Introduction

Psoriatic arthritis (PsA) represents a chronic, immune-mediated inflammatory joint disease characterised by musculoskeletal inflammation, predominantly occurring among patients with psoriasis. The onset of PsA is most common between the ages of 30 and 50, with no significant disparity between the sexes.1 Over half of the individuals diagnosed with PsA report at least one comorbidity, such as cardiovascular disease, uveitis, depression, inflammatory bowel disease, and metabolic syndrome.2 This complex association underscores the necessity for a comprehensive understanding of PsA pathogenesis. Genetic studies have provided valuable insights into the pathways and susceptibility loci associated with PsA development.3,4 Exploring the genetic connections between PsA and these comorbidities holds promise for enhancing management strategies and developing targeted therapies, offering hope for future management of this challenging and burdensome condition.

Uveitis is characterised by intraocular inflammation occurring in the uveal tract and adjacent ocular structures. It stands as a leading cause of blindness and visual impairment, significantly impacting the quality of life of affected individuals.5 Approximately 10 to 15% of uveitis cases lead to complete vision loss, while reversible vision loss occurs in up to 35% of cases. The causes of uveitis are diverse, encompassing infections, systemic diseases, and autoimmune conditions. Additionally, between 1/4 and 1/3 of individuals with uveitis may exhibit extraocular or systemic manifestations, such as PsA.6-8 In recent years, despite numerous studies having proven the association between uveitis and PsA, the causal relationship and underlying mechanisms remain unclear.9-11 Previous studies were predominantly observational in nature, making them susceptible to various biases.

Mendelian randomisation (MR) is a genetic epidemiological research methodology that employs single-nucleotide polymorphisms (SNPs) as instrumental variables (IVs) to discern potential causal relationships between exposure factors and outcomes based on the Mendelian laws of heredity.12,13 In contrast to observational studies, MR studies possess several advantages. As genetic variation is determined at the time of conception, it predates disease development and is generally unaffected by confounding factors, such as acquired factors and the social environment. Consequently, the inferred causal associations from MR studies exhibit more cogent temporality, diminish confounding bias, and mitigate reverse causation.14 This study employed a two-sample bidirectional MR analysis utilising publicly available genome-wide association studies (GWAS) databases to investigate the causal relationship between uveitis and PsA. Although previous observational studies have reported an association between uveitis and PsA, their findings are subject to potential confounding, reverse causation, and measurement bias. These inherent limitations of observational designs highlight the importance of employing causal inference approaches such as MR, which can provide more robust evidence by minimising such biases.

Methods Study design

To ascertain a causal relationship between uveitis and PsA, a two-sample bidirectional MR analysis was conducted utilising a combined dataset sourced from GWAS.15,16 In forward MR analysis, uveitis is regarded as the exposure, while PsA serves as the outcome. Conversely, in reverse MR analysis, PsA is treated as the exposure, and uveitis as the outcome. The data utilised in this investigation were acquired from previously published studies, which received ethical approval from the respective committees, thereby obviating the need for additional ethical clearance. The pictorial representation of the study design has been shown in Figure 1. This MR analysis rests upon three core assumptions: (1) relevance, where selected SNPs are strongly associated with the exposure; (2) independence, meaning the SNPs are not associated with any confounders; and (3) exclusion restriction, implying the SNPs affect the outcome only through the exposure. These assumptions were addressed through stringent SNP selection criteria, pleiotropy screening using PhenoScanner, and validation through MR-Egger intercept sensitivity analyses.

Figure 1: Study design overview. MR analyses depend on three core assumptions.

Export to PPT

Data sources

To minimise potential sample overlap and population stratification bias, we employed three independent GWAS datasets (IEU OpenGWAS, GWAScatalog, and FinnGen). A meta-analysis was conducted to integrate findings and reduce dataset-specific biases.

The data employed for MR analysis in this investigation were sourced from distinct databases. Uveitis data were acquired from the IEU OpenGWAS database, while PsA data were gathered from three major GWAS databases: IEU OpenGWAS, GWAScatalog, and the FinnGen Biobank. A comprehensive overview of the databases has been presented in Table 1.

Table 1: Details of the data sources used in this study

Trait IEU GWAS ID Source Sample size Population Uveitis ebi-a-GCST90018938 IEU OpenGWAS 480,742 European PsA ieu-b-5116 IEU OpenGWAS 217,351 European PsA GCST90044516 GWAScatalog 456,348 European PsA finngen_R10_M13_PSORIARTH FinnGen biobank 266,381 European Genetic IV selection

IV selection criteria in this study:

(1) SNPs exhibiting a genome-wide significant association with the exposure (P<5E-08).

