Integrating metabolite, BMI and genetic data in phenotypic extremes, drawn from a population of 50,000 samples, to assess causality of metabolite levels in obesity. T. Esko1,2,3, A. Metspalu1, C. Clish3, JN. Hirschhorn1,3 1) Division of Endocrinology, Children's Hospital Boston, Boston, MA; 2) Estonian Genome Center, University of Tartu, Tartu, Estonia; 3) Broad Institute, Cambridge, MA.
Obesity is a disorder of energy metabolism and signaling between tissues. As such, it may be possible to obtain additional clues about particular causal pathways in obesity by examining metabolites in whole blood from lean and obese individuals. However, although many metabolites are correlated with obesity, causality is more difficult to ascertain. Mendelian randomization offers an approach to assess causality. In order to identify metabolites correlated with obesity, we carried out both targeted (340 metabolites) and untargeted metabolite profiling using LC-MS in 100 normal weight, 100 lean and 100 obese individuals drawn from the extremes of a population of 50,000 samples. As a measure of obesity, we analyzed body mass index (BMI), using z-scores adjusted for age and gender from the entire cohort. From the targeted approach we identified 155 metabolites associated with BMI. Many associated metabolites were highly correlated and included species of di- and triacylglycerols, amino acids with their metabolic derivatives and molecules related to the choline pathway. In order to assess whether these metabolic alterations could play a causal role or are simply reflections of the obese state, we applied Mendelian randomization (MR), which uses genetic variants that directly influence a potential mediator (in this case, a metabolite level) as instrumental variables to assess causal influences on an outcome (in this case, obesity). Genome-wide association analyses were conducted with each of the 155 metabolites correlated with BMI. In total we identified 15 sequence variants at array-wide significance level for 6 compounds, which on average explained 8 to 13 percent of the respective trait variance. By using the best sequence variant per metabolite as instrumental variables, we provide initial evidence that long-chain triacylglycerols (C50:4, C52:5 and C56:5, respective MR p-values 1.8e-4, 1.3e-4 and 9.6e-5) and branched-chain amino acid valine (p-value 1.1e-4) are causally linked to obesity. In conclusion, by using a well-powered extremes design, high throughput metabolite profiling and Mendelian randomization we were able to provide evidence that most metabolites correlated with BMI are likely downstream effects of obesity but that long-chain triacylglycerols, branched-chain amino acids and related pathways may be causally linked to obesity.