We divided eQTL into separate subgroups by stepwise conditional order (first, second, and greater than second) and created sets of matched SNPs drawn from the SNPsnap31 database for each subgroup, matching on minor allele frequency, gene density (number of genes within 1 Mb of the SNP), distance from SNP to TSS of the nearest gene, and LD (number of LD-partners within r2 ≥ 0.8). For each subgroup of eQTL, we performed a logistic regression of status as eQTL or matched SNP on overlap with functional annotation, including the four SNP matching parameters as covariates. Enrichment was taken as the regression coefficient estimate, interpretable as the log-odds ratio for being an eQTL given a functional annotation. Functional annotations tested included: brain promoters and enhancers (union of all brain region TssA and Enh+EnhG intervals, respectively, from the NIH Roadmap Epigenomics Project32 ChromHMM33 core 15-state model), brain-specific promoters and enhancers (the union of all brain region TssA and Enh+EnhG intervals, excluding those present in seven other non-brain tissues/cell types: primary T helper cells from peripheral blood, osteoblast primary cells, HUES64 cells, adipose nuclei, liver, NHLF lung fibroblast primary cells, and NHEK-epidermal keratinocyte primary cells), and pre-frontal cortex (PFC) neuronal (NeuN+) and non-neuronal (NeuN−) nucleus H3K4me3 and H3K27ac ChIP-seq marks from the CMC. For each data source, active promoter and enhancer (or H3K4me3 and H3K27ac) annotations were tested for enrichment jointly. This analysis was repeated but restricting to matched SNPs located within 1 Mb of any of the 16,423 genes that were tested for eQTL, in order to determine whether the enrichment estimates were inflated due to the proximity of our primary and conditional eQTL to brain-expressed genes, which may be more likely to occur near active regulatory regions in the brain. In addition, to ensure that any enrichment patterns observed were not due to varying effect size among primary and conditional eQTL, the enrichment analyses were also carried out taking into account the variance in expression explained by each eQTL. Variance explained (R2) was estimated using the variancePartition34 R package, and eQTL were stratified into three R2 bins: bin 1, 1 × 10−2 ≤ R2 ≤ 1.75 × 10−2; bin 2, 1.75 × 10−2 ≤ R2 ≤ 2.25 × 10−2; and bin 3, 2.25 × 10−2 ≤ R2 ≤ 3 × 10−2. Logistic regression of status as eQTL or matched SNP was then carried out separately for each R2 bin, within each eQTL order.