A Genome-Wide Association Study of Diabetic Kidney Disease in Subjects With Type 2 Diabetes

Abstract

Identification of sequence variants robustly associated with predisposition to diabetic kidney disease (DKD) has the potential to provide insights into the pathophysiological mechanisms responsible. We conducted a genome-wide association study (GWAS) of DKD in type 2 diabetes (T2D) using eight complementary dichotomous and quantitative DKD phenotypes: the principal dichotomous analysis involved 5,717 T2D subjects, 3,345 with DKD. Promising association signals were evaluated in up to 26,827 subjects with T2D (12,710 with DKD). A combined T1D+T2D GWAS was performed using complementary data available for subjects with T1D, which, with replication samples, involved up to 40,340 subjects with diabetes (18,582 with DKD). Analysis of specific DKD phenotypes identified a novel signal near GABRR1 (rs9942471, P = 4.5 × 10⁻⁸) associated with microalbuminuria in European T2D case subjects. However, no replication of this signal was observed in Asian subjects with T2D or in the equivalent T1D analysis. There was only limited support, in this substantially enlarged analysis, for association at previously reported DKD signals, except for those at UMOD and PRKAG2, both associated with estimated glomerular filtration rate. We conclude that, despite challenges in addressing phenotypic heterogeneity, access to increased sample sizes will continue to provide more robust inference regarding risk variant discovery for DKD.

Introduction

Progressive loss of renal function represents one of the most serious complications of diabetes, yet strategies for prevention and management are suboptimal. One of the principal obstacles to improved clinical interventions remains rudimentary understanding of the processes whereby sustained exposure to elevated levels of glucose (and/or other manifestations of the diabetic state) leads to progressive disturbance of renal morphology and function (1).

There is considerable variation in the progression and severity of renal complications of diabetes (collectively, diabetic kidney disease [DKD]). The prevalence of DKD in subjects with type 2 diabetes (T2D) is ∼30–50%: some patients experience a relatively rapid decline in renal function, whereas others maintain normal renal function despite decades of suboptimal glycemic control (2). The factors influencing this variation in outcome have not been fully characterized, but substantial evidence supports a genetic contribution. As in type 1 diabetes (T1D), DKD in those with T2D aggregates in families (3,4), and the prevalence of DKD in T2D differs considerably between ethnic groups (5–7).

These observations indicate that the identification of genetic variants influencing DKD predisposition should accelerate characterization of the biological basis of DKD. In contrast with most complex multifactorial traits, efforts to apply candidate gene and genome-wide association study (GWAS) approaches to DKD have met with limited success (8–11). Many genetic associations have been reported, but few robustly replicated loci have emerged. This likely reflects the comparatively small sample sizes of previous studies, such that power would have been limited to detection of common loci of unusually large effect. In the case of DKD in T2D, this is likely to have been compounded by the heterogeneity of the phenotype: autopsy studies indicate that only ∼50% of chronic kidney disease (CKD) in T2D can be attributed to classic diabetic nephropathy (12). The success of equivalent GWAS efforts for CKD (for which several replicated loci have been described) provides reassurance that it is possible to identify variants with broad impact on the progression of renal disease, irrespective of the dominant pathology (13).

Reduced kidney function, reflected by the estimated glomerular filtration rate (eGFR) and end-stage renal disease (ESRD), and dysfunction of the glomerular filtration barrier, reflected by albuminuria, can develop independently. This suggests that the two cardinal features of DKD involve distinct disease mechanisms and may be subject to different genetic effects. Albuminuria is known to be a poor predictor of diabetes-related ESRD, especially in the early stages, and regression to normoalbuminuria is common in patients with microalbuminuria (14).

These observations provide confidence that the combination of increased sample size and improved definition of DKD phenotypes should enable risk variant detection and uncover mechanisms that contribute to renal dysfunction in diabetes. In particular, the separation of case subjects into phenotypic classes based on disease stage and/or phenotype manifestations, incorporating information on both albumin excretion and eGFR, can be expected to increase etiological homogeneity and augment power for locus identification (14–16).

The SUrrogate markers for Micro- and Macrovascular hard endpoints for Innovative diabetes Tools (SUMMIT) Consortium adopted such a strategy to perform a GWAS for DKD in subjects with T1D (17). Here, we report on equivalent analyses conducted in the context of T2D, as well as those from a combined T1D+T2D analysis involving up to 40,340 subjects.

Research Design and Methods

DKD Phenotype Definitions

Not all patients with DKD will develop every form of the disease or progress to the most severe stage of ESRD. Dysfunction of the glomerular barrier, represented by albuminuria, and reduced kidney function, represented by eGFR, can develop independently. To explore the disease severity spectrum and the different disease processes represented by eGFR and albuminuria, we defined seven binary phenotypes using clinical measures of albumin-to-creatinine ratio (ACR), albumin excretion rate (AER), and eGFR (Table 1 [T2D only] and Supplementary Table 8 [T1D+T2D]). The phenotype definitions were aligned to other large-scale genetic studies of T1D DKD in SUMMIT (17) and the Diabetic Nephropathy Collaborative Research Initiative (DNCRI) (18). The definition of CKD was also aligned to that used by the CKDGen Consortium (eGFR <60 mL/min/1.73 m²), although we restricted case and control subjects to those with diabetes (13).

