A total of 242 GC cases and 294 controls were included in this study, and the clinical characteristics of the cases and controls are detailed in Table 1. Males accounted for 71.9% of the cases compared with 59.5% of the controls. The mean ± SD age (at diagnosis) of the patients was 54.9 ± 12.5 years (range 21–79 years), and that of the controls was 58.4 ± 16.4 years (range 14–94 years). The distribution of gender and mean age between cases and controls indicated significant difference, so all statistical analyses were subsequently adjusted by sex and age. Of 242 patients, the most common specified sites were in cardia or spanned over two anatomical regions of the stomach, accounting for 90.1% of cases. A total of 63 patients (26.1%) were classified as well or moderately differentiated GC, 132 (54.5%) poorly differentiated GC and 19 (7.8%) signet-ring cell carcinomas, while the remaining 28 patients (11.6%) were unclassified. The cases consisted of 68 (28.1%) in the clinical stage I or II groups and 160 (71.9%) in the clinical stage III or IV groups. Only 26 patients (10.7%) were present in the early gastric cancer group, in which the depth invasion is confined to lamina propria or submucosa, 89.2% of patients were found to have advanced GC including T2, T3 and T4. For gastric cancer, infection with Helicobacter pylori is the main etiologic factor. However, infection with Helicobacter pylori had not received enough attention in gastric cancer until several years ago in China. Among these collected samples, patients collected in 2009 were not checked for the H. pylori infections, however, patients collected in 2010 were examined for the H. pylori infection. This leads to about 66% percent of samples that have lost the data of H. pylori infection. Therefore, information regarding infection with Helicobacter pylori was not present.

Sixteen SNPs in DNMTs were chosen from the public single nucleotide polymorphism database dbSNP [25], i.e., rs2114724, rs2228611, rs2228612, rs8101866 and rs16999593 in DNMT1; rs11695471 and rs11254413 in DNMT2; rs1550117, rs11887120, rs13420827, rs13428812 and rs6733301 in DNMT3A; rs2424908, rs2424913 and rs6087990 in DNMT3B; and rs113593938 in DNMT3L. Table 2 shows detailed information regarding these 16 SNPs.

Among these 16 SNPs, rs6087990 and rs11695471 had call rates <80%, and one SNP (rs2228612) showed deviation from the Hardy–Weinberg equilibrium expectations in the control population (p > 0.05). In addition, three SNPs (rs2424913, rs6733301 and rs113593938) were non-polymorphic in this study, with minor allele frequencies (MAF) of less than 0.01. The major genotypes were TT, GG and CC for the three SNPs rs2424913, rs6733301 and rs113593938, respectively. The above-mentioned six SNPs were excluded from subsequent analyses. The remaining 10 SNPs—consisting of four in DNMT1, one in DNMT2, four in DNMT3A and one in DNMT3B—were successfully genotyped.

The MAFs of the 10 SNPs investigated are summarized in Table 3. The MAFs in our study showed greater similarities with Beijing and Japanese populations, and greater differences with European and African populations. Moreover, there was a significant difference between cases and controls in our study for rs11254413 in DNMT2 (p < 0.01). Logistic regression analysis, adjusted for age and sex, showed the A allele of SNP rs11254413 with an odds ratio (OR) of 0.65 (95% confidence interval (95% CI): 0.48–0.87) for GC. However, no statistically significant associations were found between any of the other SNPs and GC.

Five genetic models (co-dominant, dominant, recessive, over-dominant, and log-additive) were chosen to evaluate the association between each SNP and GC risk. Genotype frequencies of cases and controls as well as OR and p values for the best-fitting genetic model (over-dominant) and co-dominant model are shown in Table 4. Three other models (dominant, recessive, and log-additive) are presented in Table S1. Significant associations between three of the 10 genotyped SNPs (rs1699593, rs11254413, rs13420827) and GC were confirmed.

CT heterozygosity in rs1699593 in DNMT1 was shown to be a risk factor for GC (p = 0.05 for overdominant model of SNPs analyses) with OR = 1.45 (1.00–2.11). A protective effect against GC was identified for AG heterozygosity of rs11254413 in DNMT2 (p < 0.01 for overdominant model) with OR = 0.15 (0.08–0.27). In addition, there was a significant protective effect between GC heterozygosity in rs13420827 in DNMT3A and GC in the overdominant model (p = 0.034), with OR = 0.66 (0.45–0.97). No significant associations with GC susceptibility between cases and controls were found for other SNPs, including rs2114724, rs2228611, rs8101866 in DNMT1, rs1550117, rs11887120 and rs13428812 in DNMT3A, and rs2424908 in DNMT3B (see Table 4).

Four SNPs in DNMT1 and 4 SNPs in DNMT3A were successfully genotyped. Linkage disequilibria for these SNPs in the DNMT1 and DNMT3A genes are shown in Figure 1. The four SNPs in DNMT1 formed a block within 25 kb (chr: 10265248–10291181), whereas the four in DNMT3A did not (chr: 25453968–25565907). The D′ and r² values between SNPs are presented in Table S2. Due to the strong linkage in DNMT1, tagged SNPs were run in Haploview. The results indicated that rs8101866 and rs16999593 could act as tag SNPs for the four SNPs in DNMT1. The SNP rs8101866 had higher r² with rs2114724 and rs2228611 (0.976 and 0.97, respectively).

