Global Gene Expression Analysis of Gastric Cancer by Oligonucleotide Microarrays¹

Gastric cancer is one of the leading causes of cancer death in the world (1). Advances in diagnostic and treatment technologies have enabled us to offer excellent long-term survival results for early gastric cancer, but prognosis of advanced gastric cancer still remains poor (2). Recent molecular analyses have clarified many genetic alterations in gastric carcinogenesis, such as p53 (3), β-catenin (4), E-cadherin (5), trefoil factor 1 (6), and c-met (7), but this is hardly sufficient to understand common pathway of carcinogenesis and progression of gastric cancer. Furthermore, gastric cancer shows diverse clinical properties such as histological type, metastatic status, invasiveness, and responsiveness to chemotherapy. Little is known about the genes associated with these characteristics.

Gastric cancer tissues generally contain multiple nonepithelial cell types such as fibroblast, smooth muscle cell, endothelial cell, infiltrating lymphocyte, and macrophage. Because recent advances in cancer research have revealed the relevance of epithelial-stromal interaction including ECMs,³ MMPs, and angiogenic factors in cancer progression (8, 9), we have analyzed whole cancer tissues in this study instead of focusing only on cancer cells to better describe the entire aspect of gastric cancer.

Array technologies are accurate and comprehensive ways of simultaneously analyzing the expression of thousands of genes and have been rapidly applied in many research fields (10). To clarify gene expression changes that are common in cancer tissues or differ among cancer tissues, we have analyzed gastric cancer by oligonucleotide microarray representing approximately 5600 unique genes in this study. We classified both samples and genes by a two-way clustering analysis and identified genes that were differentially expressed between cancer and noncancerous tissues. Furthermore, several genes were identified as being associated with lymph node metastasis or histological types by comparing array data with clinicopathological data.

Twenty-six pairs of advanced gastric cancer tissues (T1–T26) and corresponding adjacent noncancerous gastric tissues (N1–N26) were obtained with informed consent from patients who underwent gastrectomy at Jichi Medical College Hospital (Tochigi, Japan). Depth of invasion was more than muscularis propria for all of the cancer tissues, some of which were diagnosed microscopically to accompany lymph node metastasis. A pathologist (J-M. C.) dissected tissue samples from surgical specimens with special care for minimal contamination of nonepithelial cells, and samples were immediately snap-frozen in liquid nitrogen. Another pathologist (H. T.) determined histological classification according to Lauren’s classification (11) after H&E staining. These clinical and histopathological features are summarized in Table 1.Tissues were homogenized and lysed directly in Isogen reagent (Nippon Gene, Osaka, Japan). Total RNA was extracted according to the manufacturer’s instructions only from tumor specimens that contained >50% cancer cells.

Twenty-two gastric cancers (T1–T22) and 8 noncancerous gastric tissues (3N, 4N, 9N, 12N, 16N, and 22N–24N) were analyzed by oligonucleotide microarray (GeneChip HuGeneFL array; Affymetrix, Santa Clara, CA). This array contains 6936 probes interrogating approximately 5600 full-length human genes from the Unigene (Build 18), GenBank, and The Institute for Genomic Research. Analysis was performed essentially as described previously (12). Briefly, double-stranded cDNA was synthesized from 10 μg of total RNA with oligo(dT)₂₄ T7 primer, amplified with T7 RNA polymerase up to approximately 100 μg of cRNA, and hybridized to the oligonucleotide microarray according to manufacturer’s instructions. For normalization, the average intensity for 6936 genes in total was made equal to 100 to reliably compare variable multiple arrays.

A two-way clustering analysis of 30 samples by Pearson’s correlation was performed using the 6272 genes that passed prefiltering, which eliminated genes with an expression level <10 for all of the samples. To identify genes that were differentially expressed between the two groups, Mann-Whitney’s U test was used with significance set at P < 0.05. Genespring (Silicon Genetics, Redwood City, CA) was used for clustering and statistical analysis. The average expression level of each gene in each group (Ca, cancer tissue; N, noncancerous tissue) was calculated, value below 10 was set to 10, and then the ratio of average expression level between the two groups (Ca:N or N:Ca) was calculated. The cutoff value was set to 80 for average expression level and to 2.5 or 2.0 for the ratio.

