Epigenetic and genetic variation in GATA5 is associated with gastric disease risk

Abstract

Gastric cancer incidence varies considerably among populations, even those with comparable rates of Helicobacter pylori infection. To test the hypothesis that genetic variation plays a role in gastric disease, we assessed the relationship between genotypes and gastric histopathology in a Colombian study population, using a genotyping array of immune-related single nucleotide polymorphisms (SNPs). Two synonymous SNPs (rs6061243 and rs6587239) were associated with progression of premalignant gastric lesions in a dominant-effects model after correction for multiple comparisons (p = 2.63E–07 and p = 7.97E–07, respectively); effect sizes were β = −0.863 and β = −0.815, respectively, where β is an estimate of effect on histopathology scores, which ranged from 1 (normal) to 5 (dysplasia). In our replication cohort, a second Colombian population, both SNPs were associated with histopathology when additively modeled (β = −0.256, 95 % CI = −0.47, −0.039; and β = −0.239, 95 % CI = −0.45, −0.024), and rs6587239 was significantly associated in a dominant-effects model (β = −0.330, 95 % CI = −0.66, 0.00). Because promoter methylation of GATA5 has previously been associated with gastric cancer, we also tested for the association of methylation status with more advanced histopathology scores in our samples and found a significant relationship (p = 0.001). A multivariate regression model revealed that the effects of both the promoter methylation and the exonic SNPs in GATA5 were independent. A SNP-by-methylation interaction term was also significant. This interaction between GATA5 variants and GATA5 promoter methylation indicates that the association of either factor with gastric disease progression is modified by the other.

Introduction

Gastric cancer is the third most common cause of cancer-related mortality worldwide, accounting for an estimated of 723,000 deaths annually (Ferlay et al. 2014). Lack of symptoms early in the course of disease often leads to late diagnoses and fatal outcomes. Most gastric cancers are adenocarcinomas, which may be classified into intestinal and diffuse subtypes. Intestinal-type gastric adenocarcinoma, the most common subtype in high-risk regions, is associated with a well-defined histopathological cascade: chronic superficial gastritis, atrophic gastritis, intestinal metaplasia, dysplasia and adenocarcinoma (Correa 1988). Typically, these lesions advance over decades, affording opportunity for early detection and interventions.

Most cases of gastric cancer can be attributed to infection with Helicobacter pylori, a class I carcinogen (IARC Working Group 1994). Gastric disease progression to noncardia gastric cancer depends in part on genetic variation in the infecting H. pylori strain. In some strains, insertion of a 40-kb cag pathogenicity island containing the cagA virulence factor causes a severe inflammatory response and increases the risk of gastric cancer (Blaser et al. 1995; Covacci et al. 1993). Variants of another H. pylori gene, vacA, also increase gastric cancer risk by inducing vacuole formation, stimulating epithelial-cell apoptosis, and suppressing host T cell responses to the bacterium (Atherton et al. 1995).

Genetic variation in the host also influences the clinical trajectory of gastric disease. Polymorphisms known to modulate levels of cytokines IL-1β, TNF-α, and IL-10 have been associated with elevated risk of non-cardia gastric cancer (El-Omar et al. 2000, 2003). Additionally, genome-wide association studies (GWAS) in Asian populations have identified associations between gastric cancer and variants in PSCA and MUC1 (Abnet et al. 2010; Sakamoto et al. 2008; Shi et al. 2011) and near PRKAA1 and PTGER4 (Shi et al. 2011; Song et al. 2013). The interaction of genetic variants in the host and pathogen also affects disease progression (Kodaman et al. 2014a, b).

In addition to host genetic factors, epigenetic variation, specifically changes in DNA methylation, have been associated with the development of gastric cancer. Such epigenetic changes often involve DNA promoter methylation and subsequent inactivation of tumor suppressor genes (The Cancer Genome Atlas Research Network 2014). Many changes in DNA methylation occur non-specifically, as a function of aging or environmental factors (Teschendorff et al. 2010). Some changes may occur in premalignant lesions, making methylation changes potentially useful as biomarkers of gastric cancer risk (Schneider et al. 2015; Watanabe et al. 2009).