The haplotypes of DNMT1 and DNMT3A using global haplotype score tests for association with GC risk are shown in Table 5. Of the three individual haplotypes estimated in DNMT1, the most frequent haplotype CGTT was seen in 53.9% of the controls and 51.0% of the patients. The most frequent DNMT3A haplotype in patients and controls was CCAG (25.5% and 27.6%, respectively). There were no significant differences in haplotype frequency distribution between GC patients and controls for any of the haplotypes examined.

Three of the 16 SNPs examined showed significant associations with GC, i.e., a missense mutation of rs16999593 in DNMT1 (His97Arg substitution), a missense mutation of rs11254413 in DNMT2 (His101Tyr substitution) and a synonymous mutation of rs13420827 in DNMT3A. Previous studies indicated associations of rs16999593 with risk of infiltrating ductal breast carcinoma (IDC) [26] in the northern Chinese population (Heilongjiang Province), and of rs13420827 with ovarian cancer [27,28] in the American population (Mayo and North Carolina). However, no association between gastric cancer and DNMT polymorphisms among southern Chinese patients has been reported. To our knowledge, this is the first report of the association of these three gene variants with GC susceptibility.

Abnormal DNA methylation is thought to be a major early event in tumor development. For example, tumor suppressor genes are silenced by DNA methylation on CpG islands around their promoter regions in cancer cells, and thus the overall level of DNA methylation is higher in cancer cells than in normal cells [29,30]. SNPs of DNMTs are important indicators of genetic susceptibility to cancer development. Therefore, genetic polymorphism assays have been used to investigate the aetiology of malignant diseases [31].

In our case-control study, four of five SNPs in the DNMT1 gene were successfully genotyped (rs2228612 failed due to Hardy–Weinberg equilibrium deviation) and only rs16999593 was significantly associated with GC. CT heterozygosity in rs16999593 was associated with an increased risk of GC; C→T variation results in an Arg to His amino acid substitution in DNMT1. DNMT1 is located at the replication fork and methylates newly biosynthesised DNA strands immediately after the round of replication [32]. Therefore, the variation leading to missense mutation in rs1699593 may affect the structure and function of DNMT1. Inactivation of DNMT1 has been shown to lead to mitotic catastrophe in human cancer cells [33]. Previous studies indicated that the level of DNMT1 expression is significantly elevated in GC, and suggested that DNMT1 can be used as a predictive biomarker and potential therapeutic target for chemotherapy in GC [34–36]. Increased DNMT1 protein expression is significantly correlated with poorer tumor differentiation and frequent DNA hypermethylation of multiple CpG islands in GC [37]. However, Khatami et al. [21] found no associations between GC susceptibility and DNMT1 polymorphisms, including rs721186, rs13784, rs2228611 and rs11488, in an Iranian population. Based on the haplotype analysis in the present study, we could not confirm associations between four SNPs in DNMT1 and GC risk, while a previous study indicated that the CA haplotype (rs16999593 T, rs2228611 G) may be a risk factor for breast IDC [26]. With regard to the pivotal role of methylation in the DNMT1 promoter in tumourigenesis, we hypothesized that CT heterozygosity in rs16999593 could contribute to genetic susceptibility to GC.

The AG genotype of rs11254413 in DNMT2 had a lower frequency in patients than in controls. The SNP rs11254413 leads to missense mutation of His to Tyr, and may also alter the function of the protein encoded by the DNMT2 gene. DNMT2 is primarily localized to the cytoplasm [38] and is responsible for the methylation of aspartic acid tRNA and possesses residual DNA (cytosine-C5) methyltransferase activity [39]. Although it is the most widely conserved DNMT protein, little is known about DNMT2 compared to other DNMT proteins. El-Maarri et al. reported that rs11254413 was associated with hypermethylation at a small subset of loci, but it was unclear whether this SNP per se played a causative role [18]. The results of the present study indicate a potentially protective effect of AG heterozygosity of rs11254413 on GC risk. Therefore, these results further emphasize the importance of defining DNMT2-modulated cellular pathways in future studies.

DNMT3A and DNMT3B are localized on chromosomes 2p23 and 20q11.2, respectively. As de novo DNA methyltransferases, they play important roles in the generation of aberrant methylation in carcinogenesis. In gastric cancers, CpG island hypermethylation of tumour suppressor genes are more frequently inactivated by aberrant DNA methylation [40–42]. The polymorphism of rs1550117 in DNMT3A has been reported to be associated with susceptibility to GC among the Chinese population in Jiangsu province [22]. In the present study, we detected a possible association of rs13420827 but not rs1550117 with GC susceptibility. Thus, the potential association between missense mutation of rs13420827 and the change in DNMT3A activity remains uncertain. Kelemen et al. reported that rs13420827 was not associated with ovarian carcinoma [27].

