Causal Association Between Atopic Dermatitis and Keratoconus: A Mendelian Randomization Study

Abstract

Purpose: Although many studies have indicated that atopic dermatitis (AD) could contribute to the risk of keratoconus (KC), the causality between AD and KC remains controversial. This study aimed to explore the potential causal associations between AD and KC.

Methods: Instrumental variables for both exposures and outcomes were obtained from large-scale genome-wide association study summary statistics from previous meta-analyses. Mendelian randomization (MR) was applied to infer causal associations between AD and KC. Our main analyses were conducted by inverse-variance weighted (IVW) method multiplicative random effect model, complemented with additional five models and sensitivity analyses. Reverse MR analysis was applied to determine the direction of the causal association between AD and KC.

Results: Both IVW and weighted median methods revealed a causal effect of AD on KC (IVW odds ratio [OR], 1.475; P = 4.16 × 10⁻⁴; weighted median OR, 1.351; P = 7.65 × 10⁻³). The weighted mode, simple mode, and MR Egger methods demonstrated consistent direction of causality. Evidence from all sensitivity analyses further supported these associations. Reverse MR analyses did not suggest causal effects of KC on AD.

Conclusions: This study supported a significant causal effect of AD on KC, and reverse MR analysis proved that the causal association was unilateral.

Translational Relevance: This study provides valid evidence that regular ophthalmic examinations are recommended for patients with AD to detect and prevent KC at an early stage.

Introduction

Keratoconus (KC) is a corneal disorder characterized by progressive thinning and conical protrusion of the central cornea, leading to high myopia and irregular astigmatism.¹ The disease usually manifests in the second decade of life.² Corneal topography, as the most sensitive method of detecting corneal shape, is considered the gold standard for the diagnosis and monitoring of KC.^2–4 The etiology of KC is believed to be multifactorial, affected by genetic, environmental, and socioeconomic factors.³ Although KC is defined traditionally as a noninflammatory condition,² recent studies have indicated the involvement of inflammation in the pathogenesis of KC.^5–8 

Atopic dermatitis (AD) is a chronic inflammatory skin disease, with an incidence of 15% to 20% in developed countries,⁹ often occurring in childhood. The disease is characterized by recurrent eczematous lesions of the skin, intense itching, and discomfort.¹⁰ AD is accompanied by multiple complications, including psychological disorders and inflammatory diseases, as well as various ocular surface diseases, such as conjunctivitis, keratitis, and KC.^11,12 In recent years, several cross-sectional studies and meta-analyses have revealed a certain association between AD and KC, with a significantly higher incidence of KC in patients with AD compared with the general population.^2,13–16 However, observational studies have limited ability to establish a definitive causal relationship between AD and KC, owing to various confounding factors and biases. Previous Mendelian randomization (MR) studies have also identified controversy regarding the causality.¹⁷ 

MR is a statistical method using independently inherited single nucleotide polymorphisms (SNPs) to estimate potential causal relationships between exposure and outcomes. This approach is less susceptible to confounding factors or reverse causation present in traditional observational studies.^18,19 In this study, we conducted a two-sample MR analysis to assess the causal effects between AD and KC. 

Methods

Study Design

This study investigates the causal association of AD and KC, using instrumental variables (IVs) extracted from genome-wide association study (GWAS) summary datasets. The MR methodology was based on three assumptions: robust associations between IVs and exposures, the lack of independent association between the outcome and the IVs except via the exposures, and the lack of confounders between the exposure–outcome relationship. This study was conducted according to the recommendations of the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization (STROBE-MR) guideline.²⁰ The conceptual framework of the current MR study is reported in Figure 1. 

Figure 1.

 

Study design and framework of the MR analysis in this study. Genetic variants that are associated with the exposure at genome-wide significance (P < 5 × 10⁻⁸) were used as IVs. The causal relationship of AD and KC were investigated using two-sample MR and reverse two-sample MR.

