PERSPECTIVE
Functional Genomics and Expression Profiling
Be There or Be Square

Steven M. Albelda and Dean Sheppard

Pulmonary, Allergy, and Critical Care Division, Department of Medicine, University of Pennsylvania Medical Center, Philadelphia, Pennsylvania; and Lung Biology Center, Department of Medicine, University of California, San Francisco, San Francisco, California.

The manuscript by Hong and colleagues in this issue of the American Journal of Respiratory Cell and Molecular Biology (1) profiling the downstream genes activated by expression of the tumor suppressor gene, PTEN, represents one of the first papers published in this journal that makes use of complementary DNA (cDNA) microarrays. cDNA microarray analysis is one of the most important new areas that has emerged from the ongoing revolution in functional genomics. This paper provides some background material and highlights the importance and promise of this new technology for lung researchers.

	Definitions

In the simplest terms, a microarray is a modified Northern blot, where one analyzes the relative expression level of a gene by determining the amount of messenger RNA (mRNA) that is present. Unlike a conventional Northern blot where one can analyze one, two, or even 10 to 20 mRNAs, a cDNA microarray allows the simultaneous analysis of the expression levels of hundreds, thousands, or even tens of thousands of genes in a single experiment (2). This capability, coupled with sophisticated computer analysis tools (bioinformatics) will, quite simply, revolutionize our understanding of pulmonary biology and pulmonary diseases. It is thus important for both basic scientists and clinicians to understand and begin to use this new technology.

	Anatomy of an Array Experiment

The following section briefly describes how a typical microarray experiment would be conducted (see Figures 1 and 2). A number of recent reviews provide more details (3).

View larger version (36K): [in this window] [in a new window]   Figure 1.   Relative intensity hybridization experiment. One commonly used technique to generate arrays involves RNA isolation (A), followed by synthesis of cDNA using reverse transcriptase (B). This is usually done in the presence of labeled nucleotides to created a "tagged" probe (C). This probe is hybridized to the array and washed (D). The array is then imaged, and the intensity of each cDNA "spot" is determined (E). Gene expression intensities are corrected for background and for relative intensity by normalization with control genes or by a measure of total spot intensity.

View larger version (24K): [in this window] [in a new window]   Figure 2.   Comparative hybridization experiment. A second commonly used technique to measure gene expression involves RNA isolation of two separate samples (A). The RNA from each sample is treated with reverse transcriptase (B) and labeled with a distinct fluorescent tag (C). The two pools of labeled RNA are mixed, hybridized to the array, and washed (D). The array is imaged using a specialized fluorimeter, and the color of each spot is determined (E). In this example, genes only expressed in Sample A would be red in color, genes only expressed in Sample B would be green, and those genes expressed equally in both samples would be yellow.

A. RNA Isolation

The basic question to be asked in any type of array experiment is, how does gene transcription compare in two or more types of samples? Since genes code for proteins that are first transcribed into mRNAs, the starting point of any array experiment is the isolation of mRNA from two or more types of cells or tissues. Obviously, this represents one of the major caveats of array analysis because mRNA levels do not always reflect subsequent protein expression levels (due to issues such as differing translational efficiencies and post-translational stability and modification).

To prepare mRNA for an array assay, it must be extracted from total cellular contents. This could represent a considerable challenge, especially when one is analyzing tissue samples because RNA is quite sensitive to ubiquitous RNA-degrading enzymes. It must also be free from any contaminating genomic DNA. Thus, the quality and purity of input RNA needs to be analyzed on each sample prior to proceeding with subsequent steps.

B. cDNA generation

To prevent the fragile mRNAs from being degraded, they are next reverse-transcribed back into a more stable cDNA form using a reverse transcriptase. A key theoretical concern at this step is that the RNAs be reverse-transcribed with the same efficiency, avoiding a "reverse transcription bias." This bias is not usually a major problem when one compares the same type of RNA across different cell populations; however, it could limit quantitative comparisons of two different genes within the same population of cells.

C. Labeling of Probe

Once the cDNAs are prepared (or more correctly, while they are being prepared), they must be labeled in some fashion to allow detection after hybridization to the array. Three major systems are in current use. One system relies on measuring the absolute intensity of the labeled cDNA probe from each sample (Figure 1). This can be accomplished by incorporating radiolabeled, biotinylated, or fluorescent nucleotides into the second-strand synthesis reaction. Detection can then be performed with a PhosphorImager (for radiolabeled cDNA) with labeled streptavidin (for biotinylated cDNA) or by a variety of fluorescence detectors. Hong and associates (1) labeled their cDNA with biotinylated nucleotides, and achieved detection by using a streptavidin-linked -galactosidase enzyme rather than radioactivity.

