Tissue samples were prospectively collected from patients who underwent gastric surgery or gastroscopy at the National Cancer Center (NCC) Hospital between 2008 and 2009. All tissues were obtained according to the protocols approved by the Institutional Review Board, NCC for the human subject guideline of NCC (NCCNCS-08-127) that is in accordance with the principles of the Declaration of Helsinki. The samples were obtained by endoscopic biopsies from gastric cancer patients who gave informed consent to the protocol. The samples were stored at –80°C. The clinical and pathological features of the patients are listed in Table I.

Total RNA was extracted from gastric cancer and adjacent normal tissues from EGC patients using TRIzol reagent (Invitrogen, Carlsbad, CA, USA), followed by purification of the RNA using Qiagen RNeasy mini kit columns (Valencia, CA, USA) according to the manufacturer’s instructions. RNA quality was evaluated using the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA) and concentration measured by Nanodrop 1000 (Thermo Scientific, Wilmington, DE, USA). Only RNAs showing distinct 18S/28S ribosomal peak ratios of 1.5–2.0 in the Bioanalyzer (Agilent Technologies) and 260/280 ratios of 1.8–2.1 in the Nanodrop (Thermo Scientific) analyses were accepted for further analysis.

Genome-wide gene expression was analyzed in the 27-paired EGC tissue samples using Affymetrix GeneChip Human Exon1.0 ST Array (Santa Clara, CA, USA). Target preparation and microarray processing procedures were carried out as described in the manufacturer’s instructions, and raw data were deposited in the NCBI Gene Expression Omnibus (GSE30727). The data were preprocessed by a default robust multi-array average (RMA) method implemented in the Bioconductor (www.bioconductor.org) ‘oligo’ package. The differentially expressed genes between EGC tissues and adjacent non-cancerous gastric tissues (i.e., the up- and downregulated genes in EGC) were filtered by a fold-change cut-off of 1.5 and a P-value cut-off of 0.05.

We downloaded a Gene Ontology (GO) annotation file (gene_association.goa_human) and an ontology file (gene_ontology_ext.obo) from www.geneontology.org, as recommended by the BiNGO tutorial (9). In the BiNGO analysis, all options, except for filtering the IEA code, were set at default values. The false discovery rate (FDR) cut-off was 0.05. DAVID v6.7 software (http://david.abcc.ncifcrf.gov/) was used to summarize the over-representation of the KEGG pathways (10). The gene expression signatures of up- or downregulated genes in EGC were analyzed using the L2L microarray analysis tool (http://depts.washington.edu/l2l/) (11).

Two micrograms of total RNA were reverse transcribed with Superscript III reverse transcriptase (Invitrogen). Reverse transcription PCR (RT-PCR) was performed using 5 ng cDNA for 1 cycle at 94°C for 2 min, followed by 32–35 cycles of 94°C for 20 sec, 60°C for 40 sec and 72°C for 30 sec, using gene-specific primers (Table II). Gene expression levels were analyzed by gel electrophoresis.

Independent additional cancer datasets were obtained from NCBI GEO (www.ncbi.nlm.nih.gov/geo) and EBI ArrayExpress (www.ebi.ac.uk/arrayexpress): GSE19536 for ERα-negative breast cancers (12), and the E-MTAB-62 dataset for Ewing’s sarcoma, bladder cancer, small cell lung cancer and LNCaP prostate cancer cell lines (13). The up- and downregulated genes in the EGC tissues were compared with these 5 cancer types. We transformed the expression of all our EGC tissue samples, GSE19536 and E-MTAB-62, into standard scores (z-scores), and then performed hierarchical clustering for the 6 cancers.

We selected differentially expressed genes (i.e., up- and downregulated genes) from the 27 pairs of EGC tissue and their adjacent normal tissue. The P-value cut-off of 0.05 in t-tests, and the fold-change cut-off of 1.5 or 1/1.5 for up- and downregulated genes, respectively, was used for selection. We identified 556 upregulated genes and 417 downregulated genes. The differentially expressed genes were then fed into function, pathway and cross-experiment analyses to acquire a deeper understanding of the molecular basis of EGC.

The BiNGO plug-in on the Cytoscape platform (http://www.cytoscape.org/) was used to explore the molecular function and biological processes in GO.

