Replication Analysis for Severe Diabetic Retinopathy

Abstract

Purpose.: The purpose of this study is to attempt to replicate the top single nucleotide polymorphism (SNP) associations from a previous genome-wide association study (GWAS) for the sight-threatening complications of diabetic retinopathy in an independent cohort of diabetic subjects from the Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR).

Methods.: This study included 469 type 1 diabetic, Caucasian subjects from WESDR. Cases (n = 208) were defined by prior laser treatment for either proliferative diabetic retinopathy or diabetic macular edema. Controls (n = 261) were all other subjects in the cohort. Three hundred eighty-nine SNPs were tested for association using the Illumina GoldenGate custom array. A retinopathy-only subanalysis was conducted in 437 subjects by removing those with end-stage renal disease. Evaluation for association between cases and controls was conducted by using chi-square tests. A combined analysis incorporated the results from WESDR with the prior GWAS.

Results.: No associations were significant at a genome-wide level. The analysis did identify SNPs that can be pursued in future replication studies. The top association was at rs4865047, an intronic SNP, in the gene CEP135 (P value 2.06 × 10⁻⁵). The top association from the subanalysis was at rs1902491 (P value 2.81 × 10⁻⁵), a SNP that sits upstream of the gene NPY2R.

Conclusions.: This study nominates several novel genetic loci that may be associated with severe diabetic retinopathy. In order to confirm these findings, replication and extension in additional cohorts will be necessary as susceptibility alleles for diabetic retinopathy appear to be of modest effect.

Introduction

Diabetic retinopathy is the number one cause of irreversible vision loss in working age adults in the developed world. Over time almost all diabetic individuals will go on to develop diabetic retinopathy. ¹ Most of the visual morbidity and attendant health care expense due to diabetic retinopathy can be attributed to its two severe stages: diabetic macular edema (DME) and proliferative diabetic retinopathy (PDR). 

There is mounting evidence to support a large genetic contribution to these severe, sight-threatening manifestations of diabetic retinopathy. While epidemiologic studies indicate that the strongest predictors of retinopathy severity correlate with glycemic level and diabetes duration, ^2,3 several recent lines of inquiry now suggest that genetic factors may also predispose and mitigate retinopathy severity. Heritability estimates for the severe manifestations of diabetic retinopathy range from 25% to 50% or more. ^4,5  

While evidence exists that there is a significant genetic component to severe diabetic retinopathy, ^6–10 to date, few genes or pathways have reliably been associated with the development of this disorder. The results of genetic association studies have enjoyed only modest success in identifying genetic variation that may affect risk of retinopathy. ¹¹ These previous studies generally involved modest numbers of cases and controls or trios and have probably been uniformly underpowered to detect genes with modest effect. In addition, many examined only a small number of genes, albeit plausible candidates. ^12,13 Herein, we report a replication analysis of the largest genome-wide association study (GWAS) to date for severe diabetic retinopathy. 

Methods

Institutional Review Board (IRB) Approval

The samples were obtained from all subjects through an approved IRB protocol at the consenting institution. As all patient health information was de-identified prior to genotyping and analysis, the IRB at the University of Chicago declared this study to be non–human subjects research. 

Phenotypic Characterization

The present analysis was restricted to Caucasian type 1 diabetic (T1D) subjects, as the previous meta-analysis ¹¹ included only Caucasian T1D subjects. Severe retinopathy cases were identified on the basis of prior laser photocoagulation for either DME or PDR. Details are given in following sections. Controls were all remaining subjects in the cohort. No individuals were excluded on the basis of renal status. Rather, all individuals with the presence of nephropathy, as defined by the presence of end-stage renal disease (ESRD), were removed from a retinopathy-only subgroup analysis. The diagnosis of ESRD was based on a history of kidney transplantation and/or the use of dialysis. Subjects with ESRD were removed to reduce possible confounding given the close correlation between nephropathy and retinopathy and to facilitate assessment for genetic heterogeneity and pleiotropy ^14,15 (Table 1). 

Table 1.

