Investigating the Shared Genetic Architecture of 3 Age-Related Ocular Disorders

Abstract

Purpose

To explore the shared genetic architecture, causal relationships, and cell type–specific expression patterns of pleiotropic genes in age-related macular degeneration (AMD), cataract, and primary open-angle glaucoma (POAG), uncovering molecular mechanisms and informing targeted therapies.

Design

A genetic association study combined with cross-trait meta-analyses, Mendelian randomization analyses, and single-cell RNA sequencing (scRNA-seq) analysis.

Subjects

The data related to 3 age-related ocular diseases, including AMD, cataract, and POAG, were obtained from publicly available genome-wide association study (GWAS) databases.

Methods

We conducted a comprehensive genetic analysis utilizing GWAS summary statistics to examine genetic correlations among AMD, cataract, and POAG. Cross-trait meta-analyses were performed to identify shared risk loci. Mendelian randomization was employed to investigate potential causal relationships between these conditions. Additionally, we analyzed scRNA-seq data to examine the expression patterns of identified pleiotropic genes across different retinal cell types.

Main Outcome Measures

Identification of shared risk single nucleotide polymorphisms (SNPs) and pleiotropic loci. Causal relationships between AMD, cataract, and POAG. Cell type–specific expression patterns of pleiotropic genes in retinal cells.

Results

Our analysis revealed significant genetic correlations, with a negative correlation between AMD and POAG and a positive correlation between cataract and POAG. Cross-trait meta-analyses identified shared risk SNPs, with CDKN2B-AS1 emerging as a notable pleiotropic locus. Mendelian randomization analyses suggested causal relationships between AMD and cataract, as well as between POAG and AMD. Single-cell expression analysis demonstrated cell type–specific expression patterns of pleiotropic genes including ATXN2, HTRA1, SIX6, and THSD7A across retinal cells.

Conclusions

This study provides compelling evidence for shared genetic architecture and causal relationships among AMD, cataract, and POAG. The identification of specific pleiotropic genes and their expression patterns across retinal cell types offers new insights into the molecular mechanisms underlying these age-related ocular diseases, potentially informing the development of targeted therapeutic strategies.

Financial Disclosure(s)

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Age-related ocular diseases, such as age-related macular degeneration (AMD), cataract, and glaucoma, are degenerative eye disorders and are among the leading causes of blindness and visual impairment in adults >50 years old worldwide.¹ Glaucoma is a progressive optic neuropathy characterized by the degeneration of retinal ganglion cells (RGCs) and their axons, with primary open-angle glaucoma (POAG) being the most prevalent form.² A study by FinnGen³ reported an inverse association between the allele ε4 of apolipoprotein E4 and both POAG and AMD, suggesting that these conditions may share certain pathophysiological pathways. Moreover, genetic variations within the CAV1/2 gene locus have been linked to the risk of developing POAG, whereas the high-risk HTRA1 variant associated with AMD is thought to modulate CAV1 expression, playing a role in its pathogenic effects.⁴ These findings further revealed the possible genetic connections between these eye diseases. A previous study reported that among elderly adults undergoing cataract surgery, the proportion with ocular comorbidities is as high as 42.5%, with AMD, glaucoma, and myopic degeneration being the most common. These comorbidities primarily limit postoperative visual improvement, even when the surgical procedure itself is anatomically successful, by impairing pre-existing retinal or optic nerve function. Several studies suggested that cataract surgery may increase the risk of POAG and AMD.^4,5 Besides, glaucoma interventions, whether medical or surgical, could potentially exacerbate the formation of cataracts. Patients who have undergone trabeculectomy are at an increased risk of developing and progressing cataracts. These findings support the need for further research on the association among AMD, cataract, and POAG, potentially resulting in more secure and impactful treatments for each disorder, whether in isolation or when they present concurrently.

Genome-wide association studies (GWASs) have discovered hundreds of susceptibility loci related to AMD, cataract, and POAG, and confirmed their polygenic property.^6,7 A study conducted GWAS on 11 275 individuals of African descent (6003 cases and 5272 controls) for POAG, identifying 2 previously unreported variants (rs1666698 mapped to DBF4P2 and rs34957764 mapped to ROCK1P1).⁸ Another study found that ABCA1 was significantly associated with both POAG and AMD at the genome-wide significance level. In addition, a study found extensive genetic pleiotropy and pairwise connections among AMD, cataract, and glaucoma, but the relationship between AMD and glaucoma was not consistent between genetic correlation analysis and epidemiological evidence. And the study did not identify a causal relationship among these diseases.⁹ Considering the inconsistent findings and limited understanding of these age-related eye diseases, there is a need for further research to elucidate the connections among these age-related ocular diseases to reveal common disease mechanisms and develop potential treatments.

In this study, we leverage large-scale GWAS summary data to investigate genetic correlations and potential causal relationships among AMD, cataract, and POAG. We conduct cross-trait GWAS meta-analyses across these ocular diseases to detect shared risk single nucleotide polymorphisms (SNPs) between any 2 ocular disorders. Additionally, we integrate GWAS summary data with tissue-specific gene expression data to assess whether SNP heritability for AMD, cataract, and POAG is enriched in shared tissues. Using Mendelian randomization (MR), we inferred the potential causal relationships among 3 ocular diseases. Finally, we explored the expression of pleiotropic genes across different cell types based on single-cell RNA sequencing (scRNA-seq) data sets. A flowchart of our analysis approach is presented in Fig 1.

