Abstract
Background The epidemiological association between type 2 diabetes and cataract has been well-established. However, it remains unclear whether the two diseases share a genetic basis, and if so, whether this reflects a causal relationship.
Methods We utilized East Asian population-based genome-wide association studies (GWAS) summary statistics of type 2 diabetes (Ncase=36,614, Ncontrol=155,150) and cataract (Ncase=24,622, Ncontrol=187,831) to comprehensively investigate the shared genetics between the two diseases. We performed 1. linkage disequilibrium score regression (LDSC) and heritability estimation from summary statistics (ρ-HESS) to estimate the genetic correlation and local genetic correlation between type 2 diabetes and cataract; 2. multiple Mendelian randomization (MR) analyses to infer the putative causality between type 2 diabetes and cataract; and 3. Summary-data-based Mendelian randomization (SMR) to identify candidate risk genes underling the causality.
Results We observed a strong genetic correlation (rg=0.58; p-value=5.60×10−6) between type 2 diabetes and cataract. Both ρ-HESS and multiple MR methods consistently showed a putative causal effect of type 2 diabetes on cataract, with estimated liability-scale MR odds ratios (ORs) at around 1.10 (95% confidence interval [CI] ranging from 1.06 to 1.17). In contrast, no evidence supports a causal effect of cataract on type 2 diabetes. SMR analysis identified two novel genes MIR4453HG (βSMR=−0.34, p-value=6.41×10−8) and KCNK17 (βSMR=−0.07, p-value=2.49×10−10), whose expression levels were likely involved in the putative causality of type 2 diabetes on cataract.
Conclusions Our results provided robust evidence supporting a causal effect of type 2 diabetes on the risk of cataract in East Asians, and posed new paths on guiding prevention and early-stage diagnosis of cataract in type 2 diabetes patients.
Key Messages
We utilized genome-wide association studies of type 2 diabetes and cataract in a large Japanese population-based cohort and find a strong genetic overlap underlying the two diseases.
We performed multiple Mendelian randomization models and consistently disclosed a putative causal effect of type 2 diabetes on the development of cataract.
We revealed two candidate genes MIR4453HG and KCNK17 whose expression levelss are likely relevant to the causality between type 2 diabetes and cataract.
Our study provided theoretical fundament at the genetic level for improving early diagnosis, prevention and treatment of cataract in type 2 diabetes patients in clinical practice
Introduction
Type 2 diabetes is one of the most prevalent chronic diseases in East Asians1, and cataract is a major cause of vision impairment among patients with type 2 diabetes2. Previous studies2, 3 have revealed a strong phenotypic association between type 2 diabetes and cataract for East Asians. For instance, Foster et al. conducted a cross-sectional study of 1,206 Singapore Chinese and found patients with diabetes had higher risks for obtaining cortical cataract3. Another Asian population-based study recruited 10,033 participants and identified diabetes as a significant risk factor for elevating incidence of cataract surgery2.
The phenotypic association between type 2 diabetes and cataract could be partially explained by their shared genetics4, 5. As pieces of evidence for disclosing their shared genetics, Lee et al.4 analyzed a Hong Kong Chinese cohort and found cataract is common in patients with type 2 diabetes who carried microsatellite polymorphism around aldose reductase-related genes. Lin et al.5 identified multiple candidate genes that had significantly different expression levels in the type 2 diabetes patients with higher Lens Opacities Classification System (LOCS) score (i.e., a system used to grade age-related cataract6), comparing to the patients with zero or minor LOCS score. However, the magnitude of the genetic association between type 2 diabetes and cataract remains unclear, as does the problem of whether their genetic association reflects a causal relationship.
