Mendelian randomization study of causality between 35 blood and urine biomarkers and age-related eye diseases

Abstract

Age-related eye diseases (AREDs) are a group of age-related visual degenerative diseases characterized by insidious and gradual onset. Identification of biomarkers associated with the early stage of AREDs is particularly important to delay its progression. This study aimed to evaluate the causal relationship between 35 blood and urine biomarkers and AREDs, thereby providing new insights into disease mechanisms and potential therapeutic targets. Two-sample Mendelian randomization (MR) was used to clarify the causal relationship between blood and urine biomarkers and AREDs. We classified AREDs into 4 types: age-related cataract (SC), age-related macular degeneration, glaucoma, and diabetic retinopathy. Inverse variance weighting method was used as the main analysis method, Cochran Q test, MR-Egger intercept, and leave-one-out test were used to investigate whether there was heterogeneity and pleiotropy of MR results. And the results were corrected by false discovery rate. Through MR analysis, this study provides the first systematic genetic evidence regarding the associations between blood and urine biomarkers and 4 AREDs. Specifically, our findings suggest a causal effect of 1 biomarker on SC, support causal roles of 3 biomarkers in age-related macular degeneration, and indicate causal relationships for 3 biomarkers with diabetic retinopathy. The reliability of these results was bolstered by a series of sensitivity analyses. This study strengthened the link between specific blood and urine biomarkers and the risk of AREDs. It has enhanced our understanding of the pathogenesis of AREDs and may guide the development of targeted prevention, diagnosis, and treatment strategies.

1. Introduction

According to the Global Burden of Disease study, hundreds of millions of people suffered moderate or severe vision impairment and tens of millions had vision loss in 2020. Moreover, with the acceleration of population aging, the disease burden will continue to rise in the coming decades, putting heavy pressure on society.^[1,2] The main drivers of avoidable blindness and low vision are age-related eye diseases (AREDs). Among them, senile cataract (SC), age-related macular degeneration (AMD), glaucoma (GLC), and diabetic retinopathy (DR) are common. DR is the most representative.^[3–6] Although surgical and drug treatments (such as cataract surgery,^[7] anti-vascular endothelial growth factor therapy,^[8] etc) have achieved remarkable results in some subtypes of AREDs, they mostly target patients with middle and late stage of obvious symptoms, and lack of early intervention methods, high treatment costs and high risk of complications, and it is difficult to fundamentally reduce the risk of disease and the burden of disease at the population level.^[9–11] Against the backdrop of global aging, it is urgent to build a prevention and control strategy centered on early screening, early diagnosis and early treatment.^[12]

Systematic collection, convenience, controllable cost and repeatable blood and urine biomarkers provide a key step for population prevention and control. Compared with imaging or functional indicators, blood and urine biomarkers can sensitively reflect multi-organ-multi-pathway systemic status, such as inflammation and oxidative stress, lipid and energy metabolism, renal microvascular injury and endothelial function, one-carbon metabolism and micronutrient homeostasis, etc. These pathways have complex interactions with the occurrence and development of AREDs.^[13,14] It is of great significance for the early screening of clinical ARED. Some studies have found that microalbumin, urine protein to creatinine ratio, creatinine are closely related to the early changes of DR. Regular detection of urine microalbumin, albumin-to-creatinine ratio and other indicators can assist early warning of the risk of retinal microangiopathy and guide early intervention.^[15] Previous studies have suggested that some circulating markers are associated with the risk or progression of AMD, DR, GLC, and cataract, such as inflammatory proteins, lipid components, amino acid metabolites and renal function indicators, etc.^[16–18] However, most of these evidences come from cross-sectional or prospective observational studies, which are susceptible to confounding and reverse causality, and it is difficult to support causal inference.^[19] More importantly, there is a lack of comprehensive comparison and generalization of multiple blood and urine biomarkers and a variety of AREDs under the same methodological framework, resulting in unclear common and specific etiological axes among different diseases.

At the methodological level, Mendelian randomization (MR) provides a causal inference tool for observational epidemiology that is close to randomized controlled trials. It makes use of germline genetic variants that are strongly associated with an exposure (e.g., a blood or urine biomarker) as instrumental variables (IVs), which can largely resist the interference of traditional confounding and reverse causality under the premise of meeting the assumptions of correlation, independence, and exclusion restrictions, Figure 1 illustrates the comprehensive design of our study.^[20] With the increase in sample size for genome-wide association studies (GWAS) and the maturity of IV construction strategies, two-sample MR has been widely used to analyze the underlying causal pathways of complex diseases.^[21]

Figure 1.

