Population differentiation of polygenic score predictions under stabilizing selection

Abstract

Given the many small-effect loci uncovered by genome-wide association studies (GWAS), polygenic scores have become central to genomic medicine, and have found application in diverse settings including evolutionary studies of adaptation. Despite their promise, polygenic scores have been found to suffer from limited portability across human populations. This at first seems in conflict with the observation that most common genetic variation is shared among populations. We investigate one potential cause of this discrepancy: stabilizing selection on complex traits. Counterintuitively, while stabilizing selection constrains phenotypic evolution, it accelerates the loss and fixation of alleles underlying trait variation within populations (GWAS loci). Thus even when populations share an optimum phenotype, stabilizing selection erodes the variance contributed by their shared GWAS loci, such that predictions from GWAS in one population explain less of the phenotypic variation in another. We develop theory to quantify how stabilizing selection is expected to reduce the prediction accuracy of polygenic scores in populations not represented in GWAS samples. In addition, we find that polygenic scores can substantially overstate average genetic differences of phenotypes among populations. We emphasize stabilizing selection around a common optimum as a useful null model to connect patterns of allele frequency and polygenic score differentiation. This article is part of the theme issue ‘Celebrating 50 years since Lewontin's apportionment of human diversity’.

The background and legacy of Lewontin's apportionment of human genetic diversity

“…That said, elevated differentiation between human groups in genetic variation underlying traits can arise for several reasons: if the loci underlying a trait have experienced geographically variable selective pressures (as seen with skin pigmentation loci), or less obviously, in various cases where there is stabilizing selection on a trait, even when there is a single optimum shared between species (e.g. [100,101]). Overall, the degree to which variable selective pressures have acted on variants underlying human traits is still unknown and is often difficult to understand rigorously using techniques developed to date [101][102][103].…”

Polygenic risk, population structure and ongoing difficulties with race in human genetics

“…Both (i) and (ii) involve the possibility that differences in the genetic makeup of two populations might result in a polygenic risk score (PRS) that is predictive in one population not being predictive in the other, because of the differences in allele frequencies between the two populations (see [ 67 ] for a detailed analysis of some of the ways in which this can happen). Imagine two populations, each of which has hundreds or thousands of genes that influence a particular trait.…”

Celebrating 50 years since Lewontin's apportionment of human diversity

“…As in forensic genetics, human genetic structure, despite generally low levels of differentiation, causes complications for polygenic scores. These complications arise both in their estimation, because confounding due to population stratification can lead to errors in estimating the scores, and, as discussed by Yair & Coop [18], in their application, as polygenic scores estimated in one population do not predict phenotypes as well when used in other populations. Kaplan & Fullerton draw out these and other tensions that arise in a setting of widespread health disparities and social patterns of inequity that correlate with aspects of human population structure.…”

“…In a modelling study, Yair & Coop [18] investigate the process of population differentiation under stabilizing selection, a process that plausibly underlies the evolution of many complex traits. Empirical results such as Lewontin's have established the relatively low locus-by-locus-level genetic differentiation of human populations, and long-standing results from quantitative genetics predict that under neutral evolution, the degree of population differentiation in a heritable trait is expected to mirror population differentiation at a typical locus.…”

“…Indeed, such a conclusion follows almost immediately for single-locus phenotypes, if Lewontin's results can be taken as representative of the relevant loci. For complex traits, there are more possibilities, but subsequent phenotypic modelling under neutral models argues that a high within-population variance component for typical genotypes suggests a similarly high within-population variance component for many phenotypes-and the within-population component will be even higher on average if populations experience stabilizing selection that selects for the same optimal phenotypic value (reviewed by Yair & Coop [18]). Although Lewontin's 1972 paper demonstrates that the essentialized racial theory is unsupported genetically, philosophical discussion since the paper has asked if some other theory of biological race with a weaker predictive ability could accommodate Lewontin's result (e.g.…”