Causal association between glaucoma and risk of retinal vascular occlusion: a Mendelian randomization study

Abstract

Objective

Retinal vein occlusion (RVO) leads to visual impairment, and risk factors may include glaucoma. Although some studies suggest a relationship between glaucoma and RVO, it is unknown whether this association is causal. We performed Mendelian randomization (MR) analyses to evaluate the causal contribution to RVO of five genetically predicted glaucoma conditions.

Methods

We conducted two-sample univariable MR analysis using inverse variance weighting, weighted median, and MR-Egger methods. We obtained publicly available datasets of genome-wide association studies (GWAS) meta-analyses for glaucoma as the exposure and a GWAS for RVO in the Finn Gen biobank study as the outcome.

Results

Genetically predicted glaucoma was causally associated with RVO risk (beta = 0.267, standard error [SE] = 0.052, odds ratio [OR] = 1.306, 95% confidence interval [CI]: 1.181–1.445). This association was supported by sensitivity analyses using weighted median (beta = 0.211, SE = 0.075, OR = 1.235, 95% CI: 1.067–1.429), weighted mode (beta = 0.491, SE = 0.132, OR = 1.633, 95% CI: 1.261–2.116), and MR-Egger (beta = 0.3900, SE = 0.138, OR = 1.476, 95% CI: 1.128–1.933) methods. MR analyses using the validation dataset obtained consistent results.

Conclusion

Our results indicated that glaucoma is likely causally associated with an increased risk of RVO. The current findings may help in determining the underlying mechanisms via which glaucoma affects the risk of RVO.

Introduction

Retinal vein occlusion (RVO) is the second most common retinal vascular disorder and is a main cause of visual impairment. ¹ It is estimated that RVO affects more than 25 million people worldwide. ² Patients with RVO are at high risk of vision loss owing to a number of complications associated with interrupted blood flow and even tractional retinal detachment. ³ The pathological mechanisms of RVO are gradually being recognized, yet clinical prevention and exploration of risk factors have not been fully revealed.

Glaucoma is a group of ocular diseases characterized by progressive damage to the optic nerve. An important factor in glaucoma is the relative increase in intraocular pressure (IOP), which can lead to irreversible loss of vision. ⁴ In clinical work, ophthalmologists are concerned about IOP, visual field defects, and damage and progression of the optic nerve in the fundus of patients with glaucoma. ⁵ However, in consultations, we found that some of our patients with RVO had a history of glaucoma. Recently, a series of clinical studies demonstrated a significant association between glaucoma and the risk of RVO. ⁶ According to Galvez-Olortegui et al., a systematic review of the extensive literature on RVO found that the risk for glaucoma, as a factor in the pathogenesis of RVO, exceeded 10%. ⁷

It is important to clarify the causal relationship between glaucoma and RVO. Mendelian randomization (MR) is a method of prospective causal inference that has demonstrated success in identifying risk factors for diseases. ⁸ In MR, genetic variations are considered to be a natural experiment in which individuals are randomly assigned to higher or lower levels of exposure over their lifetime. ⁸ MR, which is not susceptible to the effects of common confounding variables and exhibits a logical causal sequence, represents a state-of-the-art analytic tool for investigating the causal relationship between exposure factors and disease at the genetically inherited level. ⁹ To date, MR analysis has not been used to detect the effect of glaucoma on RVO risk. In this study, we analyzed the causal relationship between glaucoma and RVO at the genetic level using MR.

Methods

The Strengthening the Reporting of Observational Studies in Epidemiology using MR (STROBE-MR) guideline was rigorously adhered to in the reporting of this MR study. ⁹ A series of MR analyses were conducted with the objective of establishing a causal relationship between glaucoma and the risk of RVO. This MR study was conducted using publicly available summary statistics of genome-wide association studies (GWAS); all original studies received ethical approval. In the present MR study, we only analyzed publicly available summary statistics; therefore, this study was exempt from ethical approval as well as the requirement for informed consent. All study procedures adhered to the tenets of the Declaration of Helsinki. The following assumptions were made for the purpose of making inferences based on the MR method: 1) genetic variants are strongly and causally related to exposure; 2) genetic variants are not associated with any potential confounders; 3) the relationship between genetic variants and outcomes is not consistent. For ease of understanding, the above principles are illustrated using diagrams (Figure 1).