Figure 1.

Study overview. AMD = age-related macular degeneration; CPASSOC = Cross Phenotype Association Test; GNOVA = Genetic Covariance Analyzer; LDSC = Linkage Disequilibrium Score Regression; LDSC-SEG = LDSC specifically expressed gene; MTAG = Multi-Trait Analysis of GWAS; scRNA-seq = single-cell RNA sequencing; SMR = Summary data–based MR; PLACO = pleiotropic analysis under composite null hypothesis; PheWAS = Phenome-Wide Association Study; POAG = primary open-angle glaucoma; TSMR = 2-sample Mendelian randomization.

Methods

GWAS Summary Statistics Collection

Detailed information of the GWAS summary statistics for AMD, cataract, and POAG is presented in Table S1 (available at www.ophthalmologyscience.org). Genome-wide association study summary statistics for AMD included 14 034 controls and 91 214 cases of European descent.¹⁰ Summary statistics for POAG comprised data from 216 257 participants of European descent.⁶ The summary statistics for AMD and POAG can be accessed from the GWAS Catalog (https://www.ebi.ac.uk/gwas/home). Summary statistics for cataract were obtained from the FinnGen database, encompassing 65 235 cases and 341 546 controls of European descent.¹¹ This study utilized publicly available summary-level GWAS data sets, which have received ethics board approval from their respective institutional review boards and obtained informed consent from participants.

Heritability and Genetic Correlation

Linkage Disequilibrium Score Regression (LDSC) is a statistical method widely used in genetic research to estimate the heritability of complex traits and diseases, as well as to assess genetic correlations between different traits. Linkage Disequilibrium Score Regression relies on the assumption that the GWAS effect size estimate for a given SNP reflects the combined effects of all SNPs in LD with it. Herein, we used precomputed LD scores from the 1000 Genomes Project, derived from SNPs in the HapMap 3 set. We then excluded SNPs that did not align with the reference panel, defined as having a minor allele frequency ≤0.01 or an imputation information score score ≤0.9.¹² Next, we reformatted the GWAS summary statistics and applied LDSC with the baseline-LD model (version 2.2) to estimate single-trait SNP heritability for 3 age-related ocular disorders. Finally, bivariate LDSC was used to evaluate the genetic correlations between 2 ocular disorders. The outputs from LDSC included SNP-based heritability estimates and genetic correlations. In addition, we applied Genetic Covariance Analyzer to estimate genetic covariance and correlation and complement the LDSC analysis.¹³ Similar to LDSC, Genetic Covariance Analyzer uses GWAS summary statistics as input but also incorporates an LD matrix and allows optional functional annotations. Genetic Covariance Analyzer corrects for sample overlap and estimates both genetic covariance and correlation, which are also reported in Table 1 and Table 2. These results were used to validate and support the findings obtained from LDSC.

Table 1.

Heritability for AMD, Cataract, and POAG Using LDSC and GNOVA

Trait	LDSC	GNOVA
AMD	0.023	0.033
Cataract	0.022	0.048
POAG	0.052	0.076

AMD = age-related macular degeneration; GNOVA = Genetic Covariance Analyzer; LDSC = Linkage Disequilibrium Score Regression; POAG = primary open-angle glaucoma.

Table 2.

Genetic Correlation Estimated for AMD, Cataract, and POAG Using LDSC and GNOVA

Trait1	Trait2	LDSC_rg	LDSC_rg_p	GNOVA_rg	GNOVA_rg_p
AMD	Cataract	0.038	7.053E-01	0.105	7.275E-02
AMD	POAG	–0.289	5.381E-04	–0.288	3.019E-09
Cataract	POAG	0.162	1.286E-03	0.101	4.764E-04

AMD = age-related macular degeneration; GNOVA = Genetic Covariance Analyzer; LDSC = Linkage Disequilibrium Score Regression; POAG = primary open-angle glaucoma.

Cross-Trait Meta-Analysis

To detect shared risk SNPs between any 2 ocular disorders (AMD, POAG, and cataract), we conducted 2 cross-trait meta-analyses: Multi-Trait Analysis of GWAS (MTAG) and Cross Phenotype Association Test (CPASSOC). Multi-Trait Analysis of GWAS enhances the statistical power of estimating the genotype–phenotype variance-covariance matrix. By generating cross-trait-specific estimates for each SNP, MTAG allows researchers to assess genetic associations across multiple traits simultaneously. This approach utilizes bivariate LD score regression to adjust for potential errors arising from sample overlap, ensuring more accurate results. Multi-Trait Analysis of GWAS is particularly suitable when all variants exhibit the same effect size across traits and yield trait-specific association statistics, thereby effectively integrating genetic information from various traits and improving the robustness of genetic association findings while accounting for shared underlying genetic architecture. Additionally, we used CPASSOC as a sensitivity analysis. Cross Phenotype Association Test is a statistical method that integrates association evidence from multiple traits to detect variants that influence ≥1 trait. This approach assumes cross-phenotype heterogeneity of effects and employs a weighted meta-analysis based on sample sizes from GWAS summary statistics. By estimating cross-phenotype statistical heterogeneity and corresponding P values, CPASSOC enables researchers to identify shared genetic variants across different traits, thereby enhancing the understanding of the genetic architecture underlying complex diseases. We prioritized independent SNPs with genome-wide significance (P < 5 × 10^–8) in both MTAG and CPASSOC analyses.

