Genome-wide meta-analysis and omics integration identifies novel genes associated with diabetic kidney disease

Aims/hypothesis Diabetic kidney disease (DKD) is the leading cause of kidney failure and has a substantial genetic component. Our aim was to identify novel genetic factors and genes contributing to DKD by performing meta-analysis of previous genome-wide association studies (GWAS) on DKD and by integrating the results with renal transcriptomics datasets. Methods We performed GWAS meta-analyses using ten phenotypic definitions of DKD, including nearly 27,000 individuals with diabetes. Meta-analysis results were integrated with estimated quantitative trait locus data from human glomerular (N=119) and tubular (N=121) samples to perform transcriptome-wide association study. We also performed gene aggregate tests to jointly test all available common genetic markers within a gene, and combined the results with various kidney omics datasets. Results The meta-analysis identified a novel intronic variant (rs72831309) in the TENM2 gene associated with a lower risk of the combined chronic kidney disease (eGFR<60 ml/min per 1.73 m2) and DKD (microalbuminuria or worse) phenotype (p=9.8×10−9; although not withstanding correction for multiple testing, p>9.3×10−9). Gene-level analysis identified ten genes associated with DKD (COL20A1, DCLK1, EIF4E, PTPRN–RESP18, GPR158, INIP–SNX30, LSM14A and MFF; p<2.7×10−6). Integration of GWAS with human glomerular and tubular expression data demonstrated higher tubular AKIRIN2 gene expression in individuals with vs without DKD (p=1.1×10−6). The lead SNPs within six loci significantly altered DNA methylation of a nearby CpG site in kidneys (p<1.5×10−11). Expression of lead genes in kidney tubules or glomeruli correlated with relevant pathological phenotypes (e.g. TENM2 expression correlated positively with eGFR [p=1.6×10−8] and negatively with tubulointerstitial fibrosis [p=2.0×10−9], tubular DCLK1 expression correlated positively with fibrosis [p=7.4×10−16], and SNX30 expression correlated positively with eGFR [p=5.8×10−14] and negatively with fibrosis [p<2.0×10−16]). Conclusions/interpretation Altogether, the results point to novel genes contributing to the pathogenesis of DKD. Data availability The GWAS meta-analysis results can be accessed via the type 1 and type 2 diabetes (T1D and T2D, respectively) and Common Metabolic Diseases (CMD) Knowledge Portals, and downloaded on their respective download pages (https://t1d.hugeamp.org/downloads.html; https://t2d.hugeamp.org/downloads.html; https://hugeamp.org/downloads.html). Graphical abstract Supplementary Information The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-022-05735-0.


Introduction
Diabetes is the leading cause of kidney disease. Diabetic kidney disease (DKD) is associated with high cardiovascular risk [1] and mortality [2] and, consequently, both diabetes and kidney disease are leading causes of death worldwide [3]. Both environmental and genetic factors have a major impact on the risk of developing DKD [4,5]. Although more than 300 genetic loci have been associated with chronic kidney disease (CKD) in the general population, these loci show limited effect in DKD, especially in individuals with type 1 diabetes [6]. Genomewide association studies (GWAS) have previously identified a handful of genetic loci for DKD at the genome-wide significance level (p<5×10 −8 ) [7][8][9][10][11]. Recently, a meta-analysis of GWAS, including up to 19,406 individuals with type 1 diabetes from the Diabetic Nephropathy Collaborative Research Initiative (DNCRI), identified 16 loci. The strongest association was a common missense variant in the COL4A3 gene, which also showed evidence of association in individuals with type 2 diabetes [6]. A GWAS meta-analysis from The SUrrogate markers for Micro-and Macrovascular hard endpoints for Innovative diabetes Tools (SUMMIT) consortium, including 6000 individuals with type 2 diabetes from five different studies, identified three loci for DKD, including UMOD and PRKAG2 previously identified in the general population [12]. However, meta-analysis with SUMMIT type 1 diabetes studies (SUMMIT-1, N=5156) did not yield any genome-wide significant findings. To increase the power to detect novel genetic risk factors for DKD shared among diabetes subtypes, we aggregated all available data for DKD in individuals of European ancestry with type 1 or type 2 diabetes (N~27,000). Specifically, we performed GWAS meta-analyses on ten different DKD case-control definitions, meta-analysing summary statistics from DNCRI [6], SUMMIT-1 [4] and SUMMIT type 2 diabetes studies (SUMMIT-2) [12], followed by integration with diverse biological data to improve our understanding of the underlying biological mechanisms and clinical correlations (Fig. 1).

