Genetic Association of Plasma Homocysteine Levels with Gastric Cancer Risk: A Two-Sample Mendelian Randomization Study

Gastric cancer is the fifth most frequently diagnosed malignancies worldwide with over 1,000,000 new cases and 783,000 deaths estimated in 2018 (1). In China, gastric cancer is the second incident cancer and the second leading cause of cancer death with an estimated 679,100 new gastric cancer cases and an estimate of 498,000 deaths of gastric cancer in 2015 (2). Gastric cancer is a multifactorial disease, and both environmental and genetic factors play a role in its etiology. Helicobacter pylori infection is the most important cause of gastric cancer. Moreover, tobacco smoking, low consumption of fresh fruits and vegetables, and high intake of pickled and smoked foods are established risk factors for gastric cancer (3).

Homocysteine (Hcy), a key metabolite at the intersection of the methylation, remethylation, and transsulfuration pathways, is intrinsically related to cellular methylation status (4). Hcy can be recycled into methionine with the aid of vitamin B₁₂, folic acid, or trimethylglycine or converted into cysteine with vitamin B₆ as the cofactor (5). Therefore, genetic defects of the pathway above and nutritional deficiencies of folate, vitamin B₆, and vitamin B₁₂ can lead to elevation of Hcy concentrations. High plasma homocysteine level (PHL) has been recognized as an independent risk factor for cardiovascular diseases (6). However, the relationship is still controversial for gastric cancer. Several observational studies indicated a positive association between high PHL and increased gastric cancer risk (7, 8), whereas others found no statistically significant relationship (9–11). One limitation of these studies is that it is difficult to interpret the results with confounding and reverse causation.

Mendelian randomization (MR), an established epidemiological approach, provides an opportunity for causal inference within the framework of observational research design (12). It utilizes instrumental variables (IV) such as genetic variants that proxy for environmental, social, or behavioral factors to assess the causality of an observed association between a given exposure and an outcome (such as PHL and gastric cancer; ref. 13). This inference relies on the natural, random assortment of genetic variants during meiosis yielding a random distribution of genotypes in a population (14). Thus, MR is often robust to the issues of confounding and reverse causation inherent in observational epidemiologic studies. Recent genome-wide association studies (GWAS) have identified multiple loci associated with PHL (15–17), enabling investigation of a potentially causal role of PHL in gastric cancer risk using the MR approach.

In this study, we included both individual and summary data from 7,004 subjects of Chinese Han descent to perform a two-sample MR analysis (18) to investigate the causal relationship between PHL and risk of gastric cancer. A weighted genetic risk score (wGRS) incorporating 15 SNPs taken from a meta-analysis of GWAS on plasma homocysteine concentrations was employed as a proxy of PHL.

Epidemiologic and genetic data were derived from 2,631 gastric cancer cases and 4,373 controls from three published GWAS (Nanjing-GWAS, Beijing-GWAS, and NCI-GWAS; refs. 19–21). The characteristics of the three datasets are summarized in Supplementary Table S1. The cases and cancer-free controls of each study were unrelated individuals of Chinese Han descent, and there was no overlap of participants between these studies. For Nanjing GWAS and Beijing GWAS, individuals were recruited from two separate case–control studies conducted in Nanjing (565 cases and 1,162 controls) and Beijing (468 cases and 1,123 controls), and all cases were histopathologically confirmed nonardia gastric cancer. For NCI GWAS, subjects were from Shanxi (1,368 cases and 1,650 controls) and Linxian (257 cases and 450 controls), and the cases contained both cardia and noncardia gastric cancer. We accessed the NCI GWAS data from the database of Genotype and Phenotypes (dbGaP, http://www.ncbi.nlm.nih.gov/gap) with the study accession number phs000361.v1.p1. Written informed consent was given by all participants, and studies were approved by the relevant institutional review boards.

Nanjing/Beijing study used Affymetrix Genome-Wide Human SNP Array (V.6.0) for genotyping while the NCI study applied Illumina 660W-Quad microarray. Samples were excluded on the basis of excess autosomal heterozygosity, gender discrepancy, and outlier with identical-by-state clustering analysis. SNPs were excluded on the basis of call rate (<95%), minor allele frequency (MAF < 0.01), and departure from Hardy–Weinberg equilibrium (P-value < 1 × 10⁻⁶). Imputation was performed separately for the Nanjing/Beijing study and the NCI study using SHAPEIT (V.2) and Impute2 (V.2.2.2) software. All populations from the 1,000 Genomes Project Phase 3 were taken as the reference set (22). Qualified SNPs were restricted to those with MAF >1% and overall IMPUTE2 INFO score >0.4.

A recent meta-analysis of GWAS including a total of 44,147 individuals identified 18 SNPs associated with PHL at genome-wide significance level (P < 5 × 10⁻⁸; ref. 15). For each GWAS-identified locus, a representative SNP with the lowest P-value in the original GWAS was selected (linkage disequilibrium r² < 0.01, based on 1,000 Genome Phase 3 data; ref. 23). As a result, the remaining 15 SNPs were from independent loci.

