Integrated Functional Genomic Analysis Enables Annotation of Kidney Genome-Wide Association Study Loci

Significance Statement

The absence of high-resolution epigenomic maps of key kidney cell types has hampered understanding of kidney-specific genome regulation in health and disease. Kidney-associated genetic variants, identified in genome-wide association studies, are concentrated in accessible chromatin regions containing regulatory DNA elements. The authors describe the generation and initial characterization of paired DNA maps of these regulatory regions and gene expression profiles of cells from primary human glomerular and cortex cultures. By integrating analyses of genetic and epigenomic data with genome-wide chromatin conformation data generated from freshly isolated human glomeruli, they physically and functionally connected 42 kidney genetic loci to 46 potential target genes. Applying this approach to other kidney cell types is expected to enhance understanding of genome regulation and its effects on gene expression in kidney disease.

Visual Abstract

Abstract

Background

Linking genetic risk loci identified by genome-wide association studies (GWAS) to their causal genes remains a major challenge. Disease-associated genetic variants are concentrated in regions containing regulatory DNA elements, such as promoters and enhancers. Although researchers have previously published DNA maps of these regulatory regions for kidney tubule cells and glomerular endothelial cells, maps for podocytes and mesangial cells have not been available.

Methods

We generated regulatory DNA maps (DNase-seq) and paired gene expression profiles (RNA-seq) from primary outgrowth cultures of human glomeruli that were composed mainly of podocytes and mesangial cells. We generated similar datasets from renal cortex cultures, to compare with those of the glomerular cultures. Because regulatory DNA elements can act on target genes across large genomic distances, we also generated a chromatin conformation map from freshly isolated human glomeruli.

Results

We identified thousands of unique regulatory DNA elements, many located close to transcription factor genes, which the glomerular and cortex samples expressed at different levels. We found that genetic variants associated with kidney diseases (GWAS) and kidney expression quantitative trait loci were enriched in regulatory DNA regions. By combining GWAS, epigenomic, and chromatin conformation data, we functionally annotated 46 kidney disease genes.

Conclusions

We demonstrate a powerful approach to functionally connect kidney disease-/trait–associated loci to their target genes by leveraging unique regulatory DNA maps and integrated epigenomic and genetic analysis. This process can be applied to other kidney cell types and will enhance our understanding of genome regulation and its effects on gene expression in kidney disease.

In the past 15 years, large-scale genome-wide association studies (GWAS) have successfully identified genetic variants associated with a wide variety of measurable traits and human diseases, including those related to kidney function.^1–3 The minority of GWAS variants localize to protein-coding sequences (exemplified by APOL1⁴); most lie in nonprotein-coding genomic sequences, which compose >98% of the human genome.⁵ Regulatory DNA elements, which encompass promoters, enhancers, and insulators, are small segments of the genome where DNA binding proteins recognize specific DNA sequence features leading to the recruitment of histone modifying complexes, displacement of nucleosomes, and opening of nuclear chromatin.⁶ Active regulatory DNA elements are often located in nonprotein-coding sequences^7,8 and appear to functionally and physically associate with their target gene promoters over large genomic distances, often skipping intervening genes.⁸ Furthermore, GWAS variants frequently localize to regulatory DNA elements.^5,9–14 Because the chromatin accessibility of some regulatory DNA elements can be very cell type–specific,^7,9,15–17 one mechanism by which genetic variants can contribute to disease risk is by altering the regulation of gene expression in a cell type–specific manner (Supplemental Figure 1).¹⁸ Delineating the cell type–specific gene regulatory networks for multiple important kidney cell types will therefore be paramount to dissecting kidney disease mechanisms.

Enzymatic reporters, such as deoxyribonuclease I (DNase I)^19,20 and the Tn5 transposase,²¹ when combined with next-generation sequencing methods, efficiently identify regions of open chromatin that are associated with regulatory DNA elements. However, these methodologies have only been applied to the study of a few kidney cell types and chromatin accessibility maps are lacking for many important kidney cell types such as podocytes, mesangial cells, distal tubule cells, peritubular microvascular endothelial cells, pericytes, and resident immune cells (e.g., macrophages and dendritic cells). To begin to approach this problem, we report the generation of high-resolution chromatin accessibility maps (DNase-sequencing [DNase-seq]) and paired gene expression profiles (RNA-sequencing [RNA-seq]) of primary cultures of human glomerular outgrowth cells (mixed cultures of podocytes and mesangial cells). To enable comparative analysis of cell type–specific features, we also generated datasets from primary human renal cortical cultures treated in similar fashion. To understand the basis of long-range interactions between regulatory DNA elements and their target genes, we generated a chromatin conformation (Hi-C) map from freshly isolated human glomeruli. We then integrated these diverse measures of genome function to connect kidney GWAS loci to their potential target genes, thereby gaining new insights into the genome regulatory mechanisms that underlie kidney phenotypes and disease.

Methods

Generation and Characterization of Primary Glomerular and Cortex Cultures

Kidney tissues were obtained from patients undergoing radical nephrectomy for renal tumors with informed consent for DNA sequencing obtained before surgery. Patient characteristics and resulting dataset features are listed in Supplemental Table 1. The study and consent forms were approved by the University of Washington’s Institutional Review Board (protocol #1297). Approximately 1 cm³ portions of uninvolved kidney cortex (from the pole furthest from the tumor mass) were harvested and transported in RPMI medium on ice. These tissues were then minced with a sterilized razor blade and the fragments were placed in 20 ml of prewarmed RPMI medium (without serum) supplemented with Accutase (diluted 1:10; Sigma), collagenase P (100 µg/ml; Roche), and trypsin/EDTA (0.25% solution diluted 1:10; Gibco). The tissue fragments were digested at 37°C for 20 minutes with vigorous agitation. For glomerular core isolation, the softened tissue fragments were mashed through a #60-gauge sterilized steel mesh using the bottom of sterilized glass beaker. This treatment stripped the glomerular cores of their Bowman’s capsules. The isolated glomerular cores passed through the mesh and were collected on a #140-gauge steel mesh placed below. Tubules were disrupted sufficiently that they did not collect on the #140-gauge steel mesh. The isolated glomerular cores were washed extensively with sterile room temperature PBS and were then transferred into a tissue culture flask with prewarmed culture medium (RPMI supplemented with 10% FBS and insulin-transferrin-selenite+ [ITS+] supplement; Corning). For the primary culture of cortical cells, after digestion of a separate cortical tissue fragment (as described above), the pieces were spun down and macerated in a petri dish using a sterile plunger from a 5 ml syringe. These softened tissue fragments were then transferred into tissue culture flasks with prewarmed medium containing 10% FBS and ITS+ as described above. After 2–3 days (for cortex cultures) and 10–14 days (for glomerular outgrowth cultures), the tissue fragments were decanted and the adherent cells were fed with fresh medium. At this stage, primary cortex cells grew rapidly and had an epithelioid morphology, whereas primary glomerular outgrowth colonies grew more variably with a mixture of epithelioid and spindled cells. Glomerular outgrowth cells were used at first passage for all experiments (typically approximately 2–3 weeks after initial isolation). Cortex outgrowth cells were subcultured at 1:4 when they reached 80% confluence, and used within two passages for all experiments.

Immunofluorescence Staining of Primary Cultures

Disaggregated single-cell suspensions in growth medium were applied to sterilized glass coverslips placed in a six-well plate and incubated overnight to allow cells to fully adhere. The following day, the cells were washed three times with cold PBS and then fixed with 2% paraformaldehyde with 4% sucrose for 10 minutes at room temperature. The fixative was then removed and cells washed once with PBS. Cells were permeabilized with 0.3% Triton X-100 in PBS for 10 minutes at room temperature and then washed three times with PBS. Primary antibodies were incubated for 1 hour at room temperature followed by three washes with PBS. Primary antibodies reported in this study recognized podocin (P0372 rabbit polyclonal; Sigma-Aldrich), WT1 (sc-192 rabbit polyclonal; Santa Cruz Biotechnology), synaptopodin (10r-s125a mouse monoclonal; Fitzgerald Industries International), PAX2 (716000 rabbit polyclonal; Invitrogen), claudin-1 (ab15098 rabbit polyclonal; Abcam), and smooth muscle actin (A5228 mouse monoclonal; Sigma-Aldrich). Species-specific fluorescence-labeled secondary antibodies were then applied for 1 hour at room temperature, followed by three washes with PBS (Alexa Fluor 488 anti-rabbit [A11070] or Alexa Fluor 594 anti-mouse [A11020]; both from Invitrogen). After counterstaining with DAPI (10 µg/ml) and mounting, cells were imaged with a widefield Nikon TI-E inverted epifluorescence microscope.

