Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Gene expression analysis of early and advanced gastric cancers


Gastric carcinoma is one of the major causes of cancer mortality worldwide. Early detection results in excellent prognosis for patients with early cancer (EGC), whereas the prognosis of advanced cancer (AGC) patients remains poor. It is not clear whether EGC and AGC are molecularly distinct, and whether they represent progressive stages of the same tumor or different entities ab initio. Gene expression profiles of EGC and AGC were determined by Affymetrix technology and quantitative polymerase chain reaction. Representative regulated genes were further analysed by in situ hybridization (ISH) on tissue microarrays. Expression analysis allowed the identification of a signature that differentiates AGC from EGC. In addition, comparison with normal gastric mucosa indicated that the majority of alterations associated with EGC are retained in AGC, and that further expression changes mark the transition from EGC to AGC. Finally, ISH analysis showed that representative genes, differentially expressed in the invasive areas of EGC and AGC, are not differentially expressed in the non-invasive areas of the same tumors. Our data are more directly compatible with a progression model of gastric carcinogenesis, whereby EGC and AGC may represent different molecular stages of the same tumor. Finally, the identification of an AGC-specific signature might help devising novel therapeutic strategies for advanced gastric cancer.


Gastric cancer is the fourth most frequent type of cancer in the world with a prevalence that varies greatly depending on geographic location (Parkin, 2004). In western countries, this cancer has a decreasing incidence but still accounts for significant overall mortality (Dicken et al., 2005). The prognosis of gastric carcinoma largely depends on whether it is diagnosed as an early gastric carcinoma (EGC) or advanced gastric cancer (AGC). EGC is a lesion confined to mucosa and submucosa, irrespective of lymph node involvement. Conversely, AGC progressively penetrates the muscularis propria or the subserosa, the serosa and the adjacent structures (Kajitani, 1981; Wang et al., 1997). In EGC, prognosis is favorable, after gastrectomy and lymph node dissection, accounting for 5- and 10-year survival rates of 90 and 85–90%, respectively, compared with 20–30% in AGC (Lawrence and Shiu, 1991; Dicken et al., 2005). Therefore, great attention has been directed to early detection in countries where gastric carcinoma is prevalent, such as Japan and Korea (Henderson, 1990; Parkin, 2004).

It is not clear whether early and advanced gastric cancers are distinct entities ab initio, or whether they represent progressive stages of the same tumor. Clinical evidence supports the latter contention, as non-concurrent long-term follow-up studies reported that EGCs, for the most part, progress to the advanced stage, leading to death from gastric cancer, if left untreated (Tsukuma et al., 1983, 2000). However, the molecular alterations that might differentiate EGC from AGC are not known. Therefore, molecular knowledge might prove very valuable in the management of gastric cancer, allowing, for instance, preoperative diagnosis on biopsies, or more precise classification, with impact on prognostic evaluation and possibly on therapeutic regimens.

Expression profiling might provide such knowledge, and efforts have been made in this direction. However, available studies have primarily dealt with comparisons of normal tissues, or precancerous lesions, with advanced tumors (Boussioutas et al., 2003; Meireles et al., 2004; Yu et al., 2005) or with the determination of profiles associated with the most common histotypes, intestinal and diffuse ( Hippo et al., 2001; Hasegawa et al., 2002; Kim et al., 2003; Jinawath et al., 2004; Norsett et al., 2004). Much less is known in terms of comparison between EGC and AGC, owing to the limited number of samples analysed and owing to the difficulty of evaluating the impact of non-cancerous cells within the tumor tissue (Lee et al., 2003; Oue et al., 2004).

We report here that EGC and AGC have distinct molecular profiles, and that a signature can be identified that distinguishes between the two conditions. Our data are also consistent with the notion that EGC and AGC have a common tumor origin, and that AGC-specific alterations in gene expression might thus occur later in the natural history of the tumor and be part of the mechanism of malignant progression.


EGC and AGC are molecularly distinct

Twenty-eight gastric cancers, nine EGC and 19 AGC (see Supplementary Table 1), were profiled using Affymetrix technology. To explore whether EGC and AGC are molecularly distinct, we carried out a supervised analysis to select genes whose expression levels significantly differed between the two tumor classes (2.0-fold change, P-value <0.05 with multiple testing correction). This analysis identified a signature of 415 probe sets (Supplementary Table 2) able to discriminate, by two-way hierarchical clustering, EGC from AGC with 86% accuracy (Figure 1a). A list of the top 20 up- and downregulated genes is shown in Table 1. Of note, in the dendrogram of Figure 1a, as well as in those of Figures 3 and 4, EGC and AGC samples did not cluster according to their grade (tumor grades are reported in Supplementary Table 1).

