Introduction

Epigenetic regulations play critical roles in diverse functions of genome including those related with cancer. Among human epigenetic codes, histone methylation regulates specific gene expression patterns and is manifested by interplay between methyltransferases and demethylases acting on the same histone residue. For example, trimethylation of lysine-27 residue on histone H3 (H3K27me3) is strongly associated with gene silencing and tightly regulated by three enzymes, EZH2, KDM6A and KDM6B [1].

EZH2 (enhancer of zeste 2 polycomb repressive complex 2 subunit), a.k.a. KMT6A (lysine methyltransferase 6A), is a methyltransferase subunit of polycomb repressor complex 2 (PRC2). KDM6A (lysine demethylase 6A), a.k.a. UTX (ubiquitously transcribed X-chromosome tetratricopeptide repeat protein), and KDM6B (lysine demethylase 6B), a.k.a. JMJD3 (Jumonji domain containing protein 3), are demethylases.

All three enzymes have both tumor-promoting and -suppressing functions depending on cancer types [2,3,4,5,6,7], but genetic association of cancer risk has been reported only on EZH2, not KDM6A or KDM6B yet. A recent article surveyed 12 case–control studies where EZH2 single-nucleotide polymorphisms (SNPs) have been associated with susceptibilities to diverse cancer types including lung and breast cancers, digestive system cancers (gastric, liver, colorectal, esophageal, and oral cancers), and urogenital system cancers (prostate, urothelial, and bladder cancers) [8].

Gastric cancer, a.k.a. stomach cancer, is the fourth most common cause of cancer-related death in the world, and recently the most commonly diagnosed cancer in Korea [9]. EZH2 SNPs have been associated with susceptibility to gastric cancer [10] and primary tumor invasion depth and lymph node metastasis in gastric cancer [11]. In this study, not only is EZH2 reexamined, but also KDM6A and KDM6B are examined for gastric cancer risk association of their SNPs to investigate gene–gene interaction, a.k.a. epistasis, among the three functionally related genes.

Materials and methods

Subject genotyping

Gastric cancer patients and healthy controls were recruited at Hanyang University Guri Hospital, Chungnam National University Hospital, and Pusan National University Hospital, which participate in the National Biobank of Korea. Tag SNPs were selected for genotyping using Haploview 4.2 [12], based on their statistical power estimated using the PS Power and Sample Size Calculations [13]. Genomic DNA extracted from peripheral blood samples were genotyped using the Sequenom’s MassARRAY platform.

Statistical analyses

Control sample genotypes were analyzed for Hardy–Weinberg equilibria. Odds ratio (OR), 95% confidence interval (CI) and P value were calculated for each SNP association, and OR of interaction (ORint) was calculated for each intergenic SNP pair, using PLINK software [14]. The interaction is regarded synergistic if ORint > 1, and antagonistic if ORint < 1 [15].

Results

Cancer susceptibility association tests

A total of 23 tag SNPs were genotyped in 2349 Korean participants including 1100 gastric cancer patients and 1249 healthy controls, whose demographic characteristics are summarized in Table 1. Excluding four SNPs with control sample genotypes out of Hardy–Weinberg equilibria (P < 0.05/23 = 0.0022, a Bonferroni correction threshold for 23-SNP testing), 19 SNPs including four in KDM6A, six in KDM6B, and nine in EZH2 were tested for association with gastric cancer susceptibility using multivariate logistic regression with adjustment for age and gender (Table 2) because age distribution and gender ratio were different between the cases and controls (Table 1).

Table 1 Characteristics of the study participants
Full size table
Table 2 SNPs tested for association with gastric cancer risk
Full size table

Female and male subjects were analyzed together according to additive genetic models for EZH2 and KDM6B SNPs located on human chromosomes 7 and 17, respectively. However, for SNPs of KDM6A, which is located on X chromosome but escapes from female X-inactivation [16], female and male subjects were separately analyzed according to additive and allelic genetic models, respectively.

Among all documented SNPs located in the three gene loci, each spanning from 2 kb upstream of transcription start site to 2 kb downstream of polyadenylation site, only those having a sufficient statistical power (> 0.8) were tagged with r2 ≥ 0.8. The power was estimated for expected OR of 1.4 using the Han Chinese and Japanese data of the 1000 Genomes Project. Their minor allele frequencies (MAFs) were ≥ 0.057 for EZH2 and KDM6B SNPs (1100 cases versus 1249 controls), and ≥ 0.18 for KDM6A SNPs (337 female cases versus 546 female controls).

SNPs associated with cancer susceptibility

Gastric cancer risk is significantly associated with three SNPs (Table 2). KDM6A rs5952279 (P = 1.0 × 10−4), KDM6A rs144974719 (P = 2.4 × 10−4), and KDM6B rs78633955 (P = 1.9 × 10−3) pass a Bonferroni correction threshold, α = 0.05/19 = 2.6 × 10−3. Three additional SNPs, EZH2 rs67648693 (P = 0.0028), KDM6B rs11657063 (P = 0.0036), and EZH2 rs1061037 (P = 0.023) show marginal association. Thus, two KDM6A, two KDM6B, and two EZH2 SNPs are significantly or marginally associated with gastric cancer susceptibility in additive genetic models, supporting individual associations of the three genes with gastric cancer risk.

