Selection and mutation shape genetic variation underlying human traits, but the specific evolutionary mechanisms driving complex trait variation are largely unknown. We developed a statistical method that uses polarized GWAS summary statistics from a single population to detect signals of mutational bias and selection. We found evidence for non-neutral signals on variation underlying several traits (BMI, schizophrenia, Crohn’s disease, educational attainment, and height). We then used simulations that incorporate simultaneous negative and positive selection to show that these signals are consistent with mutational bias and shifts in the fitness-phenotype relationship, but not stabilizing selection or mutational bias alone. We additionally replicate two of our top three signals (BMI and educational attainment) in an external cohort, and show that population stratification may have confounded GWAS summary statistics for height in the GIANT cohort. Our results provide a flexible and powerful framework for evolutionary analysis of complex phenotypes in humans and other species, and offer insights into the evolutionary mechanisms driving variation in human polygenic traits.Impact summaryMany traits are variable within human populations and are likely to have a substantial and complex genetic component. This implies that mutations that have a functional impact on complex human traits have arisen throughout our species’ evolutionary history. However, it remains unclear how processes such as natural selection may have acted to shape trait variation at the genetic and phenotypic level. Better understanding of the mechanisms driving trait variation could provide insights into our evolutionary past and help clarify why it has been so difficult to map the preponderance of causal variation for common heritable diseases.In this study, we developed and applied methods for detecting signatures of mutation bias (i.e., the propensity of a new variant to be either trait-increasing or trait-decreasing) and natural selection acting on trait variation. We applied our approach to several heritable traits, and found evidence for both natural selection and mutation bias, including selection for decreased BMI and decreased risk for Crohn’s disease and schizophrenia.While our results are consistent with plausible evolutionary scenarios shaping a range of traits, it should be noted that the field of polygenic selection detection is still new, and current methods (including ours) rely on data from genome-wide association studies (GWAS). The data produced by these studies may be vulnerable to certain cryptic biases, especially population stratification, which could induce false selection signals. We therefore repeated our analyses for the top three hits in a cohort that should be less susceptible to this problem – we found that two of our top three signals replicated (BMI and educational attainment), while height did not. Our results highlight both the promise and pitfalls of polygenic selection detection approaches, and suggest a need for further work disentangling stratification from selection.
Estimating heritability without environmental bias
