Comparison of gene-expression profiles between diffuse- and intestinal-type gastric cancers using a genome-wide cDNA microarray

Abstract

Gastric cancer is the fourth leading cause of cancer-related death in the world. Two histologically distinct types of gastric carcinoma, ‘intestinal’ and ‘diffuse’, have different epidemiological and pathophysiological features that suggest different mechanisms of carcinogenesis. A number of studies have investigated intestinal-type gastric cancers at the molecular level, but little is known about mechanisms involved in the diffuse type, which has a more invasive phenotype and poorer prognosis. To clarify the mechanisms that underlie its development and/or progression, we compared the expression profiles of 20 laser-microbeam-microdissected diffuse-type gastric-cancer tissues with corresponding noncancerous mucosae by means of a cDNA microarray containing 23 040 genes. We identified 153 genes that were commonly upregulated and more than 1500 that were commonly downregulated in the tumors. We also identified a number of genes related to tumor progression. Furthermore, comparison of the expression profiles of diffuse-type with those of intestinal-type gastric cancers identified 46 genes that may represent distinct molecular signatures of each histological type. The putative signature of diffuse-type cancer exhibited altered expression of genes related to cell–matrix interaction and extracellular-matrix (ECM) components, whereas that of intestinal-type cancer represented enhancement of cell growth. These data provide insight into different mechanisms underlying gastric carcinogenesis and may also serve as a starting point for identifying novel diagnostic markers and/or therapeutic targets for diffuse-type gastric cancers.

Introduction

Gastric cancer is the fourth most frequent cancer worldwide, accounting for 10.4% of cancer deaths in 2000; Japan has the highest age-standardized incidence of this disease (Parkin, 2001). Since the 5-year survival rate of patients who are diagnosed at an advanced stage is generally less than 10% (Peddanna et al., 1995), identification of sensitive diagnostic markers and development of novel therapeutic modalities other than surgery are of some urgency.

Gastric cancers are histologically classified into diffuse type (infiltrating, poorly differentiated, noncohesive cancer cells with vast fibrous stroma) and intestinal type (cohesive, glandular-like cell groups) (Lauren, 1965). These two types have different epidemiology, etiology, pathogenesis and biological behavior. The diffuse type occurs in relatively younger individuals regardless of gender, and often metastasizes to peritoneum or lymph nodes with a poorer prognosis as a result. Helicobacter pylori, which tends to be associated with gastric cancer (Sipponen, 1995) might not have a role in diffuse-type gastric carcinogenesis because prolonged infection leads to intestinal metaplasia, a precursor lesion of intestinal-type gastric cancer. The precursor lesion of diffuse-type gastric cancer is proposed to arise from hyperplastic neck cells, but that has not yet been proven (Ming, 1998).

Molecular investigations have identified multiple genetic alterations in gastric carcinomas including point mutations, deletions, loss of heterozygosity and microsatellite instability. Notably, mutations in APC, CTNNB1 (β-catenin) and/or p53 genes (Craanen et al., 1995; Park et al., 1999) are more frequently associated with the intestinal type, whereas aberrant expression of CDH1 (encoding E-cadherin) is often a feature of diffuse-type gastric cancers; somatic mutations of CDH1, or methylation in its promoter region, occur in about 50% of these tumors (Becker et al., 1994). Additionally, a limited number of genes including IQGAP1 (Takemoto et al., 2001), TGFβ1, TGFβRII (Chung et al., 1997; Maehara et al., 1999), HGF and HGFR (c-met) (Lee et al., 2000), have been reported to show altered expression in diffuse-type gastric cancers.

Since most cancer cells in the diffuse type are scattered and accompanied by marked stromal reactions such as desmoplasia and lymphocytic infiltrates, bulk tissues would necessarily yield expression data from a mixture of cancer cells and abundant noncancerous cells. However, laser-microdissection technology now allows us to obtain precise expression patterns in a pure population of diffuse-type gastric-cancer cells. Although several studies of gene expression in diffuse-type gastric cancer have been reported (Hippo et al., 2001; Hippo et al., 2002; Leung et al., 2002; Boussioutas et al., 2003; Chen et al., 2003; Kim et al., 2003; Tay et al., 2003), none of the studies used laser microdissection for the analysis. Thus, our data is the first genome-wide expression profiles of pure population of diffuse-type gastric-cancer cells.