(2) Lack of linkage disequilibrium (LD) between SNPs. Quality control standard: r2<0.001, kb=10,000.17

(3) Integration and concordance of the exposure-outcome dataset, with correction of palindromic SNPs with an ambiguous strand based on allele frequency information.18

(4) Screening the SNPs selected through previous steps in the PhenoScanner database (www.phenoscanner.medschl.cam.ac.uk/) aimed to identify variants associated with other phenotypes, such as smoking, obesity, anxiety, depression, etc. These associations may potentially influence the progression of outcomes independently of the primary exposure.

(5) Evaluation of IV strength by calculating the F-value, and exclusion of potentially weak IV bias between the IV and exposure factors (F>10). The formula for calculating the F-value: F=R2×(N-k-1)k×(1-R2), where R2 represents the proportion of variance in the exposure explained by the genetic variants, N represents the number of participants, and K represents the number of instruments.

Statistical analysis

In this study, the “Two-Sample MR,” “meta,” and “ggplot2” packages of R 4.1.0 software were utilised for analysis. We primarily utilised the Inverse Variance Weighted (IVW)19 method to compute the odds ratio (OR) and its 95% confidence interval (CI) in order to assess the potential causal relationship between uveitis and PsA. IVW is a weighted linear regression model that provides reliable results when there is no horizontal pleiotropy in the IV. The other two methods were the MR-Egger regression method and the weighted median method (WME).20,21 Given the susceptibility of IVW estimates to invalid instrumental bias or pleiotropy, a series of sensitivity analyses were conducted to ascertain the validity and robustness of the IVW results. The assessment of heterogeneity among SNPs was performed using Cochran’s Q.22 Significance was indicated by P<0.05, leading to the adoption of a random effects model. Otherwise, a fixed effects model was employed. The MR-Egger regression test is employed to evaluate the presence of horizontal pleiotropy. If the intercept significantly differs from zero, it may suggest the presence of horizontal pleiotropy.23 The leave-one-out analysis is commonly used in genetic epidemiology studies employing MR to assess the robustness of the results, particularly to explore the extent to which an SNP influences the overall causal estimation. This analytical approach helps identify whether one or more SNPs have an excessive impact on the overall outcome. Another strength of our study lies in the selection of three large-scale GWAS databases from diverse sources, and the subsequent meta-analysis of MR analysis results related to uveitis and PsA. Through this meta-analysis, we conducted a comprehensive evaluation of the outcomes to ensure the reliability of our conclusions. All statistical tests employed bilateral analysis, and P<0.05 was deemed statistically significant.

Results The outcomes of IV selection

Eight SNPs were identified as IVs for uveitis, following LD adjustments, reaching the GWAS threshold of P < 5E-08. Notably, rs62394550 was excluded due to the presence of a palindromic sequence. Additionally, all SNPs have F-values greater than 10. Please find detailed information in Supplementary Table S1. In the reverse MR study, we employed the same screening process as described above to filter SNPs. Detailed results of the SNPs can be found in Supplementary Table S2. Detailed results of the SNPs, including F-statistics for instrument strength, have been provided in Supplementary Tables S1 and S2. Notably, all SNPs had F-values greater than 10, meeting the conventional threshold for strong instruments in MR analysis.

Causal effect of uveitis on PsA

Meta-analysis results for the forward MR outcomes have been shown in Figure 2a. We discerned a causal association between uveitis and PsA, as indicated by the IVW method (OR=1.63, 95% CI: 1.22-2.19, P-value = 0.001). The reverse MR outcomes have been shown in Figure 2b (OR=1.55, 95%CI: 1.07-2.24, P-value = 0.02). The effects of a single SNP on uveitis and PsA has been depicted in the scatter plot [Supplementary Figure S1-S6 (A)] and forest plot [Supplementary Figure S1-S6 (B)]. Detailed results of the MR analysis have been shown in Supplementary Table S3.

Figure 2: Meta-analysis results for the forward MR outcomes and reverse MR outcomes. The causal estimates are presented as odds ratios (ORs) along with their corresponding 95% confidence intervals (CIs).

Export to PPT

Sensitivity analysis

This study conducted a total of six sensitivity analyses, including heterogeneity and horizontal pleiotropy analyses. In all analyses, no horizontal pleiotropy was detected. Detailed horizontal pleiotropy and heterogeneity results have been provided in Supplementary Tables S4, and leave-one-out results have been provided in Supplementary Figure S1-S6 (C).

