Unravelling the complex genetics of common kidney diseases: from variants to mechanisms

Abstract

Genome-wide association studies have identified hundreds of loci associated with kidney-related traits such as glomerular filtration rate, albuminuria, hypertension, electrolyte and metabolite levels. Nearly all of these loci are located in the non-coding region of the genome, therefore their target genes, affected cell types and regulatory mechanisms remain unknown. Genome-scale approaches that attempt to identify the associations between DNA sequence variants and disease risk, changes in gene expression levels (quantified through bulk and single-cell methods), regulation of gene expression and other molecular quantitative trait studies, such as chromatin accessibility, DNA methylation, protein expression, metabolite levels, can be used to deliver robust mechanistic inference for translational exploitation. Understanding the genetic basis of common kidney diseases means having a comprehensive picture of the genes that have a causal role in the etiology and progression of the disease, of the cells, tissues and organs in which these genes act to affect the disease, of the cellular pathways and mechanisms that drive disease and of potential targets for disease prevention, detection and therapy.

Introduction

In the past decade, advances in genetics have provided paradigm-shifting diagnostic and mechanistic insight into the development of kidney disease. Family-based studies, later coupled with exome sequencing, have identified a large number of gene mutations in the coding regions of DNA that explain the development of cystic kidney disease, nephrotic syndrome, hypertension and rare kidney diseases.^1,2 These mutations are exceedingly rare in the general population, but they have a substantial impact in affected individuals, and explain the recurrence of disease in specific families. Such genetic disorders usually manifest earlier in life, follow a clear dominant or recessive pattern of inheritance and are known as Mendelian diseases. ³ In contrast to these disorders, most common kidney pathologies that affect the adult population, such as diabetic or hypertensive kidney disease, follow a more complex inheritance pattern.⁴ These polygenic conditions are controlled by nucleotide variants that are commonly observed in the general population, (that is, the allele frequency or variant incidence is higher than 5%)⁵ and each variant only has a small effect size on the risk of disease.⁶

Genome-wide association studies (GWAS) developed in the early 2000’s enabled a systematic interrogation of the association between common sequence variants (also known as single nucleotide polymorphisms (SNPs)) and different disease states and traits in hundreds of thousands of individuals.⁷ GWAS have since reported hundreds of genetic loci that are associated with kidney function, including loci that correlate with estimated glomerular filtration rate (eGFR), albuminuria and hypertension, as well as levels of metabolites and electrolytes present in blood and urine ^{8–10,11,12,13,14}. More than 95% of disease-associated variants discovered thus far are in the non-coding region of the genome.^15,16

GWAS have been exceedingly successful in identifying mostly non-coding sequence variants associated with traits and diseases but this genomic information has not translated into improved diagnostics and therapeutics in nephrology, as our understanding of how non-coding variants cause kidney disease remains incomplete. Causal variants; defined as variants responsible for the observed complex trait, ¹⁷ are often difficult to identify within a genetic locus, as DNA sequences that are in close proximity are inherited together. ¹⁸ Furthermore, owing to secondary chromatin structure [G], genes that are nearest to the significant sequence variant is not always the causal gene.¹⁹ The relatively few loci for which mechanistic studies have been performed, suggest that most disease-causing variants localize to gene regulatory regions, which are usually found in open chromatin and to which transcription factors can bind and regulate gene expression.^20–22 As gene regulation is often cell-type specific, the effect of genotype on gene regulation and expression is also cell type-specific,²² which might explain organ-specific disease development.

In this Review we consider different methods that can be used to prioritize causal variants, causal genes and regulatory mechanisms for kidney disease, including fine mapping, epigenome annotation and analysis of molQTLs, as well as the use of single-cell sequencing analyses to identify target cell types. Finally, we will discuss studies that analyzed kidney disease risk loci and identified causal genes, cell types and underlying disease mechanisms.

Causal variant prioritization

A crucial problem in GWAS analysis is that each genetic locus of interest includes a relatively large number of genetic variants with genome-wide significance. Humans are a relatively young species, which means that only a limited number of recombination events have occurred in the genome. Nucleotide variants observed at specific genomic regions are therefore highly correlated¹⁷. GWAS take advantage of this local genetic correlation and usually only directly identifies a limited number of SNPs; the remaining SNPs are defined by imputation [G] based on data from prior whole-genome sequencing studies²³.

