FEATURE

Ten years of Nature Biotechnology research

Monya Baker

Authors of some of the most highly cited Nature Biotechnology papers from the past 10 years discuss their work and the remaining challenges for their fields.

Ten years is not a long time in the life of an The interviews below provide a record of the Affymetrix, and colleagues had already devel- industry like biotech, where drug develop- rather remarkable progress that has been made oped the technique to build oligonucleotides ment or commercialization of a crop can take over the past ten years. on a solid matrix. Lockhart and his colleagues decades. But today’s research landscape is had to figure out how to prepare the sample, almost unrecognizable from that in 1996, when pick what oligos to put on the chip, separate Nature Biotechnology was relaunched from its Dawn of the gene expression array signal from noise and analyze results. Key to

http://www.nature.com/naturebiotechnology predecessor Bio/technology. Publication of the David Lockhart this was deliberately including oligos with human genome sequence still lay five years remembers that his single-nucleotide mismatches to eliminate off, cDNA arrays had just been described, hands shook that confounding signals from cross hybridiza- photolithographically synthesized oligonu- afternoon in early tion. cleotide chips were under investigation at an 1995 as he opened In the 1996 paper, Lockhart had monitored obscure company on the West Coast of the his laboratory note- the activity of 118 genes using 130,000 oligos , the term ‘proteomics’ had just book and pressed as gene-detecting probes. Today, Affymetrix been coined two years earlier by an Australian the phone to his sells a chip packed with millions of oligos researcher at a conference in Siena, Italy, and the ear. After months of that monitors all known human exons. With first transgenic crops had just been approved work, the Affymetrix the help of laser-mediated transfer, chips can for commercial use. scientist had been given a reference sample of also read a sample extracted from just a few Nature Publishing Group Group Nature Publishing 2 6 In the following article, Nature Biotechnology ten unidentified genes expressed at four dif- cells . But despite the vastly expanded read- interviews several of the authors who con- ferent levels. Now, Gene Brown from out, the chip is fundamentally the same, says 200

© tributed some of the most highly cited papers Institute (now part of Wyeth, North Andover, Lockhart. What’s different is what you can do from the past ten years. The papers cover a MA, USA) would tell him if the gene expres- with the data. diverse range of research areas—from array sion chip Lockhart’s team had designed gave “Even early on, you could see that certain technology, proteomics, molecular beacons, the right answers. After the 40th ‘correct,’ genes would go up and down,” he recalls, fluorescent proteins, quantum dots, embry- Lockhart’s colleague and officemate, Mark but making sense of that information was onic stem cells and gene delivery to agbiotech. Chee, tried to give him a celebratory ‘high often difficult because little was known The selection of research presented here is not five’ hand-slap, but furniture got in the way. about the identified genes. For example, in intended to be an exhaustive representation Over a year into the job, the researchers’ desks 1997, Lockhart and his collaborators used a of the most important advances in biotech and file cabinets were still in the same random gene expression array, this one containing from the past decade; it is simply a snapshot places where a moving crew had left them. over 260,000 probes, to detect differences in of the past decade’s most highly cited papers That blind test led to the first gene expres- expression levels in this journal according to the Institute for sion array and the most highly cited paper in of 6000 genes in Scientific Information (Philadelphia, PA, Nature Biotechnology1. Lockhart and his col- yeast growing in USA). Clearly, research areas not included in leagues had designed over a hundred thou- rich and minimal this article, such as modeling and systems biol- sand oligonucleotides as gene detectors and media and found ogy, chemical biology, antibody and protein synthesized them directly onto a high-density large differences engineering, biosensors, expression systems, array using massively parallel combinatorial in both known biochemical engineering and environmental chemistry. The tool could quantify mRNA and uncharacter- biotechnology, will continue to make impor- molecules’ abundance in cells over three ized genes3. “That tant contributions to the biotech endeavor in orders of magnitude and unambiguously used to be one of coming years. detect RNAs with frequencies as low as one the knocks, you in 300,000 or about one copy per mamma- find all these genes David Lockhart on microarrays: “Ten years lian cell. that are associated ago, the thinking was Monya Baker is a freelance writer based in San But making the chip wasn’t the most impor- with the phenotype. that that was way too Francisco, California. tant part. After all, Stephen Fodor, founder of Well what does that complicated.”

NATURE BIOTECHNOLOGY VOLUME 24 NUMBER 3 MARCH 2006 301 FEATURE

mean? Is it a cause? An effect? Something Proteomics goes global spectrometry quantitative as well as qualita- irrelevant coming along for the ride?” The late eighties and tive. “The general advance of the paper,” says Now, he says, people are better at design- early nineties were Aebersold, was “the widely perceived need to do ing experiments that determine the gene depressing times proteomic measurements quantitatively and a changes that matter. One way to do so is to for protein chem- range of robust methods to do that.” This ability combine genotyping data with gene expres- ists, recalls John shifted the overall goals of proteomics to study sion data and compare differences between Yates, director of the relative quantities of proteins, as well as how phenotypes. Then, researchers can manipu- Mass Spectrometry protein content changes temporally. late the expression of identified genes and Laboratory at Scripps Aebersold and his colleagues also devel- monitor the effects. This is precisely what Research Institute oped a solid-phase isotope tagging reagent that Lockhart did as part of a team looking for (La Jolla, CA, USA). allowed easier automation and higher sensitiv- insights into anxiety. Using gene expression Colleagues liked to tell him how DNA sequencing ity6. Additional refinements allowed the group profiling on six inbred mouse strains that and the consequent explosion of genetic infor- to determine the relative abundances of almost exhibit different levels of anxiety, the team mation were rapidly transforming his field into 500 proteins contained in the microsomal frac- identified 17 genes that correlated with a backwater. tions of naive and differentiated human myeloid anxious behavior. Then, they homed in on But Yates found inspiration rather than leukemia cells7. Since then, Aebersold and other two genes involved in the oxidative stress despair, exploring how the idea of whole genome groups have developed methods to identify and response. Overexpression of these genes sequencing could be applied to the proteome. quantify phosphorylated peptides. These pro- increased measures of anxiety, whereas inhi- His team coupled a system of digesting proteins vide a way to investigate how proteins are modi- bition decreased them4. Lockhart says that and separating the resulting peptides by cycles of fied (e.g., by phosphorylation) and processed to few people expected gene expression to aid chromatography followed by mass spectrometry control their own and other pathways8,9. the analysis of as complex a trait as anxiety. in a technique they call MudPIT (multidimen- Both Aebersold and Yates think the biggest “Ten years ago, the thinking was that that was sional chromatographic protein identification). barriers for proteomics are not technical, but

