Samples were excluded if RIN < 5.5 or genetic information from the sample was inconsistent with subject descriptors such as sex. Of the 633 samples sent for sequencing (those with RIN ≥ 5.5), 15 samples were removed because they yielded < 50 million total reads (~25 million paired end reads) or had > 5% of reads aligning to rRNA, based on two attempts to produce quality sequence (all samples failing either QC criterion on the first attempt were re-prepped and/or re-sequenced, and those failing twice were removed); calculation of RNA-seq QC metrics is described in “Mapping, QC and Quantification of Gene Expression” below. Of the 10 Pitt control samples that were sequenced twice, only the first sequencing run was included in our analysis. Of the remaining 609 samples, two were removed because their DNA genotypes had high rates of missingness; one sibling pair was identified and the sample with the lower RNA quality (RIN) was removed; 14 samples were removed (see details below) because they were determined to be outliers based of a series of multivariate analyses of the RNA-seq data (N =10), or due to sample contamination/mix-up (N = 4). This left 592 samples for subsequent analyses.
