Trachoma and Ocular Chlamydial Infection in the Era of Genomics

Figure 2

Evidence for MMP9-IL10 epistasis in Gambians with trachomatous scarring. The protective effect of MMP9 allele (Q279R) is modulated by host genetic background at the IL10 locus such that protective effects of the G allele are lost in the presence of either of 2 minor frequency risk alleles (IL10-1082C or IL10-3575A). The interaction between these nonallelic genes (or risk genotypes) has a dominant effect over other combinations. The interaction between risk genotypes was examined by conditional likelihood ratio tests (LRT) (main effects) log p/1 − p = a + b (SNP1) + c (SNP2). Interaction terms were defined as Log p/1 − p = a + b(SNP1) + c(SNP2) + d(SNP1SNP2) and when significant identified statistical epistasis. This approach was applied to 651 Gambian case-control pairs of TS. The MMP9 Q279R and IL10-3575 loci showed strong evidence for statistical interaction affecting risk of TS (LRT , ). Carriers of the (protective) MMP9 Q279R G allele who also had the IL10-3575A minor allele were at significantly increased risk of TS (OR = 1.83 (1.06–3.19)) when compared to subjects with the IL10-3575 T allele (common allele) (OR = 0.84 (0.69–1.02)). The IL10-1082 C minor allele in combination with the MMP9 Q279R G allele had an increased risk of TS (OR = 1.51 (1.02–2.24)) relative to IL10-1082 C in the presence of the MMP9 Q279R A allele (OR = 0.82 (0.67–1.01)) (LRT , for the interaction between the IL10-1082 and MMP9 risk alleles). Interaction between IL10-1082 and MMP9 Q279R affects risk despite the null single SNP main effect for IL10-1082 [71]. The individually protective MMP9 Q279R G allele was therefore associated with an increased risk of scarring in the presence of IL10 risk alleles (IL10-1082C or IL10-3575A minor alleles) and a decreased risk in the presence of common IL10 (protective) alleles. Similar modelling at other loci previously investigated in this cohort (IFNγ-1616, +3234; LTA -252, +77; IkBL -63; IL-8 -251; GM-CSF2 27348, 27438) showed significant or marginally significant evidence for two-way interactions, at the genotype or allelic level, with the MMP9 Q279R SNP. Some of these SNPs are in high LD and therefore not all the hypotheses tested are independent. The existence both of LD between loci and of potential biological interdependence between loci raises methodological difficulties in correction for multiple testing. We did not attempt any correction for multiple testing: and therefore a contribution of chance to these results is difficult to exclude as we point out in the main text. Comparing main and additional two-way epistatic effects in the final model suggested that the inclusion of interaction terms improved the fit of the model to the data, so that the final best model included both main and epistatic effects. For TS this model suggested that two-way interactions of MMP9-Q279R with IFNγ-1616, IFNγ+3234, IL10-1082, IL8-251, LTA+77, LTA-252, and IkBL-63 improved the fit of the model (data courtesy of Natividad, Mabey, Holland, and Bailey).