Pleiotropic Analysis under Composite Null Hypothesis

Pleiotropic analysis under composite null hypothesis (PLACO) is an innovative method designed to detect pleiotropic variants between 2 phenotypes while operating under a composite null hypothesis.¹⁴ The method analyzes 1 variant at a time, utilizing 2 sets of Z statistics, and divides the composite null hypothesis into 3 subnull scenarios: H₀₀ indicates no association with either disease, H₁₀ suggests an association with the first disease but not the second, and H₀₁ suggests association with the second disease but not the first. The alternative hypothesis, H₁₁, proposes that the variant is associated with both diseases, indicating potential pleiotropic effects. By providing a structured framework for assessing pleiotropic effects, PLACO enhances the understanding of the genetic interplay between complex traits. In this study, we integrated the results from MTAG, CPASSOC, and PLACO analyses. We classified SNPs as pleiotropic if they achieved genome-wide significance (P < 5 × 10^–8) in MTAG, CPASSOC, and PLACO, ensuring robust evidence across complementary statistical models.

Mendelian Randomization

Based on the findings of genetic correlation, we utilized the R package “TwoSampleMR” to explore potential causal relationships between any 2 ocular disorders (P < 0.05). We selected SNPs as instrumental variables related to the exposure, based on the following criteria: (1) achieving genome-wide significance (P < 5 × 10^–8); (2) performing LD clumping using the 1000 Genomes Project European reference panel with an R² threshold of 0.001 and a window of 10 000 kb; and (3) calculating F-statistics (F = beta²/se²) and selecting only those with F > 10. To harmonize the effects of SNPs on both exposure and outcome, we adjusted for non-palindromic SNPs and removed all palindromic SNPs. We utilized 4 MR statistical methods: inverse variance weighting (IVW), MR-Egger, weighted median, and weighted mode.^15,16 The IVW method was employed as the primary analysis, summarizing the estimates for each genetic variant and providing an accurate causal estimate, assuming all genetic variants are valid or that the overall pleiotropy is balanced to zero. The other 3 methods served as sensitivity analyses. Additionally, we applied the MR-Egger intercept test to assess the presence of pleiotropy and Cochran's Q statistic to evaluate heterogeneity. A significant deviation of the MR-Egger intercept from zero (P < 0.05) would indicate the presence of unbalanced pleiotropy, suggesting a potential violation of MR assumptions and possible bias in the causal estimates. Conversely, a nonsignificant intercept (P > 0.05) suggests no evidence of directional pleiotropy and supports the validity of the instrumental variables.

Summary Data–Based Mendelian Randomization

Summary data–based MR (SMR) is a method that integrates summary statistics from GWAS and expression Quantitative Trait Loci (eQTL) studies within the framework of MR. This approach utilizes eQTL data alongside trait GWAS data to detect pleiotropic relationships between gene expression and complex traits. In this study, we conducted SMR analysis using eQTL summary statistics from the genotype-tissue expression (GTEx) project (including whole blood and 13 brain tissues) and the eQTLGen project (whole blood), along with additional eQTL summary data related to brain tissue.¹⁷ We aimed to explore the putative functional genes associated with AMD, POAG, and cataract through their statistical associations with GWAS summary statistics. We also employed heterogeneity in dependent instruments test to assess the presence of heterogeneity. To avoid false-positive findings due to multiple testing, we selected significant SMR probes based on the Bonferroni-corrected SMR P value threshold (0.05 divided by the number of probes) with heterogeneity in dependent instruments test P values > 0.05 considered indicative of no heterogeneity. Pathogenic genes that appeared in both traits within the same tissue and overlapped with SNP-related genes were considered shared functional genes.

Tissue-Specific Enrichment Analysis

Linkage Disequilibrium Score Regression specifically expressed gene (LDSC-SEG) analysis is a method that identifies tissue and cell-type associations with diseases by integrating gene expression data and GWAS summary statistics.¹⁸ This approach employs stratified LD score regression to test whether the genetic contribution to disease is enriched in regions surrounding genes that are most specifically expressed in a given tissue. We utilized gene expression data from the GTEx project (13 brain tissues) and GWAS summary statistics on 3 ocular disorders (AMD, POAG, and cataract) to investigate whether SNP heritability shows significant tissue-trait associations using LDSC-SEG method. The 1000 Genomes Project Phase 3 of European ancestry served as the reference panel for calculating LD scores, with a P value < 0.05 indicating suggestive significance. To comprehensively explore potential tissues involved in the genetic architecture, multiple-testing correction across tissues was not applied. We acknowledge that this may increase the risk of false-positive findings; however, this approach enables identification of a broader range of candidate tissues for further investigation.