 

WESDR Demographic Table

	WESDR Controls	WESDR Cases	WESDR P Value
Age	48.9 (9.3)	50.7 (9.1)	0.0326
Sex	51.7%	46.2%	0.2315
Duration of diabetes	34.7 (7.2)	37.2 (6.7)	1.45 × 10−4
HbA1c	9.2 (1.1)	9.8 (1.0)	7.33 × 10−12
ESRD	0.8% (2)	14.4% (30)	3.19 × 10−9
Blood pressure	121.2 (11.5)	126.9 (12.7)	2.24 × 10−7
BMI	25.4 (3.8)	25.8 (3.5)	8.50 × 10−2

 Age and duration is presented in years, with data taken from the final examination. Gender is expressed as the percentage of female subjects. HbA1c is expressed as a percentage of glycosylated hemoglobin. All individuals with ESRD had a history of hemodialysis or of kidney transplantation. Data for HbA1c, blood pressure, and BMI is an average of all study visits over the course of the 25-year study. Standard deviation indicated in parenthesis. In case of ESRD, parentheses indicate number of subjects with ESRD.

The GoKinD and EDIC Cohorts

Detailed information regarding ascertainment of the Genetics of Kidney in Diabetes (GoKinD) and Epidemiology of Diabetes Intervention and Control Trial (EDIC) cohorts has been previously published ¹¹ (see also Table 2). 

Table 2.

 

Cohort Summary

	GoKinD Cohort	EDIC Cohort	WESDR Cohort
Diabetes mellitus type	Type 1	Type 1	Type 1
Ethnicity	Caucasian	Caucasian	Caucasian
No. of samples	1830	1441	469
No. of cases	815	158	208
No. of controls	803	1053	261
No. with ESRD	672	15	32
Basis for retinopathy phenotype	Self-reported laser treatment for diabetes	7-standard field stereoscopic retinal photos graded according to Airlie House Classification system and ETDRS grading scale	7-standard field stereoscopic retinal photos graded according to Airlie House Classification system and ETDRS grading scale
Follow-up	None	16 y	25 y
Platform	Affymetrix Genome-Wide Human SNP Array 5.0	Illumina 1M chip	Illumina 1536-SNP GoldenGate custom array

 ESRD was used to define presence of nephropathy in subjects.

The Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR) Cohort

The WESDR was conducted to assess the prevalence, incidence, and progression of diabetic retinopathy and other ocular and systemic complications of diabetes and their relationship to level of glycemia and other risk factors. ² For the study 996 individuals were examined at the baseline evaluation (1980–1982) who had insulin-dependent diabetes (primarily of type 1) and were less than 30 years of age at the time of diagnosis. Almost all of these individuals were Caucasian (Supplementary Fig. 1; supplementary data). DNA samples for this cohort have been obtained. The WESDR cohort in many ways ideally complements the GoKinD and EDIC samples given its composition of Caucasian T1D subjects followed prospectively with careful phenotypic characterization for renal, ophthalmic, and related endpoints. 

Detailed clinical and epidemiologic data were collected on these individuals approximately every 5 years over a 25-year period, including that collected at the fifth follow-up exam (2005–2007). The pertinent parts of the ocular and physical examinations included measuring weight, height, blood pressure, and intraocular pressure; dilating the pupils; taking 30° stereoscopic color film fundus photographs of seven standard fields; performing a semiquantitative determination of glucose, ketone, and protein levels in the urine using Labstix (Ames, Elkhart, IN); and determining blood glucose and glycosylated hemoglobin A1 (HbA1c) levels from a capillary blood sample. A structured interview was conducted by the examiners, including questions about specific medications for control of hyperglycemia and blood pressure and use of diuretic agents, the number of aspirin used during the 30 days before the baseline examination, and smoking history. The variables available for selection in all the models for the whole cohort were age, duration of diabetes, sex, glycosylated hemoglobin, change in glycosylated hemoglobin, systolic and diastolic blood pressure, change in systolic and diastolic blood pressure, hypertension, gross proteinuria, and severity of retinopathy at baseline. ^16,17  

Data were available on 505 of these subjects for retinopathy severity, 469 of whom had usable DNA samples for this study. There were 194 participants who had proliferative diabetic retinopathy at any time during the 25-year period, and 97 who had diabetic macular edema at any time during the 25-year period (based on grading using the modified Airlie House Classification system, an Early Treatment of Diabetic Retinopathy Study [ETDRS] retinopathy severity scale ¹⁸ ). This left 261 who never had signs of PDR or DME at any time. For the purposes of this study, all subjects with either PDR or DME were considered cases, and 39.9% of subjects had both PDR and DME. All other subjects in the WESDR cohort, including those with nonproliferative diabetic retinopathy but without DME, were used as controls. In all, there were 208 cases and 261 controls (Table 2). In the subanalysis, subjects were excluded who had a history of ESRD as defined by prior hemodialysis for renal failure or kidney transplant. In all, 32 subjects in WESDR were excluded for ESRD, with the remaining cases being used in the subanalysis. 