Figure 1

Expression profiling of EGC and AGC. (a) Hierarchical clustering of 415 probe sets differentially expressed (FC>2 and P<0.05), by Affymetrix analysis, in AGCs (n=19) and EGCs (n=9). (b) Hierarchical clustering of the 10 representative genes on AGC (n=9) and EGC (n=8) samples based on Q-PCR data. (c) Data relative to both the Affymetrix analysis on 28 tumors (right) and the Q-PCR analysis on a subset of 17 tumors (left) of the 10 genes selected for validation. The probe set number and the gene symbol are indicated. For each gene in the two classes of samples median normalized average expression values were calculated, and relative fold changes and t-test P-values are displayed relative both to the Affymetrix (right) data and Q-PCR data (left). Asterisk indicates no significant P-value. Rows, probes; columns, samples. Sample code: blue, AGC; red, EGC. Topology code: black, cardias; grey, antrum; light blue, fundus; violet, body. Grade code: black, G3; dark gray, G2; light gray, G1.

Table 1 List of the top 20 up- and downregulated genes differentially expressed in EGC and AGC
Figure 3

Overall expression profile of normal gastric mucosa, EGC and AGC. Left: two-way hierarchical clustering based on 954 probe sets that varied the most across the three classes of normal (NOR), EGC and AGC samples with Spearman correlation similarity measure. Right: the expression values of the two major clusters of genes are also shown as box plots, displaying the distribution of relative gene expressions in each sample for the selected gene list. Each box extends from the 25th to the 75th percentiles; the solid line within each box represents the median. The whiskers below and above each box plot extend to the 1.5x of the interquartile range (75th–25th percentile). Sample code: cyan, Normal; blue, AGC; red, EGC. Topology code: black, cardias; grey, antrum; cyan, fundus; violet, body. Grade code: black, G3; dark gray, G2; light gray, G1; white, not applicable (NOR).

Figure 4

Statistical analysis of normal gastric mucosa, EGC and AGC. (a) Schematic representation and Venn diagram of the sequential statistical analysis performed on normal (NOR), EGC and AGC samples. (b) Hierarchical clustering of 1024 probe sets on 16 normal samples (NOR), nine EGC and 19 AGC. (c) Hierarchical clustering of 369 probe sets on 16 normal samples, nine EGC and 19 AGC. The normal sample of patient ID 18 clustered together with the AGC class, possibly due to the presence of contaminating AGC tumor cells (pT4, N2). Sample code: blue, AGC; red, EGC; light blue, NOR. Grade code: black, G3; dark gray, G2; light gray, G1; white, not applicable (NOR).

A pool of 10 representative genes, displaying a fold change regulation ranging from 2.0 to 6.5, was selected from the 415-probe set list for validation by quantitative polymerase chain reaction (Q-PCR) (Figure 1c). Q-PCR, performed on a subset of 17 carcinomas (eight EGC and nine AGC), indicated excellent concordance with the data obtained by the Affymetrix analysis (Figure 1b and c). Although additional samples should be analysed for an independent validation, together, gene profiling and the Q-PCR suggest that a molecular signature able to discriminate EGC from AGC can be identified.

Next, we performed in situ hybridization (ISH) on tissue microarrays (TMA), including normal gastric mucosa, EGC and AGC. Twelve representative genes whose major trend of regulation ranged between 2.0 and 6.0, from the 415-probe set list, were tested (Supplementary Table 3). In eight cases, positive signals were detected; the remaining four genes did not show appreciable signal, possibly due to low mRNA abundance (Supplementary Table 3). ISH-TMA analysis identified two classes of genes: (i) epithelial-specific (three genes, CLDN18, TFF2 and CLDN23) and (ii) microenvironment-associated (five genes, COL8A1, CDH11, SULF1, SFRP2 and GPNMB) (Supplementary Table 3). Remarkably, all the epithelial-specific genes were, in the Affymetrix screening, downregulated in AGC versus EGC, whereas the opposite was true for the microenvironment-associated genes (Supplementary Table 3).

To carry out a statistical analysis of the ISH-TMA data, we concentrated on two epithelial genes, CLDN18 and TFF2, and three stromal genes, COL8A1, CDH11 and SULF1. In the remaining three cases, the number of positive samples was too low for meaningful analysis. The two epithelial genes, CLDN18 and TFF2, were homogeneously expressed at high levels in almost all normal gastric mucosa samples. Interestingly, the number of positive samples gradually decreased as the disease progressed (Figure 2 and Supplementary Table 4). The three microenvironment-associated genes CDH11, COL8A1 and SULF1, were significantly upregulated in the peritumoral stroma of advanced cancers when compared with normal tissues but only in a low percentage of early invasive lesions, indicating that their expression could be specifically induced during late stages of tumor invasion (Figure 2 and Supplementary Table 4).

