Celiac Disease as a Genetic Predisposing Factor for Dermatitis Herpetiformis | CCID
Introduction
Celiac disease (CD) is a chronic autoimmune condition triggered by the ingestion of gluten, a protein complex found in wheat, rye, and barley, and represents one of the most prevalent inflammatory disorders affecting the small intestine.1,2 This disease is marked by characteristic mucosal damage and impaired nutrient absorption in genetically predisposed individuals. The pathological mechanism is linked to an immune response activated by dietary proteins rich in proline and glutamine, commonly referred to as “gluten”.3 Typically, CD manifests during early childhood, around the age of 2, although a secondary peak in diagnosis is observed in individuals around the age of 40.4 The clinical presentation of CD predominantly stems from nutrient and vitamin malabsorption, which results in a spectrum of symptoms including abdominal pain, bloating, diarrhea, unintentional weight loss, anemia, edema, and musculoskeletal complaints such as bone or joint pain.5 The severity of gastrointestinal symptoms can vary widely; some individuals may experience subtle or nonspecific complaints, while others remain asymptomatic for extended periods despite the presence of significant mucosal lesions. Furthermore, CD is not confined to gastrointestinal manifestations. Extraintestinal complications, including osteoporosis, dental enamel defects, or neurological involvement affecting either the peripheral or central nervous system, can also emerge.6 These systemic manifestations often contribute to delayed or missed diagnoses, highlighting the importance of heightened clinical awareness.
The prevalence of clinically recognized CD varies geographically, ranging from 1 in 270 in Finland to 1 in 500 in North America.7 Globally, CD affects 0.5% to 1.0% of the population, highlighting its significance as a widespread health concern.1 The onset and progression of CD are influenced by genetic, environmental, and immune factors.3 Genetic susceptibility, as evidenced by familial clustering and a 70%-75% concordance rate among monozygotic twins, plays a central role in CD pathogenesis.3 Environmental triggers, such as gluten exposure, and immune dysregulation further complicate disease development. Untreated CD is associated with life-threatening long-term complications, including increased risk of secondary autoimmune conditions, small bowel adenocarcinoma, enteropathy-associated T-cell lymphoma, and other lymphoproliferative malignancies, such as non-Hodgkin lymphoma.8–10 Epidemiological studies suggest individuals with CD have approximately double the cancer risk of the general population, underscoring the importance of early diagnosis.8,10 Beyond health implications, CD imposes significant societal burden, with systemic manifestations involving multiple organ systems. Both genetic and environmental factors are crucial to its etiology, emphasizing the need for ongoing research and public health initiatives focused on prevention and early detection.5
Dermatitis herpetiformis (DH), a cutaneous manifestation of CD, is an autoimmune disorder marked by intensely pruritic, blistering eruptions, typically on extensor surfaces like elbows, knees, and buttocks.11 The rash is polymorphic, including blisters, erythematous papules, and plaques, but severe itching and scratching often lead to excoriations, crusting, and post-inflammatory hyperpigmentation, obscuring primary lesions.12 DH and CD share genetic predisposition, small intestinal mucosal changes, and autoimmune responses involving antibodies against tissue transglutaminase, highlighting their shared pathophysiology.13 CD is often diagnosed in childhood, whereas DH typically presents around age 50, affecting adults more than children, with men showing slightly higher susceptibility.12,13 DH prevalence varies, reaching 75 cases per 100,000 in some populations, with 13% of CD patients manifesting DH.14,15 Incidence rates are 2.7 per 100,000 annually in Finland and 0.8 per 100,000 in the UK.13–15 A key diagnostic feature is immunoglobulin A (IgA) deposits in the papillary dermis, identifiable by direct immunofluorescence microscopy, aiding in differentiation from other dermatological conditions.12 Despite a nearly fourfold increase in CD prevalence, DH incidence has declined, likely due to improved detection of asymptomatic cases through serological screening.14,15 The factors determining why only some individuals with undiagnosed or untreated CD develop DH remain unclear. The deposition of IgA and the cutaneous manifestations likely involve complex interactions between genetic, immune, and environmental factors. Investigating the genetic basis of the CD-DH relationship could provide insights into their shared and divergent pathways.
