|Home | About | Journals | Submit | Contact Us | Français|
In their recent article, Castellanos-Rubio et al1 use genome-wide expression profiling combined with published data on genomic regions showing modest linkage to celiac disease as a strategy to identify inherited genetic variants predisposing to disease. They investigated 361 identified variants in a small case-control collection (262 cases, 214 controls) of Spanish origin. After quality control procedures 330 single nucleotide polymorphisms (SNPs) were available for association testing: 10 SNPs from 6 different genes/regions were stated to show evidence of association with celiac disease (uncorrected P < .005), with 4 SNPs remaining significant after statistics were corrected for multiple testing. They suggested these preliminary findings should be validated and tested in different populations.1
We recently carried out a genome-wide association study of 310,605 SNPs in 778 celiac cases and 1422 population controls from the British population.2 Such studies have recently been shown to be highly effective to identify risk variants for common disease.3 We previously directly genotyped rs365836 in our UK celiac/control genome-wide study,2 and have now imputed data for the other 9 SNP markers analyzed by Castellanos-Rubio et al1 using algorithms implemented in PLINK v184.108.40.206 We found no evidence (Table 1, uncorrected P>.05) in the UK collection for 9 of the 10 associations reported in the Spanish collection. Although rs6747096 was borderline significant (uncorrected P = .016) in the UK collection, the effect is in the opposite direction (rs6747096 G allele is more frequent in UK cases versus controls, yet less frequent in Spanish cases versus controls).1
There are a number of possible reasons for these apparent discrepancies. First, the results of the study by Castellanos-Rubio et al1 could be due to type 1 statistical error arising through the multiple hypothesis testing of the 330 SNPs investigated. The authors did correctly apply a Bonferroni correction. Second, UK data may represent a type 2 error. However, the UK collection is of much larger sample size, and power calculations suggest adequate power (>80% at P < .05 for all markers, using a multiplicative model and assuming celiac disease population prevalence of 1%) to detect effects of the allele frequencies (as observed in UK controls) and odds ratios as reported by Castellanos-Rubio et al.1 A third possible explanation is the presence of genetic heterogeneity between the British and Spanish populations. However, in our genetic investigations of celiac disease to date across Irish, British, and Dutch populations (~8000 samples genotyped for >1000 SNPs) we have not yet observed any evidence for heterogeneity at 8 non-HLA celiac disease associated regions.5 Broadly similar disease prevalence and clinical features are seen across European populations suggesting (with the possible exception of HLA-DQ) that genetic heterogeneity is at most a minor issue.
Multiple risk variants for common human diseases have recently been identified by genome-wide association studies.6 Most of these findings do not map to regions previously identified by genetic linkage studies. In celiac disease, consistent findings from linkage studies have not been obtained. We feel the strategy pursued by Castellanos-Rubio et al1 (prioritizing linkage regions for association studies) is suboptimal.
As Castellanos-Rubio et al discuss,1 there are often discrepancies between the findings of candidate gene association studies. We highlight the importance of carrying out large, well-designed, genome-wide studies with findings replicated in at least 1 other population. (Table 1)
Support and funding provided by Coeliac UK and The Wellcome Trust.