Exploiting the mediating role of the metabolome to unravel transcript-to-phenotype associations

Abstract

Despite the success of genome-wide association studies (GWASs) in identifying genetic variants associated with complex traits, understanding the mechanisms behind these statistical associations remains challenging. Several methods that integrate methylation, gene expression, and protein quantitative trait loci (QTLs) with GWAS data to determine their causal role in the path from genotype to phenotype have been proposed. Here, we developed and applied a multi-omics Mendelian randomization (MR) framework to study how metabolites mediate the effect of gene expression on complex traits. We identified 216 transcript-metabolite-trait causal triplets involving 26 medically relevant phenotypes. Among these associations, 58% were missed by classical transcriptome-wide MR, which only uses gene expression and GWAS data. This allowed the identification of biologically relevant pathways, such as between ANKH and calcium levels mediated by citrate levels and SLC6A12 and serum creatinine through modulation of the levels of the renal osmolyte betaine. We show that the signals missed by transcriptome-wide MR are found, thanks to the increase in power conferred by integrating multiple omics layer. Simulation analyses show that with larger molecular QTL studies and in case of mediated effects, our multi-omics MR framework outperforms classical MR approaches designed to detect causal relationships between single molecular traits and complex phenotypes.

Keywords: Mediation; Mendelian Randomization; gene expression; genetics; genomics; human; metabolomics.

Figure 1.. Workflow overview.

(A) Estimation of the causal transcript-to-metabolite and metabolite-to-phenotype effects through univariable Mendelian randomization (MR). (B) Estimation of the causal transcript-to-phenotype effects through univariable transcriptome-wide MR (TWMR). (C) Estimation of the direct (i.e., not mediated by the metabolites) and mediated effect of transcripts on phenotypes through multivariable MR (MVMR) by accounting for mediation through the metabolome.

Figure 1—figure supplement 1.. Number of instrumental variables (IVs) used for causal effect estimation through Mendelian randomization (MR).

Distribution of the number of IVs used by the univariable MR aiming at identifying (A) transcript-to-metabolite, (B) metabolite-to-phenotype, and (C) transcript-to-phenotype causal relations. Data related to Figure 1—figure supplement 1 panels A to C are available in Figure 1—figure supplement 1—source data 1.

Figure 2.. Direct and mediated effects.

(A) Graphical representation of the multivariable Mendelian randomization (MVMR) framework for mediation analysis: DNA represents genetic instrumental variables (IVs) chosen to be directly associated with either the exposure (transcript; ${\beta }_{eQTL}$) or the mediator (metabolite; ${\beta }_{mQTL}$) through summary statistics. The effect of these IVs on the outcome (phenotype; ${\beta }_{GWAS}$) originates from genome-wide association studies (GWASs) summary statistics. Total effects ${\alpha }_{TP}$ of transcripts on phenotypes are estimated by transcriptome-wide Mendelian randomization (TWMR), while direct effects ${\alpha }_d$ are estimated by MVMR. Total effects ${\alpha }_{TP}$ are assumed to equal the sum of the direct ${\alpha }_d$ and indirect ${\alpha }_i$ (i.e., mediated) effects, the two former being depicted in B. (B) Direct (${\alpha }_d$ ; y-axis) and total (${\alpha }_{TP}$ ; x-axis) effects for the 216 transcript-metabolite-trait causal triplets. The dashed line represents the identity, while the purple line represents the regression line with a shaded 95% confidence interval. Data related to Figure 2 panel B are available in Figure 2—source data 1.

Figure 3.. Molecular pleiotropy at the FADS locus.

(A) Genome browser (GRCh37/hg19) view of the genomic region on chromosome 11 encompassing TMEM258, FADS1, and FADS2 (red). (B) Diagram of the mediation signals detected for TMEM258, FADS1, and FADS2. Two of the implicated genes encode enzymes involved in arachidonic synthesis (purple). Involved genes impact 17 phenotypes (pink) through alteration of the levels of three metabolites, 1-arachidonoylglycerophosphocholine (LPC(20:4)), 1-arachidonoylglycerophosphoethanolamine (LPE(20:4)), and 1-arachidonoylglycerophosphoinositol (LPI(20:4)) whose structure is depicted (orange). (C) Network of the 65 transcript-metabolite-trait causal triplets involving TMEM258, FADS1, and FADS2. Nodes represent genes (purple), metabolites (orange), or phenotypes (pink). Edges indicate the direction of the effects estimated through univariable Mendelian randomization. Width of edges is proportional to effect size and color indicates if the effect is positive (red) or negative (blue).

Figure 4.. Power comparison between transcriptome-wide Mendelian randomization (TWMR) and multivariable Mendelian randomization (MVMR).

Heatmap showing the difference in statistical power between TWMR and mediation analysis through MVMR at current (A; ${\displaystyle N=8000}$) and realistic future (B; ${\displaystyle N=90,000}$) metabolic quantitative trait loci (mQTL) dataset sample sizes. The x-axis shows the proportion ($\rho$) of direct (${\alpha }_d$) to total (${\alpha }_{TP}$) effect (i.e., effect not mediated by the metabolite) ranging from –2 to 2, arrows indicating increasing proportion of direct effect. The y-axis shows the ratio ($\sigma$) between the transcript-to-metabolite (${\alpha }_{TM}$) and the metabolite-to-phenotype (${\alpha }_{MP}$) effects, ranging from 0.1 to 10. Red vs. gray indicates higher power for TWMR vs. mediation analysis, respectively, while white represents equal power between the two approaches. Data related to Figure 4 panels A and B are available in Figure 4—source data 1 and Figure 4—source data 2, respectively.

Figure 4—figure supplement 1.. Distribution of empirical causal triplets along tested regime parameters.

Distribution of the proportion ($\rho$) of direct (${\alpha }_d$) to total (${\alpha }_{TP}$) effect (i.e., effect not mediated by the metabolite with arrows indicating increasing proportion of direct effect; x-axis) and ratio ($\sigma$) between the transcript-to-metabolite (${\alpha }_{TM}$) and the metabolite-to-phenotype (${\alpha }_{MP}$) effects (top row indicates $\sigma$ larger than 10; y-axis) for the 216 identified putative causal triplets. Color indicates the number of triplets under each combination of parameters, ranging from 0 (white). Data related to Figure 4—figure supplement 1 are available in Figure 4—figure supplement 1—source data 1.

Figure 4—figure supplement 2.. Power comparison between transcriptome-wide Mendelian randomization (TWMR) and multivariable Mendelian randomization (MVMR) at smaller sample sizes.

Heatmap showing the difference in statistical power between TWMR and mediation analysis through MVMR at metabolic quantitative trait locus (mQTL) dataset sample sizes smaller than the one in the main analysis: (A) N = 1000; (B) N = 2000; (C) N = 4000. The x-axis shows the proportion ($\rho$) of direct (${\alpha }_d$) to total (${\alpha }_{TP}$) effect (i.e., effect not mediated by the metabolite) ranging from –2 to 2. The y-axis shows the ratio ($\sigma$) between the transcript-to-metabolite (${\alpha }_{TM}$) and the metabolite-to-phenotype (${\alpha }_{MP}$) effects, ranging from 0.1 to 10. Red vs. gray indicates higher power for TWMR vs. mediation analysis, respectively, while white represents equal power between the two approaches. Data related to Figure 4—figure supplement 2 panels A, B, and C are available in Figure 4—figure supplement 2—source data 1–3, respectively.