Single Nucleotide Polymorphism (SNP) Selection

SNP selection was done using single marker association analyses in typed and untyped (i.e., imputed) SNPs from the GoKinD and EDIC datasets. The follow-up study was designed based on a preliminary analysis prior to imputation of the top associations of the GoKinD-EDIC genome wide meta-analysis. ¹¹  

The 389 SNPs for the follow-up study were selected from three lists comprised of the top preliminary associations from GoKinD, EDIC, and the combined analysis (Supplementary Table 1). The joint GoKinD-EDIC associations were assessed by combining the P values from each of the single studies using a simple modification of Fisher's method that weighed the log P values according to the information available in the individual datasets. Out of the 389 SNPs approximately 180 came from GoKinD (the P value threshold corresponding to this was 2 × 10⁻⁴), 120 from EDIC (the P value threshold corresponding to this was 2 × 10⁻⁴) and the remainder from the combined analysis (these were selected such that they were different than those in the GoKinD and EDIC lists). For each SNP associated with GoKinD at the level of 2 × 10⁻⁴, a region within 50 kb on either side of the SNP was defined. SNPs that were associated in EDIC at the 0.05 significance level within this region were selected for replication. The same process was used for EDIC associated regions. Markers showing an r ² value larger than 0.8 with any other marker in the set were removed from the list. 

Genotyping and Quality Control

Genotyping for WESDR subjects was performed on the Illumina 1,536-SNP GoldenGate custom array. The platform has probe-sets for 1536 SNPs, and ancestry informative markers were added to the SNP list for an accurate evaluation of population stratification. The genotyping was done at the Center for Inherited Disease Research (CIDR) and the SNP genotypes were called using Illumina software. ¹⁹ The number of SNPs with high quality genotyping scores was 1432. Three hundred eighty-nine of these SNPs were used for the retinopathy analysis. The remaining SNPs were ancestry informative markers or were used for quality control or to evaluate other phenotypes in WESDR not part of this study, including hypertension, obesity, HbA1c, and nephropathy. 

Extensive QC analyses were performed on the WESDR data including detecting relatedness (with three duplicated samples removed), sample contamination, and deviations from the self-reported ethnicity (three samples showing African ancestry were removed, see also Supplementary Figure 1). We performed a detailed analysis of the data to untangle any population structure within the samples. The vast majority of the subjects in the WESDR cohort are of individuals reporting European ancestry, but we utilized the genotype data to determine incorrect self-reported ancestry. Population stratification was investigated with EIGENSTRAT. ²⁰ EIGENSTRAT was run on the joint HapMap and subject samples and on the subject samples separately. The analysis yielded the principal components that were used for identifying outliers, which were excluded from subsequent analyses based on ancestry (Supplementary Figure 1). The total number of samples remaining was 469. SNPs were filtered using call rate (five SNPs with call rates smaller than 95% were removed) and the test for Hardy-Weinberg equilibrium (two SNPs with Hardy-Weinberg equilibrium p < 0.001 were removed). The quality control methods used for GoKinD and EDIC have been previously published. ¹¹  

Statistical Analyses

Association testing for severe diabetic retinopathy was conducted by comparing allele frequency between cases and controls using chi-square tests. The allelic association tests were done in two groups: (1) including subjects with ESRD and (2) excluding subjects with ESRD. The conventionally accepted threshold of 5 × 10⁻⁸ was used as the correction for multiple testing to allow for all common variants in the genome. ²¹ For each SNP, in each dataset, an odds ratio (OR) was calculated. The OR preserves the direction of the effect. A normal approximation was then used to calculate the combined analysis P values. All combined analysis P values for severe retinopathy are reported in either Table 3 or Supplementary Table 1. 

Table 3.

 

Severe Retinopathy GWAS Combined Analysis Top Associations

ESRD Subjects	rs#	Chromosome	Genomic Location	Alleles (Reference First)	Gene	Left gene 3′->5′	Right gene 3′->5′	WESDR OR	WESDR P Value	GoKinD OR	GoKinD P Value	EDIC OR	EDIC P Value	Combined Analysis P Value
Included	rs4865047	4	56,516,563	C, T	CEP135	LOC644173	LOC100130647	0.65	0.11	0.87	0.26	0.44	3.63 × 10−6	2.06 × 10−5
	rs476141	1	242,243,047	G, T	NA	AKT3	ZNF238	0.95	0.70	1.27	0.0015	1.72	1.59 × 10−5	4.04 × 10−5
Excluded	rs1902491	4	156,274,783	C, A	NA	LOC729902	NPY2R	0.81	0.13	0.62	0.0003	0.75	0.034	2.81 × 10−5

 Top association signals from combined analysis (P value ≤ 10⁻⁴) for severe retinopathy. Information for the SNP, genomic location, associated P values, and OR are provided. Genomic location is based on genome build 36.1. Correction for multiple testing is P value < 5 × 10⁻⁸.