Methods
For detailed methods, please refer to the electronic supplementary material (ESM) Methods.
Participating studies and phenotype definitions A total of ten case-control definitions for DKD were included in DNCRI [6], based on either urinary AER (divided into controls with normal AER, and cases with microalbuminuria, macroalbuminuria or end-stage renal disease [ESRD]), eGFR, or both, and harmonised to match and include all seven phenotypic definitions assessed in SUMMIT-1 [4] and SUMMIT-2 [12] analyses (ESM Table 1). All individuals (both cases and controls) had diabetes (either type 1 or type 2 diabetes). For three phenotypic comparisons not initially part of the SUMMIT analysis, GWAS and metaanalysis were performed with three SUMMIT-2 studies and the Scania Diabetes Registry type 1 diabetes cohort. Individuals from the Finnish Diabetic Nephropathy Study (FinnDiane) were included in both the original DNCRI (N=6019) and SUMMIT-1 analyses (N=3415) and thus were excluded here from the SUMMIT-1 data (ESM Table 2). All contributing studies were performed in accordance with the Declaration of Helsinki and Declaration of Istanbul.

Genome-wide association study and meta-analysis
Genotyping and statistical analysis of the DNCRI [6] and SUMMIT [4,12] cohorts have been previously described. Analysis plans were similar in the cohorts (ESM Table 3 Power calculations indicated 80% power to detect associations with p<5×10 −8 and an OR of 1.20 for the combined 'all vs ctrl' phenotype for common variants with MAF≥10%, or with an OR of 1.28 and 1.73 for low-frequency (MAF 5%) and rare (MAF 1%) variants, respectively (ESM Fig. 1).
Correction for multiple testing was estimated with spectral decomposition of the ten DKD traits, suggesting 5.36 effective tests, leading to a corrected significance threshold of p<9.3×10 −9 .
Gene prioritisation analysis Gene prioritisation at the top loci was performed using two complementary similarity-based gene prioritisation approaches (Polygenic Priority Score [PoPS] v0.1 [13] and MAGMA v1.06b [14]), which integrate GWAS with gene set enrichment based on a variety of biological annotation datasets.
Gene-level analysis SNPs from the GWAS meta-analysis summary statistics were aggregated by gene-level regression analysis using two related programs, MAGMA v1.06b [14] and PASCAL v2016 [15], using default parameters. Genelevel significance thresholds were determined by a Bonferroni multiple testing correction based on the number of genes tested for each of the ten phenotypes (from p<2.7×10 −6 to p<2.3×10 −6 ).
Transcriptome-wide association study In the transcriptomewide association study (TWAS), MetaXcan [16] was applied with default parameters to integrate GWAS meta-analysis results with kidney expression quantitative trait locus (eQTL) datasets for micro-dissected human glomerular (N=119) and tubular (N=121) samples [17]. Significance threshold of p<4.1×10 −6 was determined by Bonferroni correction for two tissues and 6050 genes found in either tubular or glomerular eQTL data.
Human kidney gene expression For the 29 lead genes, we studied gene expression in kidneys in human transcriptomics data from nephrectomy samples (433 tubule and 335 glomerulus samples) [22] and kidney biopsies from the Pima Indian cohort (67 glomerular and 47 tubulointerstitial tissues) [23], and tested for correlation with relevant pathological phenotypes. The micro-dissected nephrectomy samples were from individuals with varying degree of diabetic and hypertensive kidney disease, and gene expression was defined with RNA sequencing. The study was approved by the institutional review board of the University of Pennsylvania.
In the Pima Indian cohort, gene expression profiling in the first biopsy was performed with Affymetrix gene chip arrays [23], and with Illumina RNA sequencing for the second biopsy [6]. Available phenotypes included progression to ESRD, measured GFR (mGFR), albumin/creatinine ratio (ACR), HbA 1c and six kidney morphological variables for both biopsies, and change in the phenotypes between the first and the second study biopsies (27 phenotypes in total [24]). The study was approved by the Institutional Review Board of the National Institute of Diabetes and Digestive and Kidney Diseases.

