Molecular Diagnosis of Colorectal Tumors by Expression Profiles of 50 Genes Expressed Di€Erentially in Adenomas and Carcinomas

Molecular diagnosis of colorectal tumors by expression pro®les of 50 genes expressed dierentially in adenomas and carcinomas

Yu-Min Lin1,2, Yoichi Furukawa1, Tatsuhiko Tsunoda3, Chung-Tai Yue2, Kou-Ching Yang2 and Yusuke Nakamura*,1

1Human Genome Center, Institute of Medical Science, University of Tokyo, 4-6-1 Shirokanedai, Minato-ku, Tokyo 108-8639, Japan; 2Department of Internal Medicine, Shin Kong Wu Ho-Su Memorial Hospital, 95 Wen Chang Road, Taipei 11160, Taiwan; 3SNP Research Center, Riken (Institute of Physical and Chemical Research), 4-6-1 Shirokanedai, Minato-ku, Tokyo 108-8639, Japan

Most colon cancers are thought to develop through the treatment, but many patients with advanced colorectal `adenoma-to-carcinoma sequence' model. To elucidate cancers still wait for more eective but less harmful the mechanisms underlying this pathway, we analysed therapeutic drugs. Development of a sensitive, speci®c gene-expression pro®les of 20 colorectal tumors (nine and convenient diagnostic system for detecting very adenomas and 11 dierentiated adenocarcinomas) by early-stage colorectal cancers or pre-malignant lesions means of a cDNA microarray representing 23 040 genes could ultimately eliminate this disease. Furthermore, coupled with laser-capture microdissection. A two- identi®cation of eective preventive strategies would dimensional hierarchical clustering analysis of expression release people from fear of this life-threatening illness. pro®les of the 20 tumors correctly separated the Muto et al. (1975) ®rst described pre-malignant carcinoma group from the adenoma group. Furthermore lesions of the colon on the basis of careful histological we identi®ed 51 genes whose expression was commonly examination and proposed a model of `adenoma- up-regulated, 376 that were commonly down-regulated in carcinoma sequence of the colon'. Subsequent mole- both types of tumors as opposed to normal colonic cular studies during the past two decades have opened epithelium and 50 whose expression levels were the door to a better understanding of colorectal signi®cantly dierent between adenomas and carcinomas. carcinogenesis and established a view of multi-step On the basis of expression pro®les of the 50 discriminat- carcinogenesis in the colon that agrees well with ing genes, we established a scoring system to separate Muto's. Investigations of hereditary colon cancers have adenomas from carcinomas. Application of this scoring improved the general understanding of mechanisms system for evaluating ®ve additional colorectal tumors underlying sporadic colorectal carcinomas. However, correctly predicted their histological features. The our knowledge of genes involved in colorectal tumors genome-wide information reported here should contribute is still fragmentary and awaits genome-wide informa- to a more profound understanding of colorectal tumor- tion about how expression pro®les are altered during igenesis, particularly of adenoma-carcinoma progression, multi-step carcinogenesis. and provide indicators for developing novel strategies to Analysis of expression pro®les using cDNA micro- diagnose, treat, and ultimately prevent colorectal arrays is a promising method for identifying genes carcinomas. whose expression is altered by response to various Oncogene (2002) 21, 4120 ± 4128. doi:10.1038/sj.onc. physiological conditions or drugs, or pathological 1205518 conditions (Afshari et al., 1999; Kallioniemi, 2001). This technology can also provide systematic expression Keywords: colon cancer; adenoma-carcinoma sequence; pro®les that may re¯ect the physiological or patholo- cDNA microarray; expression pro®le gical phenotype of a subset of a cell population. This approach has proven to be useful for analysing genes involved in various neoplasms and for detecting speci®c phenotypes (Ono et al., 2000; Okabe et al., Introduction 2001; Kitahara et al., 2001). We were interested in examining expression pro®les of pre-malignant and Colorectal carcinoma is a leading cause of cancer malignant lesions of the colon, because comparison of deaths in developed countries. Recent medical ad- these two groups of tumors would provide information vances have made great progress in diagnosis and about genes that undergo altered expression during progression of an adenoma to a carcinoma. We report here a genome-wide analysis of gene- expression pro®les in nine pre-malignant lesions and 11 *Correspondence: Y Nakamura; E-mail: [email protected] Received 22 October 2001; revised 11 March 2002; accepted 22 dierentiated adenocarcinomas of the colon by means March 2002 of a cDNA microarray representing 23 040 genes. We Analysis of expression profiles in colonic tumors Y-M Lin et al 4121 identi®ed a panel of genes that were commonly deregulated in both colorectal adenomas and carcinomas, and also 50 genes whose expression diered signi®cantly between benign and malignant tumors. We established a scoring system on the basis of expression of selected genes that may assist clinicians in distinguishing adenomas from carcinomas. The data reported here will add to a comprehensive understanding of colorectal carcinogenesis, facilitate development of novel diagnostic strategies, and provide clues for identi®cation of molecular targets for therapeutic drugs and preventive agents.