Processing of Cell Cultures for DNase-Seq

For each primary culture, cells were subjected to DNase I treatment, small DNA fragment isolation, and library construction per published Encyclopedia of DNA Elements (ENCODE) protocols or a modified protocol adapted for low DNA input.^22,23 Libraries were subjected to paired-end (2×36 bp) sequencing. Most datasets (Supplemental Table 1) used in this study were deemed of high quality (signal portion of tags>0.4).⁷

DNase-Seq Data Processing

Sequence reads from our DNase-seq libraries were subjected to a uniform data-processing pipeline, which was used previously for ENCODE DNase-seq datasets.⁷ Briefly, read pairs passing quality filters are trimmed of adapter sequences and aligned to the reference human genome (GRCh38/hg38) using BWA.²⁴ Genomic regions with a significant enrichment of DNase I cleavages were identified using the hotspot algorithm,⁷ and were further refined to fixed-width, 150-bp regions (“peaks”) containing the highest cleavage density (referred to as DNase I–hypersensitive sites; DHS). Hotspot and peak calling were performed using full-depth data.

Processing of Cell Cultures for RNA-Seq

We washed 500,000–1,000,000 cells of primary cortex or glomerular cultures once in PBS and stabilized them at 4°C in RNALater (Ambion). Total RNA was extracted using a mirVana RNA isolation kit (Ambion). Illumina sequencer-compatible libraries were constructed using a TruSeq Stranded Total RNA Library Prep Kit with Ribo-Zero Gold (Illumina) and subjected to paired-end (2×76 bp) sequencing.

RNA-Seq Data

RNA-seq libraries were aligned to the reference human genome (GRCh38/hg38) using TopHat 2.0.13,²⁵ and assigned to transcript models using RNA-STAR 2.3.1.²⁶

Processing of Glomeruli for Hi-C

Glomeruli were mechanically isolated as described before from a portion of uninvolved kidney cortex obtained from a 74-year-old woman undergoing nephrectomy for renal cell carcinoma. The glomeruli were fixed for 20 minutes in 10 ml of a 1:10 dilution of 10% neutral buffered formalin (Fisher) with end-over-end tumbling. At the end of this period, 0.1 g glycine (Sigma) was added directly to the tube to quench the fixation reaction. The cell pellet was delivered to Phase Genomics, Inc. (Seattle, WA) for Hi-C library preparation using a Phase Genomics Human Hi-C Kit. A total of 481,795,427 2×150 bp read pairs were sequenced from the resulting library on an Illumina HiSeq 4000. Topology-associated domain (TAD) calling was performed using the DomainCaller algorithm at 50 kbp resolution with default parameters.²⁷ Chromosomal contacts (binned to 10 kb resolution) were generated using the Phase Genomics Matlock tool (https://github.com/phasegenomics/matlock).

Availability of Data

All datasets produced for this study are available on the Gene Expression Omnibus as a SuperSeries GSE115961 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE115961). Also see Supplemental Table 1 for individual dataset accession numbers and sequencing depth.

Data Exploration and Visualization

For many of the computational analyses, R (v3.3.3) (https://www.R-project.org/) was used to format, analyze, and visualize the data. To annotate genes and transcripts, BioMart (v2.30.0)^28,29 and Bioconductor were used. Genomic coordinates were manipulated using GenomicRanges (v1.26.4).³⁰ For visualization, gplots (v3.0.1) (https://CRAN.R-project.org/package=gplots) and ggplot2 (v2.2.1) (https://ggplot2.tidyverse.org/) were used. For data manipulation, packages within the tidyverse (v1.1.1) (https://CRAN.R-project.org/package=tidyverse) were used, including tidyr (v0.6.3), plyr (v1.8.4), dplyr (v0.7.1), purr (v0.2.2.2), and tibble (v1.3.3).

Differential Gene Expression Analysis

Differential gene expression was calculated using DESeq2 (v1.14.1).³¹ All reported significant differential gene expression results are on the basis of the DESeq2 multiple testing–corrected P values ≤0.01 and >1.5 log₂ fold change in each respective culture. To filter out nonexpressed genes and ensure that the genes were expressed in at least one sample type, the differential genes were filtered to require that the median fragments per kilobase of transcript per million mapped reads (FPKM) expression in either the glomerular or cortex cultures was >1 FPKM. Using the glomerular and tubular differentially expressed genes, the gene ontology enrichment for biologic processes was performed using clusterProfiler (v3.2.14)³² with a P value cutoff of 0.05 and a Bonferroni-corrected q value cutoff of 0.01. Differential gene expression was visualized in heatmaps generated using heatmap.2 from gplots (for row normalized data) or pheatmap (for raw expression and row normalized data).

DHS Master List Generation and Differential DHS Analysis

Master lists were generated with a bedops -u command as detailed on the BEDOPS website (https://bedops.readthedocs.io/en/latest/).^33,34 We created DHS master lists for our glomerular and cortex DHS data, for ENCODE tubule datasets (RPTEC, HRE, and HRCE), and for ENCODE fetal kidney data. In most analyses, to improve the stringency of peak calling, we filtered the glomerular and cortex culture master list by only including peaks with a minimum cut count of 50 and those that were present in at least three out of the seven total sample datasets. We used a similar approach to generate the master list of 72 fetal kidney DNase-seq datasets by including peaks with a minimum cut count of 50 in at least five samples. This resulted in master DHS lists of comparable size between our datasets (335,161 DHS) and the fetal kidney datasets (353,676 DHS). We utilized the DESeq2 software package³¹ in R to identify DHS with significant differences in accessibility between replicate glomerular and cortex culture samples, analyzing each patient separately. Sites that met a false discovery rate (FDR) threshold of 1% were considered differential DHS. The distance of differential DHS to the nearest gene and the associated biologic ontologies were computed using the Stanford GREAT analysis tool.³⁵

Transcription Factor Motif Enrichment

Transcription factor motif models were curated from TRANSFAC v.11,³⁶ JASPAR,³⁷ and a SELEX-derived collection.³⁸ Instances of transcription factor recognition sequences in the human genome were identified by scanning the genome with these motif models using the FIMO tool³⁹ from the MEME Suite v.4.6,⁴⁰ with a fifth order Markov model generated from the 36 bp mappable genome used as the background model. Instances with a FIMO P<10⁻⁴ were retained and used for subsequent analyses. Families of transcription factors with highly similar recognition motifs were grouped into clusters.⁴¹ Motif enrichments were then calculated by using a custom Python script to count the number of DHS that contained an instance of any motif model within the cluster of highly similar motif models. For a given analysis, these counts were compared between a query set of DHS (e.g., those with increased accessibility in glomerular cultures) and a background set (e.g., all other DHS), and significance was determined using the hypergeometric distribution with Bonferroni correction of P values.

Correlation of Differentially Accessible DHS with Differentially Expressed Genes

We calculated the number of differential DHS within a given window from the transcription start site (TSS) of differentially expressed genes with valid Entrez and HGNC identifiers in each respective culture. Next, we calculated the number of differential DHS near these genes that were constitutively expressed (i.e., not significantly changing) in both glomerular and cortex cultures. The constitutively expressed genes were defined as >1 median FPKM in both glomerular and tubular cultures, adjusted P values >0.05, and the top four quintiles of DESeq2 variance stabilized gene expression values. To test for the enrichment of differential DHS near differentially expressed genes, the number of differentially accessible DHS in cortex and glomerular cultures within ±20 kb of the TSS for differentially expressed genes were counted and compared with the distribution for constitutively expressed genes. The Spearman correlation was calculated on the values of the log₂FoldChange and number of differential DHS.