Results

We analyzed the WESDR cohort for correlations between study variables and the presence of severe retinopathy (Table 1). As anticipated, duration of diabetes (P = 1.45 × 10⁻⁴), renal status (P = 3.19 × 10⁻⁹), age (P = 0.0326), blood pressure (P = 2.24 × 10⁻⁷), and HbA1c (P = 7.33 × 10⁻¹²) all significantly correlated with the risk for severe retinopathy. The subjects with severe retinopathy had a longer duration of diabetes (mean duration 37.1 vs. 34.7 years). The values for HbA1c represent the average over all exams during the 25-year study period for each subject. 

The analysis revealed three SNPs with a P value < 10⁻⁴. The strongest signals were at a P value of 10⁻⁵, falling below the conventional genome-wide significance threshold of 5 × 10⁻⁸. The top association was found at rs4865047 (P value 2.06 × 10⁻⁵) on Chromosome 4 at 56.5 Mb in an intron of the gene CEP135 (encoding centrosomal protein 135 kDa). Within CEP135, rs4865047 is in high linkage disequilibrium (LD) with a number of SNPs in intronic regions (Supplementary Fig. 2), but no exonic SNPs or eQTLs. The second strongest association, rs476141, on Chromosome 1 at 242 Mb represented the most significant association in the prior meta-analysis of GoKinD and EDIC. ¹¹ Its P value (4.04 × 10⁻⁵ vs. 1.2 × 10⁻⁷) did not improve with the addition of the WESDR cohort in this analysis and the OR was in the opposite direction of GoKinD and EDIC. rs476141 lies in a region between two genes, AKT3 and ZNF238. AKT3 is an intriguing candidate gene because it is a kinase implicated in insulin signaling and angiogenesis. ²² The other main finding from the previous study, rs10521145, was genotyped in WESDR separately, but failed to replicate (P = 0.631). 

rs1902491 was the strongest association from the retinopathy only subgroup analysis that excluded subjects with ESRD. rs1902491 (P value 2.81 × 10⁻⁵) is on Chromosome 4 and lies approximately 75 kb upstream of the gene NPY2R. It is in high LD with the SNP rs9790645 that lies 48 kb upstream of the gene NPY2R (r ² = 0.964). 

Discussion

In this study the top 389 preliminary genetic associations identified in a genome-wide combined analysis for severe diabetic retinopathy in Caucasian T1D subjects were tested in a separate, independent cohort from the WESDR. Findings were incorporated in a combined analysis of the WESDR, GoKinD, and EDIC cohorts. No associations were at the level of genome-wide significance. 

The strongest association for severe diabetic retinopathy generated by this combined analysis was at rs4865047 in CEP135. CEP135 is expressed in the retina and is involved in centriole adhesion and replication. It interacts with SMAD9, which is involved in TGF-β signaling, a pathway that is up-regulated in diabetes. ²³ Interestingly, rs4865047 is also 50.5 kb downstream of the gene EXOC1 (exocyst complex component 1). The exocyst complex is required for the insulin stimulated transport of Glut4 to the plasma membrane. ²⁴  

rs476141 was the strongest association from our prior meta-analysis of GoKinD and EDIC. There are several reasons that could explain why its significance did not improve with the addition of the WESDR cohort in this study. First, the association between severe diabetic retinopathy and rs476141 could be a false positive or spurious signal. Second, even though all cohorts are composed of Caucasian subjects, the LD structure of subjects could differ between the three groups of patients. Third, when the P values for each cohort are analyzed, the signal for rs476141 was primarily generated by the EDIC cohort. EDIC by its design was composed of far greater numbers of subjects with minimal to no retinopathy, creating a substantial difference in the control group composition compared to that of the WESDR and GoKinD cohorts. The signal for rs476141 could be driven by this “squeaky-clean” control group. 