Results

Two-dimensional hierarchical clustering To analyse correlation among the samples and genes, we applied a two-dimensional hierarchical clustering algorithm (http://www.microarrays.org/software) using the data obtained from the 20 tumors. We excluded genes for further analysis when average Cy3-and Cy5- ¯uorescence intensities were below 16105 units, and selected a set of genes whose values were obtained in more than 16 cases (80%). We further excluded genes with standard deviations of observed values less than 0.5. A total of 771 genes passed this ®lter for subsequent clustering analysis. In the sample axis, the 20 samples were separated into two major groups based on their expression pro®les; all of the nine tumors belonging to one group were adenomas and the other major group consisted of the 11 carcinomas (Figure 1). This result is consistent with a recent report that four colon adenomas were separated from 18 adenocarcinomas using oligonucleo- tide arrays (Notterman et al., 2001). The expression pro®les obtained on our microarray clearly demon- strated that adenomas and adenocarcinomas have Figure 1 A two-dimensional hierarchical clustering of 771 genes speci®c expression pro®les, and indicated that mole- across 20 colorectal tumors. The color in each well represents cular classi®cation of colonic tumors is feasible. Our relative expression of each gene (vertical axis) in each paired sample (horizontal axis); more intense colors re¯ect wider data also disclosed the molecular nature of each tumor dierences between tumor and normal epithelium. Red, increased type. In the gene axis, we later focused on two gene in tumor; green, decreased; black, unchanged; gray, no expression clusters, cluster A (more abundant expression in in the tumor cells. In the sample axis, carcinomas (T) and carcinomas than adenomas) and cluster B (more adenomas (P) were separated to two dierent trunks. In the gene abundant expression in adenomas than carcinomas). axis, 771 genes were clustered in dierent branches according to their similarity; the shorter the branches the greater the similarity. Sub-clusters A and B were selected for further analysis (see text Genes commonly up-regulated in colonic tumors and Figure 3) Since most colon cancers arise from adenomas, genes involved in early stages of colorectal tumorigenesis are normal epithelia (Figure 2). Among the 51 genes, 19 likely to be deregulated in both types of tumors. To were involved in RNA/protein processing; e.g. ribo- identify such genes, we selected genes from our data set somes, translation elongation/initiation factors, and of 2425 genes according to the following criteria: if the chaperonins. Other up-regulated genes detected in our Cy3/Cy5 ratio of the gene was 42 in more than 50% experiments included oncogenes (HMGIY, DEK and of the tumors, we de®ned it as a commonly up- NPM1), genes encoding cell adhesion/cytoskeleton regulated gene, and if the ratio was 50.5 in more than molecules (TUBB, K-ALPHA, TGFBI, CDH3 and 50% of the tumors, we de®ned it as commonly down- PAP), genes involved in growth control (IMPDH2 regulated. With these criteria we identi®ed 51 genes and ODC1), signal transduction (BRF1, PLAB, that were commonly up-regulated in both tumor LAP18, CD81 and MACMARCKS), and cell-cycle phenotypes as compared with their corresponding control (RAN and UBE2I); transcription factors

Oncogene Analysis of expression profiles in colonic tumors Y-M Lin et al 4122

Figure 2 Fifty-one genes commonly up-regulated in both adenomas and carcinomas. (Cy3/Cy5)ave indicates average value of Cy3/ Cy5 in the 20 paired samples. Genes that appear repeatedly represent the same genes spotted on dierent set of slides. Red, increased in tumor; green, decreased; black, unchanged; gray, no expression in the tumor cells

(HMG1 and HMG2), tumor-associated molecules of cellular homeostasis in tumors (Lawrance et al., (PPP2R1B, LDHB and SLC29A1), and others. 2001).

