The detailed map of the human genome can potentially transform future cancer therapy by merging genomics with pharmacology, thereby identifying which patients will benefit from specific therapeutic agents. Single-nucleotide polymorphisms (SNPs) provide a valuable tool for this pharmacogenetic approach to cancer therapy.
ABSTRACT: The detailed map of the human genome can potentially transform future cancer therapy by merging genomics with pharmacology, thereby identifying which patients will benefit from specific therapeutic agents. Single-nucleotide polymorphisms (SNPs) provide a valuable tool for this pharmacogenetic approach to cancer therapy. The discovery of SNPs as disease markers may facilitate identification of populations at increased risk for certain cancers. In addition, SNP genetic screening may facilitate administration of appropriate treatment modalities or reveal specific genetic profiles that have importance in drug efficacy and toxicity. In addition to SNP analysis, DNA and tissue microarray analyses have the potential to transform the future of cancer therapy. For example, DNA microarrays may improve tumor classification systems as well as provide a molecular level dissection of global gene expression changes that occur in carcinogenesis. Tissue microarrays would allow one to verify candidate genes, identified from DNA microarrays, against archival tumor specimens with known clinical outcome. In addition, both microarray technologies may be combined to rapidly validate gene targets. We will review and discuss these state-of-the-art technologies including data suggesting that the combined use of these high throughput technologies will facilitate our understanding of the genetic complexities inherent in cancer and will revolutionize cancer therapy. [ONCOLOGY 16 (Suppl 4):7-13, 2002]
The recently announced completion of the"working draft" of the human genome sequence is an important milestonein medical science.[1,2] Identifying all the genes and their regulatory regionsprovides the essential framework for the genetic blueprint of humankind and willfacilitate an understanding of the molecular basis of disease. In turn, thepostsequence era has brought forth a new foundation for a broad range of genomictools that can be applied to medical science, which will ultimately change thepractice of modern medicine.
Numerous lines of evidence have demonstrated that the risk ofdeveloping certain disorders and the metabolism of drugs used to treat theseconditions are influenced by one’s genetic makeup. With a detailedstructure of the human genome at hand, the future of cancer therapy involves amerging of genomics with pharmacology, in which underlying genomic variationscan be used to predict both the efficacy and toxicity of therapeutic agents.This pharmacogenetic approach will likely result in vastly improved patientcare.
In this article, we describe how the most abundant class ofgenetic variants in the human genome, termed single-nucleotide polymorphisms (SNPs),provides a valuable tool for pharmacogenetics in cancer therapy. We alsohighlight two high-throughput technologies, DNA and tissue microarrays, whichhave the potential to significantly augment the field of pharmacogenomics,defined by Roses as "the determination and analysis of the genome (DNA) andits products (RNA and proteins) as they relate to drug response."
The application of technologies such as SNP analysis, DNAmicroarray, and tissue microarray analysis will undoubtedly revolutionize cancertherapy. It will soon be possible to identify patients who respond or fail torespond to treatment early in the clinical drug development process. Thisinformation would provide a significant step towards "individualizing"cancer therapy and maximizing the benefits of treatment by tailoring patienttherapy.
The human genome, composed of approximately 3 billion base pairsof DNA, is commonly referred to as the "book of life." Chapters ofthis book represent individual chromosomes, the sentences represent genes, andthe words are codons made up of the DNA bases, adenine, cytosine, thymine, andguanine. It is estimated that approximately 99.9% of the genetic makeup of allindividuals is identical, leaving genomic sequence variance to less than afraction of 1% (0.01% or about 3 million bases). Though seemingly negligible,this 0.01% difference is significant indeed, because a single basechange/mutation can cause clinical disease. An individual is estimated to carryapproximately 300 to 1,200 deleterious mutations.
In addition to deleterious mutations, silent base pair changes(ie, changes that result in no apparent effect in an individual) seem to occurthroughout the genome, with an average frequency of 1 per 1,000 to 2,000bases.[6-8] Single base pair differences that occur when the DNA sequences ofindividuals are compared are SNPs (Figure 1). Intuitively, a high-density,genome-wide map of all these SNPs would help to create a fingerprint of thepolymorphic variants in each individual and would have significant implicationsfor disease gene discovery, diagnosis, and treatment.
