RNA sequencing was performed after depleting ribosomal RNA (rRNA). Following quality control, there were 258 SCZ cases and 279 controls. Fifty-five cases with affective disorder were included to increase power to detect eQTLs. The median number of paired end reads per sample was 41.6 million, with low numbers of rRNA reads (Supplementary Fig. 2). Following data normalization, 16,423 genes (based on Ensembl models) were expressed at levels sufficient for analysis, of which 14,222 were protein coding. Validation using PCR showed high correlation (r > 0.5) with normalized expression from RNA-seq for the majority of genes assessed (Supplementary Fig. 3). Gene expression measurement can be influenced by a number of variables; some are well documented (e.g., RNA integrity (RIN) and post-mortem interval (PMI)), but others may be unknown. We investigated known covariates by standard model selection procedures to find a good statistical model (Supplementary Fig. 4 and 5). Covariates for RIN, library batch, institution (brain bank), diagnosis, age of death, genetic ancestry, PMI, and sex together explained a substantial fraction (0.42) of the average variance of gene expression, and were thus employed to adjust the data for all analyses.
