Molecular Pathogenesis of Gastric Cancer

Copy-Number Variations

More than 60% of gastric cancer (GC) cases exhibit chromosomal instability (CIN) characterized by gross copy-number changes 1. Deng et al. 2 used high-resolution genomic analysis to profile somatic copy-number alterations in a panel of 233 GC (primary tumors and cell lines) and 98 matched gastric nonmalignant samples. Regarding broad chromosomal regions, the most frequently amplified included 1q, 5p, 6p, 7p, 7q, 8q, 13q, 19p, 20p, and 20q, and the most frequently deleted regions included 3p, 4p, 4q, 5q, 6q, 9p, 14q, 18q, and 21q, which were also identified in at least two of other four studies published last year addressing copy-number variation in GC 2-6. Concerning focal genomic alterations, 22 recurrently altered regions were found 2. Amplifications detected included FGFR2, ERBB2, EGFR, MET, KRAS, MYC, and CCND1 (previously known to be amplified in GC), and also GATA4, GATA6, and KLF5 transcription factors. Somatic deletions were found in FHIT, RB1, CDKN2A/B, and WWOX and in genes not previously reported in GC such as PARK2, PDE4D, PTPRD, CSMD1, and GMDS 2. These results were largely overlapping with those of Dulak et al. 6 that conducted a similar analysis in a large series of gastrointestinal tumors that included 110 GC.

Deng et al. also showed that amplifications in the receptor tyrosine kinases (RTK) genes FGFR2 (9%), EGFR (8%), ERBB2 (7%), and MET (4%) were mutually exclusive, and that KRAS amplification (9%) was also mutually exclusive to RTKs amplification. RTK amplification was shown to be a predictor of poor prognosis, independently of tumor stage and grade 2. As RTK/RAS amplifications collectively occurred in 37% of the primary GC analyzed, the authors suggest that these patients may potentially be treated with RTK/RAS-directed therapies.

Exome Sequencing

Aiming at identifying the spectrum of somatic mutations in GC, Zang et al. 7 used an exome-sequencing approach to study the coding regions of about 18,000 genes of 15 GC and matched controls. Among the most commonly mutated genes, the authors identified TP53 (11/15; 73%), PIK3CA (3/15; 20%), and CTNNB1 (2/15; 13%), which had been previously observed in GC, and 26 other genes that were mutated in at least two of the 15 GC. Interestingly, cell adhesion was the most enriched biologic pathway among the frequently mutated genes, which included PKHD1, CTNNB1, CNTN1, and FAT4. The authors then focused on FAT4, a cadherin family gene, and performed an additional screening that confirmed the presence of FAT4 mutations in 5% (6/110) and genomic deletions in 4% (3/83) of gastric tumors. In functional assays, silencing of FAT4 in wild-type GC cell lines resulted in increased cell proliferation and soft-agar colony formation, increased cell invasion and migration, and reduced cell adhesion to matrix components, suggesting that FAT4 has a tumor-suppressor role 7.

Zang et al. also observed that almost half of the tumors had mutations in chromatin remodeling genes, including ARID1A (3/15; 20%), MLL3 (2/15; 13%), MLL (1/15; 6.7%), DNMT3A (1/15; 6.7%), and SETD1A (1/15; 6.7%). In a prevalence screening, somatic mutations in the AT-rich interactive domain-containing protein 1A (ARID1A) gene were detected in 8% of GC (9/110) 7. Mutations in ARID1A gene had recently been identified in several tumor types, including GC (10/100; 10%) 8, and in another exome-sequencing study of 22 GC samples by Wang et al. 9. What both studies demonstrated was that ARID1A mutations were associated with tumor microsatellite instability (MSI) 7, 9. Tumors harboring ARID1A mutations had loss or reduced ARID1A protein expression 9, and two other studies confirmed in large series of GC cases that ARID1A expression was lost in tumors and associated with poor prognosis 10, 11. Also in agreement with Wang et al. 9 that identified higher incidence of ARID1A mutations in MSI and in MSS EBV-infected GC, in comparison with MSS EBV-noninfected GC, Abe et al. 10 showed that loss of ARID1A protein expression was more frequent in MSI and in EBV-infected tumors. Functional analysis revealed that ARID1A inactivation in wild-type GC cell lines enhanced proliferation 7, 11 and increased the levels of E2F1 and CCNE1 mRNA, suggesting that ARID1A may inhibit cell-cycle regulation 7. Altogether, these data suggest that somatic inactivation of FAT4 and ARID1A may be important tumorigenic events in a subset of GC 7.

