oncogenomics - Nature

Oncogene (2002) 21, 6549 – 6556 ª 2002 Nature Publishing Group All rights reserved 0950 – 9232/02 $25.00 www.nature.com/onc

SHORT REPORT

Comprehensive analysis of the gene expression profiles in human gastric cancer cell lines Jiafu Ji1,9, Xin Chen2,3,9, Suet Yi Leung4,9, Jen-Tsan A Chi5, Kent Man Chu6, Siu Tsan Yuen4, Rui Li2,3, Annie SY Chan4, Jiyou Li7, Nina Dunphy8 and Samuel So*,2,3 Department of Surgery, Beijing Cancer Hospital, Peking University School of Oncology, Beijing, China; 2Department of Surgery, Stanford University, Stanford, California, USA; 3Asian Liver Center, Stanford University, Stanford, California, USA; 4 Department of Pathology, The University of Hong Kong, Queen Mary Hospital, Hong Kong, China; 5Department of Biochemistry, Stanford University, Stanford, California, USA; 6Department of Surgery, The University of Hong Kong, Queen Mary Hospital, Hong Kong, China; 7Pathology Department, Beijing Cancer Hospital, Peking University School of Oncology, Beijing, China; 8Department of Genetics, Stanford University, Stanford, California, USA

Gastric adenocarcinoma is one of the major malignancies worldwide. Gastric cell lines have been widely used as the model to study the genetics, pharmacology and biochemistry of gastric cancers. Here we describe a comprehensive survey of the gene expression profiles of 12 gastric carcinoma cell lines, using cDNA microarray with 43 000 clones. For comparison, we also explored the gene expression patterns of 15 cell lines derived from lymphoid, endothelial, stromal and other epithelial cancers. Expression levels of specific genes were validated through comparison to protein expression by immunohistochemistry using cell block arrays. We found sets of genes whose expression corresponds to the molecular signature of each cell type. In the gastric cancer cell lines, apart from genes that are highly expressed corresponding to their common epithelial origin from the gastrointestinal tract, we found marked heterogeneity among the gene expression patterns of these cell lines. Some of the heterogeneity may reflect their underlying molecular characteristics or specific differentiation program. Two putative gastric carcinoma cell lines were found to be B-cell lymphoma, and another one had no epithelial specific gene expression and hence was of doubtful epithelial origin. These cell lines should no longer be used in gastric carcinoma research. In conclusion, our gene expression database can serve as a powerful resource for the study of gastric cancer using these cell lines. Oncogene (2002) 21, 6549 – 6556. doi:10.1038/sj.onc. 1205829 Keywords: gene expression profile; microarray; gastric cancer

*Correspondence: S So, Department of Surgery, 300 Pasteur Drive, Room H3680, Stanford University Medical Center, Stanford, CA 94305, USA; E-mail: [email protected] 9 These authors contributed equally to this work. Received 3 April 2002; revised 13 June 2002; accepted 28 June 2002

Gastric cancer is the second most common cancer worldwide, accounting for almost 10% of new cancer cases (Parkin et al., 1999). It is also among the leading causes of death from cancer throughout the world (Pisani et al., 1999). Gastric adenocarcinoma constitutes approximately 90% of all gastric cancers. Helicobacter pylori infection has been clearly linked to the development of gastric adenocarcinoma (Ebert et al., 2000). However, the molecular mechanism of this association remains unclear. Pathologically, there are two types of gastric cancer: intestinal type and diffuse type. The intestinal type cancer commonly arises in a background of chronic atrophic gastritis and intestinal metaplasia. As with other types of malignancies, the prognosis of gastric cancer patients depends heavily on the clinical and pathological stage of the disease at diagnosis. Patients with small and early cancer lesions who undergo surgical resection have a better chance of survival. However, most patients are diagnosed with advanced stage disease and the five-year survival rate is generally less than 10% (Peddanna et al., 1995). Recent evidence suggests that the phenotypic diversity of tumors is associated with corresponding diversity in their gene expression programs. cDNA microarray technology has been applied to study the gene expression patterns in different tumor types, providing new insight into the development and classification of these cancers (Alizadeh et al., 2000; Golub et al., 1999; Perou et al., 2000; Ramaswamy et al., 2001; Welsh et al., 2001). Cell lines have been extensively used as experimental models to study the genetics, pharmacology and biochemistry of cancer cells, as well as cellular response to different stimulators. Ross et al. (2000) systematically studied the variation in gene expression programs in 60 human cancer cell lines (NCI60). NCI60 cell lines were derived from nine broad categories of tissue of origin. They have been used in the large-scale drug screening by NCI (Monks et al., 1997). It was shown that each cell line expresses genes characteristic of its cellular origin. Specific features of these gene expression patterns seemed to be related to