In the present study, we tested for association between human genetic variation and gastric disease severity, using gastric biopsies from patients residing in Colombian regions with high and low risk for gastric cancer. Genes relevant to immunological function were examined. Identifying such variants may improve our understanding of gastric cancer etiology, guide therapy, and provide genetic markers of disease progression risk.

Materials and methods

Study participants

Adult patients presenting with dyspeptic symptoms that merited upper gastrointestinal endoscopy were recruited from two regions in the state of Nariño, Colombia: Tumaco on the Pacific coast and the Pasto region in the Andes Mountains that, respectively, have low and high incidence rates for gastric cancer (Bravo et al. 2002). Patients with chronic conditions such as diabetes, heart disease, or a prior gastrectomy were excluded, as were those who had been treated with H₂-receptor antagonists, proton pump inhibitors, or antimicrobials during the month prior to the endoscopy. Human subjects approvals were obtained from the Institutional Review Boards of the participating hospitals, the Universidad del Valle in Cali, Colombia, and Vanderbilt University. Informed consent was obtained from each subject. The discovery cohort included 130 patients, and 159 patients from an independent set comprised a replication cohort. Samples from 189 patients from both cohorts were assessed for percent DNA methylation (Methylation group).

Biopsies and histopathology scoring

Gastric mucosa biopsy samples were taken from three sites: the antrum (greater curvature, within 3 cm of the pylorus), incisura angularis (lesser curvature), and corpus (middle anterior wall). After staining with hematoxylin and eosin, tissue sections were evaluated independently by two pathologists blinded to sample origin. Diagnostic categories for histopathology scoring employed the updated Sydney system for gastritis (Dixon et al. 1996) and the Padova International Classification for Dysplasia (Rugge et al. 2000) as previously described (Kodaman et al. 2014a). Cases with discordant diagnoses were reviewed until a consensus was reached. Ordinal values for diagnostic categories were as follows: 1 = normal, 2 = non-atrophic gastritis (NAG), 3 = multifocal atrophic gastritis without intestinal metaplasia (MAG), 4 = intestinal metaplasia (IM), and 5 = dysplasia. The most severe lesion observed in all biopsies from an individual represented that person's final diagnosis in our study.

Helicobacter pylori culture and characterization

For the discovery cohort, virulence determinants cagA and vacA s were characterized as previously described (Kodaman et al. 2014a). In the replication cohort and methylation group, for 104 subjects for whom single colony cultures were not available, the vacA s region was characterized from biopsy DNA as previously described (Schneider et al. 2015).

Genotyping in the discovery cohort

Human DNA from the study participants was extracted from blood samples using Puregene kits (Qiagen, Valencia, CA, USA). DNA was genotyped using the Illumina platform Immunochip array, which assays 196,524 SNPs in genes primarily involved in immune response (Cortes and Brown 2011). If fewer than 95 % of samples could be genotyped for a given SNP, that SNP was removed from further analyses. Study participants for whom fewer than 90 % of SNPs could be genotyped were excluded from analyses. A priori power analyses suggested that a minor allele frequency threshold of 0.2 would allow us to detect a beta of 0.5 with power of 0.90, given our sample size (Table S1). SNPs in perfect linkage disequilibrium (LD) and those failing the Hardy–Weinberg test at the 0.01 level were also excluded; 39,802 markers passed quality control and were used in the final analysis (Table S2).

Characterizing human ancestry in the discovery cohort

Estimation of human ancestry was performed as previously described (Kodaman et al. 2014a). Briefly, all Immunochip SNPs in LD were removed, using an r² threshold >0.1 prior to analysis (performed in PLINK (Purcell et al. 2007)) and the remaining SNPs were used to estimate ancestry. Using an admixture model in STRUCTURE, (Falush et al. 2003; Pritchard et al. 2000) model probability was maximized using K = 3; the inferred ancestral groups matched the expected distribution of Amerindian, African, and European ancestry (Wang et al. 2007, 2008). To validate this assumption, we obtained genotype data for the European CEU and African YRI populations from Phase 2 of HapMap (International Hapmap Consortium 2005) the Spanish Iberian (IBS) population from the 1000 Genomes Project (Abecasis et al. 2012), and the H952 subset of the Karitania, Surui, and Colombian populations in the Human Genome Diversity Project (Cavalli-Sforza 2005). After merging these data with those from our own population, the final dataset contained 5947 SNPs from 514 individuals. The admixture model of STRUCTURE was run 10 times, with 50,000 iterations after a burn-in of 50,000 iterations. Replicate runs were collated in CLUMPP (Jakobsson and Rosenberg 2007) and means of individual ancestry were calculated.

