Risk Heterogeneity between PsA and PsC Is Explained by HLA-B
Next, we conducted an analysis focusing on subphenotypes of PsA and PsC. Overall, the association results of the PsA case-control analysis and of the PsC case-control analysis were similar to those of the PsV case-control analysis (Figures 1A–1C). The classical HLA-C∗06:02 allele demonstrated the lowest p value among the HLA variants in the MHC region. Stepwise association analyses, separately conducted for the PsA case-control analysis and the PsC case-control analysis, both revealed independent contributions of other class I HLA genes (HLA-B and HLA-C; Figure S2; Table S2). We did not observe independent signals for HLA class II genes in this stratified analysis, perhaps as a result of reduced statistical power.
Surprisingly, when we directly assessed comparative risk between PsA and PsC subjects, we found the lowest p value of the nominal association signal at HLA-B amino acid position 45 (pomnibus = 2.2 × 10−11; Figures 1D and 2F) rather than at HLA-C alleles. After conditioning on HLA-B amino acid position 45, or all classical HLA-B alleles, we observed no significant association in the MHC region (p > 5.0 × 10−8; Figures 2G and 3E). Of the HLA-B amino acid residues at position 45, HLA-B Glu45 increased PsA susceptibility in comparison to PsC susceptibility (OR = 1.46, 95% CI = 1.31–1.62, p = 2.9 × 10−12; Table 2; Table S2). Examining the classical alleles, we noted that HLA-B∗27, a risk allele for another arthritic disease, ankylosing spondylitis (MIM 106300),33 demonstrated the lowest p value for PsA versus PsC association (p = 1.2 × 10−4; Figure 4; Table S4) but much less significantly than HLA-B Glu45. We note that HLA-B∗27, along with HLA-B∗38, HLA-B∗39, and a number of other alleles, carries Glu at position 45.
Previous studies comparing PsA and PsC have suggested that HLA-C∗06:02 shows increased PsC risk but decreased PsA risk,13,14 and our study replicated this finding in a concordant directional effect (p = 9.4 × 10−6 for PsA versus PsC individuals). However, the differential impact of HLA-C∗06:02 on PsA and PsC risk disappeared after we conditioned on HLA-B amino acid position 45 (p = 0.12), suggesting that this reduced effect was the result of linkage to HLA-B. In contrast, the effect of HLA-B amino acid position 45 with respect to PsA versus PsC risk retained significance even after we conditioned on HLA-C∗06:02 (pomnibus = 2.2 × 10−7).
We evaluated effect-size (OR) heterogeneity in classical four-digit alleles of the HLA genes between the two association analyses of PsA-affected versus control individuals and PsC-affected versus control individuals. Among the eight class I and class II HLA genes that we evaluated in the overall PsV case-control analysis, the HLA-B and HLA-C alleles showed significant risk heterogeneity (pheterogeneity = 5.8 × 10−14 and pheterogeneity = 2.9 × 10−6, respectively, with a significance threshold of p < 0.05/8 = 0.0073). When we conditioned on HLA-B amino acid position 45, risk heterogeneity diminished in both HLA-B and HLA-C classical alleles (pheterogeneity > 0.01). In contrast, when we conditioned on HLA-C∗06:02, risk heterogeneity diminished for the classical HLA-C alleles (pheterogeneity = 0.018), but not in the classical HLA-B alleles (pheterogeneity = 6.5 × 10−5). These results demonstrate that the risk heterogeneity between PsA and PsC primarily derives from HLA-B, but not HLA-C (or other), genes.