To assess causality, we conducted a Mendelian randomization analysis using genetically predicted antibody responses to infections as exposures and the five ocular diseases as outcomes, selecting associated SNPs as instrumental variables. Summary data on antibody-mediated immune responses were obtained from the GWAS Catalog. ¹ These data originated from the study by Butler-Laporte et al. [22], who performed serological profiling on 9,724 adults of European ancestry from the UK Biobank using 13 pathogens to define 46 phenotypes. Details of all phenotypes are provided in Supplementary Table S1. Summary statistics for the five ocular diseases were sourced from the FinnGen GWAS. ² All participants were of European ancestry. Case and control numbers were as follows: chronic iridocyclitis (1,869/473,095), scleritis (441/473,095), neovascular AMD (6,699/331,070), diabetic retinopathy (14,142/82,287), and glaucoma (26,591/473,757). Details of the case and control groups for each eye disease are shown in Supplementary Table S2.

In our MR analysis, single nucleotide polymorphisms (SNPs) served as instrumental variables (IVs), following three critical assumptions: (1) association with the exposure, (2) independence from confounders, and (3) influence on the outcome exclusively through the exposure (Figure 1). Rigorous selection of instrumental variables (IVs) is essential for ensuring the robustness of Mendelian randomization (MR) analysis. Initially, instrumental variables (IVs) must exhibit a robust association with the exposure, with the standard threshold for genome-wide significance established at p < 5 × 10⁻⁸. To maintain an adequate number of instrumental variables for each phenotype, we adjusted the threshold to p < 5 × 10⁻⁶. A screening procedure was conducted to eliminate the effects of linkage disequilibrium (LD), confirming that the chosen SNPs displayed low LD (r² < 0.001 within a 10,000 kb window) [23].

The strength of the IVs was assessed using the F-statistic, defined as F = R 2   N − K − 1 / K 1 − R 2 . R² is defined as the proportion of variance in exposure explained, where N denotes the sample size and K indicates the number of instruments. The formula for calculating R² is as follows: R ^ 2 = 2 × MAF × 1 − MAF × β / SD ^ 2 . In the equation, MAF represents the minor allele frequency, β denotes the effect size, and SD indicates the standard deviation [24]. To rule out weak instrument bias, all included SNPs were ensured to have F-statistics greater than 10.

We conducted forward MR analyses to evaluate causal relationships between genetically predicted infection-related antibody responses and five ocular diseases, employing five methodologies: inverse-variance weighted (IVW), MR-Egger, weighted median, simple mode, and weighted mode. The IVW method was selected as the primary approach due to its assumption of valid instruments without horizontal pleiotropy, while the other methods functioned as complementary analyses [25]. All analyses were performed using R (version 4.5.1) with the TwoSampleMR package (version 0.6.22). The presence of 46 immune phenotypes associated with five ocular diseases as outcomes heightens the likelihood of Type I errors due to repeated testing. Consequently, we utilized Bonferroni correction, which is a more rigorous method than FDR adjustment, for multiple significance testing. A p-value of less than 2.17 × 10⁻⁴ (0.05/46/5) was considered statistically significant.

Reverse MR was conducted to assess the causal relationships between five ocular diseases (as exposures) and antibody-mediated immune responses (as outcomes), adhering to an analytical procedure aligned with the forward MR approach. The same Bonferroni correction method is employed.

To evaluate the validity of instrumental variables and robustness of the results, we conducted comprehensive sensitivity analyses. Heterogeneity was assessed using Cochran’s Q statistic from IVW and MR-Egger regression, with P > 0.05 indicating no significant heterogeneity [26]. Horizontal pleiotropy was examined via the MR-Egger intercept test, where P > 0.05 suggested no evidence of directional pleiotropy, and the MR-PRESSO global test was applied to identify and remove outliers. If outliers existed, the MR analysis was repeated after eliminating these outliers [27, 28]. Additionally, leave-one-out analyses were performed to assess whether overall associations were driven by individual influential variants and to visualize the contribution of each SNP. To evaluate the overall robustness, funnel plots were generated using R (version 4.5.1) to assess the symmetry of the SNPs.

Based on forward MR analyses, Epstein-Barr virus (EBV) ZEBRA antibody levels, varicella-zoster virus glycoprotein E and I (VZV gE & gI Ab) antibody levels, and anti–JC virus IgG seropositivity (JCV IgG+) were identified as risk factors for multiple ocular diseases. The results of sensitivity analyses and visualisations are presented in Supplementary Table S3, Supplementary Figures S1, S2.

To evaluate reverse causation, we performed reverse MR analyses to assess the effects of ocular diseases on viral antibody levels. DR had a positive causal effect on JCV VP1 antibody levels after strict multiple testing correction (OR = 1.195, 95% CI: 1.105–1.292, P = 8.150e-06), corresponding to a 19.5% increased risk of elevated JCV VP1 antibodies in DR patients. Consistency across MR-Egger, weighted median, IVW, and weighted mode methods strengthened causal credibility (Figures 3, 5). While the MR-Egger intercept indicated potential minor horizontal pleiotropy, neither the MR-PRESSO nor Cochran’s Q test showed significance (Supplementary Table S3; Supplementary Figures S1, S2).