Single-Cell RNA Sequencing Analysis

Single-cell RNA sequencing data from human ocular tissues were derived from several previously published studies. Jin Li et al¹⁹ integrated transcriptomic data from retinal samples of 20 healthy individuals, totaling 51 samples. Joshua Sanes et al²⁰ integrated transcriptomic data from human ocular anterior segment including 26 samples of trabecular meshwork and 3 samples of lens. Rui Chen et al collected a single-cell atlas of the human retinal pigment epithelium (RPE) and the choroid with 50 samples. The integrated datasets are available at the CELLxGENE collection²¹ (https://cellxgene.cziscience.com/e/be41a86a-b606-4b1c-8055-32f334898775.cxg/, https://cellxgene.cziscience.com/e/9ff99bf8-2524-4ab5-ab6e-4bc218e4a449.cxg/, https://cellxgene.cziscience.com/e/5ed90bc0-12f3-4354-b314-0ce0dbb96458.cxg/, https://cellxgene.cziscience.com/e/489318a0-24c3-4f5c-b105-f084ed0ea026.cxg/), and further details can be found in the original publications. We conducted data analysis using the R package Seurat and utilized Uniform Manifold Approximation and Projection for visualization. Compared to t-Distributed Stochastic Neighbor Embedding, Uniform Manifold Approximation and Projection offers the advantage of preserving global data structure more effectively.

Phenome-Wide Association Study

Phenome-Wide Association Study (PheWAS) is a research approach used to explore the associations between genetic variants and a wide range of phenotypes or traits. Unlike traditional GWAS, which typically focus on a single trait, PheWAS allows researchers to examine how genetic variations affect multiple phenotypic outcomes simultaneously. To investigate the relationship between pleiotropic SNPs associated with 3 ocular disorders and retinal image features, we utilized data from the UK Biobank, which included 46 OCT image characteristics. These characteristics encompassed retinal thickness across various layers, such as the retinal nerve fiber layer, inner nuclear layer, and ganglion cell and inner plexiform layer, as well as the vertical cup-to-disc ratio.²² Phenome-Wide Association Study analysis was conducted to explore these associations. To account for multiple testing, we applied Bonferroni correction, with associations considered statistically significant at P < 0.00109 (0.05/46).

Results

Estimation of Genetic Correlation

We collected complete GWAS summary statistics for AMD, cataract, and POAG from public databases. Using baseline LD modeling, we estimated the SNP heritability for AMD, cataract, and POAG (Table 1). The LDSC SNP–based heritability estimates for AMD, cataract, and POAG were 0.023, 0.022, and 0.052, respectively. Additionally, we employed bivariate LDSC to estimate the genetic correlations between pairs of ocular disorders (Table 2). The genetic correlation between AMD and POAG was significantly negative (rg = –0.289, P = 5.381 × 10^–4). In contrast, the genetic correlation between cataract and POAG was significantly positive (rg = 0.162, P = 1.286 × 10^–3). However, the genetic correlation between AMD and cataract was not significant (rg = 0.038, P = 7.053 × 10^–1). To obtain robust results, we conducted Genetic Covariance Analyzer analysis, which, after correcting for sample overlap, yielded results consistent with those from LDSC. Specifically, the genetic correlation between AMD and POAG was significantly negative (rg = –0.288, P = 3.019 × 10^–9). In contrast, the genetic correlation between cataract and POAG was significantly positive (rg = 0.101, P = 4.764 × 10^–4) (Tables 1 and 2).

Identification of Shared-Risk SNPs

Given the genetic relationships among AMD, POAG, and cataract, we employed 2 complementary cross-trait meta-analyses (MTAG and CPASSOC) to identify shared risk SNPs between each pair of these ocular disorders. A total of 11 genome-wide significant SNPs (P < 5 × 10^–8) were identified in both MTAG and CPASSOC, which are as follows: rs33912345, rs62555370, rs6475604, rs6845653, rs7137828, rs3750847, rs4658046, rs763720, rs10757265, rs2304720, and rs2526100 (Table 3). Furthermore, PLACO analysis provided additional validation of the results. Among the identified SNPs, 5 (rs33912345, rs62555370, rs6475604, rs6845653, and rs7137828) were significantly associated with the joint phenotype of AMD-POAG (P < 5 × 10^–8). Additionally, 3 SNPs (rs3750847, rs4658046, and rs763720) were significantly associated with the joint phenotype of AMD-cataract (P < 5 × 10^–8). Furthermore, 3 SNPs (rs10757265, rs2304720, and rs2526100) were significantly associated with the joint phenotype of cataract-POAG (P < 5 × 10^–8). Notably, CDKN2B-AS1 (rs62555370, rs6475604, and rs10757265) has been implicated in the pathogenesis of POAG by multiple studies, suggesting its role in mediating aging, inflammation, and extracellular matrix (ECM) accumulation.²³ However, research on the relationship between CDKN2B-AS1 and AMD and cataract remains limited.

Table 3.

