Lack of strong human replication: The lack of strong human replication is a serious limitation of our study. However, non-replication of novel signals is common in GWAS of AD and may have several causes. Specific to our study, differences in sampling, data production, QC, imputation or analysis can confound meta-analysis. The GESGA sample was imputed to an older, smaller HapMap reference panel, has many missing data points and shows the least consistent sign tests (Table 1 and Supplementary Table S1). The OZALC population sample was genotyped at multiple sites, a well-known source of systematic genotype bias. We note that our strongest replication signals come from analyses of the three samples (Irish, COGA, Yale/Penn) with the most consistent ascertainment and genotyping. More generally, this pattern of non-replication may reflect the multiple independent domains of risk for AD (Kendler et al., 2012), including variation in 1) alcohol-specific physiological measures like initial sensitivity and tolerance (Schuckit et al., 1997), 2) brain reward circuitry implicated in substance use and other pathological behaviors (Volkow et al., 2012),, and 3) personality traits like internalizing and externalizing
