Causal association between rheumatoid arthritis and interstitial lung disease: A two-sample bidirectional Mendelian randomization study

Abstract

Epidemiological observations suggest a potential causal relationship between rheumatoid arthritis (RA) and interstitial lung disease (ILD), and vice versa. Alternatively, this association could be due to simple co-occurrence of RA and ILD resulting from shared population genetic or environmental risk factors. However, the causal nature of this relationship remains uncertain. This study aims to investigate the genetic causality between RA and ILD. Two-sample bidirectional Mendelian randomization (MR) was employed to test the causal effects of RA on ILD and ILD on RA, utilizing genome-wide association studies for ILD and RA. Data statistics were collected from public data sets. Relevant single nucleotide polymorphisms were selected through quality control steps in the meta-analysis results of genome-wide association studies. Two-sample bidirectional MR analysis was conducted to assess the causal relationship between the 2 conditions. The main MR analyses utilized methods such as inverse variance weighting, weighted median, and MR-Egger regression. Sensitivity analyses, including MR-Egger, leave-one-out, MR pleiotropy tests, and heterogeneity tests, were performed to evaluate heterogeneity and pleiotropy. In the European population, the risk of ILD was found to be increased by RA (odds ratio = 1.272, 95% confidence interval: 1.186–1.363, P < .001). Conversely, ILD was associated with an increased risk of RA (odds ratio = 1.111, 95% confidence interval: 1.008–1.223, P = .033). These findings highlight a potential genetic link between rheumatoid arthritis and interstitial lung disease and may inform future research regarding potential screening strategies in genetically susceptible individuals.

1. Introduction

Rheumatoid arthritis (RA) is a systemic autoimmune disease primarily characterized by persistent joint inflammation and destruction.^[1] It imposes an increasing burden on global healthcare resources. Although RA mainly affects the joints, it can also involve multiple organ systems, such as the skin, eyes, heart, lungs, kidneys, nervous system, and gastrointestinal tract, with pulmonary involvement being the most common, affecting up to 60% of RA patients.^[2,3] The potential mortality and morbidity associated with RA have been underestimated for decades.^[4]

Interstitial lung disease (ILD) is characterized by inflammation or fibrosis in the interstitial spaces, which leads to impaired gas exchange, causing dyspnea, decreased exercise tolerance, and reduced quality of life.^[5] ILD is a common but underrecognized complication of RA, associated with significant morbidity and mortality. ILD can occur at any point during the course of RA,^[6] clinically significant ILD occurs in approximately 5% to 15% of patients with RA.^[7] Research indicates that rheumatoid arthritis-related interstitial lung disease (RA-ILD) is a life-shortening condition. Raimundo et al found that 35.9% of patients died within 5 years of their initial diagnosis of RA-ILD, with a median survival of 7.8 years in the United States.^[8]

The association between RA and ILD has been extensively discussed, with research indicating that ILD is recognized as a severe extra-articular manifestation of RA.^[9] Another study demonstrates that a significant proportion of patients often develop ILD before experiencing arthritis symptoms, suggesting a role for the lungs in the progression of RA.^[10] The link between RA and ILD is supported not only by clinical manifestations but also by pathophysiological mechanisms.^[11] Therefore, it is essential to further explore the causal relationship between RA and ILD, establishing evidence to clarify the mechanisms underlying their connection.

Mendelian randomization (MR) is an epidemiological method that infers causal relationships between exposure and outcomes by using single nucleotide polymorphisms (SNPs) as instrumental variables. This approach effectively reduces the influence of confounding factors and reverse causation, providing more reliable causal inferences.^[12] Utilizing the advantages of MR, we investigated the causal relationship between RA and ILD based on summary statistics of genetic determinants from genome-wide association studies (GWAS) across diverse populations.^[13] If MR analysis confirms a causal relationship between RA and ILD, it could guide clinicians in conducting early ILD screening in RA patients,^[14] enabling timely interventions that may improve patient outcomes.

2. Materials and methods

2.1. Research design

We employed MR analysis to investigate the causal relationship between RA and ILD, utilizing forward MR with RA as the exposure and ILD as the outcome, and reverse MR with ILD as the exposure and RA as the outcome. Data on exposures and outcomes were sourced from GWAS databases. Suitable SNPs were selected as instrumental variables, and various statistical methods were applied to analyze the association between RA and ILD. The detailed MR research design is shown in Figure 1.