Materials and Methods
Data Source
The IEU OpenGWAS database (https://gwas.mrcieu.ac.uk/) is an open-source platform maintained by the research team of the University of Edinburgh in the UK, which is mainly used to provide public data related to GWAS. The GWAS summary data for CD analyzed in this study were obtained from the IEU OpenGWAS database. The dataset comprises 15,283 participants, including 4,533 cases and 10,750 controls, and encompasses a total of 523,399 SNP loci. The cohort included both male and female participants, all the participants are of European descent. Genotyping was conducted using the Illumina GoldenGate BeadXpress assays at multiple locations, including London, Hinxton, and Groningen. Imputation of genotypes for samples initially genotyped on the Hap300 platform was performed using BEAGLE software, employing CEU, TSI, MEX, and GIH reference panels from HapMap3.20 The FinnGen database (https://www.finngen.fi/) is a large-scale genomic data resource jointly established by several universities, research institutions and hospitals in Finland. Its purpose is to explore the relationships between genetic factors and diseases, as well as health characteristics, through GWAS. For DH, GWAS summary data were sourced from the FinnGen consortium. This dataset includes 218,344 individuals, of whom 278 were identified as cases and 218,066 as controls, covering a comprehensive 16,380,466 SNP loci. Both male and female participants were included in the study, and all participants were of European descent. And DH cases were classified based on the M13 code from the International Classification of Diseases, 10th Edition (ICD-10). Genotyping utilized Illumina and Affymetrix chip arrays, supplied by Illumina Inc. (San Diego, California) and Thermo Fisher Scientific Inc. (Santa Clara, California, USA). Additional details concerning the dataset can be found on the FinnGen consortium’s official website. The analyzed GWAS summary data are publicly available and derived from European populations, thus precluding the necessity for ethical approval or informed consent. A comprehensive description of the data utilized is provided in Supplementary Table 1 (https://www.dovepress.com/article/supplementary_file/556046/556046%20Revised%20Supplementary%20File.doc) .
IVs Selection
Throughout this analysis, we strictly adhered to the core assumptions of MR to ensure the validity of our findings: 1) IVs must demonstrate a strong and significant association with the exposure factors; 2) IVs should not be associated with the outcome or confounding variables; and 3) the effect of IVs on the outcome should occur exclusively through the exposure factors. To enhance the reliability of the genetic causal inferences, we employed a series of stringent quality control measures during the IV selection process. First, IVs were required to exhibit a robust association with the exposure factors (P < 5 × 10^-8, F-statistic > 10), with the F-statistic calculated as F = R2(N−K−1)/K(1−R2).21 In instances where the number of IVs meeting this criterion was insufficient, adjustments to the p-value thresholds were guided by relevant literature. Second, to address potential biases arising from linkage disequilibrium (LD) between SNPs, we applied an LD threshold of r2 < 0.001 within a 10,000-kb genomic window.22 Third, in cases where target SNPs were unavailable in the summary GWAS data, proxy SNPs were identified using LDlink online platform.23 Fourth, we ensured that the selected IVs were independent of the outcome, applying the same significance threshold (P < 5 × 10^-8) as for exposure factors. Fifth, to control for potential confounding effects, SNPs associated with confounders were excluded. Confounders for forward MR analysis included factors such as gluten diets and iodide, which are linked to DH.12,24 For reverse MR analysis, confounders included gluten diets, vitamin D deficiency, and gastrointestinal infections, known to be associated with CD.25 Relevant confounding SNPs were identified and excluded using data from the GWAS Catalog database. Lastly, palindromic SNPs with intermediate allele frequencies were excluded to avoid strand ambiguity, ensuring alignment between the alleles associated with exposure and those linked to outcomes.26
MR Analysis
To investigate the genetic causal relationship between CD and DH, we employed a comprehensive array of MR methods. Specifically, eight distinct MR approaches were utilized: MR Egger, random-effects inverse variance weighted (IVW), weighted median, simple mode, weighted mode, maximum likelihood, penalized weighted median, and fixed-effects IVW. Among these, IVW was designated as the primary analytical framework. In cases where results from other methods diverged from those of the IVW analysis, with IVW serving as the reference standard in such cases. The IVW method can be further stratified into random-effects and fixed-effects models, depending on the heterogeneity detected in the data. When significant heterogeneity was present, the random-effects IVW model was prioritized to account for variability, while the fixed-effects IVW model was applied under conditions of lack heterogeneity. Data analysis was performed using R software (version 4.1.2), leveraging the “TwoSampleMR” package for conducting two-sample MR analyses. Statistical significance was determined using a P-value threshold of < 0.05, indicating a robust genetic causal relationship between exposure and outcome. Additionally, an odds ratio (OR) greater than 1 was indicative of a positive causal relationship, whereas an OR less than 1 suggested a negative causal association.
