Abstract
Objective
The association between myopia and diabetic retinopathy (DR) is unclear, with inconsistent results reported, and whether the association represents causality remains unknown. This study aimed to investigate the causal associations of genetically determined myopia with DR, and further explore specific mechanisms.
Methods
We conducted two-sample mendelian randomization (MR) analyses of any myopia and high myopia on six DR phenotypes, including any DR, background DR, severe background DR, proliferative DR (PDR), diabetic maculopathy and unspecific DR in the primary study. Mechanism exploration of spherical equivalent refraction (SER), corneal curvature (CC) and axial length (AL) on any DR was carried out subsequently. Single-nucleotide polymorphisms (SNPs), used as genetic instruments, were derived from UK Biobank, Genetic Epidemiology Research on Adult Health and Aging cohort (GERA) and FinnGen. The inverse variance weighted (IVW) method was mainly used to assess the causality, and was complemented with sensitivity analyses and causality direction analyses.
Results
Using SNPs that have excluded possible confounders, we discovered suggestive and positive causal associations of any myopia with any DR (IVW: odds ratio [OR] = 1.133, 95% confidence interval [95%CI]: 1.070–1.201, P = 1.91×10 −5 ) and PDR (IVW: OR = 1.182, 95%CI: 1.088–1.285, P = 8.31×10 −5 ). Similar but more significant associations were found of high myopia with any DR and PDR (IVW: OR = 1.107, 95%CI: 1.051–1.166, P = 1.39×10 −4 ; OR = 1.163, 95%CI: 1.088–1.244, P = 8.76×10 −6 , respectively). Further mechanism analyses found only AL, rather than SER or CC, was strongly and significantly associated with any DR. These associations were robust in sensitivity analyses and causality direction analyses.
Conclusions
We found significant and positive causal associations of any myopia and high myopia with the risk of DR and PDR, which might be related with AL, indicating the significance of myopia control for preventing DR development and progression.
1
Introduction
Diabetic retinopathy (DR) is one of the primary causes of visual impairment and blindness worldwide, with the affected individuals projected to rise to 161 million by 2045. This has brought a huge burden to the society and healthcare system. When effective treatment is not timely, DR may progress to irreversible central and peripheral vision loss, emphasizing the importance of identifying its risk factors for early diagnosis and management.
Epidemiological studies have reported a variety of important factors that influence the development of DR, such as duration of diabetes, glycemic control, body mass index (BMI), hypertension, etc. An association between myopia and DR has also been mentioned. Recent studies suggested that myopia and high myopia were protective factors against different stages of DR. , Moreover, most studies believed only the axial component of myopia played an important role in DR correlation. , However, the results were inconsistent and conflicting. The association of DR with refractive myopia has also been reported. Additionally, the cohort study from Man et al. indicated no association of myopia on DR. This discrepancy among previous studies might be due to biases or confounders that couldn’t be completely rule out in observational epidemiological studies, such as small sample size, heterogeneity in demographic characteristics, selection bias and reverse causality.
Mendelian randomization (MR) analysis has become a popular and practical method for causal inference. It utilizes genetic variations as instrumental variables (IVs) and takes advantage of the inherent random segregation of alleles, allowing genetic associations to be independent of confounding factors and reverse causation. Moreover, MR studies can largely imitate randomized clinical trials, and the findings are generally consistent. Since both myopia and DR are heritable, , MR is an ideal approach which can overcome some of the limitations of observational studies and establish causal link. However, there have been no MR studies evaluating the observational correlations between myopia and DR.
In this study, we first conducted a primary study on myopia and DR to see whether a causal relationship exists. A two-sample MR approach of any myopia and high myopia on the risk of six DR phenotypes, including any DR, background diabetic retinopathy (BGDR), severe background diabetic retinopathy (SBGDR), proliferative diabetic retinopathy (PDR), diabetic maculopathy (DMP) and unspecific diabetic retinopathy (UDR) was employed. To further understand whether axial, or refractive myopia, or both, was the main contributor to this causal link, we subsequently carried out mechanism exploration of spherical equivalent refraction (SER), corneal curvature (CC) and axial length (AL) on any DR using MR analysis. This might give us a better understanding of the relationship between myopia and DR, deepen our awareness of DR pathophysiology, and have public health and clinical implications for prevention and early detection of this common cause of visual disability.