Figure 1.

Key assumptions in our Mendelian randomization study evaluating the causal contribution of glaucoma type to retinal vein occlusion (RVO). (1) Assumption 1: instrument single nucleotide polymorphisms (SNPs) are strongly associated with glaucoma (p < 5.0 × 10⁻⁸); (2) Assumption 2: SNPs are not related to other confounders; (3) Assumption 3: SNPs affect RVO risk through glaucoma but not through other pathways.

Genetic instruments for glaucoma

Summary data of single nucleotide polymorphisms (SNPs) associated with glaucoma were obtained from a GWAS meta-analysis conducted by Craig et al., which included 133,492 glaucoma cases and 90,939 control participants of European ancestry. ¹⁰ The diagnosis of glaucoma was based on self-reported medical history and International Classification of Diseases codes in linked electronic medical health records. ¹⁰ We selected glaucoma-related SNPs with genome-wide significance (p < 5.0 × 10⁻⁸). ¹⁰ We used the linkage disequilibrium clumping function of the MR package in R (version 4.0.3; www.r-project.org). Following linkage disequilibrium clumping (r² = 0.001) and harmonization of the glaucoma summary-level data with the corresponding RVO summary statistics, the instrument sets were found to contain 85 instrument SNPs for glaucoma.

RVO outcome data

For outcome data, summary statistic datasets were extracted from a recent GWAS of RVO conducted by the FinnGen biobank, adjusted for sex, age, genotyping batch, and the first 10 principal components. The study included 1595 cases and 203,108 controls from among individuals of European ancestry. The Base database was available at no cost at app.mrbase.org. ¹¹ SNPs reaching GWAS (GWAS p < 5 × 10⁻⁸) with a minor allele frequency of >0.01 were included in the analysis. ¹¹ SNPs were clustered based on linkage disequilibrium (r² = 0.001) within the specified genome region (kb = 10,000). Palindromic SNPs were discarded. ¹⁰ To assess efficacy of the selected genetic predictors for RVO, an F statistic (F = beta²/standard error [SE]²; beta: beta for the SNP−exposure association, SE: variance) was calculated for each SNP. ¹² In general, an F value greater than 10 indicates that the presence of weak instrumental variables (IVs) does not result in a significant bias. We excluded SNPs with a lower statistical power (F value <10). The MR-Steiger filtering method was used to identify and remove those variations that exhibited a stronger correlation with RVO than with glaucoma. ¹¹

The causal effect in two-sample MR analysis was estimated using the standard inverse variance weighting (IVW) method. Furthermore, we used the MR-Egger and weighted median methods as supplementary approaches to the IVW method. We used Cochrane 's Q-test to determine the heterogeneity of IVs, with I². A value p < 0.05 indicated the presence of heterogeneity. ¹³ I² was categorized as low, medium, or high, with >25% of the values considered to have significant heterogeneity. Furthermore, the heterogeneity of MR-Egger analysis was evaluated by calculating Rucker's Q. ¹⁴ A p-value of <0.05 for Cochrane's Q and Rucker's Q (Q−Q′) indicated that the MR-Egger method was superior owing to horizontal multiplicity imbalance. We examined cross-sectional pleiotropy using the MR-Egger intercept and MR-PRESSO global test, with p < 0.05, suggesting that there may be other pathways via which exposure affects IV outcomes.^15–17 A p-value <0.05 was deemed to be statistically significant. All MR analyses were conducted using R version 4.1.2 and the R package MR-PRESSO.

