Signatures of negative selection in the genetic architecture of human complex traits

Methods such as LD-score regression and GREML typically assume a fixed relationship between allele frequency and effect size of single-nucleotide polymorphisms (SNPs), whereby rarer variants have larger effects than common... [ view full abstract ]

Methods such as LD-score regression and GREML typically assume a fixed relationship between allele frequency and effect size of single-nucleotide polymorphisms (SNPs), whereby rarer variants have larger effects than common variants, to exactly such an extent that each causal variant contributes by the same amount to phenotypic variance, regardless of allele frequency. However, the strength of the relation between allele frequency and effect size may differ across traits. Estimates of this relationship give us a clue about the effect selection has had on the frequency of trait-affecting variants. Therefore, such estimates may yield new insight into the genetic architecture of complex human traits, and can help to guide future GWAS efforts. To enable estimation of this relationship, we provide a Bayesian linear mixed model. This model allows us to simultaneously estimate SNP heritability and polygenicity (i.e. the proportion of SNPs with non-zero effects) as well as the strength of the relationship between effect size and allele frequency, for complex human traits. This method uses genome-wide SNP data on unrelated individuals. An application to data from the UK Biobank on 28 complex traits shows that, on average, 6% of the SNPs have non-zero effects. On average, these SNPs in total explain 22% of the phenotypic variance. For 23 traits, we find significant signatures of natural selection in the genetic architecture. Amongst others, these traits include reproductive and anthropometric traits, as well as biologically more distal outcomes such as educational attainment. As we illustrate using forward simulations, these estimates are compatible with negative selection, where deleterious variants are kept at low frequencies. We, therefore, conclude that negative selection has acted across the genome, affecting the allele frequencies of variants associated with a wide range of complex human traits.