ONCOGENOMICS

1

Gene expression of gastric cell lines J Ji et al

6550

Figure 1 (a) Hierarchical clustering of the patterns of variation in the expression of 6849 cDNA clones in 27 cell lines. The data are shown in a table format, in which rows represent individual genes and columns represent individual cell lines. The color in each cell reflects the expression level of the corresponding gene in the corresponding cell line, relative to its mean expression level across the entire set of cell lines. The scale (lower right corner) extends from fluorescence ratios of 0.25 to 4 relative to the mean level for all samples. Grey indicates missing or excluded data. (b) to (f) Features of the variation in the gene expression patterns that can be related to specific physiological or histological features of the cell lines. (b) Epithelial cell cluster; (c) B lymphocytes cluster; (d) Oncogene


6551

the physiological properties of the cell lines, including cell proliferation rate and drug metabolism. The gene expression profiles of these cell lines also helped to differentiate specific cell types in normal and tumor tissues. Furthermore, correlating the gene expression profile in each cell line with the responsiveness of each cell line to drugs allowed for the identification of genes that may be important for drug sensitivities (Scherf et al., 2000; Staunton et al., 2001). However, gastric cancer cell lines were not included in the NCI60 studies. Still little is known about the gene expression patterns of gastric cell lines on the genomic scale. Global gene expression patterns in 27 human cell lines In this study, we used cDNA microarray with 43 000 cDNA clones, representing approximately 30 000 unique genes to study the gene expression patterns of 12 gastric carcinoma cell lines. As a control to identify the gene expression signatures of the various cell types, we included 15 cell lines of other tissue origin, including T-cell, B-cell, monocyte, myelocyte, fibroblast, endothelium, colon, breast and pancreas. We used a hierarchical clustering algorithm to group genes as well as the cell lines, on the basis of similarity in their expression patterns. A total of 6849 cDNA clones were shown to have significant variation (genes with at least fourfold of expression difference from the mean in one array and 60% valid data points) among 27 cell lines (Figure 1a). The most notable feature of the clustered data was that cell lines clustered into two major branches. Cell lines derived from gastrointestinal epithelial cells, including gastric, colon and pancreas, clustered into one branch, whereas the rest of the cell lines clustered into a second branch. The second branch again consisted of three sub-branches: endothelial cells and fibroblast cluster together; the cell lines derived from