http://www.nature.com/naturebiotechnology just way too complicated. We were told a lot With this technique, peptide profiles from het- sociological. Yates cites a need to develop stan- of times that it would never work.” erogeneous mixtures of proteins can be com- dards across labora- Despite the exponential growth of genetic putationally matched with those of known tory groups to make information, expression profiling can still peptides. In 2001, it detected nearly 1,500 dis- experiments more turn up uncharacterized genes. Even changes tinct proteins in yeast cells, more than anyone repeatable. Aebersold in known genes require further study and vali- had ever seen before in a single analysis1. The thinks the field needs dation. Lockhart admits that it’s too early for previous record, of about 300 proteins, had also to shift away from the any drug to come from targets identified by been set by the Yates team using the more cum- shotgun approach. Reudi Aebersold thinks combined gene expression and genotyping bersome, and less sensitive technique of tandem “The key idea is to proteomics needs to assays, though of course gene expression is rou- coupling of two-dimensional liquid chromatog- generate once a com- shift away from the tinely used to identify drugs’ potential toxicities raphy2. plete proteomic map shotgun approach. Nature Publishing Group Group Nature Publishing

6 and off-target effects, as well as to characterize MudPIT could identify many types of pro- and then to navigate potential drug targets. teins that were previously hard to detect. “It was for all subsequent experiments in that mapped 200

© A more serious limitation is that gene expres- really good at analyzing membrane proteins,” space, rather than to perpetually rediscover the sion profiling only gives a snapshot of one level says Yates, “Just digesting them off the lipid proteome de novo in each experiment.” The two in the hierarchy of cell activity; it looks only bilayer, you could show what proteins were pres- goals are intertwined: once experiments are con- at levels of mRNA, not concentrations of pro- ent.” Subsequently, Yates refined techniques to sistently reproducible, investigators will no lon- teins, let alone proteins’ activation states. But recover more of the membrane proteins and to ger feel the need to rediscover the proteome. the fact that gene expression profiling raises identify proteins by phosphorylation state and questions that it can’t answer isn’t something subcellular location3. This ability helped identify 1. Washburn, M.P., Wolters, D. & Yates, J.R. Large-scale analysis of the yeast proteome by multidimensional that can be helped by better technology, says 67 previously uncharacterized proteins in the protein identification technology. Nat. Biotechnol. 9, Lockhart. nuclear envelope, several of which mapped to 242–247 (2001). 2. Link, A.J. et al. Direct analysis of protein complexes “The limitations really aren’t technical chromosome regions linked with various dys- using mass spectrometry. Nat. Biotechnol. 17, 676–682 4 anymore. The limitations these days are trophies . (1999). kind of old-fashioned, designing the right Identifying proteins is important, but so is 3. Wu, C.C., MacCoss, M.J., Howell, K.E. & Yates, J.R. A method for the comprehensive proteomic analysis experiments and getting access to the right quantifying them, says Ruedi Aebersold, a sys- of membrane proteins. Nat. Biotechnol. 21, 532–538 samples.” tems biologist at the Swiss Federal Institute of (2003). Technology in Zurich. Neither two-dimensional 4. Schirmer, E.C., Florens, L., Guan, T., Yates, J.R. III & Gerace, L. Nuclear membrane proteins with potential 1. Lockhart, D.J. et al. Expression monitoring by hybrid- gels nor liquid chromatography coupled to tan- disease links found by subtractive proteomics. Science ization to high-density oligonucleotide arrays. Nat. 301, 1380–1382 (2003). Biotechnol. 14, 1675–1680 (1996). dem mass spectrometry (LC-MS/MS) could 5. Gygi, S.P. et al. Quantitative analysis of complex pro- 2. Schutze, K. & Lahr, G. Identification of expressed genes do this, so Aebersold (then at the University tein mixtures using isotope-coded affinity tags. Nat. by laser-mediated manipulation of single cells. Nat. of Washington, Seattle) developed a way to Biotechnol. 17, 994–999 (1999). Biotechnol. 16, 737–742 (1998). incorporate different isotopes into otherwise 6. Zhou, H., Ranish, J.A., Watts, J.D. & Aebersold, R. 3. Wodicka, L., Dong, H., Mittmann, M., Ho, M.H. & Quantitative proteome analysis by solid-phase isotope Lockhart, D.J. Genome-wide expression monitoring chemically identical molecules to create iso- tagging and mass spectrometry. Nat. Biotechnol. 20, in Saccharomyces cerevisiae. Nat. Biotechnol. 15, tope-coded affinity tags (ICATs)5. The dilution 512–515 (2002). 1359–1367 (1997). of signal from the labeled molecule relative to a 7. Han, D.K., Eng, J., Zhou, H. & Aebersold, R. Quantitative 4. Hovatta, I. et al. Glyoxalase 1 and glutathione reduc- profiling of differentiation-induced microsomal proteins tase 1 regulate anxiety in mice. Nature 438, 662–666 known amount of unlabeled molecules reveals using isotope-coded affinity tags and mass spectrometry. (2005). how abundant that molecule is, making mass Nat. Biotechnol. 19, 946–951 (2001).