2
Methods
2.1
Study design
This study is a univariable two-sample MR analysis. Using summary statistics from genome-wide association studies (GWASs), we investigated the causal associations of any myopia and high myopia with six DR phenotypes (any DR, BGDR, SBGDR, PDR, DMP and UDR) in the primary study, and the causality of SER, CC and AL with any DR in the mechanism exploration. A schematic diagram outlining the process is presented in Fig. 1 . Furthermore, we have adhered to the MR-STROBE guidelines in reporting our findings.
2.2
GWAS data source
Brief information of the exposures and outcomes are displayed in Table 1 .
| Trait | Variable type | Sample size (case/control) | Population Ethnicity | Consortium | Study b /GWAS ID c |
|---|---|---|---|---|---|
| Any myopia | Exposure | 27993/36275 | European | UK Biobank | PMID: 35841873 |
| High myopia | Exposure | 3164/21416 | European | UK Biobank | PMID: 33830181 |
| SER | Exposure | 102117 a | European | UK Biobank | PMID: 32231278 |
| Corneal curvature | Exposure | 88218 a | European | UK Biobank | PMID: 32193507 |
| Axial length | Exposure | 16523 a | European | GERA | PMID: 37351342 |
| Any DR | Outcome | 12242/289034 | European | Finngen | finngen_R7_DM_RETINOPATHY |
| BGDR | Outcome | 3098/296912 | European | Finngen | finngen_R7_DM_BCKGRND_RETINA |
| SBGDR | Outcome | 672/296912 | European | Finngen | finngen_R7_DM_BCKGRND_RETINA_NONPROLIF |
| PDR | Outcome | 7349/296912 | European | Finngen | finngen_R7_DM_RETINA_PROLIF |
| DMP | Outcome | 2790/296454 | European | Finngen | finngen_R7_DM_MACULOPATHY |
| UDR | Outcome | 2719/296912 | European | Finngen | finngen_R7_DM_RETINA_NOS |
a SER, corneal curvature and axial length are continuous variables, contain only the total sample size.
b GWAS summary datasets of exposures are from Pubmed ( https://pubmed.ncbi.nlm.nih.gov/ ), do not have GWAS ID.
c GWAS summary datasets of outcomes are from the Finngen database ( https://r7.finngen.fi/ ), possess GWAS ID.
Summary statistics of any myopia, high myopia, SER and CC were all obtained from lately published GWASs using data from the UK Biobank (UKB, https://www.nealelab.is/uk-biobank ). The UKB is a large, prospective and population-based cohort of 502,413 participants aged 40–69 years recruited from 2006 to 2010. Details of the study design and protocols could be found elsewhere. In UKB, non-cycloplegic autorefraction and keratometry was measured directly using the Tomey RC 5000 Auto-Refractor Keratometer (Tomey Corporation, Aichi, Japan). The spherical equivalent was estimated as the sphere power (UKB codes 5084 and 5085) plus half the cylinder power (UKB codes 5086 and 5087) for each eye, with mean spherical equivalent (MSE) averaged between fellow eyes. The keratometry was reported as the maximum (UKB codes 5132 and 5135) and minimum (UKB codes 5096 and 5099) corneal power in each eye. For any myopia, participants with MSE ≤ −0.50D were identified as cases, while participants with MSE > −0.50D and didn’t have any ocular disease were controls. GWAS for high myopia (MSE: −6.00D or less) comprised 3164 cases and 21416 emmetropia controls (MSE: 0.00 to +1.00D). As for SER (continuous variable, per 1D decrease as unit), MSE was used as the outcome of the GWAS analysis. Another continuous variable, CC (per 1 mm decrease as unit), was converted by equation (337.5)/corneal power, and the average CC of the two eyes was taken as the phenotype.
