REVIEWS 40 Derman, A. I., Prinz, W.A., Belin, D. and Beckwith, J. (1993) Science 262, 1744–1747 41 Bowdish, K., Tang, Y., Hicks, J. B. and Hilvert, D. (1991) J. Biol. Chem. 266, 11901–11908 42 Tang, Y., Hicks, J. B. and Hilvert, D. (1991) Proc. Natl. Acad. Sci. U. S. A. 88, 8784–8786

43 Martineau, P., Jones, P. and Winter, G. (1998) J. Mol. Biol. 280, 117–127 44 Fields, S. and Song, O. (1989) Nature, 340, 245–247 45 Chien, C. T. et al. (1991) Proc. Natl. Acad. Sci. U. S. A. 8, 9578–9582 46 Fields, S. and Sternglantz, R. (1994) Trends Genet. 10, 286–292

Proteomics: quantitative and physical mapping of cellular proteins Walter P. Blackstock and Malcolm P. Weir Genome sequencing provides a wealth of information on predicted gene products (mostly proteins), but the majority of these have no known function. Two-dimensional gel electrophoresis and mass spectrometry have, coupled with searches in protein and EST databases, transformed the protein-identification process. The proteome is the expressed protein complement of a genome and proteomics is functional genomics at the protein level. Proteomics can be divided into expression proteomics, the study of global changes in protein expression, and cell-map proteomics, the systematic study of protein–protein interactions through the isolation of protein complexes.

he term ‘proteome’ was used for the first time in 1995, to describe the protein complement of a genome1; imperceptibly, the proteome was transmuted into a new discipline, ‘proteomics’, which at least implies that money should be spent on it! The first book on proteome research has recently been published2 and, if the spate of recent conferences is a guide, it seems that proteomics is here to stay. So what is proteomics? In essence, it is the study of protein properties (expression level, post-translational modification, interactions etc.) on a large scale to obtain a global, integrated view of disease processes, cellular processes and networks at the protein level. This field is of growing importance to the question posed by the genome-sequencing projects – what are the functions of all the proteins? Now that whole genomes can be visualized, how do all the protein products interact? As soon as the first complete genome sequence, of Haemophilus influenzae, was published3, it became clear that many putative proteins encoded by the newly found genes had no known function and, of those with surmised function, many had functions attributed by analogy only. This has been the case as more and more genomes have yielded to the massive sequencing efforts that are going on around the world, and there is no reason to believe that it will be otherwise for the human genome, originally scheduled

T

W. P. Blackstock ([email protected]) and M. P. Weir are at the Biomolecular Structure Unit, Glaxo Wellcome Research and Development, Gunnel’s Wood Road, Stevenage, UK SG1 2NY. TIBTECH MARCH 1999 (VOL 17)

for completion by 2005. If the humble baker’s yeast, Saccharomyces cerevisiae, is a guide, at least half of the proteins encoded by the human genome may have no known function. With complete genome sequences now available for at least 18 organisms and over 30 more due for completion within three years4, a shift of emphasis is taking place towards making genomics functional. Bioinformatics alone will be insufficient for this task and many experimental approaches will be needed, because the interrelationships and properties of cellular protein networks cannot yet be quantitatively predicted. Indeed, it may be that this task will never reach a finite end point, if the proponents of complexity and emergent phenomena are to be believed5. Fortunately, as Bains observed6, even if the reductionist viewpoint is an oversimplification, the pragmatic approach that has served the biotechnology and pharmaceutical industry to date is likely to continue to be successful. Yet the problem of making sense of the huge amount of sequence data that is being generated remains. Molecular biology has provided powerful techniques for high-throughput DNA analysis that are not yet reflected in the protein world. This has resulted in an emphasis on the ‘message’ (mRNA or cDNA) rather than the product of that message (protein). As most drug targets are proteins, a route to studying the genome efficiently at the protein level is of great value and this is what proteomics offers. Currently, the best-established applications of proteomics are in the clinical and biomedical fields;

0167-7799/99/$ – see front matter © 1999 Elsevier Science. All rights reserved. PII: S0167-7799(98)01245-1

