Variants in the original VCF files were renamed to format “type_chr_pos” (e.g. ‘snp_1_236887241’ or ‘indel:2D_1_18847945’), and filtered for polymorphic and bi-allelic sites with VCFtools v.0.1.12b 70 (vcftools –gzvcf IN --mac 1 --min-alleles 2 --max-alleles 2 --recode --recode-INFO-all –out OUT). The resulting VCF files were then converted to ‘012’ format (vcftools --gzvcf IN --012 --out OUT.012), where 0, 1, and 2 represent the number of non-reference alleles, and further to HDF5 format using a converter function (-g012) from LIMIX (). This resulted in a set of N = 14,644,791 variant sites. To obtain allele dosages, genotype likelihoods (GL) in the original VCF were converted to genotype probabilities (GP) with BCFtools (bcftools +tag2tag IN -O z -o OUT -- --gp-to-gl) and used to define allele dosage as follows: Dosage of alternative allele = GP(REF/ALT) + 2*GP(ALT/ALT). Genotype dosage information was converted to HDF5 format using LIMIX converter (LIMIX converter, --g012_dosage). For eQTL mapping, we included autosomal variants with a minimum minor allele frequency of 1% in our samples and maximum 10% missing values across individuals. Variant sites were further required to have