Genotyping in the replication cohort

SNP analyses were performed using Pyrosequencing^®, as follows: 20–100 ng of DNA was amplified using forward and reverse primers (rs6061243 forward primer: biotin-CTTACCTGGGGGGCCATG; reverse primer: CTCTTCGCTTCCAGGATCCAC; sequencing primer: GCTCAGCAGCCACTT; rs6587239 forward primer: CTTCCGCTTCCGTGTCTG; reverse primer: biotin-GGAATGAGGGGGTGCTGT; sequencing primer: TCCGTGTCTGGATGC; rs1046587 forward primer: biotin-GGGAGTGGAGTGGAAGAAGA; reverse primer: CACTCGGCCCTCACTTTG; sequencing primer: GGCCCTCACTTTGCT). Biotinylated strands of PCR products were purified on streptavidin-sepharose beads. Sequencing primers were annealed, following the manufacturer's recommendations (Qiagen, Valencia, CA, USA). Assignment of alleles was performed using a Pyromark MD Pyrosequencing instrument, in accordance with the manufacturer's instructions. Assignment of alleles was confirmed by comparing the Immunochip array with Pyrosequencing results, using 36 samples that were analyzed by both methods. Concordance in the 36 samples was 100 % between the two methods. Alleles were designated ancestral if identical with those in chimpanzee, or derived if different.

GATA5 methylation analysis in the methylation group

We performed a GATA5 methylation analysis on samples from both cohorts. DNA (up to 2 ug) was bisulfite-modified using a DNA Methylation Direct kit (Zymo Research Corp., Irvine, CA). Levels of DNA methylation were quantitated (in 30–60 ng modified DNA per reaction) by Pyrosequencing^®. GATA5 Pyrosequencing primers were F: GGGATTAGTATAAATTTGGAGAAGTT; R: biotin-CCACCTAACCCTAACAAACCCTACT; Sequencing primer: GGTTAGTTAGTTTAGTTGTATT. This assay quantitates methylation across chr20:61,051,250 to 61,051,265 (hg19). The initial PCR cycling program consisted of 95 °C for 15 min, then 46 cycles of 94 °C for 30 s, 56 °C for 30 s, and 72 °C for 30 s, followed by a 10 min incubation at 72 °C. The initial PCR reagents included two units of Amplitaq Gold Hot Start DNA polymerase (Life Technologies, Carlsbad, CA) per reaction. The biotinylated strand of each PCR product was isolated from 15 μl of each PCR product, using streptavidin beads (GE Healthcare Biosciences, Piscataway, NJ, USA). Pyrosequencing reactions were performed using a PyroMark MD Pyrosequencing instrument, as recommended by the manufacturer (Qiagen, Valencia, CA, USA). Bisulfite-modified methylated HeLa DNA (New England Biolabs, Ipswich, MA, USA) was employed as a positive control with each experiment; bisulfite-modified pooled normal human blood DNA (Promega Corp., Madison, WI, USA) and no-template buffer were used as negative controls, at least one of which was used with each experiment. Results were reported as percentage of DNA molecules that are methylated.

Statistical analyses of the discovery and replication cohorts

In the discovery cohort, univariate linear regression analyses modeling histopathology scores were first performed with each available variable: participant age, region of origin, sex, Amerindian ancestry, European ancestry, African ancestry, and H. pylori virulence factors cagA and vacA s. Significant variables were used in the final model. Patient genotypes were assessed using both additive and dominant models. Recessive models would have been underpowered and were not performed. A Bonferroni correction was used to determine the appropriate level of statistical significance using the number of SNPs passing quality controls (prior to any analyses) as the number of independent tests performed (0.05/(39,802) = 1.26 × 10⁻⁶).

The SNPs significant in the discovery cohort were tested for association in the replication cohort using both additive and dominant models. Ancestry estimates could not be ascertained for the replication cohort because genome-wide data were not available; therefore, analyses were adjusted for region of origin as proxy for ancestry. Analyses were also adjusted for the vacA s virulence factor.

