The R Foundation for Statistical Computing). Using principal components analysis of expression data, outlier arrays were detected. Non-pseudoautosomal Y chromosome transcript expression levels were used for a gender check. Outlier arrays, samples with inconsistent gender information, and samples designated as duplicates in our GWAS data, were removed from the raw data (n = 67). Also, non-autosomal probes were excluded (n = 2,002). The thus obtained trimmed raw dataset was again quantile normalized and log2 transformed. All probe sequences were aligned to the NCBI build 36 reference genome using UCSC’s Genome Browser function BLAT [31]. Non-specific probes, defined as no or multiple hits with a sequence homology >95%, were removed (n = 7,234). RefSeq (updated on 27 September 2010) and UniGene (build #228, release date 29 October 2010) databases were used to determine probes mapping to transcripts designated as retired and these probes were excluded as well (n = 2,449), leaving 37,118 gene-expression probes.