Study design overview.

Based on this, we need to fill the gap in the existing evidence of the association of blood and urine biomarkers with AREDs. The relative causal contributions of multiple systemic pathways to different AREDs were compared in a unified design, and an integrated theoretical framework was established to analyze the “shared mechanism-specific pathways” of etiology. At the public health and clinical levels, through the ranking and clustering of the causal effects of candidate markers, priority indicators with high testability, generalization, and intervention plasticity are screened out, which can provide a basis for the optimization of early screening strategies, lifestyle and nutrition interventions, and the priority of drug research and development. In the real dilemma of rapid aging of population and limited medical resources, cause-oriented risk management based on blood and urine biomarkers is of great significance for AREDs prevention and control system.

2. Data sources

Data on 35 blood and urine biomarkers were obtained from the GWAS study of the UK Biobank (https://www.ebi.ac.uk/gwas/), which included 363,228 participants of European ancestry. Specifically, the GWAS IDs for the 35 blood and urine biomarkers studied ranged from GCST90019492 to GCST90019526.^[22] We obtained the R11 dataset for senile cataract, AMD and GLC from the FinnGen Consortium, and the R12 dataset for DR. The UK Biobank integrates comprehensive health records and genetic data of over 500,000 volunteers to form a large-scale research database; meanwhile, the FinnGen project led by Finland innovatively links national genomic information with the national health registration system, providing a high-purity sample pool for gene–phenotype association studies in specific populations.

2.1. Instrumental variable selection

When screening IV, strict inclusion and exclusion criteria were implemented, only single nucleotide polymorphisms (SNPs) closely related to the exposure phenotype were included (P < 5e‐8), and IV was further filtered through linkage disequilibrium (kb = 10,000, r² = 0.001). The F value of IV was calculated using the formula (F = R²* (N ‐ 2)/1 ‐ R), and all weak IVs with F < 10 were excluded (detailed in File S1, Supplemental Digital Content, https://links.lww.com/MD/R246).

2.2. Statistical analysis

The study followed the STROBE-MR guidelines for two-sample MR analysis using the “TwoSampleMR (0.5.7)” software package in R software package (version 4.3.1; R Foundation for Statistical Computing, Vienna, Austria).^[23] The main methods used to assess causality are inverse variance weighting (IVW), MR-Egger, weighted median,^[24] debiased inverse-variance weighted method, and constrained maximum likelihood. Constrained maximum likelihood is used to eliminate biases caused by both relevant and irrelevant pleiotropy, and this method is considered much more powerful than MR-Egger.^[25] The debiased inverse-variance weighted method eliminates the weak instrumental bias of the IVW method and exhibits stronger robustness under many weak instruments.^[26] To ensure the robustness and validity of the findings, we performed MR sensitivity analysis to assess heterogeneity using Cochrane Q test.^[27] When significant heterogeneity was found, we used the random-effects IVW method to mitigate the bias introduced by heterogeneity IV (P < .05 indicated the presence of heterogeneity). The intercept of IVs in MR-Egger regression was used to evaluate horizontal multidimensionality, and MR multivalence residual and outlier method was used to detect and exclude SNPs that may have horizontal multidimensionality (P < .05 indicates the presence of horizontal multidimensionality).^[28] Finally, we performed an “left out one” analysis to test for the presence of bias effects by stepwise elimination of individual SNPs. Based on previous studies, we used the Steiger test^[29] to verify the direction of causal inference for each SNP, ensuring that the SNP did not have a direct effect on the results, but only influenced the results through exposure. To maximize mitigation of multi-effect bias, we not only conducted MR multivalence residual and outlier method global tests but also verified all identified IVs against the PhenoScanner V2 database. This excluded SNPs associated with known confounders, thereby ensuring the specificity of the IVs. The statistical power of each MR analysis was calculated according to the method established by Brion et al (https://shiny.cnsgenomics.com/mRnd/).^[30] Furthermore, to reduce the probability of type I errors caused by multiple comparisons, this study employed false discovery rate (FDR) correction for the results. Statistical significance was determined when the FDR q-value was <.05.

