Kidney Multiome-based Genetic Scorecard Reveals Convergent Coding and Regulatory Variants

Abstract

Kidney dysfunction is a major cause of mortality, but its genetic architecture remains elusive. Here, we conducted a multi-ancestry genome-wide association study in 2.2 million individuals and identified 1,026 (97 previously unknown) independent loci. Ancestry-specific analysis indicated an attenuation of newly identified signals on common variants in European ancestry populations and the power of population diversity for further discoveries. We defined genotype effects on allele-specific gene expression and regulatory circuitries in over 700 human kidneys and 237,000 cells. We found 1,363 coding variants disrupting 782 genes, with 601 genes also targeted by regulatory variants and convergence in 161 genes. Integrating 32 types of genetic information, we present the ‘Kidney Disease Genetic Scorecard’ for prioritizing potentially causal genes, cell types, and druggable targets for kidney disease.

Introduction

Kidney disease encompasses a range of conditions impairing the kidney’s ability to filter blood effectively. This impairment can lead to the accumulation of toxins and excess fluids in the body, which may result in serious health complications, such as hypertension (high blood pressure), electrolyte abnormalities, and cardiovascular disease. The increasing mortality rates associated with kidney disease underscore its importance as a major public health issue(1).

The etiology of kidney disease is complex, with contributing factors ranging from genetic disorders like polycystic kidney disease, to comorbidities including diabetes, high blood pressure, and autoimmune diseases. Although each cause affects the kidneys differently, they all contribute to a diminished filtration capacity. Glomerular filtration rate (GFR) is a critical measure used to assess this capacity and is typically estimated by measuring serum creatinine levels in the blood. Low eGFR is not only a marker of kidney function decline but it defines kidney disease.

Kidney function has strong heritable components(2). Pathogenic variants in the coding regions are primarily associated with monogenic forms of kidney disease, which are more commonly observed in children but very rarely detected in adults as they are presumed deleterious and subjected to the action of purifying selection at a population level. Recent exome-wide association studies (ExWAS) utilizing whole exome sequencing data have cataloged rare coding variants associated with kidney disease(3–5). While coding variants provide a strong indication of candidate causal genes – making these studies potentially fruitful for dissecting the molecular pathways underlying kidney function – they ultimately explain a very small fraction of common disease heritability in the population. In contrast, common variants identified by genome-wide association studies (GWAS) explain most of the heritability of common disease conditions. Prior GWAS have uncovered associations between common single nucleotide polymorphisms (SNPs) and kidney function(6–8). More than 90% of GWAS-identified common variants are situated in non-coding regions of the genome. Therefore pinpointing their target genes and regulatory functions is challenging—a problem often referred to as the ‘variant-to-gene’ or ‘variant-to-function’ issue.

Genetic variation associated with molecular traits – quantitative trait loci (QTLs) – has been used to investigate the regulatory functions of genetic variants across various human tissues, including the kidneys(9–14). By integrating transcript expression QTLs (eQTLs) and DNA methylation QTLs (meQTLs), our group suggested gene candidates for over 300 loci associated with kidney function(8). Despite the effectiveness of QTL mapping, pinpointing putatively regulatory variants within most GWAS loci remains a massive challenge. Allele-specific expression, which arises from genotypic differences between maternally and paternally inherited alleles, offers valuable insights for fine-mapping these elusive causal regulatory variants(15, 16). However, a comprehensive understanding of the genome-wide landscape of allelic imbalance in gene expression within human kidneys has yet to be examined.

Chromatin accessibility is pivotal for gene regulation. Single nucleus assay for transposase-accessible chromatin using sequencing (snATAC-seq) has identified cell type-specific regulatory elements in human kidneys(8, 12, 17, 18). However, investigating cell type-specific genetic effects on chromatin accessibility and their roles in regulating genes related to kidney disease remains a formidable challenge, often due to the absence of paired transcriptomic and open chromatin information at single cell level. Therefore, simultaneous profiling of chromatin accessibility and gene expression within the same cell could enhance our understanding of the regulatory variants implicated in human diseases.

The current study aimed to define the genetic architecture of GFR and use a multitude of complimentary omics datasets and tools to nominate regulatory variants, genes, and cell types for kidney function regulation. We show that each type of multiomic information can prioritize partially overlapping genes, cell types, and regulatory variants for each locus. We show the convergence of common non-coding and rare coding variants in specific genes and the key role of proximal tubules in kidney function maintenance. By integrating 32 genomic annotation and prioritization tools, we present the Kidney Disease Genetic Scorecard that can be applied for disease diagnostic and therapeutic development.

Results

Genetic discovery from 2.2 million individuals

We conducted a genetic analysis of kidney function, as measured by the estimated glomerular filtration rate based on serum creatinine (eGFRcrea), by integrating data from millions of individuals across 19 cohorts. First, we mapped genetic associations in 861,840 individuals (including 639,074 participants from European-ancestry, 136,620 East Asian-ancestry, and 34,879 African-ancestry populations) across nine cohorts (MVP, TWB, All of Us, MGB, GHS, BioMe, PMBB, eMERGE-III, and CCPM, table S1). Then, we performed a meta-analysis by combining summary data comprising 1,413,622 individuals from ten previously published studies (CKDGen, Pan-UK Biobank, HUNT, BioVU, MGI, TOPMed, UKHLS, PAGE, SUMMIT, and BBJ). Through sample size-weighted meta-analysis of z-scores using METAL(19) (materials and methods), we developed a catalog of eGFRcrea associations genome-wide for European-ancestry (n = 1,785,582), East Asian-ancestry (n = 282,852), African-ancestry populations (n = 67,943) as well as across all ancestries (n = 2,287,877; Fig. 1A, fig. S1A, and table S2). To assess the extent of population stratification in these GWAS meta-analyses, we applied linkage-disequilibrium (LD) score regression(20). The LD score intercepts ranged from 0.912 to 1.109 with low attenuation ratios (fig. S1B and table S3), which is consistent with the expectations under polygenicity(21–23), suggesting that there was no obvious excess inflation due to population stratification.

Fig. 1. eGFRcrea GWAS of 2.2 million individuals.

