Imputation of fluid intelligence scores reduces ascertainment bias and increases power for analyses of common and rare variants
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Studying the genetics of intelligence can help us understand the neurobiology of cognitive function and the aetiology of rare neurodevelopmental conditions. The largest previous genetic studies of intelligence have used ~270k individuals who completed the fluid intelligence (FI) tests in UK Biobank. Here, we integrate additional FI measures in this cohort and leverage 82 correlated variables to impute FI values for unmeasured individuals, increasing the sample size to >450k. Through population-based and within-family genome-wide association studies and genetic correlation analyses, we show that this imputation produces a phenotype that genetically resembles measured FI, as well as reduces ascertainment bias within the cohort. We further show that combining measured and imputed FI scores increases the number of independent SNP associations (p<5x10-8) from 385 to 608 and increases polygenic score accuracy in external cohorts by 15% on average. Additionally, incorporating imputed FI scores increases the number of gene-level associations with rare variants from five to twenty-six (FDR<1%). These include fourteen well-established developmental disorder-associated genes, a four-fold enrichment (p=8x10-8); for several of these (e.g. ATP1A1, CACNA1E), our results suggest that the gene has a loss-of-function mechanism as well as the previously-documented altered function mechanism. We also implicate twelve genes without strong prior evidence of association developmental disorders, of which eight have not been previously linked to intelligence (ROBO2, RB1CC1, ANK3, CHD9, TLK1, PCLO, DPP8, IPO9). These twelve genes were significantly enriched for de novo loss-of-function mutations in a set of >31k patients with developmental disorders (p=6.8x10-4). We further identify three genes showing significant rare variant associations with educational attainment but not with FI in UK Biobank, including CADPS2 in which, unusually, protein-truncating variants show a positive association. Our results demonstrate the power of phenotype imputation for genetic studies and suggest that incorporating genetic association results for cognitive phenotypes in the general population could help discover new developmental disorder genes.