The second approach is based on comparative intensity between two cDNA samples (Figure 2). In this technique, the cDNA from one sample (Sample A) is labeled with one fluorescent dye, and the cDNA from the sample to be compared (Sample B) is labeled with a dye that fluoresces at a distinct and nonoverlapping wavelength. The cDNAs are mixed and then hybridized to the array. For example, if red and green dyes are used and the expression levels are similar, both red and green cDNAs hybridize, resulting in a yellow color. If expression levels from red-labeled cDNAs are higher, the spot will appear red. In contrast, higher expression from the second sample will appear green. If detectors that can read more colors are employed, it is theoretically possible to simultaneously hybridize and compare more than two samples (as is now commonly done for flow cytometry).

The third approach incorporates a fluorescent or biotinylated nucleotide into the synthesis of complementary RNA (cRNA) generated by in vitro transcription.

D. Hybridization to the Microarray

After labeling, the cDNAs are hybridized to a microarray. An array is an ordered arrangement of genetic information that provides a platform for the hybridization of known and unknown DNA samples. A macroarray contains sample spots of 300 microns or larger and can be imaged using existing gel or blot scanners (see Reference 6 for an example). A spotted microarray contains much smaller sample spots (typically anywhere from 10 to 200 microns) and usually contains thousands of spots. Microarrays require special robotics for production and specialized imaging equipment for analysis. DNA microarrays can be produced on glass, polycarbonate, metal, or nylon substrates.

There are two major variants of microarrays. In the first system, cDNAs (often produced by polymerase chain reactions [PCR] from cDNA libraries) are immobilized onto glass slides or membranes. This was the platform used by Hong and coworkers (1). In the second system, an array of overlapping oligonucleotides is synthesized on the chip itself (7). This method, historically called the "gene chip," was developed at Affymetrix, Inc., Santa Clara, CA (for example, see Reference 8).

Of course, one major factor in any microarray analysis is the composition of the cDNAs or the oligonucleotides on the slide or membrane. The human genome is estimated to contain between 50,000 to more than 100,000 genes. Although currently available reagents only allow analysis of some subsets of these genes, complete representative arrays are already available for model organisms (i.e., S. cerevesiae), and within a few years, it is likely that probes for all of the expressed human and murine genes will be available for array analysis. The composition of the microarray defines how much possible information might be obtained, and it also determines the complexity of the data. From a practical standpoint, the density of the array will have a major impact on the cost required to purchase or produce the array and then to analyze it. There is an increasing number of options available from commercial sources, as well as the increased ability of individual labs or core laboratories to make their own arrays. Complexity of current arrays varies widely and can range from 500 to 1,000 known genes on nylon membranes (i.e., the Clonetech arrays [see Reference 6]) to tens of thousands of genes or oligonucleotides that contain many genes of unknown function (expressed sequence tags [ESTs]) on glass slides. The choice of array will represent a balance between the goal of the experiment (i.e., obviously, for new gene discovery one would want a microarray with a large number of ESTs), availability of the array and the imaging equipment, cost, and bioinformatics capability. A number of companies, as well as individual investigators, are designing arrays that include a specific subset of genes of interest. For example, arrays of "stress genes" or "inflammatory response" genes are available. Libraries of all expressed genes from a particular cell type are also becoming available (i.e., a prostate cancer cell gene library [9]).

E. Imaging

After hybridization and washing, the array must be appropriately scanned to determine how much of each probe is bound to each spot. For microarrays, this usually requires an expensive, dedicated, scanning instrument ranging in cost from $50,000 to $100,000. Scanning is also available from commercial sources or core facilities. The scanned output must then be displayed in a database suitable for analysis. The ease of obtaining such output varies greatly among currently available scanners.

F. Analysis

For all but the smallest macroarray, a single study can generate thousands or tens of thousands of data points, which cannot be simply analyzed by sorting in spreadsheets or plotting on simple graphs. Consultation with experts familiar in analysis becomes a necessity. Although providing the details are beyond the scope of this review, a number of approaches have been developed (10).

