a transcript were affected were removed and allowed on the exon-level list if they achieved exon-level study-wide significance. (2) Using stepwise linear regression (STATA/IC 10.0) associations were removed that were redundant due to LD between SNPs. Specifically, following inclusion of the most significant SNP-probeset association, only SNPs that contributed significantly above and beyond the initial association at a p-value of <10−6 were considered as a separate association. (3) Finally, if the sQTL or a SNP in LD (r 2 > 0.5) was located in a probeset for the exon-level associations, it was excluded from any list of significant associations. Step 3 was not applied to transcript level associations as these expression levels were determined over a range of exons thereby reducing the contribution of SNPs to the expression levels. Post hoc evaluation of the transcript level associations we identified were manually inspected for effects of SNPs or SNPs in LD contributing to the association by exporting the raw values, eliminating probesets that contained a SNP, and recalculating the expression level. None of the reported associations could be accounted for by a SNP in a probeset measuring the transcript.