Identifying and cataloguing these sequence variations to createa high-density SNP map of the entire human genome are the primary goals of TheSNP Consortium and the Human Genome Project (see http://www.ncbi.nlm.nih.gov/SNP/).Recently, a map was published of 1.42 million SNPs distributed throughout thehuman genome (an average density of 1 SNP per 1.9 kilobases), providing one ofthe first highly detailed marker maps of the sequence variability in humangenomes. This valuable resource continues to expand as more SNPs are added tothe SNP database.
Because of their mean density, stability, and high-throughputgenotyping capabilities, SNPs have recently emerged as genetic markers of choicefor disease gene discovery and mapping. Use of SNPs facilitates disease genemapping in two ways, genome-wide association studies and linkage disequilibriumanalysis. Single-nucleotide polymorphisms may be directly associated with adisease trait by effecting the expression or function of the gene where they arelocated. These "functional" SNPs may exist in a regulatory region, mayresult in an amino acid change in a gene product, or may alter the exon-intronsplicing pattern. Functional SNPs may be enriched in particular diseasepopulations compared with controls. It has been estimated that individuals areheterozygous for 24,000 to 40,000 polymorphisms that have been found to alteramino acid composition. However, it is thought that single disease-relatedSNP alleles can increase or modify risk for disease, but are not sufficient tocause disease.[12,13]
Alternatively, SNPs may be used as markers for linkagedisequilibrium.[14-17] Linkage disequilibrium is the measure of the degree ofassociation between two or more genetic markers that lie near each other on achromosome. Studies using linkage disequilibrium can identify regions of thegenome associated with a disease in a population. Single-nucleotidepolymorphisms that alter the risk of disease outcome will be the most predictiveof a possible clinical phenotype.
Genetic Screening for Treatment of Disease
Single-nucleotide polymorphism analysis also provides a usefultool in genetic screening for the treatment of disease. There are a number ofclinically relevant SNPs that have been shown to be associated with drugresponse and toxicity. Polymorphisms in genes that encode drug metabolizingenzymes for example, are observed at varying frequencies throughout the humanpopulation. Among the most commonly prescribed of all anticancer drugs, thethiopurines (eg, mercaptopurine [Purinethol] and thioguanine) must be convertedto thioguanine nucleotides by various enzymes in the body. These nucleotides arethen incorporated into the patient’s DNA. Polymorphisms in thedrug-metabolizing enzyme thiopurine methyltransferase have been linked to thetherapeutic efficacy of mercaptopurine, as well as to its toxicity. Patientswith two mutant thiopurine methyltransferase alleles have very low thiopurinemethyltransferase activity and, therefore, have an impaired capacity toeliminate mercaptopurine and thioguanine from the body. This results in serious,often life-threatening, toxicity.
Gene-specific polymorphisms have been observed in a number ofother drug-metabolizing enzymes, including dihydropyrimidine dehydrogenase,glucuronosyl transferase, Cyp17, glutathione transferase, cytochrome P-450, and5,10-methylenetetra-hydrofolate reductase. Furthermore, polymorphisms have alsobeen identified in genes that encode proteins involved in drug absorption,distribution, and elimination.[19-22] Thus, we are currently able to identifyinherited differences between individuals that may affect patient outcomes withanticancer drug therapies.
Apart from the known gene-specific polymorphisms that haverelevance to cancer treatment, how can nondisease/nongene specific SNP analysisbe used to predict patient response to medicine? An attractive, evolving modelis to obtain genome-wide SNP profiles from large numbers of cancer patientsreceiving anticancer drugs.[4,23] If a specific SNP pattern from patients whoresponded to therapy is compared with that of patients who failed to respond, acommon set of genetic variants between the two groups might be revealed.Additionally, SNP profiles from patients who experience adverse events duringtreatment can be compared with those patients who did not suffer adverse eventsto identify DNA regions associated with drug toxicity. Taken together, these SNPsignatures, or "medicine response profiles," provide a potentiallypowerful tool to predict whether an individual is likely to respond to a drug (Figure2).
SNP Integration Into Clinical Trials
Given this paradigm, SNP analysis has significant implicationsfor examining both drug efficacy and safety in clinical trials (Figure3).[4,24]Single-nucleotide polymorphism analysis can be implemented into clinical trialsin two ways. First, patients should be selected for phase III trials based onthe response profiles obtained from high-density SNP scoring of responders andnonresponders in phase II trials. Second, SNP profiles should be identified thatcharacterize patients who suffer serious or common drug adverse events comparedwith those patients who respond to therapy with no drug adverse events. The goalis to combine these two SNP profiles and generate a comprehensive medicalresponse to drug efficacy.