Table 1.

GWAS characteristics by DKD phenotypes in subjects with T2D and T1D

Analysis	Case definition	Control definition	Subjects with T2D		Subjects with T1D
			Case, n	Control, n	Case, n	Control, n
All DKD	Microalbuminuria OR Late DKD OR ESRD	Normoalbuminuria (AER <20 µg/min OR AER <30 mg/24 h OR ACR <2.5/3.5 mg/mmol for men/women) AND duration of T2D >10 years‡	3,345	2,372	2,563	2,593
Microalbuminuria*	At least 2 out of 3 consecutive measurements with AER ≥20 AND <200 µg/min OR AER ≥30 AND <300 mg/24 h OR ACR ≥2.5/3.5 AND <25/35 mg/mmol for men/women	Normoalbuminuria AND duration of T2D >10 years‡	1,989	2,238†	806	2,593
Late DKD	At least one measurement with AER ≥200 µg/min OR AER ≥300 mg/24 h OR ACR ≥25/35 mg/mmol for men/women OR ESRD (eGFR <15 mL/min/1.73 m2 OR kidney transplantation OR dialysis)	Normoalbuminuria AND duration of T2D >10 years‡	1,339	2,372	1,757	2,593
ESRD vs. control subjects	ESRD: eGFR <15 mL/min/1.73 m2 OR renal dialysis OR kidney transplant	No DKD AND duration of T2D >10 years‡	371	2,076	813	2,398
ESRD vs. no ESRD	ESRD (see above)	No ESRD AND duration of T2D >10 years‡	371	4,471	813	3,995
CKD	eGFR <60 mL/min/1.73 m2	No CKD AND duration of T2D >10 years‡	3,094	2,906	2,460	774
CKD and DKD	eGFR <45 mL/min/1.73 m2 AND all DKD	No CKD AND no ESRD AND normoalbuminuria AND duration of T2D >10 years‡	897	1,610	1,750	1,385
eGFR	32,788 × serum creatinine (μmol/L)−1.154 × age−0.203 × [0.742 if female] (mL/min/1.73 m2)		9,197		3,961

*Equivalent to the early DKD phenotype from Sandholm et al. (17).

†Not all studies were able to define microalbuminuria (due to limited information on microalbuminuric status) and thus the case and control group number is smaller than all DKD and late DKD.

‡The duration of diabetes for subjects with T1D was >15 years.

We used AER measured overnight (µg/min), during 24 h (mg/24 h), or as a spot measurement of ACR (mg/mmol) or eGFR calculated using the Modification of Diet Renal Disease Study (MDRD) formula (eGFR = 32,788 × serum creatinine (μmol/L)^−1.154 × age^−0.203 × [0.742 if female]) to classify disease stage and severity. We based the control definition on either AER or ACR, as most studies had measured either. In the studies that had measured both, two of the three measures for AER and ACR had to meet the control criteria (Table 1). We were unable to exclude albuminuric patients that presented as normoalbuminuric due to prescribed renin-angiotensin system blockers. As reduced kidney function (reflected by eGFR) and dysfunction of the glomerular filtration barrier (reflected by albuminuria) can develop independently, we did not exclude individuals with albuminuria from the control subjects for the eGFR-defined phenotypes and vice versa. In subjects with T2D, ∼46% of normoalbuminuric control subjects had an eGFR <60 mL/min/1.73 m² (1,098/2,372).

In all, we defined seven dichotomous phenotypes:

1. The “all DKD” phenotype, our primary phenotype, designed to capture the broadest set of DKD phenotypes.

2. The “microalbuminuria” phenotype (equivalent to early DKD from Sandholm et al. [17]) to identify variants that contribute to early dysfunction of the glomerular barrier.

3. The “late DKD” phenotype to identify variants that contribute to severe glomerular barrier dysfunction.

4. Two ESRD-related phenotypes focused on identification of variants associated with end-stage renal failure, comparing those with ESRD either to control subjects without any DKD (“ESRD vs. control subjects”) or to control subjects without ESRD (“ESRD vs. no ESRD”).

5. The “CKD” phenotype to identify variants that contribute to reduced kidney function (eGFR).

6. The “CKD and DKD” phenotype to identify any variants that may contribute to the development of kidney disease irrespective of glomerular barrier dysfunction or reduced kidney function.

7. The “eGFR” phenotype, a continuous phenotype, to identify variants that play a role in kidney function that may not be detected by the analysis of the binary DKD phenotypes. The eGFR measures were not transformed as they approximated a normal distribution (Supplementary Fig. 1).

