Diabetic retinopathy (DR) is the second leading cause of blindness due to retinal degeneration, contributing to an overall 4.8% blindness across the globe 1 . Clinically, diabetic retinopathy is characterized as basement membrane (BM) thickening, pericyte loss, endothelial cell (EC) dysfunction, microaneurysms, microvascular infarcts, and neovascularization in a patient with diabetic retinopathy 2 3 . Localized hypoxia has been implicated in this process, which can eventually lead to blindness. Although good glycemic control is considered to be a mainstay in preventing the vision loss, some patients with poor glycemic control do not develop retinopathy. This is well documented in the Diabetes Control and Complications Study (DCCT) and the United Kingdom Prospective Diabetes Study (UKPDS) 4 5 .

Several bio-chemical pathways have been proposed to mediate the pathogenesis of DR. These include polyol accumulation, formation of advanced glycation end products (AGEs), oxidative stress, activation of protein kinase C, upregulation of matrix metalloproteinases, elaboration of growth factors and secretion of adhesion molecules 6 7 . Genetic as well as environmental factors may play a crucial role in understanding the differential susceptibility to DR. To date, markers at four chromosomal loci have been linked with DR. By multipoint sib-pair analysis, loci on chromosomes 3 and 9 showed suggestive evidence for linkage to DR in American Pima Indians with type 2 diabetes yielding lod scores of 1.36 and 1.46 respectively 8 . Two other recent studies also suggested loci on chromosomes 1 and 3, and 12 with suggestive support for a number of additional loci for DR in Mexican Americans 9 10 .

DR shows a complex inheritance pattern, and genetic association studies are often used to identify the genetic factors influencing its development. In this context, polymorphic markers in or near several genes including aldose reductase (AKR1B1), the receptor for AGE (advanced glycation end products, RAGE), vascular endothelial growth factor (VEGF), intercellular adhesion molecule 1 (ICAM-1), pigment epithelium derived factor (PEDF), hemochromatosis (HFE), and alpha2beta1 integrin have been identified as being associated with DR 11 12 13 14 15 16 17 . However, with the exception of VEGF, eNOS (endothelial nitric oxide synthase), IGF (insulin like growth factor), AKR1B1 and RAGE, detailed investigations of most of these associations in the Indian population have not been carried out 13 18 19 21 23 .

Therefore, the present study aims to evaluate SNPs in nine candidate genes previously reported to be associated with DR, including the receptor for advanced glycation end products (RAGE), pigment epithelium derived factor (PEDF or SERPINF1), aldose reductase (AKR1B1), erythropoietin (EPO), intercellular adhesion molecule 1 (ICAM-1), hemochromatosis (HFE), high temperature requirement A1 or serine peptidase (HTRA1), Complement Factor H (CFH) and age related maculopathy susceptibility 2 (ARMS2) to determine whether alleles of these markers are associated with DR in an Indian population.

This study protocol was approved by both the Institutional Review board and Ethics Committee of Aravind Eye Hospital, Madurai, Tamil Nadu, India and Institutional Review board of the National Eye Institute (NEI), NIH, Bethesda, Maryland. Patients were recruited from the Aravind Eye Hospital and its clinics and the laboratory part of the study was carried out at the NEI in accordance with the tenets of Declaration of Helsinki.

The clinical characteristics of the study subjects are shown in Table 1. In this study, patients with PDR as defined by eyes with definite neovascularization and/or vitreous/preretinal hemorrhages and more severe hemorrhages were selected for the DR study group and patients with no signs of DR coupled with type 2 diabetes of at least 12 years duration were selected for diabetic without retinopathy (DNR) study group. The criteria for diagnosis of diabetes were random blood sugars over 150 mgs% and HbA1C over 6%. The clinical characteristics such as random blood sugars, HbA1C, blood pressure, and duration of diabetes were essentially similar in both groups, although the DR and DNR exploratory groups both had somewhat longer duration of diabetes in the report by Kumaramanickavel et al. 21 than the confirmatory groups (17 and 20 vs. 14 and 14 years, respectively). Of the clinical characteristic recorded only percentage males (P = 0.05) and blood urea nitrogen (P = 0.002) were significantly associated with DR. None of the clinical characteristics other than DR were associated with any tested genotypes.