Although this local genetic correlation was crucial for the feasibility and early success of GWAS, as not all variants need to measured, it also complicates the identification of causal variants because hundreds or even thousands of variants in a specific region usually correlate significantly with the observed traits. ²⁴

Based on these limitations, causal variant identification is exceedingly difficult. Identification of causal variants usually involves additional statistical methods such as fine mapping, annotation of regulatory regions and experimental validation. ²⁵ So far only a few disease-causing variants have been identified.

Fine mapping of GWAS variants

Several fine mapping [G] approaches have been developed to refine lists of disease-associated SNPs identified through GWAS. ^17,26,27 Although genetic variants that are physically close to each-other are usually inherited together, which increases the number of analyzed variants, increasing the sample size by direct measurement of a larger number of individuals, or by imputation, can help to identify variants that are more likely to be disease-causing.²⁸ Increasing the sample size and analyzing different populations is important, because alleles of SNPs encoded in close proximity within a chromosome are usually co-inherited, and genetically distinct populations therefore tend to have divergent haplotype [G] structures and frequencies. ²⁸ The analysis of large multi-ethnic cohorts and the use of whole-genome sequencing integrate these fine-mapping approaches, and should offer new powerful insight into the identification of causal disease variants. In addition to the larger sample size and diverse population better disease phenotype mapping, such as deep phenotyping [G] and focusing on endophenotypes appear to be important. Endophenotypes are heritable intermediate phenotypes associated with disease. For example, IgA1 glycosylation defects are inherited and constitute a heritable risk factor for IgA associated kidney disease. ²⁹

Epigenome annotation

In our current working model, a causal variant(s) results in quantitative changes in gene expression. Therefore, the methods most commonly used for causal variant prioritization include the annotation of regulatory regions to identify SNPs with regulatory function. Epigenomic annotation, including the identification of regions of open chromatin and histone tail modifications, as well as cytosine methylation maps, can narrow down areas in the genome that are associated with the disease state of interest.

The DNA in the nucleus is densely packaged to form chromatin, which is a DNA–protein complex. However, genomic regions that are actively involved in gene regulation must be accessible to transcription factors and are therefore located in areas of open chromatin. For a genetic variant to influence gene expression, it must be located on a gene regulatory region. ¹⁵ Several methods have been developed to identify areas of open chromatin (FIG. 1a). Careful titration of DNase I, coupled with next-generation sequencing (DNase-seq), has provided crucial insight into cell-specific and tissue-specific gene regulatory regions as open chromatin regions are more sensitive to DNaseI digestion than areas that are tightly packed.³⁰ DHS is a highly sensitive method to identify open chromatin, but difficult to perform. Transposases preferentially insert into areas of open chromatin and assay for transposase-accessible chromatin using sequencing (ATAC-seq) can also be used for genome-wide annotation of open chromatin (FIG. 1a).³¹ Although ATAC-seq might be slightly less comprehensive than DHS, it has gained popularity as a very robust and reproducible method that is fairly easy to perform; currently, ATAC-seq is probably more widely used than DHS.

Figure 1. Epigenome annotation methods for causal variant prioritization.

a)Genetic effects on the epigenome are an important component of the genetic risk of a disease. | DNase I hypersensitivity (DHS) analysis¹⁴⁴, identifies DNA that is not tightly wound around histones, ie is in an area of open chromatin. These DNA elements are more sensitive to enzymatic digestion by DNase1,¹⁴⁵ In contrast, DNA which is tightly wrapped in nucleosomes is more resistant to digestion. ¹⁴⁶ The DNAse digested ends are then enriched and sequenced. ¹⁴⁶ (1A)

b)ATAC-seq uses a hyperactive Tn5 transposase, which is an RNase that inserts sequencing adapters into open regions of the genome during a process called tagmentation. ³¹ The tagged DNA fragments are then purified, PCR-amplified and sequenced using next-generation sequencing.(1B)

c) Modifications to histone proteins, such as methylation, acetylation and trimethylation, can be detected using Chromatin Immunoprecipitation (ChIP) based methods. ³⁵ These modifications can then be used as indicators of the chromatin state which is associated with gene activation or repression. ³⁵

d) Analysis of Chromatin Conformation. Chromatin conformation experiments examine which pairs of DNA loci are in contact with each other, genome wide, and allows long distance contacts to be elucidated. The process involves DNA digestion, biotin labelling and formation of ligation products, which are then sequenced. ^147,148

These methods of epigenome annotation, identify cis-regulatory elements (CREs), which contain causal variants and associate with gene expression.