Traditional methods estimate the shared genetics by comparing the concordance between monozygotic and dizygotic twins7, and establish causal conclusions using the randomized controlled trials (RCTs)8, a widely accepted gold standard for causal inference. However, these methods are occasionally limited or impracticable due to their own methodological weakness, such as the laborious data collection process and unethical study design. With the development of genome-wide association studies (GWAS) during past decades, some alternatively feasible statistical methods have been proposed to estimate the shared genetics between focal traits directly using the GWAS summary data9. For instance, Bulik-Sullivan et al.10 developed a technique named linkage disequilibrium score regression (LDSC) to estimate the contributions of polygenic genetic effects for a focal trait (i.e., single-trait heritability) and the magnitude of shared genetic overlap underlying two traits (i.e., cross-trait genetic correlation). Shi et al.11 extended LDSC and proposed heritability estimation from summary statistics (ρ-HESS), a method to quantify the local single-trait heritability and cross-trait genetic correlation from approximately LD-independent genomic regions. For pair of traits with significant genetic correlation, Mendelian randomization (MR)12 methods are capable of inferring the potential genetic causal relationship between traits using single nucleotide polymorphisms (SNPs) as instruments. To further investigate any putative functional genes underlying the susceptibility to a trait, Zhu et al.13 proposed summary data-based Mendelian randomization (SMR), which is an approach to identify gene expressions in an association with a target trait, by integrating GWAS summary data with expression quantitative trait loci (eQTL) summary data.
In this study, we leveraged the large East Asian population-based GWAS summary statistics of type 2 diabetes and cataract from BioBank Japan Project (BBJ)14 to comprehensively investigate the shared genetics between the two diseases. We applied LDSC, ρ-HESS, and seven MR or MR-equivalent approaches to estimate the genetic correlation, local genetic correlation, and potential genetic causality between type 2 diabetes and cataract, respectively. We also conducted SMR to the single-trait GWAS (i.e., type 2 diabetes, cataract) and cross-trait GWAS meta-analyses of type 2 diabetes and cataract to explore candidate genes involved in the causality between two diseases. A brief overview of our study is summarized in Fig. 1.
Research Design and Methods
GWAS Data Source
We downloaded the GWAS summary statistics of type 2 diabetes15 and cataract16 from the BBJ (http://jenger.riken.jp/en/), a database with common diseases, complex traits, and demographic and genotype data from ~200,000 Japanese individuals (53.10% male; average baseline age at 62.70 for men and 61.50 for women14). Both type 2 diabetes and cataract were diagnosed by physicians. The GWAS of type 2 diabetes was generated using a fixed-effect inverse-variance meta-analysis via METAL17, comprising 36,614 cases and 155,150 controls of four Japanese ancestry-based cohorts15. The GWAS of cataract was generated from 24,622 cases and 187,831 controls using a linear mixed model via SAIGE18, adjusted by age, sex, and top five principal components16. Both GWAS were based on the hg19 coordinate.
LDSC of single-trait heritability and cross-trait genetic correlation
We applied LDSC10, 19 to estimate the liability-scale heritability (h2) of type 2 diabetes and cataract as well as their genetic correlation (rg). GWAS summary statistics were filtered according to the HapMap3 reference20. SNPs were excluded if they were strand-ambiguous, had minor allele frequency <0.01, or located within the major histocompatibility complex (MHC) region (chromosome 6: 28,477,797–33,448,354) due to the complicated LD structure in this region21. The LD scores were pre-computed based on the 481 East Asians in 1000 Genomes (https://alkesgroup.broadinstitute.org/LDSCORE/). Univariate LDSC was performed to estimate the liability-scale h2 of type 2 diabetes and cataract, assuming the population and sample prevalence at 7.50%15 and 19.10%15 for type 2 diabetes, and 0.09%22 and 11.59%16 for cataract, respectively. Bivariate LDSC was utilized to estimate the genetic correlation (i.e., rg) between type 2 diabetes and cataract with and without a constrained intercept, which is designed to reduce the bias from population stratification. A significant rg was determined with p-value <0.05.