Figure 2

ISH-TMA analysis of representative genes. Left: ISH images of the epithelial-associated (CLDN18 and TFF2) and stromal (SULF1, CDH11 and COL8A1) genes on EGC and AGC. The bright field panels (BF) show hematoxylin/eosin counterstaining; the dark field panels (DF) show the ISH signals (bright areas) at × 20 magnification. The boxed area is also shown at higher magnification (HM panels, only the bright field is shown), with indication of the tumor (T) and stromal (S) areas. Right: Bar graph representation of the percentage of samples in which a positive signal (ISH scores 2–3) was observed in normal mucosa, EGC and AGC. Asterisks indicate the statistical significance (P<0.05) of differential expression in EGC and AGC tumor classes vs normal samples (Fisher's exact test). Actual numbers and statistical analysis are in Supplementary Table 4.

Overall, the ISH-TMA results corroborate the Affymetrix data (taking into account the different sensitivities of the two methods), and further indicate that there are molecular differences between EGC and AGC both at the epithelial and the stromal level. Finally, they provide evidence for a progressive gain and/or loss of gene expression in the transition from normal mucosa to EGC to AGC.

Evidence for progression from EGC to AGC

Evidence presented so far is compatible with the idea that EGC and AGC, albeit molecularly distinct, represent differently progressed forms of the same tumor entity. To gain insights into this issue we performed two independent series of analyses.

First, we performed additional gene expression analyses including 16 normal tissue samples. A two-way hierarchical clustering was performed to explore the overall degree of similarity of normal (NOR), EGC and AGC samples by selecting 954 probe sets that varied most among the three classes of samples (see Materials and methods for details). The majority of normal, EGC and AGC clustered in three separate branches (Figure 3). Of note, the EGC cluster correlated more closely with normal than with AGC samples (Figure 3) and displayed a gene expression profile intermediate between normal and AGC samples (as also shown by the box plots in Figure 3). To further investigate on this finding, we performed a statistical analysis following the analytical scheme depicted in Figure 4a. Initially, we compared normal tissues with the EGC samples and we identified 1024 probe sets differentiating the two classes (Supplementary Table 5). Remarkably, only three genes of these probe set list (SPP1, FMO5 and GC) significantly changed between EGC and AGC (analysis of variance cutoff P-value 0.05 with multiple testing correction). Moreover, by clustering analysis, the 1024 probe sets were able to separate normal samples from the two EGC and AGC classes with few exceptions (Figure 4b). We next compared this list with the previously identified list of 415 probe sets that discriminates between EGC and AGC tumors. Only 46 probe sets were common in the two lists, whereas 369 of 415 probe sets (89%) were specific of the AGC condition (Figure 4a). Indeed, the 369 probe sets (Supplementary Table 6) were able to separate normal tissues and EGC from the AGC class (Figure 4c). Of note, a functional analysis of the 369 probe sets indicated that genes of the defense response and the extracellular matrix were remarkably and specifically activated in AGC tumors, whereas genes involved in cell metabolism and digestion were downregulated pointing to a gradual loss of normal gastric function, as the disease progresses (see Discussion and Supplementary Table 9).

Together, these data suggest that most of the alterations in gene expression associated with the conversion from normal mucosa to EGC are retained in AGC, and that additional changes may occur later on, during tumor progression, differentiating EGC from AGC.

We also analysed, by ISH-TMA, the expression trend of the two identified epithelial genes, TFF2 and CLDN18 (differentially regulated between EGC and AGC, see Supplementary Table 4) in high-grade dysplasia, which represents a common precursor/preinvasive lesion of both EGC and AGC. In four cases (two EGC and two AGC), normal mucosa, high-grade dysplasia and corresponding invasive tumors were available. Both CLDN18 and TFF2 displayed high expression in normal gastric mucosa. Remarkably, these two epithelial genes were expressed at similarly high levels in the high-grade dysplasia of both the EGC and the AGC (Figure 5), indicative of a common preinvasive neoplastic program. However, in invasive AGC but not in invasive EGC, the expression of the two genes was strongly downregulated. These data not only are in agreement with the overall profiles detected in the Affymetrix screening but also with the idea of a common origin of AGC and EGC, which, thereby, may represent progressive stages of the same tumor.

