6 Oral - Platform Session 1 Complex Traits
Wednesday June 08, 9:15 AM - 9:30 AM

Authors:
Yuval Simons ¹; Hakhamanesh Mostafavi ¹; Jonathan Pritchard ¹; Guy Sella ²

Affiliations:
1) Stanford University, Stanford, CA; 2) Columbia Universtiy, New York, New York

Keywords:
Complex traits

In recent years, genome-wide association studies (GWAS) have begun to systematically uncover the genetic architecture of complex traits. Complex traits seem to greatly differ in their genetic architecture, differing in the number of genome-wide significant hits discovered and the proportion of trait variance these hits explain. We sought to understand the evolutionary basis of such differences using a model of pleiotropic stabilizing selection. Using our model, we used the frequencies and z-scores of GWAS hits to infer the distribution of selection coefficients at trait-affecting loci and the number of trait-affecting loci, i.e. the target size. Using 96 continuous traits that have over 100 hits in the UK biobank, we discovered that variants underlying most traits have surprisingly similar distributions of selection coefficients. Intuitively, sufficiently polygenic traits sample selection coefficients from the same common pool of functional variation spread throughout the genome.

This common distribution of selection coefficients implies that differences in trait architectures between highly polygenic traits arise almost exclusively from differences in heritability and target size. There are order of magnitude differences in the heritability of the traits we consider due to different degrees of environmental contribution. Traits differ even more in target size: we infer that target sizes span over two orders of magnitude. However, once these two factors are accounted for, we show that the architecture of almost all of these 96 traits is essentially identical.

We explore the implications of this result. First, we show that the proportion of variants discovered in GWAS and the proportion of heritability they explain has near-identical dependency on study size for all traits, after normalizing study size by the heritability per trait-affecting site. Second, we show that the lack of portability of GWAS-derived polygenic risk scores is mainly driven by genetic drift and selection, and this reduction should be near-identical across traits. Interestingly, about 150 GWAS hits, mostly of low frequency and large effect size, stand out as outliers to our model and many of these hits affect genes well-known to influence trait biology. Thus, deviations from our model may offer clues to trait biology.