Figure 1.

The MR study design.

2.2. Data sources

GWAS summary statistics for RA and ILD were obtained from the IEU Open GWAS Project (https://gwas.mrcieu.ac.uk/).^[15] The RA dataset (GWAS ID: ebi-a-GCST90018910) was derived from a European population and included 417,256 individuals, comprising 8255 cases and 409,001 controls, with a total of 24,175,266 SNPs. The ILD dataset (GWAS ID: ebi-a-GCST90018863) included 469,827 individuals, of whom 2267 were cases and 467,560 were controls, with a total of 24,192,245 SNPs.

Basic characteristics of the RA and ILD GWAS datasets are summarized in Table 1. As this MR study was conducted using publicly available GWAS summary-level data, all ethical approvals and informed consents for participants were obtained in the original studies.

Table 1.

Basic information of the GWAS database in the 2-sample MR studies.

Disease	GWAS ID	Sample size	Population	Sex	Year	ncase	ncontrol	Number of SNPs
RA	ebi-a-GCST90018910	417,256	European	NA	2021	8255	409,001	24,175,266
ILD	ebi-a-GCST90018863	469,827	European	NA	2021	2267	467,560	24,192,245

2.3. Instrumental variables

In our MR study, we used SNPs as instrumental variables to infer the causal relationship between exposure and outcome.^[16] The selection of instrumental variables must satisfy 3 fundamental assumptions^[17]: First, the relevance assumption requires that selected SNPs are strongly associated with the exposure (P < 5 × 10⁻⁸). Second, the independence assumption stipulates that the SNPs are independent of confounding factors and do not exhibit horizontal pleiotropy. Third, the exclusion restriction assumption holds that the SNPs affect the outcome exclusively through the exposure, with no alternative pathways. To satisfy these assumptions, we selected SNPs significantly associated with RA at the genome-wide significance level (P < 5 × 10⁻⁸). We then performed linkage disequilibrium (LD) clumping using a threshold of r² < 0.001 within a 10,000 kb window, based on summary-level statistics, to ensure the independence of the instruments. To minimize potential bias from the original GWAS data, we excluded SNPs with a minor allele frequency below 0.01. Finally, to avoid weak instrument bias, we defined strong instruments as those with an F-statistic >10. The strength of each selected SNP was assessed using this criterion.

2.4. MR analysis

This study employed 3 methods—inverse variance weighting (IVW), weighted median, and MR-Egger regression—to estimate the causal relationship between exposure and outcome.^[18] The IVW method calculates the Wald ratio for each SNP and provides a summary causal estimate, allowing for overdispersion. The weighted median estimate serves as a supplement to the IVW method, ensuring reliable estimates that account for at least 50% of the analytical weight provided by the instrumental variables. The MR-Egger method was utilized to test for biases in the analysis results, which may arise from pathways other than the exposure affecting the outcome. Among these methods, IVW is the most critical method for estimating the causal relationship between exposure and outcome. It provides an accurate estimate of the causal relationship regarding the risk of the outcome.

2.5. Sensitivity analysis

To assess the robustness of the results and prevent potential pleiotropy and heterogeneity, we conducted a series of sensitivity analyses.^[19] Cochran Q test was used to quantify heterogeneity among SNPs, and a funnel plot was used for visualization. The intercept from MR-Egger regression and the global test from MR-PRESSO were applied to evaluate whether horizontal pleiotropy influenced the MR analysis results. Additionally, a leave-one-out analysis was performed to determine if a single SNP was driving the causal estimate. A P-value of <.05 was considered statistically significant.

2.6. Statistical analysis

We used R software (version 4.3.3, University of Auckland, New Zealand) along with the TwoSample MR package (version 0.5.11; MRC Integrative Epidemiology Unit, University of Bristol, Bristol, United Kingdom) for all statistical analyses. All data were analyzed using the “TwoSampleMR” package in R version 4.3.3.^[20]

3. Results

3.1. Causal effects of RA on ILD

3.1.1. Instrumental variable selection

After selecting SNPs strongly associated with RA (P < 5 × 10⁻⁸), we performed LD clumping using the European reference panel, applying a threshold of r² < 0.001 and a 10 Mb (10,000 kb) window. As a result, 65 independent SNPs were identified that satisfied the 3 core instrumental variable assumptions. Each SNP had an F-statistic >10, indicating a low risk of weak instrument bias in the causal analysis from RA to ILD.