Single-stranded cDNA was synthesized with oligo(dT) primer in a 20-μl reaction from 5 μg of total RNA using SuperScript Preamplification System for First Strand cDNA Synthesis System (Life Technologies, Inc., Rockville, MD) and diluted up to 80 μl. PCR was then performed with 1 μl of cDNA for 1 cycle of 94°C for 2 min, followed by 20–30 cycles of 94°C for 30 s, 60°C for 30 s, and 72°C for 3 min using gene-specific primers and Taq polymerase. Amplification of the right target DNA was confirmed by mobility on gel electrophoresis and sequencing after subcloning into pGEM-T easy vector (Promega, Madison, WI). β-Actin was used as an internal control to confirm equal amount of the templates.

Formalin-fixed, paraffin-embedded gastric cancer sections were immunostained with horseradish peroxidase-conjugated secondary antibodies using the DAKO LSAB2/HRP kit (DAKO JAPAN, Kyoto, Japan) following the manufacturer’s instructions. Primary antihuman antibodies were diluted 1:100 for p53 (DO-7; Novocastra Laboratories Ltd., Newcastle, United Kingdom), 1:1000 for β-catenin (BD; Transduction Laboratories, Lexington, KY), 1:500 for E-cadherin (HECD-1; Takara, Tokyo, Japan), and 1:500 for Oct-2 (C-20; Santa Cruz Biotechnology, Inc., Santa Cruz, CA).

Twenty-two advanced gastric cancer tissues and 8 noncancerous tissues were analyzed by oligonucleotide microarray, and expression data for 6936 genes were obtained. With expression data of 6272 expressed genes that passed prefiltering, a two-way clustering analysis according to Pearson’s correlation was performed. Cancer tissues and noncancerous tissues (30 samples in total) were successfully distinguished (Fig. 1). However, this algorithm using most of the genes on the array did not classify samples among cancer tissues by subgroups associated with histopathological features such as those listed in Table 1.

To identify genes that were differentially expressed between cancer and noncancerous tissues, we applied Mann-Whitney’s U test to the raw data obtained by microarray analysis. When cutoff values were set to 2.5 for the ratio and 80 for the average expression level to extract only reliable data, 162 and 129 genes were identified that showed higher expression in cancer tissues and noncancerous tissues, respectively (P < 0.05). To verify the reproducibility of these gene lists, we performed semiquantitative RT-PCR using the same RNA used for microarray analysis. Four randomly selected pairs of cancer and corresponding noncancerous tissue samples were analyzed in 80 randomly selected genes. Concordant results were obtained for four pairs in 33 genes, three pairs in 26 genes, two pairs in 17 genes, and one pair in 4 genes, showing a high concordance rate between the data obtained by RT-PCR and data obtained by microarray. Additionally, these genes were classified in terms of function by referring to the literature and web database⁴ as shown in Table 2. Supplementary tables are available at Cancer Research Online.⁵

Advanced gastric cancer often accompanies lymph node metastasis in the course of progression. Some genes such as IL8, VEGF, OPN, CD44v9, and MMP9 are reportedly related to gastric cancer metastasis and invasion in general (13). However, there are few studies that focus on lymph node metastasis. To explore genes associated with this type of metastasis, we compared 15 cancer samples with metastasis to 7 cancer samples without metastasis and subsequently to 8 noncancerous tissues. Nine genes showed a distinct expression pattern exclusively in cancer tissues with lymph node metastasis (P < 0.05), a >2-fold change as compared with any of the other groups (Fig. 2,A). These genes included matrix remodeling genes, such as FN1 and PCOLCE, and PFN2, which affects cell motility by regulating actin polymerization (14). Among 9 genes identified, association of Oct-2 with metastasis was intriguing, because it has been generally regarded not as a gastric but as a lymphoid or neuronal cell-specific transcription factor (15). To investigate which cells are expressing Oct-2, immunohistochemical analysis was performed. Strong immunoreactivity was observed in gastric cancer cells with lymph node metastasis and in some infiltrating lymphocytes, but not in cancer cells without metastasis (Fig. 3).

