Mendelian imputation of parental genotypes for genome-wide estimation of direct and indirect genetic effects

Abstract

Associations between genotype and phenotype derive from four sources: direct genetic effects, indirect genetic effects from relatives, population stratification, and correlations with other variants affecting the phenotype through assortative mating. Genome-wide association studies (GWAS) of unrelated individuals have limited ability to distinguish the different sources of genotype-phenotype association, confusing interpretation of results and potentially leading to bias when those results are applied – in genetic prediction of traits, for example. With genetic data on families, the randomisation of genetic material during meiosis can be used to distinguish direct genetic effects from other sources of genotype-phenotype association. Genetic data on siblings is the most common form of genetic data on close relatives. We develop a method that takes advantage of identity-by-descent sharing between siblings to impute missing parental genotypes. Compared to no imputation, this increases the effective sample size for estimation of direct genetic effects and indirect parental effects by up to one third and one half respectively. We develop a related method for imputing missing parental genotypes when a parent-offspring pair is observed. We provide the imputation methods in a software package, SNIPar (single nucleotide imputation of parents), that also estimates genome-wide direct and indirect effects of SNPs. We apply this to a sample of 45,826 White British individuals in the UK Biobank who have at least one genotyped first degree relative. We estimate direct and indirect genetic effects for ∼5 million genome-wide SNPs for five traits. We estimate the correlation between direct genetic effects and effects estimated by standard GWAS to be 0.61 (S.E. 0.09) for years of education, 0.68 (S.E. 0.10) for neuroticism, 0.72 (S.E. 0.09) for smoking initiation, 0.87 (S.E. 0.04) for BMI, and 0.96 (S.E. 0.01) for height. These results suggest that GWAS based on unrelated individuals provides an inaccurate picture of direct genetic effects for certain human traits.