Sensitivity Analysis
To ensure the robustness and reliability of the findings derived from MR analysis, we conducted a series of rigorous sensitivity analyses. Heterogeneity in the MR estimates was evaluated using two distinct approaches: Cochran’s Q statistic applied to the MR-IVW method and Rucker’s Q statistic used within the MR-Egger framework.27 To further assess horizontal pleiotropy, we employed the MR-Egger intercept test alongside the global test provided by the MR Pleiotropy Residual Sum and Outlier (MR-PRESSO) framework.19 Radial IVW methods were utilized to visually detect potential outliers,28 while the distortion test embedded within MR-PRESSO was applied to systematically identify outlying variants influencing MR estimates.19 If there are outliers, a second round of MR analysis is performed after removing them. Additionally, a leave-one-out analysis was conducted to evaluate the influence of individual SNPs on the overall MR results, ensuring that no single SNP disproportionately impacted the findings.29 Lastly, the Shapiro–Wilk normality test, incorporated in the MR Robust Adjusted Profile Score (MR-RAPS) method, was employed to confirm that the MR estimates followed a normal distribution, further validating the statistical properties of the results.28 Criteria for heterogeneity, pleiotropy, and distribution assessments were standardized, with P-values > 0.05 indicating the absence of significant heterogeneity and horizontal pleiotropy, as well as compliance with the assumption of normality. These comprehensive analyses reinforced the reliability of the genetic causal inferences in the study.
Results
Genetic Causality Between CD and DH (Forward MR Analysis)
Through the application of a rigorous significance threshold (P < 5 × 10−8) and ensuring the F-statistic exceeded the critical value of 10, we identified a total of 12 SNPs that demonstrated a significant association with CD. Importantly, all 12 SNPs were cross-referenced and confirmed in the GWAS summary data for DH, with no proxy SNPs detected. These genetic variants were also assessed for associations with DH and other potential confounding factors, none of which were identified as significant. Further scrutiny revealed that none of the selected SNPs exhibited palindromic properties, ensuring the integrity of the subsequent instrumental variable analysis. As a result, the final set comprised 12 SNPs deemed suitable as IVs for robust evaluations of genetic causality between CD and DH (Supplementary Table 2 (https://www.dovepress.com/article/supplementary_file/556046/556046%20Revised%20Supplementary%20File.doc) ).