This SNP signature, or fingerprint, would result in thedevelopment of phase III trials that are faster to complete, require fewerpatients, and cost less to conduct. Moreover, lead molecules targeted toclinically nonresponsive patients could be more rapidly developed. A focusedclinical trial approach is important for individualsboth in terms of responseto treatment and in sparing the patient unnecessary adverse treatment effects.
Presently, the diagnosis and classification of human cancer isbased on a pathologic evaluation of the histology and morphology of a tumor,which is an essential step in determining appropriate treatment. The pathologicassessment of tumors, however, has limitations: it is subjective and there areno formal grading systems for many tumor types. In addition, the complexmolecular heterogeneity that drives and maintains the neoplastic state limitssuccessful therapy, in that tumor response and clinical outcome can varyconsiderably despite similar histopathological appearances.
Recent work in our laboratory and in a number of others hasshown remarkable progress toward an era in which cancer diagnosis will move froma traditional histologic/morphologic approach to a more molecular-basedassessment. The evolution of a new "molecular taxonomy" of cancer isbecoming possible through the use of the high-throughput genomic technologyknown as complementary DNA (cDNA) microarrays.[27,28] DNA microarray techniquesafford simultaneous expression monitoring of thousands of genes in a singleexperiment, engendering a molecular portrait or expression profile of both tumorand normal tissues. This is an eminently valuable step toward the basicunderstanding of the genetic complexities inherent in cancer and has significantclinical implications for revealing pathways and novel targets related to theneoplastic process.
DNA microarray technology is an RNA-based method of geneexpression analysis in which there are several formats for producing microarrays.[29,30]The two most commonly employed methods are oligonucleotide arrays and cDNAarrays. Oligonucleotide arrays, pioneered by Affymetrix, Inc. (Santa Clara,Calif), can be generated using a photolithographic process in whicholigonucleotides are synthesized directly onto a glass surface. The currentAffymetrix human genome GeneChip arrays contain 1,000,000 unique oligonucleotidefeatures that represent approximately 33,000 characterized human genes.Recently, other manufacturers, such as Motorola Life Sciences (Northbrook, Ill;CodeLink Bioarray System) and Agilent Technologies (Palo Alto, Calif; Custom InSitu Oligo Microarray Kit), have also introduced arrays using mechanicalmicrospotting or ink-jet printing instead of photolithography for deposition ofthe oligonucleotides onto glass slides.
The National Human Genome Research Institute of the NationalInstitutes of Health and a number of other laboratories use a system in whichcDNA microarrays are produced by robotically printing a large number of genesonto glass slides containing a gridded array (see http://www.nhgri.nih.gov/DIR/Microarray/main.htmlfor more information).[32-34] The spotted arrays varywith a "packingdensity" of up to 50,000 elements possible, although ranges from 5,000 to30,000 cDNAs are more common.
In general, the procedure requires high-quality mRNA that isisolated and purified from two samples (test and control), differentiallylabeled using reverse transcription in the presence of fluorescent dyes (eg,Cy3dUTP and Cy5dUTP), and hybridized overnight to the glass slide containing thearrayed cDNAs. Following a series of washes, the slide is scanned andmonochromatic images of each fluorescent channel are obtained. The images aremerged, pseudo-colored (eg, red and green), normalized, and the relativeexpression (level of red versus green fluorescence [R/G ratio]) between comparedsamples is obtained.
The statistical methods used to profile the vast quantities ofexpression data generated from DNA microarray analyses must be carefullyscrutinized, as they can have significant influence on the interpretation of theresults. Several computational tools are available for clustering and organizingthe data,[35-38] some of which have been summarized by Quackenbush. It isimportant to note, however, that no consensus has been reached as to the bestmethod for revealing patterns of gene expression changes. Thus, it is criticalthat molecular targets identified by DNA array technology be validated(discussed below).
The ability to analyze the simultaneous expression of thousandsof genes should greatly improve cancer classification and have a profound effecton the way new drug targets are identified. Gene-expression profiling using DNAmicroarrays has demonstrated unparalleled progress in the ability to subclassifypreviously unknown or indistinguishable subtypes of lymphoma, leukemia,breast carcinoma,[42-44] melanoma, prostate,[45,46] and ovarian cancer.Other studies utilizing cDNA microarrays have been designed to assess geneexpression patterns in estrogen-receptor-positive and -negative breasttumors, as well as familial breast cancers, where hereditary signatures forBRCA1 and BRCA2 mutated tumors were obtained.