Genes commonly down-regulated in colonic tumors Genes expressed differently between adenomas and carcinomas We also identi®ed 376 genes (including 127 expressed sequence tags) that were commonly down-regulated in When we focused on genes in clusters A and B both types of tumor by the criteria described above (Figure 3), we identi®ed several that regulate (supplementary data). This group includes genes bioenergetics. In cluster A, PGK1 and LDHA may associated with programmed cell death (CASP8, be induced by hypoxia (Semenza et al., 1994). CASP9, CFLAR, DFFA, PAWR, TNF, TNFRSF10C Proteasomes (PSMD7 and PSMB8) have been and TNFRSF12), immunity (chemokine receptors such reported to accumulate in cancer cells by glucose as IL1RL2, IL17R and IL3RA), growth suppression starvation and hypoxia (Ogiso et al., 1999). VDAC3 (Suppressin, DCN, MADH2 and SST), and tumor is one of the voltage-dependent anion-selective suppression (TP53). Other down-regulated genes en- channel proteins that play important roles in code cell adhesion/cytoskeleton molecules (e.g. regulating mitochondrial homeostasis (Vander Heiden ADAM8, AVIL, CDH17, CEACAM1, CTNNA2, et al., 2000). GSS is a regulator of oxidative stress ICAPA, KRT9, and ARHGAP5), various metabolic (Uhlig and Wendel, 1992). Some of the genes in factors (e.g. BPHL, CA2, CA5A, HSD11B2 and cluster B encode proteins that have been reported to ECHS1), ion transporters (SLC15A2, SLC22A1, function in adaptation to low-oxygen conditions SLC4A3 and SLC5A1), a natural antimicrobial (GPX2, PPIA, GAPD, ANXA2, ALDH1 and ADAR) molecule (DEFA6), and others. Metabolic enzymes (Chu et al., 1993; Zhong and Simons, 1999; Hoeren and ion-transport mediators are key factors for et al., 1998; Denko et al., 2000), in energy maintaining pivotal cellular functions such as detoxica- consumption (ATP6A1, ATP1B1, and ATP5A1) tion (CA2, CA5A and BPHL) and acid-base balance. (Wodopia et al., 2000) or in carbohydrate metabolism Down-regulation of these genes indicates a disruption (GMDS).

Oncogene Analysis of expression profiles in colonic tumors Y-M Lin et al 4123

Figure 3 Functional clusters in the gene axis. (A) Ten of 24 genes in cluster A and (B) 12 of 29 genes in cluster B were related to bioenergetics homeostasis (labeled in blue). Genes that appear repeatedly represent the same genes spotted on dierent set of slides

