A.M., Zhao, J.-Z. & Roush, R.T. Annu. Rev. Entomol. 47, 845–881 (2002). 8. Estruch, J.J. et al. Nat. Biotechnol. 15, 137–141 (1997).
Reconstructing genetic networks in yeast Zhaolei Zhang & Mark Gerstein By combining data from gene expression and DNA-binding experiments, a computational algorithm identifies the genetic regulatory network in yeast. A central challenge in genomic biology is to determine how cells coordinate the expression of thousands of genes throughout their life cycle or in response to external stimuli, such as nutrients or pheromones. In eukaryotes, gene expression is modulated by various transcription factors that bind to the promoter regions, and different combinations of transcription factors may alternatively activate or repress
Zhaolei Zhang and Mark Gerstein are at the Department of Molecular Biophysics and Biochemistry, Yale University, 266 Whitney Avenue, New Haven, Connecticut 06520-8114, USA. e-mail:
[email protected]
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
gene expression. This is analogous to an electronic circuit, in which components are switched on and off by a network of transistors. In this issue, Bar-Joseph and colleagues1 report a computational approach to show that in yeast, genes are indeed regulated in networks that are controlled by groups of transcription factors. Furthermore, they show that these regulatory networks also have a modular structure in which groups of genes under the control of the same regulators tend to behave similarly. Genetic regulation and its mechanisms have been investigated since the days of Jacob and Monod and the discovery of the lac operon. Traditionally, such studies are labor-intensive and gene-specific and often require years of
1295
NEWS AND VIEWS
AAAAA AAAAA
Genome-wide gene expression analysis
Genome-wide Ch-IP location analysis
AAAAA
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Gene expression profiles
Transcription factor binding p-values TF1 TF2 TF3 TF4 TF5 TF6 TF7
a b c d e
a
f g h i
Core gene set [a,b,c,d,e] Re-calculate combined p-value
≥ p1 ≥ p1 ≥ p1 ≥ p1 ≥ p1 ≥ p1 ≥ p1
≥ p1
Determine core gene set, calculate center expression profile
Combined p-values
a b c d e f g h i
Final gene module G = [a,b,c,d,e,f,g] as regulated by transcription factors F= [TF1, TF2, TF3, TF4]
≥ p1 ≥ p1
Figure 1 Schematic describing the GRAM algorithm. The data from gene expression and ChIP-chip experiments are presented on the top as stacked expression profiles and a P value table, respectively. In the P value table, the confidence values that are less than the strict threshold (p1) are colored red. In the ChIP-chip experiments, P values were calculated for each spot on the microarray to represent the confidence value (the smaller the P value, the more likely the observed DNA binding is real)5. The GRAM algorithm first selects a ‘core set’ of genes that share a common group of transcription factors and also have similar expression profiles. In this example, the core set consists of genes a, b, c, d and e but not f and g because only the first five genes have P values strictly less than p1 for the subset of regulators TF1, TF2, TF3 and TF4. A center expression profile is then computed from this core set of genes. The algorithm then revisits the P value table to recompute a combined P value for every gene with respect to the subset of regulators. A gene is added to the selected set if its expression profile is close to the center expression profile and the combined P value is less than p1. The final selected set of genes is exported as a gene module. The above procedures are repeated for every possible combination of transcription factors in yeast to derive the complete regulatory network.
bench work. More recently, and with the complete genome sequences of a number of eukaryotic organisms becoming available, several high-throughput genomic technologies have been developed, which allow biologists to study gene expression and gene regulation on a whole-genome scale. The concept of determining gene expression at the whole-genome level was first introduced by Brown and colleagues2, who developed DNA microarrays to measure the expression level for every gene in yeast simultaneously.
1296
In recent years, several groups have also implemented chromatin immunoprecipitation (ChIP) DNA chips for directly mapping the in vivo physical interactions between transcription factors and their DNA binding sites3–6. Briefly, a cell line expressing a specific tagged transcription factor is constructed. After growth under experimental conditions, DNA fragments bound to the tagged transcription factors are recovered by a ChIP assay and hybridized to DNA microarrays containing the complete set of the yeast intergenic
regions. Strong hybridization in a region proximal to a gene would indicate transcription factor binding to that gene’s promoter site. Many researchers have attempted to apply statistical or computational approaches to reconstruct genetic regulatory networks based on data sets derived from these whole-genome methodologies. Most of the approaches have consisted of applying clustering algorithms to gene expression data to identify coexpressed genes, which are surmised to be coregulated by shared transcription factors7. Such approaches have also been expanded to incorporate previous knowledge about the genes, such as cellular functions or promoter sequence motifs8,9. These methods have achieved various levels of success, but an intrinsic limitation is their overreliance on expression data, which represent the result rather than the cause of genetic regulation. In addition, some of these methods assume that expression levels are correlated between the transcription factors and the genes that they regulate. This has been proven not always to be true10. Other computational methods have also been developed to extract regulatory information from whole-genome DNA-binding data sets5,6. The rationale behind these approaches is that if two genes share a common set of transcription factors, then they are probably coregulated and belong to the same gene module. Using this location-based approach, researchers have successfully identified some basic regulatory motifs in the yeast network. But this approach has its own limitations. First, location information does not indicate whether the nature of the regulation is in the positive or negative direction; second, DNAprotein interaction data are noisy owing to much nonspecific binding. As reported in the present paper, Bar-Joseph et al. improved previous algorithms incorporating both DNA-binding data and gene expression data. Their new algorithm, called GRAM (genetic regulatory modules), works in three steps, as shown in Figure 1. As described in their paper1, the authors reconstructed a yeast rich media regulatory network using DNA-binding data from 106 transcription factors and over 500 gene expression data sets. The final regulatory network contains 655 distinct genes partitioned into 106 modules, and 68 transcription factors are placed in the network representing regulatory hubs (see Figure 1 in original paper; ref. 1). They carried out gene-specific ChIP experiments to verify a number of selected regulatory interactions predicted by GRAM. The power of GRAM is evident in the fact that 40% of the 1,560 unique regulatory interactions it identifies in yeast would not have
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS AND VIEWS been detected using only the DNA-binding data. Another advantage of the combined approach is that it can also predict directionality of the edges in the network; that is, it can be inferred whether the genes in a module are upregulated or downregulated by examining their expression correlations. An important benefit of having a complete genetic network of an organism is its potential to provide clues on a gene’s role in, for example, signal transduction pathways and thereby identify its interaction partners. It is accepted that genes in the same network module generally have similar cellular functions. This has also been observed among network modules generated by GRAM. Notably, the authors found that in most cases in which a gene module is regulated by more than one transcription factor, previous evidence could always be found suggesting potential physical or functional interactions between these transcription factors. All these observations prove that the regulatory networks produced by GRAM are biologically relevant and promise to serve as a blueprint to direct future experiments. Like microarrays in the late 1990s, it is almost certain that the new ChIP-chip technology will quickly catch on with researchers worldwide, and before long, hundreds of genome-wide DNA-binding data sets will be available. Powerful and sophisticated computer algorithms, such as GRAM, will be needed to analyze these data. Finally, many other research avenues can be pursued. For example, these tools can be applied to determine the degree of conservation of modular network structures or regulatory interactions among closely related species, such as Saccharomyces cerevisiae and Schizosaccharomyces pombe. This type of comparative analysis can potentially shed light on the evolution of regulatory networks. Also, the current knowledge on genetic networks does not paint a truly dynamic picture of the processes taking place inside a cell. Existing technologies and algorithms, such as GRAM, are the first steps toward the development of tools capable of capturing the dynamics of genetic regulatory networks. 1. Bar-Joseph, Z. et al. Nat. Biotechnol. 21, 1337–1342 (2003). 2. Chu, S. et al. Science 282, 699–705 (1998). 3. Ren, B. et al. Science 290, 2306–2309 (2000). 4. Iyer, V.R. et al. Nature 409, 533–538 (2001). 5. Horak, C.E. et al. Genes Dev. 16, 3017–3033 (2002). 6. Lee, T.I. et al. Science 298, 799–804 (2002). 7. Qian, J., Dolled-Filhart, M., Lin, J., Yu, H. & Gerstein, M. J. Mol. Biol. 314, 1053–1066 (2001). 8. Ihmels, J. et al. Nat. Genet. 31, 370–377 (2002). 9. Pilpel, Y., Sudarsanam, P. & Church, G.M. Nat. Genet. 29, 153–159 (2001). 10. Yu, H., Luscombe, N.M., Qian, J. & Gerstein, M. Trends Genet. 19, 422–427 (2003).
Playing tag with the yeast proteome Brenda J Andrews, Gary D Bader & Charles Boone Two tagged proteome studies offer the most intimate and detailed view into the inner works of yeast cells to date. Proteomics—the study of the complement of expressed cellular proteins (or proteome)— has catapulted to the forefront of biological research. This advance is due to the development of enabling technologies for producing large-scale data sets of protein activities and to the increasing number of annotated genome sequences that can serve as prerequisite proteome ‘blueprints’. Pioneering methods for analysis of the proteome have been developed in yeast and have relied on the systematic cloning of open reading frames (ORFs) for subsequent expression or generation of genomic sets of strains expressing tagged proteins suitable for a variety of array-based manipulations. In two recent Nature papers, the Weissman and O’Shea groups1,2 report two notable additions to the arsenal of tools available for the comprehensive analysis of gene and protein function in yeast. The authors describe two collections of yeast strains in which each ORF is fused with affinity or fluorescence tags, thereby providing the most comprehensive and sensitive view yet of the expressed proteome and its subcellular location in a eukaryotic cell. In the past few years, myriad genetic and biochemical methods have been used to query genomic sets of proteins for biochemical activity and protein-protein interactions. Notable landmarks on the road to the functional description of the yeast proteome include large-scale two-hybrid screens, immunoprecipitation– mass spectrometric analysis of protein complexes and the generation of tagged sets of pro-
Brenda J. Andrews is at the Department of Medical Genetics & Microbiology, and Charles Boone is at the Department of Medical Genetics & Microbiology and the Banting and Best Department of Medical Research, University of Toronto, Toronto, Ontario M5S 1A8, Canada. Gary D. Bader is at the Computational Biology Center, Memorial Sloan-Kettering Cancer Center, Box 460, 1275 York Ave., New York, New York 10021, USA. e-mail:
[email protected];
[email protected]
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
teins for production of functional protein chips (reviewed in ref. 3). The generation of protein complex interaction maps and functional surveys of proteins for DNA binding and other activities are providing a rich, but relatively static, view of the yeast ‘interactome’. A more complete ‘cell biological’ view of the proteome will emerge from integration of proteomics information with functional genomics data derived from transcriptional profiling and gene disruption projects, as well as a picture of the subcellular distribution of proteins and their relative abundance. In a tour-de-force of strain construction, Ghaemmaghami et al.1 used a PCR-based homologous recombination strategy to insert a tandem affinity purification (TAP) tag at the C termini of all predicted yeast ORFs. They reasoned that an explanation of the biological properties of the proteome would require not only a description of macromolecular complexes and their subcellular location, but also an experimental description of the expressed proteome and a reasonable measure of the absolute levels of proteins in the cell. Two features of the strain collection allow both a survey of expressed proteins in a particular physiological circumstance and a measure of their cellular abundance. First, the tagged proteins are expressed from their native promoters in their endogenous chromosomal location and should be responsive to normal regulatory circuitry. Second, each ORF is marked with a common tag allowing measurement of the absolute abundance of each protein using quantitative western-blot analyses (see http://yeastgfp.ucsf.edu/). A sensible set of test cases suggests that the regulation and activity of most yeast proteins is unperturbed by the C-terminal tag, which bodes well for the utility of the strain set in future genetic and cell biological studies and is good news for the many other projects that have used convenient tags to study gene and protein function. The authors were able to successfully TAPtag 6,109 of the 6,243 predicted ORFs and observed a protein product for 4,251 or 70% of the tagged proteome in log-phase yeast cells grown in optimal laboratory conditions1. A
1297
NEWS AND VIEWS
Protein-protein interaction
Protein abundance per protein Low
High Average/compartment
Bud Transcription factor-gene regulation
Arp2/3 complex
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
TRAPP complex
Pheromone response pathway
Golgi Cytoplasm ER
TOM complex
Mitochondrion
Proteasome
Exosome complex
Nucleolus Nucleus RNA Pol II holoenzyme
APC complex
Figure 1 Integrated view of localization and protein abundance data1,2 with protein-protein and gene regulation interactions in the context of selected complexes, pathways and the cell. Cellular compartments are colored by average protein abundance, with light colors representing compartments with low protein abundance (and dark colors those with high abundance). Black lines represent proteinprotein interactions and red arrows point from transcription factors to the genes they regulate. Proteins are represented by circles colored by protein abundance continuously from blue to red indicating low to high abundance. White proteins have no abundance information. The TRAPP complex localizes to the ‘endoplasmic reticulum (ER) to Golgi’. The transport outer membrane (TOM) complex localizes to the mitochondrion. The Arp2/3 complex localizes to cortical actin patches and the cytoplasm. The proteasome, anaphase promoting complex (APC) and RNA polymerase II (Pol II) complexes mostly contain subunits that localize to the nucleus, but some subunits localize to the cytoplasm (depicted as spanning both compartments). The pheromone-response mitogen-activated protein kinase cascade spans the cell from the surface to the nucleus. It contains a cell-surface receptor and other cortical components, cytoplasmic signaling molecules and nuclear transcription factor effectors, which control the expression of genes encoding components of the pathway as part of a regulatory circuit (red arrows).
subset of proteins were only seen with a second GFP-tagged strain set (see below), and the combined data show that ∼80% of the proteome is expressed in happily growing yeast cells. This experiment considerably augments efforts to view the proteome using mass spectrometry and two-dimensional gel electrophoresis4 and provides the most comprehensive and sensitive view, so far, of the expressed proteome in a eukaryotic cell. The analysis also confirms that the range of protein expression in the cell is massive, from 50 to well over 1,000,000 molecules per cell, although an even more sensitive assay may find lower abundance proteins. The observed protein set can be used as an experimental validation of the existence of hypothetical genes and, in combination with comparative genomics information, provides a powerful means of correcting errors in gene annotation (see http://www. yeastgenome.org/chromosomeupdates/start_cha nges.shtml). In a parallel proteomics project, Huh et al.2 used an identical strategy to generate a green fluorescent protein (GFP)-tagged yeast strain
1298
collection that provides both a second powerful experimental resource for the yeast community and the first view of the native yeast proteome in living cells. The first proteomescale analysis of protein localization involved a description of the cellular location of almost half of yeast proteins using plasmid-based overexpression of epitope-tagged proteins and genome-wide transposon mutagenesis for high-throughput immunolocalization of tagged gene products5. This study affirmed the correlation between protein function and subcellular environment and highlighted the importance of generating a high-resolution and comprehensive view of protein localization. Huh et al. analyzed 6,029 strains with GFPtagged ORFs and found that three-quarters of the proteome and over two-thirds of the previously unlocalized proteins had a detectable GFP fluorescence signal in log-phase cells. In a first pass, the GFP patterns were classified as having one or more of 12 rather broad subcellular localization patterns, such as cell periphery, nucleus, mitochondrion and cytoskeleton.
A second round of colocalization experiments, using monomeric red fluorescent protein fusions to reference proteins of known localization, distinguished another 11 localization categories, including the nucleolus and spindle pole body. Several criteria suggest that this impressive two-stage binning of GFP patterns produced a high-quality view of the GFP-proteome that approximates the real situation in the cell. First, the results agree substantially with Saccharomyces Genome Database (http://www.yeastgenome.org/) annotations for the localization of ∼2,500 yeast proteins and with the previous large-scale study that examined the proteome using immunofluorescence5. Second, over 90% of proteins identified were also found within the set detected by western-blot analysis of the TAP tag collection, indicating that GFP fluorescence can be used to detect a broad range of protein expression levels. And third, this comprehensive GFP data set encompasses a set of organellar proteomics projects that aim to identify subsets of proteins in various organelles. For example, 164 nucleolar proteins were identified, 82 of which overlapped with the 127 proteins catalogued in Saccharomyces Genome Database and 82 of which were newly defined. Because many of the characterized nucleolar proteins are involved in ‘ribosomal RNA expression and processing’ and ‘ribosomal biogenesis’, most of the newly localized proteins would be expected to participate in some aspect of these processes. Given the apparent high quality of the localization data, researchers interested in the function and regulation of the nucleolus ought to get busy. The authors note that proteins with crucial C-terminal targeting signals are often mislocalized in this study and new fusions will have to be constructed to get an accurate view of the subcellular location of this group of proteins. All GFP localization information has been admirably recorded in a public database (http://yeastgfp.ucsf.edu), making the data easily accessible for further analysis. Integration of different functional genomics data sets enables biological hypotheses to be formulated with increasing levels of confidence. In an effort to leverage the biological information in their data set, Huh et al. analyzed their subcellular localization data in light of transcriptional coregulation information and combined physical and genetic interaction data sets. Both types of integrative analysis provided useful insight. In the first bioinformatics exercise, the authors investigated whether transcriptional coregulation is related to subcellular localization. To do this, they calculated the relative representation of proteins with a given localization for 33 general transcriptional
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS AND VIEWS modules, defined previously from an analysis of over 1,000 microarray data sets6. Notably, statistically significant enrichment was seen for 19 of the 22 most highly expressed modules, indicating that colocalization is highly correlated with transcriptional coexpression. Deeper analysis of the data can provide information on biological function that could not be gleaned from analysis of either data set alone. In a second computational analysis, the authors examined the relationship between colocalization and physical or genetic interaction. The relative enrichment for colocalization was assessed for the combination of protein-protein or genetic interactions in the GRID database (http://biodata.mshri.on.ca/ grid/), a collection of information derived from existing databases and large-scale data sets. As expected, because both genetic and physical interactions are indicative of a functional relationship, they were highly enriched between proteins that colocalize. An enrichment of interactions was also observed for protein pairs showing distinct localization categories, such as microtubule and spindle pole body. This illustrates the potential of this approach for identifying the network of functional relationships between subcellular local-
izations, a network that may reflect a dynamic interchange of proteins between compartments or genetic buffering of compartments. With the unveiling of these two new tools, researchers are in the privileged position of having a comprehensive description of yeast that includes positive identification of nearly all of its genes, many proteins categorized by their interactions in complexes, and, of course, data on the abundance and location of most known proteins. The generation of a global in vivo view of the yeast proteome means that we can start to assemble diagrams of its cellular pathways and complexes with unprecedented detail. For example, using colocalization and abundance data together with existing interaction data, we can overlay the architecture of complexes and signaling pathways with specific cellular compartments and environments (Fig. 1). Next, these strain sets can be used to move from a relatively static view of the proteome to an analysis of the dynamic abundance and localization of proteins during developmental programs or in response to environmental and genetic insults. For example, the GFP-tagged strains could be grown as high-density arrays in solid medium and assayed for colony fluo-
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
rescence to monitor global changes in protein levels in response to drug treatments7. Highthroughput strain construction methods8 will also allow introduction of the entire GFP- or TAP-tagged proteome into any genetic background. In this way, the genetic requirements for protein localization and protein complex formation can be systematically assessed for pathways and proteins of interest. The combination of the GFP- and TAP-tagged strain set in a matrix format would create a doubly tagged set that should allow proteome-wide coimmunoprecipitation tests as a sensitive means for assessing protein-protein interactions globally and in defined genetic backgrounds. These tools open the door for numerous other focused and large-scale analyses of the yeast proteome. 1. Ghaemmaghami, S. et al. Nature 425, 737–741 (2003). 2. Huh, W.-K. et al. Nature 425, 686–691 (2003). 3. Phizicky, E.M., Bastiaens, P.I.H., Zhu, H., Snyder, M. & Fields, S. Nature 422, 208–215 (2003). 4. Washburn, M.P., Wolters, D. & Yates, J.R Nat. Biotechnol. 19, 242–247 (2001). 5. Kumar, A. et al. Genes Dev. 16, 707–719 (2002). 6. Ihmels, J. et al. Nat. Genet. 31, 370–377 (2002). 7. Dimster-Denk, D. et al. J. Lipid Res. 40, 850–860 (1999). 8. Tong, A.H.Y. et al. Science 294, 2364–2368 (2001).
1299
RESEARCH NOTES
Green dwarfs Dwarf plants are largely credited with powering the green revolution of the 1960’s because short, stocky crops withstand adverse wind and rain conditions and can be grown at high densities. Now Johal and coworkers have solved the mysteries behind dwarf varieties of corn and sorghum. They found that Brachytic2 (br2) mutants of corn, which had been around since the 1950s, have altered auxin transport in the lower stalk, which is caused by a mutation in a P-glycoprotein transporter. These br2 plants not only have short internode distances, but also have more cells per unit area, resulting in stronger stalks. The notoriously unstable sorghum variant (dwarf3, dw3) has a duplication in the analogous transporter gene, which could explain its instability. Chromosomal analysis of dw3 mutants shows that unequal crossing over between the duplicated genes can result in some plants with three copies of the gene and others with one, suggesting that stable strains of dwarf sorghum could be isolated for use in parts of Africa where sorghum is a staple. (Science 302, 81–84, 2003) LD
Using automated, structure-based computational design, Hellinga and colleagues now report the successful reengineering of a PBP that binds ribose into a PBP that binds zinc. Taking into account the primary and secondary coordination spheres necessary for optimal zinc binding, the researchers modified a total of eight different amino acids, generating zinc-binding proteins of comparable, if not better, binding characteristics than the original ribose-binding protein (RBP). Escherichia coli containing the zinc-binding proteins together with the RBPassociated signal transduction pathway were capable of detecting zinc in solution. (Proc. Natl. Acad. Sci. USA 100, 11255–11260, 2003) GTO
Finding T cells in a haystack A new method identifies and isolates rare, tumor-killing T cells from patients for use in T-cell cancer therapy, a cancer treatment that is more specific and potentially less toxic than traditional chemotherapeutics. Current methods for measuring antigenspecific T-cell responses do not provide a way to link specificity with tumor-cell killing, and some do not allow further analysis or isolation of the cells. To overcome these obstacles, Lee and colleagues use flow cytometric analysis to isolate antigen-specific T cells expressing a marker that is upregulated on the surface of T cells during the process of cell killing (CD107a). They identify rare, tumor antigen–specific T cells that can destroy tumor targets from individuals vaccinated with tumor antigen and grow these cells in culture. A significant advantage of the method is that it is not necessary to know the tumor antigen target of the T cells to perform the assay—a requirement of most current assays. This is important because only a small number of tumor antigens have been identified to date. (Nat. Med. 9, 1377–1382, 2003) MS
Reengineering biosensors Synthetic cellular signal transduction pathways sensitive to metals could have applications ranging from metal detection in the environment to gene expression control during bioprocessing. Triggering of a signal by an extracellular molecule depends on the specificity and sensitivity of the periplasmic binding proteins (PBPs) associated with a particular signal transduction pathway.
Research Notes written by Kathy Aschheim, Laura DeFrancesco, Meeghan Sinclair and Gaspar Taroncher-Oldenburg.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
Double rats! The rat—an important animal model for human physiology and disease— has been successfully cloned for the first time. This rodent had previously proven recalcitrant to cloning efforts, mostly due to a quirk in the development of rat eggs, which start multiplying shortly after removal from the animal and thus do not allow sufficient time to perform somatic cell nuclear transfer (SCNT). Renard and colleagues have overcome this limitation with a simple and elegant approach in which a drug is added that prevents cell division of the rat eggs in culture, buying researchers enough time to perform SCNT. The authors produced two healthy pups, which in turn generated litters of healthy second-generation pups. The technology paves the way for the production of rats in which specific genes can now be manipulated to generate models relevant to medical research. (Science; published online 25 September 2003, doi:10.1126/science.1088313) GTO courtesy Science
Researchers have identified heart cells that meet most of the criteria for stem cells. The existence of heart stem cells contradicts a longstanding belief that the heart is a postmitotic organ with little capacity to regenerate itself after injury. Over the years, various clues—such as the discovery of replicating myocytes and of primitive heart cells expressing stem cell markers—have pointed to the error of this view. Now, Anversa and colleagues have identified rat cardiac Lin– c-kit+ cells that are selfrenewing and multipotent, with the ability to differentiate in vivo into myocytes, smooth muscle cells and endothelial cells. When transplanted into ischemic rat hearts, the Lin– c-kit+ cells produced new, functional myocardium. The existence of heart stem cells suggests novel therapeutic approaches to heart disease such as the development of drugs that could stimulate latent endogenous repair mechanisms. (Cell 114, 763–776, 2003) KA courtesy Cell
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Heart stem cells unmasked
1301
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
FOCUS ON OPTICAL IMAGING
B R I E F C O M M U N I C AT I O N S
Immunofluorescence stimulated emission depletion microscopy Marcus Dyba, Stefan Jakobs & Stefan W Hell We report immunofluorescence imaging with a spatial resolution well beyond the diffraction limit. An axial resolution of ∼50 nm, corresponding to 1/16 of the irradiation wavelength of 793 nm, is achieved by stimulated emission depletion through opposing lenses. We have demonstrated not only that an antibody-tagged label is stable enough to be recorded in this microscopy mode, but also that subdiffraction resolution can be obtained using a standard immunofluorescence preparation. Stimulated emission depletion (STED) microscopy, which relies on saturated depletion of excited molecules at the periphery of a scanning focal spot, has fundamentally overcome the diffraction barrier in farfield fluorescence imaging1,2. In theory, this method allows the spot to be sharpened without limit, but in practice, the effective spot size is determined by the central minimum of the STED beam and the maximum intensity that the fluorophore can withstand. Exploiting the narrow minimum of destructively interfering waves, one particular STED microscopy mode, STED-4Pi microscopy, has attained axial (z) resolutions of 33 nm and 46 nm for oil- and water-immersion lenses, respectively3. However, all STED images of biological samples reported to date have been of cells with labeled membranes. Given the dependence of the resolution on the tolerable intensity and the possibility that the lipid membrane is photoprotective to the dye, it has remained unclear whether subdiffraction resolution can be obtained under other labeling conditions. Here we show that a resolution beyond the diffraction barrier can be achieved in the STED-4Pi microscopy mode with standard immunofluorescence labeling of an intracellular structure. In contrast to previous work, the dye used can be covalently linked to antibodies and is stable under aqueous conditions. For our studies we chose the microtubular network of a human embryonic kidney (HEK) cell, which is otherwise difficult to resolve by light microscopy. Immunofluorescence decoration was performed according to standard procedures4. Cells grown on a coverslip were fixed with methanol and their microtubules were labeled with a mouse anti-β-tubulin monoclonal antibody (Boehringer Mannheim). For the secondary antibody, the red-emitting dye MR-121SE (provided by K.H. Drexhage, Department of Chemistry, University of Siegen, Germany) was coupled to an anti–mouse IgG (Jackson ImmunoResearch Laboratories) via its succinimidyl ester. The cells were mounted between two coverslips in a buffer solution (PBS: 137 mM NaCl, 2.7 mM KCl, 4.3 mM Na2HPO4, 1.4 mM KH2PO4,
pH 7.3). Figure 1a sketches the resulting dimensions for a decorated microtubule and Figure 1b gives a large-scale confocal overview image of a labeled cell. As the refractive index of PBS virtually matches that of water, we employed water-immersion lenses of 1.2 numerical aperture (Leica Microsystems). STED-4Pi imaging was carried out as described in its initial report3. The fluorophore was excited at a wavelength of 532 nm with 10-ps pulses at 76 MHz with 10 µW average power for single photon absorption. Because the emission of the dye is centered around 700 nm, STED was carried out at a wavelength of 793 nm, with synchronized 107-ps pulses and at an average power of 5.2 mW per lens. A monomolecular layer of the same fluorophore was chemically bonded to the inner side of one of the coverslips. The image of this ultrathin layer is referred to as the z response. Its full-width half-maximum (FWHM) gives the axial resolution. First, we measured the z responses of the STED-4Pi microscope and of its standard confocal counterpart under cell-free conditions. Whereas the confocal z response revealed a FWHM of 770 nm, which is typical for water immersion, the STED-4Pi z response yielded a FWHM of 48 nm, representing a 16-fold improvement in axial resolution (see Supplementary Figure 1 online). With a cellular sample, the z response beneath the cell at the applied STED power exhibited a FWHM of 71 ± 2 nm and axial side-lobes of <25 % of the main peak resulting from the secondary interference side minima of the depletion spot3. The slightly larger FWHM (as compared with the cell-free measurement) most likely stems from aberrations induced by the cell. The z response was exploited for a linear deconvolution of the data, which removed the effect of the side-lobes in the image and slightly improved the resolution, by a factor of 1.35. To investigate the resolution obtained in immunofluorescence imaging, we recorded image sections of an arbitrarily selected cell in the direction of the optic axis (xz images). Figures 1c and 1d show a typical xz image pair of the same site in the cell produced by the confocal reference and the STED-4Pi mode, respectively. Figures 1e and 1f display the intensity profiles along the marked lines in the respective images. The STED-4Pi image is substantially clearer. Whereas in the confocal image all of the adjacent microtubules appear as a single blob, in the STED-4Pi image most of them appear as distinct objects. The fact that in Figure 1d the monomolecular layer appears as a straight line indicates that the interference of the counterpropagating depleting beams is not distorted by the cell. The representation of the monomolecular layer strikingly demonstrates the resolution. Blurred in the confocal reference, the layer appears in the STED-4Pi mode as a distinct line with an FWHM of 53 nm. Moreover, the subdiffraction image shows that the brightness of the layer is of the same order as that of the microtubules. In contrast, the larger focal spot of the confocal microscope integrates the fluorescence originating from several microtubules, thus giving an improper representation of the object
Department of NanoBiophotonics, Max Planck Institute for Biophysical Chemistry, 37070 Göttingen, Germany. Correspondence should be addressed to S.W.H. (
[email protected]). Published online 19 October 2003; doi:10.1038/nbt897
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1303
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
B R I E F C O M M U N I C AT I O N S a
b
c
d
e
Figure 1 Subdiffraction immunofluorescence imaging with STED-4Pi microscopy. (a) Sketch of typical dimensions of a microtubule decorated with a fluorescently labeled secondary antibody. (b) Overview image (xy) of the microtubular network of a HEK cell. (c,d) Standard confocal and STED-4Pi xz images recorded at the same site of the cell; the straight line to the left of the cell is due to a monomolecular fluorescent layer attached to the adjacent coverslip. In both images the pixel size was 95 nm × 9.8 nm in the x and z directions, respectively; the dwell time per pixel was 2 ms. Both directions are shown at the scale given in e and f. To remove sidelobe effects, the STED-4Pi image was linearly filtered as described in the text. (e,f) Corresponding profiles of the image data along the marked dashed lines quantify the improved axial resolution of the STED-4Pi microscopy mode (f) over the confocal benchmark (e). Peaks 1–3, resulting from microtubules, are broader than the response to the monolayer. Note the ability of the STED-4Pi microscope to distinguish adjacent features.
allowed us to determine the typical diameter of the decorated microtubule to be 60–70 nm, which matches well the values reported by electron microscopy4. In conclusion, we have demonstrated immunofluorescence imaging of an intracellullar structure with spatial resolution well beyond the diffraction limit. We expect that the resolution of the STED microscope can be further increased by optimizing the depletion level or by using dyes with shorter emission wavelengths. It should be possible to obtain an additional gain of up to 22% in absolute resolution by using glycerol or oil immersion rather than water immersion. Future improvements will also include fast beam scanning, parallelization and a similar improvement in the lateral direction5.
f
Note: Supplementary information is available on the Nature Biotechnology website. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
brightness and fluorophore distribution in the image. The comparison of Figure 1c and Figure 1d underscores the importance of subdiffraction resolution for quantitative microscopy. We measured the axial FWHM of about 50 intensity peaks representing the microtubules in our STED-4Pi images. Although the microtubules were primarily oriented perpendicular to the optical section, the values inevitably varied owing to slight variations in the orientation of the microtubules. 83% of the FWHM values were in the range of 64–96 nm, which is larger than the 53-nm optical resolution obtained for the monomolecular layer. Simple quadratic subtraction
1304
ACKNOWLEDGMENTS We thank Lars Kastrup for valuable discussions and the German Ministry for Research and Education (BMBF) for partial support. Received 19 August; accepted 3 October 2003 Published online at http://www.nature.com/naturebiotechnology/ 1. Hell, S.W. & Wichmann, J. Optics Lett. 19, 780–782 (1994). 2. Klar, T.A., Jakobs, S., Dyba, M., Egner, A. & Hell, S.W. Proc. Natl. Acad. Sci. USA 97, 8206–8210 (2000). 3. Dyba, M. & Hell, S.W. Phys. Rev. Lett. 88, 163901-1–163901-4 (2002). 4. Weber, K., Rathke, P.C. & Osborn, M. Proc. Natl. Acad. Sci. USA 75, 1820–1824 (1978). 5. Klar, T.A., Engel, E. & Hell, S.W. Phys. Rev. E 64, 066611–066619 (2001).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
B R I E F C O M M U N I C AT I O N S
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Human antibodies from immunized donors are protective against anthrax toxin in vivo Martha A Wild1, Hong Xin1, Toshiaki Maruyama1, Mary Jean Nolan1, Peter M Calveley1, John D Malone2, Mark R Wallace2 & Katherine S Bowdish1 A panel of Fabs that neutralize anthrax toxin in vitro was selected from libraries generated from human donors vaccinated against anthrax. At least two of these antibodies protect rats from anthrax intoxication in vivo. Fabs 83K7C and 63L1D bind with subnanomolar affinity to protective antigen (PA) 63, and Fab 63L1D neutralizes toxin substoichiometrically, inhibits lethal factor (LF) interaction with PA63 and binds to a conformational epitope formed by PA63. Inhalation anthrax is usually fatal if not identified early enough for antibiotics to be of use. Death occurs within days of the onset of acute symptoms. The causative agent is Bacillus anthracis, and the lethality is primarily due to the effects of the bacterial toxin. This toxin is constructed from nonlethal components: PA combined with LF and/or edema factor (EF). PA83 binds to cellular receptors where it is cleaved by a protease to PA63 (ref. 1), which assembles into a heptameric pore2. The pore can bind up to three units of LF, EF or both3. Endocytosis of this structure leads to the entry of LF and/or EF into the cytosol, where each factor exerts its toxic effects. Neutralizing anthrax-toxin activity could provide time for antibacterial agents or the immune system to clear an infection. The toxin could be neutralized at several stages during entry into the cell. PA83 antibodies might inhibit PA63 processing, pore assembly or binding to receptor, LF or EF. PA63 antibodies might inhibit LF or EF binding or their translocation into the cytosol. EF or LF antibodies could also inhibit pore binding. Antibodies are highly specific and effective, making them a logical choice for an anthrax therapeutic. Mouse monoclonal antibodies neutralize anthrax toxin in vivo in rats4,5, and polyclonal guinea pig antibodies provide guinea pigs with passive protection against anthrax infection6. Human or humanized antibodies are safe and well tolerated for many therapeutic purposes. High-affinity human antibodies to anthrax toxin should therefore have therapeutic value in human anthrax infections. To obtain therapeutically useful high-affinity human antibodies, we generated phage display libraries bearing Fabs derived from the bone marrow or blood of human donors immunized against anthrax (for this and other protocols see Supplementary Methods online). Donor sera were screened by enzyme-linked immunosorbent assay (ELISA) to determine reactivity to PA83 antigen (Supplementary Fig. 1 online). Two donors were used to generate four libraries, using the gamma chains from each paired with either kappa or lambda chains. Phage bearing antibody fragments from all four libraries were prepared and panned through four rounds of enrichment against PA83. Libraries from one of the two donors were separately panned against
purified, bound PA63 in the presence of soluble PA83 to block epitope sites shared on both molecules. Phage expressing Fabs from various panning rounds in all libraries were screened for reactivity to PA83 and/or PA63 by ELISA. Positive candidates were sequenced to identify a representative panel of candidates containing unique heavy- and light-chain combinations. Sequences of 20 panel candidates are available from GenBank (accession nos. AY315900–AY315939). Panel candidates were subcloned to remove the phage gene III fusion from the heavy chain. Fabs were purified and used for in vitro neutralization assays (Supplementary Methods online). Two of 3 anti-PA63 Fabs (Fig. 1a, E–G) and 14 of 16 anti-PA83 Fabs (Fig. 1a, H–X) neutralized the effects of the anthrax toxin by >80%. Selected Fabs were then assayed for neutralization activity using serial dilutions (Fig. 1b). All the anti-PA83 Fabs shown had 50% neutralization values close to equimolar with the concentration of PA83 used in the assay (4.8 nM). Anti-PA63 Fab 63L1D, however, had a substoichiometric 50% neutralization value. This suggests that 63L1D might be acting at the level of the heptameric pore, effectively neutralizing multiple PA83 molecules at once. Consistent with this hypothesis, 63L1D did not bind to denatured PA63 (or PA83) on western blots (Supplementary Fig. 1 online), suggesting binding to a conformational epitope. In contrast, 83K7C bound to both denatured PA63 and denatured PA83, suggesting binding to a linear epitope. To further define the neutralizing epitopes, we examined the ability of selected Fabs to neutralize toxin after PA83 had bound to the receptor (Fig. 1c). Only 63L1D fully neutralized toxin under these conditions, possibly by blocking LF from binding to the heptamer. An ELISA titration of four Fabs against PA83 and PA63 is shown in Figure 1d. Note that the signal for 63L1D at saturation binding to PA63 is about one-fourth of that for 83K7C. If 63L1D does bind to a conformational epitope on the heptamer, it may, like LF, be unable to bind stoichiometrically, resulting in a reduced signal relative to 83K7C. Consistent with this observation, 63L1D competed with LF for binding to the heptamer in competitive ELISA experiments (Supplementary Fig. 1d online). Kinetic analysis of Biacore surface plasmon resonance using the CLAMP program7 produced the following apparent affinities: 63L1D to PA63, 0.13 nM; 83K7C to PA63, 0.87 nM; and 83K7C to PA83, 3.67 nM (see Supplementary Fig. 2 online). In vivo testing of Fabs 83K7C and 63L1D against recombinant toxin challenge (Fig. 2) showed that they are protective. At 6 nmol/ 250 g rat, both Fabs protected animals fully; at 2 nmol/250 g rat, both Table 1 In vivo dosage studies of IgGs Group
IgG concentrationa (nmol/250 g rat)
Survivors/ total
Average min to death
PBS
–
4/4
Toxin only
–
0/4
Ig 83K7C
1.0
4/4
0.3
4/4
NA
0.1
0/4
101 ± 5.5
0.3
4/4
0.1
3/6
Ig 63L1D
NA 81 ± 6.0 NA
NA 67 ± 3.0
aFull-length IgG1 versions of each Fab were tested in rats at concentrations indicated. PBS group, negative controls not challenged with toxin; toxin-only group, controls receiving toxin but no Fab. NA, not applicable.
1Alexion Antibody Technologies, Inc., 3985 Sorrento Valley Blvd., San Diego, California 92121, USA. 2Naval Medical Center, 34800 Bob Wilson Drive, San Diego, California 92134, USA. Correspondence should be addressed to K.S.B. (
[email protected]).
Published online 12 October 2003; doi:10.1038/nbt891
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1305
B R I E F C O M M U N I C AT I O N S
80 60 40 20 0
120 100 80 60 40 20 0
Cell death (%)
c
100
Cell death (%)
a
63L1D 83L6R 83K7C 83L8E 83L8F 83K2H 83K7H Toxin
A B C D E F G H I J K L M N O P Q R S T U V W X
120
Protection (%)
80 60 40 20
3
2
1
0
–20 0.01
0.10
1.00
10.00
0 0.00001
100.00
0.001
0.1
10
µg/ml Fab
nM Fab
Figure 1 In vitro neutralization and ELISA. (a) Fab protection at 50 nM (see Supplementary Methods online for details of these experiments). A, 83L6R alone; B, 83L5B alone; C, toxin alone. All remaining samples include toxin and the indicated Fab. D, anti–hepatitis B; E, 63L1D; F, 63K3G; G, 63K9A; H, 83K2A; I, 83K2H; J, 83K3C; K, 83K3F; L, 83K3H; M, 83K7C; N, 83K7H; O, 83K9C; P, 83L2D; Q, 83L2E; R, 83L3B; S, 83L4E; T, 83L5B; U, 83L6R; V, 83L7D; W, 83L8E; X, 83L8F. (b) Percent protection for seven serially diluted Fabs. Average 50% neutralization values (mean nM ± s.d.) as determined by these assays were: 83K7C, 5.06 ± 1.25; 83L6R, 7.11 ± 2.68; 83L8E, 5.41 ± 2.67; 83L8F, 4.31 ± 2.26, 63L1D, 1.15 ± 0.57; 63K3G, 32.4 (not repeated 4 times); 63K9A did not neutralize. (c) In vitro neutralization after PA binding. PA83 was preincubated with cells for 2 h before addition of LF and Fab. (d) ELISA titration of Fabs. Open symbols designate reactivity to PA83; black symbols, reactivity to PA63. , 63L1D; , 83K7C; , 83L8E; , 83L6R.
protected from death, but 63L1D did not prevent symptoms. At 0.6 nmol/250 g rat, both 63L1D and 83K7C significantly delayed symptoms and death (63L1D, P = 1.43E-07 and P = 2.22E-05, respectively; 83K7C, P = 0.0005 and P = 0.038, respectively), with 63L1D showing longer delays. In vivo testing of full-length IgG1 molecules derived
Note added in proof: Apparent affinities for Ig 63L1D and Ig 83K7C are 8 and 25 pM against PA63, respectively, and 118 pM for Ig 83K7C against PA83.
Note: Supplementary information is available on the Nature Biotechnology website. ACKNOWLEDGMENTS Our special thanks go to the military personnel who voluntarily donated blood and bone marrow to this effort. We thank Arthur Friedlander for his gift of PA83 antigen, Janette Delgadillo for expert assistance, and David Myszka for timely Biacore analyses. The views expressed in this article are those of the authors and do not reflect the official policies or positions of the Departments of the Army, Navy, Defense or the US government.
180
COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
150
Time (min)
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
100
4
d
63K3G 63K9A 63L1D 83L2D 83L6R 83K7C 83L8E
Absorbance (405 nm)
b
from these Fabs (Table 1) showed that both antibodies protected fully at 0.3 nmol/250 g rat. Notably, at 0.1 nmol/250 g rat, Ig 63L1D fully protected three of six rats, with no appearance of symptoms, although three died more rapidly than controls that received toxin but not IgG. Rats receiving 0.1 nmol/250 g rat of Ig 83K7C died after a short delay relative to toxin-only control. These results are consistent with the in vitro results, which showed that Fab 63L1D neutralized substoichiometrically and was more potent than anti-PA83 antibodies. This work shows that human anti–anthrax toxin antibodies that have high affinity for, and potently neutralize, toxin in vivo can be isolated from immunized donors. Our data suggest that anthrax vaccination generates a robust immune response in humans. The two antibodies described may be therapeutically useful against anthrax infection or for passive protection of unvaccinated individuals at risk of exposure.
120
Received 18 April; accepted 21 July 2003 Published online at http://www.nature.com/naturebiotechnology/
90 60 30 0
Toxin
0.6 nm 83K7C
2 nm 83K7C
6 nm 83K7C
Toxin
0.6 nm 63L1D
2 nm 63L1D
6 nm 63L1D
Figure 2 In vivo dosage studies of Fabs. Experiments were done at Perry Scientific using protocols approved by their Institutional Animal Care and Use Committee. Two Fabs were tested in separate experiments. Negative controls injected with PBS from Fab dialysis all survived and are not shown. Gray bars, time to symptoms; black bars, minutes to death; black arrows, survival for 1 week without symptoms until killed.
1306
1. Klimpel, K.R., Molloy, S.S., Thomas, G. & Leppla, S.H. Proc. Natl. Acad. Sci. USA 89, 10277–10281 (1992). 2. Milne, J.C., Furlong, D., Hanna, P.C., Wall, J.S. & Collier, R.J. J. Biol. Chem. 269, 20607–20612 (1994). 3. Mogridge, J., Cunningham, K. & Collier, R.J. Biochemistry 41, 1079–1082 (2002). 4. Little, S.F., Leppla, S.H. & Cora, E. Infect. Immun. 56, 1807–1813 (1988). 5. Little, S.F., Leppla, S.H. & Friedlander, A.M. Infect. Immun. 58, 1606–1613 (1990). 6. Little, S.F., Ivins, B.E., Fellows, P.F. & Friedlander, A.M. Infect. Immun. 65, 5171–5175 (1997). 7. Myszka, D.G. & Morton, T.A. Trends. Biochem. Sci. 23, 149–150 (1998).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES
The genome sequence of the entomopathogenic bacterium Photorhabdus luminescens Eric Duchaud1, Christophe Rusniok1, Lionel Frangeul2, Carmen Buchrieser1, Alain Givaudan5, Séad Taourit1, Stéphanie Bocs6, Caroline Boursaux-Eude2, Michael Chandler7, Jean-François Charles3, Elie Dassa4, Richard Derose8, Sylviane Derzelle3, Georges Freyssinet8, Sophie Gaudriault5, Claudine Médigue6, Anne Lanois5, Kerrie Powell9, Patricia Siguier7, Rachel Vincent5, Vincent Wingate9, Mohamed Zouine1, Philippe Glaser1, Noël Boemare5, Antoine Danchin3 & Frank Kunst1 Photorhabdus luminescens is a symbiont of nematodes and a broad-spectrum insect pathogen. The complete genome sequence of strain TT01 is 5,688,987 base pairs (bp) long and contains 4,839 predicted protein-coding genes. Strikingly, it encodes a large number of adhesins, toxins, hemolysins, proteases and lipases, and contains a wide array of antibiotic synthesizing genes. These proteins are likely to play a role in the elimination of competitors, host colonization, invasion and bioconversion of the insect cadaver, making P. luminescens a promising model for the study of symbiosis and host-pathogen interactions. Comparison with the genomes of related bacteria reveals the acquisition of virulence factors by extensive horizontal transfer and provides clues about the evolution of an insect pathogen. Moreover, newly identified insecticidal proteins may be effective alternatives for the control of insect pests. Photorhabdus luminescens is an enterobacterium that is symbiotic with soil entomopathogenic nematodes and pathogenic to a wide range of insects. P. luminescens promotes its own transmission among susceptible insect populations using its nematode host as vector1. Its life cycle comprises a symbiotic stage in the nematode’s gut and a virulent stage in the insect larvae, which it kills through toxemia and septicemia. After the nematode attacks a prey insect and P. luminescens is released, the bacterium produces a wide variety of virulence factors ensuring rapid insect killing. Bioconversion of the insect cadaver by exoenzymes produced by the bacteria allows the bacteria to multiply and the nematode to reproduce. During this process P. luminescens produces antibiotics to prevent invasion of the insect cadaver by bacterial or fungal competitors. Finally, elimination of competitors allows P. luminescens and the nematode to reassociate specifically before leaving the insect cadaver2,3. To better understand this complex life style, we determined the genome sequence of P. luminescens subspecies laumondii strain TT014, a symbiont of the nematode Heterorhabditis bacteriophora isolated on Trinidad and Tobago. RESULTS General features Strain TT01 possesses a single circular chromosome of 5,688,987 bp with an average GC content of 42.8%. No plasmid replicon was found.
A total of 4,839 protein-coding genes, including 157 pseudogenes, seven complete sets (23S, 5S and 16S) of ribosomal RNA operons and 85 tRNA genes, were predicted (Fig. 1; Supplementary Table 1 online). Toxins against insects More toxin genes were predicted in the P. luminescens genome than in any other bacterial genome sequenced yet. A large number of these toxins may be involved in the killing of a wide variety of insects. Some may act synergistically or use redundancy for ‘overkill’5, ensuring a quick death of the host. In addition, some may kill insects by interfering with their development. In the TT01 genome, two paralogs, plu4092 and plu4436, encode proteins similar to juvenile hormone esterases (JHEs) of the insect Leptinotarsa decemlineata6. Juvenile hormone maintains the insect in a larval state. Its inactivation by JHE allows metamorphosis to proceed. JHEs may be used to trigger the insect endocrine machinery at an inappropriate time and thus represents a promising approach for insect control7. These genes are located downstream of highly related orphan genes (plu4093 and plu4437), suggesting a locus duplication. The toxicity of the proteins encoded by these two loci was verified experimentally. Two Escherichia coli clones, containing the recombinant BAC1A02 and BAC8C11, were shown to be toxic toward insects. BAC1A02, which contains the locus plu4093–plu4092, exhibited substantial oral toxicity toward three mosquito species, Aedes aegypti,
1Laboratoire
de Génomique des Microorganismes Pathogènes, 2Génopole, Plate-Forme Intégration et Analyse Génomiques, 3Unité de Génétique des Génomes Bactériens, 4Unité de Programmation Moléculaire et Toxicologie Génétique, Institut Pasteur, 28 rue du Dr. Roux, 75724 Paris Cedex 15, France. 5Laboratoire EMIP, Université Montpellier II, IFR122, Institut National de la Recherche Agronomique (UMR 1133), 34095 Montpellier Cedex 5, France. 6Atelier de Génomique Comparative, Génoscope/CNRS-UMR 8030, 2, rue Gaston Crémieux, 91006 Evry Cedex 15, France. 7Laboratoire de Microbiologie et de Génétique Moléculaire, CNRS, 118 Route de Narbonne, 31062 Toulouse Cedex, France. 8Bayer CropScience, 1 rue Pierre Fontaine, 91058 Evry, France. 9Bayer CropScience, 2 T.W. Alexander Drive, Research Triangle Park, North Carolina 27709, USA. Correspondence should be addressed to F.K. (
[email protected]). Published online 5 October2003; doi:10.1038/nbt886
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1307
ARTICLES 0
5
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
1
Photorhabdus luminescens 5,688,987 bp
4
2
3 Figure 1 Circular representation of the P. luminescens genome. The outer scale is marked in megabases. Circle 1 and 2 (from outside to inside), genes transcribed clockwise and counterclockwise. Circle 3, transposases (red) and phage-related genes (green). Circle 4, GC bias (G – C/G + C). Circle 5, GC content with <32% G + C in light yellow, between 32% and 53.6% G + C in yellow and with >53.6% G + C in dark yellow. Shifts in GC bias correspond to positions of the predicted origin and terminus of replication or to putative prophage regions (green).
Culex pipiens and Anopheles gambiae, as well as toward the lepidopteran Plutella xylostella; BAC8C11, which contains the locus plu4437–plu4436, showed insecticidal activity toward P. xylostella (data not shown). Ingestion of undiluted recombinant E. coli, in which both plu4093 and plu4092 were expressed under the control of the Plac promoter (pDIA700), led to killing of 96% of P. xylostella and 100% of C. pipiens larvae after 48 h8 (Table 1). Partial deletion of plu4092 (pDIA701) abolished toxicity for both insects, indicating that the JHElike product of plu4092 is required for toxicity. This is the first time that a protein lethal for mosquitoes has been found in P. luminescens. Mcf, a gene previously identified in the P. luminescens strain W14, encodes a cytotoxin that is able to kill the insect Manduca sexta5. Strain TT01 possesses this gene as well as the paralog plu3128. Four toxin-complex loci, tca, tcb, tcc and tcd, have been identified in the P. luminescens strain W14 (ref. 9). The tca locus encodes complexes
with high oral toxicity for the insect Manduca sexta10,11. Strain TT01 contains the tcc and tcd loci, an incomplete tca locus and five newly identified tc loci (Fig. 2; Supplementary Table 2 online). The physical linkage of the tcdAB and tccC genes was found in three other insectassociated bacteria, Xenorhabdus nematophila, Serratia entomophila and Yersinia pestis. This organization may be due to gene transfer events between bacteria sharing the same niche, namely insects. The putative tcc and rhs gene products belong to one protein superfamily9. They contain repeated Tyr-Asp motifs, previously found in teneurins, transmembrane proteins of invertebrates and vertebrates12. An in silico search for the presence of Arg-Tyr × Tyr-Asp motifs (at least three, no more than one mismatch) in the TTO1 proteome identified 14 putative proteins of a superfamily including the 7 TccC paralogs, putative nematicidal proteins (Plu2222 and Plu2442) and hypothetical rhs-like elements as well as Mcf and its paralog (see Supplementary Fig. 1 and Supplementary Table 3 online). It has been suggested that Tyr-Asp-repeat proteins bind carbohydrates12. One could speculate that proteins belonging to this superfamily are exposed at the surface of the bacteria where they are important for evasion of defenses from a broad spectrum of insects. P. luminescens contains eight genes predicted to encode hemolysinor hemagglutinin-related proteins secreted by the two-partner secretion (TPS) pathway. These comprise the secreted PhlA protein and the PhlB protein, which allows secretion and activation of PhlA13. We identified six other TPS gene pairs and ten incomplete loci (Supplementary Table 3 online). Repeats-in-toxin (RTX) proteins constitute another family of toxins, including cytolytic toxins, metalloproteases and lipases. We identified eight genes whose deduced protein sequence is homologous to the Vibrio cholerae RtxA toxin (4,558 amino acids), which causes crosslinking and depolymerization of actin stress fibers in an in vivo model14. The rtxA homologs are clustered in two chromosomal regions and are tandemly organized. Four are complete genes and four are disrupted by frameshifts or insertion sequences (ISs; Supplementary Table 2 online). Furthermore, the organization of the genes predicted to encode the RTX secretion system is identical to that in V. cholerae (Fig. 3). Plu4117 and Plu3668 belong to another subfamily of RTX proteins. This large set of putative RTX toxins and related proteins probably contributes to the insect pathogenicity of P. luminescens. In addition to the RTX family we identified a new family of putative proteins containing Gly-Asp repeats (Supplementary Fig. 2 online). Genes encoding proteins similar to toxins from other bacteria were also predicted, including Plu0840, similar to the heat-stable cytotonic enterotoxin from Aeromonas hydrophila15, and Plu1537, similar to a component of the Bacillus thuringiensis δ-endotoxin16.
Toxins against competitors One of the most important questions about the ecology of this bacterium is how P. luminescens defends the insect cadaver against different microbial competitors. The TT01 genome contains 33 genes, clustered in 20 loci, encoding proteins similar to polyketide and nonribosomal peptide synthases that may be part of the biosynthetic pathway of Table 1 Oral larvicidal activities of E. coli clones on second-instar insect larvae antibiotics known to be produced by P. lumia E. coli clones Mortality of P. xylostella (%) Mortality of C. pipiens (%) nescens (Supplementary Table 4 online). As observed in strain W14 (ref. 17), ten genes of 48 h 72 h 24 h 48 h TT01 are similar to genes for the biosynthesis pDIA 700 (plu4093 + plu4092) 96 100 30 ± 8b 100 of syringomycin by Pseudomonas syringae, pDIA 701 (plu4093 + truncated plu4092) <15 <15 0 0 which acts as a pore-forming host cell cytoControl (water) ND ND 0 0 lysin18. One gene, plu2670, encodes a 16,367aPercentages of larval mortality were corrected with the negative control using the Abbott equation. bStandard error of the amino-acid putative peptide synthase that is, mean. ND, not done.
1308
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
ARTICLES a plu0516 tcaC
tcaZ
b
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
tccC4 hpaBC
tcdA5
c
tcdA2
tcdB2 tccC3 tcdA4 tchA tciR1
IS
tccA1 tccB1
tccC1
plu4166 - 4162 (tccZ)
d tccA2 tccB2 tccA3 tccB3
tccC5 tciR2
tcdA1
tcdB1
tccC2
Figure 2 Toxin complex loci identified in strain TT01. Related genes are shown in the same color. Similarities of predicted proteins to previously known Tc proteins are reported in Supplementary Table 2 online. Nomenclature is according to Waterfield et al39. (a) Locus similar to the toxin complex a (tca) locus from strain W14 (GenBank accession no. AF046867). (b) Locus similar to the toxin complex d (tcd) island from strain W14 (GenBank accession no. AY144119). The W14 tcdA3 counterpart is only weakly similar, and thus the corresponding TT01 gene has been designated tcdA5. (c) Locus similar to the toxin complex c (tcc) locus from strain W14 (GenBank accession no. AF047028). (d) Loci weakly similar to the toxin complex c (tcc) locus from strain W14 (GenBank accession no. AF047028). (e) Other loci plu2333, plu2334 and plu2335 correspond to a single pseudogene.
e plu2333 - 2334 - 2335 tccC6 tccC7
to our knowledge, the largest protein reported for a prokaryote to date. In our work here, we frequently identified genes that were adjacent to antibiotic biosynthetic genes and that presumably encode transporters involved in antibiotic efflux. Four genes (plu0250, plu2474, plu2475, plu2476) are similar to Burkholderia glumae genes involved in the biosynthesis of toxoflavin. Compounds belonging to this family are active against a variety of Gram-positive bacteria and fungi19. Toxin-antitoxin systems are efficient in killing related bacteria, including other P. luminescens strains that invade the host20. The TT01 genome encodes three putative colicin-like factors and 17 putative immunity proteins involved in self-protection. These genes are located in three clusters, which may encode pyocin S3–like factors (plu0884, plu4177) and a colicin-like factor (plu1894) (Fig. 3 and Supplementary Table 3 online). This complex pattern of immunity-protein-coding genes may provide a selective advantage to TT01 against related bacteria producing similar toxins. Host interactions Another intriguing question is how P. luminescens interacts with the nematode midgut and the insect hemocytes. We identified a large number of genes that could be involved in adhesion. plu2096 encodes a protein similar to the Pseudomonas aeruginosa lectin PA-I, which recognizes specific carbohydrates exposed on the host cells and functions as an adhesin and a cytotoxin that plays a role in the initiation of infection21. The putative plu1561 gene product is similar to a Ca2+-dependent adhesion molecule of the cadherin family in Dictyostelium discoideum22. We have further identified one gene, plu2057, encoding a protein 45% similar to inv and three ail-like paralogs (plu2480, plu2481 and plu1967). In Yersinia species, Inv and Ail play a role in surface adhesion or invasion23,24. Another P. luminescens gene, plu2433, may encode
an adhesin required for insect colonization, because Plu2433 is similar to the recently discovered Erwinia carotovora virulence factor (Evf) allowing this bacterium to colonize the gut epithelium of Drosophila melanogaster25. P. luminescens TT01 has a large repertoire of fimbrial genes. Eleven clusters could be distinguished (see Supplementary Fig. 3 online). Among these, two gene clusters encode proteins similar to MR/P (mannose-resistant) fimbriae of Proteus mirabilis26 as well as type IV pili of pathogenic E. coli and Salmonella enterica27. This large repertoire of pili may help P. luminescens colonize the nematode gut and invade the different insect compartments, as the ngrA gene (an ngrA homolog has previously been identified in Photorhabdus luminescens strain W14; see ref. 28), belonging to cluster VIII, is important for bacterial-nematode interactions28. Secreted proteins P. luminescens secretes many enzymes that contribute to insect death and result in bioconversion of the insect cadaver29. In X. nematophila a zinc metalloprotease, PrtA, is involved in the immunosuppression of the insect30. In P. luminescens, the protein is encoded by the operon prtA-inh-prtBCD and is secreted by a type I secretory system31. Other putative proteases were identified, such as Plu2820, similar to a serine protease of the subtilisin family, Plu1382, similar to an extracellular metalloproteinase, and Plu2455, similar to calpain. Lipases are another class of secreted proteins. TT01 encodes ten triacylglycerol lipase-, phospholipase A- and D-like proteins. For example, the proteins predicted to be encoded by genes plu3370 and plu3369 are similar to phospholipase A and its accessory protein, respectively. Gene plu0830 encodes a similar protein that is more distantly related to phospholipase A. Both contain the lipase-specific consensus sequence G-x-S-x-G-G in their amino-terminal moiety. Yersinia enterocolitica phospholipase A contributes to pathogenesis in a mouse model32, suggesting a role in virulence for the TT01 homolog. Plu1971 is remarkably similar to the Y. pestis plasmid (pMT1)-encoded Yersinia murine toxin (Ymt). It contains the two phospholipase D motifs H-x-K-x4-D-x6-G-G. Genes plu1971 and ymt are 56% identical, with the same GC content (38%), substantially different from the 50% GC content of the pMT1 plasmid; this suggests that ymt has been acquired
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1309
ARTICLES a
VC1449 VC1446 rtxD
rtxB
V. cholerae
rtxC
rtxA
FS
IS
FS
FS
IS
P. luminescens ybfE
plu1331 rtxD
rtxB
rtxC plu1336/37
plu1339/40
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
plu1342/43
plu1344
IS
IS ybfF
plu1341
plu3207/09
plu3217
plu3324
b recB
D
plu0634
plu0635
plu0643
argA
amiC
c plu0918
yhfS
plu0925
plu2001
plu2017
plu2019
rafZ
d mrcB
plu0884
plu0887 plu0889
plu0895
plu1892
tccA1
e
plu4170
plu1894
plu4177
flhB
flhA
plu4178
IS plu1899
plu1911 flgN
f
IS fdoI
plu4890
plu4895
fdhE
Figure 3 Schematic representation of tandemly repeated genes. Related genes are color coded: blue-purple, genes encoding putative secretion systems; gray, genes belonging to the core genome; white, genes of unknown function; dashed, pseudogenes. ↓ IS indicates insertion-sequence element. (a) Red, tandemly repeated rtxA genes and genes encoding the RTX secretory machinery as compared to V. cholerae. (b) Yellow, genes encoding unknown proteins, probably secreted by the Type I secretory system encoded by plu0634 and plu0635. (c) Green, luxR gene clusters. (d) Red, clusters encoding putative pyocin- and colicin-like factors; yellow, immunity proteins. (e) Orange, highly related genes of unknown function. (f) Brown, putative O-methyltransferaseencoding genes.
by Y. pestis from P. luminescens or a close relative. Because Ymt is essential for flea colonization by Y. pestis33, P. luminescens and Y. pestis may use similar genes for insect colonization. Besides the lip-1 gene, encoding a triacylglycerol lipase34, we predicted four triacylglycerol lipases unrelated to Lip-1, encoded by the tandemly repeated paralogs plu1519, plu1518, plu1517 and plu1516. Another unrelated lipase, Plu2266, is similar to LipP (carboxylesterase) of Pseudomonas sp. strain B11-135, and Plu2313 is highly similar to the extracellular lipase, LipA, from Serratia marcescens36. Plu2313 lacks an N-terminal signal peptide but contains instead, in its C terminus, four almost exact repeats of the glycine-rich motif L-x-G-G-x-G-x-D of RTX toxins, suggestive of a similar secretion pathway37. We predicted a chitin-binding-like protein (Plu2352) and chitinase-like proteins (Plu2235, Plu2458 and Plu2461) as expected for an insect pathogen. Metabolism Within an insect, P. luminescens needs to get access to the available nutrients and has to deal with low-iron conditions. The overall inter-
1310
mediary metabolism of P. luminescens is similar to that found in most Enterobacteriaceae, but it comprises many degradation pathways that are absent from the laboratory strain E. coli K12, including urease, proline or histidine degradation pathways that may help the bacteria to multiply in the hemocele of the insect larvae, while providing the bacteria and the nematode host with building blocks for their growth. The TT01 genome contains many genes encoding proteins similar to monooxygenases, dioxygenases and hydroxylases that could be implicated in rapid elimination of insect polyphenols, such as plu4258, adjacent to glutathione transferase (plu4259), or operons such as the hca operon or plu1434–plu1437. Many gene products may be involved in the detoxification of reactive oxygen species generated by the invaded host; the bacterium may take advantage of the reactivity of dioxygen to metabolize insect products. We also identified counterparts of a putative steroid monooxygenase (Plu4232), glycine oxidase (Plu2242) and aromatic monooxygenase (Plu1313) as well as five tandemly organized paralogs (plu4890–plu4895), possibly encoding proteins similar to O-methyltransferase that may play a similar role (Fig. 3).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES Iron acquisition seems to be of particular importance for the life cycle of P. luminescens, as this bacterium has the largest known set of iron, heme, hemin and siderophore transporters. Five loci have counterparts in Y. pestis, one of which is similar to the genes involved in biosynthesis of the siderophore yersiniabactin (Supplementary Table 3 and Supplementary Fig. 4 online). A sixth locus (fecI–fecE) is similar to the E. coli iron(III)-dicitrate transport system. Two putative binding proteins and six outer membrane receptors of the ISVH (ironsiderophores, vitamin B12 and hemin) family may be involved in iron, heme, hemin or siderophore transport38 (Supplementary Table 3 online). P. luminescens encodes four complete type I, one type II and one type III secretion systems39. The latter one is highly similar to the plasmid-encoded type III system of Y. pestis and is probably involved in the secretion of virulence proteins or in immunomodulation of the insect response. The phage elements present in the genome may participate in the release of molecules involved in pathogenesis by partial lysis of the bacteria (S.G., unpublished data). Regulation To adapt to invertebrate environments such as the nematode gut, the insect hemolymph and the insect cadaver, P. luminescens needs to sense changes in such conditions as nutrient and cation availability, osmolarity and bacterial density. Strain TT01 encodes homologs of the five E. coli sigma factors RpoD, RpoS, RpoH, FliA and RpoN. In contrast to E. coli, which has only one ECF sigma factor, TT01 possesses five such factors. We predicted 192 transcriptional regulators (Supplementary Table 5 online) from the sequence. Interestingly, the LuxR family of regulators is over-represented (32 genes) and the majority of these genes are tandemly located in two regions (Fig. 3). This tandem organization of the luxR genes may play a role in the response of TT01 to acylhomoserine lactone (AHL) signals produced by TT01 or by closely related bacteria. LuxR regulators may then, in association with AHL-responsive systems, affect diverse biological functions including bioluminescence. Although TT01 contains a complete lux operon responsible for the production of bioluminescence, we could identify only one gene, plu2238, that might be part of an AHL-responsive system. It encodes a protein highly similar (76% identity) to an Agrobacterium tumefaciens N-AHL-lactonase40. Two other highly represented classes of regulators are phage gene repressors (37) and Ner-like regulators (15), which is consistent with the large number of prophages or prophage remnants present in the genome. Bacteriophages may have affected bacterial-host interactions. It has been shown that overexpression of a Ner-like regulator switches the primary bacterial variant, which supports nematode growth, to the secondary variant, which does not41. Nineteen two-component regulators and 20 LysR–type regulators were identified. A LysR-type regulator, also called HexA or LrhA, was reported to control symbiosis negatively and pathogenicity positively, suggesting that it is involved in the transition from symbiosis to pathogenicity42. However, the relatively low number of two-component regulators identified in the P. luminescens genome (compared to the 89 systems of P. aeruginosa) might be related to the quite stable environment encountered by P. luminescens, obviating the need for metabolic adaptation to a wide range of conditions. Genome comparison The P. luminescens life cycle and a number of its phenotypic traits resemble those of Y. pestis. Both transit and colonize insects and both are pathogenic bacteria. However, P. luminescens has been used for biological control of insects without causing any harm to humans. Several common phenotypic characteristics are reflected in the genome.
We identified 2,107 genes in P. luminescens that have an ortholog in Y. pestis. In addition, 77% (1,621 out of 2,107) of the orthologous genes of TT01 and Y. pestis CO92 (ref. 43) are syntenic. Interestingly, Y. pestis and P. luminescens share not only the chromosomal backbone of Enterobacteriaceae, which both share also with E. coli, but also many putative mobile regions encoding toxins, virulence factors and proteins of unknown function (Supplementary Fig. 5 online). Orthologs of genes present on the Y. pestis plasmids (pCD1 and pMT1) were also identified in the genome of P. luminescens. The impressive number of mobile genetic elements or their remnants suggest that the TT01 genome is subject to continuously ongoing gene transfer (Supplementary Table 6 online). We identified phage remnants representing 4% of the genome, 195 ISs or IS fragments, putative transposons, and 711 inverted repeats of the enterobacterial repetitive intergenic consensus (ERIC) sequences44, in contrast to only 21 ERIC sequences present in the E. coli K12 chromosome. In addition, several gene classes are over-represented (Supplementary Table 3 online), suggesting frequent rearrangements (duplication, recombination) and a high degree of genome plasticity. This redundancy may contribute not only to the impressive arsenal of toxins but also to the generation of variability, which is an advantage for a pathogen subject to strong selective constraints. P. luminescens kills a wide variety of insects and thus has to circumvent a multitude of host defenses. DISCUSSION The availability and the functional analysis of the genome sequence of P. luminescens TT01 should lead to several useful applications, such as the development of new entomotoxins for crop protection and the genetic engineering of the bacterium-nematode pair for use as biological control agents. The identification of new antibiotic biosynthetic genes, which could be manipulated to generate new biological activities by domain shuffling, provides a promising resource for fighting microbial infections in the future. Furthermore, the availability of the first genome sequence of a symbiotic as well as entomopathogenic bacterium is an important advance in helping to decipher the relation between pathogenesis and symbiosis. METHODS Cloning, sequencing, assembly and annotation. Genome sequencing was performed using the whole-genome shotgun strategy45, as described previously46,47. Two libraries (1–2 kb and 2–3 kb inserts) were generated by random mechanical shearing of genomic DNA and cloning into pcDNA-2.1 (Invitrogen) and a medium-size insert library (5–10 kb) was generated in the low copy number vector pSYX3448. The BAC library was constructed as previously described49. Assembly and annotation are detailed in Supplementary Methods online. BAC1A02 encompassing the JHE-like toxin locus plu4093–plu4092 extends from position 4,729,455 to position 4,789,585 of the P. luminescens genome; BAC8C11 encompassing the JHE-like toxins locus plu4437–plu4436 extends from position 5,179,747 to position 5,205,606. To construct the pDIA700 plasmid, a DNA fragment containing the two genes plu4093 and plu4092, was generated by PCR with primers JHE2 (5′AACTGCAGCATTGAAGCAGAGCGTTGACAT-3′) and JHE3 (5′-CGGGATCCGACGTCGGCAAGTGCATCAAAT-3′). The amplified DNA fragment of 2,060 bp was purified and cloned into the pBluescript SK vector (Stratagene) at the EcoRV site. To inactivate plu4092, the pDIA700 was digested with EcoRI, leading to a deletion of the DNA region located between an EcoRI site located at the 270th codon of plu4092 and the pBluescript EcoRI site, and self-ligated to yield pDIA701, which contains a 3′ truncated plu4092 gene. Insecticidal assays. P. xylostella leaf bioassay. The recombinant E. coli XL1-Blue strain containing pDIA700 or pDIA701 and E. coli DH10B containing BAC1A02 or BAC8C11
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1311
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES were grown for 20 h at 28 °C in 50 ml of LB broth supplemented with 100 µg/ml ampicillin or 12.5 µg/ml chloramphenicol, respectively. For each clone, six cabbage leaves (3 cm diameter) were incubated in undiluted bacterial cultures for 1 h with 0.05% Tween 20 (Sigma). Treated leaves were put onto 6-well plates with agar beds containing 15 g/l of agar (Difco) and 30 mg/l of the fungicide nipagine (Sigma). Five second-instar larvae were placed into each well. All the assays were done at 28 °C with a photoperiod of 11 h:13 h (night/day). After 2 d the treated leaves were replaced by untreated leaves. The larval mortality was recorded at day 2 and day 3. Recombinant E. coli strains containing the pBluescript SK vector without insert and P. luminescens TT01 cultures were used as negative and positive controls, respectively. The percentage of larval mortality was corrected with respect to the negative control by using the Abbott equation50. The toxicity of each clone was tested on 30 larvae, and each experiment was done in triplicate. Mosquitocidal activity. Derivatives of E. coli TG1 containing plasmids pDIA700 and pDIA701, or E. coli DH10B containing BAC1A02 or BAC8C11, were grown at 30 °C in 50 ml LB broth supplemented with 100 µg/ml ampicillin or 12.5 µg/ml chloramphenicol, respectively, until the optical density at 600 nm (OD600) reached a value of 2, then harvested by centrifugation and resuspended at the same optical density. Ten second-instar C. pipiens larvae were placed in Petri dishes (2.5 cm diameter) containing 5 ml bacterial suspensions at OD600 = 2 (or water as a control). Yeast cells were given as a food source in all dishes (to avoid mortality in the control), and larval mortality was recorded at days 1 and 2. Two independent experiments were conducted at 24 ± 2 °C, each one in duplicate. The same procedure was used for toxicity assays involving Aedes aegypti and Anopheles gambiae. URLs. The genome sequence has been submitted to EMBL under the accession number BX470251. The complete data set of DNA and protein sequences, linked to the relevant annotations and functional assignments, is available online at http://genolist.pasteur.fr/PhotoList. Note: Supplementary information is available on the Nature Biotechnology website. ACKNOWLEDGMENTS This work received financial support from the Institut Pasteur, the Centre National de la Recherche Scientifique, the Institut National de la Recherche Agronomique and the Ministère de l’Industrie et des Finances (Après séquençage des Génomes). We wish to thank Tatiana Vallaeys and Evelyne Krin for helpful discussions, Ivan Moszer and Eduardo Rocha for help with bioinformatics, Elisabeth Couvé, Christina Nielsen-Le Roux and Hafed Nedjari for expert technical assistance and Tim Stinear for critical reading of the manuscript. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Received 7 May; accepted 18 August 2003 Published online at http://www.nature.com/naturebiotechnology/ 1. Boemare, N.E., Akhurst, R.J. & Mourant, R.G. DNA relatedness between Xenorhabdus spp. (Enterobacteriaceae), symbiotic bacteria of entomopathogenic nematodes, and a proposal to transfer Xenorhabdus luminescens to a new genus, Photorhabdus gen. nov. Int. J. Syst. Bacteriol. 43, 249–255 (1993). 2. Forst, S., Dowds, B., Boemare, N. & Stackebrandt, E. Xenorhabdus and Photorhabdus spp.: bugs that kill bugs. Annu. Rev. Microbiol. 51, 47–72 (1997). 3. ffrench-Constant, R. et al. Photorhabdus: towards a functional genomic analysis of a symbiont and pathogen. FEMS Microbiol. Rev. 26, 433–456 (2003). 4. Fischer-Le Saux, M., Viallard, V., Brunel, B., Normand, P. & Boemare, N.E. Polyphasic classification of the genus Photorhabdus and proposal of new taxa: P. luminescens subsp. luminescens subsp. nov., P. luminescens subsp. akhurstii subsp. nov., P. luminescens subsp. laumondii subsp. nov., P. temperata sp. nov., P. temperata subsp. temperata subsp. nov. and P. asymbiotica sp. nov. Int. J. Syst. Bacteriol. 49, 1645–1656 (1999). 5. Daborn, P.J. et al. A single Photorhabdus gene, makes caterpillars floppy (mcf), allows Escherichia coli to persist within and kill insects. Proc. Natl. Acad. Sci. USA 99, 10742–10747 (2002). 6. Vermunt, A.M., Koopmanschap, A.B., Vlak, J.M. & de Kort, C.A. Evidence for two juvenile hormone esterase-related genes in the Colorado potato beetle. Insect Mol. Biol. 7, 327–336 (1998). 7. Bonning, B.C. & Hammock, B.D. Development of recombinant baculoviruses for insect control. Annu. Rev. Entomol. 41, 191–210 (1996). 8. patent PCT-FR/0301238 Duchaud, E. et al. Insecticidal proteins of Photorhabdus luminescens. International patent PCT-FR03/01238. April 17, 2002.