302 VOLUME 24 NUMBER 3 MARCH 2006 NATURE BIOTECHNOLOGY FEATURE

8. Zhou, H., Watts, J.D. & Aebersold, R. A systematic abnormalities in cultured hES cells4. “If 24, 185-187 (2006). approach to the analysis of protein phosphorylation. you’re really careful with the culture condi- 4. Draper, J.S. et al. Recurrent gain of chromosomes Nat. Biotechnol. 19, 375–378 (2001). 17q and 12 in cultured human embryonic stem cells. 9. Oda, Y., Nagasu, T. & Chait, B.T. Enrichment analysis of tions, the cells are quite stable. It’s a concern Nat. Biotechnol. 22, 53–54 (2004). phosphorylated proteins as a tool for probing the phos- that has to be managed,” he adds, “but it is 5. Herszfeld, D. et al. CD30 is a survival factor and phoproteome. Nat. Biotechnol. 19, 379–382 (2001). biomarker for transformed human pluripotent ultimately manageable.” stem cells, Nat. Biotechnol. 24, 345-355 (2006). “The odd abnormal cell may not be that much of a problem if it doesn’t have an The tabula rasa of advantage” that lets it outcompete healthy cells cells, agrees Pera. Nonetheless, abnormal Beacons of light It isn’t often that a cells will arise even in ideal culture condi- Ten years ago, a scientific publication tions; the key will be identifying and remov- pair of scientists get its author on the ing potentially dangerous cells before they described a new cover of Time, but are used in patients. In this issue, Pera and kind of probe for that’s what happened colleagues5 show that five hES cell sublines quantifying nucleic to Jamie Thomson with chromosomal abnormalities all express acids that could when his laboratory the CD30 receptor and that the protein’s do it in real-time at the University of expression in normal hES cells prevents and in solution1. Wisconsin derived the first human embry- apoptosis, but Pera says additional biomark- No purification, no onic stem (hES) cell lines1. Two years later, ers will be essential to weed out cells likely to gels, no radioactivity (and no background Nature Biotechnology published the second run amok. Understanding, and preventing, smudges to contend with). They did it by paper describing the production of hES cell the process that sets them down that path is combining the specificity of nucleic acid lines by Benjamin Reubinoff, Martin Pera more important. hybridization with the ability of nucleic and their colleagues at Monash University Of course, deriving and culturing hES cells acids to take on different conformations. 2 http://www.nature.com/naturebiotechnology in Melbourne . This paper confirmed is really a means to an end. The ultimate goal This was the birth of molecular beacons. In Thomson’s results, and went further, show- is understanding and controlling differentia- the intervening years, beacons (which are ing that hES cells could be differentiated in tion well enough that the cells can be used to fluorogenic hairpin oligonucleotide probes) vitro. generate tissue for human therapies. “We just have been applied in various ways—for The first cells were derived with a medium basically need more people beavering away at quantitative PCR, for pathogen detection, containing mouse fibroblasts, “which is kind it,” says Thomson, who believes that develop- for assaying single nucleotide polymor- of a standard tissue culture that you use when mental biologists will be able to make most phism and mutations, for incorporation you don’t know what you’re doing,” explains clinically relevant into self-reporting arrays, to name a few. Thomson, “because they secrete a lot of stuff. cell types within a The simple beacon itself has evolved by add- Then over the years, we’ve found that the decade. “Ultimately, ing gold particles to broaden its specificity things that do support mouse ES cells don’t we’re going to arrive and sensitivity, primers so that it can both Nature Publishing Group Group Nature Publishing

6 support hES cells.” Pera says hES cell culture at a molecular blue- prime and detect nucleic acid amplification has progressed on many fronts, particularly print of the pluripo- in a unimolecular and hence faster reaction, 200

© “refining the culture system to make it easier tent stage and then and even enzymes so that it can potentially to expand the cells and defining the medium we’ll know what prime, amplify and detect a sequence all in to eliminate animal products.” The remain- switches we need to one tidy little package. ing challenges, he says, are scaling up the cul- tweak,” agrees Pera, Sangi Tyagi and Fred Kramer of the Public ture and creating reliable techniques to grow though he is quick Health Research Institute (Newark, NJ, USA) up an entire culture from a single ES cell. to point out that the Martin Pera: “Stem were working at the time on Qβ replicase Already, Pera says, many promising culture ability to create dif- cell culture isn’t an amplification of RNA, which, like the poly- methodologies have been described; now ferentiated cells is absolute reflection merase chain reaction that later supplanted they need to be assessed by multiple research a very far cry from of embryonic it, had unacceptably high backgrounds when development; it’s sort groups. In January of this year, Thomson’s the ability to use the of a caricature.” working with low-abundance RNAs. They group described the first fully defined xeno- cells for therapy. But had the idea, based on Paul Lizardi’s (now free medium for culturing human stem cells he warns that there at Yale University) work on conformational and isolated two cell lines derived in fully will be many bumps in the road towards changes of nucleic acids, to create probes that defined medium, though both lines had creating differentiated tissues, stemming would allow amplification only when they chromosomal abnormalities3. largely from our ignorance of basic biology. changed their shape. But Thomson thinks that the major barri- “Stem cell culture isn’t an absolute reflec- Thus, Tyagi, Kramer and Lizardi designed ers for culturing hES cells have already been tion of embryonic development: it’s sort of a probe with a stem-loop structure that overcome. “There will be improvements,” a caricature.” unfolds when it reacts with its target he says, “but it will be kind of diminish- sequence, the complement for which is ing returns from now on.” In particular, 1. Thomson, J.A. et al. Embryonic stem cell lines encoded in the loop. In the first paper, Tyagi, derived from human blastocysts. Science 282, the twin specters of genetic instability and 1145–1147 (1998). Kramer and Lizardi showed how their fluo- tumorigenicity will be mastered, though not 2. Reubinoff, B.E. et al. Embryonic stem cell lines from rogenic probe, which they dubbed molecular human blastocysts: somatic differentiation in vitro. exorcised, through better technique, says Nat. Biotechnol. 18, 399–404 (2000). beacon, could assay the accumulation of a Thomson, who was part of the team that first 3. Ludwig, T.E. et al. Derivation of human embryonic target molecule in solution (a PCR fragment described the emergence of chromosoma stem cells in defined conditions. Nat. Biotechnol. of the HIV invertase gene), one of the first

