To investigate whether the most significant variant from the meta-analysis (rs4714329) belonged to a longer haplotype, we performed LD and haplotype analyses for all of the four SNPs residing in the 6p21.2 genomic region. Two of the variants (rs6458146 and rs10498746) were located in the same haploblock (bp 40218128–40224268, GRCh37; Supplementary Figure 3). Also, rs9471290 (bp 40260515) was in a strong LD with those (D′=0.93 and 0.95, and LOD⩾2.0, respectively). Rs4714329 was, however, only in a moderate LD with the other 6p21.2 variants (D′=0.77, 0.68, and 0.66 to rs9471290, rs6458146 and rs10498746, respectively). A post hoc haplotype analysis of the SNPs with the most significant signal from the replication analysis yielded an association signal from a relatively frequent (f=0.31) haplotype of minor alleles A–G (rs9471290 and rs4714329, respectively) of OR=1.59 (P=5.6 × 10−6) in the GWAS data, and of OR=1.49 (P=0.0022) in the replication cohort (males and females combined). Thus, the signal for association from 6p21.2 was not strengthened by the haplotype and the risk genotype was best captured by the G-allele of rs4714329.