Nucleosomes are structural units of chromatin that comprise DNA coiled around histones, which are basic proteins. Gene regulatory regions have specific histone tail modifications that can be used to locate them³². Promoter [G], enhancer [G] and insulator [G] regions can all be identified by specific histone tail modifications^33,34. The most commonly used method for the systematic characterization of histone tail modifications is chromatin immunoprecipitation followed by sequencing (ChIP-seq), which uses an antibody to enrich for histone tails and analyzes DNA sequences associated with specific histone tail modifications (FIG. 1b).³⁵ Specific combinations of histone tail markers, can identify, for example, regions of repressed transcription (trimethylated histone 3 lysine9 (H3K9me3) and H3K27me3), transcriptional start sites of actively transcribed genes (H3K4me3) and acetylated H3K27 (H3K27ac)), areas of active transcription (H3K36me3) and active enhancers (monomethylated H3K4 (H3K4me1) and H3K27ac).³⁵

Methylation of the 5^th carbon of cytosine, which generates 5-methylcytosine (5mC), is a well-studied epigenetic mark³⁶. The main role of cytosine methylation is transcriptional repression, mostly for transposable elements [G]. Most cytosines are therefore highly methylated in the genome but active gene regulatory regions are usually characterized by low cytosine methylation.³⁷ Analysis of genome-wide cytosine methylation therefore enables the identification of gene regulatory regions such as active promoters and enhancers^38,37. DNA methylation can be analyzed in archived human tissue samples and can thus be more readily applied to clinical samples than histone-ChIP-Seq or ATAC-seq analysis³⁹.

Publicly available resources, which allow analysis of epigenetic material include, the Encyclopedia of coding elements (ENCODE) and the International human epigenome consortium (IHEC). These projects have generated a substantial number of tissue-specific and cell type-specific epigenomic maps, which are important reference sources. ^40,41 Unfortunately these datasets do not include a large number of human kidney samples. Epigenetic studies have highlighted the substantial enrichment of GWAS signals on regulatory regions such as enhancers in disease-relevant tissues¹⁵

Analysis of transcription factor binding

Our working hypothesis is that disease causing variants alter the binding strength of transcription factors, which leads to quantitative changes in gene expression. Classic analyses of transcription factor binding include electrophoretic mobility shift analysis (EMSA), which is used to study protein–DNA interactions⁴². EMSA can determine if a protein or mixture of proteins is capable of binding to a given sequence of DNA or RNA, and can sometimes indicate if multiple protein molecules are involved in the binding complex. Transcription factor ChIP (TF-ChIP) experiments can analyze transcription factor binding at a genome-wide level ⁴³ and are important for causal variant identification because they can directly quantify differences in transcription factor binding that are determined by the underlying sequence variants⁴³. This approach can be used to generate genome-wide maps of transcription factor binding regions, also known as cis-regulatory elements [G].⁴⁴ Importantly, TF-Chip requires an antibody for detection and this antibody must be both highly sensitive and specific to ensure that the antibody–transcription factor complex is enriched in the genomic areas where transcription factor binding occurs without significant confounding due to unspecific antibody binding⁴⁵. Several computational methods have also been developed that predict transcription factor binding based on DNA nucleotide sequences, such as the SNP effect matrix pipeline (SEMpl)⁴⁶,which estimates the affinity of transcription factor binding based on the combination of ChIp-seq data and DNA sequences mapped to areas of open chromatin. These novel methods can potentially predict alterations in transcriptional outcomes based on the identified nucleotide variations,⁴⁶ but they require experimental validation, which can be performed through the CRISPR-Cas9 mutagenesis system. ⁴⁷

Chromatin conformation

Another challenge in the analysis of gene regulatory regions is the ability to link regulatory elements in the genome to their target genes. This analysis is particularly problematic because causal variants are enriched in enhancer regions. Promoters are always located at the 5’ end of their target genes but enhancers can be located in intronic regions, as wells as upstream or downstream of the target gene.⁴⁸ Enhancer regions can be as far as 250 kilobases (kB) away from their target gene and, due to chromatin looping, the gene nearest to the enhancer is not always the target or causal gene.⁴⁹

Chromatin conformation capture experiments can be used to measure these loop interactions (FIG. 1c).⁵⁰ These methods identify 3D interactions between genomic loci that cannot be detected when the genome is studied in a single dimension, and can uncover biologically relevant promoter–enhancer interactions⁵¹. One of the most comprehensive methods of chromatin conformation capture is Hi-C, which uses high-throughput sequencing to characterize all 3D genomic interactions.⁵²

Chromatin interaction analysis by paired-end tag sequencing (ChIA-PET) is another technique used to determine long-range 3D chromatin interactions...ChIA-PET incorporates enrichment for histone tail modification by ChIP-based enrichment, therefore it can also provide information on the type of regulatory element (for example, promoter or enhancer) and not just the interaction ^53,54 (FIG. 1c). Understanding long-range genomic interactions, especially enhancer–promoter units that regulate gene expression, ⁵⁵ is important for linking GWAS variants to their target genes.