sion pro®les in two colon cancer tissues and two non- Verification of microarray data by quantitative- cancerous colonic mucosae analysed by means of Serial RT ± PCR Analysis of Gene Expression was provided by National To examine the reliability of our microarray data, we Center for Biotechnology Information (http:// selected six genes, two of which (TGFBI and LAP18) www.ncbi.nlm.nih.gov/SAGE/). Among 100 tags of were up-regulated in both adenomas and carcinomas genes expressed most dierently between the cancer and the others (HECH, NME1, TCEA1 and PSMA7) and non-cancerous tissues, 50 tags corresponded to were dierently expressed between adenomas and independent unique genes in UniGene database. carcinomas, and examined their expression in 13 Among 50 genes corresponding to these 50 tags, four additional paired aRNA samples (seven adenomas up-regulated genes and 24 down-regulated genes in and six carcinomas) by quantitative RT ± PCR cancer were contained in our microarray. One of the (QRT ± PCR). The results were highly similar to four up-regulated gene, TGFBI (transforming growth microarray data for all six genes (Figure 4). These factor, beta-induced, 68 kD) also showed elevated data veri®ed the reliability and rationality of our expression in our data (Figure 2). Eighteen of the 24 strategy to identify genes that are commonly up- down-regulated genes including TSPAN-1 (tetraspan1), regulated or dierently expressed during development GPA33 (glycoprotein A33), CA1 (carbonic anhydrase and progression of colorectal cancer. 1), MT2A (metallothionein 2A), CEACAM1 (carci- noembryonic antigen-related cell adhesion molecule 1), YF13H12 (protein expressed in thyroid), MUC13 Comparison of expression analysis data in colon cancers (mucin 13), HLAB (major histocompatibility complex We additionally compared our data with two sets of class 1B), DUSP1 (dual speci®city phosphatase 1), data reported previously. Information of gene expres- GSN (gelsolin), LGALS4 (galectin 4), CKB (creatinine

Oncogene Analysis of expression profiles in colonic tumors Y-M Lin et al 4124

Figure 4 Validation of microarray data. (A) Log2(Cy3/Cy5) values of 20 samples (11 carcinomas and nine adenomas) in cDNA microarray analysis. (B) Log2(Tumor/Normal) values for 13 additional samples (six carcinomas and seven adenomas) obtained by QPCR. The expression ratio (Tumor:Normal) of TGFBI, LAP18, HECH, NME1, TCEA1 and PSMA7 determined by QPCR were in line with the microarray data for all six genes. Data are presented here as 10th to 90th percentiles of calculated values. Statistical signi®cance was examined by Mann ± Whitney U-tests

kinase, brain type), KRT19 (keratin 19), RNASE1 of the nine adenomas was 75.9+14.4 (mean+s.d., (RNase A family 1), IFI27 (interferon, alpha-inducible P50.0001, Mann ± Whitney U-test; Figure 5B). We protein 27), PP1201, EPS8R2 (FLJ21935), and an de®ned the cut-o value for discriminating adenocarci- EST(Hs. 107139) revealed decreased expression in more noma from adenoma at 35, an average of the mean than half the cases examined by us. We also compared values of the two groups, and analysed ®ve additional a list of genes highly expressed in colon cancer tissues tumors to verify the reliability of the MDS system. compared to the matched non-cancerous tissues, which Among the ®ve samples tested, three that showed was reported by Notterman et al. (2001) using the scores greater than 35 (73.5, 63.2, and 64.6) all turned Aymetrix Human 6500 GeneChip Set, and found that out to be carcinomas by histological examination. The only two genes, GTF3A (general transcription factor two samples with scores of 10.3 and 71.4 were both IIIA) and AHCY (adenosylhomocysteine hydrolase), adenomas (Figure 5B). Since the distribution of MDSs were in our list of frequently up-regulated genes in for 14 carcinomas ranging from 57.7 to 94.1 were cancer (Figure 5A). However, among 10 other genes completely separated from that for 11 adenomas that were highly expressed in their cancer tissues and ranging from 733.4 to 16.7, both sensitivity and spotted on our array slides, eight genes, such as speci®city of the MDS system were 100% on the basis KIAA0101, PYCR1 (pyrroline 5-carboxylate reduc- of this cut-o value. In addition, a hierarchical tase), HSPE1 (heat shock 10 kD protein 1), CDC25B clustering analysis of all 25 samples correctly separated (cell division cycle 25B), CSE1L (chromosome segrega- adenomas from carcinomas based on the expression tion 1-like), CKS2 (CDC28 protein kinase 2), MMP1 pro®les of the 50 selected genes (Figure 5A). (metalloprotenase 1), and CLNS1A (chloride channel, nucleotide-sensitive, 1A), also showed enhanced expression in more than half of cancer tissues. Discussion

