If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Epidemiology and Health Statistics, School of Public Health, Capital Medical University, Beijing, ChinaDepartment of Epidemiology and Bio-statistics, School of Public Health, Peking University, Beijing, China
Department of Neurology, Beijing Tiantan Hospital, Capital Medical University, ChinaChina National Clinical Research Center for Neurological Diseases, Beijing, China
Centre for Precision Health, Edith Cowan University, Perth, WA, AustraliaSchool of Medical and Health Sciences, Edith Cowan University, Perth 6027, Australia
Corresponding author. Department of Epidemiology and Health Statistics, School of Public Health, Capital Medical University, 10 Xitoutiao, Youanmen, Fengtai District, Beijing, 100069, China.
Department of Epidemiology and Health Statistics, School of Public Health, Capital Medical University, Beijing, ChinaCenter for Primary Health Care Research, Lund University/Region Skåne, Malmö, Sweden
Mendelian randomization was used to assess the effects of lipids and drugs on diabetic retinopathy (DR).
•
HDL-C was inversely related to severe nonproliferative DR and proliferative DR.
•
No association between CETP inhibitor and DR was found.
Abstract
Background and aims
The evidence that dyslipidemia is associated with hyperglycemia calls for an investigation of whether dyslipidemia, as well as lipid-modifying agents, could affect the subsequent development of diabetic retinopathy (DR). Therefore, we aimed to address these unanswered questions by utilizing Mendelian randomization (MR) analysis.
Methods
Genetic variants were selected from the UK Biobank as instruments to serve as proxies for lipid traits [high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), triglyceride (TG), apolipoprotein A-I (APOA-I) and apolipoprotein B (APOB)]. Univariable and multivariable MR analyses were performed to examine the associations of these lipid traits with DR and different levels of severity of DR. Based on the evidence for the effects of lipids on outcomes, we estimated the causal relevance of cholesteryl ester transfer protein (CETP) inhibitors in severe nonproliferative and proliferative DR using protein quantitative trait loci (pQTLs) and expression quantitative trait loci (eQTLs) as instruments.
Results
Genetically determined HDL-C levels were inversely associated with the risk of severe nonproliferative DR (OR = 0.70, 95% CI = 0.52–0.94) and proliferative DR (OR = 0.90, 95% CI = 0.83–0.97) in the main analyses utilizing the inverse variance-weighted (IVW) MR method and a couple of sensitivity analyses. No association was noted between genetically proxied CETP inhibitors and DR.
Conclusions
This MR study suggests the casual protective roles of HDL-C in severe nonproliferative DR and proliferative DR, which calls for further studies to confirm these findings. Current lipid-modifying agents acting on HDL-C may not reduce the risk of DR and new treatments are required in the future.
]. The current treatment strategies for DR, such as anti-vascular endothelial growth factor (VEGF) agents and laser photocoagulation, have many side effects, and some patients with DR continue to lose vision [
]. The Action in Diabetes and Vascular Disease: Preterax and Diamicron-MR Controlled Evaluation (ADVANCE) study followed patients for five years and suggested that there was no statistically significant relationship between high-density lipoprotein cholesterol (HDL-C) and DR [
]. However, a global case‒control study in 13 countries found that HDL-C was associated with a decreased risk of DR after matching for diabetes duration, age, sex, and low-density lipoprotein cholesterol (LDL-C) [
Association between plasma triglycerides and high-density lipoprotein cholesterol and microvascular kidney disease and retinopathy in type 2 diabetes mellitus: a global case-control study in 13 countries.
The standard method for identifying causality is the randomized clinical trial (RCT). However, RCTs are expensive and have poor feasibility. Mendelian randomization (MR) is a method that could overcome these potential limitations. MR, a natural experiment analogous to RCTs, uses genetic variants as instruments to elucidate the causal relevance of exposures with outcomes, which can minimize the effects of potential confounding and reverse causality [
]. A previous MR study based on 2969 cases and 4096 controls showed no evidence of causal roles of the four lipid levels [HDL-C, LDL-C, total cholesterol (TC) and triglyceride (TG)] in DR [
]. However, the influence of apolipoprotein A-I (APOA-I) and apolipoprotein B (APOB) on DR was not evaluated in this study. Therefore, we assessed the relationships of lipid traits, including HDL-C, LDL-C, TG, APOA-I and APOB, with overall DR based on a large sample of 18,097 cases and 239,347 controls, as well as background DR, severe nonproliferative DR, and proliferative DR. To date, it is still unknown whether lipid-modifying agents can affect patients’ susceptibility to DR. MR can also provide important information for drugs, including predicting treatment effects and revealing adverse effects of targeted drugs [
]. Lipid-modifying agents are classified by their predominant lipid targets. Therefore, we selected lipid-modifying agents based on the evidence of the role of lipids in outcomes to investigate their association with severe nonproliferative and proliferative DR using the protein quantitative trait loci (pQTLs) and expression quantitative trait loci (eQTLs) as instruments.