leukocytes cluster together; and SNU1 and MCF7 formed the third cluster. As shown previously, these gene expression patterns were clearly related to the histological origins of the cell lines (Ross et al., 2000). Each cell type expressed the genes characteristic of its cellular origin (Figure 1b – f). For example, B lymphocyte cell lines expressed CD20, BCL2, Immunoglobin heavy and light chain, and HLA class II molecules (Figure 1c); T lymphocyte cell lines expressed CD3, CD6 and CD28 (Figure 1d); endothelial cell lines expressed Von Willebrand factor, CD31, VE-cadherin (Figure 1f); and fibroblast cell lines expressed different members of collagen family and other genes encoding cell matrix proteins (Figure 1e). Cells originated from epithelial lineage, including gastric carcinoma cell lines, clustered together and expressed a set of genes, many of which have been implicated in epithelial cell biology (Figure 1b). This cluster included genes whose products encode structural proteins, for example, cell – cell adherence complex (desmoplakin, Claudin 3 and Claudin 4), cell – matrix complex (integrin beta 4) and epithelial intermediate filament (Keratin 8 and Keratin 18). The patterns of gene expression measured in these 27 cell lines provide us with a framework for distinguishing different cell types within the histologically complex gastric tissues. It is notable that within each tissue specific gene cluster, many genes encoding signaling molecules, transcriptional regulators, and EST clones were identified (see the GeneExplore file of web supplement (http://genome-www.stanford.edu/gc_cells/ explore.shtml) for the complete data set including EST Unigen cluster ID and Accession Numbers). Of particular interest in cancer cell biology are the epithelial and endothelial gene clusters. In the epithelial cell cluster, besides those structural proteins known to be important for epithelial cell biology, the cluster also included cell surface receptors (ErbB3, MST1 receptor, DDR1), transcriptional factors (E74-like factor),

T lymphocytes cluster; (e) fibroblast cell cluster; (f) endothelial cell cluster. Due to limited space, only a few selected gene names are shown. See Supplementary Information for full data. Materials and methods: A total of 27 cell lines were used in this study. These included: twelve gastric cancer cell lines (AGS, KATO3, SNU1, SNU5, SNU16, RF1, RF48, N87, NUGC3, MKN45, BGC823, PAM82), two T cell lines (Jurkat, MOLT4), two B cell lines (LAM, HFI-1), one acute promyelocytic leukemia-like cell line (NB4+RA), one monocyte-like cell line (U937+PMA), one pancreas cancer cell line (BxPC3), four colon cancer cell lines (colo205, HCT116, SW620, HCT15), one breast cancer cell line (MCF7), one primary fibroblast and two primary endothelial cell lines. The detailed information for these cell lines is available through the web supplement. All cell lines were cultured to 80% confluence, harvested and frozen in 7808C until ready to be isolated. mRNA was extracted directly from the frozen cell pellet using FastTrack (Invitrogen) mRNA isolation kit. For the microarray production, 43 000 cDNA clones, representing about 30 000 unique genes, were mechanically printed onto treated glass microscope slides, as previously described (http://cmgm.stanford.edu/pbrown/array.html) (Perou et al., 2000). For RNA labeling, a common reference, which consisted of mixture of eleven cell lines was used (Perou et al., 2000). The hybridization procedures were performed as previously described (Alizadeh et al., 2000). A detailed protocol is available at: http://cmgm.stanford.edu/pbrown/protocols/5_hyb_human.html. Primary data collection and analysis were carried out using GenePix Pro 3.0 (Axon Instruments). Areas of the array with obvious blemishes were manually flagged and excluded from subsequent analysis. The raw data were deposited into Stanford Microarray Database (Sherlock et al., 2001) at: http://genomewww4.stanford.edu/MicroArray/SMD/index.html. For the generation of the cluster, all non-flagged array elements for which the fluorescent intensity in either channel was greater than 2.5 times the local background were considered well measured. Genes for which fewer than 60% of measurements across all the samples in this study met this standard were excluded from further analysis. We chose to further analyse genes whose expression level differed by at least fourfold, in at least one sample, from their mean expression level across all samples. We applied a hierarchical clustering algorithm both to the genes and arrays using the Pearson correlation coefficient as the measure of similarity, and average linkage clustering, as described (Eisen et al., 1998). The results were visualized and analysed with TreeView (M Eisen; http://rana.lbl.gov) Oncogene


