Multiple Superoxide Dismutase 1/Splicing Factor Serine Alanine 15 Variants Are Associated With the Development and Progression of Diabetic Nephropathy: The Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications Genetics Study

The DCCT was a multicenter randomized clinical trial to compare intensive and conventional insulin therapy on the development and progression of early vascular and neurological complications of type 1 diabetes (22,23). We studied 1,369 white probands (out of 1,441 DCCT subjects) and 2,966 of their first-degree relatives (details are provided in online appendix 26 and the online appendix text [available at http://dx.doi.org/10.2337/db07-1059]). To confirm findings from the DCCT/EDIC probands, we identified three other studies of diabetic nephropathy in type 1 diabetes (Table 5); details of the characteristics are in the online appendix text.

Outcomes were defined as time in years from DCCT baseline until the event (detailed in the online appendix text). Renal outcomes included data up to EDIC year 8 (2001): Persistent microalbuminuria was defined as the time to two consecutive albumin excretion rates (AERs) >30 mg/24 h (>20.8 μg/min); severe nephropathy was the time to AER >300 mg/24 h (>208 μg/min) with prior persistent microalbuminuria or ESRD. For retinal outcomes, data up to EDIC year 10 (2003) were used: ≥3 step progression was defined as the time to ≥3 step progression from DCCT baseline using Early Treatment of Diabetic Retinopathy Study (ETDRS) scoring; severe retinopathy was the time to ETDRS level 53/<53 or greater or scatter laser treatment; clinically significant macular edema (CSME) was the time to macular edema or focal laser treatment.

Candidate genes were selected based on their involvement in the pathophysiology of diabetic nephropathy and/or diabetic retinopathy. We considered biological evidence, including gene function, evidence for differential expression in diabetes complications, prior genetic association studies of diabetes complications and their risk factors, and genes in pathways related to these genes. For genes sequenced by SeattleSNPs (24), tagSNPs were selected using LDselect (24), requiring pairwise r² cutoff >0.64 and minor allele frequency (MAF) >0.05 in either the European American or PDR90 samples (25) (http://egp.gs.washington.edu). Otherwise, we used Haploview pairwise Tagger (26) to select SNPs with an r² cutoff >0.8 in the 30 CEU trios (CEPH samples of European descent) from release 16 (phase I) of the HapMap project (http://www.hapmap.org), including 5 kb flanking either side of the gene (27,28). Selected SNPs underwent bioinformatic evaluation to score their suitability for genotyping via the Illumina BeadArray GoldenGate custom assay (29). SNPs that did not pass criteria were replaced by other tagSNPs when possible or were excluded. In total, ∼2,750 SNPs were evaluated bioinformatically, and 1,536 were genotyped in all DCCT/EDIC probands with available DNA (n = 1,418). Additional SNPs were genotyped using Taqman technology (see online appendix text).

To fine-map the causal variant(s), we sequenced ∼19 kb of the superoxide dismutase 1 (SOD1) genomic region in all eight DCCT/EDIC probands homozygous for the rare genotype (C/C) at rs17880135. The region encompassed SOD1 and ∼5 kb at both the 5′ and 3′. In addition, we sequenced the cDNA of splicing factor serine alanine 15 (SFRS15, the gene 3′ of SOD1) in seven individuals, selected in a way similar to that of the samples for SOD1 sequencing (online appendix text and online appendix 18).

We quantified SOD1 mRNA expression, serum SOD activity, and SOD1 mass in a subset of probands and nondiabetic relatives selected by rs17880135 genotype. SOD1 mRNA expression was measured in Epstein-Barr virus–transformed lymphocytes from 18 individuals homozygous for the rare allele (8 probands and 10 nondiabetic relatives), 38 heterozygotes (16 probands and 22 nondiabetic relatives), and 38 homozygotes (16 probands and 22 nondiabetic relatives) for the common allele at rs17880135 with matching on relevant covariates (online appendix text). To determine whether serum SOD activity and SOD1 mass differed by rs17880135 genotype, we proceeded in a similar fashion (online appendix text).