GWAS Summary Datasets

Genetic associations for AD were obtained from a meta-analysis by Sliz et al.²¹ A total of 796,661 individuals with 22,474 cases of European descent were included, combining three independent datasets from the FinnGen study, the Estonian Biobank, and the UK Biobank. SNPs associated with KC were obtained from the results of a multiethnic meta-analysis of 121,216 individuals with 4669 cases by Hardcastle et al.²² The study combined datasets from three independent European cohorts (Los Angeles, Australian, and Genetic epidemiology research in adult health and aging cohort), and two non-European cohorts (Indian and African). Only genetic variants with genome-wide significance (P < 5 × 10⁻⁸) were selected as IVs for the MR analysis. IVs were clumped using the PLINK software, with a window of 10 Mb and maximal linkage disequilibrium of r² < 0.001 to ensure independence. The F statistic was used to assess the level of weak instrument bias. 

MR Analysis

MR analyses were conducted in R version 4.2.3 and RStudio Version 2023.06.0+421 using the “TwoSampleMR” package.²³ This package makes causal inferences of exposures on the outcomes, and estimates the effects using inverse-variance weighted (IVW), MR Egger regression,²⁴ weighted median methods,²⁵ and weighted mode. MR estimate outcomes are presented as odds ratios (OR) with 95% confidence intervals (CI). The significance of causal effects in the MR analyses was defined as IVW P < 8.33 × 10⁻³ and all four MR models should show consistent direction. 

Horizontal pleiotropy of IVs was evaluated by the regression intercept of MR Egger analysis. A leave-one-out analysis was performed to assess the robustness of the MR estimate. The Cochrane Q test was used to estimate the heterogeneity among the IVs. If significant heterogeneity was observed, the IVW multiplicative random effects model was used to estimate the MR effects. In addition, the MR pleiotropy residual sum and outlier (MR-PRESSO) test was performed using the “MR-PRESSO” package,²⁶ to identify potential outliers among the genetic instruments. The outliers were then excluded, and the MR effects were re-estimated to decrease heterogeneity in the analyses and provide a more robust estimate of MR effects. 

Reverse MR analyses were conducted to determine the direction of the causal association between the exposure and the outcome. The reverse MR estimates the effects of the outcome on the exposure, using IVs associated with the outcome. When both the original and the reverse MR analyses yielded positive results for a pair of traits, the causal association between this pair of traits was deemed bidirectional, instead of unilateral. 

Results

Causal Association of AD on KC

We selected 22 SNPs from the GWAS as IVs for AD against KC (Supplementary Table S1). The F statistics for AD was 1541.76, confirming that these analyses will not be affected by weak instrument bias. To investigate whether genetically determined AD increases the risk of KC, we conducted two-sample MR analyses. The IVW method was the main method used to estimate associations between AD and KC, whereas the weighted median, MR-Egger, simple mode, and weighted mode serve as supplementary methods. As the Cochrane Q tests demonstrated that significant heterogeneity exists among the effects of the genetic instruments (P < 0.05) (Supplementary Table S2), the IVW analysis with an multiplicative random effects model was used. The IVW multiplicative random effects approach showed that AD was significantly and positively associated with the incidence of KC (OR, 1.475; 95% CI, 1.189–1.830; P = 4.16 × 10⁻⁴). A significant association was generated using the weighted median method (OR, 1.351; 95% CI, 1.083–1.685; P = 7.65 × 10⁻³), but not the weighted mode, simple mode and MR-Egger methods. However, the direction of causality from these methods is consistent with IVW (Table). The results suggest a positive causal link between AD and KC. Figure 2A shows scatterplots of the MR analyses revealing the effect sizes of associations between AD and KC. 

Table.

 

Two-sample MR Estimates for Causal Associations Between AD and KC

 

Figure 2.

 

Causal association of AD on KC. (A) Scatterplots for MR estimation of the causal effect of AD on KC. The slope of each line corresponds to the estimated MR effect per method. (B) Forest plot visualizing the causal effect of each single SNP on the risk of KC. (C) Funnel plots visualizing overall heterogeneity of MR estimates for the effect of AD on KC. (D) Leave-one-out sensitivity analysis for AD on KC.