3.1.2. MR analysis

Three MR analysis methods—IVW, MR-Egger, and weighted median estimation—were utilized to assess the causal relationship between RA and the risk of ILD. The IVW results revealed a positive correlation between RA and ILD, suggesting that RA increases the risk of ILD (odds ratio = 1.272, 95% confidence interval: 1.186–1.363, P < .001), as depicted in Table 2 and Figure 2. The primary conclusions of this study are based on the IVW results, with the overall effects of the other 2 methods aligning in direction with those of IVW. Therefore, we can conclude that RA is a positive risk factor for ILD. All 3 MR methods indicate that RA increases the incidence risk of ILD, as illustrated in Figure 3.

Table 2.

Mendelian randomization analysis of the risk of interstitial lung disease in rheumatoid arthritis.

Exposure	Outcome	SNP	MR method	OR	95% CI	P	Q_p-val	pleiotropy_test
RA	ILD	65	MR-Egger	1.205	(1.081, 1.342)	.001	0.009	–
			MR IVW	1.272	(1.186, 1.363)	< .001	0.011*	0.207**
			Weighted median	1.225	(1.130, 1.327)	< .001	–	–

Odds ratio (OR) values, confidence interval (CI), P-value, Q_p-val, and pleiotropy_test of MR results were obtained by IVW, MR-Egger, and weighted median in the group of European population.

^*P < .05.

^**P > .05.

Figure 2.

Forest diagram of rheumatoid arthritis on interstitial lung disease.

Figure 3.

Scatter plot of rheumatoid arthritis on interstitial lung disease.

3.1.3. Sensitivity analysis

We employed a random effects model for analysis and found significant heterogeneity within the study (P < .05) as indicated by Cochran Q test in Table 2. The MR-Egger intercept test for horizontal pleiotropy yielded a P-value >.05, suggesting that horizontal pleiotropy is not significant and that the results are robust, as shown in Table 2. To further assess the stability of these results, a funnel plot was generated, showing that the included SNPs were largely symmetrical, representing a symmetrical distribution of causal association points. This indicates minimal differences among the instrumental variables and suggests a low likelihood of potential bias affecting the causal relationship, as illustrated in Figure 4. Additionally, the leave-one-out sensitivity analysis revealed no individual SNPs that significantly influenced the causal estimate, as shown in Figure 5.

Figure 4.

Funnel plot of rheumatoid arthritis on interstitial lung disease.

Figure 5.

Leave one method result.

3.2. Causal effects of ILD on RA

3.2.1. Instrumental variable selection

After selecting SNPs strongly associated with ILD (P < 5 × 10⁻⁸), we performed LD clumping using the European reference panel, with a threshold of r² < 0.001 and a 10 Mb (10,000 kb) window. As a result, 22 independent SNPs were identified that satisfied the 3 core instrumental variable assumptions. Each SNP had an F-statistic >10, suggesting a low risk of weak instrument bias in the causal analysis from ILD to RA.

3.2.2. MR analysis

This study employed 3 MR analysis methods—IVW, MR-Egger, and weighted median estimation—to evaluate the causal relationship between ILD and RA. According to the IVW results, a positive correlation was found, indicating that ILD increases the risk of RA (odds ratio = 1.111, 95% confidence interval: 1.008–1.223, P = .033), as shown in Table 3 and Figure 6. The primary conclusions of this study are based on the IVW results, with the overall effects of the other 2 methods aligning in direction with those of IVW. Therefore, it can be concluded that ILD is a positive risk factor for RA. All 3 MR methods indicate that ILD increases the incidence risk of RA, as illustrated in Figure 7.

Table 3.

Mendelian randomization analysis results of interstitial lung disease on the risk of rheumatoid arthritis.