(A) Manhattan plots of eGFRcrea GWAS of 2,287,877 individuals of multi-ancestry and 1,785,582 individuals of European-ancestry. The x-axis is the chromosomal location of SNP. The y-axis is the strength of association −log10(p value, GWAS meta-analysis) (the scale is capped at 300). Lead SNPs of independent loci are highlighted. Green dots represent loci overlapping or nearby (within 500 kb or LD R² > 0.001) previously reported sentinel variants, and purple represents loci not reported previously. Prioritized genes for new loci are highlighted. (B) Fraction of known and new independent loci identified in GWAS of multi-ancestry. (C) The number of independent loci with at least one variant validated by eGFRcys and BUN. (D) Number of GWAS loci identified in down-sampled GWAS of European-ancestry. The lines represent loci defined by different locus window sizes (35 kb, 50 kb, and 100 kb). (E) Overlap of credible sets by fine-mapping in eGFRcrea GWAS of 2,287,877 individuals of multi-ancestry and 1,785,582 individuals of European-ancestry. (F) Enrichment of credible set variants in non-synonymous variants, kidney enhancers, and active promoters. The y-axis is the odds ratio, and the chi-square test significance is presented in each bar. (G) Locus-zoom of credible sets by fine-mapping in the IRF5 locus in European-ancestry GWAS. The x-axis is the chromosomal location of SNP. The panels from top to bottom show the original GWAS, posterior inclusion probability, credible set 1, credible set 2, and gene annotations. (H) Pleiotropy maps of eGFRcrea and other traits. The x-axis is the number of colocalization loci between eGFRcrea and related GWAS traits listed on the y-axis.

Within the multi-ancestry GWAS encompassing 13,700,391 variants, we identified 99,595 genome-wide significant variants (P < 5×10⁻⁸), excluding 12,895 significant variants from the major histocompatibility complex locus. Further, we identified 1,026 independent loci, including 97 loci situated at least 500 kb away from any previously reported eGFRcrea-associated variants(6–8, 24–27) (Fig. 1B, figs. S1C and D, table S4). Most (86.8%, 891/1,026) of these eGFRcrea-associated loci have been validated as kidney function loci by at least one variant associated with complementary kidney function measurements, serum cystatin C-based eGFR (eGFRcys) or blood urea nitrogen (BUN)(6) (Fig. 1C and fig. S1E). These independent (including the newly identified) loci were significantly (FDR < 0.05) enriched near genomic regions associated with genes critical for kidney development and function (fig. S1F and table S5). More than half (42/78) of the prioritized genes in the 97 new loci have been previously reported as related to kidney function or disease, including Forkhead Box C2 (FOXC2)(28), Polycystin 2 (PKD2)(29), and Salt-Inducible Kinase 3 (SIK3)(30) (Fig. 1G and table S4).

Down-sampling analysis indicated that the number of identified loci rose with sample size; however, this upward trend began to attenuate for sample sizes exceeding 1.5 million in European-ancestry GWAS (Fig. 1D, fig. S1H and table S6). Although the European-ancestry GWAS accounted for most (85.4%) of the loci identified in the multi-ancestry GWAS, we identified ancestry-specific loci in African (3 loci near the APOL1 gene, known to be associated with kidney disease(31)) and East Asian populations (4 loci near the BICC1, PKD2, ABCG2 and SIM1 genes) (fig. S1, I and J), highlighting the value of increased population diversity. Our GWAS provides a genetic link between the BICC1 gene and eGFRcrea, two decades after the recognition of BICC1’s role in kidney development and disease(32, 33).

Fine-mapping and characterizing eGFRcrea-associated loci

To identify putative causal variants at eGFRcrea loci, we performed fine-mapping using CARMA, a Bayesian model for fine-mapping in meta-analyses(34), separately for European-ancestry and multi-ancestry GWAS. In the European-ancestry GWAS, we delineated 1,818 sets comprising 99% credible intervals, 45.7% (830/1,818) of which had single putative causal variants when including PolyFun annotations(35) (Fig. 1E, fig. S2A and table S7). These credible sets included 8,901 putative causal variants, which were significantly enriched for non-synonymous variants (P < 1×10⁻³⁰⁰), kidney enhancers (P = 7.7×10⁻²⁵²), and promoters (P = 1.2×10⁻⁹⁶) (Fig. 1F). Most (82.0%) of these sets were also observed in the multi-ancestry GWAS (Fig. 1E, table S8, and fig. S2, B and C). For instance, at a chromosome 7 locus (q32.1), we identified two statistically distinct signals with credible sets with distinct regulatory features: one proximal to the promoter of the IRF5 (Interferon Regulatory Factor 5) gene and colocalized with eQTLs of IRF5 (Fig. 1G), and another adjacent to an enhancer and colocalized with allelic expression of the TNPO3 (Transportin 3) gene (fig. S2D). Additionally, our colocalization analyses with eQTL and meQTL datasets led to the prioritization of 387 genes and 4,363 CpG sites, which exhibited shared genetic predispositions with kidney function (fig. S2, E and F).

To evaluate the shared genetic basis between kidney disease and other conditions, we conducted a genome-wide multi-trait colocalization analysis between eGFRcrea and 23 GWAS traits potentially related to kidney function (fig. S3A and table S9) utilizing HyPrColoc (36). This analysis resulted in the identification of pleiotropy maps across 351 loci where eGFRcrea colocalized with other traits with regional probability > 0.8, 64.4% (226/351) of which colocalized with known kidney function traits (table S10). The number of colocalizations per signal varied, ranging from four loci with fasting glucose to 183 loci with eGFRcys (Fig. 1H). Within these loci, eGFRcrea colocalized with a single trait at 164 loci, while 187 loci demonstrated colocalization with multiple traits, including loci of GCKR (15 traits including type 2 diabetes, T2D), FGF5 (11 traits including systolic blood pressure, SBP), and APOE (5 traits including parental longevity) (fig. S3, B and C). To explore causal relationships, we applied the Mendelian randomization analysis using the method GSMR(37), which revealed multiple traits, including coronary artery disease, type 2 diabetes, and low-density lipoprotein cholesterol, that are potentially implicated in the etiology of kidney disease (fig. S3, D to G).

Allele-specific expression in bulk human kidneys (bASE)

To elucidate the genetic impact on gene expression within the kidney, we conducted an allele-specific expression (ASE) analysis on two distinct kidney compartments: the tubule (with a sample size of 613 individuals) and glomeruli (with a sample size of 522 individuals). This analysis was performed using RASQUAL(15), with adjustments for technical and clinical variables as well as probabilistic estimation of expression residuals (PEER) factors (Fig. 2A, table S11, and fig. S4, A and B). We identified 10,398 ASE genes, with 8,573 in the tubule and 8,725 in the glomeruli, while maintaining a 1% false discovery rate (FDR) (Fig. 2B and table S12). As a validation, we also performed typical eQTL mapping using MatrixeQTL(38) in the same dataset. The effect sizes estimated by the two methods were significantly correlated; with RASQUAL identifying 389 additional genes not reported by typical eQTL mapping (FDR < 0.01) (Fig. 2C and fig. S4, B and C). Approximately 60% of ASE genes showed allele-specific expression in AlleleDB, an allele-specific annotation for the 1000 Genomes variant catalog(39), compared to 45.6% for genes identified solely through typical eQTL (Fig. 2C). Among these ASE genes, 1,773 (17.1%) had not been previously detected by kidney eQTL studies (table S12). This subset includes genes that are already known to play roles in kidney function, such as Glycine Amidinotransferase (GATM)(40) and HNF1 Homeobox A (HNF1A)(41) (Fig. 2D and fig. S4D). Furthermore, we observed that variants with minor allele frequency (MAF) < 10% tended to exhibit a greater impact on gene expression, which is consistent with previous eQTL studies(42) (fig. S4E).