Testing for Enrichment of Kidney GWAS Variants in Glomerular and Tubular DHS

The GWAS catalog⁴² was downloaded on August 24, 2017 (Supplemental Table 2), and filtered for all kidney traits and removing sex chromosome SNPs; this returned 636 entries. SNPs with the same position were then removed, keeping the SNP with the most significant P value, reducing the list to 477 entries (Supplemental Table 3). To collapse SNPs within linkage disequilibrium, we generated probabilistic identification of casual SNPs credible sets⁴³ for each SNP and estimated colocalization with the credible sets using PICCOLO.⁴⁴ We removed GWAS catalog SNPs with the higher P value if the two credible sets had a H4 posterior probability of colocalization ≥0.80 (see Supplemental Table 4). This resulted in 430 kidney GWAS catalog loci remaining, of which 200 achieved a genome-wide significance P value ≤5×10⁻⁸ (Supplemental Table 5). The proportions of the GWAS SNPs that overlapped kidney master list DHS within the specified windows (i.e., padded intervals around the SNP) were then calculated. In addition, to determine that single SNPs were not outliers, we built a distribution for the kidney GWAS SNPs using a leave-one-out analysis. The kidney GWAS SNPs were then split into 200 groups, with each group leaving out one kidney GWAS SNP. The proportions of the GWAS SNPs that overlapped kidney master list DHS within the specified windows were then calculated for the 200 groups. The kidney proportions were compared with a randomized distribution of 10,000 sets of 200 SNPs matched with the SNPsnap⁴⁵ tool that has been previously used by CKDGen.¹ SNPsnap matched the kidney GWAS SNPs on minor allele frequency, linkage disequilibrium, distance to nearest gene, and local gene density. FDR P values were calculated on the basis of this bootstrapped background distribution.

Enrichment of Kidney Expression Quantitative Trail Loci SNPs in Kidney DHS

We selected all expression quantitative trait loci (eQTL) SNPs (P value <1×10⁻⁴) that were eGenes (i.e. genes whose expression is associated with an eQTL SNP with a q value <0.05), from Genotype-Tissue Expression project (GTEx v7⁴⁶) and NephQTL⁴⁷ tissues. These SNPs were then tested for overlap of the kidney master list DHS. For each tissue, the proportion of all eGene eQTL SNPs (eSNPs) in kidney DHS relative to the total number of eSNPs within the tissue was calculated. With the number of tissue-specific eSNPs ranging from 43,751 to 2,144,827, we had increased power to detect small changes in the proportion of eSNPs found in kidney DHS, resulting in a significant difference between tissue proportions using a test of equal proportions (chi-squared test=5861.5, degrees of freedom=47, P value <2.2×10⁻¹⁶). For the GTEx samples, these proportions were normally distributed with a mean proportion of 0.0174 with SD 0.0015 (Shapiro test P=0.2207). To test if there was enrichment of the proportion of eSNPs within kidney DHS for a given tissue, we compared its proportion with the distribution of proportions across the GTEx samples and tested for significance with a one-sided Z test. In addition to the kidney master list DHS, we evaluated for enrichment in the subset of DHS with greater accessibility in either glomeruli or cortex samples.

Integrated Functional Genomic Mapping of GWAS SNPs to Putative Target Genes

For these analyses, we utilized all unique kidney disease/trait SNPs (Supplemental Table 5) because even SNPs below the genome-wide significance threshold (P<5×10⁻⁸) can contain genetic signals that contribute to cell type–specific regulation,¹ especially when combined with epigenomic data.^5,48 We linked unique autosomal GWAS SNP positions to DHS with differential accessibility within a ±20 kb window centered on the SNP. Each differential DHS was then linked to all potential differentially expressed genes within ±500 kb. For this analysis, differentially expressed genes lacking Entrez gene identifiers were removed. Using the 10 kb windows encompassing the SNP position and the putative target gene TSS as “baits,” we tallied long-range chromatin contacts identified by Hi-C in human glomeruli. We visually confirmed interaction or lack thereof between all GWAS SNP–target gene pairs. Lastly, we tested whether the genomic range between each GWAS SNP and its target gene TSS was wholly contained within TADs identified in human glomeruli using the BEDOPS³³ command bedops -e 100%.

LocusZoom Plots

Genetic association and regional linkage disequilibrium plots were rendered using the LocusZoom⁴⁹ webtool (locuszoom.org), using the default 1000 Genomes EUR database (November 2014) for linkage data. Genome-wide association summary statistics for serum creatinine were obtained from the CKDGen website (https://fox.nhlbi.nih.gov/CKDGen/)¹ and summary statistics for circulating parathyroid hormone levels were obtained directly from the authors of a previous study.⁵⁰

Results

Generation and Characterization of Primary Glomerular and Cortex Cultures

Glomeruli were isolated by mechanical sieving from patients undergoing nephrectomy for renal tumors (Figure 1A). The isolated glomerular cores were grown in tissue culture dishes until attached and primary outgrowth cultures were established. Genome-wide chromatin accessibility profiling with DNase-seq and gene expression profiling with RNA-seq was performed in parallel on these primary glomerular outgrowth cultures. Similar datasets were also generated from primary cultures of normal human renal cortex^51–53 (three patients) grown under similar conditions in the presence of 10% FBS (Supplemental Table 1). In primary culture, the cells growing out of the attached glomerular cores showed a mixed epithelioid and spindled morphology characteristic of podocytes and mesangial cells, respectively (Figure 1B). These cultures exhibited patchy staining for the podocyte marker WT1, compared with moderate and more uniform staining for podocin (Figure 1C). The parietal epithelial cell markers PAX2 and claudin-1 were consistently identified by immunofluorescence staining in the glomerular cultures (Figure 1C). Although the expression of these markers may indicate the presence of parietal epithelial cells within our glomerular cultures, mechanical isolation almost uniformly stripped the Bowman’s capsule away from the isolated glomeruli, and therefore these markers were most likely being expressed on cultured and proliferating podocytes, as has been reported by others.^54–58 Very few cells (<1%) demonstrated staining for smooth muscle actin (data not shown) despite this gene (ACTA2) being strongly expressed in the primary glomerular cultures (see below). Consistent with the idea that high exogenous VEGFA supplementation is required to support the growth of primary endothelial cells, staining for the endothelial marker CD31 was not detected in any of our cultures (data not shown). These immunofluorescence findings demonstrated that the primary glomerular outgrowths consisted of mixed cultures of podocytes and mesangial cells.

Figure 1.

Primary human glomerular cell outgrowth cultures and primary human cortex cultures express podocyte/mesangial and tubular markers respectively. (A) Schematic of samples and datasets generated for this study. (B) Phase contrast micrograph of adherent human glomeruli with outgrowth of epithelioid (podocytes) and spindled cells (mesangial cells) after 14 days in culture. (C) Immunofluorescence staining for indicated proteins (green) in human glomerular cell outgrowth cultures. Nuclei are counterstained with DAPI (blue). (D) Unsupervised clustering and row-normalized expression heatmap of all differentially expressed genes in glomerular and cortex samples. (E) Heatmap of raw FPKM expression values of selected lineage-defining genes in RNA-seq datasets.

Gene Expression Signatures of Glomerular Outgrowth and Cortex Cultures

Next, we statistically compared the global gene expression profiles of the glomerular and cortex cultures using DESeq2³¹ (Supplemental Table 6). This identified 507 differentially expressed genes with higher levels in the glomerular cultures and 257 genes with higher expression in the cortex cultures. Unsupervised clustering of these differentially expressed genes clearly demarcated the two types of primary cultures (Figure 1D). Examination of the ontologies linked to genes with higher expression in the glomerular cultures versus the cortex cultures revealed marked enrichment for genes associated with extracellular matrix organization and for renal/urogenital system development (Supplemental Figure 2).