121

REVIEWS

alterations in the proteome of tissues or body fluids such as serum can be measured directly and changes in the levels or modifications of proteins can be correlated with disease, drug or hormone action and toxicological studies. This area has recently been reviewed7. It should be emphasized that, owing to the technological limitations outlined below, the field of proteomics is in its infancy; nonetheless, in our view, the coincidence of genome-sequencing projects with improvements in protein-analysis methods makes its goals attainable. Proteomics Since O’Farrell8 and Klose9 demonstrated that it was possible to separate proteins based on their isoelectric points and molecular weights by electrophoresis on polyacrylamide gels, two-dimensional polyacrylamidegel electrophoresis (2D PAGE) has remained unchallenged as the most efficient way of separating complex protein mixtures. Modern large-format gels (e.g. 20 3 20 cm) are highly reproducible, and Coomassieblue or silver staining allows protein quantitation; fluorescent dyes can be used to visualize thousands of proteins quantitatively over a wide dynamic range. Visualization is not the same as identification, of course, and, until recently, this was the bottleneck because identification relied on western blotting or, for the more abundant proteins, classical Edman sequencing, and these are relatively low-throughput methods. Two changes broke this impasse: the development of highly sensitive mass-spectrometric techniques and the parallel growth of protein and, in particular, expressed sequence tag (EST) databases. Identification of a protein by mass spectrometry is now more a problem of locating the protein or cognate EST in a database10 than of de novo sequencing, although small amounts of sample and the incomplete coverage of the EST databases still limit protein identification to some extent. However, we foresee the development of large-scale protein-characterization activities by pharmaceutical companies and specialist service providers, which will make protein identification more routine and of much higher throughput. Dual threads Proteomics is often used as a catch-all term and so lacks definition: we discern two related but distinct ways to apply this technology. First, the creation of quantitative maps of protein expression from cell or tissue extracts, akin to the EST maps commercially available. This approach relies on 2D gel maps and image analysis, and opens up the possibility of studying cellular pathways and their perturbation by disease, drug action or other biological stimuli at the wholeproteome level, thereby offering the potential to find disease markers and elucidate biological pathways. We refer to this as ‘expression proteomics’, and it holds particular promise in the fields of disease-marker discovery, toxicology and, perhaps, in drug-target validation. Second, the determination of the subcellular location of proteins and of protein–protein interactions by the purification of organelles or protein complexes followed by mass-spectrometric identification of the components11,12. Most proteins are thought to exist in the

122

cell not as free entities but as part of ‘cellular machines’13, which perform cellular functions cooperatively. Systematic identification of protein complexes would permit these machines to be defined and allow ‘physical maps’ to be created for a variety of cell types and states. Such information is of great value for the assignment of protein function, the principal problem facing the post-genome era, and for the validation of new disease targets. We refer to this as ‘cell-map proteomics’. Technology for proteomics Separating proteins Unlike genomics, proteomics has no equivalent of amplification by PCR. Thus, sample handling and sensitivity are critical issues that are not completely overcome. 2D PAGE remains the most effective means of separating large numbers of proteins and, with protein labelling by fluorescent dyes14, some of the problems of the limited dynamic range of silver- and Coomassiestaining methods are overcome. However, the problem of visualizing proteins expressed at a low level (‘lowcopy-number’ proteins, 10–1000 copies per cell) remains, and highly expressed proteins (so-called ‘housekeeping’ proteins, .10 000 copies per cell) can frequently obscure minor components. Some measure of prefractionation, such as centrifugation or free-flow electrophoresis15 is often required; combining such approaches with modified gel loading16 promises to bring even low-abundance proteins into view. Gels can also be selective, and special procedures are being devised for very basic proteins, membrane proteins and other poorly soluble proteins17. Post-translationally and proteolytically modified proteins are usually separated from the ‘parent’ protein by 2D PAGE and, although potentially useful, this can put demands on subsequent identification methods. Indeed, it has been suggested that up to a quarter of the spots on a gel may be modified proteins18. It is not uncommon to isolate what is essentially the same protein in several places on a gel. Traditionally, proteins were visualized by radiolabelling or by staining with Coomassie brilliant blue or, more sensitively, with silver. More recently, fluorescent dyes have been used that are both sensitive and offer a wider dynamic range. Ideally, the staining intensity should correlate with the amount of protein and be independent of the nature of the protein. In addition, mass spectrometry requires stains that do not compromise subsequent analysis. For example, silver staining has to be modified to avoid cross-linking agents19. Even with the most sensitive unmodified silver stains, proteins present at ,1000 copies per cell may be detectable but impossible to characterize by physical methods in the amounts typically loaded onto a gel 20. Although commercial software packages are available for automated spot detection, they were all designed for analysing only a few gels per day and can require 1–8 h additional manual editing per gel. This is an impossible limitation when 200–400 gels per week are being analysed or if the software is to be used to drive a spot-excision robot. There is currently a need for better image analysis that can be integrated with subsequent robotics to allow gels to be imaged and spots excised and characterized in one unattended operation. TIBTECH MARCH 1999 (VOL 17)