Sensitivity and Robustness Analyses

To assess the robustness of the causal association between AD and KC, we conducted comprehensive sensitivity analyses. Calculation of the Egger intercept suggested no evidence of directional horizontal pleiotropy effects (intercept = −0.009; P > 0.05) (Supplementary Table S2). The MR-PRESSO models were implemented to identify potential outliers and re-estimate the causal effects after outlier correction. The MR-PRESSO test identified one outlier, rs11786685. The P values of the raw and outlier-corrected MR analyses were 1.96 × 10⁻³ and 2.55 × 10⁻³, respectively. The distortion test showed no significant difference (P = 0.426). A leave-one-out analysis was also performed, and no outlier other than rs11786685 was observed (Fig. 2D). The funnel plot showed no evidence of asymmetry (Fig. 2C). Altogether, our initial results are supported by further evidence from all sensitivity analyses. 

Effect of KC on AD

The 37 SNPs independently associated with KC were used as IVs (Supplementary Table S1), with an F statistic of 5047.24. We found no evidence for a causal effect of KC on AD (Table): IVW (OR, 1.005; 95% CI, 0.979–1.032; P = 0.711), MR-Egger (OR, 0.991; 95% CI, 0.921–1.067; P = 0.817), weighted median method (OR, 1.055; 95% CI, 0.982–1.135; P = 0.892), simple mode method (OR, 0.967; 95% CI, 0.918–1.018; P = 0.155), and weighted mode method (OR, 1.005; 95% CI, 0.979–1.032; P = 0.213). The scatterplots of the MR analyses revealing the effect sizes of associations between KC and AD are shown in Figure 3. 

Figure 3.

 

Lack of causal association of KC on AD. (A) Scatterplots for MR analyses of the causal effect of KC on AD. The slope of each line corresponds to the estimated MR effect per method. (B) Forest plot to visualize causal effect of each single SNP on the risk of AD. (C) Funnel plots to visualize overall heterogeneity of MR estimates for the effect of KC on AD. (D) Leave-one-out sensitivity analysis for KC on AD.

Discussion

This study used an MR analysis to evaluate the causal associations between AD and KC using large GWAS datasets. We found evidence for a causal association of AD on KC, which was consistent across different MR methodologies and was proved robust in sensitivity analyses. Reverse MR analyses yielded no significant causality of KC on AD, demonstrating that the association was unilateral instead of bidirectional. 

The causal relationship between AD and KC remains a matter of debate. Previous observational studies have consistently reported the association between AD and KC. A retrospective study carried out at Johns Hopkins Hospital revealed a notably elevated prevalence of KC in patients with AD compared with the general population.¹⁵ Another cross-sectional study showed that the incidence of all ocular surface diseases, including KC, increased with the severity of AD in adult patients.²⁷ These findings are consistent with epidemiological research based on nationwide registry data from Denmark,¹⁶ which reported that the hazard ratio for developing KC is 3.06 for patients with mild AD compared with the general population, whereas it increases to 10.01 for patients with severe AD. Another nationwide study from Denmark compared 2679 patients with KC with 26,790 controls and found that the odds of AD risk in patients with KC were over 7.3 times of that in the control group (OR, 7.32; 95% CI, 5.73–9.35).²⁸ Similarly, a cohort study involving adolescents and adults in Taiwan identified KC as an independent risk factor for AD. In particular, the probability of developing AD is significantly increased among female patients with KC aged 20 to 29 years and 12 to 19 years who reside in southern regions.²⁹ Furthermore, some correlative studies also suggested significant associations between KC and AD.^12,14 In addition, some studies explored the differences in corneal topography in patients with KC with or without atopic syndromes, including AD, allergic asthma, and allergic rhinitis. Kaya et al.³⁰ performed corneal examinations with the Orbscan II device and observed that atopic KC eyes exhibited steeper cones, thinner central corneas, and a more peripheral location of both the thinnest point and the cone apex compared with the control group. Shajari et al.³¹ used the Pentacam HR for assessment and found no topographical differences between the two groups. However, patients with KC in the atopic group were significantly younger and had notably higher corneal density in the anterior 120 µm of the cornea compared with the controls. 

Meanwhile, several studies reported a lack of a correlation between AD and KC. In a systematic review and meta-analysis, AD and KC displayed no significant associations.³² Another longitudinal cohort study observed that the incidence of KC among young women with AD showed no difference compared with the general population within 10 years.³³ Zhou et al.¹⁷ performed a two-sample Mendelian analysis to investigate the relationship between AD and ocular surface diseases and did not find causality between them, possibly owing to a limited sample size for KC. 