6552

signaling molecules (SH3BP1, ARHEGF5, ARHEGF16), and many ESTs of unknown function. In the endothelial specific gene cluster, many angiogenic factors and receptors which were recently implicated in neovascularization were identified: for example, placental growth factor (PGF), Angiopoietin 2 (ANGPT2), fms-related tyrosine kinase 1 (FLT1), and neuropilins (NRP1 and NRP2). Moreover, many signaling molecules (RGS5, RGS4), cell cycle regulator (CDKN1B), as well as ESTs of unknown function were also seen. Further investigation into the function of these genes is clearly needed. Our gene expression data has provided preliminary information about these genes and further study of each of these genes may reveal novel biological function in epithelial cell growth, angiogenesis, extracellular matrix formation, and host immune response. Also, examination of the gene repertoire in the tumor cells, fibroblasts and endothelial cells may reveal potential autocrine and paracrine interactions between these cell types in vivo. Several interesting features regarding gastric cell lines emerged from the hierarchical clustering analysis. Two of the gastric carcinoma cell lines (RF1 and RF48) were found to co-cluster with the B lymphoma cell lines, LAM and HFI-1 (Figure 1), and showed dissimilar gene expression patterns compared with other gastric cell lines (Figures 1 and 3). They

expressed genes that were characteristic of the B-cell lineage, including CD20, BCL2, immunoglobin and HLA class II molecules (Figure 1c). A closer look at the cell morphology also revealed that the RF1 and RF48 were relatively small cells with vesicular nuclei, multiple peripheral attached nucleoli, consistent with the cytological features of immature B-cells. This raised the question that RF1 and RF48 were possibly misidentified. Both cell lines were derived from the same patient, RF1 from the primary gastric adenocarcinoma and RF48 from the metastases in ascitic fluid. To further investigate the cellular origin of these two cell lines, immunohistochemical staining was performed on these two cell lines together with another gastric carcinoma cell line, SNU5, with Leukocyte Common Antigen (LCA, CD45, Dako), B-cell marker CD20 (Dako) and pan-epithelial marker Cam5.2 (Becton Dickinson) (Figure 2). Both RF1 and RF48 stained positive for LCA and CD20, but negative for Cam5.2, whereas SNU5 showed strong positive staining for Cam5.2 and negative staining for LCA and CD20. Together with the microarray data, it is conclusive that RF1 and RF48 were in fact B-cell lymphoma. SNU1 appeared to be very different from all other epithelial derived cell lines. It loosely clustered with MCF7, which is a breast carcinoma cell line (Figure 1).

Figure 2 Immunohistochemical staining shows that RF1 and RF48 are derived from B cell lineage. From left to right are RF1, RF48 and SNU5. Each cell line was stained with the following antibodies: (from top to bottom) LCA (CD45, Dako), CD20 (L26, Dako) and Cam5.2 (Becton-Dickinson). Immunohistochemical staining was performed using the standard streptavidin-biotin peroxidase method with heat-mediated antigen retrieval Oncogene


6553

We found that it lacked the expression of the set of genes characteristic of epithelial cells (Figure 1b). It expressed a set of unique genes (Figure 3a), however, most of these genes lacked tissue specificity. To further characterize this cell line, we performed immunohistochemical staining. We found that it stained negative for all the pan-epithelial markers available, including Cam5.2 (Becton Dickinson), AE1/AE3 (Dako), MNF116 (Dako), BerEP4 (Dako), EMA (Dako), and carcinoembryonic antigen (mCEA, Zymed). SNU1 also stained negative for gastrointestinal stromal tumor markers including desmin (Dako), smooth muscle actin (Dako), c-kit (Dako) and CD34 (Becton Dickinson), connective tissue marker vimentin (Dako), and melanoma marker HMB45 (Dako). It stained positive for neuronal specific enolase (Zymed) and S100 (Dako), raising the possibility of neuroendocrine origin. However, ultrastructural examination revealed very primitive tumor cells with no specific differentiation features. In particular, there was no neurosecretory granule so the possibility of a neuroendocrine tumor cannot be confirmed (data not shown). Together with our microarray data, it suggests that SNU1 may not be derived from gastric epithelial cells. However, we cannot be conclusive about its cellular origin and can only consider it to be a poorly differentiated tumor cell line.