In this study, we disclosed common nature of adenoma Development of a `Molecular Diagnosis Score' (MDS) and adenocarcinoma of the colon through the analysis system of genome-wide expression pro®les. The 51 genes Among the genes expressed dierently between adeno- commonly up-regulated in both adenomas and carcinomas and carcinomas, we identi®ed 50 whose expression mas included 19 involved in RNA/protein processing; showed statistically signi®cant dierences between the e.g. ribosomes, translation elongation/initiation factors, two types of tumors (P40.01, Mann ± Whitney U-test; and chaperonins. Ribosomes are the molecular machines Figure 5A). We developed a `Molecular Diagnosis that manufacture proteins according to blueprints of Score' (MDS) system based on expression pro®les of mRNAs that encode them. Interactions of the ribosome these 50 genes (see Materials and methods) as a way to with mRNAs, tRNAs, and a number of non-ribosomal apply that information to clinical diagnosis. The mean protein cofactors such as translation initiation/elonga- score of the 11 carcinomas was 77.4+11.6, while that tion factors guarantee that polypeptide chains are

Oncogene Analysis of expression profiles in colonic tumors Y-M Lin et al 4125

Figure 5 `Molecular diagnosis score' (MDS) based on expression of 50 genes whose levels of expression were statistically dierent between adenomas and carcinomas. (A) Clustering analysis. The data for each gene were ®rst median-centered, and an `Average Linkage Clustering' was subsequently applied to the data set (red, data 4median value; green, data 5median value). In the sample axis, 25 tumors were separated to two trunks (adenoma group and carcinoma group). Asterisks (*) indicate additional test samples. A sample, 056P3, was diagnosed as an early adenocarcinoma by histological examination. In the gene axis, the 18 genes on the top showed higher expression in carcinoma than in adenoma, and were labeled with `1' as a sign. The 32 genes at the bottom showed higher expression in adenoma than in carcinoma, and were labeled with `71' as a sign (see Materials and methods). Statistical signi®cance was examined by the Mann ± Whitney U-test. (B) MDSs were presented as 10 ± 90th percentiles of the calculated values. Asterisks denote the ®ve additional samples for validating the MDS system. Tumor 056P3 is a well-dierentiated adenocarcinoma

Oncogene Analysis of expression profiles in colonic tumors Y-M Lin et al 4126 Table 1 Genes, sequences of primers and probes used for quantitative RT ± PCR Symbol Primer Probe

MDH1 F:5'-TCCCTGTTGTAATCAAGAATAAGACCT-3' 5'-Vic-TTGTTGAAGGTCTCCCTATTAATGATTTCTCACGTG- R:5'-CAGTTCCTTTGCAGTAAGATCCATC-3' Tamra-3' TGFBI F:5'-GCAGACTCTGCGCTTGAGATC-3' 5'-Fam-AACAAGCATCAGCGTTTTCCAGGGCT-Tamra-3' R:5'-GGGCTAGTCGCACAGACCTC-3' LAP18 F:5'-CAAATGGCTGCCAAACTGG-3' 5'-Fam-CGTTTGCGAGAGAAGGATAAGCACATTGAAG-Tamra-3' R:5'-GGATTCTTTGTTCTTCCGCACT-3' HECH F:5'-GACAGCAGTGGAGAATTGATGTTT-3' 5'-Fam-TGAGGCAGACTTGGTGCTGGCG-Tamra-3' R:5'-ACAATTTGAGGACACTTCATATTTGC-3' MNE1 F:5'-GCATACAAGTTGGCAGGAACATTA-3' 5'-Fam-CATGGCAGTGATTCTGTGGAGAGTGCA-Tamra-3' R:5'-ACCACAAGCCGATCTCCTTCT-3' TCEA1 F:5'-AGATGCGGAAAAACTTGACCA-3' 5'-Fam-AAGCCATCAGAGAGCATCAGATGGCC-Tamra-3' R:5'-AGTCAGTCTGGGTCCCACCA-3' PSMA7 F:5'-GCAGCGTTATACGCAGAGCA-3' 5'-Fam-TGGGCGCAGGCCGTTTGG-Tamra-3' R:5'-ACCCACGATGAGGGCAGAG-3'

