Coeliac disease, or gluten-sensitive enteropathy (GSE), and inflammatory bowel disease (IBD) are two common gastrointestinal inflammatory disorders, both showing enhanced intestinal epithelial permeability.1 2 Some healthy first-degree relatives of IBD patients also display an impaired barrier function, suggesting that this is a heritable feature rather than acquired.3 In further support of this, it has been shown that in IBD cases in human4 and mouse,5 and in GSE cases in human6 and dog,7 an impaired intestinal permeability is present long before the onset of disease. Moreover, treatment of GSE patients on a gluten-free diet reverses the disease process but does not completely reduce the increased intestinal permeability.8 Functional abnormalities of tight junctions have also been observed in non-inflamed ileum of Crohn’s disease patients, compatible with a role in IBD pathogenesis.9 Similarly, altered expression, localisation and phosphorylation of epithelial junctional proteins have been observed in GSE.10 This could mean that the barrier defect observed in GSE and IBD is mediated through epithelial tight junctions.

Both GSE and IBD have a strong genetic component and there are currently multiple susceptibility loci in the human genome linked to either IBD (reviewed in Newman et al11) or GSE (reviewed in Van Heel et al12). These two studies show sharing of certain chromosomal regions that may predispose to both disorders—for example, a locus on 5q31–q33 and a locus on 19p13.13 GSE and IBD co-occur in families and patients, with an approximately fivefold increased prevalence of IBD in GSE patients.14^–17 This suggests that IBD and GSE share part of their genetic susceptibility.

Based on the above, genes encoding tight junction proteins should be considered highly relevant functional candidate genes for GSE and IBD. In this study we focused on 41 genes from the tight junction pathway, including those encoding transmembrane, adaptor, signal transduction and transcriptional regulatory proteins.18 19 By using genetic association analysis with a tag single nucleotide polymorphism (SNP) approach, we identified two adaptor protein-coding genes involved in coeliac disease in Dutch and British patients. One of these genes was also associated with ulcerative colitis in a Dutch patient cohort. This suggests that both disorders share genetic risk factors that point to the involvement of the epithelial intestinal barrier.

METHODS

Patients and controls

The cohorts used in this study are summarised in Supplementary table 1. DNA, isolated from whole blood, was available from a cohort of 463 Dutch coeliac disease patients20 and 752 British coeliac disease patients.21 The Dutch IBD case cohort has been described in detail elsewhere.22 23 Two cohorts of random Dutch blood bank donors were available: set 1 controls (n = 470, described by Monsuur et al20) and set 2 controls (n = 459, described by van Bodegraven et al23). The 1185 British controls were lymphoblastoid cell line DNA from the 1958 British Birth Cohort. All Dutch cases and controls were from The Netherlands, of European descent, and with at least three of the four grandparents also born in The Netherlands. This study was approved by the Medical Ethical Committee of the University Medical Center Utrecht, the VU Medical Center Amsterdam Ethics Committee, University Medical Center Groningen Ethics Committee and the Oxford Research Ethics Committee.

Comprehensive screen: tag SNP selection and genotyping

We selected 44 tight junction genes based on the tight junction network described in the literature,18 19 24 and the KEGG pathway database (www.genome.jp/kegg/). Three genes were excluded from further study: no tag SNPs could be selected for PARD6A and RAB13, while CLDN2 was localised on the X-chromosome. SNPs were selected by downloading all the SNPs typed in the CEPH (Centre d'Etude du Polymorphisme Humain) population (Utah residents with ancestry from northern and western Europe) that were located in the genomic sequence of the remaining 41 genes from the HapMap database (November 2004, Phase I, http://www.hapmap.org/).25 The program Tagger (available at http://www.broad.mit.edu/mpg/tagger/) was subsequently used to select tag SNPs, such that all SNPs with a minor allele frequency (MAF) ⩾10% were captured with r²⩾0.7 (excluding SNPs with low Illumina quality design scores). A few genes required too many tag SNPs to cover the entire genomic sequence. For these we tagged only the exons and exon–intron boundaries, and used a MAF ⩾20% (see Supplementary table 2). A final set of 211 tag SNPs was obtained for genotype analysis of 463 Dutch coeliac cases and set 1 controls (n = 470). See Supplementary table 3 for a detailed list of the tag SNPs used.

SNP genotyping was performed using the GoldenGate assay on an Illumina BeadStation 500 GX (Illumina Inc., San Diego, CA). All tag SNPs were examined for their resulting quality, and 13 SNPs which had a low signal or too wide clusters were excluded. Five tag SNPs were not in Hardy–Weinberg equilibrium (HWE) in the controls and were therefore excluded. This yielded a total of 197 SNPs that were successfully analysed (92%).