Gastric cancer is generally classified into two major histological types according to Lauren’s classification: intestinal type and diffuse type, which roughly correspond to the highly and poorly differentiated type, respectively (16). Many previous works indicate distinct genetic changes and expression pattern of a subset of genes between these two types. Loss of E-cadherin expression and K-sam amplification are predominant in diffuse-type cancer, and mutation or nuclear accumulation of β-catenin and amplification of c-erbB2 are predominant in intestinal-type cancer, whereas mutation or nuclear accumulation of p53 is frequently observed irrespective of histological type (3, 4, 16). As described in Table 1, immunohistochemical analysis of E-cadherin, β-catenin, and p53 in this study is consistent with the findings of these previous works. Moreover, extremely high expression of c-erbB2 in the microarray data, which is suggestive of gene amplification, was observed exclusively in intestinal-type cancer (data not shown). Gastric cancer samples used in the current study are therefore quite adequate for further analysis. To identify novel genes associated with histological types based on transcription analysis, we compared gene expression between the two types. Fifteen genes showed >2-fold differential expression between the two types (P < 0.05; Fig. 2 B). Overexpression of intestinal enzymes GALC (17), GUCY2C (18), and GPX2 (19) and reduced expression of gastric protein MSMB (20) in intestinal-type cancer were identified, reflecting intestinal differentiation in intestinal-type gastric cancer. Additionally, LI-cadherin (CDH17), one of the cadherin family genes, which have crucial roles in cell-cell adhesion, showed preferential expression in intestinal-type cancer. Because expression of LI-cadherin is observed in intestinal cells and hepatocytes (21), but not in gastric epithelium (22), it can also be regarded as one of the intestinal differentiation markers.

In this study, we have globally analyzed gene expression of gastric cancer tissues and noncancerous tissues to elucidate characteristic changes associated with carcinogenesis and progression in gastric cancer. Cancer tissues and noncancerous tissues were distinguished by gene expression profiling alone, indicating that array analysis of whole cancer tissues can efficiently detect characteristics of gastric cancer by integrating alteration of gene expression in cancer cells and stromal cells. When we reviewed genes that were highly expressed in gastric cancer tissues (Table 2),⁵ we could readily extract two major features: (a) high proliferative status of cancer cells; and (b) reactive status of stromal cells. Genes classified in the cell cycle, growth factor-related, DNA synthesis, chromosome, and transcription category were related to high proliferative status of cancer cells and expressed predominantly by cancer cells. Some genes have been previously reported to show high expression in gastric cancer, such as TOP2A, CKS1, CKS2, CDK4, and PCNA (23), FGFR4 (24), IGF2 (25), CDC25B, ERBB2 (3), and GRB7 (26). On the other hand, genes classified in the ECM, ECM remodeling, and angiogenesis category were related to the reactive status of stromal cells and expressed mainly by stromal cells and partly by cancer cells. When we referred to other comprehensive studies on gene expression specific to endothelium and cancer invasion, these genes could be characterized more precisely. Genes expressed predominantly in the endothelium have recently been identified with serial analysis of gene expression (27). Among the genes listed in Table 2,⁵ VWF, SPARC, COL18A1, COL4A2, and GEM were expressed in both normal and tumor endothelium, whereas CST4, THY1, MMP11, COL1A2, COL6A3, COL3A1, and COL1A1 were expressed exclusively in tumor endothelium. Interestingly, the most abundant six of nine collagen genes were of endothelial origin, highlighting a crucial role of angiogenesis in the formation of desmoplasia, a fibrotic change seen frequently in gastric cancer. Genes related to cancer invasion included most of endothelium-expressed genes mentioned above and cancer-expressed genes CALD1 (28), HSPA1A (29), NNMT (30) and LRP1 (23, 31). Besides, high expression of MAGE3 (32), VIL1 (33), and SOD2G (34) in gastric cancer has been reported previously. Many other genes identified here were also associated with various types of cancer. For example, high expression of MSLN (35) and KLK6 (36) in ovarian cancer, GRO1 in malignant melanoma (37), and H19 (38), MK (39), and chaperone genes (40) in many types of cancer have been reported previously.