REVIEWS

Identifying proteins Mass spectrometry is now the method of choice for protein identification and for the characterization of post-translational modifications. The molecular weight of a protein of itself is insufficiently discriminating, so most approaches to protein identification rely on proteolysis of the separated protein with trypsin and analysis of the resulting peptides, often without any further separation. Two new ionization methods for the mass spectrometry of proteins and derived peptides combined with the rapid growth of protein and EST databases now make this a very powerful approach. This subject has been recently reviewed by Patterson21. Matrix-assisted laser-desorption-ionization–time-offlight (MALDI–TOF) mass spectrometry is usually used for peptide mass fingerprinting (PMF)22. The masses of peptides derived from an in-gel proteolytic digestion are measured and searched against a computer-generated list formed from the simulated digestion of a protein database using the same enzyme (usually trypsin). Using the mass accuracy now available (better than 10 ppm), this technique alone may be sufficient to characterize proteins from completely sequenced genomes such as yeast23, as the masses of all the tryptic peptides from the predicted open reading frames can be precisely calculated. If full-length sequences are not available then electrospray, particularly nano-electrospray (nanoES)24, is used to generate additional partial sequence information to complement the mass information. In the form of a ‘peptide-sequence tag’25, this may be used to search protein and EST databases with very high specificity. A conceptually different approach, which seeks to avoid any user interpretation, uses a correlation algorithm to match uninterpreted tandem mass spectra with computer-generated mass spectra from protein and EST databases26. The peptide-sequence-tag approach achieves a similar goal by carrying out proteolysis of gel spots in 18O-labelled water, thereby specifically labelling peptides from the C terminus. The resulting pattern of doublets arising from the 16O- and 18O-label is readily analysed by computer27. Both MALDI–TOF and electrospray can detect low levels of protein and are suitable for automation. MALDI–TOF is potentially capable of analysing several thousand proteins isolated from gels per week, if the necessary automation for spot excision and proteolysis and peptide extraction is developed. Many groups are developing robots specifically for this purpose. Electrospray methods, although of lower throughput, offer the additional information of partial sequence as well as peptide mass and are especially useful for posttranslational modifications28. It is most efficient to use both methods in a hierarchical approach (Fig. 1)29. Using mass-spectrometrically derived peptide masses alone (e.g. PMF) or mass and sequence together (e.g. peptide-sequence tags), current protein or EST databases can now be searched in under a minute, allowing real-time feedback to direct further mass spectrometry for rapidly confirming a predicted hit. It should be noted that mass-spectrometry approaches are inherently capable of dealing with mixtures of 2–3 proteins, which often arise from incomplete separation on a gel30. Non-redundant protein databases now have around 350 000 entries, but human EST databases with over TIBTECH MARCH 1999 (VOL 17)