In a previous study, we carried out a genome-wide analysis of gene-expression profiles of 20 intestinal-type gastric-cancer tissues using a cDNA microarray containing 23 040 genes (Hasegawa et al., 2002). In those experiments, a large number of genes showed altered expression in the cancer tissues compared to corresponding noncancerous gastric mucosae. Since diffuse- and intestinal-type gastric cancers show distinctly different clinicopathological features in many aspects, we were encouraged to compare gene-expression profiles of diffuse-type tumors with those obtained earlier in intestinal-type gastric cancers.

For the present study, we used the same cDNA microarray system and microdissected cell populations from 20 diffuse-type gastric cancers. We identified genes showing altered expression in this type of tumor including some associated with tumor progression. We were also able to identify numerous genes that were differentially expressed between diffuse- and intestinal-type gastric cancers. The data presented here provide important information regarding the mechanisms of gastric carcinogenesis and may contribute to the development of type-specific diagnostic markers and/or therapeutic targets.

Results

Identification of genes commonly up- or downregulated in diffuse-type gastric cancers

Comparison of the expression profiles of 20 tumors with their corresponding noncancerous mucosae identified 153 genes (including 23 of unknown function) that were upregulated (Supplementary Material: 1) (Table 1 and http://www.ims.u-tokyo.ac.jp/nakamura/furukawa/microarray.html) and 1553 (including 395 of unknown function) that were downregulated (Supplementary Material: 2) (Table 2 and http://www.ims.u-tokyo.ac.jp/nakamura/furukawa/microarray.html) in 50% or more of the samples examined. The upregulated genes represented a variety of functions including genes associated with signal-transduction pathways (S100A10, ANXA1, GNAI2, LY6E, RAI3 and PLAB), genes encoding transcription factors (TGIF2 and ETV4), or genes involved in various metabolic pathways (GPX1, BACH, NNMT and GNPI), transport systems (ATP1B3 and KCNA3), cell proliferation (TGFBI), apoptosis (CARD4 and BIRC5), protein translation and RNA processing (PRKDC, HSPCB and FBL), cell-cycle regulation (PRC1, CDC25B and CDC20), cell adhesion and cytoskeleton (ZYX, LCP1, ARHGDIB and MSLN), cell motility and extracellular matrix (CD81, MMP7, SPARC, COL3A1 and FN1), immunity (MIF, IFITM2 and G1P2), and other functions (CTSB).

Table 1 The 75 most upregulated genes in diffuse-type gastric cancer^aa

Table 2 The 75 most downregulated genes in diffuse-type gastric cancer^aa

Commonly downregulated elements included genes associated with various metabolic pathways (ALDH3A1, GSTA1, FBP1, CA2, AKR1C3 and CYP3A7), small-molecule or heavy-metal transport (ATP2A3, GIF, MT1F and SLC7A8), defense response (TFF1, TFF2 and GSTA3), immunity (LTF, IL1R2, FCGBP and C5), signal transduction and cell-cycle regulation (MAL and ERBB2IP), cell proliferation (PAP and REG1A) or other functions (SYTL2 and PGC).

To validate our microarray data, we selected eight representative genes (TGFBI, SPARC, COL3A1, MSLN, FLJ20736, GW112 and two ESTs) that were commonly upregulated in diffuse-type tumors, and performed semiquantitative RT–PCR using eight pairs of the same RNA samples that had served for the microarray analysis (Figure 1). Among the 42 microarray data that showed normalized Cy3 and Cy5 signal intensities above cutoff value, 39 were consistent with those obtained by semiquantitative RT–PCR. Hence, we estimated the concordance between the microarray and semiquantitative RT–PCR data to be 92.85%, corroborating the high reliability of our microarray data.