ρ-HESS of local genetic correlation
To explore whether type 2 diabetes had significant genetic overlap with cataract in some specific independent genomic regions, we performed ρ-HESS11 to estimate the local genetic correlations between type 2 diabetes and cataract according to the hg19-based 1000 Genomes East Asian reference. A total of 1,439 approximately LD-independent genomic regions (with the exclusion of the MHC region)23 were utilized in our analysis. The regions were excluded if the estimated local single-trait heritability was negative because of the insufficient study power. The estimated local genetic correlations were divided into four regional types: 1. the regions harboring significant type 2 diabetes-specific SNPs (i.e., ‘type 2 diabetes-specific’); 2. the regions harboring significant cataract-specific SNPs (i.e., ‘cataract-specific’); 3. the regions harboring shared SNPs significantly associated with both type 2 diabetes and cataract (i.e., ‘intersection’); and 4. other regions (i.e., ‘neither’). Three GWAS p-value thresholds, 5×10−8, 1×10−5, and 1×10−3, were used to define the significant SNPs. For these four regional types occupied by more than 10 regions, we calculated the mean and standard error of local genetic correlations within each type. A potential causal effect of type 2 diabetes on cataract is suggested if the average local genetic correlation at type 2 diabetes-specific regions and cataract-specific regions were significantly and non-significantly different from zero, respectively. The opposite is true for inferring potential causal effect of cataract on type 2 diabetes. Besides, the existence of pleiotropic effect may be implicated if there is a non-zero average local genetic correlation at intersection regions.
MR analyses for genetic causality inference
The causal relationship between type 2 diabetes and cataract was evaluated by six MR approaches (i.e., inverse variance weighted [IVW] model24, MR-Egger model25, generalized summary-data-based Mendelian randomization [GSMR]26, weighed median model27, and weighted mode model28, and the causal analysis using summary effect estimates [CAUSE]29) and one MR-equivalent latent causal variable (LCV) model30. Multiple methods were employed because they have different assumptions on horizontal pleiotropy, a term defined as the instrumental SNPs with effects on both exposure and outcome through non-causal pathways12. Horizontal pleiotropy is a potential confounding factor for inferring causality and can be divided into uncorrelated pleiotropy if the instrumental SNPs influence exposure and outcome via independent mechanisms, and correlated pleiotropy if the instrumental SNPs affect exposure and outcome through shared factors12. The consistent results of multiple MR methods are expected to effectively minimize the impact of horizontal pleiotropy31 from putative causality and thus reduce the false-positive rate.
Among the seven methods, IVW measures the causal effect by integrating ratios of variant effects (ratio estimates) between exposure and outcome, assuming a balanced uncorrelated pleiotropy (with mean zero) and no correlated pleiotropy. MR-Egger assumes no correlated pleiotropy and a non-zero uncorrelated pleiotropy, which adds an extra intercept compared to IVW to represent the magnitude of uncorrelated pleiotropy. GSMR assumes the presence of uncorrelated pleiotropy and excludes such effect by outlier removal using heterogeneity in dependent instrument (HEIDI) approach. The weighted median model assumes the proportion of pleiotropic (both uncorrelated and correlated) instrumental SNPs is less than half, and calculates the causal effect using the weighted median of the SNP ratio. The weighted mode model greatly loosens the assumptions on uncorrelated and correlated pleiotropy and measures the causal effect only from the most frequent (the mode) SNP set with consistent effect. At least 10 independent instrumental SNPs are required for maintaining study power using these five MR methods. Independent instrumental SNPs are selected from the exposure-specific genome-wide significant (GWAS p-value<5×10−8) SNPs that are also merged with outcome GWAS, and then clumped by LD r2<0.05 within 1,000 kb window using PLINK version 1.932 according to the reference genome of 1000 Genomes East Asian33. If the number of independent instrumental SNPs was less than 10, we selected the ‘proxy’ instrumental SNPs by relaxing the exposure GWAS p-value threshold to 1×10−5 to maintain study power. The LCV model30 is a MR-equivalent method that assumes the genetic correlation between two traits is mediated by a latent variable, and distinguishes causality from uncorrelated and correlated pleiotropy by measuring the genetic causality proportion (GCP) using all genetic variants30. CAUSE29 is a method that is more powerful and sensitive to identify the causality from uncorrelated and correlated pleiotropy compared to other models. CAUSE increases MR detection power by recruiting more approximately independent instrumental SNPs with GWAS p-value <1×10−3 and pruned by LD r2 <0.1. CAUSE also provides an expected log pointwise posterior density (ELPD) test to compare the overall fitness among a causal model (i.e., instrumental SNPs act on exposure and outcome through a causal pathway and shared factors), a sharing model (i.e., instrumental SNPs act on exposure and outcome only through shared factors), and a null model (i.e., no causal pathway or shared factors underlying exposure and outcome).