GWAS meta-analysis
The GWAS meta-analysis of the DNCRI (type 1 diabetes), and SUMMIT-1 and SUMMIT-2 metaanalyses included up to 26,785 individuals with either type 1 or type 2 diabetes from 25 studies; 11,380 individuals had any DKD (micro-or macroalbuminuria or ESRD) and 15,405 individuals had normal AER (ESM Table 2). QQ plots, λ genomic control inflation factor (λ GC ) and LDSR intercepts of the meta-analysis indicated no marked inflation or population stratification bias of the results (ESM Fig. 2).
The meta-analysis identified a novel association between the combined CKD-DKD phenotype (cases with eGFR <45 ml/ min per 1.73 m 2 and microalbuminuria or worse, vs controls with normal AER and eGFR ≥60 ml/min per 1.73 m 2 ) and rs72831309 (MAF=4%; OR 2.08, p=9.8×10 −9 ; Fig. 2a, Table 1 and ESM Table 5). Of note, the association was barely above the threshold after correction for multiple testing due to multiple phenotypes (p>9.3×10 −9 ). We observed no heterogeneity between individuals with type 1 or type 2 diabetes (p HET =0.88). The variant is located in an intron of the TENM2 gene encoding the teneurin transmembrane protein 2. The variant was imputed with moderate imputation quality across cohorts (INFO score 0.38-0.66). In the FinnDiane cohort with the strongest statistical significance, the association remained (though slightly attenuated) after re-imputation with a population-specific panel ( At the previously identified COL4A3 locus, we identified a secondary association peak (rs6436688, p=1.8×10 −7 for severe DKD; ESM Fig. 3) in only partial LD (D′=0.51, r 2 =0.08) with the lead variant rs55703767. The association at rs6436688 remained nominally significant after conditional analysis for rs55703767 (p=0.002).
In addition to the COL4A3 locus, nine other previously identified, mostly low-frequency or rare variants were associated with various kidney phenotypes (Table 1 and ESM Fig. 4). None of these variants were found in the SUMMIT metaanalyses (filtered to MAF≥1%), and thus, these associations represent the originally reported associations from the DNCRI [6]. One common (chr14q12), and four of our previously identified rare DNCRI loci (TAMM41, HAND2-AS1, DDR1-VARS2, BMP7; MAF~1%) associated with microalbuminuria demonstrated attenuated association when combined with the SUMMIT meta-analyses (with the rare variants only found in SUMMIT-2). The lack of replication across diabetes subtypes suggest either false positives or a lack of power to detect an association for rare variants in individuals with type 2 diabetes DKD (a more heterogeneous disease, particularly for the early  . Alternatively, this could simply represent a lack of shared biology across diabetes subtypes, again possibly due to the different underlying causes of kidney damage in individuals with type 1 vs type 2 diabetes. Two variants, chr3:141792314:I and rs186434345, were associated with ESRD (N=940, p=4.6×10 −10 ), and with the CKD-DKD phenotype (N=2571, p=4.0×10 −8 ) in the SUMMIT-2 and SUMMIT-1 cohorts, but were absent in DNCRI. When the original SUMMIT-FinnDiane GWAS was included in the analysis, both associations were non-significant (chr3:141792314:I, p=0.056, N=3207; rs186434345, p=0.002, N=4782) and thus excluded from further consideration.
Gene prioritisation To identify the underlying causal genes within each of our top loci, we used the PoPS [13] method that leverages genome-wide enrichment of biological annotations in combination with GWAS summary statistics to prioritise candidate genes. To increase precision, we intersected the results with both the simple nearest-gene approach and MAGMA gene prioritisation. Four genes (COL4A3, PLEKHA7, CNTN6 and TENM2) were both the PoPS prioritised gene and the nearest protein coding gene to the lead SNP (Table 1). Of note, the CNTN6 locus contained only two protein coding genes and the TENM2 locus only one. When taking the intersect between PoPS genes and genes that were within MAGMA's top 10% of prioritised genes genomewide, COL4A3 was the only prioritised gene (ESM Fig. 5). The gene set that prioritised COL4A3 for severe DKD was the fibulin 2 protein-protein interaction network ('FBLN2 PPI subnetwork'), which together with 26 correlated reconstituted gene-sets makes up the 'basement membrane' meta-gene set derived in Marouli et al [28] (ESM Table 6).