Pleiotropic SNPs Associated with 3 Ocular Disorders (MTAG, CPASSOC, and PLACO)

SNP	Trait 1	Trait 2	CHR	BP Position	Effect Allele	Noneffect Allele	Nearest Gene	MTAG Trait 1 P value	MTAG Trait 2 P value	CPASSOC P value	PLACO P value
rs33912345	AMD	POAG	14	60 976 537	A	C	SIX6,C14orf39	6.698E-11	1.958E-39	4.066E-42	7.679E-15
rs62555370	AMD	POAG	9	22 102 437	A	G	CDKN2B-AS1	7.306E-09	3.625E-14	2.667E-16	2.900E-12
rs6475604	AMD	POAG	9	22 052 734	T	C	CDKN2B-AS1	1.520E-20	4.157E-91	4.671E-100	1.555E-26
rs6845653	AMD	POAG	4	7 899 379	T	C	AFAP1	4.339E-10	7.701E-40	7.456E-43	4.098E-13
rs7137828	AMD	POAG	12	111 932 800	T	C	ATXN2	2.698E-08	6.093E-11	5.691E-13	6.001E-11
rs3750847	AMD	Cataract	10	124 215 421	T	C	ARMS2	9.313E-127	4.659E-19	1.786E-123	9.531E-76
rs4658046	AMD	Cataract	1	196 670 757	T	C	CFH	3.178E-121	4.512E-08	1.130E-121	8.190E-43
rs763720	AMD	Cataract	10	124 262 444	A	G	HTRA1	2.026E-22	2.557E-08	3.088E-20	6.456E-21
rs10757265	Cataract	POAG	9	22 048 860	C	T	CDKN2B-AS1	2.297E-09	6.003E-83	3.581E-78	4.457E-30
rs2304720	Cataract	POAG	15	73 946 393	C	T	LOXL1	1.375E-17	1.268E-10	1.493E-21	3.266E-20
rs2526100	Cataract	POAG	7	11 638 151	C	T	THSD7A	1.006E-12	1.844E-09	3.937E-16	4.226E-16

AMD = age-related macular degeneration; BP = base-pair; CHR = chromosome; POAG = primary open-angle glaucoma; SNP = single nucleotide polymorphism; MTAG = Multi-Trait Analysis of CPASSOC = Cross Phenotype Association Test; PLACO = pleiotropic analysis under composite null hypothesis.

Causal Inferences among 3 Ocular Disorders

After investigating the shared genetic background among the 3 ocular disorders, we further explored their potential causal relationships through MR analysis. The F-statistics for the instrumental variables were all greater than 10, thereby avoiding biases associated with weak instruments. We found evidence of a causal relationship between AMD and cataract (P_IVW = 1.883 × 10^–7; odds ratio [OR] = 1.145; 95% confidence interval [CI]: 1.088–1.204), as well as between POAG and AMD (P_IVW = 3.891 × 10^–8; OR = 0.876; 95% CI: 0.835–0.918). These findings were further supported by MR-Egger, weighted median, and weighted mode methods (Fig 2 and Table S2, available at www.ophthalmologyscience.org). However, we did not find evidence for a causal relationship between POAG and cataract. Additionally, all results indicated no evidence of horizontal pleiotropy (Tables S4 and S5, available at www.ophthalmologyscience.org). In addition, to obtain robust results, we conducted reverse MR analysis. Our findings did not indicate a reverse causal relationship between AMD and cataract nor between POAG and AMD (Table S3, available at www.ophthalmologyscience.org).

Figure 2.

Forest plots of MR analysis for 3 ocular disorders. The plots display the nSNP and the effect size estimates (OR with 95% confidence intervals). AMD = age-related macular degeneration; CI = confidence interval; MR = Mendelian randomization; nSNP = number of SNP; OR = odds ratio; POAG = primary open-angle glaucoma.

Identification of Shared Functional Genes

There is a close interrelationship between the eyes and the brain, as they share developmental origins from neural ectoderm and potentially overlapping molecular pathways.²⁴ The retina, often considered an extension of the CNS, contains neurons that process visual information before transmitting it to the brain via the optic nerve. This neuroanatomical connection is further evidenced by the presence of similar neurotransmitter systems and signaling molecules in both tissues. Recent research has also identified shared genetic risk factors between ocular disorders and various neurodegenerative conditions, suggesting common pathophysiological mechanisms.²⁵ Additionally, because eye tissue samples are typically scarce and difficult to obtain, we aimed to gain insights into the potential interactions among the 3 ocular disorders by studying genes associated with brain tissue. By integrating GWAS summary data for AMD, POAG, and cataract with eQTL summary data from the eQTLGen (whole blood), GTEx (whole blood and 13 brain tissues), and other brain tissue studies, we applied SMR to identify potential functional genes associated with AMD, POAG, and cataract. Under the tissue-specific Bonferroni correction threshold, we identified 5 genes associated with AMD, 15 genes associated with cataract, and 17 genes associated with POAG (Table S6, available at www.ophthalmologyscience.org). Although no shared functional genes met significance, LOXL1 emerged as a strong candidate for POAG. Notably, we observed a significant association between LOXL1 and POAG in brain tissue (P = 4.123 × 10^–6). This finding aligns with the results from the shared risk SNPs analysis. Studies have shown that mutations in LOXL1 gene are a risk factor for pseudoexfoliation syndrome (PXFS). Pseudoexfoliation syndrome is an age-related ocular and systemic disorder characterized by the production and accumulation of abnormal extracellular material in various tissues. It is the most identifiable cause of open-angle glaucoma. Additionally, PXFS has been linked to an increased risk of cataract formation.²⁶