Among the 15 SNPs studied, only one, rs2070600 (G>A) in exon 3 of RAGE, which leads to a change of amino acid glycine to serine at codon 82 displays significant association with DR in the initial screening group (Table 2). The G allele (P = 0.016), the homozygous GG (P = 0.012), and the heterozygous GA (P-0.011) genotypes are significantly associated with DR. Individuals with the homozygous GG genotype of this SNP have 1.78 times higher risk of developing DR (OR = 1.78: 95% CI-1.13 - 2.82). A dominant model also shows association with DR (P = 0.012, Table 3), with the A allele in either heterozygous of homozygous state giving an odds ratio (OR) = 0.56 with 95% CI = 0.35 - 0.89. In contrast, the other RAGE gene polymorphisms tested, -374 T/A and -429 T/C, do not show any significant association in this population. No other SNPs previously reported to show association could be confirmed in this population, with the possible exceptions of rs11200638 (G>A) in HTRA1 and the LOC387715 SNP rs10490924, which show marginally significant associations with DR (allelic P = 0.055 and P = 0.07, respectively).

When association of rs2070600 is tested in the confirmatory group the association is confirmed. Although the allelic association test gives only a suggestive P = 0.068 (A allele OR = 0.54: 95% CI = 0.38 - 0.76), the whole SNP P = 0.035 and a dominant model for the A allele gives a P = 0.032 with an OR = 0.34 (95% CI = 0.13 - 0.94). Combining the two groups for analysis continues to show significant association of rs2070600 with DR (allelic P < 0.003). Similarly, the genotypic test (P = 0.003), the dominant model (P = 0.0013, OR = 0.51, 95% CI = 0.34 - 0.82) and the additive model (P = 0.003) all show significant association (Table 4).

The RAGE gene was selected as a candidate based on previous analysis showing association in an Indian population by Kumaramanickavel et al 21 . Because of the ethnic and clinical similarities between the ARAVIND and Sankara Nethralaya study populations (Table 1) and because they represented the only two studies examining RAGE gene polymorphisms in the Indian population, we decided to combine data from the two studies in a single analysis. It was noted that our results for the RAGE gene were remarkably similar to those reported by Kumaramanickavel et al. 21 with major allele frequencies of .95 vs. .95 and .92 vs. .89 for the DR and DNR groups, respectively. The combined results once more confirmed association of the G allele with DR: allelic P < 0.0003 with the A allele OR = 0.54 (0.38 - 0.76). The whole SNP association is also significant with a P = 0.0003, as are the results of analysis under specific models with the A allele dominant P = 0.00011 with an OR = 0.49 (95% CI = 0.34 - 0.7) and the additive model P = 0.00047 (Table 4).

None of the remaining SNPs, including, four SNPs rs1136287 (C/T), rs58697961 (G/A), rs12150053 (T/C) and rs12948385 (G/A) in the PEDF gene, rs1617640 (G/T) in EPO, rs1800562 (G/A) in HFE, rs5498 (A/G) in ICAM, rs1061170 and rs3753394 in CFH, rs10490924 in ARMS2/LOC387715 and rs35839483 (CA)n in AKR1B1 showed significant association with DR. The SNP rs1800562 (G/A) in HFE was found not to be polymorphic in the study groups. Additionally, 12 alleles 138 bps-162 bps in rs35839483 (CA)n were identified. The allele frequencies of the AKR1B1 (CA)n dinucleotide repeat polymorphism are given in Table 5. None of the alleles show significant association with DR in this population.