In the subanalysis without ESRD subjects, an association was identified with an intergenic SNP, rs1902491, that lies upstream of the gene NPY2R. Neuropeptide Y Receptor subtype 2 (NPY2R) is a G protein coupled receptor for NPY, a neurotransmitter released by endothelial cells implicated in ischemic angiogenesis. ²⁵ There is substantial genetic, biologic, and functional data supporting a role for neuropeptide Y signaling in diabetic retinopathy. ^19–22  

A significant advantage of this study design is the comparability between cases and controls between the three cohorts. We have made a very concerted effort to harmonize the retinopathy phenotype and risk factors between the three cohorts. Three separate, large cohorts of Caucasian T1D subjects of northern European ancestry should help to minimize heterogeneity, exposure to environmental differences, and population substructure. 

Since cases were defined as those individuals who had received treatment for PDR or DME, it is possible that there was a small amount of misclassification because some individuals may have been graded as having PDR or DME, but did not have treatment. It is also possible based on the study design that the case–control groups are not distinct using the phenotypic parameters of DME and PDR. If there was a systematic pattern to this misclassification, it would simply blunt the study's ability to identify an associated allele for severe diabetic retinopathy. It would not introduce a false-positive association. A similar blunting of the power of the study may result from the possibility of a difference in underlying pathophysiology of PDR and DME, though this, again, would not create a type I error. 

From a clinical perspective the phenotype of diabetic retinopathy and its characteristic ophthalmoscopic appearance are essentially identical regardless of ethnicity and diabetes type. We anticipate that there may be several genetic variants that may transcend both ethnicity as well as diabetes type. The most important genetic variants may be those that are found across populations, which if present would certainly increase confidence in any positive association. Subsequent analysis of ethnically diverse type 2 diabetes cohorts will be required to evaluate candidate genes in depth and their associated common and uncommon variants that may be markers for the retinopathy phenotype in order to substantiate this hypothesis. 

The present study represents one of the largest reported association analyses for diabetic retinopathy to date. Effect size for most significant associations of complex diseases to date has been on the order of a genetic relative risk (GRR) of 1.2. ²⁶ In comparison this study had limited power to detect an allele whose GRR is less than 1.5 unless the risk allele was fairly common in the population. A major lesson of the Wellcome Trust Case Control Consortium (WTCCC) is that not only are large scale genetic association studies feasible and that humans may be the best model organism for the study of complex disease, but also that these studies should in the future be performed on an even a larger scale to increase the analytical power of detection of even more genetic variants. ^27,28 A greater number of subjects may also be necessary to further refine the regions associated with diabetic retinopathy than are currently available from these samples. 

Supplementary Materials

PDF S1 

PDF S2 

Suppl Table 1 

Acknowledgments

The authors are indebted to Graeme Bell, University of Chicago, for many helpful discussions; to the lab of Andrew Paterson for graciously sharing genotyping and statistical analyses for the WESDR cohort; and to two anonymous reviewers for their critical input. 

Footnotes

 Supported by the following grants: NIH K08 EY019089-02, NIH R01 DK077489, NIH National Center for Research Resources #UL1RR024999, Diabetes Research and Training Center P60 DK020595-32, Illinois Society for the Prevention of Blindness, Fight for Sight, OneSight, the Louis Block fund, and GAIN (Genetic Association Information Network). This study was also supported by the National Eye Institute Core Grant EY001792 for Vision Research and an unrestricted grant from Research to Prevent Blindness, Inc., New York, NY; and by Senior Scientific Investigator Awards to Barbara E. Klein and Ron Klein, awarded by Research to Prevent Blindness. The Center for Inherited Disease Research (CIDR) is funded through a federal contract from the National Institutes of Health to The Johns Hopkins University, contract number HHSN268200782096C. The Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR) has been supported by the National Institutes of Health, Bethesda, Maryland (grant nos.: EY03083 and EY016379 [RK, BEKK]); and, in part, by Research to Prevent Blindness, Inc., New York, New York (RK and BEKK, Senior Scientific Investigator Awards). Genotyping services were provided by the Center for Inherited Disease Research (CIDR). The Genetics of Kidneys in Diabetes (GoKinD) Study was conducted by the GoKinD Investigators and supported by the Juvenile Diabetes Research Foundation, the CDC, and the Special Statutory Funding Program for Type 1 Diabetes Research administered by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). This manuscript was not prepared in collaboration with Investigators of the GoKinD study and does not necessarily reflect the opinions or views of the GoKinD study or the NIDDK.

Footnotes

 Disclosure: M.A. Grassi, None; A. Tikhomirov, None; S. Ramalingam, None; K.E. Lee, None; S. M. Hosseini, None; B.E.K. Klein, None; R. Klein, None; Y.A. Lussier, None; N.J. Cox, None; D.L. Nicolae, None