Fig. 2. Allele-specific expression (ASE) analysis in tubular and glomerular compartments.

(A) Analytical scheme and Manhattan plots of the ASE analyses in tubule and glomeruli compartments. In the Manhattan plot, the x-axis is the chromosomal location of SNP. The y-axis represents the strength of association −log10(p value, ASE) (the scale is capped at 300). (B) Overlap of ASE genes identified in tubule and glomerular compartments using RASQUAL. (C) Number of ASE genes identified by RASQUAL and eQTL genes identified by MatrixeQTL in the same dataset (left panel) and fraction of genes validated by AlleleDB ASEs (right panel). (D) Locus-zoom of eGFRcrea GWAS, tubule ASE, glomeruli ASE, and kidney eQTL in GATM locus. The y-axis represents the strength of association −log10(p value). PP.H4 is the posterior probability of colocalization between ASE and eGFRcrea GWAS. (E) Number of eGFRcrea-associated variants with ASEs/eQTLs (left panel) and the fraction of ASE-only/eQTL-only variants in fine-mapping credible sets of eGFRcrea GWAS (right panel). (F) Number of genes whose ASEs/eQTLs are colocalized with eGFRcrea GWAS (left panel) and the fraction of ASE-only/eQTL-only genes reported as nephropathy genes (right panel). (G) Locus-zoom of eGFRcrea GWAS, tubule ASE, glomeruli ASE, and kidney eQTL in the SLC22A2 locus. The y-axis represents the strength of association −log10(p value). PP.H4 is the posterior probability of colocalization between ASE/eQTL and eGFRcrea GWAS.

Furthermore, we used the ASE associations to prioritize variants identified in eGFRcrea GWAS. This led to the nomination of 32,610 variants targeting 1,212 genes across 470 (45.8% of 1,026) eGFRcrea loci improving the eQTL-eGFRcrea annotation (Fig. 2E and fig. S4F). To further refine the list of potential causal genes, we carried out a colocalization analysis and highlighted 531 genes whose ASE associations colocalized with eGFRcrea GWAS signals (Fig. 2F and table S13). These genes were significantly enriched in metabolic processes (P = 2.4×10⁻⁷, fig. S4G). While most (76.7%, 407/531) of ASE-GWAS colocalization genes also showed eQTL-GWAS colocalization (Fig. 2F and fig. S4H), 124 genes did not show colocalization between eQTLs and eGFRcrea GWAS (Fig. 2F). For instance, SLC22A2 (also known as OCT2, a kidney function gene(43)) exhibited significant ASEs (top variant rs8177512, P = 3.5×10⁻⁴⁰) and eQTLs (top variant rs316018, P = 1.7×10⁻⁷); however, colocalization was observed exclusively between ASEs and eGFRcrea GWAS (Fig. 2G), underscoring the efficacy of ASE analysis in uncovering signals related to kidney function.

Allele-specific chromatin accessibility in bulk kidneys (bASA)

Chromatin accessibility is vital for gene regulation, yet the genotype influence on chromatin accessibility within the kidney has not been characterized. To address this gap, we conducted whole-genome sequencing paired with high-throughput transposase-accessible chromatin sequencing (ATAC-seq) on human kidney samples (table S14). Initially, we established comprehensive maps of chromatin accessibility, which included 446,385 open chromatin peaks (fig. S5A). Following stringent quality control measures (fig. S5, B and C), we mapped allele-specific chromatin accessibility in 82 bulk kidney samples (bASA). This was done using RASQUAL, with adjustments for both technical and clinical variables, as well as PEER factors (fig. S5D). Our analysis identified 19,083 peaks with 523,700 single nucleotide polymorphisms (SNPs) with MAF > 5% that were associated with changes in chromatin accessibility by RASQUAL with validation by at least one of two other models, MatrixeQTL(38) and stratAS(44) (FDR < 0.05, Fig. 3A and table S15). This effort provided an understanding of the genotype effect on open chromatin across the human kidney.

Fig. 3. Allele-specific accessibility analysis in human kidneys.

(A) Analytical workflow and Manhattan plot of the bASA analyses in human kidneys. Manhattan plot: the x-axis is the chromosomal location of SNP, and the y-axis is the strength of association −log10(p value, bASA). (B) Enrichment of bASA peaks in kidney chromatin states which were estimated using histone modifications. The x-axis is the log2-transformed odds ratio. The y-axis is enrichment significance −log10(p value). (C) Association of effect sizes between kidney bASAs (x-axis) and meQTLs (y-axis) (Spearman correlation test). (E) Prioritization of GWAS variants and loci using bASAs in conjunction with ASEs and meQTLs. The upper panel represents the analytical workflow to identify bASA and GWAS variants associated with open chromatin and kidney function. The middle panel shows the fraction of bASA and GWAS variants overlapped with ASEs and/or meQTLs. The middle panel also shows the fraction of bASA and GWAS colocalization loci, which also colocalized with ASEs and/or meQTLs. (F) Regulatory elements and locus-zoom of eGFRcrea GWAS, kidney bASA, meQTL, and tubule ASE in USP24 locus. Chromatin states are represented by different colors. In the locus-zoom plots, the y-axis is the strength of association −log10(p value). PP.H4 is the posterior probability of colocalization between eGFRcrea GWAS and molecular traits.

Human kidney bASA peaks were significantly enriched in regions identified as kidney enhancers, with an odds ratio (OR) of 1.7 (chi-square test P = 2.4×10⁻¹⁷¹), and active promoters, with an OR of 2.2 (chi-square test P = 2.3×10⁻²⁵⁹), in comparison to non-bASA peaks (Fig. 3B and fig. S5E). Most of the bASA SNPs identified above were found within 50 kb of the associated peaks (fig. S5F). We discovered that 8,317 bASA peaks had associations with both local (within peak) and distal (outside peak) SNPs, which exerted consistent effects on the same peak (fig. S5, G and H). Variants with MAF < 10% had a more substantial impact on chromatin accessibility, aligning with the findings from the ASE analysis in this study (fig. S5I). Furthermore, the effect sizes of bASA negatively correlated with methylation QTLs (correlation = −0.78, P < 2.2×10⁻¹⁶) and positively correlated with ASEs (correlation = 0.51, P < 2.2×10⁻¹⁶), suggesting a regulatory effect of open chromatin on gene expression (Fig. 3, C and D).