1 200 000 entries exist in the public domain (e.g. dbEST). Although only 60 million base pairs (2%) of the 3 billion present in the human genome have been sequenced so far, over 50% of all human genes are thought to be represented in dbEST31, with up to 80% coverage in private EST databases. Thus, the ability to use mass-spectrometry data to mine EST data efficiently is vitally important in the search for those ESTs that are of the greatest biological or pharmaceutical importance. At the moment, several hundred femtomoles of protein per spot on the gel are realistically needed for characterization, and picomolar amounts are needed for the analysis of post-translational modifications32. Lower levels may occasionally be attained for particular ‘oneoff ’ situations but this should not be considered to be routine. This takes us into the realm of 1000 copies of a protein per cell for isolates from 108–109 cells, but there are currently very few laboratories capable of consistently working at this level. Although improvements in the sensitivity of mass spectrometers reflected in new instruments such as ion traps and quadrupole time-of-flight mass spectrometers will continue, we foresee more-immediate gains from better integrated software, using bioinformatics tools to control decision points on whether and where further mass spectrometry is needed. In addition, we believe that, in most cases, mass-spectrometer sensitivity is not limiting at present but that the signal to noise ratio (S:N), the crux of any measurement, is currently burdened by N, especially ‘chemical’ noise, from upstream protein preparation, gels and proteolysis, and the general environment. Practitioners in the field are experts on the mass-spectrometry signature of the many keratins that surround us! Clean rooms, automation and reduced sample handling are expected to make inroads into problems of contamination, but we would also like to see continued research into better gel materials specifically purified for mass spectrometry, as they are likely to remain the principal means of high-resolution protein separation for some years to come. In some cases, it may be possible to dispense with gels completely and use multidimensional chromatography (such as ion-exchange–size-exclusion) and tandem mass spectrometry. The reduced protein-separation capability is offset by exhaustive data analysis of the sequence data from tandem mass spectrometry. This approach looks promising for small bacterial genomes33 and may also be efficient for protein complexes. The approach selected will probably be determined by the amount of sample available. Applying the technology As outlined above, proteomics may, in our opinion, be divided into two areas – protein-expression mapping and protein-complex identification – which, although overlapping in the techniques used, seek to answer different questions. This distinction is particularly important in clarifying the role of proteomics in the pharmaceutical industry. Protein-expression mapping Protein-expression mapping may be defined as the quantitative study of global changes in protein expression in tissues, cells or body fluids using 2D gels and

123

REVIEWS

Immunoprecipitate or affinity purify

Protein complex isolated using tagged genes

1D or 2D gel electrophoresis

Robot picking In-gel enzyme digest Peptide mass fingerprint

Peptide mass fingerprint by MALDI-TOF

Protein-database search

Mass units If unidentified, select mass and fragment

m1

V

S

F

Nanoelectrospray for peptide sequence tag

m2

Peptide tag generated

EST database search

EST hit Identify

Add to the cell map

Add to cell map

Iterate

Figure 1 Protein identification by mass spectrometry. Genes of interest are tagged, transfected into cells and the proteins associated with the cognate tagged protein are purified by affinity methods. Separation of the protein complex is carried out by 1D or 2D gel electrophoresis. A hierarchical mass-spectrometry approach using low-cost and high-throughput methods [e.g. matrix-assisted-laser-desorption-ionization–timeof-flight (MALDI–TOF)] is used for initial peptide mass fingerprinting; for completely sequenced genomes, this alone may be sufficient to identify the proteins in the complex. When necessary, electrospray methods are then used to generate peptide-sequence tags for searching protein and EST databases. The schematic diagram illustrates the approach developed by Mann25. In Yates’ approach26, however, the protein mixture is digested without gel separation and analysed by combined liquid chromatography and tandem mass spectrometry. Database searching uses uninterpreted fragment-ion mass spectra and a cross-correlation algorithm to compare them with the predicted spectra.

124

TIBTECH MARCH 1999 (VOL 17)

REVIEWS

Protein-expression proteomics DNA-sequence databases

Bioinformatics

Differential gene expression (DGE)

Tissue distribution

Biological researchers

100 000 Genes

1000 Drug targets

Human genetics Yeast genetics

Caenorhabditis elegans genetics Knockouts

Yeast two-hybrid system Cell-map proteomics

Figure 2 Target validation and functional analysis. For a novel protein to be ‘validated’ as a drug target, it must be assigned a biological function and plausibly linked to a disease. Only a small proportion of the estimated 100 000 human genes are likely to be drug targets, and systematic use of many functional-analysis techniques, including proteomics, will be needed to find them.