One of the most direct questions asked is whether a set of genes is expressed at a higher or lower level when comparing two conditions, an approach called transcription profiling (5). For example, in the paper by Hong and colleagues (1), genes with different intensities of expression were sought in cells expressing or lacking the tumor suppressor, PTEN. In this paper, cDNA clones that consistently showed at least a two-fold difference were "arbitrarily" selected for further analysis. Using this approach, a number of known and unidentified genes that were modulated by PTEN expression were identified. Although this two-fold "cutoff" has been used by many investigators, it does not confront the problem of defining what level of difference is statistically significant, believable, and/or meaningful. This is an important problem because there are significant amounts of variability that accumulate from each step in the process. As in all biologic experiments, probably the most important source of variability arises from variability in the cells or tissues being studied. Thus, as many as 20% of the genes analyzed are likely to differ by levels of two-fold or more among dishes of primary cells obtained from splitting a single dish (D. S., unpublished observations). Variability also arises during RNA harvesting, reverse transcription, labeling, hybridization, and scanning. There may also be considerable variability among the arrays themselves. Importantly, when one deals with very large numbers of comparisons, a large number of "differences" will appear solely because of chance. A preferred approach to arbitrary cutoffs involves analyzing a suitable number of identical samples and using statistical approaches to define a standard deviation for each spot of interest. Using these data, genes with expression levels outside of statistically defined ranges can be defined (i.e., see Reference 6). However, even with this approach, statistical methods to determine what changes should be considered meaningful have not yet been developed. As a general rule, the smaller the differences observed, the smaller the sample size; and the lower the level of baseline expression, the less confidence one can have in the observed differences. Such considerations should not discourage investigators from using arrays, as long as it is recognized that this approach should be used primarily to generate hypotheses rather than definitively test biologic postulates.

In addition to the simple identification of genes that are upregulated or downregulated under defined experimental conditions, a great deal more information can be extracted from arrays (13). One of the most important of these analyses, called cluster analysis, involves determining which genes are correlated in their responses over a range of perturbations. For example, Kaminski and colleagues (8) used cluster analysis to identify two distinct groups of genes involved in the inflammatory and fibrotic responses of murine lungs to bleomycin instillation. Cluster analysis must be done by computer, and most current analyses use an agglomerative approach, where single expression profiles are successively joined to form related groups that are plotted in a hierarchical tree structure, similar to that used to describe evolutionary relationships (see Reference 12 for further review).

G. Validation

Given the inherent variability of the microarray data, it is important for investigators to validate important findings using alternative approaches. It is not certain that other methods to measure RNA expression levels (i.e., Northern blot analysis, quantitative PCR, in situ hybridization) are required; ultimately, findings that are confirmed by measurements of protein expression, and especially by experiments that demonstrate biologic significance, will be most convincing. The paper by Hong and associates (1) nicely illustrates this principle by confirming results of specific candidate genes identified by microarray using Northern blots and flow cytometry.

	How Can Microarrays Be Used?

The potential applications of microarray technology are almost limitless (see Reference 13), but only a few examples will be provided here.

Observational Studies

One of the most exciting early uses of arrays would be to simply use them to "see what happens" after a specific perturbation or to analyze what is different between two cell types. Arrays will allow the observation of the previously unobservable and have therefore been likened to the discovery of the telescope or microscope (14). Of course, this type of observational, broad-scale inquiry is the antithesis of hypothesis-driven science so loved by current study sections and has the risk of being criticized as the ultimate "fishing expedition." However, it is highly likely that microarrays will soon be one of the most powerful hypothesis-generating approaches available today (see subsequent text).

New Gene Discovery

Microarrays that include genes with currently unknown functions (ESTs) offer the exciting possibility of new gene discovery and will likely play a key role in the translation of genomics into functional genomics. For example, the study by Hong and his colleagues (1) identified a number of new genes that were highly expressed after transfection with the PTEN genes, thus potentially implicating them in the invasion/metastasis cascade. This approach has particular appeal to the pharmaceutical industry, where new drug targets are actively being identified using microarrays (15). For example, previously unknown genes selectively expressed in T helper (Th) 2-type lymphocytes could provide new targets for asthma (16). It is also possible to use comprehensive arrays to perform deletion mapping to identify specific regions of chromosomes whose genes are absent in a large microarray.