Figure 1

Re-evaluation of elevated expression of eight genes in diffuse-type gastric cancers (T) and corresponding noncancerous epithelia (N) by semiquantitative RT–PCR using eight pairs of the aRNAs utilized in microarray analysis. Expression of FDFT1 served as an internal control

Cluster analysis of diffuse- and intestinal-type gastric cancers

To arrange the samples and genes on the basis of similarities, we carried out a two-way unsupervised hierarchical clustering algorithm using a total of 1051 genes that had passed the first filter described in Materials and methods. The cluster analysis classified the 40 tumor samples into two major classes that correctly corresponded to their histological subtypes, namely 20 diffuse- and 20 intestinal-type gastric cancers (Figure 2). The data indicated that diffuse-type cancer has a distinctly different expression profile from that of intestinal-type cancer. Regarding the gene axis, the algorithm classified the genes into several clusters that represented different expression signatures among the tumors. We identified clusters that included a number of genes associated with similar biological processes. For example, clusters A, B and C, which contained genes that were relatively upregulated in diffuse-type cancers, included a number of genes associated with cell adhesion or migration and the extracellular matrix (ECM), for example, SPARC, TUBB2, ITGB1, ARHGDIA, COL1A1 and COL1A2, and genes associated with immune response or metabolism (IGHG3, IGKC, C3, APOA1, GPX1 and LDHA). On the other hand, genes in groups D and F, which were relatively upregulated in intestinal-type cancers, contained genes associated with growth-factor receptors such as GRB10, IRS2 and EPS15R, and genes related to cellular proliferation or mitochondrial function such as EGFR, EGR1, PCNA, CDK2, PSORT, TIMM10 and BCAT2. These data appear to reflect the different natures of the two types of tumor. Notably, clones corresponding to the same gene that had been spotted at different positions on the microarray slides were present in the same branch of the gene-cluster dendrogram, indicating a high degree of reliability for our microarray data.

Figure 2

An unsupervised two-way hierarchical clustering analysis of 1051 genes across 40 gastric-cancer samples. In the horizontal axis, the 40 tumors were classified into two major trunks, while in the vertical axis, the 1051 genes were clustered according to similarities in relative expression ratios. The color of each cell in the matrix represents the expression level of a single gene, with red and green indicating expression levels, respectively, above and below the median derived from log-transformed relative expression ratios for that gene across all samples. Black represents unchanged expression, gray indicates no or insignificant expression (intensities of both Cy3 and Cy5 under the cutoff value)

Identification of genes that were differentially expressed between diffuse- and intestinal-type gastric cancers

We carried out random permutation tests to identify genes that could discriminate diffuse-type cancer from intestinal-type cancer. The tests identified a total of 46 genes with P-values of less than 0.01. We further performed an additional, supervised hierarchical clustering analysis using the data for the 46 genes that had clearly separated the two classes of tumors (Figure 3). Of these 46 genes, 14 (including three of unknown function) showed relatively higher expression levels, and 32 (including eight of unknown function) showed relatively lower expression levels in diffuse-type cancer than intestinal-type gastric cancer. The 14 upregulated elements included genes encoding chaperones (CCT3 and TOR1B) and genes associated with cell motility and cytoskeleton (CD81 and TUBA3), glycosylation (RPN2, MGAT1 and MPI), or other functions (SDCCAG8 and HRMT1L2). The 32 genes with relatively lower expression levels in diffuse-type cancers included some involved in signal transduction and transcriptional regulation (RHBDL, SFRS8, MLL5, LDB3 and GFRA2), nuclear transportation (KPNB2 and NUP133), cell adhesion (PSK-1, ITGB5, SRPX and IBSP), or other functions (HRG and TG737) (Table 3). We also carried out quantitative RT–PCR experiments with three randomly selected discriminating genes (SLC25A4, GFRA2, HSPCB) to verify the microarray data. The results of RT–PCR experiments using 16 pairs of RNA from eight intestinal-type and eight diffuse-type cancers consistently showed significantly different levels of expression between these two types of cancer, supporting the reliability of our strategy (Figure 4).

