To conduct haplotype analyses raw genotype data is needed. Chromosome 6 genotypes were available from UHS, VIDUS, ODB, Yale-Penn, CATS and Kreek (Supplementary Table 1). These cohorts’ data were phased with Eagle v2.4 via the Michigan Imputation Server65. Haplotypes for samples from each study were constructed by extracting OPRM1 variants that were GWS in the gSEM analysis and concatenating ordered by genomic position. Supplementary Table 24 provides counts for the various haplotypes observed across the studies. The 3 most common haplotypes, which accounted for 98% of observed haplotypes, were tested for association with OA in R adjusting for sex and genotype principal components. Only individuals carrying exclusively these haplotypes were included in the analysis. Two models were run, one in which the haplotype containing all major alleles served as the reference haplotype and one in which the haplotype containing the minor rs1799971-G allele served as the reference. This approach provided three effective comparisons: (a) rs1799971-G haplotype versus the major allele haplotype; (b) minor allele+rs1799971-A haplotype versus the major allele haplotype; (c) minor allele+rs1799971-A haplotype versus rs1799971-G haplotype. Individual cohort results were combined in an inverse variance-weighted meta-analysis using METAL (N = 21,037).
