Fine-mapping in the MHC region accounts for 18% additional genetic risk for celiac disease

Abstract

Although dietary gluten is the trigger, celiac disease risk is strongly influenced by genetic variation in the major histocompatibility complex (MHC) region. We fine-mapped the MHC association signal to identify additional risk factors independent of the HLA-DQ alleles and observed five novel associations that account for 18% of the genetic risk. Together with the 57 known non-MHC loci, genetic variation can now explain up to 48% of celiac disease heritability.

Celiac disease is an autoimmune disease triggered by the consumption of dietary gluten in which the primary organ affected is the small intestine. The major histocompatibility complex (MHC) is the main genetic factor in disease development, with the strongest effects mapped to the classical HLA-DQA1 and HLA-DQB1 genes. Specifically, the common haplotypes DQ2.5, DQ2.2 and DQ8 have been shown to increase disease risk by 6-fold on average^1,2. Genome-wide association analyses have identified another 57 independent SNPs located in 39 risk loci outside the MHC^3,4. In this study, we hypothesized that a substantial fraction of celiac disease heritability might be explained by unknown independent variants across the MHC (but outside the classical HLA-DQ region). These variants might have remained undetected in previous genome-wide association studies due to the complexities of fine-mapping association signals in the extended MHC region. We therefore imputed the region and tested for associated SNPs, amino acid polymorphisms, and classical HLA alleles in 12,016 cases and 11,920 controls of European ancestry (in six cohorts from five countries), all genotyped on the Illumina Immunochip array (Supplementary Table 1, Supplementary Methods).

Because HLA class II haplotypes are established risk factors for celiac disease, we first focused on the region encompassing HLA-DRB1, HLA-DQA1 and HLA-DQB1. Consistent with functional studies, we confirmed strong associations for HLA-DQ (p = 10⁻¹⁷⁷ for a direct comparison between all classical 4-digit alleles of HLA-DQA1/DQB1 and of HLA-DRB1; Supplementary Methods, Supplementary Table 2). We also observed remarkable allele frequency differences across European populations (Supplementary Table 3, Supplementary Fig. 1). In a step-wise conditional analysis within the HLA-DQ heterodimer molecule, we identified amino acid positions 25 and 47 in HLA-DQα1, and positions 57 and 74 in HLA-DQβ1, as the strongest independent associations with celiac disease risk (Supplementary Fig. 2, Supplementary Table 4). The haplotype model based on these four associated amino acid positions, of which β57 and β74 are thought to make direct contact with the epitope in the HLA-DQ binding groove, almost perfectly recapitulates the associations with the classical HLA-DQA1/DQB1 alleles (99.6% of the explained variance). For example, the haplotype defined by HLA-DQA1-Tyr25/Cys47 and HLA-DQB1-Ala57/Ala74 completely mirrors the risk conferred by HLA-DQ2.5 (amino acid haplotype OR=19.13, P= 10^−1,754; HLA-DQ2.5 OR=21.56, P=10^−1,801) (Supplementary Tables 2 and 4).

In order to identify additional associations beyond HLA-DQ, we tested all the other variants across the MHC region after correcting for the collective effects of all classical HLA-DQA1 and DQB1 4-digit alleles. We identified five further independent variants with genome-wide significance: amino acid position 9 in HLA-DPβ1 (omnibus p = 10⁻⁴¹), classical alleles HLA-B*08:01 (p = 10⁻²²) and HLA-B*39:06 (p = 10⁻⁸), and two SNPs, rs1611710 (p = 10⁻⁹) and rs2301226 (p = 10⁻⁹) (Fig. 1, Supplementary Fig. 3, Supplementary Table 5). eQTL analysis of publicly available datasets⁵ demonstrated a cis-eQTL effect of rs1611710 on HLA-F expression (eQTL p = 1.22×10⁻⁵⁶) and of rs2301226 on B3GALT4 (eQTL p = 1.67×10⁻³⁹) and on HLA-DPB1 expression (eQTL p = 3.87×10⁻¹³). When we did not adjust a priori for the effects of HLA-DQ alleles, we found the same five independent effects, in addition to several strong effects in the class II region (Supplementary Fig. 4 and Supplementary Table 5), providing support for the robustness of these novel findings.

Figure 1.

Schematic representation of the stepwise analysis applied in fine-mapping the HLA region. HLA-DQA1 and HLA-DQB1 4-digit alleles were included in the model to control for their effect (red arrow). A green flag represents each of the steps in the analysis and an independent association.

Based on all the genetic associations for celiac disease risk, we estimated the genetic variance explained (h²) with a liability threshold model, assuming a disease prevalence of 1%^6–10 (Table 1 and Supplementary Methods). The classical HLA-DQ loci alone explain h² = 0.232 on the liability scale, whereas the five novel variants reported here explain an additional h² = 0.180, which is more than twice that explained by the 57 independent SNPs outside the MHC³ (h² = 0.065). In total, all the variants identified to date as being robustly associated with celiac disease risk explain approximately h² = 0.477. These liability estimates are more modest than those previously published (~87% explained genetic risk¹¹). This discrepancy illustrates the difficulty of comparing explained variance estimates between studies if the model assumptions are not the same¹². When we changed the assumed prevalence to 2% and 3%, as reported in some populations (Supplementary Methods), we observed substantially increased heritability estimates, which illustrates not only how sensitive the estimates are to this parameter, but also how estimates may change in populations with an empirically higher disease prevalence (Supplementary Table 6).

Table 1.

Heritability estimates for HLA and non-HLA variants in celiac disease.

	h2	h2
Model	Observed scale	Liability scale
Classical HLA-DQ alleles (4-digit)	0.316	0.232
Independent MHC variants outside HLA-DQ	0.357	0.180
Non-MHC variants (57 SNPs)	0.15	0.065
Combined	0.823	0.477

Although the importance of the MHC region in celiac disease and other autoimmune diseases is widely recognized, the particular properties of the region¹³ have made it difficult to identify independent loci and pinpoint functional variants that play a role in disease pathology. Taking advantage of high-density imputation¹⁴, we implicated a set of five variants (Fig. 1, Supplementary Figs. 2 and 3) that act independently of HLA-DQ and which can explain an additional 18% of celiac disease heritability. On the basis of these results, it seems likely that fine-mapping the MHC region in other diseases with classical HLA associations will reveal some of the “missing” heritability.

Finally, this study provides evidence that the MHC class I region is also an important locus in celiac disease etiology. This is an interesting observation given the presumed role for intraepithelial lymphocytes (IELs) TCRαβ⁺CD8αβ⁺, which have been shown to exacerbate celiac disease and are restricted to MHC class I recognition¹⁵. These cells represent a subtype of antigen-experienced “innate-like” T cells and, under conditions of homeostasis, they are located within the intestinal mucosa, maintaining epithelium integrity and preventing pathogen incursion. The TCRαβ⁺CD8αβ⁺ are most abundant in the intestinal mucosa (70–80% of the IEL subset) and are a hallmark of celiac disease due to their correlation with gluten intake, thereby providing further support for the role of IELs in the etiology of celiac disease.