This MR study used large GWAS datasets and identified a causal association between AD and KC. Previous research has proposed several potential mechanisms. The first is the mechanical hypothesis: repeated eye rubbing caused by AD induces corneal deformation. One of the symptoms of AD is itching, which can lead patients to rub their eyes. Some studies have reported significant associations between KC and eye rubbing,^32,34 but the exact mechanisms remain to be known. The possible mechanisms involve mechanical damage to the epithelium, the release of inflammatory mediators, increased rate of corneal cell apoptosis, and disrupted collagen maintenance, leading to decreased corneal shear strength. Rubbing the eyes may also increase corneal temperature, thereby upregulating these mechanisms and leading to abnormal enzyme activity. Enzymatic tissue degradation can promote the sliding of collagen fibers, and the increase in intraocular pressure during rubbing can cause the cornea to bend steeply, together exacerbating corneal ectasia.^35–37 

Second, some researchers believe that the systemic inflammation caused by AD is related to KC, with increasing evidence of the involvement of the immune system in the pathogenesis of KC. A recent study found that KC is correlated positively with various immune-mediated diseases.³⁸ Increased levels of proinflammatory proteins, such as interleukins, tumor necrosis factors, and matrix metalloproteinases, can be detected in the tears of patients with KC.^5–8 Toprak et al.³⁹ reported that the total oxidative status and oxidative stress index in the serum of patients with KC are significantly higher than in the control group, suggesting the microenvironment of KC may also be influenced by oxidative stress, the exact reasons for which are not yet clear. 

The third point is that genetics may play a role in the causal association between AD and KC. The filaggrin gene (FLG) encodes an important protein that promotes epidermal differentiation and maintains the skin barrier, which is also expressed in the cornea.^40,41 The mutations of this gene lead to AD. A study indicated that the expression levels of filaggrin in KC group and normal corneal groups did not differ, but patients carrying FLG mutations exhibited more severe KC.⁴² Subsequent research may investigate additional FLG mutations or other genetic variations shared between AD and KC. 

Our study has important strengths. Our analysis, for the first time, was based on independent and large GWAS summary statistics for both exposures and outcomes, which provided sufficient statistical power to estimate the causal effects precisely. Reverse MR analysis restricted reverse causality and sensitivity analyses were conducted to examine the robustness of the results. In contrast with observational studies, the MR methodology is less likely to be affected by confounding factors and reverse causation. The primary MR analyses in our study were performed by an IVW method, which provides the most precise estimates. When the IVW method yields significant results (P < 0.05), even if the other methods do not, the findings can be considered robustly positive provided that the OR values of the other methods align in the same direction. Therefore, through this MR study, we have sufficient evidence to show the causal effect of AD on KC. However, some limitations of the study should be noted. The uniting biobank for AD that we used was restricted to participants of European ancestry, and the GWAS database for KC only included a small proportion of non-European populations. Therefore, the results may underestimate the causality. This limitation could be addressed in future research as genetic data of other ethnic groups becomes more available. Additionally, the possibility of pleiotropic effects cannot be entirely ruled out, despite the evidence against departure from the MR assumptions. 

In conclusion, using a two-sample MR approach, our study strongly supported previous observational studies, suggesting that AD has unidirectional causal effects on KC. We recommend that individuals with AD undergo regular ophthalmic examinations, such as corneal thickness and corneal topography, to detect any early signs of KC. Especially for young patients, receiving ophthalmic examinations and treatment upon the diagnosis of AD can help to prevent permanent ocular damage threatening vision. 

Acknowledgments

Supported by the National Natural Science Foundation of China (Grant No. 82000863) and National High-Level Hospital Clinical Research Funding (Grant No. 2022-PUMCH-A-198). 

Authors’ Contributions: DC devised the idea and designed the study. YC, TZH, and SY analyzed the data and wrote the manuscript. DC and YL supervised the study and revised the manuscript. 

Ethics Approval and Consent to Participate: This study adheres to the Declarations of Helsinki. Specific ethics approval was not required for all data was obtained from sources available to the public. 

Data Availability and Materials: The datasets supporting the conclusions of this article were obtained from sources available to the public. 

Disclosure: Y. Chang, None; T. Huang, None; S. Yang, None; Y. Li, None; D. Chen, None