SNP allele and genotype frequencies were calculated and deviation from Hardy-Weinberg equilibrium (HWE) was determined using exact tests (online appendix 3). Details about the association analysis for SNPs with DCCT/EDIC outcomes are provided in the online appendix text (17,30). Briefly, SNP genotypes were tested for association with DCCT baseline design and biological covariates (online appendix 4). Cox proportional hazards analysis of discrete time-to-event outcomes was performed using two models with additive genotype coding (31). The first model was “univariate,” including DCCT cohort, treatment, and cohort-treatment interaction as covariates, and stratified by the year of entry into DCCT. Our primary analysis was a multivariate Cox model, including cohort, treatment, cohort-treatment interaction, age of diagnosis, DCCT baseline duration, sex, BMI, mean blood pressure, triglyceride, HDL-C, total cholesterol, baseline smoking, A1C eligibility, time-dependent updated mean A1C, and time-dependent indicators for hypertension diagnosis and/or treatment of hypertension, with stratification by DCCT year of entry (online appendix text and online appendix 5). X chromosome SNPs were analyzed separately in men and women (online appendixes 6–9). SNPs with MAF <1% were analyzed separately (online appendixes 10–12) using a modified Cox model designed for sparse data (32,33), including cohort, treatment, cohort-treatment interaction, and sex as covariates.

To address the inherent problem of multiple hypotheses testing, we adopted the false discovery rate (FDR) framework and applied the q value method (34) (http://faculty.washington.edu/jstorey/qvalue/). The q value is a measure of statistical significance in terms of FDR, and a test with q value of 0.5 means that the expected FDR is 0.5 if we declare that test and all other tests with smaller q values significant. To use the available phenotype information, we applied the stratified FDR (35) (http://faculty.washington.edu/jstorey/qvalue/) designed to increase power; q values were hence calculated separately for each outcome.

To detect population structure, we used STRUCTURE v2.0 to analyze genotype data for SNPs chosen to have low pairwise linkage disequilibrium from the Illumina assay (36). We first included both white and nonwhite probands (n = 50) to identify white probands whose subgroup classification as determined by STRUCTURE disagreed with their self-reported ethnicity; then we analyzed the white individuals alone to identify subgroups within them. Using either 180 SNPs with the highest heterozygosity or 164 ancestry informative SNPs (37), there was no evidence for outliers, and the white probands could not be separated into subgroups. In addition, to further exclude that the association of SOD1 SNPs with persistent microalbuminuria and severe nephropathy was due solely to population stratification, we genotyped three SOD1 SNPs associated with renal outcomes (Table 3) in all available DCCT/EDIC relatives (online appendix 26) and performed family-based analysis (online appendix 25).

Martingale residuals (38) obtained from the multivariate Cox proportional hazards models without genotype, were used as continuous traits in subsequent family-based association analyses (FBAs; see online appendix 26 for family structures). FBA conditions on the parental genotypes were used to test for departures between the observed offspring genotypes and those expected under Mendelian inheritance (39). Analyses were performed assuming an additive genetic model and used FBAT v1.7.3 to perform single-SNP and haplotype analyses (40–42).

Multi-SNP selection analyses were also performed to determine the most likely etiological variant(s) for each of persistent microalbuminuria and severe nephropathy. These included multiple Cox proportional hazards regression models with forward and backward selection of main effects and two-way interactions of SOD1 SNPs (online appendix text) to capture un-genotyped variants and to avoid the uncertainties of phasing unrelated probands. To examine the association of SOD1/SFRS15 SNPs with both renal outcomes, we assumed a three-state progressive model, focusing on the rate of transition from normal to persistent microalbuminuria, and from persistent microalbuminuria to severe nephropathy. The multistate model provided separate estimates of the transition intensities between states and separate estimates of the SNP and covariate effects (43–45). Note that our model differs from the primary multivariate model above in that the repeated indicator for hypertension was not included.

Replicate log SOD1 expression values were averaged and compared across genotypes using separate general linear models for probands and relatives, taking into account the matched design. Log SOD activity and log SOD1 mass were analyzed similarly. To address concerns about residual non-normality in the latter, comparisons were repeated in conditional and unconditional logistic regression models that treated the genotype as a categorical response and log activity or log mass as an explanatory covariate (online appendix text).

