As with other phenotypes, the first GWAS findings for imaging phenotypes were difficult to replicate, probably due to the limited availability of replication samples. Prior work had cautioned that, using the most standard kind of univariate analysis of SNP effects, very large sample sizes would be required—far larger than a typical neuroimaging study—to discover influential genetic variants, unless their effect sizes on the brain phenotypes being analyzed were unusually large (Potkin et al. 2009b). The power of a GWAS study depends on the number of null genetic variants assessed, the expected effect size of the genetic variant (typically less than 1 % of the trait variance), and the population frequency of the variant (Potkin et al. 2009b; Flint et al. 2010; Paus et al. 2012). The power obviously also depends on the genetic architecture of the trait—key factors are the large number of effectively independent LD regions in the genome (regions of linkage disequilibrium, with correlated SNPs) and the number of detectable causal loci per phenotype (which is typically small). The first series of studies identifying genetic markers for MRI