Experimental validation of regulatory elements

Genome-wide mapping and sequencing approaches such as DNase-seq, ATAC-seq for open chromatin, ChIP-seq for annotation of histone tail modifications and Hi-C data on 3D chromatin interactions are excellent tools for identifying regulatory regions, but the effect of the genetic variant on regulatory activity is still difficult to predict. Experimental validation of regulatory activity detected by high-throughput techniques is therefore crucial. Self-transcribing active regulatory region sequencing (STARR-seq) is a popular massively parallel reporter assay [G] (MPRA) method⁵⁶ that enables genome-wide quantitative measurement of enhancer activity. ⁵⁵ In STARR-seq, following fragmentation of genomic DNA, potential regulatory regions are placed downstream of a minimal promoter in a reporter construct.⁵⁷ Strong regulatory regions can self-transcribe and their abundance therefore increases, which is measured in a high-throughput manner and allows the simultaneous testing of the activity of hundreds of enhancer regions to identify active enhancers^58,57,59...

The best and most direct approach for causal variant testing is the use of CRISPR-Cas9-mediated single nucleotide genome editing followed by read-outs of gene expression (FIG. 2)⁶⁰ Although CRISPR-Cas9-based methods require substantial optimization, as they can currently only test a few variants at a time and can cause off- target effects,⁶¹ they can provide the most direct evidence for the functional importance of a specific SNP. In the past few years, several groups have developed large-scale editing methods with single-nucleotide resolution that enable the screening of a large number of SNPs at a specific genomic region, and evaluation of their role in the regulation of gene expression in a cell type-specific manner.⁶² A limitation of such analyses is that they use cultured cells, which often lack expression of the target gene regulated by the SNP of interest. Tissue-specific organoids can also be used for such genome editing studies but organoid cells remain poorly differentiated, (FIG. 2)^63,64, which can make the extrapolation of results to fully differentiated human tissues challenging. Further complications arise from the likely presence of more than one single causal variant in any given region, as well as the potential existence of causal variants that are unique to specific ethnicities and cannot be generalised. ⁶⁵ Overall, experimental validation of computationally-prioritized causal variants remains crucial. However, these studies are difficult to perform and have low throughput — prioritization efforts are therefore essential.

Figure 2. Function validation of causal variants.

Several methods can be used to validate causal variants and elucidate the underlying mechanisms that link the sequence variant to disease risk — variant-to-function analysis. Modelling risk variants using Crispr-Cas9 system in single cells or in 3D structures (organoids) and animal models (zebra fish and mice) can provide information on the effect of a sequence variant on gene expression levels.

Analyses of healthy tissue, which can be microdissected into glomeruli and tubules to generate kidney tissue compartment-specific data, or analysed by single-cell technologies can provide information on homeostatic gene regulation, including information on the cell types that express genes whose expression associates with sequence variants. Further validation of the potential mechanisms underlying the modulation of disease risk can obtained by analysing samples obtained from patient with kidney disease.

Target gene identification

For biologists, the most important element of investigating how complex traits contribute to disease is the identification of the target gene. Traditionally, GWAS loci were annotated by gene names based on the proximity of the signal to those genes, or according to subjective author preference. Although the precision of the heuristic approach for target gene identification is still fiercely debated, for several loci, the target gene of the causal variant is clearly located at a distant location. ⁶⁶ In nephrology, the most widely known example is the chromosome 22 locus, which was initially thought to be associated with MYH9, although it later became clear that APOL1 was the correct target gene.⁶⁷ This example highlights the need to develop novel datasets that can be used to determine the effect of genetic variation on gene expression — such datasets should ideally be cell type-specific. Although small cell type-specific datasets have been developed for isolated blood cells⁶⁸, at present most available datasets are based on bulk tissue. Cellular heterogeneity and changes in cell proportions remain an important confounder in such bulk analyses. ⁶⁹

Molecular quantitative trait loci

Molecular quantitative trait loci [G] (MolQTLs) are genetic variants that are associated with molecular traits and are classed according to the quantitative trait being analysed. Examples of molQTLs include expression QTLs (eQTLs), splicing QTLs (sQTLs), chromatin accessibility QTLs (caQTLs), methylation QTLs (meQTLs), protein QTLs (pQTLs) and metabolic QTLs (mQTLs)..⁷⁰