The latter study was particularly important in demonstrating thepower of gene-expression profiling. In this study, one of the tumors in which noBRCA1 gene mutations were identified displayed a molecular profile that wascharacteristic of tumors with BRCA1 mutations. Subsequent analysis of thisobscure tumor revealed BRCA1 gene silencing through abnormal methylation in thepromoter region. These studies clearly illustrate that molecular profilingof tumors using DNA microarrays shows tremendous promise for improving tumorclassification.
Implicit in the use of DNA microarrays is the ability tomolecularly dissect global gene expression changes that occur in thecarcinogenic process. For example, gene expression changes that distinguishbenign prostatic hyperplasia from prostate cancer may reveal genes that areimportant in prostate tumorigenesis.[45,46] Not surprisingly, striking degreesof molecular variation are often revealed among samples, particularly whencomparing normal tissue with its cancerous counterpart, making it is essentialto establish a database that permits the data management and comparison of largequantities of information. As discussed, the relationships between the geneexpression changes require sophisticated data mining techniques, such ashierarchical clustering, multidimensional scaling, artificial neural networks,and self-organizing maps.
Global gene expression analysis has provided great insight intothe molecular heterogeneity of specific types of cancer. However, because it isoften difficult to assign functions to a large number of identified genes,cancer pathway discovery remains in its infancy. Still, molecular profilingholds tremendous promise for the identification of novel targets for therapy andthe development of drugs tailored to the genetic changes critical for malignanttransformation and tumor progression. Several studies are now beginning to usetumor samples with available follow-up information to assess the prognosticsignificance of gene-expression profiles.
An elegant example illustrating the clinical relevance of DNAmicroarray technology was provided by Alizadeh et al. Using a "lymphochip"containing 17,856 genes expressed in lymphoid cells, Alizadeh and colleaguesperformed a global gene expression analysis on diffuse large B-cell lymphoma, acancer that displays great clinical heterogeneity (only 40% of patients respondto therapy). Here, an unsupervised clustering of the data was able to dissecttwo molecularly distinct subsets of the disease into germinal center B-likediffuse large B-cell lymphoma and activated B-like diffuse large B-celllymphoma. More importantly, the authors demonstrated that patients classifiedwith germinal center B-like diffuse large B-cell lymphoma had better overallsurvival than those patients with the molecular profile characteristic of B-likediffuse large B-cell lymphoma. This study illustrates how specific molecularportraits may be able to predict patient outcome.
Similarly, an important study by Shipp et al utilizedgene-expression profiling to predict the outcome in a series of diffuse largeB-cell lymphoma patients. Thirteen key genes were identified as topprognosis discriminators for patients who had a high chance of dying andpatients that demonstrated reduced recurrence following treatment. These resultsdiffer from the former study in that Alizadeh et al hypothesized a cell oforigin (ie, germinal center B-like or activated B-like) might be predictive of aclinical response, whereas Shipp and colleagues used supervised learning methodsto correlate outcome with gene expression.
Another example of the clinical relevance of cDNA microarraytechnology was recently demonstrated with a set of pediatric tumors, thechildhood small-, round-, blue-cell tumors,  which include Ewing’ssarcoma, neuroblastoma, rhabdomyosarcoma, and non-Hodgkin’s lymphoma. Thisclass of tumors is often difficult to diagnose and shows variable response tochemotherapy. Khan et al identified a set of 80 known genes and 13anonymously expressed sequence tags that discriminate the four distinct subtypesof small-, round-, blue-cell tumors. Of interest, one commonly used marker todiagnose Ewing’s sarcoma, MIC2, was highly expressed in severalrhabdomyosarcoma samples. These results suggest MIC2 alone may not represent anappropriate diagnostic marker for Ewing’s sarcoma.
DNA microarray technology shows great promise for revealing geneexpression changes that are predictive of tumor response and sensitivity totherapeutic agents. However, the majority of drug response studies, includingstudies of drug resistance, have been limited to cell culture analyses.[52-54] Amassive effort to identify in vitro growth inhibitory activities of 60,000compounds has been established at the National Cancer Institute. It isanticipated that a DNA microarray approach to examine gene expression changes ina panel of 60 cancer cell lines exposed to various compounds will help toidentify new agents for clinical trials.