Furthermore, we used bASA associations to prioritize open chromatin peaks affected by variants from the eGFRcrea GWAS. We identified 20,035 SNPs that were associated with both eGFRcrea and human kidney chromatin accessibility within 997 bASA peaks across 445 eGFR GWAS loci (Fig. 3E). Most bASA SNPs (88.9%) also had substantial associations with DNA methylation and/or allele-specific expression.

To investigate the shared genetic architecture between open chromatin and kidney function, we carried out a colocalization analysis, which identified 209 loci where eGFRcrea GWAS signals colocalized with the accessibility of 268 peaks (Fig. 3E and table S16). Notably, most (82.3%) of these also colocalized with meQTLs and/or ASEs (Fig. 3E). Specifically, we found colocalization of eGFRcrea with all three molecular traits (bASA, meQTL, and ASE) at 79 loci for 104 genes, which include genes previously linked to kidney disease (such as UMOD and UNCX) as well as potential new disease genes (like USP24 and ACSM2A) (fig. S5J). For instance, genetic variants at the USP24 locus showed significant associations with eGFRcrea (P = 7.9×10⁻²⁰), open chromatin (P = 2.4×10⁻²⁶), CpG methylation (P = 6.3×10⁻⁸⁸), and gene expression (P = 2.6×10⁻⁴⁶) at shared variant rs2802865 (Fig. 3F). Most (91.9%, 192/209) of these eGFRcrea~bASA colocalization loci have been validated as kidney function loci by complementary kidney function measurements (eGFRcys and/or BUN), underscoring the intersection of genetic variation and epigenetic regulatory elements in influencing kidney function.

Allele-specific chromatin accessibility in kidney cell types

A critical limitation of bulk open chromatin and gene expression analysis is its constraint in nominating cell type-specific changes. This is particularly challenging given the intricate complexity of the kidney, which has more than 30 cell types. To discern the genotype effects on chromatin accessibility in specific kidney cell types, we profiled individual cells using single nuclear ATAC sequencing (snATAC-seq) in 30 human kidneys (17 newly profiled and 13 publicly available(8, 12, 45, 46), table S17). Following rigorous quality control, we compiled an extensive atlas of open chromatin across 237,274 cells, spanning 19 cell types, with a median of 5,302 cells per type (fig. S6A). We found a significant enrichment of eGFRcrea GWAS variants within proximal tubule-specific open chromatin peaks (FDR = 0.021), correcting for 836 tests performed across 19 cell types and 44 GWAS traits (fig. S6B).

We then assessed the allelic impact on chromatin accessibility within each cell type by conducting snATAC-based allele-specific chromatin accessibility (snASA) analyses (Fig. 4A and fig. S7A). In total, we identified 25,766 SNPs exhibiting significant snASA (FDR < 0.05) in at least one cell type (Fig. 4B and table S18). Most of these snASA SNPs (93.2%, or 24,015 out of 25,766) were located within open chromatin peaks (Fig. 4C), and non-peak snASA SNPs showed significantly higher read coverage than shuffled SNPs from non-peak regions (Wilcoxon ranked sum test P = 6.8×10⁻³⁰⁷, fig. S7B). Furthermore, we observed that variants with MAF < 10% tended to exhibit a greater impact on chromatin accessibility (Wilcoxon ranked sum test P < 1×10⁻¹⁴⁰), which is consistent with the ASE and bASA analysis (fig. S7C).

Fig. 4. Single cell-level allele-specific accessibility analysis in human kidneys.

(A) Analytical workflow of snASA analyses in human kidney cell types. (B) Number of cells (x-axis) and snASA SNPs (y-axis) identified in each cell type. (C) Fraction of snASA SNPs within open chromatin peaks. (D) Cell type-specific and shared snASA SNPs (rows) across kidney cell types (columns). The color represents the effect size of ASA from low (blue) to high (red). GWAS variants were marked in the right column. (E) Fraction of cell type-specific and shared snASA SNPs validated by bASA. The significance of shared snASA SNPs with bASA was calculated using the chi-squared test. (F) Association of effect sizes between PT snASAs (x-axis) and kidney bASAs (y-axis). The correlation of effect size was calculated by the Spearman correlation test. (G) Enrichment of cell type-specific snASA SNPs on TFBSs. White block; non-significant. Yellow to red indicates the significance of TF enrichment (enrichment score > 3 and p value < 3.68×10⁻⁵, Bonferroni’s threshold accounting for 1,358 tests of 7 cell types and 194 TFs). (H) Fraction of snASA & GWAS variants overlapped with TF binding sites. (I) Examples of snASA & GWAS variants overlapped with cell type-specific open chromatin peaks and TFBS.

Most of the snASA SNPs (22,151 out of 25,766, 86%) were specific to individual cell types, while 3,615 snASA SNPs were shared across at least two cell types (Fig. 4D). We then assessed the potential of snASA to identify new regulatory variants by analyzing 18,727 common SNPs that were tested in both snASA and bulk kidney allele-specific accessibility (bASA) in the 82 bulk kidney samples mentioned earlier. While most (68.4%) of the snASA SNPs that were common across cell types were also detected by bASA, a large proportion (58.7%) of cell type-specific snASA SNPs did not show significant bASA associations (Chi-square test P = 1.4×10⁻¹⁴⁸, Fig. 4E), indicating an advantage of snASA in identifying cell type-specific regulatory variants. Furthermore, we observed strong positive correlations of allelic effect sizes both across different cell types (median Spearman’s rho = 0.99) and between snASA and bASA (Spearman’s rho = 0.90, P < 1.0×10⁻³⁰⁰) (Fig. 4F and fig. S7D). These correlations indicate robust biological mechanisms regulating chromatin accessibility. For instance, allele C of the variant rs10875467, situated in an intron of the SLC45A4 gene, was significantly associated with decreased chromatin accessibility in bulk kidney samples (bASA P = 4.8×10⁻¹⁵) and various cell types, including PT (snASA P = 3.5×10⁻¹⁰¹) and LOH (snASA P = 2.6×10⁻⁵) (fig. S7E).

To investigate the transcription factors (TFs) mediating allele-specific chromatin accessibility in kidney cell types, we conducted a TF motif enrichment analysis for snASA SNPs specific to each cell type. We identified 29 TF motifs that were preferentially disrupted by snASA SNPs, with an enrichment score greater than 3 and a p-value < 3.7×10⁻⁵ (Bonferroni adjustment for 1,358 tests overall, across 7 cell types and 194 TFs at a 5% significance level, Fig. 4G and table S19). Among these motifs, some showed cell type-specific disruption, such as the HNF4A and HNF4G motifs in the proximal tubule and the ESRRB and ESRRA motifs in the loop of Henle. Others, like the FOSL2 and NFKB1 motifs, were enriched in injured proximal tubule cells. Additionally, we found motifs with shared enrichment across various cell types, such as those binding the CCCTC-binding factor (CTCF), an architectural protein involved in regulating three-dimensional genome structure(47) (Fig. 4G). eGFRcrea GWAS signals were found within a 10kb radius of 8 genes encoding these disrupted TFs, including HNF4G, HNF4A, CTCF, NFKB1, and PPARG (fig. S7F).

