Unveiling the Genetics Architectures of Rare Coding Variants in Blood Lipids Levels via Large Scale Meta-analysis. D. Liu on behalf of the Global Lipids Genetics Consortium Department of Public Health Sciences, Pennsylvania State University, Hershey, PA.
In order to understand the impact of rare coding variants on plasma lipids levels, we are performing large scale meta-analysis of exome-array data in the Global Lipids Genetics Consortium and developing novel statistical methods that summarize and describe the underlying genetic architecture. Association statistics were aggregated across 289,500 individuals from 94 studies with plasma lipids levels and exome-array genotypes. Meta-analysis of single variant and gene-level association tests were performed centrally using RAREMETAL for HDL, LDL, triglyceride and total cholesterol levels. After quality control, a total of 235,526 variants segregated in the aggregate dataset. Of these, 89.5% and 69.9% have minor allele frequency <1% and <0.1%, respectively. Among the coding variants, 204,618 are nonsynonymous (NS) and 7,438 are loss-of-function (LOF) variants. Using this data, we identified known and novel lipids genes, e.g. LDLR, APOC3, NPC1L1, APOH, ABCA6, as well as novel variants within known lipids loci. The impact of these new genes on MI are being investigated. In addition to identifying novel loci, large datasets provide us a unique opportunity to learn the genetic architecture of rare coding variants. To achieve this goal, we develop a novel empirical Bayesian model, which estimates genetic effect size distributions and posterior probability of associations for different classes of rare variants. The method only requires summary level data as input, allows the presence of multiple causative variants in the same gene, and accurately models the dependence between variants induced by linkage disequilibrium and relatedness. To apply the model to whole exome data, we also developed an efficient hybrid Expectation-Maximization and MCMC algorithm. Fitting the model to whole-exome data and adjusting for winners curse, we showed that there is significant enrichment of causative variants among LOF alleles relative to NS variants (p<1e-5). The genetic effect sizes for LOF variants are ~3X larger on average than for NS variants. A fraction of >33% of the GWAS signals can be explained in full or in partial by rare coding variants. Jointly modeling effects of common and rare variants increases estimated heritability at GWAS loci by 20%. Taken together, our method and analysis provide one of the first comprehensive investigations of genetic architectures in >280,000 individuals, and establish the significance of studying rare coding variants for lipids traits.
You may contact the first author (during and after the meeting) at