Epigenetics meets metabolomics: an epigenome-wide association study with blood serum metabolic traits

Abstract

Previously, we reported strong influences of genetic variants on metabolic phenotypes, some of them with clinical relevance. Here, we hypothesize that DNA methylation may have an important and potentially independent effect on human metabolism. To test this hypothesis, we conducted what is to the best of our knowledge the first epigenome-wide association study (EWAS) between DNA methylation and metabolic traits (metabotypes) in human blood. We assess 649 blood metabolic traits from 1814 participants of the Kooperative Gesundheitsforschung in der Region Augsburg (KORA) population study for association with methylation of 457 004 CpG sites, determined on the Infinium HumanMethylation450 BeadChip platform. Using the EWAS approach, we identified two types of methylome-metabotype associations. One type is driven by an underlying genetic effect; the other type is independent of genetic variation and potentially driven by common environmental and life-style-dependent factors. We report eight CpG loci at genome-wide significance that have a genetic variant as confounder (P = 3.9 × 10(-20) to 2.0 × 10(-108), r(2) = 0.036 to 0.221). Seven loci display CpG site-specific associations to metabotypes, but do not exhibit any underlying genetic signals (P = 9.2 × 10(-14) to 2.7 × 10(-27), r(2) = 0.008 to 0.107). We further identify several groups of CpG loci that associate with a same metabotype, such as 4-vinylphenol sulfate and 4-androsten-3-beta,17-beta-diol disulfate. In these cases, the association between CpG-methylation and metabotype is likely the result of a common external environmental factor, including smoking. Our study shows that analysis of EWAS with large numbers of metabolic traits in large population cohorts are, in principle, feasible. Taken together, our data suggest that DNA methylation plays an important role in regulating human metabolism.

Figure 1.

Schematic view of processes that link genetic variance and CpG–methylation to metabolic phenotypes. Possible feedback reactions are depicted by dashed lines, such as transcription activity leaving traces on the DNA by CpG–methylation, alosteric inactivation of enzymatic reactions or transcription regulation by metabolite-mRNA binding. Other potential regulatory and feedback mechanisms, involving for instance microRNA silencing and histone modifications, may exist but are not depicted here.

Figure 2.

Study design.

Figure 3.

Manhattan plots of CpG–metabotype associations without (top) and including (bottom) three SNPs into the model to account confounding genetic factors. Associations with P-values < 10⁻¹³ are indicated by vertical lines. Associations with P-values < 10⁻²⁵ are indicated by red dots. Manhattan plots comparing these CpG–metabotype associations to previously published SNP–metabotype associations are provided as Supplementary Material, Figure S1.

Figure 4.

Association between genotype, CpG–methylation and metabolic phenotype at the ACADM locus. (A) Scatterplot of b-values at cg10523679 and hexanoylcarnitine, colored by the genotype of SNP rs12134854; (B) correlation between methylation of cg10523679 determined by EpiTYPER and by the Infinium HumanMethylation450 BeadChip for selected samples (r² = 0.954); (C) as in (A), but for cg10523679 methylation determined on a subset of samples using the EpiTYPER system (fragment 4, which contains cg10523679); (D) boxplots of hexanoylcarnitine concentrations as a function of rs12134854 genotype; (E) methylation of cg10523679 determined using the Infinium HumanMethylation450 BeadChip as a function of the rs12134854 genotype. This figure shows that there is a strong three-way association between genotype, CpG methylation, and hexanoylcarnitine concentrations at the ACADM locus. Note that hexanoylcarnitine is essentially a substrate of the ACADM enzyme, rs12134854 is in linkage equilibrium of the ACADM gene, and cg10523679 is located in the promoter region of the ACADM gene.

Figure 5.

Possible scenarios that may result in an observed CpG–metabotype association induced by a confounding genetic variant or by an external environmental factor.