The numbers of individuals with each of the renal and retinal events or who were censored are listed in online appendix 24. Twenty-three percent (n = 212) and 8% (n = 98) of the probands developed persistent microalbuminuria and severe nephropathy, respectively, during a follow-up of 13.5 ± 2.6 years (means ± SD) in DCCT/EDIC with 10 ± 2 AER measures. The effects of covariates in the Cox proportional hazards models, without the inclusion of SNPs, are provided for the univariate (online appendix 1) and multivariate models (online appendix 2), with HRs calculated for specific changes in quantitative covariates for the multivariate models (online appendix 23).

Of the 1,536 SNPs attempted on the Illumina chip, no genotypes were generated for 86 SNPs, leaving 1,441 autosomal and 9 X chromosome SNPs successfully genotyped. Thirty-seven autosomal and two X chromosome SNPs were nonpolymorphic in white DCCT/EDIC probands. Excepting these SNPs, genotype data were 99.95% complete with only six samples having no genotypes. Sixty-two autosomal SNPs had MAF <1% (online appendix 10), leaving 1,342 autosomal and 7 X chromosome SNPs with MAF >1% (online appendix 3). One proband was discrepant between reported and genotyped sex. Based on 18 duplicate samples (n ∼ 52,000 genotypes), genotype agreement was 100%. Marked deviation from HWE (online appendixes 3 and 6) was observed for both rs12431885 (triallelic) and rs2069579 in BDKRB2 and the triallelic rs3091244 (CRP) for which we only genotyped alleles A and G.

From the multivariate association analysis with outcomes, there were four SNPs with q < 0.5 for persistent microalbuminuria, 16 for severe nephropathy, and none for ≥3 step progression, severe retinopathy, or CSME (Table 1; online appendixes 5, 9, and 12). Of interest, most positive SNPs showed detectable associations with either persistent microalbuminuria or severe nephropathy, but not both. Two exceptions were fms-related tyrosine kinase 4 (FLT4; also known as vascular endothelial growth factor receptor 3) and SOD1. For FLT4, rs307806 was associated with both persistent microalbuminuria and severe nephropathy, whereas two other SNPs in this gene were separately associated with persistent microalbuminuria and severe nephropathy. Also of note, three SNPs in CNDP1 (online appendix 5) had P < 0.05 for persistent microalbuminuria, but their linkage disequilibrium with CNDP1 variants associated with diabetic nephropathy is unknown (11). In this report, we focus on SOD1 because rs17880135 had the lowest q value for any outcome (severe nephropathy, q = 0.06), and the same SNP also showed the most significant association with persistent microalbuminuria (q = 0.48).

In the fine-mapping stage, we followed up the association of rs17880135 by examining a total of 11 SNPs in SOD1/SFRS15 (4 on the Illumina chip and 7 genotyped using Taqman). In designing the Ilumina chip, we had selected five tagSNPs with MAF >5% in SOD1 from the SeattleSNPs project PDR90 sample (24), which identified 80 SNPs including 55 represented in 19 bins and an additional 25 single SNPs bins (online appendix 21). Many of the bins contained SNPs with low MAF. Of the five tag SNPs selected, four passed the Illumina bioinformatic design criteria and were genotyped. Rs17880135 lies 3′ of SOD1 and is in the same linkage disequilibrium bin in SeattleSNPs as rs17881180 in intron 1 and was therefore genotyped in all DCCT/EDIC probands. As expected, rs17881180 was in strong linkage disequilibrium with rs17880135 (r² = 0.912; Table 2) and had similar associations with renal outcomes (Table 3).

We sequenced 19 kb across the genomic region of SOD1 in eight individuals who were homozygous for the rare (high-risk) genotype at rs17880135 (online appendix 19) and found no coding variations. Of note, all eight individuals were homozygous for the rare allele at rs202446 (46), and 4 of 16 chromosomes carried the rare allele at rs17881203 (which had MAF = 5% in HapMap) and were therefore genotyped in DCCT/EDIC probands (Tables 2 and 3). No variants were found in the cDNA of SFRS15 compared with the reference sequence.