Figure 3

A supervised two-way hierarchical clustering analysis using 46 discriminating genes. The color of each cell in the matrix represents the expression level of a single gene in an individual sample, with red and green indicating expression levels, respectively, above and below the median derived from log-transformed relative expression ratios for that gene across all samples. Black represents unchanged expression

Table 3 Genes differentially expressed between diffuse- and intestinal-type gastric cancers

Figure 4

Quantitative RT–PCR analysis of genes differentially expressed between intestinal- and diffuse-type gastric cancers. Each dot represents a log 2-transformed expression ratio (tumor : normal) of the three selected genes in the eight intestinal-type and eight diffuse-type tumors. Short horizontal lines represent the median ratios for each type of tumor

Identification of genes related to LN metastasis, venous invasion, or lymphatic-vessel involvement in diffuse-type gastric cancer

Additional random permutation tests identified 13 genes (including three of unknown function) associated with venous invasion and 11 (including three of unknown function) associated with lymphatic vessel invasion. Discriminators in both invasion groups included genes involved in differentiation (S100P), cell adhesion (SDCBP), transport (SNX2, HCN3 and ATP6V0A1), transcription (FOXD1, ZFP36), or signal transduction (SYK, PPP3CA and PTPRJ). We also identified 31 genes that were differentially expressed between tumors with and without LN metastasis; 13 (including six of unknown function) showed elevated expression and 18 (including two of unknown function) showed reduced expression in tumors with LN metastasis. Discriminators in the LN metastasis group included genes involved in differentiation (NOTCH2), cytoskeleton (TUBB2), metabolism (FACL5 and ACAA1), signal transduction (EPS15), transcription, or protein synthesis (ZFP103, RPS23, RPS10, RPL31 and EIF4G2) (Figure 5).

Figure 5

A supervised hierarchical clustering analysis of expression profiles associated with (a) venous invasion, (b) lymphatic vessel involvement, or (c) LN metastasis in diffuse-type gastric cancers. In the horizontal axis, the 20 diffuse-type tumors were correctly classified into two major trunks corresponding to the histopathological data. In the vertical axis, the genes were clustered according to their similarities in expression patterns. The color of each cell in the matrix represents the expression level of a given gene in an individual sample: red indicates elevated expression, green reduced expression and black unchanged expression. Color standards are indicated on the right of the cluster

Discussion

The advent of laser-microdissection technology has brought about a great improvement in our ability to isolate cancer cells from interstitial tissues. The proportion of contaminated cells using this method is estimated to be less than 0.3% (Yanagawa et al., 2001) and 0.29% (Nakamura et al., 2004) in our earlier studies, which suggests that our data should represent the expression profile of a highly pure population of diffuse-type gastric-cancer cells.

Some of the genes that were commonly upregulated in our experiments reflected the distinct nature of diffuse-type gastric cancer. A number of genes involved in the TGF-β signaling pathway, such as TGFBI and PLAB, were upregulated in this type of gastric cancer. TGFBI, first isolated from human lung-adenocarcinoma cell line A549 (Skonier et al., 1992), has a recognition site for integrin, a key molecule for adhesion and migration of cancer cells. PLAB helps regulate tissue differentiation and maintenance, and also inactivates macrophages by suppressing production of TNF-α (Bootcov et al., 1997). Since previous analyses of gene expression in intestinal-type gastric cancers (Hasegawa et al., 2002), colorectal cancers (Kitahara et al., 2001) and pancreatic cancers (lacobuzio-Donahue et al., 2003) also showed elevated expression of TGFBI or PLAB, enhanced activity of the TGF-β signaling pathway seems to play an important carcinogenetic role in a wide range of human tissues.