We obtained the AL data of 16523 European participants from a large multiethnic GWAS consisting of 19420 individuals from the Genetic Epidemiology Research on Adult Health and Aging (GERA) cohort. The GERA cohort, containing 110266 adults, is established by Kaiser Permanente Medical Care Plan, Northern California Region (KPNC), an integrated health care delivery system with >3 million members ( https://researchbank.kaiserpermanente.org/ ). Haag-Streit Lenstar 900 was applied to measure AL, and the mean AL (per 1 mm increase as unit) of an individuals’ two eyes was used for the analysis.
The GWAS summary statistics for DR were sourced from the FinnGen ( https://r7.finngen.fi/ ), a large-scale research project including genome and health data from 500000 Finnish biobank participants. DR was identified using the International Classification of Diseases-Revision 10 (ICD-10) criteria from the hospital discharge registry. Participants in any DR (ICD-10: H36.0) analysis included 12242 cases and 289034 controls. Based on different levels of DR severity, four DR phenotypes were selected: BGDR (ICD-10: H36.00), DMP (ICD-10: H36.01), SBGDR (ICD-10: H36.02) and PDR (ICD-10: H36.03). Additionally, we also used UDR (ICD-10: H36.09) as our outcome. As for the controls, people without DR including healthy and individuals with diabetes were enrolled. Further details have been described elsewhere ( https://finngen.gitbook.io/documentation/v/r7/ ).
The data and information we used in this article were all searched and downloaded from the public database. No ethical review was required for this study.
2.3
Selection of SNPs
Single-nucleotide polymorphisms (SNPs) were selected to represent instrumental variables (IVs) in our study. To attain unbiased causal effects, MR must be in accordance with three core assumptions ( Fig. 1 ): (1) SNPs are strongly associated with the exposure; (2) SNPs must influence the outcome only through the exposure of interest; (3) SNPs are not related to potential confounders.
SNPs must meet the following criteria to fulfil the three basic MR assumptions: (1) SNPs for exposure must satisfy P < 5×10 −8 ; (2) F > 10 to establish a strong association between SNPs and exposure, and avoid weak instrument bias ; (3) Linkage disequilibrium r 2 < 0.001 and linkage disequilibrium distance >10000 kb to ensure independence between SNPs ; (4) SNPs of P < 1×10 −5 with the outcome must be removed to exclude association with outcome ; (5) Exclude SNPs associated with potential confounders ( P < 1×10 −5 ) , using the GWAS Catalog database ( https://www.ebi.ac.uk/gwas/ ) and the IEU Open GWAS Project database ( https://gwas.mrcieu.ac.uk/ ). Only GWASs restricted to European ethnicity were considered. , We have included BMI, , blood glucose, , blood pressure, diabetes, intraocular pressure and smoking status as potential confounders, since these factors have been identified to be related with myopia, or DR, or both.
2.4
MR analysis
All analyses were performed using the TwoSampleMR package (version 0.5.8) and MendelianRandomization package (version 0.9.0) in R (R Foundation for Statistical Computing, Vienna, Austria, version 4.3.2). The packages harmonize exposure and outcome datasets, and make causal inference, sensitivity analysis and directional analysis using GWAS summary statistics.
We used inverse variance weighted (IVW), the most efficient (greatest statistical power) method, as our primary result. This method rules out the presence of intercept and uses inverse variance of the outcome effect for weighted regression. Then, we presented the weighted median, a method that provides valid estimates when at least 50% of information is derived from valid SNPs. Moreover, we have also applied three additional methods including MR Egger, simple mode and weighted mode. However, these methods have less statistical power than IVW, leading to very wide confidence interval (CI), and hence we focused more on the consistency of the estimate direction. Ultimately, the causal estimates were expressed as odds ratio (OR) along with corresponding 95%CI. On the premise that all five MR methods had effects in the same direction (all OR > 1 / < 1), MR results were considered significant for strong evidence if the IVW and weighted median methods satisfied a two-sided Bonferroni-corrected P < 4.17×10 −3 [0.05/(2×6)] for primary study or P < 1.67×10 −2 [0.05/(3×1)] for mechanism exploration, and considered suggestive when 4.17×10 −3 ≤ P < 0.05 for primary study or 1.67×10 −2 ≤ P < 0.05 for mechanism exploration.