Tissue-Level SNP Heritability Enrichment

To identify brain tissues associated with 3 ocular disorders, we conducted LDSC-SEG analysis by integrating gene expression data from 13 brain tissues in the GTEx project with GWAS data for AMD, cataract, and POAG. At a threshold P value of 0.05, AMD showed significant enrichment in the caudate, substantia nigra, and putamen. Cataract demonstrated significant enrichment in the nucleus accumbens, whereas POAG was significantly enriched in the putamen and caudate. The degree of SNP heritability enrichment across the 13 brain tissues is shown in Figure 3 and Table S7 (available at www.ophthalmologyscience.org). All 3 ocular disorders exhibited notable enrichment in the basal ganglia, suggesting potential involvement of motor regulation mechanisms associated with this brain region. Previous studies have linked basal ganglia calcification to increased risk of cataract.²⁷

Figure 3.

Tissue type–specific enrichment SNP heritability for 3 ocular disorders. A, AMD, (B) cataract, and (C) POAG. AMD = age-related macular degeneration; GTEx = genotype-tissue expression; POAG = primary open-angle glaucoma; SNP = single nucleotide polymorphism.

Single-Cell RNA Sequencing Reveals Pleiotropic Gene Expression in Different Ocular Tissues

To explore the cell type–specific context of disease-related genes, we then used scRNA-seq data sets of ocular tissues from the CELLxGENE portal. CELLxGENE provides high-resolution cellular context specific to the retina, choroid, lens, and trabecular meshwork, whereas GTEx data set was used for bulk tissue-level eQTL data from brain tissue.

We analyzed scRNA-seq data from 130 samples collected from 85 healthy individuals across 4 tissues including retina, choroid, lens, and trabecular meshwork. This analysis yielded a total of 527 794 cells and identified 46 distinct cell types, including microglial cells, amacrine cells, retinal cone cells, retinal rod cells, RGCs, rod bipolar cells, and other cell types (Fig 4, Fig S1, and Fig S2, available at www.ophthalmologyscience.org). We observed that ATXN2 gene, associated with both AMD and POAG; HTRA1 gene, associated with AMD and cataract; SIX6 gene, linked to AMD and POAG; and THSD7A gene, associated with cataract and POAG, exhibited varying expression levels across most cell types (Fig 5 and Figs S3–S5, available at www.ophthalmologyscience.org). In addition, ATXN2 showed differential expression across distinct tissue types, with the significant highest expression observed in the retinal pigment epithelial cell from retinal tissue (log2 fold-change = 0.28, P < 2.2 × 10^–16) and endothelial cell of venule from choroidal tissue (log2 fold-change = 0.41, P < 2.2 × 10^–16) (Table S8, available at www.ophthalmologyscience.org). HTRA1 gene displayed higher expression levels in H1 (log2 fold-change = 1.44, P < 2.2 × 10^–16) and H2 horizontal cells (log2 fold-change = 1.99, P < 2.2 × 10^–16) as well as in retinal pigment epithelial cells (log2 fold-change = 0.87, P < 2.2 × 10^–16) from retinal tissue, endothelial cell from trabecular meshwork tissue (log2 fold-change = 1.37, P < 2.2 × 10^–16), lens fiber cell from lens tissue (log2 fold-change = 1.31, P < 2.2 × 10^–16), and fibroblast from choroid tissue (log2 fold-change = 0.33, P < 2.2 × 10^–16) with significant difference (Table S8). SIX6 gene showed notably widespread expression in Müller cells (log2 fold-change = 1.03, P < 2.2 × 10^–16) of the retina (Table S8). THSD7A gene exhibited higher expression levels in retinal bipolar neurons (log2 fold-change = 1.79, P < 2.2 × 10^–16), blood vessel endothelial cell of trabecular meshwork (log2 fold-change = 1.92, P < 2.2 × 10^–16), endothelial cell of choroid (log2 fold-change = 1.68, P < 2.2 × 10^–16), and lens epithelial cell (log2 fold-change = 0.61, P < 2.2 × 10^–16, Table S8).

Figure 4.

Uniform Manifold Approximation and Projection plot of 265 767 single cells from 51 retinal samples and 182 151 single cells from 50 choroidal samples. Different colors represent different cell types. UMAP = Uniform Manifold Approximation and Projection.

Figure 5.

Expression of ATXN2, HTRA1, SIX6, and THSD7A across different retinal cell types. A, Uniform Manifold Approximation and Projection visualization of gene expression, with color intensity representing expression levels. Each point represents a single cell, and cells are grouped according to type. B, Heatmap showing the distribution of gene expression levels across 18 distinct retinal cell types, with each color representing a different cell type. UMAP = Uniform Manifold Approximation and Projection.

