Integrated analysis of direct and proxy genome wide association studies highlights polygenicity of Alzheimer's disease outside of the APOE region

Recent meta-analyses combining direct genome-wide association studies (GWAS) with those of family history (GWAX) have indicated very low SNP heritability of Alzheimer's disease (AD). These low estimates may call into question the prospects of continued progress in genetic discovery for AD within the spectrum of common variants. We highlight dramatic downward biases in previous methods, and we validate a novel method for the estimation of SNP heritability via integration of GWAS and GWAX summary data. We apply our method to investigate the genetic architecture of AD using GWAX from UK Biobank and direct case-control GWAS from the International Genomics of Alzheimer's Project (IGAP). We estimate the liability scale common variant SNP heritability of Clinical AD outside of APOE region at similar to 7-11%, and we project the corresponding estimate for AD pathology to be up to approximately 23%. We estimate that nearly 90% of common variant SNP heritability of Clinical AD exists outside the APOE region. Rare variants not tagged in standard GWAS may account for additional variance. Our results indicate that, while GWAX for AD in UK Biobank may result in greater attenuation of genetic effects beyond that conventionally assumed, it does not introduce appreciable contamination of signal by genetically distinct traits relative to direct case-control GWAS in IGAP. Genetic risk for AD represents a strong effect of APOE superimposed upon a highly polygenic background. Author summary In this article we demonstrate that common approaches for combining direct and proxy GWAS data produce dramatic underestimates of heritability, and we introduce a novel multivariate method for recovering unbiased estimates. Applying our method to direct and proxy GWAS data for Alzheimer's disease we obtain estimates of liability-scale SNP heritability of AD outside of APOE region that are more than double the most recent estimates. Our analysis of local SNP heritability indicates that nearly 90% of common variant risk for Clinical AD represent a polygenic signal that is relatively diffusely distributed across the genome outside of the APOE region. We add substantial clarity to the scientific understanding the genetic architecture of AD, by introducing and validating new methodology that can be widely applied by researchers seeking to incorporate family information into GWAS.