Several studies have demonstrated the remarkable power of DNAmicroarrays to reveal the underlying molecular diversity among subtypes ofcancer. These applications will undoubtedly have tremendous impact on efforts toindividualize cancer therapy and improve treatment success. Molecular targetsidentified by global gene expression analysis may be useful diagnostic orprognostic indicators and may be important indicators for cancer therapy failureor toxicity. Of significant importance, however, is that the future goal ofusing high-throughput array technologies as a step toward a tailored treatmentapproach relies heavily on the careful selection and evaluation of patients whohave undergone treatment in clinical trials and on access to large, publiclyavailable data sets that can define clinical objectives.
It is necessary to validate the genes that are leadingindicators from array experiments to determine if they can provide meaningfulinformation for clinical applications. One important process may be to verifythe gene against archival tumor specimens with known clinical outcome. However,gene-expression profiling reveals large numbers of genes (eg, 10 to 300 genesper tumor) that must be validated in large sets of tumors to obtainstatistically significant data. Validation would take years to complete in atraditional setting and could potentially exhaust banked archival specimens. Forthese reasons, tissue microarray technology is a vital technological advancethat permits high-throughput validation of candidate target genes.
Tissue microarrays can contain up to 2,500 different archivaltumor samples on a single glass slide. In this method, a core biopsy of tissue(0.6 mm in diameter) is punched from a preselected region of a paraffin-embeddedtumor (the donor block) and placed into a recipient paraffin block containingpremade holes in a defined array (Figure 4). Each hole receives a differenttumor specimen. The filled recipient block is sectioned and used for thesimultaneous in situ analysis at the DNA, RNA, and protein level.
Several studies have combined DNA and tissue microarraytechnologies to validate gene targets. For example, Hedenfalk et al used tissuemicroarray technology to demonstrate that the protein levels (determined byimmunohistochemistry) of two selected genes, cyclin D1 and mitogen-activatedprotein kinase kinase-1, correlated with the DNA microarray results (Figure5). Similarly, Bubendorf et al used tissue microarrays containing a broadspectrum of prostate cancer samples to validate overexpression of insulin-likegrowth factor-binding protein 2 and heat-shock protein (HSP27) genes identifiedin DNA microarray analysis of hormone-refractory prostate cancer. Theseresults demonstrate the power of combining two high-throughput technologies tovalidate candidate target genes more efficiently and rapidly.
The growing body of data shows convincing evidence that thecombined use of the high-throughput technologies described here is certain tounravel the genetic complexities inherent in cancer and revolutionize cancertherapy. Single-nucleotide polymorphism analysis, combined with cDNA microarraysand tissue microarrays, should help to reveal specific genetic profiles andnovel molecular targets that have importance in drug efficacy and toxicity.Linking these technologies could be a key element in the quest to broadly andrapidly develop approaches to test individual patients’ tumors for specifictherapeutic targets. These technologies may, indeed, revolutionize cancerdiagnosis and treatment.
1. International Human Genome Sequencing Consortium: Initialsequencing and analysis of the human genome. Nature 409:860-921, 2001.
2. Venter JC, Adams MD, Myers EW, et al: The sequence of thehuman genome. Science 291:1304-1351, 2001.
3. Evans WE, Relling MV: Pharmacogenomics: Translatingfunctional genomics into rational therapeutics. Science 286:487-491, 1999.
4. Roses AD: Pharmacogenetics and the practice of medicine.Nature 405:857-865, 2000.
5. Fay JC, Wyckoff GJ, Wu CI: Positive and negative selection onthe human genome. Genetics 158:1227-1234, 2001.
6. Altshuler D, Pollara VJ, Cowles CR, et al: An SNP map of thehuman genome generated by reduced representation shotgun sequencing. Nature407:513-516, 2000.
7. The International SNP Map Working Group: A map of humangenome sequence variation containing 1.42 million single-nucleotidepolymorphisms. Nature 409:928-933, 2001.
8. Wang DG, Fan JB, Siao CJ, et al: Large-scale identification,mapping, and genotyping of single-nucleotide polymorphisms in the human genome.Science 280:1077-1082, 1998.