Phenome-Wide Association Study Analysis of Pleiotropic SNPs and Retinal Imaging Traits

To investigate the association between pleiotropic SNPs related to the 3 ocular disorders and retinal imaging traits, we conducted PheWAS analysis using data on 46 OCT imaging features from the UK Biobank (Table S9, available at www.ophthalmologyscience.org). Our analysis revealed that HTRA1 gene (rs763720), associated with both AMD and cataract, was linked to the thickness between the inner and outer segments of photoreceptor cells and the RPE (Fig 6). Additionally, CDKN2B-AS1 gene (rs62555370), associated with both AMD and POAG, was significantly associated with the vertical cup-to-disc ratio in the left eye, as well as the average thickness measured between inner nuclear layer and RPE in the central subfield of the right eye (Fig S5 and S6).

Figure 6.

The PheWAS analysis illustrates the association between HTRA1 gene (rs763720) and OCT imaging features. ISOS = inner segment and outer segment; PheWAS = Phenome-Wide Association Study; RPE = retinal pigment epithelium.

Discussion

This study investigated causality and the shared genetic architecture among cataract, POAG, and AMD in terms of global genetic correlation, tissue and cell specificity, and pleiotropy, leveraging multiple large GWAS summary statistics. Our results provide new insights into their comorbidity and might contribute to the prediction, diagnosis, and treatment of these diseases. We identified a positive genetic correlation between POAG and cataract suggesting potential shared biological mechanisms, as well as the negative genetic correlations between POAG and AMD. Moreover, MR analysis emphasizes a higher risk for AMD leads to an increased risk for cataract, but POAG is a protective factor against AMD. The association between POAG and AMD was consistent between genetic correlation analysis and MR analyses.

In cross-trait meta-analysis, we identified 5, 3, and 3 pleiotropic SNPs shared between AMD and POAG, AMD and cataract, and POAG and cataract, respectively. Among these pleiotropic SNPs, rs62555370, rs6475604, and rs10757265 are variants of CDKN2B-AS1, which resides in the 9p21 region and is associated with susceptibility to POAG. Previous studies have suggested that pathogenesis may involve RGC and trabecular meshwork cells senescence.^28,29 Another study suggests that transforming growth factor-beta (TGF-β) signaling may contribute to cellular senescence, inflammation, and ECM accumulation mediated by CDKN2B-AS1, affecting trabecular meshwork cell function and leading to increased intraocular pressure (IOP). Additionally, TGF-β signaling is associated with both AMD and cataract.³⁰ In cataract, TGF-β signaling induces lens opacity formation by promoting ECM accumulation.

Through SMR analysis with brain tissue, we observed a significant association between LOXL1 and POAG. Previous studies have indicated a significant relationship between LOXL1 and PXFS, a common cause of POAG, revealing that LOXL1 contributes to the synthesis of elastic fibers, which increases resistance to aqueous humor outflow and thereby leads to PXFS. LOXL1 is involved in the formation of the ECM, and its abnormalities can affect the structure and function of the RPE, thereby promoting the progression of AMD.³¹ When LOXL1 undergoes hypermethylation, it leads to reduced gene expression levels, which may affect the elasticity and structure of the lens suspensory ligament, destabilize the lens, and thereby increase the risk of cataract formation. The LOXL1 gene increases the risk of AMD, cataracts, and PXF by affecting the formation of the ECM and elastic fibers, and its variations are closely related to the progression and complexity of these diseases. To achieve complementary analytical objectives, our study integrated data from both GTEx and CELLxGENE. GTEx-based SMR and LDSC-SEG analyses allowed us to identify putative functional genes and heritability enrichment at the brain tissue level, which is relevant given the shared neuroectodermal origins of the eye and brain. However, brain-tissue-level eQTLs may not fully reflect gene activity in ocular tissues. Therefore, we complemented this with single-cell transcriptomic data from CELLxGENE, which provided cellular resolution to determine where pleiotropic genes are most active within the eye. This two-tiered strategy enabled both systemic and local interpretation of genetic findings.

Tissue specific enrichment analysis revealed that basal ganglia might play a pivotal role in the pathogenesis of cataract, POAG, and AMD. Previous studies have reported that basal ganglia lesions may be a risk factor for glaucomatous optic nerve damage, characterized by an increased vertical cup-to-disc ratio.^32,33 Additionally, basal ganglia calcification has been associated with an increased risk of cataract. These findings highlight the importance of the basal ganglia in visual degenerative diseases. Unfortunately, there is currently no evidence linking basal ganglia lesions to the occurrence of AMD. Further research is warranted to investigate the shared neuropathogenesis underlying cataract, POAG, and AMD.