With regard to DNMT3B polymorphisms, previous studies focused on the transcriptional start site within the promoter region, especially the SNPs rs2424913 and rs1569686. Hu et al. [23] reported associations of rs2424913 and rs1569686 with GC in a Chinese population. In the present study, the frequency of allele C on rs2424913 was lower than 0.01, which was consistent with reports from Japan and Hebei, China [24,43]. In addition, we found no significant differences in rs1569686 and rs2424908, although these three SNPs have been reported to be related to other types of tumor, such as oesophageal cancer, breast cancer, colorectal cancer, etc. [44–46].

DNMT3L is an inactive accessory factor among the DNMTs. It cannot directly methylate DNA as it lacks the amino acid residues necessary for methyltransferase activity [47]. However, DNMT3L acts as a regulator of de novo methyltransferases by stimulating the activities of DNMT3A and DNMT3B through protein–protein interactions [48]. In addition, DNMT3L is important for embryonic development, imprinting and X-chromosome inactivation, which may lead to the genome-wide nuclear reprogramming observed in tumor cells [49]. With regard to rs113593938 in DNMT3L, the C→T allele leads to the missense mutation of Arg to Gln. While the locus lacked gene polymorphisms among the southern Chinese population in the present study, the distribution of CC was 100%, in accordance with the data from dbSNP in a cohort population [25]. Recent reports have suggested that hypomethylation in the DNMT3L promoter may be a novel marker of cervical cancer [50]. However, few association studies of DNMT3L polymorphisms have been reported, with the exception of that between rs7354779 and human intelligence [51].

All subjects included in the present case-control study were recruited from the outpatient and inpatient clinics of the First Affiliated Hospital of Nanchang University, Nanchang, Jiangxi Province, China, from 2009 to 2010. Patients selected for the study were those with histologically-confirmed GC. Controls were randomly selected cancer-free individuals. The collection of samples and their utilization for research purposes were approved by the Southern Medical University Ethics Committee, and all subjects provided written informed consent prior to enrolment.

Peripheral blood samples were drawn from participants at the First Affiliated Hospital of Nanchang University. The samples were delivered in the frozen state by express mail to the School of Biotechnology, Southern Medical University, and stored at −70 °C until DNA extraction. Genomic DNA was extracted using a commercial blood DNA kit (TIANamp Genomic DNA Purification Kit; Tiangen Biotech, Beijing, China), according to the manufacturer’s instructions, and stored at −70 °C until use.

The inclusion criteria of candidate SNPs selection are SNPs among DNMT genes (DNMT1, DNMT2, DNMT3A, DNMT3B and DNMT3L) and previous literatures [18,21–24] have been reported the association with cancers, especially gastric cancer; SNPs data from the Han Chinese Population included in the HapMap, were also used to as referred [52]. Furthermore, SNPs failed in assay design were excluded. A total of 16 frequency-validated SNPs were selected for genotyping.

These 16 SNPs were genotyped using the SEQUENOM MassARRAY matrix-assisted laser desorption ionisation-time of flight mass spectrometry platform (Sequenom, San Diego, CA, USA). Primers for multiplex PCR and extended reactions were designed using proprietary software (Assay Designer, version 3.1; Sequenom: San Diego, CA, USA, 2006) provided by Sequenom Inc. In accordance with the manufacturer’s instructions, the first step was to amplify the genomic sequence containing the SNP by a standard PCR protocol, which would produce amplicons 80–120 bp in length. Excess PCR primers and unincorporated dNTPs were neutralized by incubation with shrimp alkaline phosphatase (SAP). Subsequently, the single-base extension (SBE) reaction was performed on the genomic amplification product using iPLEX enzyme and mass-modified terminators. The process was designed to generate SNP-specified DNA products of different lengths with predictable masses. The products of the iPLEX reaction were desalted and transferred onto a SpectroCHIP by the MassARRAY nanodispenser. The SpectroCHIP was then analyzed by the MassARRAY Analyzer Compact.

The genotypes were examined separately by a detailed quality control (QC) process. The SNP genotype threshold was set at 80%. SNPs not in Hardy–Weinberg equilibrium (p < 0.05) in the control samples were excluded from all analyses. SNPs with MAF ≤ 0.01 were also excluded from the subsequent analyses.

The results are expressed as proportions or means and standard deviation (SD). Differences in genotypes and alleles between GC patients and healthy controls were assessed using the χ² test. The associations between GC and individual DNMT SNPs were estimated by logistic regression with adjustment for age and sex. Statistical analyses and haplotyping were performed using the web-based tool SNPstats [53]. Five genetic models (co-dominant, dominant, recessive, over-dominant, and log-additive) were chosen to evaluate the association between each SNP and GC risk. Two alternative models (co-dominant and overdominant) were fitted for each outcome, adjusting for age and gender as covariates. The linkage disequilibrium (LD) and pairwise LD coefficients in DNMT1 and DNMT3A were implemented with Haploview 4.2 (Daly Lab: Cambridge, MA, USA, 2008). All comparisons were two tailed, and p < 0.05 was considered to indicate statistical significance.