We showed also that MMP-7 and TIMP1, genes associated with cell–ECM interaction, were highly expressed in diffuse-type but not in intestinal-type gastric cancers (Hasegawa et al., 2002). As MMPs catalyse extracellular matrices including collagen and fibronectin, they are thought to be involved in invasion and/or metastasis of cancer cells. Enhanced expression of MMP7, a uterine matrilysin, has been reported in cancers of the pancreas (Crnogorac-Jurcevic et al., 2002), ovary (Schwartz et al., 2003), bladder (Sumi et al., 2003) and colon (Zeng et al., 2002). Immunohistochemical staining has demonstrated marked accumulation of MMP-7 in the tumor-front areas of metastatic colorectal-cancer cells, suggesting involvement of MMP-7 activation in metastasis (Zeng et al., 2002). Despite the fact that TIMP-1 was initially identified as a gene encoding an inhibitor of metalloproteinases, recent studies have pointed out its paradoxical effect in carcinogenesis through regulation of cell growth, apoptosis and angiogenesis. The upregulation of TIMP-1 in our data agrees with earlier findings that elevated expression of this gene is associated with poor prognosis among patients with several types of human tumors (Jiang et al., 2002), since the diffuse type of gastric cancer often shows poorer prognosis than the intestinal type.

We also observed enhanced expression of genes involved in cell migration or production of ECM components (e.g. SPARC, FN1, TUBB2, ITGB1, COL1A1 and COL1A2) in diffuse-type cancers but not in intestinal-type cancers. Upregulation of SPARC has been implicated in invasive and/or metastatic phenotypes (De et al., 2003). The importance of genes encoding ECM components was emphasized in a recent study that identified COL1A1 and COL1A2 as members of a metastasis-associated gene signature (Ramaswamy et al., 2003). Although collagens were reported to be expressed specifically in tumor endothelial cells (St Croix et al., 2000), COL1A1 and COL1A2 were abundantly expressed in MKN45 and St-4 diffuse-type gastric cancer cell lines when we analysed expression profiles of 39 human cancer cell lines using the same microarray system (Dan et al., 2002). Therefore, augmented expression of these collagens should not result from contaminated stromal cells, but from elevated expression in cancer cells. Our data strongly suggest that elevated expression of genes encoding ECM components is in fact a characteristic feature of diffuse-type gastric-cancer cells.

Genes involved in glycosylation (MGAT1, MPI and RPN2) were significantly enhanced in diffuse-type cancers, but not in intestinal-type cancers. Increased activity of these genes may confer properties of invasion and/or metastasis to diffuse-type gastric-cancer cells, since invasion and metastasis are highly dependent on alterations in the ECM and cell–cell interactions that, in part, involve structural changes in cell-surface components including glycosylation (Couldrey and Green, 2000). Our data imply that dysregulated cell–matrix interaction is a common feature of diffuse-type gastric cancers. Interestingly, SDCCAG8, a tumor antigen whose elevated expression was associated with poor survival in gastric-cancer patients (Chen et al., 2003), was also upregulated in our diffuse-type cancers. Hence, overexpression of SDCCAG8 in diffuse-type cancer may contribute to its poor-prognosis characteristic and may be used for prediction of prognosis in patients.

The majority of downregulated elements in our experiments represent a set of genes that function in various metabolic pathways and transport systems. Among them were several genes with specific functions in gastric epithelium, such as PGC and GIF, implying that dedifferentiation is a common feature of carcinogenesis. The trefoil factor family 1 gene (TFF1), which participates in stabilizing the mucous gel overlying the gastrointestinal mucosa, was among the downregulated genes in both diffuse- and intestinal-type gastric cancers (Hasegawa et al., 2002). Since TFF1-knockout mice develop multiple gastric adenomas and carcinomas, and somatic mutations in TFF1 have been reported in human gastric cancers (Park et al., 2000), the reduced expression of TFF1 in our populations of gastric-cancer cells is in line with a gastric-specific tumor-suppressive role. These findings are consistent with genes depicted in the normal gastric-tissue cluster compared to tumor cluster previously reported (Chen et al., 2003).