The scRNA-seq analysis suggests a shared etiology among ocular disorders. The genes ATXN2, HTRA1, SIX6, and THSD7A are linked to ≥2 types of ocular diseases, and their differential expression in specific cells from retina, choroid, lens, and trabecular meshwork may indicate mechanisms underlying disease comorbidity. The ATXN2 gene is associated with AMD and POAG, particularly showing higher expression in retinal pigment epithelial cells. Previous study indicated that ATXN2, were associated with elevated IOP or POAG risk.³⁴ Mutations in the ATXN2 gene may lead to the aggregation of intracellular proteins and cellular stress responses, thereby causing damage and death of RGCs. Furthermore, these mutations may also affect intracellular calcium homeostasis and mechanosensitive, impairing the function of trabecular meshwork cells and leading to obstruction of aqueous humor outflow, which are pathological characteristics of POAG. The ATXN2 gene is associated with cellular stress responses, particularly oxidative stress. Lens cells are particularly susceptible to oxidative damage, leading to the occurrence of cataract.³⁵ An important pathological feature of AMD is oxidative stress, especially in RPE cells. The ATXN2 gene regulates the autophagy process in cells, which is crucial for the clearance of damaged proteins and organelles within the cell. If mutations occur in ATXN2 or its expression is dysregulated, the autophagy function is impaired, leading to the accumulation of abnormal proteins in the lens, which may promote the formation of cataracts. In AMD, the accumulation of undetoxified cellular waste and lipids in RPE cells is an important pathological change.³⁶ The ATXN2 gene may influence the onset of POAG, cataract, and AMD through aspects such as oxidative stress and autophagy regulation. The HTRA1 gene, associated with cataracts and AMD, exhibits higher expression in H1 and H2 horizontal cells. HTRA1 mutations activate the TGF-β/Smad signaling pathway, regulating the proliferation and migration of human lens epithelial cells, and thereby causing cataract. Research findings indicate that HTRA1 mutations may disrupt the integrity of the RPE-Bruch membrane, leading to ECM protein deposition or affecting TGF-β activity, which promotes the development of AMD.³⁷ Mutations in the SIX6 gene may disrupt early ocular cell growth and differentiation, potentially reducing the number of RGCs, leading to optic nerve degeneration, and thereby increasing susceptibility to POAG. The SIX6 gene is widely expressed in Müller cells and retinal midget ganglion cells (retinal support and neuronal cells), suggesting that SIX6 may be involved in the common pathological processes of AMD and POAG by influencing neuronal health and retinal structural stability. The THSD7A gene is highly expressed in retinal bipolar neurons, which may reflect a key role for THSD7A in retinal neural transmission and cell communication, thereby impacting visual function and the progression of ocular diseases.

However, only 4 pleiotropic genes were detected in single-cell analysis. It may be because of the testing efficiency. Similar situation appeared when analyzing genes in ocular tissues except retina. SIX6 gene showed extremely low proportion of expressed cells in choroid (Fig S3) and trabecular meshwork (Fig S5) and exactly no expression in lens (Fig S4), which might be another condition that some genes only express in certain cells. The pervasive lack of expression for numerous pleiotropic genes in adult ocular cell types underscores that their disease associations are likely mediated through indirect trans-acting mechanisms, such as altering developmental programming, functioning in rare transient cell states, or exerting effects noncell autonomously, rather than through direct action within the majority of differentiated ocular cells. Therefore, we can only confirm the potential effect of ATXN2, HTRA1, SIX6, and THSD7A genes in this study.

Previous studies have investigated causal relationships and pleiotropic mechanisms underlying several ocular diseases using similar methodologies. For example, Seo et al³⁸ demonstrated a bidirectional causal link between glaucoma and cataract using two-sample MR. Cuellar-Partida et al³⁹ assessed shared polygenic effects between AMD and POAG. Xue et al⁴⁰ identified common genetic risk loci and evaluated their dynamic expression in the retina using scRNA-seq. Of note, our study integrates MR, single-cell transcriptomics, and PheWAS across multiple ocular tissues to provide novel insights into the genetic architecture underlying age-related ocular disorders. Specifically, we incorporated expression data from nonretinal tissues such as the choroid, lens, and trabecular meshwork, which are highly relevant to glaucoma and cataract. This multitissue approach expands our understanding of disease-specific mechanisms and highlights tissue-selective gene expression patterns that may inform future therapeutic strategies.

There are several limitations in our study. Firstly, the GWAS summary statistics were limited to the European population, which could limit the generalizability of our results to other ethnic populations. Secondly, we evaluated tissue-specific heritability enrichments in the brain rather than the eye. Therefore, larger-scale data sets for eye tissue in the future may provide new insights into cataract, POAG, and AMD. We did not stratify expression patterns by donor age or sex because of the limited sample size. Thirdly, the summary GWAS data used for AMD analysis were exclusively derived from early AMD studies. This limitation may not fully capture the comprehensive genetic architecture of AMD. Future analyses incorporating more extensive AMD genetic data would likely yield a more complete understanding of shared genetic mechanisms among these ocular disorders. Lastly, although we have revealed the potential shared genetic architecture, further animal and clinical research is warranted to investigate the potential mechanisms underlying cataract, POAG, and AMD.

Data Availability

All data used in this study are publicly available data at the summary level, with citations to the relevant studies. GWAS summary statistics for available proteins were obtained from the MR-Base NHGRI-EBI GWAS Catalog (https://gwas.mrcieu.ac.uk/). The summary-level data download links were displayed in Table S1.

Manuscript no. XOPS-D-25-00190.