Figure 5

ISH-TMA analysis of epithelial-specific genes in normal mucosa, high-grade dysplasia and invasive carcinomas. Serial TMA sections of normal, high-grade dysplasia, and invasive tumor areas of the same EGC and AGC samples of two representative patients (two EGC and two AGC) are shown. Expression levels of the epithelial-associated TFF2 and CLDN18 genes (black/brown signal in bright field; silver grain signal in dark field) varied according to the tissue-type and gene studied, sometimes totally obscuring the underlining cell morphology. Normal: normal gastric mucosa; dysplasia: high grade dysplasia; invasive: invasive gastric carcinoma.


Several long-term follow-up clinical studies support the notion that human gastric carcinogenesis is a multistep and multifactorial process (Tsukuma et al., 1983, 2000; Rugge et al., 2003). Progression of low- and high-grade non-invasive neoplastic lesions to invasive adenocarcinoma, and of early gastric lesions to advanced stages has been clinically documented (Tsukuma et al., 1983, 2000; Rugge et al., 2003).

The results of our expression profile analysis are in agreement with the clinical evidence, and further argue that although EGC and AGC are molecularly distinct, they, probably, represent temporal variations in the natural history of the same tumor entity. We identified two major signatures: the first, of 1024 probe sets, characterizes the transition from normal mucosa to EGC. Importantly, most of the genes in this signature are also retained in the transition from normal mucosa to AGC. The second signature, of 415 probe sets, distinguishes EGC from AGC. Within this second signature, the greatest majority of the genes (369 probe sets, see Figure 4a) are completely AGC-specific, and they also separate normal tissues from AGC (Figure 4c). In addition, ISH-TMA analysis of selected genes was also compatible with a progression model.

One important caveat to address was the possibility that the differentiation status of the analysed tumors could affect the outcome of expression profiles. This is unlikely to be the case, in that tumor grade did not appear to have a great impact on the clusters reported in Figures 1, 3 and 4. Thus, in the reported analyses, tumor stage (AGC vs EGC) seems to be more important than tumor grade in separating the samples. In addition, ISH analysis showed that representative genes, differentially expressed in the invasive areas of EGC and AGC, are not differentially expressed in the non-invasive areas of the same tumors indicative of a common preinvasive neoplastic program (Figure 5). As a final note of caution, however, we note that there are scenarios under which our data would be compatible with a separate origin ab initio of EGC and AGC. For instance, genetic alterations, separately initiating EGC and AGC, might occur in overlapping but not entirely identical regions of the genome. Further studies will be needed to rule out, or to confirm, this possibility.

A plausible corollary of the ‘progression’ model is that, in some cases, tumors should be identified that present an intermediate expression profile between EGC and AGC. This would be the result of a process in which the more aggressive (progressed) component of the tumor has not yet substituted completely the less aggressive component. This possibility draws attention to the ‘outlier’ cases, that is those cases that are not clearly grouped with EGC or AGC. In our study, three EGC samples (EGC ID 02, 05 and 31) display an expression pattern, considering the 369 AGC-specific probe sets, intermediate between normal and AGC (see Figure 4c). Clearly, a larger series of cases will be needed to establish whether the detection of an intermediate signature marks the initial phases of the transition from EGC to AGC.

What is the nature of the gene expression changes associated with the transition from normal to EGC to AGC? Clues could be derived from a functional classification based on Gene Ontology (Supplementary Tables 7–9). In the normal mucosa → EGC transition (1024 probe set list), up- and downregulation events were equally distributed (52 and 48%, respectively). Upregulated genes/pathways were significantly, and not unexpectedly, enriched for those related to cell cycle, RNA processing, ribosome biogenesis and cytoskeleton organization. Conversely, downregulation events affected, in many cases, genes/pathways implicated in specific functions of the gastric mucosa (digestion, lipid metabolism and G-protein coupled receptor protein signaling pathway), consistent with a loss of normal gastric function and differentiation (Supplementary Table 7).