A nominally significant causal relationship was observed between CIR and EBV ZEBRA antibody levels (OR = 1.082, 95% CI: 1.009–1.161, P = 0.025). However, this association did not survive strict multiple testing correction (Bonferroni-corrected threshold P < 0.0021) and should therefore be interpreted with caution.

EBV is a prevalent pathogen, infecting more than 90% of the global population, and is linked to several autoimmune diseases and malignancies [29]. We present genetic evidence indicating that the ZEBRA antibody, derived from the immediate-early gene BZLF1 that triggers the viral lytic cascade, causally increases the risk of CIR through various potential mechanisms.

One key mechanism is molecular mimicry. EBV proteins contain epitopes that resemble human proteins, and viral reactivation causes cell death and self-antigen release [30]. Upon presentation by antigen-presenting cells (APCs), these antigens may activate autoreactive T cells, which can cross-react with ocular tissues and trigger iridocyclitis. Studies have shown that in uveitis patients, Th1 cells activate macrophages via IFN-γ, while Th17 cells sustain chronic inflammation through IL-17. Together, these responses promote leukocyte recruitment and local inflammatory cascades, ultimately leading to tissue damage via oxidative stress and pro-inflammatory cytokine release [31, 32].

Secondly, by modulating signaling pathways, EBV promotes B cell differentiation into T-bet⁺ B cells [30]. These cells subsequently produce pathogenic IgG1/IgG3 autoantibodies and IFN-γ, establishing a pro-inflammatory feedback loop [33]. Their additional role in disrupting tissue barriers may permit CXCR3⁺ B cell migration across the blood-aqueous barrier, initiating ocular inflammation. This is corroborated by the marked increase in IL-6 and chemokines (CXCL13, CCL8, CCL13, CCL20) in the aqueous humor of uveitis patients [34].

Additionally, the EBV early lytic protein BALF0/1 hijacks host caveolin-mediated endocytosis and ERAD pathways during viral replication to degrade the B-cell receptor complex [35]. Thus, BZLF1 (ZEBRA)-driven promoter activation likely upregulates BALF0/1, disrupting B-cell function, promoting viral release, and ultimately causing local ocular immune dysregulation that contributes to chronic iridocyclitis.

The link between EBV specific immune responses and autoimmune uveitis is further supported by Hendrikse et al. [17], who observed elevated antibody levels against the RRPFFHPV motif of Epstein–Barr virus nuclear antigen 1 (EBNA-1) in pediatric uveitis patients carrying the *HLA-DRB1*15:01* risk allele, with the strongest IgG signals detected in those with non-anterior uveitis. This strongly suggests a role for EBNA-1–directed immunity in certain forms of autoimmune uveitis. In contrast, our analysis supports EBV EBNA-1 antibody levels as a protective factor against DR, wet AMD, and glaucoma. We thus propose that beyond its potential role as an initial trigger or cross-reactive target, EBNA-1 may also induce a state of immune tolerance or modulate dominant inflammatory pathways, thereby conferring a protective effect in these ocular disorders.

EBV may exhibit differential impacts on ocular diseases depending on the stage of the viral life cycle. In our study, latent infection marked by EBNA-1 antibodies conferred protection against DR, wet AMD, and glaucoma, implicating a modulatory role in the chronic immune dysregulation common to these diseases. Chronic inflammation and immune dysregulation are fundamental pathogenic drivers common to these three conditions [36, 37]. In DR and wet AMD, activated glial cells and macrophages secrete critical mediators including VEGF, TNF-α, IL-1β, and IL-6, which facilitate inflammation and angiogenesis [11, 38]. In glaucoma, T-cell infiltration and microglial activation play a role in the loss of RGCs, which is further aggravated by cytokines such as TNF-α, IL-6, and IL-8 [39, 40]. The protective role of EBNA-1 indicates its potential involvement in the modulation of inflammatory pathways.

Research indicates that EBNA-1 can suppress the canonical NF-κB pathway by inhibiting phosphorylation of IKKα/β [41]—a key signaling axis driving inflammation in diseases such as DR and AMD. By dampening NF-κB activation, EBNA-1 may reduce the expression of pro-inflammatory genes, albeit in a manner that also favors viral persistence and oncogenesis. Furthermore, latent EBNA-1 infection evades cytotoxic immune responses and may promote immune tolerance, possibly through the induction of regulatory T cells or specific antibody subtypes that mitigate auto-reactive immunity against ocular antigens [42]. This modulated immune environment may also restrain excessive macrophage and microglial activation in the retina and choroid [43], thereby limiting VEGF and pro-inflammatory cytokine production and potentially slowing disease progression.