Genetic association analyses in the discovery and replication cohorts were performed using PLINK (Purcell et al. 2007) and STATA v11.1, (Statacorp, College Station, TX, USA), respectively. LD of significant variants was assessed using Haploview (Barrett et al. 2005). The qqman package in R (Turner 2014) was used to create QQ and Manhattan plots. A 50 kb region surrounding the SNP with the strongest association was plotted with Locus Zoom (Pruim et al. 2010). Variants in regions of interest (within 0.5 megabase) were imputed with IMPUTE2 (v2.3.1), using one multipopulation phased reference panel from the 1000 Genomes project (Abecasis et al. 2012; Howie et al. 2009). Linear regression on histopathology scores in the discovery cohort was also performed using imputed variants, adjusting for the same covariates as in the original analysis.

Statistical analysis of the methylation group

Linear regression was used to test the methylation level of the GATA5 promoter for association with gastric histopathology, using STATA v11.1. Histopathology was modeled in univariate analyses with participant age, region of origin, sex, vacA s status, dominant rs6587239 genotype, and GATA5 methylation as a continuous variable. Variables significant in univariate analyses were included in the final model. To assess whether (dominant) genotype or methylation had non-linear effects on histopathology, an interaction term (genotype-by-methylation) was, then added to the multivariate model. Similar analyses were performed using ordinal regression.

Results

Discovery cohort

The discovery cohort consisted of 130 patients (mean age 48.7 ± 7.0), of whom 78 lived in the low risk coastal region and 52 in the high-risk mountain region (Table 1). Human ancestry and H. pylori virulence factors were both significantly associated with histopathology score and advanced gastric disease in this cohort (Table 2). As expected, the presence of cagA in H. pylori was associated with higher histopathology scores (β = 0.55, p = 0.037), while vacA s2 alleles were associated with less severe lesions (β = −0.76, p = 0.001). In Colombia, continental human ancestry is known to be primarily African, Amerindian, and European (Kodaman et al. 2014a); in our study population, we found that human African ancestry was associated with less severe lesions (β = −0.71, p = 0.005) and European and Amerindian ancestry with more severe lesions (β = 1.65 and 0.65, p = 0.004 and 0.046, respectively). To adjust for all three ancestries in subsequent regression models, we included two of them, which by default defined the third. Neither region of origin, sex, nor age was independently associated with histopathology scores (Table 2). The independent variables in the final model were therefore genotype, African ancestry, Amerindian ancestry, and cagA/vacA s. The genomic inflation factor was 1.02 in these analyses, indicating that population stratification was adequately controlled (Figure S1).

Table 1.

Characteristics of study participants

Variable	Discovery cohort N = 130	Replication cohort N = 159	Methylation group N = 189
Sex: # malesa	86 (66 %)	100 (63 %)	132 (70 %)
Ageb	48.66 ± 6.95	50.02 ± 7.31	49.53 ± 7.27
Agec	47 (39, 64)	49 (31, 67)	48 (31, 67)
Region of origin: # from mountainsa	52 (40 %)	133 (84 %)	126 (67 %)
Histopathology scoreb	2.92 ± 0.92	3.41 ± 0.98	3.37 ± 0.97
GATA5 methylation: %	NA	NA	21.45 ± 8.37
cagA: # positivea	115 (89 %)	NA	NA
vacA: # s2 allelea	18 (14 %)	11 (7 %)	19 (10 %)

^aValues are expressed as counts (percent total)

^bValues are expressed as mean ± standard deviation

^cValues are expressed as median (min, max)

Table 2.

Association of gastric histopathology scores in the discovery cohort

Variable	Coefficient	95 % confidence interval	p value for difference in histopathology score
Sex (male)a	0.18	(–0.16, 0.52)	0.291
Region of origin (mountain)a	0.30	(–0.023, 0.62)	0.068
Agea	–0.0004	(–0.024, 0.023)	0.970
cagAb	0.55	(0.033, 1.06)	0.037
vacA s2c	–0.76	(–1.21, –0.31)	0.001
European ancestrya	1.65	(0.54, 2.75)	0.004
African ancestrya	–0.71	(–1.19, –0.22)	0.005
Amerindian ancestrya	0.65	(0.013, 1.29)	0.046