The functions of the upregulated genes of EGC tissues were significantly associated with C-X-C and other chemokine-related signaling, interleukin-8 binding, growth factor binding, collagen binding, and the extracellular matrix (ECM) (Fig. 1A). Moreover, the biological processes involved in the upregulated genes in the EGC tissues were strongly related to cell proliferation, mitosis, apoptosis and cell-matrix adhesion (Fig. 1B). Additionally, wound healing terms, cell migration terms and cell motility terms were also listed in the upregulated genes with statistical significance (Table III). Since cell migration, cell motility and wound healing are typically observed in late-stage, metastatic cancer, this may indicate that EGC tissues could possess intrinsic aggressiveness, despite their early detection. Conversely, the downregulated genes were strongly linked to oxidoreductase activity (e.g., oxidoreductase activity acting on the CH or CH2 groups, quinones) in GO molecular function (Fig. 1C). Furthermore, the downregulated genes were enriched in various terms related to metabolic processes (e.g., flavone and flavonoid metabolic pathways) in GO biological processes (Fig. 1D). The GO terms of the downregulated genes clearly indicate dysregulation of metabolism in EGC, which is one of the emerging cancer hallmarks (14).

The DAVID tool (http://david.abcc.ncifcrf.gov/) was used to inspect the KEGG biological pathways associated with the differently expressed genes in EGC.

The upregulated genes in EGC tissues were intrinsically associated with cytokine-cytokine receptor interactions, ECM-receptor interactions, the cell cycle, hematopoietic cell lineage and Toll-like receptor signaling pathways (Table IV). In addition, focal adhesion and cell adhesion molecule pathways were highlighted. Thus, similar to the functional enrichment analysis, upregulated pathways in these tumor tissues suggest a strong potential for cell motility and metastasis, despite early detection. In contrast, the downregulated genes in the EGC tissues were strongly associated with xenobiotics-, drug-, retinol-, starch- and sucrose-related metabolism, steroid hormone biosynthesis, as well as pentose and glucuronate interconversion pathways (Table IV).

Our functional and pathway analyses demonstrated that the significantly upregulated genes in EGC tissues are associated with cell migration and metastasis, events typical of late-stage cancer. To verify our findings, we further analyzed the expression pattern of matrix metalloproteinases (MMPs), which are well known cell migration-related genes. MMPs are also known to play critical roles in the regulation of cell invasion by ECM proteolysis, as well as by processing cytokine precursors in the chemokine network.

We analyzed the expression pattern of 7 MMPs (MMP1, −3, −7, −9, −10, −12 and −13) within the upregulated gene data in EGC tissues. As expected, and consistent with the microarray data where MMPs were upregulated 1.56- to 8.68-fold (Fig. 2A), RT-PCR indicated that MMP mRNA expression was highly upregulated in the patients’ EGC tissues (Fig. 2B).

In order to investigate similar molecular signatures between EGC and other cancer types, we compared our data of differentially expressed genes with a public gene expression signature warehouse, L2L. This revealed that the upregulated genes in EGC most significantly correlated with the gene expression signature of ERα-negative breast cancer (Table V). As summarized in Table V, the upregulated genes in EGC were also similar to the gene expression signature related to an undifferentiated cancer status (cancer_undifferentiated_meta_up: 69 genes commonly upregulated in undifferentiated cancer relative to well-differentiated cancer, from a meta-analysis of the OncoMine gene expression database), stemness (stemcell_embryonic_up: enriched in mouse embryonic stem cells, compared to differentiated brain and bone marrow cells) and survival (dox_resist_gastric_up: upregulated in gastric cancer cell lines resistant to doxorubicin, compared to parent chemosensitive lines). Together, the EGC tissues reflect various facets of cancer-related phenotypes, viz., strong survival, stem-like and morphology.

The same L2L analysis was applied to the downregulated genes in EGC (Table V). Interestingly, epigenetic-related cancer gene expression signature terms (5azac_hepg2_up and 5azac-tsa_hepg2_up in Table V) were highly ranked. This suggests that global alterations in DNA methylation and histone modification occur in EGCs, as it does in other cancers.

To validate the result of the L2L analysis showing a relationship between EGC and ERα-negative breast cancer, we performed a hierarchical clustering analysis. The expression datasets of the differently expressed genes in EGC (556 upregulated gene symbols and 417 downregulated gene symbols), ERα-negative breast cancer and 4 additional cancers (small cell lung cancer, LNCaP prostate cancer cell lines, bladder cancer and Ewing sarcoma) were used in an unsupervised hierarchical clustering analysis. As in the L2L analysis, the results indicated that EGC correlated most closely with the ERα-negative cancer than with the other 4 cancers (Fig. 3).

When we inspected the expression levels (z-scores) of the 7 MMP genes (Fig. 4), the results indicated that the ERα-negative cancer, above all other observed cancers, showed the most similar expression patterns for the 7 MMPs. Overall, the hierarchical clustering was consistent with the cross-experimental analysis and strongly supported the molecular similarity between EGC and ERα-negative breast cancer in terms of carcinogenesis.