reports of genomic deletion of p53 gene in KATO3 (Yokozaki, 2000). Amplification of FGFR2 (K-sam) is known in KATO3 (Yokozaki, 2000), and we have found a correspondingly high level of FGFR2 expression in KATO3 and SNU16 (Figure 3c). It is known that amplification of the ErbB2 gene, which is associated with over-expression of the mRNA and protein, is present in a subset of gastric adenocarcinoma (Tokunaga et al., 1995). N87 showed a very high expression level of ErbB2 compared with all other gastric cancer cell lines. Interestingly, a group of genes mapped to the ErbB2 locus at chromosome 17q11-21, including PPARBP, CrkRS, MLN64, MLLT6 and MCG953, was also highly expressed in N87 cell (Figure 3d). This supports the idea that there is amplification at 17q11-21 and this amplification results in the high expression of ErbB2 oncogene in N87 cell. Examination of the gene expression of individual gastric cell lines revealed marked heterogeneity with sets of genes being up-regulated or down-regulated in each cell line (Figure 3a). Further study will be needed to investigate if these represent results of chromosomal aberrations, alteration in transcriptional regulators or other underlying molecular events; and the biological significance of these unique gene expression patterns. Gene expression patterns correlating with immunohistochemical staining of cell block arrays

Heterogeneity among the gene expression patterns of gastric cancer cell lines To further analyse the gene expression patterns of gastric cancer cell lines, we clustered 12 gastric cell lines using 3499 cDNA clones which had significant variation among these 12 cell lines (Figure 3). Apart from RF1, RF48 and SNU1, which are probably nonepithelial in origin, the remaining nine gastric cancer cell lines were divided into two main groups. SNU5, SNU16, KATO3, MKN45 and AGS clustered into one branch, whereas NUGC3, N87, PAM82 and BGC823 clustered into a second branch. The major genes that distinguish the two groups included Villin 1, LGALS4 and LI-cadherin (Figure 3e). All of these genes are known to be expressed in intestinal epithelial cells. Moreover, Villin1 and LI-cadherin has been shown to be expressed in gastric mucosa with intestinal metaplasia (Grotzinger et al., 2001; Osborn et al., 1988). It is most likely that the cell lines from the first branch are derived from tumors that progress from intestinal metaplasia. As these two groups of cell lines adopt a different differentiation program, further investigation is needed to clarify whether they may possess distinct biological and genetic properties, or different responsiveness to chemotherapeutic agents. Some oncogenes and tumor suppressor genes are known to be altered by various mechanisms in gastric cancers. They are also reflected in the gene expression data. The expression of p53 protein was dramatically downregulated in KATO3 compared with other gastric cell lines (Figure 3b). This result is consistent with

DNA microarray analysis showed heterogeneity among the gene expression patterns within human gastric cell lines. However, microarray data only reviews the expression level of mRNA. Because proteins are the major components that carry out most of the cellular functions, it is important to study the corresponding expression level at the protein level. To facilitate the characterization of large number of gene products identified from DNA microarray studies in an efficient and economic way, we constructed a cell block array containing all twelve gastric cancer cell lines (including RF1, RF48 and SNU1) and performed immunohistochemical staining on these arrays. Several antiserums, including Villin-1, ErbB2, E-cadherin (CDH1), Keratin 7, Keratin 20 and Muc5AC, were used in the cell block array staining (Table 1). In general, we noticed a close association of mRNA expression and protein level. This observation supported the reliability of our gene expression data obtained from cDNA microarray experiments. Two of the examples are shown in Figure 4. ErbB2 is highly expressed in N87 cell by our microarray analysis and we found that N87 is the only cell line that shows a positive staining of ErbB2 (Figure 4a). Ecadherin is one of the major molecules involved in cell – cell adhesion. Loss of expression of E-cadherin has been linked to the development of tumor (Nollet et al., 1999), and may result in the diffuse type of gastric cancer (Debruyne et al., 1999). Microarray analysis revealed that E-cadherin (CDH1) is expressed at very low levels in PAM82, BGC823, SNU5 and AGS cells. Oncogene