NATURE BIOTECHNOLOGY VOLUME 24 NUMBER 3 MARCH 2006 303 FEATURE

real-time PCR assays in the literature1. The 1. Tyagi, S. & Kramer, F.R. Molecular beacons: probes Russian Academy of Sciences. Whereas oth- choice of the two dyes used in the original that fluoresce upon hybridization. Nat. Biotechnol. ers were combing beaches for biolumines- 14, 303–308 (1996). molecular beacons turned out to be provi- 2. Matayoshi, E.D., Wang, G.T., Krafft, G.A. & Erickson, cent organisms, Lukyanov and colleagues dential, according to Tyagi. By scouring the J. Novel fluorogenic substrates for assaying retroviral probed non-bioluminescent corals obtained proteases by resonance energy transfer. Science 247, literature, he found a fluorogenic peptide 954–958 (1999). from aquarium shops in Moscow and very 4 assay that used the fluorophore/quencher 3. Tyagi, S., Bratu, D.P. & Kramer, F.R. Multicolor quickly cloned six new GFP homologs . pair adenosine-5´-EDANS (5-[(2-aminoeth molecular beacons for allele discrimination. Nat. Though these proteins had only 30% or less Biotechnol. 16, 49–53 (1998). yl)amino]naphthalene-1 sulfonic acid) and 4. Nitin, N., Santangelo, P.J., Kim, G., Nie, S. & Bao, sequence homology with GFP, they still had DABCYL (4-(4-dimethylaminophenylazo) G. Peptide-linked molecular beacons for efficient the barrel conformation. Furthermore, one benzoic acid), used in the design of a prote- delivery and rapid mRNA detection in living cells. of these proteins, DsRed, fluoresced red, an Nucleic Acids Res. 32, e58 (2004). ase assay by scientists at Abbott Laboratories unprecedented color of emission, expand- (Deerfield, IL, USA)2. Later, when he started ing the breadth of wavelengths that could be experimenting with different dye combi- used. A few years later, two groups, headed by nations, he was able to create a palette of More bright ideas Atsushi Miyawaki and Benjamin Glick, used molecular beacons with different colored Fluorescent pro- random and directed mutagenesis to isolate dyes, the fluorescence from which could all teins have been used variants of yellow and the red fluorescent be quenched by a single ‘universal’ quencher. as tools to address a protein DsRed, leading to the development Thus, a few years after the first paper, Tyagi range of important of functionally improved forms5,6. and Kramer, together with Diana Bratu, had problems: as tags for Meanwhile, Lukyanov’s team had been a second Nature Biotechnology paper in which cell sorting to agents optimizing and expanding the uses of fluo- they described their palette, and showed how for tracking organ- rescent proteins. Realizing that even very multicolored molecular beacons could be elles, and as probes close homologs can have very different spec- used simultaneously in a single tube3. for studying the tral wavelengths, the team created a ‘photo- 7 http://www.nature.com/naturebiotechnology Since 1996, over 350 papers have been localization, expres- switchable’ protein that changes color from published on molecular beacons, five US sion and interaction of proteins. Dozens of cyan to green when irradiated by light of 405 patents have been issued, 45 companies have varieties exist, offering a range of colors and nm wavelength. This change in the ratio of licensed the technology, and several diagnos- intensities, some that can be activated and cyan to green can be used to track the move- tics have been approved, including the first deactivated at will. Some fluorescent proteins ment of tagged proteins, if fused with pro- real-time probe for HIV viral load measure- can be converted from one color to another teins and then activated in select subcellular ments. There were some pitfalls, particularly and back again. Multiple fluorescent proteins locations. for in vivo use that led some to question their can be used in a single cell. Lukyanov thinks fluorescent proteins will reliability, and sensitivity and penetrability It’s easy to forget that the complex applica- still yield many and more-effective tools. For remain challenging. Some scientists have tions and permutations began with a small, example, available far-red fluorescent pro- designed around the permeability prob- bioluminescent that lives off the teins are dim, and brighter versions need to Nature Publishing Group Group Nature Publishing

6 lem—adding a cell-permeating peptide, for coast of Washington State. Over a decade be invented or discovered. In addition, natu- example, has allowed Gang Bao of Emory ago, Douglas Prasher and colleagues showed ral fluorescent proteins usually function as 200

© University to detect cancer-related mRNAs that green fluorescent protein (GFP) from dimers or tetramers, limiting their uses in in live cells4—Tyagi has persisted with the Aequorea victoria could be used to track gene tagging proteins; monomeric variants of original design and instead found situations expression without any need for exogenous some of these have still not been made. that exploit its inherent properties. In fruit- substrates or cofactors1. However, the tech- Above all, Lukyanov thinks the biologi- fly oocytes, for example, specific mRNAs nique had practical limitations. cal function and evolution of these proteins are concentrated in different regions of the For one thing, natural GFP emits too little should be better studied, as the natural world cell. These clusters of RNAs can be detected light to be efficient for tracking proteins in could still yield exploitable ideas, including with molecular beacons injected into the eukaryotic cells. By shuffling GFP DNA in natural versions of photoswitchable, photo- oocyte. In his most recent study, Tyagi and and selecting for the bright- activatable and phototiming proteins. colleagues engineered an RNA molecule est colonies, Willem Stemmer and colleagues And although he expects existing protein with a kite tail-like addition possessing 96 boosted the protein’s signal about 40-fold variants to improve, tandem molecular beacon target sites, which and created a more soluble protein to boot2. he also believes enabled him to track That same year, George Phillips and his col- entirely new cat- individual mRNA leagues solved a high-resolution structure of egories will emerge. molecules from the GFP3. The structure showed the fluorophore In January this year, site of transcription near the center of a barrel-like β-can fold and his team described through the nucleus identified several adjacent amino acids that KillerRed, a geneti- to the cytoplasm. affect spectral properties, opening the door cally encoded pho- “Cell biologists are to rational engineering and mutagenesis. tosensitizer8 derived really excited,” he But human innovation is no match for from the chromo- says, as this work nature’s, and researchers were eager for protein of the hydro- Sanjay Tyagi, who Sergey Lukyanov, who with Fred Kramer, has provided new natural homologs to illuminate the mecha- zoan Anemonia was part of the original 8 developed the original insights into the nism of fluorescence and to expand proteins’ sulcata . When team that isolated concept of molecular dynamics of RNA spectral properties. One breakthrough came exposed to green fluorescent proteins beacons. metabolism. from Sergey Lukyanov and colleagues at the light, the protein can from corals.

