Despite the large sample size of MDD2000+, a number of limitations may reduce the realized power for detection of association. First, our study combines data sets genotyped on different primary platforms, and the power of our sample will vary between SNPs depending on the linkage disequilibrium structure between genotyped and imputed SNPs. Second, some of our cases and controls were genotyped separately because of economic constraints. We were fortunate that we had >300 QIMR cases genotyped on both the Affymetrix and Illumina platforms, which we used extensively for QC. Third, when combining cases and controls genotyped as separate sets, it was necessary to impose more stringent QC constraints than for cases and controls genotyped on the same platform. These constraints were successful in removing artefactual differences between the case–control sets, but could also have eliminated true associations. Fourth, in MDD2000+ with the aim of maximizing sample size we have combined the clinical cases from the University of Edinburgh cohort with the community samples from QIMR and NTR. The extent to which this is important depends on the unknown genetic etiology