1312
9. Waterfield, N.R., Bowen, D.J., Fetherston, J.D., Perry, R.D. & ffrench-Constant, R.H. The tc genes of Photorhabdus: a growing family. Trends Microbiol. 9, 185–191 (2001). 10. Bowen, D. et al. Insecticidal toxins from the bacterium Photorhabdus luminescens. Science 280, 2129–2132 (1998). 11. Marokhazi, J. et al. Using a DNA microarray to investigate the distribution of insect virulence factors in strains of Photorhabdus. J. Bacteriol. 185, 4648–4656 (2003). 12. Minet, A.D. & Chiquet-Ehrismann, R. Phylogenetic analysis of teneurin genes and comparison to the rearrangement hot spot elements of E. coli. Gene 257, 87–97 (2000). 13. Brillard, J., Duchaud, E., Boemare, N., Kunst, F. & Givaudan, A. The PhlA hemolysin from the entomopathogenic bacterium Photorhabdus luminescens belongs to the two-partner secretion family of hemolysins. J. Bacteriol. 184, 3871–3878 (2002). 14. Fullner, K.J. & Mekalanos, J.J. In vivo covalent cross-linking of cellular actin by the Vibrio cholerae RTX toxin. EMBO J. 19, 5315–5323 (2000). 15. Sha, J., Kozlova, E.V. & Chopra, A.K. Role of various enterotoxins in Aeromonas hydrophila-induced gastroenteritis: generation of enterotoxin gene-deficient mutants and evaluation of their enterotoxic activity. Infect. Immun. 70, 1924–1935 (2002). 16. Moellenbeck, D.J. et al. Insecticidal proteins from Bacillus thuringiensis protect corn from corn rootworms. Nat. Biotechnol. 19, 668–672 (2001). 17. ffrench-Constant, R.H. et al. A genomic sample sequence of the entomopathogenic bacterium Photorhabdus luminescens W14: potential implications for virulence. Appl. Environ. Microbiol. 66, 3310–3329 (2000). 18. Bender, C.L., Alarcon-Chaidez, F. & Gross, D.C. Pseudomonas syringae phytotoxins: mode of action, regulation, and biosynthesis by peptide and polyketide synthetases. Microbiol. Mol. Biol. Rev. 63, 266–292 (1999). 19. Nagamatsu, T. et al. Syntheses of 3-substituted 1-methyl-6-phenylpyrimido[5,4-e]1,2,4-triazine-5,7(1H,6H)-diones (6-phenyl analogs of toxoflavin) and their 4-oxides, and evaluation of antimicrobial activity of toxoflavins and their analogs. Chem. Pharm. Bull. (Tokyo) 41, 362–368 (1993). 20. Sharma, S. et al. The lumicins: novel bacteriocins from Photorhabdus luminescens with similarity to the uropathogenic-specific protein (USP) from uropathogenic Escherichia coli. FEMS Microbiol. Lett. 214, 241 (2002). 21. Avichezer, D., Katcoff, D.J., Garber, N.C. & Gilboa-Garber, N. Analysis of the amino acid sequence of the Pseudomonas aeruginosa galactophilic PA-I lectin. J. Biol. Chem. 267, 23023–23027 (1992). 22. Wong, E.F., Brar, S.K., Sesaki, H., Yang, C. & Siu, C.H. Molecular cloning and characterization of DdCAD-1, a Ca2+-dependent cell-cell adhesion molecule, in Dictyostelium discoideum. J. Biol. Chem. 271, 16399–16408 (1996). 23. Gustavsson, A. et al. Role of the β1-integrin cytoplasmic tail in mediating invasinpromoted internalization of Yersinia. J. Cell. Sci. 115, 2669–2678 (2002). 24. Miller, V.L., Beer, K.B., Heusipp, G., Young, B.M. & Wachtel, M.R. Identification of regions of Ail required for the invasion and serum resistance phenotypes. Mol. Microbiol. 41, 1053–1062 (2001). 25. Basset, A., Tzou, P., Lemaitre, B. & Boccard, F. A single gene that promotes interaction of a phytopathogenic bacterium with its insect vector, Drosophila melanogaster. EMBO Rep. 4, 205–209 (2003). 26. Zhao, H., Li, X., Johnson, D.E., Blomfield, I. & Mobley, H.L. In vivo phase variation of MR/P fimbrial gene expression in Proteus mirabilis infecting the urinary tract. Mol. Microbiol. 23, 1009–1019 (1997). 27. Srimanote, P., Paton, A.W. & Paton, J.C. Characterization of a novel type IV pilus locus encoded on the large plasmid of locus of enterocyte effacement-negative Shigatoxigenic Escherichia coli strains that are virulent for humans. Infect. Immun. 70, 3094–3100 (2002). 28. Ciche, T.A., Bintrim, S.B., Horswill, A.R. & Ensign, J.C. A Phosphopantetheinyl transferase homolog is essential for Photorhabdus luminescens to support growth and reproduction of the entomopathogenic nematode Heterorhabditis bacteriophora. J. Bacteriol. 183, 3117–3126 (2001). 29. Bowen, D., Blackburn, M., Rocheleau, T., Grutzmacher, C. & ffrench-Constant, R.H. Secreted proteases from Photorhabdus luminescens: separation of the extracellular proteases from the insecticidal Tc toxin complexes. Insect. Biochem. Mol. Biol. 30, 69–74 (2000). 30. Caldas, C., Cherqui, A., Pereira, A. & Simoes, N. Purification and characterization of an extracellular protease from Xenorhabdus nematophila involved in insect immunosuppression. Appl. Environ. Microbiol. 68, 1297–1304 (2002). 31. Valens, M., Broutelle, A.C., Lefebvre, M. & Blight, M.A. A zinc metalloprotease inhibitor, Inh, from the insect pathogen Photorhabdus luminescens. Microbiology 148, 2427–2437 (2002). 32. Schmiel, D.H., Wagar, E., Karamanou, L., Weeks, D. & Miller, V.L. Phospholipase A of Yersinia enterocolitica contributes to pathogenesis in a mouse model. Infect. Immun. 66, 3941–3951 (1998). 33. Hinnebusch, B.J. et al. Role of Yersinia murine toxin in survival of Yersinia pestis in the midgut of the flea vector. Science, 296, 733–735 (2002). 34. Thaler, J.O., Duvic, B., Givaudan, A. & Boemare, N. Isolation and entomotoxic properties of the Xenorhabdus nematophilus F1 lecithinase. Appl. Environ. Microbiol. 64, 2367–2373 (1998). 35. Choo, D.W., Kurihara, T., Suzuki, T., Soda, K. & Esaki, N. A cold-adapted lipase of an Alaskan psychrotroph, Pseudomonas sp. strain B11-1: gene cloning and enzyme purification and characterization. Appl. Environ. Microbiol. 64, 486–491 (1998). 36. Akatsuka, H. et al. The lipA gene of Serratia marcescens which encodes an extracellular lipase having no N-terminal signal peptide. J. Bacteriol. 176, 1949–1956 (1994). 37. Li, X., Tetling, S., Winkler, U.K., Jaeger, K.E. & Benedik, M.J. Gene cloning, sequence analysis, purification, and secretion by Escherichia coli of an extracellular
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES lipase from Serratia marcescens. Appl. Environ. Microbiol. 61, 2674–2680 (1995). 38. Dassa, E. & Bouige, P. The ABC of ABCS: a phylogenetic and functional classification of ABC systems in living organisms. Res. Microbiol. 152, 211–229 (2001). 39. Waterfield, N.R., Daborn, P.J. & ffrench-Constant, R.H. Genomic islands in Photorhabdus. Trends Microbiol. 10, 541–545 (2002). 40. Zhang, H.B., Wang, L.H. & Zhang, L.H. Genetic control of quorum-sensing signal turnover in Agrobacterium tumefaciens. Proc. Natl. Acad. Sci. USA 99, 4638–4643 (2002). 41. O’Neill, K.H., Roche, D.M., Clarke, D.J. & Dowds, B.C. The ner gene of Photorhabdus: effects on primary-form-specific phenotypes and outer membrane protein composition. J. Bacteriol. 184, 3096–3105 (2002). 42. Joyce, S.A. & Clarke, D.J. A hexA homologue from Photorhabdus regulates pathogenicity, symbiosis and phenotypic variation. Mol. Microbiol. 47, 1445–1457 (2003). 43. Parkhill, J. et al. Genome sequence of Yersinia pestis, the causative agent of plague. Nature 413, 523–527 (2001).
44. Hulton, C.S.J., Higgins, C.F. & Sharp, P.M. ERIC sequences: a novel family of repetitive elements in the genomes of Escherichia coli, Salmonella typhimurium and other enterobacteria. Mol. Microbiol. 5, 825–834 (1991). 45. Fleischmann, R.D. et al. Whole-genome random sequencing and assembly of Haemophilus influenzae Rd. Science 269, 496–512 (1995). 46. Frangeul, L. et al. Cloning and assembly strategies in microbial genome projects. Microbiology 145, 2625–2634 (1999). 47. Frangeul, L. et al. CAAT-Box, Contigs-Assembly and Annotation tool-box for genome sequencing projects. Bioinformatics, in the press. 48. Xu, S.Y. & Fomenkov, A. Construction of pSC101 derivatives with Camr and Tetr for selection or LacZ’ for blue/white screening. Biotechniques 17, 57 (1994). 49. Buchrieser, C. et al. The 102-kilobase pgm locus of Yersinia pestis: sequence analysis and comparison of selected regions among different Yersinia pestis and Yersinia pseudotuberculosis strains. Infect. Immun. 67, 4851–4861 (1999). 50. Abbott, W.S. A method for computing the effectiveness of an insecticide. J. Econ. Entomol. 18, 265–267 (1925).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1313
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES
Targeting cytokines to inflammation sites Gill Adams1,3, Sandrine Vessillier1,3, Hanna Dreja2 & Yuti Chernajovsky1 To increase the half-life of a cytokine and target its activation specifically to disease sites, we have engineered a latent cytokine using the latency-associated protein (LAP) of transforming growth factor-β1 (TGF-β1) fused via a matrix metalloproteinase (MMP) cleavage site to interferon (IFN)-β at either its N or C terminus. The configuration LAP-MMP-IFN-β resembles native TGF-β and lacks biological activity until cleaved by MMPs, whereas the configuration IFN-β-MMP-LAP is active. LAP provides for a disulfide-linked shell hindering interaction of the cytokine with its cellular receptors, conferring a very long half-life of 55 h in vivo. Mutations of the disulfide bonds in LAP abolish this latency. Samples of cerebrospinal fluid (CSF) or synovial fluid from patients with inflammatory diseases specifically activate the latent cytokine, whereas serum samples do not. Intramuscular injection in arthritic mice of plasmid DNA encoding these constructs demonstrated a greater therapeutic effect of the latent as compared to the active forms.
Cytokine gene expression is generally controlled at the transcriptional and post-transcriptional levels1,2. Another level of regulation is found in cytokines that interact with the extracellular matrix (ECM)3, such as TGF-β, which is secreted in a latent form, binds to the ECM and becomes ‘activated’ by releasing the cytokine moiety at sites where processes of inflammation, wound healing and tissue repair take place4. Cytokines are natural products that serve as soluble local mediators of cell-cell interactions. They have short half-lives and a variety of pleiotropic actions, some of which can be harnessed for therapeutic purposes5,6. Unfortunately, these very potent biological agents have to be administered at very high concentrations systemically to achieve biologically meaningful concentrations in the tissue being targeted. This gives rise to toxic systemic effects that limit their use and efficacy7–9. To bypass the toxic effects of systemic administration, we have engineered a cytokine that uses the LAP of TGF-β1 as a protective ‘shell’ preventing it from interacting with its receptors. Our engineered latent cytokine also possesses an MMP cleavage site that is cut at sites of inflammation and tissue remodeling, releasing the active cytokine and enabling it to act on cells locally. To our knowledge, this is the first description of a fusion protein enabling action of a cytokine at sites of inflammation. Other cytokine fusion proteins, such as immunocytokines, target cytokines using antibodies to specific cell types expressing defined antigens10; previous fusion proteins of TGF-β11,12 have focused on the active cytokine moiety but not its LAP. RESULTS Structural considerations To develop a latent cytokine using the LAP domain of TGF-β, we built fusion proteins in two conformations, one containing LAP at the N terminus of mouse IFN-β and the other at the C terminus. Cysteines
224 and 226 are important in the intermolecular disulfide bond between two LAPs. Their mutation to serine renders the molecule ‘active’13–15. The RGD motif (residues 245–247) facilitates the interaction with integrins16,17. Cysteine 33 is important for the disulfide bridge with the third eight-cysteine-rich repeat of latent TGF-β binding protein (LTBP)18. Modification of LTBP by other enzymes such as thrombospondin19,20, transglutaminase21,22, and MMP9 and MMP2 (ref. 23) could release the active, noncovalently bound portion of TGF-β from the latent complex. To prevent processing of the LAP-IFN-β protein at Arg278 of LAP, we cloned LAP spanning amino acids Met1–Ser273. This sequence was followed by a flexible linker (GGGGS), a putative MMP9 (refs. 24,25) or putative MMP1 (ref. 26) cleavage site (PLGLWA), and another flexible portion (GGGGSAAA) followed by mature IFN-β (starting at Ile22). We expected that embedding the MMP cleavage site in a hydrophilic area would facilitate access to enzymatic attack. The core of the cleavage site (PLGL) is cleaved as a peptide by MMP2 and, in a different version (PLGI), also by MMP3, MMP7 and MMP8 (ref. 26). The IFN-β-LAP molecule consisted of the precursor IFN-β sequence, with its stop codon mutated to allow readthrough of the flexible linker and the MMP site, followed by the mature sequence of LAP (Leu29–Ser273). The unprocessed LAP-IFN-β and IFN-β-LAP fusion proteins have expected molecular weights of 52.375 and 51.768 kDa, respectively. The primary sequences of these fusion proteins each contain four possible N-glycosylation sites. A schematic representation of the primary structure and putative folding of these proteins and their interaction with LTBP are shown in Figure 1a,b. We expressed these recombinant proteins in dihydrofolate reductase–deficient Chinese hamster ovary (DHFR− CHO) cells using a permanent DNA-transfection process. Secreted IFN-β-LAP had a low residual biological activity (210 U/ml), whereas LAP-IFN-β was completely ‘latent’ or inactive. The expression of these recombinant
1Bone and Joint Research Unit, William Harvey Research Institute, St. Bartholomew’s and Royal London School of Medicine and Dentistry, Queen Mary, University of London, Charterhouse Square, London EC1M 6BQ, UK. 2Present address: Institut de Génétique Moléculaire, CNRS, 1919 route de Mende, 34293 Montpellier, Cedex 05, France. 3These authors contributed equally to this work. Correspondence should be addressed to Y.C. (
[email protected]).
Published online 5 October 2003; corrected 19 October 2003 (details online); doi:10.1038/nbt888
1314
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
ARTICLES a
c
N
LAP
N
IFN-β
b
MMP
MMP
M.W.
Fusion protein
C
LAP
N
N
LTBP
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
C
IFN-β
LTBP IFN-β-LAP LAP-IFN-β
C N
N
C C N C
N C
C
Figure 1 Schematic representation of fusion proteins and their putative folding. (a) Primary structure of recombinant latent proteins. The linear sequence arrangement of the constituents in the two configurations is shown. The green box at the N terminus depicts the native signal sequence peptide for secretion of either TGF-β or IFN-β, respectively. (b) Putative folding and interactions with LTBP of latent cytokine. Right, folding of LAP-IFN-β, resembling the folding of native TGF-β. Near the N terminus of the protein, Cys33 interacts with the third eight-cysteine-rich repeat of LTBP, whereas Cys224 and Cys226 dimerize the protein through intermolecular disulfide bonds27. Left, structure of IFN-β-LAP. Cys33 is now located behind the MMP cleavage site and Cys224 and Cys226 are closer to the C terminus of the protein. In LTBP, the epidermal growth factor–like repeats are shown as small, square yellow boxes, the cysteine-rich repeats and hybrid domain as blue circles, and the ‘hinge region,’ which is sensitive to proteolytic cleavage, as a line. Disulfide bonds are shown as blue lines. (c) Detection of recombinant fusion proteins in cell supernatants. Nondenaturing SDS-PAGE of supernatants from nontransfected (lane 1), LAP-IFN-β-transfected (lane 2) and IFN-β-LAP-transfected CHO cells (lane 3). Positions of the double bands of newly expressed fusion proteins are marked by a double arrow. Positions of the molecular weight markers (M.W.) in kilodaltons are shown.
After gene amplification with methotrexate (MTX), the minor 75-kDa component (Fig. 2a, lanes 1, 3, 5) became the major component recognized by anti-LAP antibody (Fig. 2b, lanes 1–4). The approximately sixto eightfold increase of the fusion protein over LTBP indicates that interaction with LTBP is not a requirement for latency of LAPIFN-β (see basal levels of activity in Fig. 3 and Table 1). Interestingly, the monoclonal antiIFN-β antibody does not seem to recognize the 75-kDa glycosylated product (Fig. 2b,c, lanes 5–8) and the anti-LAP antibody recognizes it poorly in the IFN-β-LAP configuration (Fig. 2b, lanes 5–8), indicating that these fusion proteins have different conformations. This may explain the different sensitivities of these proteins to various MMPs (see below) and their differing degree of latency. The calculated molecular weight of the secreted recombinant proteins is 49.376 kDa for both LAP-IFN-β and IFN-β-LAP. The higher molecular weight is due to glycosylation of these proteins. Incubation with N-glycosidase F yields two major proteins of molecular weights 70 kDa and 51 kDa, corresponding to LTBP and fusion protein, respectively (data not shown).
MMP cleavage of recombinant proteins Immunoprecipitated complexes were treated proteins was similar, as confirmed by Western blotting with an anti- overnight with MMPs either singly or in combination. MMP1 did not cleave the 57-kDa recombinant product very efficiently (Fig. 2a). LAP antibody (data not shown). MMP1 cleaved the glycosylated form of the fusion protein (Fig. 2a, lanes 5, 6 and Fig. 2b, lane 2), whereas MMP3 digested it into several Biochemical characterization of recombinant proteins Secreted proteins from permanently transfected CHO cells were discrete bands (Fig. 2a, lanes 3, 4 and Fig. 2b, lanes 3, 7). The LTBP band labeled with [35S]methionine and [35S]cysteine. When subjected to was also cleaved by MMP3 (Fig. 2a, lanes 3, 4 and Fig. 2c, lanes 3, 7), electrophoresis under nonreducing conditions, the labeled LAP-IFN-β giving rise to a 78-kDa product. Two of the digested products (43 kDa and IFN-β-LAP proteins each produced two major bands of more than and 32 kDa) corresponded to the expected LAP and IFN-β polypeptide 97 kDa that were not seen in supernatants from nontransfected CHO fragments, respectively, as assessed by western blotting with the respeccells (Fig. 1c). Upon immunoprecipitation with anti-LAP antibody in tive antibodies (data not shown). The specificity shown in these in vitro experiments does not fully reducing conditions, supernatants from LAP-IFN-β- and IFN-β-LAPtransfected cells showed three bands, one of 57 kDa and another reflect the antiviral activity measured in cell supernatants after MMP of 135 kDa and a minor component at around 75 kDa (Fig. 2a). The treatment. Cell supernatants were already activated to a certain 135-kDa protein is probably the CHO-derived LTBP, which is linked by extent, indicating that other proteinases in the supernatant may activate the latent cytokine moiety. We did not see increased proteolysis disulfide bonds to LAP27. of the fusion polypeptides after immunoprecipitation using a combination of recombiTable 1 IFN-β activity of supernatants of MTX-amplified, fusion protein–transfected nant pro-MMP9 with MMP1 or MMP3, or CHO cells with aminophenylmercuric acetate (APMA)IFN-β biological activity (U/ml) after treatment with: activated pro-MMP9 on its own in vitro (data not shown), suggesting that MMP9 does not Cytokine No MMP MMP1 MMP3 Pro-MMP9 Pro-MMP9 Pro-MMP9 RASF No SPI construct + MMP1 + MMP3 cleave the fusion proteins directly. Cell supernatants were concentrated 100LAP-IFN-β 288 6,144 9,216 288 1,536 768 1,152 768 fold by centrifugation through porous mem(50 nM MTX) branes to allow for MMP activity at a higher IFN-β-LAP 1,536 6,144 3,072 1,536 1,536 4,608 6,144 3,072 substrate concentration. LAP-IFN-β super(12.5 nM MTX) natant showed antiviral activity without any Supernatants were supplemented with or without (last column) serine protease inhibitors (SPI) and MMPs as further treatment (163 U/ml). This is not surindicated. Incubations were overnight at 37 °C and tested for IFN-β activity. Supernatants from nontransfected CHO prising, as CHO cells are reported to secrete a cells had no IFN-β activity even after treatment with MMPs or synovial fluid from patients with rheumatoid arthritis (RASF) at 1/5 of final volume (data not shown). The RASF is the same as used in Figure 3. variety of proteinases28,29, including MMPs30.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1315
ARTICLES a
LAP- IFN-β IFN-β-LAP + MMP1 + MMP3
+ – – –
– + – –
1
2
M.W.
+ – – +
– + – +
+ – + –
– + + –
3
4
5
6
b
Anti-LAP Anti-IFN MMP1 MMP3 SF
+ – – – –
+ – + – –
+ – – + –
+ – – – +
– + – – –
– + + – –
– + – + –
– + – – +
M.W.
220
220
LTBP 97
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
97 66
Fusion protein
45
*
LTBP
*
LAP
66
LTBP LAP-IFN-β
45
LAP
*
IFN- β IFN-β
30
21
c
Anti-LAP Anti-IFN MMP1 MMP3 SF
+ – – – –
+ – + – –
+ – – + –
+ – – – +
– + – – –
– + + – –
– + – + –
– + – – +
M.W.
1
2
3
4
5
6
7
8
*
30
*
21
*
Figure 2 Detection and characterization of fusion proteins by immunoprecipitation. (a) Immunoprecipitation with anti-LAP antibody of supernatants from LAP-IFN-β- (lanes 1, 3, 5) and IFN-β-LAP-transfected CHO cells (lanes 2, 4, 6) that were either not treated (controls; lanes 1, 2), treated with MMP3 (lanes 3, 4) or treated with MMP1 (lanes 5, 6) overnight before SDS-PAGE. SDS-PAGE was carried out under denaturing conditions. Locations of LTBP and fusion proteins are indicated. Asterisks (*) indicate MMP cleavage products. MMP3 appears to cleave LTBP (see also Fig. 3c, lanes 3, 7). Positions of the molecular weight markers (M.W.) in kilodaltons are shown. (b,c) Immunoprecipitation of MTX-selected CHO cell supernatants with anti-LAP and anti-IFN-β antibodies after cleavage with MMP1, MMP3 and synovial fluid. Supernatants from MTX-selected cells were treated with MMPs or synovial fluid (1:5) overnight, and the reactions were stopped with 10 mM EDTA and then immunoprecipitated. CHO cells transfected with LAP-IFN-β (b) or IFN-β-LAP (c), either untreated (lanes 1, 5) or treated with MMP1 (lanes 2, 6), with MMP3 (lanes 3, 7) or with synovial fluid from a rheumatoid arthritis patient (lanes 4, 8). Immunoprecipitation was done with anti-LAP (lanes 1–4) and anti-IFN-β monoclonal antibody (lanes 5–8). The positions of LTBP and fusion proteins are indicated by arrows. Asterisks (*) indicate MMP cleavage products. The new MMP3 cleavage product clearly detected in the IFN-β-LAP samples (b, lanes 3, 7) appears to be a cleavage product of LTBP. A similar pattern of LTBP cleavage appears in Figure 2a (lanes 3, 4). Positions of LAP and IFN-β were independently assessed by western blotting and are indicated.
220
97
LTBP
66
*
IFN-β-LAP
* 45
LAP IFN-β
30
* 21
Possibly the concentration step removed some natural inhibitors of MMPs (tissue inhibitor of metalloproteinase), facilitating their activity. Addition of MMP1 to concentrated supernatants slightly increased the biological activity, as did addition of both MMP1 and pro-MMP9 or MMP3 and pro-MMP9 (data not shown). Interestingly, treatment of IFN-β-LAP with MMP1 and pro-MMP9 led to three- and sixfold increases in antiviral activity, respectively (from 1,300 U/ml to 3,480 and 7,740 U/ml), indicating that further activation of this molecule is possible. Using nonconcentrated supernatants from MTX-amplified cells, we showed that both MMP1 and MMP3 activated LAP-IFN-β by 21and 32-fold, respectively, and that synovial fluid from rheumatoid arthritis patients activated it by up to 4-fold (Table 1). Synovial fluid from patients with rheumatoid arthritis also cleaved the fusion proteins into discrete products of expected size (Fig. 2b,c, lanes 4 and 8). The sensitivity of the two fusion proteins to the presence of MMP9 is different: IFN-β-LAP can be activated, whereas LAP-IFN-β appears
1316
to be inhibited, perhaps inducing its further degradation by other enzymes present in the CHO cell supernatants. Disulfide bonds are required for latency To assess whether the latency detected with LAP-IFN-β required the formation of a putative closed ‘shell’ structure bounded by the double disulfide–linked LAP, we constructed a fusion protein using the porcine LAP in which Cys223 and Cys225 were mutated to serines. We compared this PorcLAP-IFN-β to the other constructs with respect to its biological activity in vitro after transient transfection into COS-7 cells. PorcLAP-IFN-β was as active as IFN-β-LAP in this assay (256 U/ml) (representative result of two independent experiments done in duplicate) whereas LAP-IFN-β did not show any biological activity. This result demonstrated that the ‘shell’ structure requires the double disulfide bond of LAP. Activation of latent IFN-β with fluids from inflamed sites To determine whether long-term incubation of the fusion proteins
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
ARTICLES denatured protein was purified on nickel chelate columns (data not shown). However, 2,500 2,500 its resuspension and folding were unsuccessful. Finally, serum-free supernatants from 2,000 2,000 CHO cells expressing LAP-IFN-β and amplified in 500 nM MTX were supplemented with 1,500 1,500 EDTA to avoid degradation by endogenous 1,000 1,000 MMPs; the fusion protein was purified by affinity chromatography on goat anti–human 500 500 LAP antibodies crosslinked to a solid support (Fig. 4a). The eluted single peak contained 0 0 98% pure LAP-IFN-β as assessed by western 0 24 48 72 96 120 0 24 48 72 96 120 blotting (Fig. 4b) and iodination (Fig. 4c). Time (h) Time (h) The iodinated protein was then injected Figure 3 Kinetics of IFN activity after incubation of cell supernatants in medium alone or with synovial intraperitoneally into Wistar rats, and blood fluid from rheumatoid arthritis patients. (a) LAP-IFN-β-transfected cells. (b) IFN-β-LAP-transfected samples were drawn at the specified times cells. Samples were incubated for up to 5 d at 37 °C and then applied to the IFN assay. Open symbols (Fig. 4d). The average half-life of the protein indicate samples incubated in medium with 10% FBS, and filled symbols indicate samples incubated with 1:5 (vol/vol) rheumatoid arthritis patient–derived synovial fluid (RASF). The RASF is different than in blood was 55.02 h. The distribution of the one used in Table 1 or Figure 2b,c. Data is representative of two independent experiments. radioactivity in tissues, as compared to the values in blood, were lower in heart (0.44fold) and lungs (0.76-fold), and higher in would lead to their degradation or accumulation into active com- liver, spleen (1.3-fold) and kidneys (4-fold). These data suggest that pound, we incubated both LAP-IFN-β and IFN-β-LAP for 5 d in the the protein is secreted mainly via the urine. presence or absence of synovial fluid from a rheumatoid arthritis patient and took samples at 24-h intervals. Incubation resulted in up Gene delivery of LAP-IFN-β is the most efficient treatment of CIA to tenfold increased activity of the LAP-IFN-β during the first 24–48 h, We have previously shown that IFN-β gene delivery has therapeutic followed by a steady decrease afterward (Fig. 3). The IFN-β-LAP was effects in models of multiple sclerosis31 and rheumatoid arthritis32. not activated; only a decrease in its activity was seen. These data indi- Thus, we assessed whether in vitro data obtained with our fusion procate the conformation LAP-IFN-β is protected from degradation dur- teins could be verified and further substantiated in vivo. We determined ing its incubation in serum-containing medium and can be activated that our plasmid constructs could affect established arthritis when by synovial fluid of rheumatoid arthritis patients, and that in contrast, delivered by a single intramuscular injection of DNA. The control plasthe IFN-β-LAP conformation is very sensitive to degradation in both. mid without insert (pcDNA3) has no effect on arthritis development as To assess whether samples from patients with other pathological assessed from paw swelling (Fig. 5a) or clinical score (Fig. 5b). On the inflammatory conditions could activate LAP-IFN-β, we tested, using a other hand, both the active IFN-β-LAP and PorcLAP-IFN-β inhibited blind study design, synovial fluid or CSF and paired sera from patients established disease. More importantly, the latent construct, LAP-IFN-β, with osteoarthritis or neurological diseases. Serum samples did not was more potent as a therapeutic agent in inhibiting paw swelling and activate the latent cytokine above background. Thus, fluid samples reducing clinical score (P < 0.05 from day 8 onward as compared to causing more than fourfold greater IFN biological activity than the pcDNA3). Similar results were obtained in an independent experiment paired serum were considered positive. After our collaborators dis- using 10 µg DNA (data not shown). closed the origins of the samples (Table 2), we determined that three of eight human CSF samples tested were positive, and these correlated DISCUSSION with four samples from patients with oligoclonal antibodies in the We have shown that an active cytokine molecule could be designed to CSF. Of these samples, three were from patients with multiple sclerosis become ‘latent’ by addition of the latency domain of TGF-β at either and one from a patient with meningitis. We were unable to detect its N or C terminus. The LAP domain of TGF-β conferred ‘latency’ to activity in one of the multiple sclerosis samples. Four CSF samples IFN-β, which could be abrogated by incubating the fusion protein were negative and all were from patients with noninflammatory neu- with recombinant MMPs. The latency is a result of steric hindrance rological conditions. Two of three synovial fluid samples from osteo- inhibiting the interaction between IFN-β and its cellular receptors. arthritis patients were positive in this assay. The sensitivity of the Despite the fact that both the N and C termini of the molecule are in activation in vitro could have varied because of differing storage conditions as well as the extent of disease activity at the time of sampling. Table 2 Activation of LAP-IFN-β with fluids from patients with In addition, three parallel samples of serum and CSF from rhesus inflammatory diseases monkeys with experimental autoimmune encephalomyelitis (EAE) or Number of positive samples/total samples collagen-induced arthritis (CIA) were tested for activation of LAP- Sample (fluid type) IFN-β. CSF from monkeys with CIA did not activate the latent moleMultiple sclerosis (CSF) 2/3 cule, whereas CSF from monkeys with EAE did in all samples. Neither Meningitis (CSF) 1/1 EAE nor CSF serum samples activated LAP-IFN above background.
a
3,000
LAP-IFN
b
3,000
IFN -LAP IFN -LAP + RASF
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
IFN activity (U/ml)
IFN activity (U/ml)
LAP-IFN +RASF
Affinity purification of LAP-IFN-β and pharmacokinetic studies Attempts to purify the fusion protein by conventional chromatographic procedures failed. Hence, the protein was tagged with a polyhistidine tail and expressed in inclusion bodies in Escherichia coli. The
Noninflammatory CNS disease (CSF)
0/4
Osteoarthritis (SF)
2/3
Paired samples of serum, cerebrospinal fluid (CSF; 1:5, vol/vol) or synovial fluid (SF) – were incubated with LAP-IFN-β overnight at 37 °C and assayed on LTK cells for IFN-β activity. Increased activity of CSF or SF over paired serum (greater than fourfold) was considered positive.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1317
ARTICLES a
Wash
b
Elution
150
75 kDa 100 50 0
160
165
170
175
180
185
190
Volume (ml)
d
100,000
LAP-IFN-β (CPM/gr)
c 75 kDa
10,000
1,000 0
20
40
60
80
Time (h)
Figure 4 Affinity chromatography of LAP-IFN-β and pharmacokinetics. (a) 280 nm absorbance profile of fractions collected; arrows indicate when loading, washing and elution were done. (b) Western blot of peak fraction with anti-IFN-β antibody. (c) Autoradiograph of 125I-labeled LAP-IFN-β from peak fraction. (d) Pharmacokinetic data from two Wistar rats injected intraperitoneally with iodinated LAP-IFN-β. Plot of c.p.m. iodinated LAP-IFN-β/g blood (log scale) were measured from 1 h to 78 h.
close proximity in the crystal structure of IFN-β33, it is clear that a better ‘shell’ is conferred by fusing the LAP domain at its N terminus, as it is found in native TGF-β. It is interesting to note that the ‘shell’ structure can encapsidate a substantially bigger IFN-β (32 kDa) than the native TGF-β (12.5 kDa). The MMP site located between LAP and IFN-β could be cleaved in vitro by MMP-3 and MMP-1. This is not surprising, as these MMPs have homologous regions in their active site34. Other MMPs could also cleave this site as shown by the activation occurring in concentrated serum-free supernatants of CHO cells or fluids from inflamed sites. MMP9 could not cleave our fusion proteins. Using fluorogenic peptide
substrates with the sequence PLGLWA-D-R, the value of the rate of hydrolysis (kcat/KM) of MMPs appears to follow the order MMP9 > MMP2 > MMP7 > MMP3 > MMP1 (ref. 26). This discrepancy in hydrolysis sensitivity between the peptide substrate and our engineered proteins may be related to their tertiary structures. The ‘latent’ cytokine design has a longer half-life. LAP-containing TGF-β has a longer half-life than free TGF-β in vivo35. We showed here that LAP-IFN has a half-life of 55.02 h when injected intraperitoneally in rats. Thus, the half-life of LAP-IFN is about 37.6 times longer than the reported half-life of IFN-β alone (1.46 h) and 5.08 times longer than that of improved pegylated IFN-β (10.82 h) injected by the same route36. The increase in the half-life of IFN-β (37.6-fold) in the context of LAP-IFN is of the same order of magnitude (30- to 50-fold) as that reported for TGF-β when it was associated with LAP in vitro before injection35. Thus, because of this longer half-life, the cytokine could be administered systemically using lower dosages. Secondly, the cytokine will not be released to interact with cellular receptors unless inflammatory or tissue-remodeling processes involving MMP activity are taking place. Expression of MMPs is very tightly regulated37. MMP activity is found in osteoarthritis, rheumatoid arthritis38–40 and other chronic diseases such as inflammatory bowel disease41,42, multiple sclerosis43,44, atherosclerosis45 and cancer during its invasive phase46. Thus, therapeutic agents, engineered as LAP-IFN-β was, could be used to treat these conditions. We found that, when delivered by gene therapy using intramuscular injection, the latent cytokine was more effective than the active counterparts in the treatment of established arthritis. Upon cleavage, the release of LAP could have antagonistic effects for TGF-β, as LAP can inhibit active TGF-β action in vitro47. We expect, however, that our LAP fusion protein will act at sites of inflammation where free radicals abound. Nitrosylation of LAP disables its capacity for binding to TGF-β48. Thus, it is unlikely that, in sites of inflammation, the released LAP will antagonize TGF-β function. The applications of this targeting approach for biologically active compounds are broad. Moreover, additional modifications to the MMP cleavage site may provide additional tissue or disease specificity. Such engineered latent cytokines, having a longer half-life and increased specificity, will help in reducing the cost of cytokine treatment for patients, as they will be more efficient at lower concentrations than the free cytokines used today. METHODS
b
32.5
12.5 IFN-β-LAP
Arthritis score (scale 1–12)
a Hind paw swelling (mm × 10 –1)
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Absorbance (280 nm)
Loading 200
27.5
22.5
17.5
LAP-IFN-β
10
por-IFN-β 7.5
pCDNA3
5
2.5
0 0
5
10
15
Days from onset
20
0
5
10
15
20
Days from onset
Figure 5 Inhibition of established collagen-induced arthritis by DNA injection with LAP-IFN-β. (a,b) Male DBA/1 mice (n = 4 per group) were immunized with CII and, at onset of disease (clinical score of 1), injected once intramuscularly with plasmid DNA as shown. Hind paw swelling (a) and arthritis score (b) were assessed as described in Methods. Data from one independent representative experiment of two are shown. The data shown are the mean ± s.e.m. Where not seen, the s.e.m. bars are smaller than the symbol used.
1318
Cloning of GS-MMP-GS linker into EcoRI-NotI sites of pcDNA3. A double-stranded deoxyoligonucleotide coding for GGGGSPLGLWAGGGS was designed as follows: sense: 5′-AATTCGGGGGAG GCGGATCCCCGCTCGGGCTTTGGGCGGG AGGGGGCTCAGC-3′; antisense: 5′-GGCCGCT GAGCCCCCTCCCGCCCAAAGCCCGAGC GGGGATCCGCCTCCCCCG -3′. Annealed deoxyoligonucleotides were cloned into pcDNA3 cleaved with EcoRI and NotI. The correct clone was named GS-MMP-GS. Construction of TGF-β-LAP at the N terminus followed by GS-MMP-GS and mature IFN-β. Human TGF-β LAP as 5′ unit with its signal peptide and with HindIII and EcoRI ends was cloned by PCR from plasmid TGF-β-Babe neo49. The following primers were used: sense, 5′-CCAAGCTTATGC CGCCCTCCGGGCTGCGG-3′; antisense, 5′-CCG AATTCGCTTTGCAGATGCTGGGCCCT-3′. The 820-base pair (bp) product was cloned into GS-
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
ARTICLES
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
MMP-GS plasmid cut with HindIII and EcoRI. The clone was named TGF-βGS-MMP-GS linker. Mature IFN-β with 5′ NotI and 3′ XbaI sites was synthesized by PCR from clone Aphrodite50 using the following primers: sense, 5′-CGCGGCCGCAATCAACTATAAGCAGCTCCAG-3′; antisense, 5′-GGTC TAGATCAGTTTTGGAAGTTTCTGGTAAG-3′. It was then cloned into NotI and XbaI sites of TGF-β-GS-MMP-GS linker plasmid. The name of the correct clone was LAP-IFN-β. Construction of IFN-β at the N terminus followed by GS-MMP-GS and mature TGF-β-LAP. Pre-IFN-β with signal peptide and without stop codon was synthesized by PCR as above using the following primers: sense, 5′-CCAA GCTTATGAACAACAGGTGGATCCTC-3′; antisense, 5′-CCGAATTCGTTTT GGAAGTTTCTGGTAAG-3′. The DNA product was cloned into plasmid pcDNA3 GS-MMP-GS digested with HindIII and EcoRI. The clone was named IFN-β + GS-MMP-GS linker. Mature TGF-β-LAP with stop codon was synthesized by PCR as above using the following primers: sense, 5′-CGCGGCCGC ACTATCCACCTGCAAGACTATC-3′; antisense, 5′-GGTCTAGATCAGCTTT GCAGATGCTGGGCCCT-3′. The DNA fragment was cloned into plasmid IFN-β-GS-MMP-GS linker digested with NotI and XbaI to obtain the fusion protein. The correct clone was named IFN-β-LAP. Cloning of porcine LAP in front of IFN-β. The doubly mutated porcine TGFβ1 cDNA Cys223 and Cys225 to Ser223 and Ser225 as plasmid pPK14 (ref. 13), was kindly provided by P.J. Wirth, National Institutes of Health, Bethesda, Maryland, USA. Cloning was done by PCR, using the following set of primers: sense (starting at signal peptide), 5′-CGCCCATGGCGCCTTCGGGGCCT-3′ (this primer has a modified sequence around the initiator ATG to create a NcoI site); antisense, 5′-CCGAATTCGCTGTGCAGGTGCTGGGCCCT-3′. The PCR product was blunted with the Klenow fragment of DNA polymerase, cut with EcoRI, and then cloned into LAP-IFN-β plasmid that had been cut with HindIII (filled-in) and cut with EcoRI (replacing human LAP). The clone was named PorcLAP-IFN-β. Permanent transfection into DHFR− CHO cells. DHFR− CHO cells were maintained and permanently transfected with plasmids (20 µg) expressing LAP-IFN-β or IFN-β-LAP, each linearized with PvuI and ligated separately with PvuI-cut pSV2DHFR (1 µg)51. The DNA was added as 1 ml calcium phosphate coprecipitate on CHO cells and selected as described51 with the addition of 1 mg/ml G418. For gene amplification, cells were selected with MTX at 50 nM (LAP-IFN-β) or 12.5 nM (IFN-β-LAP), respectively. Further amplification of LAP-IFN-β in 500 nM MTX facilitated its purification by affinity chromatography. Transient transfection into monkey COS-7 cells. Plasmid DNA (20 µg) was transfected as described above, in duplicate, into 0.5 × 106 COS-7 cells. The supernatants were collected for IFN assay 48 h after glycerol shock. IFN-β biological assay. Mouse IFN-β biological activity was assessed on mouse LTK or L929 cells by inhibition of the cytopathic effect of EMC virus (kindly provided by I. Kerr, Imperial Cancer Research Fund, London) as described50. Where indicated, serum-free CHO cell supernatants were concentrated using filters with a cut off of 30.0 kDa. Metabolic labeling of CHO cells. Permanently transfected or nontransfected CHO cells were washed with cysteine- and methionine-free medium containing 10% dialyzed FBS and supplemented with thymidine, glutamine, penicillin, streptomycin and L-proline. Labeling was carried out overnight or for 48 h in the presence of a [35S]methionine-[35S]cysteine mix at 1 Ci/mmol using 250 mCi in 5 ml medium. Supernatants were collected and supplemented where indicated with serine-protease inhibitors (SPI) (pepstatin-A at 10 µg/ml, aprotinin at 1 µg/ml, chymostatin at 10 µg/ml, leupeptin at 10 µg/ml and AEBSF (4-(2aminoethyl) benzene sulfonyl fluoride, HCl) at 200 µM. Immunoprecipitation. 35S-labeled supernatants were precleared with protein G–Sepharose in PBS with 0.1% NP-40. Supernatants were incubated with goat anti–human-LAP antibody (at 0.9 µg/ml; R&D Systems) or monoclonal rat anti–IFN-β (monoclonal 7F-D3 at a dilution of 1:250; Yamasa) for 3–4 h at 4 °C. The antigen-antibody complexes were bound to protein G–Sepharose overnight at 4 °C and washed three times with 5 ml 0.1% NP-40 in PBS.
Proteins bound to beads were directly resuspended in Laemmli loading buffer (Tris-HCl 0.1 M, pH 6.8, 5% SDS wt/wt, 0.1% bromophenol blue wt/vol, 50% glycerol wt/vol) or used in MMP reactions before electrophoresis on a 10% SDS-polyacrylamide gel. Alternatively, supernatants were treated with MMPs as described below and then immunoprecipitated. Gels were fixed and treated with 1 M sodium salicylate before being dried and exposed to autoradiography. MMP digestion. Recombinants pro-MMP9 (kindly provided by R. Fridman, Wayne State University, Detroit, Michigan, USA) or active MMP1 and MMP3 (kindly provided by H. Nagase, Kennedy Institute of Rheumatology, Imperial College, London) either were incubated overnight at 37 °C with immunoprecipitated supernatants from CHO cells in 20 mM Tris-HCl, pH 7.4, 5 mM CaCl2, 140 mM NaCl and 0.1% Brij 35 in 50 µl at 1 µg/ml or were directly added to cell supernatants (at 4 µg/ ml). APMA at 10 µM was used in certain experiments to activate pro-MMP9 overnight at 37 °C52. Purification of LAP-IFN-β by affinity chromatography. Goat polyclonal, affinity-purified anti–human LAP antibody (500 µg; R&D Systems) was bound to CarboLink (Perbio Science) by chemical crosslinking using manufacturer’s instructions. We obtained 70% crosslinking of the antibody in 2 ml of resin. Supernatant (168 ml) from CHO cells expressing LAP-IFN-β and amplified with 500 nM MTX in CD medium (Invitrogen Life Technologies) was loaded onto the column. Before loading, using fast performance liquid chromatography (AKTA), the CD medium was supplemented with 5 mM EDTA (pH 8.0) at a flow rate of 0.22 ml/min. Bound material was washed with 10 mM phosphate, 140 mM NaCl and 5 mM EDTA, pH 6.8, and eluted with 100 mM glycine and 140 mM NaCl, pH 2.5. Fractions of 1 ml were collected and their pH neutralized with 50 µl of 1 M phosphate buffer, 5 mM EDTA, pH 8.0, and were later dialyzed against PBS at 4 °C. After analysis by 10% SDS-PAGE on Hybond-P membranes (Amersham Biosciences), the presence of LAP-IFN was assessed by western blotting using 1% casein (Fisher Scientific) as blocking agent in Trisbuffered saline (10 mM Tris, 150 mM NaCl, pH 7.5) with 0.2 % Triton X-100 and 0.05% Tween 20 (vol/vol). Next, the blots were probed with a primary rat anti-IFN-β antibody (at 1/1,000) and a secondary goat anti–rat IgG F(ab′)2HRP antibody (at 1:1,000; ImmunoPure, Perbio Science). Using anti-LAP (at 1:1,000; R&D Systems), westerns were subjected to detection using as secondary antibody mouse anti-goat-IgG–horseradish peroxidase conjugate (at 1:1,000; Santa Cruz Biotechnology). The western blots were developed using enhanced chemiluminescence (ECL) reagents (Amersham Biosciences) and exposed to autoradiography using Hyperfilm (Amersham Biosciences). Films were developed using an AGFA Curix 60 developer (Gevaert). Protein iodination and pharmacokinetic studies. Purified LAP-IFN (7 µg) was iodinated in PBS using iodogen and 125NaI (3.7 GBq/ml; Amersham Biosciences) in a final volume of 400 µl for 10 min in ice. The reaction was quenched with 2.5 mM L-tyrosine and after 10 min BSA was added to a final concentration of 0.2 mg/ml. The solution was centrifuged through a 10-ml packed G-25 Sephadex spin column equilibrated in PBS with 1 mg/ml BSA at 5,000g for 1 min. The protein was eluted in 1 ml and had a specific activity of 30,312 c.p.m./µl as assessed by counting on a γ counter (LKB Wallac 1282, Pharmacia). The labeled protein was assessed for integrity and correct molecular weight by SDS-PAGE and autoradiography. Wistar rats (A. Tuck & Son) (n = 2) weighing more than 300 g were injected intraperitoneally with either 400 or 500 µl of iodinated protein and blood samples taken by tail vein puncture at the times indicated. All samples were weighed to correct c.p.m. per sample size. At the end of the experiment organs were also removed, weighed and counted as above. Pharmacokinetic data for each rat were analyzed using the program Kinetica (http://www.innaphase.com). The half-life reported is the average of the data obtained from each rat. CIA and plasmid DNA injection. All animals used in this study were kept according to Institutional and Home Office guidelines. DBA/1 mice were immunized with collagen type II (CII) as described53 and 3 weeks later were boosted with 100 µg CII in incomplete Freund’s adjuvant. We injected mice with 100 µg plasmid DNA in PBS intramuscularly at the onset of arthritis (arthritis score = 1). Subsequently, mice were scored every other day by a blinded observer for clinical arthritis. Each paw was given a clinical score from 0 to 3 as follows: 0 = normal; 1 = slight swelling and/or erythema; 3 = pronounc-
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1319
ARTICLES
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ed edematous swelling; 3 = ankylosis. Inflammation was assessed by hind paw swelling measured with calipers as previously described52. Statistical analysis was done by the nonparametric Mann-Whitney test. ACKNOWLEDGMENTS This work was funded by the Multiple Sclerosis Society of Great Britain and Northern Ireland, the Arthritis Research Campaign (UK) and Kinetique Biomedical Research Seed Fund UK. We thank Irene Theoharidou for her expert secretarial assistance, G. Scott and G. Giovannoni for clinical samples, S. Amor for samples from rhesus monkeys, S. Mather and D. Ellison for help with protein iodination, J. Wilson and A. Mustafa for technical help, A. Johnston for the analysis of pharmacokinetic data, L. Layward, D. Willoughby, O.L. Podhajcer and P. Colville-Nash for reviewing the manuscript and H. Nagase for helpful and encouraging discussions. COMPETING INTERESTS STATEMENT The authors declare competing financial interests (see the Nature Biotechnology website for details). Received 11 June; accepted 21 August 2003 Published online at http://www.nature.com/naturebiotechnology/ 1. Waldmann, T.A. & Tagaya, Y. The multifaceted regulation of IL-15 expression and the role of this cytokine in NK cell differentiation and host response to intracellular pathogens. Annu. Rev. Immunol. 17, 19–49 (1999). 2. Taniguchi, T. Regulation of cytokine gene expression. Annu. Rev. Immunol. 6, 439–464 (1988). 3. Taipale, J. & Keski-Oja, J. Growth factors in the extracellular matrix. FASEB J. 11, 51–59 (1997). 4. Khalil, N. TGF-β: from latent to active. Microbes Infect. 1, 1255–1263 (1999). 5. Aulitzky, W., Schuler, M., Peschel, C. & Huber, C. Interleukins: clinical pharmacology and therapeutic use. Drugs 48, 667–677 (1994). 6. Gutterman, J. Cytokine therapeutics: lessons from IFN α. Proc. Natl. Acad. Sci. USA 91, 1198–1205 (1994). 7. Golab, J. & Zagozdon, R. Antitumor effects of IL-12 in pre-clinical and early clinical studies. Int. J. Mol. Med. 3, 537–544 (1999). 8. Atkins, M. et al. High dose recombinant IL-2 therapy for patients with metastatic melanoma: analysis of 270 patients treated between 1985 and 1993. J. Clin. Oncol. 17, 2105–2116 (1999). 9. Margolin, K.A. Interleukin-2 in the treatment of renal cancer. Semin. Oncol. 27, 194–203 (2000). 10. Lode, H.N., Xiang, R., Becker, J.C., Gillies, S.D. & Reisfeld, R.A. Immunocytokines: a promising approach to cancer immunotherapy. Pharmacol. Ther. 80, 277–292 (1998). 11. Han, B., Hall, F.L. & Nimni, M.E. Refolding of a recombinant collagen-targeted TGF β 2 fusion protein expressed in E. coli. Protein Expr. Purif. 11, 169–178 (1997). 12. Gordon, E.M. et al. Capture and expansion of bone marrow-derived mesenchymal progenitor cells with a TGF β1-von Willebrand’s factor fusion protein for retrovirus-mediated delivery of coagulation factor IX. Hum. Gene Ther. 8, 1385–1394 (1997). 13. Sanderson, N. et al. Hepatic expression of mature TGF-β1 in transgenic mice results in multiple tissue lesions. Proc. Natl. Acad. Sci. USA 92, 2572–2576 (1995). 14. Brunner, A. et al. Site-directed mutagenesis of glycosylation sites in the transforming growth factor-β 1 (TGF β 1) and TGF-β 2 (414) precursors and of cysteine residues within mature TGF β 1: effects on secretion and bioactivity. Mol. Endocrinol. 6, 1691–1700 (1992). 15. Brunner, A., Marquardt, H., Malacko, A., Lioubin, M. & Purchio, A. Site-directed mutagenesis of cysteine residues in the pro region of the TGF β 1 precursor. J. Biol. Chem. 264, 13660–13664 (1989). 16. Munger, J.S., Harpel, J.G., Giancotti, F.G. & Rifkin, D.B. Interactions between growth factors and integrins: latent forms of TGF- β are ligands for the integrin αVβ 1. Mol. Cell Biol. 9, 2627–2638 (1998). 17. Derynck, R. TGF-β-receptor-mediated signaling. TIBS 19, 548–553 (1994). 18. Saharinen, J., Taipale, J. & Keski-Oja, J. Association of the small latent TGF-β with an eight cysteine repeat of its binding protein LTBP-1. EMBO J. 15, 245–253 (1996). 19. Schultz-Cherry, S., Lawler, J. & Murphy-Ullrich, J.E. The Type 1 repeats of thrombospondin 1 activate latent TGF-β. J. Biol. Chem. 269, 26783–26788 (1994). 20. Crawford, S.E. et al. Thrombospondin-1 is a major activator of TGF-β 1 in vivo. Cell 93, 1159–1170 (1998). 21. Nunes, I., Gleizes, P.-E., Metz, C.N. & Rifkin, D.B. Latent TGF-β binding protein domains involved in activation and transglutaminase-dependent cross-linking of latent TGF-β. J. Cell Biol. 136, 1151–1163 (1997). 22. Kojima, S., Nara, K. & Rifkin, D.B. Requirement for transglutaminase in the activation of latent TGF-β in bovine endothelial cells. J. Cell Biol. 121, 439–448 (1993). 23. Yu, Q. & Stamenkovic, I. Cell surface-localized matrix metalloproteinase-9 proteolytically activates TGF-β and promotes tumor invasion and angiogenesis. Genes Dev. 14, 163–176 (2000). 24. Peng, K.-W., Morling, F.J., Cosset, F.-L., Murphy, G. & Russell, S.J. A gene delivery system activable by disease-associated matrix metalloproteinases. Hum. Gene Ther. 8, 729–738 (1997).
1320
25. Ye, Q.Z., Johnson, L.L., Yu, A.E. & Hupe, D. Reconstructed 19 kDa catalytic domain of gelatinase A is an active proteinase. Biochemistry 34, 4702–4708 (1995). 26. Nagase, H. & Fields, G.B. Human matrix metalloproteinase specificity studies using collagen sequence-based synthetic peptides. Biopolymers 40, 399–416 (1996). 27. Saharinen, J., Hyyatiäinen, M., Taipale, J. & Keski-Oja, J. Latent TGF-β binding proteins (LTBPs)-structural extracellular matrix proteins for targeting TGF-β action. Cytokine Growth Factor Rev. 10, 99–117 (1999). 28. Goldman, M.H., James, D.C., Ison, A.P. & Bull, A.T. Monitoring proteolysis of recombinant human IFN-γ during batch culture of Chinese hamster ovary cells. Cytotechnology 23, 103–111 (1997). 29. Satoh, M., Hosoi, S., Miyaji, H., Itoh, S. & Sato, S. Stable production of recombinant pro-urokinase by human lymphoblastoid Namalwa KJM-1 cells: host-cell dependency of the expressed-protein stability. Cytotechnology 13, 79–88 (1993). 30. Masure, S. et al. Production and characterisation of recombinant active mouse gelatinase B from eukaryotic cells and in vivo effects after intravenous administration. Eur. J. Biochem. 244, 21–30 (1997). 31. Croxford, J.L. et al. Cytokine gene therapy in experimental allergic encephalomyelitis by injection of plasmid DNA-cationic liposome complex into the central nervous system. J. Immunol. 160, 5181–5187 (1998). 32. Triantaphyllopoulos, K.A., Williams, R.O., Tailor, H. & Chernajovsky, Y. Amelioration of collagen-induced arthritis and suppression of IFN-γ, IL-12, and TNF-α production by IF- β gene therapy. Arth. Rheum. 42, 90–99 (1999). 33. Karpusas, M. et al. The crystal structure of human IFN β at 2.2-Å resolution. Biochemistry 94, 11813–11818 (1997). 34. Massova, I., Fridman, R. & Mobashery, S. Structural insights into the catalytic domains of human matrix metalloprotease-2 and human matrix metalloprotease-9: implications for substrate specificities. J. Mol. Model. 3, 17–30 (1997). 35. Wakefield, L.M. et al. Recombinant latent TGF-β1 has a longer plasma half life in rats than active TGF-β1, and a different tissue distribution. J. Clin. Invest. 86, 1976–1984 (1990). 36. Pepinsky, R. et al. Improved pharmacokinetic properties of a polyethylene glycolmodified form of IFN-β-1-a with preserved in vitro bioactivity. J. Pharmacol. Exp. Ther. 297, 1059–1066 (2001). 37. Han, Z., Boyle, D.L., Manning, A.M. & Firestein, G.S. AP-1 and NF-κB regulation in rheumatoid arthritis and murine collagen-induced arthritis. Autoimmunity 28, 197–208 (1998). 38. van Meurs, J. et al. Cleavage of aggrecan at the Asn341-Phe342 site coincides with the initiation of collagen damage in murine antigen-induced arthritis: a pivotal role for stromelysin 1 in matrix metalloproteinase activity. Arth. Rheum. 42, 2074–2084 (1999). 39. Singer, I. et al. Aggrecanase and metalloproteinase-specific aggrecan neo-epitopes are induced in the articular cartilage of mice with collagen II-induced arthritis. Osteoarth. Cartil. 5, 407–418 (1997). 40. Kubota, E., Imamura, H., Kubota, T., Shibata, T. & Murakami, K. Interleukin 1β and stromelysin (MMP3) activity of synovial fluid as possible markers of osteoarthritis in the temporomandibular joint. J. Oral Maxillofac. Surg. 55, 20–27 (1997). 41. Louis, E. et al. Increased production of matrix metalloproteinase-3 and tissue inhibitor of metalloproteinase-1 by inflamed mucosa in inflammatory bowel disease. Clin. Exp. Immunol. 120, 241–246 (2000). 42. Baugh, M. et al. Matrix metalloproteinase levels are elevated in inflammatory bowel disease. Gastroenterology 117, 814–822 (1999). 43. Leppert, D. et al. Matrix metalloproteinase-9 (gelatinase B) is selectively elevated in CSF during relapses and stable phases of multiple sclerosis. Brain 121, 2327–2334 (1998). 44. Anthony, D.C. et al. Differential matrix metalloproteinase expression in cases of multiple sclerosis and stroke. Neuropath. Appl. Neurobiol. 23, 406–415 (1997). 45. Libby, P. The interface of atherosclerosis and thrombosis: basic mechanisms. Vasc. Med. 3, 225–229 (1998). 46. DeClerck, Y.A. et al. Inhibition of invasion and metastasis in cells transfected with an inhibitor of metalloproteinases. Cancer Res. 52, 701–708 (1992). 47. Wakefield, L. et al. Recombinant TGF-β1 is synthesized as a two-component latent complex that shares some structural features with the native platelet latent TGF-β1 complex. Growth Factors 1, 203–218 (1989). 48. Vodovotz, Y. et al. Regulation of TGF β1 by nitric oxide. Cancer Res. 59, 2142–2149 (1999). 49. Chernajovsky, Y., Adams, G., Triantaphyllopoulos, K., Ledda, M.F. & Podhajcer, O.L. Pathogenic lymphoid cells engineered to express TGF β1 ameliorate disease in a collagen-induced arthritis model. Gene Ther. 4, 553–559 (1997). 50. Triantaphyllopoulos, K., Croxford, J.L., Baker, D. & Chernajovsky, Y. Cloning and expression of murine IFN β and a TNF α antagonist for retrovirus gene therapy of experimental allergic encephalomyelitis. Gene Ther. 5, 253–263 (1998). 51. Chernajovsky, Y. et al. Efficient constitutive production of human fibroblast interferon by hamster cells transformed with the IFN-β 1 gene fused to an SV40 early promoter. DNA 3, 297–308 (1984). 52. Ogata, Y., Itoh, Y. & Nagase, H. Steps involved in activation of the pro-matrix metalloprotease 9 (progelatinase B)-tissue inhibitor of metalloproteinases-1 complex by 4aminophenylmercuric acetate and proteinases. J. Biol. Chem. 270, 18506–18511 (1995). 53. Dreja, H., Annenkov, A. & Chernajovsky, Y. A novel truncated form of human soluble complement receptor 1 delivered by gene therapy prevents and ameliorates collageninduced arthritis by inhibiting B and T cell responses. Arth. Rheum. 43, 1698–1709 (2000).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES
A genetic approach to inactivating chemokine receptors using a modified viral protein V McNeil Coffield1,3, Qi Jiang3 & Lishan Su1–3 We have developed a genetic system, called degrakine, that specifically and stably inactivates chemokine receptors (CKR) by redirecting the ability of the HIV-1 protein, Vpu, to degrade CD4 in the endoplasmic reticulum (ER) via the host proteasome machinery. To harness Vpu’s proteolytic targeting capability to degrade new receptors, we fused a chemokine with the C terminal region of Vpu. The fusion protein, or degrakine, accumulates in the ER, trapping and functionally inactivating its target CKR. We have demonstrated that degrakines based on SDF-1 (CXCL12), MDC (CCL22) and RANTES (CCL5) specifically inactivate their respective receptor functions. Using a retroviral vector expressing the SDF-1 degrakine, we have established that CXCR4 is required for the homing of hematopoietic stem/progenitor cells (HSPC) to the bone marrow immediately after transplantation. Thus the degrakine provides an effective genetic tool to dissect receptor functions in a number of biological systems in vitro and in vivo.
CKRs are G protein–coupled receptors (GPCRs) that have critical roles in controlling the migration of HSPCs and mature leukocytes and that participate in important pathophysiological processes including inflammation, allograft rejection, autoimmunity, viral infection and lymphoid development1–5. Approximately 20 CKRs and over 40 chemokines are encoded in the mouse or human genome1,2. Chemokines are subdivided into CXC (alpha), CC (beta), C (gamma) and CX3C (delta) classes, depending on the position of conserved cystine residues in the polypeptide. Each CKR is specific to one or multiple chemokines, and each chemokine can bind to one or multiple CKRs. Despite recent interest in chemokines and CKRs, their involvement in HSPC and leukocyte migration remains poorly defined because of the large number of family members, cross-reactivity and a lack of efficient genetic tools to dissect their functions. Gene targeting in mice has been done for a limited number of CKRs, but it is time-consuming and expensive to generate knockout mice. In addition, it is difficult to breed mice with mutations in multiple CKRs because the genes encoding them are closely linked at two chromosomal loci2. A common approach to studying CKR function is through the use of blocking antibodies, an approach that has limitations including toxicity and the lack of target specificity and in vivo efficacy6. Another method, the intrakine system, inhibits the level of CKR surface expression by sequestration of the target CKR in the ER7. However, the intrakine seems to have a paracrine activity, probably as a result of secretion of the intrakine protein out of the cell8, and this method may also lead to toxic accumulation of intrakine-CKR protein complexes in the ER. More recently, RNA interference technology has been used to knock down gene function in vitro and in vivo9–13. Many viruses have intricate, naturally evolved mechanisms to interact with receptors on infected cells and thereby elude the host’s immune
system and facilitate replication14,15. HIV-1 uses its envelope and Vpu proteins to specifically target CD4 for degradation in the ER14–16. The C terminal region of Vpu mediates this interaction by binding with the β-TrCP–Skp1 protein complex, which leads to CD4 ubiquitination and degradation16,17. Phosphorylation of the serine residues Ser52 and Ser56 in Vpu is critical for β-TrCP binding and CD4 degradation, although Vpu itself is resistant to ubiquitination and degradation16,18. To harness Vpu’s proteolytic targeting capability to degrade new receptors, we fused a chemokine with the C terminal region of Vpu. The chemokine–VpuC fusion protein (degrakine) is designed to specifically downregulate the surface expression and inhibit the function of its target CKR. Combining the degrakine with retroviral gene transfer, we have developed a genetic system to rapidly and specifically inactivate CKRs in either human or mouse cells in vitro and in vivo. The prototype degrakine is designed with the chemokine SDF-1α (CXCL12), which specifically binds CXCR4 (refs. 19–22). SDF-1 is expressed from a single gene, but has two isoforms, SDF-1α and SDF1β, that are generated by alternate splicing23. Although originally identified as a pre-B-cell growth factor expressed in mouse bone marrow21,23–25, SDF-1α elicits a potent migratory response from monocytes, T cells and HSPCs21,25,26. Mutations in either SDF-1 or CXCR4 in mice lead to embryonic lethality with neurological and hematological abnormalities27,28. Additionally, CXCR4 is required for stable engraftment of human HSPC cells into the bone marrow of NOD-SCID mice and of mouse fetal liver cells into lethally irradiated recipient mice28,29. However, it remains unclear whether CXCR4 mediates HSPC homing to the bone marrow, retention and proliferation within the bone marrow, or long-term maintenance of progenitor cells. Using the SDF-1 degrakine, we show that CXCR4 plays a role in HSPC homing to the bone marrow immediately after HSPC transplantation.
1Curriculum
in Genetics and Molecular Biology, 2Department of Microbiology and Immunology, and 322-048 Lineberger Cancer Center, School of Medicine, The University of North Carolina, Chapel Hill, NC 27599-7295, USA. Correspondence should be addressed to L.S. (
[email protected]). Published online 12 October 2003; doi:10.1038/nbt889
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1321
ARTICLES a
Membrane domain N term
C term
** S52/S56 Critical for the interaction between βTrCP and Vpu
CRBD
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
LTR
SDF-1 HA
VpuC
In vitro biology (cell lines)
b Lumen ER mem.
CX CR 4 CRBD (SDF-1)
VpuC
Cytosol WD F Box h-βTrCP
Staining of human osteosarcoma U2OS cells with anti-hemagglutinin showed that the SDF-1 degrakine colocalized with YFP-KDEL in the ER (Fig. 2a). SDF-1 contains a signal peptide sequence that directs its transport to the ER and into the secretory pathway21. Thus the SDF-1 degrakine could be secreted from the cell. To monitor this possibility, we analyzed human embryonic kidney 293T cells and SupT1 T-lymphocyte cells expressing the indicated vectors for SDF-1 secretion by enzyme-linked immunosorbent assay (ELISA). A control vector expressing SDF-1 was used to determine maximum secretion without ER retention. Additionally, a construct expressing the SDF-1 intrakine, SDF-1KDEL7, was used to demonstrate KDEL-mediated retention of SDF-1 within the ER. 293T cells transiently expressing SDF-1 and SDF-1KDEL both secreted SDF-1 efficiently into the medium (∼2.5 nM, Fig. 2b). In contrast, SDF-1VpuC and SDF-1VpuC2/6 secreted 66% less SDF-1 (Fig. 2b). In stably transduced SupT1 cells, the presence of SDF-1VpuC construct resulted in an 89% and 67% decrease in secreted SDF-1 as compared with SDF-1 and SDF-1KDEL, respectively (Fig. 2c). The SDF-1VpuC2/6 protein with mutations inhibiting Vpu’s interaction with βTrCP was secreted in amounts similar to those of SDF-1VpuC (Fig. 2b,c). Thus, the SDF-1 degrakine is retained within the cell and may contain an unidentified ER retention motif because no known ER retention signal has been identified in the VpuC domain. Expression of the SDF-1 degrakine did not affect cell viability or proliferation in a number of human cell lines, primary mouse thymocytes and mouse HSPCs. Comparable rates of proliferation were observed over 9 d in culture in parental SupT1 cells and cells transduced with vector or SDF-1VpuC (Fig. 2d). Cells transduced with vector or with SDF-1VpuC were also monitored by fluorescence-activated cell sorting (FACS) for EGFP expression at 2 d and 3 weeks after transduction. Constant levels of EGFP expression were observed at both time points (Fig. 2e), showing that the degrakine protein is not cytotoxic. In addition, stable expression of the degrakine was also observed in primary mouse bone marrow cells cultured for 12 d in vitro (Fig. 2f). Therefore, the degrakine fusion protein does not affect cell viability or proliferation in various types of cells.
Cytoplasmic domain (VpuC)
CMV
GFP
Ex vivo/in vivo biology (progenitor cells) CX CR 4 CRBD (SDF-1)
VpuC WD F Box
CX CR 4 CRBD (SDF-1)
VpuC WD F Box SKP1
SKP1
+ Ub
Proteasome
Figure 1 Design and function of the degrakine system. (a) The SDF-1 degrakine consists of SDF-1 fused to the C terminus of Vpu (aa28–81). The SDF-1 degrakine was cloned into a MSCV retroviral vector, which also encodes EGFP under the control of a CMV promoter. (b) The SDF-1 degrakine localizes to the ER, binds CXCR4 and prevents CXCR4 translocation to the cell surface. Additionally, VpuC interacts with β-TrCP and targets CXCR4 for degradation via the proteasome. HA, hemagglutinin.
RESULTS Degrakine design The degrakine protein comprises two domains. The C-terminal domain is aa28–81 (VpuC) of Vpu, which recruits the proteasome16. The N-terminal domain consists of a chemokine (or a chemokine receptor binding domain, CRBD) (Fig. 1a). The prototype degrakine is constructed with SDF-1 and a hemagglutinin tag as the CRBD. The murine stem cell virus long terminal repeat (MSCV LTR) drives expression of the SDF-1 degrakine fusion protein, and a cytomegalovirus (CMV)-driven enhanced green fluorescent protein (EGFP) marker is included in the retroviral vector (Fig. 1a). In the proposed SDF-1 degrakine pathway, the degrakine localizes to the ER, traps the target CKR (CXCR4) in the ER and recruits the proteasome to degrade the target CKR (Fig. 1b). Degrakine ER localization, stable expression and noncytotoxicity The SDF-1 degrakine protein was expressed and localized to the ER. Protein expression was confirmed by western blotting with anti-Vpu and anti-hemagglutinin (data not shown). ER localization was determined by coexpression of the ER-accumulated YFP-KDEL protein.
1322
Stable and specific suppression of CXCR4 expression The SDF-1 degrakine induced rapid and stable downregulation of CXCR4 surface expression in SupT1 cells and Jurkat lymphoblastic leukemia cells. Three days after transduction, SupT1 cells showed a ∼70% reduction in CXCR4 surface expression compared with the vector control, and after 1 week in culture, CXCR4 was downregulated by ∼90% (Fig. 3a). At 3 and 6 weeks in culture, the SDF-1 degrakine maintained stable CXCR4 downregulation (Fig. 3b). Surface staining of CD4 and MHC-I confirmed that the VpuC-based degrakine did not alter the expression of CD4 or MHC-I as full-length Vpu14,17 (Fig. 3c). β-TrCP is involved in receptor downregulation The degrakine system requires the interaction of VpuC with β-TrCP for efficient CKR downregulation. This was demonstrated using a VpuC mutant, VpuC2/6, which contains alanine substitutions at Ser52 and Ser56. These mutations block the interaction between Vpu and β-TrCP, which leads to the loss of Vpu-mediated CD4 degradation16,18. SDF-1VpuC2/6 therefore binds and sequesters CXCR4 in the ER, but cannot recruit the proteasome. We also used the SDF-1 intrakine, SDF-1KDEL7, as a control for CXCR4 downregulation by ER sequestration without degradation. Cells expressing vector, SDF1KDEL, SDF-1VpuC or SDF-1VpuC2/6 were analyzed for CXCR4 expression by FACS. SDF-1VpuC inhibited CXCR4 expression by ∼90% (Figs. 3a and 4a). In contrast, SDF-1VpuC2/6 and SDF-1KDEL
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
ARTICLES
Vector
a
SDF-1 VpuC
b
c 0.5
3.0
YFP 0.4 SDF-1 (nM)
SDF-1 (nM)
HA-Rho
2.0 1.5 1.0
0.3 0.2 0.1
0.5 0.0
0 SDF-1
SDF-1 KDEL
SDF-1
SDF-1 SDF-1 VpuC VpuC2/6
SDF-1 KDEL
SDF-1 SDF-1 VpuC VpuC2/6
Overlay
20 90
0.15 0.1 Mock Vector SDF-1 VpuC
0.05
Day 3
Day 6
Day 9
80
2d
70
21 d
14
60 50 40
12 10 8
30
6
20
4
10
2
0
SDF-1 VpuC
Vector
0
SDF-1 VpuC
Vector
GFP (%)
0.2
Vector
18 16
GFP+ (%)
0.25
0
f
SDF-1 VpuC
e
d
OD reading
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
2.5
SDF-1 VpuC
3d
6d
9d
12 d
Figure 2 The SDF-1 degrakine localizes to the ER and is stably expressed. (a) The EGFP in the SDF-1 degrakine vector was replaced with YFP-KDEL. U2OS cells were transfected, stained with anti-HA monoclonal antibody, followed by anti-mIgG rhodamine (degrakine) and DAPI and visualized by confocal microscopy. (b,c) 293T cells were transfected (b) and SupT1 cells were transduced (c) and monitored by ELISA for SDF-1 secretion into the medium. Similar results were observed in two independent experiments. (d) The proliferation of transduced SupT1 cells was measured at the indicated time points using a MTT assay (OD reading). Results are representative of two independent experiments with each sample in triplicate. (e) Transduced SupT1 cells assayed by FACS for EGFP percentage at 2 d and 3 weeks in culture. (f) GFP (%) Transduced cKit+ bone marrow cells were monitored for EGFP percentage at 3, 6, 9 and 12 d in vitro. At day 12, 1 × 106 cells were harvested and assayed for degrakine expression by western blotting (anti-Vpu).
(constructs that only sequester CXCR4 in the ER) showed a ∼60–70% decrease in CXCR4 surface expression (Fig. 4a). These results demonstrate that the interaction between Vpu and β-TrCP is important in efficient CXCR4 downregulation by the SDF-1 degrakine. To demonstrate VpuC-mediated degradation of CXCR4, we measured total CXCR4 protein in transduced Jurkat cells by western blotting. NIH3T3 cells stably expressing human CXCR4 were used as a positive control. Total levels of CXCR4 were similar in cells expressing vector or SDF-1VpuC2/6, but lower in cells expressing SDF-1VpuC (Fig. 4b). The expression levels of SDF-1VpuC and SDF-1VpuC2/6 were similar, as determined by western blotting with anti-VpuC (data not shown). Therefore, the interaction between VpuC and β-TrCP leads to the degradation of CXCR4 by the SDF-1 degrakine. Functional inactivation of target CKRs by degrakines SDF-1 binding to CXCR4 resulted in the release of intracellular calcium into the cytosol (calcium influx)30. Intracellular calcium influx in response to SDF-1 in cells expressing the indicated proteins was measured by FACS. SDF-1VpuC inhibited the ability of SDF-1 to
induce calcium influx even at 100 nM, indicating that the SDF-1 degrakine efficiently inactivated CXCR4 function (Fig. 5a). CXCR4 mediated cell migration in response to SDF-1 (ref. 21,22). Mock or vector control cells migrated efficiently in response to 1 nM and 10 nM of SDF-1, with a typical oversaturation response at 100 nM. In contrast, cells expressing SDF-1VpuC showed no response to 1 nM and 10 nM of SDF-1 and only a low migration response to 100 nM of SDF-1 (Fig. 5b), a result consistent with low CXCR4 expression. The reduced migration of SDF-1VpuC cells in the absence of exogenous SDF-1 was probably a result of SDF-1-like activity in the culture medium. These experiments show that both CXCR4-mediated calcium influx and chemotaxis are inhibited by the SDF-1 degrakine. Two additional degrakine constructs were tested to demonstrate the specificity and general utility of the system. RantesVpuC and MDCVpuC were designed to inhibit CCR5 and CCR4, respectively. Human acute monocytic leukemia THP-1 cells expressing both CXCR4 and CCR5 (ref. 31) were used to demonstrate the specific targeting of CXCR4 and CCR5 by SDF-1VpuC and RantesVpuC, respectively. The migration response to SDF-1 was inhibited by 81% only in
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1323
ARTICLES a
b
c
120
Vector
125
10e3
10e3
10e2
10e2
10e1
10e1
10e1 10e2 10e3 10e4
SDF-1 VpuC
10e1 10e2 10e3 10e4
Receptor expression (%)
CXCR4
100 80 3 weeks 6 weeks
60 40 20
100 75 50 25
GFP 0
0
Vector
SDF-1 VpuC
CD4
MHC-I
Figure 3 CXCR4 expression is stably and specifically downregulated by the SDF-1 degrakine. (a) A representative FACS plot of independent SupT1 infections showing CXCR4 surface expression 1 week after transduction. (b) The cells were reanalyzed for CXCR4 expression at 3 and 6 weeks after transduction. (c) CD4 and MHC-I expression was monitored at the 3-week time point. Identical results were observed in two independent experiments with SupT1 cells and Jurkat cells.
to downregulate and desensitize CXCR4. These cells showed the same homing defect as SDF-1VpuC-transduced HSPC (Fig. 6b). HSPCs expressing the MDC degrakine, which inhibited thymocyte migration to MDC in vitro (Fig. 5e), showed a bone marrow homing percentage similar to the control cells (Fig. 6b). This is consistent with the fact that CCR4 is not expressed on HSPC, and MDC is not expressed in the bone marrow34. These results demonstrate that CXCR4 plays an important role in HSPC homing to the bone marrow immediately after transplantation.
THP-1 cells expressing SDF-1VpuC (Fig. 5c), whereas response to MIP-1β was reduced 65% in THP-1 cells expressing RantesVpuC (Fig. 5d). Similarly, MDCVpuC-transduced mouse thymocytes, which express CCR4 (ref. 32), showed a 66% reduction in migration response to 100 nM MDC as compared with vector or SDF-1VpuC (Fig. 5e). Thus, the degrakines specifically inactivated their target CKR functions. Taken together, these results suggest that the degrakine system is applicable to inactivating chemokine receptors in general.