Exposure	Outcome	SNP	MR method	OR	95% CI	P	Q_p-val	pleiotropy_test
ILD	RA	22	MR-Egger	1.069	(0.908, 1.260)	.432	< .001	–
			MR IVW	1.111	(1.008, 1.223)	.033	< .001*	0.581**
			Weighted median	1.064	(1.015, 1.114)	.017	–	–

Odds ratio (OR) values, confidence interval (CI), P-value, Q_p-val, and pleiotropy_test of MR results were obtained by IVW, MR-Egger, and weighted median in the group of European population.

^*P < .05.

^**P > .05.

Figure 6.

The forest of interstitial lung disease and rheumatoid arthritis.

Figure 7.

Scatter plot of interstitial lung disease on rheumatoid arthritis.

3.2.3. Sensitivity analysis

We used a random effects model for analysis. Cochran Q test showed significant heterogeneity within the study (P < .05), as displayed in Table 3. The MR-Egger intercept test for horizontal pleiotropy had a P-value above .05, indicating that horizontal pleiotropy is not significant and the results are robust, as shown in Table 3. To further assess the stability of these results, a funnel plot was generated, showing that the included SNPs were largely symmetrical, representing a symmetrical distribution of causal association points. This indicates minimal differences among the instrumental variables, suggesting a low likelihood of potential bias affecting the causal relationship, as illustrated in Figure 8. Additionally, the leave-one-out sensitivity analysis revealed no individual SNPs that significantly influenced the causal estimate, as shown in Figure 9.

Figure 8.

Funnel plot of interstitial lung disease on rheumatoid arthritis.

Figure 9.

Leave one method result.

4. Discussion

Our study demonstrates the use of 2-sample MR methods based on GWAS statistics to explore the bidirectional causal relationship between RA and ILD in European populations, providing evidence for the genetic link between these conditions. Our findings indicate that RA is a risk factor for ILD, while ILD is also a risk factor for RA, suggesting that these 2 common autoimmune diseases may share similar underlying pathophysiological mechanisms.

Multiple clinical studies have identified significant clinical symptoms associated with RA-ILD, including exertional dyspnea, persistent dry cough, fatigue, and weakness.^[2,21] A longitudinal cohort study of Chinese RA-ILD patients revealed that 69% of cases were diagnosed with ILD following their RA diagnosis.^[22] However, not all studies have reached the same conclusions regarding the association between RA and ILD. Research indicates that a substantial proportion of RA-ILD patients develop ILD prior to joint symptoms, highlighting the central role of the lungs in the progression of RA.^[10] A nationwide cohort study found that RA-ILD is associated with increased mortality (compared to RA patients without ILD), particularly among those under 75 years of age, patients with ILD preceding RA, and males.^[23] Given the complex mechanisms of RA-ILD, the causal relationship remains debated. Our bidirectional MR study further complements previous research, suggesting that RA and ILD may have similar risk profiles and mechanisms of pathogenesis.

Several factors may contribute to the increased risk of ILD in patients with RA. Studies have shown that the development of RA-ILD is associated with genetic factors, age, gender, smoking, environmental pollutants, and the presence of autoantibodies such as rheumatoid factor and anti-cyclic citrullinated peptide antibodies (ACPA).^[24] In patients with RA, the prevalence of ILD is higher in individuals carrying the MUC5B variant, with a risk of 16.8%, compared to 6.1% in those who do not carry the variant.^[3] The MUC5B promoter variant rs35705950 is strongly linked to RA-ILD, particularly in patients exhibiting the usual interstitial pneumonia pattern, leading to a significantly elevated risk. Male patients have twice the risk compared to female patients, and they also experience higher mortality rates^[25] Smoking is a major predictor of RA-ILD, with long-term smokers facing a higher risk.^[26,27] Exposure to air pollutants such as PM2.5 and silica has been shown to increase hospitalization risk for RA-ILD.^[28] Additionally, ACPA and rheumatoid factor levels are significantly correlated with the incidence of RA-ILD, with patients exhibiting high ACPA titers being at increased risk of developing ILD.^[29]