304 VOLUME 24 NUMBER 3 MARCH 2006 NATURE BIOTECHNOLOGY FEATURE

kill cells and inactivate enzymes to which it dot, and provides have the greatest potential for in vitro bio- has been fused. Such innovations mark the a general platform medical applications, particularly more sen- rise of “a principally novel type of fluores- to add molecules sitive detection of biomarkers. In fact, Nie’s cent protein,” says Lukyanov, “that can be that target tumors, group has already developed microbeads used as active light-operated ‘manipulators’ enhance permeabil- encrusted with biodetecting quantum dots rather than passive reporters.” ity or bestow other emitting different colors and intensities5. properties. In nude Theoretically, these beads could generate 1. Chalfie, M., Tu, Y., Euskirchen, G., Ward, W.W. mice with human millions of distinct colors, a capacity beyond & Prasher, D.C. Green fluorescent protein as a marker for gene expression. Science 263, 802–805 prostate cancer any imaginable practical applications. So far, (1994). xenografts, the dots about 20 different types can be detected at 2. Crameri, A., Whitehorn, E.A., Tate, E. & Stemmer, accumulated in once, though Nie thinks the ultimate num- W.P. Improved green fluorescent protein by molecular Shuming Nie: “We evolution using DNA shuffling. Nat. Biotechnol. 14, tumors, allowing want something that is ber could be as high as a thousand. The 315–319 (1996). sensitive and mul- technology has already been commercial- 3. Yang, F., Moss, L.G. & Phillips, G.N. Jr. The molec- clinically realistic.” ular structure of green fluorescent protein. Nat. ticolor fluorescence ized by Crystalplex (Pittsburgh, PA, USA). Biotechnol. 14, 1246–1251 (1996). imaging. Nie thinks issues of penetration and toxicity 4. Matz, M.V. et al. Fluorescent proteins from nonbio- luminescent Anthozoa species. Nat. Biotechnol. 17, Others also started to explore the applica- will likely keep in vivo applications at bay for 969–973 (1999). tions of quantum dots in imaging. In per- several years, but he’s not giving up. In fact, 5. Nagai, T. et al. A variant of yellow fluorescent protein with fast and efficient maturation for cell-biological haps the first paper to apply quantum dots he is already putting his coatings onto near- applications. Nat. Biotechnol. 20, 87–90 (1999). to a surgical procedure, Harvard oncologist infrared quantum dots like the ones designed 6. Bevis, B.J. & Glick, B.S. Rapidly maturing variants of John Frangioni showed that quantum dots by Frangioni. the Discosoma red fluorescent protein (DsRed). Nat. Biotechnol. 20, 83–87 (2002). could transform a difficult, error-prone sur- 7. Chudakov, D.M. et al. Photoswitchable cyan fluores- gery used to detect cancer spread4—sentinel 1. Wu, X. et al. Immunofluorescent labeling of cancer cent protein for protein tracking. Nat. Biotechnol. marker Her2 and other cellular targets with semicon- 22, 1435–1439 (2004). lymph node biopsy—into a straightforward ductor quantum dots. Nat. Biotechnol. 22, 41–46 http://www.nature.com/naturebiotechnology 8. Bulina, M.E. et al. A genetically encoded photosen- procedure that no longer requires radioactive (2003). sitizer. Nat. Biotechnol. 24, 95–99 (2006). 2. Jaiswal, J.K. et al. Long-term multiple color imaging tracers. Key to this was the design of quan- of live cells using quantum dot bioconjugates. Nat tum dots that emit near-infrared light, which Biotechnol. 21, 47-51 (2003). is scattered less by tissues than is visible light. 3. Gao X. et al. In vivo cancer targeting and imaging with semiconductor quantum dots. Nat. Biotechnol. Quantum leaps Combined with an imaging system that inte- 22, 969–976 (2004). Quantum dots have grated visible and near-infrared light, the 4. Kim S. et al. Near-infrared fluorescent type II quan- properties imagers dots clearly identified sentinel lymph nodes tum dots for sentinel lymph node mapping. Nat. Biotechnol. 22, 93–97 (2004). long for. The nano- as deep as 1 cm below the skin in a living, 5. Han, M., Gao, X., Su, J.Z. & Nie, S. Quantum-dot- crystals absorb a 35-kg pig, allowing the nodes to be removed tagged microbeads for multiplexed optical coding of biomolecules. Nat. Biotechnol. 19, 631–635 broad spectrum of with minimal dissection. (2001). light, but the light But Frangioni doesn’t expect quantum Nature Publishing Group Group Nature Publishing

6 they emit can be dots to guide human surgery anytime soon, tuned to any narrow if ever. Instead, he hopes to replace the dots 200

© frequency. Compared with albumin attached to indocyanine green, The delivery problem to organic dyes, they a near infrared fluorescent dye. These probes Delivering drugs, can be 20 times brighter and orders of mag- won’t work as well as the quantum dots, nucleic acids and nitude more stable to photobleaching. But which can be engineered to a size that will proteins into cells in at several nanometers across, the dots are lodge in the sentinel nodes, but that’s not the a safe and efficient closer in size to proteins than small mol- point. “Now that we’ve proved principle,” he manner has been, ecules; a quality that could limit their uses says, “we want something that’s clinically and remains, one in cells. In 2003, researchers at Quantum realistic.” of the most chal- Dot (Hayward, CA, USA) showed that dots Frangioni thinks near-infrared systems lenging problems could label cancer markers, actin and nuclear will eventually image tissues as deep as 4 in biotech. Robert antigens inside cancer cells1. The same year, cm below the skin, but that penetrance will Debs still wants to researchers at Rockefeller University (New depend not on fluorophores, but on detec- get DNA into cells, but he’s largely through York) showed that antibodies could be con- tion systems. For quantum dots, he says, the fiddling with cationic liposomes. In 1997, jugated to quantum dots, and that cells tak- most significant barrier is potential toxicity. Debs, of San Francisco’s California Pacific ing in the conjugates grew for over a week2. Dots are typically made from heavy metals. Medical Center Research Institute, reported In 2004, in what one commentator hailed as They are too big to be removed by the kid- that big, multilayered vesicles studded with a possible precursor to an ‘optical biopsy,’ ney and too foreign to be metabolized by the cholesterol could improve the efficiency of a team led by Shuming Nie, director of the liver, so they remain in the body indefinitely. gene delivery in living animals by as much Emory-Georgia Tech Nanotechnology Center To create a formulation suitable for humans, as 1,740-fold1. A few months later, Nancy (Atlanta, GA, USA) demonstrated for the Frangioni and Massachusetts Institute of Templeton and her colleagues, at Baylor first time that quantum dots could be used to Technology collaborator Moungi Bawendi College of Medicine in Austin, Texas, showed image cancer in living animals3. To do so, Nie are developing smaller dots made from dif- how processes like sonication, heating and and his team encased the quantum dots in an ferent elements. optimizing DNA/liposome ratios could boost amphiphilic triblock shell. This allows mul- Over the next five to ten years, both systemic gene delivery2. Debs and colleagues tiple functionalities to be attached to each Frangioni and Nie believe quantum dots were subsequently able to increase efficiency