After this global analysis of gene expression, we queried our data for known cell type–specific marker genes. First, we examined the expression of lineage-defining genes in the gene expression data. The podocyte-selective genes WT1,⁵⁹ BMP7,⁶⁰ SYNPO,⁶¹ ARHGAP24,⁶² and RAP1GAP⁶³ were highly expressed in all four primary glomerular outgrowth cultures, but were not expressed in primary cortex cultures (Figure 1E). Interestingly, although podocin gene (NPHS2) transcripts were undetectable by RNA-seq, podocin protein expression was seen by immunofluorescence microscopy (Figure 1C), which is consistent with the known long t_1/2 of podocin protein.⁶⁴ Genes encoding the mesangial cell markers ACTA2 (smooth muscle actin), PDGFRB, and MEIS1 were also expressed in the primary glomerular cultures; TBX18⁶⁵ and DES were not. TLR4 is normally expressed in both podocytes and mesangial cells,^66,67 and this gene showed strong expression in the primary glomerular cultures. Endothelial cell–associated marker genes (CD34, VWF, KDR, FLT1, and FLT4)⁶⁸ were not detected in any of the cell types examined. Tubule-associated genes (ANPEP, GGT1, MME, GDA, KL, and BHMT)^69,70 showed stronger and more consistent expression in the primary cortex culture datasets, although other known tubule markers (AQP1, AQP7, SLC5A2, and SLC34A1) were not detected. Undetectable uromodulin gene expression is consistent with minimal contamination by distal nephron components in the primary cortex cultures (Figure 1D). Minimal expression of some podocyte (e.g., SYNPO) and mesangial cell genes (e.g., ACTA2) in the cortical tubule cultures may reflect low-level contamination by glomerular outgrowths in these primary cultures, although care was taken to reduce glomerular attachment by early and frequent media exchanges.

At the time of the study, the cellular input for DNase-seq required millions of cells necessitating culture and expansion of the primary cell populations described in this study. Culture of glomerular outgrowth cells is a long-established method,⁷¹ typically performed in the presence of serum, and has been used to generate conditionally immortalized human podocyte,⁷² mesangial,⁷³ and glomerular endothelial cell lines.⁷⁴ However, this two dimensional culture method leads to downregulation of key podocyte markers,⁷⁵ and consistent with this, the primary glomerular outgrowth cultures did not express PODXL, NPHS1, NPHS2, or TRPC6.⁷⁶ The expression of mesangial markers and a mixed epithelioid and spindled morphology is consistent with the presence of mesangial cells in the glomerular outgrowth cultures. The absence of glomerular endothelial cells is not surprising given that these cells require the presence of VEGF in the growth medium for ex vivo survival and expansion.⁷⁴ Taken together, these analyses of phenotype and the gene expression data confirmed that the primary glomerular outgrowth cultures comprise a mixed population of podocytes and mesangial cells with few, if any endothelial cells. Furthermore, the expression signatures of podocytes (embedded within the whole glomerular cell gene expression data) are consistent with an injured/proliferative podocyte phenotype.

In contrast to glomerular outgrowths, primary culture of cortical tubular epithelial cells is usually performed under serum-free growth conditions.^77–79 Exposure of primary cortical tubular epithelial cells to serum stimulates rapid growth,⁸⁰ but results in the downregulation of functional markers and may promote fibroblast outgrowth.⁸¹ In this study, we cultured the cortex samples under identical growth conditions to the glomerular outgrowth cultures primarily to remove the effect of serum as analytic confounder. Cognizant of this limitation of cell preparation, hereafter we refer to these as “primary cortex cultures” to reflect their injured tubular epithelial cell phenotype and potentially complex cellular composition.

Examination of Chromatin Accessibility and Gene Expression Profiles at WT1 and PDGFRB Gene Loci

The RNA-seq data showed significantly higher expression of WT1 and PDGFRB in the glomerular cultures than the primary cortex cultures (Figure 2, A and B, top panels). Examination of the chromatin accessibility patterns around these two genes revealed numerous regulatory elements, which appear as DHS (Figure 2, A and B, bottom panels, shaded bars). The patterns of accessibility of these DHS are consistent both across the independently derived samples and within cell types. Although many of these DHS cluster around the TSSs, consistent with a role in promoter function, many others are located tens of kilobases away (Figure 2, red arrows) and represent putative distal regulatory elements.

Figure 2.

Correlating gene expression (RNA-seq) with chromatin accessibility (DNase-seq) reveals patterns of genome regulation for two key glomerular genes, (A) WT1 and (B) PDGFRB. Each horizontal “track” represents either RNA-seq (top panel) or DNase-seq (bottom panel) data for each of four primary glomerular outgrowth cultures and three primary cortex cultures. The RNA-seq tracks and DNase-seq tracks are shown for the same genomic region. For reference, the position of the TSS of WT1 and PDGFRB gene is indicated by the vertical black dashed line. For RNA-seq tracks, increased gene expression is visualized by greater transcript tag density over the exons of the predicted gene model (exons are taller vertical bars, direction of transcription is indicated by arrowheads). The quantified RNA expression level (FPKM) for the gene in that sample is shown to the left of each track. For DNase-seq tracks, increased chromatin accessibility is visualized by greater sequence tag density, resulting in a “peak.” Numerous DHS show significantly differential accessibility between glomerular cells and tubular cells, which are highlighted by the vertical shaded bars. There is dense differential DHS activity around the TSS of both genes consistent with regulation of gene promoters. This correlates with elevated expression of both genes in the glomerular cells. However, several DHS located several kilobases from the promoter region may also be contributing to gene regulation (red arrows), consistent with their role as distal regulatory DNA elements. RNA-seq and DNase-seq tracks are shown at the same vertical scale for both panels and for all samples.

Global Characterization of Chromatin Accessibility Profiles

Next, we analyzed the chromatin accessibility landscapes of the glomerular and cortex cultures genome-wide to compare them with published DHS data already available in ENCODE, to characterize their positioning with respect to chromatin landmarks and to identify DHS that were significantly differentially accessible in the glomerular and cortex cultures. Before this study, the only kidney-focused DHS maps available through ENCODE were for fetal kidney (primarily from the second trimester) and from cultured proximal tubules (RPTEC), whole renal epithelium cells (HRE), or renal cortical epithelial cells (HRCE). Because we report the first glomerular DHS maps, we compared their profiles to ENCODE fetal kidney DHS data as the closest approximation to previously published data. Combining all of the unique DHS detected in our glomerular cultures produced 402,684 elements that showed substantial overlap (51.6%) with fetal kidney DHS maps (Supplemental Figure 3A). However, many DHS identified in our glomerular DNase-seq data were not present in the fetal kidney dataset. This may be because of adult–fetal differences and a more glomerular cell–enriched composition in our samples. Similarly, comparison of our cortex culture DHS profiles to ENCODE tubule datasets (RPTEC, HRE, and HRCE) revealed a large overlap (57.7%), suggesting that a large portion of the tubule regulatory landscape is captured by our data (Supplemental Figure 3B) despite differences in culture conditions (as described above). Overall, the glomerular and cortex culture DHS maps generated in this study contained 176,462 distinct DHS that were not present in the previously published ENCODE kidney datasets.

Next, to further increase the stringency of our analysis, we filtered the glomerular and cortex culture DHS list by only including high-confidence DHS with a minimum cut count of 50 and those that were present in at least three out of the seven total datasets. Using these criteria, 335,161 DHS were present in either the glomerular or cortex cultures (i.e., a master list of all DHS), a far greater number than the number of regulatory elements identified by chromatin immunoprecipitation sequencing for histone modifications (H3K4Me3, H3K27Me3) or CTCF binding (Figure 3, A and B). In contrast to H3K4Me3, which is associated with promoters,⁸² the majority of DHS are located >5 kb from known TSS of genes, consistent with their role as distal regulatory elements.⁷

Figure 3.