2. Materials and methods
2.1 Study design
We conducted two MR analyses. First, univariable MR and multivariable MR were performed to analyze the direct effects of lipids, including HDL-C, LDL-C, TG, APOA-I, and TG, on DR and different severities of DR. Second, based on the evidence of the effects of lipids on outcomes, we selected lipid-modifying agents to evaluate the associations with outcomes by performing MR using the pQTL and eQTL data. The study design is shown in Fig. 1. The protocol and data collection were approved by the ethics committee for the original genome-wide association studies (GWASs), pQTL, and eQTL, and participants provided informed consent. This study conforms to the ethical guidelines of the 1975 Declaration of Helsinki.
The largest available GWASs of LDL-C (440,546 samples), HDL-C (403,943 samples), TG (441,016 samples), APOA-I (393,193 samples) and APOB (439,214 samples) were obtained from the UK Biobank [
Evaluating the relationship between circulating lipoprotein lipids and apolipoproteins with risk of coronary heart disease: a multivariable Mendelian randomisation analysis.
]. These GWAS analyses were adjusted for age, sex, and a binary variable denoting the genotyping chip individuals were allocated to in UKB. Each trait was standardized with a mean of zero and a standard deviation (SD) of one. The details of the data are presented in Table 1. For the main univariable analysis using inverse variance-weighted (IVW) MR, we selected genetic variants associated with specific lipid fractions at the genome-wide significance level (p < 5 × 10−8) and filtered for linkage disequilibrium with a cutoff of r2 < 0.001 based on the European 1000 Genomes panel (http://fileserve.mrcieu.ac.uk/ld/1kg.v3.tgz). Single nucleotide polymorphisms (SNPs) were excluded if they had a minor allele frequency (<0.01), or were located in the major histocompatibility complex region (chromosome 6: 28,477,797–33,448,354) because of their complicated linkage disequilibrium structure [
]. Finally, we used radial MR to detect and remove outliers by setting a threshold for identifying outliers (0.05 in our model) and using modified second-order weights [
]. The remaining SNPs were used as instruments to perform MR analyses. The details of the selected SNPs are described in Supplementary Data 1. For the multivariable MR analyses, we pooled all genome-wide significant (p < 5.0 × 10−8) SNPs that were associated with any exposure, and further clumped by linkage disequilibrium (r2 < 0.001 within a window of 10,000 kb) to ensure that the SNPs were independent.
Based on the evidence of the effects of genetically determined lipids on outcomes, we initially selected two lipid-modifying agents that significantly elevate HDL-C levels, including nicotinic acids [
Pharmacological inhibition of CETP (cholesteryl ester transfer protein) increases HDL (High-Density lipoprotein) that contains ApoC3 and other HDL subspecies associated with higher risk of coronary heart disease.
], we only assessed the effects of genetically determined CETP inhibitors on DR by using significantly related pQTLs and eQTLs as instruments. First, the significant pQTLs (p < 5.0 × 10−8) that determined CETP concentration [
CETP (cholesteryl ester transfer protein) concentration: a genome-wide association study followed by mendelian randomization on coronary artery disease.
Circulation Genomic and precision medicine.2018; 11e002034
] were selected to index CETP inhibitors and conduct the MR analyses. Second, the eQTL data on CETP (the target gene of CETP inhibitors) gene expression levels were obtained from the eQTLGen consortium (https://eqtlgen.org/). We used a common (minor allele frequency >0.01) SNP top-associated with the expression of the target gene in blood as an instrumental variable to perform summary data-based MR (SMR) analyses. The details of the genetic instruments are shown in Supplementary Table S1.