initiated, elongated and terminated (Maguire and self-destruction machinery, alteration of cell structure, Zimmerman, 2001). After translation, polypeptides and adaptation to micro-environmental changes, are of emerge from the ribosomes and enter the endoplasmic great importance for the development and progression reticulum where chaperonins may remodel the polypep- of normal colonic mucosal cells to adenocarcinomas. tides (Boon et al., 2001). Therefore accelerated protein These six features suggest that adenoma and carcinoma synthesis appears to be a common feature of adenomas cells share several genetic characteristics but have and carcinomas and re¯ects a heavy proliferative burden unique expression pro®les, and this information may in both tumor types. In addition, our data suggest that guide eorts to identify novel targets for blocking activation of oncogenes, aberrant transduction of malignant transformation. signals, deregulation of the cell cycle, impaired growth The results of MDS system using expression pro®les control, and remodeling of cytoskeletal structures are of the 50 genes corroborated its feasibility for predicting general features of tumor cells (Hanahan and Weinberg, the histological features of colorectal tumors. Analysis 2000). Targeting these genes or the molecules they of gene-expression pro®les of very early colon cancers encode could be an important strategy for developing will be necessary for de®ning a more precise cut-o novel approaches to prevention, diagnosis, and treat- value to distinguish between benign and malignant ment of colorectal cancer. lesions; moreover, the scoring system must be validated The commonly down-regulated genes included a using bulk tumors before it can become a tool for number of genes associated with cell death, which clinical diagnosis. Nevertheless, since histological diag- may imply that broad repression of programed cell- nosis of adenomas and carcinomas is sometimes very death pathways is a crucial step for colorectal dicult and may vary among pathologists (Schlemper tumorigenesis. In addition, the list of commonly et al., 2000), the MDS system may ultimately be useful down-regulated genes suggests that reduction of in distinguishing benign from malignant tumors because growth-suppressive signals and/or tumor-suppressive it is an objective quanti®cation of each tumor on the functions may confer continuous proliferative proper- basis of a genome-wide database. ties to neoplastic cells. Therefore, induction of genes In conclusion, the gene-expression analysis of color- categorized in this cluster may be a potential approach ectal adenomas and carcinomas reported here, obtained for development of novel cancer therapies. through a combination of laser-capture dissection and a We also revealed that a number of genes discrimi- genome-wide cDNA microarray, provides a large body nating carcinoma from adenoma were relevant to of information to clarify the mechanisms of colorectal hypoxia. Conceivably, carcinoma cells may be more tumorigenesis and should contribute to identi®cation of exposed to starved and hypoxic conditions, where molecular targets for prevention, diagnosis and treat- carbohydrate/oxygen homeostasis is impaired, than are ment of colorectal tumors. Moreover, the novel MDS adenoma cells. Therefore our data may imply that system that we developed, based on a subset of 50 genes, cancer cells are changing their expression pro®les in may facilitate more sensitive, speci®c and precise response to low-nutrient and hypoxic conditions. Since diagnosis of such tumors. hypoxia is a prognostic indicator in a number of tumors (Denko et al., 2000), targeting the genes in this category may help us to clarify micro-environmental changes during malignant transformation, and to Materials and methods identify novel prognostic predictors for colon cancer. Consideration of the nature of the genes described Tissue samples and laser-capture microdissection (LCM) above leads us to conclude that activation of Eleven dierentiated adenocarcinomas, nine adenomas, and oncogenes, augmentation of proliferation signals, their corresponding normal mucosae of the colon were attenuation of anti-proliferative signals, avoidance of obtained with informed consent from 16 patients who