image analysis. This method has the advantages of direct determination of protein abundance and detection of post-translational modifications such as glycosylation or phosphorylation, which result in a shift in mobility; mass spectrometry may be used for the subsequent characterization of proteins of interest. Because thousands of proteins are imaged in one experiment, a picture of the protein profile of the sample at a given point in time is obtained, enabling comparative proteome analysis. There are at least three companies that are capable of preparing cell or tissue samples and running several hundred gels per week: Hochstrasser has reviewed proteomics as a diagnostic tool in the clinic34; Celis is using proteomics to search for markers for bladder cancers in the urine35; and Mose Larsen has compared normal and b cells from the islets of Langerhans to search for changes in insulin-dependent diabetes mellitus36 – this study suggested that protein-expression changes may give clues to the role of certain proteins in this disease, and also showed that some of the identified proteins map to known diabetes genetic loci. 2D gel electrophoresis and image analysis can also be used to study drug action and toxicology: for example, a group of peroxisomal proliferator compounds was found to cause coordinated changes in mouse liverprotein expression that correlated with peroxisomal b oxidation37. A study comparing 2D-gel protein-expression measurements with the corresponding message data derived from differential gene expression showed that the correlation between mRNA and the cognate proTIBTECH MARCH 1999 (VOL 17)

tein is poor38. There is no a priori reason to expect a 1:1 correlation between mRNA input and the protein output so, for accurate quantitative work, it may be preferable to work at the protein level. Although protein-expression mapping is currently carried out by 2D gel electrophoresis, it must strive to be competitive in ease of use and throughput with differential gene expression if scientists are, on a regular basis, to measure changes in the proteome rather than the transcriptome. A promising alternative approach to 2D gel electrophoresis for serum markers involves use of phage antibody libraries to derive specific antipeptide antibodies, providing specific and sensitive reagents for immunocytochemistry. Using motif searching to identify secreted proteins from EST databases, this technique can be biased towards potential serum diagnostic markers. Published information on antibody-array approaches is currently very limited but seems to hold promise. Protein-complex identification: towards a physical map of the cell Assigning function to a novel protein requires the integration of many techniques (Fig. 2) and is currently a bottleneck in the drug-discovery process. There may be as few as a thousand drug targets for major diseases in the 100 000 genes that make up the human genome39, and finding these ‘validated’ targets is a considerable challenge. Much can be inferred about a protein’s function by the proteins with which it associates (its ‘protein partners’), its cellular location and any changes in these parameters introduced by stimuli. Direct

125

REVIEWS

measurement of protein–protein interactions by purification of protein complexes and mass-spectrometric identification has recently proved to be very valuable in discovering novel signalling proteins (for example FLICE40 and IKK41) and elucidating the components of the spliceosome complex11. Following the proposal of Mann and Lamond42, the systematic use of gene tagging with peptide sequences that enable affinity purification, subcellular fractionation, affinity purification and mass spectrometry (Fig. 3) could deliver a map of many protein complexes and their cellular locations. This would effectively be a physical map of the cell that would increase in depth over time and could incorporate different cell types and conditions; it could even be biased towards proteins of interest. A conceptually similar exercise is already being undertaken for all 6200 proteins of the yeast proteome using the two-hybrid system43; this technique identifies pairwise interactions rather than whole complexes and relies on proteins entering and folding in the nucleus, but it will undoubtedly continue to deliver information that complements that from mass spectrometry. There will be technology overlap for the two flavours of proteomics (protein-expression mapping and physical mapping of the cell). In both cases, automation and systematization on a scale similar to what has happened for genome sequencing may be needed, but we are optimistic that this can be achieved adapting the many tools already developed for the genome world to apply to protein biochemistry.

Selected genes Tagged versions Cell culture Subcellular fractions Purification Isolated complexes Protein identification by mass spectrometry Informatics Populate the cell map Figure 3 Towards a physical map of the cell. A systematic high-throughput approach to the characterization of protein complexes is proposed, in which protein complexes are isolated in parallel experiments by standardized gene tagging and affinity-based isolation, followed by 1D or 2D gel electrophoresis and mass-spectrometric identification of the proteins. This will augment the current two-hybrid approach to mapping pairwise protein–protein interactions.