KDM6A rs5952279 showing the lowest P value in the additive model analyses has minor-allele protective association in a dominant (P = 2.0 × 10−8, OR 0.42) rather than recessive (P = 0.72) genetic model. Two-copy carriage of the risk-associated major allele has 2.9-fold higher risk than one-copy carriage. Additionally, this SNP shows association with diffuse-type (P = 2.6 × 10−6, OR 0.47) rather than intestinal-type (P = 0.075) gastric cancer risk.

The lowest-P KDM6B SNP rs78633955 fits to a dominant model (P = 0.0051, OR 1.35) better than a recessive model (P = 0.032). This SNP is not associated with diffuse (P = 0.75) or intestinal (P = 0.31) type risk. The lowest-P EZH2 SNP rs67648693 favors a recessive (P = 0.0070, OR 0.67) over a dominant model (P = 0.021, OR 0.80). Intestinal-type risk is most significantly associated with this SNP rs67648693 (P = 2.5 × 10−3, OR 0.76), while diffuse-type risk is most significantly associated with EZH2 rs73158295 (P = 1.9 × 10−8).

Synergistic triad epistasis

Interactions between two SNPs in separate genes were calculated to discover gene–gene interactions that contribute to cancer susceptibility. We tested 114 SNP pairs including 36 EZH2–KDM6A pairs (9 × 4), 24 KDM6A–KDM6B pairs (4 × 6), and 54 KDM6B–EZH2 pairs (6 × 9). Interactions of a KDM6A SNP (on X chromosome) with others were tested only with females.

OR values were recalculated for risk-associated versus nonrisk-associated alleles rather than for minor versus major alleles according to additive genetic models, and used in calculation of ORint = (OR for SNP pair) ÷ (OR for one SNP × OR for the other SNP). Now, new OR ≥ 1 for every SNP, but ORint for each SNP pair can be higher or lower than 1, respectively, indicating synergistic or antagonistic interaction, if not equal to 1 for null interaction [15].

Six SNP pairs show significant or marginal interaction after Bonferroni correction with α = 0.05/114 = 4.4 × 10−4 (Table 3). A significant interaction between EZH2 rs73158295 and KDM6B rs78633955 is synergistic (ORint = 1.7, P = 3.0 × 10−4) in gastric cancer risk. Four SNP pairs between EZH2 and KDM6A have marginal synergistic interaction (0.00066 ≤ P ≤ 0.046), with the lowest P value for synergistic interaction of EZH2 rs58579167 and KDM6A rs5952279 (ORint = 3.2). Additionally, a marginal KDM6A–KDM6B interaction is also synergistic as revealed between their respective SNPs, rs2230018 and rs78633955 (ORint = 1.9, P = 0.044). All these inter-SNP interactions among the three genes together form a synergistic triad epistasis network of ring-type topology (Fig. 1).

Table 3 Intergenic SNP interactions in gastric cancer risk
Full size table
Fig. 1
figure 1

Synergistic triad epistasis of EZH2, KDM6A, and KDM6B. The three pairwise interactions are all synergistic in gastric cancer risk and together form a ring-type network. Between EZH2 and KDM6A, only one exemplary SNP pair is shown among the four epistasis pairs listed in Table 3

Full size image

Discussion

We report here that the three H3K27me modifier genes, EZH2, KDM6A, and KDM6B, are individually associated with gastric cancer risk and synergistically interact with each other in conferring the risk. Therefore, super-synergism can be expected for carriage of all three risk-associated alleles that form a triad interaction network of ring-type topology where all links are synergistic (Fig. 1). Inclusion of this synergistic effect in weighted genetic risk score would improve risk prediction.

No cancer risk has been previously associated with KDM6A or KDM6B SNPs, but several cancer types with EZH2 SNPs [8]. Among them, gastric cancer risk association of EZH2 was reported from a Chinese study with Han people (311 cases and 425 controls), but their cancer-associated SNPs [10] are different from ours. The Chinese study reported association of a linkage disequilibrium block of rs12670401 and rs6464926 and another block of rs2072407, rs734005, and rs734004, but rs12670401 (P = 0.95) of the former block is not associated in this study with Koreans (Table 2). This discrepancy could be due to differences in ethnicity, gender ratio, or age distribution between the two study populations, among other differences.

Six SNPs in the three genes are associated with cancer risk, and thereby additional SNPs highly correlated with them would have the same association. KDM6B rs11657063 (genotyped) is perfectly correlated (r2 = 1.0) with rs3744252 and rs3840058 (ungenotyped) in the Han Chinese and Japanese populations of the 1000 Genomes Project. Both rs11657063 and rs3744252 overlap with promoter histone marks (H3K4me3 and H3K9ac), enhancer histone marks (H3K4me1 and H3K27ac), and DNase hypersensitivity regions according to the HaploReg v4.1 [17], possibly affecting KDM6B transcription.

At least one SNP in each gene is significantly associated with susceptibility to gastric cancer, diffuse type, or intestinal type (Table 2). The EZH2–KDM6B interaction is significant, but EZH2–KDM6A and KDM6B–KDM6A interactions are merely marginal (Table 3). It warrants replication studies with other populations to verify cancer risk association of the H3K27me modifier genes and their synergistic triad interaction.