^a130 observations

^b129 observations

^c128 observations

In tests of dominant effects, two SNPs in the GATA5 gene (rs6061243 and rs6587239) were significantly associated with histopathology scores after correction for multiple testing (p = 2.63 × 10⁻⁷ and 7.97 × 10⁻⁷, respectively) (Table 3; Fig. 1). Ordinal regression modeling of histopathology scores produced similar associations (Table S3). Both rs6061243 (C/G) and rs6587239 (C/T) are synonymous exonic SNPs in strong linkage disequilibrium with each other (r² = 0.98), consistent with a single association at this locus (Figure S2A). These variants correspond to c.981G>C, Ser327Ser in the coding strand in exon 5 and c.852G>A, Lys284Lys in the coding strand in exon 4, respectively. In both cases, presence of the derived allele (G for rs6061243 and T for rs6587239) was associated with lower histopathology scores (β = −0.86 and β = −0.82, respectively, in the dominant model) where β estimates the change in histopathology score with presence of the minor allele. The effect sizes in additive-effects models, where β estimates the change in histopathology score with additional copies of the minor allele, were similar, but not as strong (β = −0.41, p = 4.92 × 10⁻⁴; and β = −0.40, p = 7.35 × 10⁻⁴, respectively).

Table 3.

Single nucleotide polymorphisms associating with gastric histopathology scores in the discovery cohort

SNP	Chr	Positiona	Minor allele	MAF	Model	n	β	p value	Gene
rs6061243b	20	61,040,453	G	0.446	Dom	126	–0.86	2.63 × 10−7	GATA5
rs6587239b	20	61,040,951	T	0.438	Dom	127	–0.82	7.97 × 10−7	GATA5
rs1046587	14	24,768,060	A	0.216	Dom	127	0.71	1.29 × 10−5	LTB4R
rs158250	5	102,406,779	C	0.215	Add	127	–0.57	4.10 × 10−5	near PAM
rs1046587	14	24,768,060	A	0.216	Add	127	0.51	6.51 × 10−5	LTB4R
rs464720	5	102,406,887	T	0.227	Add	127	–0.52	8.30 × 10−5	near PAM
rs464720	5	102,406,887	T	0.227	Dom	127	–0.61	8.54 × 10−5	near PAM
rs1113665	6	109,579,675	G	0.396	Dom	127	0.62	9.64 × 10−5	LOC100996634

Associations <1 × 10⁻⁴ are shown; analyses were adjusted for African and Amerindian ancestry and H. pylori virulence factors cagA and vacA s

MAF minor allele frequency

^aPosition in the hg19 assembly

^bPasses Bonferroni multiple testing adjustment

Fig. 1.

a From discovery cohort data, SNPs were imputed within a 50 kb region around GATA5, using a dominant model. The SNP with the most significant association, rs6061243, is shown as a purple square. Other colors represent linkage disequilibrium (measured with r²) between imputed variants and the index SNP. b Annotation of DNA methylation in the region shown is from the UCSC Genome Browser (hg19). c Positions of the regions assayed for DNA methylation in the methylation group are located within the black boxes. d Gastric histopathology scores versus genotypes of rs6587239, in tertiles of GATA5 promoter methylation. Dotted lines indicate mean histopathology scores for each tertile (color figure online)

Linear regression on histopathology scores in the discovery cohort was also performed using imputed variants, adjusting for the same covariates as in the original analysis (Table S4; Fig. 1). No additional imputed SNPs were identified as more significantly associated than rs6587239.

Four other SNPs were significant at p < 10⁻⁴ (Table 3). The minor allele (A) of rs1046587 in the 3′ UTR region of LTB4R was associated with higher histopathology scores (β = 0.71) in the dominant-effects model. An intergenic locus rs1113665 was also significant (p = 9.64 × 10⁻⁵). Two SNPs near PAM on chromosome 5 (rs158250 and rs464720) were significant at this threshold in tests using an additive genetic model (β = −0.57 and −0.52, respectively). The SNPs near PAM were in strong linkage disequilibrium (r² = 0.93) (Figure S2B).

