Single-marker analyses were implemented in PLINK 1.07 (Purcell, http://pngu.mgh.harvard.edu/~purcell/plink/) using logistic regression for the binary trait of AD and linear regression for the quantitative traits in cases only to calculate effect size (either OR or regression coefficient) and significance level. We used sex as a covariate in the logistic regressions and both age and sex as covariates in the linear regressions. To address the possibility of type I error because of multiple testing of several SNPs within individual genes, we permuted each p-value 10,000 times using the gene-based set test in PLINK and only reported the empirical p-value here if it was significant after this correction. We reasoned that gene-based correction was sufficiently conservative because all selected genes have a priori evidence of association with AD and/or related phenotypes. We used the set-based test in PLINK for multiple test correction because this method allows for identification of independent SNPs determined by a selected threshold. We changed the default threshold for LD from r2 = 0.5 to r2 = 0.8 because our LD-tagging SNPs were selected based on r2 ≥ 0.8.