One of the mechanisms that tumor cells use to escape from the immune response is to compromise the antigen-presenting functions of infiltrating dendritic cells by secreting tumor-derived factors. One of the upregulated genes on our list, G1P2 (or ISG15), encodes a key cytokine that is secreted by melanoma cells to induce E-cadherin expression on dendritic cells and thereby impair migratory behavior (Padovan et al., 2002). This finding may reflect a novel survival mechanism of diffuse-type gastric-cancer cells. Furthermore, in our experiments MHC class II genes such as CD74, HLA-DRA and HLA-DQB1 were expressed to a lesser degree in diffuse-type tumors than intestinal-type tumors (Figure 2, group E). Since these genes tend to be expressed abundantly in the new molecular subtypes of gastric cancer that show better prognosis (Tay et al., 2003), our data support the notion that the ability to escape from immunological surveillance confers aggressive properties on diffuse-type tumors.

Since diffuse-type gastric cancer has more invasive and metastasizing potential and poorer prognosis than intestinal-type cancer, it is in our interest to clarify the mechanisms underlying progression of this tumor. Therefore, we attempted to identify genes specifically associated with venous invasion, lymphatic vessel involvement or LN metastasis in diffuse-type cancers. Among such genes, SYK, a nonreceptor type of protein tyrosine kinase, showed lower expression in tumors having lymphatic vessel involvement than in tumors without that feature. Since reduced expression of wild-type SYK has been correlated with poor prognosis and with distant metastases in breast cancers (Toyama et al., 2003), and its overexpression suppresses motility and invasiveness of breast-cancer cells, SYK is regarded as a candidate metastasis-suppressor gene (Mahabeleshwar and Kundu, 2003). Hence, its decreased expression in gastric-cancer cells may also play a role in metastasis through invasion of lymphatic vessels. Expression of CCL25, encoding a chemokine that functions as an effector of lymphocyte migration, was higher in tumors with venous invasion than in the noninvasion group. Tumor-associated chemokines are thought to play roles in invasion and/or metastasis including control of the movement of tumor cells themselves (Balkwill, 2003). Since CCL25 mediates lymphocyte entry into the small intestine by promoting their rapid adhesion to vascular endothelium or transmigration into lamina propria (Campbell and Butcher, 2002), cancer cells with upregulated CCL25 may promote hematogenous metastasis by the same mechanism.

In this study, we identified many genes that were commonly altered in diffuse-type gastric cancer, as well as genes that were differentially expressed between diffuse- and intestinal-type tumors. This large body of information has not only clarified mechanisms involved in diffuse-type gastric cancer, but has revealed distinct molecular signatures of both types of gastric cancer. Further characterization of the genes identified in this study may lead to a more profound understanding of gastric carcinogenesis and facilitate the development of novel diagnostic markers and more effective therapeutic modalities for each histological type of gastric cancer.

Materials and methods

Patients and tissue samples

Primary gastric cancers and corresponding noncancerous gastric mucosae were obtained with informed consent from 20 patients who underwent gastrectomy at Kyoto University Hospital, Kitano Hospital, Cancer Institute Hospital, Sakai City Hospital, and Hospitals of Kinki University, Japan. Patient profiles were obtained from medical records. Histopathological classification of each tumor, performed according to Lauren's classification (Lauren, 1965), diagnosed all samples as diffuse-type gastric adenocarcinomas. Clinical stage was determined according to the UICC TNM classification. The 20 cases analysed consisted of 19 advanced (T2–T4) and one early (T1) tumors. All cancer tissues were dissected from the margin of tumor mass, while their corresponding noncancerous tissues were collected from mucosa in the same region of the stomach. Clinicopathological data for these 20 samples and for the 20 intestinal-type gastric-cancer samples analysed previously are summarized in Table 4. No significant differences were observed in terms of gender, age of patients, size of tumor, depth of invasion, vessel invasion, or node involvement between the two groups. All samples were immediately frozen and embedded in TissueTek OCT medium (Sakura, Tokyo, Japan) and stored at −80°C until used for microarray analysis.