Having identified and characterized allelic imbalance in chromatin accessibility across kidney cell types, we next assessed whether snASA SNPs could mediate the genotype effect on eGFRcrea. Our analysis revealed that 669 snASA SNPs were significantly (P < 5×10⁻⁸) associated with eGFRcrea across 337 GWAS loci, which included high-probability variants for eGFRcrea association at the DAB2, SHROOM3, and SLC47A1 loci (Fig. 4D and fig. S7G). Most (79.5%, 532/669) of these snASA-GWAS SNPs have been validated by complementary kidney function measurements (eGFRcys and/or BUN). A substantial proportion (63.4%, 448/738) of these snASA-GWAS SNPs resided in transcription factor binding sites, a frequency significantly greater than the 18.8% observed in GWAS SNPs not associated with snASAs (odds ratio = 7.5, chi-square test P = 1.9×10⁻²⁰⁶) (Fig. 4, H and I). A TF enrichment analysis further emphasized the potential involvement of CTCF, HNF4A, PAX5, and KLF5 in the pathogenesis of kidney disease (fig. S7H). Our findings suggest that allele-specific chromatin accessibility maps of human kidneys are instrumental in identifying both the cell types and regulatory variants implicated in kidney function.

Gene regulatory circuits via single nuclear multiomics

While we observed associations between gene regulatory elements, gene expression, and kidney function, we wanted to understand a potential causal relationship. We simultaneously profiled chromatin accessibility and gene expression in nine human kidney samples using single-nucleus multiome sequencing, which combines RNA and ATAC sequencing. Following rigorous quality control procedures, we compiled a comprehensive atlas of joint chromatin accessibility and gene expression across 62,278 cells spanning 19 kidney cell types (Fig. 5A and fig. S8A). To pinpoint open chromatin regions involved in gene activation, we devised a statistical approach named Open4Gene. This method uses the hurdle negative binomial regression model(48, 49) to estimate the association between chromatin accessibility and gene expression, specifically designed to handle the overabundance of zeros in single-nucleus RNA data (Fig. 5A and fig. S8B). For validation, we applied Open4Gene alongside two widely-used single-cell multimodal peak-to-gene linkage methods, ArchR(50) and Signac(51), to analyze a set of 903,472 peak-to-gene pairings (spanning 16,239 genes and 347,480 open chromatin peaks within a 100 kb cis-regulatory range) based on the joint profiles from the 62,278 kidney cells. By using HiC-based chromatin interactions as a benchmark for replication, we observed that the top peak-to-gene pairs identified by Open4Gene had higher validation rates compared to those identified by the alternative methods (fig. S8C). These top pairs identified by Open4Gene exhibited significantly (Wilcoxon rank sum test P < 1×10⁻³⁰⁰) greater correlation scores between peak accessibility and gene expression than those by other methods. Additionally, simulation analysis indicated a low rate of false positives identified by Open4Gene (fig. S8D). These findings suggest that Open4Gene is an effective tool for discerning functional cis-regulatory elements within single-cell multimodal datasets.

Fig. 5. Human kidney single cell multi-omics Open4Gene associations link GWAS variants to target genes.

(A) Analytical workflow of single-nucleus multiome (RNA+ATAC) and Open4Gene analyses in human kidney cells. (B) Number of cells (x-axis) and Open4Gene links (y-axis) identified in each cell type. (C) Fraction of open chromatin peaks targeting unique genes and multiple genes. (D) Prioritization of target genes of GWAS variants using Open4Gene links. (E) eGFRcrea GWAS strength (y-axis) and distance from GWAS variant to Open4Gene peaks (x-axis). (F) The number of GWAS variants identified for each gene using Open4Gene links. The y-axis is the number of GWAS variants, and the x-axis shows the genes with a given number of GWAS variants. Top genes and known kidney disease genes were labeled. (G) Locus-zoom of eGFRcrea GWAS, Open4Gene links, and regulatory elements in the LRP2 locus. In the locus-zoom plot, the y-axis is the strength of association −log10(p value). Chromatin states are represented by different colors. The tracks on the left panel show the open chromatin in each cell type, and the right panel shows the expression of LRP2 in each cell type.

Using our Open4Gene tool, we delineated a definitive set of peak-to-gene associations for each human kidney cell type (FDR< 0.01). This process yielded 35,388 unique Open4Gene links, with 40.6% being distinct from those obtained when analyzing a combination of all cell types (table S20). The quantity of Open4Gene links detected within each cell type was directly proportional to the cell count recorded for that specific type (Fig. 5B). By integrating the findings from each cell type with those from the collective cell analysis, we altogether identified 142,452 peak-to-gene connections, encompassing 125,699 peaks, nearly 90% of which were specifically associated with a single gene (Fig. 5C). The derived Open4Gene associations indicated a median of eight linked peaks per gene, with gene PKHD1—implicated in kidney disease(52)—having the highest number of peaks (fig. S8, E and F). Furthermore, an overwhelming 97% of Open4Gene associations were positively linked (effect estimate beta > 0) with the expression of their target genes (fig. S8, G and H). Additionally, Open4Gene beta values exhibited high correlations across different cell types (fig. S8I), reinforcing the crucial role of open chromatin in gene activation.

Furthermore, Open4Gene was instrumental in linking variants to gene expression and function by identifying 1,351 target genes for 7,137 genetic variants located within Open4Gene peaks (Fig. 5D and table S21). Genetic variants situated in these Open4Gene peaks exhibited strong associations with kidney function (Fig. 5E). We observed that 80.2% (5,723/7,137) of these variants were exclusive to a single gene (fig. S9A). There were 253 genes targeted by 10 or more GWAS variants, which displayed significant enrichment in pathways associated with abnormal renal system physiology (FDR = 1.1×10⁻⁷, fig. S9, B and C). This group includes genes previously linked to kidney disease, such as LRP2, UMOD, DAB2, SLC34A1, PAX8, and KLF5 (Fig. 5, F and G; and fig. S9D). For the genes without known kidney association in the literature, we focused on ACSM2A (Acyl-CoA Synthetase Medium Chain Family Member 2A), which encodes a mitochondrial enzyme involved in fatty acid metabolism, and explored its potential function in kidneys. The expression of ACSM2A was predominantly found in proximal tubules and appeared to be correlated with open chromatin, as indicated by Open4Gene links (fig. S9D). Leveraging RNA-sequencing data from 433 human kidneys, we established that ACSM2A expression was significantly positively correlated with kidney function (eGFRcrea, Spearman’s rho = 0.31, P = 7.7×10⁻¹¹) and inversely correlated with kidney fibrosis (Spearman’s rho = −0.58, P < 2.2×10⁻¹⁶) (fig. S9E). Additionally, ACSM2A expression was notably reduced in chronic kidney disease (ANOVA P = 1.9×10⁻⁵), a finding corroborated in three mouse models: unilateral ureteral obstruction (UUO, P = 3.4×10⁻²⁵), folic acid nephropathy (FAN, P = 3.7×10⁻¹¹), and a cisplatin model (P < 1×10⁻³⁰⁰) (fig. S9, E and F). These results underscore the efficacy of the Open4Gene framework in elucidating the regulatory mechanisms of disease-associated genes and resolving variant-to-function issues in GWAS.