Replication cohort

The replication cohort consisted of 159 patients (see Table 1 for demographic data). Distributions of sex and age did not differ significantly from those of the discovery cohort (p = 0.565 and 0.109, respectively), but the mean histopathology score was significantly higher (p = 1.98 × 10⁻⁵) (Tables 1, S5), and a greater percentage of individuals were residents of the mountain region (p = 1.58 × 10⁻¹⁴). The vacA s2 allele frequency among H. pylori samples was also greater in this cohort, although not statistically significant (p = 0.051). All models were subsequently adjusted for vacA s and region of origin, which were significant in univariate analyses of gastric histopathology scores. We replicated the association between rs6587239 and gastric histopathology scores in tests of both additive and dominant effects (β = −0.26, p = 0.021; β = −0.33, p = 0.050; respectively; Table 4), with the derived allele (T) being associated with lower histopathology scores. While rs6061243 was significantly associated with histopathology only in the additive-effects model (β = −0.24, p = 0.029), the direction of effect was consistent in the dominant model (β = −0.30, p = 0.076). Ordinal regression modeling of histopathology scores produced similar results (Table S6). The p value of the association of SNP rs1046587 on chromosome 14 (Figure S3) and its proximity to LTB4R, a leukotriene receptor, prompted us to test this SNP in the Replication cohort. However, it was not significant (p = 0.427).

Table 4.

Association of gastric histopathology scores with additive and dominant models of GATA5 SNPs in the Replication cohort

Independent variables	Minor allele	MAF	n	β	95 % CI	p value
Add rs6061243	G	0.460	149	–0.24	(–0.45, –0.024)	0.029
Region of origin				0.11	(–0.29, 0.50)	0.6
vacA s2				–0.63	(–1.082, –0.18)	0.007
Add rs6587239	T	0.457	150	–0.26	(–0.47, –0.039)	0.021
Region of origin				0.20	(–0.20, 0.59)	0.328
vacA s2				–0.62	(–1.08, –0.16)	0.009
Dom rs6061243	G	0.460	149	–0.30	(–0.62, 0.032)	0.076
Region of origin				0.12	(–0.28, 0.52)	0.564
vacA s2				–0.63	(–1.08, –0.17)	0.007
Dom rs6587239	T	0.457	150	–0.33	(–0.66, 0.00)	0.050
Region of origin				0.22	(–0.18, 0.62)	0.276
vacA s2				–0.62	(–1.08, –0.16)	0.009

Analyses were adjusted for region of origin and the H. pylori virulence factor vacA s

Add additive model, Dom dominant model, MAF minor allele frequency

Methylation group

Because promoter methylation in GATA5 has been previously associated with gastric cancer, we assessed whether methylation was associated with severity of gastric lesions in our study. We quantitated the levels of GATA5 promoter methylation in 189 individuals (Table 1), and found a significant positive association between the degree of methylation and histopathology scores in an unadjusted model (β = 0.028, p = 0.001) (Table 5). Homozygosity of the rs6587239-C allele was correlated with significantly higher methylation (Spearman ρ = −0.157, p = 0.031).

Table 5.

Variables affecting gastric histopathology scores in the methylation group

Independent variable(s)	β	95% CI	p value
Age	0.026	(0.006, 0.045)	0.009
Region of origin	0.40	(0.10, 0.69)	0.008
Dominant rs6587239	−0.45	(−0.76, −0.15)	0.004
GATA5 methylation	0.028	(0.012, 0.045)	0.001
Sex (male)	−0.28	(−0.58, 0.023)	0.070
vacA s	−0.59	(−0.96, −0.22)	0.002
Age	0.022	(0.005, 0.040)	0.013
Region of origin	0.18	(0.093, 0.46)	0.194
Dominant rs6587239	−1.20	(−2.07, −0.33)	0.007
GATA methylation	−0.012	(−0.044, 0.020)	0.463
vacA s2	−0.73	(−1.08, −0.39)	4.28 ×10−5
Interaction of Dom rs6587239 * GATA5 methylation	0.040	(0.003, 0.076)	0.033

n = 189

The top half of the Table shows independent covariates in univariate tests.

The lower half shows covariates in a single regression model.

Multivariate tests of genetic and epigenetic variation in GATA5

Given that SNPs in GATA5 and DNA methylation patterns in its promoter were both significantly associated with disease severity, we tested both variables in a single regression model. Region of origin, age, and vacA s were included as covariates, as they were all associated with histopathology scores in univariate tests (p = 0.008, 0.009, and 0.002, respectively). Associations with histopathology scores were significant for rs6587239 and GATA5 promoter methylation in the multivariate test (p = 0.039 and p = 0.018, respectively; Table S7), indicating independent effects.