A wGRS was constructed as a genetic proxy for PHL by adding up the dosages for PHL-raising alleles across the 15 variants in each individual following the formula: |${\rm{wGRS}} = \sum\nolimits_{i = 1}^{15} {{\beta _i}{\rm{SN}}{{\rm{P}}_i}}$|⁠, where |{\beta _i}$| is the beta coefficient of the |i$|th SNP for PHL from the published GWAS and SNP_i is the imputed dosage of the individual variant (recoded as 0, 1 and 2, according to the number of PHL-increasing allele). Associations of PHL-related SNPs and wGRS with risk of gastric cancer were estimated using logistic regression model adjusted for age, sex, and study site, as well as the first principal component. We also categorized the PHL wGRS into four groups according to its quartile distribution in controls. The association between the PHL wGRS (as a continuous variable scaled per SD and as a categorical variable taken the bottom quarter as reference) and gastric cancer risk was assessed using logistic regression with potential confounders adjusted. We also performed stratified analyses based on sex, age, and study site.

In addition, we also applied an MR inverse-variance weighted (IVW) method (24) to estimate the causal effect using summary statistics from gastric cancer GWAS. IVW regression implemented in the MendelianRandomization package (0.3.0; ref. 25) in R was performed to assess the combined association of the 15 PHL-related variants with gastric cancer risk. The Wald estimator (26) and delta method (27) were used to calculate the causal effect and SE, respectively. MR Egger regression and weighted median regression were also performed to detect potential pleiotropy of the instruments, which were described elsewhere (28, 29). Briefly, MR-Egger is a modified form of standard IVW approach, using a weighted regression with an unconstrained intercept to relax the assumption that the effects of the IVs on the outcome are entirely mediated via the exposure. A significant, nonzero intercept implies directional bias among the genetic instruments. Weighted median, which provides valid estimates when up to 50% of the statistical weight comes from valid IVs, orders, and weights the MR estimates by the inverse of their variance. We also excluded four SNPs (rs1047891, rs1801222, rs234709, and rs838133) associated with other traits (P < 5 × 10⁻⁸) in GWAS Catalog to eliminate the potential pleiotropy. Analyses were conducted using PLINK (v 1.90) and R (v.3.3.2). Two-sided P values less than 0.05 were considered statistically significant.

All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of the 1964 and later versions.

Informed consent to be included in the study, or the equivalent, was obtained from all patients.

A total of 2,631 gastric cancer cases and 4,373 cancer-free controls of Chinese Han descent were included in the current analyses, comprising 1,006 cases and 2,273 controls from the Nanjing/Beijing study, and 1,625 cases and 2,100 controls from the NCI study. Selected characteristics of cases and controls included in the final analyses are summarized in Supplementary Table S1.

We constructed the instruments based on 15 LD-independent SNPs that achieved genome-wide significance for PHL. All of these SNPs were genotyped or imputed, and the INFO values of imputed SNPs were more than 0.4 in all datasets. The information of these 15 SNPs were shown in Table 1.

Using wGRS with the 15 SNPs as IVs, per SD increase in PHL (μmol/L) was associated with a 7% increase in odds of gastric cancer [OR = 1.07; 95% confidence interval (CI), 1.01–1.12; P = 0.011; Table 2]. Compared with individuals in the bottom quartile of the PHL wGRS, those in the top quartile had an 19% (95% CI, 1.04–1.37) increased gastric cancer risk (P_trend = 7.08 × 10⁻³; Table 3).

In addition, we also estimated the potential causal effect of PHL on gastric cancer using MR IVW with summary statistics from gastric cancer GWAS. As shown in Table 2, similar positive association of PHL with gastric cancer risk (OR = 1.09 per SD increase; 95% CI, 1.01–1.18; P = 0.034) was observed in comparison with that using wGRS.

Subgroup analyses by sex showed different effect size for men (OR = 1.03; 95% CI, 0.97–1.09; P = 0.379) and for women [(OR = 1.17; 95% CI, 1.07–1.29; P = 0.001); P_{heterogeneity} = 0.023]. This result suggests that high PHL may have a stronger causal effect on gastric cancer risk in women than men. Stratification analyses by age group and study site did not show significant heterogeneity between different strata (Table 4).

The MR-Egger method suggested that no pleiotropy was present (the intercept was centered at 0.000 (95% CI, −0.014–0.015; P = 0.994; Table 2) and the MR-Egger slope was consistent with previous estimates. Results from the weighted median analyses were also consistent with the main findings, indicating a low probability that pleiotropy influenced the results (Fig. 1). In addition, we excluded four SNPs that were associated with other traits (P < 5 × 10⁻⁸) in GWAS Catalog and the results remained consistent (Table 2).