The mechanisms by which RA increases the risk of ILD are linked to abnormal tissue responses in the alveolar walls and lung parenchyma, initiated by epithelial cell damage. Persistent damage to epithelial cells leads to the activation of immune cells, including neutrophils, dendritic cells, and macrophages, resulting in the excessive accumulation of extracellular matrix components in lung tissue.^{[21,28,30,31]} Research indicates that multiple cytokines play a role in the pathogenesis of RA-ILD, particularly MCP-1/CCL2, SDF-1α, and IL-18.^[32] TNF-α is a major pro-inflammatory factor involved in the lung fibrosis process, promoting disease progression by triggering fibroblast proliferation and the secretion of various factors.^[33] IL-23 contributes to fibrosis in RA-ILD by stimulating epithelial-to-mesenchymal transition, whereas IL-11 is associated with disease activity and the development of ILD in RA patients.^[34,35] Furthermore, research has shown that HDAC3 promotes fibrosis in RA-ILD by upregulating miR-19a-3p-targeted IL17RA.^[36–38] Proteomic analyses have identified changes in the expression of various proteins related to lung fibrosis, confirming the role of CX3CL1 in this process.^[39,40] Furthermore, studies suggest that mitochondrial DNA mutations and ROS accumulation in RA patients are also implicated in the pathogenesis of ILD.^[41,42] Wang et al^[37] reported that elevated IL-11 levels activate the JAK/STAT3, ERK, and PI3K/Akt/mTORC1 signaling pathways, which are associated with the onset of ILD.

The mechanisms by which RA increases the risk of ILD include the identification of matrix metalloproteinase (MMP)-7 as a potential biomarker for RA-ILD, with significantly elevated levels found in the serum of RA patients with clinical and subclinical ILD.^[43] A prospective cohort study evaluated the association between plasma MMPs and the incidence of interstitial lung disease in a large multicenter RA cohort, further supporting the potential pathogenic role of MMP-7 in RA-ILD. The association between RA-induced pulmonary fibrosis also involves major histocompatibility complex gene loci on chromosome 6, which are linked to an increased risk of ILD in RA patients.^[7] While the mechanisms by which ILD increases the risk of RA remain unclear, several possibilities exist. First, RA-related autoimmune diseases may begin at mucosal sites, including the lungs, and subsequently transition to involve synovial joints.^[44,45] In genetically susceptible individuals, factors such as smoking can lead to damage of alveolar and airway epithelium, resulting in protein citrullination, the formation of neutrophil extracellular traps, and the production of RA-related autoantibodies in local lung mucosa, ultimately establishing systemic autoimmunity.^[46–48] Second, chronic inflammation of the lung mucosa leads to elevated levels of cytokines and chemokines, which can affect lymphocyte subsets and ultimately impact other tissues involved in autoimmune responses. Additionally, McFarlane et al^[49] noted conflicting reports regarding the racial and gender distribution of RA-ILD. These factors may significantly influence outcomes, highlighting the need for further research to determine genetic differences.

Despite the robust design and multiple sensitivity analyses, some methodological limitations should be acknowledged. First, while MR reduces bias from confounding and reverse causality, it still relies on key assumptions (e.g., absence of horizontal pleiotropy), which may not be fully testable. Second, the limited sample size for ILD cases may reduce the power of the reverse MR analysis. Third, potential shared genetic architecture between RA and ILD may complicate the interpretation of directionality. Therefore, clinical implications, particularly regarding screening or management, should be interpreted with caution until further validated by functional studies or prospective clinical data.

Our study has 3 main strengths. First, it provides genetic evidence suggestive of a bidirectional causal relationship between RA and ILD through MR analysis. However, these findings should be interpreted with caution due to potential methodological limitations, such as pleiotropy or shared genetic architecture. Second, instrumental variables were selected using strict criteria, ensuring strong instrument strength, sufficient statistical power, and increased robustness of the causal inference. Third, unlike observational studies that are prone to confounding and reverse causation, MR analysis leverages genetic variants as natural proxies for exposures, thereby offering a more reliable assessment of causality.

Author contributions

Conceptualization: Haozhu Chen.

Data curation: Jinxiang Peng.

Formal analysis: Feng Wu.

Funding acquisition: Feng Wu.

Investigation: Feng Wu.

Methodology: Haozhu Chen.

Project administration: Haozhu Chen.

Software: Jinxiang Peng, Jinmei Tan.

Validation: Zhuang Chen.

Visualization: Jinmei Tan.

Writing – original draft: Jinxiang Peng.

Writing – review & editing: Jinxiang Peng, Haozhu Chen, Jinmei Tan, Zhuang Chen, Feng Wu.