Differentially accessible DHS are enriched for distal location, cell-type specific gene ontologies and are located near transcription factors with changing gene expression. (A) The number of DHS or ChIP-seq peaks in human renal epithelial (HRE) cells from ENCODE as a function of distance from gene TSSs is shown. (B) The percentage of DHS or ChIP-seq peaks in human renal epithelial (HRE) cells from ENCODE as a function of distance from gene TSS is shown. (C) The number of DHS with greater accessibility in glomerular cells (maroon) or cortex cells (blue) is shown as a function of distance to the nearest gene’s TSS. The majority of differentially accessible DHS in either cell type are located >5 kb from the nearest gene TSS. (D) The nearest genes to the differentially accessible DHS in glomerular (maroon) or cortex cells (blue) were annotated for gene ontology enrichment using the Stanford GREAT analysis tool. Results shown are for the ontologies with highest enrichment rank ordered by negative log₁₀ binomial P value. (E) Correlation of the magnitude of differential gene expression versus the number of differentially accessible DHS. For each gene that is expressed significantly higher in glomerular cells or cortex cells, the number of DHS with greater accessibility in glomerular cells (maroon dot) and those with greater accessibility in cortex cells (blue dot) is shown. Therefore, each gene is represented by two dots on the horizontal axis. Genes with greater expression in glomeruli tend to have greater numbers of DHS with higher accessibility in glomeruli (and vice versa for genes expressed at a higher level in cortex cultures). This enrichment is significant: Spearman correlation=0.45, P=4.2×10⁻⁵³. (F) Enrichment of transcription factor DNA recognition motifs in constitutive (left column) and differentially accessible DHS (middle and right columns) between the two sample types. The Bonferroni-corrected P values of the enrichments using a hypergeometric distribution are shown. ChIP-seq, chromatin immunoprecipitation sequencing.

Of the master list DHS, 296,041 (88.3%) represented regulatory elements with stable chromatin accessibility between the glomerular and tubular samples. 21,059 (6.3%) of the master list DHS showed significantly greater accessibility in the glomerular cultures versus the cortex cultures (FDR<0.01). Conversely, 18,061 (5.4%) of the master list DHS showed significantly greater accessibility in the cortex cultures than the glomerular cultures. We calculated the distance of these differentially accessible DHS to the closest gene TSS. Similar to the global DHS distribution (Figure 3, A and B), very few (<10%) of these differentially accessible DHS were located within 5 kb of gene TSS (Figure 3C). Examination of the genes located closest to the DHS with greater accessibility in the glomerular cultures uncovered enrichment for ontologies associated with glomerular function (e.g., kidney mesenchyme development, glomerulus development, etc.) (Figure 3D). Conversely, examination of the genes located closest to the differential DHS with greater accessibility in the cortex cultures uncovered enrichment for ontologies associated with tubular function (e.g., nephron epithelium development, renal tubule development, etc.). Taken together, these analyses support the idea that the differentially accessible DHS drive the differences in gene expression programs between the primary glomerular and cortex cell cultures.

Linking Differentially Accessible DHS to Differentially Expressed Genes: A Role for Transcription Factors

Because the regulation of gene expression may be influenced by multiple distal regulatory elements, we enumerated differentially accessible DHS in windows around the TSS of differentially expressed genes. We found that the number of differential DHS located near differentially expressed genes was significantly correlated with the magnitude of the differential gene expression (Spearman correlation=0.45, P=4.2×10⁻⁵³) (Figure 3E). Examination of the top ten glomerular-expressed genes with the greatest number of nearby differentially accessible DHS uncovered genes with known roles in glomerular cell biology (e.g., WT1,⁸³ MEIS2,^84,85 BMP7,^60,86 and FOXC2^87,88) and other genes with unknown or poorly characterized roles (e.g., TENM3 and FOXL1) (Figure 3E). At least five of these genes are transcription factors that can exert broad influence on gene expression programs by regulating the expression of numerous target genes.⁸⁹ Similarly, of the top ten cortex-expressed genes with nearby differentially accessible DHS, four (POU3F3,^90–92 SIM2,⁹³ SIM1,^93,94 and TFCP2L1⁹⁵) are transcription factor genes. DHS represent sites where transcription factors bind to sequence-specific DNA recognition motifs.⁸⁹ Therefore, we asked if the DNA recognition motifs of particular families of transcription factors were enriched in DHS showing different patterns of accessibility in the glomerular and cortex cultures (Figure 3F). DHS with greater accessibility in glomerular cultures were enriched for motifs of forkhead, GATA, TEAD, and the SOX family of transcription factors. Conversely, DHS with greater accessibility in cortex cultures were enriched for motifs of CTF, E-box, NF-I, and HNF transcription factor families. These studies showed that transcription factor genes with cell type–specific function are at the apex of regulatory control in our DHS maps. Furthermore, the differentially accessible DHS encode distinct patterns of transcription factor binding motif enrichment between the glomerular and cortex cultures.

Kidney Disease Risk Loci Are Enriched in Glomerular and Tubular DHS and eQTL

GWAS have revolutionized the identification of genetic loci that associate with disease and other complex traits. However, most SNPs associated with traits identified by these studies are in nonprotein-coding DNA sequence (i.e., introns and intergenic regions). Given that these loci are nonprotein-coding, it has been difficult to assign a target gene to the genetic signal that is responsible for the risk allele’s phenotype. It is now becoming clear that many of these genetic risk variants are enriched in genomic regions containing regulatory elements that control gene expression programs which can influence disease.⁵ This has also been recently demonstrated for kidney disease traits by the CKDGen consortium.¹ However, those analyses did not test for enrichment in fetal kidney DHS data and did not include chromatin accessibility data from primary glomerular outgrowth cells, which we report here. Therefore, we revisited the question to ask if loci associated with kidney disease and related traits were enriched in our newly generated regulatory DNA maps and in the fetal kidney DHS data available through ENCODE. We examined expanding windows surrounding genome-wide significant loci (Supplemental Table 5) for glomerular or tubular DHS from our master DHS list or from ENCODE fetal kidney DHS. We compared these to bootstrapped distributions (10,000 iterations) of randomized SNPs using the SNPsnap tool⁴⁵ to match SNPs on the basis of the allele frequency, number of SNPs in linkage disequilibrium, distance to the nearest gene, and gene density of SNPs associated with kidney disease/traits. Compared with the background distribution, the kidney disease/trait SNPs were significantly more likely than the matched random SNPs to have a master list DHS or fetal kidney DHS in their vicinity, regardless of the window surrounding the SNP (Figure 4A; FDR≤0.001 for all padded window sizes tested). For example, 37.5% (75 out of 200) and 32% (64 out of 200) of genome-wide significant kidney disease-/trait-associated SNPs were located within ±2 kb of a master list and fetal kidney DHS respectively. Next, we asked if our datasets were identifying novel loci compared with the fetal kidney DHS data. For this, we pooled all kidney disease/trait GWAS SNPs identified at various window sizes in Figure 4A and asked how many of these were unique to either our data versus the ENCODE fetal kidney data, or jointly identified by both (i.e., shared). Although some kidney disease/trait GWAS SNPs are jointly identified, many are uniquely identified by DHS seen in our data (Figure 4B). This analysis again demonstrated that kidney disease-/trait-associated SNPs are enriched in regulatory DNA elements, but also showed that the DHS maps reported in this study identify distinct genetic loci.

Figure 4.