Process Profiling

Identification of complex gene patterns using cluster analysis is another burgeoning application of array technology. There are many potential areas for study, ranging from examination of natural processes, including cell division or aging (i.e., see References 10, 17, and 18), disease progression (i.e., see Kaminski and colleagues [8] in a model of pulmonary fibrosis or McCaffrey and associates [6], who studied atherosclerosis), pharmacologic interventions (i.e., Jelinsky and coworkers [19], who studied response of yeast to an alkylating agent), carcinogen identification, toxicology, and drug safety evaluation (20). One obvious area of interest has been the comparison of gene profiles in normal versus malignant tissues (i.e., Sgroi and associates [21], who profiled human breast cancer progression or Alon and colleagues [11], who identified broad patterns of gene expression in normal tissues versus colon tumors). One could, of course, envision many applications in pulmonary diseases. Preliminary work has been published studying inflammatory airways disease (22). At the most recent American Thoracic Society meeting in Toronto, a number of abstracts and presentations described gene profiles in lungs exposed to oxidative injury and airway inflammation and in lung cells exposed to cytokines. Expression profiles in lung cancer cells and tumors, mesotheliomas, and pulmonary fibrosis were also described. Identifying specific patterns of gene expression could possibly provide important clues to the etiology of diseases that remain mysterious, like sarcoidosis.

Molecular Classification of Diseases

Gene profiling and cluster analysis promise to revolutionize the classification of disease, greatly extending the current diagnostic limitations of histochemical, or even immunohistochemical, analysis. For example, data have been published showing markedly different gene expression profiles in tissues from inflammatory diseases such as rheumatoid arthritis and inflammatory bowel disease (23). Pioneering work in this field has been done with hematologic malignancies where specific clusters of genes have been identified that can characterize specific subtypes of leukemias (24) or lymphomas (25). In addition to diagnosis, this sort of profiling could be helpful in understanding disease pathogenesis. For example, a recent study has examined profiles from true neuroendocrine brain tumors, small-cell lung cancers, lung carcinoids, and bronchial epithelial cells and has concluded that small-cell carcincomas were much more like epithelial tumors than neuroendocrine tumors. In contrast, carcinoids were more closely related to neuroendocrine tumors (26).

Finally, expression profiling and cluster analysis will doubtless be extremely useful for predicting prognosis and for identifying subgroups of patients that will respond in a certain way to therapy. Thus, one could search for specific clusters of gene expression patterns of malignant lung tumors that might predict which tumors are likely to recur, as well as provide information about optimal adjuvant therapy. Similarly, one could imagine expression profiling becoming extremely useful in the complex classification of interstitial lung diseases, which by light microscopy look very similar.

	What Are the Limitations and Challenges in This Technology?

Like most deals in life, if something sounds too good to be true, it probably is. Despite tremendous potential, there are still many difficulties and challenges in microarray technology. One theoretical limitation is the fact that cDNA expression profiles derived from RNA do not necessarily mirror protein expression levels. In addition, many critical cellular events involve signal transduction events, such as protein phosphorylation, that cannot be detected directly by array technology. From a technical standpoint, linear reverse transcription of the RNA is also important.

Independent of technical issues, a key problem is the quality and purity of the tissues analyzed ("garbage in, garbage out"). When one harvests whole organs or even pieces of tumor tissues, the RNA extracted is clearly from a wide variety of different cell types, including the normal constituents of the tissue, plus whatever inflammatory cells are present. Conclusions from these studies must be limited by this heterogeneity of starting material. However, solutions are developing. The use of microdissection using laser capture microscopy (27), coupled with new technologies to amplify RNA from very small quantities of tissue (28), promises to allow analysis from specific cell types.

Another major concern relates to the accessibility and expense of microarrays. Until very recently, large arrays were only available from specialized academic centers (the N.C.I. leading the way) or commercial sources at extremely high prices. Coupled with the expense of specialized imaging equipment, most early studies have been done by pharmaceutical clients or specialized collaborative arrangements. Fortunately, these financial constraints appear to be lessening. A number of companies now offer macroarrays or microarrays (along with analysis, if requested) for reasonable prices. Many universities are making investments in the necessary infrastructure to support microarray "cores," which both produce arrays and analyze them at relatively low cost.

Finally, the informatics and data analysis requirements of microarrays are considerable. Rapid progress is being made in this area, and a number of free and commercial software tools are becoming available. Again, inclusion of informatics as part of microarray cores will be critical for making this technology accessible and easy to use for the average investigator.

	Future Directions

Functional genomics and microarray technology are in their infancy. Based on the rapid progress in the past five years, it is almost certain that this technology will be increasingly accessible and useful. Advances are likely to be even more impressive with the development of the closely related, but less developed, field of proteomics (proteomics will array cellular proteins instead of genes). As discussed above, these approaches are likely to revolutionize our thinking about disease processes and molecular diagnosis. Over the next few years, many lung investigators will be using microarrays and expression profiling to study the entire range of respiratory disorders and processes. Be there or be square!