Oncogene Analysis of expression profiles in colonic tumors Y-M Lin et al 4127 underwent colectomy. In four cases, both adenomas and values were obtained in more than 16 cases (80%) and their carcinomas had arisen in the same patient. All 20-paired standard deviations of observed values were grater than 0.5. samples were embedded in TissueTek OCT medium (Miles, Inc.) and frozen at 7808C. Procedures of ®xation, staining Calculation of `Molecular Diagnosis Score' (MDS) and LCM were performed as described previously (Kitahara et al., 2001). About 10 000 cells were selectively collected by The MDS of each tumor was de®ned as the sum of weighted LCM from each tissue sample. log ratios of expression pro®les of the 50 genes selected as dierentially expressed in adenomas vs carcinomas: MDS =SS log (r ), where r is the expression ratio (Cy3/ RNA extraction, T7-based RNA amplification and cDNA i k 2 ik ik Cy5) of gene k of patient i, and S is the sign for gene k microarray k which was determined as follows: we ®rst calculated the Extraction of total RNA and T7-based RNA ampli®cation average log ratio log2(rik) for gene k in the 11 adenocarci- were carried out as described previously (Kitahara et al., nomas and the nine adenomas (avecarcinoma=Slog2(rik)/ 2001). Two rounds of ampli®cation yielded 15 ± 80 mgof ncarcinoma and aveadenoma=Slog2(rik)/nadenoma). Then, we ampli®ed RNA (aRNA) from each sample. A 2.5 mg aliquot determined the sign for each gene: Sk=+1, if avecarcinoma of aRNA from each tumor and normal epithelium were 4 aveadenoma, and Sk=71, if avecarcinoma 5 aveadenoma labeled with Cy3-dCTP and Cy5-dCTP respectively (Amer- (Figure 4A). sham Pharmacia Biotech). To reduce experimental ¯uctua- tion, we used duplicate sets of cDNA microarray slides Real-time quantitative RT ± PCR containing 23 040 cDNAs for each analysis. Fabrication of the cDNA microarray slides, hybridization, washing, and To verify our microarray data, we selected six genes and detection of signals were carried out as described elsewhere examined their expression levels in 13 additional samples (Ono et al., 2000). The 23 040 genes were selected from (seven adenomas and six carcinomas) by means of real-time UniGene database (National Center for Biotechnology quantitative RT ± PCR (TaqMan PCR, Perkin-Elmer) (Heid Information) and their cDNA fragments were ampli®ed by et al., 1996), using a 7700 Sequence Detector (Perkin-Elmer). RT ± PCR using gene-speci®c primers for each gene and a Each single-stranded cDNA was reverse-transcribed from variety of human poly(A)+ RNAs as template (Clontech). ampli®ed RNA and diluted for subsequent PCR ampli®cation. Malate dehydrogenase 1 (MDH1) served as a relative quantitative control since it showed the smallest Cy3/Cy5 Data analysis ¯uctuations in 100 hybridizations. Each PCR was carried out The intensity of each signal of Cy3 and Cy5 was evaluated in a 25-ml volume and ampli®ed for 10 min at 958C for photometrically using ArrayVision software (Imaging Re- activation of AmpliTaq GoldTM, followed by 40 cycles of search Inc., St. Catharines, Ont. Canada) and normalized 958C for 15 s and 608C for 1 min. The genes and sequences according to the expression of 52 housekeeping genes of the primers and probes used for quantitative RT ± PCR are (Kitahara et al., 2001). After normalization, each gene was listed in Table 1. separated into one of four categories based on the average Cy3/Cy5 ratio (r): up-regulated (r42), down-regulated Statistics (r50.5), unchanged (0.55r52) and low (expression level below cuto level for detection). Excel, Cluster and TreeView Assessment of statistical dierences of gene expression in software packages were used for subsequent analysis. carcinomas vs adenomas was determined by Mann ± Whitney U-tests. A P value 40.05 was considered statistically signi®cant. Statistical analyses were performed using Stat Validation of data View software. To assess the reproducibility of hierarchical clustering, we compared clustering results in the sample axis by using dierent sets of genes (23 040 target sequences were spotted on ®ve slides, and we performed clustering analysis for the 20 samples in all ®ve sets). When one sample consistently fell into the same cluster in dierent sets of genes, we de®ned the data as reproducible. The reproducibility was more than 80% Acknowledgments when Cy3 or Cy5 ¯uorescent units were above 100 000. We appreciate the help of Drs Yoshinao Yamada, Therefore, we ®rst calculated average Cy3- and Cy5- Yoshiyuki Akiyama, Masataka Sakaguchi, Osamu Kita- ¯uorescence intensities of each gene in all 20 cases, excluded hara, and Renpei Yanagawa for helpful discussions, and genes for further analysis when both intensities were below the assistance of Dr Meiko Takahashi in preparing the 16105 units, and selected a total of 2425 genes. A great manuscript. This work was supported in part by Research majority of their spots (more than 99% of spots) had S/N for the Future Program Grant #00L01402 from the Japan ratios greater than 3.0. We further chose 771 genes whose Society for the Promotion of Science.