To assess whether the effects of the derived allele of rs6587239 on histopathology score varied non-linearly with the level of methylation, we added a methylation-by-genotype interaction term to the above model. The interaction was significant (p = 0.033), as were age, vacA s, and rs6587239. However, methylation was not significant in the model that included the interaction term (Table 5). These results demonstrated that the combined effect of genotype and methylation on histopathology scores was non-linear; at low levels of methylation, individuals who were homozygous for the C allele at rs6587239 presented with more advanced lesions (>1 unit, on average, corresponding to progression from multifocal atrophic gastritis to intestinal metaplasia) compared to individuals who had at least one T allele. This effect on progression was attenuated at higher levels of methylation, where those with CC genotypes had scores less than a half unit more advanced, on average (Fig. 1d). Analyses of rs6061243 produced similar results to that of rs6587239 with the ancestral allele (C) associating with higher scores (Table S8). Ordinal regression analyses showed a similar interaction effect and loss of significance of methylation (Supplemental Table S9).

Validation of previous GWAS results in the discovery cohort

In our analyses of the discovery cohort, we replicated several genes previously reported to associate with intestinal gastric cancer in published GWAS (Shi et al. 2011; Song et al. 2013). Three intronic SNPs in ZBTB20, (rs4399857, rs6438214, and rs9860965) had p values below 0.05 when modeled both additively (p = 8.57 × 10⁻³, 0.0160 and 6.19 × 10⁻⁴, respectively) and dominantly (p = 6.97 × 10⁻³, 0.014, and 4.23 × 10⁻⁴, respectively) (Table S10). The SNPs rs4399857 and rs6438214 were in strong LD (r² = 0.94) with each other, but in weaker LD with rs9860965 (r² = 0.46 and 0.44, respectively) (Figure S2C). Additionally, rs7725052, a SNP in the previously associated 5p13.1 region that includes genes PTGER4 and PRKAA1 (Shi et al. 2011; Song et al. 2013), was significant in the dominant-effects model (p = 0.037). Another gene previously associated with gastric cancer, MUC1 (Saeki et al. 2013), was not well tagged by the SNPs in this analysis, but intronic SNPs in genes on either side of MUC1 (rs11264341 in TRIM46 and rs2075571 in THBS3), were significant in additive-effects models (p = 0.031 and 0.023, respectively); rs11264341 is ~10 kb upstream of MUC1 while rs2075571 is ~12 kb downstream, and they are in strong LD with each other (r² = 0.95) (Figure S2D). In a meta-analysis of gene association studies for various subtypes of gastric cancer in different populations, Mocellin et al. (2015) reported other genes in addition to those listed above. Concordant with their results, we found these variants significant in dominant models: rs7635274 near TGFBR2 (p = 0.027), rs7743193 near TNF (p = 0.039), and rs10882332 and rs4304697 near PCLE1 (p = 0.044 and 0.022, respectively). We were unable to replicate the association of advanced disease with rs2294008 in PSCA (Table S10) (Saeki et al. 2013; Sakamoto et al. 2008; Sala et al. 2012). We detected no significant associations of variants near MTX1, which is related to risk of diffuse gastric cancer; or to variants near CASP8 or to those near GSTP1, which was found significantly related to gastric cancer risk in Asian but not Caucasian populations (Mocellin et al. 2015).

Discussion

We observed and replicated a novel association between severity of gastric lesions and exonic SNPs in GATA5. Although the effects of these synonymous variants are unclear, it is possible that they induce differential gene expression through changes in transcript stability, editing, and binding affinity of protein to DNA (Fernandez-Calero et al. 2014). GATA5 is a gene involved in cell and tissue differentiation and gastrointestinal development. It encodes a lineage-restricted transcription factor, expressed in the stem cell compartment in the gut, and this transcription factor is upregulated as these cells undergo terminal differentiation (Gao et al. 1998; Laverriere et al. 1994). Importantly, GATA5 is expressed in gastric epithelial cells, as indicated by in situ hybridization and immunohistochemistry studies (Laverriere et al. 1994; Wen et al. 2010); data from the Genotype-Tissue Expression (GTEx) study identify stomach as one of the tissues with the highest expression of GATA5 among the 53 tested (Mele et al. 2015). Consistent with the role of promoting differentiation, GATA5 activates the promoter of pepsinogen A, a digestive enzyme produced by gastric chief cells (Sakamoto et al. 2000). Other genes regulated by GATA5 include trefoil factors 1 and 2 (TFF1, TFF2), and inhibin α (INHA) (Akiyama et al. 2003).