Table 4 Histological data of the 40 gastric cancer clinical samples used for cDNA microarray analysis

Laser-microbeam microdissection, extraction of RNA, and T7-based RNA amplification

Frozen tissues were prepared in 8-μm sections, as described previously (Kitahara et al., 2001) except for elimination of the final dehydration in xylene. A total of 30 000–40 000 cancer cells or noncancerous gastric epithelial cells were collected selectively using the EZ cut system (SL Microtest GmbH, Germany) according to the manufacturer's protocol. Extraction of total RNA, T7-based amplification, and labeling of probes were performed as described previously (Kitahara et al., 2001). A measure of 2.5-μg aliquots of twice-amplified RNA (aRNA) from each cancerous and noncancerous tissue were then labeled, respectively, with Cy3-dCTP or Cy5-dCTP (Amersham Biosciences).

cDNA microarray and analysis of data

We used the same cDNA microarray system containing 23 040 cDNAs in duplicate and the same procedures for hybridization, washing, photometric quantification of signal intensities of each spot, and normalization of data as in our previous analysis of 20 intestinal-type gastric cancers (Hasegawa et al., 2002). To filter less reliable data derived from low signal intensities, we determined the cutoff value as the intensity of spot whose S/N (signal to noise) ratio was 3. Genes were categorized into three groups according to their expression ratios (Cy3/Cy5): upregulated (ratio equal to or greater than 2.0), downregulated (ratio equal to or less than 0.5), and unchanged expression (ratios between 0.5 and 2.0). Genes with Cy3/Cy5 ratios greater than 2.0 or less than 0.5 in more than 50% of the cases examined were defined, respectively, as commonly up- or downregulated genes.

Semiquantitative RT–PCR

We selected eight representative genes (TGFBI, SPARC, COL3A1, MSLN, FLJ20736, GW112 and two ESTs) that were commonly upregulated in diffuse-type tumors and examined their expression levels by semiquantitative RT–PCR. A housekeeping gene, FDFT1, served as an internal control because it had shown the smallest Cy3/Cy5 fluctuation in our experiments. Reverse transcription was carried out as described previously (Kitahara et al., 2001). The PCR reaction was preceded by 95°C for 2 min, then underwent 25 cycles of 95°C for 30 s, 60°C for 30 s, and 72°C for 30 s followed by 72°C for 5 min. The primer sequences were as follows: FDFT1 forward, 5′-TGTGTGGCTGGGACCTTTAGGAA-3′ and reverse, 5′-TCATTCTAGCCAGGATCATACTAAG-3′; TGFBI forward, 5′-TCCCTGGAAAAGGAGCTTCAGTA-3′ and reverse, 5′-ACACCATGGCTCTGTCACAATAG-3′; SPARC forward, 5′-CAAGAGTGAGATGTAGAAAGTTGT-3′ and reverse, 5′-CTTCACATCATGGTGAGAGTTTG-3′; COL3A1 forward, 5′-AGACGCATGTTATGGTGCTAATGTA-3′ and reverse, 5′-GATCAACAACCACATACAAGCTTAC-3′; MSLN forward, 5′-AACGGCTACCTGGTCCTAGAC-3′ and reverse, 5′-GTTTACTGAGCGCGAGTTCTCT-3′; FLJ20736 forward, 5′-ATATGAGCAGGAACTCTGGGAG-3′ and reverse, 5′-CTCAGGATAGAGGGCAAAGAGA-3′; GW112 forward, 5′-GAAAATCTGATGGCAGTGACAA-3′ and reverse, 5′-AAGGTTTCCAACTACTGCACTGA-3′; EST (GenBank Accession No. AA143060) forward, 5′-TGTTGCTCTTCCTTGTGGAGCT-3′ and reverse, 5′-GCAAATCCTACTTTCAACTGCAC-3′; EST (GenBank Accession No. AA430699) forward, 5′-TTTAACGCTGGTGGGCAGCA-3′ and reverse, 5′-ATAAACAGAACCCATCCCAAAG-3′.