Gene Expression Profiling

Based on gene expression profiling from 248 gastric tumors, Lei et al. 12 identified three gastric tumor subtypes classified as proliferative, mesenchymal, and metabolic. The proliferative subtype, associated with Lauren's intestinal type, was characterized by gene sets related to cell cycle and DNA replication and had high activities for oncogenic pathways such as E2F, MYC, and RAS. These tumors had high levels of TP53 mutations, DNA hypomethylation, and genomic instability. Proliferative subtype tumors also showed enrichment in copy-number alterations that included regions of recurrent amplifications of the oncogenes CCNE1, MYC, ERBB2, and KRAS and of deletions of the genes PDE4D and PTPRD that were previously reported 2, 12. Tumors of the mesenchymal subtype, which were associated with Lauren's diffuse type, had high activity of the EMT, TGF-β, VEGF, NFκB, mTOR, and SHH pathways and contained features of cancer stem cells. Mesenchymal subtype tumors also showed a high proportion of aberrantly hypermethylated CpGs. Interestingly, cell lines of this tumor subtype were sensitive to inhibitors of the PI3K-AKT-mTOR pathway 12. Tumors of the metabolic subtype were characterized by gene sets from metabolism pathways and by high activity of the spasmolytic polypeptide-expressing metaplasia (SPEM) pathway. Another interesting finding was that not only cell lines of the metabolic subtype were more sensitive to 5-fluorouracil (5-FU) than cells of the other subtypes, but also patients with metabolic subtype tumors appeared to have had benefits from 5-FU treatment in terms of cancer-specific and disease-free survival 12.

Methylation Profiling

Zouridis et al. 13 extensively characterized the repertoire of DNA methylation events associated with GC on a genome-wide scale in 240 tumors and 94 matched samples of adjacent normal tissue. Overall, they found tumor-specific hypermethylation (in most cases and preferentially located in CpG islands) and hypomethylation (in a smaller proportion of cases and in sites outside CpG islands). Their data also allowed the identification of a cluster with CpG island methylator phenotype (CIMP), which was associated with prevalent hypermethylation, younger age of patients, and adverse patient outcome independently of the disease-stage, confirming previous studies proposing that a subset of GC display CIMP features 14. Analysis of the hypermethylated CpG sites in CIMP tumors showed enrichment in genes related to stem cells. Another interesting finding was that CIMP cell lines were sensitive to treatment with the DNA methylation inhibitor 5-Aza-2′-deoxycytidine (5-Aza-dC) and in vivo a significant reduction in tumor growth was observed in the cisplatin/5-Aza-dC-treated cell xenografts. Additionally, the authors found large chromosomal regions of tumor hypermethylation (LRESs), which were associated with the CIMP status, and which appeared to regulate silencing of specific genes instead of all genes within the region 13. Furthermore, they have identified large chromosomal regions of tumor hypomethylation, which were associated with increased CIN. This study is an important contributor also to the understanding of the epigenetic influence of Helicobacter pylori on gastric epithelial cells in order to increase the risk for GC.

In this regard, Cheng et al. 15 undertook genome-wide methylation profiling analyses of human GC specimens and of gastric samples of a mouse model of H. pylori infection. They used an integrative approach by overlapping the two microarray lists of hypermethylated genes, which revealed that forkhead box D3 (FOXD3) was the common hypermethylated gene in H. pylori-infected gastric mucosa and GC. The authors also observed progressive FOXD3 promoter methylation along the gastric carcinogenesis cascade. There were increased methylation levels in H. pylori-positive gastritis and intestinal metaplasia (IM) tissues in comparison with normal uninfected controls and further elevation of the methylation levels in GC tissues. Additionally, FOXD3 methylation was associated with shorter survival of GC patients. Gain- and loss-of-function assays showed that FOXD3 reduced GC cell proliferation and subcutaneous tumor growth in nude mice, and this was associated with increased cell apoptosis. The authors also showed that FOXD3 binds to the promoters and influences the transcriptional activity of the pro-apoptotic genes CYFIP2 and RARB, which show reduced transcriptional levels in gastric tumors 15.