Integrated Kidney Disease Genetic Scorecard concept for combined variant and gene prioritization

In this study, we have generated a multitude of multi-omics datasets for the annotation of regulatory variants, encompassing allele-specific expression and chromatin accessibility at both bulk and single-cell levels, Open4Gene links, complementing existing datasets of eQTLs, meQTLs, exome-wide association analyses and used complementary trait GWAS studies (eGFRcys, BUN). We observed important differences in prioritizing different genes and variants with no single tool or dataset enabling variant-to-function annotation. We, therefore, reason that a Kidney Disease Genetic Scorecard that includes all available datasets and tools (currently 32) and uses a simple additive approach to nominate kidney disease genes and variants. Here we included eight datasets, including bulk allele-specific accessibility (bASA), single-nucleus allele-specific accessibility (snASA), and fine-mapping credible sets to evaluate regulatory variants. Additionally, we assessed the disease risk of genes using 12 annotations, such as coding sequence (CDS), exome-wide association studies (ExWAS), and phenome-wide association studies (PheWAS). We examined the influence of non-coding variants on target genes through 12 types of cis associations, including Open4Gene links, allele-specific expression (ASE), and colocalization analyses (Fig. 6A). The number of prioritized variants diminished with increasing prioritization scores; however, the proportion of variants targeting unique genes showed a marked increase, indicating the effectiveness of our prioritization strategy in pinpointing authentic regulatory variants (Fig. 6B and fig. S10, A and B).

Fig. 6. Kidney Disease Genetic Scorecard indicates the convergence of coding and regulatory variants in GWAS loci.

(A) Analytical workflow of prioritization of eGFRcrea GWAS variants by integrating kidney multi-omics. (B) Prioritization of regulatory variants by integrating kidney multi-omics. The left y-axis is the number (bar chart) of non-coding variants prioritized using each prioritization score cutoff on the x-axis. The right y-axis shows the fraction (red dot plot) of variants targeting unique genes and the fraction (blue dot plot) of validated nephropathy genes prioritized by each prioritization score cutoff on the x-axis. The gray dashed line shows the prioritization threshold that covers all validated nephropathy genes. (C) The number of eGFRcrea GWAS variants located in different genome features (left panel), and enrichment test with log-transformed odds ratio on the x-axis and significance labeled for each feature (right panel). (D) Fraction of coding GWAS variants annotated as pLOF, missense, and synonymous. (E) Number of eGFRcrea GWAS loci prioritized by regulatory variants (prioritization score ≥ 10) and coding variants. (F) Number of genes validated by eGFRcys/BUN, with the convergence of coding and regulatory variants, cell type-specific expression, and targeted by FDA-approved drugs. (G) Kidney Disease Genetic Scorecard of genes with convergence of coding and regulatory variants (prioritization score ≥ 12). This figure only shows the genes with the highest expression and open chromatin on their regulatory variants in the same cell type. The full list is available in table S25.

Focusing on 24,437 regulatory variants that target 1,060 genes, identified by prioritization scores of 10 or higher (tables S22 and S23), we noted the top prioritized regulatory variant rs35716097 (with a prioritization score of 27) targeting the SLC34A1 gene (also known as NPT2), a known kidney disease-causing gene(53) (fig. S10C). 64.4% of prioritized regulatory variants were located in intronic regions, a proportion greater than expected (odds ratio = 1.76, chi-square test P = 1.9×10⁻²⁹⁵, fig. S10D), and nearly half (44.1%) were associated with distal genes (fig. S10E). Moreover, 29.2% of these regulatory variants coincided with transcription factor binding sites (TFBSs), which was a significantly higher overlap than the 14.5% observed in non-regulatory GWAS variants (odds ratio = 2.42, chi-square test P < 1×10⁻³⁰⁰) (fig. S10F), underscoring the regulatory impact of non-coding variants on gene regulation.

The convergence of regulatory and coding variants for kidney disease

Both coding and regulatory variants contribute to disease risk(54), yet their collective presence at GWAS loci remains largely unexplored. To address this gap, we integrated multi-omics datasets to prioritize genetic variants linked to kidney function and disease (Fig. 6A). Initially, we annotated eGFRcrea GWAS variants according to their proximity to gene regions and discovered significant (P < 1×10⁻⁷⁰) enrichment in protein-coding sequences (CDS), five prime untranslated regions (5’UTR), three prime untranslated regions (3’UTR), and introns (Fig. 6C and fig. S11A). Furthermore, exonic variants showed significant enrichment within the fine mapping credible set (odds ratio = 1.53, chi-square test P = 1.4×10⁻¹⁶, fig. S11B).

When we annotated GWAS variants based on chromatin states—defined by histone modifications in human kidneys—we noted significant enrichment of these variants in genic enhancers (odds ratio = 3.2, chi-square test P < 1×10⁻³⁰⁰), active promoters (odds ratio = 2.3, chi-square test P < 1×10⁻³⁰⁰), enhancers (odds ratio = 2.1, chi-square test P < 1×10⁻³⁰⁰), and other regulatory elements (fig. S11C). To determine whether this enrichment was unique to the kidney, we conducted stratified LD score regression(55) using eGFRcrea GWAS summary statistics alongside expression and chromatin data from the Roadmap Epigenomics project(56). This analysis confirmed a consistent enrichment of eGFRcrea GWAS variants in kidney-specific active genes, chromatin marks, and enhancers (fig. S11D). These findings suggest a convergence of coding and regulatory variants at GWAS loci, highlighting their collective contribution to the genetic landscape of kidney disease.