In the EGC → AGC transition (415 probe set list), 59% of the probe sets were upregulated and 41% downregulated. It is remarkable that in our detailed ISH-TMA analysis, all upregulated genes were of stromal origin, whereas the downregulated ones were epithelial specific. This indicates an important participation of the tumor microenvironment to tumor progression, in agreement with the concept that cancer cells might recruit stromal cells essential for tumor spread (Cunha et al., 2003; Bhowmick et al., 2004). This notion was further reinforced by the Gene Ontology analysis (Supplementary Tables 8 and 9), which showed significant upregulation in AGC of genes belonging to stress/defense response, response to wounding and cell motility. Moreover, many genes of the extracellular matrix, in particular collagens, were specifically identified in the 415-gene list (Supplementary Table 2). These findings underscore the importance of the immune response in the pathogenesis of gastric cancer, as it is known that chronic injury or inflammation predisposes to neoplastic progression (Wang and Fox, 1998; Balkwill and Mantovani, 2001). Moreover, the importance of genes encoding extracellular matrix components was recently emphasized in a study that identified COL1A1 and COL1A2 as members of a metastasis-associated gene signature (Ramaswamy et al., 2003). Finally, two transcription factors, TWIST1 and SNAI2, were found upregulated in AGC (Supplementary Table 2). Recently, these two genes were reported to play an essential role in tumor recurrence and metastasis by promoting epithelial–mesenchymal transition (Yang et al., 2004; Moody et al., 2005).

With regard to genes downregulated in the EGC → AGC transition (172 probe sets in the 415 probe set list), we speculate that the majority of them might be epithelial specific, based on our ISH-TMA survey. This possibility is supported by the functional classification, which evidenced how downregulated genes included genes involved in digestion and cell metabolism (as already seen for the normal → EGC transition, Supplementary Tables 7 and 8) pointing to a further loss of gastric differentiation, as the disease progresses. In four cases, however, the combined analysis of Affymetrix and ISH-TMA data allowed the unequivocal identification of epithelial-specific genes as significantly downregulated in AGC. One of them, TFF2, is a member of the trefoil family of peptides, which displays a well-established protective effect of the gastrointestinal tract by promoting the healing of injured mucosa (Farrell et al., 2002). TFF2 and the other member of the family, TFF1, have been reported to be frequently downregulated in primary gastric cancers (Kirikoshi and Katoh, 2002; Kim et al., 2003), and TFF1 might be a gastric-specific tumor suppressor gene (Lefebvre et al., 1996). Other two epithelial-specific genes, downregulated in AGC, are those encoding claudin 18 and 23. In addition, claudin 15 is part of the 415 probe set list. Claudins constitute a family of more than 20 proteins involved in the organization of tight junctions (Van Itallie and Anderson, 2006). Loss of normal tight junction functions constitutes a hallmark of human carcinomas (Mullin, 2004), and subversion of claudin function has been implicated in various solid malignancies (Katoh, 2003; Swisshelm et al., 2005).

Overall, our data suggest that EGC and AGC share many initial alterations (comprised in the 1024 probe sets list) centered on activation of cell proliferation pathways (and related functional categories), and downregulation of normal gastric differentiation. Regardless of the mechanism (progression or ab initio), the more aggressive biological phenotype in AGC seems to be linked to massive upregulation of stromal components and to downregulation of epithelial-specific functions connected, at least in part, with protection of mucosal integrity and functionality.

Whether a particular cell type is the tumor progenitor of both EGC and AGC, that is a bone marrow cell as recently reported in a Helicobacter pylori infection mouse model (Houghton et al., 2004), remains to be established. Finally, our results have implications for the diagnosis and the treatment of gastric cancers. First, the 1024 probe set list, marking the progression from normal to EGC and also to AGC, represents a reservoir of candidate markers for gastric cancer early detection. Second, some of the gene products of the AGC-specific signature (415 probe sets) might represent potential therapeutic targets for this poor prognosis malignancy.

Materials and methods

Tissue samples

Specimens were from 32 patients who underwent surgery for gastric cancer at San Paolo Hospital and Fondazione IRCCS Ospedale Maggiore Policlinico, Mangiagalli e Regina Elena hospital (Milan, Italy). Written informed consent was obtained from all patients. Samples were snap-frozen in liquid nitrogen and stored in Killik frozen section medium (Bioptica, Milan, Italy) at −80°C. Part of the sample was formalin fixed and paraffin embedded. Clinical and pathological features are reported in Supplementary Table 1. Additional details are in Supplementary Information.

RNA extraction, GeneChip hybridization and quantitative real-time PCR assay

Total RNA was extracted using commercial homogenization (QIAshredder) and purification (RNeasy Mini Kit) reagents (Qiagen, Valencia, CA, USA). RNA quality was analysed with the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). A total of 500 ng RNA were amplified using the T7-polymerase-based double linear amplification protocol (Eberwine et al., 1992). Twenty-five micrograms of cRNA probe was hybridized onto the Affymetrix (Santa Clara, CA, USA) HG-U133 ChipSet, and processed according to Affymetrix technical protocols. The average intensity of every array was scaled to a predefined value (target intensity) of 500, in order to make arrays comparable.