126

Prospects Expression proteomics Several companies are setting up to offer contract and collaborative services to the pharmaceutical industry. The problems of low-copy-number proteins do not seem to be insurmountable when combining cellular fractionation and higher gel loading with sensitive fluorescent protein labels and mass spectrometry. Gel production is being automated and robots are being developed for spot excision, proteolysis and mass spectrometry. Problems with membrane proteins or very basic proteins remain but can be addressed to some extent by special preparation methods. Expression mapping is a valuable tool in the discovery of disease markers and its use in gaining information in toxicological and drug-action studies seems assured. It is unclear at present how successful this approach will be in elucidating cellular pathways and their importance in disease processes, and how much the precise measurement of protein levels matters when compared with the rough guide provided by the measurement of mRNA levels. Although there are few examples at present, the ability to measure protein-level changes directly would seem to carry inherent advantages and it seems likely that expression proteomics will be a useful tool in drug target discovery and in studying the effects of various biological stimuli on the cell. Cell-map proteomics Currently, affinity purification of protein complexes is done on a one-off basis. A move to a large-scale proteomic approach to mapping protein complexes using mass spectrometry combined with biological and genomic information would provide the framework of a physical map of the cell that can be filled in with everincreasing detail to encompass metabolic and signalling pathways. It would have the immediate advantage that genes of interest can be targeted and worked on in parallel in pursuit of biological understanding and drugtarget validation. The ultimate aim would be to incorporate structural data on the domain organization of the complexes from, for example, the use of cryoelectron microscopy44 and atomic-resolution structures solved by X-ray analysis, all within the known framework of the cell as defined by optical and electron microscopy. This ‘virtual cell’ would include an inventory of cell components and structures, and their approximate location in three dimensions. Such a framework would provide a sound basis to handle, for example, diseaseor drug-induced perturbations. There are, however, potential drawbacks to this approach: protein-expression and -purification stages may introduce artefacts, and further validation experiments will usually be required; in addition, not all complexes will be detected – some may only be formed transiently and may not be sufficiently stable to be isolated. Both proteomic approaches will benefit from the improved speed and sensitivity of protein identification. We foresee that just as, in the genome world, the price per base sequenced has fallen from US$5 to US$0.50 over five years, a similar inroad will be made into protein characterization by similarly improving the detail rather than by waiting for some novel technology. Immediate gains are there to be made from automation, TIBTECH MARCH 1999 (VOL 17)

REVIEWS

better software integration and the industrialization of the necessary biology. References 1 Wasinger, V. C. et al. (1995) Electrophoresis 16, 1090–1099 2 Wilkins, M. R., Williams, K. L., Appel, R. D. and Hochstrasser, D. F., eds (1997) Proteome Research: New Frontiers in Functional Genomics, Springer-Verlag 3 Fleischmann, R. D. et al. (1995) Science 269, 496–512 4 Tomb, J-F. (1998) Nat. Biotechnol. 16, 23 5 Strohmann, R. (1997) Nat. Biotechnol. 15, 194–200 6 Bains, W. (1997) Nat. Biotechnol. 15, 396 7 Hochstrasser, D. F. (1997) in Proteome Research: New Frontiers in Functional Genomics (Wilkins, M. R., Williams, K. L., Appel, R. D. and Hochstrasser, D. F., eds), pp. 187–221, Springer-Verlag 8 O’Farrell, P. H. (1975) J. Biol. Chem. 250, 4007–4021 9 Klose, J. (1975) Humangenetik 26, 231–243 10 Mann, M. (1996) Trends Biol. Sci. 21, 494–495 11 Neubauer, G. et al. (1997) Proc. Natl. Acad. Sci. U. S. A. 94, 385–390 12 Mitchell, P., Petfalski, E., Shevchenko, A., Mann, M. and Tollervey, D. (1997) Cell 91, 457–466 13 Alberts, B. and Miake-Lye, R. (1992) Cell 68, 415–420 14 Steinberg, T. H., Haugland, R. P. and Singer, V. L. (1996) Anal. Biochem. 239, 238–245 15 Völkl, A., Mohr, H., Weber, G. and Dariush Fahimi, H. (1997) Electrophoresis 18, 774–780 16 Sanchez, J. C. et al. (1997) Electrophoresis 18, 324–327 17 Rabilloud, T., Adessi, C., Giraudel, A. and Lunardi, J. (1997) Electrophoresis 18, 307–316 18 Celis, J. E. et al. (1995) Electrophoresis 12, 2177–2240 19 Shevchenko, A., Wilm, M., Vorm, O. and Mann, M. (1996) Anal. Chem. 68, 850–858 20 Herbert, B., Sanchez, J-C. and Bini, L. (1997) in Proteome Research: New Frontiers in Functional Genomics, (Wilkins, M. R., Williams, K. L., Appel, R. D. and Hochstrasser, D. F., eds), pp. 13–61, Springer-Verlag

