DNase-seq libraries were created as previously described29, with small modifications. Each library was sequenced on at least two lanes of an Illumina GAIIx. Resulting 20bp sequencing reads were mapped to the human genome sequence (hg18) using an algorithm that we designed specifically to eliminate mappability biases between sequence variants. We divided the genome into 100 bp windows and selected the top 5% in terms of total DNaseI sensitivity. DNaseI sensitivity for each individual in each window was normalized by the total number of mapped reads for that individual. For QTL mapping, the data were further rescaled within and across individuals, and we adjusted the data for an observed individual × GC interaction, as well as for the top four principal components of the DNaseI sensitivity matrix. Genotypes for all available SNPs and indels were obtained from HapMap and 1000 Genomes data and imputed where necessary6,7,30. We performed DNase-seq association mapping by regressing the adjusted sensitivity in each window against the genotypes at variants in a 40 kb region centered on each DHS. As validation, we used our DNase-seq reads as