We performed these models using R packages “cause” (version: 1.0.0), “LCV”, “gsmr” (version: 1.0.9) and “TwoSampleMR” (version: 0.5.4). Any instrumental SNPs located within the MHC region were excluded21. A significant causal relationship was determined if the causal effect estimates were consistent and significant at the Bonferroni-corrected level (with p-value<0.05/13≈3.85×10−3, including six bi-directional MR methods and LCV model). The causal effects (i.e., β) were converted from logit-scale to liability-scale using the method described by Byrne et al34: where Kx and Ky are the population prevalence of exposure and outcome, and and are the height of standard normal distribution at such prevalence. We assumed the population prevalence for type 2 diabetes and cataract are 7.5%15 and 0.09%22 that are same as the population prevalence used in estimating the liability heritability. The liability β was then transformed into odds ratio (OR).
SMR analysis to identify candidate genes underling the genetic causality
When a causal relationship between type 2 diabetes and cataract is established, we applied SMR13 to identify candidate risk genes underling the causality between the two diseases. SMR leverages GWAS and eQTL summary statistics to explore the association between gene expression and a target disease or trait using a MR equivalent analysis. SMR further utilizes HEIDI-outlier test to check if the significant association between gene expression and the target trait/disease is due to the causality (i.e., causal SNPs drive disease by regulating gene expression levels) or pleiotropy (i.e., causal SNPs influences both disease and gene expression via shared effects) rather than linkage (i.e., different causal SNPs in LD influences disease and gene expression, respectively).
In our study, we performed SMR using cis-eQTLgen summary data (19,250 expression probes in blood; URL: https://eqtlgen.org/cis-eqtls.html)35 to the single-trait GWAS (i.e., type 2 diabetes, cataract) and cross-trait GWAS of type 2 diabetes and cataract generated by inverse-variance-based meta-analysis via METAL36. Significant gene expressions due to causality or pleiotropy were determined if with a study-wise Bonferroni-corrected SMR p-value <0.05/19,250/3 ≈ 8.66×10−7 and a HEIDI-outlier p-value >0.05 calculated from minimum 10 SNPs. To identify any candidate risk genes possibly involved in the causality between type 2 diabetes and cataract, we focused on the gene expressions that were significantly associated with cross-trait meta-analysis of type 2 diabetes and cataract, but not with the original single-trait disease (i.e., type 2 diabetes or cataract).
Results
Strong genetic association between type 2 diabetes and cataract
As shown in Table 1, the estimated SNP-based liability-scale h2 for type 2 diabetes and cataract were 21.47% (standard error [SE]=1.17%, p-value=3.28×10−75) and 1.63% (SE=0.13%, p-value=4.60×10−36) with constrained LDSC intercept, respectively. These h2 decreased to 15.12% (SE=1.37%, p-value=2.55×10−28) and 0.54% (SE=0.19%, p-value=4.48×10−3) without constrained LDSC intercept, suggesting the mild inflations in both diseases GWAS. We then performed the bivariate LDSC with and without constrained intercept, and identified the significant genetic correlation rg between type 2 diabetes and cataract at 0.28 (SE=0.05, p-value=3.25×10−9) and 0.58 (SE=0.13, p-value=5.60×10−6), respectively, indicating strong shared genetics between type 2 diabetes and cataract.
