Controlling for genetic Principal Components is now standard practice in fields such as sociogenomics and population genetics. Most critiques of these approaches revolve around the concerns that principal components do not adequately capture subtle population structure and certainly do not control for assortative mating or genetic nurture. However, in this paper, we ask the converse question: whether there phenotypes for which PCs represent "true" causal effects? Specifically, we address the following questions: 1) whether there is residual signal in genetic principal components causally linked to different phenotypes, and 2) if so, whether our estimates of the effects of polygenic indices on phenotypic outcomes are being biased by them. We do so across nine clinical, anthropomorphic, and social phenotypes. We use white, British sibling pairs in the UKB (n = 19,290) and impute their parental genomes. Since within-family genetic variation is random, by controlling for parental genotypes in our models, we provide estimates of the direct genetic effects of PCs and PGIs, free from environmental confounding. Our results show that PCs are predictive of anthropomorphic traits like height and birthweight, above and beyond parental PCs, suggesting a causal association between ancestry and these traits. Even so, controlling for causal PCs does not appear to bias PGI estimates, which are robust in both their significance and magnitude across models.
Ramina Sotoudeh, University of Oxford
Sam Trejo, Princeton University
Dalton Conley, Princeton University