1324
0
1
0
1
2
2
3
3
4
4
0
1
2
3
4
0
1
2
3
4
CXCR4+
WT
SDF-1 VpuC2/6
SDF-1 VpuC
Vector
CXCR4 is required for the homing of HSPC to the bone marrow SDF-1 is a potent migratory chemokine for HSPC cells in vitro26,33,34. DISCUSSION We used the SDF-1 degrakine to determine whether CXCR4 is A number of chemokines and their receptors are involved in normal required for the homing of HSPC to the bone marrow in vivo. Bone hemato-lymphopoiesis and in the immune response to pathogens. We marrow cells from B6/Ly5.2 mice were enriched for cKit+ progenitor have developed a genetic approach for the specific functional inactivacells35 and transduced with the MSCV-based retroviral vectors. Stable tion of CKRs in various cell types in vitro and in vivo. In stably transEGFP expression in vector or SDF-1 degrakine–transduced pro- duced cell lines and mouse thymocytes or HSPCs, the degrakine does genitor cells was observed in vitro, and degrakine expression was not affect cell proliferation or viability. The functional inactivation of confirmed by anti-Vpu western blotting 12 d after transduction (Fig. 2f). The SDF-1 Jurkat 10e4 3T3 degrakine is therefore not cytotoxic to mouse a 10e4 Vector b SDF-1 HSPCs. Progenitor cells expressing SDF-1 or 10e3 10e3 KDEL RANTES degrakines were assayed for their 10e2 10e2 migratory response to 10 nM and 50 nM of SDF-1. Cells expressing SDF-1VpuC showed 10e1 10e1 a 60–70% reduction in migration response to SDF-1 whereas cells expressing the RANTES 10e1 10e2 10e3 10e4 10e1 10e2 10e3 10e4 degrakine showed normal responses to SDF-1 10e4 10e4 66 kDa SDF-1 SDF-1 (Fig. 6a). To investigate CXCR4’s role in VpuC VpuC 10e3 10e3 2/6 HSPC homing during transplantation, we CXCR4 injected intravenously B6/Ly5.2 bone marrow 10e2 10e2 cKit+ cells expressing vector, SDF-1VpuC or 10e1 10e1 MDCVpuC into lethally irradiated B6/Ly5.1 45 kDa recipient mice. Three hours after transplanta10e1 10e2 10e3 10e4 10e1 10e2 10e3 10e4 tion, bone marrow cells were harvested and GFP analyzed by FACS for Ly5.2 (donor) and EGFP (degrakine-positive) expression. SDFFigure 4 The interaction between VpuC and β-TrCP is important for efficient CXCR4 downregulation. 1VpuC inhibited short-term homing of pro(a) A representative FACS plot of independent SupT1 infections showing CXCR4 surface expression 1 genitor cells by ∼40% (Fig. 6b). As a control, week after transduction. Similar results were observed in Jurkat cells. (b) Total cytoplasmic lysates were vector-transduced progenitor cells were analyzed by western blotting with a polyclonal anti-CXCR4. NIH3T3 cells stably expressing human pulsed with SDF-1 (10 nM) before injection CXCR4 were used as a positive control. Similar results were observed in two independent experiments. CXCR4
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Vector
10e4
CXCR4 expression (%)
SDF-1 VpuC 10e4
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
ARTICLES
Vector
192
192
192
128
128
128
64
64
64
0 14 min 0 s 256
0 14 min 0 s 256
192
192
128
128
64
64
0s
SDF-1 VpuC
b
SDF-1 (100 nM) 256
0
30
Mock Vector
25
SDF-1 VpuC
14 min
20 15 10 5
0
0s
14 min
0
0s
0
14 min
0 nM
1 nM
10 nM
100 nM
Computed time
d
25
e
30
7 6
25 Migration (%)
20 15 10
** 5
20 15
* 10 5
0 SDF-1 VpuC
Rantes VpuC
5 4
**
3 2 1 0
0 Vector
Migration (%)
c Migration (%)
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
SDF-1 (10 nM) 256
Migration (%)
Ionomycin 256
Indo 405/Indo 485
a
Vector
Rantes VpuC
Vector
SDF-1 VpuC
MDC VpuC
Figure 5 Degrakines specifically inhibit their target CKR function. (a) Transduced SupT1 cells were loaded with Indo-1AM, stimulated with SDF-1 and monitored by FACS for the change in Indo-1AM fluorescence. (b) Parental SupT1 cells (Mock), cells expressing vector or SDF-1VpuC were assayed for migration response to SDF-1 in a trans-well assay. Each sample was performed in triplicate and the assay is representative of two independent experiments. (c,d) THP-1 cells were transduced, cultured for 1 week and assayed for migration in response to 100 nM SDF-1 (c), or 100 nM MIP-1β (d) in a trans-well migration assay. (e) Mouse thymocytes were transduced, cultured for 1 week and measured for their migration response to 100 nM MDC in a trans-well assay. *, P < 0.05. **, P < 0.005.
CXCR4 by the SDF-1 degrakine was verified by a reduction in CXCR4 expression, inhibition of SDF-1 induced calcium influx and blockage of chemotaxis in human T cell lines, mouse thymocytes and mouse HSPCs. The specificity and general utility of the degrakines were illustrated by the specific inactivation of their respective CKRs by SDF-1, MDC and RANTES degrakines. Furthermore, we used the SDF-1 degrakine to demonstrate that CXCR4 is involved in the homing of mouse HSPCs immediately after bone marrow transplantation, a finding that may partly explain the requirement of CXCR4 for the engraftment of human or mouse HSPCs in the mouse bone marrow compartment27–29. The degrakine system offers a number of advantages for studying the function of CKRs in HSPC functions. First, it may be applicable in human HSPCs where conventional genetic approaches are not feasible. Second, multiple CKRs can be targeted by a single degrakine or by a combination of coexpressed degrakines. Finally, it is also feasible to combine the degrakine system with existing knockout mice to inactivate multiple CKRs. Genetic manipulation of HSPCs from mature animals is directly relevant to clinical settings. The use of degrakines in these cells may elucidate the functions of specific CKRs during hemato-lymphopoiesis and immune responses in adults. In addition to blocking antibodies, which may have limited target specificity and in vivo efficacy, the intrakine system has also been used to inhibit CKR expression7. However, the intrakine secretes high levels of the intrakine protein out of the cell8
(Fig. 2b,c), is less efficient than the degrakine in downregulating CKR expression (Fig. 4a) and may cause an accumulation of intrakine-CKR proteins in the ER. In contrast, the degrakine is designed to trap the CKR in the ER and target it for degradation. We envision several possible modifications of the degrakine system. First, one might be able to derive the CRBD from single-chain antibodies or peptides that specifically bind a target CKR. Such derivatives with enhanced CKR binding activity, but no agonistic or antagonistic activity, should substantially improve degrakine efficiency and potential applications. Second, the approach could be adapted to receptors other than chemokine receptors. Third, degrakines could be combined with stable hairpin RNAi in the retroviral system, so that the CKR is targeted for inactivation at both the mRNA and protein levels. This might overcome the variable success rate of RNAi alone. Finally, in combination with recent advances in in vivo imaging technology36,37, the degrakine system might be used to analyze the migration of HSPCs, T cells and dendritic cells in vitro and in vivo. METHODS Animals. All animals were housed at the University of North Carolina at Chapel Hill in a sterile animal facility. Total bone marrow was harvested from the tibias and femurs of 1- to 2-month-old B6/Ly5.2 animals. Three-month-old recipient B6/Ly5.1 animals were administered sterile pH 2.0 water supplemented with neomycin sulfate. Lethal irradiation of animals (1,000 rads) was administered with a cesium source irradiator. The Institutional Animal Care and Use Committee approved all experiments.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1325
ARTICLES 25
Migration (%)
20
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
b
Vector SDF-1 VpuC
0.8
Rantes VpuC
0.7
15
* 10 5
added and the wells were assayed with a microplate reader using a test wavelength of 570 nm, reference wavelength of 630 nm and calibration setting of 1.99 (ref. 40). All optical density (OD) readings were done in triplicate.
0.9
Migration (%)
a
0.6
**
**
0.5 0.4 0.3 0.2
*
0.1 0
0 10 nM SDF-1
50 nM SDF-1
Vector
SDF-1 VpuC
MDC VpuC
Vector (+SDF-1)
Figure 6 The SDF-1 degrakine inhibits CXCR4-mediated HSPC migration in vitro and homing to the bone marrow in vivo. (a) Transduced HSPCs (cKit+) were assayed for migration to 10 nM and 50 nM of SDF-1 in a trans-well assay. Each sample was performed in triplicate and the results are representative of two independent experiments. (b) B6/Ly5.2+ HSPCs (cKit+) were transduced, injected into lethally irradiated B6/Ly5.1 recipients and harvested from the recipient’s bone marrow 3 h after transplantation. Vector + SDF-1 cells were vector-transduced cells pulsed with 10 nM of SDF-1 immediately before injection. Three mice were used for each vector and the assay is representative of two independent experiments. *, P < 0.005. **, P < 0.005.
Recombinant DNA. SDF-1 was cloned by PCR from the SDF-1 KDEL7 template, so that it is in-frame with amino acids 28–81 of Vpu (forward primer: 5′-CGCGAATTCGCGCCATGAACGCCAAGGTCG-3′; reverse primer: 5′-CAT GCGGCCGCTAGCATAATCTGGAACATC-3′). Amino acids 28–81 of Vpu were cloned by PCR from the NL4-3 genome (forward primer: 5′-CTAGCGGCCGCGAATATAGGAAAATATTAAGAC-3′; reverse primer: 5′-CAGCTCGAGCACTACAGATCATCAATATCC-3′). VpuC2/6 was cloned with the same primers as VpuC, but using Vpu2/6 (ref. 18) as a template. All constructs were cloned into a high-titer retroviral vector derived from the MSCV LTR38. Cell culture. SupT1 cells were cultured in RPMI 1640 with 10% (vol/vol) FBS and penicillin/streptomycin. U2OS cells and 293T cells were cultured in DMEM-H with 10% (vol/vol) FBS and penicillin/streptomycin. Primary bone marrow was flushed from femurs and tibias, enriched for cKit+ cells35, cultured in IMDM containing 20% FBS (Gibco), 50 ng/ml stem cell factor (SCF), 50 ng/ml interleukin-6 (IL-6) and 10 ng/ml IL-11 (Peprotech) for 48 h, and then transduced with the indicated retroviruses. HSPC cells were maintained in IMDM with 10 ng/ml of SCF, IL-6 and IL-11. Thymocytes were harvested and cultured in IMDM, 20% ES FBS, 50 ng/ml SCF and 15 ng/ml IL-7 (Peprotech). Retrovirus production and infection. 293T cells were transfected with plasmids encoding MSCV retroviral DNA, VSVg and gag/pol as described39. Retroviral infections by spinoculation were performed in 2 ml Eppendorf tubes with 1 × 106 cells with 500 µl RPMI 1640 with 10% (vol/vol) FBS, 500 µl of viral supernatant and 8 µg/ml polybrene. Cells were incubated at 22–24 oC for 20 min and centrifuged at 2000g for 3 h. Cells were either cultured as a pool or sorted for EGFP expression. Immunofluorescence. The EGFP cassette in the degrakine vector was replaced with YFP-KDEL (Clontech). U2OS cells were transfected (Effectene, Qiagen) with the indicated DNAs. At 48 h after transfection, cells were fixed in 3% paraformaldehyde and permeabilized with 0.1% Triton X-100. Cells were stained with anti-hemagglutinin (12CA5) monoclonal and goat antimIgG–phycoerythrin (Pharmingen) and visualized by confocal microscopy (100×). MTT assay. 1 × 103 cells per well of EGFP-sorted SupT1 cells were plated into a 96-well plate (triplicate). At the indicated time points, the cells were transferred into a new 96-well plate containing MTT (methylthiazolydiphenyl tetrazolium bromide; Sigma) solution and incubated at 37 °C for 4 h. Acid-isopropanol was
1326
FACS. Each sample (1 × 105 cells) was washed in PBS with 2% (vol/vol) FBS and stained with the indicated antibody in PBS with 2% (vol/vol) FBS, and 1–2 × 104 cells were collected through a Becton Dickinson FACScan. CXCR4 expression was detected with anti-12G5 monoclonal antibody (NIH AIDS Reagent) and anti-mIgG–phycoerythrin (Caltag). Phycoerythrin–anti-CD4 and phycoerythrin–anti-MHC-I antibodies were purchased from Caltag. GFP-positive cells were sorted on a Cytomation MoFlo FACS. Calcium influx assays were performed using Indo-1 AM (Molecular Probes) as described41. ELISA. Secreted SDF-1 was measured by ELISA (R&D Systems). 293T cells were transfected using Effectene (Qiagen) and cultured for 48 h. Stably transduced SupT1 cells were washed, plated at equal concentrations and cultured for 48 h.
Western blotting. Transduced mouse HSPCs (1 × 106 cells) or Jurkat cells (1 × 106 cells) were lysed and resolved by SDS-PAGE42. Gels were transferred to PVDF membranes (Amersham Pharmacia), blocked with 5% nonfat dry milk, probed with either anti-Vpu (1:1,000) or anti-CXCR442 (1:100) and anti-rabbit-IgG–horseradish peroxidase (1:10,000) (Amersham Pharmacia). Blots were visualized using an ECL Kit (Amersham Pharmacia). In vitro migration assay. Sorted SupT1 cells (2 × 104/well in triplicate) were assayed for migration response to SDF-1 over 2 h using a 0.6 µm Neuroprobe ChemoTx 96-well assay plate and cell numbers were quantified with Packard’s ATPLite kit. 5 × 105 THP-1 cells, thymocytes or HSPCs were assayed for migration over a 3-h period (24-well Corning trans-well plate, 0.5 µM membrane) to 100 nM SDF-1 (NIH AIDS Reagent) and 100 nM Mip-1β (Peprotech). The percentage of migrating degrakine expressing cells was determined by FACS. (Migration % = migrating GFP+ cells/input GFP+ cells.) HSPC homing assay in vivo. cKit+ B6/Ly5.2 cells were cultured for 3–6 d after retroviral transduction. In vivo homing to the bone marrow was determined by injection of 2 × 106 degrakines expressing cKit+ B6/Ly5.2 cells into lethally irradiated B6/Ly5.1 mice (1,000 rads, 24 h before injection). At 3 h after injection the bone marrow was flushed from femurs and tibias and monitored for Ly5.2 and GFP with a BD FACScan43. (Migration % = harvested Ly5.2+GFP+ cells/input Ly5.2+GFP+ cells.) ACKNOWLEDGMENTS We thank S. Chen, K. Strebel, H. Golding and M. Zaitseva for kindly providing the SDF-1 intrakine construct, the Vpu2/6 mutant, anti-CXCR4-4N and CXCR4 western protocol, respectively; J. Smith, Y. Kim, M. Townsend and A. Elms for technical assistance; E. Meissner and W. Helms for critical reading of the manuscript; R. Bagnell for assistance with confocal microscopy; and L. Arnold of the UNC Flow Cytometry Facility for assistance with FACS. Anti-Vpu and anti-CXCR4 (12G5) were obtained through the NIH AIDS Reagent program. The project was partially supported by grants from the US National Institutes of Health (nos. HL72240 and AI53804 to L.S.). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Received 5 March; accepted 22 August 2003 Published online at http://www.nature.com/naturebiotechnology/ 1. Rossi, D. & Zlotnik, A. The biology of chemokines and their receptors. Annu. Rev. Immunol. 18, 217–242 (2000).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES 2. Yoshie, O., Imai, T. & Nomiyama, H. Chemokines in immunity. Adv. Immunol. 78, 57–110 (2001). 3. Sallusto, F., Mackay, C.R. & Lanzavecchia, A. The role of chemokine receptors in primary, effector, and memory immune responses. Annu. Rev. Immunol. 18, 593–620 (2000). 4. Nelson, P.J. & Krensky, A.M. Chemokines, chemokine receptors, and allograft rejection. Immunity 14, 377–386 (2001). 5. Baggiolini, M. Chemokines and leukocyte traffic. Nature 392, 565–568 (1998). 6. Marsal, J. et al. Involvement of CCL25 (TECK) in the generation of the murine smallintestinal CD8α α+CD3+ intraepithelial lymphocyte compartment. Eur. J. Immunol. 32, 3488–3497 (2002). 7. Chen, J.D., Bai, X., Yang, A.G., Cong, Y. & Chen, S.Y. Inactivation of HIV-1 chemokine co-receptor CXCR-4 by a novel intrakine strategy. Nat. Med. 3, 1110–1116 (1997). 8. Engel, B.C. et al. Intrakines—evidence for a trans-cellular mechanism of action. Mol. Ther. 1, 165–170 (2000). 9. McManus, M.T. et al. Small interfering RNA-mediated gene silencing in T lymphocytes. J. Immunol. 169, 5754–5760 (2002). 10. McManus, M.T. & Sharp, P.A. Gene silencing in mammals by small interfering RNAs. Nat. Rev. Genet. 3, 737–747 (2002). 11. Cullen, B.R. RNA interference: antiviral defense and genetic tool. Nat. Immunol. 3, 597–599 (2002). 12. Brummelkamp, T.R., Bernards, R. & Agami, R. A system for stable expression of short interfering RNAs in mammalian cells. Science 296, 550–553 (2002). 13. Tiscornia, G., Singer, O., Ikawa, M. & Verma, I.M. A general method for gene knockdown in mice by using lentiviral vectors expressing small interfering RNA. Proc. Natl. Acad. Sci. USA 27, 27 (2003). 14. Tortorella, D., Gewurz, B.E., Furman, M.H., Schust, D.J. & Ploegh, H.L. Viral subversion of the immune system. Annu. Rev. Immunol. 18, 861–926 (2000). 15. Hirsch, C. & Ploegh, H.L. Intracellular targeting of the proteasome. Trends Cell Biol. 10, 268–272 (2000). 16. Margottin, F. et al. A novel human WD protein, h-beta TrCp, that interacts with HIV-1 Vpu connects CD4 to the ER degradation pathway through an F-box motif. Mol. Cell. 1, 565–574 (1998). 17. Willey, R.L., Maldarelli, F., Martin, M.A. & Strebel, K. Human immunodeficiency virus type 1 Vpu protein induces rapid degradation of CD4. J. Virol. 66, 7193–7200 (1992). 18. Schubert, U. & Strebel, K. Differential activities of the human immunodeficiency virus type 1–encoded Vpu protein are regulated by phosphorylation and occur in different cellular compartments. J. Virol. 68, 2260–2271 (1994). 19. Loetscher, M. et al. Cloning of a human seven-transmembrane domain receptor, LESTR, that is highly expressed in leukocytes. J. Biol. Chem. 269, 232–237 (1994). 20. Bleul, C.C. et al. The lymphocyte chemoattractant SDF-1 is a ligand for LESTR/fusin and blocks HIV-1 entry. Nature 382, 829–833 (1996). 21. Bleul, C.C., Fuhlbrigge, R.C., Casasnovas, J.M., Aiuti, A. & Springer, T.A. A highly efficacious lymphocyte chemoattractant, stromal cell-derived factor 1 (SDF-1). J. Exp. Med. 184, 1101–1109 (1996). 22. Oberlin, E. et al. The CXC chemokine SDF-1 is the ligand for LESTR/fusin and prevents infection by T-cell-line-adapted HIV-1. Nature 382, 833–835 (1996). 23. Shirozu, M. et al. Structure and chromosomal localization of the human stromal cellderived factor 1 (SDF1) gene. Genomics 28, 495–500 (1995). 24. Tashiro, K. et al. Signal sequence trap: a cloning strategy for secreted proteins and type I membrane proteins. Science 261, 600–603 (1993). 25. Nagasawa, T. et al. Molecular cloning and characterization of a murine pre-B-cell
growth–stimulating factor/stromal cell-derived factor 1 receptor, a murine homolog of the human immunodeficiency virus 1 entry coreceptor fusin. Proc. Natl. Acad. Sci. USA 93, 14726–14729 (1996). 26. Aiuti, A., Webb, I.J., Bleul, C., Springer, T. & Gutierrez-Ramos, J.C. The chemokine SDF-1 is a chemoattractant for human CD34+ hematopoietic progenitor cells and provides a new mechanism to explain the mobilization of CD34+ progenitors to peripheral blood. J. Exp. Med. 185, 111–120 (1997). 27. Zou, Y.R., Kottmann, A.H., Kuroda, M., Taniuchi, I. & Littman, D.R. Function of the chemokine receptor CXCR4 in haematopoiesis and in cerebellar development. Nature 393, 595–599 (1998). 28. Kawabata, K. et al. A cell-autonomous requirement for CXCR4 in long-term lymphoid and myeloid reconstitution. Proc. Natl. Acad. Sci. USA 96, 5663–5667 (1999). 29. Peled, A. et al. Dependence of human stem cell engraftment and repopulation of NOD/SCID mice on CXCR4. Science 283, 845–848 (1999). 30. D’Apuzzo, M. et al. The chemokine SDF-1, stromal cell-derived factor 1, attracts early stage B cell precursors via the chemokine receptor CXCR4. Eur. J. Immunol. 27, 1788–1793 (1997). 31. Schols, D. et al. Inhibition of T-tropic HIV strains by selective antagonization of the chemokine receptor CXCR4. J. Exp. Med. 186, 1383–1388 (1997). 32. Chantry, D. et al. Macrophage-derived chemokine is localized to thymic medullary epithelial cells and is a chemoattractant for CD3+, CD4+, CD8low thymocytes. Blood 94, 1890–1898 (1999). 33. Kim, C.H. & Broxmeyer, H.E. In vitro behavior of hematopoietic progenitor cells under the influence of chemoattractants: stromal cell-derived factor-1, steel factor, and the bone marrow environment. Blood 91, 100–110 (1998). 34. Wright, D.E., Bowman, E.P., Wagers, A.J., Butcher, E.C. & Weissman, I.L. Hematopoietic stem cells are uniquely selective in their migratory response to chemokines. J. Exp. Med. 195, 1145–1154 (2002). 35. Spangrude, G.J., Heimfeld, S. & Weissman, I.L. Purification and characterization of mouse hematopoietic stem cells. Science 241, 58–62 (1988). 36. Hardy, J. et al. Bioluminescence imaging of lymphocyte trafficking in vivo. Exp. Hematol. 29, 1353–1360 (2001). 37. Contag, P.R., Olomu, I.N., Stevenson, D.K. & Contag, C.H. Bioluminescent indicators in living mammals. Nat. Med. 4, 245–247 (1998). 38. Cheng, L. et al. Sustained gene expression in retrovirally transduced, engrafting human hematopoietic stem cells and their lympho-myeloid progeny. Blood 92, 83–92 (1998). 39. Pear, W.S., Nolan, G.P., Scott, M.L. & Baltimore, D. Production of high-titer helperfree retroviruses by transient transfection. Proc. Natl. Acad. Sci. USA 90, 8392–8396 (1993). 40. Mosmann, T. Rapid colorimetric assay for cellular growth and survival: application to proliferation and cytotoxicity assays. J. Immunol. Methods 65, 55–63 (1983). 41. Vilen, B.J., Burke, K.M., Sleater, M. & Cambier, J.C. Transmodulation of BCR signaling by transduction-incompetent antigen receptors: implications for impaired signaling in anergic B cells. J. Immunol. 168, 4344–4351 (2002). 42. Lapham, C.K., Zaitseva, M.B., Lee, S., Romanstseva, T. & Golding, H. Fusion of monocyctes and macrophages with HIV-1 correlates with biochemical properties of CXCR4 and CCR5. Nature 5, 303–308 (1999). 43. Szilvassy, S.J., Meyerrose, T.E., Ragland, P.L. & Grimes, B. Differential homing and engraftment properties of hematopoietic progenitor cells from murine bone marrow, mobilized peripheral blood, and fetal liver. Blood 98, 2108–2115 (2001).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1327
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES
E3 gene manipulations affect oncolytic adenovirus activity in immunocompetent tumor models Yaohe Wang1, Gunnel Hallden1, Richard Hill1, Arthi Anand1, Ta-Chiang Liu1, Jennelle Francis1, Gabriel Brooks1, Nick Lemoine1 & David Kirn1,2 Oncolytic replication-selective adenoviruses constitute a rapidly growing therapeutic platform for cancer. However, the role of the host immune response and the E3 immunoregulatory genes of the human adenovirus were unknown until now. We identified four mouse carcinoma lines of variable permissivity for adenoviral gene expression, cytopathic effects and/or burst size. To determine E3 gene effects in immunocompetent tumor-bearing hosts, we injected tumors with one of three adenoviruses: Ad5 (E3 wild type), dl 309 (del. E3 10.4/14.5, 14.7 kDa) or dl 704 (del. E3 gp19 kDa). Compared with Ad5 and dl 704, dl 309 was cleared much more rapidly and/or its activity was lower in all four models. Intratumoral injection with dl 309 resulted in markedly greater macrophage infiltration and expression of both tumor necrosis factor and interferon-γ. Adenovirus replication, CD8+ lymphocyte infiltration and efficacy were similar upon intratumoral injection with either dl 704 or Ad5. E3-dependent differences were not evident in athymic mice. These findings have important implications for the design of oncolytic adenoviruses and may explain the rapid clearance of E3-10.4/14.5,14.7-deleted adenoviruses in patients.
Replication-selective oncolytic viruses are a rapidly expanding therapeutic platform for cancer1,2. Agents from numerous viral species have been engineered for, or have shown inherent, tumor selectivity and efficacy. The most commonly described engineered oncolytic viruses over the past ten years have been adenoviruses3. Adenoviruses have efficacy, safety and manufacturing characteristics that make them attractive oncolytic virus candidates2. Over 25 different oncolytic adenoviruses have been described as potential cancer treatments to date. Examples include dl1520 (also known as Onyx-015)4,5, CV706 (now CG7060)6, CV787 (now CG7870)7, CV8840 (now CG8840), dl922–947 (ref. 8), Ad5-CD/tk-rep9, Ad-delta24 (ref. 10), Ad.DF3E1 (ref. 11), Onyx-411 (ref. 12), OAV001 (ref. 13), KD3 (ref. 14) and 01/PEME15. Data have now been reported from more than ten clinical trials with at least four different oncolytic adenoviruses: dl1520 (Onyx Pharmaceuticals)16, Ad5-CD/tkrep, CV787 and CV706 (Cell Genesys)17. Tumor-specific replication of dl1520 was demonstrated, but viral replication was cleared in <10 d18; tumor regressions were typically transient19 or were not seen16,20. Likewise, CV706 replication appeared to be short-lived (≤7 d) and complete tumor responses were not reported17. Therefore, data from immunodeficient nude mouse– human tumor xenograft models have not been predictive for results in cancer patients to date. The lack of an immunocompetent tumor efficacy model has been a critical limitation for the field. Adenoviruses have numerous interactions with immune response effectors21,22. To date, all published animal tumor model efficacy data with replication-selective adenoviruses have come from immunodeficient mouse–human tumor xenograft
models23–25. Based on data from adenovirus in normal mouse lung26, it was assumed by many that viral replication would not occur in mouse tumors. In addition, the E3 region of human adenovirus encodes a variety of proteins involved with immune-response evasion27–28. The gp19 kDa protein inhibits major histocompatibility complex (MHC) class I expression on the cell surface (that is, avoidance of cytotoxic T-lymphocyte-mediated killing)29, and the E3B region proteins (10.4/14.5 kDa-RID and 14.7 kDa) inhibit apoptosis mediated by FasL, tumor necrosis factor–related apoptosis-inducing ligand (TRAIL) and/or tumor necrosis factor (TNF)30–32. Notably, the vast majority of the adenoviruses either entering or already in human clinical trials have major deletions within the adenovirus E3B region3. We hypothesized that some mouse tumors might be permissive to adenovirus. First, adenovirus replication had been demonstrated in mouse liver in vivo33. In addition, replication might be enhanced by any of the diverse genetic alterations occurring during carcinogenesis. It had been shown previously that cytopathic effects can be induced in some mouse cancer cell lines in vitro34. We screened a large panel of mouse carcinoma cell lines for their ability to support adenovirus uptake, early gene expression, late gene expression, replication and cytopathic effects. We subsequently focused on four carcinoma lines representing a spectrum of cytopathic effect sensitivities, burst sizes, tumor types and genetic backgrounds. Ad5 efficacy, replication and/or late gene expression was demonstrated in all four tumor types in immunocompetent mice. We next sought to evaluate the role of the E3-RID and E3 gp19 kDa gene products, and that of the T cell–dependent immune response, in
1Viral and Genetic Therapy Program, Cancer Research UK and Imperial College School of Medicine, Hammersmith Hospital, London, UK. 2Present address: Department of Pharmacology, Oxford University Medical School, Oxford, UK. Correspondence should be addressed to D.K. (
[email protected]).
Published online 12 October 2003; doi:10.1038/nbt887
1328
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES a
b
c
d
others were substantially lower. We then compared the replication time course for a subset of cell lines (three mouse, one human) and obtained similar results (Fig. 1b). Next, we compared the peak viral burst (96 h) in mouse lines to that in the human H460 line with a panel of viruses (Table 1): Ad5, the E1A mutant dl312 (E1A–; replication-deficient) and E3 mutant viruses dl309 (E3B–) and dl704 (Fig. 1c). We saw no significant differences between these viruses in vitro, although we noted a trend towards enhanced replication of dl309. Finally, we had previously detected an increase in the potency of cytopathic effects with dl309 against human lines in vitro. To confirm that the relative potency of these viruses in mouse lines mirrored that in human lines, we compared Ad5 and dl309 in a panel of human and mouse carcinoma lines (Fig. 1d); dl309 was more potent than Ad5 in all lines tested.
Ad5, dl309 or dl 704 in immunocompetent and athymic mice We initially assessed early gene expression (E1A), late gene expression (Ad5 coat proteins), viral DNA synthesis and replication Figure 1 Replication of wild-type and E3-mutant adenovirus in murine and human cancer cell lines in over time after intratumoral injection of vitro. (a) PFUs produced by murine (black) and human (white) cell lines 96 h after infection with 1,000 CMT-64 (Fig. 2a,b), JC (Fig. 2c,d) and CMTparticles per cell (p.p.c.). (b) PFU/ml over time after Ad5 infection of murine carcinoma cells (CMT-64, 93 (Fig. 2e). The extent and duration of dl309 JC, CMT-93) versus the H460 human cell line infected at 1,000 p.p.c. (c) The number of PFU/cell gene expression between days 5 and 22 were 96 h after dl 312, Ad5, dl 309 and dl 704 infection of murine carcinoma cell lines (CMT-64, JC, CMT-93) versus the H460 human carcinoma cell line infected at 1,000 p.p.c. (d) The ratio of the EC50 significantly lower than those of Ad5 and (dose required for 50% destruction of the target cells) for dl 309 (E3B–) versus Ad5 (E3B+) in murine dl704 (P < 0.01, Fisher’s exact test; Fig. 2a–e). (black) and human (white) cancer cell lines 6 d after infection; ratios <100% indicate that the potency A representative tumor cross section is shown of dl 309 was greater than that of Ad5 in the cell line tested. in Figure 2f. Notably, by day 15, dl309 E1A expression was undetectable in CMT-64, JC and CMT-93 tumors whereas Ad5 and dl704 the context of oncolytic adenovirus therapy. Multiple reports have were still detectable (P < 0.001 for paired t-test versus Ad5 or dl704). confirmed that E3B (E3 10.4/14.5, 14.7) and E3 gp19 kDa genes bind Finally, adenovirus DNA replication was assessed over time in vivo and inactivate the mouse versions of their cellular targets in vitro and (Fig. 2g). These results mirrored those for E1A and coat proteins, conin vivo35–41. In this manuscript we describe E3 gene–dependent and firming the rapid clearance of dl309. Finally, to confirm that clearance of dl309 gene expression and DNA T cell–dependent effects on oncolytic adenovirus replication, persistence, inflammation induction and efficacy in immunocompetent host replication resulted in reduced infectious or plaque-forming units animals. These results have important implications for oncolytic aden- (PFUs) in the tumors, we evaluated PFU levels at various time points after injection into tumors induced by two carcinoma cell lines that supovirus design. ported replication in vitro (Fig. 3a,b), and in mouse liver after intraRESULTS venous administration (Fig. 3c); published data demonstrated Ad5 replication in mouse liver43. dl309 titers were significantly lower than In vitro adenovirus replication with and without E3 genes We previously demonstrated expression of E1A and coat protein those for Ad5 and/or dl704 (P < 0.05 at all time points; Fig. 3a–c), and (hexon, fiber, penton) and associated induction of cytopathic effects in the differences increased between day 7 and 15 after treatment (Fig. 3b). seven mouse carcinoma cell lines in vitro42; viral burst sizes varied from high levels (similar to that in some human cells) to relatively low levels. For this publication we focused on four mouse carcinoma lines Table 1 Adenovirus strains used in experiments: E1A and E3 that represented a spectrum of cytopathic effect sensitivities, burst genes present and functional sizes, tumor types and genetic backgrounds. The relative sensitivities Virus E1A E3Ba E3 gp19 kDa to cytopathic effects were as follows: CMT-93 (highest), CMT-64, PDV + + + and JC (lowest). In contrast, the relative burst sizes were as follows: Ad5 dl309 + – + CMT-64 and JC (equivalent), PDV and CMT-93 (lowest). + + – We compared Ad5 replication in the four mouse carcinoma lines to dl704 dl312 – – + that in the human carcinoma lines (96 h; Fig. 1a); the peak viral burst in CMT-64 and JC cells was similar to that in one human line whereas aE3B consists of E3-RID (10.4/14.5) and E3 14.7.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1329
ARTICLES
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
a
b
c
d
e
f
g
Figure 2 Adenovirus gene expression and DNA replication after injection with Ad5, nonreplicating control adenovirus, dl 309 or dl 704 in immunocompetent mice. (a–e) Mouse tumors growing in the flanks of immunocompetent mice were directly injected with the viruses listed on the x-axis (including replicationdeficient control, PUV-dl 312). Between days 5 and 15 the tumors were harvested and immunohistochemical staining (IHC) was performed for adenoviral coat proteins or E1A: CMT-64 day 5 (a); CMT-64, day 15 (b); JC, day 5 (c); JC, day 15 (d); CMT-93, day 5 (e). The mean (±s.e.m.) IHC score for each group is shown on the y-axis (scoring system in Methods). (f) Representative E1A staining (brown-staining nuclei in Ad5, dl 309 and dl 704 sections) from tumor cross-sections (JC) after injection (day 5) with the viruses listed. (g) PDV tumor fragments were grown in the flanks of immunocompetent (C57B/6) mice and the resulting tumors were subsequently injected with Ad5 (black) or dl 309 (white). On days 1, 3, 7 and 22 after treatment, tumors were harvested, sectioned and the percentage of tumors cells staining positively for viral DNA synthesis was estimated. *, undetectable.
We subsequently compared viral replication and clearance over time after intratumoral injection of Ad5 or dl309 in immunocompetent versus athymic mice (Fig. 4a,b). In contrast to the differences in suppression and clearance between Ad5 and dl309 in immunocompetent models, Ad5 and dl309 behaved similarly in tumors grown in athymic mice; dl309 gene expression and DNA replication were similar to those for Ad5 in immunocompetent mice. Therefore, the enhanced clearance of dl309 compared to Ad5 was demonstrated to be T cell dependent. Intratumoral infiltration in immunocompetent mice To evaluate the intratumoral immune response to adenovirus treatment of tumors in an immunocompetent host, we harvested and analyzed mouse tumors over time after intratumoral injection with Ad5 and E3 mutant adenoviruses (Fig. 5). In comparison to the other adenoviruses, dl309 induced an intense early (day 5–8) macrophage infiltration (Fig. 5a–c) that disappeared by day 22 when the virus had been cleared. In contrast, there was substantially less macrophage infiltra-
1330
tion induced by Ad5 and dl704 than by dl309 at early time points (day 8); by day 22, both viruses had induced significant (P < 0.05) intratumoral macrophage infiltration (e.g., CMT-93; Fig. 5b). In CMT-93 tumors, dl704 and Ad5 were associated with enhanced eosinophil infiltration compared to dl309 or dl312 (data not shown). Because TNF is a critical macrophage-derived mediator of adenovirus clearance, we carried out immunohistochemical staining for TNF at early time points. dl309-injected tumors were associated with higher TNF expression compared to Ad5-injected tumors (Fig. 5d). Similar results were obtained for interferon-γ staining (Fig. 5e). The accelerated clearance of dl309 was associated with markedly increased early macrophage recruitment, TNF induction and interferon-γ expression. We analyzed injected JC and CMT-93 tumors for CD8+ lymphocyte infiltration (Fig. 5f). In comparison to PBS, dl312 (replication-deficient control) and dl309, Ad5 and dl704 induced significant intratumoral infiltration by CD8+ lymphocytes by day 15 (P < 0.001, t-test; Fig. 5f). To confirm CD8+ cell infiltration into tumors by a second method, we performed fluorescence-activated cell sorting (FACS)
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
ARTICLES
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
a
b
c
Figure 3 Adenovirus infectious unit titers over time in murine tumors and liver. (a) PFU/gm tumor (JC, CMT-64) on day 5 after injection with Ad5 (black), dl 309 (dark gray), dl 704 (light gray) and PUV-dl 312 (white; replication-deficient control) (±s.e.m.). (b) Same in JC tumors on days 7 and 15 after injection. (c) PFU/gm liver in murine liver 48 h after intravenous administration of viruses as described in methods. *, undetectable.
analysis of peripheral blood mononuclear cells (PBMCs) at baseline and through day 22 on all treatment and control groups; a decrease in CD8+ cells in the blood suggests extravasation out into infected tissues. Levels of CD8+ PBMCs were significantly (P = 0.05) lower after Ad5 and dl704 treatment (26% and 34% lower than controls, respectively), whereas levels were not significantly lower after dl309 treatment (P > 0.2). Tumors infiltrated with CD8+ cells also expressed interferon-γ in almost all cases. Activated dendritic cells were demonstrated in the Ad5- and dl704-treated tumors but not in dl309-treated tumors (data not shown). Efficacy in immunocompetent versus athymic mice Finally, we compared the antitumoral efficacy of Ad5 with dl309 and/or dl704 in immunocompetent mice bearing CMT-64, PDV and CMT-93 tumors. In addition, two cell lines were also treated in athymic tumor-bearing mice to determine the role of T cell–dependent immunity on efficacy (PDV, CMT-93). Of note, complete tumor regressions were only seen in the CMT-93 tumors; the other tumors showed substantial necrosis and growth delay without regressions (D.K., unpublished data). Ad5 efficacy was significantly greater than for dl309 (log rank test, P = 0.01), PBS (P = 0.005) or dl312 (P = 0.01) (Fig. 6a). Similar data were obtained in the CMT-93 model (Fig. 6b,c); Ad5 and dl704 efficacy were significantly greater than the nonreplicat-
a
ing particle control (log rank test, P = 0.04 and 0.001, respectively) whereas dl309 efficacy was not (Fig. 6b). Inactive adenovirus particles did have some efficacy in this model, however, in contrast to the other models described42. In addition, complete responses were more common in the Ad5- and dl704-treated groups than in those treated with dl309 (data not shown). Ad5 efficacy was also superior to that with dl309 in the immunocompetent PDV model (Fig. 6d); the mean tumor volume on day 21 after Ad5 treatment was significantly less than for the dl309-treated group (P = 0.005, t-test). Markedly different results were obtained when the efficacy of these viruses was tested against tumors from the same cell lines grown in athymic mice. In contrast to the superiority in efficacy of Ad5 over dl309 in immunocompetent mice, Ad5 and dl309 had equivalent efficacy in athymic mice (Fig. 6c,d). Both Ad5 and dl309 had significant efficacy compared to controls against PDV tumors in athymic mice (t-test, P = 0.001; Fig. 6d). Interestingly, the efficacy of Ad5 against CMT-93 tumors was significantly less in athymic mice than in immunocompetent mice (Fig. 6c); for example, the fraction of tumors undergoing complete responses with Ad5 and dl704 decreased from over 50% to zero (data not shown). Therefore, the efficacy advantage associated with E3B expression was demonstrable only in immunocompetent mice, and T cell–dependent immune responses either had no effect or actually increased Ad5-associated efficacy.
b
Figure 4 Adenovirus gene expression and DNA replication over time in murine tumors after injection with Ad5, dl 309 or dl 704 in immunocompetent versus athymic mice. (a,b) PDV (a) or CMT-93 (b) tumor fragments were grown in the flanks of immunocompetent (black) or athymic (white) mice and the resulting tumors were subsequently injected with Ad5 or dl 309 as shown. On the days noted, tumors were harvested, sectioned and either viral DNA synthesis (PDV tumors) or coat protein staining (CMT-93) was scored as described; the score (±s.e.m.) is shown on the y-axis (scoring system is detailed in Methods). ISH, in situ hybridization.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
DISCUSSION Replication-selective oncolytic adenoviruses represent a rapidly growing experimental therapeutic platform. A major limitation, however, has been the lack of an immunocompetent model of antitumoral efficacy. Deletions in E3 immunomodulatory genes are present but largely ignored in first-generation adenoviruses. Here we demonstrate that deletion of the E3B gene region results in decreased viral gene expression, decreased replication, accelerated virus clearance and/or reduced antitumoral efficacy. In contrast, E3 gp19 kDa deletion led to increased replication and/or antitumoral efficacy. These findings have important implications for the field of oncolytic viral therapy. The biology of the immune response and of viral infections can vary between species. Nevertheless, data from rodents can be highly
1331
ARTICLES
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
a
b
c
d
f
e
Figure 5 Acute intratumoral inflammatory response and CD8+ lymphocyte infiltration after injection with Ad5, dl 309, dl 704 or control adenovirus in immunocompetent murine tumor models. (a,b) JC and CMT-93 tumors growing in the flanks of immunocompetent mice were directly injected with the viruses listed on the x-axis. Between days 5 and 22, the tumors were harvested and scored for macrophage immunohistochemical (IHC) staining; the mean (±s.e.m.) score for each treatment group is shown as a bar with the score on the y-axis (scoring system is detailed in Methods) for JC (a) and CMT-93 (b). (c) Representative macrophage marker (F4/80) IHC-stained sections from injected JC and CMT-93 tumors (day 5). (d,e) Representative TNF (d) and interferon-γ (IFN-γ; e) IHC-stained sections from injected CMT-93 tumors (day 5). (f) CMT-64 tumors growing in the flanks of immunocompetent mice were directly injected with the viruses listed on the x-axis, and on days 8 and 15 the tumors were harvested and scored for CD8+ IHC staining (score on y-axis). The mean (±s.e.m.) score for each treatment group is shown as a bar with the score on the y-axis (scoring system is detailed in Methods). HPF, high-power field.
illuminating in many systems. We believe that the findings presented here will be applicable to cancer patients for a number of reasons. E3B region genes (10.4/14.5-RID and 14.7) protect infected cells from premature apoptosis owing to TNF, TRAIL and/or Fas activation30,40,41; these genes are known to function in a similar fashion in human and mouse cells. For example, deletion of these genes resulted in increased viral clearance and acute inflammation in the lungs and livers of infected mice22,44,45. Deletion of the E3B genes also leads to increased phospholipase-A2 activity, arachidonic acid release and acute inflammation in mice46. The E3 gp19 kDa protein binds to MHC class I molecules and leads to avoidance of cytolytic T-cell recognition47. Mouse MHC class I molecules are downregulated in similar fashion; mouse strain variation in binding affinities have been described38. Expression of the gene encoding the E3 gp19 kDa protein from the native adenovirus promoter results in increased gene expression and virus persistence in murine livers in vivo38. We have confirmed E3 gene expression and E3B-dependent downregulation of EGF receptors, as well. Finally, our results were consistent between mice with two different genetic backgrounds (C57B/6, BALB/c) and despite varied levels of Ad5 replication in the different tumor types; therefore, these models represent a broad spectrum of tumor/host interactions in which to test human adenoviruses. Of note, clinical trial results with two different E3B gene–deleted oncolytic adenoviruses are consistent with these murine model results1,16–18. In addition to rapid clearance, biopsies taken after injection of dl1520 showed induction of macrophage infiltration (S. Morley, Beatson Institute, Glasgow, UK, personal communication); TNF was induced by intravascular treatment16,48.
1332
The mechanisms responsible for the accelerated clearance of E3B deletion mutants are under investigation. Increased clearance of E3B region–deleted viruses was associated with increased macrophage infiltration and antiviral cytokine production (TNF and interferon-γ). In contrast, in athymic mice these parameters were equivalent with and without E3B. Therefore, the accelerated clearance in the absence of E3B is at least partially T-helper cell dependent. The exact mechanisms involved are not clear at this time, but it is highly likely that the increased expression levels of antiviral cytokines such as TNF play an important role49. Immunohistochemical staining for cytokines localized to areas of macrophage infiltration. T cells themselves may both stimulate macrophages and secrete antiviral cytokines themselves. Similar studies are planned in TNF and interferon-γ knockout mice to determine the role of these cytokines. In marked contrast to E3B deletion, deletion of the E3 gp19 kDa gene (dl704) resulted in enhanced viral gene expression, no increase in acute inflammatory cell infiltration, increased cytotoxic T lymphocyte infiltration and/or improved efficacy in each immunocompetent model tested. These differences were not demonstrated in athymic mice with dl704. The mechanism by which E3 gp19 kDa deletion results in improved intratumoral gene expression and/or replication is not clear at this time. Oncolytic adenovirus efficacy will be influenced by numerous factors in addition to E3 genes. In addition to replication completion with subsequent lysis, numerous other mechanisms are used by oncolytic adenoviruses to destroy tumors3. Viral proteins expressed late in infection are directly cytotoxic, including the E3 11.6 kDa adenovirus death protein and E4ORF4. Deletion of these gene products results in a sub-
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
ARTICLES a
b
c
d
oncolytic adenoviruses can be improved through coexpression of therapeutic genes50. Because the packaging of adenoviruses is limited, it is critical to identify gene regions that can be deleted without reducing efficacy (e.g., E3 gp19 kDa). Finally, unlike the dl1520 virus, selectivity must be achieved with genetic modifications that do not attenuate the virus in cancer cells5—for example, E1A-CR28 and E1B-19kDa gene deletions (T.-C.L., unpublished data). The potential synergy reported with chemotherapy in both patients and preclinical models may be optimally studied in an immunocompetent model system51. It is feasible that the interaction with chemotherapy and/or radiotherapy is mediated by the immune response and by E3 genes; some of the cytokines induced by oncolytic adenoviruses are chemo- and radiosensitizers (e.g., TNF, interferon-γ). These hypotheses are currently being tested in these immunocompetent tumor models. METHODS
Adenoviruses and cancer cell lines. Mouse CMT-64 lung and CMT–93 colorectal carcinoma cell lines were provided by Cell Services at Cancer Research, as were human carcinoma lines PT45, A2780 and A2780CP. H460 (human non–small cell lung) Figure 6 Relative antitumoral efficacy after treatment with Ad5, dl 309, dl 704 or control adenovirus and JC (murine mammary) carcinoma lines were in immunocompetent or athymic (nu/nu) mice. (a) CMT-64 tumors were grown subcutaneously in the obtained from the American Type Culture flanks of immunocompetent mice and were subsequently injected directly with PBS (), Ad5 (), Collection. PDVc57.subclone.1 (murine squamous PUV-312 (nonreplicating particle; ) and dl 309 (). (b) CMT-93 tumors were grown subcutaneously cell carcinoma) was obtained from the Beatson in the flanks of immunocompetent mice and were subsequently injected directly with PBS (), Ad5 Institute. All cell lines were maintained in DMEM (), PUV-312 (), dl 312 (E1A– control; ), dl 309 () or dl 704 (). The percentage of mice alive supplemented with 10% fetal calf serum (FCS). at each time point is estimated using the Kaplan-Meier method (± s.e.m.). (c,d) Tumors were injected Ad5, dl312 (E1A-deleted), dl309 (E3 10.4/14.5 directly with PBS (white), Ad5 (black) or dl 309 (gray) in both immunocompetent (C57B/6) and kDa– and 14.7 kDa–deleted) and dl704 (E3 gp19 athymic (nu/nu) mice harboring CMT-93 tumors (c) or PDV tumors (d), and on day 21 after treatment, kDa–deleted) were kindly provided by W. Wold, St. tumors from each group were measured. Tumor sizes are expressed as a percentage of the PBS-injected Louis University, St. Louis, MO, USA. Psoralen-UVcontrol tumors (± s.e.m.). * P < 0.05 versus PBS- or dl 309-treated tumors. inactivated dl312 adenovirus was used as nonreplicating control virus particle. All viruses were grown on HEK-293 cells as previously described5. stantial delay in cell death. In addition, E1A expression early during Psoralen UV-inactivation was carried out as previously described5. In brief, dl312 the adenovirus life cycle induces cell sensitivity to TNF-mediated was reacted with 8-methoxypsoralen at a final concentration of 0.33 µg/µl, and killing. This effect is inhibited by the E1B 19 kDa and E3B region pro- the mixture was exposed to a UV light source at 365 nm (Stratagene Linker) on teins; deletion of the E3 proteins leads to increased TNF expression in ice for 3 exposures of 10 min each. The reaction mixture was desalted on a Sxmice in vivo and enhanced cell sensitivity to TNF22. E1A has also been G-50 (10 × 1 cm) column. Inactivation of virus was verified by plaque and cytoshown to downregulate expression of vascular endothelial growth fac- pathic effect assays on HEK293 cells and was more than 1,000 times less potent than the original viral preparation.
tor and therefore antiangiogenic factors may play a role. Finally, viral gene expression and/or replication in and lysis of tumor cells have been shown to promote the induction of cell-mediated immunity to uninfected tumor cells in model systems with other viruses1; we have also demonstrated that oncolytic adenoviruses can induce tumorspecific immunity and rejection of uninfected cells after initial treatment. Finally, TNF and interferons are induced by adenovirus; these cytokines have potent antivascular and cell cycle–inhibitory properties. It is therefore not surprising that in vivo sensitivity to Ad5 cannot be predicted by a single variable (e.g., burst size). Replication-selective oncolytic adenoviruses hold promise as potential cancer therapeutics, but the single-agent efficacy of first-generation viruses has been relatively disappointing to date. These viruses might be improved in several ways. Replacement of the E3B region appears to result in improved replication and gene expression. In addition,
In vitro evaluation of adenoviruses. Adenovirus replication. Cells were seeded at 1 × 106 cells/60-mm-diameter culture dish in 10% FCS in DMEM. One day after seeding, cells were infected at 1,000 particles per cell (p.p.c.) with different viruses for 1 h in serum-free medium and incubated for another 96 h in 10% FCS in DMEM. Both cells and lysate were collected, freeze-thawed three times and titered on HEK293 cells by the limiting dilution method (determination of 50% tissue culture infective dose (TCID50)). Briefly, HEK293 cells were seeded in 96-well plates at 1 × 104 cells/well in 200 µl of 10% FCS in DMEM. Harvested samples were serially diluted and aliquots of 22 µl added to the HEK293 cells. The plates were incubated at 37 °C and cytopathic effect determined 10–12 d after infection. For each cell line, all cells infected with the different viruses and harvested at the same time points were analyzed within the same assay. Each sample was determined in duplicates or triplicates in the TCID50 assay and data from two to three separate infection studies were averaged and expressed as PFU/ml.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1333
ARTICLES
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Cell survival assay. Mouse and human cells were seeded at 1 × 104 cells/well in 96-well plates and infected with Ad5 and dl309 24 h later. After an additional 5–7 d of incubation, cell survival was determined by the MTS assay according to the protocol provided by the manufacturer. All assays were performed at least three times and EC50 values were determined as the viral dose killing 50% of cells expressed in p.p.c. The ratio of the EC50 values, dl309/Ad5, was determined as a measure of sensitivity to virally induced cytopathic effects. Establishment of in vivo tumor model. All murine experiments were approved by the Cancer Research UK animal use and safety committee before initiation. 105 (CMT-64) to 106 (JC, CMT-93) cancer cells were suspended in 100 µl normal saline and injected subcutaneously into the flanks of either C57B/6 (all lines except JC) or BALB/c (JC only) or athymic (nu/nu) mice (4–6 weeks old) and allowed to grow to an estimated volume of 40–100 µl before treatment. Because of extremely variable tumor take rates and growth kinetics with the PDV tumor line, it was necessary to harvest tumors formed from cell suspension and mince them into tumor fragments (1–2 mm diameter) that subsequently were inoculated into the flanks of mice via trocar5. Volume was estimated twice weekly using the formula: volume = (length × width × width)(3.142/6). Once enough tumors reached the size range of 40–100 µl (n = 10–12 per group), mice were stratified by tumor size and then randomly allocated to treatment groups; treatment groups were balanced by tumor size at the time of treatment initiation in all cases (t-test for tumor volumes, P ≥ 0.8). Assessment of biological endpoints and antitumoral efficacy. Tumors were injected directly with 80 µl PBS with or without virus (Ad5 or replication-incompetent control adenovirus). The injections were introduced through a single central tumor puncture site and 3–4 needle tracts were made radially from the center; virus was injected as the needle was withdrawn. For efficacy experiments, tumors were injected (1010 particles/day) on days 1, 3 and 5 for 1 (PDV, CMT93, first CMT-64 study) or 2 weeks (second CMT-64 study). Tumors were injected on days 1, 2 and 3 (1010 particles/day) for histopathology endpoint experiments. A single injection (1010 particles) was made on day 1 for in vivo viral replication experiments. Tumors were harvested at the stated time points after injection (days 5–22) for biological/histopathology endpoints (see below). For tumor growth and progression experiments, mice were followed until tumors reached ≥ 400 µl estimated volume or until symptomatic tumor ulceration occurred, whichever came first. Tumor ulcerations requiring the animal to be killed before tumor volume was ≥ 400 µl were necessarily censored from the analysis. Survival analysis was performed according to the method of KaplanMeier (log rank test for statistical significance). Ad5 gene expression, cytopathic effects and inflammatory response in vivo. Histopathology techniques. The harvested tumor samples (three tumors per group per time point) were immediately dipped into chilled (4 °C) isopentane for 3 min, frozen in liquid nitrogen and stored at –70 °C. Each sample was cut into at least 40–90 cryostat sections (5–7 µm) serially. The first three and last three were processed for hematoxylin and eosin staining to assess the general histopathology. The remaining sections were stained immunohistochemically using a streptavidin-peroxidase complex technique5. Briefly, cryostat sections were air-dried and fixed in 4% buffered formaldehyde for 15 min (E1A, hexon, H and E) or acetone for 10 min (CD4, CD8, macrophage), and then rinsed in PBS (pH 7.1). They were next blocked with normal serum for 30 min and incubated for 1 h with the relevant primary antibody at the optimized dilution (E1A, 1:50; hexon, 1:100; CD8, 1:100; F4/80, 1:100). The following primary antibodies were used: the anti-E1A antibody was a rabbit anti-human Ad2 E1A polyclonal antibody (Santa Cruz); anti-hexon was a goat anti-human Ad5 hexon protein polyclonal antibody (Autocleave Chemical); anti-CD4 and antiCD8 were purified rat anti-mouse antibodies (BD Biosciences, PharMingen). Macrophages were stained with rat anti-mouse F4/80 antigen (Serotech). The sections were then incubated with the secondary antibody, biotin-labeled swine anti-rabbit IgG (E1A), or rabbit anti-goat IgG (hexon) or rabbit anti-rat (CD8) for 30 min (all secondary antibodies from Dako; 1:300 dilution). To block endogenous peroxidase, sections were also incubated in methanol solution containing 0.3% (v/v) H2O2 for 30 min. After three washes with PBS, the streptavidin-peroxidase complex (Dako) was added, and sections were incubated for 30 min. The reaction product was developed for 3 min using a solution of PBS
1334
containing 0.1% 3-amino-9-ethylcarbazole and 0.03% H2O2 as substrates. After color development, sections were counterstained with hematoxylin and mounted. Negative control experiments were performed by omitting the primary antibody. For assessment of the necrotic proportion of tumors, each sample was cut in at least three sections at different, equally spaced depths. The degree of necrosis was assessed semi-quantitatively with the following scoring system: 0 (<25% of the cross-section on average is necrotic), 1 (25–50% necrotic) and 2 (>50% necrotic). Early and late gene expression from adenovirus were assessed by immunohistochemical analysis of tumor cross-sections. Quantitative analysis of immunohistochemical staining for E1A and hexon or in situ hybridization was as follows. All sections were analyzed by light microscopy to evaluate quantitatively the presence or absence of immunostaining and its distribution (n = 3 sections per assay per tumor). The percentage of positive cells was analyzed as follows: –, no positive cells; 1+, <10% positive cells with nuclear staining; 2+, 10–20% positive cells with nuclear staining; 3+, 21–30% positive cells with nuclear staining; 4+, >30% positive cells with nuclear staining. dl312-injected tumors were used as a negative control for E1A staining; dl312-injected tumors were uniformly negative staining. Psoralen-UV-inactivated dl312-injected tumors were used as a negative control for hexon staining; these were uniformly negative for hexon staining in the nucleus (rare, very low-intensity surface or cytoplasmic staining was seen in <10% of cells in the minority of tumors). All immunohistochemical scoring was performed by a board-certified pathologist who was blinded to treatment group, and was confirmed by a second independent pathologist. Scoring of CD8+ infiltration was performed according to the ‘running mean’ method52. The scoring was performed within the tumor stroma and in the peritumoral region; necrotic areas were avoided. A total of 15 high-powered fields were counted and the exact number of positive cells determined per field. For macrophages, tumors were scored semiquantitatively according to the extent of infiltration: 0, no positive cells; 1, a few scattered individual cells; 2, a few scattered clusters of cells; 3, moderate infiltration involving 25–50% of the tumor mass; 4, >50% of the tumor mass was densely infiltrated. All immunohistochemical scoring was performed by a board-certified pathologist who was blinded to treatment group, and was confirmed by a second independent pathologist. Cytometric phenotyping of PBMCs. Blood was collected in heparinized tubes when animals were killed. PBMCs were separated on a Ficoll-Hypaque gradient as follows. Blood was diluted 1:1 with complete medium (RPMI with 10% fetal calf serum) and layered in an equal volume of diluted blood on an equal volume of histopaque (Ficoll-Hypaque). Tubes were centrifuged at 800g for 30 min at 20–25 °C. The red blood cells were pelleted and the PBMCs collected at the interface between histopaque and layered blood. The PBMCs were washed with excess medium two or three times at 2,000g for 15 min each time. The pelleted cells from the last wash were suspended in a small volume (200–1,000 µl of 10% DMSO in FCS) depending on cell yield and transferred to cryotubes. The tubes were stored in isopropanol-filled containers at –80 °C for 1 week and transferred to liquid nitrogen containers later. The anti-CD8 antibody PE-Cy5 antimouse CD8a was used as a primary antibody; fluorescein isothiocyanate, PE and PE-Cy5 IgG1 isotype controls were also used (eBioscience). Cells were plated on 96-well plates (1 × 106/well). Antibodies were added at appropriate dilutions and incubated on ice for 40 min in the dark. Cells were washed three times with PBS/BSA in a plate centrifuge and fixed with 2% paraformaldehyde. Analysis was done on BD FACScaliber. Assessment of in vivo replication. After intratumoral injection of 1010 particles as above (JC, CMT-64) tumors were harvested 24 h, 5 d or 15 d later, weighed and flash frozen (n = 3 per time point per group). After gradual thaw on ice, the tumors were physically dispersed by a tissue grinder (Kendall) and subsequently freeze-thawed three times in DMEM. After centrifugation at 1,000g for 5 min, the supernatant was titered on HEK293 cells by plaque assay. HEK293 cells were seeded at 1 × 105 cells/well in 6-well plates 24 h before infection. Samples were diluted in serum-free medium and added to the cells. One hour after infection, the cells were overlayed with DMEM supplemented with 2% FCS and 1% SeaPlaque agarose (1 ml) (Invitrogen). After 6 d incubation at 37 °C, a second layer of agarose solution (1 ml) containing 0.1% neutral red (Sigma) as indicator
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
ARTICLES
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
was added. Formation of plaques was evaluated after 15 d. Each tumor sample was assayed in triplicate and each study was repeated twice. Finally, for assessment of viral titers in the liver, mice were injected intravenously (109 virus particles in 0.1 ml) and the liver was harvested as described above. ACKNOWLEDGMENTS We thank Cancer Research UK for support of the Molecular Oncology Unit, Hammersmith Hospital (London, UK), Russell Foxall for production and titration of viruses and Lynda Hawkins for laboratory personnel training and support. We also would like to thank Suzanne Forry-Schaudies, David Ennist and Paul Hallenbeck for helpful insights and reading of this manuscript. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Received 7 April 2003; accepted 23 July 2003 Published online at http://www.nature.com/naturebiotechnology/ 1. Kirn, D., Martuza, R.L. & Zwiebel, J. Replication-selective virotherapy for cancer: biological principles, risk management and future directions. Nat. Med. 7, 781–787 (2001). 2. Kirn, D. Replication-selective micro-organisms: fighting cancer with targeted germ warfare. J. Clin. Invest. 105, 836–838 (2000). 3. Kirn, D. Replication-selective oncolytic adenoviruses: virotherapy aimed at genetic targets in cancer. Oncogene 19, 6660–6668 (2000). 4. Bischoff, J.R. et al. An adenovirus mutant that replicates selectively in p53-deficient human tumor cells [see comments]. Science 274, 373–376 (1996). 5. Heise, C. et al. ONYX-015, an E1B gene-attenuated adenovirus, causes tumor-specific cytolysis and antitumoral efficacy that can be augmented by standard chemotherapeutic agents [see comments]. Nat. Med. 3, 639–645 (1997). 6. Yu, D., Sakamoto, G. & Henderson, D.R. Identification of the transcriptional regulatory sequences of human kallikrein 2 and their use in the construction of calydon virus 764, an attenuated replication competent adenovirus for prostate cancer therapy. Cancer Res. 59, 1498–1504 (1999). 7. Yu, D., Chen, Y., Seng, M., Dilley, J. & Henderson, D.R. The addition of adenovirus region E3 enables calydon virus 787 to eliminate distant prostate tumor xenografts. Cancer Res. 59, 4200–4203 (1999). 8. Heise, C. et al. An adenovirus E1A mutant that demonstrates potent and selective antitumoral efficacy. Nat. Med. 6, 1134–1139 (2000). 9. Freytag, S.O., Rogulski, K.R., Paielli, D.L., Gilbert, J.D. & Kim, J.H. A novel threepronged approach to kill cancer cells selectively: concomitant viral, double suicide gene, and radiotherapy [see comments]. Hum. Gene Ther. 9, 1323–1333 (1998). 10. Fueyo, J. et al. A mutant oncolytic adenovirus targeting the Rb pathway produces anti-glioma effect in vivo. Oncogene 19, 2–12 (2000). 11. Kurihara, T., Brough, D.E., Kovesdi, I. & Kufe, D.W. Selectivity of a replicationcompetent adenovirus for human breast carcinoma cells expressing the MUC1 antigen. J. Clin. Invest. 106, 763–771 (2000). 12. Johnson, L. et al. Selectively-replicating adenovirus targeting deregulated E2F activity are potent antitumor agents. Cancer Cell 1, 325–337 (2002). 13. Jakubczak, J. et al. An oncolytic adenovirus selective for retinoblastoma proteindefective tumors. Cancer Res. 63, 1490–1499 (2003). 14. Doronin, K. et al. Tumor-specific, replication-competent adenovirus vectors overexpressing the adenovirus death protein. J. Virol. 74, 6147–6155 (2000). 15. Ramachandra, M. et al. Re-engineering adenovirus regulatory pathways to enhance oncolytic specificity and efficacy. Nat. Biotechnol. 19, 1035–1041 (2001). 16. Kirn, D. Clinical research results with dl1520 (Onyx-015), a replication-selective adenovirus for the treatment of cancer: what have we learned? Gene Ther. 8, 89–98 (2001). 17. DeWeese, T. et al. A phase I trial of CV706, a replication-competent, PSA selective oncolytic adenovirus, for the treatment of locally recurrent prostate cancer following radiation therapy. Cancer Res. 61, 7464–7472 (2001). 18. Nemunaitis, J. et al. Selective replication and oncolysis in p53 mutant tumors with Onyx-015, an E1B-55kDa gene-deleted adenovirus, in patients with advanced head and neck cancer: a phase II trial. Cancer Res. 60, 6359–6366 (2000). 19. Nemunaitis, J. et al. Phase II trial of intratumoral administration of ONYX-015, a replication-selective adenovirus, in patients with refractory head and neck cancer. J. Clin. Oncol. 19, 289–298 (2001). 20. Vasey, P., Shulman, L., Gore, M., Kirn, D. & Kaye, S. Phase I trial of intraperitoneal Onyx-015 adenovirus in patients with recurrent ovarian carcinoma. J. Clin. Oncol. 20, 1562–1569 (2002). 21. Wold, W.S., Hermiston, T.W. & Tollefson, A.E. Adenovirus proteins that subvert host defenses. Trends Microbiol. 2, 437–443 (1994). 22. Sparer, T.E. et al. The role of human adenovirus early region 3 proteins (gp19K, 10.4K, 14.5K, and 14.7K) in a murine pneumonia model. J. Virol. 70, 2431–2439 (1996). 23. Rodriguez, R. et al. Prostate attenuated replication competent adenovirus (ARCA) CN706: a selective cytotoxic for prostate-specific antigen-positive prostate cancer cells. Cancer Res. 57, 2559–2563 (1997). 24. Heise, C., Williams, A., Xue, S., Propst, M. & Kirn, D. Intravenous administration of ONYX-015, a selectively-replicating adenovirus, induces antitumoral efficacy. Cancer
Res. 59, 2623–2628 (1999). 25. Heise, C., Williams, A., Olesch, J. & Kirn, D. Efficacy of a replication-competent adenovirus (ONYX-015) following intratumoral injection: intratumoral spread and distribution effects. Cancer Gene Ther. 6, 499–504 (1999). 26. Ginsberg, H.S. et al. A mouse model for investigating the molecular pathogenesis of adenovirus pneumonia. Proc. Natl. Acad. Sci. USA 88, 1651–1655 (1991). 27. Wold, W.S., Tollefson, A.E. & Hermiston, T.W. E3 transcription unit of adenovirus. Curr. Top. Microbiol. Immunol. 199, 237–274 (1995). 28. Dimitrov, T. et al. Adenovirus E3-10.4K/14.5K protein complex inhibits tumor necrosis factor-induced translocation of cytosolic phospholipase A2 to membranes. J. Virol. 71, 2830–2837 (1997). 29. Hermiston, T.W., Tripp, R.A., Sparer, T., Gooding, L.R. & Wold, W.S. Deletion mutation analysis of the adenovirus type 2 E3-gp19K protein: identification of sequences within the endoplasmic reticulum lumenal domain that are required for class I antigen binding and protection from adenovirus-specific cytotoxic T lymphocytes. J. Virol. 67, 5289–5298 (1993). 30. Lichtenstein, D.L., Krajcsi, P., Esteban, D.J., Tollefson, A.E. & Wold, W.S. Adenovirus RIDβ subunit contains a tyrosine residue that is critical for RID-mediated receptor internalization and inhibition of Fas- and TRAIL-induced apoptosis. J. Virol. 76, 11329–11342 (2002). 31. Gooding, L.R. Regulation of TNF-mediated cell death and inflammation by human adenoviruses. Infect. Agents Dis. 3, 106–115 (1994). 32. Shisler, J., Duerksen, H.P., Hermiston, T.M., Wold, W.S. & Gooding, L.R. Induction of susceptibility to tumor necrosis factor by E1A is dependent on binding to either p300 or p105-Rb and induction of DNA synthesis. J. Virol. 70, 68–77 (1996). 33. Duncan, S. et al. Infection of mouse liver by human adenovirus type 5. J. Gen. Virol. 40, 45–61 (1978). 34. Ganly, I., Mauntner, V. & Balmain, A. Productive replication of human adenoviruses in mouse epidermal cells. J. Virol. 74, 2895–2899 (2000). 35. Hayder, H. et al. Adenovirus-induced liver pathology is mediated through TNF receptors I and II but is independent of TNF or lymphotoxin. J. Immunol. 163, 1516–1520 (1999). 36. Efrat, S. et al. Adenovirus early region 3(E3) immunomodulatory genes decrease the incidence of autoimmune diabetes in NOD mice. Diabetes 50, 980–984 (2001). 37. Tufariello, J., Cho, S. & Horwitz, M.S. The adenovirus E3 14.7-kilodalton protein which inhibits cytolysis by tumor necrosis factor increases the virulence of vaccinia virus in a murine pneumonia model. J. Virol. 68, 453–462 (1994). 38. Schowalter, D.B., Tubb, J.C., Liu, M., Wilson, C.B. & Kay, M.A. Heterologous expression of adenovirus E3-gp19K in an E1a-deleted adenovirus vector inhibits MHC I expression in vitro, but does not prolong transgene expression in vivo. Gene Ther. 4, 351–360 (1997). 39. Liu, Z.X., Govindarajan, S., Okamoto, S. & Dennert, G. Fas- and tumor necrosis factor receptor 1-dependent but not perforin-dependent pathways cause injury in livers infected with an adenovirus construct in mice. Hepatology 31, 665–673 (2000). 40. Krajcsi, P. et al. The adenovirus E3-14.7K protein and the E3-10.4K/14.5K complex of proteins, which independently inhibit tumor necrosis factor (TNF)-induced apoptosis, also independently inhibit TNF-induced release of arachidonic acid. J. Virol. 70, 4904–4913 (1996). 41. Gooding, L.R. et al. The 10,400- and 14,500-dalton proteins encoded by region E3 of adenovirus function together to protect many but not all mouse cell lines against lysis by tumor necrosis factor. J. Virol. 65, 4114–4123 (1991). 42. Hallden, G. et al. Novel immunocompetent models for assessment of oncolytic adenovirus. Molecular Therapy 8, 412–424 (2003). 43. Duncan, S. et al. Infection of mouse liver by human adenovirus type 5. J. Gen. Virol. 40, 45–61 (1978). 44. Wold, W.S., Hermiston, T.W. & Tollefson, A.E. Adenovirus proteins that subvert host defenses. Trends Microbiol. 2, 437–443 (1994). 45. Hayder, H. et al. Adenovirus-induced liver pathology is mediated through TNF receptors I and II but is independent of TNF or lymphotoxin. J. Immunol. 163, 1516–1520 (1999). 46. Krajcsi, P. et al. The adenovirus E3-14.7K protein and the E3-10.4K/14.5K complex of proteins, which independently inhibit tumor necrosis factor (TNF)-induced apoptosis, also independently inhibit TNF-induced release of arachidonic acid. J. Virol. 70, 4904–4913 (1996). 47. Hermiston, T.W., Tripp, R.A., Sparer, T., Gooding, L.R. & Wold, W.S. Deletion mutation analysis of the adenovirus type 2 E3-gp19K protein: identification of sequences within the endoplasmic reticulum lumenal domain that are required for class I antigen binding and protection from adenovirus-specific cytotoxic T lymphocytes. J. Virol. 67, 5289–5298 (1993). 48. Reid, T. et al. Intra-arterial administration of a replication-selective adenovirus (dl1520) in patients with colorectal carcinoma metastatic to the liver: a phase I trial. Gene Ther. 8, 1618–1626 (2001). 49. Day, D.B., Zachariades, N.A. & Gooding, L.R. Cytolysis of adenovirus-infected murine fibroblasts by IFN-γ-primed macrophages is TNF- and contact-dependent. Cell Immunol. 157, 223–238 (1994). 50. Hermiston, T. Gene delivery from replication-selective viruses: arming guided missiles in the war against cancer. J. Clin. Invest. 105, 1169–1175 (2000). 51. Khuri, F. et al. A controlled trial of Onyx-015, an E1B gene-deleted adenovirus, in combination with chemotherapy in patients with recurrent head and neck cancer. Nat. Med. 6, 879–885 (2000). 52. Nagtegaal, I. et al. Local and distant recurrences in rectal cancer patients predicted by the immune response; a histopathological and immunohistochemical study. BMC Cancer 1, 7–11 (2001).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1335
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
LETTERS
Computational discovery of gene modules and regulatory networks Ziv Bar-Joseph1,4, Georg K Gerber1,4, Tong Ihn Lee2,4, Nicola J Rinaldi2,3, Jane Y Yoo2, François Robert2, D Benjamin Gordon2, Ernest Fraenkel2, Tommi S Jaakkola1, Richard A Young2,3 & David K Gifford1 We describe an algorithm for discovering regulatory networks of gene modules, GRAM (Genetic Regulatory Modules), that combines information from genome-wide location and expression data sets. A gene module is defined as a set of coexpressed genes to which the same set of transcription factors binds. Unlike previous approaches1–5 that relied primarily on functional information from expression data, the GRAM algorithm explicitly links genes to the factors that regulate them by incorporating DNA binding data, which provide direct physical evidence of regulatory interactions. We use the GRAM algorithm to describe a genome-wide regulatory network in Saccharomyces cerevisiae using binding information for 106 transcription factors profiled in rich medium conditions and data from over 500 expression experiments. We also present a genome-wide location analysis data set for regulators in yeast cells treated with rapamycin, and use the GRAM algorithm to provide biological insights into this regulatory network. High-throughput biological data sources hold the promise of revolutionizing molecular biology by providing large-scale views of genetic regulatory networks. Many genome-wide expression data sets are now readily available, and typical computational analyses have applied clustering algorithms to expression data to find sets of coexpressed and potentially coregulated genes1. Recent approaches have used more sophisticated algorithms; one group of researchers constructed a probabilistic model that uses expression data to link regulators to regulated genes2. Their method relies on the assumption that the expression levels of regulated genes will depend on the expression levels of regulators, which is a limitation in cases in which the expression level of the regulator does not change appropriately (e.g., cases of post-transcriptional modification). Other approaches have combined expression data with additional information, such as shared DNA binding motifs or Munich Information Center for Protein Sequences (MIPS) categories3–5, but the use of these data sources provides essentially only functional or indirect evidence of genetic regulatory interactions. These methods cannot reliably distinguish among genes that have similar expression patterns but are under the control of different regulatory networks (see Supplementary Note online for further details).