Based on the sequencing, SeattleSNPs, and HapMap, we genotyped six additional SNPs in SOD1/SFRS15—selected predominantly to capture linkage disequilibrium bins across SOD1 (online appendixes 21 and 13–15; Tables 2 and 3). Of note, 5 of the 11 SOD1/SFRS15 SNPs (rs17880135, rs17881180, rs202446, rs9974610, and rs204732) were associated with persistent microalbuminuria (with P < 10⁻³), whereas rs17880135 and rs17881180 were most strongly associated with severe nephropathy (Table 3). We found no evidence for significant nonadditivity for the association of SOD1/SFRS15 SNPs with renal outcomes (online appendix 22). There was no significant association (P < 0.01) between any of the SOD1 SNPs and any of the retinal outcomes (online appendixes 5 and 15).

To determine linkage disequilibrium between rs17880135 and rs17881180 and other SNPs genotyped by HapMap in the SOD1 region, we genotyped them in the CEU HapMap samples (27,28). We found that none of the SNPs within 100 kb SOD1 genotyped by HapMap are in linkage disequilibrium (r² > 0.5) with any of the five SNPs associated with persistent microalbuminuria (online appendixes 28 and 29).

Because some of the SNPs in SOD1 that are associated with persistent microalbuminuria and severe nephropathy are in strong linkage disequilibrium with each other (Table 2), we used multi-SNP regression analyses to attempt to identify the most statistically important SNP(s). For persistent microalbuminuria, either rs202446 (backward selection) or rs204732 (forward selection) remained in the model, suggesting that either of them or other variant(s) in strong linkage disequilibrium with both of them account for this association. For severe nephropathy, only rs17881180 remained after selection, but because of the strong linkage disequilibrium with rs17880135, we cannot distinguish between them. Examination of all possible pairwise interactions between SNPs provided no compelling evidence for interactions for either persistent microalbuminuria or severe nephropathy (online appendix 16).

Initially, we analyzed persistent microalbuminuria and severe nephropathy as separate traits; however, our definition of severe nephropathy required that individuals first progress through persistent microalbuminuria. To compare the association of each of the SOD1 SNPs with the different stages of disease, we examined multistate models. rs17880135 and rs17881180 were significantly associated with both persistent microalbuminuria and severe nephropathy and with the progression from persistent microalbuminuria to severe nephropathy (P < 0.05; Table 4). In contrast, rs204732 was associated with persistent microalbuminuria but not with the subsequent development of severe nephropathy (heterogeneity P = 0.044).

Although there was no significant preferential allele transmission with severe nephropathy detected for any of the three SNPs genotyped in the relatives (rs17880135, rs17881180, and rs202446), rs202446 was associated with persistent microalbuminuria (P = 0.036) in the same direction as the individual-based analysis, consistent with haplotype analysis (data not shown). Of note, the family-level test for severe nephropathy with rs17880135 and rs17881180 suffered from low statistical power due to MAF = 5% and subsequently only 152–163 informative families contributing to the test statistic.

Generally, we observed no significant difference (P < 0.05) by genotype at rs17880135 of SOD1 expression in EBV-transformed lymphoblasts, serum SOD1 mass, or SOD activity (Fig. 1), although there was a borderline higher SOD1 mass in probands with A/C compared with A/A genotypes. This effect was not observed in the nondiabetic relatives. Findings for rs17881180 were identical, due to nearly complete linkage disequilibrium, and results for the other three SOD1 SNPs associated with persistent microalbuminuria were nonsignificant (online appendixes 30 and 31).

We attempted to confirm the association of SOD1 SNPs with diabetic nephropathy in type 1 diabetes in three cross-sectional studies (Table 5). None of the SOD1 SNPs were significantly (P < 0.05) associated in any of the studies, but the 95% CIs were wide, suggesting modest power to detect association. Although each study had different definitions of diabetic nephropathy (online appendix text), meta-analysis did not detect significant between-study heterogeneity. The pooled 95% CI could not exclude an odds ratio as large as 1.84.