ρ-HESS analyses of local genetic correlations
We conducted ρ-HESS to estimate the local heritability of type 2 diabetes and cataract (detailed results in Table S1 and Fig. 2A). We also estimated the local genetic covariance and correlation between type 2 diabetes and cataract in 824 regions (detailed in Table S1) after excluding the regions with negative local heritability. As shown in Table 2, we identified six genomic regions at a nominal significance level (p-value<0.05) from different chromosomes, with estimated local rg at [0.48, 1).
We further investigated the distribution of local genetic correlations in four regional types. As shown in Fig. 2B, regions harboring type 2 diabetes-specific SNPs were identified with average local rg significantly higher than zero. In reverse, regions harboring cataract-specific SNPs showed a non-significant average local rg close to zero. Therefore, the distribution of local rg revealed by ρ-HESS suggested a potential putative causal relationship of type 2 diabetes on cataract. Besides, the average local rg from the ‘intersection’ regions harboring shared significant SNPs with GWAS p-value<1×10−3 was estimated at 0.22 (SE=0.03, p-value=1.19×10−15), suggesting the mild pleiotropic effects of ‘less significant’ genetic variants may be underlying type 2 diabetes and cataract, while we cannot further distinguish the causality of type 2 diabetes on cataract from such pleiotropy here using ρ-HESS.
Putative causality of type 2 diabetes on cataract
Application of seven MR or MR-equivalent methods consistently detected a causal effect of type 2 diabetes on cataract at Bonferroni-corrected significance level (p-value<3.85×10−3), detailed in Table S2. In reverse, there was no or modest evidence for a causal effect of cataract on type 2 diabetes. As shown in Fig. 3, six MR methods provided consistent evidence for a causal effect of type 2 diabetes on cataract with estimated liability-scale ORs ranging from 1.07 to 1.13, under the assumption of the population prevalence at 7.50% of type 2 diabetes and 0.09% of cataract, respectively. These results indicated that individuals with type 2 diabetes had approximately 1.06 to 1.17 times of risks for developing cataract compared to healthy individuals. Besides, LCV provided a GCP at 0.87 (p-value=1.54×10−9), suggesting that the strong genetic correlation between type 2 diabetes and cataract can be largely explained by the causality of type 2 diabetes on cataract. Remarkably, our putative causality of type 2 diabetes on cataract was less likely to be influenced by the horizontal pleiotropy because of the close-to-zero MR-Egger intercept (−0.002, p-value =0.47) and the better model fitness of causal model (with non-significant effects of correlated pleiotropy [η=−0.02, 95% CI=−0.53 to 0.30] and uncorrelated pleiotropy [q=0.04, 95% CI=0 to 0.25]) compared to the sharing model (ELPD p-value=8.80×10−3) and the null model (ELPD p-value=4.69×10−13) revealed by CAUSE (Fig. S1).
Two candidate genes likely involved in the causality of type 2 diabetes on cataract
As shown in Table S3, we performed a cross-trait meta-analysis of type 2 diabetes and cataract using METAL, and identified 9 independent ‘novel’ SNPs that were associated with cross-trait of type 2 diabetes and cataract but not with the original GWAS of type 2 diabetes or cataract. Next, we applied SMR to the single-trait GWAS and the cross-trait meta-analysis GWAS of type 2 diabetes and cataract, and identified two candidate risk genes (Table 3), MIR4453HG (βSMR=−0.34; SMR p-value=6.41×10−8; HEIDI p-value=0.08 from 13 SNPs) and KCNK17 (βSMR=−0.07; SMR p-value=2.49×10−10; HEIDI p-value=0.08 from 17 SNPs), whose expression levels were negatively associated (i.e., lower gene expression level increases the disease risk) with the susceptibility to co-morbid type 2 diabetes and cataract but not with the single-traits. These genes likely play crucial roles in the casual effects of type 2 diabetes on cataract.
Discussion
To our knowledge, this is the first study to quantify the genetic correlation and explore the potential causality between type 2 diabetes and cataract specifically using East Asian population-based GWAS summary statistics. Our results have highly enriched our current knowledge on the shared genetic architecture between type 2 diabetes and cataract.
