Global DNA hypomethylation is mostly seen in GC (19–21), even at the early steps of carcinogenesis (22,23). An enhanced hypomethylation was associated with a more invasive and advanced stage type of GC (23). Previous studies have shown that global DNA hypomethylation accumulates with patient age and is associated with copy number alterations in gastrointestinal cancers, possibly due to gradual acquisition of aberrance during DNA methylation and correlates to genomic damage (23). LINE-1 is frequently hypo-methylated in GC (24) and is associated with poor prognosis (25). Moreover, LINE-1 hypomethylation in GC patients is significantly correlated with HP infection (24). Hypomethylation of Alu elements, a member of the short interspersed nuclear element (SINE) family of repetitive elements, was also observed in GC (26). Alu and Satα hypomethylation is induced in gastric mucosae by H. pylori infection during gastric carcinogenesis (21). LOI is another example of an epigenetic alteration related to aberrant hypomethylation (21). LOI of the insulin-like growth factor-2 gene (IGF2) was shown to be associated with increased GC risks (21,27).

Regions of lower-density methylation near CpG islands (CGIs), defined as 'shores', exhibit great variation in methylation, including hypomethylation and hypermethylation, across diverse types of cancers (28). Currently, a number of hypomethylated tumor-promoting genes and hypermethylated tumor-suppressor genes (TSG) have also been found in GC and are associated with oncogene positive transcriptional regulation during a variety of cellular processes (Table I). The DNA methylation of chromatin-modifying enzymes (CMEs) that cause changes in chromatin structure can affect diverse pathways involved in multiple aspects of GC development and progression. For example, SMARCA5 and MGMT (O-6-methylguanine-DNA methyltransferase), SWI/SNF related, are downregulated in GC as a consequence of its promoter methylation (54,55). The methylation HLTF, encoding a member of the SWI/SNF family, has been reported in 50% of cases of GC (56). DNA methylation biomarkers of certain genes that are abnormally repressed in GC are appealing in normal gastric epithelium or premalignant lesions as well as in other body samples (e.g. stool and plasma) early in GC development. Alterations in DNA methylation can also influence treatment response in GC. For example, methylation of SULF2 and BMP4 have been linked to chemotherapeutic responses (57,58).

GC-associated alterations in DNA methylation were widespread and tend to localize at CGIs (59). Patients with widespread gene hypermethylation tend to have poor overall survival (59). However, the methylation profile differs between the intestinal and diffuse types of GC (60). The epithelial cadherin gene CDH1, a tumor suppressor gene which is downregulated in gastric tumors, is hypermethylated more frequently in the diffuse type of GC than in the intestinal type (61,62). Zouridis et al (59) showed that patients with CIMP-positive GC tend to be younger, with less-differentiated tumors, and their cases are associated with diffuse histology and tend to exhibit worse survival outcomes. One potent cause is that genes methylated by CIMP in GC overlap with those corresponding to PRC2-targets in embryonic stem cells, suggesting that it has been present since normal, early development. The strong relationship between CIMP and H. pylori, EBV, and MSI has been emphasized in a meta-analysis, but CIMP cannot be used as a prognostic marker for GC (63).

Notably, the increased expression of DNMT3A in GC is significantly higher than that of DNMT1 and DNMT3B (64,65). A recent study demonstrated that the poor overall survival rate of GC patients is associated with elevated DNMT3A expression, but not with increased expression of DNMT1 or DNMT3B (66). DNMT3A contributes to the dysregulation of the cell cycle by repressing p18^INK4C in a DNA methylation-dependent manner (67). DNA methyltransferase genes have been shown to be mutated in certain cancers. For example, the DNMT3A gene is mutated in acute myelogenous leukemia, myeloproliferative disease and myelodysplastic syndrome (68). Zang et al (69) also observed that GC had mutations in DNMT3A (1/15; 6.7%). In addition, 23.5% allelic loss of DNMT3A was observed in GC (70). Increment of functional polymorphism of DNMT1, DNMT3A and DNMT3B were found in gastric neoplasm and have been found to be associated with development and progression of GC (65,71–73). One meta-analysis showed that DNMT1 rs16999593 and DNMT3A rs1550117 could result in GC and that DNMT3B rs1569686 may be a protective factor against gastric carcinogenesis (73). Furthermore, DNMT1, DNMT3A and DNMT3B proteins are downregulated through overexpression of miR-200b and miR-200c, resulting in the global DNA hypomethylation in GC cell lines (74). In addition, H. pylori infection could increase DNMT activity via upregulation of the epidermal growth factor (EGF) and its receptor or via the release of inflammatory mediators, such as NO (75,76). In particular, overexpression of DNMT1 and DNMT3b was found to be associated with EBV infection in GC (77,78).