Gene-level analysis To improve power and jointly test all available common genetic markers within a gene, SNPs from the GWAS meta-analysis summary statistics were aggregated PODO, podocytes; T2D meta, meta-analysis of type 2 diabetes cohorts by gene and tested jointly for association using two similar programs, MAGMA and PASCAL. In addition to COL20A1 and SNX30 identified previously [6], we identified eight novel gene associations (p<2.7×10 −6 ; Table 2 and ESM Fig. 6). The lead variants in these loci indicated no significant heterogeneity between type 1 and type 2 diabetes apart from the GPR158 locus (p HET =0.005; ESM Table 7). MAGMA's gene-level analysis type 1 error was well controlled, with all but one λ GC inflation factor under 1.05 (MAGMA's ESRD vs all λ GC =1.07). The genome-wide gene-level results from PASCAL showed slightly more inflation (λ GC up to 1.15 for ESRD vs all).
Integration of GWAS with kidney eQTL data We performed TWAS for each of the ten DKD meta-analyses to predict differential gene expression between cases and controls based on eQTL data in glomerular and tubulointerstitial samples from histologically normal kidneys [17]. The type 1 error was well controlled (λ GC 0.968-1.097; ESM Fig. 7). Expression levels of AKIRIN2 were predicted to be higher in the tubular tissue of cases with severe DKD (or macroalbuminuria alone), as compared with controls with normal AER (p values 1.1×10 −6 and 1.7×10 −6 , respectively; Fig.  3a and ESM Tables 8, 9).   Table 10). Altogether, 17 variants were significantly associated with kidney DNA methylation levels at six CpG sites (p<1.5×10 −11 ; ESM Table 11), of which the mQTL colocalised with the GWAS association in LSM14A, DCLK1 and COL20A1 (posterior probability for colocalisation >0.80). SNPs in the LSM14A gene were associated with severe DKD and cg14143166 methylation levels (p=1.9×10 −28 ). Interestingly, cg14143166 methylation in blood was nominally associated with DKD status in our EWAS in the UK-ROI and FinnDiane cohorts (p=0.03), suggesting that the DKD association at LSM14A is mediated through methylation changes. Similarly, blood methylation levels at significant kidney mQTL CpG sites (rs7664964-cg25974308 p=1.1×10 −11 ) in EIF4E were nominally associated with eGFR slope in diabetes (p=0.04) [29].
Genetic correlation of DKD between type 1 and type 2 diabetes and general population kidney traits We performed LDSR to study the genetic correlation of DKD traits Association with DKD (diabetic nephropathy) was queried in two data sets (Woroniecka et al [36] and Ju et al [35]), with p<4.3×10 −4 or p<0.05 and fold change>1.5. BX1 Fibr, fibrosis at first biopsy; BX1 ΔMesV, change in the mesangial volume between the first and the second biopsies; DN Wor, diabetic nephropathy in Woroniecka et al [36]; DN Ju, diabetic nephropathy in Ju et al [35]; GlomScl, glomerulosclerosis between individuals with type 1 and type 2 diabetes but, likely due to limited sample size, no significant correlations were observed. When compared with the kidney traits from the CKDgen consortium, the 'All vs Ctrl' phenotype was correlated with microalbuminuria in the general population [30] and ACR in diabetes [30], both in the main meta-analysis and for type 1 and type 2 diabetes separately (p<0.01). In addition, microalbuminuria in type 2 diabetes was correlated with microalbuminuria in the general population [30], and CKD in type 2 diabetes Fig. 5 Genetic correlation between DKD phenotypes (y-axis) and kidney phenotypes in the general population (x-axis). Correlations were calculated with LD score regression for the whole meta-analysis (any diabetes, purple), type 2 diabetes only (red), and type 1 diabetes only (blue). The first column (purple) indicates genetic correlation for the DKD phenotypes between individuals with type 1 or type 2 diabetes (none significant). Only significant correlations (p<0.01) are shown. General population GWAS results were taken from CKDgen consortium: ACR [30]; ACR in diabetes [30]; microalbuminuria [30]; eGFR [32]; and CKD [31].  inverse variance-weighted method for the traits significant in LDSR ('mother's age at death' had fewer than than five genome-wide significant SNPs and thus, was not included in MR). Horizontal bars represent 95% CI. Ctrl, control was positively correlated with CKD in the general population [31] and negatively with eGFR in the general population [32]; these were not significantly correlated in individuals with type 1 diabetes despite a larger number of samples (Fig. 5).