Expression QTLs

eQTL analysis focuses on the statistical relationship between SNPs and gene expression⁷¹ and involves high-throughput genome-wide measurement of gene expression and its associated genetic variants, which is usually performed through the use of genotyping arrays. ⁷¹ Most eQTL analyses are limited to cis-eQTLs, meaning that the genetic variants are within 1 megabase (Mb) of the expressed genes.⁷² Cis-eQTLs directly alter the expression of a gene through local promoter–enhancer interactions, whereas trans-eQTLs affect the expression of a regulator, usually a transcription factor, which alters the expression of large number of target genes by directly binding to their local promoter region.^73,74 The focus on cis-eQTLs is related to the issue that current eQTL datasets are relatively small and underpowered for trans-eQTL analysis, in which the variant is further away from the target gene and the mode of regulation is likely indirect.⁷³ Analysis of trans-eQTLs might be highly informative for target gene identification and to highlight critical disease-causing pathways, as it can reveal which genes are co-regulated by the same disease-associated transcription factor, and are thus likely to be involved in the pathogenesis of the disease⁷⁵. At present, eQTL datasets that are large enough for trans-eQTL identification are only available for blood samples.^76,77.

The Genotype Tissue Expression (GTEx) consortium has generated eQTL datasets for ~ 50 tissue types from ~1,000 samples.⁷⁸ The systematic multi-tissue cis-eQTL analysis in GTEx showed that ~70% of cis-eQTLs are observed in multiple cell and tissue types (that is, they are not tissue-specific), whereas some cis-eQTLs show tissue or cell-type specificity.⁷⁹ By contrast, trans-eQTLs might show greater cell-type specifity.⁷⁹ However, due to the low number of genes and variants that have been validated experimentally, the true positives or negatives in these analyses remain uncertain and further validation is required.

The human genome contains two copies of each allele and each copy is expected to contribute to 50% of the total transcript level. However, allelic imbalance, whereby two alleles of a gene are expressed at different levels instead of the expected 50:50, can also be observed and the analysis of this allele-specific expression (ASE) can substantially boost the power of eQTL analyses⁸⁰. Allelic imbalance is caused by sequence variations that lead to differences in the strength of transcription factor binding between the two alleles^81,82. Compared with classic eQTL analysis, which is based on an association analysis between different individuals, ASE analysis investigates the genotype effect on gene expression within the same sample, when the individual is heterozygous for the specific allele. ⁸⁰ As ASE analysis focuses on two alleles from the same individual, which can reciprocally serve as internal controls, minimizing the effects of potential confounding factors and environmental biases.⁸⁰ Although ASE analysis is associated with multiple technical difficulties, such as bias in the mapping of RNA sequencing (RNA-seq) reads towards the reference allele⁸³, its ability to quantify the effect of gene expression in the same sample is a key advantage.⁸⁰

Splicing QTLs

Another critical consideration for the identification of target transcripts is the analysis of alternative splicing [G]. Splicing is an important factor in the diversification of gene expression and each gene has, on average, at least 3–4 isoforms.⁸⁴ Although early reports did not suggest a key role of alternative splicing in disease development, subsequent datasets indicate that ~25% of genotype-mediated regulation of gene expression is mediated through effects on gene splicing.⁸⁵ sQTL analyses use data on protein isoforms and can identify genotype-driven changes in isoform expression. An sQTL dataset generated for the prefrontal cortex region of the human brain indicated that schizophrenia GWAS variants associated with changes in the transcripts of splice isoforms. ⁸⁶ This finding supports the concept that alternate splicing might be a mechanism whereby genetic risk variants lead to phenotype development.

Chromatin accessibility QTLs

caQTLs can be identified when genotype assessment and quantitative measure of open chromatin can be performed in the same sample⁸⁷. The integration of caQTL data with GWAS can be extremely useful for causal variant prioritization.⁸⁷ The GWAS identifies the SNPs that are associated with disease risk, whereas the caQTL highlights which of those SNPs are likely to be non-coding regulators of gene expression, by identifying which SNPs are not only located in areas of open chromatin, but also correlate with changes in gene expression patterns. These SNPs are therefore likely causal variants that drive the increased or decreased risk of disease. In 2018, researchers generated a caQTL dataset for unstimulated and stimulated human CD4 T cells.⁸⁷ This study reported that genotype significantly affected caQTL and that these areas of open chromatin overlapped with previously reported autoimmune disease GWAS risk variants, which could be mapped and prioritized based on the caQTL data. ⁸⁷

Computational integration of GWAS and molQTL datasets

Given that disease-causing variants are mostly unknown, analysing single SNP–eQTL or SNP–epigenome datasets might lead to misleading conclusions. Several methods have been developed to integrate genotype, disease risk, gene expression and epigenome datasets, and the development of novel computational methods for high-dimensional data integration remains an active area of research.