The present study evaluates the potential association of fifteen polymorphisms in nine previously identified DR candidate genes with type 2 diabetes patients in a South Indian population. In this case-controlled study only one SNP, rs2070600 in the RAGE gene, could be shown to be significantly associated with retinopathy in the presence of DR. Because of the relatively small number of patients and controls analyzed, this study cannot necessarily exclude the possibility of alleles of the remaining SNPs showing some association with DR, especially those in HTRA1 and LOC387715. However, it is unlikely that the level of association would be greater than the 95% confidence limits of the odds ratios shown in Table 3, which are generally less than 1.5 for allelic association with the exception of the RAGE related SNPs, and less than 2-2.5 for the genotype odds ratios. Finally, it is possible that associations seen in other populations might not be present in southern Indians because of different population histories, which could alter haplotype block structure and thus non-causative allelic association, and possibly differing genetic and environmental contributions to the disease risk in various ethnic and geographic groups. In addition, this study did not address the use of oral hypoglycemic agents on the development of DR directly. However, since the main effects of these agents would be expected to result from their enhancement of glycemic control, to some extent one might expect their effects to be reflected in the fasting blood glucose and HbA1C levels, which were similar in the DR and DNR subjects.

RAGE gene polymorphisms are attractive candidates to influence DR because of pathophysiological data correlating retinopathy and advanced glycation end products (AGEs). RAGE, which maps to chromosome 6p21.3, is a member of the immunoglobulin super family. AGEs result from the non-enzymatic glycation of proteins and lipids. They are found at increased levels in diabetes and can lead to increased oxidative stress and receptor-mediated activation and secretion of various cytokines 24 25 . Accumulation of AGEs has been suggested to contribute to vasculopathy by increasing retinal endothelial cell permeability 26 . Specific AGE-binding receptors include the AGE-receptor complex, the macrophage scavenger receptors, and the receptor for AGE (RAGE). A Gly82Ser polymorphism in RAGE is potentially interesting since it occurs at a predicted N-linked glycosylation motif in the AGE binding site, thereby influencing AGE-RAGE interactions 12 . Kumaramanickavel et al. 21 previously reported that this polymorphism is associated with decreased risk for the development of DR in a Southern Indian population essentially similar to the one studied here. However, additional studies investigating the same locus did not show significant association for this polymorphism in Chinese and Japanese populations 20 27 , and a recent study showed association of the A allele (serine) with DR in a Han Chinese population 28 . Association of the opposite allele of a polymorphism with disease has been reported before, and an additional example with an elegant theoretical analysis of possible causes for this has been presented by Lin et al. 29 30 . As also suggested by Zhang et al. 28 in their report, this might represent either different molecular pathogenesis of the disease in the two populations or simply a different population history resulting in different haplotype structure of the associated region. A particularly interesting example is that of the LOXL1 polymorphisms rs3825942 and rs1048661, which are in linkage disequilibrium and both show association with exfoliative glaucoma (XFG) 31 32 . While the G allele of rs3825942 is associated with XFG in both Caucasian and Japanese populations, the G allele of rs1048661 (GG haplotype of rs1048661 and rs3825942) is associated with XFG in Caucasians but the T allele (TG haplotype of rs1048661 and rs3825942) is associated with XFG in Japanese.