NATURE BIOTECHNOLOGY VOLUME 24 NUMBER 3 MARCH 2006 305 FEATURE

by a couple of orders of magnitude more, have done this job for millions of years, and I Silence is golden but found that modifications that made the trust more evolution than [rational] design” Molecular biolo- liposomes more efficient at delivering DNA no matter how smart the creator of the non- gist John Rossi also made them more toxic. viral vector, he says. Of course, virologists fought HIV with The solution, Debs says, is not figuring out are doing plenty of designing themselves, ribozymes—cata- how to put more DNA into a cell but to put and important advances have appeared in lytic RNAs that tar- more powerful DNA into a cell. “Rather than Nature Biotechnology. get specific nucleic looking at the DNA carrier, we’re looking at Daniel Curiel and colleagues at the acid sequences—for the DNA vector.” Finding the best vector will University of Alabama, Birmingham dem- years with little suc- eventually mean figuring out, tissue by tissue onstrated that adenoviral vectors could be cess. Then, in 2001 and condition by condition, what regulates tuned to enter specific cell types and not oth- he heard a lecture by DNA delivery, cellular processing and, most ers. They used a neutralizing antibody frag- Tom Tuschl, then of the Max Plank Institute of all, expression. ment to stop the virus from latching onto the for Biophysical Chemistry in Göttingen, about That understanding would improve not widespread cellular receptor for the adeno- a new kind of gene silencing that functions in just liposome-mediated gene delivery, but virus fiber, and redirected the virus to infect mammalian cells. RNA interference (RNAi) also delivery of naked DNA and, in some cells expressing the folate receptor by conju- uses short RNA molecules to waylay mRNA cases, viral vectors. The type of delivery gating folate to the neutralizing fragment4. before it can be translated into protein. The mode used, whether injection or electro- Compared with adenoviruses, lentivi- technique works exquisitely—assuming the poration or adsorption, will also depend on ruses promise to be more broadly applicable necessary short RNA molecules can get into the target tissue. “Each target organ requires because they can infect non-proliferating cells. cells. Rossi rushed back to the lab to map out a very different gene delivery method,” says In 1996, Trono and colleagues published the plans to engineer cells to make the appro- Jun-ichi Miyazaki, professor of stem cell first lentiviral vector system based on HIV-1; priate RNA themselves. “We’d never seen regulation at Osaka University Graduate however, many researchers feared using this knockdown that was so potent,” he recalls.

http://www.nature.com/naturebiotechnology School of Medicine (Osaka, Japan). In 1998, system because of the possibility that HIV-1 “We immediately shifted to going after the Miyazaki and colleagues transferred genes could be reconstituted inside cells5. The next virus [with small interfering RNAs].” into muscle by delivering electric pulses year, Trono modified the system by deleting In 2002, Rossi’s group at City of Hope close to the injection site of the plasmid about two-thirds of the HIV genetic infor- Beckman Research Institute (Duarte, CA, DNA, increasing the rate of gene expres- mation6. This multiply attenuated virus USA) described, for the first time, the use of sion by about 1,000-fold over simple DNA could still transduce differentiated cell RNAi against HIV in human cells. In trans- injection3. But since electroporation won’t types efficiently, and lentiviral vectors have fected cells, a plasmid designed to produce work in inaccessible internal organs like since become a widely popular research tool small interfering RNAs (siRNAs) inhibited the liver where blood flow quickly sweeps for expressing transgenes, siRNA and even the expression of HIV DNA by four orders of away injected DNA, other methods are nec- for creating transgenic animals. Indeed, magnitude1. Another paper in the same issue essary. “None of the current methods,” says VIRxSYS Corporation (Gaithersburg, MD, of Nature Biotechnology showed that intra- Nature Publishing Group Group Nature Publishing

6 Miyazaki “is sufficient in the specificity and USA) currently has a lentiviral gene delivery cellular gene expression could be targeted by duration of expression.” vector in clinical trials for HIV/AIDS. siRNAs produced from plasmid DNA2. 200

© “What limits us now in our nonviral Trono says that one issue keeping lenti- The following year, Mark Kay, a Stanford applications is that we have an almost infi- viruses from wider use is the possibility for pediatrician and geneticist, showed that nite number of combinations and permuta- them to cause insertional mutagenesis just as RNAi could inhibit viral replication in an tions we can test,” says Debs, but he believes has been shown for simple retroviral vectors. animal model3. Kay and colleagues used that this limitation will eventually become But he says lentiviral vectors have already one plasmid to introduce genes from the a strength. “The nonviral approaches are overcome bigger barriers. “In just a few years, hepatitis B virus (which is not infectious in inherently more powerful because they are a proposal that was initially viewed by many mice) and another to produce small hairpin more versatile,” he contends. as rather bold, not to say irresponsible, led RNAs that silenced viral genes. About 40% Viral vectors, though, with their higher to a tool that is routinely used.” of hepatocytes expressed the viral transgenes transduction efficiencies and ability to sta- in response to the first plasmid. If the second bly transduce cells, currently have the upper 1. Liu, Y. et al. Factors influencing the efficiency of cat- plasmid was included, levels of detectable hand in therapeutic applications. But, ionic liposome-mediated intravenous gene delivery. hepatitis B virus antigens in hepatocytes fell Nat. Biotechnol. 15, 167–173 (1997). according to Miyazaki , plasmids have other 2. Douglas, J.T. et al. Targeted gene delivery by tro- by 99%, and replication of the viral genomes advantages in that they are faster and cheaper pism-modified adenoviral vectors. Nat. Biotechnol. was inhibited. Kay is now looking at using to produce, can carry larger genes and can 14, 1574–1578 (1996). adeno-associated virus as a vector in mouse be designed to ameliorate autoimmune reac- 3. Aihara, H. & Miyazaki, J. Gene transfer into muscle models of human hepatitis infection. tions. Deb sums this up: “If you’re looking by electroporation in vivo. Nat. Biotechnol. 16, 867– Meanwhile, Rossi’s laboratory has developed 870 (1998). short-term, the majority of applications will 4. Douglas, J.T. et al. Targeted gene delivery by tro- a lentiviral vector for delivering siRNAs be viral; in the long term, none will,” he says. pism-modified adenoviral vectors. Nat. Biotechnol. and hopes to move it into clinical trials. “One of the reasons we’re alive is that the 14, 1574–1578 (1996). When bone marrow cells are transduced immune system has learned to recognize and 5. Naldini, L. et al. In vivo gene delivery and stable with sequences encoding siRNAs against blunt these viruses.” transduction of nondividing cells by a lentiviral vec- HIV’s TAT-REV gene, they differentiate into tor. Science 272, 263–267 (1996). But Didier Trono, at the Swiss Institute 6. Zufferey, R. et al. Multiply attenuated lentiviral immune cells that produce siRNAs that tar- of Technology Lausanne, thinks evolution vector achieves efficient gene delivery in vivo. Nat. get HIV’s replication machinery; the inclu- argues on the side of the microbes. “Viruses Biotechnol. 15, 871–875 (1997). sion of sequences encoding an RNA decoy