Runx3

The role of RUNX3 as a tumor suppressor in GC is now well established 16. Lu et al. 17 showed (in 1056 samples from 854 patients) an increase in the proportion of RUNX3 promoter methylation along gastric carcinogenesis: 16% in chronic atrophic gastritis, 37% in IM, 42% in gastric adenoma, 55% in dysplasia, and 75% in GC tissues. This increase was best observed in H. pylori-positive patients, whereas in H. pylori-negative patients, RUNX3 methylation was only observed in severe dysplasia and cancer.

It has been suggested that GC promotion by the loss of RUNX3 may occur by enhancement of the Akt1-mediated signaling pathway 18. Lin et al. demonstrated that RUNX3 directly binds to the Akt1 promoter and represses Akt1 transcription. RUNX3-mediated Akt1 inhibition promotes GSK-3β activation and β-catenin degradation followed by cyclin D1 downregulation. The authors also demonstrated that cyclin D1 suppression had an important role in RUNX3-mediated cell cycle arrest and inhibition of cell proliferation.

The cellular consequences of RUNX3 loss of function were also addressed by Voon et al. 19. They showed that loss of RUNX3 (and P53) resulted in spontaneous and TGF-β-induced epithelial-mesenchymal transition (EMT) in a subpopulation of cells. This tumorigenic subpopulation expressed stem cell-like markers such as LGR5, OCT4, NANOG, and SOX9 and showed enhanced tumor sphere-initiation and colony-forming capacities. Taken together, their results suggest that RUNX3 plays a protective role in gastric epithelial cell differentiation against EMT-induced plasticity and tumorigenicity 19.

Following on their previous work suggesting that hypoxia silences RUNX3 by epigenetic histone regulation in GC 20, Lee et al. 21 explored the cross-talk between RUNX3 and HIF-1α under hypoxia. They showed that RUNX3 decreased the stability of HIF-1α and prevented HIF-1α-mediated angiogenesis in GC cells under hypoxic conditions. They further demonstrate that RUNX3 destabilized HIF-1α by direct interaction with the C-terminal transactivation domain of HIF-1α and by stimulating its ubiquitination through proline hydroxylation. Their data suggest that molecular strategies aimed at re-expressing RUNX3 might inhibit HIF-1α expression in GC angiogenesis 21. Following this line of thought, Lim et al. 22 developed cell-permeable forms of biologically active RUNX3 (CP-RUNX3) that were able to suppress cell-cycle progression, wound healing, and survival and to induce changes consistent with effects of RUNX3 on TGF-β signaling. CP-RUNX3 also suppressed the growth of subcutaneous human gastric tumor xenografts, suggesting a therapeutic potential for these molecules, especially when administered locally 22.

