For the three brain data sets36-38, expression levels were inverse normal transformed and regressed on the number of rs4583255-T alleles with gender, age at death, post-mortem interval, brain source, expression experiment batch, pH (Colantuoni et al36 only), sample expression level based on the total number of transcripts detected (Webster et al38 only) and Alzheimer’s disease patient status (Webster et al38 only) as covariates. To incorporate data from different brain regions (Gibbs et al37) or different probes (KCTD13 in Colantuoni et al36) derived from the same individual, a mixed-effects model with individual as a random effect was used. Results from the three data sets were combined using inverse-variance weighted meta-analysis. The Dutch whole blood data set included control samples from two studies39, 40. Analysis was performed using linear regression in Plink41 taking age and gender as covariates. The Icelandic blood data set has been described previously42, and analysis was carried out as detailed in that work42.