Results

Eighty-five independent SNPs were selected from GWASs on glaucoma as IVs. All selected SNPs were significantly associated with glaucoma at the genome-wide level (Table 1, Figure 2). Thirty-eight of the 68 SNPs were found to be positively associated with RVO, although the results were not statistically significant (Table 1). Genetic variants serving as IVs explained 1.4% of the variance in the exposure (value of the r² statistic). A p-value 5 × 10⁻⁸ corresponds to an F statistic greater than 30 for each single variant. A threshold of F = 10 was used to define a “weak IV.” Consequently, the bias resulting from use of an inadequate instrument was insignificant.

Table 1.

MR estimates from each method assessing the causal effect of glaucoma on the risk of RVO.

MR method	Number of SNPs	Beta	SE	95%confidence interval	Association p‐value	Cochran Q statistic	I2	Heterogeneity p‐value
Inverse variance weighting	85	0.2673	0.0515	1.181–1.445	0.003	99.7	0.168	0.10
MR-Egger	85	0.3896	0.1375	1.128–1.933	0.624	100.8	0.167	0.11
Weighted median	85	0.2111	0.0745	1.067–1.429	0.112	–	–	–
Weighted mode	85	0.4905	0.1321	1.261–2.116	0.053	–	–	–

a

I² = (Q−df)/Q. ²⁴

SNP, single nucleotide polymorphism; MR, Mendelian randomization; Beta, beta coefficient; SE, standard error.

Figure 2.

Forest plot of the causal effect on retinal vein occlusion of single nucleotide polymorphisms associated with glaucoma. Significance of the red line is the Mendelian randomization (MR)-Egger test and MR results of the inverse variance-weighting method.

The results of our analysis using the IVW method showed evidence to support a causal association between glaucoma and RVO. We found that for every unit increase in glaucoma, there was a corresponding increase in RVO (beta = 0.267, SE = 0.052, p = 0.003). This indicated that glaucoma is likely to have a direct effect on the development of RVO. Additionally, we used the MR-Egger test to assess the presence of directional pleiotropy, which occurs when the genetic variant affects both the exposure and the outcome through different pathways. Our analysis revealed an intercept value of −0.013 with a p-value of 0.34. This suggests that directed pleiotropy was unlikely to bias our result, which supported the validity of a causal association between glaucoma and RVO. However, when we performed MR-Egger regression analysis, we found no evidence of a causal association between glaucoma and RVO (beta = 0.390, SE = 0.137). Similarly, the weighted median approach also yielded no evidence of a causal association (beta = 0.211, SE = 0.074, p = 0.004). Interestingly, when we used the weighted mode method, evidence was found to support a causal association between glaucoma and RVO (beta = 0.491, SE = 0.132, p = 0.05). This method considers the most common estimate across multiple instrument variables, suggesting that glaucoma has a causal effect on the risk of developing RVO. Overall, our findings highlight the discrepancy between different methods used to assess causal associations. Although the IVW and weighted mode methods suggest a causal effect, the MR-Egger method suggests a null causal effect. It is important to note that IVW has the advantage of retaining greater precision in the estimates compared with the MR-Egger analysis. This further emphasizes the need for careful consideration and interpretation of the results when determining causal associations.

