Introduction

Diabetic nephropathy is a frequent condition affecting a substantial proportion of diabetic patients, and is a leading cause of end-stage renal disease. The determinants of diabetic nephropathy are complex and include genetic factors, as supported by a strong familial aggregation of diabetic nephropathy [1].

The precise nature of the genetic burden involved in diabetic nephropathy remains poorly understood. An alternative approach to candidate gene studies is to perform unbiased genome-wide association studies (GWAS). To date, only a small number of GWAS for diabetic nephropathy have been performed. Recently, the Genetics of Nephropathy an International Effort (GENIE) consortium identified two loci associated with a population attributable risk of 0.5–10.5% [2]. Therefore, additional studies are needed to discover novel loci associated with diabetic nephropathy.

The aim of our research was to conduct a GWAS in a large-scale case–control study including French and Danish patients, based on the latest 1000 Genomes (1000G) imputation techniques, and to replicate our findings in available cohorts of patients with type 1 diabetes from European ancestry.

Methods

Studied participants were of white European origin [2–4]. Briefly, patients were classified as cases or controls at the time of recruitment in all but the Finnish Diabetic Nephropathy (FinnDiane) study, where they were selected among a population-based consecutive collection of people with type 1 diabetes, and the nephropathy status of the patients was defined based on the latest available data. All cohorts were multi-centric except the single-centre LEwis blood type and Angiotensin Converting Enzyme (LEACE) study. Local ethics committees approved the study protocols. All participants gave written informed consent.

Phenotypic determination

Type 1 diabetes

Type 1 diabetes was diagnosed using American Diabetes Association criteria: rapid definitive insulin requirement (within 1 year of diagnosis) and age at diabetes onset below 31 or 36 years, differing according to each cohort (Electronic supplementary material [ESM] Table 1).

Diabetic nephropathy status

Phenotype determination was based on the clinical criteria proposed by the Genetics of Kidneys in Diabetes US (US-GoKinD) study: cases had proteinuria with or without renal failure. Proteinuria was uniformly defined on at least two of three sterile urine collections (ESM Table 1). Controls had long-term duration of diabetes (over 15 years), normo-albuminuria without renal failure and were not prescribed drugs that block the renin-angiotensin system. Microalbuminuric patients were not considered for the current analyses.

Study organisation

The general research strategy adopted in this project is summarised in ESM Fig. 1. Brief descriptions of the studied cohorts and patients can be found in ESM Tables 1 and 2.

We ultimately assessed expression of the identified gene in a post-hoc analysis of micro-dissected glomerular and tubule samples [5].

Genotype, quality control and imputation

Genotyping platforms, quality control and imputation are detailed in ESM Methods.

Statistical analysis

A logistic regression analysis was conducted to evaluate the association of each imputed single nucleotide polymorphism (SNP) with diabetic nephropathy under an additive genetic model (allele dosage was used as the covariate for characterising the tested SNP). Analyses were adjusted for age, sex and the first four principal components derived from the genome-wide genotyped SNPs computed by the Eigenstrat programme (http://www.hsph.harvard.edu/alkes-price/software/) [6]. Association analyses were performed using the mach2dat (version 1.0.19) software (http://www.unc.edu/~yunmli/software.html).

All SNPs with suggestive evidence of an association with diabetic nephropathy (p < 10−5) were moved forward in the first-stage study. The centre- and sex-adjusted association of these SNPs with diabetic nephropathy was investigated using the SNPTEST programme (https://mathgen.stats.ox.ac.uk/genetics_software/snptest/snptest.html#introduction).

In the second-stage samples, the association of candidate SNP with diabetic nephropathy was tested using the Cochran–Armitage (CA) trend test. ORs derived from application of the CA test in the four case–control samples were finally combined into a random-effect model based meta-analysis using the GWAMA programme (http://www.well.ox.ac.uk/gwama/download.shtml) [7].

Results

Discovery analysis

None of the 11,133,962 tested SNPs achieved the statistical threshold of 5 × 10−8 for declaring genome-wide significance (ESM Figs 2 and 3). All SNPs with suggestive evidence for association at p < 10−5 (n = 46) were selected for further association testing in a first independent sample of 820 cases and 885 controls in the US-GoKinD study (ESM Fig. 1).

First-stage study

After Bonferroni correction, three SNPs demonstrated significant association with diabetic nephropathy: rs11188343 (p = 9.06 × 10−5), rs1326934 (p = 9.85 × 10−5) and rs4917695 (p = 1.27 × 10−4) (ESM Table 3). These SNPs mapped to the SORBS1 gene and were in near complete linkage disequilibrium (r 2 ~ 1).

In the discovery GWAS, the minor rs1326934-C allele was less frequent in cases than in controls (0.34 vs 0.43) and was associated with a decreased risk for diabetic nephropathy (OR 0.70 [95% CI 0.60, 0.82]; p = 7.87 × 10−6). In the first-stage population, a consistent association was observed with the rs1326934-C allele at a lower frequency in cases (combined statistical evidence for association of p = 3.52 × 10−9) compared with controls (Table 1).

Table 1 Association of SORBS1 rs1326934 with diabetic nephropathy in the four studied cohorts
Full size table

Second-stage study

To validate further the association of rs13626934 with diabetic nephropathy, we studied this SNP in two additional case–control studies composed of 823 cases and 903 controls (All Ireland-Warren 3-Genetics of Kidneys in Diabetes UK and Republic of Ireland [UK-ROI]) and 1,335 cases and 1,633 controls (FinnDiane; Table 1). In UK-ROI, the same trend of association, though not significant (p = 0.15), was observed: the rs13626934-C allele was less frequent in cases than in controls (0.42 vs 0.45). However, no association was observed in the FinnDiane population (p = 0.44) where the allele frequency of the C allele was similar in cases and controls (0.27 vs 0.28).