SOD1 (Copper-zinc super oxide dismutase) is an antioxidant enzyme present in the cytosol, nuclei, peroxisomes, and inner mitochondrial space. Its primary function is to lower intracellular concentrations of superoxide anion, a reactive oxygen species. There is ample evidence from animal models that oxidative stress is involved in the pathogenesis of diabetic nephropathy (47–49). In a rat model of type 1 diabetes, renal mitochondria produced significantly increased quantities of superoxide anion and showed evidence of oxidative damage (50). Moreover, mRNA expression of SOD1 is increased in the kidneys of diabetic rats and is normalized upon the administration of insulin and return of metabolic control (51). The role of SOD1 in diabetic nephropathy is supported in murine models of both type 1 and type 2 diabetes. When SOD1 is overexpressed, there is a reduction in renal cell injury, including albuminuria (52,53), leading to the conclusion that increases in cellular SOD1 activity attenuate diabetic renal injury. Consistent with this, when SOD1 is knocked out, diabetic renal disease develops faster (54). Figure 2 shows the distribution of SOD1 protein in human kidney, with the strongest staining in podocytes and parietal cells, with minimal expression in mesangial cells. The somatic mutation frequency has been shown to be reduced by overexpression of SOD1 (55). Mutations in SOD1 are present in ∼10% of patients with familial amyotrophic lateral sclerosis (56), with most mutations nonsynonymous and genetically dominant (57,58). Although there are multiple transcripts of SOD1, the functional relevance of most is unclear (59–61). There is no known functional effect of the SOD1/SFRS15 SNPs that are significantly associated with persistent microalbuminuria and severe nephropathy, and little is known about SFRS15 (62).

We initially tested 1,411 common variations in 212 candidate genes for nephropathy and retinopathy in the DCCT/EDIC probands using time-to-event analysis, including all appropriate covariates. The most striking finding was for rs17800135 in SOD1 with severe nephropathy and persistent microalbuminuria, and we proceeded to fine-map this by DNA sequencing and genotyping seven additional SNPs. We found multiple SOD1 SNPs to be associated with the development of each of persistent microalbuminuria and severe nephropathy, and evidence for differential association of rs204732 with persistent microalbuminuria and severe nephropathy from multistate models. Allele and haplotype frequencies of four SOD1 SNPs vary across global populations (online appendix 20; Fig. 3) and are important information for power calculations for replication studies in other populations. Specifically, the rare (risk) alleles at rs17800135 and rs17881180 were present solely in the European and Indian population samples and also showed variation in frequency within Europe. The argument that the association of SOD1 SNPs with diabetic nephropathy is solely due to population stratification is mitigated by the observed association from family-based analyses of rs202446 with persistent microalbuminuria.

We observed a HR of 2.62 (95% CI 1.64–4.18; P = 5.6 × 10⁻⁵) for the association of rs17880135 with time to severe nephropathy. This SNP was chosen by selecting the smallest P value after testing 1,411 SNPs for association, leading to upward bias in the magnitude of the genetic effect estimate (i.e., HR) (63,64), a phenomenon known as the Beavis effect or Winner's curse. To obtain more realistic estimates of the effect size expected in a replication sample, we calculated bootstrap estimates based on repeated sample splitting in which approximately two-thirds of the individuals were used to identify the SNP with the smallest P value and the HR for that SNP was estimated in the remaining individuals (65–68), yielding a weighted bootstrap estimate of 1.65 for the HR for rs17880135. To detect a HR of 1.65 in a study of similar design to DCCT/EDIC, assuming MAF = 5%, 8% of probands with the outcome, and power of >80%, a study sample size of 4,375 individuals (350 with diabetic nephropathy) would be required for α = 0.05 (6,250 with 500 case subjects for α = 0.01). This suggests that large sample sizes are needed to achieve replication studies with sufficient power. The point estimates for population attributable risk in for the time-to-event analysis in DCCT/EDIC for rs17881180 with severe nephropathy was 13.8%, whereas for rs202446 with persistent microalbuminuria, it was 14.2% (69).