2.3 Data sources for outcomes
The GWASs for DR (18,097 cases, 239,347 controls) and the different severities of DR, including background DR (2510 cases, 242,308 controls), severe nonproliferative DR (568 cases, 242,308 controls) and proliferative DR (10,860 cases, 242,308 controls), were obtained from the FinnGen consortium (release 6, https://r6.finngen.fi/) [
] and were identified based on International Classification of Diseases, Revision 10 (H36.0, H36.00, H36.02, and H36.03, respectively) (Table 1). Control subjects in the GWASs of DR were defined as subjects without DR and diabetic complications. There were 214,308 general population without diabetes (89.54%) in the control group, which decreased collider bias [
]. These GWAS analyses were adjusted for age, sex, genetic relatedness, genotyping batch, and first principal components.
2.4 Statistical analyses
2.4.1 Analyses for genetically determined lipid traits and outcomes
To assess the validation of GWASs on outcomes, positive control analyses focusing on diabetes mellitus and DR based on the IVW MR method were undertaken (Supplementary Data 2).
IVW MR was the main analysis method used to estimate the causal effects of genetically determined lipid traits on DR. The multiplicative random-effects meta-analysis combined the Wald ratio estimates of every SNP to obtain causal IVW estimates [
]. This method can obtain accurate causal effects, but these might be affected by pleiotropy. Therefore, we used several sensitivity analyses to evaluate the robustness of the results and check for pleiotropy. These analyses included Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) and multivariate MR. The MR-PRESSO method evaluates whether the exclusion of potential outlier SNPs influences the results, which is an indication of potential pleiotropy [
]. Multivariate MR analyses were implemented for observed associations. In order to avoid multicollinearity issues because of their high correlations, we did not simultaneously adjust for APOA-I and HDL-C, as well as APOB and LDL-C, in the multivariateMR analyses [
]. Therefore, we conducted multivariate MR models to assess the potential independent association of lipid traits with outcomes by first adjusting for other lipid-related traits and then additionally adjusting for hemoglobin A1c (HbA1c) and systolic blood pressure (SBP).
To ensure the robustness of the results, we performed three sensitivity analyses. First, we conducted Cochran's Q test to evaluate the heterogeneity between SNPs and performed MR‒Egger regression to assess horizontal pleiotropy. The strength of the genetic instruments for each lipid trait was estimated using the mean F-statistic with the approximation method described by Bowden et al. [
]. All SNPs included in the study had an F-statistic of more than 10, suggesting that weak instrument bias was minimized. Second, we changed the p value threshold to 0.1 and 0.2 in radial MR to exclude more outliers and reanalyze the relationships. Third, to ensure that the lipid traits were associated with DR and not with diabetes, that is, to ensure that the selected instrumental variables were not related to diabetes, we excluded the SNPs associated with related glycemic parameters using the phenoscanner website (http://www.phenoscanner.medschl.cam.ac.uk/) (Supplementary Table S2) and then reestimated the associations between lipid traits and DR using the IVW MR method.
2.4.2 Analyses of drug target gene expression and outcomes
We first used the IVW method as the primary MR approach to determine the effects of genetically predicted CETP concentrations on DR or different severities of DR. The F-statistic, Cochran's Q statistic and MR‒Egger were used to test for weak instrument bias, heterogeneity and pleiotropy, separately. Second, to validate the robustness of the results using the pQTLs as instruments, we additionally proposed an instrument by using the SNP top associated with CETP gene expression levels to proxy the exposure. The SMR method was adopted to generate effect estimates of CETP gene expression levels on DR or different severities of DR by combining summary-level data from GWAS and eQTL studies [
]. We used SMR software to harmonize alleles and explore the effects of exposures on outcomes. For SMR analysis, the strength of the genetic instrument was also evaluated using the F-statistic. The HEIDI test (p < 0.05) was applied to distinguish pleiotropy from linkage.