In its capacity as a transcription factor, GATA5 is highly pleiotropic, and thus, polymorphisms that modulate its expression may influence gastric disease in multiple ways. One function of possible relevance for gastric carcinogenesis is that GATA5 activates transcription of TFF1 and TFF2 (Akiyama et al. 2003). TFF1 in particular is a well-established tumor suppressor in gastric tissue: TFF1 knockout mice develop adenocarcinomas of the gastric antrum, and human gastric biopsies show a decrease in TFF1 expression as gastric lesions progress along the multi-step carcinogenic cascade (Soutto et al. 2011). This decrease in TFF1 expression is accompanied by an increase in NFκB activation, which promotes the inflammatory response to H. pylori (Soutto et al. 2015). Furthermore, TFF1 induces p53 protein expression by downregulating its negative regulator, miR-504 (Soutto et al. 2014).

In vitro evidence indicates that GATA5 acts as a tumor suppressor in gut and in other cell types. Over-expression of GATA5 reduces colony formation, cell proliferation, migration, invasion and anchorage-independent growth in colorectal cancer cells (Hellebrekers et al. 2009); decreased expression has been implicated in carcinogenesis in both gastric and renal cancer cell lines as well as in colorectal and invasive ductal breast cancer tissue specimens (Hellebrekers et al. 2009; Peters et al. 2014; Wang et al. 2014). Inactivation of the likely tumor suppressor GATA5 by promoter methylation was previously reported in gastric and colorectal cell lines, gastric cancers and dysplasias (Akiyama et al. 2003; Wang et al. 2011; Wen et al. 2010). Methylation of CpG sites within enhancer sequences near GATA5 alters a transcription factor USF1 binding site, downregulating GATA5 expression in the gut (Chen et al. 2012). Wen et al. (2010) found that GATA5 promoter methylation was associated with loss of GATA5 protein in human gastric adenocarcinomas. These observations are consistent with our methylation analyses in which increased methylation was associated with higher histopathology scores, as tissues progressed in the premalignant cascade. The mechanism by which GATA5 promoter methylation is associated with gastric disease progression may operate directly or indirectly via TFF1, TFF2 or other target genes.

A multivariate test adjusted for relevant covariates indicated that GATA5 promoter methylation and rs6587239 genotype CC were independently associated with higher histopathology scores. We then added a methylation-by-genotype interaction term, as a significant interaction would indicate that the effect of one variable was modified by the levels of the other. The interaction was significant, which may indicate that the association of the CC genotype with more advanced lesions is attenuated when cells with this GATA5 genotype become inactivated by promoter methylation. Alternatively, the relationship between SNP status and methylation may signify that one or both of our associating SNPs is a methylation quantitative trait locus (meQTL), a variant associated with quantitative differences in methylation levels (Kerkel et al. 2008). Indeed, the association between the CC genotype and increased methylation supports this possibility. Future studies examining meQTLs in stomach and other portions of the gastrointestinal tract would be of interest, especially because there is a relatively high incidence of methylation defects in tumors of the GI tract, as shown in Figure S4.6 of the Supplement to The Cancer Genome Atlas paper on gastric adenocarcinoma (The Cancer Genome Atlas Research Network 2014).

In conclusion, we showed genetic and epigenetic data that together provide compelling evidence to support GATA5 promoter methylation and exonic variants as risk factors in the development of gastric disease. The associations we presented for each of these data types were also affected by an interaction between them. This observed interaction between genotype and methylation status together with the negative correlation between genotype and methylation indicate a complex relationship between genotype and epigenetics that likely contributes to risk of gastric disease progression.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Cancer Institute, the National Center for Advancing Translational Sciences, the National Institutes of Health, or Vanderbilt University Medical Center.