Studies conducted on histone methylation of GC have focused mainly on H3 and H4 (56,79). Studies have reported that certain histone methylation markers (H3K4me3, H3K9me3 and H3K27me3) are positively correlated with clinicopathological characteristics in GC, including tumor stage and survival (80–83). As a sequence, specific histone methylation can result in dysregulation of many genes with important roles in GC (Table II). For instance, EZH2 mediates H3K27 trimethylation to maintain transcriptional silencing. EZH2 knockdown represses cell growth, cell proliferation (RUX3 and ANXA6) (90,91), invasion and migration (E-cadherin and ArgBP2) (92,93), and induces cell cycle arrest in GC cells (p53, p21, p14 and p16) (94,95). In addition, EZH2 promotes the activation of Wnt and MAPK signaling through downregulation of CXXC4 expression, figuring an epigenetic mechanism of wnt signaling activation in GC cells (96,97).

Histone methylation is processed by lysine methyltransferases. A number of HMTs and histone demethylases (HDTs) have been found to mediate the addition and removal of methyl groups from different lysine residues on histones. Indeed, methylation at different lysine residues on histones has been shown to display differential functions (69). A current model suggests that methylated histones are recognized by chromatin effector molecules (readers), leading to the recruitment of other molecules to alter the chromatin and/or transcription states (106). In particular, histone methylation reader proteins such as WD40 repeats, chromo- and bromodomain proteins, and the PHD finger domain were shown to recruit HMTs to their target sites. During the past decade, several HMTs and HDTs have been identified in humans (Fig. 2). All HMTs except KMT4 (DOT1L), contain the SET-domain, which provides interaction and recognition sites for lysine substrates and cofactors for general catalysis. SET-domain-containing protein, a class of HMTs, has been regarded as an important factor in carcinogenesis (107). HDTs include amine oxidase-like (AOL) domain-containing demethylases and Jumonji C (JmjC) domain-containing demethylases (108,109). These HMTs and HDTs have been shown to methylate histones incorporated in chromatin, free histones and non-histone proteins (110).

Beyond alteration of the histone code, HMT interacts with various molecules including DNMT1. For example, EZH2 can recruit and regulate other epigenetic silencing enzymes, such as DNMTs, by influencing the binding capacity between DNMTs and the promoter of target genes (111). DNMT1-mediated promoter methylation is indispensably maintained by EZH2-mediated methylation (112,113). LSD family members contain a SWIRM domain and an AOL domain. The SWIRM domain is present in many chromatin-interacting enzymes and it can interact with DNA. Several studies have shown that the expression of several genes in GC can be regulated by DNA methylation and histone modification simultaneously (83,114,115). For example, gene expression of PRDM5 (PR domain containing 5), a member of the kruppel-like zinc finger family, is downregulated via DNA methylation and H3K27 trimethylation, alleviating the cell-growth-suppressive effect of PRDM5 (116).

The biological and physiological significance of non-histone lysine methylation in human tumorigenesis has been recently explored (117,118). For example, SET and MYND domain-containing protein 2 (SMYD2) was identified as a lysine methyltransferase for K370 of p53 (119), K860 and K810 of Rb (120,121) and K528 of PARP1 (122). SMYD2-dependent methylation of Rb at K810 promotes the cell cycle progression of cancer cells (121). Moreover, methylation of K528 on PARP1 enhances its poly(ADP-ribose) activity in cancer cells (122). Mazur and colleagues (123) found that SMYD3 methylated the lysine 260 of MAP3K2 gene, leading to the activation of the signaling of the Ras/Raf/MEK/ERK in the development of pancreatic ductal adenocarcinoma and lung adenocarcinoma.

As mentioned above, it has been well established that many genes are hypermethylated in GC (Table I). Two DNA methylation inhibitors, azacytidine (5-azacitidine) and decitabine (5-aza-deoxycytidine) have been established for the treatment of myeloid malignancies (145,146). Recently, more and more DNMT inhibitors, categorized as nucleoside analogs (decitabine, zebularine, SGI-110 and tetrahydrouridine) and non-nucleoside compounds (SGI-1027, procainamide, flavonoids, RG108 and their derivatives), are actively being explored in the clinical and preclinical trials as novel treatments for cancer (Fig. 1) (147).

Azacytidine and decitabine are nucleoside analogs of cytosine that cannot accept a methyl donor at the 5′ position of the pyrimidine ring and depletes cellular DNMT1. Intracellularly, azacytidine can convert to decitabine and subsequently be incorporated into DNA (148). Results showed that azacitidine inhibited the proliferation and decreased the level of DNA methylation in GC cell lines (149). Decitabine treatment resulted in growth suppression and reduced the levels of DNMT3A and DNMT3B accompanied with the demethylation of P16^INK4A gene (140). CIMP-positive GC lines appeared to exhibit significant reductions in proliferation after treatment of DNMT inhibitors such as decitabine compared to non-CIMP lines (59). The greatest challenge associated with decitabine is the compound's instability, which limits its efficacy in treating solid tumors (150).