Study Populations

We identified DKD case and control subjects with T2D from the Scania Diabetes Registry (SDR) (19), Genetics of Diabetes Audit and Research in Tayside Scotland (GoDARTS) study (20), Steno Diabetes Centre (21), and Bergamo Nephrologic Diabetes Complications Trial (BENEDICT) A and B studies (22). We identified independent replication studies in populations of European descent (Diabetes Epidemiology: Collaborative analysis Of Diagnostic criteria in Europe [DECODE], Family Investigation of Nephropathy and Diabetes [FIND], Diabetes Register Vaasa [DIREVA], Diagnostic Optimization and Treatment of Diabetes and Its Complications in the Chernihiv Region [DOLCE], Malmö Diet and Cancer [MDC], Inter99, Vejle Diabetes Biobank, and the Anglo-Danish-Dutch Study of Intensive Treatment In People with Screen Detected Diabetes in Primary Care [ADDITION] studies) and Asian descent (RIKEN, the Singapore Diabetic Cohort Study [SDCS], the Hong Kong Diabetes Registry [HKDR], and the Singapore Study of Macro-angiopathy and Microvascular Reactivity in Type 2 Diabetes [SMART2D] study) (Supplementary Table 1).

We combined the subjects with T2D with nonoverlapping samples from the study of DKD in subjects with T1D (17). Replication studies (of DKD in subjects with T1D) (17) were also used for replication in the combined analysis of T1D+T2D (Supplementary Table 1). None of these studies overlapped with samples included in the analysis of eGFR and CKD by the CKDGen Consortium (13).

Genome-Wide Genotyping and Imputation

The T2D discovery cohorts were genotyped on the Affymetrix SNP 6.0, the Illumina Omni express array, and the Illumina 610Quad arrays (Supplementary Table 4). Individual study centers excluded single nucleotide polymorphisms (SNPs) for minor allele frequency (MAF) <1%. SNPs with a MAF 1–5% were excluded if the Hardy-Weinberg equilibrium test P < 1 × 10⁻⁴ or the call rate <99%. SNPs with MAF ≥5% were excluded if the Hardy-Weinberg equilibrium test P < 5.7 × 10⁻⁷ or the call rate <95% (23). Samples were excluded if their call rate was <95%, genotype heterozygosity was >3 SD from the study sample mean, or they failed sex checks. Based on principal component analysis, population outliers were removed if they were not of European descent (compared with the 1000 Genomes Project [1000G] populations) or fell >3 SD away from the population means of the first two principal components for samples of European descent. Duplicates were removed but related individuals were retained for genotype imputation.

Genotypes were prephased using SHAPE-IT (v2) (24) and imputed using IMPUTEv2 (25) against the March 2013 1000G version 1 reference panel using standard protocols and recommended settings.

Replication Genotyping

Direct typing of twelve SNPs (rs11622435, rs12917707, rs17421627, rs1989248, rs2194025, rs2206136, rs4977388, rs61277444, rs6865390, rs7222331, rs9939609, and rs9942471) was performed in DIREVA samples using TaqMan allelic discrimination assays, according to the manufacturer’s protocol (Applied Biosystems, Carlsbad, CA). Sequenom multiplex genotyping was performed for the same SNPs in DOLCE, using the standard protocol (26).

Statistical Analysis

Heritability of DKD Phenotypes

Narrow sense heritability was estimated by GCTA (v1.26) (27) from 4.5 million directly typed and imputed markers (info >0.75) in GoDARTS (Supplementary Table 1) for all DKD, CKD, and eGFR. The sample size for these phenotypes exceeded the recommended threshold for reliable heritability estimates (N = 3,160 based on SE ≤0.1) (28).

Genome-Wide Association Analysis

Genome-wide association analyses were performed by individual study centers using an additive model while correcting for age, sex, and duration of diabetes. We estimated allelic effects using the score test from SNPTESTv2 in unrelated samples for dichotomous traits (29). Association P values were calculated using EMMAX from a larger sample of related individuals while correcting for a kinship matrix (30). For eGFR phenotype, we estimated allelic effects and association P values using EMMAX (30).

Power Calculations

We performed power calculations for dichotomous traits based on a MAF of 8%, an allelic odds ratio (OR) range of 1.05–2.00, and α = 5 × 10⁻⁸ (genome-wide significance). The power calculations were performed for the discovery meta-analysis of all DKD and separately for the meta-analyses of T2D only (3,345 DKD case and 2,372 DKD control subjects) and the combined T1D+T2D (5,908 DKD case and 4,965 DKD control subjects).

At α ≤ 5 × 10⁻⁸, we had >80% power to detect an allelic OR >1.40 in the T2D-only discovery analysis (Supplementary Fig. 2C) and an allelic OR >1.25 in the T1D+T2D discovery analysis (Supplementary Fig. 2B). We also performed power calculations for the reported DKD loci, as above, but using α = 9 × 10⁻⁴ (this α accounts for the number of loci tested but not the number of phenotypes analyzed). In T1D+T2D analysis, we had >80% power to detect variants with an allelic OR >1.20 (Supplementary Fig. 2A).

Discovery Meta-analysis

Two discovery meta-analyses were performed: one that included summary statistics estimated from subjects with T2D only and a second that combined T2D-only analyses with equivalent analyses in subjects with T1D (17). Individual study summary statistics were centrally filtered for a minor allele count in either case or control subjects <10 and an info score <0.4 for imputed variants.