306 VOLUME 24 NUMBER 3 MARCH 2006 NATURE BIOTECHNOLOGY FEATURE

and a ribozyme that interfere with other HIV small molecules. That was kind of a shock to lematic gene could targets further enhances efficacy. Rossi hopes us.” Moreover, she says, the siRNAs are probably have a therapeutic to file an Investigational New Drug applica- also acting through the microRNA pathway, for effect, but the RNAi tion by mid-year. which expression analysis is inadequate. “There therapy will prob- RNAi therapies are still so early in develop- might be off-target effects that we aren’t pick- ably need to persist ment that many issues surrounding safety and ing up because they are occurring at the protein for a lifetime, either efficacy remain unclear, says Rossi. RNAi simply level,” she says. by permanently hasn’t been studied long enough to quell con- There is some evidence that chemically changing cells, cerns about long-term toxicities, for one thing. modifying RNA will make it more specific, chronically admin- Yet, Rossi cannot hide his enthusiasm for the she says, but such modification would work istering therapy, or Beverly Davidson: technology. “This is just a blessing,” he says, “it’s only for siRNAs delivered directly to cells, both. Viral therapies, “Can we hit enough given my lab a whole new direction.” not for siRNAs cells made endogenously. on the other hand, of the target cells in a Beverly Davidson feels the same way. Kay and Rossi say off-target issues are less might not work brain that’s the size of The University of Iowa neurologist works worrisome for viral genes, which share fewer unless viral genes a human brain?” on therapies for genetic neurodegenerative sequences with human genes. Nonetheless, are almost com- diseases. “I knew I could get genetic mate- Rossi is currently analyzing genome-wide pletely suppressed, but RNAi might not need rial into the regions of the brain that were microarrays in his siRNA-producing immune to operate over a long period of time. Also affected, the hurdle was what do we put cells for potential problems. Just as impor- unknown is how many cells RNAi would there?” Davidson and colleagues delivered tant as minimizing side effects will be deter- have to reach. Mouse brains are awfully genes for siRNA in living mice by injecting mining what off-target effects are acceptable. small, says Davidson. “Can we hit enough of recombinant adenoviruses into their brains. After all, most small-molecule drugs have the target cells in a brain that’s the size of a “We were able to see silencing of a transgenic some off-target effects, says Kay. human brain?” allele in the mouse brain,” recalls Davidson, Even those problems presuppose that Kay says delivery is the biggest problem

http://www.nature.com/naturebiotechnology “That was a jump-up-and-down kind of siRNAs can be safely and reliably delivered to right now, but he’s already seen what prog- day.” The team also showed silencing in liver cells. And although the mechanism of RNAi ress persistence can yield. When he first cells and of both endogenous and exogenous seems to hold constant from worms to mice started in gene therapy about fifteen years genes4. Since then, Davidson has been able to to humans, the means of using RNAi to fight ago, transducing liver cells in mice took use siRNA to ameliorate disease symptoms disease will vary. In dominant genetic dis- luck and persistence. High concentrations in mice models for Huntington disease and eases, even a partial knockdown of a prob- of the virus had to be infused directly into spinocerebellar ataxia5,6. Despite their enthusiasm, all three are quick to rattle off potential barriers to using Box 1 Digesting the implications of metabolomics RNAi as a therapy. One is finding the best sequence of RNA to use. Researchers have In 2000, Nature Biotechnology published one of the first papers Nature Publishing Group Group Nature Publishing

6 largely switched from randomly screening demonstrating that metabolic profiling (metabolomics) could sequences to using algorithms to select the 1

200 be used to compare genotypes . Oliver Fiehn and colleagues at

© best candidates. For example, researchers at Max Planck Institute of Molecular Plant Physiology (Potsdam, Dharmacon (Lafayette, CO, USA) system- Germany) developed a high-throughput way to analyze hundreds atically analyzed 180 siRNAs targeting two of compounds in leaves and showed that genes to find eight characteristics that can differences in these small molecules can clearly distinguish help identify potent siRNAs and showed that different genotypes. Now he’s applying these techniques for an algorithm incorporating all the character- diverse applications including figuring out how cigarette smoke istics improved selection7. harms a developing fetus, finding functions for orphan plant Rational designs have led to big improve- genes, and generally mapping metabolic networks. ments, says Rossi, but there’s still a long way Key to Fiehn and colleagues’ 2000 innovation was a way to volatalize metabolites to go. “You go through all those gyrations so that they could be separated by gas chromatography before mass spectrometry. But of trying to find a good target site–siRNA although this increased the number of compounds that can be analyzed, many still are not combination, and you still don’t get some- identified by molecular structure. “You see things going up and down and you don’t know thing that works at the subnanomolar con- the name for it,” says Fiehn, now a molecular biologist at the University of California, centration.” Davis. “This is a clear bottleneck.” A more worrisome challenge is off-target In most other areas, says Fiehn, metabolomics has surged ahead. He estimates that effects, In fact, Aimee Jackson and her col- experiments and analyses that required three weeks in 2000 can now be completed in two leagues at Merck (Whitehouse Station, NJ, days. And the faster experiments include more quality controls, catalog more compounds USA) found that siRNA can silence genes, and permit instant comparisons across data sets. Now, Fiehn and colleagues can look even if they share as few as 11 complementary at the metabolites from individual organisms, identify correlations between metabolites, nucleotides, a number she has currently revised elucidate metabolic networks and potentially link these networks back to genes. Best of down to as few as eight8. “The dogma was that all, says Fiehn, today’s data can fuel tomorrow’s experiments. “The whole idea of having RNAi was so specific we could use the RNAi to data that is consistent, structured and can be queried wasn’t there in 2000, but it’s there identify and predict off-target effects of small today.” molecules. What we found was that the siRNAs 1. Fiehn, O. et al. Metabolite profiling for plant functional genomics. Nat. Biotechnol. 18, 1157–1161 (2000). had significantly more off-target effects than