Sensitivity analysis is crucial for ensuring the robustness of association results in MR research. We employed Cochran’s Q test and I 2 calculation to evaluate potential heterogeneity among SNPs. Cochran Q-derived P > 0.05 in IVW method and I 2 < 25% indicates no heterogeneity. The relative symmetry around the vertical line corresponding to IVW method of funnel plots was also used to visualize the absence of heterogeneity. Horizontal pleiotropy was detected utilizing Egger intercept calculation by examining whether the intercept significantly deviated from zero. If the intercept shows no significant difference from zero ( P > 0.05), there’s no pleiotropy. Moreover, Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) test was applied to evaluate horizontal pleiotropy (Global Test P > 0.05 indicates no pleiotropy), detect outliers and re-assess causality after excluding the outliers. To determine whether a single SNP dominated the causal association, we also conducted a leave-one-out analysis in which IVW analysis was repeated with the omission of each exposure-related SNP in turn. Two additional approaches for assessing robustness included the use of income, education and intelligence level as negative control outcomes, and the swap of exposed and unexposed populations.
Finally, we applied MR-Steiger methods (Steiger test and Steiger filtering) to explore whether a reverse causal relationship existed, and whether the association of SNP with outcome was stronger than that with the exposure.
3
Results
3.1
SNPs for myopia and DR
Our strict SNPs screening process was based on the five standards mentioned above to meet the three major assumptions. Beyond that, some SNPs were removed due to no corresponding data in outcome GWAS datasets, incompatible alleles with outcome GWAS datasets and containing palindromic sequences. After clumping, 7 SNPs for high myopia and 26 SNPs for any myopia were chosen for final MR analysis on DR (any DR, BGDR, SBGDR, PDR, DMP and UDR) in the primary study. As for the mechanism exploration, 11 SNPs for AL, 20 SNPs for CC and 84 SNPs for SER were used for causal inference with any DR. Detailed information of the SNPs is provided in Tables S1–S5 .
3.2
Causal effects of myopia on DR in the primary study
The associations of myopia with DR are demonstrated in Supplementary Fig. 1 . Only suggestive evidence indicated that the genetically predicted incidence of any myopia was significantly and positively associated with any DR and PDR, since IVW alone satisfied P < 4.17×10 −3 (OR = 1.133, 95%CI: 1.070–1.201, P = 1.91×10 −5 ; OR = 1.182, 95%CI: 1.088–1.285, P = 8.31×10 −5 , respectively) while weighted median satisfied 4.17×10 −3 ≤ P < 0.05 (OR = 1.100, 95%CI: 1.010–1.198, P = 0.028; OR = 1.121, 95%CI: 1.003–1.253, P = 0.043, respectively). The uniform direction of ORs (all >1) using MR Egger, simple mode and weighted median further confirmed its adverse role against any DR and PDR.
We have found strong evidence for causal effects of high myopia on both any DR risk and PDR risk, as the existence of high myopia was more likely to result in higher risk of any DR (IVW: OR = 1.107, 95%CI: 1.051–1.166, P = 1.39×10 −4 ; weighted median: OR = 1.122, 95%CI: 1.049–1.201, P = 7.88×10 −4 ) and PDR (IVW: OR = 1.163, 95%CI: 1.088–1.244, P = 8.76×10 −6 ; weighted median: OR = 1.164, 95%CI: 1.067–1.270, P = 6.56×10 −4 ), respectively. The other three MR methods all revealed OR > 1, which further supported this conclusion. No other associations were found between myopia and DR.
Scatter plots and forest plots were applied in our study to provide a more intuitive representation of the correlation between myopia and DR, as well as the effect size and 95%CI for each SNP ( Fig. 2 ; Fig. S2 ).
