Multivariate Modeling of Direct and Proxy GWAS Indicates Substantial Common Variant Heritability of Alzheimer's Disease
Genome-wide association studies (GWAS) of proxy-phenotypes using family history of disease (GWAX) substantially boost power for genetic discovery when combined with direct case-control GWAS, most prominently in the context of Alzheimer's Disease (AD). However, despite twin study heritability estimates of approximately 60%, recent SNP-based estimates of common variant heritability of AD from meta-analyzed GWAS-GWAX data have been particularly low (2.5%), calling into question the prospects of continued progress in AD genetics. We demonstrate that commonly used approaches for combining GWAX and GWAS data produce dramatic underestimates of heritability, and we introduce a multivariate method for estimating individual SNP effects and recovering unbiased estimates of SNP heritability when combining GWAS and GWAX summary data. We estimate the SNP heritability of Clinical AD diagnoses excluding the APOE region at ~6-10%, with the corresponding estimate for biological AD (based on prevalence rate estimates from recently published molecular imaging data) as high as ~20%. Common variant risk for AD appears to represent a very strong effect of APOE superimposed upon a relatively diffuse polygenic signal that is distributed across the genome.