Previously, researchers preferred to define co-occurrence of cataract and diabetes as a single outcome (i.e., diabetic cataract) and explored its genetics straightforwardly. For example, Lin et al.5 performed a GWAS using 758 Chinese cases with type 2 diabetic cataract and 649 healthy controls and identified 15 independent genome-wide significant SNPs, which are associated with blood sugar regulation and cataract development. Another study37 recruited 2,501 Scottish cases and 3,032 controls and found a significant role of rs2283290 in triggering diabetic cataract. Instead of using a single GWAS dataset with a small number of diabetic cataract patients, we leveraged large population-based GWAS summary statistics of type 2 diabetes and cataract, which is more powerful and provided robust evidence supporting the shared genetics between type 2 diabetes and cataract3,4,38.
Using ρ-HESS, we identified six genomic regions with a significant local genetic correlation between type 2 diabetes and cataract. Assuming these regions might contribute to the causal effect of type 2 diabetes on cataract, any SNPs or genes that are located within such regions and associated with type 2 diabetes and/or cataract risks are of great interest to understand the mechanisms underlying the regions. Therefore, we collected information from a total of 254 SNPs in ClinVar39 and genes in Malacards (supported by trustworthy sources or Cochrane based reviews40) for further analyses (Table S5). We identified gene HNF1B (hepatocyte nuclear factor 1β; Chromosome: 17: 36,046,434–36,105,096) and two SNPs (i.e., rs121918673 [Chromosome: 17: 36,061,127] and rs1555818071 [Chromosome: 17: 36,047,338]; located within gene HNF1B) that located within the significant genomic region on chromosome 17: 34,395,061–36,495,389 and reported to be associated with type 2 diabetes41, 42. In contrast, no SNPs or genes located in the ρ-HESS estimated significant genomic regions were found to be associated with cataract risk. Additionally, this result revealed a large proportion of shared genetics between type 2 diabetes and cataract were from the ‘type 2 diabetes-specific’ regions. These findings provided further evidence that the strong genetic correlation between type 2 diabetes and cataract is due to the type 2 diabetes-specific variants.
Application of seven MR and MR-equivalent methods provided consistent results for a causal effect of type 2 diabetes on cataract. Our findings raise an important clinical concern in prevention and early-diagnosis of cataract in patients with type 2 diabetes. We provided theoretical basis at genetic level for suggesting that assessing the development and severity of type 2 diabetes is likely yielding new targets for early-diagnosis of cataract, while further studies are required to pinpoint the potential aetiology underlying type 2 diabetes and cataract.
We also tried to replicate our findings in the European cohort using the European population-based publicly available GWAS summary statistics of type 2 diabetes43 (Ncase=62,892, Ncontrol=596,424) and cataract (Ncase=5,045, Ncontrol=356,096; UKB field ID: 20002; accessed from URL: http://www.nealelab.is/uk-biobank). However, LDSC analysis indicated a non-significant genetic correlation between the two diseases according to either European or East Asian reference (see Table S6). This result suggests the shared genetic variance between type 2 diabetes and cataract in East Asians may have strong genetic heterogeneity compared to Europeans. Future investigations are required for a better understanding of such difference.
To identify any blood-based biomarkers that may contribute to the causal effect of type 2 diabetes, we performed the multi-trait-based conditional & joint analysis (mtCOJO)44 to adjust both type 2 diabetes and cataract GWAS on each blood-based biomarker and then conducted post-mtCOJO MR analysis on adjusted type 2 diabetes and cataract GWAS (see Supplementary note, Table S7-9, and Fig. S2-3). We found that HbA1c (i.e., Hemoglobin A1c) may be involved in the causality of type 2 diabetes on cataract, standing in line with previous RCTs showing the impact of glycemic control on the prevention of ocular complications45–47. However, this result was possibly caused by the high genetic correlation between HbA1c and type 2 diabetes (rg=0.57 and 0.84 with and without constrained intercept) which may greatly decrease the heritability of type 2 diabetes and thus reduced the genetic correlation and putative causal relationship between type 2 diabetes and cataract. Future investigations should focus on this finding with the recruitment of a larger sample size.