Large-scale, genome-wide location analysis for DNA-binding regulators offers a second means for identifying regulatory relationships6. Location analysis identifies physical interactions between regulators and DNA regions, providing strong direct evidence for genetic regulation. Although helpful, the usefulness of binding information is also limited, as the presence of the regulator at a promoter region indicates binding but not function. The regulator may act positively, negatively or not at all. In addition, as with all microarray-based data sources, location analysis data contain substantial experimental noise. Because expression and location analysis data provide complementary information, our goal was to develop an efficient computational method for integrating these data sources. We expected that such an algorithm could assign groups of genes to regulators more accurately than methods based on either data source alone. The GRAM algorithm begins by performing an efficient, exhaustive search over all possible combinations of transcriptional regulators indicated by the DNA-binding data with a stringent criterion for determining binding. Once a set of genes to which a common set of transcriptional regulators binds is found, the algorithm identifies a subset of these genes with highly correlated expression, which serves as a ‘seed’ for a gene module. The algorithm then revisits the binding data and, using a relaxed binding criterion, seeks to add additional genes to the module that are similarly expressed and to which the same set of transcriptional regulators binds. Our algorithm allows genes to belong to more than one module. (See the Methods section for a complete description of the GRAM algorithm.) The GRAM algorithm was applied to genome-wide location data for 106 transcription factors and over 500 expression experiments (details on the data used are available in Supplementary Table 1 online). We identified 106 gene modules, containing 655 distinct genes and regulated by 68 of the transcription factors. Figure 1 presents a visualization of these results as a graph with edges between gene modules and regulators. The gene modules abstraction allowed us to label regulator-module edges in the graph to indicate whether there is significant evidence (P < 0.05) that regulators may be functioning as activators. Because a gene module provides a link between a set of regulators and the common expression pattern of a set of genes to which the regulators bind, we can use the relationship between a regulator’s expression pattern
1MIT
Computer Science and Artificial Intelligence Laboratory, 200 Technology Square, Cambridge, Massachusetts 02139, USA. 2Whitehead Institute for Biomedical Research, Nine Cambridge Center, Cambridge, Massachusetts 02142, USA. 3MIT Department of Biology, 31 Ames Street, Room 68-132, Cambridge, Massachusetts 02139, USA. 4These authors contributed equally to this work. Correspondence should be addressed to D.G. (
[email protected]). Published online 12 October 2003; corrected 19 October 2003 (details online); doi:10.1038/nbt890
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1337
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
LETTERS Figure 1 Rich medium gene modules network.Visualization of the transcriptional regulatory network discovered by the GRAM algorithm as a graph with edges between gene modules and regulators shows that there are many groups of connected gene modules and regulators involved in similar biological processes. The network consists of 106 modules containing 655 distinct genes regulated by 68 transcription factors. In most cases in which a gene module is controlled by one or more regulators, there was previous evidence suggesting that these regulators interact physically or functionally (see Supplementary Table 3 online for details). The directed arrows point from transcription factors to the gene modules that they regulate. Blue arrows indicate discovered activator regulatory relationships (see Supplementary Table 2 online and the text for details). Gene modules are colored according to the MIPS category to which a significant number of genes belong (significance test using the hypergeometric distribution P < 0.005). Modules containing many genes with unknown function or an insignificant number belonging to the same MIPS category are colored black. When the gene modules discovered by the GRAM algorithm were compared to results generated using location data alone, the GRAM algorithm yielded almost three times as many modules significantly enriched for genes in the same MIPS category.
and the common expression pattern of genes in a module to infer whether a regulator acts as an activator. In contrast, the use of genomic location data alone allows us only to infer the presence of regulators at promoters, but not to determine the type of interaction. We searched for activator relationships by examining regulators with expression profiles that are positively correlated with the expression profiles of genes in the corresponding modules. Positive correlation indicates that higher levels of regulator expression correlate with higher levels of expression of genes in the module and suggests that the transcription factor positively regulates the expression of genes in the module. We determined the statistical significance of the activator relationships by computing correlation coefficients between all transcriptional regulators studied and all gene modules and taking the 5% positive tail of the distribution of correlation coefficients. Supplementary Table 2 online presents the 11 activators identified using the method described above. Ten of these were previously identified in the literature, suggesting that this analysis produces biologically meaningful results. Several findings obtained by analysis of the discovered gene modules suggest that the algorithm identifies biologically relevant groupings of genes. First, we found that gene modules generally identify groups of genes that function in a similar biological pathway as defined by the MIPS functional categorization7 (see Fig. 1 and Supplementary Table 3 online for details). Second, we found the gene modules to be generally accurate in assigning regulators to sets of genes whose functions are consistent with the regulators’ known roles. As an example, Gcr1 is a well-characterized regulator of glucose metabolism8,9; six of the seven genes identified in the Gcr1 module are enzymes involved in glycolysis and gluconeogenesis. Additionally, we found that in most cases in which a gene module is controlled by one or more regulators, there was previous evidence suggesting that these regulators interact physically or functionally (see Supplementary Table 4 online). For example, gene modules identify Hap2-Hap3-Hap4-Hap5, Hap4-Abf1, Ino2-Ino4,
1338
Hir1-Hir2, Mbp1-Swi6 and Swi4-Swi6 interactions. Taken together, these results provide evidence that the GRAM algorithm identifies not only biologically related sets of genes, but also relevant factors that are interacting to control the genes. Although genome-wide location data alone are potentially useful for deriving transcriptional regulatory networks, a key feature of the GRAM algorithm is its ability to compensate for technical limitations in the location data through the integration of expression data. To determine binding events in location data, researchers have previously used a statistical model and chosen a relatively stringent P-value threshold (0.001) with the intention of reducing false positives at the expense of false negatives6. The GRAM algorithm presents a useful alternative to using a single P-value threshold to predict binding events, because our method allows the P-value cutoff to be relaxed if there is sufficient supporting evidence from expression data. As an example, consider Hap4, a well-characterized regulator of genes involved in oxidative phosphorylation and respiration10. The Hap4 modules contain 28 genes that are involved in respiration and show a high degree of coregulation over the collected expression data sets (Fig. 2). Six of these genes (PET9, ATP16, KGD2, QCR6, SDH1 and NDI1) would not have been identified as Hap4 targets using the stringent 0.001 P-value threshold (P-values range from 0.0011 to 0.0036). Overall, 627 of 1,560 unique regulator-gene interactions (40%) in the rich medium network discovered by the GRAM algorithm would not have been detected using only location data and the stringent P-value cutoff. To further verify the ability of the GRAM algorithm to lower the rate of false negatives without substantially increasing the rate of false positives, we performed gene-specific chromatin-immunoprecipitation (IP) experiments for the factor Stb1 and 36 genes. The profiled genes were picked randomly from the full set of yeast genes, with representatives selected from four P-value ranges. In these experiments, we found that Stb1 bound to three additional genes that had P-values between
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
LETTERS
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
a
b
0.001 and 0.01 in the genomic location experiments and had thus been excluded under the stringent cutoff. The GRAM algorithm identified all three as genes to which Stb1 binds without adding any additional genes that were not detected in the gene-specific chromatin-IP experiments (see Supplementary Table 5 and Supplementary Methods online for full details). We also expected that the gene modules derived by the GRAM algorithm would improve on the biological relevance of gene groupings that could be inferred from location data only. Because genes that participate in the same biological pathway often have similar expression patterns, and genes in a module share not only a common set of transcription factors but also similar expression patterns, we expected that genes in modules would be more likely to be functionally related than sets of genes identified by location data alone. Indeed, we found that gene modules derived using the GRAM algorithm were almost three times more likely to show enrichment for genes in the same MIPS functional category than were sets of genes derived solely from location data. Similarly, we expected that genes in modules derived by the GRAM algorithm would be more likely to show independent evidence of coregulation by the regulators assigned to the module than would sets of genes obtained using location data alone. One line of evidence for such an improvement would be enrichment for specific DNA sequence motifs. We identified 34 transcriptional regulators that bind to genes in at least one module and have well-characterized DNA binding motifs in the Transcription Factor (TRANSFAC) database11. For each of these 34 transcriptional regulators, we constructed two lists of genes, the first using modules to which the regulator binds (generated by the GRAM algorithm) and the second using location data alone (stringent P-value cutoff of 0.001). We then computed from each list the percentage of genes that contained the appropriate known motif in the upstream region of DNA. We found that in most cases the percentage of genes containing the correct motif was higher when we used modules generated using the GRAM algorithm than when we used sets of genes generated from location data alone (see Fig. 3 and Supplementary Table 6).
Figure 2 The GRAM algorithm integrates genome-wide binding and expression data and improves on either data source alone. (a) Binding data: the GRAM algorithm can improve the quality of DNA-binding information because it uses expression data to avoid a strict statistical significance threshold. Shown is DNA-binding and expression information for the 99 genes bound by the regulator Hap4 with a P value < 0.01 using an earlier statistical model6. The blue-white column on the left indicates binding P values, and the horizontal yellow line denotes the strict significance threshold of 0.001. As can be seen, the P values form a continuum and a strict threshold is unlikely to produce good results. The blue horizontal lines on the right indicate the 28 genes that were selected for modules by the GRAM algorithm. As can be seen, 22 (79%) have a P value < 0.001, but 6 (21%) have P values above this threshold. The lower portion of the figure shows together the 28 genes selected by the GRAM algorithm, and it can be seen that they exhibit coherent expression. Further, all the selected genes are involved in respiration. Six of these genes (PET9, ATP16, KGD2, QCR6, SDH1 and NDI1) would not have been identified as Hap4 targets using the stringent 0.001 P-value threshold (P values range from 0.0011 to 0.0036). (b) Expression data: the GRAM algorithm can assign different regulators to genes with similar expression patterns that cannot be distinguished reliably using expression clustering methods alone. Hierarchical clustering of expression data was used to obtain the subtree on the left. On the right, the regulators assigned to genes by the GRAM algorithm are color coded. As can be seen, many genes with very similar expression patterns are regulated by different transcription factors.
The use of a very large set of genome-wide location and expression data allowed us to validate the results of the GRAM algorithm comprehensively for the gene modules discussed above through literature searches, independent chromatin-IP experiments, and analysis for enrichment for genes in the same MIPS category and for known DNAbinding motifs. The results of this large-scale validation gave us confidence that the GRAM algorithm would be useful in analyzing new data sources. Because biological insights are often gained by examining responses to specialized treatments or environmental conditions, we were interested in exploring the performance of the GRAM algorithm on a data set that was smaller and more biologically targeted than the rich medium data. So, we chose to examine a transcriptional regulatory subnetwork involved in the response to Tor kinase signaling. The Tor proteins are highly conserved and function as critical regulators in the response to nutrient stress12–15. Tor kinase signaling can be inhibited by the addition of the small macrolide rapamycin, which mimics nutrient starvation and results in a wide range of physiological responses including cytoskeleton reorganization, decreased translation initiation, decreased ribosome biogenesis, amino acid permease regulation and autophagy16–19. Expression analysis indicates that Tor signaling also controls transcriptional regulation of metabolic pathways involving nitrogen metabolism, glycolysis and the tricarboxylic acid (TCA) cycle15–17. The rapamycin response presented an ideal opportunity for applying the GRAM algorithm to the analysis of a novel transcriptional regulatory subnetwork. Previous studies suggest a specific set of regulators that are likely to function in the transcriptional response to rapamycin15,16. Also, several publicly available genome-wide expression data sets measuring response after rapamycin treatment are available15,16. More importantly, the fact that there is little information available about the transcriptional regulatory network involved and how this transcriptional network may contribute to the overall response to rapamycin treatment presented an opportunity for new biological insights. We selected 14 transcriptional regulators that seemed likely to function in the rapamycin response in S. cerevisiae based on evidence from
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1339
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
LETTERS
Figure 3 Motif enrichment. Genes in modules discovered by the GRAM algorithm are more likely to show independent evidence of coregulation by the regulators assigned to the module when compared to sets of genes obtained using genomic location analysis data alone, as demonstrated by an enrichment for the presence of known DNA-binding motifs. We identified 34 transcriptional regulators that bind to genes in at least one module and have well-characterized DNA binding motifs in the TRANSFAC database11. For each of these 34 transcriptional regulators, we generated a list of genes in modules bound by the regulator and a second list of genes bound by the regulator using location analysis data alone (stringent P value cutoff of 0.001). We then computed the percentage of genes from each list that contained the appropriate known motif in the upstream region of DNA. In most cases, the percentage of genes containing the correct motif was higher when we used modules generated by the GRAM algorithm than when we used sets of genes generated by location analysis data alone. See Supplementary Table 6 online for a complete list of transcription factors analyzed.
the literature, and performed genome-wide location analysis experiments (see Methods and Supplementary Table 7 online for full details). We ran the GRAM algorithm using the location data for the 14 transcription factors in rapamycin and 22 previously published expression experiments relevant to rapamycin conditions. We discovered 39 gene modules containing 317 unique genes and regulated by 13 transcription factors (see Fig. 4 and Supplementary Table 8 online). The GRAM algorithm added 192 pairs of gene-regulator interactions that would not have been identified with a strict P value (0.001) in the location analysis experiments. Because genome-wide binding experiments for the rapamycin regulatory network have not been performed before, it was not possible to verify these interactions comprehensively using literature searches. As with the rich medium gene modules network, the rapamycin regulatory network discovered by the GRAM algorithm had many features that were consistent with expectations from the literature. Twenty-three of the gene modules were found to contain a significant number of genes (P < 0.05) belonging to a single MIPS category. There were a total of nine categories, all corresponding to biological responses associated with rapamycin treatment12–14. We also found that, in general, regulators were assigned to genes that reflect functions described in previously published results. In addition to identifying established regulatory interactions, analysis of the rapamycin gene modules suggested several unexpected interactions in which regulators typically assigned to a particular biological response also appear to bind genes acting in different biological pathways. Below we give several examples of such regulatory interactions. These findings suggest models of transcriptional regulation of the rapamycin response that can be validated in further, more directed studies. A first example of an unexpected regulatory interaction involves the factors Msn2 and Msn4, which are generally regarded as
1340
Figure 4 Rapamycin gene modules network. Analysis of the rapamycin transcriptional regulatory subnetwork revealed a number of novel biological insights, including evidence that some transcriptional regulators may control genes involved in biological pathways different from those generally associated with these regulators. Further, analysis of the network suggested more complex regulatory interactions in which there is communication among modules. Such complicated network topologies may be important for facilitating rapid and flexible responses to changing environmental conditions. See the text for further details. Thirty-nine modules containing 317 unique genes and regulated by 13 transcription factors were discovered. Red arrows between transcriptional regulators indicate that the source transcription factor binds at least one module containing the target transcription factor. Modules are colored according to the MIPS category to which a significant number of genes belong (significance test using the hypergeometric distribution P < 0.05).
stress response factors and have been well studied as activators of stressrelated responses18–21. Unexpectedly, there were three gene modules in which Msn2 and Msn4 bound to a significant number of genes involved in the mating pheromone response pathway (P < 0.006). A second example involves the factor Rtg3, which is generally thought to regulate directly genes of the TCA cycle and indirectly contribute to nitrogen metabolism22–25 (products of the TCA cycle are shunted to nitrogen metabolism pathways in low- or poor-nitrogen conditions). The gene modules network suggests that Rtg3 may directly regulate genes involved in amino acid metabolism, and more specifically in nitrogen metabolism. A third example of an unexpected regulatory interaction involves Hap2, a part of the Hap2-Hap3-Hap4-Hap5 complex that has been well characterized as a regulator of genes involved in respiration22,26. Indeed, in the rich medium gene modules network, members of the Hap complex are unique among the 106 regulators profiled as the only regulators controlling modules that are significantly enriched for genes involved in respiration (P < 0.005). As expected, Hap2 regulates a module of respiration genes under rapamycin conditions. Unexpectedly, Hap2 was also found to regulate two modules containing genes involved in nitrogen metabolism. There is some genetic evidence for such cross-pathway regulation, as Hap2 was previously implicated as a regulator of two nitrogen metabolism genes27,28. Our results indicate that Hap2 participates in cross-pathway regulation more extensively than previously reported. In addition to suggesting that some transcriptional regulators may control genes involved in biological pathways different from those
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
LETTERS generally associated with these regulators, analysis of the gene modules network suggests more complex regulatory interactions in which there is communication among gene modules. Such complicated network topologies may be important for facilitating rapid and flexible responses to changing environmental conditions. As an example, we found that several transcriptional regulators may be involved in a feedforward regulatory loop in which the gene encoding a regulator is bound by another regulator and both regulators bind to a set of common genes6,29. The regulator Gat1 has been previously identified as a general activator of nitrogen-responsive genes30. We found that Gat1 is itself contained in several modules along with genes involved in nitrogen metabolism. The transcriptional regulators Dal81, Dal82, Gln3 and Hap2 bind to these gene modules. Interestingly, Gat1 also binds to several gene modules along with Dal81, Dal82 and Gln3 (see Fig. 4). Feed-forward mechanisms may be important in regulatory responses (such as the response to rapamycin) by modulating regulatory sensitivity to sustained rather than transient inputs, providing temporal control or amplifying the transcriptional response29. These findings can be validated in further directed experimental studies. The above analyses indicate that the GRAM algorithm can be useful for studying transcriptional regulatory networks using genome-wide location and expression data sources. We have made a Java implementation of the algorithm publicly available (see Supplementary Methods online), and believe that as new genome-wide location data become increasingly available, other researchers will find the algorithm helpful. As demonstrated, the algorithm can integrate sources of genome-wide location and expression data to help compensate for technical limitations in the data. Further, the inferred gene modules networks can give a clearer view of regulation than can either location or expression data sources alone. We have found that the algorithm is particularly useful for uncovering how certain regulators may act in multiple biological pathways. Overall, the GRAM algorithm facilitates a genome-wide approach to analysis of transcriptional regulatory networks that can suggest specific novel regulatory models, which can then be validated in more directed experimental studies. METHODS The GRAM (Genetic Regulatory Modules) algorithm. Below we describe the operation of the algorithm. Some details are omitted owing to space constraints; see the Supplementary Methods online for complete information as well as a Java implementation of the algorithm. Let ei denote an expression vector and bi a vector of binding P values for gene i, where there are ng genes. Let B(i,t) denote the set of all transcription factors that bind to gene i with a P value less than t, that is, the list of indices j such that bij < t. Let F ⊆ B(i,t) denote a subset of the transcription factors that bind to i. Let G(F,t) be the set of all genes i such that for any gene i ∈ G(F,t), F ⊆ B(i,t), that is, genes to which all the factors in F bind with a given significance threshold. The algorithm begins by going over all genes, and assigning each gene i to all possible sets G(F,t), where t1 is a high-stringency binding threshold and F ranges over all subsets of B(i,t). For every set of transcription factors F, the genes in G(F,t1) serve as candidates for a module regulated by F. For each such set G(F,t1) with a sufficient number n of genes (e.g., n ≥ 5), the algorithm attempts to find a ‘core’ expression profile. That is, we are seeking a point c in expression space such that for an expression similarity threshold sn, the ball centered at c of radius sn contains as many genes in G(F,t1) as possible. Denote by C(F,t1 ,c) the ‘core’ set of genes such that C(F,t1 ,c) ⊆ G(F,t1) and for each gene i ∈C(F,t1 ,c), d(ei ,c) < sn, where d is the Euclidian distance between two points. The threshold sn is determined by using all genes, and randomly sampling subsets of size n to determine the distribution of expression distances from a subset to all genes. The problem of finding a point c for a set of expression vectors is nontrivial, and cannot be optimally solved in a reasonable time given the dimensionality of the expression space (>500). Thus, we use a theoretically motivated approximation algorithm that
looks for the central point in all triplets of genes in G(F,t1) (see Supplementary Methods online for more details). The genes in C(F,t1 ,c) are used to initialize a module M(F). Conceptually, we would like to expand this module by relaxing our criteria for binding if a gene’s expression profile is sufficiently similar to those in the ‘core.’ To do so, the algorithm calculates a combined P value pi for each gene i that belongs to the expanded set C(F,t2 ,c) and does not belong to C(F,t1 ,c), where t2 > t1. The P value pi is arrived at by computing independent P values for gene i and each transcription factor in F and then combining the P values using the Fisher method. A gene i from C(F,t2 ,c) is then included in M(F) if pi < t1. This module initialization and expansion is completed for each feasible F, starting with the sets containing the largest number of factors and proceeding to the smallest. If a gene is included in a module M(F), it is masked out (not considered) when forming modules with factor subsets, M(F′) where F′ ⊆ F. That is, the algorithm will seek to explain a gene’s expression using the most specific regulatory patterns. The thresholds t1 = 0.001 and t2 = 0.01 were chosen based on experiments6 that suggested very low false positive rates for a significance threshold of 0.001. Further, the rate of false negatives was found to be relatively high for P values between 0.01 and 0.001, but decreased markedly (to <3%) thereafter. Strains. Epitope-tagged strains were generated as described6. Briefly, regulators were tagged at the C terminus by using homologous recombination to insert multiple copies of the Myc epitope coding sequence into the normal chromosomal loci of these genes. Insertion of the epitope coding sequence was confirmed by PCR and expression of the epitope-tagged protein was confirmed by western blotting analysis. Growth conditions. Strains containing epitope-tagged regulators were grown in 50 ml YPD broth (yeast extract, peptone, dextrose) at 30 °C. Cells were grown to an OD600 of 0.7–0.8 and rapamycin was then added to a final concentration of 100 nM. Cells were grown for 20 min at 30 °C in the presence of rapamycin. Genome-wide location analysis. Genome-wide location analysis was done as previously described6. Briefly, cells containing an epitope-tagged regulator were fixed with formaldehyde (1% final concentration) and then harvested by centrifugation. Cells were lysed and then sonicated to shear DNA. DNA fragments representing chromosomal regions crosslinked to a protein of interest were enriched by immunoprecipitation with an anti-epitope antibody. After reversal of crosslinking, enriched DNA was purified. The ends of DNA fragments were then blunted using T4 DNA polymerase and ligated to previously prepared linkers. The enriched DNA was then amplified and labeled with a fluorescent dye by ligation-mediated PCR. A sample of control DNA was similarly processed and labeled with a different fluorophore. Both IP-enriched and control DNA were then hybridized to a single DNA microarray. For each factor, three independently grown cell cultures were processed and scanned to generate binding information as previously described (see Supplementary Materials online for complete binding data for the rapamycin experiments). URL. The latest version of the Java implementation of the GRAM algorithm may be obtained from the authors’ website at http://psrg.lcs.mit.edu/ GRAM/Index.html. Note: Supplementary information is available on the Nature Biotechnology website. ACKNOWLEDGMENTS Z.B-J. is supported by the Program in Mathematics and Molecular Biology at Florida State University through the Burroughs Wellcome Fund Interfaces Program. G.G. is supported by a National Defense Engineering and Science graduate fellowship. This work was partially funded by a US National Institutes of Health grant. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Received 9 June; accepted 5 August 2003 Published online at http://www.nature.com/naturebiotechnology/ 1. Eisen, M.B., Spellman, P.T., Brown, P.O. & Botstein, D. Cluster analysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci. USA 95, 14863–14868 (1998).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1341
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
LETTERS 2. Segal, E. et al. Module networks: identifying regulatory modules and their condition-specific regulators from gene expression data. Nat. Genet. 34, 166–176 (2003). 3. Ihmels, J. et al. Revealing modular organization in the yeast transcriptional network. Nat. Genet. 31, 370–377 (2002). 4. Pilpel, Y., Sudarsanam, P. & Church, G.M. Identifying regulatory networks by combinatorial analysis of promoter elements. Nat. Genet. 29, 153–159 (2001). 5. Berman, B.P. et al. Exploiting transcription factor binding site clustering to identify cis-regulatory modules involved in pattern formation in the Drosophila genome. Proc. Natl. Acad. Sci. USA 99, 757–762 (2002). 6. Lee, T.I. et al. Transcriptional regulatory networks in Saccharomyces cerevisiae. Science 298, 799–804 (2002). 7. Mewes, H.W. et al. MIPS: a database for genomes and protein sequences. Nucleic Acids Res. 30, 31–34 (2002). 8. Holland, M.J., Yokoi, T., Holland, J.P., Myambo, K. & Innis, M.A. The GCR1 gene encodes a positive transcriptional regulator of the enolase and glyceraldehyde-3phosphate dehydrogenase gene families in Saccharomyces cerevisiae. Mol. Cell Biol. 7, 813–820 (1987). 9. Baker, H.V. Glycolytic gene expression in Saccharomyces cerevisiae: nucleotide sequence of GCR1, null mutants, and evidence for expression. Mol. Cell Biol. 6, 3774–3784 (1986). 10. Forsburg, S.L. & Guarente, L. Identification and characterization of HAP4: a third component of the CCAAT-bound HAP2/HAP3 heteromer. Genes Dev. 3, 1166–1178 (1989). 11. Matys, V. et al. TRANSFAC: transcriptional regulation, from patterns to profiles. Nucleic Acids Res. 31, 374–378 (2003). 12. Jacinto, E. & Hall, M.N. Tor signalling in bugs, brain and brawn. Nat. Rev. Mol. Cell Biol. 4, 117–126 (2003). 13. Crespo, J.L. & Hall, M.N. Elucidating TOR signaling and rapamycin action: lessons from Saccharomyces cerevisiae. Microbiol. Mol. Biol. Rev. 66, 579–591 (2002). 14. Raught, B., Gingras, A.C. & Sonenberg, N. The target of rapamycin (TOR) proteins. Proc. Natl. Acad. Sci. USA 98, 7037–7044 (2001). 15. Hardwick, J.S., Kuruvilla, F.G., Tong, J.K., Shamji, A.F. & Schreiber, S.L. Rapamycinmodulated transcription defines the subset of nutrient-sensitive signaling pathways directly controlled by the Tor proteins. Proc. Natl. Acad. Sci. USA 96, 14866–14870 (1999). 16. Shamji, A.F., Kuruvilla, F.G. & Schreiber, S.L. Partitioning the transcriptional program induced by rapamycin among the effectors of the Tor proteins. Curr. Biol. 10, 1574–1581 (2000). 17. Cardenas, M.E., Cutler, N.S., Lorenz, M.C., Di Como, C.J. & Heitman, J. The TOR signaling cascade regulates gene expression in response to nutrients. Genes Dev. 13,
1342
3271–3279 (1999). 18. Hasan, R. et al. The control of the yeast H2O2 response by the Msn2/4 transcription factors. Mol. Microbiol. 45, 233–241 (2002). 19. Rep, M., Krantz, M., Thevelein, J.M. & Hohmann, S. The transcriptional response of Saccharomyces cerevisiae to osmotic shock. Hot1p and Msn2p/Msn4p are required for the induction of subsets of high osmolarity glycerol pathway-dependent genes. J. Biol. Chem. 275, 8290–8300 (2000). 20. Boy-Marcotte, E., Perrot, M., Bussereau, F., Boucherie, H. & Jacquet, M. Msn2p and Msn4p control a large number of genes induced at the diauxic transition which are repressed by cyclic AMP in Saccharomyces cerevisiae. J. Bacteriol. 180, 1044–1052 (1998). 21. Martinez-Pastor, M.T. et al. The Saccharomyces cerevisiae zinc finger proteins Msn2p and Msn4p are required for transcriptional induction through the stress response element (STRE). EMBO J. 15, 2227–2235 (1996). 22. Schuller, H.J. Transcriptional control of nonfermentative metabolism in the yeast Saccharomyces cerevisiae. Curr. Genet. 43, 139–160 (2003). 23. Crespo, J.L., Powers, T., Fowler, B. & Hall, M.N. The TOR-controlled transcription activators GLN3, RTG1, and RTG3 are regulated in response to intracellular levels of glutamine. Proc. Natl. Acad. Sci. USA 99, 6784–6789 (2002). 24. Komeili, A., Wedaman, K.P., O’Shea, E.K. & Powers, T. Mechanism of metabolic control: target of rapamycin signaling links nitrogen quality to the activity of the Rtg1 and Rtg3 transcription factors. J. Cell Biol. 151, 863–878 (2000). 25. Liao, X. & Butow, R.A. RTG1 and RTG2: two yeast genes required for a novel path of communication from mitochondria to the nucleus. Cell 72, 61–71 (1993). 26. Pinkham, J.L. & Guarente, L. Cloning and molecular analysis of the HAP2 locus: a global regulator of respiratory genes in Saccharomyces cerevisiae. Mol. Cell Biol. 5, 3410–3416 (1985). 27. Dang, V.D., Bohn, C., Bolotin-Fukuhara, M. & Daignan-Fornier, B. The CCAAT boxbinding factor stimulates ammonium assimilation in Saccharomyces cerevisiae, defining a new cross-pathway regulation between nitrogen and carbon metabolisms. J. Bacteriol. 178, 1842–1849 (1996). 28. Dang, V.D., Valens, M., Bolotin-Fukuhara, M. & Daignan-Fornier, B. Cloning of the ASN1 and ASN2 genes encoding asparagine synthetases in Saccharomyces cerevisiae: differential regulation by the CCAAT-box-binding factor. Mol. Microbiol. 22, 681–692 (1996). 29. Shen-Orr, S.S., Milo, R., Mangan, S. & Alon, U. Network motifs in the transcriptional regulation network of Escherichia coli. Nat. Genet. 31, 64–68 (2002). 30. Coffman, J.A., Rai, R., Cunningham, T., Svetlov, V. & Cooper, T.G. Gat1p, a GATA family protein whose production is sensitive to nitrogen catabolite repression, participates in transcriptional activation of nitrogen-catabolic genes in Saccharomyces cerevisiae. Mol. Cell Biol. 16, 847–858 (1996).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
LETTERS
Enzymatic synthesis of antithrombin III–binding heparan sulfate pentasaccharide Balagurunathan Kuberan1, Miroslaw Z Lech1, David L Beeler1, Zhengliang L Wu1 & Robert D Rosenberg1,2 Heparan sulfate (HS) proteoglycans are crucial to numerous biological processes and pathological conditions, but to date only a few HS structures have been synthesized and characterized with regard to structure-function relationships. Because HS proteoglycans are highly diverse in structure, there are substantial limitations on their synthesis by classical chemical means, and thus new methods to rapidly assemble bioactive HS structures are needed. Here we report the biosynthesis of bioactive HS oligosaccharides using an engineered set of cloned enzymes that mimics the Golgi apparatus in vitro. We rapidly and efficiently assembled the antithrombin III–binding pentasaccharide in just 6 steps, in contrast to the approximately 60 steps needed for its chemical synthesis, with an overall yield at least twofold greater and a completion time at least 100 times faster than for the chemical process. HS proteoglycans are negatively charged, linear polysaccharides that interact with a variety of proteins at the cell surface. They regulate many different biological systems, including blood coagulation, viral infection, cell growth, tumor metastasis and various developmental processes1. Unfortunately, structure-function relationships for HS have been elucidated in only a few cases, including antithrombin III (ATIII) and fibroblast growth factor1,2. The best known of these structure-function relationships involves the pentasaccharide that binds to and accelerates the action of ATIII3,4. A major barrier to determining HS structure-function relationships, and uncovering the many biological roles of these compounds, is the great difficulty of synthesizing biologically active HS fragments. We have previously cloned and expressed virtually all of the enzymes, and their isoforms, involved in HS biosynthesis. We envisioned that these enzymes could be used to synthesize homogeneous bioactive HS oligosaccharides in vitro in a relatively simple and rapid manner. Here we report the successful implementation of this synthetic approach, in which we generated a well-characterized ATIII-binding HS pentasaccharide. This pentasaccharide binds to ATIII highly specifically and induces a conformational change sufficient to promote rapid inhibition of blood coagulation3. The structural features of the pentasaccharide required for its binding are a 3-O and a 6-O sulfate at two sugar residues (Fig. 1; 3 and 1, respectively)5,6.
In HS, glucosamine is linked by α-glycosylation to an adjacent glucuronic acid. It is difficult to generate this α-linkage with high stereoselectivity, and a nonparticipating C-2 group such as an azido group is also required during chemical glycosylation. In addition, the carboxyl group is generally masked during glycosylation as a protected hydroxyl group, because the presence of carboxyl group makes uronic acid a poor glycosyl donor or acceptor. Multiple protection and deprotection steps are thus necessary to introduce sulfate groups in a regioselective manner. Hence, chemical synthesis of the ATIII-binding pentasaccharide (first accomplished by Sinay and coworker7,8) is a daunting task that requires as many as 60 steps and has a yield of less than ∼0.5%7. The biosynthetic approach described here requires substantially fewer steps and has at least a twofold greater yield (see Supplementary Tables 1 and 2 online). Our results suggest that this new enzymatic method can be used to rapidly generate homogenous and biologically active heparan oligosaccharides of any size or structure. This simplified approach should aid in establishing structure-function relationships for these molecules and identifying their roles in many biological systems. A nonsulfated polysaccharide 1 of Escherichia coli strain K5, resembling the unmodified nascent HS chain, was used as a starting material in the enzymatic synthesis (Fig. 2; ref. 9). We treated polysaccharide 1 with N-deacetylase-N-sulfotransferase 2 (NDST2) in the presence of 3′-phosphoadenosine-5′-phosphosulfate (PAPS) to prepare polysaccharide 2, which contains both N-sulfated glucosamine and unmodified, intact N-acetylglucosamine residues10. We ascertained the extent of modification of polysaccharide 1 by NDST2 by treating a small portion of the reaction mixture with heparitinases and analyzing this mixture by capillary high-performance liquid chromatography coupled to mass spectrometry (LC-MS)11. The reaction was quenched when 70% of N-acetylglucosamine units were N-deacetylated and N-sulfated. Next, polysaccharide 2 was partially cleaved to produce hexasaccharide 3. Polysaccharide 2 was treated with heparitinase I and the resulting oligosaccharide mixture of different sizes and different composition was purified to homogeneity by preparative high-performance liquid chromatography (HPLC). Each fraction was analyzed by mass spectrometry. The fractions containing hexasaccharide 3 (molecular weight 1,213 Da) were subjected to further analysis. Sequential treatment of hexasaccharide 3 with ∆4,5-glycuronidase and α-N-acetylglucosaminidase resulted in a tetrasaccharide with a molecular weight of 852 Da, confirming the hexasaccharide’s structure12,13. If the penulti-
1Department
of Biology, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. 2Division of Molecular and Vascular Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts 02215, USA. Correspondence should be addressed to R.R. (
[email protected]). Published online 5 October 2003; doi:10.1038/nbt885
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1343
LETTERS 1
OH
2
OSO 3–
O HO
– OOC
RHN O R = acetyl or SO3– R′ = H or SO3–
A
3
O
O OH
HO A
OH
1
OH O
O
N O HO cH
HO OH
AcHN O
OR′
O
HO
–OOC O
O
NDST2
4
O
– O SHN 3
O 3SO
O HO
5
O –
OH
– OO C
– OOC
– O SO 3
O
OH O
– OO C
O
O
N O
O
AcH
HO
O
HO
OH
OH O O HO
– OOC
OH
2
OH O
O HN O HO –O3S
HO
O
HO O3SHN
OH
OR′
O
Heparitinase I – OOC
O –
OH
O
HO
O 3 SHN
O
OH O
–OOC
O
OH
O
N O HO
– OOC
OH
O
O
HO O
O
HO
– OOC
HO
O
O
HO
HO – O SHN 3
– OOC
O N O HO AcH
OH
OH
Epimerase 2-OST1 –OOC
O
O N O HO – O 3SH
HO
– O SHN 3
OH
OH
O
O
4
OH
OH
HO
3
OH
O
A cH
HO
OH
Figure 1 ATIII-binding heparan sulfate pentasaccharide. Critical sulfate groups are shown in red. The α-glycosidic linkage, the most difficult to make by chemical methods, is shown in blue. (1) N-sulfoglucosamine carrying the critical 6-O sulfate group; (2) glucuronic acid flanked by two critical glucosamines; (3) N-sulfoglucosamine residue carrying the most critical 3-O sulfate group; (4) 2-O-sulfated iduronic acid; (5) N-sulfoglucosamine located at the reducing end.
O O
OH HO
– O SHN 3
– O SO 3
6-OST1 and 6-OST2a – OOC
HO
mate residue of the hexasaccharide had been N-sulfoglucosamine, it would have been resistant to α-N-acetylglucosaminidase treatment and would not have yielded the tetrasaccharide with a molecular weight of 852 Da. Thus the loss of 361 Da upon the treatment of hexasaccharide 3, which contains two N-sulfoglucosamine units, with these exoenzymes clearly indicates that the N-acetylglucosamine unit is desirably located at the nonreducing end (see Supplementary Methods online). After comprehensive structural analysis, hexasaccharide 3 was treated with C-5 epimerase and 2-O-sulfotransferase 1 (2-OST1) to prepare hexasaccharide 4 (ref.14). The C-5 epimerase can act only on the glucuronic acid flanked by N-sulfoglucosamine units, converting it to iduronic acid (Fig. 2, in blue)15,16. The glucuronic acid located at the reducing side of N-acetylglucosamine is resistant to epimerase, and thus treatment of hexasaccharide 3 with the epimerase led to the exclusive formation of a single product containing iduronic acid next to N-sulfoglucosamine units located at the reducing end. 2-OST1 preferentially sulfates iduronic acid located at the reducing side of N-sulfoglucosamine17. Thus, immediately after epimerase acted on hexasaccharide 3, preferential sulfation of the newly generated iduronic acid by 2-OST1 resulted in the formation of hexasaccharide 4. This tandem modification has a yield of ∼15%. Hexasaccharide 4 was then treated with 6-OST1 and 6-OST2a to prepare hexasaccharide 5 (ref. 18). There are three 6-O sulfation sites available for 6-O-sulfotransferases. The sites of 6-O sulfation were determined by disaccharide analysis using LC-MS. Only two glucosamine units, located at the nonreducing end and in the middle of the hexasaccharide, were 6-O sulfated, whereas the glucosamine unit at the reducing end was not modified. Next, hexasaccharide 5 was treated with ∆4,5-glycuronidase to selectively remove the terminal unsaturated uronic acid residue (generated by the action of heparitinases on polysaccharide 2) at the nonreducing end12. This resulted in quantitative generation of pentasaccharide 6. The final step was 3-O sulfation of pentasaccharide 6 by 3-OST1 to generate the anticoagulant pentasaccharide 7 (refs. 19–21). All of the sulfotransferase-mediated reactions gave quantitative conversion. The final modification was carried out using PAPS enriched either with the stable 34S isotope (PAP34S) for structural characterization or with the radioactive 35S isotope (PAP35S) for gel mobility shift analysis22. The identity and purity of the final product were verified by LC-MS (Fig. 3a), which can resolve oligosaccharides that differ in number
O
– O SO 3 O
1344
O
–OOC N O HO cH
OH
O
HO HO
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
O
HO
OH
O
– OOC O
O
HO
N AcH
– OOC
– O SO 3
O
O
N – O 3SH O
OH
O O
OH HO
– O SHN 3
– O SO 3
∆
HO HO
O
O
– OOC
O
– O SO 3
4, 5
O
-Glycuronidase
– OOC
O N O HO AcH
HO OH
5
OH
OH
HO
OH
–O SO 3
– OOC
O O HO
O
6
OH
OH – O SHN O 3
O O
OH HO
– O SO 3
– O SHN 3
3-OST1, PAP34S – O SO 3 HO HO
O
– OOC
N O HO AcH
O
OH
– O SO 3 O – O 34SO 3
O
– OOC
– O 3SHN O
– O SO 3
7
OH OH O
O O
OH HO
– O SHN 3
Figure 2 Synthesis of ATIII-binding heparan sulfate pentasaccharide. Hexasaccharide 3 was synthesized from K5 polysaccharide in two steps using N-deacetylase-N-sulfotransferase (NDST2), a bifunctional enzyme, and heparitinase I, a bacterial lyase enzyme. Pentasaccharide 6, a precursor structure for 3-OST1, was synthesized from hexasaccharide 3 in three consecutive steps using recombinant HS biosynthetic enzymes (C5-epimerase, 2-OST1, 6-OST1 and 6-OST2a) and a bacterial enzyme (∆4,5- glycuronidase) that removes the terminal unsaturated uronic acid. The final modification by 3-OST1 generated ATIII binding pentasaccharide 7.
and/or pattern of sulfate groups11. The observed single peak showed an abundant quasimolecular ion of m/z 752.16 corresponding to [M+1(DBA)-3H]2–, consistent with the calculated molecular weight for pentasaccharide 7 with the stable isotope incorporated (Fig. 3b). The use of the stable isotope as a mass spectrometric probe further strengthened our claim about the identity and purity of the final product. Gel mobility shift analysis confirmed that the radiolabeled pentasaccharide 7 can specifically bind to ATIII (Fig. 3c). Furthermore, omission of either the 3-O sulfate or the 6-O sulfate at the nonreducing end completely eliminated oligosaccharide binding to ATIII, as predicted by our earlier biochemical studies22. The specific binding of pentasaccharide 7 to ATIII, as visualized by gel mobility shift analysis, confirmed the correct placement and alignment of these two critical groups during synthesis. The observed broad, fast-migrating band corresponded to 35S-labeled APS (AP35S) derived from the degradation of PAP35S (Fig. 3c, right lane). We chose to carry out the binding analysis at this stage, without any further purification that might bias the characteristics of the bioactive product. Thus, enzymatic synthesis of ATIII-binding heparan sulfate pentasaccharide was accomplished in six steps, far fewer than in chemical approaches, and with at least a twofold better yield (∼1.1%).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
LETTERS a
proteins to further the understanding of various biological systems.
c
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
METHODS Materials. Heparan sulfate sulfotransferases (NDST2, 2-OST1, 3-OST1, 6-OST1 and 6-OST2a) and C-5 epimerase were all cloned and expressed in a baculovirus system. PAPS was purchased from Calbiochem. PAP34S and PAP35S were prepared using a reported procedure22. All other chemicals were purchased from Sigma. Solvents were from Aldrich and HPLC columns were from Vydac. The labeling 2× sulfotransferase buffer22 consists of 50 mM 2-[N-morpholino]ethanesulfonic acid (MES; pH 7.0), 1% (wt/vol) Triton X-100, 5 mM MgCl2, 5 mM MnCl2, 2.5 mM CaCl2, 0.075 mg/ml protamine chloride and 1.5 mg/ml BSA. α-N-acetylglucosaminidase was obtained from Glyko. ∆4,5-glycuronidase was a generous gift from from K. Yoshida (Seikagaku Corp.) and adenosine 5′-phosphosulfate (APS) kinase was a generous gift from I. Segel (Univ. California, Davis).
b
LC-MS analysis. An Ultimate capillary HPLC workstation (Dionex) was used for microseparation. UltiChrom software was used in data acquisition and analysis. A gradient elution was performed, Figure 3 Structural and functional analysis of synthetic pentasaccharide 7. (a) Extractive ion using a binary solvent system composed of water chromatogram from LC-MS analysis of pentasaccharide 7. The peak at 65.8 min corresponds (eluent A) and 70% aqueous methanol (eluent B), to pentasaccharide 7; the single peak suggests that pentasaccharide 7 should have a single both containing 8 mM acetic acid and 5 mM sulfation pattern and be homogeneous. (b) Mass spectrum of pentasaccharide 7. The observed dibutylamine as an ion-pairing agent. HPLC separaabundant quasimolecular ion for pentasaccharide 7 was at m/z 752.15, which corresponds to tions were performed on a 0.3 mm × 250 mm C18 2– [M+1(DBA)-3H] . Two other minor molecular ions observed were 687.57 and 816.74, which silica column (Vydac), with the flow rate set at 5 2– 2– correspond to [M-2H] and [M+2(DBA)-4H] , respectively. Analytes are often observed in µl/min. Mass spectra were acquired on a Mariner adduction with the ion-pairing agent dibutyl ammonium acetate (DBA). (c) Gel mobility shift BioSpectrometry Workstation electrospray ionizaanalysis of ATIII binding pentasaccharide 7. Radiolabeled pentasaccharide 7 was mixed with tion (ESI) time-of-flight mass spectrometer 1 µg of ATIII. Complex formation was analyzed by nondenaturing gel electrophoresis (4.5% (PerSeptive Biosystems). In the negative-ion mode, polyacrylamide). The mobility of radiolabeled pentasaccharide 7 was compared without and with N2 was used as a desolvation gas as well as a nebuATIII (left and right lanes, respectively). lizer. Conditions for ESI-MS were as follows: nebulizer flow 0.75 l/min, nozzle temperature 140 °C, drying gas (N2) flow 1.2 l/min, spray tip potential We have shown that enzymatic synthesis of the bioactive HS pen- 2.8 kV, nozzle potential 70 V and skimmer potential 12 V. Negative ion spectra tasaccharide structure that specifically binds to ATIII can be accom- were generated by scanning the range of m/z 40–2,000.
plished relatively simply and rapidly as compared to classical chemical synthesis. The ATIII-binding pentasaccharide was chosen as an initial synthetic target because more is known about this product than any other oligosaccharide. This includes not only its structure but also which residues within the oligosaccharide are functionally important. A set of analytical techniques for evaluating the structure and function of this oligosaccharide is also presented here. Our approach can be generalized to oligosaccharides with structures different from that of the ATIII-binding pentasaccharide by making minor alterations in the enzyme-synthetic method, including changes in the order of the addition of enzymes, the use of different isoforms, or the use of exocatabolic enzymes along with glycosyltransferases to remodel oligosaccharides from the nonreducing end. These latter enzymes are available commercially or can be produced using published procedures. In addition, HS polymerase could be employed to synthesize the starting E. coli K5 polysaccharide or oligosaccharide. Thus, our approach can be generalized to synthesize oligosaccharides of virtually any size or structure. Although our procedure was carried out at the microgram scale, on the basis of experiments carried out during this investigation we envision that this enzymatic approach should be scalable by at least 1,000-fold. The ready availability of these structures should allow the identification of ligand proteins that recognize specific HS structures and the use of these ligand
Expression of heparan sulfate sulfotransferases and epimerase. All of the HS biosynthetic enzymes were expressed and purified in a baculovirus system as described previously19–21. In brief, a donor plasmid for the preparation of recombinant baculovirus expressing a soluble form of the epimerase was constructed in pFastBac HT plasmid modified by the insertion of honeybee mellitin signal peptide ahead of the histidine tag. HS biosynthetic enzyme–recombinant baculovirus was prepared using the donor and the Bacto-Bac baculovirus expression system (Life Technologies) according to the manufacturer’s protocol, except that recombinant bacmid DNA was purified with an endotoxin-free plasmid purification kit (Qiagen) and transfection of Sf9 cells was scaled up to use ∼15 µg of bacmid DNA and ∼2.5 × 107 exponentially growing cells in four 100-mm dishes and amplified twice. The resulting hightiter viral stock was stored in aliquots (0.75 ml) sufficient to infect ∼3 × 108 cells, as determined by Western blotting of medium from infected cells using anti(His)6 antibody (Qiagen). Infected cells were plated and incubated at 26 °C for 90–96 h. The pooled medium was subjected to further purification to obtain pure enzymes. Preparation of polysaccharide 2. Polysaccharide 1 was harvested from E. coli K5 bacterial cells as reported earlier9. Polysaccharide 1 (0.625 mM) was taken in 100 µl of 2× sulfotransferase buffer22. Next, 90 µl water, 10 µl PAPS and 1 µl of NDST2 were added. The N-sulfation was carried out at 37 °C, and small aliquots were withdrawn and treated with heparitinases and analyzed by LC-MS to quantify the extent of N-sulfation in polysaccharide 2.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1345
LETTERS
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Isolation of hexasaccharide 3 using HPLC. Polysaccharide 2 (2.5 mM) was subjected to partial digestion using 1.0 mU of heparitinase I (EC 4.2.2.8) in 100 µl of 40 mM ammonium acetate buffer (pH 7.0) containing 3.3 mM CaCl2 at 37 °C for 30 min to obtain oligosaccharides. These were then purified to homogeneity on a C18 reverse-phase column (0.46 × 25 cm). The individual fractions were analyzed by mass spectrometry and the fractions containing hexasaccharide 3 were pooled, dried and treated sequentially with ∆4,5-glycuronidase and α-N-acetylglucosaminidase to confirm the hexasaccharide 3 structure. Preparation of hexasaccharide 4. Hexasaccharide 3 (0.125 mM) was incubated with purified glucuronyl C-5 epimerase, 2-OST1 and PAPS at 37 °C in a volume of 200 µl containing either 25 mM HEPES, 40 mM CaCl2, pH 6.5, or alternatively 25 mM MES (pH 7.0). After incubation for 24 h, the fresh epimerase, 2-O-sulfotransferase 1 and PAPS were added and incubation was extended for an additional 24 h. At the end of 48 h, the reaction mixture was purified using a Biogel P6 column (Bio-Rad) or DEAE ion-exchange column (Pharmacia) and analyzed by LC-MS for structural elucidation. The overall yield of this tandem modification step, which was calculated from mass spectrometric analysis, was about 15 ± 5%. Synthesis of pentasaccharide 7. Hexasaccharide 4 (0.1 mM) was treated with 500 µl of 2× sulfotransferase buffer22, 1 µl 6-OST1 enzyme and 1 µl 6-OST2a enzyme, 25 µl PAPS or 25 µl PAP35S, and 473 µl water. The reaction was incubated at 37 °C overnight, then diluted to 1 ml with water and purified on a DEAE column to obtain hexasaccharide 5. Hexasaccharide 5 was then treated with glycuronidase (0.1 mU) in 200 µl of 40 mM ammonium acetate containing 1 mM calcium chloride buffer (pH 7.0) at 37 °C for 36 h, yielding pentasaccharide 6. After purification, this was mixed with 100 µl of 2× sulfotransferase buffer22, 1 µl 3-OST1 enzyme, 10 µl PAP34S or PAP35S, and 90 µl water. The reaction was incubated at 37 °C overnight, then diluted to 1 ml with DEAE wash buffer and purified on a DEAE column. Alternatively, the reaction, which was carried out with radioactive PAP35S, was stopped by heating at 70 °C; the reaction mixture was then centrifuged at 10,000g for 3 min and the supernatant used directly for gel mobility shift analysis. Gel mobility shift assay. The heparin-ATIII binding buffer contained 12% glycerol, 20 mM Tris-HCl (pH 7.9), 100 mM KCl, 1 mM EDTA and 1 mM DTT. For a typical 20 µl binding reaction, radiolabeled polysaccharide (∼10,000 cpm) was mixed with ATIII (1 µg) in binding buffer. The reaction mixture was incubated at room temperature (23 °C) for 20 min and then applied to a 4.5% native polyacrylamide gel (with 0.1% of bisacrylamide). The gel buffer was 10 mM Tris (pH 7.4), 1 mM EDTA, and the electrophoresis buffer was 40 mM Tris (pH 8.0), 40 mM acetic acid, 1 mM EDTA. The gel was run at 6 V/cm for 1–2 h with an SE 250 Mighty Small II gel apparatus (Hoefer Scientific Instruments). After electrophoresis, the gel was transferred to 3MM-grade gel blotting paper (Schleicher & Schuell) and dried under vacuum. The dried gel was autoradiographed with a PhosphorImager 445SI (Molecular Dynamics). The image was analyzed with NIH Image 1.60 and the band intensities were evaluated. Note: Supplementary information is available on the Nature Biotechnology website. ACKNOWLEDGMENTS We thank Irvin Segel for his generous gift of APS kinase. We thank Keiichi Yoshida, Seikagaku Corporation, for generously providing us with ∆4,5-glycuronidase.
1346
COMPETING INTERESTS STATEMENT The authors declare competing financial interests; see the Nature Biotechnology website for details. Received 20 May; accepted 7 August 2003 Published online at http://www.nature.com/naturebiotechnology/ 1. Bernfield, M. et al. Functions of cell surface heparan sulfate proteoglycans. Annu. Rev. Biochem. 68, 729–777 (1999). 2. Capila, I. & Linhardt, R.J. Heparin–protein interactions. Angew. Chem. Intl. Edn. 41, 391–412 (2002). 3. Damus, P.S., Hicks, M. & Rosenberg, R.D. Anticoagulant action of heparin. Nature 246, 355–357 (1973). 4. Rosenberg, R.D. & Damus, P.S. The purification and mechanism of action of human antithrombin-heparin cofactor. J. Biol. Chem. 248, 6490–6505 (1973). 5. Atha, D.H., Stephens, A.W. & Rosenberg, R.D. Evaluation of critical groups required for the binding of heparin to antithrombin. Proc. Natl. Acad. Sci. USA 81, 1030–1034 (1984). 6. Desai, U.R., Petitou, M., Bjork, I. & Olson, S.T. Mechanism of heparin activation of antithrombin. Role of individual residues of the pentasaccharide activating sequence in the recognition of native and activated states of antithrombin. J. Biol. Chem. 273, 7478–7487 (1998). 7. Sinay, P. et al. Total synthesis of a heparin pentasaccharide fragment having high affinity for antithrombin-III. Carbohyd. Res. 132, C5–C9 (1984). 8. Petitou, M. et al. Synthesis of thrombin-inhibiting heparin mimetics without side effects. Nature 398, 417–422 (1999). 9. Vann, W.F., Schmidt, M.A., Jann, B. & Jann, K. The structure of the capsular polysaccharide (K5 antigen) of urinary-tract-infective Escherichia coli 010:K5:H4. A polymer similar to desulfo-heparin. Eur. J. Biochem. 116, 359–364 (1981). 10. Orellana, A., Hirschberg, C.B., Wei, Z., Swiedler, S.J. & Ishihara, M. Molecular cloning and expression of a glycosaminoglycan N-acetylglucosaminyl N-deacetylase/N-sulfotransferase from a heparin-producing cell line. J. Biol. Chem. 269, 2270–2276 (1994). 11. Kuberan, B. et al. Analysis of heparan sulfate oligosaccharides with ion pair-reverse phase capillary high performance liquid chromatography-microelectrospray ionization time-of-flight mass spectrometry. J. Amer. Chem. Soc. 124, 8707–8718 (2002). 12. Warnick, C.T. & Linker, A. Purification of an unusual ∆4,5-glycuronidase from flavobacteria. Biochemistry 11, 568–572 (1972). 13. Weber, B., Blanch, L., Clements, P.R., Scott, H.S. & Hopwood, J.J. Cloning and expression of the gene involved in Sanfilippo B syndrome (mucopolysaccharidosis III B). Hum. Mol. Genet. 5, 771–777 (1996). 14. Li, J. et al. Biosynthesis of heparin/heparan sulfate. cDNA cloning and expression of D-glucuronyl C5-epimerase from bovine lung. J. Biol. Chem. 272, 28158–28163 (1997). 15. Kusche, M., Hannesson, H.H. & Lindahl, U. Biosynthesis of heparin. Use of Escherichia coli K5 capsular polysaccharide as a model substrate in enzymic polymermodification reactions. Biochem. J. 275, 151–158 (1991). 16. Razi, N. et al. Structural and functional properties of heparin analogues obtained by chemical sulphation of Escherichia coli K5 capsular polysaccharide. Biochem. J. 309, 465–472 (1995). 17. Rong, J., Habuchi, H., Kimata, K., Lindahl, U. & Kusche-Gullberg, M. Substrate specificity of the heparan sulfate hexuronic acid 2-O-sulfotransferase. Biochemistry 40, 5548–5555 (2001). 18. Habuchi, H. et al. The occurrence of three isoforms of heparan sulfate 6-O-sulfotransferase having different specificities for hexuronic acid adjacent to the targeted N-sulfoglucosamine. J. Biol. Chem. 275, 2859–2868 (2000). 19. Liu, J., Shworak, N.W., Fritze, L.M.S., Edelberg, J.M. & Rosenberg, R.D. Purification of heparan sulfate D-glucosaminyl 3-O-sulfotransferase. J. Biol. Chem. 271, 27072–27082 (1996). 20. Liu, J. et al. Heparan sulfate D-glucosaminyl 3-O-sulfotransferase-3A sulfates N-unsubstituted glucosamine residues. J. Biol. Chem. 274, 38155–38162 (1999). 21. Shworak, N.W. et al. Multiple isoforms of heparan sulfate D-glucosaminyl 3-O-sulfotransferase—isolation, characterization, and expression of human cDNAs and identification of distinct genomic loci. J. Biol. Chem. 274, 5170–5184 (1999). 22. Wu, Z.L., Zhang, L., Beeler, D.L., Kuberan, B. & Rosenberg, R.D. A new strategy for defining critical functional groups on heparan sulfate. FASEB J. 16, 539–545 (2002).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
FOCUS ON OPTICAL IMAGING
PERSPECTIVE
Toward fluorescence nanoscopy Stefan W Hell For more than a century, the resolution of focusing light microscopy has been limited by diffraction to 180 nm in the focal plane and to 500 nm along the optic axis. Recently, microscopes have been reported that provide three- to sevenfold improved axial resolution in live cells. Moreover, a family of concepts has emerged that overcomes the diffraction barrier altogether. Its first exponent, stimulated emission depletion microscopy, has so far displayed a resolution down to 28 nm. Relying on saturated optical transitions, these concepts are limited only by the attainable saturation level. As strong saturation should be feasible at low light intensities, nanoscale imaging with focused light may be closer than ever. In 1873, Ernst Abbe discovered that the resolution of a focusing light microscope is limited by diffraction1. This physical insight became one of the most prominent paradigms in the natural sciences, with paramount importance in biology. Although the advent of confocal and multiphoton fluorescence microscopes facilitated three-dimensional imaging, they did not really improve the resolution2–4. In the best case, these and other established focusing microscopes resolve 180 nm in the focal plane (x,y) and only 500–800 nm along the optic axis (z)5. Fluorescence microscopes routinely detect single molecules if their fellow molecules are far enough apart6. By the same token, they discern several molecules at arbitrary distance, provided none of them is of the same kind. Telling apart fluorescent labels that are spectrally distinct is not challenged by diffraction. Therefore, resolution must be confused neither with single-molecule sensitivity7 nor with measuring of distances between distinct fluorescent markers8–12. Notwithstanding the importance of these issues, this review is concerned with improving the ability of a light microscope to distinguish identical fluorescent items at high spatial density, such as the distribution of a green fluorescent protein (GFP) fusion protein in a cell. Likewise, it is concerned with methods for producing fluorescent volumes that are fundamentally smaller than those of confocal and multiphoton microscopy. The approaches discussed rely on visible light and regular objective lenses. Moreover, they are designed for operation at 18–37 °C and are applicable to the imaging of live cells. According to diffraction theory, the resolution of a focusing light microscope is related to the size of its focal spot. The spot size can be decreased by using shorter wavelengths and larger aperture angles1,13 (Box 1), but this strategy has the shortcoming that wavelengths λ < 350 nm are incompatible with live cell imaging and the lens half-aperture Max-Planck-Institute for Biophysical Chemistry, Department of NanoBiophotonics, Am Fassberg 11, 37077 Göttingen, Germany. Correspondence should be addressed to S.W.H. (
[email protected]). Published online 31 October 2003; doi:10.1038/nbt895
is technically limited to 70°. The restricted aperture angle is also responsible for the poorer resolution along the optic axis. Therefore, during the past decade, concepts have appeared for improving the axial resolution14–16 by combining the aperture of opposing lenses. The up to sevenfold-improved axial sectioning capability of these techniques, termed 4Pi microscopy17,18 and I5M microscopy19, should be a strong incentive to map organelles, the nucleus and protein distributions at higher resolution. The notion of the virtually insurmountable diffraction barrier stems from the fact that focusing always results in a blurred spot of light. Consequently, near-field optical microscopes abandon focusing altogether20. To localize the interaction of the light with the object to subdiffraction dimensions, near-field microscopes use ultrasharp tips or tiny apertures that confine imaging to surfaces. Consequently, this approach does not allow the noninvasive imaging of live cells. Defeating the resolution limit without defeating diffraction per se is evidently preferable. Although this formidable problem has challenged many physicists21,22, feasible proposals did not emerge in the past. Nevertheless, it had long been clear that the crossing of the diffraction barrier would be enabled by a nonlinear relationship between the intensity of the illumination light and the signal to be measured. Such a nonlinear relationship is offered by the fluorescence induced by m-photon absorption (see article by Webb, this issue), which has led to the long-standing popular notion that superresolution is readily attained by the cooperative absorption of many photons. But as we now know, m-photon excitation (m > 1) of a fluorophore has not opened up the nanoscale yet and is unlikely to do so in the future. Although it is true that m-photon absorption occurs mainly at the center of the spot, the concomitant narrowing of the effective spot is spoiled by the fact that m-photon excitation usually entails photons of m times lower energy (that is, m times longer wavelength) and thus m times larger focal spots to begin with. In addition, this approach requires very high intensities23. Therefore, my collaborators and I24-26 devised m-photon excitation concepts for working at lower intensities and without photon-energy subdivision, but they rely on very specific fluorophores. Moreover, in spite of being higher than with standard m-photon excitation, the resolution promised by these concepts is still modest. It was not until the mid 1990s that the first viable concepts to break the diffraction barrier appeared27,28. They all share a common principle, that is, the spatially modulated and saturable transition between two molecular states. This principle establishes a whole family of methods for achieving nanoscale resolution in all directions29–31. Although the full potential of these approaches remains to be explored, their fundamental nature, pertinence to biotechnology and potential synergy with protein engineering make their review timely. Moreover, recently reported results demonstrate that considerable
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1347
PERSPECTIVE decreasing aperture angle15,33. Therefore this concept requires focused wavefronts of highSample angle lenses15,37. Although accurate alignment of two lenses did initially pose challenges, the real physical problem in the development of this concept was avoidance of lobe-induced artifacts. DM BS To solve this problem, three lobe-suppression mechanisms were introduced: first, conr se focalization14,15; second, two-photon La CC excitation17; and third, use of excitation/fluoD rescence wavelength disparities15–17. The last of these is particularly efficient if both wave2 µm 2 µm front pairs are brought to interfere in the sample and at the detector15,16, respectively, Figure 1 Examples of 4Pi confocal microscopy. (a) MMM-4Pi microscopy for live cell imaging. The because the respective side-lobes no longer sample is placed between two opposing water-immersion lenses that are jointly used for multiphoton coincide in space. A single mechanism may be excitation with up to 64 pairs of 4Pi spots. These spots are produced by splitting an array of pulsed sufficient; so far, however, the implementalaser beamlets at the beamsplitter (BS). The 4Pi focal array is brushed across the specimen by fast tion of at least two mechanisms has proved to scanning (not shown). Fluorescence from the spots is imaged onto a charge-coupled device (CCD) be more reliable. After an initial demonstracamera, after being deflected by a dichroic mirror (DM). The system provides fourfold improved tion35, my laboratory first applied superresectioning over a comparable confocal microscope. Nonlinear image restoration results in ∼100-nm three-dimensional resolution. Recording times, currently ∼100 s per 20 × 20 × 5 µm stack, are solved axial separation with two-photon determined by sample brightness and will be decreased by emerging new CCD camera technology 4Pi-confocal microscopy to fixed cells36. The (sketch slightly simplified). (b) GFP-labeled mitochondrial compartment of live Saccharomyces images can be further augmented by applying cerevisiae. The organelle displays strong tubular ramification of a single large body that is exclusively nonlinear restoration38–40, which under biolocated beneath the plasma membrane (counterstained in blue). Inset, a mitochondrial tubule that can logical imaging conditions typically improves be followed through the thickened cell wall at the budding site. (c) Golgi apparatus, as represented by the resolution up to a factor of 2 in both GalTase-EGFP expression in a live Vero cell44. Note the convoluted structure of the Golgi apparatus, transverse and axial directions. Therefore, in featuring ribbons and fractionated stacks, as well as smaller tubular and vesicular subcompartments. Inset, an epifluorescence overview image of the same cell, which colocalizes the organelle with the combination with image restoration, twonucleus counterstained in blue. (Data in a and b are adapted from ref. 18; data in c reprinted by photon 4Pi confocal microscopy has resulted permission of J. Struct. Biol. from ref. 44.) in a resolution of ∼100 nm in all directions, as first demonstrated in images obtained by my group of filamentous actin36 and immunofluprogress is being made, such as the first demonstration of spatial reso- orescently labeled microtubules41,42 in mouse fibroblasts. lution of λ/25 with focused light and with regular lenses32. Recently, my colleagues and I have introduced a multifocal variant, In this article, I outline strategies, implementations and initial appli- termed MMM-4Pi18, enabling 100-nm three-dimensional resolution cations of superresolution microscopy, and finally discuss a potential to be translated into live cell imaging43. This method has provided road map toward imaging with nanoscale resolution in live cells. superior three-dimensional images of the reticular network of GFPBridging the gap between electron and current light microscopy, a labeled mitochondria in live budding yeast cells (Fig. 1). Cell-induced ‘nanoscope’ working with focused light should be a powerful tool for phase changes have proved more benign than anticipated18, but they unraveling the relationship between structure and function in cell are likely to confine these methods to the imaging of individual cells or biology. thin cell layers. The deep modulation of the focal spot resulting from the joint action of multiphoton excitation and interference provides a Axial resolution improvement with two lenses new tool to measure thicknesses of cellular constituents in the 50- to Conventional and confocal microscopy fail in distinguishing objects 500-nm range with a precision of a few nanometers18. This property that are more closely stacked than 500–800 nm because the focal spot has been used to detect changes of ∼20 nm in the diameter of mitoof a lens is at least three- to fourfold longer than it is wide3,5,13. An chondrial tubules on a change of growth conditions18. explanation is that the focusing angle of a lens is not a full solid angle 4Pi confocal microscopy requires the sample to be mounted of 4π. If it were, the focal spot would be spherical and the axial resolu- between two coverslips, unless one of the lenses is a dipping lens. The tion as good as its lateral counterpart. Therefore, an obvious way to recent development of sample chambers with appropriate air and CO2 decrease the axial spot size is to enlarge the focusing angle of the sys- conditions has allowed cell viability to be sustained over periods up to tem by synthesizing a larger wavefront with two opposing lenses14–16. 48 h and enabled 4Pi imaging in live mammalian cells44. By imaging Wavefront synthesis requires the addition of wave amplitude and the Golgi-resident proteins uridinyl-diphosphate-galactosyltransphase (that is, interference). A first effort to exploit interference for ferase and heparan sulfate-2-O-sulfotransferase as enhanced GFP axial resolution improvement with flat standing waves33 was limited to (EGFP) fusion proteins, this work has enabled the first three-dimen200-nm-thin samples34, and thus it was not until the advent of spot- sional representation of the Golgi apparatus of a live mammalian cell scanning 4Pi confocal14,17 and widefield I5M microscopy16 that the use at ∼100 nm resolution in all directions (Fig. 1c). The results indicate of interference led to improved axial resolution in three-dimensional that ∼100-nm three-dimensional resolution can be obtained in the imaging19,35,36. The reason is that, while interference readily gives a imaging of protein distributions in the cytosol and probably also in the focal spot of ∼λ/4n width, it also spawns off periodic side-lobes at nucleus. Extending the technique to multicolor detection will improve ∼λ/2n ≈ 200 nm distance, which increase in height and number with the microscope’s ability to axially colocalize differently tagged proteins
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
a
1348
b
c
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
PERSPECTIVE a
b
Monolayer
2 µm
Monolayer
Confocal
c
170 nm
STED-4Pi
b
d
Counts
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
a
Figure 2 Actin cytoskeleton at the rim of a HeLa cell. (a,b) Imaged by conventional epifluorescence (a) and structured illumination microscopy supported by linear deconvolution (b). (c,d) Insets show the difference in definition possible in a and b, respectively. Whereas a and c have a lateral resolution of 280–300 nm, the superresolved images in b and d have a resolution of 110–120 nm. Note the clear separation of two parallel fibers at 170 nm distance. (Data reprinted by permission of J. Microsc. from ref. 48.)
c
d
400 200 z/µm
0 0
1
2
3
z/µm 40
1
2
3
4
Microtubules
by the same factor of 3–7; this is likely to become important in protein interaction studies. The otherwise very effective lobe-reducing measures of confocalization and two-photon excitation are to some extent restrictive. Clearly, nonconfocal wide-field detection and regular illumination would make 4Pi microscopy more versatile. Therefore, the related approach of I5M (refs. 16,19,45,46) confines itself to using the simultaneous interference of both the excitation and the (Stokes-shifted) fluorescence wavefront pairs. Impressive work has demonstrated that this method yields three-dimensional images of actin filaments with slightly better than 100-nm axial resolution in fixed cells19. To remove the side-lobe artifacts, I5M-recorded data are deconvoluted offline with a linear mathematical filter. The benefits of I5M are readily stated: single photon excitation with arguably less photobleaching, an additional 20–50% gain in fluorescence signal, and lower cost. However, the relaxation of the side-lobe suppression comes at the expense of increased vulnerability to sampleinduced aberrations, especially with nonsparse objects37,47. Thus I5M imaging, which has so far relied on oil-immersion lenses, has required mounting of the cell in a medium with n = 1.5 (ref. 19). Live cells inevitably necessitate aqueous media (n = 1.34). Moreover, waterimmersion lenses have a poorer focusing angle and therefore larger lobes to begin with43. Potential strategies for improving the tolerance of I5M are the implementation of a nonlinear excitation mode and the combination with pseudo-confocal or patterned illumination48. Although these measures again add physical complexity, they may have the potential to render I5M more suitable for live cells. Up to now, however, live cell imaging has been the prerogative of two-photon 4Pi confocal microscopy. Recently, confocalization, twophoton excitation, and the use of excitation/fluorescence wavelength disparities have been synergistically implemented in a compact 4Pi unit that was interlaced with a state-of-the-art confocal scanning microscope (Leica TCS-SP2 AOBS, Mannheim, Germany). Consequently, a sevenfold-improved axial resolution (80 nm) over confocal microscopy has been achieved in live cell imaging, with a rugged system (unpublished data).
Figure 3 Optical sections from the microtubular network of a human embryonic kidney cell labeled by immunofluorescence. (a,b) Standard confocal (a) and STED-4Pi xz sections (b) from the same site. The straight vertical line serving as a resolution reference stems from a monomolecular fluorescent layer on the coverslip. The STED-4Pi image was linearly filtered to remove the effect of side-lobes. Note the fundamentally improved clarity in b. (c,d) Profiles of the image data along the marked lines, quantifying an ∼15-fold improved axial resolution of the STED-4Pi microscope over its confocal counterpart. The profiles of the microtubules (FWHM 60–70 nm) are broader than the response to the monolayer (∼50 nm). The STED-4Pi microscope can distinguish spatially dense features and reveals weak objects next to bright clusters. As the cell was mounted in an aqueous buffer and recorded with water-immersion lenses, the results indicate that the optical conditions for obtaining subdiffraction resolution can be met in live cells as well58.
Lateral resolution improvement In theory, the resolution of a confocal microscope slightly surpasses that of the standard epifluorescence microscope. Confocal fluorescence microscopes feature an effective focal spot that is narrower by 40%, and their optical transfer function (OTF) has twice the bandwidth3. This is because in a confocal microscope the focusing ability of the objective lens is used twice: first for focusing the excitation light onto a spot on the sample, and second for focusing the fluorescence onto a point-like detector3. Thus, in contrast to epifluorescence microscopy, confocal microscopy illuminates and detects selectively in space. As spatially selected detection is achieved by a pinhole, some photons are discarded, meaning that the slight improvement of resolution is gained by losing some of the light, not only from above and below the focal plane, but also from the focal plane itself. This is disadvantageous if one wishes to use the additional higher frequencies for resolution improvement through deconvolution (Box 1). However, spatially selected detection can also be performed with a camera49,50, in which case all photons are detected. Provided that they are properly reassigned, they may all contribute to the image. Therefore, in the 1980s and early 1990s, the groups of Bertero and
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1349
PERSPECTIVE
Box 1 The concept of resolution a
b
PSF
c
OTF 1
Separation of molecules
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
STED 40 nm
62 nm STED deconv.
Convent. 222 nm
STED
0.1
33 nm
Mol. no.1
Convent.
Mol. no.2
0.01
–300 –200 –100
0
100
200
300
10
0
20
∆x –1 (µm –1)
x (nm)
30
40
–100
0
100
x (nm)
Figure 4 Quantifying resolution. (a) The intensity profile of the effective point-spread-function (PSF) quantifies the focal blur in the microscope. Identical fluorescent objects that are closer than the FWHM of the PSF cannot be distinguished. (b) The optical transfer function (OTF) is an equivalent representation of the resolution, giving the bandwidth of the spatial frequencies passed to the image; the broader the OTF, the better the resolution. The data plotted in (a,b) are gained by probing the fluorescent spot of a scanning microscope with a single molecule of the fluorophore JA 26, both in the conventional (Abbe type) mode and with STED. Employed conditions: n = 1.5, α =67°, wavelengths λ: 635 nm (excitation), 650 –720 nm (fluorescence collection), and 790 nm (STED). Note the 5.5-fold sharper PSF (a) and the equally broader OTF (b) of STED compared to the diffraction-limited conventional microscope. (b) Linear deconvolution is equivalent to multiplying the higher frequencies of the OTF that are not masked by noise (see arrows). (c) Subdiffraction resolution with STED microscopy. Two identical molecules located in the focal plane that are only 62 nm apart can be entirely separated by their intensity profile in the image. A similarly clear separation by conventional microscopy would require the molecules to be at least 300 nm apart. Data adapted from ref. 32.
with the approximately fivefold-enlarged OTF of the STED fluorescence microscope. The marked bandwidth enlargement over that of the conventional microscope signifies a fundamental breaking of Abbe’s diffraction barrier in the focal plane in the case of STED. The FWHM of the PSF and the bandwidth of the OTF of the microscope are just estimates; a thorough description of the resolution requires the complete functions. Moreover, knowing these functions in full also makes it possible to improve the resolution by deconvolution. Note that the OTF falls off with larger spatial frequencies (Fig. 4b). Provided that in the image these frequencies are not swamped by noise, they can be artificially elevated by multiplication (see arrows). Mathematically, this amounts to a (de)convolution in real space. As the higher frequencies are responsible for small details in the image, deconvolution results in a further image sharpening. As an example of linearly deconvoluted STED microscopy, two molecules at 62 nm distance are distinguished in full by two sharp peaks (Fig. 4c)32. The individual peaks are sharper (33 nm) than the initial peak of 40 nm, as a result of deconvolution. The effective OTF after deconvolution is slightly augmented at lower frequencies, as indicated by the arrows in Figure 4b.
Light microscopy resolution can be described either in real space or in spatial frequencies. In real space, the resolution is assessed by the full-width half-maximum (FWHM) of the focal spot, referred to as the point-spread function (PSF). Loosely speaking, if identical molecules are within the FWHM distance, the molecules cannot be separated in the image. Therefore, improving the resolution is equivalent to reducing the FWHM of the PSF. In a conventional microscope, the FWHM of the PSF is about λ/(2n sin α), with λ denoting the wavelength, n the refractive index and α the semiaperture angle of the lens. Figure 4a shows the measured profile of the PSF in the focal plane (x) for a conventional fluorescence microscope along with its sharper subdiffraction STED fluorescence counterpart. Note the 5.5-fold improvement of resolution with STED. In the frequency world, the sample is described as being composed of spatial frequencies. Therefore, the microscope’s resolution is given by the OTF describing the strength with which these frequencies are transferred to the image3,68. Thus, the resolution limit is given by the highest frequency passed. PSF and OTF are intertwined by Fourier mathematics: the sharper the PSF, the broader the OTF. Figure 4b shows the OTF of a conventional microscope along
Pike proposed concepts using (camera-based) spatially weighted detection in conjunction with scanning point-like illumination49,51,52. However, the combination of computation with scanned point-like illumination rendered these systems not very effective. Therefore, it was not until Gustafsson recently implemented highfrequency line-patterned illumination that this approach has been brought to fruition48. Wide-field camera detection allows fast data acquisition, except that the pattern needs to be scanned and rotated.
1350
Because it targets the high object frequencies with the line pattern and features a somewhat improved signal, this approach has the prerequisites to yield the lateral resolution promised by ideal confocal microscopy (∼100 nm). Sequential pattern alteration combined with data processing may render it more prone to movement artifacts. Thus far, superior images of the actin cytoskeleton (Fig. 2) have been achieved in fixed cells48, but the coming years will show whether this method will be applicable to live cells as well.
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE Breaking the diffraction barrier Confocal and related imaging modalities may, in the ideal case, surpass the diffraction barrier by a factor of two, but they do not break it. Breaking implies the potential of featuring an infinitely sharp focal spot, or an infinitely large OTF bandwidth (Box 1). In 1994, together with Jan Wichmann, I published a theoretical paper that detailed a concept to eliminate the resolution-limiting effect of diffraction without eliminating diffraction itself27. It was termed stimulated emission depletion (STED) microscopy because the depletion of the molecular fluorescent state through stimulated emission was exploited. Shortly after, together with Mathias Kroug, I proposed ground-state depletion microscopy as a further concept with molecular resolution potential28. Both apply the same principle to break Abbe’s barrier: a focal intensity distribution with a zero-point in space effects a saturated depletion of a molecular state that is essential to the fluorescence process28,29. As any saturable process is a potential candidate29,31,53, the choice of process is solely determined by practical
conditions, such as the required intensities, available light sources, photobleaching and, with respect to applications in cell biology, compatibility with live cell imaging. Box 2 discusses the principles of this radically different approach to overcoming the diffraction barrier. STED microscopy is a special case of this approach (Box 3). The fluorophore in the fluorescent state S1 (state A) is stimulated to the ground state S0 (state B) with a doughnut-shaped beam. Saturated depletion of S1 confines fluorescence to the central naught. With typical Isat ranging from 1 to 100 MW/cm2, saturation factors up to ς ≈ 120 have been reported54,55. Doughnut imperfections have so far confined the up to tenfold possible improvement to a five- to sevenfold observed improvement over the diffraction barrier55 (Box 1). Using STED wavelengths of λ = 750–800 nm, a lateral resolution of up to 28 nm has been reached in experiments with single molecules32. My laboratory has obtained subdiffraction images with threefold axial and doubled lateral resolution with membrane-labeled bacteria and live budding yeast cells54. Although there is preliminary evidence
Box 2 The principles of breaking the diffraction barrier The basic idea underlying stimulated emission and ground state depletion microscopy can be generalized as follows. Let us assume two arbitrary fluorophore states A and B between which the molecule can be transferred; typical examples are the ground and an excited state or conformational and isomeric states. Transition A→B is induced by light, but no restriction is made on transition B→A. It may be spontaneous, but also be effected by light, heat, and so on. The only further assumption is that at least one of the states is critical to molecular fluorescence. By denoting the rates of A→B and B→A by kAB and kBA, respectively, changes of the normalized populations NA and NB are subject to the relationship dNA /dt = –kABNA + kBANB = – dNB /dt. If the molecule first resides in A, or any other state, after t ≈ 5(kAB + kBA)–1, the equilibrium population of state A is NA∞ = kBA/(kAB + kBA). We are now interested in depleting state A by light via the transition A→B, whose rate is given by kAB = σI, with σ and I denoting the molecular cross section and the photon flux per unit area, respectively. Hence, the equilibrium population is NA∞ = kBA/(σI + kBA). If I >> I sat = kBA/σ, it follows that NA∞→0, that is, all the molecules end up in B. I sat is referred to as the saturation intensity. (We note that if A decays with kAB to B, I sat = kAB/σ.) If we now elect a spatial intensity distribution I(→ r ) >> I sat with a → ro. Thus naught at r o, all molecules end up in B, except for those at → we can create arbitrarily sharp regions of state A (Fig. 5). Written more formally, I(→ r ) = I maxf (→ r ), where f (→ r ) is a diffraction-limited → spatial function featuring f ( r o) = 0. For I max → ∞, the region in which the molecule can be found in A is squeezed to a point r ). regardless of the details of f (→ r ) cannot be neglected. If I max and I sat are finite, the details of f (→ For example, the minima of a standing wave f(x) = sin2(2πnx/λ) create regions of A with an FWHM of λ arcsin ∆x = — πn
——– kBA λ— ——– ≈ —— – σImax πn ς
ς = I max/I sat denotes the saturation factor. ς =1,000 yields ∆x ≈ λ/(100n), but in principle the spot of ‘A molecules’ can be continuously squeezed by increasing ς.