The Cochran's Q test, conducted on the individual variants listed in Table 1, revealed no evidence of heterogeneity among the IV estimates. Heterogeneity refers to the variability observed in the causal estimates obtained for each SNP, which essentially indicates the consistency of causal estimates across all SNPs. The low heterogeneity observed in this study suggests that the MR estimates are more reliable. This implies that the instruments used in this study yielded consistent results, increasing the confidence in the causal estimates obtained. Furthermore, a “leave-one-out” analysis was conducted to identify any single SNP that may be driving IVW point estimates. However, no specific SNP was found to have a significant influence on the overall point estimate. This suggests that the results obtained using the IVW method were not heavily influenced by any one particular SNP, strengthening the validity of the overall estimate. To assess the presence of directional horizontal pleiotropy, which can introduce bias in MR methods, we examined the asymmetry in a funnel plot (Figure 3). However, neither the funnel plot nor the MR-Egger regression test showed signs of asymmetry, indicating that directional horizontal pleiotropy was not a major concern in this study. This suggests that the MR estimates were not significantly biased by confounding factors or alternate pathways, which further enhances the credibility of our study findings. The results in Table 1 show low heterogeneity among the IV estimates, indicating increased reliability of MR estimates. Additionally, the leave-one-out analysis showed that no single SNP had a significant impact on the overall estimate, providing further support for the validity of the results. Additionally, no asymmetry was found in the funnel plot and MR Egger regression tests, suggesting that there was no evidence of directed level pleiotropy, suggesting that the MR estimates were not substantially biased by confounding factors (Figures 3 and 4).

Figure 3.

Scatterplot of glaucoma genetic associations versus retinal vein occlusion genetic associations. The slope of each line represents the causality of each method. Blue line represents the inverse variance-weighted estimate, green line the weighted median estimate, and dark blue line represents the Mendelian randomization-Egger estimate.

Figure 4.

Funnel plot for assessing heterogeneity. Blue line represents the inverse variance-weighted estimate, and dark blue line represents the Mendelian randomization-Egger estimate.

Discussion

According to the current literature, ours is the first study to discuss the causal relationship between glaucoma and RVO at the genetic level. Glaucoma has been identified as a potential risk factor for the development of RVO. ³ Nevertheless, the precise relationship between glaucoma and RVO remains uncertain. Four estimation methods were used in the MR analyses, namely, the IVW method, weighted median method, MR-Egger regression, and weighted mode. Two previous studies on risk factors in patients with RVO have used these methods, making the feasibility of this study clear. Zheng et al. found a causal relationship between lipids and RVO using MR and concluded that high-density lipoprotein cholesterol is a significant risk factor for the development of to RVO, as compared with other lipid types. ¹⁸ Using the MR method, Huang found that type 2 diabetes mellitus is a risk factor for RVO. ¹⁹ The results of our study indicated that the associations between glaucoma and RVO may be causal. Although the MR estimates derived from analysis using IVW, weighted median, MR Egger, and weighted mode were not consistent, the results of IVW and weighted mode analyses supported a causal association between glaucoma and RVO. In comparison to MR-Egger analysis, IVW is advantageous in that it retains greater precision in the estimates. The present MR analysis indicated a potential causal role of glaucoma in the risk of RVO. Furthermore, the beta values calculated using all four MR methods were positive and in the same direction, which also demonstrated that glaucoma is a risk factor for the occurrence of RVO. Thus, our study corroborates the association found in previous observational studies.

A meta-analysis demonstrated that glaucoma represents a significant risk factor for RVO. ²⁰ It has been proposed that chronic or primary open angle glaucoma (POAG) may increase the incidence of RVO, particularly central RVO. ²¹ Concurrently, a reduced correlation was found between primary angle-closure glaucoma and RVO, particularly branch RVO; however, heterogeneity in that study was unavoidable. ⁷ In the Korean National Health and Nutritional Examination Survey, a cross-sectional study confirmed the non-association between glaucoma and RVO. ¹⁹ Thus, previous studies of glaucoma and RVO have yielded correlations, but the existence of an association between the two remains controversial. Furthermore, ours was the first study to discuss the causal relationship between glaucoma and RVO ate genetic level. In eyes with narrow angles, intermittent angle closure episodes may result in the onset of venous occlusion owing to collapse of the vein. This, in turn, may lead to endothelial lesions, which may subsequently give rise to intimal proliferation at the level of the lamina cribrosa. ²⁰ Călugăru et al. elaborated three risk factors for central/hemicentral RVO are angle closure, ²² permanently high IOP, and glaucomatous cupping of the optic disc. These risk factors corroborate that remaining alert for RVO is important in patients with glaucoma.

