A genome-wide metabolic QTL analysis in Europeans implicates two loci shaped by recent positive selection

Abstract

We have performed a metabolite quantitative trait locus (mQTL) study of the (1)H nuclear magnetic resonance spectroscopy ((1)H NMR) metabolome in humans, building on recent targeted knowledge of genetic drivers of metabolic regulation. Urine and plasma samples were collected from two cohorts of individuals of European descent, with one cohort comprised of female twins donating samples longitudinally. Sample metabolite concentrations were quantified by (1)H NMR and tested for association with genome-wide single-nucleotide polymorphisms (SNPs). Four metabolites' concentrations exhibited significant, replicable association with SNP variation (8.6×10(-11)<p<2.8×10(-23)). Three of these-trimethylamine, 3-amino-isobutyrate, and an N-acetylated compound-were measured in urine. The other-dimethylamine-was measured in plasma. Trimethylamine and dimethylamine mapped to a single genetic region (hence we report a total of three implicated genomic regions). Two of the three hit regions lie within haplotype blocks (at 2p13.1 and 10q24.2) that carry the genetic signature of strong, recent, positive selection in European populations. Genes NAT8 and PYROXD2, both with relatively uncharacterized functional roles, are good candidates for mediating the corresponding mQTL associations. The study's longitudinal twin design allowed detailed variance-components analysis of the sources of population variation in metabolite levels. The mQTLs explained 40%-64% of biological population variation in the corresponding metabolites' concentrations. These effect sizes are stronger than those reported in a recent, targeted mQTL study of metabolites in serum using the targeted-metabolomics Biocrates platform. By re-analysing our plasma samples using the Biocrates platform, we replicated the mQTL findings of the previous study and discovered a previously uncharacterized yet substantial familial component of variation in metabolite levels in addition to the heritability contribution from the corresponding mQTL effects.

Figure 1. Hit region for N-ACu.

Top: location of genes, with rectangles denoting the position of exons. Middle: log-transformed p-values () for the test of association of the metabolite's concentration with each SNP in the region. Bottom: LD between each pair of SNPs in the region, with the colour scale for superimposed.

Figure 2. Hit region for TMAu.

Top: location of genes, with rectangles denoting the position of exons. Middle: log-transformed p-values () for the test of association of the metabolite's concentration with each SNP in the region. Bottom: LD between each pair of SNPs in the region, with the colour scale for superimposed.

Figure 3. Relative metabolite concentrations against genotypes at their most significantly associated mQTL SNP.

Each point corresponds to a study participant's mQTL genotype and corresponding metabolite concentration. Metabolite identifiers are labelled at top. Genotypic classes for each mQTL are shown on the horizontal axis (random horizontal variation within each genotypic class is introduced for clarity); dbSNP identifiers are labelled at bottom. At each metabolite peak, the transformed data vector shown in the plot is , where denotes the vector of normalized peak heights at that peak (prior to any logarithmic transformation, as described in Materials and Methods ). So, the transformation maps to zero the lowest observed concentration of each metabolite, and log₂(fold change) can be visually quantified relative to this baseline level. In particular, the maximum observed log₂(fold change) in a metabolite's concentration is easily accessible from the plot. Within-participant replicate observations (biological and technical) were averaged on log₂ scale.

Figure 4. Biological variance decomposition for metabolic traits driven by mQTLs featuring in the current paper.

Results from the current paper's replication of on the Biocrates platform are shown in the bottom section of the plot. Results for ¹H NMR mQTLs identified in the current study are shown in the upper section. For each metabolic trait (labelled right), the plot displays estimates of the proportion of biological variance explained by five complementary sources (labelled top; see Materials and Methods for explanation), including the mQTL SNP genotypes, familial variation excluding the mQTL SNP variation, individual environmental variation, and two types of visit variation (individual and common). Posterior distributions for proportions are represented as follows: the central tick in a box marks the posterior mean, the ends of a box mark the posterior quartiles, and the whiskers represent a 95% credible interval (extending to the 2.5 and 97.5 posterior percentiles).

Figure 5. Relationship between sample size and the size of effect detectable with 80% power in each study (shown by solid lines).

The effect size is parameterized by , which is the proportion of total population variance in metabolite concentration explained by the mQTL genotype (or, equivalently, the squared correlation between genotype and trait). It is assumed that the family-wise error rate in each study is controlled at 0.05 using the Bonferroni method. The number of tests performed is calculated as the product of the SNP and metabolite counts, as shown in the legend. Dashed lines relate the actual sample size of each study to that study's detectable effect size.

Figure 6. TMAu's mQTL effect may be mediated by variation in mRNA transcription at PYROXD2.

Each point represents, for a single study participant, their concentration of TMAu (vertical axis), their expression of PYROXD2 in adipose tissue (horizontal axis), and their mQTL genotype (point colour). The intensity data, , on each of the vertical and horizontal axes have been transformed . This transformation sets the minimum observation to zero on log₂ scale, and presents log₂(fold change) relative to the minimum value.