(1)
Fluorescent Absorbing A M
I(r)
1.0
NA∞
σI
Translation
Imax
FWHM << λ
0.5
Non-fluorescent Non-absorbing M
B
kBA
B A
C(r) 0 0
0.5
r0
1.0
r /λ Figure 5 Diffraction-unlimited spatial resolution with a reversible, saturable optical transition: the principle. A standing wave of intensity I(r) and wavelength λ is used for photoswitching molecules from state A into a state B. If only a small fraction of the total intensity I max is sufficient for transferring the molecule to B, the probability NA∞(r) of finding it in A is confined to the nodal points (blue). If state A, but not B, is involved in fluorescence the signal originates from the narrow region defined by NA∞(r) only. This simple concept enables fluorescence imaging with diffractionunlimited resolution. For this purpose, one or several nodes are scanned across the sample C (r). Except for the nodes, the molecules are transiently transferred to B, so that the fluorescence from the molecules in A maps out the object C (r). The FWHM and thus the resolution are determined solely by the ‘saturation factor’, that is, the factor by which I max surpasses the required intensity threshold at which, say, 50% or more of the molecules are already in the nonfluorescent state B. The idea is readily extended to all directions in space and thus to three-dimensional imaging. Conventional camera-based detection is possible if the nodes are farther apart than the classical resolution limit of the microscope. Complete depletion of A is not required. It is sufficient that the non-nodal region features a large enough population B, so that it can be distinguished from its sharp counterpart. If not A but B is the fluorescent state, one reads out B and may obtain the same super-resolved image after subtraction.
The sharp regions of A can be used to map out the fluorophores with arbitrary resolution, as explained in Figure 5.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1351
PERSPECTIVE
Box 3 Stimulated emission depletion microscopy c
Fluor.
STED
Exc.
S1
B
B
T1
λ/2phaseplate
STED
S0 A
∆t
I
0.5
SAT
= 80 MW/cm
x y
2
97 nm
1.0
490 nm
Excitation
1.0
Confocal
STED
b Fluorescence (a.u.)
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
A
d Detector
a
z
0.5 0.0
0
1
Z
2
X Saturated depletion
104 nm
0.0 0
3
6
ISTED (GW/cm2)
244 nm
9
Excitation STED Saturated spot + spot + depletion
Figure 6 Physical conditions, setup and typical focal spot for STED. (a) Energy diagram of an organic fluorophore. Molecules in the excited state S1 return to the ground state S0 by spontaneous fluorescence emission. Return to S0 may also be enforced by light through stimulated emission67, a phenomenon with the same cross section and intensity dependence as normal absorption. To prevail over the spontaneous return, STED requires intense light pulses with duration of a fraction of the S1 lifetime. Tuning the STED wavelength to the red edge of the emission spectrum prevents re-excitation by the same pulses. T1 is a dark triplet state that can be accessed through S1 and then returns to S0 within 1–104 µs. (b) Saturated depletion of the S1 with increasing STED pulse intensity ISTED, as measured by the remaining fluorescence of an organic fluorophore. Depletion of the S1 saturates with increasing ISTED and therefore establishes a nonlinear relationship between the fluorescence and the intensity applied for STED. The saturation is the essential element for the breaking of the diffraction barrier, as explained in Box 2; the inset highlights the saturation intensity Isat. (c) Sketch of a point-scanning STED microscope. Excitation and STED are accomplished with synchronized laser pulses focused by a lens into the sample, sketched as green and red beams, respectively. Fluorescence is registered by a detector. Below, note the panels outlining the corresponding spots at the focal plane: the excitation spot (left) is overlapped with the STED spot featuring a central naught (center). Saturated depletion by the STED beam reduces the region of excited molecules (right) to the very zero point, leaving a fluorescent spot of subdiffraction dimensions shown in panel d. (d) Fluorescent spot in the STED and in the confocal microscope. Note the doubled lateral and fivefold-improved axial resolution. The reduction in dimensions (x,y,z) yields an ultrasmall volume of subdiffraction size, here 0.67 attoliter54, corresponding to 6% of its confocal counterpart. The spot size is not limited on principle but by practical circumstances such as the quality of the naught and the saturation factor of depletion.
STED microscopy produces subdiffraction resolution and subdiffraction-sized fluorescence volumes through the saturated depletion of the fluorescent state of the dye. The nonlinear intensity dependence brought about by saturation is radically different from the nonlinearity connected with, for example, m-photon excitation, m th harmonics generation and coherent antiStokes-Raman scattering2,66. In the last two cases, the nonlinear signal stems from the simultaneous action of more than one photon at the sample, which would only work at high focal intensities. In contrast, the nonlinearity brought about by saturation and depletion stems from a change in the population of the involved states, which is effected by a single-photon process, namely stimulated emission. Therefore, unlike in m-photon processes, strong nonlinearities are achieved at comparatively low intensities. Ultrasmall volumes of detection are critical to several sensitive
1352
bioanalytical techniques. For example, fluorescence correlation spectroscopy69 relies on small focal volumes to detect rare molecular species or interactions in concentrated solutions70,71. Although volume reduction can be obtained by nanofabricated structures72, STED may prove instrumental to attaining spherical volumes at the nanoscale. Published results imply the possibility of a further decrease of the volume by another order of magnitude53,55. Initial applications may be hampered by the requirement of an additional pulsed laser that is tuned to the red edge of the emission spectrum of the dye. Nevertheless, STED is so far the only known method to squeeze a fluorescence volume to the zeptoliter scale without mechanical contact. Ultrasmall volumes with dimensions tens of nanometers in diameter created by STED may provide a pathway to improving the sensitivity of fluorescence-based bioanalytical techniques73,74.
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE for increased nonlinear photobleaching of certain markers with the elevated intensities56, there is no indication that the intensities applied currently would exclude the imaging of live cells. This is not surprising as the intensities are lower by two to three orders of magnitude than those used in multiphoton microscopy4. Moreover, STED has proved to be sensitive to single molecules, despite the proximity of the STED wavelength to the emission peak. In fact, my group has been able to switch individual molecules on and off by STED on command57. The power of STED and 4Pi microscopy has been synergistically combined29 to demonstrate the first axial resolution of 30–40 nm in focusing light microscopy53. Initial studies in my laboratory have enabled xz images of membrane-labeled bacteria to be obtained53. More recent studies are extending STED-4Pi microscopy to immunofluorescence imaging58. In this work, we have demonstrated a spatial resolution of ∼50 nm in the imaging of the microtubular meshwork of a mammalian cell (Fig. 3). These results indicate that the basic physical hurdles have been overcome in attaining a three-dimensional resolution to the order of a few tens of nanometers. Because the samples were mounted in an aqueous buffer53,58, the results indicate that the optical conditions for obtaining subdiffraction resolution are met under the physical conditions encountered in live cell imaging. STED microscopy is still at an early stage of development. Our work57 has demonstrated the suitability of laser diodes for both excitation and depletion, but further efforts are required to implement STED into fast-scanning systems. The lack of compact tunable pulsed light sources in the visible range has so far confined STED investigations to red-emitting dyes. As more efficient light sources become available, however, both visible blue, green and yellow fluorophores as well as fluorescent proteins will be interesting candidates for saturated depletion59,60. Shorter wavelengths will also lead to higher spatial resolution. However, a further increase of the intensity might be barred in aqueous media by intolerable photobleaching. Although STED pulses >300 ps recently improved dye photostability56, saturation factors (ς) of >200 might not be readily attainable. Fortunately, this limitation can be counteracted by lowering I sat through kBA (Box 2). Thus, it has been proposed to deplete the ground state (now state A) by targeting an excited state (B) with a comparatively long lifetime28,29, such as the metastable triplet state T1 (Fig. 6a). In many fluorophores, T1 can be reached through the S1 with a quantum efficiency of 1–10%61. A forbidden transition, the relaxation of sat the T1 is 103- to 105-fold slower than that of the S1, thus giving IAB = 2 0.1–100 kW/cm . The signal to be measured (from the naught) is the fluorescence of the molecules that remained in the singlet system, through a synchronized further excitation28. The disadvantage here is the involvement of the T1 in photobleaching. Potential alternatives are metastable states of rare earth metal ions that are fed through chelates. Another option is to deplete the S0 by saturating the S1 (now B), as has been proposed recently30. This is perhaps the simplest realization of saturated depletion because it requires just excitation wavelength matching (Fig. 6). However, as the fluorescence emission maps the spatially extended ‘majority population’ in state B, the superresolved images (represented by state A) are hidden under a bright signal from B (Box 2). Thus, acquiring these images requires computational extraction, which makes this approach prone to noise, unless the sample is very sparse. Nevertheless, the simplicity of raw data acquisition may render it attractive for the imaging of fixed cells. I sat is of the same order as with STED, because the saturation of fluorescence also competes against the spontaneous decay of S1. Therefore attaining ς > 200 might involve similar photostability issues. Importantly, the quest for large ς at low I sat should be solved by compounds with two (semi-) stable states31,58. If the rate kBA (and the
spontaneous rate kAB) almost vanishes, large ς values are attained at low intensities. The lowest useful intensity is set by the concomitant increase in switching time. In the ideal case, the marker is a bistable fluorescent compound that can be photoswitched, at separate wavelengths, from a fluorescent state to a dark state, and vice versa. A photoswitchable coupled molecular system, based on a photochromic diarylethene derivative and a fluorophore, has been reported62. Using equation (1), one can determine that focusing less than 100 µW of deep-blue ‘switch-off light’ to an area of 10–8 cm2 for 50 µs should yield better than 5-nm spatial resolution. Targeted optimization of photochromic or other compounds toward fatigue-free switching and visible light operation could therefore open up radically new avenues in microscopy and data storage31. For live cell imaging, fluorescent proteins are more advantageous. Any fluorescent protein that can be pushed to a dark state29,31 (and vice versa), and has a lifetime longer than 10 ns, may result in larger ς; however, fluorescent proteins that can be switched ‘on’ and ‘off ’ at different wavelengths are a more attractive option31. An example is Anemonia sulcata purple protein (asFP595), which, according to the published data63, may allow saturated depletion of the fluorescence state with intensities of less than a few watts per square centimeter. Under favorable switching conditions, such or similar fluorescent proteins should allow a spatial resolution of better than 10 nm31 at very low intensities. The low power involved should also enable parallelization of saturation through an array of minima or dark lines. Initial realization of very low intensity depletion microscopes may, however, be challenged by switching fatigue62 and overlapping action spectra63. However, the prospect of attaining nanoscale resolution with regular lenses and focused light is an incentive to surmount these challenges by strategic fluorophore modification31. Conclusions Although most textbooks still portray light microscopy as limited by resolution, in recent years concepts have emerged that are poised to radically change this view. Featuring the aperture angle of two opposing lenses, 4Pi and its widefield cousin I5M microscopy have displayed 80–100 nm resolution along the optic axis. In particular, compact 4Pi confocal microscopes are emerging that feature the same scanning and detection amenities as commercial confocal systems, but with a sevenfold improved optical sectioning in live cells. Moreover, we have begun to map out concrete physical concepts for overcoming the resolution limit altogether. By exploiting these concepts, my group has been able to break the diffraction barrier in several imaging experiments64, including experiments with single fluorescent molecules, simple fixed specimens and live biological specimens. STED microscopy has so far witnessed a resolution improvement by up to a factor of 6, resulting in the smallest fluorescent volumes that have been created with focused light so far. Combined with 4Pimicroscopy, STED has provided the first demonstration of immunofluorescence imaging with an (axial) resolution of 50 nm, and this value is more of a starting point than a limit. Future research on its spectroscopy conditions and on practical aspects65 will reveal its full potential. Relying on saturated optical transitions, the spatial resolution of the new concepts is determined by the attainable saturation level. The nonlinear intensity dependence brought about by saturation is fundamentally different from the nonlinearity connected with m-photon excitation, or with mth harmonics generation, coherent anti-StokesRaman scattering2,66, etc. In the latter cases, the nonlinear signal stems from the action of more than one photon at the sample at the same time, which demands high focal intensities. In contrast, the nonlinear-
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1353
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE ity brought about by saturation and depletion stems from a change in the population of the involved states, which opens the door to lowintensity implementations. I expect this detail to be essential to opening up the cellular nanoscale with visible light and regular lenses. Because of their population kinetics, switchable dyes and fluorescent proteins should allow high levels of saturation at low light intensities31. Although first candidates have been named, dedicated synthesis or protein engineering might uncover a whole new range of suitable markers. Physics undoubtedly has greatly contributed to the emergence of molecular and cell biology in the past century. Paradoxically, the development of nanoscale imaging with focused light in molecular and cell biology might now topple a longstanding paradigm of physics. ACKNOWLEDGMENTS I thank M. Dyba, A. Egner, S. Jakobs, J. Jethwa, L. Kastrup, J. Keller and A. Schönle for constructive reading. The work of this laboratory has been funded by the Max Planck Society, with further support by the German Ministry of Research and Education, the Deutsche Forschungsgemeinschaft, and the Volkswagen Foundation. COMPETING INTERESTS STATEMENT The author declares that he has no competing financial interests. Published online at http://www.nature.com/naturebiotechnology/ 1. Abbe, E. Beiträge zur Theorie des Mikroskops und der mikroskopischen Wahrnehmung. Arch. Mikroskop. Anat. 9, 413–420 (1873). 2. Sheppard, C.J.R. & Kompfner, R. Resonant scanning optical microscope. Appl. Opt. 17, 2879–2882 (1978). 3. Wilson, T. & Sheppard, C.J.R. Theory and Practice of Scanning Optical Microscopy (Academic Press, New York, 1984). 4. Denk, W., Strickler, J.H. & Webb, W.W. Two-photon laser scanning fluorescence microscopy. Science 248, 73–76 (1990). 5. Pawley, J. Handbook of Biological Confocal Microscopy (Plenum, New York, 1995). 6. Basché, T., Moerner, W.E., Orrit, M. & Wild, U.P. Single-Molecule Optical Detection, Imaging and Spectroscopy (VCH, Weinheim, New York, Basel, Tokyo, 1997). 7. Weiss, S. Fluorescence spectroscopy of single biomolecules. Science 283, 1676–1683 (1999). 8. Ha, T., Enderle, T., Chemla, D.S. & Weiss, S. Dual-molecule spectroscopy: molecular rulers for the study of biological macromolecules. IEEE J. Select. Top. Quantum Electron. 2, 1115–1128 (1996). 9. Bornfleth, H., Sätzler, K., Eils, R. & Cremer, C. High-precision distance measurements and volume-conserving segmentation of objects near and below the resolution limit in three-dimensional confocal fluorescence microscopy. J. Microsc. 189, 118–136 (1998). 10. Oijen, M.v., Köhler, J., Schmidt, J., Müller, M. & Brakenhoff, G.J. 3-Dimensional superresolution by spectrally selective imaging. Chem. Phys. Lett. 292, 183–187 (1998). 11. Lacoste, T.D. et al. Ultrahigh-resolution multicolor colocalization of single fluorescent probes. Proc. Natl. Acad. Sci. USA 97, 9461–9466 (2000). 12. Hettich, C. et al. Nanometer resolution and coherent optical dipole coupling of two individual molecules. Science 298, 385–389 (2002). 13. Born, M. & Wolf, E. Principles of Optics 6th edn. (Pergamon, Oxford, 1993). 14. Hell, S.W. Double-confocal microscope. European Patent 0491289 (1990). 15. Hell, S. & Stelzer, E.H.K. Properties of a 4Pi-confocal fluorescence microscope. J. Opt. Soc. Am. A 9, 2159–2166 (1992). 16. Gustafsson, M.G.L., Agard, D.A. & Sedat, J.W. Sevenfold improvement of axial resolution in 3D widefield microscopy using two objective lenses. Proc. Soc. PhotoOptical Instrumentation Engineers 2412, 147–156 (1995). 17. Hell, S.W. & Stelzer, E.H.K. Fundamental improvement of resolution with a 4Pi-confocal fluorescence microscope using two-photon excitation. Opt. Commun. 93, 277–282 (1992). 18. Egner, A., Jakobs, S. & Hell, S.W. Fast 100-nm resolution 3D-microscope reveals structural plasticity of mitochondria in live yeast. Proc. Natl. Acad. Sci. USA 99, 3370–3375 (2002). 19. Gustafsson, M.G.L., Agard, D.A. & Sedat, J.W. I5M: 3D widefield light microscopy with better than 100 nm axial resolution. J. Microsc. 195, 10–16 (1999). 20. Pohl, D.W. & Courjon, D. Near Field Optics (Kluwer, Dordrecht, 1993). 21. Toraldo di Francia, G. Supergain antennas and optical resolving power. Nuovo Cimento Suppl. 9, 426–435 (1952). 22. Lukosz, W. Optical systems with resolving powers exceeding the classical limit. J. Opt. Soc. Am. 56, 1463–1472 (1966). 23. Xu, C., Zipfel, W., Shear, J.B., Williams, R.M. & Webb, W.W. Multiphoton fluorescence excitation: new spectral windows for biological nonlinear microscopy. Proc. Natl. Acad. Sci. USA 93, 10763–10768 (1996). 24. Hänninen, P.E., Lehtelä, L. & Hell, S.W. Two- and multiphoton excitation of conjugate dyes with continuous wave lasers. Opt. Commun. 130, 29–33 (1996). 25. Schönle, A., Hänninen, P.E. & Hell, S.W. Nonlinear fluorescence through intermolec-
1354
ular energy transfer and resolution increase in fluorescence microscopy. Ann. Phys. (Leipzig) 8, 115–133 (1999). 26. Schönle, A. & Hell, S.W. Far-field fluorescence microscopy with repetitive excitation. Eur. Phys. J. D 6, 283–290 (1999). 27. Hell, S.W. & Wichmann, J. Breaking the diffraction resolution limit by stimulated emission: stimulated emission depletion microscopy. Opt. Lett. 19, 780–782 (1994). 28. Hell, S.W. & Kroug, M. Ground-state depletion fluorescence microscopy, a concept for breaking the diffraction resolution limit. Appl. Phys. B 60, 495–497 (1995). 29. Hell, S.W. in Increasing the resolution of far-field fluorescence light microscopy by point-spread-function engineering in Topics in Fluorescence Spectroscopy Vol. 5. (ed. Lakowicz, J.R.) 361–422 (Plenum, New York, 1997). 30. Heintzmann, R., Jovin, T.M. & Cremer, C. Saturated patterned excitation microscopy—a concept for optical resolution improvement. J. Opt. Soc. Am. A 19, 1599–1609 (2002). 31. Hell, S.W., Jakobs, S. & Kastrup, L. Imaging and writing at the nanoscale with focused visible light through saturable optical transitions. Appl. Phys. A 77, 859–860 (2003). 32. Westphal, V., Kastrup, L. & Hell, S.W. Lateral resolution of 28nm (λ/25) in far-field fluorescence microscopy. Appl. Phys. B 77, 377–380 (2003). 33. Lanni, F. Applications of Fluorescence in the Biomedical Sciences 1st edn. (Liss, New York, 1986). 34. Bailey, B., Farkas, D.L., Taylor, D.L. & Lanni, F. Enhancement of axial resolution in fluorescence microscopy by standing-wave excitation. Nature 366, 44–48 (1993). 35. Schrader, M. & Hell, S.W. 4Pi-confocal images with axial superresolution. J. Microsc. 183, 189–193 (1996). 36. Hell, S.W., Schrader, M. & van der Voort, H.T.M. Far-field fluorescence microscopy with three-dimensional resolution in the 100 nm range. J. Microsc. 185, 1–5 (1997). 37. Nagorni, M. & Hell, S.W. Coherent use of opposing lenses for axial resolution increase in fluorescence microscopy. I. Comparative study of concepts. J. Opt. Soc. Am. A 18, 36–48 (2001). 38. Holmes, T.J. Maximum-likelihood image restoration adapted for non-coherent optical imaging. J. Opt. Soc. Am. A 5, 666–673 (1988). 39. Carrington, W.A. et al. Superresolution in three-dimensional images of fluorescence in cells with minimal light exposure. Science 268, 1483–1487 (1995). 40. Holmes, T.J. et al. Light microscopic images reconstructed by maximum likelihood deconvolution in Handbook of Biological Confocal Microscopy (ed. Pawley, J.) 389–400 (Plenum, New York, 1995). 41. Nagorni, M. & Hell, S.W. 4Pi-confocal microscopy provides three-dimensional images of the microtubule network with 100- to 150-nm resolution. J. Struct. Biol. 123, 236–247 (1998). 42. Hell, S.W. & Nagorni, M. 4Pi confocal microscopy with alternate interference. Optics Lett. 23, 1567–1569 (1998). 43. Bahlmann, K., Jakobs, S. & Hell, S.W. 4Pi-confocal microscopy of live cells. Ultramicroscopy 87, 155–164 (2001). 44. Egner, A., Goroshkov, A., Verrier, S., Söling, H.-D. & Hell, S.W. Golgi apparatus of live mammalian cell at 100 nm resolution. J. Struct. Biol. in the press (2003). 45. Gustafsson, M.G., Agard, D.A. & Sedat, J.W. 3D widefield microscopy with two objective lenses: experimental verification of improved axial resolution. in ThreeDimensional Microscopy: Image Acquisition and Processing III (eds. Cogswell, C., Kino, G.S. & Wilson, T.) 62–66 (SPIE, New York, 1996). 46. Gustafsson, M.G.L. Extended resolution fluorescence microscopy. Curr. Opin. Struct. Biol. 9, 627–634 (1999). 47. Nagorni, M. & Hell, S.W. Coherent use of opposing lenses for axial resolution increase in fluorescence microscopy. II. Power and limitation of nonlinear image restoration. J. Opt. Soc. Am. A 18, 49–54 (2001). 48. Gustafsson, M.G.L. Surpassing the lateral resolution limit by a factor of two using structured illumination microscopy. J. Microsc. 198, 82–87 (2000). 49. Bertero, M., De Mol, C., Pike, E.R. & Walker, J.G. Resolution in diffraction-limited imaging, a singular value analysis. IV. The case of uncertain localization or non-uniform illumination of the object. Opt. Acta 31, 923–946 (1984). 50. Barth, M. & Stelzer, E. Boosting the optical transfer function with a spatially resolving detector in a high numerical aperture confocal reflection microscope. Optik 96, 53–58 (1994). 51. Walker, J.G. et al. Superresolving scanning optical microscopy using holographic optical processing. J. Opt. Soc. Am. A 10, 59–64 (1993). 52. Young, M.R., Davies, R.E., Pike, E.R., Walker, J.G. & Bertero, M. Superresolution in confocal scanning microscopy: experimental confirmation in the 1D coherent case. Europhys. Lett. 9, 773–778 (1989). 53. Dyba, M. & Hell, S.W. Focal spots of size λ/23 open up far-field fluorescence microscopy at 33 nm axial resolution. Phys. Rev. Lett. 88, 163901 (2002). 54. Klar, T.A., Jakobs, S., Dyba, M., Egner, A. & Hell, S.W. Fluorescence microscopy with diffraction resolution limit broken by stimulated emission. Proc. Natl. Acad. Sci. USA 97, 8206–8210 (2000). 55. Klar, T.A., Engel, E. & Hell, S.W. Breaking Abbe’s diffraction resolution limit in fluorescence microscopy with stimulated emission depletion beams of various shapes. Phys. Rev. E 64, 066613, 066611–066619 (2001). 56. Dyba, M. & Hell, S.W. Photostability of a fluorescent marker under pulsed excited–state depletion through stimulated emission. Appl. Opt. 42, 5123–5129 (2003). 57. Westphal, V., Blanca, C.M., Dyba, M., Kastrup, L. & Hell, S.W. Laser-diode–stimulated emission depletion microscopy. Appl. Phys. Lett. 82, 3125–3127 (2003). 58. Dyba, M., Jakobs, S. & Hell, S.W. Immunofluorescence stimulated emission depletion microscopy. Nat. Biotechnol. 21, 1303–1304 (2003).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE 59. Gryczynski, I., Bogdanov, V. & Lakowicz, J.R. Light quenching and depolarization of fluorescence observed with laser pulses. A new experimental opportunity in timeresolved fluorescence spectroscopy. Biophys. Chem. 49, 223–232 (1994). 60. Lakowicz, J.R. & Gryczynski, I. in Topics in Fluorescence Spectroscopy Vol. 5 (ed. Lakowicz, J.R.) 305–355 (Plenum, New York, 1997). 61. Lakowicz, J.R. Principles of Fluorescence Spectroscopy (Plenum, New York, 1983). 62. Irie, M., Fukaminato, T., Sasaki, T., Tamai, N. & Kawai, T. A digital fluorescent molecular photoswitch. Nature 420, 759–760 (2002). 63. Lukyanov, K.A. et al. Natural animal coloration can be determined by a nonfluorescent green fluorescent protein homolog. J. Biol. Chem. 275, 25879–25882 (2000). 64. Hänninen, P. Beyond the diffraction limit. Nature 419, 802 (2002). 65. Stephens, D.J. & Allen, V.J. Light microscopy techniques for live cell imaging. Science 300, 82–91 (2003). 66. Shen, Y.R. The Principles of Nonlinear Optics Edn. 1 (Wiley, New York, 1984).
67. Einstein, A. Zur Quantentheorie der Strahlung. Physik. Zeitschr. 18, 121–128 (1917). 68. Goodman, J.W. Introduction to Fourier Optics (McGraw-Hill, New York, 1968). 69. Magde, D., Elson, E.L. & Webb, W.W. Thermodynamic fluctuations in a reacting system—measurement by fluorescence correlation spectroscopy. Phys. Rev. Lett. 29, 705–708 (1972). 70. Eigen, M. & Rigler, R. Sorting single molecules: applications to diagnostics and evolutionary biotechnology. Proc. Natl. Acad. Sci. USA 91, 5740–5747 (1994). 71. Elson, E.L. & Rigler, R. (eds.) Fluorescence Correlation Spectroscopy. Theory and Applications (Springer, Berlin, 2001). 72. Levene, M.J. et al. Zero-mode waveguides for single-molecule analysis at high concentrations. Science 299, 682–686 (2003). 73. Weiss, S. Shattering the diffraction limit of light: a revolution in fluorescence microscopy? Proc. Nat. Acad. Sc. USA 97, 8747–8749 (2000). 74. Laurence, T.A. & Weiss, S. How to detect weak pairs. Science 299, 667–668 (2003).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1355
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE
FOCUS ON OPTICAL IMAGING
Second-harmonic imaging microscopy for visualizing biomolecular arrays in cells, tissues and organisms Paul J Campagnola & Leslie M Loew Although the nonlinear optical effect known as secondharmonic generation (SHG) has been recognized since the earliest days of laser physics and was demonstrated through a microscope over 25 years ago, only in the past few years has it begun to emerge as a viable microscope imaging contrast mechanism for visualization of cell and tissue structure and function. Only small modifications are required to equip a standard laser-scanning two-photon microscope for secondharmonic imaging microscopy (SHIM). Recent studies of the three-dimensional in vivo structures of well-ordered protein assemblies, such as collagen, microtubules and muscle myosin, are beginning to establish SHIM as a nondestructive imaging modality that holds promise for both basic research and clinical pathology. Thus far the best signals have been obtained in a transmitted light geometry that precludes in vivo measurements on large living animals. This drawback may be addressed through improvements in the collection of SHG signals via an epi-illumination microscope configuration. In addition, SHG signals from certain membrane-bound dyes have been shown to be highly sensitive to membrane potential. Although this indicates that SHIM may become a valuable tool for probing cell physiology, the small signal size would limit the number of photons that could be collected during the course of a fast action potential. Better dyes and optimized microscope optics could ultimately lead to the imaging of neuronal electrical activity with SHIM. SHIM is based on the familiar nonlinear optical effect called SHG, also commonly called frequency doubling. This phenomenon requires intense laser light passing through a highly polarizable material with a noncentrosymmetric molecular organization—most typically an inorganic crystal. The second-harmonic light emerging from the material is at precisely half the wavelength of the light entering the material. Thus, the SHG process within the nonlinear optical material changes two near-infrared incident photons into one emerging visible photon at exactly twice the energy (and half the wavelength). As opposed to two-photon-excited fluorescence (TPEF), in which some of the incoming energy is lost during relaxation of the excited state, SHG does not involve an excited state, is energy conserving and
Center for Biomedical Imaging Technology, Department of Physiology, University of Connecticut Health Center, Farmington, Connecticut 06030, USA. Correspondence should be addressed to L.M.L. (
[email protected]). Published online 31 October 2003; doi:10.1038/nbt894
1356
preserves the coherence of the laser light. Frequency-doubling crystals are commonly used to produce visible laser light from near-infrared lasers or UV light from visible lasers. Box 1 provides a simplified summary of key equations underlying the theory of SHG. Biological materials can be highly polarizable and often assemble into large, ordered noncentrosymmetric structures. Indeed, it has been known for 20 years that collagen can produce SHG signals1 and for 15 years that biological membranes might be good general scaffolds for noncentrosymmetric arrays of SHG-active molecules2,3. However, it is only in the past few years that it has been shown that high-resolution SHG images can be obtained with instrumentation similar to that used for TPEF microscopy4–11. Like that of two-photon absorption, the amplitude of SHG is proportional to the square of the incident light intensity. Therefore, SHIM has the same intrinsic optical sectioning characteristic as TPEF microscopy. Thus, a new threedimensional microscope contrast mechanism that does not require excitation of fluorescent molecules has been made available to the biological community. The properties of SHG offer several advantages for live cell or tissue imaging. Because SHG does not involve excitation of molecules, it should not suffer, in principle, from phototoxicity effects or photobleaching, both of which limit the usefulness of fluorescence microscopy, including two-photon fluorescence microscopy, for the imaging of living specimens. (There can be collateral damage, however, if the incident laser light also produces two-photon excitation of chromophores in the specimen.) Another advantage is that many intrinsic structures produce strong SHG, so labeling with exogenous molecular probes is not required. On the other hand, electrochromic membrane dyes can be used to produce SHG that is highly sensitive to membrane potential. This may allow new optical approaches to be developed for mapping electrical activity in complex neuronal systems. Excitation uses near-infrared wavelengths, allowing excellent depth penetration, and thus this method is well suited for studying thick tissue samples. Information about the molecular organization of chromophores, including dyes and structural proteins, can be extracted from SHG imaging data in several ways. SHG signals have well-defined polarizations, and thus SHG polarization anisotropy can be used to determine the absolute orientation and degree of organization of proteins in tissues. In addition, TPEF images can be collected in a separate data channel simultaneously with SHG. Correlation between the SHG and TPEF images provides the basis not only for molecular identification of the SHG source but also for probing radial and lateral symmetry within structures of interest. These special characteristics of SHIM as well as their current limitations are reviewed in this article.
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
PERSPECTIVE
Box 1 Theoretical background and fundamental equations The nonlinear polarization for a material is defined by:
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
P = χ(1) ⊗ E + χ(2) ⊗ E ⊗ E + χ(3) ⊗ E ⊗ E ⊗ E + ...
(1)
where P represents the induced polarization vector, E represents the vector electric field, χ(i) is the i th-order nonlinear susceptibility tensor and ⊗ represents a combined tensor product and integral over frequencies. The χ(i) corresponds to optical effects as follows: • 1st-order process: absorption and reflection • 2nd-order process: SHG, sum and difference frequency generation, hyper-Rayleigh • 3rd-order process: multiphoton absorption, third harmonic generation, coherent anti-Stokes Raman scattering The second nonlinear susceptibility is a bulk property and related to the molecular hyperpolarizability, β, by: χ(2) = Ns < β >
(2)
where Ns is the density of molecules and the brackets denote an orientational average, which shows the need for an environment lacking a center of symmetry. The second-harmonic intensities in such media will scale as follows: SHGsig ∝ p2τ(χ(2))2
(3)
where p and τ are the laser pulse energy and pulse width, respectively. The magnitude of the SHG intensity can be strongly enhanced when the energy of the – ω) is in resonance with an electronic absorption band (h – ω ). Within the SHG signal (2h ge two-level system model, the first hyperpolarizability, β, and thus SHG efficiency is given by ωge ƒge ∆µge 3e 2 β ≈ —– —————————–– 3 2– 2 – ω ][ωge2 – 4ω2] [ω 2h ge
(4)
where e is electron charge and ωge, ƒge and ∆µge are the energy difference, oscillator strength (that is, integral of the absorption spectrum) and change in dipole moment between the ground and excited states, respectively. Because of the denominator in this equation, resonance-enhanced SHG has a dependence on the wavelength of the incident light similar to the two-photon excitation spectrum.
History and instrumentation The first reports of the integration of SHG and microscopy appear to have been by Hellwarth and Christensen12 in 1974 and Sheppard et al.13 in 1977. To the best of our knowledge, the first biological SHG imaging experiments were done by Freund and colleagues in 1986 (ref. 1). The researchers used SHG to study the orientation of collagen fibers in rat tail tendon at ∼50-µm resolution and showed that the collagen fibers formed highly dipolar structures at this scale. Initial reports by one of us (L.M.L.) and Lewis examined the secondharmonic response of styryl dyes in electric fields14 and showed the possibility of imaging live cells by SHG15. In all this earlier work, stage scanning with a picosecond laser source focused through the microscope was used and frame rates of minutes to hours were required. To speed up the process, we modified a laser-scanning two-photon microscope to obtain SHG images with pixel density similar to that obtained with a standard confocal microscope with similar acquisition rates4. The enabling technologies that contributed to this advance were improved dye development, commercially available femtosecond titanium lasers and the use of single-photon counting for data acquisition. The last technology is required, especially for imaging dye-stained cells, because the SHG signal power per incident photon is much smaller than that for TPEF16.
Because SHG is a coherent process, most of the signal wave propagates with the laser. The exact ratio of the forward to backward signal is dependent upon the sample characteristics. For thin samples, such as tissue culture cells, essentially the entire signal is directed forward. At higher sample turbidity, some of the SHG is scattered backward. The use of an upright microscope has some advantages for SHG imaging over that of an inverted geometry. The optical path is simpler for trans collection and this geometry also makes it more straightforward to implement the additional external optics required for polarization anisotropy measurements (described below). Commercial titanium sapphire femtosecond lasers are ideal for SHG (and TPEF) because of the high repetition rate (80 MHz) and high peak powers (and low pulse energies) and the broad tunability throughout the near infrared (700–1,000 nm). Because SHG intensities are typically smaller than those of TPEF, it is important to optimize the collection and detection efficiency for both signal isolation and detection electronics. SHG imaging can also determine molecular symmetries by use of polarization analysis. SHG polarization anisotropy measurements are made with a Glan Laser Polarizer through which the data is obtained by maintaining the same input laser polarization and obtaining images with the analyzing polarizer oriented both parallel and perpendicular to the laser polarization. In addition, radial and lateral symmetries are probed by rotating the plane of polarization of the laser with half (λ/2) and quarter (λ/4) wave plates.
Endogenous imaging of structural protein arrays Recently, it has been observed that very large SHG signals are directly obtainable from several structural protein arrays in tissues, without the addition of fluorescent dyes6–11,17–22. Our group6,17,20,21 has examined structural proteins, including collagen, actomyosin complexes and tubulin, from several animal sources, among them mouse, tetra fish and Caenorhabditis elegans. Historically, these protein structures have been studied with other imaging modalities, including electron microscopy and polarization microscopy23, and they are known to form arrays that are highly ordered and birefringent. The ability to image completely off resonance has the great benefit of the virtual elimination of photobleaching and phototoxicity, especially at longer wavelengths (λ > 850 nm). In addition, more detailed molecular information is readily elucidated than is possible through the use of fluorescent labels, even when dyes are conjugated to proteins. This is because dyes only infer details of protein assembly and motion, whereas SHG directly visualizes the submicrometer- and micrometerscale assemblies. One focus of our work—whose long-term goal is the study and diagnosis of muscle-related diseases—has been to investigate the use of SHG in studying the assembly of actomyosin complexes. It is important to determine the possible depth of penetration into muscle
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1357
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE a
a
b
b
Figure 1 Endogenous SHG images of native mouse leg. (a,b) Within a 550-µm-thick tissue stack, images obtained from depths of ∼200 µm (a) and 300 µm (b), respectively. The sarcomere repeat patterns are observed in both slices.
Figure 2 Endogenous SHG imaging of a living adult C. elegans nematode, showing two distinct axial slices. (a) The sarcomeres in the body wall muscles are seen at the edges of the animal, as well as in a portion of the chewing mechanism. (b) An optical section further into the same animal, where only the chewing mechanism is observed with substantial SHG intensity.
tissue as well as the molecular source of the SHG contrast. We address the first issue in Figure 1, which shows two slices of an explanted mouse leg muscle from regions approximately 200 and 300 µm deep from within a stack of tissue 550 µm thick. Despite the turbidity of muscle, these images display, with high contrast, the sarcomere repeat pattern that is characteristic of striated muscle. Indeed, only a fourfold loss of intensity was observed through the range of the entire stack. It should be noted that in all our work, we use the forward-detected geometry. In the present case, epidetection would have resulted in decreased signal intensity because less second harmonic is produced by backscattering, and the resulting lower-wavelength SHG light would undergo greater scattering losses as well. Nonetheless, the epidetection geometry may be the only viable method for imaging entire organs or entire, intact animals. Although the SHG contrast of the sarcomeres resembles that seen by polarization microscopy, we have previously shown that the contrast is not identical. This is because the contrast in a polarization microscope arises from linear birefringence in the sample, whereas that of SHG arises from a quadratic process. The squared dependence on the protein concentration can then give rise to different features in the image. We see SHG as a more powerful imaging modality than polarization microscopy because of SHG’s intrinsic sectioning and because polarization microscopy does not readily yield quantitative molecular level properties. In contrast, using the appropriate combinations of laser polarization and signal polarizations, all the relevant matrix elements of χ(2) that give rise to the SHG signals can be determined, and thus yield the absolute orientation of fibrous structures and the degree of organization of the molecular dipoles. For example, using SHG polarization anisotropy, in which the laser polarization is kept fixed and the SHG signal is analyzed by a polarizer, we have observed that the collagen fibers in a tetra fish scale are highly anisotropic6. We determined the anisotropy parameter r using the expression for electric dipole distributions
1358
Ipar – Iperp r = ————– Ipar + 2Iperp where Ipar and Iperp are the intensities of the signals whose polarizations are parallel and perpendicular, respectively, to the polarization of the incident laser. The limiting values of r = 0 and 1 correspond to the completely isotropic and aligned cases, respectively. We found an anisotropy parameter of r = 0.7, indicating that the dipoles in the collagen fibers form well-aligned structures. This type of data will be critical in the use of SHG in probing diseased states and differentiating between normal and abnormal tissue. For example, we are currently examining such differences in the micrometer-scale morphology of collagen fibers in the diseases osteogenesis imperfecta and tight skin through SHG imaging (P.J.C., unpublished data). One aspect to be determined in this work is the orientation of the molecular dipole relative to the long axis of the fibers. The SHG anisotropy approach of interrogating protein organization cannot achieve this directly by fluorescence anisotropy of dyeconjugated proteins because the loosely attached dye can rotate and ‘wash out’ some of the encoded structural information content. We have also been similarly examining the molecular source of SHG in actomyosin complexes. In this work, we have largely used C. elegans, as these animals are optically clear and easy to manipulate genetically. Two frames of a three-dimensional stack of a live C. elegans nematode show the body wall muscles (Fig. 2a) and the chewing mechanism (Fig. 2b). To compare the SHG and fluorescence contrast, we raised C. elegans with green fluorescent protein (GFP)-labeled myosin and with myosin mutations, and showed that the SHG arises from the thick filaments. In particular, the data were consistent with the myosin heavy chain B isoform being the dominant source of SHG, with little contribution from either myosin heavy chain A or actin filaments. This example shows how the combination of simultaneous SHG and TPEF
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
PERSPECTIVE a
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
b
c
Figure 3 Membrane potential sensitivity of SHG from di-4-ANEPPS on a voltage-clamped N1E-115 neuroblastoma cell. (a) Structure of di-4-ANEPPS. (b) The results from a single experiment in which the cell is cycled between 0 mV and –100 mV every three image frames. Each point represents the integral of SHG intensity around the cell periphery. The dye is introduced shortly before the start of the image sequence and a small upward drift is seen as dye translocates from the bathing solution to the cell membrane. The incident laser is at 850 nm and SHG is detected at 425 nm. (c) The mean change in SHG as a function of membrane potential, with 0 mV as the reference. (Adapted from Millard et al.29, by permission of the Optical Society of America.)
can be used to provide detailed structural data not possible by either alone. As our understanding of SHG in tissues expands, however, it will not always be necessary to use this combination imaging approach. Although SHG is technologically much more complex, we see it as being a viable alternative to normal histologic analysis, as samples can be imaged in their natural state without fixation, labeling (and consequent photobleaching) and sectioning. Furthermore, SHG provides more complete structural information than polarization microscopy, as all the matrix elements of the second-order susceptibility that give rise to the contrast can be determined through the appropriate combinations of excitation and analyzer polarizations. We further expect that SHG will have a substantial impact on in vivo studies in various fields of biology and medicine, including tissue organization, wound healing, myofilament assembly, muscle development and disease, aging, and the division cycle of normal and cancerous cells in situ. The determination of the fundamental properties of the SHG contrast in simple model systems is the first, critical step in making the technique broadly useful. Already, we and other researchers have begun to extend these methods to the analysis of fibrillar species in connective tissue and studies of
skin and muscle and brain pathology. For example, Jain and coworkers24 have recently used SHG imaging to compare collagen content in tumors, identifying primarily type 1 collagen as the SHG source. Cox et al.22 found that SHG can discriminate between type I and type III collagen in several tissue specimens. Similarly, Tromberg and colleagues8 used SHG to probe the assembly of collagen in explanted rabbit corneas, to explore if SHG could be used as a nondestructive ophthalmological imaging tool. In other work, Webb and coworkers10 showed that the polarity of microtubules is uniform in native brain and suggested that SHG could perhaps be used not only to probe neuronal polarity development, but also to investigate Alzheimer disease. SHG imaging enabled researchers in these studies to visualize the structural protein arrays directly (and quantitatively), rather than indirectly as with fluorescently tagged proteins or antibodies. As our understanding of the SHG process continues to expand, this may well become a useful tool complementary to ultraresolution structural techniques such as electron microscopy or X-ray diffraction, which will provide a complete picture of tissue assembly. SHIM of electrochromic membrane dyes A consequence of equation (4) (Box 1) for resonance-enhanced SHG is that β depends on a large difference in the ground- and excited-state electron distribution (that is, a large µe − µg). This is also a key requirement for electrochromism—the sensitivity of the dye linear spectra to electric field. Our laboratory25 has designed and synthesized a large number of electrochromic styryl dyes as fluorescent probes of membrane potential, and this led us to initially explore, in a fruitful collaboration with the laboratory of A. Lewis, the possibility that the dyes may produce large SHG signals26. In subsequent studies, we have used specially synthesized chiral dyes to boost the SHG signal4,15,20,27. It should be noted, however, that because population of the dye-excited state is a byproduct (via two-photon absorption) of resonance-enhanced SHG, some collateral photobleaching and phototoxicity may occur. We found that a strong SHG signal with achiral dyes is possible only if the dyes stain just one leaflet of the membrane bilayer—usually the outer leaflet because the dyes are applied from the external bath; if the dye equilibrates across the membrane, the requirement for a noncentrosymmetric distribution is violated and the SHG signal is abolished. Using this class of dyes, SHIM has also been investigated in the laboratory of Mertz5,28. In addition to a thorough theoretical analysis of SHG cross-sections and polarization effects, they have shown that for aggregated lipid vesicles, regions where two membranes are in close apposition show no SHG, even though TPEF from these regions is strong; again, the close (less than the wavelength of the SHG signal) apposition of two stained membranes produces a symmetric distribution of dyes that does not allow SHG. Thus, TPEF and SHIM even from the same labeling dye can produce usefully different information, because the fluorescence signal and the second-harmonic signal arise from fundamentally different physics. In dual labeling experiments, complementary information is to be anticipated, as the SHG dyes can provide information about local structural organization, whereas fluorescent labels can provide information about molecular distributions. The relationship between SHG and electrochromism shown in equation (4) (Box 1) also prompted us to ask whether SHG from membranes stained with our dyes could be sensitive to membrane potential. In our initial experiments, SHG from a dye-stained lipid bilayer was highly sensitive to membrane potential14. Experiments on live cells have confirmed this effect4,15,29. The sensitivity to membrane potential was most precisely revealed in recent voltage-clamp studies of neuroblastoma cells stained with one of our standard voltagesensitive dyes, di-4-ANEPPS, in which the voltage was cycled from 0 to –100 mV every three frames29 (Fig. 3). The SHG of di-4-ANEPPS was
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1359
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE linearly dependent on membrane potential, showing a relative change of 18% per 100 mV with a laser wavelength of 850 nm. The Mertz laboratory found similar sensitivities in a study of lipid vesicles placed in external uniform electric fields, and this group has also provided a detailed theory for the modulation of SHG by electric fields30. We have found that increasing the wavelength to 950 nm results in sensitivities of 40% per 100 mV (L.M.L., unpublished data). Such sensitivities are fourfold greater than can be achieved with the best fluorescence recordings of electrical activity with voltage-sensitive dyes. Even so, a major challenge for the practical application of SHIM to the recording of electrical activity in neuronal systems is the SHG signal size. Although we have made great strides in reducing the acquisition time required to accumulate enough signal from many minutes for a 100 × 100 pixel image in the early experiments to just one second for a 512 × 512 pixel image in our current SHIM apparatus, this is still at least two orders of magnitude too slow to image action potentials. One approach is to limit the number of spatial points sampled in the laser scanning microscope, most commonly by scanning just single lines, so that data can be acquired more rapidly; of course, this will limit the ability to map electrical activity in a morphologically complex neuronal preparation. Another approach is to develop improved dyes with greater electrochromism and hyperpolarizabilities. A radically different strategy for dye development that has shown some preliminary promise27,31,32 is to attach available electrochromic chromophores, such as aminonaphthyl derivatives (ANEPs), to silver or gold nanoparticles; the metal particles locally focus the laser electromagnetic field via a plasmon resonance effect, thereby enhancing SHG signals from neighboring dye molecules. We are working to develop easily deployable metal-dye conjugates to fully exploit this idea. Conclusions The promise of SHIM as a new tool for dynamic imaging of biological structure and function is now apparent. Nondestructive imaging of several endogenous proteins has been demonstrated and more examples are likely to be discovered. Detailed structural information as well as indications of pathology can be obtained from these images. Although the largest signals can be obtained in transmission configurations, it has also been shown that collagen can be imaged in a backscattering geometry. This will further extend the applicability of SHIM to studies of tissues in intact animals where transmitted light cannot be collected. For imaging membrane potential, the high sensitivity of the SHG signal promises to expand the application of optical recording of electrical activity in neuronal systems to the mapping of more complex preparations than had been previously possible. However, better dyes and improved optics will be needed to permit routine detection of action potentials by SHG. The simplicity of the modifications required to enable SHG detection in a multiphoton laser scanning microscope lead us to anticipate that this feature will shortly be offered by all the commercial manufacturers of these instruments. This will, of course, assure the dissemination of this exciting technology and prompt the discovery of new applications for both the basic research laboratory and the clinic. ACKNOWLEDGMENTS The authors would like to thank their collaborators Aaron Lewis, Andrew Millard, William Mohler, Heather Clark and David Boudreau for their contributions to this work. We gratefully acknowledge financial support under US Office of Naval Research grant no. N0014-98-1-0703, National Institute of Biomedical Imaging and Bioengineering grant no. R01EB00196 and National Center for Research Resources grant no. R21 RR13472. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
1360
Published online at http://www.nature.com/naturebiotechnology/ 1. Freund, I., Deutsch, M. & Sprecher, A. Connective tissue polarity. Optical second harmonic microscopy, crossed-beam summation, and small-angle scattering in rat tail tendon. Biophys. J. 50, 693–712 (1986). 2. Huang, Y.A., Lewis, A. & Loew, L.M. Nonlinear optical properties of potential sensitive styryl dyes. Biophys. J. 53, 665–670 (1988). 3. Rasing, T., Huang, J., Lewis, A., Stehlin, T. & Shen, Y.R. In situ determination of induced dipole moments of pure and membrane-bound retinal chromophores. Phys. Rev. A 40, 1684–1687 (1989). 4. Campagnola, P.J., Wei, M.–d., Lewis, A. & Loew, L.M. High-resolution optical imaging of live cells by second harmonic generation. Biophys. J. 77, 3341–3349 (1999). 5. Moreaux, L., Sandre, O., Blanchard–Desce, M. & Mertz, J. Membrane imaging by simultaneous second harmonic generation and two-photo microscopy. Optics Lett. 25, 320–322 (2000). 6. Campagnola, P.J. et al. Three-dimensional high-resolution second harmonic generation imaging of endogenous structural proteins in biological tissues. Biophys. J. 82, 493–508 (2002). 7. Zoumi, A., Yeh, A. & Tromberg, B.J. Imaging cells and extracellular matrix in vivo by using second harmonic generation and two-photon excited fluorescence. Proc. Natl. Acad. Sci. USA 99, 11014–11019 (2002). 8. Yeh, A.T., Nassif, N., Zoumi, A. & Tromberg, B.J. Selective corneal imaging using combined second harmonic generation and two-photon excited fluorescence. Optics Lett. 27, 2082–2084 (2002). 9. Zipfel, W.R. et al. Live tissue intrinsic emission microscopy using multiphoton–excited native fluorescence and second harmonic generation. Proc. Natl. Acad. Sci. USA 100, 7075–7080 (2003). 10. Dombeck, D.A. et al. Uniform polarity microtubule assemblies imaged in native brain tissue by second harmonic generation microscopy. Proc. Natl. Acad. Sci. USA 100, 7081–7086 (2003). 11. Brown, E. et al. Dynamic imaging of collagen and its modulation in tumors in vivo using second harmonic generation. Nat. Med. 9, 796–800 (2003). 12. Hellwarth, R. & Christensen, P. Nonlinear optical microscopic examination of structure in polycrystalline ZnSe. Optics Comm. 12, 318–322 (1974). 13. Sheppard, C.J.R., Kompfner, R., Gannaway, J. & Walsh, D. Scanning harmonic optical microscope. IEEE J. Quantum Electron. 13E, 100D (1977). 14. Bouevitch, O., Lewis, A., Pinevsky, I., Wuskell, J.P. & Loew, L.M. Probing membrane potential with non-linear optics. Biophys. J. 65, 672–679 (1993). 15. Ben-Oren, I., Peleg, G., Lewis, A., Minke, B. & Loew, L.M. Infrared nonlinear optical measurements of membrane potential in photoreceptor cells. Biophys. J. 71, 1616–1620 (1996). 16. Moreaux, L., Sandre, O. & Mertz, J. Membrane imaging by second harmonic generation microscopy. J. Opt. Soc. Am. B 17, 1685–1694 (2000). 17. Mohler, W., Millard, A.C. & Campagnola, P.J. Second harmonic generation imaging of endogenous structural proteins. Methods 29, 97–109 (2003). 18. Stoller, P., Reiser, K.M., Celliers, P.M. & Rubenchik, A.M. Polarization-modulated second harmonic generation in collagen. Biophys. J. 82, 3330–3342 (2002). 19. Stoller, P., Kim, B.M., Rubenchik, A.M., Reiser, K.M. & Da Silva, L.B. Polarization-dependent optical second harmonic imaging of a rat-tail tendon. J. Biomed. Opt. 7, 205–214 (2002). 20. Campagnola, P.J., Clark, H.A., Mohler, W.A., Lewis, A. & Loew, L.M. Second harmonic imaging microscopy of living cells. J. Biomed. Opt. 6, 277–286 (2001). 21. Millard, A.C., Campagnola, P.J., Mohler, W., Lewis, A. & Loew, L.M. Second harmonic imaging microscopy. Methods Enzymol. 361, 47–69 (2003). 22. Cox, G. et al. 3-dimensional imaging of collagen using second harmonic generation. J. Struct. Biol. 141, 53–62 (2003). 23. Inoue, S. Video Microscopy (Plenum Press, New York, 1986). 24. Brown, E. et al. Dynamic imaging of collagen and its modulation in tumors in vivo using second harmonic generation. Nat. Med. 9, 796–800 (2003). 25. Loew, L.M. in Fluorescent and Luminescent Probes for Biological Activity (ed. Mason, W.T.) 210–221 (Academic Press, London, 1999). 26. Huang, J.Y., Lewis, A. & Loew, L.M. Non-linear optical properties of potential sensitive styryl dyes. Biophys. J. 53, 665–670 (1988). 27. Clark, H.A., Campagnola, P.J., Wuskell, J.P., Lewis, A. & Loew, L.M. Second harmonic generation properties of fluorescent polymer encapsulated gold nanoparticles. J. Am. Chem. Soc. 122, 10234–10235 (2000). 28. Moreaux, L., Sandre, O., Charpak, S., Blanchard–Desce, M. & Mertz, J. Coherent scattering in multi-harmonic light microscopy. Biophys. J. 80, 1568–1574 (2001). 29. Millard, A.C., Jin, L., Lewis, A. & Loew, L.M. Direct measurement of the voltage sensitivity of second harmonic generation from a membrane dye in patchclamped cells. Optics Lett. 28, 1221–1223 (2003). 30. Moreaux, L., Pons, T., Dambrin, V., Blanchard–Desce, M. & Mertz, J. Electro–optic response of second harmonic generation membrane potential sensors. Optics Lett. 28, 625–627 (2003). 31. Peleg, G. et al. Gigantic optical non-linearities from nanoparticle enhanced molecular probes with potential for selectively imaging the structure and physiology of nanometric regions in cellular systems. Bioimaging 4, 215–224 (1996). 32. Peleg, G., Lewis, A., Linial, M. & Loew, L.M. Nonlinear optical measurement of membrane potential around single molecules at selected cellular sites. Proc. Natl. Acad. Sci. USA 96, 6700–6704 (1999).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
PERSPECTIVE
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
FOCUS ON OPTICAL IMAGING
Optical coherence tomography for ultrahigh resolution in vivo imaging James G Fujimoto Optical coherence tomography (OCT) is an emerging biomedical optical imaging technique that performs highresolution, cross-sectional tomographic imaging of microstructure in biological systems. OCT can achieve image resolutions of 1–15 µm, one to two orders of magnitude finer than standard ultrasound. The image penetration depth of OCT is determined by the optical scattering and is up to 2–3 mm in tissue. OCT functions as a type of ‘optical biopsy’ to provide cross-sectional images of tissue structure on the micron scale. It is a promising imaging technology because it can provide images of tissue in situ and in real time, without the need for excision and processing of specimens. Since its introduction the early 1990s, OCT has become a powerful method for imaging the internal structure of biological systems and materials1. OCT is analogous to ultrasound, except that it measures the echo time delay and magnitude of light rather than sound. It provides images of tissue pathology in situ and in real time, without the need to remove and process specimens, as in conventional excisional biopsy and histopathology2. The technology promises to have applications in a wide range of clinical situations: imaging tissue pathology when excisional biopsy is hazardous or impossible; guiding surgical procedures; and reducing sampling errors associated with excisional biopsy3–5. This article discusses the principles and applications of OCT. It also briefly describes the clinical development of the technology in the context of ophthalmic imaging, the most clinically advanced application of OCT to date. Optical coherence tomography imaging In contrast to conventional microscopy, OCT provides cross-sectional images of structure below the tissue surface in analogy to histopathology. Standard-resolution OCT can achieve axial resolutions of 10–15 µm. Recently, using state-of-the-art lasers as light sources, ultrahigh-resolution imaging with axial resolutions as fine as 1–2 µm has been demonstrated6. The maximum imaging depth in most tissues (other than transparent tissues such as the eye) is limited by optical attenuation and scattering to approximately 2–3 mm2,7. Although this depth is shallow compared with other clinical imaging Department of Electrical Engineering and Computer Science and Research Laboratory of Electronics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. Correspondence should be addressed to J.G.F (
[email protected]). Published online 31 October 2003; doi:10.1038/nbt892
techniques, the image resolution of OCT is 10–100 times finer than conventional ultrasound imaging, magnetic resonance imaging or computed tomography. Several features of OCT imaging make it attractive for imagingbased diagnostics and surgical guidance. First, OCT can provide resolutions approaching that of conventional histopathology. Second, OCT imaging can be done in situ and in real time, allowing surgical guidance, guidance of biopsy or investigation of dynamic response to therapeutic agents. Third, OCT imaging may be carried out using noninvasive or minimally invasive delivery systems such as microscopes, handheld probes, endoscopes, catheters, laparoscopes and needles. Fourth, OCT can perform functional imaging, such as spectroscopic imaging of tissue properties, Doppler blood flow measurement and quantification of blood oxygenation or tissue birefringence. And finally, imaging processing techniques and intelligent algorithms can be used to assess OCT images quantitatively and extract diagnostic information. Principles of operation of OCT The concept of ‘seeing through tissue’ using time-resolved measurement of light was proposed almost 30 years ago8 in the field of picosecond optics. The principle of OCT imaging is analogous to that of ultrasound B-mode imaging except that OCT uses light rather than acoustic waves. Cross-sectional images are generated by scanning an optical beam across the tissue and measuring the echo time delay and intensity of backscattered light (Fig. 1). Because the velocity of light is extremely high, optical echoes cannot be measured by direct electronic detection. Instead OCT is based on a technique called low-coherence interferometry, which is similar to white-light interferometry, first described by Sir Isaac Newton. In low-coherence interferometry, light reflected or backscattered from inside the specimen is measured by correlating with light that has traveled a known reference path. This technique has been applied in optoelectronic devices to perform micron-resolution optical ranging and measurement9–11. Some of the first biomedical applications of low-coherence interferometry were measurements of axial eye length and corneal thickness12,13. OCT imaging is performed using a fiber-optic Michelson interferometer with a low-coherence-length light source. The fiber-optic implementation provides a compact and robust system that can be interfaced to a variety of clinical imaging instruments. Low-coherence light can be generated by compact superluminescent semiconductor diodes or other sources, such as solid-state lasers. One interferometer arm contains a modular probe that focuses and scans the light onto the sample, also collecting the backscattered light. The second interferometer arm is a reference path with a translating mirror or scanning delay
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1361
PERSPECTIVE Axial (longitudinal) measurement
a
b Transverse scan
Distance
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Reflectivity
c
d Reference path
Scanning optical delay
Light source Sample Detector
Light source
Transverse scanning Beam splitter
Receiver demodulator Detector
Computer image generation Slitlamp biomicroscope
Figure 1 Principles of OCT imaging. (a) OCT performs cross-sectional imaging of internal microstructure in tissue by measuring the echo time delay and magnitude of reflected light. (b) Cross-sectional images are generated by scanning the incident light beam at different transverse positions. The resulting two-dimensional data set can be displayed as a gray-scale or false-color image. (c) Echo time delays of light are measured using low coherence interferometry with a Michelson-type interferometer and low-coherence light source. (d) Fiber-optic implementations provide a compact and robust system that can be interfaced to a wide range of imaging instruments.
line. Optical interference between the light from the sample and reference paths occurs only when the distance traveled by the light in both paths matches to within the coherence length of the light14. The interference fringes are detected and demodulated to produce a measurement of the magnitude and echo delay time of light backscattered from structures inside the tissue. Low-coherence interferometry thus enables femtosecond time resolution of optical echoes, corresponding to micron-scale distance measurement. In OCT, two-dimensional cross-sectional images of internal tissue microstructure are constructed by scanning the optical beam and performing multiple axial measurements of backscattered light at different transverse positions. The resulting data set is a two-dimensional array that represents the optical backscattering or reflection within a cross-sectional slice of the tissue specimen. These data can be digitally filtered, processed and displayed as a two-dimensional gray-scale or false-color image. In addition to imaging in a cross-sectional plane, imaging in en face planes at a given depth is also possible15–17. The axial resolution of an OCT image depends on the coherence length of the light and is independent of beam focusing conditions and numerical aperture. The transverse resolution for OCT imaging is determined by the focused spot size, as in microscopy. Because OCT uses interferometry, it performs optical heterodyne detection, and extremely high sensitivities, near the quantum limit, are possible14. For
1362
typical image sizes and resolutions, OCT systems have a sensitivity approaching –100 dB, meaning that reflected signals as small 10–10 may be detected. The image penetration depth is determined by the absorption and scattering of tissue. Using wavelengths in the near infrared, where hemoglobin and melanin absorption are low and scattering is reduced, permits imaging depths of up to 2–3 mm in tissues. The ability to perform in situ, real-time imaging of tissue pathology enables a wide range of clinical applications in three general categories: (i) imaging tissue where conventional excisional biopsy is hazardous or impossible, (ii) guiding surgical or microsurgical procedures and (iii) guiding excisional biopsy to reduce false negatives caused by sampling errors. Ophthalmic and arterial imaging Ophthalmic imaging is an example of a clinical situation in which conventional excisional biopsy is not possible. OCT can provide information on retinal pathology in vivo that cannot be obtained by any other method. OCT was first used to image the eye, and currently the most successful clinical application of OCT is in ophthalmology18–20. Although the retina is virtually transparent, with extremely low optical backscattering, the high sensitivity of OCT allows extremely weak signals to be detected. Figure 2 shows examples of OCT retinal imaging and illustrates the improvement in OCT technology from the first
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
PERSPECTIVE a
Retina
d
BV Vitreous
0°
90°
BV SFR
RPE Temporal
Nasal Inferior
30°
Superior
120°
Sclera
250 m
300 µm
60°
150°
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Log reflection
b
Fovea
NFL Optic disc
250 µm
Log reflection
OCT normal macular thickness (µm) Sup
250 um
c
Tem
Nas 238 ± 14
Inf
GCL
IPL
Fovea
OPL
264 ± 15
NFL 229 ± 15
ONL Center
Choriocapillaris
255 ± 16
263 ± 16
Mean 152 ± 21 s.d. 11 ± 6
Photoreceptors
248 174 260 ± 15 ± 18 ± 16
225 ± 15
Figure 2 Ophthalmic OCT imaging and its development. (a) The first demonstration of OCT imaging showing imaging of the human retina ex vivo acquired at 800 nm with 10-µm resolution (reprinted by permission of Science from ref. 1). (b) Example of current OCT clinical retinal image produced by the StratusOCT instrument (Carl Zeiss Meditec) showing the papillary-macular axis of the retina between the fovea and optics disc. (c) Image of the fovea showing normal architectural morphology of the retina. Visible features include NFL, nerve fiber layer; GCL, ganglion cell layer; IPL, inner plexiform layer; OPL, outer plexiform layer; and INL, inner nuclear layer. (d) Topographic map of the macula constructed by segmenting multiple OCT images and showing thicknesses using a false color scale (reprinted by permission of Opthalmology from ref. 29). Image processing methods such as these enable quantitative analysis of images to aid disease diagnosis.
150 200 250 300 350 400 450 500 µm
ex vivo images to current clinical applications. OCT enables the internal architectural morphology of the retina to be visualized noninvasively and in real time. The retinal nerve fiber layer is visible as a scattering layer originating from the optic disk, becoming thinner as it approaches the fovea. The plexiform layers are visible as highly scattering, whereas the nuclear layers are low scattering. Numerous clinical studies have been carried out to investigate the use of OCT for diagnosing and monitoring retinal diseases, such as glaucoma21,22, macular hole23,24, macular edema25, central serous chorioretinopathy26, age-related macular degeneration27 and epiretinal membranes28. Images can be analyzed quantitatively and processed using intelligent algorithms to extract such features as retinal or retinal nerve fiber layer thickness21,22. Mapping and display techniques have been developed to represent the tomographic data in alternative forms, such as thickness maps, to aid interpretation29,30. OCT is especially promising for disease diagnosis and monitoring because it can provide quantitative information about retinal pathology and monitor disease progression30–34. OCT has the potential to detect and diagnose early stages of disease before physical symptoms and irreversible vision loss can occur. Another situation where excisional biopsy is not possible is imaging of atherosclerotic plaque morphology. Recent research has shown that most myocardial infarctions result from ‘unstable plaques,’ small to moderately sized, cholesterol-laden plaques that have a high risk of rupturing, causing acute thrombosis and vessel occlusion35–37. These unstable plaques have a structurally weak fibrous cap and are difficult to visualize by conventional radiologic techniques such as angiography. OCT is a promising tool in diagnostic intravascular imaging for the identification of unstable plaques as well as for the guidance of interventional procedures such as atherectomy and stent placement. Figure 3 provides examples of OCT arterial imaging and its development. In Figure 3a, the OCT image and histology of a surgical specimen show an unstable plaque with a thin intimal layer covering a large atherosclerotic plaque that is heavily calcified, with relatively low lipid content38. Imaging can be performed using fiber-optic scanning catheter-endoscopes39,40. One prototype catheter consists of a single-mode optical fiber in a hollow rotating or translating cable that emits and scans the OCT beam radially from the catheter axis (Fig. 3b). Catheters with diameters as small as 0.356 mm (0.014 in.)
are commercially available (LightLab Imaging; Westford, MA, USA). The first catheter-based arterial imaging studies, conducted in the mid 1990s, demonstrated that OCT provided superior visualization of the intima, media and adventitia as compared to intravascular ultrasound (IVUS)41,42. In vivo imaging studies have been performed in animal models (Fig. 3c)43,44, and several groups have carried out in vivo human clinical studies of OCT imaging for the assessment of coronary plaques and stenting (Fig. 3d)45–47. Because blood is optically scattering, balloon occlusion or saline flushing is required before
a
b
500 µm
c
d Stent strut
Figure 3 Intravascular OCT imaging and its development. (a) OCT image ex vivo and histology showing unstable plaque with a thin intimal cap layer (reprinted by permission of Circulation from ref. 38). (b) Prototype fiberoptic OCT catheter consisting of a rotating fiber and microlens assembly encased in a transparent sheath. (c) OCT catheter-based image of a stented rabbit aorta in vivo. (d) Clinical OCT image of right coronary artery of a human subject in vivo, from a follow-up study of drug-eluting stents. (Courtesy of E. Grube, Siegburg Heart Center, Siegburg, Germany, and LightLab Imaging, Westford, MA, USA.)
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1363
PERSPECTIVE a
c m
Barrett’s epithelium
mm
OCT imaging probe
sm
500 µm
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
m mm sm
d
b
ep lp
mm sm
c
mp
Figure 4 Imaging neoplastic changes. (a) OCT images and corresponding histology of gastrointestinal tissues in vitro. The normal colon has columnar epithelial morphology with crypt structures (reprinted by permission of American Journal of Gastroenterology from ref. 48). (b) Image of carcinoma shows loss of normal epithelial structure with disruption and disorganization of the crypts. (c) Endoscopic imaging of the esophagus: endoscopic view of normal esophagus and Barrett’s esophagus (from ref. 51). (d) OCT image of normal esophagus, showing squamous epithelial structure, and Barrett’s esophagus, showing disruption of squamous epithelium and formation of glandular structures (c,d reprinted by permission of Endoscopy from ref. 51).
500 µm 500 µm
c g
intravascular imaging. Balloon imaging devices permit image-guided atherectomy or stent placement. Detecting neoplastic changes Another important category of OCT imaging applications is the screening and detection of neoplastic changes, where conventional excisional biopsy can have unacceptably high false-negative rates
because of sampling error. OCT can resolve changes in architectural morphology that are associated with many neoplastic changes. Figure 4 provides examples of OCT imaging in the gastrointestinal tract48. Many carcinomas are associated with disruption or loss of architectural morphology and glandular organization. The development of highspeed OCT imaging combined with small fiber-optic probes enabled in vivo endoscopic OCT imaging40. Endoscopic OCT can be performed using fiber-optic imaging probes introduced into the biopsy port of standard endoscopes49–51. Figure 4c,d show examples of endoscopic OCT imaging of the human esophagus51. The OCT image of the normal esophagus shows the normal layered structure of squamous epithelium with a well-differentiated epithelium, lamina pro-
Box 1 From research laboratory to clinical practice To have an impact on health care, a technology must be transferred from the laboratory to the clinic. Medical instrumentation is challenging to develop because multiple hurdles must be overcome before widespread clinical use can be achieved. The stages in this process include fundamental studies to demonstrate technical feasibility, laboratory prototype development, clinical feasibility studies, commercial prototype development, multicenter clinical trials, US Food and Drug Administration (FDA; Rockville, MD, USA) and regulatory approval, sales to early adopters (clinical research sites), demonstration of clinical efficacy, health-care reimbursement, and finally, sales to the clinical community. Although the development of technology is less costly than that of pharmaceuticals, it remains challenging because of the long development process, rapid evolution of technology and inherently small market associated with medical instrumentation. OCT has had the most significant clinical impact in ophthalmology. However, the transition of OCT from research to clinical practice in ophthalmology required over a decade. OCT was developed in the early 1990s and the first in vitro retinal imaging was performed in 1991. The first demonstration of in vivo retinal imaging took place in 1993 and clinical studies got underway in 1994. By 1995, crosssectional clinical studies had already been performed to investigate the major retinal pathologies. OCT was transferred to industry in 1993 and was introduced to the ophthalmic marketplace by Carl
1364
Zeiss Meditec (formerly named Humphrey Systems; Dublin, CA) in 1996. The adoption of OCT proceeded slowly, with sales of only a few hundred instruments during a period of several years. A secondgeneration instrument was introduced in 2000. The instrument was then re-engineered to achieve a fourfold increase in imaging speed and a third-generation instrument (the StratusOCT) was introduced in 2002. With the availability of extensive clinical data as well as improved technology, OCT ophthalmic imaging is now rapidly being adopted in ophthalmology. There are currently more than 1,500 StratusOCT systems in use in ophthalmic clinics internationally and OCT technology is rapidly becoming the standard of care in ophthalmology. Biomedical instrumentation represents somewhat of a paradox. Whereas most technologies, such as consumer electronics and personal computers, evolve quickly and have very short product development cycles, the transition of medical technology from research to clinical practice proceeds exceptionally slowly. Although the technology development is challenging and requires patience, from a broader health-care perspective technology affords many important benefits. Diagnostic and therapeutic instrumentation is a capital cost with low recurring cost, patient throughput can be high, techniques for early diagnosis and treatment can significantly improve patient outcome, and significant cost savings can be achieved.