Genetic correlation and MR with related traits LDSR of related metabolic traits revealed significant genetic correlation (p<6.4×10 −4 ) between DKD and 15 traits including multiple obesity-related traits, mother's age at death, type 2 diabetes, coronary artery disease, HDL-cholesterol, urate, and two smoking-related traits ( Fig. 6 and ESM Fig. 9). MR of these traits suggested that being overweight or obese was a causal risk factor for DKD ( Fig. 6b and ESM Table 14). The causal effects were directionally consistent across methods, with no evidence of heterogeneity (I 2 =0-42.9%, p>0.05; ESM Table 14) or unbalanced horizontal pleiotropy (ESM Table 15). The MR Egger method, more robust for pleiotropic effects, further supported a causal role for higher BMI, waist circumference and hip circumference in DKD risk (p<0.05; ESM Table 14, ESM Fig. 10).

Discussion
We have performed the largest GWAS meta-analysis to date on kidney complications in diabetes, including ten different phenotypic definitions in up to 26,785 individuals with either type 1 or type 2 diabetes, and integrated the results with emerging kidney omics data (Fig. 1). In the single-variant analysis with the combined CKD-DKD phenotype, we identified one novel locus, rs72831309, intronic in TENM2. TENM2 encodes the teneurin transmembrane protein 2 involved in cell-cell adhesion. The variant rs72831309 alters a predicted CREB1 transcription factor binding site (Fig. 2c), and is nominally associated with expression of a TENM2 antisense transcript TENM2-AS1 in kidneys (p=0.007; ESM Table 10). Furthermore, chromatin conformation data in the GM12878 cell line indicated that the rs72831309 region interacts with the TENM2 transcription start site, as well as with three antisense transcripts ( Whereby kidney single-cell RNA sequencing indicated TENM2 expression particularly in podocytes (Fig. 2d) [34], lower tubular TENM2 expression was associated with renal fibrosis (p=2.0×10 −9 ) and lower eGFR (p=1.6×10 −8 ) in the nephrectomy samples. Despite multiple supporting lines of evidence, the locus still needs further validation as the imputation quality of rs72831309 was on the low end across our cohorts (0.38-0.66), and the association did not remain significant after correction for multiple testing (p>9.3×10 -9 ). Gene-level analysis identified ten genes associated with DKD. The DCLK1 gene encodes a doublecortin-like kinase. The histone modification-based ChromHMM 15-state model for fetal kidney indicated strong transcription overlapping one of the three lead SNPs in the DCLK1 locus (rs61948262), and ChIP-seq data supported ZSCAN4 binding to the locus in the HEK293 kidney epithelial cell line. In addition, the lead SNPs were kidney mQTLs for DCLK1 CpG sites (p=6.8×10 -22 ). Furthermore, multiple lines of evidence highlight the importance of DCLK1 in DKD. The correlation between tubular DCLK1 expression and fibrosis was among the strongest correlations both in the nephrectomy samples (p=7.4×10 −16 ; Fig. 7) and in the Pima Indian biopsies (p=3.0×10 −4 ), and glomerular DCLK1 expression was nominally associated with glomerular width, mesangial volume and podocyte foot process width in the Pima Indian biopsies (p<0.05; ESM Table 13). Furthermore, both glomerular and tubular DCLK1 expression were elevated in DKD in two additional datasets (fold change 1.98, p=1.2×10 −4 for glomeruli [35]; fold change 2.09, p=0.003 for tubules [36]; Fig. 7). Finally, we previously identified a subset of transcripts, including DCLK1, targeted by the early growth response-1 transcription factor in a murine model of DKD. In that study, Dclk1 expression was upregulated in diabetic vs non-diabetic Apoe −/− mouse kidneys [37]. Taken together, these expression data in human and experimental DKD identify DCLK1 as a novel target.
The TWAS analysis based on our GWAS results, integrated with micro-dissected tubular and glomerular eQTL data, predicted that AKIRIN2 gene expression is elevated in tubules in individuals with severe DKD compared with individuals with normal AER (p=1.1×10 -6 ). AKIRIN2 gene expression was highly correlated with the level of fibrosis (p=2.8×10 −7 ). AKIRIN2 encodes a conserved nuclear factor that is a downstream effector of the toll-like receptor, TNF and IL-1β signalling pathways, involved in stimulating proinflammatory pathways [38]. This factor binds to nuclear NF-κB complexes and is required for the transcription of a subset of NF-κB-dependent genes such as IL6, CXCL10 and CCL5 [39]; NF-κB activation drives inflammatory responses and is activated in DKD [40].