Bonferroni correction was used to adjust for multiple testing. For the association between lipid traits and DR, strong evidence was defined as p < 0.0025 (5 exposures and 4 outcomes), moderate evidence was defined as 0.0025 ≤ p < 0.0125 (1 exposure and 4 outcomes), and suggestive evidence was defined as 0.0125 ≤ p < 0.05. For the association between CETP inhibitors and DR, strong evidence was defined as p < 0.025 (1 exposure and 2 outcomes), and suggestive evidence was defined as 0.025 ≤ p < 0.05. Statistical analyses were conducted using the TwoSampleMR (version 0.5.6), MR-PRESSO (version 1.0), RadialMR (version 1.0), and phenoscanner (version 1.0) packages in R (version 4.1.2) and SMR software (version 1.03) (https://yanglab.westlake.edu.cn/software/smr/#Download).
3. Results
3.1 Genetically determined lipid traits and risks of outcomes using univariate MR and multivariate MR
Positive control studies showed that both type 1 diabetes and type 2 diabetes had significant associations with DR and the different severities of DR, which indicated that the GWASs of outcomes were effective (Supplementary Table S3).
Among the five lipid traits, we found that only genetically determined HDL-C levels were associated with a decreased risk of severe nonproliferative and proliferative DR in both univariate and multivariate MR (p < 0.05) (Fig. 2). For a one-SD increase in HDL-C, the ORs of severe nonproliferative DR and proliferative DR were 0.70 (95% CI = 0.52–0.94, p = 0.02) and 0.90 (95% CI = 0.83–0.97, p = 0.005), respectively, using the IVW MR method (Fig. 3). Neither of these ORs reached statistical significance for strong evidence, indicating suggestive and moderate evidence, respectively. Based on the MR-PRESSO method, no pleiotropic outlier was tested and the results were consistent with the IVW MR results. Specifically, a one-SD increase in HDL-C levels was also related to severe nonproliferative DR (OR = 0.70, 95% CI = 0.52–0.94, p = 0.02) and proliferative DR (OR = 0.90, 95% CI = 0.84–0.96, p = 0.001) (Supplementary Table S4). Notably, the association between HDL-C and proliferative DR reached statistical significance for strong evidence (p < 0.0025).
Fig. 2Results of the Mendelian randomization study on lipids and diabetic retinopathy.
The inverse variance-weighted method was used for main analysis. Model 1–8 are multivariable Mendelian randomization (MVMR). Model 1: adjusted for APOB, HDL-C, and TG; model 2: adjusted for APOA-I, APOB, and TG; model 3: adjusted for APOA-I, TG, and LDL-C; model 4: HDL-C, TG, and LDL-C. Model 5–8: adjusted for the variables in model 1–4, as well as for HbA1c and SBP, separately.
Fig. 3Associations of lipids with diabetic retinopathy and different severities of diabetic retinopathy using the inverse variance-weighted Mendelian randomization method.
Odds ratios represent the association of per standard deviation unit change of lipids with diabetic retinopathy and different severities of diabetic retinopathy. OR: odds ratio; CI: confidence interval; DR: diabetic retinopathy; LDL-C: low-density lipoprotein cholesterol; HDL-C: high-density lipoprotein cholesterol; TG: triglycerides; APOA-I: apolipoprotein A-I; APOB: apolipoprotein B.
In multivariate MR models, elevated HDL-C levels were associated with a lower risk of severe nonproliferative DR (after adjusting for TG and LDL-C: OR = 0.62, 95% CI = 0.43–0.91, p = 0.01; after adjusting for TG and APOB: OR = 0.67, 95% CI = 0.46–0.96, p = 0.03) and proliferative DR (after adjusting for TG and LDL-C: OR = 0.85, 95% CI = 0.76–0.95, p = 0.01; after adjusting for TG and APOB: OR = 0.87, 95% CI = 0.78–0.97, p = 0.01) (Supplementary Table S5). When HDL-C, TG, LDL-C (or APOB), HbA1c, and SBP were assessed together by performing multivariate MR analyses, we still found that a one-SD increase in genetically proxied HDL-C levels was associated with a decreased risk of severe nonproliferative DR and proliferative DR (Supplementary Table S6).
Other lipid fractions, including LDL-C, TG, APOA-I, and APOB, had partly consistent or no evidence with overall DR and the different severities of DR, as shown in Fig. 2. Notably, although the findings in the main analysis provided strong evidence of the associations between TG and DR, horizontal pleiotropy was detected. Additionally, multivariate MR analyses and the results obtained by changing the threshold for identifying outliers to 0.2 in radial MR showed that genetically proxied TG was not related to DR. (Fig. 2 and Supplementary Tables S5–S6).