We identified several genes associated with lymph node metastasis (Fig. 2 A). FN1 and PCOLCE are genes related to matrix remodeling (41). The involvement of FN1 in cell migration and metastasis has been well documented (42, 43). PFN2 affects cell motility by regulating actin polymerization in response to outer signals (14). Growth factor IGF2 also promotes cell motility (44). It is likely that cell motility enhanced by these genes can lead to metastasis. Unexpectedly, Oct-2 was highly expressed by cancer cells with lymph node metastasis. Oct-2 is a POU domain transcription factor that shows a restricted expression pattern in lymphoid cells and neuronal cells and is involved in transcription of immunoglobulin genes in B cells (15, 45). There is only one report of Oct-2 expression by cancer cells (46); however, constitutive expression in vitro of Oct-2 was confirmed by RT-PCR in 7 of 11 gastric cancer cells examined (data not shown), suggesting its frequent ectopic expression by cancer cells. We have previously reported overexpression of MHC class II genes via up-regulation of CIITA, a transactivator of MHC class II genes, in a gastric cancer cell line with high metastatic potential to lymph nodes in a nude mouse model (47). It is extremely intriguing that these lymphoid cell-specific genes are associated with lymph node metastasis. It remains to be investigated whether these genes are functionally relevant to lymph node metastasis of gastric cancer.

We further identified genes associated with histological type of gastric cancer (Fig. 2 B). Because GALC (17), GUCY2C (18), and GPX2 (19) are expressed predominantly in the intestine, overexpression of these genes can be regarded as intestinal differentiation of cancer cells. On the other hand, MSMB is predominantly expressed in gastric antrum (20), and its selective down-regulation can be viewed as dedifferentiation from the gastric phenotype. Consistent with the current study, HSPA1B (29) and CDKN2A (48) have been reported to show differential expression between the two types. Moreover, LI-cadherin showed high expression in intestinal-type gastric cancer, which is in line with a recent immunohistochemistry study (49). Because LI-cadherin could already be detected in intestinal metaplasia, a cancer-predisposed lesion for intestinal-type gastric cancer (50), the transcriptional regulator of LI-cadherin may have crucial roles in the multistep carcinogenesis of intestinal-type gastric cancer.

Advanced gastric cancer is generally refractory to chemotherapy by anticancer drugs, which leads to poor prognosis. Accordingly, targets of gastric cancer therapeutics have been recently extended from molecules of cancer origin to molecules of stroma origin, such as those related to angiogenesis and matrix remodeling (51, 52). Because our study was based on whole tissue samples, the list of genes up-regulated in cancer tissues contained and may still contain many genes for therapeutic target molecules of stroma. Precise prediction of metastases in neighboring lymph nodes remains very difficult but can provide evidence for selecting optimal therapy between surgical and endoscopic resection or optimal extent of lymph node dissection in case of surgery. If examination of the metastasis-associated genes identified in this study were applicable in the future to predict lymph node metastasis from biopsy samples, it would undoubtedly be of great clinical value. In conclusion, the genes described in the current study should therefore provide valuable resources not only for basic studies, such as understanding molecular mechanism of carcinogenesis, progression, and metastasis, but also for clinical applications, such as development of novel diagnostic markers and identification of therapeutic targets in gastric cancer.

We thank Akima Harada and Kaori Shiina for immunohistochemical analysis, Hiroko Meguro for GeneChip analysis, and Saori Fukui and Erio Fujita for excellent technical support.