E-cadherin

The CDH1 gene, encoding E-cadherin, is now established as a tumor suppressor in GC 23. E-cadherin dysfunction may occur through several mechanisms, including CDH1 mutations, epigenetic silencing by promoter hypermethylation, loss of heterozygosity (LOH), transcriptional silencing by a variety of transcriptional repressors that target the CDH1 promoter, and microRNAs that regulate E-cadherin expression 23. A comprehensive analysis of the prevalence of CDH1 somatic alterations was published by Corso et al. 24 in a large series, comprising 246 patients with sporadic and familial GC (negative for CDH1 germline mutations) and including intestinal and diffuse histologic types. Overall, approximately 30% of the tumors had CDH1 alterations (20% epigenetic and 10% structural) that were present in all clinical settings and histotypes. The frequency of CDH1 alterations was similar in sporadic and familial gastric tumors, and patients with tumors harboring structural alterations showed the worst survival rate. Regarding histologic types, while intestinal tumors had similar frequencies of epigenetic and structural alterations, tumors of the diffuse type had more often epigenetic than structural alterations. The authors found that CDH1 alterations were not associated with specific patterns of E-cadherin expression, suggesting that other transcriptional/post-transcriptional regulatory mechanisms exist. In fact, Pinheiro et al. 25 have searched for transcriptional events arising from CDH1 intron 2 and discovered several new transcripts, including CDH1a, which was translated into a novel E-cadherin isoform. CDH1a was absent from the normal stomach and expressed de novo in GC cell lines. Functionally, CDH1a replaced canonical protein interactions and functions in an E-cadherin negative context. However, when co-expressed with canonical E-cadherin, CDH1a increased the expression of interferon-induced gene IFITM1 and IFI27, increased cell invasion, and angiogenesis, which were reverted upon CDH1a knockdown. Another alternative mechanism for E-cadherin loss of function in GC was described in the study by Carvalho et al. 26. The authors used microRNA microarray expression profiling and array-CGH and observed that miR-101 was significantly downregulated in GC in comparison with the normal gastric mucosa. This miR-101 downregulation was caused by (micro)deletions at miR-101 genomic loci and resulted in EZH2 overexpression and aberrant E-cadherin expression, preferentially in intestinal-type GC. It has also been proposed that Smad3 may regulate E-cadherin via transcriptional regulation of miR-200 family members 27, which in turn target the E-cadherin transcriptional repressor ZEB2 28.

Novel Putative Tumor Suppressors

Several novel putative tumor-suppressor genes in GC have been identified last year 29-40. The cytoplasmic polyadenylation element binding protein 1 (CPEB1) was identified in a screen for novel GC genes in a transgenic Drosophila model and was frequently silenced by methylation in GC cell lines and in primary tumors, especially of the diffuse type. Functionally, CPEB1 was shown to inhibit invasion as well as angiogenesis via downregulation of MMP14 and VEGFA 35.

The disintegrin-like metalloprotease with thrombospondin type 1 motif 9 (ADAMTS9) was shown to exert tumor-suppressor functions by inhibiting cell proliferation, subcutaneous tumor growth in nude mice, and angiogenesis, and by inducing apoptosis. Tumor inhibition by ADAMTS9 occurred by suppression of the oncogenic AKT/mTOR signaling pathway 36.

Several transcription factors/regulators were also proposed to act as tumor suppressors in GC. For example, the transcription factor paired box gene 5 (PAX5), the zinc-finger protein 545 (ZNF545), and the B-cell CLL/lymphoma 6 member B (BCL6B) were commonly silenced or downregulated by promoter hypermethylation in GC cell lines as well as in primary gastric tumors compared with the adjacent noncancer tissues 37, 39, 40. Gain- and loss-of-function assays showed that PAX5 inhibited tumor cell growth, arrested cell cycle, and induced apoptosis. Accordingly, gene expression profiling showed upregulation of the pro-apoptotic genes TP53 and BAX, and of the cell-cycle regulator CDKN1A, and downregulation of the anti-apoptotic gene BCL2. Functional assays also revealed that PAX5 might act as a suppressor of cell migration and invasion, through upregulation of metastasis suppressors MTSS1 and TIMP1 and downregulation of MET and MMP1. TP53 and MET were shown to be the targets of PAX5 that directly binds to the promoters of these genes 37.

Functionally, ZNF545 inhibited GC cell proliferation and induced apoptosis. Interestingly, in GC cells, ZNF545 was shown to be localized at the nucleolus and to act as a suppressor of rRNA transcription, by binding directly to the rDNA promoter and recruiting the co-repressor HP1β, and by diminishing the level of H3Lys4 trimethylation (H3K4me3), a promoter-specific histone modification associated with active transcription 39, 41.

Concerning BCL6B, its tumor-suppressor effects on GC cell inhibition and induction of apoptosis were attributed to upregulation of the proapoptosis genes TNF1RSF1A and CDKN1A, and of the active forms of caspases-3, -7, -8, -9, and PARP, to downregulation of the anti-apoptotic genes S100A4 and VEGFA, and to the induction of tumor-suppressor proteins ATM and p53 40. Clinically, promoter methylation of PAX5, ZNF545, BCL6B, and ADAMTS9 was shown to be associated with poor GC patient survival, and the authors of these publications suggest that methylation of these genes may serve as biomarkers for the prognosis of these patients 36, 37, 39, 40.