Our study provides evidence that genetically elevated PHL is causally associated with an increased risk of gastric cancer, where per SD increase in PHL conferred a 7% increase in the odds of gastric cancer in Chinese Han descent. Multiple MR techniques and sensitivity analyses demonstrated consistent results, suggesting the robustness of our findings. The results from the summary data were in agreement with those from individual-level genotype data. The findings hint that elevated PHL may have a causal role in the etiology of gastric cancer. In addition, stratified analysis suggested sex-specific associations between PHL and gastric cancer risk.

We previously reported that PHL was positively associated with gastric cancer risk in an observational study (7). Using both individual and summary data of the identified SNPs associated with PHL, we further confirmed the causal relationship in an MR approach. Although some studies reported a null association (9–11), the positive association results from MR analysis may offer more robust evidence to evaluate the causal role of PHL in gastric cancer etiology, which was free from confounding and reverse causation of traditional observational epidemiological study designs. By applying the two-sample MR approach, we were able to increase statistical power by extracting data from large sample size GWAS for PHL (n = 44,147) and gastric cancer (n = 2.631 cases and 4.373 controls).

Although our MR analyses indicate a causal relationship between PHL and gastric cancer risk, the underlying mechanism remains to be fully understood. Several possible explanations have been proposed to offer some mechanistic insights. First, homocysteine can disrupt methionine cycle and change cytosine methylation in CpG islands of DNA resulting in the repression of tumor suppressor genes and activation of proto-oncogenes, which may promote cancer development (30). PHL was found to modulate the expression of tumor suppressor genes RASSF1 and BRCA1 and therefore played a role in the breast cancer initiation (31). Moreover, inflammatory remodeling of gastrointestinal tract due to high PHL increases production of reactive oxygen species (32), which can cause several disorders including carcinogenesis in excessive accumulation (33). Finally, homocysteine was reported to contribute to perturbations in the endoplasmic reticulum protein folding machinery via altered cytosolic redox metabolism and it was tested in models of gastrointestinal tract cancer, including gastric cancer (34).

In this study, we observed a significant association of genetically elevated PHL with gastric cancer risk in women, but not in men. Sex-specific relationships were also reported between PHL and risk of cardiovascular disease (35), bipolar disorder (36), and nonalcoholic fatty liver disease (37). A recent study suggested the PHL-related risk of ischemic stroke prone to be significant in women, even though the PHL were higher in men than in women (38). Zhong and colleagues found that elevated PHL could be an independent risk factor for prognosis of acute ischemic stroke only in women (39). PHL are usually higher in men than in women, and it mainly attributes to differences of estradiol and homocysteine metabolism pathway (40). Of interest, GWAS also reported sex-specific genetic effects on PHL. In African Americans, the CPS1 locus was associated with PHL only in women (41). The SNPs rs18011131 of MTHFR and rs838892 of SCARB1 were also associated with PHL for women only (42). Nevertheless, further studies are warranted to clarify the mechanism of sex-specific relationship between elevated PHL and gastric cancer risk.

The major strength of this study was the availability of individual-level data in gastric cancer cases and controls, allowing us to perform analysis to control the effect of potential confounders. Furthermore, we adjusted for principal components, which accounted for potential confounding by population stratification. There are several limitations to this study. First, the potentially residual pleiotropy could not be fully tested while multiple sensitivity analyses were applied to evaluate pleiotropy. However, consistency across these approaches, and the fact that the MR-Egger intercept was centered at the origin, suggest that our results were not substantially influenced by pleiotropy. Second, this study only includes individuals of Chinese ancestry, which means our results may not apply to other races. However, this also avoids the potential confounding by population stratification. Third, we have used all PHL-related SNPs reported by the meta-analysis from different populations to make the causal inference. Some of these SNPs may be not appropriate instrument variables for Chinese, but so far there was no GWAS on PHL in Chinese population.

In summary, our findings provide evidence supporting a causal role for elevated PHL in gastric cancer etiology. Further investigations are warranted to uncover the underlying biological mechanisms for the associations observed in this study.

No potential conflicts of interest were disclosed.

Conception and design: T. Wang, J. Dai, Y. Ding, G. Jin

Development of methodology: T. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Ren, C. Yan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Wang, C. Ren, J. Ni, H. Ding, Q. Qi

Writing, review, and/or revision of the manuscript: T. Wang, H. Ding, Q. Qi, Y. Ding, G. Jin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Deng, G. Li

Study supervision: Y. Ding, G. Jin

The authors thank all the participants of the GWAS from Nanjing/Beijing and the NCI studies. This work was supported by grants from the National Natural Science Foundation of China (81872702 and 81521004); National Major Research and Development Program (2016YFC1302703); Key Research and Development Program of Jiangsu Province (BE2019698); Key Grant of Natural Science Foundation of Jiangsu Higher Education Institutions (15KJA330002); Top-notch Academic Programs Project of Jiangsu Higher Education Institutions (PPZY2015A067); Jiangsu Province "333" project (BRA2018057); and Priority Academic Program for the Development of Jiangsu Higher Education Institutions (Public Health and Preventive Medicine).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.