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
Doppler shift (kHz)
pria, muscularis mucosa, submucosa and b a muscularis propria. In contrast, Barrett’s esophagus is associated with the loss of squamous epithelial organization and the formation of glandular structure. The imaging depth of OCT is 2–3 mm, which is less than that of ultrasound. For diseases that originate 50 µm 100 µm from or involve the mucosa, submucosa and muscular layers, however, imaging the micro2X c NFL scopic structure of small lesions is well within GCL the range of OCT. Imaging can also be perIPL INL ELM IS/OS formed in solid tissues or tumors using neeOPL dle-based OCT imaging devices52. These ELM IS/OS ONL devices can image a 4–5 mm diameter cylindrical volume of tissue and can be coupled to RPE 250 µm 500 µm excisional biopsy devices. RPE Imaging neoplastic changes is an active area 1.5 d 2x of research. Studies have been performed to investigate OCT imaging in vitro and estab1.0 lish its correspondence with histology in the 48,53–56 57–59 Systole gastrointestinal , urinary , respira–4 tory60 and female reproductive tracts61,62. 0.5 –3 Endoscopic imaging studies in human Time –2 63,64 patients began more than five years ago –1 0.0 and have included investigations of upper and Diastole 0 Depth Time (s) 0.0 1.0 2.0 3.0 4.0 5.0 lower gastrointestinal pathology50,65–68, comparative studies with ultrasound69 and biliary imaging70,71. High sensitivities and specifici- Figure 5 Developments in OCT technology. (a) Ultrahigh-resolution OCT image of African frog tadpole ties have been reported for the visualization of (Xenopus laevis) with 1-µm axial resolution at 800 nm obtained using a femtosecond laser light source. Barrett’s metaplasia72. OCT imaging has also (b) Optical coherence microscopy (OCM) en face image of human skin in vivo, demonstrating cellular been investigated in the oral cavity73, larynx74 resolution, performed at 800 nm with a 3-µm coherence gate and a transverse resolution of <2 µm. (c) Ultrahigh-resolution image of the normal human retina with 3 µm resolution. Architectural and bladder75 and is promising for the identimorphology can be visualized, including NFL, nerve fiber layer; GCL, ganglion cell layer; IPL, inner fication of tumor margins. plexiform layer; INL, inner nuclear layer; OPL, outer plexiform layer; ONL, outer nuclear layer; ELM, These studies show that changes in archi- external limiting membrane; and IS/OS, junction between photoreceptor inner and outer segments. tectural morphology can be used to identify This example shows the image resolution that is possible in future-generation clinical ophthalmic regions of abnormality; however, with cur- instruments. (d) A sequence of Doppler OCT images of a retinal blood vessel measuring flow dynamics. rent image resolutions, sensitivity and speci- Extracted flow profiles showing parabolic laminar flow at different phases of the cardiac cycle. (reprinted ficity for the detection of high-grade dysplasia by permission of Archives of Opthalmology from ref. 96) (Courtesy of J. Izatt, Duke University.) or cancer is limited. For this reason, OCT will most likely be used in conjunction with excisional biopsy and histopathology to guide excisional biopsy and internal retinal architectural morphology and promising to improve reduce sampling errors. This would improve the sensitivity of the exci- the accuracy and reproducibility of retinal morphometry81,82. To sional biopsy and would allow clinical diagnosis to be made using resolve human cells, the transverse resolution or focused spot size, as histopathology, a well-accepted standard. With future improvements well as the axial resolution, must be improved. Hybrid techniques, such in OCT imaging performance, direct image-based diagnosis may be as optical coherence microscopy (OCM), that combine coherencepossible in selected clinical applications. gated detection of OCT with high-numerical-aperture imaging of confocal microscopy provide improved imaging depth and contrast as New developments in OCT technology compared with confocal microscopy and promise to enable endoscopOCT draws heavily from the rapidly developing technology base in ically based cellular imaging15,53,83. Figure 5b,c shows examples of celphotonics and lasers. OCT technology is an extremely active and lular imaging using OCM. dynamic area of research in its own right. The development of highAnother active area of research is functional OCT imaging. speed real-time imaging was important to reduce motion artifacts and Spectroscopic OCT imaging using broadband light sources enables the enable clinical imaging applications40,76,77. Recently, efforts have been spectrum of the backscattered or back-reflected light from each pixel directed at improving image resolution. Short-pulse laser light sources to be measured84. Because tissues have different optical properties, can generate broadband light and provide a quantum leap in axial spectroscopic imaging allows tissue contrast to be enhanced. In addiimage resolution, from 10–15 µm to 1–5 µm6,78,79. tion, it is possible to perform quantitative imaging of tissue chroFigure 5 provides some examples of ultrahigh-resolution imaging. mophores or functional state. Metabolic indicators such as tissue Cellular-level image resolutions as fine as 1 µm have been demon- hydration85 or oxygenation of hemoglobin86 in vascular tissue can be strated in biological specimens, allowing visualization of the mitotic measured. Other tissue properties, such as birefringence, can also be cycle and tracking cell migration6,80. Ultrahigh resolutions of 3 µm imaged and quantitatively measured87,88. Functional imaging of brain have been achieved in ophthalmic imaging, enabling visualization of activity induced by visual stimulation has also been demonstrated89. Depth (mm)
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE
1365
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE Tissue birefringence can be used to assess effects such as early osteoarthritic changes in cartilage90, burn depth91 or retinal nerve fiber layer thickness92. Doppler OCT has also emerged as a powerful imaging modality93–100. Doppler flow measurements can be performed by measuring the Doppler shift of light scattered from blood. A wide range of Doppler imaging technologies now exist, including high-speed and high-sensitivity techniques. Figure 5d provides an example of Doppler imaging in retinal vasculature showing extracted flow profiles as a function of cardiac cycle96. Doppler OCT imaging is also promising as a method for quantitatively assessing capillary density and angiogenesis. Conclusions OCT is a powerful imaging technology because it enables the real-time, in situ visualization of tissue microstructure without the need to excise and process specimens as in conventional biopsy and histopathology. Image resolutions of 1–15 µm are possible and imaging can be performed using catheters, endoscopes, laparoscopes and needles. Functional imaging based on spectroscopy, tissue birefringence and Doppler flow is possible. The ability to perform ‘optical biopsy,’ visualizing tissue morphology in real time under operator guidance, promises to have numerous research and clinical applications. However, more comprehensive biomedical and clinical studies are required to develop these applications (see Box 1). These studies will be challenging because statistical performance criteria, such as sensitivity and specificity, must be evaluated on an application-by-application basis. Technology development also remains an important area of research in itself. Many applications will depend critically on improvements in imaging performance and functionality as well as the development of new paradigms for imaging. The unique features of OCT imaging suggest that it will find use in many medical applications ranging from the visualization of tissue pathology, where excisional biopsy is impossible, to improvements in the screening and diagnosis of neoplasia. ACKNOWLEDGMENTS The contributions of A. Aguirre, S. Boppart, B. Bouma, S. Bourquin, M. Brezinski, W. Drexler, J. Duker, C. Chudoba, I. Hartl, P. Herz, P. Hsiung, T. Ko, X. Li, H. Mashimo, C. Pitris, J. Schuman, G. Tearney, J. Van Dam and J. Wei are gratefully acknowledged. We thank E. Grube of the Heart Center Siegborg, LightLab Imaging, and J. Izatt of Duke University for granting permission to present the images shown in this paper. This research is supported in part by the US National Institutes of Health, contracts NIH-9-R01-CA75289-05 and NIH-9-R01-EY1128916; the Medical Free Electron Laser Program, contract F49620-01-1-0186; the Air Force Office of Scientific Research, contract F49620-98-01-0084; the US Army Medical Research Material Command, contract DAMD 17-01-1-156; the National Science Foundation, contracts ECS-0119452 and BES-0119494; the Poduska Family Foundation Fund; and the philanthropy of G. Andlinger. COMPETING FINANCIAL INTERESTS The authors declare competing financial interests (see the Nature Biotechnology website for details). Published online at http://www.nature.com/naturebiotechnology/ 1. Huang, D. et al. Optical coherence tomography. Science 254, 1178–1181 (1991). 2. Fujimoto, J.G. et al. Optical biopsy and imaging using optical coherence tomography. Nat. Med. 1, 970–972 (1995). 3. Schmitt, J.M. Optical coherence tomography (OCT): a review. IEEE J. Selected Topics Quantum Electron. 5, 1205–1215 (1999). 4. Brezinski, M.E. & Fujimoto, J.G. Optical coherence tomography: high-resolution imaging in nontransparent tissue. IEEE J. Selected Topics Quantum Electron. 5, 1185–1192 (1999). 5. Fujimoto, J.G., Pitris, C., Boppart, S.A. & Brezinski, M.E. Optical coherence tomography: an emerging technology for biomedical imaging and optical biopsy. Neoplasia 2, 9–25 (2000). 6. Drexler, W. et al. In vivo ultrahigh-resolution optical coherence tomography. Optics Lett. 24, 1221–1223 (1999). 7. Schmitt, J.M., Knuttel, A., Yadlowsky, M. & Eckhaus, M.A. Optical-coherence tomog-
1366
raphy of a dense tissue: statistics of attenuation and backscattering. Physics Med. Biol. 39, 1705–1720 (1994). 8. Duguay, M.A. Light photographed in flight. Am. Sci. 59, 551–556 (1971). 9. Youngquist, R., Carr, S. & Davies, D. Optical coherence-domain reflectometry: a new optical evaluation technique. Optics Lett. 12, 158–160 (1987). 10. Takada, K., Yokohama, I., Chida, K. & Noda, J. New measurement system for fault location in optical waveguide devices based on an interferometric technique. Appl. Optics 26, 1603–1608 (1987). 11. Gilgen, H.H., Novak, R.P., Salathe, R.P., Hodel, W. & Beaud, P. Submillimeter optical reflectometry. IEEE J. Lightwave Technol. 7, 1225–1233 (1989). 12. Fercher, A.F., Mengedoht, K. & Werner, W. Eye-length measurement by interferometry with partially coherent light. Optics Lett. 13, 1867–1869 (1988). 13. Huang, D., Wang, J., Lin, C.P., Puliafito, C.A. & Fujimoto, J.G. Micron-resolution ranging of cornea and anterior chamber by optical reflectometry. Lasers Surgery Med. 11, 419–425 (1991). 14. Swanson, E.A. et al. High-speed optical coherence domain reflectometry. Optics Lett. 17, 151–153 (1992). 15. Izatt, J.A., Hee, M.R., Owen, G.M., Swanson, E.A. & Fujimoto, J.G. Optical coherence microscopy in scattering media. Optics Lett. 19, 590–592 (1994). 16. Podoleanu, A.G. et al. Transversal and longitudinal images from the retina of the living eye using low coherence reflectometry. J. Biomed. Optics 3, 12–20 (1998). 17. Podoleanu, A.G., Rogers, J.A., Jackson, D.A. & Dunne, S. Three-dimensional OCT images from retina and skin. Optics Express 7 (2000). 18. Swanson, E.A. et al. In vivo retinal imaging by optical coherence tomography. Optics Lett. 18, 1864–1866 (1993). 19. Hee, M.R. et al. Optical coherence tomography of the human retina. Arch. Ophthalmol. 113, 325–332 (1995). 20. Puliafito, C.A. et al. Imaging of macular diseases with optical coherence tomography. Ophthalmology 102, 217–229 (1995). 21. Schuman, J.S. et al. Quantification of nerve fiber layer thickness in normal and glaucomatous eyes using optical coherence tomography. Arch. Ophthalmol. 113, 586–596 (1995). 22. Schuman, J.S. et al. Reproducibility of nerve fiber layer thickness measurements using optical coherence tomography. Ophthalmology 103, 1889–1898 (1996). 23. Hee, M.R. et al. Optical coherence tomography of macular holes. Ophthalmology 102, 748–756 (1995). 24. Gaudric, A. et al. Macular hole formation: new data provided by optical coherence tomography. Arch. Ophthalmol. 117, 744–751 (1999). 25. Hee, M.R. et al. Quantitative assessment of macular edema with optical coherence tomography. Arch. Ophthalmol. 113, 1019–1029 (1995). 26. Hee, M.R. et al. Optical coherence tomography of central serous chorioretinopathy. Am. J. Ophthalmol. 120, 65–74 (1995). 27. Hee, M.R. et al. Optical coherence tomography of age-related macular degeneration and choroidal neovascularization. Ophthalmology 103, 1260–1270 (1996). 28. Wilkins, J.R. et al. Characterization of epiretinal membranes using optical coherence tomography. Ophthalmology 103, 2142–2151 (1996). 29. Hee, M.R. et al. Topography of diabetic macular edema with optical coherence tomography. Ophthalmology 105, 360–370 (1998). 30. Lederer, D.E. et al. Analysis of macular volume in normal and glaucomatous eyes using optical coherence tomography. Am. J. Ophthalmol. 135, 838–843 (2003). 31. Zangwill, L.M., Williams, J., Berry, C.C., Knauer, S. & Weinreb, R.N. A comparison of optical coherence tomography and retinal nerve fiber layer photography for detection of nerve fiber layer damage in glaucoma. Ophthalmology 107, 1309–1315 (2000). 32. Zangwill, L.M. et al. Discriminating between normal and glaucomatous eyes using the Heidelberg Retina Tomograph, GDx Nerve Fiber Analyzer, and Optical Coherence Tomograph. Arch. Ophthalmol. 119, 985–993 (2001). 33. Williams, Z.Y. et al. Optical coherence tomography measurement of nerve fiber layer thickness and the likelihood of a visual field defect. Am. J. Ophthalmol. 134, 538–546 (2002). 34. Schuman, J.S. et al. Comparison of optic nerve head measurements obtained by optical coherence tomography and confocal scanning laser ophthalmoscopy. Am. J. Ophthalmol. 135, 504–512 (2003). 35. Falk, E. Plaque rupture with severe pre-existing stenosis precipitating coronary thrombosis, characteristics of coronary atherosclerotic plaques underlying fatal occlusive thrombi. Br. Heart J. 50, 127–134 (1983). 36. Davies, M.J. & Thomas, A.C. Plaque fissuring—the cause of acute myocardial infarction, sudden ischemic death, and crescendo angina. Br. Heart J. 53, 363–373 (1983). 37. Fuster, V.L et al. The pathogenesis of coronary artery disease and the acute coronary syndromes. New Engl. J. Med. 326, 242–249 (1992). 38. Brezinski, M.E. et al. Optical coherence tomography for optical biopsy. Properties and demonstration of vascular pathology. Circulation 93, 1206–1213 (1996). 39. Tearney, G.J. et al. Scanning single-mode fiber optic catheter–endoscope for optical coherence tomography. Optics Lett. 21, 543–545 (1996). 40. Tearney, G.J. et al. In vivo endoscopic optical biopsy with optical coherence tomography. Science 276, 2037–2039 (1997). 41. Tearney, G.J. et al. Images in cardiovascular medicine. Images in cardiovascular medicine. Catheter-based optical imaging of a human coronary artery. Circulation 94, 3013 (1996). 42. Brezinski, M.E. et al. Assessing atherosclerotic plaque morphology: comparison of optical coherence tomography and high frequency intravascular ultrasound. Heart 77, 397–403 (1997). 43. Fujimoto, J.G. et al. High-resolution in vivo intra-arterial imaging with optical coherence tomography. Heart 82, 128–133 (1999).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PERSPECTIVE 44. Tearney, G.J. et al. Porcine coronary imaging in vivo by optical coherence tomography. Acta Cardiol. 55, 233–237 (2000). 45. Jang, I.K., Tearney, G. & Bouma, B. Visualization of tissue prolapse between coronary stent struts by optical coherence tomography: comparison with intravascular ultrasound. Circulation 104, 2754 (2001). 46. Jang, I.K. et al. Visualization of coronary atherosclerotic plaques in patients using optical coherence tomography: comparison with intravascular ultrasound. J. Am. Coll. Cardiol. 39, 604–609 (2002). 47. Grube, E., Gerckens, U., Buellesfeld, L. & Fitzgerald, P.J. Images in cardiovascular medicine. Intracoronary imaging with optical coherence tomography: a new high-resolution technology providing striking visualization in the coronary artery. Circulation 106, 2409–2410 (2002). 48. Pitris, C. et al. Feasibility of optical coherence tomography for high-resolution imaging of human gastrointestinal tract malignancies. J. Gastroenterol. 35, 87–92 (2000). 49. Bouma, B.E. & Tearney, G.J. Power-efficient nonreciprocal interferometer and linearscanning fiber-optic catheter for optical coherence tomography. Optics Lett. 24, 531–533 (1999). 50. Rollins, A.M. et al. Real-time in vivo imaging of human gastrointestinal ultrastructure by use of endoscopic optical coherence tomography with a novel efficient interferometer design. Optics Lett. 24, 1358–1360 (1999). 51. Li, X.D. et al. Optical coherence tomography: advanced technology for the endoscopic imaging of Barrett’s esophagus. Endoscopy 32, 921–930 (2000). 52. Li, X., Chudoba, C., Ko, T., Pitris, C. & Fujimoto, J.G. Imaging needle for optical coherence tomography. Optics Lett. 25, 1520–1522 (2000). 53. Izatt, J.A., Kulkarni, M.D., Wang, H.-W., Kobayashi, K. & Sivak, M.V. Jr. Optical coherence tomography and microscopy in gastrointestinal tissues. IEEE J. Selected Topics Quantum Electron. 2, 1017–1028 (1996). 54. Tearney, G.J. et al. Optical biopsy in human gastrointestinal tissue using optical coherence tomography. Am. J. Gastroenterol. 92, 1800–1804 (1997). 55. Kobayashi, K., Izatt, J.A., Kulkarni, M.D., Willis, J. & Sivak, M.V. Jr. High-resolution cross-sectional imaging of the gastrointestinal tract using optical coherence tomography: preliminary results. Gastrointest. Endosc. 47, 515–523 (1998). 56. Tearney, G.J. et al. Optical biopsy in human pancreatobiliary tissue using optical coherence tomography. Dig. Dis. Sci. 43, 1193–1199 (1998). 57. Tearney, G.J. et al. Optical biopsy in human urologic tissue using optical coherence tomography. J. Urol. 157, 1915–1919 (1997). 58. Jesser, C.A. et al. High resolution imaging of transitional cell carcinoma with optical coherence tomography: feasibility for the evaluation of bladder pathology. Br. J. Radiol. 72, 1170–1176 (1999). 59. D’Amico, A.V., Weinstein, M., Li, X., Richie, J.P. & Fujimoto, J. Optical coherence tomography as a method for identifying benign and malignant microscopic structures in the prostate gland. Urology 55, 783–787 (2000). 60. Pitris, C. et al. High resolution imaging of the upper respiratory tract with optical coherence tomography: a feasibility study. Am. J. Respir. Crit. Care Med. 157(5) Pt 1, 1640–1644 (1998). 61. Pitris, C. et al. High-resolution imaging of gynecologic neoplasms using optical coherence tomography. Obstet. Gynecol. 93, 135–139 (1999). 62. Boppart, S.A. et al. High-resolution imaging of endometriosis and ovarian carcinoma with optical coherence tomography: feasibility for laparoscopic-based imaging. Br. J. Obstet. Gynaecol. 106, 1071–1077 (1999). 63. Sergeev, A.M. et al. In vivo endoscopic OCT imaging of precancer and cancer states of human mucosa. Optics Express [online] 1, 432–440 (1997). 64. Feldchtein, F.I. et al. Endoscopic applications of optical coherence tomography. Optics Express [online] 3, 257–370 (1998). 65. Brand, S. et al. Optical coherence tomography in the gastrointestinal tract. Endoscopy 32, 796–803 (2000). 66. Jäckle, S. et al. In vivo endoscopic optical coherence tomography of the human gastrointestinal tract—toward optical biopsy. Endoscopy 32, 743–749 (2000). 67. Jäckle, S. et al. In vivo endoscopic optical coherence tomography of esophagitis, Barrett’s esophagus, and adenocarcinoma of the esophagus. Endoscopy 32, 750–755 (2000). 68. Sivak, M.V., Jr. et al. High-resolution endoscopic imaging of the GI tract using optical coherence tomography. Gastrointest. Endosc. 51(4) Pt 1, 474–479 (2000). 69. Das, A. et al. High-resolution endoscopic imaging of the GI tract: a comparative study of optical coherence tomography versus high-frequency catheter probe EUS. Gastrointest. Endosc. 54, 219–224 (2001). 70. Seitz, U. et al. First in vivo optical coherence tomography in the human bile duct. Endoscopy 33, 1018–1021 (2001). 71. Poneros, J.M. et al. Optical coherence tomography of the biliary tree during ERCP. Gastrointest. Endosc. 55, 84–88 (2002). 72. Poneros, J.M. et al. Diagnosis of specialized intestinal metaplasia by optical coherence tomography. Gastroenterology 120, 7–12 (2001).
73. Feldchtein, F.I. et al. In vivo OCT imaging of hard and soft tissue of the oral cavity. Optics Express [online] 3, 239–250 (1998). 74. Shakhov, A.V. et al. Optical coherence tomography monitoring for laser surgery of laryngeal carcinoma. J. Surg. Oncol. 77, 253–258 (2001). 75. Zagaynova, E.V. et al. In vivo optical coherence tomography feasibility for bladder disease. J. Urol. 167, 1492–1496 (2002). 76. Tearney, G.J., Bouma, B.E. & Fujimoto, J.G. High-speed phase- and group-delay scanning with a grating-based phase control delay line. Optics Lett. 22, 1811–1813 (1997). 77. Rollins, A.M., Kulkarni, M.D., Yazdanfar, S., Ung-arunyawee, R. & Izatt, J.A. In vivo video rate optical coherence tomography. Optics Express [online] 3, 219–229 (1998). 78. Bouma, B. et al. High-resolution optical coherence tomographic imaging using a mode-locked Ti:Al2/O3 laser source. Optics Lett. 20, 1486–1488 (1995). 79. Bouma, B.E., Tearney, G.J., Bilinsky, I.P., Golubovic, B. & Fujimoto, J.G. Self-phasemodulated Kerr-lens mode-locked Cr:forsterite laser source for optical coherence tomography. Optics Lett. 21, 1839–1841 (1996). 80. Boppart, S.A. et al. In vivo cellular optical coherence tomography imaging. Nat. Med. 4, 861–865 (1998). 81. Drexler, W. et al. Ultrahigh-resolution ophthalmic optical coherence tomography. Nat. Med. 7, 502–507 (2001). 82. Drexler, W. et al. Enhanced visualization of macular pathology with the use of ultrahigh-resolution optical coherence tomography. Arch. Ophthalmol. 121, 695–706 (2003). 83. Aguirre, A.D., Hsiung, P., Ko, T.H., Hartl, I. & Fujimoto, J.G. High-resolution optical coherence microscopy for high-speed, in vivo cellular imaging. Optics Lett. 2064–2066 28 (2003). 84. Morgner, U. et al. Spectroscopic optical coherence tomography. Optics Lett. 25, 111–113 (2000). 85. Schmitt, J.M., Xiang, S.H. & Yung, K.M. Differential absorption imaging with optical coherence tomography. J. Opt. Soc. Am. A 15, 2288–2296 (1998). 86. Faber, D.L., Mik, E.G., Aalders, M.C.G. & van Leeuwen, T.G. Light absorption of (oxy)hemoglobin assessed by spectroscopic optical coherence tomography. Optics Lett. 28, 1436–1438 (2003). 87. De Boer, J.F., Milner, T.E., van Gemert, M.J.C. & Nelson, J.S. Two-dimensional birefringence imaging in biological tissue by polarization-sensitive optical coherence tomography. Optics Lett. 22, 934–936 (1997). 88. de Boer, J.F. & Milner, T.E. Review of polarization sensitive optical coherence tomography and Stokes vector determination. J. Biomed. Optics 7, 359–371 (2002). 89. Maheswari, R.U., Takaoka, H., Kadono, H., Homma, R. & Tanifuji, M. Novel functional imaging technique from brain surface with optical coherence tomography enabling visualization of depth resolved functional structure in vivo. J. Neurosci. Methods 124, 83–92 (2003). 90. Herrmann, J.M. et al. High-resolution imaging of normal and osteoarthritic cartilage with optical coherence tomography. J. Rheumatol. 26, 627–635 (1999). 91. de Boer, J.F., Srinivas, S.M., Malekafzali, A., Chen, Z. & Nelson, J.S. Imaging thermally damaged tissue by polarization sensitive optical coherence tomography. Optics Express [online] 3, 212–218 (1998). 92. Cense, B., Chen, T.C., Hyle Park, B., Pierce, M.C. & de Boer, J.F. In vivo depthresolved birefringence measurements of the human retinal nerve fiber layer by polarization-sensitive optical coherence tomography. Optics Lett. 27, 1610–1612 (2002). 93. Izatt, J.A., Kulkarni, M.D., Yazdanfar, S., Barton, J.K. & Welch, A.J. In vivo bidirectional color Doppler flow imaging of picoliter blood volumes using optical coherence tomography. Optics Lett. 22, 1439–1441 (1997). 94. Westphal, V., Yazdanfar, S., Rollins, A.M. & Izatt, J.A. Real-time, high velocity-resolution color Doppler optical coherence tomography. Optics Lett. 27, 34–36 (2002). 95. Wong, R.C. et al. Visualization of subsurface blood vessels by color Doppler optical coherence tomography in rats: before and after hemostatic therapy. Gastrointest. Endosc. 55, 88–95 (2002). 96. Yazdanfar, S., Rollins, A.M. & Izatt, J.A. In vivo imaging of human retinal flow dynamics by color Doppler optical coherence tomography. Arch. Ophthalmol. 121, 235–239 (2003). 97. Chen, Z., Milner, T.E., Dave, D. & Nelson, J.S. Optical Doppler tomographic imaging of fluid flow velocity in highly scattering media. Optics Lett. 22, 64–66 (1997). 98. Zhao, Y. et al. Phase-resolved optical coherence tomography and optical Doppler tomography for imaging blood flow in human skin with fast scanning speed and high velocity sensitivity. Optics Lett. 25, 114–116 (2000). 99. Ding, Z., Zhao, Y., Ren, H., Nelson, J.S. & Chen, Z. Real-time phase-resolved optical coherence tomography and optical Doppler tomography. Optics Express [online] 10, 236–245 (2002). 100. Ren, H. et al. Phase-resolved functional optical coherence tomography: simultaneous imaging of in situ tissue structure, blood flow velocity, standard deviation, birefringence, and Stokes vectors in human skin. Optics Lett. 27, 1702–1704 (2002).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1367
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
FOCUS ON OPTICAL IMAGING
REVIEW
Nonlinear magic: multiphoton microscopy in the biosciences Warren R Zipfel, Rebecca M Williams & Watt W Webb Multiphoton microscopy (MPM) has found a niche in the world of biological imaging as the best noninvasive means of fluorescence microscopy in tissue explants and living animals. Coupled with transgenic mouse models of disease and ‘smart’ genetically encoded fluorescent indicators, its use is now increasing exponentially. Properly applied, it is capable of measuring calcium transients 500 µm deep in a mouse brain, or quantifying blood flow by imaging shadows of blood cells as they race through capillaries. With the multitude of possibilities afforded by variations of nonlinear optics and localized photochemistry, it is possible to image collagen fibrils directly within tissue through nonlinear scattering, or release caged compounds in subfemtoliter volumes.
MPM is a form of laser-scanning microscopy that uses localized ‘nonlinear’ excitation to excite fluorescence only within a thin raster-scanned plane and nowhere else. Since its first demonstration by our group over a decade ago1, MPM has been applied to a variety of imaging tasks and has now become the technique of choice for fluorescence microscopy in thick tissue and in live animals. Neuroscientists have used it to measure calcium dynamics deep in brain slices2–11 and in live animals12–14 (reviewed in ref. 15), to study neuronal plasticity16 and to monitor neurodegenerative disease models in brain slices17 and in living mice18–21. MPM has proved invaluable to cancer researchers for in vivo studies of angiogenesis22,23 and metastasis24,25, to immunologists for investigating lymphocyte trafficking26–30 and to embryologists for visualizing a day in the life of a developing hamster embryo31. These types of applications define the most important niche for MPM—high-resolution imaging of physiology, morphology and cell-cell interactions in intact tissues or live animals. Although two-photon excited fluorescence is usually the primary signal source in MPM, three-photon excited fluorescence32–37 and second37–45 and third-harmonic generation (SHG, THG)46–48 can also be used for imaging. In fact, SHG imaging was one of the earliest forms of biological nonlinear microscopy, proposed49 and demonstrated38 decades ago. Notably, one of the most complex forms of nonlinear imaging, coherent anti-Stokes Raman scattering (CARS) microscopy, was developed even earlier50. CARS microscopy derives contrast directly from Raman-active vibrational modes within molecules and requires two synchronized pulsed lasers operating at different wavelengths, rather than a single pulsed laser as in two- (or three)-photon and SHG and THG microscopy. Like SHG microscopy, CARS microscopy lay dormant for decades but has recently been markedly improved51,52 with the help of tunable, pulsed lasers in the infrared (IR) wavelength range. School of Applied and Engineering Physics, 212 Clark Hall, Cornell University, Ithaca, New York 14853, USA. Correspondence should be addressed to W.W.W. (
[email protected]). Published online 31 October 2002; doi:10.1038/nbt899
Nonlinear excitation also has ‘nonimaging’ uses in biological research, such as the three-dimensional photolysis of caged molecules in femtoliter volumes16,53–56, diffusion measurements by multiphoton fluorescence correlation spectroscopy (MP-FCS)57,58 and multiphoton fluorescence photobleaching recovery (MP-FPR or MP-FRAP)16,59–61, and detecting bimolecular interactions using multiphoton two-color cross-correlation spectroscopy (MP-FCCS)62. Targeted, localized multiphoton excitation has also been used for transfection of single cells by opening a transient nanoscopic pore in the cell membrane with a parked femtosecond laser63. Precise ablation and cutting is possible on the subcellular level. For example, a small region from a single chromosome from a fixed cell can be excised64, opening the possibility of sub-chromosomal ablation in a living cell to study the effect of knocking out specific regions of a targeted chromosome. Here, we review applications of MPM and provide a simplified, practical view of the optical, technological and photophysical aspects of this type of microscopy that may be foreign to biologists but is fundamental to an understanding of how best to apply this technology. Moore’s law of multiphoton microscopy A search of publications referencing MPM (and its various synonyms) reveals several facts about the integration of this relatively new imaging technology into biological research. Publications involving MPM have increased exponentially over the past decade (Fig. 1) as femtosecond laser sources became robust and commercially available, and as the first commercial multiphoton microscopes were introduced in 1996 by BioRad Microscience (Hemel Hampstead, UK). About half of the total references have been consistently devoted to technique and instrumentation development rather than focused on specific biological questions. This is at least partly attributable to the unexplored potential of nonlinear optical processes for biological research (pointed out as early as 1978; ref. 49). A survey of the instrumentation used in the studies employing MPM for biological research (255 out of 560 total references) indicates that 66% of these studies made use of laboratory-built systems, usually based
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1369
REVIEW
References/year
200 150 100 50
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
0 1990
1992
1994
1996
1998
2000
2002
2004
Year Figure 1 Publications employing, developing or reviewing MPM (from PubMed and ISI). Bar height (white and black) indicates the total number of references for the given year; black bars represent publications focusing on instrumentation development, the remainder being work in which MPM was used to help clarify a specific biological research goal. Gray bar is the estimated number based on twice the 2003 half-year total; dotted line is an exponential fit of the data.
on modified confocal microscopes. The remaining one-third employed commercial systems—27% from BioRad and 3.5% each from Zeiss (Oberkochen, Germany) and Leica (Wetzler, Germany). Furthermore, the majority (∼80%) of the publications are from a small number of research groups (∼12) who have developed the required expertise to use the technique effectively. Taken together, these statistics indicate that MPM is still a specialized technology, used successfully by some, but apparently not yet at the level of routine use characteristic of conventional (single-photon) confocal microscopy. Three-dimensionally localized excitation Early in the development of quantum mechanics, it was shown theoretically by Maria Göppert-Mayer65 that photons of lesser energy together can cause an excitation ‘normally’ produced by the absorption of a single photon of higher energy in a process called multiphoton or two-photon excitation. Two-photon microscopy, as normally practiced, uses the simplest version of her theoretical prediction: two photons of about equal energy (from the same laser) interact with a molecule, producing an excitation equivalent to the absorption of a single photon possessing twice the energy. If the excited molecule is fluorescent, it can emit a single photon of fluorescence as if it were excited by a single higher energy photon (Fig. 2). This event depends on the two photons both interacting with the molecule nearly simultaneously (∼10–16 s), resulting in a quadratic dependence on the light intensity rather than the linear dependence of conventional fluorescence. Multiphoton processes such as twophoton excitation (TPE) are often termed ‘nonlinear’ because the rate at which they occur depends nonlinearly on the intensity. The intensitysquared dependence is the basis of the localized nature of two-photon excitation: doubling the intensity produces four times the fluorescence. In MPM, as in conventional laser-scanning confocal microscopy, a laser is focused and raster-scanned across the sample. The image consists of a matrix of fluorescence intensity measurements made by digitizing the detector signal as the laser sweeps back and forth across the sample. TPE probabilities are extremely small, and focusing increases the local intensity at the focal point. Intensity (I) is the number of photons passing through a unit area per unit time (usually in photons cm–2 s –1), whereas power is energy per second (1 W = 1 J s –1). Because intensity depends on the area, it is greater at the focus than a distance away, whereas the total power is the same everywhere along the beam. To calculate the intensity from measured laser power readings, 1 mW = λ × 5 × 1012 photons s –1 nm–1 can be used (derived from E = hc/λ). For example, 1 mW at 960 nm is ∼5 × 1015 photons s –1.
1370
Dividing this by the area of the beam in the focal plane (for high numerical aperture (NA) ∼10–9 cm2) gives an intensity at the focus of ∼5 × 1024 photons cm–2 s –1. The intensity squared is 25 × 1048 photons2 cm–4 s –2, increasing the TPE probability in the focal plane by ∼107 compared with the unfocused beam. Away from the focal plane, the TPE probability drops off rapidly so that no appreciable fluorescence is emitted (Fig. 2b), and intrinsic three-dimensional resolution is achieved. Focusing alone is still not enough to make two-photon microscopy practical. For example, 1 mW of 960-nm light focused into 10 µM fluorescein generates only ∼20,000 fluorescence photons per second. Further reduced by a total collection efficiency of only a few percent, this intensity would be much less than one fluorescence photon per pixel on a typical laser-scanning microscope with a ∼1-µs pixel dwell time. To generate enough TPE fluorescence for imaging, a pulsed laser is used to increase further the probability that two photons will simultaneously interact with a molecule, while still keeping the average power relatively low. A mode-locked titanium sapphire (Ti:S) laser—the most common laser used in MPM—produces ∼80 million pulses per second, each with pulse duration of ∼100 fs. With a pulsed laser, the two-photon fluorescence depends on the average squared intensity (
) rather than the squared average intensity (2). The average intensity is equal to the number of pulses per second (R) times the integrated instantaneous intensity during a pulse, which yields36,66 = gP 2 / (Rτ)
(1)
where gp is a unitless factor that depends on the temporal laser pulse shape (0.66 for a Gaussian pulse shape), τ is the full-width half-maximum (FWHM) of the pulse and < > denotes the time-averaged value. For 100-fs pulses and R = 80 MHz, the TPE probability is increased by gp/Rτ ≈ 105, and 20,000 emitted photons becomes 2 billion photons per second with a mode-locked laser. This translates into around 100 photons per pixel in the imaginary multiphoton microscope just described. Fluorescence excitation and two-photon action cross-sections The two-photon cross-section (σ2p) is a quantitative measure of the probability of a two-photon absorption. σ2p has units of cm4 s, with 10–50 cm4 s called a Göppert-Mayer or ‘GM’. Because it is difficult to measure σ2p directly, the two-photon ‘action’ cross-section is usually measured; this is the product of the fluorescence quantum yield (φF) and the absolute two-photon absorption cross-section (σ2p)36,66,67. Both the wavelength dependence and the absolute values of φFσ2p are important
a
488 nm
b
960 nm
Focal plane
Figure 2 Localization of excitation by two-photon excitation. (a) Single-photon excitation of fluorescein by focused 488-nm light (0.16 NA). (b) Two-photon excitation using focused (0.16 NA) femtosecond pulses of 960-nm light.
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
The optical and effective resolution of MPM The optical resolution of a multiphoton microscope seems at first thought to be worse by a factor of 2 than a single-photon confocal microscope, because the illumination used is about twice the wavelength. However, this assessment is based on comparison to a hypothetical perfect confocal microscope that has an infinitely small pinhole94. As the pinhole is opened, the resolution of a confocal microscope decreases and the resolution difference becomes less. In practice, the effective resolution achieved is a function of many complex factors, such as the absolute number of photons collected per pixel (pixel noise scales as the square root of the number of photons), and, in a confocal microscope, the fraction of true signal photons relative to scattered photons from outside the observation volume (that is, contrast). Because fluorescence only arises from the focus in MPM, when compared with confocal in scattering samples the effective resolution of the former often seems far superior95–97. Knowing the dimensions of the two-photon focal volume is useful, for example, to estimate the thickness of an optical section or to calculate the number of caged neurotransmitters one might expect to photolyse per laser pulse. The illumination point spread function, IPSF(x,y,z), describes intensity everywhere in space near the focus, and in MPM only IPSF2 is needed to define the true optical resolution98 (assuming confocal detection is not used). IPSF2 can be calculated (Fig. 4a) based on the work of Richards and Wolf99, and fits of the lateral and axial intensitysquared profiles to a Gaussian function (Fig. 4b) yield expressions for estimating the diffraction-limited lateral (ωxy) and axial (ωz) 1/e radii of IPSF2 (Fig. 4c). For NAs 0.7, ωxy is inversely proportional to NA; however, at higher NAs, this dependence deviates, and a better estimate
a
700
750
800
850
900
950
1,000
350
375
400
425
450
475
500
700
750
800
850
900
950
1,000
700
750
800
850
900
950
1,000
Arbitrary units
1.0 0.8 0.6 0.4 0.2 0.0
b 2p (GM)
200
φF σ
in MPM. To determine an optimal TPE wavelength, doubling the maximum single-photon excitation wavelength is often a good approximation. However, many molecules exhibit obvious deviations from this rule because the selection rules for two-photon processes are different from single-photon selection rules. For symmetrical molecules, initial single-photon excited states are two-photon forbidden68, and therefore fluorophores such as Rhodamine B67 show a clear difference between their one- and two-photon absorption spectra, as shown in Figure 3a. Intrinsic molecules such as NADH have extremely low action crosssections (<10–4 GM)36,37,69 yet have been successfully used with MPM imaging70–73. The other extreme, CdSe-ZnS quantum dots, have crosssections approaching 50,000 GM74 and allow multiphoton imaging with a few microwatts of laser power. Most common fluorescent dyes have φFσ2p values in the range of 1–300 GM36,67, although it is possible to design organic molecules specifically for high nonlinear absorption75–77. Intrinsically fluorescent proteins, such as green fluorescent protein (GFP), have large action cross-sections (Fig. 3b and refs. 36, 78–81), and are particularly well suited for MPM in tissue explants and live animals24,82–88. Although relative two-photon excitation spectra (λ dependence)89,90 can be useful, it is often necessary to know φFσ2p(λ) explicitly when designing a particular MPM experiment. For example, cyan fluorescent protein (CFP) and yellow fluorescent protein (YFP) can be used for multiphoton fluorescence resonance energy transfer (MP-FRET)91–93 but also have potential for MP-FCCS62. For MP-FRET, CFP must be excited without exciting YFP; for MP-FCCS, by contrast, it is optimal to excite both proteins equally. The relative excitation spectra alone do not provide the needed information, and an absolute measure of the expected fluorescence from each species at a given wavelength is necessary. From Figure 3c, it is clear that MP-FRET is best carried out at 800 nm, whereas MP-FCCS would work best where the fluorescence from both species is about equal (∼910 nm).
150 100 50 0
c
10 3
σ2p YFP/σ2p CFP
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW
10 2 10 1 10 0 10 –1 10 –2
Wavelength (nm) Figure 3 Two-photon action cross-sections. (a) Two-photon action spectrum of Rhodamine B (black) compared to the one-photon absorption spectrum (red). (b) Two-photon action spectra of five common fluorescent proteins (FPs): enhanced GFP (eGFP; green), CFP (cyan), YFP (yellow), Discosoma Red (dsRed; red) and wild type GFP (wtGFP; black). Note: our measurements of the FP cross-sections were normalized to the concentration of fluorescing protein as determined by FCS, and therefore our absolute values (but not line shape) for CFP and YFP differ from refs. 79 and 81. (c) Semilog plot of the cross-section ratio of YFP to CFP.
(∼1% error) can be found by assuming a slight inverse power dependence of NA. Analogous to the nonparaxial derivations of ref. 100, an expression for ωz of the two-photon focal volume can also be formulated (Fig. 4c). Conversions to the FWHM and 1/e2 radius (Fig. 4b) can be obtained by multiplication by 2√ln2 and √2, respectively. The FWHM is a more common measure of optical resolution, whereas the 1/e2 radius is, for example, needed to recover diffusion coefficients from MPFCS and MP-FPR measurements. Note that the objective lenses must be uniformly illuminated (overfilled) to achieve a diffraction-limited focus—a condition well approximated if, in practice, the 1/e beam diameter is no less than the diameter of the objective lens back aperture. By approximating the IPSF2 as a three-dimensional Gaussian volume, analytical integration over all space yields the TPE focal volume. This is not a volume with distinct walls but rather one based on averaging the TPE potential over all space. Integrating the three-dimensional Gaussian yields VTPE = π ⁄2 ω2xy ωz 3
(2)
where ωxy and ωz are calculated as in Figure 4c. Although these expressions provide a good estimate of the central lobe of IPSF2, they do not fully take the wings of IPSF2 into account at regions far from the focus and the integrated Gaussian volume yields a value 68% of that obtained from numerical integration of the full vectorial approach of ref. 99. With
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1371
REVIEW a I(x,y,z)
focus yield an effective TPE volume 40 times larger than the nonsaturated focal volume.
b
I(x,y,z)2
FWHM
ω
1/e2 x
c
d
0.320 λ 2 NA NA ≤ 0.7 ωxy= 0.325 λ NA > 0.7 2 NA0.91 Relative increase in effective two-photon excitation volume
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
z
ωz = 0.532 λ 2
1
n − n − NA 2
2
3.0
300 GM 2.5 2.0 1.5
1 GM
1.0 0
5
10
15
20
Power (mW)
Figure 4 The two-photon excitation volume. (a) Axial and lateral views of IPSF and IPSF2. Squaring the IPSF results in minimal wings relative to the center. (b) Axial profile (0,0,z) of IPSF (dashed red line) and ISPF2(0,0,z) (solid red). Dashed black line is a fit to a Gaussian function. (c) Equations for calculating the 1/e widths (ω) of the lateral (xy) and axial (z) intensitysquared profiles. (d) Excitation probability saturation of the TPE focal volume for a 1-GM and 300-GM fluorophore calculated at 1.2 NA for a 200-fs, 80-MHz source. Insert shows the lateral plane of the effective IPSF2 at the focus for 1 GM (left) and 300 GM (right) at 20 mW.
this correction, equation (2) provides a good estimate of the TPE volume. For example, the effective TPE volume for a 1.2-NA lens at 900 nm is (5.57) (0.175 µm)2 (0.451 µm) / 0.68 = 0.113 µm3 or ∼100 attoliters. In using a Gaussian approximation for the focal volume, it is assumed that the laser intensity is far from a level that would cause fluorophore excitation saturation. Assuming the lifetime of a fluorophore is less than the time between laser pulses, the TPE probability per laser pulse per fluorophore is 1 exp(ασ2PP2IPSF2(x,y,z)/(R2 τ)), where P is the laser power and α is a conversion constant1,74. Depending on the laser power and the value of σ2P, this probability can saturate near the focal center (i.e., become 1.0), while continuing to increase in the wings of the focal volume, resulting in a marked deviation from the excitation volume predicted by equation (2). Figure 4d shows calculations carried out for molecules with cross-sections of 1 and 300 GM for laser powers up to 20 mW (R = 80 MHz and 1.2 NA), indicating that with a relatively high cross-section dye the optical resolution may begin to deviate from the optimal possible value. This effect was clearly seen with quantum dots74, which have extremely high TPE cross sections. It also may become significant in cases where the peak laser intensity is unusually high, such as in applications of high-pulse power regenerative amplifier systems101, 102 to increase imaging depth. Similar calculations carried out with R = 200 KHz for only 0.5 mW of power at the
1372
Lasers and laser optics for MPM The application of mode-locked Ti:S crystal-based lasers to MPM, first demonstrated in 1992 (ref. 103), was really the beginning of practical MPM. Before that, MPM involved femtosecond dye lasers, which were temperamental and required constant tweaking. Commercial Ti:S lasers from Spectra-Physics (Mountain View, CA, USA) and Coherent (Sunnyvale, CA), now with broadband optics permitting use of the full tunable range (∼700 to 1,000 nm) without changing laser mirrors, have made MPM an accessible tool for biology. The recent introduction of computer-controlled, single-box Ti:S lasers from both companies continues this trend. Although other types of mode-locked lasers have been applied31,95,104,105, Ti:S lasers are presently the most common and robust excitation sources for MPM. The term ‘mode-locked’ refers to a laser operating with only a certain set of frequencies (modes) propagating in the laser cavity, with the phase between these modes locked so that there is destructive interference between the propagating frequencies everywhere in the cavity except at one point where the waves add constructively106. This results in a single short pulse of light traveling in the cavity with the repetition rate dictated by the distance between the two cavity end mirrors and the speed of light (Fig. 5a,b). Femtosecond Ti:S lasers require a relatively large number of intracavity frequencies to achieve 100-fs pulses, and therefore the pulses have a significantly large spectral bandwidth. The laser spectrum is simpler to measure than the pulsewidth, so it is commonly monitored in a MPM system as an indication of proper mode-locking. The spectrum should be a symmetrical Gaussian shape (Fig. 5c) devoid of spikes that indicate an unwanted continuous-wave component (termed ‘CW breakthrough’). The product of the pulsewidth and the spectral width (bandwidth) is called the time-bandwidth product (TBWP), which has a defined minimum value (the actual value depends on the temporal pulse shape). If the TBWP is equal to this minimum value, the pulses are ‘transform-limited’ and, assuming a Gaussian pulse shape, the temporal and spectral widths are related by τ = 0.44λ 02/c λFWHM, where λ0 is the peak of the spectrum (that is, 960 nm in Fig. 5c) and c = 300 nm/fs (speed of light). In a transform-limited pulse the wavelengths are randomized so any available color is equally probable anytime during the pulse. However, if this pulse passes through a dispersive material such as glass, it becomes ‘chirped’; the longer wavelength components travel faster so the pulsewidth lengthens, but the spectrum remains unchanged. In the visible and near-IR region, all materials have positive dispersion (red leads blue), so that femtosecond pulses passing through optics are always ‘positively chirped’ and thus longer than they were directly out of the laser. Positive dispersion can be offset by adding negative dispersion before the beam travels through the optics. This is known as dispersion compensation, or pre-chirping, and has been applied to MPM107,108 but leads to an instrument that is more complicated to operate. From equation (1) it can be shown that the increase in average power needed to maintain the same TPE with the longer pulse scales as the square root of the ratio of the chirped pulsewidth to the pulsewidth before the dispersive optics. For example, 2-fold increases in pulsewidth require 1.41-fold more laser power to achieve the same level of fluorescence. Pulse broadening resulting from dispersion can be estimated using τout = τin (1 + 7.68(D/τ2in)2) ⁄2 1
(3)
where D is the total dispersion in femtoseconds squared. A typical MPM system can have between 3,000 and 20,000 fs2 of dispersion in
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
REVIEW a
Laser pulse train
d EOM
~12 ns
Time (ns)
e
If you can read this, you have really good eyes.
Telescope
EOM driver
Beam scanner
Spectral analyzer
Detectors
–200 –100
0
100 200
Ti:S laser
Tube lens
Time (fs)
Filters
f
c 960
Scattering
980
GaAsP
40
QE (%)
Bandwidth
940
No scatter
GaAsP PMT
Microscope
Pump laser
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
b
30
"Meshless" MA
BA
20
GaAs
MA 10 0
Wavelength (nm)
300
400
500
600
700
800
900
Wavelength (nm) Figure 5 Components of a multiphoton microscope. (a) Pulse train from a laser mode-locked Ti:S at 80 MHz. (b) Pulses out of the laser typically have a FWHM duration of 100 fs and a spectral FWHM (c) of ∼10 nm. (d) Schematic of a MPM system (see text). (e) Rays generated at the focus deep in a specimen may be scattered and enter the objective lens as skew rays, resulting in diverging epifluorescence. (f) Comparison of PMT photocathode efficiencies.
total, depending on the operating wavelength, objective lens used109–111 and other components such as beam modulators and beam-expanding lenses. Laser pulsewidths vary as well, typically ranging from 80 to 150 fs, depending on the manufacturer and wavelength. At 5,000 fs2, this translates into pulsewidths of 190 and 177 fs, respectively, with both values being well within the range for efficient TPE. For a more dispersive system at 20,000 fs2, the respective pulsewidths become 697 fs and 403 fs, requiring 50–80% more average power to have the same TPE efficiency as the ∼180-fs pulses available at the focus of the less dispersive system. The complete multiphoton microscope Most multiphoton microscopes, including all commercial versions, essentially consist of a laser-scanning confocal adapted to reflect the near IR, a pulsed laser and a few required multiphoton peripherals. Figure 5d diagrams the components of a functional MPM system, similar to instruments reported in the literature112,113. Ti:S laser. A choice exists between a pair of two separate lasers (Nd:YVO4 pump laser and Ti:S laser) or a ‘single-box’ version, which has both the pump laser and Ti:S oscillator in the same factory-sealed box. The two-laser combination presently has a broader tuning range than a single-box laser (as much as 690–1,020 nm compared with 720–920 nm) and the ability to stop mode-locking, which is sometimes useful to verify that the signal is actually due to two-photon fluorescence. Laser power available varies depending on the size of the pump laser, with 5 W pumped systems providing up to 1 W at the Ti:S peak wavelengths (∼800 nm) and a few hundred milliwatts near the edges of the tuning curve (700 nm and 1,000 nm). Systems with a 10 W pump source produce ∼50% more power across the wavelength range. For maximum flexibility and tuning range, the pump laser–Ti:S pair is as yet unsurpassed; however, simplicity of operation makes single-box lasers ideal for many situations, such as imaging facilities at which user friendliness is a priority.
Beam intensity control. There are several choices for controlling the laser intensity: a collection of neutral-density (ND) filters, a rotatable polarizer, an electro-optic modulator (EOM or Pockels cell) or an acousto-optic modulator (AOM). The two modulators have the ability to blank the beam during scanner turnaround and flyback; however, EOMs are typically less dispersive. For example, an EOM with a 50- to 80-mm-long KD*P crystal has between 2,000 and 4,000 fs2 of dispersion, whereas an AOM made of TeO2 can be 4-fold more dispersive114. Beam telescope. A beam telescope can be used to adjust the size of the beam at the back aperture of the objective to be sure the lens is ‘overfilled’ for a diffraction-limited focus, or in some cases, underfilled for an axially extended focus. Note that in the position shown in Figure 5d, the range of useable beam diameters is ultimately limited by the size of the XY scanner mirrors, which may be only a few millimeters in diameter. With the availability of the new high-NA, low-magnification objectives, such as the Olympus 20 0.95 NA water immersion lens, which has a 17-mm back aperture, a better solution may be necessary. For example, beam expansion after the scanner is more optically complex to incorporate, but could better provide the range of beam sizes needed to optimize the focus. Beam scanner. Of the various XY scanner designs available115, the most common is the nonresonant point scanner, which scans the focused beam across the specimen with an adjustable scan speed, permitting software ‘zooming’ (variable apparent magnification by scanning a smaller region, more slowly) and the important ability to rotate the scan axis. Resonant galvanometers capable of faster (video) scan rates116,117 can also be used, but these operate at only one frequency and cannot zoom, pan or rotate. Also useful is the ability to park the beam stably at any specified XY position, allowing point measurements such as MP-FCS58 and MP-FPR measurements59 using the same optics for both imaging and single-point measurements. This is critical for intracellular measurements60 so that the measurement location is accurately known.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1373
REVIEW a
b
c
d Voltage pulse
5 µm
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
5 µm
e
f
g h
i
∆x ∆t 50 µm
50 µm
5 µm
Figure 6 Applications showing various capabilities of MPM. (a) Simultaneous 780-nm excitation of three different fluorophores in RBL-2H3 cells labeled with 4’,6-diamidino-2-phenylindole (DAPI) (DNA, blue pseudo color), PATMAN (plasma membrane, green) and tetramethylrhodamine (mitochondria, red). (b) Cell stimulation-induced granule-granule fusion142, as assessed by PATMAN membrane staining (gray scale and blue) and acridine orange-stained granules (green and red). (c) and (d) Calcium dynamics 300 µm deep into the neuropil of the lobster stomatogastric ganglion143. (c) Processes of a ‘PD’ neuron filled with calcium green-1N. (d) A line scan shows that upon depolarization (pulse), in a dopamine-dependent manner, calcium enters through varicosities on the neurites, copyright 2000 (reprinted by permission of the Society for Neuroscience, ©2000) (e) Measuring blood flow in a vessel plexus under the intact tibial growth plate perichondrium in the mouse. Red blood cells appear as shadows within the fluorescence-containing blood vessels (methodology from ref. 144, and work carried out in collaboration with C. Farnum, Cornell University, and W. Horton, Shriner’s Hospital, Portland, Oregon). (f) The angle of the shadow traces in the line scan can be translated to the blood flow velocity. (g) Two-dimensional image autocorrelation is used to determine the average angle, in this case yielding an average blood flow of ∆x/∆t = 226 µm s–1. (h) Intrinsic emissions (yellow) detailing the tissue structure of a mouse’s ovary can be imaged simultaneously with genetically incorporated GFPs. Ovarian surface epithelial cells were labeled by an intrabursal injection of virally incorporated eGFP. (Work done in collaboration with A. Flesken-Nikitin and A. Nikitin, Cornell University.) (i) Because collagen SHG is coherent, information about the specimen can be derived from the directionality of the SHG emission. In this 10-day-old rat tail tendon, immature fibril segments scatter backward (green), whereas mature fibrils scatter forward (red).
Detectors. The most efficient fluorescence collection scheme is obtained by the use of ‘non-descanned’ or direct detectors, because confocal detection is not needed with multiphoton excitation. Although it has been shown that a confocal aperture can improve the resolution of a two-photon microscope under certain conditions94,118, for applications in which MPM is most advantageous (such as deep imaging in scattering specimens) a confocal pinhole will degrade performance because scattered emission photons will be rejected, even though they originated in the focal volume. In fact, highly scattered emissions may even be randomly divergent leaving the objective lens (Fig. 5e) and thus difficult to focus101,119. For this reason, a particularly efficient detector design involves the use of a large-area photomultiplier tubes (PMT) close to the objective lens. PMTs are the dominant detectors for both confocal and MPM because an imaging detector is not needed for point-scanning systems, and the high gain and absence of readout noise favors PMTs (pixel integration times are usually in the microsecond range). Charge-coupled detectors (CCDs) have not found widespread use in MPM, other than in situations requiring imaging detectors120, but are rapidly improving with onchip avalanche amplification to reduce the effect of readout noise. Avalanche photodiodes (APDs) have also been used in MPM and may be superior to conventional PMTs at extremely low fluorescence levels121. A recent improvement in detectors for MPM are GaAsP photocathode PMTs (Hamamatsu H7422P), which offer high quantum efficiency
1374
(QE) values in the important 400- to 650-nm range compared with PMTs using conventional photocathode materials (Fig. 5f). Peaking at ∼42% QE, GaAsP detectors are close to the theoretical 50% maximum QE of a photocathode-based device, making them well suited for two-photon FCS, as well as multiphoton imaging. They also have a small transient-time spread (∼150 ps) and a 1-ns output pulse, making them attractive for fluorescence lifetime imaging122. Two minor limitations are a relatively small active area of 5 mm diameter (this is, however, still much larger than an APD active area), making focusing optics a requirement, and a lower damage threshold than conventional photocathodes. Overall, they are a significant advancement in MPM detectors because higher detection efficiency can translate to lower excitation power and improved viability with live specimens. For wavelengths beyond 700 nm, GaAs PMTs can be used, in addition to enhanced ‘meshless’ multialkali PMTs (such as the Hamamatsu H7732-10), which have a relatively high QE, even above 650 nm (Fig. 5f).
Applications of MPM The advantages of MPM, for the most part, arise from two basic attributes of the nonlinear excitation: localized excitation and the expanded wavelength accessibility of most fluorophores. The restriction of multiphoton excitation to the focal plane completely alleviates out-of-focus photobleaching and photodamage. All photons generated are signal; there is no background, so that emission collection can be both simple and efficient. The lack of out-of-focus fluorescence, coupled with the use of IR light, explains the technique’s successful use for fluorescence imaging in thick specimens. The second aspect, enhanced UV bands under TPE (Fig. 3a), simplifies multicolor imaging by allowing excitation of different fluorophores with the same laser, avoiding chromatic aberrations and providing a broad uninterrupted emission collection bandwidth. Figure 6 shows a panorama of applications that highlight the strengths of MPM, including the simultaneous excitation of fluorophores whose emission spectra vary by hundreds of nanometers (Fig 6a,b), deep imaging in live preparations (Fig. 6c,d) and live mice (Fig. 6e–h), as well as the use of intrinsic fluorophores and other nonlinear signals, such as SHG (Fig. 6h,i). In addition to multiphoton fluorescence, an MPM system can easily be modified to collect light from nonlinear scattering processes, such as SHG. In harmonic generation, multiple photons simultaneously interact with non-centrosymmetrical structures without absorption, producing radiation at exactly half of the exciting wavelength (Fig. 7a). Because harmonic generation is a coherent process (scattered photons maintain phase information), the scattered beam must satisfy phasematching constraints producing highly directed radiation rather than isotropic emission (Fig. 7b,c). SHG directionality is dependent on the distribution and directionality of the induced dipoles within the focal volume40,123. SHG imaging is presently being investigated to image membrane potential sensing dyes with improved signal-to-
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
REVIEW a
SHG
Fluorescence
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
b
Illumination
c SHG Fluorescence Scatterer extent along optic axis
Figure 7 Two-photon fluorescence and SHG. (a) SHG is always at half the fundamental wavelength and tunes with laser, whereas fluorescence emission is unchanged, except in its magnitude as the laser tunes through the TPE absorption peak. (b,c) SHG is directional depending on the distribution and orientation of the nonlinear dipoles (b), whereas fluorescence is, in general, isotropically emitted (c).
noise39,124–126, to visualize microtubule polarity42,44,127 and to obtain high-resolution images of unstained collagen structures in mouse models of disease24,37,127. Viability, photodamage and photobleaching Although multiphoton excitation limits photodamage to the focal plane, the possibility of bleaching and damage within this region remains. Photodamage and decreased viability may be especially pronounced at the shorter Ti:S wavelengths where TPE of intrinsic tissue chromophores37 is high. Clear enhancements in cellular viability with MPM over single-photon confocal microscopy are easiest to obtain at wavelengths >800 nm31,128. It is generally thought that overall in the 700- to 1,000-nm range, single- and three-photon excitation damage is usually negligible, and photodamage in the focal plane is primarily due to two-photon processes129, implying that pulsewidth is usually not a critical parameter in viability. This finding is contradicted by work that suggests highly nonlinear dependence of phototoxicity130; however, the excitation doses (squared power × dwell time) used in the work were nearly 1,000-fold higher than levels normally used in MPM. Probe photobleaching, like cellular photodamage, can only occur in the focal plane, but there is evidence that some molecules bleach more easily under TPE131 and may even show higher order (>2) photobleaching132. Nonetheless, reduced damage outside of the focal volume can have tremendous advantages in optically thick specimens. The key to using MPM safely and successfully is not any different from that for any other form of microscopy; one must understand the effects and limits on the specific biological system under investigation, and always optimize to reduce the excitation intensity. In MPM, this means using efficient fluorescence detectors and blanking the laser while image data are not being acquired. Conclusions and future areas of development MPM is quickly becoming a standard tool for determining the molecular mechanisms of cell-based processes in basic biological research, tissue engineering and transgenic mouse models of disease and development. The number of publications focused on nonlinear microscopy development indicates that MPM has created its own field of research. Areas of development that hold particular promise range from fabrication of minimally dispersive objective lenses designed to optimize collection of scattered emissions, to the application of adaptive
optics133–135 to MPM for correcting aberrations of the point spread function. As presently implemented, MPM can image hundreds of microns deep. We believe the ultimate depth limitation is not often a result of a lack of laser power but rather difficulty in collecting the generated fluorescence due to both absorption and scattering, which leads to collection losses (Fig. 5e), as well as reduced fluorescence due to degradation of the IPSF2. A simple test for inadequate two-photon excitation is to measure the power dependence of the fluorescence deep within the specimen to test for saturation. If it scales as less than the power squared, saturation is occurring, indicating that more than sufficient excitation exists in the focal plane. In practice, we often find blurriness and a significant reduction in contrast in many specimens when the imaging depth is increased past several hundred microns, well before all signal is lost. Aberrations of IPSF2 can be caused by either heterogeneity in the index of refraction in tissue or TPE focal volume saturation (see Fig. 4d). It may be possible to overcome the former difficulty by adaptive optics, in which the spatial phase of the beam is modified to pre-compensate for the path the rays take through the tissue. Curing the latter problem requires lower power, compensated by better detection. Nonlinear optics, the ‘magical’ area of modern physics that helped spawn MPM, still has more to offer. Femtosecond lasers are being devised that can operate in the 1,000- to 1,300-nm range, just beyond that which a Ti:S laser can conveniently reach136,137. This region is especially important for high-viability imaging of redder dyes and fluorescent proteins. Several groups are developing forms of multiphoton endoscopy138,139, and new photonic crystal fibers now allow fiber delivery of 100-fs pulses through optical fibers with more than enough power for MPM and multiphoton endoscopy140. Finally, the trick of modifying nonlinear optical responses by changing the phase of the spectral components that make up a femtosecond pulse—a field known as ‘coherent control’—may make it possible to fine-tune multiphoton excitation141 to increase multiphoton absorption, reduce two-photon photobleaching or improve photo-uncaging efficiency. Since its introduction a little over a decade ago, MPM has evolved from a photonic novelty to a tool for visualizing cellular and subcellular events within living tissue. As we focus on understanding the physiological and developmental consequences of the multitude of new genetic sequences uncovered, but not understood, MPM will surely continue to play an important part. COMPETING INTERESTS STATEMENT The authors declare competing financial interests (see the Nature Biotechnology website for details). Published online at http://www.nature.com/naturebiotechnology/ 1. Denk, W., Strickler, J.H. & Webb, W.W. Two-photon laser scanning fluorescence microscopy. Science 248, 73–76 (1990). 2. Yuste, R. & Denk, W. Dendritic spines as basic functional units of neuronal integration. Nature 375, 682–684 (1995). 3. Mainen, Z.F., Malinow, R. & Svoboda, K. Synaptic calcium transients in single spines indicate that NMDA receptors are not saturated. Nature 399, 151–155 (1999). 4. Rose, C.R., Kovalchuk, Y., Eilers, J. & Konnerth, A. Two-photon Na+ imaging in spines and fine dendrites of central neurons. Pflugers Arch. 439, 201–207 (1999). 5. Tan, Y.P. & Llano, I. Modulation by K+ channels of action potential-evoked intracellular Ca2+ concentration rises in rat cerebellar basket cell axons. J. Physiol. 520 Pt 1, 65–78 (1999). 6. Cox, C.L., Denk, W., Tank, D.W. & Svoboda, K. Action potentials reliably invade axonal arbors of rat neocortical neurons. Proc. Natl. Acad. Sci. USA 97, 9724–9728 (2000). 7. Majewska, A., Tashiro, A. & Yuste, R. Regulation of spine calcium dynamics by rapid spine motility. J. Neurosci. 20, 8262–8268 (2000). 8. Oertner, T.G. Functional imaging of single synapses in brain slices. Exp. Physiol. 87, 733–736 (2002). 9. Frick, A., Magee, J., Koester, H.J., Migliore, M. & Johnston, D. Normalization of Ca2+ signals by small oblique dendrites of CA1 pyramidal neurons. J. Neurosci. 23, 3243–3250 (2003). 10. Lendvai, B., Zelles, T., Rozsa, B. & Vizi, E.S. A vinca alkaloid enhances morphological
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1375
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW dynamics of dendritic spines of neocortical layer 2/3 pyramidal cells. Brain Res. Bull. 59, 257–260 (2003). 11. Sabatini, B.L. & Svoboda, K. Analysis of calcium channels in single spines using optical fluctuation analysis. Nature 408, 589–593 (2000). 12. Svoboda, K., Denk, W., Kleinfeld, D. & Tank, D.W. In vivo dendritic calcium dynamics in neocortical pyramidal neurons. Nature 385, 161–165 (1997). 13. Helmchen, F., Svoboda, K., Denk, W. & Tank, D.W. In vivo dendritic calcium dynamics in deep-layer cortical pyramidal neurons. Nat. Neurosci. 2, 989–996 (1999). 14. Stosiek, C., Garaschuk, O., Holthoff, K. & Konnerth, A. In vivo two-photon calcium imaging of neuronal networks. Proc. Natl. Acad. Sci. USA 100, 7319–7324 (2003). 15. Helmchen, F. & Waters, J. Ca(2+) imaging in the mammalian brain in vivo. Eur. J. Pharmacol. 447, 119–129 (2002). 16. Svoboda, K., Tank, D.W. & Denk, W. Direct measurement of coupling between dendritic spines and shafts. Science 272, 716–719 (1996). 17. Ladewig, T. et al. Spatial profiles of store-dependent calcium release in motoneurones of the nucleus hypoglossus from newborn mouse. J. Physiol. 547, 775–787 (2003). 18. Christie, R.H. et al. Growth arrest of individual senile plaques in a model of Alzheimer’s disease observed by in vivo multiphoton microscopy. J. Neurosci. 21, 858–864 (2001). 19. Bacskai, B.J. et al. Non-Fc-mediated mechanisms are involved in clearance of amyloidbeta in vivo by immunotherapy. J. Neurosci. 22, 7873–7878 (2002). 20. D’Amore, J.D. et al. In vivo multiphoton imaging of a transgenic mouse model of Alzheimer disease reveals marked thioflavine-S-associated alterations in neurite trajectories. J. Neuropathol. Exp. Neurol. 62, 137–145 (2003). 21. Bacskai, B.J. et al. Imaging of amyloid-beta deposits in brains of living mice permits direct observation of clearance of plaques with immunotherapy. Nat. Med. 7, 369–372 (2001). 22. Brown, E.B. et al. In vivo measurement of gene expression, angiogenesis and physiological function in tumors using multiphoton laser scanning microscopy. Nat. Med. 7, 864–868 (2001). 23. McDonald, D.M. & Choyke, P.L. Imaging of angiogenesis: from microscope to clinic. Nat. Med. 9, 713–725 (2003). 24. Wang, W. et al. Single cell behavior in metastatic primary mammary tumors correlated with gene expression patterns revealed by molecular profiling. Cancer Res. 62, 6278–6288 (2002). 25. Wolf, K. et al. Compensation mechanism in tumor cell migration: mesenchymal-amoeboid transition after blocking of pericellular proteolysis. J. Cell Biol. 160, 267–277 (2003). 26. Cahalan, M.D., Parker, I., Wei, S.H. & Miller, M.J. Two-photon tissue imaging: seeing the immune system in a fresh light. Nat. Rev. Immunol. 2, 872–880 (2002). 27. Miller, M.J., Wei, S.H., Parker, I. & Cahalan, M.D. Two-photon imaging of lymphocyte motility and antigen response in intact lymph node. Science 296, 1869–1873 (2002). 28. Wei, S.H., Miller, M.J., Cahalan, M.D. & Parker, I. Two-photon imaging in intact lymphoid tissue. Adv. Exp. Med. Biol. 512, 203–208 (2002). 29. Miller, M.J., Wei, S.H., Cahalan, M.D. & Parker, I. Autonomous T cell trafficking examined in vivo with intravital two-photon microscopy. Proc. Natl. Acad. Sci. USA 100, 2604–2609 (2003). 30. Acuto, O. T cell–dendritic cell interaction in vivo: random encounters favor development of long-lasting ties. Science STKE 2003, PE28 (2003). 31. Squirrell, J.M., Wokosin, D.L., White, J.G. & Bavister, B.D. Long-term two-photon fluorescence imaging of mammalian embryos without compromising viability. Nat. Biotechnol. 17, 763–767 (1999). 32. Gryczynski, I., Szmacinski, H. & Lakowicz, J.R. On the possibility of calcium imaging using Indo-1 with three-photon excitation. Photochem. Photobiol. 62, 804–808 (1995). 33. Lakowicz, J.R. et al. Time-resolved fluorescence spectroscopy and imaging of DNA labeled with DAPI and Hoechst 33342 using three-photon excitation. Biophys. J. 72, 567–578 (1997). 34. Maiti, S., Shear, J.B., Williams, R.M., Zipfel, W.R. & Webb, W.W. Measuring serotonin distribution in live cells with three-photon excitation. Science 275, 530–532 (1997). 35. Williams, R.M., Shear, J.B., Zipfel, W.R., Maiti, S. & Webb, W.W. Mucosal mast cell secretion processes imaged using three-photon microscopy of 5-hydroxytryptamine autofluorescence. Biophys. J. 76, 1835–1846 (1999). 36. Xu, C., Zipfel, W., Shear, J.B., Williams, R.M. & Webb, W.W. Multiphoton fluorescence excitation: new spectral windows for biological nonlinear microscopy. Proc. Natl. Acad. Sci. USA 93, 10763–10768 (1996). 37. Zipfel, W.R. et al. Live tissue intrinsic emission microscopy using multiphoton-excited native fluorescence and second harmonic generation. Proc. Natl. Acad. Sci. USA 100, 7075–7080 (2003). 38. Freund, I. & Deutsch, M. 2nd-harmonic microscopy of biological tissue. Opt. Lett. 11, 94–96 (1986). 39. Campagnola, P.J., Clark, H.A., Mohler, W.A., Lewis, A. & Loew, L.M. Second-harmonic imaging microscopy of living cells. J. Biomed. Opt. 6, 277–286 (2001). 40. Mertz, J. & Moreaux, L. Second-harmonic generation by focused excitation of inhomogeneously distributed scatterers. Opt. Commun. 196, 325–330 (2001). 41. Moreaux, L., Sandre, O., Charpak, S., Blanchard–Desce, M. & Mertz, J. Coherent scattering in multi-harmonic light microscopy. Biophys. J. 80, 1568–1574 (2001). 42. Campagnola, P.J. et al. Three-dimensional high-resolution second-harmonic generation imaging of endogenous structural proteins in biological tissues. Biophys. J. 82, 493–508 (2002). 43. Campagnola, P.J., Mohler, W. & Millard, A.E. 3-dimensional high-resolution second harmonic generation imaging of endogenous structural proteins in biological tissues. Biophys. J. 82, 175a–175a (2002). 44. Dombeck, D.A. et al. Uniform polarity microtubule assemblies imaged in native brain tissue by second-harmonic generation microscopy. Proc. Natl. Acad. Sci. USA 100,
1376
7081–7086 (2003). 45. Zoumi, A., Yeh, A. & Tromberg, B.J. Imaging cells and extracellular matrix in vivo by using second-harmonic generation and two-photon excited fluorescence. Proc. Natl. Acad. Sci. USA 99, 11014–11019 (2002). 46. Barad, Y., Eisenberg, H., Horowitz, M. & Silberberg, Y. Nonlinear scanning laser microscopy by third harmonic generation. Appl. Phys. Lett. 70, 922–924 (1997). 47. Muller, M., Squier, J., Wilson, K.R. & Brakenhoff, G.J. 3D microscopy of transparent objects using third-harmonic generation. J. Microsc. 191, 266–274 (1998). 48. Yelin, D., Oron, D., Korkotian, E., Segal, M. & Silbergerg, Y. Third-harmonic microscopy with a titanium-sapphire laser. Appl. Phys. B–Lasers O 74, S97–S101 (2002). 49. Sheppard, C.J.R. & Kompfner, R. Resonant scanning optical microscope. Appl. Optics 17, 2879–2882 (1978). 50. Duncan, M.D., Reintjes, J. & Manuccia, T.J. Scanning coherent anti-Stokes Raman microscope. Opt. Lett. 7, 350–352 (1982). 51. Zumbusch, A., Holtom, G.R. & Xie, X.S. Vibrational mircoscopy using coherent antiStokes Raman scattering (1999). Phys. Rev. Lett. 82, 4014 –4017 (1999). 52. Muller, M., Squier, J., De Lange, C.A. & Brakenhoff, G.J. CARS microscopy with folded BoxCARS phasematching. J. Microsc. 197 (Pt 2), 150–158 (2000). 53. Piston, D.W., Summers, R.G., Knobel, S.M. & Morrill, J.B. Characterization of involution during sea urchin gastrulation using two-photon excited photorelease and confocal microscopy. Microsc. Microanal. 4, 404–414 (1998). 54. Furuta, T. et al. Brominated 7-hydroxycoumarin-4-ylmethyls: photolabile protecting groups with biologically useful cross-sections for two photon photolysis. Proc. Natl. Acad. Sci. USA 96, 1193–1200 (1999). 55. Matsuzaki, M. et al. Dendritic spine geometry is critical for AMPA receptor expression in hippocampal CA1 pyramidal neurons. Nat. Neurosci. 4, 1086–1092 (2001). 56. Echevarria, W., Leite, M.F., Guerra, M.T., Zipfel, W.R. & Nathanson, M.H. Regulation of calcium signals in the nucleus by a nucleoplasmic reticulum. Nat. Cell. Biol. 5, 440–446 (2003). 57. Berland, K.M., So, P.T. & Gratton, E. Two-photon fluorescence correlation spectroscopy: method and application to the intracellular environment. Biophys. J. 68, 694–701 (1995). 58. Schwille, P., Haupts, U., Maiti, S. & Webb, W.W. Molecular dynamics in living cells observed by fluorescence correlation spectroscopy with one- and two-photon excitation. Biophys. J. 77, 2251–2265 (1999). 59. Brown, E.B., Wu, E.S., Zipfel, W. & Webb, W.W. Measurement of molecular diffusion in solution by multiphoton fluorescence photobleaching recovery. Biophys. J. 77, 2837–2849 (1999). 60. Zipfel, W.R. & Webb, W.W. In vivo diffusion measurements using multiphoton-excited fluorescence photobleaching recovery (MPFPR) and fluorescence correlation spectroscopy (MPFCS) in Methods in Cellular Imaging (ed. Periasamy, A.) 345–376 (Oxford University Press, Oxford, UK, 2001). 61. Stroh, M., Zipfel, W.R., Williams, R.M., Webb, W.W. & Saltzman, W.M. Diffusion of nerve growth factor in rat striatum as determined by multiphoton microscopy. Biophys. J. 85, 581–588 (2003). 62. Heinze, K.G., Koltermann, A. & Schwille, P. Simultaneous two-photon excitation of distinct labels for dual-color fluorescence cross correlation analysis. Proc. Natl. Acad. Sci. USA 97, 10377–10382 (2000). 63. Tirlapur, U.K. & Konig, K. Targeted transfection by femtosecond laser. Nature 418, 290–291 (2002). 64. Konig, K., Riemann, I. & Fritzsche, W. Nanodissection of human chromosomes with near-infrared femtosecond laser pulses. Opt. Lett. 26, 819–821 (2001). 65. Göppert-Mayer, M. Uber elementarakte mit zwei quantensprüngen. Ann. Phys. 9, 273–294 (1931). 66. Xu, C. & Webb, W.W. Multiphoton excitation of molecular fluorophores and nonlinear laser microscopy in Topics in Fluorescence Spectroscopy: Volume 5: Nonlinear and TwoPhoton-Induced Fluorescence. (ed. Lakowicz, J.) 471–540 (Plenum Press, New York, 1997). 67. Xu, C. & Webb, W.W. Measurement of two-photon excitation cross sections of molecular fluorophores with data from 690 nm to 1050 nm. J. Opt. Soc. Am. B 13, 481–491 (1996). 68. Steinfeld, J.I. Molecules and Radiation. (MIT Press, Cambridge, MA, 1989). 69. Huang, S., Heikal, A.A. & Webb, W.W. Two-photon fluorescence spectroscopy and microscopy of NAD(P)H and flavoprotein. Biophys. J. 82, 2811–2825 (2002). 70. Piston, D.W., Masters, B.R. & Webb, W.W. Three-dimensionally resolved NAD(P)H cellular metabolic redox imaging of the in situ cornea with two-photon excitation laser scanning microscopy. J. Microsc. 178 (Pt 1), 20–27 (1995). 71. Wong, B.J., Wallace, V., Coleno, M., Benton, H.P. & Tromberg, B.J. Two-photon excitation laser scanning microscopy of human, porcine, and rabbit nasal septal cartilage. Tissue Eng. 7, 599–606 (2001). 72. Noda, M. et al. Switch to anaerobic glucose metabolism with NADH accumulation in the beta-cell model of mitochondrial diabetes. Characteristics of betaHC9 cells deficient in mitochondrial DNA transcription. J. Biol. Chem. 277, 41817–41826 (2002). 73. Zhang, Q., Piston, D.W. & Goodman, R.H. Regulation of corepressor function by nuclear NADH. Science 295, 1895–1897 (2002). 74. Larson, D.R. et al. Water-soluble quantum dots for multiphoton fluorescence imaging in vivo. Science 300, 1434–1436 (2003). 75. Albota, M. et al. Design of organic molecules with large two-photon absorption cross sections. Science 281, 1653–1656 (1998). 76. Wang, X.M. et al. Synthesis of new symmetrically substituted stilbenes with large multiphoton absorption cross section and strong two-photon–induced blue fluorescence. Bull. Chem. Soc. Jpn 74, 1977–1982 (2001). 77. Zhou, X. et al. One- and two-photon absorption properties of novel multi-branched molecules. Phys. Chem. Chem. Phys. 4, 4346–4352 (2002).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW 78. Heikal, A.A., Hess, S.T. & Webb, W.W. Multiphoton molecular spectroscopy and excited-state dynamics of enhanced green fluorescent protein (EGFP): acid-base specificity. Chem. Phys. 274, 37–55 (2001). 79. Blab, G.A., Lommerse, P.H.M., Cognet, L., Harms, G.S. & Schmidt, T. Two-photon excitation action cross-sections of the autofluorescent proteins. Chem. Phys. Lett. 350, 71–77 (2001). 80. Hanson, G.T. et al. Green fluorescent protein variants as ratiometric dual emission pH sensors. 1. Structural characterization and preliminary application. Biochemistry 41, 15477–15488 (2002). 81. Tsai, P.S. et al. All-optical histology using ultrashort laser pulses. Neuron 39, 27–41 (2003). 82. Mainen, Z.F. et al. Two-photon imaging in living brain slices. Methods 18, 231–239, (1999). 83. Shi, S.H. et al. Rapid spine delivery and redistribution of AMPA receptors after synaptic NMDA receptor activation. Science 284, 1811–1816 (1999). 84. D’Apuzzo, M., Mandolesi, G., Reis, G. & Schuman, E.M. Abundant GFP expression and LTP in hippocampal acute slices by in vivo injection of Sindbis virus. J. Neurophysiol. 86, 1037–1042 (2001). 85. Potter, S.M. et al. Structure and emergence of specific olfactory glomeruli in the mouse. J. Neurosci. 21, 9713–9723 (2001). 86. Strome, S. et al. Spindle dynamics and the role of gamma-tubulin in early Caenorhabditis elegans embryos. Mol. Biol. Cell 12, 1751–1764 (2001). 87. Ahmed, F. et al. GFP expression in the mammary gland for imaging of mammary tumor cells in transgenic mice. Cancer Res. 62, 7166–7169 (2002). 88. Lawson, N.D. & Weinstein, B.M. In vivo imaging of embryonic vascular development using transgenic zebrafish. Dev. Biol. 248, 307–318 (2002). 89. Bestvater, F. et al. Two-photon fluorescence absorption and emission spectra of dyes relevant for cell imaging. J. Microsc. 208, 108–115 (2002). 90. Dickinson, M.E., Simbuerger, E., Zimmermann, B., Waters, C.W. & Fraser, S.E. Multiphoton excitation spectra in biological samples. J. Biomed. Opt. 8, 329–338 (2003). 91. Periasamy, A. Fluorescence resonance energy transfer microscopy: a mini review. J. Biomed. Opt. 6, 287–291 (2001). 92. Majoul, I., Straub, M., Duden, R., Hell, S.W. & Soling, H.D. Fluorescence resonance energy transfer analysis of protein-protein interactions in single living cells by multifocal multiphoton microscopy. J. Biotechnol. 82, 267–277 (2002). 93. Bacskai, B.J., Skoch, J., Hickey, G.A., Allen, R. & Hyman, B.T. Fluorescence resonance energy transfer determinations using multiphoton fluorescence lifetime imaging microscopy to characterize amyloid-beta plaques. J. Biomed. Opt. 8, 368–375 (2003). 94. Gu, M. & Sheppard, C.J.R. Comparison of three-dimensional imaging properties between two-photon and single-photon fluorescence microscopy. J. Microsc. 177, 128–137 (1995). 95. Centonze, V.E. & White, J.G. Multiphoton excitation provides optical sections from deeper within scattering specimens than confocal imaging. Biophys. J. 75, 2015–2024 (1998). 96. Periasamy, A., Skoglund, P., Noakes, C. & Keller, R. An evaluation of two-photon excitation versus confocal and digital deconvolution fluorescence microscopy imaging in Xenopus morphogenesis. Microsc. Res. Technol. 47, 172–181 (1999). 97. Schilders, S.P. & Gu, M. Limiting factors on image quality in imaging through turbid media under single-photon and two-photon excitation. Microsc. Microanal. 6, 156–160 (2000). 98. Sheppard, C.J.R. & Gu, M. Image-formation in 2-photon fluorescence microscopy. Optik 86, 104–106 (1990). 99. Richards, B. & Wolf, E. Electromagnetic Diffraction in Optical Systems. 2. Structure of the Image Field in an Aplanatic System. Proc. R. Soc. Lon. Ser. –A 253, 358–379 (1959). 100.Sheppard, C.J.R. & Matthews, H.J. Imaging in high-aperture optical systems. J. Opt. Soc. Am. A 4, 1354–1360 (1987). 101.Beaurepaire, E., Oheim, M. & Mertz, J. Ultra-deep two-photon fluorescence excitation in turbid media. Opt. Commun. 188, 25–29 (2001). 102.Theer, P., Hasan, M.T. & Denk, W. Two-photon imaging to a depth of 1000 microm in living brains by use of a Ti:Al2O3 regenerative amplifier. Opt. Lett. 28, 1022–1024 (2003). 103.Curley, P.F., Ferguson, A.I., White, J.G. & Amos, W.B. Application of a femtosecond self-sustaining mode-locked Ti:sapphire laser to the field of laser scanning confocal microscopy. Opt. Quant. Electron. 24, 851–859 (1992). 104.Hockberger, P.E. et al. Activation of flavin-containing oxidases underlies light-induced production of H2O2 in mammalian cells. Proc. Natl. Acad. Sci. USA 96, 6255–6260 (1999). 105.Wokosin, D.L., Squirrell, J.M., Eliceiri, K.W. & White, J.G. Optical workstation with concurrent, independent multiphoton imaging and experimental laser microbeam capabilities. Rev. Sci. Instrum. 74, 193–201 (2003). 106.Hopkins, J. & Sibbett, W. Ultrashort lasers: big payoff in a flash. Sci. Am. 283, 73–79 (2000). 107.Soeller, C. & Cannell, M.B. Construction of a two-photon microscope and optimization of illumination pulse duration. Pflugers Arch. 432, 555–561 (1996). 108.Squier, J. & Muller, M. High resolution nonlinear microscopy: A review of sources and methods for achieving optimal imaging. Rev. Sci. Instrum. 72, 2855–2867 (2001). 109.Muller, D., Squier, J. & Brakenhoff, G.J. Measurement of femtosecond pulses in the focal point of a high-numerical-aperture lens by two-photon absorption. Opt. Lett. 20, 1038–1040 (1995). 110.Guild, J.B., Xu, C. & Webb, W.W. Measurement of group delay dispersion of high numerical aperture objective lenses using two-photon excited fluorescence. Appl. Optics 36, 397–401 (1997).