We identified two candidate functional genes MIR4453HG and KCNK17 that are likely relevant to the genetic causality between type 2 diabetes and cataract. Interestingly, both genes described a significant association with single-trait type 2 diabetes due to linkage (i.e., not passed HEIDI-outlier test), and then showed a more significant association with cross-trait type 2 diabetes and cataract due to causality or pleiotropy, further suggesting that cataract is likely an outcome triggered by the genetic mutations of type 2 diabetes. MIR4453HG is an IncRNA gene and located nearby some risk genes that have been reported to be associated with blood protein level (gene ARFIP148) and lipoprotein cholesterol levels (gene TRIM249). Both traits are highly relevant to the risk for type 2 diabetes50,51 and cataract52,53. KCNK17 encoded a protein in the family of potassium channel54. The mutation of KCNK17 may cause the abnormal opening of potassium channels and is associated with cardiovascular diseases (e.g., ischemic stroke and cerebral hemorrhage)54, which are known to be involved in the susceptibility to both type 2 diabetes55 and cataract56. These results provided novel insights on the genetic mechanisms underlying the causality between type 2 diabetes and cataract. Further wet-lab experiments were required to approve the roles of these two genes in increasing cataract risks in type 2 diabetes patients.
Our study has several limitations. First, the heritability of cataract was tiny with an estimate less than 2%, which may bias the estimate of genetic correlation between type 2 diabetes and cataract. Nevertheless, this effect should be negligible as the heritability of cataract is significantly different from zero. Secondly, the number of instrumental SNPs using cataract as exposure is less than 10. Instead, we selected the ‘proxy’ instrumental SNPs with p-value <1×10−5, which may violate assumptions of some MR methods (e.g., GSMR). However, the MR effects of these MR methods are highly consistent with CAUSE, suggesting the feasible application of using the ‘proxy’ instrumental SNPs. Thirdly, due to the limitation of our statistical models, we did not investigate the genetic contributions of the MHC region on the susceptibility to co-morbid type 2 diabetes and cataract, which possibly underestimated the shared genetic between the two diseases.
In summary, we provide robust evidence for a strong genetic association between type 2 diabetes and cataract, and a putative causal effect of type 2 diabetes on cataract particularly in East Asians. Lower expression of two novel candidate genes MIR4453HG and KCNK17 were identified to be possibly involved in the causality between type 2 diabetes and cataract. Our results provided theoretical fundament at the genetic level for improving early diagnosis, prevention and treatment of cataract in type 2 diabetes patients in clinical practice.
Data availability statement
Summary statistics are publicly available at http://jenger.riken.jp/en/.
Supplementary Data
Supplementary data are available at IJE online.
Funding
The work was funded by the Natural Science Foundation of China (81801132, and 81971190; HY.Zhao, Sun Yat-sen Memorial Hospital; 61772566, 62041209, and U1611261; YD.Y, Sun Yat-sen University), Guangdong Key Field R&D Plan (2019B020228001 and 2018B010109006; YD.Y, Sun Yat-sen University), Introducing Innovative and Entrepreneurial Teams (2016ZT06D211, YD.Y, Sun Yat-sen University), Guangzhou S&T Research Plan (202007030010, YD.Y, Sun Yat-sen University), and Mater Foundation (YH.Y, Mater Research).
Author contributions
YH.Y and H.Zhao designed the study. H.Zhang and X.X conducted analyses, with assistance from YH.Y, H.Zhao and A.X. H.Zhang, YH.Y, and H.Zhao wrote the manuscript. YD.Y, YH.Y, and H.Zhao supervised the study. All authors contributed to the final revision of the paper.
Conflict of Interest
All authors state they have no competing interests.
Acknowledgments
The authors thank the BBJ project for making data available.