We then narrowed our focus to coding variants. Comprehensive GWAS mapping facilitated the identification of 1,363 coding variants linked to kidney function, with over 90% being previously undetected by ExWAS studies based on whole exome sequencing in about 450,000 UK Biobank participants(3–5) (fig. S12A and table S24). The common coding variants concurrently identified by both GWAS and ExWAS exhibited highly correlated effect sizes with respect to kidney function (fig. S12B). Using the Variant Effect Predictor(57), we pinpointed 83 loss-of-function variants (such as those resulting in a gain of stop codon, splice region variation, and frameshift mutations), 776 missense variants, and 505 synonymous variants (Fig. 6D). Among 782 genes hosting these coding variants, over 95% had not been identified by either ExWAS or phenome-wide association studies (PheWAS) which primarily focused on rare variants with MAF < 10% (58) (fig. S12, C and D). Within these genes, GWAS, ExWAS, and PheWAS collectively identified 21 kidney disease-related genes, including solute carrier genes (such as SLC47A1 and SLC6A19), multiligand receptors (LRP2 and CUBN), and genes implicated in polycystic kidney disease (PKD1, PKD2, and PKHD1) (fig. S12E). Among these, 289 genes (37.0% of the 782 total) possessed multiple coding variants, with ALMS1 (Centrosome and Basal Body Associated Protein)—a gene whose malfunction can lead to a progressive decline in kidney filtration function(59)—harboring the highest number (16) of coding variants (fig. S12F). As expected, the analysis revealed that while most common coding variants exhibited small effect sizes, rare coding variants tended to have larger effect sizes (fig. S12G).

In our concluding analysis, we concentrated on 601 genes identified by both coding and regulatory variants across 316 GWAS loci (Fig. 6E, fig. S13A, and table S25). Most (93.8%, 563/600) of these genes were prioritized by at least one variant validated by complementary kidney function measurements (Fig.6F, fig. S13B). Functional enrichment analysis of 601 genes indicated their roles in proximal tubule transport (P = 9.9×10⁻¹⁰, table S26) and kidney-specific phenotypes such as abnormal renal system physiology (P = 4.0×10⁻⁹, fig. S13C and table S27). Notably, 74 of these genes are documented nephropathy genes(60), including 10 genes (PKD1, PKD2, PKHD1, CUBN, GATM, LRP2, SLC34A1, SLC34A3, SLC6A19, and SLC7A9) validated by both ExWAS and PheWAS (fig. S13D).

For each of these genes, we performed conditional analysis on the top variant, resulting in 161 genes with the convergence of coding and regulatory variants, including 29 genes that have already been published to cause renal abnormalities in knockout mice (Fig. 6, F and G; fig. S13E and table S28). Compared with non-converged coding variants, converged coding variants showed significantly lower minor allele frequency (P = 1.9×10⁻¹³), higher effect size (P = 1.5×10⁻¹⁵), and higher significance (P = 1.4×10⁻⁵) in eGFRcrea GWAS (fig. S13F).

These genes also demonstrated pronounced cell type-specific expression and chromatin accessibility at their regulatory variants (fig. S14A). To delineate cell types contributing to kidney disease etiology, we assessed the cell type-specific expression of these genes. We found 316 genes with significant cell type-specific expression (adjusted P < 0.05, accounting for all tested genes), with an average of 16 specific genes per cell type and the highest number, 94, in proximal tubule cells (fig. S14B and table S29). Among these cell type-specific genes, 63.3% (200 out of 316) exhibited significant associations (Open4Gene FDR < 0.01) with open chromatin at their top regulatory variants, and 54.1% (171 out of 316) showed concordant expression and open chromatin profiles within the same cell type. For instance, the SLC34A1 gene displayed specifically high expression and open chromatin in proximal tubule cells (fig. S10C), while the KNG1 gene showed these features in the loop of Henle and distal convoluted tubule cells (fig. S14C).

Furthermore, in exploring therapeutic implications, we identified 124 genes amenable to targeting by FDA-approved drugs, offering potential prospects for drug repurposing and the development of therapeutic approaches (fig. S14D and tables S30 and S31). In conclusion, our integrative maps of coding and regulatory variants facilitate the identification of causal genes, implicated cell types, and potential druggable targets.

Discussion

By conducting a multi-ancestry GWAS in 2.2 million individuals, we provide insights into the genetic architectures of human kidney function, with a primary focus on common genetic variation. We generated genetic maps for eGFRcrea GWAS (n=2.2 million), allele-specific gene expression in tubule and glomerulus compartments, allele-specific chromatin in both bulk kidney and single-cell level, and regulatory circuits based on single-nucleus multiome information. By constructing the Kidney Disease Genetic Scorecard, we developed a new prioritization strategy and identified 601 kidney disease genes targeted by both coding variants and regulatory variants, including 161 genes with converged coding and regulatory variants. In these analyses, we also developed several strategies, including snASA and Open4Gene, to integrate single-cell omics and GWAS studies. Using these tools, we found critical cell types (including proximal tubules), transcriptional regulators (including HNF4A), and potential drug targets for kidney disease, providing an overall foundation to understand kidney disease.

Over the past two decades, GWAS has been instrumental in identifying thousands of loci associated with human diseases. Nonetheless, capturing nearly all disease-related loci remains challenging, often due to limited sample sizes. In this study, we have created what is likely a complete GWAS map of loci associated with kidney function within the European population based on a sample size exceeding 1.7 million individuals. This extensive GWAS resource has led to the discovery of 97 loci related to kidney function. The identification of specific GWAS signals in underrepresented populations, such as BICC1 in the East Asian population and APOL1 in the African population, underscores the importance of increasing sample sizes in these groups. Expanding our GWAS efforts to include diverse populations is crucial for uncovering genetic insights and enhancing our understanding of disease mechanisms in the future.

Recent studies have proposed that eQTLs account for only a modest proportion (on average, 0.20) of the heritability of GWAS traits(8, 61), suggesting that there is additional, unexplored information not captured. Allele-specific expression and open chromatin could provide more powerful information as these datasets are able to define both the reference and alternate allele effect on gene expression and gene regulation within the same sample, likely eliminating critical confounders. By capitalizing on allelic differences in the transcriptome, we have identified a greater number of disease-related genes. For instance, the genetic associations of HNF1A were only detected through allele-specific expression (ASE) analysis within the tubule compartment. Further investigations into allelic imbalance have illuminated the epigenetic alterations attributable to heterozygosity that contribute to allele-specific expression. The development of additional epigenetic datasets will be instrumental for the functional follow-up studies of GWAS.

The integration of single-cell multiomics data has greatly advanced our ability to pinpoint causal variants, target genes, and their relevant cell types. We found a pronounced enrichment of cell type-specific regulatory elements, including open chromatin peaks and genic enhancers, as well as transcription factors such as HNF4G and HNF4A. The presence of eGFRcrea GWAS signals in proximity to genes encoding these transcription factors provides compelling evidence for their involvement in kidney function. Delving deeper into transcription factor binding events at the single-cell level promises to unravel the intricate regulatory mechanisms underlying human diseases. Our findings indicate that the cell number is crucial for mapping snASAs and peak-to-gene associations effectively. Therefore, increasing the coverage of rare cell types will further improve our understanding of the regulatory circuits in these cell types.