3. Result

3.1. A causal relationship between blood and urine biomarkers and age-related eye diseases

We employed the MR analysis method to investigate causal relationships between blood and urine biomarkers and AREDs. Results indicate that following analysis using the IVW method and FDR correction, 7 blood and urine biomarkers were significantly associated with AREDs (at a significance level of P < .05). Among these, glycated hemoglobin (HbA1c) (OR = 1.048, 95% CI = 1.019–1.077, P = .001) was associated with an elevated risk of SC. Three biomarkers were linked to AMD, including high-density lipoprotein (HDL) (OR = 1.187, 95% CI = 1.099–1.283, P < .00001) and apolipoprotein A (ApoA) (OR = 1.145, 95% CI = 1.045–1.256, P = .003) were associated with increased risk of AMD, whilst triglycerides (TRIGs) (OR = 0.868, 95% CI = 0.796–0.947, P = .001) was associated with a reduced risk of AMD. Three biomarkers were associated with DR, specifically: alanine aminotransferase (ALT) (OR = 1.274, 95% CI = 1.099–1.478, P = .002), neutrophil alkaline phosphatase (OR = 1.336, 95% CI = 1.149–1.553, P < .0001) were associated with increased DR risk, while blood urea nitrogen (OR = 0.788, 95% CI = 0.684–0.908, P < .0001) was associated with reduced DR risk, as detailed in Figure 2. post hoc power calculations support the robust capability of our MR analysis (>85%) to detect causal relationships, as detailed in File S2, Supplemental Digital Content, https://links.lww.com/MD/R246.

Figure 2.

Mendelian randomization analysis reveals 35 blood and urine biomarkers with causal associations to age-related eye disease risk.

3.2. Sensitivity analysis results

To further verify causality, sensitivity analyses were performed to assess the pleiotropy and heterogeneity of the MR results. No significant heterogeneity was found according to the Cochran Q test, and no significant level pleiotropy was identified according to the MR-Egger intercept test (Table 1). To ensure the reliability of the results of the MR analyses, we performed “leave-one-out” analyses to show that no single SNP was driving the identified causal relationship (see File S3, Supplemental Digital Content, https://links.lww.com/MD/R247).

Table 1.

The results of sensitivity analysis.

Expose	Outcome	Heterogeneity test P value		MR-Egger pleiotropy test
		MR-Egger	IVW	Intercept	P value
HBA1C	Senile cataract	2.73e‐08*	3.34e‐08*	0.0003	.763
HDL	Age-related macular degeneration	9.27e‐05*	0.0001*	‐0.0008	.714
TRIG		3.69e‐06*	4.54e-06*	‐0.0004	.861
ApoA		5.11e‐09*	3.03e-09*	‐0.0041	.132
ALT	Diabetic retinopathy	0.482	0.503	‐0.0009	.838
NAP		0.044*	0.048*	‐0.0011	.778
BUN		0.467	0.488	‐0.0016	.812

ApoA = apolipoprotein A, BUN = blood urea nitrogen, HbA1c = glycated haemoglobin, HDL = high-density lipoprotein, IVW = inverse variance weighting, NAP = neutrophil alkaline phosphatase, TRIGs = triglycerides.

^*P < .05 was considered statistically significant.

4. Discussion

Age-related eye diseases, such as macular degeneration, cataract and GLC, are important causes of vision loss in the elderly.^[31] Therefore, it is important to develop new strategies for the treatment of age-related eye diseases. This study used MR analysis to systematically explore the causal relationship between 35 common blood and urine biomarkers and 4 age-related eye diseases, which provided a direction for the prevention and treatment of age-related eye diseases.

The elevated level of HbA1c is positively correlated with the risk of senile cataract, which is mainly due to the cumulative damage of the lens and the metabolic disorders caused by hyperglycemia.^[32] There is a significant positive correlation between aqueous glucose level and blood glucose level in diabetic patients, and the increase of HbA1c level is closely related to the increase of aqueous glucose/blood glucose ratio, which reflects the increase of glucose permeability in the anterior chamber.^[33] This metabolic abnormality in diabetic patients may accelerate cataract formation. One study showed that the incidence of cataract increased with age, HbA1c level, and duration of diabetes in diabetic patients.^[34] This suggests that a prolonged hyperglycemic state may cause chronic damage to ocular tissues, thereby increasing the risk of cataract. The positive correlation between HbA1c and age-related cataract highlights the key role of glycemic management in disease prevention.