EMMAX P values were combined in a sample size weighted z-statistic meta-analysis using METAL (version 25/03/2011) (31). Effect estimates were combined in a fixed-effect inverse-variance weighted meta-analysis using GWAMA (v2.1) (32). Meta-analysis results were restricted to allelic effects estimated in two or more studies. For binary traits, independent variants (>100 kb apart) were selected for replication from the T2D-only analysis based on association P ≤ 5 × 10⁻⁶ and from the T1D+T2D analysis based on P ≤ 1 × 10⁻⁶. For eGFR phenotype, SNPs were chosen for replication based on association P ≤ 5 × 10⁻⁶ in subjects with T2D or P < 1 × 10⁻⁶ in the T1D+T2D analysis. SNPs associated with eGFR at P ≤ 5 × 10⁻⁴ in either eGFR analysis (T2D only or T1D+T2D) that had also been reported at P ≤ 5 × 10⁻⁸ with eGFR by the CKDGen Consortium were also included in the list of SNPs for replication (13).

Replication

We sought replication for 164 lead variants in 13 studies of T2D DKD for which it was possible to obtain in silico replication from available GWAS data or replication from de novo genotyping (DIREVA and DOLCE) (Fig. 1). Replication studies aligned their DKD phenotypes with those used in the SUMMIT GWAS. Although association results for the lead variants were recovered for all compatible DKD phenotypes available in the replication samples (Supplementary Table 1), joint meta-analysis results were reported for those phenotypes where the primary GWAS associations exceeded the thresholds above.

Figure 1.

Eight DKD phenotypes were analyzed in subjects with T2D (blue boxes) and in a combined (green boxes) analysis of subjects with T2D or T1D (yellow box). N indicates the total sample count for either the all DKD (number of case subjects is given in parentheses) or the eGFR phenotype and may vary by variant as well as by DKD phenotype. Replication was sought for 164 loci and 47 loci from each analysis, respectively, in subjects of European and Asian ancestry with either T1D or T2D.

As with the discovery, meta-analysis effect estimates from replication studies were combined using GWAMA (v2.1) (32), and EMMAX P values, using METAL (version 25/03/2011) (31).

Known DKD Variants

We examined the literature for variants that have been associated with DKD from candidate gene (P < 0.05) and GWA (P ≤ 5 × 10⁻⁸) studies. Sixty-one variants were identified and aligned to the reported risk allele for binary traits (or the trait-raising allele for quantitative traits). We assessed both direction of effect and strength of association in the current study for those phenotypes that most closely matched the original report (but irrespective of type of diabetes).

Genetic Risk Score Analysis

We included variants (P ≤ 5 × 10⁻⁸) from GWAS to generate genetic risk scores (GRS) for coronary artery disease (33), BMI (34), waist-to-hip ratio adjusted for BMI (35), LDL cholesterol, triglycerides, HDL cholesterol (36), fasting insulin, insulin resistance (37–39), fasting glucose (38), T1D (40), T2D (41), and systolic blood pressure (42). The relationship between the GRS and DKD phenotype was calculated using an inverse-variance weighted method described in Ehret et al. (42).

Results

DKD Definitions

We considered seven dichotomous phenotypes designed to capture the spectrum of DKD (see research design and methods) and eGFR. We aimed to identify variants that influence multiple stages in DKD progression, as well as those that have more stage-specific effects. The principal definition (all DKD) included 3,345 T2D subjects with any form of DKD (ranging from microalbuminuria to ESRD) as case subjects and 2,372 T2D subjects, normoalbuminuric despite >10 years duration of diabetes, as control subjects. The other six dichotomous phenotypic comparisons are described in Table 1 (see research design and methods).

Contribution of Genetic Variants to DKD

The genetic variation, explained by the SNPs on the genotyping array and estimated using GCTA (v1.26) (30) in up to 6,335 subjects with T2D from GoDARTS, was highest in CKD (h² = 0.12) and similar for all DKD (h² = 0.08) and eGFR (h² = 0.07) (Supplementary Table 2). We restricted analyses to phenotypes with sample sizes deemed sufficient for accurate estimation of heritability (N ≥3,160 to obtain SE ≤0.1) (28).

GWAS for DKD in T2D

The DKD discovery analysis combined GWAS data from four studies of European descent: GoDARTS (20), SDR (19), Steno (21), and BENEDICT (phases A and B) (22) (Table 1 and Supplementary Table 3). For the principal (all DKD) analysis, the sample size of the discovery T2D-only meta-analysis had >80% power to detect variants with MAF ≥8% and allelic OR >1.40 (Supplementary Fig. 2C). The number of variants meta-analyzed for each DKD phenotype varied between 5,864,445 in the ESRD vs. no ESRD phenotype and 9,263,264 in the all DKD phenotype (Supplementary Table 4). These differences reflect the minor allele count exclusion filter.

