Heritability estimates for 361 blood metabolites across 40 genome-wide association studies

Abstract

Metabolomics examines the small molecules involved in cellular metabolism. Approximately 50% of total phenotypic differences in metabolite levels is due to genetic variance, but heritability estimates differ across metabolite classes. We perform a review of all genome-wide association and (exome-) sequencing studies published between November 2008 and October 2018, and identify >800 class-specific metabolite loci associated with metabolite levels. In a twin-family cohort (N = 5117), these metabolite loci are leveraged to simultaneously estimate total heritability (h²_total), and the proportion of heritability captured by known metabolite loci (h²_{Metabolite-hits}) for 309 lipids and 52 organic acids. Our study reveals significant differences in h²_{Metabolite-hits} among different classes of lipids and organic acids. Furthermore, phosphatidylcholines with a high degree of unsaturation have higher h²_{Metabolite-hits} estimates than phosphatidylcholines with low degrees of unsaturation. This study highlights the importance of common genetic variants for metabolite levels, and elucidates the genetic architecture of metabolite classes.

Fig. 1. Overview of the four-variance component models.

Overview of the SNP-filtering and GRM construction can be found in Supplementary Fig. 1 and is explained in details in the Methods. This figure describes which GRMs (black boxes) are used to calculate which variance components (orange boxes) by drawing black arrows from the GRMs to the variance components. The variance components give rise to the four different heritability estimates: h²_ped, h²_g, h²_Class-hits, and h²_{Notclass-hits} (see Methods). The orange arrows indicate how the various variance components are summed to obtain estimates for h²_{metabolite-hits}, h²_SNP, and h²_total (see Methods).

Fig. 2. Heritability of all 52 carboxylic acids by class.

Box- and dotplots of the h²_total, and h²_{Metabolite-hits} for all 52 successfully analyzed carboxylic acids and derivatives across all metabolomics platforms by class. The left-hand side of the figure is a close-up of the −0.08 to 0.15 part of the heritability range, focusing on the h²_Class-hits and h²_{Notclass-hits} estimates. The boxes denote the 25th and 75th percentile (bottom and top of box), and median value (horizontal band inside box). The whiskers indicate the values observed within up to 1.5 times the interquartile range above and below the box. The purple, orange and green boxes denote the keto acid, hydroxyl acid and carboxylic acid classes, respectively. Supplementary Data 3 provides the estimates for each of the individual metabolites.

Fig. 3. Heritability of all 309 lipids by class.

Box- and dotplots of the h²_total, and h²_{Metabolite-hits} for all 309 successfully analyzed lipids and lipid-like molecules across all metabolomics platforms by class. The left-hand side of the figure is a close-up of the −0.06 to 0.17 part of the heritability range, focusing on the h²_Class-hits and h²_{Notclass-hits} estimates. The boxes denote the 25th and 75th percentile (bottom and top of box), and median value (horizontal band inside box). The whiskers indicate the values observed within up to 1.5 times the interquartile range above and below the box. The yellow, pink, orange, light green, purple, and dark green boxes denote the steroids, lipoprotein, glycerolipid, sphingolipid, glycerophospholipid, and fatty acyl classes, respectively. Supplementary Data 3 provides the estimates for each of the individual metabolites.