In the sensitivity analyses, all F-statistics of the genetic instruments were larger than 10, which avoided the existence of weak instrument bias (Supplementary Data 1). We did not find evidence of instrumental heterogeneity based on the Cochran Q test (all p > 0.05). However, the results indicated horizontal pleiotropy of HDL-C (pintercept < 0.05) on DR; and TG on DR, background DR, and proliferative DR (all pintercept < 0.05) (Supplementary Table S7). When the thresholds for identifying outliers were 0.1 and 0.2 in radial MR, pleiotropy weakened to insignificant. The relationships of HDL-C with severe nonproliferative DR and proliferative DR were basically consistent with the main results (Supplementary Tables S8–S9). After excluding the SNPs related to glycemic traits, the IVW MR method showed that there was evidence for the causal effect of genetically determined HDL-C levels on severe nonproliferative DR (OR = 0.70, 95% CI = 0.52–0.94) and proliferative DR (OR = 0.91, 95% CI = 0.84–0.98) (Supplementary Fig. S1).
3.2 Genetically proxied CETP inhibitors and the risks of outcomes
Based on the evidence that HDL-C was associated with severe nonproliferative DR and proliferative DR, we investigated the causal relevance of CETP inhibitors, which can elevate HDL-C levels, in severe nonproliferative and proliferative DR. Using pQTL effect estimates for protein levels from Lisanne L. Blauw et al. [
CETP (cholesteryl ester transfer protein) concentration: a genome-wide association study followed by mendelian randomization on coronary artery disease.
Circulation Genomic and precision medicine.2018; 11e002034
], increases in genetically determined CETP levels were not associated with the decreased risk of severe nonproliferative DR (OR = 0.99, 95% CI = 0.66–1.50, p = 0.97) and proliferative DR (OR = 0.94, 95% CI = 0.85–1.05, p = 0.28) (Fig. 4). For all analyses, no evidence of instrumental heterogeneity was found (p > 0.05) (Supplementary Table S10). The F statistics for selected SNPs were >10.
Fig. 4Associations of genetically determined CETP inhibitors with severe nonproliferative DR and proliferative DR.
Additionally, the SMR analyses using a significant eQTL as a proxy of exposure showed no associations of CETP gene expression levels with severe nonproliferative and proliferative DR (Fig. 4 and Supplementary Table S10). The most significant SNPs selected as genetic instruments had an F statistic of 260.89 (Supplementary Table S1). The HEIDI test indicated that all associations were not due to linkage (p > 0.05) (Supplementary Table S10).
4. Discussion
4.1 Main findings
Using an integrated approach, including univariate MR and multivariate MR, our study found that there was consistent evidence for the associations of elevated levels of genetically determined HDL-C with a decreased risk of severe nonproliferative DR and proliferative DR. For MR analyses of genetically determined CETP inhibitors, the results indicated that neither CETP concentrations nor CETP gene expression levels were associated with DR. Therefore, it is recommended that new medicines acting on HDL-C should be developed in the future.
4.2 Comparison with previous studies and possible explanations
Current studies have examined the relationship between lipid concentrations and the risk of DR, but no definitive conclusions have been drawn. A population-based case‒control study indicated that HDL-C was a risk factor for DR [
]. In our study, we found that HDL-C was not related to the overall DR after adjusting for other lipid traits, HbA1c, and SBP. Consistent with our results, the ADVANCE study reported that there was no association between baseline HDL-C and DR after adjustment for a wide range of potential confounders [
]. However, observational studies on whether HDL-C could affect different severities of DR are limited. Our study showed that HDL-C is an independent protective factor for the development of severe nonproliferative DR and proliferative DR, which was different from results in the previous MR analyses. Compared with earlier MR analyses with 1277 cases of severe DR [
], the recently released datasets from FinnGen had a sample size of eight times larger than the previous one. The previous MR study cannot exclude the possibility that causal relationships with more modest effect sizes exist [
], which may lead to inconsistencies in the results. APOA-I is the main functional protein accounting for approximately 70% of HDL-C, and plays a central role in reverse cholesterol transport [
Low serum apolipoprotein A1/B ratio is associated with proliferative diabetic retinopathy in type 2 diabetes.