Manhattan and quantile-quantile plots of discovery P values for each of the eight DKD phenotypes were well calibrated, and several showed a modest excess of significant associations (Supplementary Fig. 3). In the discovery GWAS, only one locus reached genome-wide significance: PLCB4 (encoding 1-phosphatidylinositol 4,5-bisphosphate phosphodiesterase β-4) on chromosome 20. The lead variant rs2206136 was associated with the CKD phenotype (effect allele frequency [EAF] 42%, OR 1.20 [95% CI 1.08, 1.34]; P = 2.1 × 10⁻⁸) (Table 2 and Supplementary Fig. 3A).

Table 2.

Five loci were associated (P ≤ 5 × 10⁻⁸) with CKD, microalbuminuria, and eGFR phenotypes in T2D or T1D+T2D subjects

CHR:BP	Phenotype	SNP (locus)	Discovery			Replication			Joint analysis
			EA/NEA (Info), EAF	OR/β (95% CI)	P	Ancestry	OR/β (95% CI)	P	OR/β (95% CI)	P	N
20: 9351150	T2D CKD	rs2206136 (PLCB4)	A/T (0.98), 0.42	OR 1.20 (1.08, 1.34)	2.1 × 10−8	European	OR 1.02 (0.91, 1.15)	0.69	OR 1.13 (1.05, 1.21)	9.0 × 10−5	11,900
						Asian and European	OR 1.03 (0.94, 1.13)	0.68	OR 1.12 (1.05, 1.19)	2.1 × 10−4	13,813
6: 89948232	T2D microalbu-minuria	rs9942471 (GABRR1)	A/C (0.99), 0.64	OR 1.24 (1.15, 1.34)	2.1 × 10−7	European	OR 1.32 (0.99, 1.75)	0.06	OR 1.25 (1.16, 1.34)	4.5 × 10−8	4,801
						Asian and European	OR 1.11 (0.99, 1.23)	0.12	OR 1.15 (1.08, 1.23)	1.2 × 10−5	5,559
16: 20400839	T2D eGFR	rs11864909 (UMOD)	T/C (1.00), 0.28	β 2.42 (1.28, 3.56)	2.7 × 10−5	European	β 2.22 (1.16, 3.28)	4.1 × 10−5	β 2.31 (1.54, 3.09)	4.6 × 10−9	12,343
						Asian and European	β 2.30 (1.48, 3.12)	3.6 × 10−8	β 2.34 (1.68, 3.00)	4.4 × 10−12	19,747
2: 170646916	T1D+T2D eGFR	rs1974990* (SSB)	G/T (0.98), 0.08	β 4.07 (2.61, 5.52)	4.8 × 10−8	European	No replication available		β 4.07 (2.61, 5.52)	4.8 × 10−8	13,158
						Asian and European	β 0.04 (−2.69, 2.76)	0.98	β 3.17 (1.88, 4.45)	1.4 × 10−6	14,828
7: 151415041	T1D+T2D eGFR	rs10224002 (PRKAG2)	A/G (0.92), 0.74	β 1.75 (0.85, 2.66)	1.5 × 10−4	European	β 2.15 (0.93, 3.37)	5.8 × 10−4	β 1.89 (1.17, 2.62)	3.4 × 10−7	20,495
						Asian and European	β 2.42 (1.28, 3.56)	3.2 × 10−5	β 2.01 (1.30, 2.72)	2.7 × 10−8	22,165
16: 20400839	T1D+T2D eGFR	rs11864909 (UMOD)	T/C (0.99), 0.29	β 1.90 (1.05, 2.74)	1.1 × 10−5	European	β 2.22 (1.16, 3.28)	4.1 × 10−5	β 2.02 (1.36, 2.69)	2.1 × 10−9	16,304
						Asian and European	β 2.30 (1.48, 3.12)	3.6 × 10−8	β 2.11 (1.52, 2.70)	2.3 × 10−12	23,708

BP, base pair; CHR, chromosome; EA, effect allele; EAF, effect allele frequency; NEA, non-effect allele.

*rs1974490 was only available in the 1000G reference panel and was not imputed in the European studies used in the replications.

To extend power to detect associations of lesser effect and to replicate the PLCB4 association, we identified 139 loci with SNP associations exceeding P ≤ 5 × 10⁻⁶ in at least one of the seven dichotomous DKD analyses. We also identified 22 loci (25 lead variants) for replication from the eGFR analysis (based on either P < 5 × 10⁻⁶ in our eGFR analyses alone or P < 5 × 10⁻⁴ in our analysis and a genome-wide association [P < 5 × 10⁻⁸] reported by the CKDGen Consortium) (Supplementary Fig. 3Q) (13). We sought replication for 164 lead variants in 13 studies of T2D DKD (9 involving European subjects and 4 involving Asian subjects) for which it was possible to obtain association analyses based on either in silico (from existing GWAS) or de novo genotyping (Fig. 1). Replication studies recoded their DKD phenotypes to align definitions with those used in the SUMMIT GWAS. Although association results for the lead variants were recovered for all compatible DKD phenotypes available in the replication samples (Supplementary Table 1), joint meta-analysis results are reported for only those phenotypes where the primary GWAS associations exceeded the thresholds above (Supplementary Table 5). The replication samples available for the all DKD phenotype included up to 3,999 T2D subjects of European ancestry (1,270 case subjects) and 17,111 (8,095 case subjects) from Asia (Supplementary Table 1).