Final Note: The reader is referred to an excellent web site (http://www.gene-chips.com/) where one can find basics on DNA microarray technology and a list of academic and industrial links.

	Footnotes

Abbreviations: complementary DNA, cDNA; expressed sequence tags, ESTs; messenger RNA, mRNA.

(Received in original form June 21, 2000).

Acknowledgments: The authors would like to thank Dr. Naftali Kaminski for helpful comments and suggestions and Ms. Linda Kalb for helping to prepare the manuscript.

	References

1. Hong, T.-M., P.-C. Yang, K. Peck, J. J. W. Chen, S.-C. Yang, Y.-C. Chen, and C.-W. Wu. 2000. Profiling the downstream genes of tumor suppressor PTEN in lung cancer cells by cDNA microarray. Am. J. Respir. Cell Mol. Biol. 23: 355-363 [Abstract/Free Full Text].

2. Schena, M., R. A. Heller, T. P. Theriault, K. Konrad, E. Lachenmeier, and R. W. David. 1998. Microarrays: biotechnology's discovery platform for functional genomics. Trends Biotechnol. 16: 301-306 [Medline].

3. Bowtell, D. D. L.. 1999. Options availablefrom start to finishfor obtaining expression data by microarray. Nat. Genet. Suppl. 21: 25-32 .

4. Cheung, V. G., M. Morley, F. Aguilar, A. Massimi, R. Kucherlapati, and G. Childs. 1999. Making and reading microarrays. Nat. Genet. Suppl. 21: 15-19 .

5. Duggan, D. J., M. Bittner, Y. Chen, P. Meltzer, and J. M. Trent. 1999. Expression profiling using cDNA microarrays. Nat. Genet. Suppl. 21: 10-14 .

6. McCaffrey, T. A., C. Fu, B. Du, S. Eksinar, K. C. Kent, H. Bush Jr., K. Kreiger, T. Rosengart, M. I. Cybulsky, E. S. Silverman, and T. Collins. 2000. High-level expression of Egr-1 and Egr-1-inducible genes in mouse and human atherosclerosis. J. Clin. Invest. 105: 653-661 [Medline].

7. Lipshutz, R. J., S. P. A. Fodor, T. R. Gingeras, and D. J. Lockhart. 1999. High density synthetic oligonucleotide arrays. Nat. Genet. 21: 20-24 [Medline].

8. Kaminski, N., J. D. Allard, J. F. Pittet, F. Zuo, M. J. D. Griffiths, D. Morris, X. Huang, D. Sheppard, and R. A. Heller. 2000. Global analysis of gene expression in pulmonary fibrosis reveals distinct programs regulating lung inflammation and fibrosis. PNAS 97: 1778-1783 [Abstract/Free Full Text].

9. Emmert-Buck, M. R., R. L. Strausberg, D. B. Krizman, M. F. Bonaldo, R. F. Bonner, D. G. Bostwick, M. R. Brown, K. H. Buetow, R. F. Chuaqui, K. A. Cole, P. H. Duray, C. R. Englert, J. W. Gillespie, S. Greenhut, L. Grouse, L. W. Hiller, K. S. Katz, R. D. Klausner, V. Kuznetzov, A. E. Lash, G. Lennon, W. M. Lineham, L. A. Liotta, M. A. Marra, P. J. Munson, D. K. Ornstein, V. V. Prabhu, C. Prange, G. D. Schuler, M. B. Soares, C. M. Tolstoshev, C. D. Vocke, and R. H. Waterston. 2000. Molecular profiling of clinical tissue specimens: feasibility and applications. J. Mol. Diagn. 2: 60-66 . [Free Full Text]

10. Eisen, M. B., P. T. Spellman, P. O. Brown, and D. Botstein. 1998. Cluster analysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci. USA 95: 14863-14868 [Abstract/Free Full Text].

11. Alon, U., N. Barkai, D. A. Notterman, K. Gish, S. Ybarra, D. Mack, and A. J. Levine. 1999. Broad patterns of gene expression revealed by clustering analysis of tumor and normal colon tissues probed by oligonucleotide arrays. Proc. Natl. Acad. Sci. USA 96: 6745-6750 [Abstract/Free Full Text].

