We combined 14 eQTL datasets into the ‘MetaBrain’ resource to maximize statistical power to detect eQTLs and create a brain-specific gene co-regulation network (Fig. 2, Supplementary Figs. 1–7 and Supplementary Table 1). Previous to quality control (QC), MetaBrain includes 7,604 RNA-seq samples and accompanying genotypes from the AMP-AD consortium (AMP-AD MAYO, ROSMAP and MSBB)6, Braineac7, the PsychENCODE consortium8 (Bipseq4, BrainGVEX4, CMC9, CMC_HBCC and UCLA_ASD4), BrainSeq10, NABEC11, TargetALS12 and GTEx3. In addition, we carefully selected 1,759 brain RNA-seq samples from the ENA13, which we subsequently genotyped and imputed (Fig. 2a, Supplementary Note and Supplementary Figs. 1–3). After realignment, removal of duplicate samples and stringent QC, 8,613 RNA-seq samples remained (Methods and Supplementary Figs. 4,5). Using slightly different QC thresholds, we created a gene network using 8,544 samples (Supplementary Note). For both datasets, we corrected the RNA-seq data for technical covariates and defined seven major tissue groups (amygdala, basal ganglia, cerebellum, cortex, hippocampus, hypothalamus and spinal cord): principal component analysis (PCA) on the RNA-seq data showed clear clustering by these major tissue groups, resembling brain physiology (Fig. 2b and Supplementary Fig. 6).