The five SNPs that revealed p<0.01 in the comprehensive screen were genotyped using Taqman assays (Assays on Demand; Applied Biosystems, Foster City, CA) in set 2 of the Dutch controls (n = 459), the British coeliac cases (n = 752) and controls (n = 1185), and the Dutch IBD cases (n = 1112). Due to technical problems, the second part of the IBD cohort failed for SNP rs1496770. Therefore, we only show the results for the first part of the cohort (IBD n = 588, Crohn’s disease n = 298, ulcerative colitis n = 290) for this particular SNP. No allele frequency differences were observed between the Dutch control cohorts (see Supplementary table 4). In addition, no population stratification was observed between Dutch GSE cases and Dutch controls (De Kovel et al, unpublished observations).

Statistical analysis

Association χ² values and two-tailed p values were calculated using the Haploview program (freely available at http://www.hapmap.org), for each stage of the study. SNPs that were not in HWE (p⩽0.01) in the controls were excluded from further analysis. We used multiple logistic regression analysis to estimate allelic and genotypic odds ratios (ORs) and the corresponding 95% CIs for the five SNPs tested in the follow-up studies. The analyses were performed using STATA statistical software, version 8.0 for MS Windows. The pooled analysis of the Dutch and British GSE cohorts was performed using Mantel–Haenszel statistics. This study has an 80% power to detect association assuming a relative risk of 1.5 and an allele frequency of ⩾10%.

RESULTS

Primary genetic screen with Dutch GSE patients

In a first screen, 197 tag SNPs selected from 41 genes from the tight junction network were genotyped in a cohort of 463 Dutch GSE cases and 470 Dutch controls (see Supplementary table 5). We observed evidence in favour of association (p_uncorrected<0.01) for five SNPs in two genes (PARD3 located on 10p11, and MAGI2 located on 7q21) where two would have been expected based on chance (rs10763976, rs4379776, rs6962966, rs9640699 and rs1496770; see table 1). PARD3 and MAGI2 both encode tight junction adaptor proteins that act as membrane-associated scaffolds. Given that these 41 tight junction genes were selected based on prior knowledge, it is not certain what level of proof one would require to prove association as the SNPs used in this study were selected from a limited number of candidate loci in close proximity to each other and therefore clearly not completely independent. Hence, a Bonferroni correction for the number of SNPs tested would be overly conservative. When corrected for multiple testing for the 41 functional candidate genes, SNP rs6962966 in MAGI2 remained significant (p_c<0.05/41 = 0.0012). To distinguish between true and false observations, we genotyped extra Dutch controls in stage 2 of the study as well as a cohort of British patients and controls.

Follow-up genetic analysis including British GSE patients

A follow-up study of the five SNPs with p<0.01 was performed by typing 459 extra Dutch controls, and further a second, fully independent, cohort of 752 unrelated British GSE patients and 1185 British controls. In table 1 we present the results for the separate and combined cohorts. Association analysis of the 463 Dutch GSE cases versus 929 Dutch controls improved the p values for two of the SNPs, rs10763976 in PARD3 and rs6962966 in MAGI2 (see table 1). The three SNPs in MAGI2 were not in linkage disequilibrium with each other (D′<0.06, r²<0.003), and only a weak correlation was observed between the two SNPs in PARD3 (D′ = 0.85, r² = 0.40). Genotype calculations suggested a dominant model for rs10763976 (p = 6×10⁻⁵) (see Supplementary table 6). Although the change in allele frequencies in the British GSE cases was in a similar direction to the Dutch GSE cases, it was less pronounced. The PARD3 SNP rs10763976 (p = 0.032) showed significant association, and rs4379776 (p = 0.082) showed a trend towards significance as did the MAGI2 SNP rs6962966 (p = 0.10). Combining the Dutch and British cohorts using a Mantel–Haenszel meta-analysis (in total 1215 cases vs 2114 controls) strengthened the association considerably and revealed a highly significant association for three of the five SNPs (table 1). We observed the smallest p value for rs10763976 (p = 6.4×10⁻⁵) located in intron 21 of PARD3. Individuals carrying the A allele have a modest but significantly higher risk of developing GSE (OR 1.23; 95% CI 1.11 to 1.37). The most significantly associated SNP in MAGI2 was rs6962966 (p = 7.6×10⁻⁴) located in intron 14; this SNP was associated with a 1.2-fold increased risk for GSE (OR 1.19; 95% CI 1.08 to 1.32). Since both these associated SNPs are intronic and no direct influence on protein function was predicted, they are more likely to be markers of disease susceptibility rather than being the actual causal functional variants.