Combining the results of the four studies into a fixed-effect meta-analysis led to an overall OR for diabetic nephropathy of 0.84 (95% CI 0.79, 0.90; p = 5.69 × 10−7). However, this association was statistically heterogeneous across the four samples (p = 3.94 × 10−3) and the random-effect meta-analysed OR for diabetic nephropathy was, therefore, 0.83 (95% CI 0.72, 0.96; p = 0.009). When FinnDiane patients were excluded the meta-analysed OR was 0.795 (95% CI 0.733, 0.861; p = 2.40 × 10−8).

Gene expression analysis

SORBS1 gene overexpression was observed in tubules of type 2 diabetes patients compared with controls (Fig. 1a, ESM Tables 4 and 5). We also observed a significant inverse correlation between SORBS1 expression and estimated glomerular filtration rate (eGFR) (R = −0.493; p = 1.44 × 10−3) (Fig. 1b). Among the other genes mapped within 250 kb of rs1326934, SORBS1 demonstrated the strongest correlation with eGFR in both control and diabetic nephropathy tubule samples (Fig. 1c). Furthermore, at a protein level, sorbin and SH3 domain-containing protein 1 (SORBS1) was also highly expressed in renal tubules and medium expressed in glomeruli (www.proteinatlas.org).

Fig. 1
figure 1

Expression profile of SORBS1 in diabetic nephropathy. (a) Tubular expression of SORBS1 is significantly upregulated in diabetic kidney disease (black bars) compared with control samples (white bars; p = 0.0006). (b) Each dot represents transcript levels and eGFR values from a single kidney sample. The line represents the fitted linear correlation values. (c) The genomic position of each gene on chromosome 10 (q24.1) in the 500 kb vicinity of rs1326934 and rs11188343 loci (triangle) is shown. Not only the SORBS1 transcript, but also other transcripts, correlate with renal function in the vicinity of rs1326934 and rs11188343 loci

Full size image

Discussion

In this study, a multi-stage based GWAS looked for novel susceptibility genes associated with diabetic nephropathy in patients with type 1 diabetes. Although we did not detect any new loci passing genome-wide significance, we observed promising evidence for an association between SORBS1 rs1326934 (or any SNP in complete linkage disequilibrium with it) and diabetic nephropathy in three of four of the studied populations. Given the frequency of the risk allele observed in our control populations (~0.55) and the associated OR for diabetic nephropathy (~1.25), the population attributable risk of the identified polymorphisms would be 12%.

No association was observed in the Finish population, which exhibited a marked difference in allele frequency at the candidate SNP compared with the three other studied populations. We further interrogated databases with genome-wide genotype and gene expression but did not observe any association in monocytes, macrophages, hepatocytes, adipocytes [8] or endothelial cells [9] with SORBS1 gene expression. Nevertheless, the SORBS1 rs1188343 (in complete linkage disequilibrium with the rs1326934) was predicted in the RegulomeDB database (http://regulome.stanford.edu) to be located at a potential binding site for either RFX3 or histone interaction.

The sorbin protein, coded by SORBS1, was found to be differentially upregulated in glomeruli of rats with diabetic nephropathy compared with rats without diabetic nephropathy [10]. High tubular and moderate glomerular expression of sorbin protein was observed in human kidney samples. These findings from type 2 diabetes patients provide additional support for the association of SORBS1 with diabetic nephropathy. Gene expression changes of SORBS1 were easier to detect in tubules, as SORBS1 has a higher tubular expression. Although SORBS1 expression was significantly upregulated only in tubules, we cannot exclude the importance of glomerular SORBS1. Diabetic nephropathy not only involves glomeruli but also tubules [11], and genes identified through GWAS are likely to impact both renal structures.

The function of sorbin is not fully established but we speculate it plays a key role in several processes involved in diabetic nephropathy, including insulin resistance and cytoskeleton architecture. Sorbin acts in the genesis of stress fibres and might, therefore, be involved in podocyte alterations of the slit diaphragm barrier.

The lack of homogenous replicated associations in our work is unlikely to be due to clinical heterogeneity of the studied populations as the same definitions for type 1 diabetes and diabetic nephropathy were used. Conversely, the different patterns of allelic heterogeneity and linkage disequilibrium across European populations may explain the lack of replication of the SORBS1 signal in the Finnish population. Characterising the exact variability at the SORBS1 locus is needed to validate SORBS1 as a new susceptibility gene for diabetic nephropathy and to identify the disease-associated functional variant(s).

The main limitation of the present study is its design. We adopted a multi-stage strategy using all available GWAS resources imputed for 1000G reference dataset at the time this work was launched. A more powerful approach would have been to conduct a comprehensive meta-analysis of the four populations. An international initiative has been set up to overcome this limitation. Another caveat is that the diabetic nephropathy phenotype was defined by the presence of proteinuria, regardless of renal function, and genetic susceptibility to proteinuria might differ from genetic predisposition to renal failure. However, the diabetic nephropathy cohorts we used all had this combined diabetic nephropathy phenotype permitting international collaboration.

In conclusion, our study provides preliminary support for SORBS1 as a new susceptibility gene for diabetic nephropathy.