Comparison of this study’s DHS data to ENCODE and eQTL datasets reveals enrichment in kidney GWAS and eQTL. (A) Kidney GWAS SNPs achieving genome-wide significance (P<5×10⁻⁸) are enriched in the chromatin accessibility sites of glomerular and cortex cells. The proportion of kidney GWAS SNPs that overlap an element in the master list comprising DHS from both glomerular and cortex cells (left panel) or ENCODE fetal kidney datasets (right panel) is shown as a function of increasing padding around that SNP. This proportion is compared with a background distribution of SNPs matched for genomic context using the SNPSnap⁴⁵ tool (10,000 iterations). The bounds of the boxes represent the 25th–75th percentile range, with the median as the central hatch within the box; the whiskers represent 1.5× the interquartile range. Regardless of the padding interval, kidney GWAS SNPs (red line) are more likely to be localized near a kidney DHS both in this study’s and the ENCODE fetal kidney datasets. For each comparison, FDR<0.001. See Supplemental Table 5 for the list of SNPs used for this analysis. (B) Kidney GWAS SNPs overlap shared and unique DHS in this study’s and ENCODE fetal kidney DHS data. The distribution of kidney GWAS SNPs that are identified by either this study or ENCODE fetal kidney DHS is shown at each padding interval depicted in (A). Although many kidney GWAS SNPs overlap both this study’s DHS and the ENCODE fetal kidney DHS at various padding sizes, many are unique to each dataset. (C) Kidney DHS are enriched in kidney eQTL. The proportion of SNPs/variants (i.e., eSNPs) associated with eQTL from GTEx samples or a recently published dataset⁴⁷ from isolated human glomeruli and tubulointerstitium is depicted. The red line indicates the mean proportion of overlapping eSNPs in the GTEx samples and the gray area indicates ±2 SD from the mean. Overlap with this study’s kidney DHS master list is significant for both glomerular eSNPs (P=0.01) and tubulointerstitial eSNPs (P=1.61×10⁻⁸).

Next, we sought to compare our regulatory DNA maps with the complementary method of eQTL analysis. In this approach, genetic variants are correlated with tissue-specific gene expression profiles from hundreds to thousands of individuals.^96–99 Most variants marked by eQTL (eSNPs) act in cis with their potential target genes (eGenes), consistent with their role as tissue-specific regulatory DNA elements.¹⁰⁰ The GTEx consortium has generated dense eQTL maps for a wide range of human tissues,⁴⁶ but maps of kidney tissues and cells had not been generated until recently.^47,101,102 When we compared the proportion of eSNPs in kidney eQTL and GTEx data that overlapped with our master DHS list, we found that the proportion of overlap was highest for the glomerular and tubulointerstitial eQTL samples (Figure 4C). Both the glomerular and tubulointerstitial eSNP overlap proportions were significantly higher than the mean overlap proportion of GTEx tissues (P_{tubulointerstitial}=1.61×10⁻⁸, P_glomeruli=0.001), indicating that glomerular and tubulointerstitial eSNPs were significantly enriched in kidney DHS master list. We also found that glomerular eSNPs were significantly enriched in DHS with increased accessibility in glomeruli (P=6.59×10⁻⁵), and tubulointerstitial eSNPs were significantly enriched in DHS with increased accessibility in tubules (P=2.27×10⁻¹⁶). These findings demonstrate that our chromatin accessibility maps for glomerular and cortex cultures both corroborate and complement recently published kidney eQTL data.

Genome Conformation Links GWAS SNPs to Their Putative Target Genes

Thus far, we have demonstrated that the differentially accessible DHS cluster in proximity to differentially expressed genes (Figure 3E), and that kidney GWAS loci and eQTL are enriched in the accessible chromatin regions of primary glomerular and cortical cultures (Figure 4). Next, we sought to combine these orthogonal views of genome function to link GWAS loci to their target genes. To do this, we first identified kidney disease-/trait-associated SNPs that were located within 20 kb of DHS with differential accessibility between glomerular and cortex cells. Then we asked if any protein-coding genes ±500 kb from these differential DHS also exhibited significant changes in gene expression between glomerular and cortex cells. This identified 57 unique SNP positions associating with 64 unique potential target genes resulting in 73 unique SNP–target gene interaction pairs (Supplemental Table 7). This analysis also underscored the complexity of correlating differentially accessible regulatory elements to differentially expressed genes. Some GWAS loci (e.g., rs12826808) could potentially affect multiple target genes (e.g., MGP, ERP27, and RERG). Conversely, some genes (e.g., SFTA2 and VARS2) may be regulated by DHS near more than one GWAS locus (rs3130544 and rs9263871).

Although our target gene search interval (±500 kb) appears large, the majority of differentially accessible DHS in our data were located between 5 and 500 kb from gene TSSs (Figure 3C), and regulatory elements such as enhancers are well known to physically associate and act over large distances on their target genes (Supplemental Figure 1).¹⁰³ However, to validate physical associations between GWAS SNPs and their potential target genes, we generated genome-wide chromatin conformation (Hi-C) data from freshly isolated human glomeruli. In this method, long-range genomic interactions in cells can be mapped by stabilizing physical contacts with gentle fixation.¹⁰⁴ In the glomerular Hi-C data, we visualized long-range chromatin interactions originating from either the GWAS SNP or putative target gene TSS. All 73 SNP–target gene pairs had such interaction data available (list of chromatin contacts provided as a University of California, Santa Cruz Genome Browser interact track file in Supplemental Table 8). Physical association between the GWAS SNP and the target gene TSS was visually confirmed for 52 SNP–target gene interaction pairs (Supplemental Figure 4, Supplemental Table 7, and Table 1). Using this integrated approach, in total, we assigned 42 unique GWAS loci to 46 putative target genes. Rather than being randomly oriented, the linear genome is organized into compact, three-dimensional units termed TADs.²⁷ Although gene regulatory interactions may occur across TAD boundaries, they are much less frequent, thus defining TADs as a compartmentalized functional unit of chromatin.¹⁰⁵ We computed TADs from our glomerular Hi-C data (Supplemental Table 9) and found that the majority (90%, 47 out of 52) of the physically interacting GWAS SNP–target gene pairs were contained within TADs. This provides further evidence for the localization of SNP–target gene interactions within functional chromatin compartments.

Table 1.

Kidney GWAS SNPs and their putative target genes inferred by integrated analysis of genetic, chromatin accessibility, gene expression, and chromatin conformation data

GWAS rs ID	SNP Chromosome	SNP Position	Target Gene	GLOM/CORTEX log2 Fold Change	Within Glomerular TAD?
rs2786111	Chr1	197336969	CFHR1	4.28	Yes
rs2049805	Chr1	155225189	EFNA1	−2.09	Yes
rs12568771	Chr1	17287314	MFAP2	2.58	Yes
rs267738	Chr1	150968149	MLLT11	2.29	Yes
rs2049805	Chr1	155225189	MUC1	−3.08	Yes
rs2352039	Chr1	78354261	PTGFR	−2.73	Yes
rs1800615	Chr1	15505786	TMEM51-AS1	−2.81	No
rs84178	Chr11	2753144	CDKN1C	3.04	Yes
rs11604462	Chr11	65784177	GAL3ST3	2.40	Yes
rs3741414	Chr12	57450266	ARHGEF25	2.36	Yes
rs3741414	Chr12	57450266	DTX3	1.74	Yes
rs12826808	Chr12	15170446	ERP27	3.61	Yes
rs7956634	Chr12	15168260	ERP27	3.61	Yes
rs12826808	Chr12	15170446	MGP	4.55	Yes
rs7956634	Chr12	15168260	MGP	4.55	Yes
rs12826808	Chr12	15170446	RERG	3.78	Yes
rs7956634	Chr12	15168260	RERG	3.78	Yes
rs490049	Chr13	32990725	KL	−1.89	Yes
rs12589674	Chr14	53779635	BMP4	3.23	No
rs2071047	Chr14	53951693	BMP4	3.23	No
rs8056893	Chr16	68270489	ESRP2	−2.07	Yes
rs889472	Chr16	79612092	MAF	3.27	Yes
rs344364	Chr16	1923515	SLC9A3R2	1.67	Yes
rs9895232	Chr17	4615844	SMTNL2	4.55	Yes
rs12975033	Chr19	48746186	CA11	1.97	Yes
rs12975033	Chr19	48746186	FUT1	−1.52	Yes
rs7562121	Chr2	99767892	AFF3	2.88	Yes
rs7583877	Chr2	99844192	AFF3	2.88	Yes
rs815815	Chr2	47171925	EPCAM	−2.67	Yes
rs187355703	Chr2	176128855	HOXD1	−1.74	Yes
rs13421350	Chr2	172453843	ITGA6	−1.74	Yes
rs1260326	Chr2	27508073	KRTCAP3	−2.12	Yes
rs35612822	Chr2	216505032	MREG	−1.98	Yes
rs12105918	Chr2	144450626	ZEB2	1.94	Yes
rs6127099	Chr20	54114863	CYP24A1	−2.95	Yes
rs9977499	Chr21	27362678	ADAMTS5	2.49	Yes
rs219780	Chr21	36461009	SIM2	−2.89	Yes
rs9839909	Chr3	171790673	TNIK	1.51	Yes
rs6816344	Chr4	72917687	ADAMTS3	2.09	Yes
rs7674118	Chr4	165929313	CPE	2.01	No
rs700236	Chr5	39367637	DAB2	2.29	Yes
rs6910061	Chr6	11101685	GCNT2	−1.99	No
rs9263871	Chr6	31202751	PSORS1C3	−2.17	Yes
rs3130544	Chr6	31090563	SFTA2	−3.18	Yes
rs9263871	Chr6	31202751	SFTA2	−3.18	Yes
rs955333	Chr6	154626274	TIAM2	1.80	Yes
rs9263871	Chr6	31202751	VARS2	−2.03	Yes
rs1799884	Chr7	44189469	AEBP1	3.03	Yes
rs10954650	Chr7	139641806	KLRG2	3.11	Yes
rs6977660	Chr7	19765857	MACC1	−3.23	Yes
rs10954650	Chr7	139641806	PARP12	−1.51	Yes
rs7805747	Chr7	151710715	SMARCD3	2.04	Yes