Cluster analysis and identification of genes able to discriminate between diffuse- and intestinal-type gastric cancers

We undertook an unsupervised cluster analysis to compare the expression profiles of 20 diffuse-type gastric cancers with profiles of the 20 intestinal-type gastric tumors analysed previously (Hasegawa et al., 2002). For this procedure, we selected 1051 genes that had given valid values in more than 50% of all samples and standard deviations of the values greater than 2. After log-transforming and median-centering the data, we performed average-linkage hierarchical clustering and viewed the results using web-available software packages ‘Cluster’ and ‘TreeView’ (http://genome-www5.stanford.edu/MicroArray/SMD/restech.html) (Eisen et al., 1998). We further carried out random permutation tests using log-transformed data of genes showing Cy3 or Cy5 intensities above cutoff values, and estimated the ability of each gene to distinguish diffuse- from intestinal-type tumors. Samples were randomly permuted 10 000 times between the two classes to calculate a P-value for each gene. Genes were considered significant when signal intensities were higher than cutoff values in more than 50% of the cases, P-values were less than 0.01, and ∣Med_d-Med_i∣ was greater than or equal to 0.5, where Med indicates the median derived from log-transformed relative expression ratios in diffuse- or intestinal type. A total of 46 genes passed this filter and were selected for an additional, supervised clustering algorithm.

Quantitative RT–PCR

We selected three of the discriminating genes (SLC25A4, GFRA2, HSPCB) and examined their expression levels by means of real-time PCR experiments (Taqman PCR; Applied Biosystems, Foster City, CA, USA). QARS served as an internal control because it had shown the smallest Cy3/Cy5 fluctuation in our data for both diffuse- and intestinal-type cancers. The Taqman assay was carried out according to the manufacturer's protocol, with the same aRNAs that had been used for microarray analysis. The PCR reaction involved initial denaturation at 95°C for 10 min, followed by 45 cycles at 95°C for 15 s and 60°C for 1 min. The sequences of primers and probes were as follows: QARS forward, 5′-GGTGGATGCAGCATTAGTGGA-3′ and reverse, 5′-AAGACGCTCAAACTGGAACTTGTC-3′; probe, 5′-VIC-CTCTGTGGCCCTGGCAAAACCCTT-TAMRA-3′; SLC25A4 forward, 5′-AGTTGACTGCTGGAGGAAGATTG-3′ and reverse, 5′-ATTGGACCAGGCACCTTTGA-3′; probe, 5′-FAM-CTTGGCTCCTTCGTCTTTT-TAMRA-3′; GFRA2 forward, 5′-GTGGCGAGGCATTAAAACTTG-3′ and reverse, 5′-GGACCGTTTCTCTCTGACTTCAA-3′; probe, 5′-FAM-TTCTGCCACCGAGAAAGAA-TAMRA-3′; HSPCB forward, 5′-CCCCTGCTGGTGTCTAGTGTTT-3′ and reverse, 5′-CCAATCCTGCTGTCAAGAGTAGAG-3′; probe, 5′-FAM-ACACCCTTAGTTTACTGCCT-TAMRA-3′.

Identification of gene-expression profiles related to histological data

We divided the 20 samples into three groups of two opposing characteristics according to histopathological data, namely (1) vessel invasion-positive or -negative, (2) lymphatic vessel-involvement-positive or -negative, and (3) lymph node-metastasis-positive or -negative, and performed random-permutation analyses. Genes having signal intensities higher than the cutoff value in more than 50% of the cases, P-values less than 0.01, and ∣Med_d-Med_i∣⩾1.3, 1.2 and 0.5 were selected as discriminating genes for venous invasion, lymphatic vessel-involvement and LN metastasis, respectively. These procedures selected for subsequent cluster analysis 13 genes that distinguished group 1, 11 for group 2, and 31 for group 3.