In addition, we incorporated detailed pleiotropy and heterogeneity analyses into the main text. For the uveitis → PsA direction, MR-Egger intercepts were not statistically significant across datasets (e.g., IEU-GWAS to GWAScatalog: intercept = 0.1738, SE = 0.1746, P = 0.393), indicating no evidence of directional pleiotropy. Cochran’s Q tests revealed moderate heterogeneity (Q = 24.84, df = 4, P = 5.42×10⁻⁵ for GWAScatalog dataset), more pronounced in FinnGen (Q = 92.74, df = 5, P = 1.79×10⁻1⁸). In the PsA → uveitis analysis, MR-Egger intercepts similarly did not indicate horizontal pleiotropy (e.g., intercept = 0.0673, P = 0.231 in IEU-GWAS). However, Cochran’s Q statistics suggested significant heterogeneity in some datasets (Q = 154.69, df = 15, P = 2.81×10⁻2⁵). These findings suggest that while instrument heterogeneity exists, the risk of directional pleiotropy bias is low.

Discussion

Multiple investigations have demonstrated an association between uveitis and PsA, with the prevalence of uveitis being significantly higher in patients with PsA compared to the control group.24 Chen et al.10 found an increased risk of subsequent PsA in uveitis patients. Additionally, a nationwide cohort study in Korea revealed a higher risk of uveitis in patients with PsA.25 These findings suggest a bidirectional association between uveitis and PsA. Interestingly, a nationwide study in Denmark demonstrated a similar bidirectional relationship, indicating an increased risk of PsA in uveitis patients, and an increased risk of uveitis in patients with PsA.26 It is important to note that MR estimates reflect the effects of lifelong genetic predisposition to a trait, rather than the impact of short-term exposures or treatments. This should be considered when interpreting the directionality and clinical relevance of the observed associations.

In our MR analysis, we explored the causal relationship between uveitis and PsA. Our findings suggest that uveitis may serve as a risk factor for the development of PsA. The reverse MR analysis also suggested that PsA might also predispose individuals to uveitis, indicating a bidirectional causal relationship between the two conditions. This discovery aligns with previous observational studies.

More and more evidence suggests that some loci significantly associated with PsA are unrelated to psoriasis at the genome-wide significance threshold, including loci at CSF2, PTPN22, TNFAIP3, and HLA-B. Interestingly, a meta-analysis revealed that PTPN22 rs2488457 conferred strong susceptibility to uveitis in general.27 Numerous studies have demonstrated a close association between HLA-B27 and PsA, with HLA-B27 playing a significant aetiological role in PsA. Haroon et al. found that HLA-B27 and HLA-B39 are significantly associated with psoriatic arthritis.28 In addition to its strong association with PsA, HLA-B27 is closely associated with anterior uveitis, representing 18% to 32% of cases, while anterior uveitis constitutes 50% to 92% of total uveitis cases.29 Furthermore, a transcriptomic study found an upregulation of immunoglobulin genes in PsA patients compared to those with psoriasis, potentially promoting the spread of inflammation from the skin compartment to other tissues. Considering previous research and our study results, we propose a hypothesis that the bidirectional positive causal relationship between uveitis and PsA may be associated with HLA-B27, PTPN22, and the upregulation of immunoglobulin genes. The precise underlying mechanisms require further confirmation.

The primary strength of this study lies in its pioneering application of MR to explore the causal relationship between genetically predicted uveitis and PsA. By leveraging the principle of random allele allocation at conception, MR helps mitigate confounding and reverse causation—two major limitations of observational studies. Additionally, all participants were of European ancestry, which minimises the impact of population stratification within the analysed datasets.

However, several limitations should be acknowledged. First, the study population was restricted to individuals of European descent, which may limit the generalisability of the findings to non-European populations. Second, uveitis is a heterogeneous condition encompassing various subtypes, such as anterior, intermediate, posterior, and panuveitis. Due to the lack of subtype-specific GWAS datasets, subgroup analyses were not feasible. Third, although we applied sensitivity analyses such as MR-Egger and heterogeneity tests, residual horizontal pleiotropy cannot be fully ruled out. Fourth, some analyses used a relatively small number of instrumental SNPs, which may raise concerns regarding weak instrument bias. Lastly, this study relied on summary-level GWAS data, precluding adjustment for individual-level confounders such as medication use, disease duration, or clinical severity. These methodological limitations are inherent to the MR framework and should be considered when interpreting the results.

Limitations

Restricted to European ancestry, lack of uveitis subtypes, potential pleiotropy, limited SNPs, and no individual-level adjustment.

Conclusion

In summary, this study concludes that the presence of uveitis increases the risk of PsA, and conversely, the onset of PsA also promotes the occurrence of uveitis, indicating a bidirectional causal relationship between the two. This research provides new insights into understanding the relationship and potential mechanisms between uveitis and PsA, offering clues for the prevention and treatment of both uveitis and PsA.