Q-PCR was performed with TaqMan methodology, using ABI Prism 7900HT Sequence Detection System (Applied Biosystems, Foster City, CA, USA) (see Supplementary Information for details). Q-PCR data analysis and expression normalization were performed using the geNorm software and methodology (Vandesompele et al., 2002). Relative abundance for each gene was imported in GeneSpring and normalized to the median of all relative abundances; the hierarchical clustering analysis was then performed using the Standard Correlation as Similarity Measure.

Statistical analysis

Expression profiles, preprocessed with MAS5, were exported to GeneSpring GX software version 7.3 (Agilent Technologies,). According to the GeneSpring normalization procedure, in each analysis the 50th percentile of all measurements was used as a positive control, within each hybridization array, and each measurement for each gene was divided by this control. The bottom 10th percentile was used for background subtraction. Among different hybridization arrays, each gene was divided by the median of its measurements in all samples. Data were then log transformed for subsequent analysis.

Expression data were prefiltered by considering both MAS5 ‘Absolute Call’ flags and average expression measurements within each group analysed. We selected probe sets called present or marginal (P or M) at least once across all samples, and whose mean raw expression levels were 200 within at least one class of samples. The prefiltering method removed those probe sets whose expression signal was constantly too close to the background throughout the entire set of samples.

In order to find genes whose expression levels significantly differed between EGC (n=9) and AGC (n=19), we adopted a supervised method of analysis, using the GeneSpring software. Mean values were calculated within the two classes for each probe set, and fold-change ratios between the EGC and the AGC were derived. A difference of twofold cutoff was applied to select upregulated and downregulated genes. A further statistical analysis was performed using Welch's approximate t-test and ANOVA, with P-value cutoff of 0.05, without the assumption of equality of variances. Benjamini and Hochberg false discovery rate (FDR) was used for multiple testing correction in this and all subsequent analyses. By this analysis, 415 probe sets were found to be significantly regulated between the two classes of samples.

For the direct comparison of NOR versus EGC, we applied the same statistical criteria as described before and we identified a list of 1024 probe sets differentially regulated between normal mucosa and EGC.

To analyse the overall degree of similarity of 16 normal samples (NOR), nine EGC and 19 AGC, in addition to the prefiltering procedure previously described, we selected those probe sets whose average expression values had a 2.0-fold change difference in either direction from the gene median value within at least one class of samples. A total of 954 probe sets passed the filtering criteria.

Unless differently specified, two-way hierarchical clustering analyses were performed with the GeneSpring GX 7.3 (Agilent Technologies) using Pearson correlation as similarity metric.

TMA analysis

A gastric TMA was prepared as previously described (Kononen et al., 1998), which contained eight early (pT1) and 23 advanced (pT2, pT3 and pT4) primary gastric carcinomas, including the samples used for the Affymetrix gene chip hybridization (with the exception of one EGC, sample ID 19). Normal gastric mucosa from the same patients was also arrayed in 23 cases (see Supplementary Table 1 for details). For four patients (two EGC and two AGC), two representative cores of high-grade dysplasia were also included. Two representative cores of all samples were arrayed. A detailed description of the ISH methodology is in Supplementary Information. To evaluate gene expression, two pathologists independently counted all tumor cells in each core using a dark field condenser to visualize the silver grains. An initial semiquantitative scale was established as follows: 0 (no staining), 1 (1–25 grains/cell; weak staining), 2 (26–50 grains/cell, moderate staining) and 3 (>50 grains/cell, strong staining). Scores of 2 or 3 were considered unequivocally positive. When differences in signal intensity occurred between the two representative cores, the highest score was considered for data analysis.

Accession codes




  1. Balkwill F, Mantovani A . (2001). Inflammation and cancer: back to Virchow? Lancet 357: 539–545.

    CAS  Article  Google Scholar 

  2. Bhowmick NA, Chytil A, Plieth D, Gorska AE, Dumont N, Shappell S et al. (2004). TGF-beta signaling in fibroblasts modulates the oncogenic potential of adjacent epithelia. Science 303: 848–851.

    CAS  Article  Google Scholar 

  3. Boussioutas A, Li H, Liu J, Waring P, Lade S, Holloway AJ et al. (2003). Distinctive patterns of gene expression in premalignant gastric mucosa and gastric cancer. Cancer Res 63: 2569–2577.

    CAS  Google Scholar 

  4. Cunha GR, Hayward SW, Wang YZ, Ricke WA . (2003). Role of the stromal microenvironment in carcinogenesis of the prostate. Int J Cancer 107: 1–10.