The random-effects IVW analysis revealed a significant positive genetic causal relationship between CD and DH (P = 0.001, OR 95% confidence interval [CI]: 1.546 [1.195–1.999]). Consistent results were obtained using fixed-effects IVW analysis, which also identified a significant positive genetic causal association (P = 0.001, OR 95% CI: 1.546 [1.195–1.999]). Among the six additional MR methods, only MR Egger and the weighted mode indicated no evidence of a causal relationship, while the remaining four methods provided support for a positive genetic causal relationship between CD and DH (Figures 1 and 2A). Cochran’s Q statistic from the MR-IVW method and Rucker’s Q statistic from MR Egger suggested no significant heterogeneity (P > 0.05). Additionally, assessments for horizontal pleiotropy, including the intercept test from MR Egger and the global test from MR-PRESSO, found no evidence of pleiotropy (P > 0.05) (Table 1). Radial MR evaluations, conducted through IVW and MR Egger delineations, confirmed the absence of outliers in the genetic instrumental variables used for MR analyses (Figures 2B). Moreover, MR-PRESSO’s distortion test detected no outlier variants (Table 1). The robustness of these findings was further supported by leave-one-out analysis, which demonstrated that no individual SNP unduly influenced the observed positive genetic causal relationship between CD and DH (Figure 2C). Furthermore, MR-RAPS analysis confirmed the normality of the causal estimate distribution based on the Shapiro–Wilk test (P > 0.05) (Figure 2D and Table 1). In summary, both random-effects and fixed-effects IVW analyses robustly demonstrate a significant positive genetic causal relationship between CD and DH. The comprehensive sensitivity analyses, showing no evidence of heterogeneity, horizontal pleiotropy, or undue SNP influence, confirm the reliability of these findings. Based on these results, we validate the fixed-effects IVW analysis as a strong indicator of a positive genetic causal relationship between CD and DH.
Table 1 Sensitivity Analysis of the MR Analysis Results of Celiac Disease and Dermatitis Herpetiformis
Figure 1 Genetic causal analysis of the relationship between celiac disease and dermatitis herpetiformis was conducted using eight distinct methods: MR Egger, random-effects IVW, weighted median, simple mode, weighted mode, maximum likelihood, penalized weighted median, and fixed-effects IVW. *Second round of genetic causal assessment of dermatitis herpetiformis to celiac disease after removing an outlier.
Figure 2 Genetic causal assessment of celiac disease to dermatitis herpetiformis. (A) scatter plot; (B) radial plot; (C) leave-one-out analysis; (D) normal distribution
Genetic Causality Between DH and CD (Reverse MR Analysis)
Given the stringent significance threshold of P < 5 × 10−8, the number of SNPs identified for MR analysis was insufficient. To address this limitation, we relaxed the significance threshold to P < 6 × 10−5, following established methodologies.30,31 This adjustment resulted in the identification of 86 SNPs significantly associated with DH, all meeting the criteria of P < 6 × 10−5 and demonstrating F-statistics greater than 10. Subsequent matching of these SNPs to the GWAS summary data for CD yielded 13 SNPs suitable for further analysis. A thorough examination revealed no proxy SNPs and no associations of these SNPs with either CD or potential confounding factors. Additionally, three palindromic SNPs (rs1012753, rs12515176, rs312026) were excluded to ensure the robustness of the analysis. As a result, a final set of 10 SNPs was identified and selected as IVs for subsequent causal inference studies (Supplementary Table 3 (https://www.dovepress.com/article/supplementary_file/556046/556046%20Revised%20Supplementary%20File.doc) ).
The random-effects IVW analysis revealed no significant evidence of a genetic causal relationship between DH and CD (P = 0.555, OR 95% CI: 1.017 [0.962–1.075]). Similarly, the fixed-effects IVW analysis did not identify a genetic causal association between these conditions (P = 0.475, OR 95% CI: 1.017 [0.971–1.064]). These findings were consistent across six additional MR methods, further corroborating the lack of genetic causality between DH and CD (Figures 1 and 3A). Assessments of heterogeneity using Cochran’s Q statistic (IVW) and Rucker’s Q statistic (MR-Egger) demonstrated no significant heterogeneity (P > 0.05). While MR-Egger’s intercept test showed no indication of horizontal pleiotropy (P > 0.05), the global test from MR-PRESSO did detect evidence of horizontal pleiotropy (P < 0.05) (Table 1). Detailed examination using radial MR methods (IVW and MR-Egger) identified one outlier (Figures 3B), which was confirmed by MR-PRESSO’s distortion test as SNP rs7674113 (Table 1). Despite these findings, leave-one-out analysis revealed that the results of the reverse MR analysis between DH and CD