The CKD association near PLCB4 did not replicate in either European or Asian data (joint analysis, OR_Asian+Euro 1.12 [95% CI 1.05, 1.19]; P = 2.1 × 10⁻⁴) (Table 2). Joint analysis of dichotomous DKD phenotypes identified one novel SNP association that marginally exceeded genome-wide significance (P = 5 × 10⁻⁸, without adjustment for the multiple GWAS we performed) (Table 2). This signal, on chromosome 6, is centered on rs9942471 and lies ∼7 kb upstream of GABRR1 (encoding the rho1 subunit of the GABA type a receptor). The major allele was associated with increased risk of microalbuminuria in subjects of European ancestry (joint analysis, EAF 64%, OR_Euro 1.25 [95% CI 1.16, 1.34]; P = 4.5 × 10⁻⁸) (Fig. 2 and Table 2). Associations of rs9942471 with other DKD phenotypes are given in Supplementary Table 6.

Figure 2.

A: Manhattan plot of P values from the meta-analysis of allelic effects on early DKD in subjects with T2D of European descent. The red line represents genome-wide significance (P < 5 × 10⁻⁸) and the blue line suggestive significance (P < 1 × 10⁻⁶). The peak represented by rs9942471 (P = 4.5 × 10⁻⁸) near GABRR1 is highlighted in orange. B: A forest plot of allelic OR and imputation information scores (RSQ) from individual studies that contributed to the discovery and replication analyses of rs9942471 in microalbuminuria phenotype; rs9942471 genotypes were not available in Steno. C: A LocusZoom plot of the signal near GABRR1 led by rs9942471 that was associated with microalbuminuria in European subjects with T2D.

rs9942471 is in high linkage disequilibrium (LD) (r² >0.8) with the lead expression quantitative trait locus variant for GABRR1 expression in the artery, esophagus, and skin (P ≤ 4 × 10⁻⁸), and the major allele is associated with decreased expression (42). However, there was no evidence for replication of this SNP in T2D subjects of Asian ancestry only (EAF 90%, OR_Asian 0.99 [95% CI 0.87, 1.13]; P = 0.91), although the higher frequency of the effect allele in Asians (90%) compared with Europeans (64%) reduces the power to detect an effect in subjects of Asian descent. Ethnic differences in regional LD could have contributed to failed replication: rs9942471 may be a better marker of the shared causal variant in subjects of European descent. However, this seems unlikely given broad similarity of LD patterns across subjects of European and Asian descent (estimated separately from the 1000G population).

Replication samples for the eGFR phenotype included 8,749 subjects of European and 9,071 subjects of Asian ancestry with T2D (Fig. 1). Joint analysis of discovery and replication results captured the well-established association with variants near UMOD (uromodulin), centered on rs11864909 (β_Asian+Euro 2.34 [95% CI 1.68, 3.00] mL/min/1.73 m²; P = 4.4 × 10⁻¹²) (Table 2). There was no difference in effect by diabetes type: the effect estimate in subjects with T1D (β_T1D 1.23 (−0.05, 2.51); P = 0.06) overlapped the effect size in subjects with T2D (17). We also compared the effects of variants associated with DKD phenotypes in subjects with T2D (Table 1) with their effects in equivalent DKD phenotypes in subjects with T1D (17) (Supplementary Table 7).

Combined T1D+T2D Analysis

To increase power to detect loci that contribute to processes involved in the development of DKD irrespective of diabetes subtype, we combined the results from the primary GWAS meta-analysis for T2D-DKD phenotypes with those for the corresponding T1D-DKD phenotypes (Supplementary Tables 4 and 9) (17). The combined discovery meta-analysis of all DKD included 10,873 subjects with diabetes of European descent (5,908 case subjects) and provided >80% power (α = 5 × 10⁻⁸) to detect a SNP association with an allelic OR >1.25 for variants with MAF >8% (Supplementary Fig. 2B). The number of variants meta-analyzed ranged from 7,959,015 for ESRD vs. no ESRD to 9,364,702 for the all DKD phenotype (Supplementary Table 4).

No significant associations were detected for dichotomous DKD phenotypes in the combined T1D+T2D meta-analysis (Supplementary Fig. 4 and Supplementary Table 9). The combined meta-analysis for eGFR highlighted a novel genome-wide significant association involving a cluster of variants on chromosome 2 led by rs1974990 (EAF 8%, β 4.07 [95% CI 2.61, 5.52] mL/min/1.73 m²; P = 4.8 × 10⁻⁸) and mapping near SSB (encoding Sjogren syndrome antigen B) (Table 2).