Our study supports association of the 82Gly allele and DR (P = 0.0003) with an odds ratio of 0.54 (95% confidence interval = 0.38 - 0.76) for the 82Ser (A) allele in a southern Indian population. Although the AA genotype appears to have an increased OR in the exploratory cohort (1.04, Table 2), this is due to the presence of a single AA individual in both the DR and DNR groups and probably represents a chance variation. Overall, the risk for DR decreases with the number of 82Ser (A) alleles, with both the additive model and dominant models supporting association. This supports the original findings of Kumaramanickavel et al. that 82Ser is a protective allele for diabetic retinopathy in the Indian population, and the similarities between the study populations in terms of ethnic origin and clinical criteria suggested that results from the subjects of both studies might be combined in a single analysis 21 33 . The individuals in the study by Kumaramanickavel et al. 21 were slightly older (63 years vs. 57 years average), had a slightly longer history of diabetes (18 years vs. 14 years average) and had somewhat higher HbA1C values (8.5 vs. 6.4 average). However, the similarities in the diagnostic and exclusion criteria and particularly the Indian ethnic composition of both groups allowed a combined analysis. The combined analysis was consistent with the results of subjects recruited from both hospitals, giving a similar odds ratios for the G allele (1.8 vs. 1.7), in the combined and ARAVIND patients respectively. The combined results, however, showed increased significance, with improvements in both the allelic (0.0003 vs. 0.003) whole SNP genotypic (0.0003 vs. 0.003) dominant (0.00011 vs 0.0013) and additive (0.00047 vs. 0.003) p-values in the combined and Aravind groups respectively. While we have not included a Bonferroni correction directly because of uncertainty about the a priori likelihood of each locus being true, a Bonferroni correction for multiple testing considering tests for 15 markers were performed might be a P < 0.0033, which is met in the combined ARAVIND and the final combined data sets.

In contrast to the results with rs2070600, neither the -374 T/A SNP nor the -429 T/C SNP in RAGE were found to be associated with DR, even though this association had been reported in the British population 34 . This might reflect population differences in the pathogenesis of DR in the Indian and other populations, but might also be a reflection of the relatively low minor allele frequencies in the Indian population, as the 95% confidence intervals for these SNPs, and especially -374 T/A, are very wide.

A putative transcription factor binding site of the HTRA1 gene in the 10q26 region was shown to confer significant risk for choroidal neovascularization in age related macular degeneration (AMD) and SNPs rs1061170, rs3753394 in CFH and rs10490924 in ARMS2/LOC387715 showed significant association 35 36 37 38 39 40 . Because neovascularization is common to both AMD and DR, we hypothesized that these genes may play a role in DR. To the best of our knowledge, this is the first report correlating the association of these SNPs with DR (however, our results indicate only marginal significance of the SNP rs11200638 in HTRA1 gene with DR).

We have attempted to avoid population substructure in our study. However, it remains possible that the positive and negative results could potentially be due to subtle population stratification and should be taken as suggestive rather than definitive.

In conclusion, we are able to confirm association of the G82S polymorphism of RAGE with diabetic retinopathy in the population of southern India. We studied the association of SNPs rs11200638 in the HTRA1 gene, rs1061170, rs3753394 in CFH and rs10490924 in ARMS2/LOC387715 with DR for the first time. Our findings suggest that these three polymorphisms may not be associated with DR in the southern Indian population. Our data also strongly suggest that patients with the serine allele of the Gly82Ser (rs2070600) polymorphism in the RAGE gene show a decreased risk for diabetic retinopathy in the Indian population, although association of this allele with DR in a Chinese population suggests that this might represent linkage disequilibrium with a nearby causative rather than implying that glycation at this site plays a directly causative role in the development of diabetic retinopathy.

DR: Diabetic Retinopathy; DM: Diabetes Mellitus; T2DM: Type 2 Diabetes Mellitus; DNR: Diabetic patients without retinopathy; SNPs: Single Nucleotide Polymorphisms; RAGE: Receptor for Advanced Glycation End Products; PEDF: Pigment Epithelium Derived Factor; AKR1B1: Aldose Reductase; CFH: Complement Factor H; ARMS2: Age Related Maculopathy Susceptibility 2; EPO: Erythropoietin; HTRA1: High Temperature Requirement A1; ICAM: Intercellular Adhesion Molecule1; HFE: Hemochromatosis; OR: Odds Ratio.

The authors declare that they have no competing interests.

SB performed the sample collection, genotyping, statistical analysis, interpreted the data and drafted the manuscript. PS was involved in the study design and critically reading the manuscript. AR, KR and NP recruited and clinically characterized study subjects. GG was involved in genotyping of newly recruited samples. JFH was involved in the study design, data analysis, interpretation of the data, and preparation of the final version of the manuscript. All authors read and approved the final manuscript.