111.Muller, M., Squier, J., Wolleschensky, R., Simon, U. & Brakenhoff, G.J. Dispersion precompensation of 15 femtosecond optical pulses for high-numerical-aperture objectives. J. Microsc. 191, 141–150 (1998). 112.Majewska, A., Yiu, G. & Yuste, R. A custom-made two-photon microscope and deconvolution system. Pflugers Arch. 441, 398–408 (2000). 113.Tsai, P.S. et al. Principles, design and construction of a two photon scanning microscope for in vitro and in vivo studies in Methods for In Vivo Optical Imaging (ed. Frostig, R.) 113–171 (CRC Press, Boca Raton, FL, 2002). 114.Iyer, V., Losavio, B.E. & Saggau, P. Compensation of spatial and temporal dispersion for acousto-optic multiphoton laser-scanning microscopy. J. Biomed. Opt. 8, 460–471 (2003). 115.Pawley, J.B. Handbook of Biological Confocal Microscopy, edn 2. (Plenum Press, New York, 1995). 116.Fan, G.Y. et al. Video-rate scanning two-photon excitation fluorescence microscopy and ratio imaging with cameleons. Biophys. J. 76, 2412–2420 (1999). 117.Nguyen, Q.T., Callamaras, N., Hsieh, C. & Parker, I. Construction of a two-photon microscope for video-rate Ca(2+) imaging. Cell Calcium 30, 383–393 (2001). 118.Gauderon, R., Lukins, P.B. & Sheppard, C.J. Effect of a confocal pinhole in two-photon microscopy. Microsc. Res. Technol. 47, 210–214 (1999). 119.Oheim, M., Beaurepaire, E., Chaigneau, E., Mertz, J. & Charpak, S. Two-photon microscopy in brain tissue: parameters influencing the imaging depth. J. Neurosci. Methods 111, 29–37 (2001). 120.Egner, A., Jakobs, S. & Hell, S.W. Fast 100-nm resolution three-dimensional microscope reveals structural plasticity of mitochondria in live yeast. Proc. Natl. Acad. Sci. USA 99, 3370–3375 (2002). 121.Tan, Y.P., Llano, I., Hopt, A., Wurriehausen, F. & Neher, E. Fast scanning and efficient photodetection in a simple two-photon microscope. J. Neurosci. Methods 92, 123–135 (1999). 122.Gratton, E., Breusegem, S., Sutin, J., Ruan, Q. & Barry, N. Fluorescence lifetime imaging for the two-photon microscope: time-domain and frequency-domain methods. J. Biomed. Opt. 8, 381–390 (2003). 123.Moreaux, L., Sandre, O. & Mertz, J. Membrane imaging by second-harmonic generation microscopy. J. Opt. Soc. Am. B 17, 1685–1694 (2000). 124.Peleg, G., Lewis, A., Linial, M. & Loew, L.M. Nonlinear optical measurement of membrane potential around single molecules at selected cellular sites. Proc. Natl. Acad. Sci. USA 96, 6700–6704 (1999). 125.Moreaux, L., Sandre, O., Blanchard–Desce, M. & Mertz, J. Membrane imaging by simultaneous second-harmonic generation and two-photon microscopy. Opt. Lett. 25, 320–322 (2000). 126.Millard, A.C., Jin, L., Lewis, A. & Loew, L.M. Direct measurement of the voltage sensitivity of second-harmonic generation from a membrane dye in patch-clamped cells. Opt. Lett. 28, 1221–1223 (2003). 127.Mohler, W., Millard, A.C. & Campagnola, P.J. Second harmonic generation imaging of endogenous structural proteins. Methods 29, 97–109 (2003). 128.Konig, K., So, P.T., Mantulin, W.W., Tromberg, B.J. & Gratton, E. Two-photon excited lifetime imaging of autofluorescence in cells during UVA and NIR photostress. J. Microsc. 183 (Pt 3), 197–204 (1996). 129.Koester, H.J., Baur, D., Uhl, R. & Hell, S.W. Ca2+ fluorescence imaging with pico– and femtosecond two-photon excitation: signal and photodamage. Biophys. J. 77, 2226–2236 (1999). 130.Hopt, A. & Neher, E. Highly nonlinear photodamage in two-photon fluorescence microscopy. Biophys. J. 80, 2029–2036 (2001). 131.Dittrich, P.S. & Schwille, P. Photobleaching and stabilization of fluorophores used for single-molecule analysis with one- and two-photon excitation. Appl. Phys. B. Lasers O 73, 829–837 (2001). 132.Patterson, G.H. & Piston, D.W. Photobleaching in two-photon excitation microscopy. Biophys. J. 78, 2159–2162 (2000). 133.Neil, M.A. et al. Adaptive aberration correction in a two-photon microscope. J. Microsc. 200 (Pt 2), 105–108 (2000). 134.Booth, M.J., Neil, M.A. & Wilson, T. New modal wave-front sensor: application to adaptive confocal fluorescence microscopy and two-photon excitation fluorescence microscopy. J. Opt. Soc. Am. A Opt. Image Sci. Vis. 19, 2112–2120 (2002). 135.Marsh, P.N., Burns, D. & Girkin, J.M. Practical implementation of adaptive optics in multiphoton microscopy. Opt. Express 11, 1123–1130 (2003). 136.Brunner, F. et al. Diode-pumped femtosecond Yb:KGd(WO/sub 4/)/sub 2/ laser with 1.1-W average power. Opt. Lett. 25, 1119–1121 (2000). 137.Ilday, F.O., Lim, H., Buckley, J.R. & Wise, F.W. Practical all-fiber source of high-power, 120-fs pulses at 1 micron. Opt. Lett. 28, 1362–1364 (2003). 138.Jung, J.C. & Schnitzer, M.J. Multiphoton endoscopy. Opt. Lett. 28, 902–904 (2003). 139.Bird, D. & Gu, M. Two-photon fluorescence endoscopy with a mirco-optic scanning head. Opt. Lett. 28, 1552–1554 (2003). 140.Ouzounov, D.G. et al. Delivery of nanojoule femtosecond pulses through large-core microstructured fibers. Opt. Lett. 27, 1513–1515 (2002). 141.Pastirk, I., Dela Cruz, J.M., Walowicz, K.A., Lozovoy, V.V. & Dantus, M. Selective twophoton microscopy with shaped femtosecond pulses. Opt. Express 11, 1695–1701 (2003). 142.Williams, R.M. & Webb, W.W. Single granule pH cycling in antigen-induced mast cell secretion. J. Cell Sci. 113 (Pt 21), 3839–3850 (2000). 143.Kloppenburg, P., Zipfel, W.R., Webb, W.W. & Harris–Warrick, R.M. Highly localized Ca(2+) accumulation revealed by multiphoton microscopy in an identified motoneuron and its modulation by dopamine. J. Neurosci. 20, 2523–2533 (2000). 144.Kleinfeld, D., Mitra, P.P., Helmchen, F. & Denk, W. Fluctuations and stimulus-induced changes in blood flow observed in individual capillaries in layers 2 through 4 of rat neocortex. Proc. Natl. Acad. Sci. USA 95, 15741–15746 (1998).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1377
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW
FOCUS ON OPTICAL IMAGING
Near-field optics: from subwavelength illumination to nanometric shadowing Aaron Lewis, Hesham Taha, Alina Strinkovski, Alexandra Manevitch, Artium Khatchatouriants, Rima Dekhter & Erich Ammann Near-field optics uniquely addresses problems of x, y and z resolution by spatially confining the effect of a light source to nanometric domains. The problems in using far-field optics (conventional optical imaging through a lens) to achieve nanometric spatial resolution are formidable. Near-field optics serves a bridging role in biology between optical imaging and scanned probe microscopy. The integration of near-field and scanned probe imaging with far-field optics thus holds promise for solving the so-called inverse problem of optical imaging.
The world of microscopy can be divided into two defined approaches: lens-based imaging and lensless imaging (an overview of microscopy today is shown in Fig. 1). This article describes a method of lensless optical imaging whereby optical resolution can be increased by as much as 10-fold in the x and y dimensions and >100-fold in the z dimension. The technique is called near-field scanning optical microscopy (NSOM). To place this new approach within the constellation of microscopic methodologies available today, a starting point is lens-based imaging, which was the first approach to be developed for microscopic characterization with the development of the optical microscope, invented some 400 years ago. In general, lens-based microscopes are an example of far-field imaging in which the imaging element, the lens, is placed many wavelengths away from the object and focuses and images the appropriate radiation onto or from an object under analysis. All lensbased imaging is limited by criteria such as the Rayleigh resolution limit. The Rayleigh criterion tries to define the xy resolution that can be achieved based on the wavelength of the radiation that is used, the acceptance angle of the lens and the index of refraction of the medium in which the radiation is propagating. As is well known for lens-based optical microscopy, this means that single-wavelength illumination can resolve the rod-like geometry of a 1-µm bacterium but not the geometry of even the largest virus (pox virus, with a size of 0.25 µm)—which requires a significant reduction in the wavelength of the radiation, to the point where an election microscope has to be used. None of the most advanced lens-based, purely optical techniques, even the most advanced methods of short-pulse excitation, have achieved the goal of imaging a virus. In addition to the problem of diffraction, which results in the Rayleigh resolution limit, another critical problem in all lens-based microscopy techniques is out-of-focus light (Fig. 2). This problem is Division of Applied Physics, The Hebrew University of Jerusalem, Jerusalem 93707 Israel. Correspondence should be addressed to A.L. ([email protected]). Published online 31 October 2003; doi:10.1038/nbt898
1378
clearly highlighted by the fact that the z resolution of an optical microscope, ∼1.6 µm, is much poorer than the xy resolution noted above. A powerful method to reduce this interference is confocal imaging, which places apertures in the illumination path before the lens and the detection path after the lens (Fig. 2). This limits the detected rays to those close to the plane of focus, within 0.7 µm. The reduction in out-offocus light also has an effect on the xy resolution, which as a result is brought too close to the theoretical limit of the Rayleigh criterion, which is 0.25 µm in x and y. The increase in resolution, relative to conventional optical imaging, approached by confocal imaging is only a factor of ∼2 in the x, y and z dimensions (Table 1). Yet confocal microscopy has transformed fields from biology to semiconductor manufacturing—highlighting the importance of even moderate improvements in the resolving power of light. An alternative way to reduce out-of-focus light is to use a prism to create an evanescent field of radiation. This is a field of light that is attached to the surface of the prism and decays in intensity exponentially as a function of distance from the prism surface. Parts of an object that sit on the prism surface are illuminated to an extent of <0.3 µm, which is the maximal penetration into the object of such an evanescent field. This approach to illuminating objects effectively halves the z illumination relative to confocal microscopy. However, this means that only the underside of samples are effectively illuminated, and only with samples that are transparent can the illuminated region be imaged with a lens that is placed above the prism surface. The xy resolution in such a methodology is of course limited by the same criteria that limit lensbased image formation in x and y. Near-field optics was developed in our laboratory1 to break the resolution limit in x, y and z by sending light through an aperture that is much smaller than the wavelength of light and then scanning the aperture or the sample relative to each other at a distance much smaller than a wavelength. This ensures that the light interaction occurs before the enormous effects of diffraction come into play (see Fig. 2). The xy resolution is limited to the dimension of the aperture, whereas the z extent of effective illumination is defined by the diffraction of the aperture. An
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
REVIEW Microscopy today Lens Based
Lensless Prisms
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Ion optical microscopes
Optical microscope
Electron optical microscopes
X-ray microscopy Scanning tunneling microscopy
Aberration
Diffraction
Waveguides
Scanned probe microscopy Near-field (atomic force optical related)
Out-of-focus light
FIB SEM TEM STEM Confocal
Non-linear femtosecond imaging
3rd-harmonic generation
2nd-harmonic generation
Two- and three-photon fluorescence
Computational methods
is the case for all the nonlinear methods listed in Figure 1 aside from SHG. In the 1980s, our group realized that the requirement in SHG for asymmetric distributions could effectively be applied to selectively investigate molecules at biological interfaces, such as membranes (see ref. 4 and references therein). Our extensions of these ideas led to functional nonlinear optical imaging of membrane potential5. These advances, along with the contributions of others, are described elsewhere in this issue (see Loew and Campagnola, p. 1356). In the following review, we place near-field optics and its development and achievements within the context of state-of-the-art far-field optical imaging and its application in the life sciences. The synergism resulting from combining near-field optics and SHG is described later in this review.
Aperture-based NSOM Today, near-field optical instrumentation is Integrated imaging available that places this technique at the threshold of achieving its enormous potential Figure 1 An overview of microscopic imaging. In general, two major trends in microscopy have evolved. for nanometric imaging with light. In essence, One is based on the lens as the optical element and the other is based on lensless methodologies. near-field optics is a bridging technique Optical microscopy has representation on both sides of this divide. From an intellectual, instrumentation and computational point of view, near-field scanning optical microscopy and its between two enormous worlds of biologically associated techniques form a bridge that can interconnect in a most synergistic way the variety of relevant imaging, optics on the one hand and micro- and nanocharacterization tools available today. Arrowed lines show areas where the integration scanned probe microscopy (SPM) on the of scanned probe and near-field optics could be important. other. This bridging role needed the appropriate instrumentation to bring the technique of near-field optics to where it now finds itself in evanescent field emanates from the aperture and diffraction limits the the expanding world of microscopy. To give the reader a true sense of region in z that has sufficient fluence to generate optical contrast to what has been accomplished and what is the potential for the future, it maybe 10 nm from the surface of a 50-nm aperture. This is 300-fold is necessary to give some background of previous approaches that led better than evanescent wave imaging. Greater z penetration can be to the present instrumental capabilities (see Box 1). The present instruachieved with larger apertures. For example, an aperture of 0.25 µm, mentation, developed in our laboratories, can be used with any probe which is the limit of lens-based imaging, provides an enormous signal that is available today and any mode of operation known today. This in near-field optics and achieves penetrations in z close to those of flexibility is crucial to enable NSOM to assume a bridging, interrelating role in which it can be transparently integrated with other microscopic evanescent field imaging. Thus, near-field optics1–3 is a means of spatially confining light that is techniques, including optical and spectral (e.g., Raman) as well as elecrightly classified as lensless, but is within the intellectual genre of tron and ion optical. In the early 1980s, when we started putting our ideas of near-field approaches in lens-based imaging discussed above, such as confocal microscopy. This is in contrast to the techniques of lens-based nonlinear optics into practice, the first goal was to clearly define how much light microscopies, such as two-photon microscopy, three-photon could come out of a well-characterized subwavelength aperture2. There microscopy and second-harmonic generation (SHG), which can be clas- was no knowledge and a great deal of skepticism as to how much light sified as spectral, rather than spatial, approaches to confine the effects of could emanate from, say, a 50-nm aperture. There was even skepticism light. They use ultrafast lasers interacting with materials to limit the that such an aperture emanating light could be fabricated in a silicon effective fluence in z by the probability of multiple photon interactions wafer using nanolithography, which was then in its infancy. Our initial that are required to elicit the spectral phenomena being monitored. This experiments published in 1984 were a huge success2, however, which encouraged us to move forward. The next step was to address the problem of how to design a simple, reproducible aperture that could track real surfaces, which are Table 1 Resolution presently achievable in optical imaging markedly rough. Our first apertures2 used silicon-processing techInstrument xy resolution z resolution niques, but these techniques were in their infancy and incapable of handling this complicated task. We therefore invented a simple method, Standard microscope ∼0.5 µm ∼1.6 µm based squarely in biology, that gave a simple, cheap and reliable Confocal/two-photon microscope ∼0.25 µm ∼0.7 µm methodology for fabricating such apertures6. This technique is used in Evanescent wave ∼0.5 µm ∼0.3 µm a majority of near-field optical measurements today is based on elecSecond-harmonic microscope ∼0.25 µm <0.01 nm Near-field imaging ∼0.05 µm ∼0.01 µm trophysiological methods for producing intracellular electrodes. Glass
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1379
REVIEW Confocal principle
Evanescent wave principle Lens Evanescent wave
Laser
<300 nm Prism
Out-of-focus Confocal pinholes
Excitation
Near-field optical principle Illumination mode
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Diffraction Nonlinear optical confinement
Objective
Near-field Focal plane Not in focal plane
Subwavelength aperture
Far-field One wavelength
Figure 2 A comparison of optical imaging approaches to reduce out-of-focus light and diffraction-induced reduction in resolution. Either apertures can be placed in the excitation and in the detection path (confocal principle), prism illumination can be used (evanescent wave principle) or ultrashort pulse excitation can be applied for nonlinear optical spectral confinement of light (near-field optical principle). Within this constellation of approaches is the near-field optical principle (lower right box). The red dots are not resolvable in far-field optics but are resolvable using near-field optics by illumination or collection of light through a subwavelength aperture. The sample or the aperture can be scanned relative to the other. Near-field optics is a method based on spatial confinement of light and reduction in background.
atomic force technique that was applied to the problem of near-field optics was shear or lateral force. Of the four versions of the technique that were originally independently developed, the one based on the tuning fork or emulations of the tuning fork is nearly universally employed today8. In this technique, invented by Karrai and Grober8, a straight fiber probe is mounted on one of the tines of the tuning fork and modulated by several nanometers. When the probe tip approaches a surface, the frequency of the tine on which the fiber is mounted is altered relative to the tine that is free (Fig. 3a). This produces a change in the amplitude and phase of the modulation of the fiber attached relative to the free tine, and results in a signal that can be used by the electronics to keep the probe tip in relative proximity to the surface, by altering piezoelectrically the sample and/or tip position to keep the signal from the tuning fork constant. The ease of use of shear force has always left something to be desired. This has less to do with the tuning-fork technique—which is thought to be superior even to conventional AFM feedback when applied in standard AFM geometries of normal force9—than with the fact that the forces in the normal force direction are larger and more defined than the forces in the lateral direction. Thus, for example, even a water layer, which is found on all (even dry) surfaces, contributes greatly to a lateral force signal. This leaves doubt as to the exact distance that the probe tip is from a surface. Also of considerable importance for biological applications is that liquid NSOM imaging is not available in any commercial instrument using straight fiber probes. In addition, data obtained in shear-force mode are not directly comparable, in terms of AFM interactions, with conventional AFM imaging, where the standard is normal force feedback for obvious reasons, as detailed above. Furthermore, straight fiber probe geometries seriously perturb integration with standard optical or other microscopes by blocking the optical axis from above or below. To circumvent these difficulties, in the early 1990s, our group10 showed that such straight glass structures could readily be cantilevered
structures with apertures as small as 50 nm after metal coating were first constructed using a Flaming/Brown glass puller (Sutter, Novato, CA, USA), which relies on heating, tension and controlled pulling to taper a piece of glass. The initial glass structures pulled by this method were micropipettes, because the nichrome heaters used in the first Flaming/Brown pullers were simply not capable of melting quartz fibers. The year was 1986, the year the invention of the ground-breaking technique called atomic force microscopy (AFM). AFM was certainly an enabling technology for the confinement, manipulation a c b and analysis of light in nanometric domains. In terms of light propagation, subwavelength apertures illuminated through an air Fiber Tuning medium (as in a tapered glass nanopipette) probe fork and silicon apertures show substantial scatterg ing, because of the large mismatch between Direction of the refractive indices of glass and air and the modulation absence of guiding of the light in the tapered e Sample region where a fiber would still allow light propagation. Thus, when carbon dioxide d lasers were introduced to the technology of pipette pulling, a natural progression to quartz fibers ensued7. This resulted in higher f h Lens Lens throughput and the ability to guide light effectively in the illumination or collection mode of operation. Today, the Flaming/Brown puller based on carbon dioxide laser heating is the puller of choice. By the early 1990s, the full import of AFM was being felt, and AFM was a natural means to bring a subwavelength aperture into the Figure 3 A flowering of probes and modes. (a) Straight fiber probe in shear-force mode of feedback. near-field in a controlled fashion if the aper- (b) Cantilevered fiber probe for normal-force mode of NSOM. (c) Conventional normal-force silicon cantilever. (d) Apertured silicon cantilever. (e) Cantilevered nanopipette aperture with free access ture was placed at the tip of a force sensing from above the aperture. (f) Dye held in a nanopipette nanoaperture excited with epi-illumination. structure. The glass-tapering technology (g) Nanopipette probe with a gold nanoparticle at the tip. (h) Diagrammatic representation of the described above did exactly this, and the first difference imaging method of shadow NSOM.
1380
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
REVIEW
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Box 1 Instrumentation An NSOM system is essentially a scanned probe microscope (SPM). Standard scanned probe microscope geometries (Fig. 7a) are composed of a cylindrical piezoelectric device that can move, with appropriately placed voltages, a sample or a probe in the x and y dimensions at right angles to the axis of the cylindrical piezo, and to a limited extent in the z dimension along the piezo axis. In general, SPM technology is not optically friendly, as such a piezo geometry blocks the lens of a standard upright microscope. Thus, there is no SPM that can sit under a standard upright microscope. Often, such SPM designs also block the optical axis from below. Moreover, the standard SPM probe uses a silicon cantilever, consisting of an opaque silicon wafer, that also blocks the sample from above, by preventing the light from reaching the sample and impeding an accurate view of the probe tip. Both these impediments to optical integration can be overcome with a design that was developed in our laboratory12,43. The scanner uses the same cylindrical piezoelectric crystals as SPM scanners, but contains four of them in a square geometry (Fig. 7b). The SPM system based on this scanner (Fig. 7c) provides not only standard SPM/AFM x,y scan ranges, but also millimeters of rough scanning and, with the same scanner, a z scan range seven- to tenfold larger (70 µm) than that of any known AFM scanning system. This is of great importance for on-line (with AFM) confocal and CCD optical sectioning. In addition, such sample z motion is preferable to z motion of the lens for optical sectioning, both in terms of using any objective in the far-field microscope and in terms of not altering the position of the far-field optical element. This last point is especially important in dual-lens 4Pi geometries for advanced nonlinear optical techniques. In addition, cantilevered optical fiber or other transparent cantilevered glass probes, such as nanoparticle-containing nanopipette probes, have their tips exposed to the optical axis (Fig. 3). This is important for reflection imaging
and formed excellent normal-force AFM sensors. Fujihara and coworkers11, our laboratory12 and Dunn13 showed that excellent near-field optical images could be obtained with cantilevered optical fibers (Fig. 3b) together with on-line normal-force topographic imaging. The probes can be used in either contact or noncontact mode with nanoNewton forces between the tip and the surface. Noncontact imaging requires modulating the probe with high resonance frequencies and detecting the change in amplitude and phase in the feedback signal as the probe tip interacts with a surface. These cantilevers are remarkably similar, in their geometrical and force characteristics, to the ideal in atomic force sensors that is being sought but is hard to achieve in standard silicon AFM sensors14. These features include a single-beam, rather than the standard dual-beam, cantilever structure, which has been shown to reduce cantilever twisting14, and a probe tip that is exposed to the optical axis of the microscope (compare Fig. 3b and Fig. 3c). Other probes whose application will be discussed below are shown in Fig. 3d–h. An example of imaging with such fiber probes completed with the system described in Box 1 is shown in Figure 4. The images starting from the top left (Fig. 4a–c) are, in sequence, a low-resolution (10×) optical image of yeast cells taken with the inverted section of the dual far-field optical microscope (seen in the box), then a higher-resolution (20×) optical image to the left of the same field of view, and then a higher-resolution (50×) image with the fiber probe in place illuminat-
and for irradiating nanoparticle probe tips with the lens of the microscope, as in surface-enhanced applications or shadow applications for Raman scattering (discussed in the text). In fact, in terms of AFM, these are the only probes available today that allow the probe tip to sit on the optical axis of the microscope and to be viewed freely. These optical fiber probes also can be coated with metal to form a subwavelength point of light (Fig. 3b). The combination of the three-dimensional flat scanning stage and the cantilevered optical fiber probe makes for an optically friendly geometry (Fig. 7c). The scan head can even be inserted into a 4Pi configuration, with lenses above and below the sample in a dual microscope geometry (Fig. 7d,e). Using the lenses of the dual microscope, the transmitted and reflected light are collected from below and above the sample, respectively (the light in the fiber is shown in red, transmitted light in green and reflected light in yellow). The basis of feedback of the sample probe position is seen in the three-dimensional cutaway in the illustration of the scanner in Figure 7c, in which a laser is reflected off the cantilever onto a position sensitive detector (PSD). As the topography of the sample changes, the cantilever alters its position and the reflected signal changes; the scanner is thus adjusted in z to return the PSD value to its value before topographic alteration. The system displayed in Figure 7e is an integrated, near-field, far-field, confocal and multiphoton microscope. From the point of view of NSOM, this system can employ any probe and perform any mode of operation presently known. From the point of view of far-field confocal microscopy, the z feedback on the sample rigidly maintains the sample surface relative to the optical element and eliminates all effects of sample topography on confocal imaging. From the point of view of SPM, imaging can be performed at the limits of what is now capable in SPM platforms (see image of carbon nanotube, Fig. 7f).
ing the sample. At right is a view from the upright microscope portion of the dual microscope of the probe out of contact (Fig. 4d) and in contact with the yeast cells (Fig. 4e). Without any further adjustments, the AFM image in Figure 4f is obtained with overlapping fields of view with the optical far-field image. Simultaneously, in another channel a green fluorescent protein (GFP) fluorescence NSOM image of a form of GFP expressed in this budding yeast is recorded with the same probe (Fig. 4g). An NSOM transmission optical image is also shown in Figure 4h. Accurate placement of the AFM probe within the field of view of the optical images, permits imaging from eyeball to nanometers in one integrated instrument and adds detailed topographic information to the optical information that lacks such detail. Also, in terms of SPM in general, there is significant information added about the distribution of GFP fluorescence and absorption within the AFM topography. AFM, of course, has no such biologically relevant information. NSOM optical fiber probes can be used in all aperture-based NSOM applications. These include the excellent transmission imaging that is shown in Figure 4, reflection imaging15 and collection-mode operation16. The cantilevered nature of the probe allows viewing online with the lens along with NSOM and AFM imaging (Fig. 4). The ability to have separate channels of illumination is significant for reflection NSOM imaging, where the light reflected by a sample illuminated with the probe is collected by the lens of the upright microscope. This is
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1381
REVIEW metal-coated transparent silicon nitride cantilever21. A prerequisite for producing light at the tip of such silicon-based NSOM apertures is the ability to bring a lens in close proximity to the back side of the probe. The platform described in Box 1 allows such illumination. Essentially, e f g h this platform enables any probe and any mode that is presently known for NSOM to be effectively used. Silicon probes do not have light guiding and light concentration properties, as in tapered fibers do, which creates several problems. First, they require large fluences of Figure 4 Images of yeast obtained in three modes of operation. (a–c) Far-field optical microscope light from a lens to illuminate the aperture, images obtained with the inverted portion of a dual microscope with 10× (a), 20× (b) and 50× (c) resulting in a reduction in signal-to-noise magnification, respectively. (d,e) Frames obtained with the upright portion of the dual microscope ratio. Second, when illuminated, they suffer where the probe is, respectively, out of contact (d) and in contact (e) with the yeast cell. The red from the index mismatch problems noted reflection off the probe is the feedback laser. (f–h) Simultaneously obtained AFM (f), transmission above. Third, they do not allow an on-line fluorescence NSOM (g) and transmission absorption NSOM (h). The last two images were obtained separate illumination or collection channel with 488 nm excitation of GFP and the fluorescence was monitored at 515 nm. All images courtesy of Patrick Degenaar, associated with Eiichi Tamiya’s laboratory at the Japan Advanced Institute of Science with the lens of the optical microscope, which and Technology, Ishikawa, Japan, obtained with MultiView 1000 (Nanonics Imaging). Scale bars, 50 has to be dedicated to illuminate the aperture µm for a–c; 100 µm for d,e; 1 µm for f–h. (as a result, reflection imaging is difficult and the lack of light-guiding capacity prevents scanning of the probe in collection-mode important in the imaging of opaque biochips and tissue samples in imaging). Fourth, noncontact or intermittent contact imaging for such which transmission NSOM is not possible. It is also important for the an illuminated aperture has been problematic and liquid imaging has integration of NSOM measurements with such modalities as differen- remained difficult. Nonetheless, for femtosecond ablation experiments, tial interference contrast (DIC) imaging. Independent illumination of a where signal-to-noise concerns are not critical, these probes or their lens of an upright microscope with a probe in place is also of particular higher-damage-threshold glass counterparts (Fig. 3e) may be preferimportance when one considers new modalities of NSOM imaging able. Finally, a new development applicable to all aperture-based NSOM is (described below). Today there is a general consensus that more light can be obtained work that indicates a gold nanoparticle or gold-coated asperity placed from a straight than from a cantilevered fiber. However, our group has on the aperture can increase the boundary conditions for light the most extensive comparisons to date, and these indicate that, as throughput through such apertures22. Preliminary results from this would be expected, most light loss occurs at the subwavelength aper- research indicate that there is a chance of considerable improvement in ture, which is common to both straight and cantilevered NSOM fiber resolution for such aperture-based NSOM, to below the nominal 50 probes. The bend in the cantilevered probe exhibits less loss, by several nm that is often quoted. orders of magnitude, than the aperture and this can be compensated for. In this regard, it should be mentioned that there are two ways to Exponential growth in NSOM methodologies produce subwavelength apertures in fibers. One is the pulling method, Aside from standard aperture-based NSOM, there are a whole host of as described above, and the other is based on the etching approach pio- near-field optical methods aimed at obtaining nanometric optical neered by Ohtsu and his group in Japan17. Etched probes have a struc- information. Most of these techniques show the considerable importure that should theoretically give higher throughput. However, the tance of full optical integration of lensless and lens-based imaging. External illumination—active light sources. One technique that etching causes difficulty in coating, and thus today most work is done with pulled fiber probes. Zenobi, Deckert and coworkers18 have also highlights such integration, and indicates the considerable potential for developed a specialized etching technique that reduces extraneous pin- functional near-field optical imaging of samples with biological imporholes in the coating. All such probes, however, require quite flat sample tance, calls for externally illuminating an aperture containing a fluorophore that emits a point of light as an active light source for NSOM geometries, because etching produces very large, flat probe tips. The importance of normal force feedback has stimulated several imaging. This technique was first described by our group23. More workers to try to develop an NSOM aperture with normal force sensing recently, however, elegant studies have been performed by Sandoghdar in silicon-based materials19,20 (see Fig. 3d). Using standard silicon and coworkers demonstrating imaging using an active light source conmicrofabrication to produce such NSOM apertures in silicon AFM sen- sisting of a single molecule (for a review, see ref. 24). Optical analog of patch clamping. We have been investigating the sors has been difficult and progress has been slow. Although it is still not possible to emulate the guiding of light that is inherent to the potential of this technique for monitoring ion concentrations in and nature of an optical fiber, at least two techniques have surfaced in addi- around cellular membranes. The basis of our approach (Fig. 5) is a cantion to traditional microprocessing technology for producing such tilevered micropipette that is used as a nanovessel for an ion-sensing apertures. The methods include simply coating a silicon nitride can- dye, which is excited using an epi-illumination geometry. The AFM tilever, which is inherently transparent, with metal and then producing function of the system described in Box 1 is used to provide unprecean aperture with focused ion-beam techniques. Alternatively, a most dented (<1 nm) control of a dye at any position in and around a cell. In innovative technique based on the evanescent field on the surface of a the past, we have used a pyranine dye fluorescence to monitor pH with prism (as described above) has been used to create an aperture in a nanometric spatial control in and around charged surfaces25. We have
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
a
1382
b
c
d
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
REVIEW a
AFM feedback
b
Calcium sensing dye embedded in sol gel glass contained in cantilevered nanopipette probe tip
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Sample
Lens Dichroic mirror
Epi-illumination
Detector
Figure 5 Neuronal calcium effusion with caffeine excitation. (a) Experimental arrangement of external illumination NSOM mode for ion sensing. A cantilevered nanopipette tip is used as a nanovessel for an ion-sensing dye, which is excited in epi-illumination. The AFM function is used to provide unprecedented control of a dye in an optical microscope. (b) Calcium ion efflux through the membrane of a neuronal cell, induced by the addition of caffeine, imaged using this probe . Each frame in the image is one confocal scan and the distance of the tip to the surface of the cell is maintained to 1 nm with the AFM control.
now extended these studies and have combined a Multiview 1000 (Nanonics Imaging) system illustrated in Figure 7 with a confocal beam-scanning system (Leica Microsystems). This confocal beam scanner was used to image a neuron while the probe was held at a constant distance above the cell surface (Fig. 5b). The neuronal cell line that we investigated consisted of primary cultures of hippocampal neurons26. For these studies, the probe tip containing calcium green is held in place with AFM control over a neuronal cell membrane of a cell that is also filled with calcium green. The probe tip and the cell are excited in the epi-illumination geometry (illustrated in Fig. 5a by a beam-scanning confocal microscope). Each frame in the image is one confocal scan of the cell and the probe tip that takes 250 µs to record. The distance of the tip from the surface of the cell is maintained to 1 nm with the AFM control during all of these scans. Between the fourth and the fifth confocal scans, cell calcium starts to be released from internal stores in response to a puff of caffeine injected just before the first confocal image (left image, top row). In confocal scans five and six (right images, top and middle row of Fig. 5b), the cell continues to respond, and in seventh scan the calcium is effusing through calcium channels and the tip begins to respond. The tip continues to respond in the eighth scan where some tip pixels are even red (which indicates the highest fluorescence intensity). At this point the cell response begins to die down. As each confocal scan takes 250 µs, the first response of the cell is at 1.25 ms, whereas the first tip response is between 1.50 ms and 1.75 ms. The tip response begins to decrease between 2.00 ms and 2.25 ms. Such times are quite reasonable for such a calcium release. The technique can potentially be used to monitor ionic alterations in the dendritic spines of neurons without the mechanically perturbing suction that results from patch clamping. Also, ionic fluxes around synaptic terminals can be monitored together in conjunction with ultrastructural detection of membrane movement accompanying such ion fluxes. At least one suggested theory of neuron learning proposes a physical movement of the synapse that increases the synaptic strength, and this technology could potentially be applied to investigate such processes. External illumination—passive concentrators. An alternative approach to an external illumination protocol is to illuminate a standard silicon cantilever from the side, with the polarization of the inci-
dent light being along the axis of the tip of the probe. In general, such experiments have used standard silicon cantilevers, and light concentration is achieved simply by the nanometric tip that concentrates the light field at the tip of the probe. It is possible that glass probes, with their very long, slender profiles with gold nanoparticles (Fig. 3g), would be even better than silicon AFM sensors for such an application, but this has not been tested in this mode of operation, which is generally termed apertureless NSOM (ANSOM). Pioneering studies in this area were performed by Wickramasinghe and co-workers27 and Knoll and Keilmann28, among others. These investigators showed contrast that could be associated with alterations in optical interactions. The problem of coupling of topography with optics in this technique has been particularly severe and has plagued many studies in the field. In all of these external illumination schemes it is possible to tag alterations caused by the tip-sample interaction by modulating the probe at close to its resonance frequency. This makes it possible to electronically tag the scattered light from the tip-sample interaction. In spite of this, the scattered signal monitored at the resonance frequency is still highly contaminated with spurious signals that are not associated with the tipsample interactions. The problem can be significantly improved by monitoring the signal from this interaction at a frequency that is the second harmonic of the fundamental frequency at which the probe is modulated28,29. Probably the most exciting application of this sort of external illumination protocol is the imaging of chemical alterations in a sample by monitoring the scattering of infrared radiation within the region of the electromagnetic spectrum where vibrational modes of surface molecules absorb light in chemically specific ways. Such chemical identification with high spatial resolution is very important for numerous areas of interest in biology. These extend from the chemical identification of molecular entities on biochips to the spatially resolved nanometric imaging of highly compartmentalized cell membranes. Of course, application of this latter methodology to biological imaging is subject to the problem of high absorption of infrared radiation by water. Superresolution Raman spectral imaging. Raman spectral imaging is a way around the absorption of infrared light by water in superresolution infrared imaging. It involves shining a laser at a sample and then
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1383
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW monitoring the scattered light that emanates from the sample at frequencies or wavelengths different from the excitation. Of course, most of the scattered light occurs at the same frequency; this is called Rayleigh scattering, which explains why the sky is blue (shorter blue wavelengths scatter better). Raman scattering is seven orders of magnitude smaller than same-frequency Rayleigh scattering. The problem is obvious: conventional aperture NSOM, which excites a small signal from a small aperture, has little probability of exciting detectable Raman scattering. Nonetheless, researchers have been successful under extreme conditions (long scan times and high Raman scatterers) in obtaining near-field Raman scattering30,31. Raman enhancement. Several findings from the past year indicate that near-field Raman scattering could be made a practical reality. The first is that of surface-enhanced Raman scattering (SERS). It was shown nearly 30 years ago that roughened silver surfaces result in enormous enhancements in Raman signals because of the excitation of surface plasmons in metals, such as silver or gold32. More recently, Emory and Nie33 and Feld and coworkers34 have demonstrated that, at certain locations, these surfaces can produce Raman spectra from single molecules. With this observation, interest has focused on trying either to attach a metallic nanoparticle to an AFM sensor or to produce a roughened silver AFM sensor. Using a unique method for producing a silver or gold nanoparticle at the tip of a force-sensing cantilevered nanopipette35, Sun and Shen36 have reported enhancements of 104 in the silicon Raman signal, using a MultiView 1000 and a standard micro-Raman 180º scattering geometry in which the laser beam is illuminated through the lens of the far-field microscope and this same lens collects the scattered light. In addition, Hartschuh et al.37 have reported that a silver metal wire roughened with a focused ion beam and attached to a tuning fork for feedback is capable of enhancing, by a factor of 103, the signal of carbon nanotubes, signals similar to those observed by Sun and Shen36 for a silicon sample. For their measurements, Hartschuh et al.37 used an inverted microscope that illuminated a transparent sample, with the tip in close proximity; in contrast, Sun and Shen36 used a standard micro-Raman geometry RM Raman Series (Renishaw) with an upright microscope complexed to the instrumentation described in Box 1. It should be noted that both silicon and carbon nanotubes are very strong Raman scatterers; nonetheless, the results obtained thus far are impressive. Shadow NSOM—a contrary approach to NSOM All of the studies in NSOM thus far have considered the problem of how one produces a point of light. No one has considered the problem from the point of view of shadowing a surface of a sample from a farfield light source and obtaining the information on the nano-optical properties of the shadowed region by difference imaging (Fig. 3h). For many applications, using linear optical processes such as one-photon absorption or fluorescence, shadow NSOM is most probably not applicable either because a very large background resulting from scattering may obscure the shadowed signal in the difference image or because, in the case of fluorescence, a very strong bleaching would occur in the exposed regions. However, when the unique features of Raman spectroscopy are combined with the integrated platform described Box 1 and the special properties of glass probes, shadow NSOM becomes an appealing possibility. Shadow NSOM requires a clear optical axis for the NSOM platform, a probe tip that is exposed to the optical axis, and independent motion of the sample and the probe tip. For these measurements a MultiView 2000 (Nanonics Imaging Ltd, Jerusalem, Israel) was used. From a micro-Raman point of view, shadow NSOM makes use of the excellent rejection of Rayleigh-scattered light found in Raman spectrometers
1384
a
Micrometers
b
Micrometers Figure 6 Shadow NSOM results. (a) Si-SiO2 difference Raman edge line scan of the tip in contact versus tip out of contact at each pixel. Each step is 200 nm and the marked edge has resolution between 100 nm and 200 nm. Raman data was obtained with a 50× objective, giving a 3-µm spot size of the exciting laser beam and a slit width of 50 µm. (b) A difference secondharmonic line scan of a bacteriorhodopsin (bR) feature that produces SHG. Once again, with a 50× illuminating objective, an edge resolution of between 100 nm and 200 nm is seen.
and of the superior capabilities of charge-coupled detectors (CCDs) with exceptional linearity in response, high dynamic range, high quantum efficiency and essentially zero dark noise. With such an integration, a shadow NSOM Raman image is obtained by placing on the sample a glass probe coated with any metal that has high opacity but does not show enhancement phenomena (Fig. 3h). This probe tip is brought into contact with a point on the sample with subnanometric AFM control, and a CCD spectrum is stored. Using the independent motion of the probe, it is then retracted without movement of the sample. At this point, another CCD spectrum is recorded and stored for the same pixel and the difference spectrum is computed to give the spectrum of what was shadowed by the probe tip. In one initial test with this approach, we obtained a 100-nm resolution in the line scan (Fig. 6a). This is still not close to what was discussed above for enhancement phenomena. What should be taken into consideration, however, is that the focal spot of the laser through the lens in this experiment was 3 µm because the objective used had only a 50× magnification. With higher-magnification objectives, focal spots as small as 1 µm or less should be readily achievable with the Renishaw InVia Raman microSpectrometer (Warsash Scientific) that was used. This should increase the resolution of the shadow NSOM image, because the size of the focal spot defines the noise in the measurement and a reduction in the noise by a factor of at least 4, as would occur with a reduction of laser spot size by 2, should bring the resolution of shadow NSOM close to what is possible with SERS enhancement methods. This number is also consistent with estimates based on rules of thumb for the sensitivity of difference Raman spectroscopy, which is estimated to be 1 part in 105.
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
REVIEW the effective z extent in near-field optics as compared with confocal or multiphoton imaging (zo of ∼0.7 µm can be achieved in Lens Lens fluorescence mode for a 50-µm detection aperture and a high-numerical-aperture Cylindrical piezo objective). Thus, from equation (1), an MDE xy value can be achieved at least ∼1,000-fold higher than what is presently obtained. As a er Scan head Las result, the FCS signal increases by a factor of e ∼1,000 and, in addition, the noise is diminished by the marked reduction in the illuminated volume. This latter point has been shown in initial experiments by Levene et al.40. d In addition, on-line AFM control in the illuCylindrical piezo mination reduces any noise that is due to f mechanical motion. Furthermore, there is a great advantage to having the illumination, rather than the detection, limited by an aperture. By limiting the Optical fiber illumination, one not only reduces wo but also, without the need for a detection aperture, zo. This reduction in z also reduces background Figure 7 Instrumentation of SPM and NSOM systems. (a) SPM instrument. (b–f) Integrated near-field, noise and thereby increases detection effifar-field, confocal and multiphoton microscope (e), its components (b–d) and carbon nanotube image (f). ciency while reducing out-of-focus bleaching. Finally, the outer diameters of cantilevered optical fibers are similar to glass intracellular Shadowing techniques should work for a variety of problems in electrodes. Thus, these probes can be inserted into cells with the great which the signal observed is removed from the exciting laser beam. This control of AFM for FCS measurements at any point within the cell. The could include photoluminescence either in systems that do not experi- control is achievable by the highly controlled applications of atomic ence bleaching (e.g., in inorganic materials) or in nonlinear optical phe- force, where the force per unit area (F/A) can be accurately monitored nomena (e.g., second-harmonic or third-harmonic generation, which to penetrate without damage and great precision into a region of are produced, unlike two- and three-photon fluorescence, through choice. excited levels that do not need to be populated). Initial results are preConclusions—a crucial bridge in integrated microscopy sented in Figure 6b. In this regard, it should be mentioned that exciting activity in non- Near-field optics has resulted in the highest resolution ever achieved in linear optical near-field phenomena includes very strong enhancement optical imaging, including a wide variety of imaging modalities from phenomena that have been observed in second-harmonic generation fluorescence imaging, to reflection and collection imaging, to Raman and even aperture-based approaches to such nonlinear near-field imag- imaging and even nonlinear imaging. There is no debate about this. ing (for excellent reviews, see refs. 35,38). In terms of nonlinear optical Achieving such high resolution was akin to climbing Mount Everest. Unheard of two decades ago, near-field optics is today not only a recenhancement phenomena, credit should be given to Wessel, who clearly ognized field, but also (and this is what is so exciting) an area in which pointed out the potential of such an approach39. an exponentially growing group of ideas is coalescing on how to obtain nanometer-scale optical information using the concepts of the optical Fluorescence correlation spectroscopy Some defined areas in which near-field optics should play a dramatic near-field in its broadest sense. To achieve these ideas, as has been docurole are still to be effectively explored. One such area is fluorescence mented in this article, requires SPM platforms that are totally intecorrelation spectroscopy (FCS), which can be used to monitor the grated into upright or inverted optical microscopes (preferably both). dynamics and concentration of fluorescent species. Near-field optics We have found interesting solutions to such integration and are sure has the potential for increasing the molecular detection efficiency that other solutions will be found. These platforms will open up the (MDE) in FCS by orders of magnitude. The MDE is given by the world of integrated microscopy, which is critical for the future of optical imaging. The important issue is not the question of what is the more expression: appropriate technique—whether NSOM, AFM, two-photon 2 2 (1) microscopy, second-harmonic generation, 4Pi or FCS—but the reality MDE ∝ e –2r /wo2 e –2z /zo2 that there is a great synergism in imaging modalities. A concept not touched upon in this paper in any detail, but which we where the beam size, wo , and the z penetration, zo, is much smaller in near-field optics than what is possible in confocal imaging, which is have addressed in other investigations41, is the classic problem of optipresently used for higher sensitivity. Near-field optics can achieve a cal imaging: optical imaging fits into a general class of physics problems beam size of 0.05 µm, as compared with 0.5 µm in confocal imaging— that are called inverse problems. Such problems have considerable a factor of 10 improvement in the MDE. import in a variety of areas, even those that are purely biological, such A further increase in the MDE, of a factor of over a 100, is achieved in as neural network analysis. These problems suffer from a lack of the near-field optical illumination because the radiation exits from the required information to obtain a complete solution. In terms of optical probe tip with a large divergence. This limits by a factor of at least 100 imaging, the required information is not provided by even the most Fiber probe
Silicon probe
a
b
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
PS
D
c
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1385
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW advanced emulations of far-field optics. On the other hand, an inverse problem is one in which if something is independently known about the object, then the imaging modality that is being applied can provide much more information on the object than without this additional information. For example, consider the problem of optical deconvolution, which transcends the mode of imaging that is being employed. It is applicable to confocal, CCD, 4Pi or multiphoton microscopy and requires accurate information about where an object ends. Because such accurate information is lacking, certain deconvolution algorithms rely on what is called blind deconvolution. In contrast, an online NSOM/AFM platform that does not perturb the imaging technique being used readily provides essential nanometric information on the optical borders of the object. With this on-line information, a closedloop form of deconvolution could be implemented in which the information provided yields a calculated object that is tested on-line with the NSOM/AFM system and further iterations are undertaken to improve the results of the calculation. Such an integration of imaging with computational methodologies should have an enormous impact on all forms of optical imaging, and near-field optics will be a bridge in this future evolution of the field. In summary, near-field optics, and its associated advances in instrumentation, are at the cusp of a rapid advance, much like what happened with AFM in the early 1990s. AFM and the techniques it has spawned have played a crucial role in nanotechnology; near-field optical imaging and the technologies it is spawning (e.g., nanopipette-based nanofountain-pen nanoprotein printing42) are likely to be equally crucial in providing information about molecules in cells, their interactions in space and time, and their involvement in the fundamental processes of biology. ACKNOWLEDGMENTS I would like to thank Menachem Segal for the use of his confocal microscope and for supplying a neuronal cell line, and The Horowitz Foundation, Israel Ministry of Science and Israel Science Foundation for their support. COMPETING INTERESTS STATEMENT The authors declare competing financial interests; see the Nature Biotechnology website for details. Published online at http://www.nature.com/naturebiotechnology/
1. Lewis, A. Isaacson, M. Harootunian, A. & Muray, A. Development of a 500-Å spatialresolution light-microscope. Biophys. J. 41, 405–406 (1983). 2. Lewis, A. Isaacson, M., Harootunian, A. & Muray, A. Development of a 500-Å spatialresolution light-microscope. 1. Light is efficiently transmitted through λ/16 diameter apertures. Ultramicroscopy 13, 227–231 (1984). 3. Pohl, D.W., Denk, W. & Lanz, M. Optical stethoscopy: image recording with a resolution λ/20. Appl. Phys. Lett. 44, 651–653 (1984). 4. Bouevich, O. Lewis, A. Pinnevsky, I. & Loew, L. Probing membrane potential with nonlinear optics. Biophys. J. 65, 672–682 (1993). 5. Lewis, A. et al. Second harmonic generation of biological interfaces: probing membrane proteins and imaging membrane potential around GFP molecules at specific sites in neuronal cells of C. elegans. Chem. Physics 245, 133–144 (1999). 6. Harootunian, A. Betzig, E., Isaacson, M.S. & Lewis, A. Superresolution fluorescence near-field scanning optical microscopy (NSOM). Appl. Phys. Lett. 49, 674–676 (1986). 7. Betzig, E. Trautman, J.K., Harris, T.D., Weiner, J.S. & Kostelak, R.L. Breaking the diffraction barrier—optical microscopy on a nanometric scale. Science 251, 1468–1470 (1991). 8. Karrai, K. & Grober, R.D. Piezoelectric tip-sample distance control for near-field optical microscopes. Appl. Phys. Lett. 66, 1842–1844 (1995). 9. Giessibl, F.J. Advances in atomic force microscopy. Rev. Mod. Phys. 75, 949–983 (2003). 10. Shalom, S., Lieberman, K., Lewis, A. & Cohen, S.R. A micropipette force probe suitable for near-field scanning optical microscopy. Rev. Sci. Instr. 63, 4061–4065 (1992). 11. Muramatsu, H. et al. Development of near-field optic atomic-force microscope for biological materials in aqueous solutions. Ultramicroscopy 61, 266–269 (1995). 12. Lewis, A. et al. New design and imaging concepts in NSOM. Ultramicroscopy 61, 215–221 (1995).
1386
13. Dunn, R.C. Near-field scanning optical microscopy. Chem. Rev. 99, 2891–2927 (1999). 14. Sader, J.E. Susceptibility of atomic force microscope cantilevers to lateral forces. Rev. Sci. Instr. 74, 2438–2443 (2003). 15. Lewis, A. et al. Failure analysis of integrated circuits beyond the diffraction limit: contact mode near-field scanning optical microscopy with integrated resistance, capacitance, and UV confocal imaging. Proc. Inst. Electric. Electron. Eng. 88, 1471–1481 (2000). 16. Benami, U. et al. Near-infrared contact mode collection near-field optical and normal force microscopy of modulated multiple quantum well lasers. Appl. Phys. Lett. 68, 2337–2339 (1996). 17. Toda, Y., Kourogi, M., Ohtsu, M., Nagamune, Y. & Arakawa, Y. Spatially and spectrally resolved imaging of GaAs quantum-dot structures using near-field optical technique. Appl. Phys. Lett. 69, 827–829 (1996). 18. Stuckle, R.M. et al. High quality near-field optical probes by tube etching. Appl. Phys. Lett. 75, 160–162 (1999). 19. Zhou, H., Midha, A., Mills, G., Donaldson, L. & Weaver, J.M.R. Scanning near-field optical spectroscopy and imaging using nanofabricated probes. Appl. Phys. Lett. 75, 1824–1826 (1999). 20. Oesterschulze, E., Rudow, O., Mihalcea, C., Scholz, W. & Werner, S. Cantilever probes for SNOM applications with single and double aperture tips. Ultramicroscopy 71, 85–92 (1998). 21. Haesliger, D. & Stemmer, A. Subwavelength-sized aperture fabrication in aluminum by a self-terminated corrosion process in the evanescent field. Appl. Phys. Lett. 80, 3397–3399 (2002). 22. Frey, H.G., Keilmann, F., Kriele, A. & Guckenberger, R. Enhancing the resolution of scanning near-field optical microscopy by a metal tip grown on an aperture probe. Appl. Phys. Lett. 81, 5030–5032 (2002). 23. Lewis, A. & Lieberman, K. Near-field optical imaging with a non-evanescently excited high-brightness light source of sub-wavelength dimensions. Nature 354, 214–217 (1991). 24. Sandoghdar, V. Beating the diffraction limit. Phys. World 14, 29–33 (2001). 25. Strinkovski, A. et al. Chemical applications of near-field scanning optical microscopy: Surface and near surface chemical imaging with conventional near-field optical probes and externally illuminated chemically active ion sensors. Israel J. Chem. 41, 129–137 (2001). 26. Papa, M., Bundmann, M.C., Greenberger, V. & Segal, M. Morphological analysis of dendritic spine development in primary cultures of hippocampal neurons. J. Neurosci. 15, 1–11 (1995). 27. Zenhausern, F., Oboyle, M.P. & Wickramasinghe, H.K. Apertureless near-field optical microscope. Appl. Phys. Lett. 65, 1623–1625 (1994). 28. Knoll, B. & Keilmann, F. Near-field probing of vibrational absorption for chemical microscopy. Nature 399, 134–137 (1999). 29. Labardi, M., Patane, S. & Allegrini, M. Artifact-free near-field optical imaging by apertureless microscopy. Appl. Phys. Lett. 77, 621–623 (2000). 30. Smith, D.A. et al. Development of a scanning near-field optical probe for localised Raman spectroscopy. Ultramicroscopy 61, 247–252 (1995). 31. Prikulis, J., Murty, K.V.G.K., Olin, H. & Kall, K. Large-area topography analysis and near-field Raman spectroscopy using bent fibre probes. J. Micros. 210, 269–273 (2003). 32. Chang, R.K. & Furtak, T.E. Surface Enhanced Raman Scattering (Plenum, New York, 1982). 33. Nie, S.M. & Emory, S.R. Probing single molecules and single nanoparticles by surface enhanced Raman scattering. Science 275, 1102–1106 (1997). 34. Kneipp, K. et al. Single molecule detection using surface-enhanced Raman scattering (SERS). Phys. Rev. Lett. 78, 1667–1670 (1997). 35. Barsegova, I. et al. Controlled fabrication of silver or gold nanoparticle atomic force probes: enhancement of second harmonic generation. Appl. Phys. Lett. 81, 3461–3463 (2002). 36. Sun, W.X. & Shen, Z.X. A practical nanoscopic Raman imaging technique realized by near-field enhancement. Mater. Phys. Mech. 4, 17 (2001). 37. Hartschuh, A., Sanchez, E.J. & Xie, X.S. Novotny high-resolution near-field Raman microscopy of single-walled carbon nanotubes. Phys. Rev. Lett. 90, 95503–95507 (2003). 38. Schaller, R.D. et al. The nature of interchain excitations in conjugated polymers: spatially-varying interfacial solvatochromism of annealed MEH-PPV films studied by nearfield scanning optical microscopy (NSOM). J. Phys. Chem. B 106, 5143–5154 (2002). 39. Wessel, J. Surface-enhanced optical microscopy. J. Opt. Soc. Am. B 2, 1538–1541 (1985). 40. Levene, M.J. et al. Zero-mode waveguides for single molecule analysis at high fluorophore concentrations. Science 299, 682–686 (2003). 41. Axelrod, N. et al. Near-field optical and atomic force constraints for superresolution 3D deconvolution in far field optical microscopy. in Three-Dimensional And Multidimensional Microscopy: Image Acquisition Processing VII. Proceedings of the Society of Photo-Optical Instrumentation Engineers Vol. 3919 (J.-A. Conchello, C.J. Cogswell, A.G. Tescher & T. Wilson, eds.) 161–169 (SPIE, New York, USA, 2000). 42. Taha, H. et al. Protein printing with an atomic force sensing nanofountainpen. Appl. Phys. Lett. 83, 1031–1033 (2003). 43. Lieberman, K., Ben-Ami, N. & Lewis, A. A fully integrated near-field optical, far-field optical, confocal and normal-force scanned probe microscope. Rev. Sci. Instr. 67, 3567–3576 (1996).
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
REVIEW
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
FOCUS ON OPTICAL IMAGING
FRET imaging Elizabeth A Jares-Erijman1 & Thomas M Jovin2 Förster (or Fluorescence) Resonance Energy Transfer (FRET) is unique in generating fluorescence signals sensitive to molecular conformation, association, and separation in the 1–10 nm range. We introduce a revised photophysical framework for the phenomenon and provide a systematic catalog of FRET techniques adapted to imaging systems, including new approaches proposed as suitable prospects for implementation. Applications extending from a single molecule to live cells will benefit from multidimensional microscopy techniques, particularly those adapted for optical sectioning and incorporating new algorithms for resolving the component contributions to images of complex molecular systems.
Biological phenomena are based on the fundamental physico-chemical processes of molecular binding, association, conformational change, diffusion and catalysis. The structural hierarchy established at the level of organelles, cells, tissues and organisms is imposed via an extensive network of cascade and feedback mechanisms based on these reactions. Thus, in order to perform ‘biochemistry in the cell’ it is imperative to elucidate the spatio-temporal distributions and functional states of the constituent molecules. Fluorescence microscopy is ideally suited to this task because it generates contrast by exploiting the many manifestations of light emission: sensitivity, selectivity, and modulation via reactions in the ground and excited electronic states. Of these, FRET (for a discussion of nomenclature, see Table 1.1 in ref. 1) is unique in providing signals sensitive to intra- and intermolecular distances in the 1–10 nm range. Thus, FRET is capable of resolving molecular interactions and conformations with a spatial resolution far exceeding the inherent diffraction limit (∼λ/2) of conventional optical microscopy, yet is also compatible with super-resolution techniques. This report is intended primarily as a guide to FRET in the imaging environment, although most of the concepts are applicable to solution studies as well. Space limitations preclude a survey of applications, for which the reader is referred to recent reports and reviews2–11. We present a somewhat revised formalism for the FRET phenomenon that offers certain advantages over standard analyses and provide a systematic classification of 22 different FRET methods (see also ref. 11). These include several new approaches of potential utility in the research and biotechnological laboratories. We conclude with a brief discussion of selected probe issues and anticipated future developments extending from single molecule to live cell applications. Fluorescence resonance energy transfer FRET is a process in which energy is transferred nonradiatively (that is, via long-range dipole-dipole coupling) from a fluorophore in an 1Departamento de Química Orgánica, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, 1428 Buenos Aires, Argentina. 2Department of Molecular Biology, Max Planck Institute for Biophysical Chemistry, 37077 Göttingen, Germany. Correspondence should be addressed to E.A.J.-E. ([email protected]) or T.M.J. ([email protected]).
Published online 31 October 2002; doi:10.1038/nbt896
electronic excited state serving as a donor, to another chromophore or acceptor. The latter may, but need not, be fluorescent. Recent monographs1,9,10,12,13 and two reviews by R. Clegg14,15 provide excellent and extensive coverage of this topic. The transfer rate kt (see Box 1 for definitions and basics) varies inversely with the 6th power of the donor-acceptor separation (r 6) over the range of 1–10 nm, as first demonstrated with peptides 40 years ago16. Such distances are relevant for most biomolecules or their constituent domains engaged in complex formation and conformational transition. The transfer rate also depends on three parameters: (i) the overlap of the donor emission and acceptor absorption spectra (parameter: overlap integral J); (ii) the relative-orientation of the donor absorption and acceptor transitions moments (parameter: κ2, range 0–4); and (iii) the refractive index (parameter: n–4, range ≅ 1/3–1/5). The quantitative treatment of FRET originated with Theodor Förster and is embodied in widely disseminated formulas for kt , the ‘Förster constant’ Ro, and the transfer quantum yield generally denoted as the energy transfer efficiency E (equation (1)).
(Ro/r)6 1 (R /r)6; R 6 = c κ 2Jn–4Q = c κ 2Jn–4(k τ ); E = k τ = ———–— kt = — ; o o o o f o t τo o 1 + (R /r)6 o
τo–1 = kf + knr + kisc + kpb; τ–1 = τo–1 + k t
(1)
where co = 8.8 × 10–28 for Ro in nm and J = 1017 ∫ qd,λ εa,λλ4dλ in nm6 mol–1; qd,λ is the normalized donor emission spectrum. As shown in equation (1), the unperturbed lifetime of the donor, τo, appears both in the denominator and in the numerator (second expression for Ro6). Thus, upon canceling terms one is left only with the radiative rate constant kf in the numerator. This quantity reflects inherent properties of the fluorophore, including solvation, and can generally be regarded as invariant under given experimental conditions13,17. It follows that the fundamental relationship established by Förster between k t and kf bears no necessary relationship to the reference donor lifetime τo or to the derived quantum yield Qo. That is, the inclusion of these quantities in the definition of Ro is arbitrary, and justified only because most, but not all, estimations of the transfer efficiency (E) are made by comparisons with the properties of the unperturbed donor. (It is interesting that in his widely cited English review18, Förster made no mention of Ro,
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1387
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW plasma membrane, and conformational rearrangements. Of central relevance in the latter instance is the orientational factor, κ 2, to which one almost universally assigns the value of 2/3. Unfortunately, this procedure is valid only if the donor and acceptor molecules are oriented randomly and rotate rapidly and isotropically during the donor excited-state lifetime. Such a condition may often or generally fail to exist, as with the visible fluorescent proteins (VFPs), the rotational correlation times (see below) of which are ∼fivefold their lifetimes, thereby greatly limiting the extent of rotational relaxation19,20. (It follows that Ros reported for the various VFPs (ref. 21) must be used with caution.) For a random yet static molecular distribution, the ensemble κ 2 is not even a constant but
although he commented on the remarkable absence of Planck’s constant from na→b, his expression for k t.) A major problem with cell biological applications of FRET, particularly those involving imaging techniques, is that the reference value τo is generally unknown and may vary continuously and arbitrarily throughout the sample, for example as a result of changes in the generally environment-sensitive knr. In addition, although the donor-separation distance r is of primary interest in most FRET experiments, one or more of the other parameters incorporated in the definition of Ro may also change or be of even greater functional significance. Possible examples would be molecular translocations between the cytoplasm and the Table 1 Methods for determining FRET in fluorescence microscopy Category
Method
Resonance energy transfer parameters
References
I. Donor quenching and/or acceptor sensitization Ia. Combined donor (D) and acceptor (A) emission signals Ia1
2,3 signals; spectra
Calibrated functions
3,11,14,15,36,50,66–69
Ia2
Normalized D/A ratio
a θ = QaRda; Rda ∝ I dda,d /I da,d
67,70
Ia3
Bioluminescence RET (BRET)
θ ∝ Ib /Ia
71
Ib. Fluorescence-detected excited state lifetime(s) (FLIM) Ib1
D lifetime
ρ = τ/τo
2,9,17,31,63,72–83
Ib2
Luminescence RET (LRET)
ρ = τada,d /τo
70
Ib3
Combined D,A lifetimes
Frequency domain: τφ, τm correlations
20,75
Ib4
Spectral FLIM (sFLIM)
D and A lifetimes as functions of λexc, λem
77,79,81
Ic. Donor intensity and intensity ratios Ic1
Intensity
ρ = I/Io
Ic2
On-off ratio
ρ = OnOffo /OnOff
Proposed
Excited state saturation
Ψ2 /Ψ1 – I2 /I1 ρ = ζ/ζ o ; ζ = —————— I2 /I1 – 1
Proposed
1 Qf (1 – ϕ)(1 – ϕo) Ilong θ = ——– —— ——————– ; ϕ = —— στTΨ Q isc ϕ – ϕo Ishort
Proposed
25,26,84,85
Ic3
Ic4
Ground state depletion (triplet)
Id. Donor depletion kinetics Id1
D pb kinetics (pbFRET)
ρ = τpb,o /τpb
Id2 Id3
Integrated D pb Intersystem crossing
ρ = ζ/ζ o ; ζ = I(t = 0)/∫o I(t)dt ρ = τisc,o /τisc
∞
25,84 Proposed
Ie. Acceptor depletion (adFRET) Ie1
Direct A pb (irreversible)
Combination with Ib1–2, Ic1–4, Id1–3, IIa1
26,86,87
Ie2
Photochromic A (pcFRET)
Combinations as in Ie1; example (with Ic1): 1 – Ion /Ioff ρ = 1 – ———————— αon – αoff(Ion /Ioff)
27,28
Ie3
A saturation (frustrated FRET)
Combinations as in Ie1; example (with Ic1): ρ = αsat (Iαsat /I + αsat –1)–1
Proposed
Ie4
Sensitized A pb kinetics (PES)
σ τpb,o ρ = 1 – —a ——– –1 σd τpb
30
II. Emission anisotropy IIa. Steady-state anisotropy IIa1
Donor anisotropy r¯
10,12,88
σ r¯o ρ = 1 – —a — –1 σd r¯
Proposed
IIa2
Acceptor anisotropy r¯
ro – r¯ ρ = ζ/ζ o ; ζ = ——— r¯ – r∞
IIb. Homotransfer, energy migration FRET (emFRET, P-FRET)
1⁄2
IIb1
Steady-state anisotropy r¯
ro (1 – γe γ2 π ⁄2erfc[γ]) Qo r¯ = —————————— ; γ = ——— 1 + τ/φ 1 + τ/φ
IIb2
Dynamic r (rFLIM, P-FRET)
Functions of ro, r∞, φ, τ
1
Γo3c —— 750
1,9,12,31,32,64,88–90 9,20,31,63,72,83,91
Subscripts refer either to species composition, excitation or photophysical process; a subscript ‘o’ refers to a reference state of either D (assumed unless otherwise indicated) or A, in which the other component is absent. Superscripts indicate whether emission is measured in the D or A spectral regions and assumes correction for spectral crosstalk (for example, D→A). See equations (1–4) for definition of terms and symbols (σ, Ψ, Q, θ, ρ). I, signal intensity; (Ia3) Ib, bioluminescence signal; (Ic4) Ilong and Ishort, signals at end and beginning, respectively, of a given exposure time. In some cases, the need for calibration (scaling) factors is indicated by the symbol ‘∝’; (Ie2) αon, fractional transition to the FRET-competent photochromic form of the A upon UV irradiation, and αoff, fractional transition to the FRET-incompetent photochromic form of the A upon visible irradiation; (Ie3) αsat, degree of light-induced formation of the FRET-incompetent excited state of the A; Iαsat , D intensity corresponding to αsat; (IIb1) see text for definitions; c in mM units.