Coding variants, particularly those that lead to a loss of function in proteins, are the major cause of mortality and morbidity at an early age. However, coding variant mutations are extremely rare due to the selective pressure acting to eliminate them in the general population. Most prior exome sequencing studies indicated that coding mutations are not major contributors to common disease heritability due to their rarity. In our large GWAS, we have been able to identify a large number of coding variants (n = 1,363) associated with kidney function. These coding variants influence 782 genes implicated in disease, including 601 genes also affected by non-coding variants in cis-regulatory regions. Our observations underscore the convergence of coding and non-coding variants in hundreds of genes and enable the identification of key disease-causing genes and pathways. Unraveling the complex interplay between these variant types holds great promise for informing the development of innovative therapeutic strategies.

Limitations of our work should also be noted. First, the study primarily focuses on common autosomal variants. On these variants, we observed the eGFRcrea signals began to attenuate in the European-ancestry populations. Further improvement in kidney function measurement, rare variant detection, non-additive genetic model, and coverage of underrepresented populations should enable the capture of additional kidney function variants. Due to differences in variants analyzed by GWAS and ExWAS, integrating these two analyses could generate complementary information, while the comparisons between them may be biased. Another complication is the heterogeneity in the meta-analysis GWAS due to differences in quality control, adjustment for non-technical covariates (body mass index and duration of diabetes), and eGFR ascertainment (data transformation etc.). To minimize heterogeneity, we performed stringent variant filtering in meta-analysis and downstream analysis, resulting in a high-quality map of the genetic architecture of eGFRcrea in 2 million individuals. To prioritize these variants, we defined genotype effects on allele-specific gene expression and regulatory circuitries over 700 human kidneys and 237,000 cells, while the analyses in non-kidney tissues are necessary to assess the function specificity of these variants. Furthermore, the presented kidney function genetic scorecard integrates 32 different tools and datasets based on an additive principle. Carefully designed weights could improve the power of prioritization, but such analysis is not feasible as we currently lack information on the ground truth. Additional datasets and analytical tools should enable a more comprehensive and precise prioritization.

In summary, we present a large-scale analysis of the genetic architecture of human kidney function, a multitude of omics datasets, and information to highlight biological insights, including 601 prioritized disease genes. This study emphasizes the convergence of coding and regulatory variants on key disease genes and presents the concept of a Kidney Disease Genetic Scorecard for disease diagnostics and therapeutics development.

Materials and methods summary

The collection of human kidney samples was approved by the institutional review board of the University of Pennsylvania, and the data from population-based cohorts were approved by relevant ethics committees (table S1). We first performed de novo mapping of genome-wide associations for eGFRcrea in each of 9 population-based cohorts, such as MVP(27), TWB(62), All of Us(63), MGB(64), GHS(65), PMBB(66), BioMe(67), eMERGE-III(68), and CCPM(69), Additionally, we obtained ten more GWAS datasets from published studies, including CKDGen(6), Pan-UK Biobank (https://pan.ukbb.broadinstitute.org), HUNT(70), BioVU(27), MGI(71), TOPMed(72), UKHLS(73), PAGE(25), SUMMIT(26), and BBJ(74)). After quality control, we conducted a meta-analysis of eGFRcrea GWAS in multi-ancestry populations with a total sample size of 2,287,877, using METAL (version 2011–03-25)(19). Similarly, we performed ancestry-group-specific GWAS meta-analyses for 1,785,582 European individuals from 15 studies, 282,852 East Asian individuals from five studies, and 67,943 African individuals from seven studies, respectively (table S2). Variants associated with eGFRcrea were defined by the genome-wide significance threshold p < 5×10⁻⁸, with validation using two complementary kidney function markers. We used a clumping method to define independent loci and identified new loci by comparing them with previous studies. Fine mapping analysis was conducted in European-ancestry and multi-ancestry GWAS using the Bayesian model CARMA (version 1.0)(34). Pleiotropy maps between eGFRcrea and other traits were assessed using HyPrColoc (version 1.0)(36) and bi-directional GSMR (version 1.1.0) based on GWAS summary statistics of European-ancestry individuals.

To explore the genetic effect of eGFRcrea variants on gene expression and chromatin accessibility, we conducted multiomic profiling in over 1000 kidney samples, including genotyping 729 human samples using Axiom arrays (n = 641) and whole genome sequencing (n = 88), transcriptome analysis of 613 tubule samples and 522 glomerulus samples by RNA sequencing, single-nucleus multiome sequencing (joint RNA and ATAC) in nine kidneys and snATAC sequencing in eight kidneys. After data processing and quality control, we explored the genetic effect on gene expression via ASE mapping using RASQUAL(15) and typical eQTL mapping using MatrixeQTL(38). Significant associations were determined after multiple testing corrections based on eigenMT(75) and used to prioritize target genes of eGFRcrea variants. To investigate the effect of genetic variants on chromatin accessibility, we created comprehensive maps of kidney open chromatin regions by integrating bulk and single-nucleus ATAC-seq datasets and mapped bASA associations using RASQUAL(15). Validated associations were used to prioritize chromatin accessibility affected by eGFRcrea variants. Next, we constructed open chromatin maps of 237,274 high-quality cells in 30 human kidneys and explored the genetic effect on chromatin accessibility at the cell type level by developing a snATAC-based allele-specific accessibility analysis strategy, snASA, based on WASP (version 0.3.4)(76) and two-tailed Binomial test. Further, we developed a new statistical method (Open4Gene) based on the hurdle negative binomial regression model(48) for mapping peak-to-gene links in multiome data. Using Open4Gene, we mapped the regulatory circuits in different kidney cell types and explored the genetic effect on open chromatin and gene expression. By integrating these multiomic datasets, we proposed a Kidney Disease Genetic Scorecard and prioritized coding variants and regulatory variants associated with eGFRcrea. Finally, we explored the convergence of coding and regulatory variants using GCTA-COJO (version 1.93)(77), and identified potential drugs targeting these genes using DGIdb (v4.2.0)(78) and DrugBank(79).

Data and materials availability:

The data of eGFRcrea GWAS, tubule ASE, glomerulus ASE, kidney bASA, snASA, Open4Gene links, and prioritization produced in this study is publicly available online at Figshare (https://figshare.com/s/4f6595fb3ca5d9c455f5, Doi: https://figshare.com/articles/dataset/26299093). The Kidney Disease Genetic Scorecard and the prioritization evidence used in this study are publicly available at the Susztaklab Kidney Biobank, https://susztaklab.com/GWAS2M. The human kidney RNA-seq, bulk ATAC-seq, and single-nucleus data are deposited with the Common Metabolic Diseases Genome Atlas (https://cmdga.org/search/?type=Experiment&searchTerm=FNIH0000000). The Open4Gene package developed in this study is available on GitHub (https://github.com/hbliu/Open4Gene) and Zenodo(80).