    CAS  Article  Google Scholar 

  5. Dicken BJ, Bigam DL, Cass C, Mackey JR, Joy AA, Hamilton SM . (2005). Gastric adenocarcinoma: review and considerations for future directions. Ann Surg 241: 27–39.

    Google Scholar 

  6. Eberwine J, Yeh H, Miyashiro K, Cao Y, Nair S, Finnell R et al. (1992). Analysis of gene expression in single live neurons. Proc Natl Acad Sci U S A 89: 3010–3014.

    CAS  Article  Google Scholar 

  7. Farrell JJ, Taupin D, Koh TJ, Chen D, Zhao CM, Podolsky DK et al. (2002). TFF2/SP-deficient mice show decreased gastric proliferation, increased acid secretion, and increased susceptibility to NSAID injury. J Clin Invest 109: 193–204.

    CAS  Article  Google Scholar 

  8. Hasegawa S, Furukawa Y, Li M, Satoh S, Kato T, Watanabe T et al. (2002). Genome-wide analysis of gene expression in intestinal-type gastric cancers using a complementary DNA microarray representing 23,040 genes. Cancer Res 62: 7012–7017.

    CAS  Google Scholar 

  9. Henderson BE . (1990). Summary report of the sixth symposium on cancer registries and epidemiology in the Pacific Basin. J Natl Cancer Inst 82: 1186–1190.

    CAS  Article  Google Scholar 

  10. Hippo Y, Yashiro M, Ishii M, Taniguchi H, Tsutsumi S, Hirakawa K et al. (2001). Differential gene expression profiles of scirrhous gastric cancer cells with high metastatic potential to peritoneum or lymph nodes. Cancer Res 61: 889–895.

    CAS  Google Scholar 

  11. Houghton J, Stoicov C, Nomura S, Rogers AB, Carlson J, Li H et al. (2004). Gastric cancer originating from bone marrow-derived cells. Science 306: 1568–1571.

    CAS  Article  Google Scholar 

  12. Jinawath N, Furukawa Y, Hasegawa S, Li M, Tsunoda T, Satoh S et al. (2004). Comparison of gene-expression profiles between diffuse- and intestinal-type gastric cancers using a genome-wide cDNA microarray. Oncogene 23: 6830–6844.

    CAS  Article  Google Scholar 

  13. Kajitani T . (1981). The general rules for the gastric cancer study in surgery and pathology. Part I. Clinical classification. Jpn J Surg 11: 127–139.

    CAS  Article  Google Scholar 

  14. Katoh M . (2003). CLDN23 gene, frequently down-regulated in intestinal-type gastric cancer, is a novel member of CLAUDIN gene family. Int J Mol Med 11: 683–689.

    CAS  Google Scholar 

  15. Kim B, Bang S, Lee S, Kim S, Jung Y, Lee C et al. (2003). Expression profiling and subtype-specific expression of stomach cancer. Cancer Res 63: 8248–8255.

    CAS  Google Scholar 

  16. Kirikoshi H, Katoh M . (2002). Expression of TFF1, TFF2 and TFF3 in gastric cancer. Int J Oncol 21: 655–659.

    CAS  Google Scholar 

  17. Kononen J, Bubendorf L, Kallioniemi A, Barlund M, Schraml P, Leighton S et al. (1998). Tissue microarrays for high-throughput molecular profiling of tumor specimens. Nat Med 4: 844–847.

    CAS  Article  Google Scholar 

  18. Lawrence M, Shiu MH . (1991). Early gastric cancer. Twenty-eight-year experience. Ann Surg 213: 327–334.

    CAS  Article  Google Scholar 

  19. Lee JH, Choi SR, Hwang TH, Kim MC, Jung GJ, Roh MS et al. (2003). A gene expression in study human gastric adenocarcinoma using a cDNA microarray. Korean J Gastroenterol 42: 484–495.

    Google Scholar 

  20. Lefebvre O, Chenard MP, Masson R, Linares J, Dierich A, LeMeur M et al. (1996). Gastric mucosa abnormalities and tumorigenesis in mice lacking the pS2 trefoil protein. Science 274: 259–262.

    CAS  Article  Google Scholar 

  21. Meireles SI, Cristo EB, Carvalho AF, Hirata R, Jr, Pelosof A, Gomes LI et al. (2004). Molecular classifiers for gastric cancer and nonmalignant diseases of the gastric mucosa. Cancer Res 64: 1255–1265.