6554

Figure 3 (a) Hierarchical clustering of the patterns of variation in the expression of 3499 cDNA clones in 12 gastric cell lines based on similarity in gene expression patterns. The scale is the same as in Figure 1. To the right, lines and cell line names labeled in red represent the corresponding gene cluster that is highly expressed in the cell line. Alternatively, lines and cell line names labeled in green represent the corresponding gene cluster that is expressed in the cell line at low level. (b) to (e) Features of the variation in the gene expression patterns in these gastric cell lines. (b) Expression of TP53 genes; (c) Expression of FGFR2; (d) ErbB2 cluster; (e) Villin 1 cluster. See Supplementary Information for full data. Materials and methods: All methods in this figure, including data selection and clustering, are the same as for Figure 1 Oncogene


6555 Table 1 Summary of the correlation of the gene expression data from cDNA microarray with immunohistochemical staining SNU5 SNU16 KATO3 MKN45 AGS NUGC3 Villin

Microarray 1.73 3.22 IHC 2.00 3.00 ErbB2 Microarray 70.10 0.31 IHC 0.00 0.00 Ecad Microarray 71.00 2.05 IHC 0.00 2.00 KRT7 Microarray 0.24 1.02 IHC 0.00 2.00 KRT20 Microarray 5.31 70.51 IHC 3.00 1.00 Muc5AC Microarray 70.99 70.52 IHC 0.00 2.00

1.36 3.00 0.47 0.00 1.66 1.00 2.67 3.00 70.19 1.00 1.05 1.00

N87

PAM82 BGC823 SNU1

2.17 1.51 0.25 71.26 71.87 3.00 3.00 0.00 1.00 0.00 0.58 0.47 70.51 5.06 70.61 0.00 0.00 0.00 3.00 0.00 2.88 71.20 1.94 2.48 71.89 2.00 0.00 1.00 2.00 0.00 0.94 70.75 3.26 1.67 2.00 0.00 3.00 3.00 2.00 72.55 71.16 2.95 1.00 1.00 0.00 3.00 0.00 1.60 1.57 2.00 0.00 0.00 0.00 2.00

71.57 0.00 71.10 0.00 71.27 0.00 1.84 2.00 71.01 0.00 0.30 1.00

71.89 0.00 70.60 0.00 71.96 0.00 74.18 0.00

RF1

71.37 0.00 72.21 0.00 72.34 0.00 73.37 0.00 71.43 0.00 0.00 70.65 70.96 0.00 0.00

RF48 71.67 0.00 72.16 0.00 72.46 0.00 73.34 0.00 71.40 0.00 71.41 0.00

Pearson correlation coefficient 0.90** 0.86** 0.94** 0.88** 0.87** 0.77*

The following antibodies were used in the staining: Villin (Immunotech); E-cadherin (Zymed); ErbB2 (Dako); Keratin 7 (Dako); Keratin 20 (Dako); Muc 5AC (Neomarkers). Immunohistochemical staining was performed using the standard strepavidin – biotin peroxidase method with heat-mediated antogen retrieval. The percentage of positive cells and intensity of staining were taken into account and graded in a scale of 0 to 3 (0, negative; 1, occasional positive cells or weak staining intensity; 2, moderate number of positive cells or moderate intensity of staining; 3, most cells positive with intense staining) without knowledge of the microarray data, the mean expression levels were taken if there were more than one cDNA clone per gene in the array. The Pearson correlation coefficient of the mRNA level and protein expression was computed for each gene (**Correlation is significant at 0.01 level; *Correlation is significant at 0.05 level)