9. Marshall E: Drug firms to create public database of geneticmutations. Science 284:406-407, 1999.
10. Collins FS, Guyer MS, Charkravarti A: Variations on a theme:Cataloging human DNA sequence variation. Science 278:1580-1581, 1997.
11. Cargill M, Altshuler D, Ireland J, et al: Characterizationof single-nucleotide polymorphisms in coding regions of human genes. Nat Genet22:231-238, 1999.
12. Brookes AJ: The essence of SNPs. Gene 234:177-186, 1999.
13. Pritchard JK: Are rare variants responsible forsusceptibility to complex diseases? Am J Hum Genet 69:124-137, 2001.
14. Gray IC, Campbell DA, Spurr NK: Single-nucleotidepolymorphisms as tools in human genetics. Hum Mol Genet 9:2403-2408, 2000.
15. Kruglyak L: Prospects for whole-genome linkagedisequilibrium mapping of common disease genes. Nat Genet 22:139-144, 1999.
16. Reich DE, Cargill M, Bolk S, et al: Linkage disequilibriumin the human genome. Nature 411:199-204, 2001.
17. Riley JH, Allan CJ, Lai E, et al: The use ofsingle-nucleotide polymorphisms in the isolation of common disease genes.Pharmacogenomics 1:39-47, 2000.
18. Krynetski EY, Evans WE: Genetic polymorphism of thiopurineS-methyltransferase: Molecular mechanisms and clinical importance. Pharmacology61:136-146, 2000.
19. Caldwell J, Gardner I, Swales N: An introduction to drugdisposition: The basic principles of absorption, distribution, metabolism, andexcretion. Toxicol Pathol 23:102-114, 1995.
20. Kleyn PW, Vesell ES: Genetic variation as a guide to drugdevelopment. Science 281:1820-1821, 1998.
21. Relling MV, Dervieux T: Pharmacogenetics and cancer therapy.Nature Rev Cancer 1:99-108, 2001.
22. Vesell ES: Advances in pharmacogenetics and pharmacogenomics.J Clin Pharmacol 40:930-938, 2000.
23. Pfost DR, Boyce-Jacino MT, Grant DM: A SNPshot:Pharmacogenetics and the future of drug therapy. Trends Biotechnol 18:334-338,2000.
24. McLeod HL, Evans WE: Pharmacogenomics: Unlocking the humangenome for better drug therapy. Annu Rev Pharmacol Toxicol 41:101-121, 2001.
25. Fijal BA, Hall JM, Witte JS: Clinical trials in the genomicera: Effects of protective genotypes on sample size and duration of trial.Control Clin Trials 21:7-20, 2000.
26. Elston CW, Sloane JP, Amendoeira I, et al: Causes ofinconsistency in diagnosing and classifying intraductal proliferations of thebreast. European Commission Working Group on Breast Screening Pathology. Eur JCancer 36:1769-1772, 2000.
27. Duggan DJ, Bittner M, Chen Y, et al: Expression profilingusing cDNA microarrays. Nat Genet 21:10-14, 1999.
28. Khan J, Saal LH, Bittner ML, et al: Expression profiling incancer using cDNA microarrays. Electrophoresis 20:223-229, 1999.
29. Burgess JK: Gene expression studies using microarrays. ClinExp Pharmacol Physiol 28:321-328, 2001.
30. Cunningham MJ: Genomics and proteomics: The new millenniumof drug discovery and development. J Pharmacol Toxicol Methods 44:291-300, 2000.
31. Lockhart DJ, Dong H, Byrne MC, et al: Expression monitoringby hybridization to high-density oligonucleotide arrays. Nat Biotechnol14:1675-1680, 1996.
32. DeRisi J, Penland L, Brown PO, et al: Use of a cDNAmicroarray to analyse gene expression patterns in human cancer. Nat Genet14:457-460, 1996.
33. Schena M, Shalon D, Davis RW, et al: Quantitative monitoringof gene expression patterns with a complementary DNA microarray. Science270:467-470, 1995.
34. Shalon D, Smith SJ, Brown PO: A DNA microarray system foranalyzing complex DNA samples using two-color fluorescent probe hybridization.Genome Res 6:639-645, 1996.
35. Bittner M, Meltzer P, Chen Y, et al: Molecularclassification of cutaneous malignant melanoma by gene expression profiling.Nature 406:536-540, 2000.
