Although this study has several strengths, there are limitations. First, and most notably, the SNPs with the best evidence for replication were not the top SNP associations from the discovery analysis. For replication, we took all SNPs with P<1x10-3 that were within 3MB of the top discovery SNP for each signal based on the recognition that variants with the top statistical association signals and the underlying true causal variants may not be the same.[29,30] Although this is a broad replication strategy and the meta-analysis P value does not meet genome-wide significance (P = 4.47x10-7 vs. P<5.0x10-8), we applied appropriate multiple testing correction and identified a SNP association that surpassed the significance threshold for replication. Haplotype analyses of the top replication SNP (rs4878712) and the discovery SNP on chromosome 9 (rs1329568) suggested a stronger association when considering the paired protective alleles (meta-analysis P = 5.44x10-8) than rs4878712 alone (meta-analysis P = 4.47x10-7), which may indicate a shared haplotype with a causal variant representing a single signal. Second, although different types of HIV exposure were present in both the discovery and replication
