Association of DLG5 and inflammatory bowel disease across populations

In a recent report Daly et al¹ replicated a previously reported association² of the R30Q variant in the DLG5 gene with inflammatory bowel disease (IBD) in case–control samples from pooled Canadian and Italian populations. However, the groups were unable to replicate the findings in cases and controls from the United Kingdom. Similarly, Noble et al³ could not replicate the original association of R30Q to IBD reported by Stoll et al.² Here, we present a combined analysis of the four data sets from cases and controls from Scotland, UK (excluding Scotland), pooled Canada/Italy and Germany. We used the meta-package of R software to perform a meta-analysis of all these studies. We found a substantial (I²=83.35; 95% CI: 57.5–93.4%) amount of heterogeneity among studies. This heterogeneity was statistically significant (Cochran's Q statistic=17.96; P=0.0004). A weak, but significant (P=0.035), association of the R30Q variant and IBD was found under a fixed effect model but not under a random effects model. Under the fixed effect model, the R30Q variant was associated with an odds ratio (OR) of 1.19 (95% confidence interval (CI): 1.01–1.39). The numbers used in the analysis are shown in Table 1.

Table 1 Number of alleles reported for each population in the original studies

We further investigated the source of the heterogeneity among studies by comparing the distribution of alleles in the different populations for both cases and controls. Overall there were significant differences in the distribution of alleles in the four populations (Fisher Exact test based on a 4 (populations) × 2 (alleles) contingency table) for both cases (P=0.022) and controls (P=0.003). Table 2 shows the significance level of the six possible pairwise comparisons among the four populations using Fisher Exact test. It is worth noting that allele frequency differences among cases could be detected only for the UK/Germany comparison, whereas significant (or almost significant) differences could be detected among controls (again with the exception of the German/UK comparison). In summary, there are significant population allele frequency differences at the DLG5 gene (assuming the control sample is a representative sample of the general population) among the populations presented here. In fact allele frequencies at the DLG5 gene among IBD patients in different populations tend to be more similar than among controls. This is, in our view, inconsistent with the view of Daly et al¹ that heterogeneous replication results might be due to phenotypic differences among studies.

Table 2 Pairwise comparisons among populations

These population differences in allele frequency in both cases and particularly in controls are intriguing because they have relevance to pooling data from different populations in an attempt to identify novel loci contributing to complex disease and because genetic effects (measured as the allelic OR) will change with the population allele frequency (unless the locus mode of inheritance is multiplicative). This analysis suggests that pooling of genotyping data from studies in different populations might mask true but different effects among populations and that researchers will need to be very careful when pooling data from different populations.