1388
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
REVIEW A catalog of FRET microscopy methods In devising methods for exploiting FRET in r 5 microscopy, one is faced with two fundamental θ = Γ 10 o OnOff challenges: first, the formalism must be appro4 priate for quantifying FRET under conditions β = 10 1 0.1 of arbitrary, generally unknown, intramolecu3 1 lar and/or intermolecular stoichiometries, disQo = 0.9 tributions and microenvironments of donor 0.5 2 and acceptor; second, continuous methods of 0.1 0.1 observation (by FRET) are desirable in most 1 0.01 studies of live cells. Numerous other considerations dictate the choice of FRET techniques 0 0 0.2 0.4 0.6 0.8 1 0 1 2 3 4 5 6 for imaging purposes and lead us to the classiρ α = t/τ fication scheme given in Table 1. We include several new strategies (Ic2–4, Figure 1 Parametric FRET functions. (a) Förster factor θ as a function of a ratio function ρ. The latter involves combinations of parameters or functions equivalent to the ratio of the FRET-quenched and Id3, Ie2–3, Ie4, and IIa2) with potential for unquenched donor quantum yields, (equation (4) and Table 1). Operation in the linear irradiance implementation in fluorescence microscopy. regime (Fig. 2) is assumed. (b) FRET technique based on OnOff ratio (equation (4) in Box 1), which The methods are assigned to two groups (I and directly yields the fluorescence lifetime (via α) for any degree of saturation (β). II) depending on whether they are based on intensity and kinetic donor-acceptor relationships or on emission anisotropy. In many instances, as in the ‘ρ methods’ (Ib1–2, Ic1–3, Id1–3, Ie1–4, and IIa1–2), instead a function of the donor-acceptor separation r. A further issue is the donor-acceptor stoichiometry. In the case of a the reference measurement alluded to above (e.g., donor Io or τo) is donor surrounded by n equivalent acceptors, the operational Ro6 is ∼n required. It can be provided either by a separate region or sample, if times that for the single donor–acceptor pair, and for point-to-plane available, or by recourse to the various acceptor depletion strategies (Ie). transfer, the distance dependence varies with the 4th, and not the 6th, The determination of the Förster Factor θ by some other techniques power of the separation22. In short, the concept of a ‘constant’ reso- (Ia2–3, IIb1–2) does not require Qo as a scaling factor. nance energy transfer (RET) parameter, such as Ro, is not universally We stress the desirability of quantitative determinations by supplying equations based on the formalism introduced above. That is, we favor applicable. In our estimation, the relationship of k t to many of the experimental the view that the generation of secondary images representing FRETmeans for its determination (see Table 1) may be expressed more natu- related or FRET-derived parameters is the primary goal. The formulas rally by recourse to alternative formulations, and we propose the one differ as to whether they are restricted to the linear irradiance regime, and apply either to a single donor-acceptor pair or to arbitrary donorrepresented in equation (2). 6 acceptor stoichiometries and thus an ensemble of molecular species. k r (2) Generalization is possible but beyond the scope of this report; for Monte θ ≡ —f = — ; Γo6 = coκ 2Jn–4 ; Ro6 = QoΓo6 kt Γo Carlo simulations of some cases of Table 1, see ref. 11. In the following text, we provide brief explanations of the different ρ Q entries outlined in Table 1. Two points are worth emphasizing at the ; ρ = 1 – E = — or any other equivalent ratio or θ = Qo —— 1–ρ Qo outset. First, the methods with greatest sensitivity for low transfer effifunction. (3) ciencies—in some cases coupled with fast acquisition capability— include Ia2–3, Ib2 and Ie2–4. Second, methods differ with respect to We define the inverse proportionality constant between k t and kf as a their applicability in point-scanning as opposed to wide-field micro‘Förster Factor’ θ, and equate it to the 6th power of the ratio of the sepa- scopes. In well-defined (usually intramolecular) single-donor, single-acceptor ration distance r and a Förster constant Γo, in which Qo is absent. It is important to recognize that the other parameters defining Γo may also systems, fluorescence ratio measurements involving different spectral vary in particular experiments, either by experimental design or nature components (donor and acceptor signals) can be calibrated so as to yield of particular targets, or from changes in the inherent population dis- the FRET efficiency. These techniques (Ia1) are difficult to implement tribution of molecular states. In the latter case, appropriate ensemble because they require acquisition and registration of multiple images, averaging formalisms must be employed1,3,11,12–15. We now extend the formalism further (equation (3)) by relating θ to k ex = σ Ψ kf ρ, a ratio of experimental quantities (Table 1) proportional to the donor F + hc /λ1 F* hc /λ 2 quantum yields corresponding to the two conditions: donor with accepknr heat, other k d = τ –1 tor (presence of RET); and donor without acceptor (absence of RET). Ψ From equation (3) and Fig. 1a, it is seen that a reduction in donor Qo has two consequences. First, it displaces the transition inflection point, and high Ψ k f Ψ low Ψ thus the greatest sensitivity of ρ, from θ = 1 to smaller values (smaller r). Fluorescence (hc / λ 2 ) ∝ kf σQ Ψ; −1 ; Ψ + [σ τ ] And second, it restricts the operative dynamic range of the determinations to higher values of ρ. A further consideration, already stated earlier, is that the nonradiative decay pathway and thus Qo can also change Figure 2 The ‘Michaelis-Menten’ view of a fluorophore as a photon dramatically between alternative molecular states represented in a par- conversion catalyst or ‘enzyme’ (see Box 1). The two saturation curves depicted differ in στ by a factor of 3 (lower value in red). ticular FRET experiment.
a
b
6
Fluorescence
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
6
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1389
REVIEW F
F F
F
Donor fluorescence
F
F
F
S
S
F
F
Donor fluorescence
F
F
F
S
S
Absorbance
Absorbance Overlap kernel
300
400
500
600
700
200
300
Wavelength (nm)
400
500
600
700
Wavelength (nm)
FRET
Visible D
D
A
NA
UV FRET Off
FRET On
Off Fluorescence
Off Fluorescence
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
200
On
On 480
520
560
600
Wavelength (nm)
640
0
10
20
30
40
Cycles
Figure 3 Photochromic FRET (pcFRET). The chemical structures depicted correspond to the photochromic dithienylethene in the colorless open form (right upper) and colored closed form (left upper). The absorption spectrum of the latter overlaps well with the emission spectrum of the donor; the kernel of the overlap integral (striped) corresponds to the lucifer yellow donor selected for a model compound27. Ultraviolet light induces the photochromic transition to the closed form (On), and visible (green) light reverses the process to the open form (Off). Bottom: corresponding donor spectra and multiple cycles between the two states of the system. The photophysical scheme is represented in Figure 5c.
correction for spectrally overlapping donor and acceptor signals and direct excitation of the acceptor, as well as due consideration of variable donor-acceptor stoichiometries in the equations used to compute ρ and θ. The ratio of the quenched donor and sensitized emission signals (Ia2) is unique in being scaled by the acceptor instead of the donor quantum yield in the calculation of θ. By employing a bioluminescent donor (Ia3)—for example, luciferase as an expressed fusion protein—excitation by light is not required, thereby suppressing autofluorescence background and photobleaching. The direct determination of fluorescence lifetime (Ib1), either in the time or in the frequency domain, is one of the most direct measures of FRET. It is also relatively insensitive to variations in concentration and optical path length. Selection of a donor with a long lifetime and high transfer efficiency (Ib2), such as lanthanide-chelates, permit sensitive measurements of donor quenching (that is, of very low transfer efficiencies) because the (quenched) donor decay is monitored via the sensitized (shorter-lived) acceptor emission, thereby ensuring a low background. Sample/microenvironmental heterogeneity can be assessed with FRET by exploiting numerous formalisms for time- and frequency-domain measurements (Ib3), particularly if the fluorescent lifetimes can be correlated with continuous emission and/or excitation spectra via multiplexed, transform-encoded acquisition (Ib4). FRET determinations derived from intensity relationships (Ic1) require an accurate reference for the acceptor-free donor signals and are difficult to achieve in practice except, for example, in combination with acceptor depletion schemes (Ie) or recourse to appropriately fabricated
1390
nano-microstructures. The use of ‘dark’ acceptors falls into this general category. We introduce here three new approaches (Ic2–4) for the exploitation of intensity measurements by use of the photophysical principles outlined earlier and constituting indirect estimations of the fluorescence lifetime. The first (Ic2) is based on the OnOff function (see Box 1, equation (4)). This relationship provides a direct determination of τ for any degree of donor saturation (at saturation, the relationship is linear; see Fig. 1b). Implementation should be simple (e.g., using a detector with swinging dual integrated outputs). A second method (Ic3) is based on the displacement to the ‘right’ of the singlet saturation curve due to the FRET-induced shortening of the fluorescence lifetime (Fig. 2). Measurements are performed at two levels of irradiance (denoted by subscripts 1,2 in Table 1, Ic3), one of which has to be sufficiently high (that is, above the linear range). The indicated function involving the ratios of irradiances and signals yields στ. The acceptor should have a short lifetime to prevent its saturation via FRET. Distortions of the imaging point-spread-function may arise from donor saturation, particularly when using scanning systems; these effects can either reduce or enhance23,24 spatial resolution. In wide-field systems, such problems vanish although high-energy sources are required. Finally, in Ic4 we exploit the triplet lifetime τT, which can be extended from the microsecond characteristic of oxygen-saturated systems to the millisecond domain (depending on the fluorophore) by oxygen depletion via argon flushing or chemical reductants. Measurements performed after short and long exposure times, defined in relation to τT, differ by virtue of depletion of the singlet manifold to an extent reflecting FRET-induced changes in the fluorescence lifetime. FRET determinations based on donor depletion kinetics are also in widespread use. Donor photobleaching (Id1) occurs with a time constant that is inversely related to the donor quantum yield. Inasmuch as the process generally occurs on a timescale 6–12 orders of magnitude greater than the usual nanosecond range of fluorescence decay, the method is easy to implement. It also circumvents the registration problem of disparate images (assuming no sample movement) and the need for spectral overlap factors (as in Ic1), and does not require a fluorescent acceptor, although the latter must be photostable. Another reason accounting for the popularity of pbFRET is its good performance at low transfer efficiencies. A variant of pbFRET (Id2) introduced at the same time as Id1 (ref. 25) has not been adopted generally, despite its ideal suitability for detection with charge-coupled dipole (CCD) cameras. The underlying principle, first announced by the remarkable spectroscopist, the late Thomas Hirschfeld, is the invariance (quantum yield independence) of the total integrated emission during quantitative photobleaching of a fluorophore. The integrated image serves to normalize a corresponding initial quenched donor image of the same area. A third, new kinetic method (Id3) involves the measurement of the kinetics of ground-state depletion via intersystem crossing to the triplet state of the donor. As in Ic3, one requires conditions favoring the maintenance of a long triplet lifetime. The methods we have combined under the designation ‘acceptor depletion’ FRET (adFRET; Ie) are of fundamental importance because they permit the generation ‘in situ,’ that is, at every sample position, of the reference state required for many of the other techniques. Note that this category corresponds to ‘Γo (or Ro) engineering’ in the sense that one perturbs the system by altering the value of J (see equation (1)). In 1995, we implemented irreversible acceptor photobleaching (Ie1) upon noting the difficulty of performing donor pbFRET (Id1) with the very good FRET donor-acceptor pair of cyanine dyes Cy3–Cy5 (ref. 26). Cy3 is highly photostable (the newer version Cy3b even more so), whereas Cy5 photobleaches readily; it became apparent that
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW photobleaching the latter provided ‘before’ and ‘after’ images from which the FRET efficiency could be readily determined. This technique is used extensively due to its many virtues: (i) it is simple and rapid; (ii) only donor images are required, avoiding registration problems; (iii) there is automatic correction for pixel-by-pixel variations in the reference donor quantum yield (lifetime); (iv) it is very effective for high-transfer efficiencies (‘disappearing’ donor; see Figures 2 and 3 in ref. 26); (v) it works well with ‘dirty,’ that is, relatively impure acceptors; and (vi) it can be combined with Id1,2 (ref. 26). Photochromic FRET (pcFRET; Ie2) is the reversible equivalent of Ie1 and thus offers the prospect of continuous measurements with cellular samples. A photochromic acceptor is cycled repeatedly between FRET ‘competent (on)’ and FRET ‘incompetent (off)’ states by alternative exposures to visible and UV light27,28 (Fig. 3). Besides being reversible, pcFRET is superior to Ie1 in having a high quantum yield for photoconversion. That is, few absorbed photons are required to induce the interconversions between states, in contrast to ∼104–6 photons for irreversible photobleaching. We are adapting pcFRET for microscopy by optimized chemical design of the photochromic probes and incorporation of modulated light sources and detectors to permit very sensitive detection, especially of low FRET efficiencies. In addition, we have devised a scheme for applying the pcFRET concept in determinations of reaction kinetics (pcRelKin, patent applied for by the authors). This relaxation technique is potentially suited for very small volumes and high speed. Two new adFRET techniques are proposed here. In the first (Ie3), the acceptor is driven into saturation so as to ‘frustrate’ FRET23,29, and thereby restore the donor emission to its unquenched level. The use of modulated light sources (of which two are required) and phase-sensitive detection (with a lock-in amplifier) should provide a very sensitive measurement, particularly in laser spot scanning systems. A photostable long-lived acceptor is required. Another new and intriguing FRET method (Ie4), to our knowledge not yet applied in microscopy, was designed to detect extremely low FRET efficiencies in solution (the author claimed the potential for detecting an E of 10–4 over a distance of 20 nm30). This technique is based on the measurement of the photobleaching kinetics of a photolabile acceptor excited by a donor via FRET. The low background and high sensitivity are achieved by exciting the donor, preferably a fluorophore with a large Stokes shift, at a wavelength of minimal absorption by the acceptor. Emission anisotropy, a dimensionless quantity defined in terms of the two polarized emission signals arising from polarized excitation, provides a steady-state (r¯) or time-dependent measure of rotational diffusion and is thus sensitive to size, shape, association and motion. The parametric descriptors are the fluorescence lifetime, the rotational correlation time(s) φ, and the initial (ro ) and final (r∞ , limiting) anisotropies dictated by the intrinsic transition moments and molecular asymmetry, and the environmental anisotropy, respectively. The determination is based on signal ratios and thus shares with the lifetime a relative insensitivity to optical thickness, light intensity and concentration. We can identify at least two FRET methods based on fluorescence anisotropy. By virtue of the relationship between the rotational diffusion parameters, the donor r¯ is a function of the ratio τ/φ31; thus, a change in the lifetime will be reflected in r¯ (IIa1). We propose a second technique (IIa2) involving the selection of an excitation wavelength capable of exciting both the donor and acceptor and thus leading to both direct and indirect (FRET-sensitized) emissions of the latter. The sensitized component is virtually depolarized and thus the mean r¯ for the interrogated molecular population provides a measure of FRET. An acceptor with a large Stokes shift is required. There exists an obvious relationship to method Ie4. The emission anisotropy is also the basis of FRET determinations
FlAsH-Tetracysteine complex
Streptavidin VFP QDot Atto565 DAE
Cy3
DTPA-CS124-Eu Alexa488
IgG
2 nm
Figure 4 Comparative sizes of common fluorophores and protein carriers used in FRET imaging. Small molecules are represented with Chem3D Ultra (CambridgeSoft). Molecular structures of the small dyes were obtained from available crystallographic data or by minimization using molecular modeling (MM2). DAE, DTPA-CS-EU and ATTO correspond to the dithienylethene depicted in Figure 3, the complex of DTPA-carbostyryl 124 with europium70 and a representative of the Atto dye family (Atto-Tec), respectively. The FlAsH compound is shown as a complex with a 32 amino acid peptide containing a CCGPCC target39. The scale bar applies to all molecules. Information required for the depiction of the quantum dot (core, shell, and cap of Qdot 585 Streptavidin Conjugate) was kindly provided by Marcel Bruchez of Quantum Dot. We are greatly indebted to Reinhard Klement for the protein and peptide representations.
measuring the distribution of excited-state energy between identical molecules in close proximity, a process termed homotransfer (or energy migration) RET. EmFRET (our notation31) constitutes a sensitive measure of bulk concentration (in the 0.1–10 mM range) and/or of molecular association and clustering in solution (three dimensions) or in planar membranes (two dimensions). One can readily distinguish, for example, between dimerizing and monomeric VFPs with this technique32. The depolarization is due to the loss of orientational correlation between excitation and emission but leaving the ensemble lifetime and spectra unaltered. We have demonstrated emFRET in bacteria expressing VFPs20,31 and in signal transduction mediated by growth factors and their cognate receptor tyrosine kinases fused with VFPs32 (for other applications, see references in Table 1; IIb1,2). One great advantage of emFRET is the requirement for the expression in vivo of only a single VFP or other expression probe, as opposed to the requirement for two distinct donor and acceptor molecules in heterotransfer RET. In biotechnological applications, the ability to determine concentrations at the microscopic scale using a dimensionless parameter should be of considerable interest. In summary, both static (IIb1) and dynamic (IIb2) anisotropies can
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1391
REVIEW
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Box 1 Photophysical primer Q β e –α(1 + β) – 1 + α(1 + β) Q β(1 – e –α(1 + β)) = ————————————– ; PYoff = ———————– PYon 1+ β (1+ β)2
One can regard a given fluorophore as a photophysical catalyst that in a real functional sense shares many attributes of a protein catalyst (an enzyme). That is, the steady-state formalism of the familiar Michaelis-Menten kinetics applies directly to the transformation by a fluorophore F of its ‘substrate’—a photon (of wavelength λ1 and energy hc/λ1; h, Planck’s constant; and c, velocity of light)—that ‘binds’ (is absorbed) into a ‘product’— a photon at longer wavelength λ2 (and of lesser energy hc/λ2) with an efficiency dictated by alternative nonradiative pathways (Fig. 2). At low ‘substrate concentration,’ the rate of photon emission is linearly dependent on light intensity, or more precisely, photon flux Ψ (photons s–1 cm–2) = 5 × 1015 irradiance (W cm–2) × wavelength (nm). The photonic ‘KM’, the value of Ψ yielding half the maximal fluorescence signal, is given by [στ]–1, where σ is the absorption cross-section (a measure of photon capture probability, a quantity proportional to the decadic molar absorption coefficient ε; σ = 3.8 ×10–21ε), and τ is the first excited singlet state (S1 = F*) lifetime; τ = kd–1 = [kf + knr]–1; kf and knr are the radiative and nonradiative deactivation rate constants, respectively, excluding for the moment other competing processes described below. The initial slope (emission versus excitation photon flux), equivalent to the enzymatic kcat/KM, is given by σQ; the fluorescence quantum yield (Q) is defined as the ratio of emitted to absorbed photons or by the equivalent expression kf /kd = kfτ. The process saturates at high ‘substrate concentration’ (irradiance; Fig. 2) because the fluorophore is maintained in the excited singlet state (assuming the absence of a finite triplet steady-state population), thus yielding a maximal ‘turnover’ rate equal to kf. This maximal rate of fluorescence emission, given by the reciprocal of the radiative lifetime, is independent of Q and of the excitation light intensity and stability, implying that the most sensitive, quantitative, rapid and possibly simplest determination of molecular number, local density or concentration may often be achieved by operating at saturation instead of in the low, linear, range universally espoused for quantitative biological microscopy. The total photon yield/fluorophore/pulse (PY) for a rectangular excitation pulse of length t = ατ, and photon flux ψ = β/(στ) is given by equation (4), in which PYon and PYoff are the integrated photon emissions during the light (irradiation) and dark (post-irradiation decay) phases, respectively. For β >> 1 (the saturation condition): PY→ Q(1+α) and the ratio function OnOff →α. If α is also >>1 (that is, t >> τ) PY→ Qα ≡ kft, confirming the result derived above from the steady-state solution in Figure 2.
α 1 PYon – ——– OnOff = —— = —————– PYoff 1 – e –α(1 + β) 1+ β
Saturation can be achieved to any desired degree by selection of light pulses of a given repetition rate, duty cycle and duration. These parameters are generally selected so as to reduce background, triplet state buildup, photodestruction and generation of potentially cytotoxic photoproducts (see valuable discussions in refs. 23,92). One can minimize the latter two reactions by limiting the photon dose (irradiance × exposure time ∝ αβ). According to equation (4), a single fluorescein-like molecule (ε = 105 M–1 cm–1, τ = 4 ns, Q = 0.4) exposed to an 8-ns pulse of 0.2 nJ at 488 nm focused to an area of 1 µm2 (α = 2, β = 9.3) will on average emit 0.69 photons in the light phase and 0.36 photons in the dark phase; the OnOff ratio, 1.9, is very close to α, in accordance with the limiting cases given for equation (4) (see also Fig. 1b). The ratio of PY to a given irradiation ‘dose’ (αβ = 18.5 photons in the above example) constitutes a measure of ‘photon conversion efficiency’ and thus of signal-to-background contrast. In the event of significant contributions from scattering and short-lived luminescent components, one may wish to gate detection during the pulsed excitation cycle, thereby restricting the signal to PYoff. Photobleaching limits the number of cycles (photon ‘turnovers’) to ∼τpb/τ, in which τpb is the reciprocal photobleaching rate (Fig. 5). A typical value is ∼105 cycles (fluorescein), implying that ∼102 repetitions would be possible for single determinations based on 103 excitation pulses. On the other hand, photobleaching can also be exploited to obtain information, as in determinations of FRET (pbFRET, Id1 and Id2, Table 1) and of translational diffusion (FRAP, FLIP86 and FLAP93). To explore quantitatively the region of saturation (that is, depletion of the ground state), we are obliged to expand the formalism to account for transitions to and from the triplet state, RET between donor and acceptor fluorophores and photobleaching (Fig. 5a). The corresponding rate equations for a complete kinetic scheme are first-order except for the virtual second-order RET reaction involving donor* (D*) and acceptor (A) in the forward and acceptor* (A*) and donor (D) in the reverse direction. We circumvent this difficulty by representing the system in terms of transitions between donor-acceptor pairs in the different electronic states (Fig. 5b), thereby obtaining analytical expressions that permit the exploration of arbitrary degrees of saturation of both donor and acceptor (see also ref. 29).
Q β α + (1 + α – e –α(1 + β))β = PYon + PYoff PY = ————————————— (1+ β)2
report changes in conformation, association and FRET. These techniques are being implemented in numerous microscope systems, most recently in a confocal laser scanning microscope adapted with dual channel polarization detection32. Probes and strategies Solely from the standpoint of stability and brilliance of a fluorophore, –1 –1 one can define a ‘figure of merit,’ such as the product σkf Qf Q pb Q isc = 2 –1 σQf [τQblQisc] . The commercial sources stress σ and Qf, as exemplified by the cyanines (Amersham Biosciences), Alexa (Molecular Probes) and
1392
(4)
the long-wavelength Atto (Atto-Tec) series of dyes. However, other considerations apply depending on the FRET method adopted for use (Table 1). For example, in donor pbFRET (Id1,Id2) excessive photostability is undesirable, whereas for the methods based on ground state depletion by intersystem crossing (Ic4,Id3), Qisc must be finite. It may also be necessary to tailor the donor lifetime in relation to the dynamics of the particular process under investigation, and although in most cases a large Stokes shift is desirable so as to minimize crossover of the donor fluorescence into the acceptor emission band, a small Stokes shift is required for homotransfer FRET (IIb). Another consideration is the
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
REVIEW
k pb
a
B
D* k isc
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
A*
kt
k ex
T*
k nr kT
D
kf
A
b
tt ts
st tg sg
c sg
gs
ss
gg
gt
st
sgoff
ggoff
ts
tg
gt
ss gs gg tt
pc
gsoff
gson FRET
pc
ggon
sgon
Figure 5 Photophysical cycles including triplet state, FRET, and photobleaching. (a) Coupling of donor and acceptor. For simplicity, only the triplet and bleached states of the donor, D, are indicated; the rate constants for the acceptor, A, are also omitted. Definition of rate constants: kex = σψ; kisc, intersystem crossing (D*→T* transfer); kT = τ T-1, triplet decay; kt, FRET (D*→A* transfer); kpb, photobleaching (can also involve T*). For the usual nanosecond fluorophores lacking heavy atom substituents, kisc, kpb, kT are << kf, knr. (b) Photophysical state diagram based on twofold D-A combinations. Each pair consists of a D (left) and A (right) in any of four states: ground (g), excited singlet (blue, red and green s), excited triplet (t) and bleached (not shown). All transitions except FRET are depicted. (c) Cycle involving acceptor states excited exclusively via FRET (top), shown as the double-headed arrow (forward and reverse transfer) between sg and gs, and (bottom) of FRET-linked photochromic cycle depicted in Figure 3. Analytical, albeit complex, solutions for the corresponding time-dependent and stationary states for the schemes of b and c have been obtained with Mathematica, including arbitrary degrees of saturation of the D and/or A and explicit consideration of the FRET-excited acceptor species, thereby permitting the calculation of the time course of acceptor anisotropy.
means by which the probe is linked to the target molecule. One might wish to constrain the κ 2 conundrum by introducing flexibility, yet in anisotropy-based FRET determinations (IIb) rigidity is preferable, such as with the bisarsenylated fluorescein (FlAsH; see below) and bifunctional33 probes. Similar considerations apply to the acceptor, with which the spectral overlap and absorptivity of the acceptor dictate the sensitivity to distance modifications in a given range. However, in cases where the range of lower donor-acceptor separations is of interest, a smaller Γo (that is, J ) may be required so as to define a ‘working region’ optimized around the value θ = 1; see Figure 1a.
Current expression probes for determining molecular concentration, distribution and ‘age’ in vivo are primarily based on protein fusions with VFPs of jellyfish or coral origin (reviewed in refs. 4,5,7); other molecular FRET-active VFP derivatives are also available34,35. Recent developments in the VFP field, all of which are relevant for FRET, include mutants with increased spectral range, photoconversion capabilities, improved photostability and brightness, faster maturation rate and suppressed tendency to oligomerize. The advent of a photoactivatable VFP5 is of particular significance. Localized application of blue light to cells—two-photon activation may also be possible—permits the activation of a fluorescence signal at an arbitrary location and time, an invaluable feature for studies of protein translocation and association. Many new reporters based on donor-acceptor FRET pairs linked by a moiety that undergoes a conformational change upon binding or modification events have been devised4,5,7,34. The resulting perturbation of the FRET signal serves as a monitor of the underlying time-dependent process such as protein (de)phosphorylation or ion binding in the specific cellular compartment. A recent FRET-like addition to the VFP toolbox is bimolecular fluorescence complementation (BiFC), conceived as a means for assessing multiple protein-protein interactions in vivo with very low background36. Nonfluorescent fragments of spectrally distinct VFPs are fused to different proteins of interest. If the latter associate in the cell, the coupled VFP fragments associate and exhibit fluorescence after a maturation period. This technique joins related complementation strategies for studying protein-protein interactions, such as the protein fragment complementation assay (PCA), and the generation of fluorescence or bioluminescence by intein-mediated protein splicing of fragmented VFP or luciferase, respectively37. All of these approaches have potential for FRET-based enhancements or implementations, such as with bioluminescence resonance energy transfer in the case of luciferase (Table 1; Ia3). In all the VFP-based techniques, two fundamental problems must be faced, namely the possible functional consequences of overexpression and the need to distinguish between VFP-fused proteins delivered to their natural cellular compartment from nascent, reclaimed and degraded molecules elsewhere in the cell. Total internal reflection microscopy10,38 and other superresolution techniques (S. Hell, this issue) should be of great utility in this respect. New strategies and dyes with improved properties will expand the capabilities for in vivo FRET applications. The combination of exogenous probes and the expression of small peptide targets offer the advantage of greatly reduced size compared to VFPs (Fig. 4) and the versatility offered by the ligand in terms of lifetime, large Stokes shift or other property. One such system is based on specific hexapeptide sequences containing four cysteines and introduced into a target protein of interest. Application of an exogenous, membrane permeable, nonfluorescent probe, such as FlAsH or resorufin (ReAsH) derivatives, leads to binding to the tag and the generation of a specific fluorescent signal39. We have devoted great effort to developing functional derivatives of these very promising reagents but consider that additional chemical modifications are required to reduce the background in cellular applications (see also ref. 40). Other potential routes might exploit protein fusions with an anti-fluorophore single-chain antibody fragment41, novel protein scaffolds (‘affibodies’42), and a newly reported strategy for introducing unnatural amino acid side chains into proteins43. The latter may offer targets for a range of chemical and spectroscopic probes, including those suitable for FRET. Bioconjugated semiconductor quantum dots offer an alternative to organic molecules as fluorescence probes. Quantum dots are finding widespread application as labeling reagents for cells and macromolecules because of their unique properties and commercial availability44,45. Appropriately designed quantum dots are very photostable and
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1393
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW nontoxic, can be excited with one or more46 photons over a wide spectral range, yet emit in a narrow and programmable spectral range. Quantum dots are typically capped with a polymer bearing specific binding moieties, such as streptavidin, protein G, biotin or conjugatable chemical groups (available through such companies as Quantum Dot or Evident Technologies). We have demonstrated the utility of quantum dots as FRET donors in aqueous systems, a further property of these extraordinary materials that will undoubtedly lead to many applications. An important issue relates to size—whether ‘the tail wags the dog’—such that a quantum dot–linked probe would interfere or even abrogate the process under study. A graphical representation designed to convey the relative sizes of various common fluorophores and protein probes compared with a quantum dot is provided in Figure 4. One is struck by the extent of the IgG molecule, particularly considering the multistep signal amplification schemes in common use. Several other probes and small particles are of relevance to FRET applications: (i) transfer probes consisting of diffusible, tyramide-linked fluorophores or haptens rendered reactive by peroxidase fused to a target protein or antibody (e.g., the tyramide signal amplification system offered by Molecular Probes); (ii) photoreactive probes that phototransfer from a given protein to an (unknown) partner, for example, in a photocrosslinking reaction47 mediated by the PES (photochemical enhancement of sensitivity)-FRET mechanism (Table 1, Ie4); (iii) microspheres, nanocrystals and phosphors that can be evaluated on and in cells by virtue of attached ligands, morphology, spectroscopy and/or localization and functional effects; (iv) the photochromic probes described above (Table 1, Ie2; Fig. 3); (v) cascade FRET23,48 and photoinducible intramolecular charge transfer49 probes; and (vi) many versions of ‘dark’ acceptors with very high σ and consequently large Γo values. An important issue with these and other nonexpression probes is the method of introduction into the molecule and/or the cell. Perspectives Single-molecule spectroscopy based on fluorescence has developed since the pioneering work Thomas Hirshfeld in the 1970s and in many instances is implemented with imaging technology. FRET is an essential tool10,50,51 in this field, and should augment the high-resolution techniques recently exemplified in very elegant studies of myosin V dynamics33,52. Attempts to confine single-molecule measurements to nanocavities have succeeded in extending the operational range of fluorescence correlation spectroscopy (FCS) to the heretofore inaccessible micromolar range53 (W. Webb, this issue). Such cellular structures may be suitable for new FRET implementations in FCS54 and image correlation spectroscopy55, as well as two-dimensional FCS spectroscopy based on modulated excitation56, perhaps exploiting the dramatic enhancements of excitation energy transfer achieved in Fabry–Perot microresonators57. Scanning near-field optical microscopy (SNOM) provides many interesting imaging possibilities (A. Lewis, this issue), including a donor-coated ‘self-sharpening’ scanning tip limiting the extent of the corresponding acceptor array to tens of molecules58. The same laboratory has recently reported the integration of quantum dots as FRETSNOM sources with the prospect of single molecule resolution59, and a coherent mode of operation extending the transfer distance to 20 nm with implications for quantum computing60. In the case of cellular imaging based on FRET, one can predict great utility for highly integrated nanochambers61 and cellular microarrays62. We anticipate that many of the newer approaches for FRET microscopy outlined in Table 1 will come to fruition, hopefully integrated into the full array of emerging multidimensional microscopy techniques, particularly those adapted for optical sectioning. A concrete application representing a core research activity of our groups32 is the
1394
elucidation of ligand-mediated modulation of receptor-receptor distributions and dynamics63,64,65 on cell surfaces. The ongoing challenge is to expand further the algorithmic repertoire for dissecting such complex molecular systems into their component contributions. It is axiomatic that the full panoply of temporal, spatial and spectral resolution will be required. ACKNOWLEDGMENTS E.A.J.-E. is indebted to the Agencia Nacional de Promoción de la Ciencia y Tecnología (ANPCyT), Fundación Antorchas, Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), Secretaría de Ciencia, Tecnología e Innovación Productiva (SECyT) and the Universidad de Buenos Aires (UBA) for financial support. T.M.J. was supported by the Max Planck Society, European Union FP5 Projects QLG1-2000-01260 and QLG2-CT-2001-02278, and the Center of the Molecular Physiology of the Brain funded by the German Research Council (DFG). The authors were the recipients of a joint grant from the Volkswagen Foundation for their work on photochromic compounds and acknowledge the contribution of graduate student Luciana Giordano to the research depicted in Figure 4, as well as the efforts of many colleagues over the years in the general area represented by this review. They are also indebted professionally and personally for the inspiration offered by the late Gregorio Weber, the acknowledged father of fluorescence in biology. We thank Rainer Heintzmann, Pedro Aramendía, Carla Spagnuolo and Vinod Subramaniam for critical reading of the manuscript. COMPETING INTEREST STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/naturebiotechnology/
1. Wieb Van Der Meer, B., Coker, G. III & Simon Chen, S.-Y. Resonance Energy Transfer: Theory and Data (VCH, New York, 1994). 2. Hink, M.A., Bisselin, T. & Visser, A.J. Imaging protein–protein interactions in living cells. Plant Mol. Biol. 50, 871–883 (2002). 3. Hoppe, A., Christensen, K. & Swanson, J.A. Fluorescence resonance energy transferbased stoichiometry in living cells. Biophys. J. 83, 3652–3664 (2002). 4. Zhang, J., Campbell, R.E., Ting, A.Y. & Tsien, R.Y. Creating new fluorescent probes for cell biology. Nat. Rev. Mol. Cell Biol. 3, 906–918 (2002). 5. Lippincott-Schwartz, J. & Patterson, G.H. Development and use of fluorescent protein markers in living cells. Science 300, 87–91 (2003). 6. Meyer, T. & Teruel, M.N. Fluorescence imaging of signaling networks. Trends Cell Biol. 13, 101–106 (2003). 7. Miyawaki, A. Visualization of the spatial and temporal dynamics of intracellular signaling. Dev. Cell 4, 295–305 (2003). 8. Sekar, R.B. & Periasamy, A. Fluorescence resonance energy transfer (FRET) microscopy imaging of live cell protein localizations. J. Cell Biol. 160, 629–633 (2003). 9. Marriott, G. & Parker, I. (eds.). Biophotonics, Part A. Methods in Enzymology, vol. 360 (Academic Press, San Diego, CA, 2003). 10. Marriott, G. & Parker, I. (eds.). Biophotonics, Part B. Methods in Enzymology, vol. 361 (Academic Press, San Diego, CA, 2003). 11. Berney, C. & Danuser, G. FRET or no FRET: a quantitative comparison. Biophys. J. 84, 3992–4010 (2003). 12. Andrews, D.L. & Demidov, A.A. (eds.). Resonance Energy Transfer (John Wiley & Sons, Chicester, UK, 1999). 13. Valeur, B. Molecular Fluorescence: Principles and Applications (Wiley-VCH, Weinheim, 2002). 14. Clegg, R.M. Fluorescence resonance energy transfer and nucleic acids. Methods Enzymol. 211, 353–388 (1992). 15. Clegg, R.M. Fluorescence resonance energy transfer (FRET) in Fluorescence Imaging Spectroscopy and Microscopy (eds. Wang, X.F. & Herman, B.) 179–252 (John Wiley & Sons, New York, 1996). 16. Edelhoch, H., Brand, L. & Wilchek, M. Fluorescence studies with tryptophyl peptides. Isr. J. Chem. 1, 216–217 (1963). 17. Clegg, R.M., Holub, O. & Gohlke, C. Fluorescence lifetime-resolved imaging: measuring lifetimes in an image. Methods Enzymol. 360, 509–542 (2003). 18. Förster, T. Delocalized excitation and excitation transer in Modern Quantum Chemistry Part III: Action of Light and Organic Crystals (ed. Sinanoglu, O.) 93–137 (Academic Press, New York, 1965). 19. Volkmer, A., Subramaniam, V., Birch, D.J. & Jovin, T.M. One- and two-photon excited fluorescence lifetimes and anisotropy decays of green fluorescent proteins. Biophys. J. 78, 1589–1598 (2000). 20. Subramaniam, V., Hanley, Q.S., Clayton, A.H.A. & Jovin, T.M. Photophysics of green and red fluorescent proteins: implications for quantitative microscopy. Methods Enzymol. 360, 178–201 (2003). 21. Patterson, G.H., Piston, D.W. & Barisas, B.G. Förster distances between green fluorescent protein pairs. Anal. Biochem. 284, 438–440 (2000). 22. Kuhn, H. in Physical Methods of Chemistry, vol. 1 (eds. Weissberger, A. & Rossiter, B.) 579–650 (John Wiley & Sons, New York, 1972). 23. Schönle, A., Hänninen, P.E. & Hell, S.W. Nonlinear fluorescence through intermolecular
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
REVIEW energy transfer and resolution increase in fluorescence microscopy. Ann. Phys. (Leipzig) 8, 115–133 (1999). 24. Heintzmann, R., Jovin, T.M. & Cremer, C. Saturated patterned excitation microscopy (SPEM)—a novel concept for optical resolution improvement. J. Opt. Soc. Am. A 19, 1599–1609 (2002). 25. Jovin, T.M. & Arndt-Jovin, D.J. FRET microscopy: digital imaging of fluorescence resonance energy transfer. in Cell Structure and Function by Microspectrofluometry (eds. Kohen, E., Hirschberg, J.G. & Ploem, J.S.) 99–117 (Academic Press, London, 1989). 26. Bastiaens, P.I.H. & Jovin, T.M. Fluorescence resonance energy transfer microscopy in Cell Biology: A Laboratory Handbook, vol. 3, edn. 2 (ed. Celis, J.E.) 136–146 (Academic Press, New York, 1998). 27. Giordano, L., Jovin, T.M., Irie, M. & Jares-Erijman, E.A. Diheteroarylethenes as thermally stable photoswitchable acceptors in photochromic fluorescence resonance energy transfer (pcFRET). J. Am. Chem. Soc. 124, 7481–7489 (2002). 28. Song, L., Jares-Erijman, E.A. & Jovin, T.M. A photochromic acceptor as a reversible lightdriven switch in fluorescence resonance energy transfer (FRET). J. Photochem. Photobiol. A 150, 177–185 (2002). 29. Hänninen, P.E., Lehtelä, L. & Hell, S.W. Two- and multiphoton excitation of conjugatedyes using a continuous wave laser. Optics Comm. 130, 29–33 (1996). 30. Mekler, V.M. A photochemical technique to enhance sensitivity of detection of fluorescence resonance energy transfer. Photochem. Photobiol. 39, 615–620 (1994). 31. Clayton, A.H.A., Hanley, Q.S., Arndt-Jovin, D.J., Subramaniam, V. & Jovin, T.M. Dynamic fluorescence anisotropy imaging microscopy in the frequency domain (rFLIM). Biophys. J. 83, 1631–1649 (2002). 32. Lidke, D.S. et al. Imaging molecular interactions in cells by dynamic and static fluorescence anisotropy (rFLIM and emFRET). Biochem. Soc. Trans., 31, 1020–1027 (2003). 33. Forkey, J.N., Quinlan, M.E., Shaw, M.A., Corrie, J.E.T. & Goldman, Y.E. Three-dimensional structural dynamics of myosin V by single-molecule fluorescence polarization. Nature 422, 399–404 (2003). 34. Sato, M., Ozawa, T., Inukai, K., Asano, T. & Umezawa, Y. Fluorescent indicators for imaging protein phosphorylation in single living cells. Nat. Biotechnol. 20, 287–294 (2002). 35. Zacharias, D.A., Violin, J.D., Newton, A.C. & Tsien, R.Y. Partitioning of lipid-modified monomeric GFPs into membrane microdomains of live cells. Science 296, 913–916 (2002). 36. Hu, C.D. & Kerppola, T.K. Simultaneous visualization of multiple protein interactions in living cells using multicolor fluorescence complementation analysis. Nat. Biotechnol. 21, 539–545 (2003). 37. Ozawa, T. & Umezawa, Y. Peptide assemblies in living cells. Methods for detecting protein–protein interactions. Supramol. Chem. 14, 271–280 (2002). 38. Riven, I., Kalmanzon, E., Segev, L. & Reuveny, E. Conformational rearrangements associated with the gating of the G protein-coupled potassium channel revealed. Neuron 38, 225–235 (2003). 39. Gaietta, G. et al. Multicolor and electron microscopic imaging of connexin trafficking. Science 296, 503–507 (2002). 40. Falk, M.M. Genetic tags for labelling live cells: gap junctions and beyond. Trends Cell Biol. 12, 399–404 (2002). 41. Farinas, J. & Verkman, A.S. Receptor-mediated targeting of fluorescent probes in living cells. J. Biol. Chem. 274, 7603–7606 (1999). 42. Karlström, A. & Nygren, P.-A. Dual labeling of a binding protein allows for specific fluorescence detection of native protein. Anal. Biochem. 295, 22–30 (2001). 43. Chin, J.W. et al. An expanded eukaryotic genetic code. Science 301, 964–967 (2003). 44. Wu, X.Y. et al. Immunofluorescent labeling of cancer marker Her2 and other cellular targets with semiconductor quantum dots. Nat. Biotechnol. 21, 41–46 (2003). 45. Jaiswal, J.K., Mattoussi, H., Mauro, J.M. & Simon, S.M. Long-term multiple color imaging of live cells using quantum dot bioconjugates. Nat. Biotechnol. 21, 47–51 (2003). 46. Larson, D.R. et al. Water-soluble quantum dots for multiphoton fluorescence imaging in vivo. Science 300, 1434–1436 (2003). 47. Fancy, D.A. et al. Scope, limitations and mechanistic aspects of the photo-induced crosslinking of proteins by water-soluble metal complexes. Chem. Biol. 7, 697–708 (2000). 48. Haustein, E., Jahnz, M. & Schwille, P. Triple FRET: a tool for studying long-range molecular interactions. Chemphyschem 4, 745–748 (2003). 49. Sauer, M. Single-molecule-sensitive fluorescent sensors based on photoinduced intramolecular charge transfer. Angew. Chem. Int. Ed. Engl. 42, 1790–1793 (2003). 50. Michalet, X. & Weiss, S. Single-molecule spectroscopy and microscopy. C.R. Phys. 3, 619–644 (2002). 51. Ishijima, A. & Yanagida, T. Single molecule nanobioscience. Trends Biochem. Sci. 26, 438–444 (2001). 52. Yildiz, A. et al. Myosin V walks hand-over-hand: single fluorophore imaging with 1.5-nm localization. Science 300, 2061–2065 (2003). 53. Levene, M.J. et al. Zero-mode waveguides for single-molecule analysis at high concentrations. Science 299, 682–686 (2003). 54. Widengren, J., Schweinberger, E., Berger, S. & Seidel, C.A.M. Two new concepts to measure fluorescence resonance energy transfer via fluorescence correlation spectroscopy: theory and experimental realizations. J. Phys. Chem. A 105, 6851–6866 (2001). 55. Rocheleau, J.V., Wiseman, P.W. & Petersen, N.O. Isolation of bright aggregate fluctuations in a multipopulation image correlation spectroscopy system using intensity subtraction. Biophys. J. 84, 4011–4022 (2003). 56. He, Y., Wang, G., Cox, J. & Geng, L. Two-dimensional fluorescence correlation spectroscopy with modulated excitation. Anal. Chem. 73, 2302–2309 (2001). 57. Hopmeier, M., Guss, W., Deussen, M., Gobel, E.O. & Mahrt, R.F. Control of the energy transfer with the optical microcavity. Int. J. Mod. Phys. B 15, 3704–3708 (2001). 58. Shubeita, G.T., Sekatskii, S.K., Dietler, G. & Letokhov, V.S. Local fluorescent probes for the fluorescence resonance energy transfer scanning near-field optical microscopy. Appl. Phys. Lett. 80, 2625–2627 (2002).
59. Shubeita, G.T. et al. Scanning near-field optical microscopy using semiconductor nanocrystals as a local fluorescence and fluorescence resonance energy transfer source. J. Microsc. 210, 274–278 (2003). 60. Sekatskii, S.K., Chergui, M. & Dietler, G. Coherent fluorescence resonance energy transfer: construction of nonlocal multiparticle entangled states and quantum computing. Europhys. Lett. 63, 21–27 (2003). 61. Guijt-van Duijn, R.A. et al. Miniaturized analytical assays in biotechnology. Biotechnol. Adv. 21, 431–444 (2003). 62. Ziauddin, J. & Sabatini, D.M. Microarrays of cells expressing defined cDNAs. Nature 411, 107–110 (2001). 63. Tramier, M. et al. Homo-FRET versus hetero-FRET to probe homodimers in living cells. Methods Enzymol. 360, 580–597 (2003). 64. Krishnan, R.V., Varma, R. & Mayor, S. Fluorescence methods to probe nanometer-scale organization of molecules in living cell membranes. J. Fluoresc. 11, 211–226 (2001). 65. Wallrabe, H., Elangovan, M.A.B., Periasamy, A. & Barroso, M. Confocal FRET microscopy to measure clustering of ligand–receptor complexes in endocytic membranes. Biophys. J. 85, 559–571 (2003). 66. Garini, Y., Katzir, N., Cabib, D. & Buckwald, R.A. Spectral bio-imaging in Fluorescence Imaging Spectroscopy and Microscopy (eds. Wang, X.F. & Herman, B.) 87–124 (John Wiley & Sons, New York, 1996). 67. Jares-Erijman, E. & Jovin, T.M. Determination of DNA helical handedness by fluorescence resonance energy transfer. J. Mol. Biol. 257, 597–617 (1996). 68. Hiraoka, Y., Shimi, T. & Haraguchi, T. Multispectral imaging fluorescence microscopy for living cells. Cell Struct. Funct. 27, 367–374 (2002). 69. Elangovan, M. et al. Characterization of one- and two-photon excitation fluorescence resonance energy transfer microscopy. Methods 29, 58–73 (2003). 70. Selvin, P.R. Principles and biophysical applications of lanthanide-based probes. Annu. Rev. Biophys. Biomol. Struct. 31, 275–302 (2002). 71. Xu, Y., Piston, D.W. & Johnson, C.H. A bioluminescence resonance energy transfer (BRET) system: application to interacting circadian clock proteins. Proc. Natl. Acad. Sci. USA 96, 151–156 (1999). 72. Gadella, T.W.J. Jr., van der Krogt, G.N.M. & Bisseling, T. GFP-based FRET microscopy in living plant cells. Trends Plant Sci. 4, 287–291 (1999). 73. Schönle, A., Glatz, M. & Hell, S.W. Four-dimensional multiphoton microscopy with timecorrelated single-photon counting. Appl. Opt. 39, 6306–6311 (2000). 74. Yu, W., Mantulin, W.W. & Gratton, E. Fluorescence lifetime imaging: new microscopy techniques in Emerging Tools for Single Cell Analysis (eds. Durack, G. & Robinson, J.P.) 139–173 (Wiley-Liss, New York, 2000). 75. Harpur, A.G., Wouters, F.S. & Bastiaens, P.I.H. Imaging FRET between spectrally similar GFP molecules in single cells. Nat. Biotechnol. 19, 167–169 (2001). 76. Carlsson, K. & Philip, J. Theoretical investigation of the signal-to-noise-ratio for different fluorescence lifetime imaging techniques. SPIE Proc. 4622, 70–78 (2002). 77. Elson, D.S. et al. Wide-field fluorescence lifetime imaging with optical sectioning and spectral resolution applied to biological samples. J. Mod. Opt. 49, 985–995 (2002). 78. Gerritsen, H.C., Asselbergs, M.A.H., Agronskaia, A.V. & van Sark, W.G.J.H.M. Fluorescence lifetime imaging in scanning microscopes: acquisition speed, photon economy and lifetime resolution. J. Microsc. 206, 218–224 (2002). 79. Hanley, Q.S., Arndt-Jovin, D.J. & Jovin, T.M. Spectrally resolved fluorescence lifetime imaging microscopy. Appl. Spectrosc. 56, 155–166 (2002). 80. Calleja, V. et al. Monitoring conformational changes of proteins in cells by fluorescence lifetime imaging microscopy. Biochem. J. 372, 33–40 (2003). 81. Knemeyer, J.-P., Herten, D.-P. & Sauer, M. Detection and identification of single molecules in living cells using spectrally resolved fluorescence lifetime imaging microscopy. Anal. Chem. 75, 2147–2153 (2003). 82. Krishnan, R.V., Saitoh, H., Terada, H., Centonze, V.E. & Herman, B. Development of a multiphoton fluorescence lifetime imaging microscopy (FLIM) system using a streak camera. Rev. Sci. Instrum. 74, 2714–2721 (2003). 83. Siegel, J. et al. Wide-field time-resolved fluorescence anisotropy imaging (TR-FAIM): imaging the rotational mobility of a fluorophore. Rev. Sci. Instrum. 74 (2003). 84. Jovin, T.M. & Arndt-Jovin, D.J. Luminescence digital imaging microscopy. Annu. Rev. Biophys. Biophys. Chem. 18, 271–308 (1989). 85. Young, R.M., Arnette, J.K., Roess, D.A. & Barisas, B.G. Quantitation of fluorescence energy transfer between cell surface proteins via fluorescence donor photobleaching kinetics. Biophys. J. 67, 881–888 (1994). 86. Lippincott-Schwartz, J., Snapp, E. & Kenworthy, A. Studying protein dynamics in living cells. Nat. Rev. Mol. Cell Biol. 2, 444–456 (2001). 87. Kenworthy, A.K. Imaging protein–protein interactions using fluorescence resonance energy transfer microscopy. Methods 24, 289–296 (2001). 88. Matkó, J., Jenei, A., Matyus, L., Ameloot, M. & Damjanovich, S. Mapping of cell surface protein-patterns by combined fluorescence anisotropy and energy transfer measurements. J. Photochem. Photobiol. B 19, 71–73 (1993). 89. Runnels, L.W. & Scarlata, S.F. Theory and application of fluorescence homotransfer to melittin oligomerization. Biophys. J. 69, 1569–1583 (1995). 90. Yan, Y. & Marriott, G. Fluorescence resonance energy transfer imaging microscopy and fluorescence polarization imaging microscopy. Methods Enzymol. 360, 561–580 (2003). 91. Buehler, C., Dong, C.Y., So, P.T.C., French, T. & Gratton, E. Time-resolved polarization imaging by pump-probe (stimulated emission) fluorescence microscopy. Biophys. J. 79, 536–549 (2000). 92. Mathies, R.A., Peck, K. & Stryer, L. Optimization of high-sensitivity fluorescence detection. Anal. Chem. 62, 1786–1791 (1990). 93. Dunn, G.A., Dobbie, I.M., Monypenny, J., Holt, M.R. & Zicha, D. Fluorescence localization after photobleaching (FLAP): a new method for studying protein dynamics in living cells. J. Microsc. 205, 109–112 (2002).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1395
E R R AT A A N D C O R R I G E N D A
Erratum: Biopharmaceutical benchmarks—2003
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Gary Walsh Nat. Biotechnol. 21, 865–870 (2003) In Box 3 of the article on p. 868, it was stated that the Dutch company Pharming NV went bankrupt in 2001. In that year the company did go into receivership and its subsidary (Pharming intellectual property BV) was declared bankrupt. We fully accept, however, that Pharming itself did not go bankrupt. Pharming continues to trade successfully and currently has a number of products in both preclinical and clinical trials. NBT regrets the error.
Corrigendum: Phosphospecific proteolysis for mapping sites of protein phosphorylation Zachary A Knight, Birgit Schilling, Richard H Row, Denise M Kenski, Bradford W Gibson & Kevan M Shokat Nat. Biotechnol. 21, 1047–1054 (2003) The above report failed to cite a relevant paper by Gary Hathaway and colleagues at the California Institute of Technology (Pasadena, CA, USA) entitled “Identification of Phosphorylated and Glycosylated Sites in Peptides by Chemically Targeted Proteolysis,” which was published in the December 2002 issue of the Journal of Biomolecular Techniques (13, 228–237, 2002; http://jbt.abrf.org/cgi/reprint/13/4/228) while the Nature Biotechnology paper was in review. The two papers describe essentially the same chemistry aimed at producing aminoethylcysteine in place of phosphoserine or phosphothreonine residues for the purpose of generating proteolytic cleavage at sites of phosphorylation.
1396
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
P AT E N T S
Past versus present: the importance of tense in patent application examples Jane E R Potter & Gargi Talukder The Federal Circuit questions an example’s validity in Hoffmann-La Roche, Inc. v. Promega. arly this spring, in Hoffmann-La Roche, Inc. v. Promega, the United States Court of Appeals for the Federal Circuit reviewed a district court opinion on the validity of an issued patent1. The court found that drafting an example in the past tense, when the experimental work had not been performed as written, could be construed as inequitable conduct. Under some circumstances, inequitable conduct can render a patent unenforceable and even result in a monetary award against the patent owner. The court remanded the case to the district court to determine if the incidents of inequitable conduct sustained by the Federal Circuit on appeal justified finding the patent at issue unenforceable. This article addresses one aspect of the decision: the importance of accurately drafting the examples in a patent application. The required contents of a patent application2 include the specification, title, abstract, summary, drawings (if appropriate to the invention) and claims. A patent application need not have an examples section. However, most patents in the life sciences, biotechnology and pharmaceutical areas include one or more examples. Often the examples are written in a manner similar to the ‘materials and methods’ section of a scientific publication. Recently issued US patents including examples sections are 6,054,121; 6,251,594; and 6,414,130.
E
Examples in a patent application The examples serve at least two purposes.
Jane E.R. Potter and Gargi Talukder (summer 2003) are at Davis Wright Tremaine LLP, 2600 Century Square, 1501 Fourth Avenue, Seattle, Washington 98101-1688, USA. e-mail: [email protected]
First, they can provide the details of how experiments were performed, supporting data the applicant may rely on for utility (usefulness) and enablement (operability) of the claimed invention. Second, they can provide an outline of experiments that may be performed in the future. The past tense is used for examples disclosing experiments that have been performed as described. In contrast, ‘prophetic’ or ‘paper’ examples are written in the present or future tense. If these pro-
phetic experiments subsequently are performed as described in the patent application, the inventors can file the experimental data during prosecution of the patent application. Often this data is submitted in the form of a declaration under 37 CFR § 1.131 or 1.132, and the patent examiner will consider it in allowing the claims. Examples usually will not, and in fact should not, combine tenses, unless it is to illustrate variations in how an experiment could be performed.
Box 1 Dos and don’ts Scientists • Do use particular care when assembling protocols for use as examples. Make sure you tell the person preparing the application which protocols were actually performed. Distinguish these from the protocols for future experiments. • Don’t use the past tense when combining protocols. Combinations of protocols often do represent the best way of performing an experiment, but if the steps of the resulting experiment have not been performed as written by the time the patent application is filed, the example cannot be written in the past tense. It must be written in the present tense to signal to the examiner that it is a prophetic example.
Attorneys and agents • Do use the present tense for prophetic examples and the past tense for examples that were performed in the way they were written. • Don’t mix tenses within a single example. If an example is a prophetic example, use the present tense throughout. If the example was performed, stay with the past tense. Mixing tenses can lead to confusion about whether the experiments were performed as written, and this can be construed against the applicant.
Attorneys, agents and scientists • Do discuss future plans for further experimentation, and in the patent application, include one or more examples that track methods the way the scientist plans to perform them. Once the patent application is filed, inventors should let the attorney know when the experiments have been conducted. Experimental data may be helpful during prosecution of the patent application and can be presented to the examiner after the application is filed. For example, the data may help overcome prior art cited against the application, or may support utility or enablement of the invention as claimed.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1397
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
P AT E N T S The Manual of Patent Examining Procedure For interpreting the language in the examples, patent examiners refer to the Manual of Patent Examining Procedure (MPEP). In a section entitled “Simulated or predicted test results or prophetic examples,” the MPEP clearly distinguishes between prophetic and working examples: “Simulated or predicted test results and prophetical examples (paper examples) are permitted in patent applications. Working examples correspond to work actually performed and may describe tests which have actually been conducted and results that were achieved. Paper examples should not be represented as work actually done.” Moreover, “Paper examples should not be described in the past tense”3. Past versus present tense Hoffmann-La Roche, Inc. v. Promega emphasizes that any example written in the past tense is a signal that the experiment was actually performed in the way the example describes. More specifically, this means that any example written in the past tense must
1398
have been performed in the order in which the steps are written. After Hoffmann La Roche, Inc. v. Promega, hypothetical examples are still acceptable, but, as indicated in the MPEP, and by the court, they must be written “prophetically,”
After Hoffmann La Roche, Inc. v. Promega, hypothetical examples are still acceptable, but, as indicated in the MPEP, and by the court, they must be written “prophetically,” meaning in the present tense.
meaning in the present tense. The present tense is a signal to the examiner that the protocol may not have been performed, but that it nonetheless portrays a sequence of steps and materials for conducting the
experiment. Any example written in the past tense must have data supporting the implicit assertion that all steps of the example were performed in the order and way that they were written. If substantial evidence does not exist to support the fact that the example was performed as written, this misrepresentation can potentially render the patent unenforceable. Dos and don’ts Hoffmann-La Roche, Inc. v. Promega provides some guidelines for scientists to use when providing information to a patent attorney or agent, and for attorneys and patent agents to keep in mind when drafting the examples section of a patent application (see Box 1). Most importantly, there should be good communication between the groups to eliminate confusion about whether an example was performed or is planned for the future. 1. Hoffmann-La Roche, Inc. v. Promega, 323 F.3d 1354 (CAFC, 2003). 2. Described in 37 CFR § 1.71. 3. Manual of Patent Examining Procedure § 608.01(p)(II), August 2001.
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
P AT E N T S
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Recent patent applications in gene expression Patent #
WO 200364626
Subject
Assignee
Inventor(s)
Priority application date
Publication date
Double-stranded oligonucleotide compositions useful for inhibiting
Sequitur
Wiederholt KA, 1/7/2003
protein or gene expression, particularly in disorders such as psoriasis,
(Natick, MA, USA)
Woolf TM
A DNA sequence comprising a ribonucleotide reductase 1b promoter
Centre National
Chaboute M
1/25/2002
7/31/2003
E2F binding motif, useful in the genetic engineering of plant genomes,
Recherche
1/18/2002
7/31/2003
Palli SR
1/14/2002
7/24/2003
1/9/2002
7/24/2003
3/21/2002
7/17/2003
Markowitz VM
5/23/2000
5/29/2003
5/8/1996
4/24/2003
8/7/2003
HIV, hepatitis, cancer, autoimmune, inflammatory and cardiovascular diseases. WO 200362433
WO 200362450
particularly for inducing a S-phase or a meristem-specific gene
Scientifique
expression in plants.
(Paris, France)
Methods and compositions for correcting oligo probes by measuring
Syngenta
Chang H,
probes from each oligo probe during multiple hybridizations within a
Participations
Fan Y, Long F,
linear range and calculating a correction coefficient; useful for
(Basel, Switzerland)
detecting gene expression levels using probes. WO 200360103
WO 200360117
WO 200357169
Wang X, Zhu T, Zou G
A novel minimal DNA binding domain polypeptide comprising an
Rohm & Haas
amino acid sequence having a 67-residue amino acid sequence;
(Philadelphia,
useful for regulating gene expression in a host cell.
PA, USA)
A method for monitoring the expression of a specific gene in vivo
Japan Science and
Kokubo T,
consisting of measuring the nuclear magnetic resonance (NMR) signal
Technology Agency
Shirakawa M,
modification on addition of a material of external origin, which modifies
(Kawaguchi City,
Tame JRH
the NMR signal by accumulating a molecule detectable by NMR.
Japan)
Substituted tetracycline compounds useful for treating tetracycline
Paratek
Nelson ML,
responsive states, e.g., cancer, lung injury and stroke. The substituted
Pharmaceuticals
Ohemeng K
tetracycline compound blocks tetracycline efflux and modulation of
(Boston, MA, USA)
gene expression. US 20030100999
The integration of a multiple gene database, by defining a new class
Markowitz VM
and association for gene expression data, gene annotation data and sample data, grouping gene fragments into two groups, and annotating gene fragments to permit a database analytical engine to analyze the data. US 20030077639
US 20030077611
A synthetic ribozyme capable of selectively targeting androgen
University of Texas
Chen S,
receptor mRNA; useful in inactivating androgen receptor gene
System (Austin, TX,
Roy AK
expression and for treating prostate hyperplasia.
USA)
Methods for comparing gene expression profiles of two or more samples
Sention (Providence,
by using sample-specific primers for cDNA synthesis and amplification
RI, USA)
Slepnev VI
10/24/2001 4/24/2003
3/4/1998
of the synthesized cDNAs, and comparing levels of abundance of genes between samples. US 6553317
A computer system for biological applications with a relational database
Incyte
Akerblom IE,
having records containing information identifying the sequence of
Pharmaceuticals
Ament AD,
reagent clones nominated based on specific criteria; useful for studying
(Palo Alto, CA,
Cathcart R,
gene expression in different tissues or cells, e.g., normal and cancerous
USA)
Hodgson DM,
tissue.
4/22/2003
Ito LY, Jolley HE, Klinger TM, Lincoln SE, McKelligon BM, Panzer SR, Thanawala MK, Wong LC
Source: Derwent Information, Alexandria, VA. The status of each application is slightly different from country to country. For further details, contact Derwent Information, 1725 Duke Street, Suite 250, Alexandria, Va 22314. Tel: 1 (800) DERWENT ([email protected]).
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1399
PEOPLE
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Inpharmatica (London) has appointed Andrew Ayscough as director of chemistry. Dr. Ayscough was previously director of research at British Biotech Pharmaceuticals. In addition, the company has announced that a team of seven medicinal chemists has been recruited from Millennium Pharmaceuticals. The team will be based at Inpharmatica’s recently acquired Cambridge, UK, facility. Active Pass Pharmaceuticals (Vancouver, BC, Canada) has announced that Rod Campbell is leaving his position as CFO to pursue other opportunities. Concurrent with his departure, Jeff Tomlinson has been appointed chief business officer. Mr. Tomlinson served as director of strategic marketing and business development at Gene Logic, and in 2001, moved to GrowthWorks Capital, where he was vice president of investments. Resverlogix (Calgary, Alberta, Canada) has appointed William A. Cochrane chairman of its board of directors. A board member since May 2002, Dr. Cochrane is a former president of the University of Calgary, and has served as CEO and chairman of Connaught Laboratories and Stressgen Biotechnologies. He currently serves on the boards of Oncolytics Biotech, Andre’s Wines, Pheromone Sciences, Nucleus Bio Sciences, Medicure and Fox Energy. Mark Egerton has joined Oxagen (Abingdon, UK) as chief commercial officer and a member of the board. His most recent position was CEO of Cyprotex, and he also served as vice president of European business operations for Incyte. The Biotechnology Industry Organization (BIO; Washington, DC, USA) has named Barbara Glenn director of animal biotechnology and Hannah Highfill director for international market access. Additionally, Michael J. Phillips has been promoted to vice president, food and agriculture, science and regulatory policy. Dr. Glenn comes to BIO from the Federation of Animal Science Societies, where she served as executive vice president and chief scientific advisor. Ms. Highfill served most recently as the manager of international biotechnology education at the US Grains Council, and previously as a legislative correspondent for US Senator Max Baucus (DMT). Dr. Phillips most recently served as BIO’s executive director, food and agriculture. Paul H. Johnson has joined Nastech Pharmaceutical (Bothell, WA, USA) as senior vice president, research and development,
and chief scientific officer. Dr. Johnson was previously vice president, research and development, and chief scientific officer at EpiGenX Pharmaceuticals. Argenta Discovery (Harlow, UK) has announced the appointment of Richard Lingard as vice president, business development. He was previously director of business development at Inpharmatica. Five Prime Therapeutics (S. San Francisco, CA, USA) has named Gail J. Maderis president and CEO. Ms. Maderis was previously corporate vice president of Genzyme and president of Genzyme Molecular Oncology. She is a member of the advisory board of the Harvard Business School Alumni Biotechnology Roundtable, and previously served on the board of directors of the New York Pharma Forum and the scientific advisory board of the Canadian Medical Discoveries Funds. Lewis T. Williams, the founder of Five Prime Therapeutics, will continue in his role as executive chairman and as leader of the company’s research efforts. Sean Marett has been named vice president of business development and a member of the board at Lorantis (Cambridge, UK). He joins the company from Evotec OAI, where he served as chief business officer. ARIAD Pharmaceuticals (Cambridge, MA, USA) has announced that Paul J. Sekhri, former senior vice president in the global business development and licensing group at Novartis Pharma, has been appointed to the newly created position of president and chief business officer. Most recently, he was a partner at the Sprout Group, the venture capital arm of Credit Suisse First Boston. ARIAD has also named Athanase Lavidas, chairman and CEO of the Lavipharm Group, to its board of directors. Finally, the following promotions have been announced: Timothy Clackson is ARIAD’s new chief scientific officer, John Iuliucci has been named chief development officer, and Tomi Sawyer has been promoted to senior vice president, drug discovery. Microscience (Berkshire, UK) has appointed John St. Clair Roberts as medical director and as a member of the board of directors. Dr. St. Clair Roberts previously held the same positions at Xenova, and he served as medical director at Cantab.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1401
FOCUS ON OPTICAL IMAGING
NEW ON THE MARKET
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Imaging & microscopy
High-performance photomicrography
GeneFlash, from Syngene, is an accurate, lowcost gel documentation system featuring a camera, zoom lens and a compact darkroom that can accommodate any transilluminator. The camera is integrated to a graphical user interface rather than a computer, and enables users to see accurate images on a large color screen. To store images, GeneFlash uses the latest compact flash card technology with a USB flash card reader, eliminating the need to download files to a separate disc.
The Olympus Camedia C-5050 has a fivemillion-pixel, 1/1.8 inch CCD for high-performance photomicrography recording and archiving. Manual controls over aperture, shutter speed and focus offer maximum flexibility, and sharpness can also be left to an automatic contrast detection focusing system. The image-processing engine enables continuous shooting of either 11 images at 1.7 frames per second, or 4 shots at 3.0 frames per second. A tilting LCD monitor located on the rear of the camera makes image viewing easy when the camera is mounted on a vertical microscope port. Integrated DP-SOFT software eases image acquisition, archiving and analysis, and rapid transmission of images to coresearchers.
http://www.syngene.com/
http://www.olympus.com/
Desktop digital microscope
Fluorescent imaging solution
Nikon’s COOLSCOPE is the first self-contained, eyepiece-free, fully automated digital microscope in one tower. Set up is plug-andplay, and the system is network-enabled and offers remote control access and use. Users simply insert the specimen slide into the front of the unit, which automatically loads the slide and selects the appropriate settings for optimum brightfield viewing. The highly intuitive software allows users to observe the entire slide, magnify and mark specific parts of the specimen, and digitally capture, store and share images.
UVP’s GelDoc-It imaging system is designed to produce high-resolution images with excellent sensitivity and real-time viewing for acquisition and analysis of gels, plates, membranes and more. Features include a scientific-grade camera, FireWire connectivity and auto-exposure modes. The benchtop darkroom includes ultraviolet, white and blue transillumination, a gel viewer and a filter tray for exchange of filters to fit your application. Advanced LabWorks software completes the package for generating accurate quantitative data.
http://www.coolscope.com/
http://www.uvp.com/
Gel documentation
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1403
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEW ON THE MARKET
Forensic comparison
System for cell research
The Leica FS4000 combines the latest optomechanical developments in light microscopy with unsurpassed ease of use and ergonomics in an integrated solution. It allows highly precise comparison of two objects at magnifications up to 1,500× and gives reliable proof of slightest differences in microstructure, texture and color for applications such as forensic science. Different viewing modes, including side-by-side and superimposed image comparisons can be applied by pressing one button.
The DeltaVision Spectris, from Applied Precision, is a high-resolution image restoration microscopy system packaged in an ergonomically designed work envelope that is space and operator efficient. Proprietary XYZ stage design, excitation intensity monitoring, and a highly optimized optical path combined with sophisticated deconvolution algorithms allow scientists to acquire quantifiable fluorescence data. The system is particularly effective at imaging small, low-contrast, weakly fluorescent specimens, for quantitative comparisons of cellular compartments and time-lapse studies for cell motility, intracellular mechanics and molecular movement.
http://www.leica-microsystems.com/
http://www.appliedprecision.com/
Protein crystal imaging To eliminate time-consuming and error-prone manual protein crystallization experiments, RoboDesign has announced CrystalMation, a fully integrated, automated system for protein crystal storage, inspection and analysis. The system includes the RoboMicroScope II+, which features a high-resolution imaging system, cool diffuse fiber-optic backlighting, and RoboDesign’s patent-pending image processing method as well as a complete stand-alone plate-handling and machine vision system for protein crystal imaging. Once loaded with four plates, the RoboMicroScope II+ identifies each plate by reading its unique barcode, locates the protein drop within each well, zooms in on the drop and auto-focuses on it, images the drop, indexes each plate for imaging and stores each image to a database for later viewing.
Image analysis software
http://www.robodesign.com/
http://www.soft-imaging.net/
The analySIS Cell Imaging software series offers imaging solutions for the cell and molecular biology fields. The packages offer intelligent control of peripheral devices, efficient in-depth processing, automatic analysis, wellstructured image and data archiving, and automatic report generation. Using Cell Imaging, proteins or molecules in cells and tissues can be acquired and detected via fluorescence immunostaining. Fluorescence acquisitions can also be combined with DIC, brightfield or phase-contrast images. In addition, Cell Imaging offers measurement of fluorescence intensity, determination of colocalizations, deconvolution methods, standard-well analysis and also acquisition and evaluation of timelapse series.
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
1405
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
CAREERS AND RECRUITMENT
Professional master’s degree programs in regulatory affairs and biomedical quality systems E Dale Sevier1, Robert Wang2, Larry E Gundersen2 & A Stephen Dahms1,2 Two new programs are designed to supply biotechnology professionals with industry-specific skills required for successful biomedical product development.
ollowing years of research and development activities, a growing number of biotechnology companies are introducing new diagnostic and therapeutic products that are changing how medicine is practiced and thereby enhancing the quality of healthcare worldwide. Most of these involved are relatively small and are less than 15 years old1. As more companies continue to mature and shift their focus from R&D to product and manufacturing, the demand for individuals with industry-specific technical, management and team-based problem-solving skills has outpaced the supply of candidates. There are over 370 biotechnology companies involved in latestage clinical trials of new drugs, with many of them soon to be approved and ready for fullscale production2.
F
Biomedical workforce development issues Because all of the activities related to the development, testing, manufacture and marketing of these new biotech products are regulated by the US Food and Drug Administration (FDA; Washington, DC, USA) and its international counterparts, an awareness and understanding of these regulations are required on the part of not only regulatory affairs professionals, but also upper management, research scientists, process development technologists, manufacturing management and quality systems personnel. Often in the past, this knowledge was acquired on the job and under fire. It was not
1California State University Program for
Education and Research in Biotechnology and 2Center for Bio/Pharmaceutical and Biodevice Development, San Diego State University, San Diego, California 92182, USA. e-mail: [email protected]
unusual for a researcher to be told to “find out what is required to get FDA approval for our new product,” or a process development supervisor to be asked to “find out what is meant by ‘current Good Manufacturing Practices’ and see if we are able to comply.” In today’s world, such casual approaches to addressing regulatory requirements may be far less frequent, but still very few university graduates in the sciences are exposed to topics such as biotech product and process development, FDA regulations, quality systems or good clinical, laboratory or manufacturing practices, let alone management concepts, as part of their formal university education. A new approach to the master’s degree curriculum A master’s of science degree was established in 1990 at Northwestern University (Evanston, IL, USA) that included both the science and business of biotechnology, and serves as the model for the idea of a professional master’s degree that includes industry-specific skills, management skills and team-based approaches3. Since 1997, the Alfred P. Sloan Foundation has offered grants to support twoyear professional master’s degree programs in the sciences that prepare students for industrial careers4. Additionally in 1997, the Keck Graduate Institute (Claremont, CA, USA) was founded as a master’s degree–only graduate school designed to supply the biotech industry with skilled professionals in sciences who were knowledgeable in management, finance, regulatory, quality and intellectual property issues5. There are many other professional master’s degree programs that fit this model6. Instead of producing academicians, which is the traditional intention of graduate education, the professional master’s degree strives to
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
produce graduates who can apply their academic knowledge to practical work problems and who will be skilled in addressing aspects of both management and emerging problem areas of their chosen field. The aim of the professional master’s degree is to give students knowledge and then challenge them to use it in situations that mimic what they would encounter in the workplace. Regulatory affairs and biomedical quality systems At San Diego State University, two such professional science master’s degree programs have been developed by the Center for Bio/Pharmaceutical and Biodevice Development7 in the College of Sciences. The first program, introduced in 1999, is the master’s of science in regulatory affairs. This twoyear, 40-unit program has graduated 19 students and presently has 46 students enrolled working toward the degree. There are an additional 107 students who are not enrolled in the degree program but have enrolled in various courses. Most students are full-time employees of pharmaceutical, biotech or medical device companies and are seeking career opportunities or advancement in the regulatory affairs profession. The curriculum is designed to give the students a general appreciation of the background and regulations of the FDA relating to the approval of pharmaceutical, biopharmaceutical and medical device products and to their application to the research, development, approval and marketing of these products. Included in the curriculum are business courses that address practical issues in organizational behavior, operations management, communications and project management specifically applicable to the biotech industry. All but three of the required classes
1407
CAREERS AND RECRUITMENT
Table 1 Activities and FDA regulations applying to each stage of drug development Segment
Activities
Regulations
Prototype
Discovery research
GLP
Preclinical
Toxicity, dose studies
GLP, cGMP
Submit:
IND
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
Clinical
Human trials, safe and effective
GCP, cGMP
Submit:
NDA
Manufacturing
Process development and quality control
cGMP
Marketing
Labeling, advertising
NDA
are available as distance-learning courses. The program is designed for science managers who will be responsible for regulatory affairs. The second program, which is planned to commence in the fall of 2004, is the master’s of science in biomedical quality systems. The format is nearly identical to that of the regulatory affairs degree. It is a two-year, 43-unit program available through distance learning that will give students an understanding of the concepts of biomedical quality systems and science management and their practical application in an industry work environment. It is designed for science managers who will be responsible for process and product development, validation issues, quality systems and process improvements8. For both programs, the instructional faculty consists of experienced industry professionals who clearly understand and appreciate the biotech industry’s education and training needs and possess the necessary knowledge and practical experience to prepare students efficiently for an industry career. Biomedical product development The pharmaceutical, biotech and medical device industries are unique in that the FDA stringently regulates the development, testing, manufacturing and marketing of their products. This creates the need for scientists and technicians involved to acquire industry-specific skills that most often have not been a part of their academic training. What specific skills do scientists in these industries need to effectively fulfill the job responsibilities from discovery research through full-scale production? To answer this question one needs to consider (i) the regulatory requirements that need to be addressed as well as (ii) the tasks required to move a research project from concept through the various stages to scale-up and production. The regulatory requirements FDA regulations apply as soon as the research involves testing in human subjects. All of the regulations are designed not only to protect the patients in the clinical trials but also to ensure
1408
the continuous safety and efficacy of the drug over its lifetime. There are five regulations we will consider for this discussion: investigational new drug application (IND), new drug application (NDA), good laboratory practices (GLP), good clinical practices (GCP) and current good manufacturing practices (cGMP). Table 1 shows the activities and regulations required for each segment in the development of a pharmaceutical product. There are five main segments that will be discussed with respect to required regulations and the skills needed to complete the various tasks. Once the prototype is identified and preclinical studies begin, GLP regulations apply to the production of any reagents used in this and the clinical segment. After toxicity and dose studies are completed, an IND is submitted. The clinical segment is regulated by GCP and cGMP, and upon its successful completion, an NDA is submitted to the FDA. Investigational new drug. The IND requires a protocol for each specific clinical trial that includes a dosing plan, toxicity studies, risk analyses and documentation of any changes9. It also requires proper identification and documentation of the quality, purity, stability and strength of the investigational drug and its manufacturing process, including all labeling, reagents, solvents, catalysts and testing systems. A flow diagram of the manufacturing process is also desirable. The IND must contain descriptions and document the performance of test methods used to assure the identity, strength, quality and purity of the drug. New drug application. The NDA contains data and descriptions demonstrating that the drug is safe and effective in its proposed use(s) and that its benefits outweigh its risks. It also contains the drug’s proposed labeling (package insert), which outlines the proposed uses and limitations (these data are the basis for the marketing campaign, and all future marketing materials are defined as ‘labeling’ as far as the FDA is concerned). Data are also required that show conclusively that the methods used in manufacturing the drug, and the controls used to maintain the drug’s quality, are adequate to
preserve the drug’s identity, strength, quality and purity. Good laboratory practices. GLP establishes standards for the conduct and reporting of nonclinical laboratory studies, procedures and documentation intended to assure the quality and integrity of safety data submitted to FDA. As stated in the GLP regulations, the scope of these regulations “prescribes good laboratory practices for conducting nonclinical laboratory studies that support…applications for research or marketing permits for products regulated by the FDA, including food and color additives, animal food additives, human and animal drugs, medical devices for human use, biological products, and electronic products.”10 These regulations involve testing facilities, personnel, equipment design, reagents, animal care and laboratory operations. The responsibilities of the study director and the quality assurance unit are described in detail. The focus is on establishment of standard operating procedures (SOP) and operation of preclinical studies in a manner that conform to these SOP and the GLP regulations. Good clinical practices. In a similar manner, clinical trials must be performed in a manner that conforms to a series of regulations, collectively known as good clinical practices. These regulations are intended to assure the safety of human test subjects and to provide welldesigned and well-controlled clinical trials that will result in well-documented safety and efficacy data11,12. This involves clinical investigators and institutional review boards, as well as members of the study sponsor’s clinical research staff. Current good manufacturing practices. cGMP outlines the requirements for written procedures to approve or reject all components, facilities, equipment, instruments, containers and labels13,14. This includes written procedures for production and process control, which are designed to assure that the drug products have the identity, strength, quality and purity they purport or are represented to possess. cGMP also establish requirements for any specifications, standards, sampling plans, test procedures or other laboratory control mechanisms, including changes made in specifications, standards, sampling plans, test procedures or other laboratory control mechanisms. Master production and control records for each drug product must be prepared, dated and signed and independently verified, dated and signed by a second person. Again, conformance with these regulations involves the SOP and standardized testing methods. In addition to all marketed products, these regulations also apply to clinical supplies used in human clinical trials (the FDA recognizes that some limita-
VOLUME 21 NUMBER 11 NOVEMBER 2003 NATURE BIOTECHNOLOGY
CAREERS AND RECRUITMENT
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
tions of the application of these regulations to phase 1 and early phase 2 clinical trials exist). The tasks As stated earlier, new pharmaceutical products are subject to FDA regulations as soon as it is determined that the testing should begin in human subjects. Closely associated but preceding the preclinical segment is the prototype discovery segment (resembling ‘pure’ research, but goal oriented rather than knowledge oriented). Here, the lead drug candidates are identified, typically after an exhaustive screening of tens to hundreds of thousands of compounds. In the preclinical segment the lead drug candidates are further investigated for potential therapeutic activity, adsorption, distribution, metabolism and excretion in animal models. Results from the investigations of the preclinical segment form the foundation of an IND application that is submitted to the FDA. The IND is a request for permission to begin the clinical phases of the development project, a further investigation of the compound in human subjects. There are typically three clinical phases. Although studying the safety of the test compound is an objective in all clinical phases, in phase 1 clinical studies it is the primary focus. In phase 2 clinical studies, the primary focus is usually determining the proper dosing and demonstrating preliminary efficacy in humans. In phase 3 clinical studies, the primary objective is to demonstrate drug efficacy in a larger population of patients. If the clinical phases of a project are successful completed, an NDA is submitted to the FDA for final approval to market the product. The development process for medical devices is similar to that for pharmaceutical products, except that instead of a discovery segment and a preclinical segment, there is a design concept segment followed by a prototype design development segment and then submission of an investigational device exemption (IDE) to the FDA for permission to conduct clinical studies. Usually there is only a single clinical phase required for medical devices. Clearly, academia has provided the necessary level of technical training and understanding that industry expects their scientists to possess. In addition, though, there are other important skills that greatly enhance the contributions of industry scientists to a product development project. These include having the ability to be flexible and accept change; to be prepared to learn new disciplines; to be able to ‘multitask’; to participate in interdisciplinary teams; to communicate effectively both orally and in writing; to understand the business impact of their projects; and to understand and use project management tools.
It is often the case that a good scientist is promoted into his or her first supervisory position without ever having had any training to be a supervisor. Generally this will occur more frequently at smaller biotech companies than at larger ones. It is assumed that a person with an advanced technical degree is able to effectively prepare budgets, plan and manage projects, conduct employee evaluations and interface and communicate with various levels of management. Specific skills required Each of the five main segments shown in Table 1 has associated FDA regulations and tasks that are affected by those regulations. This section attempts to describe the relationship between the tasks, the regulations and the subject matter of the two professional master’s degrees, in regulatory affairs (MS-RA) and biomedical quality systems (MS-BQS). Prototype and preclinical segments. Also referred to as the discovery phase, the preclinical segment is basically the identification of a candidate drug. The MS-RA curriculum includes modules that explain how to classify new products, which departments of the FDA regulates them, and what documentation is required for each type and class. There are also modules that define and describe the mechanics of successful design and completion of toxicity and dosing experiments. The MS-BQS curriculum includes methods development strategies for the analysis, characterization, monitoring and documentation of the new drug and its components. Clinical segment. Here is where all of the human clinical trials are accomplished. The MS-RA curriculum has modules that describe how to design, run and document clinical trials at each of the three phases using GCP. The curriculum also includes a review of cGMP methods, documentation requirements and the mechanics of IND and NDA submissions. The MS-BQS curriculum includes process and product development strategies to ensure consistent manufacture of the drug at each phase by including statistical experimental design and statistical process control methodologies. Additionally, it describes the procedures for validation and auditing the early lots of a drug. Manufacturing and marketing segments. Each clinical trial phase described here must use materials and methods that are characterized and documented. FDA regulations require that any change in methods, components or venue of manufacture must be documented and justified. The MS-RA curriculum includes descriptions of post-approval activities such as marketing and labeling, compliance and FDA inspections. The MS-BQS curriculum includes
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 11 NOVEMBER 2003
process capability analysis and methods of process improvement and documentation. As mentioned earlier, science graduates receive little, if any, management training. Both of the master’s degrees require graduate management courses, including classes on organizational behavior, operations management and communication strategies along with other elective business classes. Just as important as the skills mentioned earlier, within an FDA-regulated industry it is critical for scientists involved in a product development project to have familiarity with the regulatory requirements pertaining to their activities. Without complete and adequate compliance with FDA requirements for preclinical studies, an IND will not be approved. Without complete and adequate compliance with FDA requirements for the conduct of clinical studies and process scale-up, the NDA may be rejected. Even after the FDA approves a product for the market, adequate compliance with FDA requirements for manufacturing is essential. Noncompliance with manufacturing requirements can result in a product being removed from the market and monetary fines into the hundreds of millions being exacted. Conclusions The biomedical industry can no longer afford to invest in long-term, in-house training efforts to produce managers with a solid science background as well as a knowledge of regulatory and quality issues. The development of professional master’s degree programs will enhance both science and management skills that the biomedical industry needs to continue the quest for new medicines, treatments and products that will improve our daily lives and our health. 1. Ernst & Young. Beyond Borders: The Global Biotechnology Report 2003. 2. Biotechnology Industry Organization. Editors’ and Reporters’ Guide to Biotechnology 2003–2004. 3. http://www.northwestern.edu/research/catalyst/ 1997/nucb.html 4. http://www.sloan.org/main.shtml 5. http://www.kgi.edu/ 6. h t t p : / / w w w. s c i e n c e m a s t e r s . c o m / s t u d e n t s / salphabetical.php 7. http://www.cbbd.sdsu.edu/regaffairs/index.html 8. Sevier, D.E. Producing a quality product: The importance of statistical design and control systems, teamwork, and definitions. Biochem. Educ. 30, 424–426 (2002). 9. http://www.fda.gov/cder/regulatory/applications/ind_ page_1.htm 10. http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/ CFRSearch.cfm?FR = 58.1 11. http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/ CFRSearch.cfm?CFRPart=50&showFR = 1 12. http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/ CFRSearch.cfm?CFRPart=56&showFR = 1 13. http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/ CFRSearch.cfm?CFRPart=210&showFR = 1 14. http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/ CFRSearch.cfm?CFRPart=211&showFR = 1
1409