We found that elevated levels of HDL and ApoA increased the risk of AMD, While TRIGs were associated with decreased risk of AMD, our results were consistent with a previous study.^[35] Studies from EYE-RISK and the European Consortium for Ocular Epidemiology have shown that elevated HDL levels are significantly associated with increased risk of AMD. Specifically, AMD risk increased by 21% for each 1 mmol/L increase in HDL level.^[36] On the contrary, elevated TRIG levels are associated with a reduced risk of AMD. In another study, A significant causal association between ApoA and AMD was found by MR analysis, indicating that elevated ApoA levels may increase the risk of AMD,^[37] a result consistent with our findings on ApoA and AMD risk. The positive association between HDL/ApoA and AMD risk appears to contradict the conventional view of HDL’s antiatherogenic role. However, this paradox may be explained by shifting the focus from the quantity to the functional quality of HDL particles. In the context of chronic inflammation and oxidative stress characteristic of the aging retina, HDL can become dysfunctional. Such dysfunctional HDL may lose its beneficial properties and even acquire pro-inflammatory and pro-angiogenic capacities, potentially contributing to AMD pathogenesis. Therefore, our genetic findings, which reflect lifelong elevations in HDL-C/ApoA-I levels, might be capturing the effect of an increased pool of HDL that is more prone to becoming dysfunctional in the ocular environment, thereby paradoxically increasing risk. These results highlight the role of lipid metabolism in the pathogenesis of AMD and suggest that modulation of HDL metabolism may be a novel strategy to delay or prevent AMD. These results suggest that comprehensive regulation of lipid profile may delay the progression of AMD, and those at high risk should be tested regularly.

Our study did not identify any blood and urine markers with a significant causal association with GLC. But it is generally accepted that diabetes is an independent risk factor for GLC and is associated with neurodegeneration caused by central insulin resistance.^[38] A meta-analysis of 2,981,342 subjects from 16 countries showed that the combined relative risk of GLC in diabetic patients compared with normal controls was 1.48.^[39] Another cross-sectional study involving 3229 cases showed that diabetes was closely related to the prevalence of GLC.^[40,41] Elevated ALT levels reflect hepatocyte damage, and its positive correlation with the risk of DR may be related to metabolic syndrome. Several studies have shown that elevated ALT levels are closely related to various components of metabolic syndrome, such as obesity, insulin resistance, and abnormal lipid metabolism.^[42,43] The elevated level of neutrophil alkaline phosphatase indicates the activation of neutrophils, and the reactive oxygen species and inflammatory factors (such as IL-6) released by neutrophils destroy the blood-retinal barrier, leading to retinal edema and microvascular leakage, and increasing the risk of DR.^[44] Although our results show that blood urea nitrogen is associated with a reduced risk of DR, this result needs to be viewed with caution as it does not match the actual situation.

Our study has several limitations. First, as all GWAS data were derived from individuals of European ancestry, the generalizability of our findings to other populations is uncertain. This constraint is not merely a matter of sample demographics but is underpinned by fundamental genetic differences. Specifically, the allele frequencies of the genetic instruments we used for the biomarkers may differ in populations of non-European ancestry, such as Asian groups, potentially rendering some instruments invalid. Furthermore, differences in linkage disequilibrium patterns between populations could mean that our genetic variants tag the underlying causal exposures differently, leading to biased causal estimates when transferred to a different genetic context. These genetic considerations are particularly relevant given the well-documented differences in the prevalence and risk factor profiles for AREDs between European and Asian populations. Therefore, validating our findings requires future studies with well-powered GWAS and MR analyses in diverse ancestral cohorts.

5. Conclusion

This study establishes causal relationships between specific blood and urine markers and subtypes of age-related eye disease. These findings significantly advance our understanding of metabolic aspects of disease pathogenesis and identify potential biomarkers for risk stratification, providing promising directions for future research and clinical applications.

Acknowledgments

We sincerely express our profound gratitude to all the researchers and participants involved in the GWAS and biobank studies. It is their contributions of summary statistical data that have made this research possible.

Author contributions

Conceptualization: Jiaqi Chen.

Data curation: Cong Zhao.

Funding acquisition: Hongji Liu.

Investigation: Jiaqi Chen, Tongtong Chen.

Methodology: Zheyu Bao.

Resources: Hongji Liu.

Supervision: Hongji Liu.

Writing – original draft: Jiaqi Chen.

Writing – review & editing: Jiaqi Chen, Tongtong Chen, Cong Zhao, Zheyu Bao.