21 Patterson, S. D. et al. (1998) Electrophoresis 19, 883–1054 22 Henzel, W. J. et al. (1993) Proc. Natl. Acad. Sci. U. S. A. 90, 5011–5015 23 Shevchenko, A. et al. (1996) Proc. Natl. Acad. Sci. U. S. A. 93, 14440–14445 24 Mann, M. and Wilm, M. (1996) Anal. Chem. 68, 1–8 25 Mann, M. and Wilm, M. (1994) Anal. Chem. 66, 4390–4399 26 Eng, J. K., McCormack, A. L. and Yates, J. R., III (1994) J. Am. Soc. Mass Spectrom. 5, 976–989 27 Shevchenko, A. et al. (1997) Rapid Commun. Mass Spectrom. 11, 1015–1024 28 Betts, J. C., Blackstock, W. P., Ward, M. A. and Anderton, B. H. (1997) J. Biol. Chem. 272, 12922–12927 29 Humphery-Smith, I., Cordwell, S. J. and Blackstock, W. P. (1997) Electrophoresis 18, 1217–1242 30 Jensen, O. N., Podtelejnikov, A. V. and Mann, M. (1997) Anal. Chem. 69, 4741–4750 31 Rowen, L., Mahairas, G. and Hood, L. (1997) Science 278, 605–606 32 Neubauer, G. and Mann, M. (1999) Anal. Chem. 71, 235–242 33 Opiteck, G. J., Ramirez, S. M., Jorgenson, J. W. and Moseley, M. A., III (1998) Anal. Biochem. 258, 349–361 34 Hochstrasser, D. (1997) in Proteome Research: New Frontiers in Functional Genomics (Wilkins, M. R., Williams, K. L., Appel, R. D. and Hochstrasser, D. F., eds), pp. 187–219, Springer-Verlag 35 Rasmussen, H. H., Ji, H., Wolf, H. and Celis, J. E. (1996) J. Urol. 155, 2113–2119 36 Aanstoot, H. J. et al. (1996) J. Clin. Invest. 97, 2772–2783 37 Anderson, N. L., Esquer-Blasco, R., Richardson, F., Foxworthy, P. and Eacho, P. (1996) Toxicol. Appl. Pharmacol. 137, 75–89 38 Anderson, L. and Seilhamer, J. (1997) Electrophoresis 18, 533–537 39 Drews, J. (1996) Nat. Biotechnol. 14, 1516–1518 40 Muzio, M. et al. (1996) Cell 85, 817–827 41 Mercurio, F. et al. (1997) Science, 278, 860–866 42 Lamond, A. I. and Mann, M. (1997) Trends Cell Biol. 7, 139–142 43 Fromont-Racine, M., Rain, J-C. and Legrain, P. (1997) Nat. Genet. 16, 277–282 44 Stark, H. et al. (1997) Cell 88, 19–28

Powerful tools for genetic analysis come of age David J. Graves Microscopic arrays of oligonucleotides or cDNA containing up to several hundred thousand different sequences are starting to influence methodologies and paths to discovery in genomics. Gene polymorphisms and mutations can be found and gene expression measured with unprecedented speed and parallelism. The principles of this modern technology and some of the problems awaiting further study are discussed.

icroarrays of DNA and oligonucleotides are beginning to have the same impact on the biological sciences that integrated circuits have already produced on the physical sciences, and for similar reasons: they can do many things in parallel, with very little material and with a modest investment of labor. It is too early to say whether complex micro-

M

D. J. Graves ([email protected]) is at the Department of Chemical Engineering, 311A Towne Building, University of Pennsylvania, Philadelphia, PA 19104-6393, USA. TIBTECH MARCH 1999 (VOL 17)

fabricated chemical systems will result in anything as revolutionary as the personal computers that grew out of the development of electronic devices on a chip, but arrays and other microfabricated devices are already beginning to generate sufficient interest to make them look very promising. These arrays consist of many microscopic spots, each of which contains identical single-stranded polymeric molecules of deoxyribonucleotide (typically oligonucleotides or cDNAs) attached to a solid support such as glass or a polymer. Each spot contains many copies

0167-7799/99/$ – see front matter © 1999 Elsevier Science. All rights reserved. PII: S0167-7799(98)01241-4

127