As in the T2D-only analysis, we selected 47 loci for replication (30 with P < 1 × 10⁻⁶ with at least one of the DKD phenotypes) from the combined T1D+T2D GWAS and an additional 17 loci from the equivalent analysis of eGFR. The combined association P value for rs9942471 (microalbuminuria, OR 1.10 [95% CI 1.02, 1.19]; P = 0.001) did not reach the threshold for replication. Lead variants at these 47 loci were tested for all DKD phenotypes available in the relevant replication samples in subjects with T1D or T2D (Supplementary Table 1). Meta-analysis results were only reported for those phenotypes that contributed to discovery-stage associations. This joint, combined T1D+T2D analysis generated a substantially enlarged data set for the all DKD phenotype (40,640 subjects [18,582 case subjects]) (Fig. 1). However, none of the variants selected for replication from the dichotomous phenotypes reached genome-wide significance (P ≤ 5 × 10⁻⁸).

The joint, combined analysis for eGFR in subjects of European and Asian descent included 31,562 subjects and replicated known associations near UMOD (rs11864909, β_Asian+Euro 2.11 [95% CI 1.52, 2.70]; P = 2.3 × 10⁻¹²) and PRKAG2 (rs10224002, β_Asian+Euro 2.01 [1.30, 2.72]; P = 2.7 × 10⁻⁸) (Table 2 and Supplementary Figs. 5 and 6). The PRKAG2 was nonsignificant (P ≤ 5 × 10⁻⁸) in individual analyses of eGFR in T2D-only (β_Euro 2.13 [95% CI 1.28, 2.98]; P = 8.5 × 10⁻⁷) or T1D-only (β_Euro 1.23 [−0.19, 2.65]; P = 0.09) analyses, and effect sizes did not differ by type of diabetes.

The association at SSB, detected in the combined eGFR analysis, did not replicate (rs1974990, β 0.04 [95% CI −2.69, 2.76] mL/min/1.73 m²; P = 0.98) and was no longer genome-wide significant in the joint, combined analysis (β_Asian+Euro 3.17 [1.88, 4.45] mL/min/1.73 m²; P = 1.4 × 10⁻⁶) (Table 2).

Evaluating Previous Association Claims

Of the 61 published loci, for which there are published claims of association with T1D DKD or T2D DKD (8), 55 of these associations were represented by variants contributing to our meta-analyses of DKD phenotypes in either subjects with T2D only or T1D+T2D. Two of these, the eGFR associations at UMOD and PRKAG2, replicate at genome-wide significance in our data (Table 2). We tested the association of the remaining 53 lead variants in the T2D-only and combined analyses (Supplementary Fig. 6). Fourteen variants were associated with a DKD phenotype corresponding to the original report at nominal significance (P < 0.05), but only 10 of these were directionally consistent with previous reports (Supplementary Table 10). At a more stringent significance level (P < 9 × 10⁻⁴) that accounts for the 55 variants tested (but not the multiple phenotypic categories), only 2 variants were associated with a DKD phenotype that corresponded to the original report, both in the combined T1D+T2D analysis and both directionally consistent with previous reports. These two SNPs were rs2838302, near SIK1, associated with ESRD vs. no ESRD (EAF 8%, OR 1.39 [95% CI 1.12, 1.74]; P = 3.9 × 10⁻⁴), and rs7583877, near AFF3, associated with ESRD vs. no ESRD (OR 1.22 [95% CI 1.13, 1.32]; P = 4.8 × 10⁻⁴) (Supplementary Table 10). When we took account of the substantial participant overlap between the original reports and the samples in the current study, apparent replications failed to reach nominal (P < 0.05) significance (though, for these, the sample sizes available for independent replication were often small). Thus, other than the eGFR associations at UMOD and PRKAG2, we found limited evidence in this study to corroborate previously reported DKD associations, despite, for most variants, sample sizes considerably larger than those included in the original report. Validation of previously reported DKD associations could be complicated by differences in phenotype definitions and/or analytical methods between this study and published reports. We could not assess whether the UMOD or PRKAG2 allelic effects were different in this study compared with those reported by CKDGen Consortium as the allelic effects were not on the same scale (e.g., untransformed vs. log transformed).

Genetic Overlap With Risk Factors

Several exposures and diseases have been reported to increase DKD risk in epidemiological studies (1,2,43). To explore the extent to which these reflect shared genetic background, we constructed weighted GRS for 20 traits related to diabetes (37,39–41), insulin resistance (38), obesity (34,35), hypertension (42), coronary artery disease (33), and lipids (36). These GRS, constructed from signals identified (P < 5 × 10⁻⁸) in previously published GWAS, included between 10 and 96 SNPs per phenotype. We tested the association of these GRS with each of the DKD phenotypes from this study in both T2D-only and T1D+T2D data sets (42).