Conversely, our analysis supports a causal role for JCV IgG+ seropositivity in DR, age-AMD, and glaucoma. This ubiquitous virus, which persists asymptomatically in most adults, may reactivate under immunosuppression, enter circulation, and promote ocular neurovascular pathology [44, 45]. Critically, studies have confirmed that JCV can infect human cerebral vascular pericytes to breach the blood-brain barrier and subsequently invade astrocytes, causing neural damage [46]. As an extension of the central nervous system, the retina shares key developmental and structural features with the brain, and its blood-retinal barrier (BRB) relies on pericytes and Müller cells as essential components. Therefore, JCV may similarly target the BRB through comparable mechanisms, contributing to the vascular leakage pathology observed in DR and AMD.

Beyond direct viral invasion, virus-induced immune responses may constitute a key pathogenic mechanism. Murinello et al. [47] observed co-deposition of IgG, C1q, and membrane attack complexes in early AMD eyes, along with increased numbers of FcγRIIa- and FcγRIIb-expressing inflammatory cells in the choroid. Thus, JCV-specific antibodies may form local immune complexes in the retina, activating macrophages or microglia via Fcγ receptors and thereby driving chronic inflammation. Neurotropic viruses such as VZV and HSV-1 promote Aβ aggregation—a key component of AMD-associated drusen—through oxidative stress, calcium dysregulation, and impaired autophagy [48, 49]. As a neurotropic virus, JCV may similarly enhance AMD susceptibility via analogous Aβ-inducing pathways. In glaucoma, JCV could directly damage RGCs, supported by detections of JCPyV DNA and viral inclusions in intraocular fluid and RGCs [50]. These proposed mechanisms, however, await confirmation through larger clinicopathological studies.

Moreover, genetically predicted levels of VZV gE & gI antibodies showed positive causal associations with scleritis and DR. VZV, a double-stranded DNA virus, establishes latency in trigeminal ganglia after primary infection. Upon immunosuppression, reactivated virus can travel along axons to the eye, causing herpes zoster ophthalmicus, which clinically manifests as uveitis, scleritis, or retinal and optic nerve inflammation [18, 51].

VZV may increase susceptibility to DR and scleritis by targeting ocular vasculature. Studies demonstrate that VZV infects vessels via the ophthalmic branch of the trigeminal nerve, subsequently spreading trans-adventitially, disrupting the internal elastic lamina, and triggering intimal hyperplasia and a pro-inflammatory state [52].

Furthermore, the VZV glycoprotein gE acts as a key regulator of PINK1/Parkin-dependent mitophagy. By interacting with the autophagy protein LC3, gE induces substantial mitochondrial reactive oxygen species (mtROS) production [53]. Given that oxidative stress and mitochondrial dysfunction are central to DR pathogenesis, VZV likely exacerbates DR progression by modulating the PINK1/Parkin pathway.

Immune dysregulation further contributes to disease pathogenesis. VZV-induced mitophagy suppresses interferon production by inhibiting the STING and MAVS pathways, leading to elevated pro-inflammatory cytokines and NLRP3 inflammasome activation [53]. The release of these mediators attacks the vascular sclera and compromises retinal vascular stability. Clinical evidence confirms VZV as a cause of scleritis, as demonstrated by the detection of viral DNA in aqueous humour. [16, 54]. The efficacy of immunosuppressants such as methotrexate in some scleritis patients further supports the role of VZV-driven immune dysregulation in disease progression.

Reverse MR analysis indicates a genetically determined positive causal relationship between DR and JCV VP1 antibody levels. Critically, This finding indicates that genetic susceptibility to DR may predispose individuals to JCV reactivation by modifying the immune environment, rather than suggesting a direct, acute impact on antibody titers.

In diabetic patients, hyperglycemia/advanced glycation end products impair T-cell and B-cell function, leading to immune dysregulation [11]. The advancement of DR entails the disintegration of the BRB and the liberation of inflammatory mediators, including TNF-α. This state of chronic inflammation and immune dysregulation may create a favorable environment for the reactivation of JCV latent in peripheral organs, such as the kidneys. At the molecular level, studies indicate that NF-κB and C/EBPβ regulate JCV transcription through the κB motif within the viral genome, with both transcription factors being modulated by proinflammatory cytokines [55]. Furthermore, latent viruses in oligodendrocytes and astrocytes can be reactivated by inflammatory stimuli, enabling viral protein expression and subsequent replication [56]. Our findings suggest that JCV infection may worsen ocular pathology via mechanisms like BRB disruption and that the DR microenvironment may facilitate JCV reactivation. This bidirectional relationship, marked by mutual reinforcement, provides a more thorough and dynamic understanding of the intricate interaction between viral infection and ocular disorders.

This study has several limitations. Its generalizability is constrained by the use of genomic data primarily from European ancestry, and importantly, the MR design itself captures the effect of lifelong genetic exposure to antibody levels, which differs from the impact of acute infection or transient serological changes. Moreover, despite thorough sensitivity analyses including Cochran’s Q, MR-Egger, and MR-PRESSO to account for pleiotropy and heterogeneity, residual confounding may still occur. The application of a strict Bonferroni correction may, to a certain degree, elevate the likelihood of Type II errors. Finally, the mechanistic insights regarding specific pathogens are still lacking and require additional experimental and clinical investigation.