NATURE BIOTECHNOLOGY VOLUME 24 NUMBER 3 MARCH 2006 307 FEATURE

the liver and even then very few hepatocytes one by one into transgenic plants. Shinozaki were transduced. Today, the technique has and colleagues engineered Arabidopsis improved so much that nearly 100% of the thaliana to constitutively overexpress the liver can be transduced simply by injecting transcription factor DREB1A, which they the vector into a mouse’s tail, a particularly had previously found to upregulate several convenient site. genes in response to cold and dehydration. The solution, say the researchers, is to test The transgenic plants showed remarkable many approaches, realizing that many will increases in tolerance to freezing, water stress fail. “RNAi is sort of in a honeymoon period, and salinity, but the plants were dwarf in and there will be some limitations that come phenotype. The researchers got around this out of this,” says Kay, “that doesn’t mean we by expressing the protective gene just when Kazuo Shinozaki and Kazuko Yamaguchi- shouldn’t push the technology.” it was needed. When they used the promoter Shinozaki of RIKEN Plant Science Center from a dehydration-induced gene to control showed that overexpression of transcription factor 1. Lee, N.S. et. al. Expression of small interfering RNAs expression of DREB1A, the resulting plants DREB1A confers remarkable increases in plant targeted against HIV-1 rev transcripts in human cells. tolerance to freezing, water stress and salinity. Nat Biotechnol. 20, 500–505 (2002). were even more stress tolerant than consti- 2. Paul, C.P., Good, P.D., Winer, I. & Engelke, D.R. tutively active transgenics and furthermore, Effective expression of small interfering RNA in human cells. Nat. Biotechnol. 20, 505–508 grew to a normal size. work of Rachel Drake and her colleagues at (2002). “The DREB1A transcription factor is Syngenta (Basel) in enhancing β-carotene 3. McCaffrey, A.P. et al. Inhibition of hepatitis B virus a master switch of stress-inducible gene enriched rice3 (also known as ‘Golden Rice’) in mice by RNA interference. Nat. Biotechnol. 21, 639–644 (2003). expression,” explains Shinozaki. Though the represented a significant step forward. The 4. Xia, H., Mao, Q., Paulson, H.L. & Davidson, B.L. research was carried out in A. thaliana, he work continued on from a landmark study in siRNA-mediated gene silencing in vitro and in vivo. says, “our method [for] molecular breeding of 2000 by Ingo Potrykus and colleagues4 that Nat. Biotechnol. 20, 1006–1010 (2002). 5. Harper, S.Q. et al. RNA interference improves abiotic stress tolerance can be applied to real aimed to produce a rice strain that could help

http://www.nature.com/naturebiotechnology motor and neuropathological abnormalities in a breeding of crops and trees.” Shinozaki’s col- alleviate vitamin A deficiencies, which can Huntington’s disease mouse model. Proc. Natl. Acad. Sci. USA 102, 5820–5825 (2005). leagues and collaborators are currently using cause blindness among malnourished popu- 6. Xia, H. et al. RNAi suppresses polyglutamine-induced these new insights to probe stress responses lations. The problem with the original paper neurodegeneration in a model of spinocerebellar in a wide variety of crop plants, including was that the levels of β-carotene (a precur- ataxia. Nat. Med. 10, 816–820 (2004). 7. Reynolds, A. et al. Rational siRNA design for rice, wheat, tobacco, corn, petunia, poplar, sor to vitamin A) produced were too low to RNA interference. Nat Biotechnol. 22, 326–330 canola, tomato, tea, and soybean. It’s possible be a practical source of the nutrient. Drake (2004). that farmers breeding crops for higher yields and her colleagues examined the metabolic 8. Jackson, A.L. et al. Expression profiling reveals off- target gene regulation by RNAi. Nat. Biotechnol. 21, have, over the centuries, selected variants less pathway in transgenic rice and found that the 635–637 (2003). likely to express protective transcription fac- original daffodil gene for phytoene synthase tors2. Now, plant scientists have insights on was the limiting step in β-carotene accumu- how to tweak coordinated gene expression lation. By systematically substituting the daf- Nature Publishing Group Group Nature Publishing

6 Golden harvests systems to improve stress tolerance. fodil gene with homologs from other plants, Technologists are used Although creating more stress-tolerant the Syngenta team boosted pro-vitamin A 200

© to trade-offs. Better crops is likely to benefit farmers everywhere, levels in the rice by more than 20-fold, which sensitivity means public mistrust of agricultural biotechnol- might be enough to make a practical differ- worse specificity. ogy remains a constraint to acceptance and ence in nutrition. What’s more, Syngenta has Higher purity means expansion of the field. This is fuelled by the donated the rights to this rice to the Golden lower yield. And for view that large agribusiness is the primary Rice Humanitarian Board. plants, greater toler- beneficiary of some of the current crop of ance to stress means transgenics, such as those displaying insect- 1. Kasuga, M., Liu, Q., Miura, S., Yamaguchi-Shinozaki, K. & Shinozaki, K. Improving plant drought, salt, stunted growth. and pesticide-resistance. Efforts to produce and freezing tolerance by gene transfer of a single Kazuo Shinozaki and low trans fat soybean oils or to enhance the stress-inducible transcription factor. Nat Biotechnol. Kazuko Yamaguchi-Shinozaki, a husband-wife nutritional content of products are therefore 17, 287–291 (1999). team, and colleagues at RIKEN Plant Science likely to be of increasing importance, but will 2. Smirnoff, N. & Bryant, J.A. DREB takes the stress out of growing up. Nat. Biotechnol. 17, 229–230 Center (Yokohama, Japan), demonstrated that require more complex genetic engineering (1999). the last trade-off can be overcome1. than was previously required for single gene 3. Paine, J.A. et al. Improving the nutritional value of By the late 1990s, many genes that protect traits (e.g., in certain Bacillus thuringiensis Golden Rice through increased pro-vitamin A con- plants against harsh environmental condi- (Bt) or herbicide-resistant crops) as well as tent. Nat. Biotechnol. 23, 482–487 (2005). 4. Ye, X. et al. Engineering the provitamin A (beta-caro- tions had been identified, but researchers a comprehensive analysis of plant metabolic tene) biosynthetic pathway into (carotenoid-free) rice balked at the prospect of transferring genes pathways (see Box 1). In this context, the endosperm. Science 287, 303−305 (2000).

308 VOLUME 24 NUMBER 3 MARCH 2006 NATURE BIOTECHNOLOGY