Genetic association in Dutch IBD patients

We were then interested to test if these two tight junction adaptor proteins were also associated with IBD as a group, or its clinical subphenotypes Crohn’s disease and ulcerative colitis. In total, 607 Dutch Crohn’s disease and 505 Dutch ulcerative colitis cases were genotyped with the same five tag SNPs in PARD3 and MAGI2. The ulcerative colitis group showed significant association with rs6962966 in MAGI2 (p = 0.0036; OR 1.26, 95% CI 1.08 to 1.47) and a trend toward association with rs4379776 in PARD3 (p = 0.068; OR 1.17, 95% CI 0.99 to 1.38) (table 2). No association was observed for Crohn’s disease.

DISCUSSION

Weakening of the barrier function of tight junctions in the intestinal wall can result from natural interaction with pathogens,26 27 as well as from food components such as gluten.28 29 Our study shows for the first time that variants in tight junction genes contribute to the pathogenesis of the dissimilar gastrointestinal disorders coeliac disease and ulcerative colitis. The tagging strategy that was applied to test the 41 tight junction genes was robust for 36 genes. However, for five of the genes an alternative tagging strategy was applied with a MAF of >20% and only including exons and exon–intron boundaries, to accommodate the large size of the genes. Although both PARD3 and MAGI2 were among the five alternatively tagged genes, it cannot be excluded that association with the remaining three alternatively tagged genes was missed. In addition, this study has relatively high power to pick up associations of modest effect (OR>1.5) but might have missed weaker associations.

The two associated genes, PARD3 and MAGI2, encode adaptor proteins that are involved as scaffolding proteins in tight junction assembly. Several of the membrane-associated proteins of the tight junction complex interact with the actin cytoskeleton. Signal transduction to the actin cytoskeleton is important in regulating both tight junction assembly and function, with a pivotal role for the small GTPase RhoA.30 31 Myosin IXB might also play a role in this process, as it possesses a Rho-GTPase-activating domain (GAP) that negatively regulates Rho proteins. Previously, we demonstrated that the MYO9B gene is associated with GSE in Dutch patients20; although this was recently replicated in Finnish and Hungarian cohorts (Dr P Holopainen, personal communication), it could not be confirmed in studies in British,21 32 Swedish/Norwegian33 and Italian34 GSE patients. We further showed that MYO9B, which localises to the IBD6 locus on chromosome 19p13, was associated with ulcerative colitis and to a lesser extent with Crohn’s disease (which together comprise the IBD phenotype) in four different Caucasian populations—that is, Dutch, British, Italian and French-Canadian.23 We hypothesised that myosin IXB—which contains a Rho-GAP domain—is involved in intestinal permeability through remodelling of the epithelial cytoskeleton and tight junction assembly, through its interaction with RhoA.20 23 Since the primary barrier defects are expected to be subtle, we presume that the cycle of barrier malfunction is sustained by the subsequent local inflammations, since it is well established that proinflammatory cytokines—such as interferon γ and tumour necrosis factor—trigger further barrier dysfunction.

Both GSE and ulcerative colitis are characterised by a superficial inflammation restricted to the intestinal mucosa. In this respect it is interesting to note that the epithelial tight junction-related genes MAGI2 and MYO9B (and possibly PARD3) are collectively associated with both GSE and ulcerative colitis. Replication of these new associations in other populations, particularly that for ulcerative colitis, is needed to validate our findings further and to confirm the contribution to intestinal disease aetiology.

A healthy gut mucosa is characterised by a robust and selective barrier, maintained by properly differentiated and polarised epithelial cells. Weakening of the barrier due to external factors such as gluten or intestinal bacteria, in conjunction with enhanced sensitivity of the host, not only results in inflammation but also leads to a loss of epithelial differentiation and polarisation, resulting in clinical phenotypes characterised by nutrient malabsorption and diarrhoea. The outcome of the interaction between potentially harmful external factors and the mucosal barrier may be influenced by genetic variants in genes that code for the tight junction complex. In support of our finding is the reported association between IBD and the ABCB1/MDR1 gene, which encodes the P-glycoprotein 170, involved in defence against xenobiotics and bacterial products, and barrier maintenance.35 Additional support comes from the reported association between IBD and the gene DLG5, which encodes a scaffold protein involved in maintenance of epithelial cell contacts and polarity.36 Its impairment is also considered to interfere with the gut barrier function. The exact involvement of DLG5 is, however, not clear, as both positive and negative replication studies appeared in the literature.37^–40

In conclusion, we have demonstrated that both PARD3 and MAGI2 (together with the previously reported MYO9B20) are genetically associated with coeliac disease. Additionally, ulcerative colitis showed association with MAGI2 (and MYO9B23) and a trend of association with PARD3. This suggests that both intestinal disorders share a common aetiology through tight junction-mediated intestinal barrier impairment. The validity of this observation will, however, require replication in additional populations.