36. Chen Y, Dougherty ER, Bittner ML: Ratio-based decisions andthe quantitative analysis of cDNA microarray images. J Biomed Opt 2:364-374,1997.
37. Ermolaeva O, Rastogi M, Pruitt KD, et al: Data managementand analysis for gene expression arrays. Nat Genet 20:19-23, 1998.
38. Kim S, Dougherty ER, Chen Y, et al: Multivariate measurementof gene expression relationships. Genomics 67:201-209, 2000.
39. Quackenbush J: Computational analysis of microarray data.Nat Rev Genet 2:418-827, 2001.
40. Alizadeh AA, Eisen MB, Davis RE, et al: Distinct types ofdiffuse large B-cell lymphoma identified by gene expression profiling. Nature403:503-511, 2000.
41. Golub TR, Slonim DK, Tamayo P, et al: Molecularclassification of cancer: Class discovery and class prediction by geneexpression monitoring. Science 286:531-537, 1999.
42. Ahr A, Holtrich U, Solbach C, et al: Molecularclassification of breast cancer patients by gene expression profiling. J Pathol195:312-320, 2001.
43. Hedenfalk I, Duggan D, Chen Y, et al: Gene-expressionprofiles in hereditary breast cancer. N Engl J Med 344:539-548, 2001.
44. Perou CM, Jeffrey SS, van de Rijn M, et al: Distinctive geneexpression patterns in human mammary epithelial cells and breast cancers. ProcNatl Acad Sci U S A 96:9212-9217, 1999.
45. Dhanasekaran SM, Barrette TR, Ghosh D, et al: Delineation ofprognostic biomarkers in prostate cancer. Nature 412:822-826, 2001.
46. Luo J, Duggan DJ, Chen Y, et al: Human prostate cancer andbenign prostatic hyperplasia: Molecular dissection by gene expression profiling.Cancer Res 61:4683-4688, 2001.
47. Welsh JB, Zarrinkar PP, Sapinoso LM, et al: Analysis ofgene-expression profiles in normal and neoplastic ovarian tissue samplesidentifies candidate molecular markers of epithelial ovarian cancer. Proc NatlAcad Sci U S A 98:1176-1181, 2001.
48. Gruvberger S, Ringner M, Chen Y, et al: Estrogen receptorstatus in breast cancer is associated with remarkably distinct gene expressionpatterns. Cancer Res 61:5979-5984, 2001.
49. Shipp MA, Ross KN, Tamayo P, et al: Diffuse large B-celllymphoma outcome prediction by gene-expression profiling and supervised machinelearning. Nat Med 8:68-74, 2002.
50. Khan J, Wei JS, Ringner M, et al: Classification anddiagnostic prediction of cancers using gene expression profiling and artificialneural networks. Nat Med 7:673-679, 2001.
51. Triche TJ: Diagnosis of small-, round-cell tumors ofchildhood. Bull Cancer 75:297-310, 1988.
52. Chang BD, Swift ME, Shen M, et al: Molecular determinants ofterminal growth arrest induced in tumor cells by a chemotherapeutic agent. ProcNatl Acad Sci U S A 99:389-394, 2002.
53. Clarke PA, Hostein I, Banerji U, et al: Gene expressionprofiling of human colon cancer cells following inhibition of signaltransduction by 17-allylamino-17-demethoxygeldanamycin, an inhibitor of thehsp90 molecular chaperone. Oncogene 19:4125-4133, 2000.
54. Kudoh K, Ramanna M, Ravatn R, et al: Monitoring theexpression profiles of doxorubicin-induced and doxorubicin-resistant cancercells by cDNA microarray. Cancer Res 60:4161-4166, 2000.
55. Scherf U, Ross DT, Waltham M, et al: A gene expressiondatabase for the molecular pharmacology of cancer. Nat Genet 24:236-244, 2000.
56. Kononen J, Bubendorf L, Kallioniemi A, et al: Tissuemicroarrays for high-throughput molecular profiling of tumor specimens. Nat Med4:844-847, 1998.
57. Bubendorf L, Nocito A, Moch H, et al: Tissue microarray(TMA) technology: Miniaturized pathology archives for high-throughput in situstudies. J Pathol 195:72-79, 2001.
58. Bubendorf L, Kolmer M, Kononen J, et al: Hormone therapyfailure in human prostate cancer: Analysis by complementary DNA and tissuemicroarrays. J Natl Cancer Inst 91:1758-1764, 1999.