The strongest regulatory evidence in RegulomeDB was obtained for rs1260634 intronic in the LSM14A gene: rs1260634 exerts strong transcription in 125 tissues including fetal kidney chromatin state model, is located in a ZNF362 binding sequence in HEK293 cell line, and affects a predicted transcription factor binding motif for Kruppel-like factors 4 and 12 (KLF4 and KLF12) and Sp8 transcription factor (ESM Fig. 11). Furthermore, in our kidney mQTL data, rs1260634 showed strong association (p=2.1×10 −28 ) with cg14143166, where methylation in blood was associated with DKD in our EWAS data (p=0.03). Tubular LSM14A expression correlated with higher eGFR (p=2.9×10 −6 ), and glomerular expression with the decrease in mesangial volume (p=6.5×10 −4 ; significant after correction for 29 tested genes and two tissues, but not for 27 phenotypes). LSM14A encodes an Sm-like protein, thought to participate in pre-mRNA splicing, and implicated in innate antiviral responses [41].
Other noteworthy novel genes include EIF4E and PTPRN. EIF4E encodes a common mRNA translation initiation factor; its activation and/or suppression are influenced by mTOR signalling cascades involved in DKD [42] as well as high glucose and high insulin environments in renal epithelial cells [43]. PTPRN encodes islet antigen 2 (IA-2), a major type 1 diabetes autoantigen involved in glucose-stimulated insulin secretion [44]. In mice, IA-2 is required to maintain normal levels of renin expression in kidneys [45]. Finally, the MFF gene identified in our gene-level analysis has been previously related to DKD [46]. However, the association may be driven by the neighbouring COL4A3 association, as suggested previously [6].
Indeed, one limitation of our gene-level analysis is the inability to confidently assign genes to a given set of correlated SNPs within a region. While it is reasonable to prioritise the gene in which the SNPs lie, it remains possible that extended LD patterns are tagging other nearby genes. Similarly, assigning a causal gene for the lead SNPs is not straightforward. We have utilised the eQTL and other data when available but also used a simple nearest-gene approach to name the associated region.
In the transcriptomics analyses, as expected, all 13 significant correlations with the level of fibrosis were observed for tubular gene expression, and the two observed correlations for glomerulosclerosis were for glomerular expression of SNX30 and COLEC11 (Fig. 4). Interestingly, eight out of ten correlations with eGFR were obtained for tubular rather than glomerular gene expression, supporting the importance of tubular damage in the loss of renal function.
While there is a strong epidemiological link between DKD and coronary artery disease in diabetes [1], our LDSR is the first study to report a genetic correlation between these major diabetic complications. Among the lipid traits, significant correlation with DKD was found only for lower HDL-cholesterol, despite previous MR of kidney disease in the general population implicating HDL as a marker of dyslipidaemia rather than a causal factor [47]. Indeed, our subsequent MR found no evidence of causality between HDL and DKD; in concordance with our previous MR on BMI [48], only obesity-related traits were causal risk factors for DKD. However, we cannot exclude that the associations detected in our studies might partly reflect collider bias. Of note, our current MR was in line with our previous MR in type 1 diabetes suggesting that serum urate levels are not a causal risk factor for DKD [49]; similar negative results were also reported for non-diabetic CKD [50].
Most kidney disease in individuals with type 1 diabetes is considered to occur due to diabetic nephropathy, histologically characterised by thickening of the glomerular basement membrane and mesangial expansion, as well as renal tubular, interstitial and arteriolar lesions. In individuals with type 2 diabetes, only a proportion of DKD is purely due to diabetic nephropathy, whereas ageing, obesity and hypertension also contribute to kidney decline. These differences were also seen in our genetic correlation analysis, with CKD in type 2 diabetes genetically resembling CKD and eGFR in the general population, and no significant correlation observed in individuals with type 1 diabetes. Thus, including individuals with type 2 diabetes in the meta-analysis increases the heterogeneity of the underlying disease. However, as type 2 diabetes represents 95% of all diabetes cases, including those individuals increases statistical power for our current work and future GWAS meta-analyses integrating multiple subtypes of diabetes to identify shared genetic risk factors for DKD.