Beaumont and Kang studied the association between glaucoma and RVO from an anatomical point of view and concluded that there was no evidence to suggest that RVO with optic nerve head swelling within the optic nerve head behind the lamina cribrosa is associated with high values of the cup-to-disc ratio, IOP, or the prevalence of POAG. ²³ The appearance of optic nerve head swelling in RVO indicates that venous occlusion results in sufficient ischemia immediately behind the lamina cribrosa, which effectively blocks axoplasmic transport. ²⁴ Shin et al. explored the mechanisms involved from a cell biological perspective. It has been postulated that cell proliferation in the lumen of the vein at the level of the lamina cribrosa and displacement of the vessel in the anterior portion of the optic nerve may be causative factors in the development of central/hemicentral RVO. ²⁴ In recent years, as the study of biomechanics in ocular diseases has advanced, the biomechanical mechanisms associated with RVO have been gradually revealed. ²⁵ Physical obstruction of the retinal venous system is generally caused by a series of pathological processes including thrombosis, vein wall deformation, and external biomechanical compression secondary to glaucoma, suggesting a biomechanical correlation. ²¹ Mechanistically, we hypothesize that elevated IOP compresses the cranial lamina propria and the optic disc and surrounding tissue structures, which leads to stretching and weakening of the vessel walls, making the retinal veins more inclined to occlusion. ²⁶ Concurrently, the direct biomechanical impact of elevated IOP impedes drainage of the retinal veins, inducing venous stasis, which in turn exacerbates the intimal proliferation in the vein. ²⁷ However, none of the above studies have argued for a causal relationship between glaucoma and RVO.

MR minimizes the bias inherent in observational studies. ²⁸ However, MR studies are highly susceptible to pleiotropy, which we define as the association of a genetic variant with more than one variable. Genetic variants may be associated with multiple phenotypes, a phenomenon known as pleiotropy, which may confound the estimates derived from MR and lead to biased results owing to biased estimates of causality. Whereas the inclusion of multiple variants in MR analysis enhances the statistical power, this may also result in the inclusion of pleiotropic genetic variants that are not valid IVs. Consequently, it is necessary to conduct sensitivity analyses to ascertain the validity of the MR findings. To eliminate pleiotropy, weighted mode estimation was used, which provides valid estimates, even in the event that 50% of SNPs are not valid instruments. MR-Egger regression was applied to test for the presence of unbalanced pleiotropic effects and to provide causal estimates of the effect of exposure on outcomes in the presence of pleiotropic effects. The results of these four methods were inconsistent. The data presented here provide further support for the association between glaucoma and RVO, as previously reported in observational studies. The results of the current study may provide clearer evidence and ideas for determining the mechanisms via which glaucoma affects RVO risk.

To the best of our knowledge, this was the first study to use MR to reveal a causal relationship between glaucoma and RVO. Nevertheless, it is important to acknowledge that our MR study has some limitations, which must be taken into account when interpreting the results. As mentioned earlier, the biggest concern in MR studies is horizontal pleiotropy, where genetic variation affects the outcome of more than one pathway. ¹⁴ Additionally, the pooled statistics of patients with glaucoma and RVO in this study were obtained from individuals of European origin, which may have biased the results of the sample. This reduced the impact of population structure bias, but limited the generalizability of our findings to other ethnic groups. Further investigation is required to determine whether there is a causal relationship between genetically predicted glaucoma and RVO in Asian populations, given the availability of relevant data. Furthermore, we did not undertake a separate assessment of the genetic prediction of associations between different types of glaucoma and RVO because the gene pool currently lacks this information. It is important to note that although the most common risk factors for RVO were taken into consideration in our analysis, it was not possible to completely exclude the influence of other factors that have not yet been explored.

Conclusion

The results of MR analysis indicated that glaucoma is likely to be causally associated with an increased risk of RVO. The current findings may help in determining the underlying mechanisms by which glaucoma affects the risk of RVO.