12. Sherlock, G.. 2000. Analysis of large-scale gene expression data. Curr. Opin. Immunol. 12: 201-205 [Medline].

13. Epstein, C. B., and R. A. Butow. 2000. Microarray technology: enhanced versatility, persistent challenge. Curr. Opin. Biotechnol. 11: 36-41 [Medline].

14. Brent, R.. 1999. Functional genomics: learning to think about gene expression data. Curr. Biol. 9: R338-R341 [Medline].

15. Debouck, C., and P. N. Goodfellow. 1999. DNA microarrays in drug discovery and development. Nat. Genet. Suppl. 21: 48-50 .

16. Rogge, L., E. Bianchi, M. Biffi, E. Bono, S. Y. Chang, H. Alexander, C. Santini, G. Ferrari, L. Sinigaglia, M. Seiler, M. Neeb, J. Mous, F. Sinigaglia, and U. Certa. 2000. Transcript imaging of the development of human T helper cells using oligonucleotides arrays. Nat. Genet. 25: 96-101 [Medline].

17. Lee, C.-K., R. G. Klopp, R. Weindruch, and T. A. Prolla. 1999. Gene expression profile of aging and its retardation by caloric restriction. Science 285: 1390-1393 [Abstract/Free Full Text].

18. Ly, D. H., D. J. Lockhart, R. A. Lerner, and P. G. Schultz. 2000. Mitotic misregulation and human aging. Science 287: 2486-2492 [Abstract/Free Full Text].

19. Jelinsky, S. A., and L. D. Samson. 1999. Global response of Saccharomyces cerevisiae to an alkylating agent. Proc. Natl. Acad. Sci. USA 96: 1486-1491 [Abstract/Free Full Text].

20. Afshari, C. A., E. F. Nuwaysir, and J. C. Barrett. 1999. Application of complementary DNA microarray technology to carcinogen identification, toxicology, and drug safety evaluation. Cancer Res. 59: 4759-4760 [Abstract/Free Full Text].

21. Sgroi, D. C., S. Teng, G. Robinson, R. LeVangie, J. R. Hudson, and A. G. Elkahloun. 1999. In vivo gene expression profile analysis of human breast cancer progression. Cancer Res. 59: 5656-5661 [Abstract/Free Full Text].

22. Davies, D. E., R. Djukanovic, and S. T. Holgate. 1999. Application of functional genomics to study of inflammatory airways disease. Thorax 54: 79-81 [Free Full Text].

23. Heller, R. A., M. Schena, A. Chai, D. Shalon, T. Bedilion, J. Gilmore, D. E. Woolley, and R. W. Davis. 1997. Discovery and analysis of inflammatory disease-related genes using cDNA microarrays. Proc. Natl. Acad. Sci. USA 94: 2150-2155 [Abstract/Free Full Text].

24. Golub, T. R., D. K. Slonim, P. Tamayo, C. Huard, M. Gaasenbeek, J. P. Mesirov, H. Coller, M. L. Loh, J. R. Downing, M. A. Caligiuri, C. D. Bloomfield, and E. S. Lander. 1999. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 286: 531-537 [Abstract/Free Full Text].

25. Alizadeh, A. A., M. B. Eisen, R. E. Davis, C. Ma, I. S. Lossos, A. Rosenwald, J. G. Boldrick, H. Sabet, T. Tran, X. Yi, J. I. Powell, L. Yang, G. E. Marti, T. Moore, J. Hudson Jr., L. Lu, D. B. Lewis, R. Tibshrani, G. Sherlock, W. C. Chan, T. C. Greiner, D. D. Weisenberger, J. O. Armitage, R. Warnke, R. Levy, W. Wilson, M. R. Grever, J. C. Byrd, D. Botstein, P. O. Brown, and L. M. Staudt. 2000. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403: 503-511 [Medline].

26. Anbazhagan, R., T. Tihan, D. M. Bornman, J. C. Johnston, J. H. Saltz, A. Weigering, S. Piantadosi, and E. Gabrielson. 1999. Classification of small cell lung cancer and pulmonary carcinoid by gene expression profiles. Cancer Res. 59: 5119-5122 [Abstract/Free Full Text].

27. Simone, N. L., R. F. Bonner, J. W. Gillespie, M. R. Emert-Buck, and L. A. Liotta. 1998. Laser-capture microdissection: opening the microscopic frontier to molecular analysis. Trends Genet. 14: 272-276 [Medline].

28. Mahadevappa, M., and J. A. Warrington. 1999. A high-density probe array sample preparation method using 10- to 100-fold fewer cells. Nat. Biotechnol. 17: 1134-1136 .