Figure 4 Correlating the gene expression data from cDNA microarray with immunohistochemical staining using cell block array. The upper panel shows the relative mRNA expression level measured by cDNA microarray. The scale of the color is the same as in Figure 1. The lower panel shows the immunohistochemical staining. (a) ErbB2 (Dako); (b) E-cadherin (Zymed). N87 shows deep brown membrane staining for ErbB2 whilst all other cell lines are negative. SNU16, KATO3, MKN45, NUGC3 and N87 expressed E-cadherin as deep brown membrane or granular cytoplasmic staining. All other cell lines are negative for E-cadherin protein. Enlarged figures for immunohistochemical staining of each cell line are available through the web supplement. Materials and methods: Twelve gastric cell lines were cultured to 80% confluence, washed in PBS and fixed in 4% paraformaldehyde. Paraffin cell blocks were prepared and a tissue microarray block containing all the cell lines was constructed. Immunohistochemical staining was performed using the standard streptavidin-biotin peroxidase method with heat-mediated antigen retrieval

Cell block array immunohistochemistry showed a corresponding negative staining of E-cadherin protein in all these cell lines (Figure 4b). Completion of the human genome project has provided the basic structure of all human genes. It also provides us an exciting opportunity to study gene expression and function on the genomic scale. In this study, by profiling the expression of 27 cell lines, we were able to identify sets of tissue-specific genes that are of potential importance in carcinoma cell growth, angiogenesis, extracellular matrix formation and

immune response. We also found a marked heterogeneity of gene expression among the gastric cancer cell lines, which may reflect their underlying differences in histiogenetic origin, differentiation program or molecular pathway of tumor evolution. By their gene expression signature, we were able to redefine the histiogenetic origin of two gastric cell lines (RF1 and RF48) as B-cell lymphomas. Although the histiogenetic origin of SNU1 remains elusive, our data has raised sufficient doubt about its epithelial origin. Because these three cell lines are available from American Oncogene


6556

Typed Cell Culture (ATCC), they have been widely used as models to study the genetics and biology of gastric carcinoma. We believe that it is not appropriate to continue to use them in future studies. In this study, we found that each gastric cell line has its own characteristic gene expression program (Figure 3a). Some of them may be the result of chromosomal abnormalities. For example, the high expression of the genes in the ErbB2 locus suggests the amplification of the 17q11-21 region in N87 cell. It would be extremely interesting to further study the DNA copy number variation in these cell lines by array based comprehensive genomic hybridization (aCGH) (Pollack et al., 1999). Correlating aCGH data with our gene expression results will certainly provide new candidates for oncogenes or tumor suppressor genes, as well as new insight into the molecular genetics of gastric cancer. Moreover, examination of the drug sensitivity of these gastric cell lines and correlation with the gene expression profile may lead to identification of genes responsible for drug resistance in gastric cancer. Our study has provided the first comprehensive view of the gene expression patterns in gastric cell lines on a

genomic scale. It serves as a powerful resource for further study of gastric cancer using these cell lines as models.

Note added in proof Supplementary Information is available through the author’s web supplement site at: http://genome_www.stanford.edu/GCcells.

Acknowledgments We are grateful for the members of the Patrick Brown Laboratory at the Department of Biochemistry, Stanford University. We are especially thankful for the advice of this project provided by Drs Patrick Brown and David Botstein. We also thank the Stanford Functional Genomic Center, Stanford Microarray database, and Stanford Asian Liver Center for their support, and Wijan Prapong for his help in the preparation of this manuscript. This work is supported by the HM Lui Foundation (X Chen, R Li, and S So); Research Grants Council of the Hong Kong Special Administrative Region (HKU 7264/01M); and China ‘973’ grant (G1998051203).

References Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, Boldrick JC, Sabet H, Tran T, Yu X, Powell JI, Yang L, Marti GE, Moore T, Hudson J, Lu L, Lewis DB, Tibshirani R, Sherlock G, Chan WC, Greiner TC, Weisenburger DD, Armitage JO, Warnke R and Staudt LM et al. (2000). Nature, 403, 503 – 511. Debruyne P, Vermeulen S and Mareel M. (1999). Acta. Gastroenterol. Belg., 62, 393 – 402. Ebert MP, Yu J, Sung JJ and Malfertheiner P. (2000). Eur. J. Gastroenterol. Hepatol., 12, 795 – 798. Eisen MB, Spellman PT, Brown PO and Botstein D. (1998). Proc. Natl. Acad. Sci. USA, 95, 14863 – 14868. Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA, Bloomfield CD and Lander ES. (1999). Science, 286, 531 – 537. Grotzinger C, Kneifel J, Patschan D, Schnoy N, Anagnostopoulos I, Faiss S, Tauber R, Wiedenmann B and Gessner R. (2001). Gut, 49, 73 – 81. Monks A, Scudiero DA, Johnson GS, Paull KD and Sausville EA. (1997). Anticancer Drug Des., 12, 533 – 541. Nollet F, Berx G and van Roy F. (1999). Mol. Cell Biol. Res. Commun., 2, 77 – 85. Osborn M, Mazzoleni G, Santini D, Marrano D, Martinelli G and Weber K. (1988). Virchows Arch. A. Pathol. Anat. Histopathol., 413, 303 – 312. Parkin DM, Pisani P and Ferlay J. (1999). Int. J. Cancer, 80, 827 – 841. Peddanna N, Holt S and Verma RS. (1995). Anticancer Res., 15, 2055 – 2064. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA, Fluge O, Pergamenschikov A, Williams C, Zhu SX, Lonning PE, Borresen-Dale AL, Brown PO and Botstein D. (2000). Nature, 406, 747 – 752.

Oncogene

Pisani P, Parkin DM, Bray F and Ferlay J. (1999). Int. J. Cancer, 83, 18 – 29. Pollack JR, Perou CM, Alizadeh AA, Eisen MB, Pergamenschikov A, Williams CF, Jeffrey SS, Botstein D and Brown PO. (1999). Nat. Genet., 23, 41 – 46. Ramaswamy S, Tamayo P, Rifkin R, Mukherjee S, Yeang CH, Angelo M, Ladd C, Reich M, Latulippe E, Mesirov JP, Poggio T, Gerald W, Loda M, Lander ES and Golub TR. (2001). Proc. Natl. Acad. Sci. USA, 98, 15149 – 15154. Ross DT, Scherf U, Eisen MB, Perou CM, Rees C, Spellman P, Iyer V, Jeffrey SS, Van de Rijn M, Waltham M, Pergamenschikov A, Lee JC, Lashkari D, Shalon D, Myers TG, Weinstein JN, Botstein D and Brown PO. (2000). Nat. Genet., 24, 227 – 235. Scherf U, Ross DT, Waltham M, Smith LH, Lee JK, Tanabe L, Kohn KW, Reinhold WC, Myers TG, Andrews DT, Scudiero DA, Eisen MB, Sausville EA, Pommier Y, Botstein D, Brown PO and Weinstein JN. (2000). Nat. Genet., 24, 236 – 244. Sherlock G, Hernandez-Boussard T, Kasarskis A, Binkley G, Matese JC, Dwight SS, Kaloper M, Weng S, Jin H, Ball CA, Eisen MB, Spellman PT, Brown PO, Botstein D and Cherry JM. (2001). Nucleic Acids Res., 29, 152 – 155. Staunton JE, Slonim DK, Coller HA, Tamayo P, Angelo MJ, Park J, Scherf U, Lee JK, Reinhold WO, Weinstein JN, Mesirov JP, Lander ES and Golub TR. (2001). Proc. Natl. Acad. Sci. USA, 98, 10787 – 10792. Tokunaga A, Onda M, Okuda T, Teramoto T, Fujita I, Mizutani T, Kiyama T, Yoshiyuki T, Nishi K and Matsukura N. (1995). Cancer, 75, 1418 – 1425. Welsh JB, Zarrinkar PP, Sapinoso LM, Kern SG, Behling CA, Monk BJ, Lockhart DJ, Burger RA and Hampton GM. (2001). Proc. Natl. Acad. Sci. USA, 98, 1176 – 1181. Yokozaki H. (2000). Pathol. Int., 50, 767 – 777.