GLOM/CORTEX, Ratio of RNA-seq gene expression in glomerular cultures versus cortex cultures Chr, chromosome.

Integrated Functional Genomic Snapshots of Exemplar GWAS Loci

Finally, we applied our integrated functional genomic approach to examine our annotation of two GWAS loci to their respective target genes in greater detail. In this analysis, we found that in some instances, the GWAS SNP was predicted to regulate a distant gene, skipping the nearest genes. The SNPs rs219779 and rs219780 have recently been linked to circulating parathyroid hormone levels, kidney stones, and bone mineral density presumably via their control of calcium metabolism within kidney tubules (Figure 5A).^50,106 rs219779/rs219780 are in the last coding exon of the CLDN14 gene, and are approximately 500 bp from an intronic regulatory element with greater chromatin accessibility in tubular cultures compared with glomerular cultures (Figure 5B). The CLDN14 gene was not strongly expressed in either the glomerular or cortex cultures (FPKM<1). However, the nearest gene with strongly differential gene expression is SIM2 (8.0-fold higher expression in cortex cultures), which suggested that it may in fact be the regulatory target of the SNP-associated, differentially accessible DHS near CLDN14. A long-range physical interaction between rs219779/rs219780 and SIM2 was also supported by Hi-C data in human glomeruli and this entire region was contained with a TAD (Figure 5C).

Figure 5.

Integration of genetics and functional genomic data identifies SIM2 as a potential target gene of the rs219779/rs219780 locus linked to circulating parathyroid hormone and calcium levels. (A) A LocusZoom style plot of genetic association and regional linkage is shown for chr21: 36,341,453–36,871,451 (hg38). Variants with no linkage data are shown in gray. (B) Chromatin immunoprecipitation sequencing, RNA-seq, and DNase-seq tracks of the genomic region surrounding the rs219779/rs219780 locus are shown. The inset shows the location of a DHS located very close to rs219779 with greater accessibility in cortex cultures compared with the glomerular cultures. Previous studies have suggested that the nearest gene, CLDN14, plays a functional role in determining circulating parathyroid hormone levels and calcium levels; however, this gene is not expressed in either the primary cortex or glomerular cultures. In fact, the only gene with a significant change in gene expression in a 500 kb window around this locus is SIM2. Other DHS in this interval with higher accessibility in cortex cultures may also be contributing to SIM2 regulation. (C) Examination of long-range genomic contacts in human glomerular Hi-C data using either rs219779/rs219780 (orange arcs) or SIM2 gene TSS (purple arcs) as “baits.” This entire region lies within a TAD identified in human glomeruli (maroon bar).

In another instance, an SNP associated with a complex and composite phenotype such as increased eGFR (rs84178)¹⁰⁷ was found within 20 kb of a DHS with differentially higher accessibility in glomerular cultures compared with cortex cultures (Figure 6A). This DHS was located within an intron of the KCNQ1 gene that was not expressed in either the glomerular or cortex cultures (Figure 6B). Instead, CDKN1C was the only protein-coding gene that was differentially expressed within a 500 kb interval around rs84178. The CDKN1C gene product p57 is an important marker of podocyte maturation and function.^108–112 Chromatin conformation data again supported a long-range interaction between the GWAS locus and the CDKN1C gene, both of which were localized within a TAD found in human glomeruli (Figure 6C).

Figure 6.

Integration of genetics and functional genomic data identifies CDKN1C as a potential target gene of the rs84178 locus linked to elevated eGFR. (A) A LocusZoom style plot of genetic association and regional linkage is shown for chr11: 2,668,770–2,938,770 (hg38). Variants with no linkage data are shown in gray. (B) Chromatin immunoprecipitation sequencing, RNA-seq, and DNase-seq tracks of the genomic region surrounding the rs84178 locus are shown. The inset shows the location of a DHS overlapping rs84178 with greater accessibility in glomerular cultures compared with the cortex cultures. This DHS and rs84178 lie within an intron of the KCNQ1 gene which is not expressed in either the primary tubular or glomerular cultures. The only protein-coding gene with a significant change in gene expression in a 500 kb window around this locus is CDKN1C, the protein product of which, p57, plays an important role in podocyte biology. Together with the other DHS in this interval showing higher chromatin accessibility in glomerular cells, this suggests that the mechanism by which rs84178 contributes to elevated eGFR is by regulation of CDKN1C in glomerular cells (likely in podocytes). (C) Examination of long-range genomic contacts in human glomerular Hi-C data using either rs84178 (orange arcs) or CDKN1C gene TSS (purple arcs) as “baits.” This span between rs84178 and the CDKN1C gene lies within a TAD identified in human glomeruli (maroon bar). Some long-range contacts from both rs84178 and the CDKN1C gene TSS appear to cross this TAD’s boundary (gray dashed line) into the neighboring TAD. Also note the sharp drop-off in genetic linkage across the TAD boundary in (A).

Discussion

Our data reports reference quality genome-wide regulatory DNA maps for primary human glomerular cultures comprising podocytes and mesangial cells. Analysis of these data together with similar maps generated for primary cortex cultures revealed thousands of differentially accessible DHS genome-wide. Most of these (>90%) were located >5 kb from the TSSs of known genes and were associated with gene ontologies corresponding to the cell type from which they were derived. Many of these DHS were distinct from those previously available through the ENCODE consortium. Furthermore, the subset of DHS that were differentially accessible were enriched in the vicinity of transcription factors with critical lineage-specific functionality. We also found that kidney disease-/trait-associated genetic variants were enriched in our newly described regulatory maps, even though they comprised only a few kidney cell types. Finally, we show a powerful approach integrating genetic data (GWAS) with functional genomic readouts (chromatin immunoprecipitation sequencing, DNase-seq, and RNA-seq) and chromatin conformation (Hi-C) to assign genetic signals to putative target genes with greater confidence.