After Bonferroni correction (P ≤ 2.5 × 10⁻³, which accounts for the number of trait GRS but not the number of DKD phenotypes), a GRS for increased waist-to-hip ratio (P = 4.8 × 10⁻⁴) was associated with increased risk of ESRD vs. no ESRD phenotype and a GRS for increased BMI was associated with all DKD (P = 1.8 × 10⁻⁴) and late DKD (P = 1.8 × 10⁻³) phenotypes in subjects with T2D. A similar pattern of association for the BMI GRS was observed in the combined T1D+T2D all DKD analysis (P = 2.4 × 10⁻⁵) (Supplementary Table 11 and Fig. 3). This last result survives additional correction (α = 1.6 × 10⁻⁴) for the 16 DKD phenotypic comparisons considered.

Figure 3.

A heat map of GRS associations with DKD phenotypes in subjects with either T1D or T2D. A GRS for BMI was significant after correction for multiple testing, whereas other traits, including systolic blood pressure, were not associated with DKD phenotypes. adj., adjusted; FG, fasting glucose; In, insulin; IR, insulin resistance.

There is evidence implicating insulin resistance in the pathogenesis of DKD, and we wanted to understand whether the BMI GRS associations might reflect obesity-related insulin resistance (44,45). We focused on the effects of two alternative GRS for insulin resistance on DKD. The first, comprising lead variants (N = 10) associated with increased fasting insulin (BMI adjusted) (37), was associated with increased risk of ESRD in subjects with T2D (ESRD vs. no ESRD P = 1.6 × 10⁻³; ESRD vs. control subjects P = 1.7 × 10⁻³) (Supplementary Table 11 and Fig. 3). The second, comprising lead variants from 53 loci associated with high fasting insulin (BMI adjusted), low HDL cholesterol, and high triglycerides (39), failed to show any association with DKD phenotypes. These findings provide some support for the causal contribution of insulin resistance and obesity to DKD pathogenesis. However, there is potential that some of these effects reflect collider bias (46), and additional larger studies will be required to substantiate this inference.

Discussion

This study represents the largest study of the genetic basis of DKD in subjects with T2D to date, extending previous reports with respect to sample size and range of DKD phenotypes. We aimed to overcome some of the limitations of earlier studies in this area and to develop insights into the pathogenesis of DKD. Despite sample sizes that exceeded 40,000, the yield of novel discoveries was modest. There were no significant (P < 5 × 10⁻⁸) genetic associations with all DKD that was best-powered definition on sample size. The relatively large sample size came with increased phenotypic (and likely genetic) heterogeneity: it was for this reason that we examined a range of DKD phenotypes that might offer better power to detect genetic associations with more restricted phenotypic impacts.

This approach successfully identified a novel locus, GABRR1 (led by rs9942471), for a microalbuminuria phenotype in European subjects with T2D. The variants, near GABRR1, reached a level of significance (P < 5 × 10⁻⁸) that has typically been associated with robust, reproducible association in common disease GWAS. GABRR1 expression is upregulated in renal biopsies from DKD subjects (compared with control subjects) and in other non-DKD subjects characterized by glomerular scarring and inflammation (47). The variants were associated with GABRR1 expression in aorta, esophageal mucosa, and skin in the Genotype-Tissue Expression (GTEx) project. However, we found no replication of the GABRR1 association in subjects of European ancestry with T1D DKD or in subjects of Asian ancestry with T2D, though differences in risk allele frequencies between these two ancestries and the modest size of the replication data sets at this locus reduce the power of the latter analysis. Our overall assessment is that this association should be considered provisional until it is possible to undertake further rounds of adequately powered replication that could establish the definitive status of this variant and that this locus should also be assessed for effects on DKD progression in longitudinal studies.

Even in the absence of specific signals of association with DKD, it is possible to use the aggregate pattern of association across the genome to identify more subtle genetic effects. The GRS analyses described here provide genetic support for the causal contribution of obesity to the development of T2D DKD. This echoes strong epidemiological data, and mirrors equivalent analyses in T1D DKD (48,49). However, we cannot exclude that these associations may partly reflect collider bias (46): subjects with high BMI are likely to have a longer duration of diabetes and thus a higher chance of developing complications. Analyses using genetic instruments (GRS) for variation in insulin sensitivity produced variable results with respect to T2D DKD but indicate that the BMI effects may be partially mediated via obesity-related insulin resistance (37). There are substantial epidemiological data to support this link between insulin resistance and DKD risk (44,45).

The modest yield of association signals and the limited replication of previous claims of DKD association emphasizes challenges associated with the identification of DKD risk variants. For many complex traits, these have been overcome through a combination of increased sample size and phenotypic precision. Published genetic association studies of DKD have often used different definitions of DKD, which makes replication of previous findings difficult. In this study, we used phenotype definitions aligned to those used in the study of DKD in subjects with T1D (17). Standardizing the phenotype definitions in this way allowed for seamless combination of the GWAS data across the two studies and may streamline subsequent efforts to study the genetics of DKD. The phenotype definitions applied to this study address some of the challenges associated with increasing sample size while maintaining phenotype precision and should, in due course, support the identification of robust associations with DKD. It is clear that these phenotype definitions are not without limitations: in the absence of strong genetic signals, we have few clues to which particular diagnostic configurations will be most productive for genetic discovery. Targeting the phenotypes that show the greatest heritability may provide a guide (14).