A recent experimental study showed that the treatment of DNMT inhibitors alone may be inadequate to reactivate gene expression (151). Neither azacytidine nor decitabine has been established as a single-agent therapy for GC in clinical settings. However, there is evidence that DNMT inhibitors can increase chemosensitization when combined with traditional chemotherapy because the reversal of gene methylation can reactivate the expression of genes involved in resistance to chemotherapeutic drugs (152). For example, the use of decitabine and 5-FU in combination showed reactive expression of TFAP2E in GC by demethylating activity and enhancing the sensitivity of GC cells to 5-FU (153). Furthermore, a phase 1 trial of azacytidine combined to EOX (epirubicin, oxaliplatin and capecitabine) showed that this combination was well-tolerated with significant clinical and epigenetic responses (67%) in patients with locally advanced gastric and esophageal adenocarcinoma (139). More prospective randomized studies will be needed to further investigate whether this combination is superior to EOX.

Compared to other DNMTi, zebularine (2(1H)-pyrimidinone riboside) is a novel DNMT inhibitor with high oral bioavailability, slight toxicity and high stability (154). Several preclinical studies have shown that zebularine can form a tight covalent complex with DNMTs and so reverse the hypermethylation of TSGs in cancer cell lines (155). Treatment with zebularine effectively inhibits GC cells proliferation by inducing cell death, causing apoptosis in a dose-dependent manner. Moreover, zebularine depletes expression of DNMT protein with re-expression of epigenetically silenced genes, such as p16 (142). These results indicated that zebularine is a promising drug for GC therapy, but further exploration is needed.

In contrast to DNMTi, the agents targeting histone methylation is still at a primitive level (156). In 2005, the first study identified the effect of chaetocin, an inhibitor of HMT SUV39H1 (157). Subsequently, more and more inhibitors of various HMTs and HDTs, such as DOT1L (EPZ004777), EZH2 (EPZ-6438 and GSK126, EI1) and G9a (BIX-01294), were investigated as antitumor drugs (Fig. 1). Several inhibitors targeting DOT1L, EZH2 and LSD1, have entered phase 1/2 human clinical trials to assess the safety and maximum tolerated dose (Table IV). For example, EPZ-5676, the most advanced inhibitor for DOT1L that specifically catalyzes the mono-, di- and tri-methylation of H3K79, is currently in clinical trials for MLL-rearranged leukemia (NCT02141828 and NCT01684150). Major clinical studies that have been conducted so far are summarized in Table IV. Here, we summarize some common research targets in cancer, especially in GC.

The first widely used EZH2 inhibitor, 3-deazaneplanocin A (DZNep), is a cyclopentanyl analog of 3-deazaneplanocine that potently interferes with S-adenosyl-l-homocysteine hydrolase (SAH), which increases cellular SAH levels, repressing the activity of SAH-dependent HMT (158). It can inhibit PRC2 and remove H3K27me3 markers in various cancer types (159). Upon exposure to DZNep, EZH2 is depleted and ubiquitination of wild-type p53 protein is inhibited, resulting in p53 stabilization and activation of downstream p53 pathways involved in apoptosis, cell cycle arrest and senescence in GC cell lines (Table III) (144). However, DZNep has a short half-life and is both non-specific and toxic in animal models (160). GSK126 is an AdoMet-competitive chemical compound targeting Y641-and A677-mutated EZH2 for the treatment of diffuse large B-cell lymphoma (DLBCL) (161). A recent study has shown that GSK126 suppressed tumor migration and angiogenesis via downregulation of VEGF-A expression in GC cell lines and rodent animals (162).

The subsequently developed small-molecule EZH2 inhibitor EPZ005687, has shown dose-dependent inhibition of H3K27me3 in EZH2-wild-type and Y641- and A677-mutant lymphoma cells as well as in cell lines of other cancer types, including breast and prostate cancer. In June 2013, a phase 1/2 clinical trial of EPZ-6438, with better oral bioavailability than EPZ005687 (163), was explored in patients with advanced solid tumors or B cell lymphomas (NCT01897571). EI1, a third SAM-competitive inhibitor, inhibits both wild-type and mutant EZH2 and alters H3K27me2 and H3K27me3 levels in EZH2-mutant DLBCL cells in a SMARCB1-mutant rhabdoid tumor cell line (164). EZH2 silencing promoted growth arrest by low concentrations of doxorubicin in p53 mutant gastric cancer cells (165). The combination of EZH2 inhibitors and doxorubicin may be a potential novel approach to GC treatment.

LSD1, which demethylates both H3K4 and H3K9 residues, is overexpressed in GCs (Table II) (109). It may be a promising target for GC therapy. Tranylcypromine (TCP), an inhibitor of FAD-dependent monoamine oxidases (MAO), was the first small molecule used to target LSD1 because of the close homology between LSD1 and the MAO (166,167). Recent reports have highlighted the potential efficacy of TCP alone or combination with all-trans-retinoic acid (ATRA) in acute myeloid leukemia (AML) (Table IV). After that, TCP derivatives such as GSK2879552 and ORY-1001 were initiated for patients with relapsed/refractory small-cell lung cancer and AML in clinical trials (Table IV). Although no reports have shown whether these inhibitors can be used in GC, LSD1 inhibitors are worthy of further research into GC.