    CAS  Article  Google Scholar 

  22. Moody SE, Perez D, Pan TC, Sarkisian CJ, Portocarrero CP, Sterner CJ et al. (2005). The transcriptional repressor Snail promotes mammary tumor recurrence. Cancer Cell 8: 197–209.

    CAS  Article  Google Scholar 

  23. Mullin JM . (2004). Epithelial barriers, compartmentation, and cancer. Sci STKE 2004: pe2.

    CAS  Google Scholar 

  24. Norsett KG, Laegreid A, Midelfart H, Yadetie F, Erlandsen SE, Falkmer S et al. (2004). Gene expression based classification of gastric carcinoma. Cancer Lett 210: 227–237.

    CAS  Article  Google Scholar 

  25. Oue N, Hamai Y, Mitani Y, Matsumura S, Oshimo Y, Aung PP et al. (2004). Gene expression profile of gastric carcinoma: identification of genes and tags potentially involved in invasion, metastasis, and carcinogenesis by serial analysis of gene expression. Cancer Res 64: 2397–2405.

    CAS  Article  Google Scholar 

  26. Parkin DM . (2004). International variation. Oncogene 23: 6329–6340.

    CAS  Article  Google Scholar 

  27. Ramaswamy S, Ross KN, Lander ES, Golub TR . (2003). A molecular signature of metastasis in primary solid tumors. Nat Genet 33: 49–54.

    CAS  Article  Google Scholar 

  28. Rugge M, Cassaro M, Di Mario F, Leo G, Leandro G, Russo VM et al. (2003). The long term outcome of gastric non-invasive neoplasia. Gut 52: 1111–1116.

    CAS  Article  Google Scholar 

  29. Swisshelm K, Macek R, Kubbies M . (2005). Role of claudins in tumorigenesis. Adv Drug Deliv Rev 57: 919–928.

    CAS  Article  Google Scholar 

  30. Tsukuma H, Mishima T, Oshima A . (1983). Prospective study of "early" gastric cancer. Int J Cancer 31: 421–426.

    CAS  Article  Google Scholar 

  31. Tsukuma H, Oshima A, Narahara H, Morii T . (2000). Natural history of early gastric cancer: a non-concurrent, long term, follow up study. Gut 47: 618–621.

    CAS  Article  Google Scholar 

  32. Van Itallie CM, Anderson JM . (2006). Claudins and Epithelial Paracellular Transport. Annu Rev Physiol 68: 403–429.

    CAS  Article  Google Scholar 

  33. Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N, De Paepe A et al. (2002). Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biol 3: research0034.1–research0034.11.

    Article  Google Scholar 

  34. Wang CS, Hsueh S, Chao TC, Jeng LB, Jan YY, Chen SC et al. (1997). Prognostic study of gastric cancer without serosal invasion: reevaluation of the definition of early gastric cancer. J Am Coll Surg 185: 476–480.

    CAS  Article  Google Scholar 

  35. Wang TC, Fox JG . (1998). Helicobacter pylori and gastric cancer: Koch's postulates fulfilled? Gastroenterology 115: 780–783.

    CAS  Article  Google Scholar 

  36. Yang J, Mani SA, Donaher JL, Ramaswamy S, Itzykson RA, Come C et al. (2004). Twist, a master regulator of morphogenesis, plays an essential role in tumor metastasis. Cell 117: 927–939.

    CAS  Article  Google Scholar 

  37. Yu CD, Xu SH, Mou HZ, Jiang ZM, Zhu CH, Liu XL . (2005). Gene expression profile differences in gastric cancer, pericancerous epithelium and normal gastric mucosa by gene chip. World J Gastroenterol 11: 2390–2397.

    CAS  Article  Google Scholar 

Download references


We thank D Nicosia, A Hurtado, G Jodice, the Sequencing and the Affymetrix Services at IFOM for technical assistance. This work was supported by a grant from Cariplo Foundation to GC, and by AIRC (Italian Association for Cancer Research), the European Community (VI Framework), the Italian Ministries of Health, and of Education and University, the Monzino Foundation and the Ferrari Foundation to PPDF.

Author information



Corresponding authors

Correspondence to P P Di Fiore or S Bosari.

Additional information

Supplementary Information accompanies the paper on the Oncogene website (

Supplementary information

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Vecchi, M., Nuciforo, P., Romagnoli, S. et al. Gene expression analysis of early and advanced gastric cancers. Oncogene 26, 4284–4294 (2007).

Download citation


  • gastric cancer
  • early gastric cancer
  • advanced gastric cancer
  • gene expression profiling
  • tissue microarrays

Further reading


Quick links