Graefe's archive for clinical and experimental ophthalmology = Albrecht von Graefes Archiv fur klinische und experimentelle Ophthalmologie.2012; 250: 957-962
Although the mechanisms underlying the observed associations are still unknown, it is presumed that HDL-C plays an important role in antioxidative stress, which may be involved in the pathogenesis of severe nonproliferative DR and proliferative DR [
]. When the HDL-C concentration is insufficient, the oxides cannot be removed promptly, resulting in a pathological state in which cells produce toxic effects and increased apoptosis of retinal endothelial cells, pericytes and ganglion cells [
]. Therefore, we speculate that high HDL-C levels could protect retinal vascular cells to maintain the integrity of retinal vessels and decrease the risk of severe DR.
CETP is involved in the transfer of cholesteryl ester from HDL-C to other lipoproteins. In our study, there was no evidence of an association between genetically predicted CETP inhibitors and the risk of DR. Although no epidemiological study focused on the relationships of CETP inhibitors with the risk of DR in a European population was found, the CETP polymorphism was reported to have no association with DR in a Taiwanese population [
This study has several notable advantages. First, this is the first study to systematically investigate the association between lipid traits and DR with a large sample size by integrating univariate and multivariate MR analyses. Second, we used radial MR to exclude potential outliers and performed a wide variety of sensitivity analyses to validate the robustness of the results. Third, based on the evidence for the effect of HDL-C on severe nonproliferative DR and proliferative DR, we estimated the association of genetically proxied CETP inhibitors with outcomes by selecting significant pQTLs and eQTLs as instruments. Finally, our results were less prone to population stratification bias because we limited the population to individuals of European ancestry.
However, the present study has potential limitations. First, MR analyses evaluate the lifetime effects of genetically proxied lipid traits and CETP inhibitors on DR, which might differ from the short-term effects of clinical interventions. Second, because the data were limited to participants of European ancestry, the results may not apply to other populations. Third, for partly consistent evidence about the association between some lipid traits and DR or different severities of DR, further clinical trials are needed to confirm our findings. Fourth, we were not able to estimate the relationships in the liver or retina, as there are no significant eQTLs available for these tissues. Finally, residual confounders potentially existed in these analyses, despite using different methods to ensure the robustness of the results.
4.4 Conclusions
In summary, using genetic data, our results demonstrated that genetically determined HDL-C protects against severe nonproliferative DR and proliferative DR. However, the associations of CETP inhibitors (which can elevate HDL-C levels) with DR were not observed. Thus, further studies are required to confirm our findings.
Financial support
This research was supported by grants from the Beijing Municipal Health System Special Funds of High-Level Medical Personnel Construction [grant number 2022-3-042] and China Scholarship Council [grant number 202008110060]. The sponsors had no role in the study design, data collection, data analysis and interpretation, writing of the report of the decision to submit the article for publication.
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The study was based on summary statistics provided by the UK Biobank, FinnGen, and eQTLGen. We thank all investigators and consortium for sharing valuable summary data.
Appendix A. Supplementary data
The following are the Supplementary data to this article.
Association between plasma triglycerides and high-density lipoprotein cholesterol and microvascular kidney disease and retinopathy in type 2 diabetes mellitus: a global case-control study in 13 countries.
Evaluating the relationship between circulating lipoprotein lipids and apolipoproteins with risk of coronary heart disease: a multivariable Mendelian randomisation analysis.
Pharmacological inhibition of CETP (cholesteryl ester transfer protein) increases HDL (High-Density lipoprotein) that contains ApoC3 and other HDL subspecies associated with higher risk of coronary heart disease.
CETP (cholesteryl ester transfer protein) concentration: a genome-wide association study followed by mendelian randomization on coronary artery disease.
Circulation Genomic and precision medicine.2018; 11e002034
Low serum apolipoprotein A1/B ratio is associated with proliferative diabetic retinopathy in type 2 diabetes.
Graefe's archive for clinical and experimental ophthalmology = Albrecht von Graefes Archiv fur klinische und experimentelle Ophthalmologie.2012; 250: 957-962
00064-3&id=ga1.jpg){kind=link}
00064-3&id=gr1.jpg){kind=link}
00064-3&id=gr2.jpg){kind=link}
00064-3&id=gr3.jpg){kind=link}
00064-3&id=gr4.jpg){kind=link}