Genome-wide search identified DNA methylation sites that regulate the metabolome

Abstract

Background: Identifying DNA methylation sites that regulate the metabolome is important for several purposes. In this study, publicly available GWAS data were integrated to find methylation sites that impact metabolome through a discovery and replication scheme and by using Mendelian randomization. Results: The outcome of analyses revealed 107 methylation sites associated with 84 metabolites at the genome-wide significance level (p<5e^-8) at both the discovery and replication stages. A large percentage of the observed associations (85%) were with lipids, significantly higher than expected (p = 0.0003). A number of CpG (methylation) sites showed specificity e.g., cg20133200 within PFKP was associated with glucose only and cg10760299 within GATM impacted the level of creatinine; in contrast, there were sites associated with numerous metabolites e.g., cg20102877 on the 2p23.3 region was associated with 39 metabolites. Integrating transcriptome data enabled identifying genes (N = 82) mediating the impact of methylation sites on the metabolome and cardiometabolic traits. For example, PABPC4 mediated the impact of cg15123755-HDL on type-2 diabetes. KCNK7 mediated the impact of cg21033440-lipids on hypertension. POC5, ILRUN, FDFT1, and NEIL2 mediated the impact of CpG sites on obesity through metabolic pathways. Conclusion: This study provides a catalog of DNA methylation sites that regulate the metabolome for downstream applications.

Keywords: DNA methylation; Mendelian randomization; biomarker; epigenomics; metabolimics; metabolite.

FIGURE 1

Summary of the analyses performed in the current study to generate the results. Through a discovery and a replication stage, CpG site-metabolite pairs that shared at least a SNP and showed causality (please see the methods section for details) were identified. Integrating eQTL data provided the possibility to investigate the intermediary genes. Finally, by integrating the identified biomarkers with GWAS data for cardimetabolic traits, a search was conducted to identify CpG sites that impact these traits through metabolic pathways and to investigate the underlying molecular mechanisms.

FIGURE 2

Distribution of the identified CpG sites across the genome and their associations with metabolites. X-axis represents physical positions of CpG sites (based on the hg19), y-axis indicates the number of metabolites associated with each site. A total of 107 CpG sites associated with 84 metabolites were identified. Most methylation sites were associated with more than one metabolite. CpG sites with the highest number of associated metabolites are annotated by their underlying genes (according to the ANNOVAR software). CpG sites on adjacent chromosomes are colored differently to aid viewing.

FIGURE 3

Overview of the genes identified in this study and their relations with metabolites. The analyses revealed 82 genes associated with 84 metabolites. To better visualize the findings, the metabolites were classified into four categories (amino acids, proteins, lipids and carbohydrates). 40 of the genes were specifically associated with lipids. Several genes were associated with multiple categories of metabolites. Notably, KRTCAP3 and NRBP1 on chromosome 2p23.3 region, that were associated with all 4 categories of metabolites. Detailed information on the nature of association of the identified genes with metabolites and underlying CpG sites are provided in the Supplementary Table S3.

FIGURE 4

The mechanism whereby cg15123755 site impacts the risk of T2D. Higher methylation at cg15123755 site lowers the risk of T2D by changing the levels of PABPC4 and HDL. As cg15123755 site becomes methylated the expression of PABPC4 increases, this leads to higher level of HDL and consequently lowers the risk of T2D. Plots provide a graphical display of Mendelian randomization results. Each point on a plot represents a SNP. The x-value of the SNP is its effect size on the exposure, and the horizontal error bar represents the standard error around the effect size. The y-value of the SNP is its effect size on the outcome, and the vertical error bar represents the standard error around the effect size. The dashed line represents the line of best fit (i.e., a line with an intercept of 0 and slope of the effect size (β) from the Mendelian randomization test). A positive slope (+β) indicates as the level of exposure increases the level of outcome increases as well, whereas a negative slope (-β) indicates a negative association.

FIGURE 5

cg20102877 methylation site impacts the risk of T2D by changing the expression of NRBP1 and KRTCAP3. (A) These plots illustrate the co-localization of eQTLs for NRBP1, KRTCAP3 with mQTLs for cg20102877. mQTLs for cg20102877 showed congruence with eQTLs for NRBP1 but incongruence with eQTLs for KRTCAP3. This indicates higher methylation at cg20102877 increases the expression of NRBP1 but lowers the level of KRTCAP3 (please see Supplementary Table S3 for statistical evidence). Each point represents a SNP. The color of the point indicates its association with cg20102877 in reverse logarithmic scale. i.e., −log10(P_mQTL). The x-coordinate indicates the physical position of SNP in 2p23.3 region; whereas, the y-coordinate indicates the association of SNP with gene expression in reverse logarithmic scale. The plots were generated with the eQTpLot R package. (B) Mendelian randomization plots indicate change in expression of these genes consequently impacts the risk of T2D. Higher expression of NRBP1 decreases (-β) the risk of T2D; whereas, higher expression of KRTCAP3 increases (+β) the risk of T2D. Additional information on a Mendelian randomization plot is provided in the Figure 4 description.

FIGURE 6

Higher methylation at cg00908766 site increases the risk of CAD through PSRC1-lipids path (A) Regional association plots for mQTLs of cg00908766, and eQTLs of PSRC1 overlap. (B) MR analysis revealed, as cg00908766 becomes hypermethylated, the expression of PSRC1 decreases. Additional information on a Mendelian randomization plot is provided in the Figure 4 description. (C) Lower expression of PSRC1 contributes to higher levels of lipids and consequently higher risk of CAD.

FIGURE 7

Genes that mediate the impact of methylation sites on obesity through lipid pathways. I identified methylation sites that contributed to obesity by changing the levels of several metabolites. By integrating eQTL data, the underlying genes were identified as POC5, ILRUN, FDFT1 and NEIL2. Higher expression of FDFT1, NEIL2, and ILRUN was associated with higher risk of obesity; however, higher expression of POC5 protected against obesity. Additional information on a Mendelian randomization plot is provided in the Figure 4 description.