Analyses of three cross-sectional studies with a total of 1,301 diabetic nephropathy case subjects and 1,585 control subjects failed to provide supportive evidence for the association of specific SOD1 SNPs with diabetic nephropathy. However, these studies were of modest size and cross-sectional in nature. Furthermore, the case definitions and characteristics of diabetic nephropathy varied considerably among studies. For example, in DCCT/EDIC, only 10% of severe nephropathy case subjects had developed ESRD by EDIC year 8, whereas 57 and 65% of case subjects in the Irish and GoKinD met this criterion, respectively. Therefore, lack of confirmation, assuming that the DCCT/EDIC result is a true positive, may be due to different phenotypes among these studies.

In addition to the Beavis effect and different phenotypes, failure to replicate the SOD1 association in the cross-sectional studies may be in part due to differences in study design. We performed a cross-sectional analysis of DCCT/EDIC based on the repeated AER measures to compare with our primary time-to-event analysis that included covariates (Table 5). We defined persistent microalbuminuria case subjects as ever having two consecutive AERs >30 mg/day during DCCT/EDIC, severe nephropathy to be two consecutive AERs >30 mg/day and one AER >300 mg/day or ESRD, whereas control subjects had to have 15 years of diabetes and never two consecutive AERs >30. In this cross-sectional analysis, the evidence for association between rs17880135 and persistent microalbuminuria was borderline significant (P = 0.059), whereas the association between rs17880135 and severe nephropathy (P = 0.0014) remained at a weaker level. However, the latter was 2 orders of magnitude less significant than the results of the primary analysis, and a similar pattern of results was observed for rs202446. Thus, even using this generous cross-sectional definition of renal disease (renal measures collected during the whole of DCCT/EDIC) produces a reduction in the evidence for association. Attempts to replicate the DCCT/EDIC associations in cross-sectional studies may have inherent limitations, particularly if the genetic effect operates to modify the time when diabetic nephropathy develops, rather than just influencing susceptibility.

We did not identify functional differences between the associated SOD1 SNPs, but these studies were limited by the low frequency of some genotypes and by the use of available samples. Specifically lymphoblasts were used for expression studies, and SOD activity and SOD1 mass were measured in serum, whereas renal tissue would have been preferred. In conclusion, using tagSNPs, sequencing, fine-mapping, and detailed statistical analysis, we have found that multiple variants in SOD1 are associated with the time to development of persistent microalbuminuria and subsequent progression to severe nephropathy in the phenotypically rich DCCT/EDIC study. The involvement of SOD1 is consistent with the known molecular mechanisms for the development of diabetic complications (49). Although we were unable to observe confirmation in three other studies, we cannot exclude an effect size similar to that obtained from cross-sectional analysis of the DCCT/EDIC data. This indicates that even the combined sample size of these studies may have been insufficient to reliably detect the SOD1/SFRS15 association, and that studies of additional populations is warranted.

H.A.-K. has received support from Premier's Research Excellence Award, Government of Ontario. L.M. has received a CIHR Research Scholarship. R.S. has received support from the Samuel Lunenfeld Research Institute Interface Training Program: Applying Genomics to Human Health (CIHR). K.K.K. has received grant P01-GM57672 toward typing the global sample from the National Institute of General Medical Sciences. S.B.B. has received a Canadian Institutes of Health Research (CIHR) Senior Investigator award. A.D.P. holds a Canada Research Chair in the Genetics of Complex Diseases. This work has received National Institute of Diabetes and Digestive and Kidney Diseases Contract N01-DK-6-2204, National Institute of Diabetes and Digestive and Kidney Diseases Grant R01-DK-077510, and Juvenile Diabetes Research Foundation Grant 1-2006-757 and support from the Canadian Network of Centres of Excellence in Mathematics. The GoKinD study sample collection has received support from the Juvenile Diabetes Research Foundation in collaboration with the Joslin Diabetes Center and George Washington University and from the U.S. Centers for Disease Control and Prevention.

Clinical data and DNA from the DCCT/EDIC study will be made available through the National Institute of Diabetes and Digestive and Kidney Diseases repository at https://www.niddkrepository.org/niddk/home.do. Clinical data and DNA from the Genetics of Kidneys in Diabetes (GoKinD) study are available at www.gokind.org.

The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Diabetes and Digestive and Kidney Diseases or the National Institutes of Health.

The contributions to the global sample of many researchers from around the world and thousands of individuals who volunteered are gratefully acknowledged.