References

Afshari CA, Nuwaysir EF and Barrett JC. (1999). Cancer Chu FF, Doroshow JH and Esworthy RS. (1993). J. Biol. Res., 59, 4759 ± 4760. Chem., 268, 2571 ± 2576. Boon K, Caron HN, van Asperen R, Valentijn L, Hermus M- Denko N, Schindler C, Koong A, Laderoute K, Green C and C,vanSluisP,RoobeekI,WeisI,VoutePA,SchwabM Giaccia A. (2000). Clin. Cancer Res., 6, 480 ± 487. and Versteeg R. (2001). EMBO J., 20, 1383 ± 1393. Hanahan D and Weinberg RA. (2000). Cell, 100, 57 ± 70.

Oncogene Analysis of expression profiles in colonic tumors Y-M Lin et al 4128 Heid CA, Stevens J, Livak KJ and Williams PM. (1996). Schlemper RJ, Riddell RH, Kato Y, Borchard F, Cooper Genome Res., 6, 986 ± 994. HS, Dawsey SM, Dixon MF, Fenoglio-Preiser CM, Flejou Hoeren FU, Dolferus R, Wu Y, Peacock WJ and Dennis ES. JF, Geboes K, Hattori T, Hirota T, Itabashi M, Iwafuchi (1998). Genetics, 149, 479 ± 490. M, Iwashita A, Kim YI, Kirchner T, Klimp®nger M, Kallioniemi OP. (2001). Ann. Med., 33, 142 ± 147. KoikeM,LauwersGY,LewinKJ,OberhuberG,Oner Kitahara O, Furukawa Y, Tanaka T, Kihara C, Ono K, F, Price AB, Rubio CA, Shimizu M, Shimoda T, Sipponen Yanagawa R, Nita M, Takagi T, Nakamura Y and P, Solcia E, Stolte M, Watanabe H and Yamabe H. (2000). Tsunoda T. (2001). Cancer Res., 61, 3544 ± 3549. Gut, 47, 251 ± 255. Lawrance IC, Fiocchi C and Chakravarti S. (2001). Hum. Semenza GL, Roth PH, Fang HM and Wang GL. (1994). J. Mol. Genet., 10, 445 ± 456. Biol. Chem., 269, 23757 ± 23763. Maguire BA and Zimmermann RA. (2001). Cell, 104, 813 ± Uhlig S and Wendel A. (1992). Life Sci., 51, 1083 ± 1094. 816. Vander Heiden MG, Chandel NS, Li XX, Schumacker PT, Muto T, Bussey HJ and Morson BC. (1975). Cancer, 36, Colombini M and Thompson CB. (2000). Proc. Natl. 2251 ± 2270. Acad.Sci.USA,97, 4666 ± 4671. Notterman DA, Alon U, Sierk AJ and Levine AJ. (2001). Wodopia R, Ko HS, Billian J, Wiesner R, Bartsch P and Cancer Res., 61, 3124 ± 3130. Mairbaurl H. (2000). Am. J. Physiol. Lung Cell Mol. Ogiso Y, Tomida A, Kim HD and Tsuruo T. (1999). Physiol., 279, L1110 ± L1119. Biochem. Biophys. Res. Commun., 258, 448 ± 452. Zhong H and Simons JW. (1999). Biochem. Biophys. Res. Okabe H, Satoh S, Kato T, Kitahara O, Yanagawa R, Commun., 259, 523 ± 526. Yamaoka Y, Tsunoda T, Furukawa Y and Nakamura. (2001). Cancer Res., 61, 2129 ± 2137. Ono K, Tanaka T, Tsunoda T, Kitahara O, Kihara C, Okamoto A, Ochiai K, Takagi T and Nakamura Y. (2000). Cancer Res., 60, 5007 ± 5011.

Oncogene