One limitation of our study is artifacts induced by expanding primary cells in two-dimensional culture and in the presence of serum. At the time of data production in this study, DNase-seq protocols¹¹³ typically required inputs of approximately 1,000,000 cells, which necessitated the use of mixed primary glomerular outgrowth cultures to achieve the required cell numbers. Advances in the protocol and library construction^22,23 enabled reduction in cell input numbers even during the course of this study. Recently, an alternative chromatin accessibility profiling technique using Tn5-transposase “tagmentation” and sequencing (ATAC-seq) has been described.²¹ Although all chromatin accessibility profiling methods have their unique advantages and limitations,¹¹⁴ the ATAC-seq protocol does involve fewer steps than the DNase-seq protocol and can generate data from as few as 500–50,000 cells. Because the problem of high mitochondrial read contamination in ATAC-seq appears to have been recently solved,¹¹⁵ this method could now be combined with gentler psychrophilic dissociation strategies¹¹⁶ and affinity-based approaches to generate epigenomic profiles of rare cell populations derived from freshly dissociated single cell suspensions of kidney tissue. More facile epigenomic profiling of directly isolated cells would enable generation of datasets from many more individuals and would overcome the artifacts introduced by in vitro culture and expansion.

A second potential limitation is that our functional annotation strategy only examines genetic risk variants that lie within/near differentially accessible DHS and ignores variants that lie in nonchanging DHS. We acknowledge that this correlative association method also does not reveal whether the identified SNP causally influences DHS regulation of target gene function, whether it pleiotropically influences the measured phenotype by affecting gene expression outside our examination window, or if it is simply linked to a causal SNP that is not within a differentially accessible DHS.¹¹⁷ Even with these caveats, we have annotated 46 target genes that may be contributing to kidney disease/traits. Because, by definition, these genes exhibit a kidney cell type–specific pattern of expression (either in glomerular cells or cortex cells), it will be interesting to study their contribution to kidney disease in relevant model systems. To this end, we also use Hi-C^104,118,119 as another layer of evidence to reveal the physical arrangement of GWAS loci and target genes in the three-dimensional environment of nuclear chromatin. A related method has been recently applied to study cultured glomerular endothelial cells (HRGEC) and kidney tubule cells (RPTEC).¹²⁰ In the future, true functional validation of DHS-target gene interactions may be definitively assessed with CRISPR-modulation approaches,^121,122 or perhaps with genome-editing of the SNP and/or DHS to test for the effects on target gene regulation, as has been done only for a handful of genetic variants identified for other human phenotypes.^13,14,123

Many kidney diseases encompass an interplay of kidney-intrinsic and kidney-extrinsic (e.g., cardiovascular) pathobiologic mechanisms. GWAS do not discriminate between these mechanisms and inherently cannot identify the relevant cell types that are contributing to the disease process. However, because GWAS-identified loci (even those that do not achieve genome-wide significance thresholds) are enriched for regulatory DNA elements that show a highly cell type–specific pattern of activity,^7,9,15–17 obtaining high-resolution chromatin accessibility maps for unique kidney cell types will be essential to assess their contribution to complex kidney traits and disease processes.¹²⁴ Our findings also support the idea that functional mapping of kidney disease-/trait-associated loci can generate hypotheses regarding the specific kidney cell types (e.g., glomerular cells, Figure 6) that play a key role in the disease process as has been suggested recently for other diseases.^9,125,126 Because many studies of CKD use imperfect and nonspecific biomarkers of kidney function (e.g., serum creatinine), deployment of cell type–specific biomarkers together with chromatin accessibility maps of important kidney cell types will improve the assignment of genetic risk to specific cells. Also, our finding that cell type–specific regulatory activity is concentrated around transcription factor genes (which, in turn, can regulate other genes with pleiotropic effects), provides an explanation for how small GWAS effect sizes may result in an amplified phenotypic response. In this context, complex traits and kidney disease may not simply be explained by stepwise accumulation of risk-predisposing genetic variants that only marginally affect the expression of their target genes. Our finding instead lends some support to the recently proposed omnigenic model of inherited risk by which genetic variants perturb cell type–specific gene regulatory networks.¹²⁴

In summary, we describe the generation and initial characterization of paired chromatin accessibility (DNase-seq) and gene expression profiles (RNA-seq) from primary human glomerular and cortical cultures from multiple individuals. We show that combined analysis of genetic, functional genomic, and chromatin conformation data can shed new light on the complex regulation of gene expression that predisposes to kidney traits and disease. These data will be useful to annotate and compare with future kidney GWAS and emerging eQTL studies.^47,101,102 In the future, generation of truly cell type–specific chromatin accessibility maps for a wide range of kidney cell types would enable high-resolution functional annotation of genetic risk variants and generation of kidney gene regulatory networks that have only been previously achieved in limited fashion and in model organisms.^127,128

Disclosures

K.B.S. and D.W. are full-time employees of GlaxoSmithKline, LLC. S.S. is a full-time employee of Phase Genomics, Inc.

Supplemental Material

This article contains the following supplemental material online at http://cjasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2018030277/-/DCSupplemental.

Supplemental Figure 1. Regulatory DNA elements stimulate the transcription of target gene(s). While regulatory elements may be located at great distances (10-100’s of kilobases) from their target gene(s) when measured by linear genomic distance, activated elements are frequently in close physical proximity with their actively transcribed target genes.⁸ Activated regulatory elements are characterized by the displacement of nucleosomes and an open nuclear chromatin state, which is particularly susceptible to DNase I-mediated cleavage.⁶ The resulting released DNA fragments can be sequenced and mapped to identify accessible chromatin regions genomewide (DNase-seq). In parallel, expression of the genes can be measured by RNA-seq, which enables correlation of regulatory DNA activity with gene transcription. Techniques such as Hi-C^104,118 reveal genome topology and can be used to test if regulatory elements and their putative target genes are in physical proximity within nuclear chromatin. Interestingly, many single nucleotide polymorphisms (SNPs) linked to human disease that have been identified through large population-based studies (GWAS) are concentrated within or near regulatory elements. Given the above model of distal regulation of gene transcription by cell-type specific regulatory elements, this leads to our current understanding that GWAS SNPs can identify/modify the function of these elements and thereby control the expression of critical cell-type specific genes. Therefore, we predict that integrated analysis of cell-type specific chromatin accessibility, gene expression and genome topology maps will help to localize kidney disease-associated genetic variants to cell-type specific regulatory elements and to functionally annotate their regulated target genes.

Supplemental Figure 2. Enriched gene ontologies in genes that are expressed at a higher level in glomerular cell cultures vs. cortex cell cultures (all p<1x10^-7).

Supplemental Figure 3. Comparison of DHS profiles from this study to fetal kidney and tubule DHS data previously available from ENCODE. (A) Overlap of all unique DHS from glomerular cultures from this study and the fetal kidney DHS data. (B) Overlap of all unique DHS from cortex cultures from this study and the 6 kidney tubule data sets available through ENCODE (proximal tubule cells, RPTEC; human renal epithelium, HRE; human cortical renal epithelium, HRCE).

Supplemental Figure 4. Physical interactions between kidney GWAS loci associated with differentially accessible DHS and their putative target gene identified by differential gene expression. Long-range genomic contacts deduced from glomerular Hi-C data using the GWAS SNP (red hash mark and orange arcs) and target gene TSS (blue hash mark and purple arcs) as “baits”. Topologically associated domains (TAD) in human glomeruli are indicated when present by the horizontal maroon bars; breaks in the bars represent the boundary between adjacent TADs. Also compare to Table 1 and Supplemental Table 7.

Supplemental Table 1. Patient characteristics, sample identifications and sequencing depth.

Supplemental Table 2. GWAS catalog download on August 24, 2017 selecting for kidney traits.

Supplemental Table 3. Unique SNP positions from GWAS catalog download.

Supplemental Table 4. H3/H4 coloc data from PICCOLO used to prune SNP positions (if prune=TRUE).

Supplemental Table 5. Kidney GWAS SNPs used in this analysis for analysis (n=430) and the subset (n=200) achieving the genome-wide significance threshold (p<5x10-8).

Supplemental Table 6. DESeq2 output for differential gene expression between primary glomerular and cortex cultures.

Supplemental Table 7. Expanded functional annotation data of kidney GWAS-putative target gene pairs.

Supplemental Table 8. Hi-C contacts between kidney GWAS SNPs and putative target gene transcription start sites (UCSC Genome Browser Custom Tract Interact file).

Supplemental Table 9. DomainCaller topologically associated domains (TADs) from freshly isolated human glomeruli.