Integration of GWAS and QTL datasets is the current gold standard for the prioritization of target genes (FIG. 3). Although QTL–GWAS methods do not perform causal inference, they are useful in prioritizing GWAS genes for experimental follow-up. The goal of these methods is to estimate the properties of the genetic relationship between a molecular trait and the GWAS trait (FIG. 3a). Methods such as transcriptome-wide association studies (TWAS)⁸⁸, summary Mendelian randomization [G] (SMR)⁸⁹, PrediXcan and MetaXcan⁹⁰, for example, test the strength of the genetic correlation between gene expression and SNPs at specific genomic regions.

Figure 3. Quantitative trait analysis for causal gene prioritization.

Expression of Quantitative trait analysis determines the association between genetic variations and gene expression in the same tissue samples.

a| In this example, GWAS identified an association between the sequence variant T/T and chronic kidney disease. This variant is located in an open area of chromatin within a region in which the lysine (K) at position 27 of histone 3 (H3) is acetylated (ac) H3K27ac, which is a DNA enhancer and therefore suggests active gene transcription. The T/T sequence variant affects binding of the transcription factor to the enhancer and results in a decrease in the expression of DAB adaptor protein 2 (DAB2) compared with the T/A and A/A variants. In turn, this decrease in DAB2 expression correlates with a decreased of estimated glomerular filtration rate.

b| Genetic correlation can result from direct causality, in which a variant has a direct effect on a quantitative trait, but it can also result from pleiotropy, in which a variant affects multiple traits, or linkage, which refers to the existence of two variants that are associated by linkage disequilibrium (that is, a non-random association of alleles at different loci) and have independent effects on different phenotypes. These different underlying causes of genetic correlation complicate the identification of causal variants.

c| [ Integration of GWAS–molecular QTLs (molQTLs). Overlap between a GWAS variant, an eQTL and a chromatin accessibility QTL (caQTL) indicates that the sequence variant not only correlates with a disease risk trait (for example, eGFR) but also with gene expression and chromatin accessibility. This overlap potentially indicates that the sequence variant is located in gene regulatory region (that is, an area of open chromatin) where it can modulate expression of a causal gene involved in disease risk.

TWAS use eQTL data to ‘impute’ the total expression of a given gene across a large cohort of genotyped individuals, followed by a test of association with disease risk. In TWAS, direct causation is not easily distinguished from genetic correlation due to pleiotropy, as a single SNP might affect multiple molecular phenotypes, including the expression of more than one gene, or differentially affect the expression of the same gene in different tissues, as well as regulate multiple processes, including gene expression and DNA methylation (FIG. 3b). ⁷⁰ Many molecular processes, such as gene expression, DNA methylation and histone variation, are controlled by the same cis-regulatory elements. ^91,92 Therefore, TWAS serves as a first step in the prioritization of putative target genes but experimental validation is still required to establish causality⁸⁸. Discriminating these multiple potential phenotypes can be facilitated by the integration of orthogonal datasets [G], which is a process known as triangulation. ⁹³

Other methods, such as the Coloc and Moloc, estimate the posterior probability [G] of colocalization, which is defined as the presence of at least one shared causal variant between the eQTL datasets and GWAS (FIG. 3c). ⁹⁴ This method requires prior probabilities to be defined at a SNP level. ⁹⁴ The output of colocalization consists of the posterior probabilities of variants that affect both traits, such as kidney disease (GWAS) and gene expression (eQTL) or SNPs that independently associate with each trait. The posterior probability values have an upper limit of 1.0 and do not give an indication of the strength of the correlation between the SNPs and the traits. However, TWAS measures the strength of the association(s)⁸⁸ and can therefore complement a colocalization analysis ...

SMR is another tool that assesses causality of an association between a risk factor and a clinically relevant outcome. ⁹⁵ In contrast to observational studies, with SMR, genetic variants are used as an instrumental variable [G] and the effect of these variants on a trait, disease or gene expression is investigated. ^96–98,99 The use of genetic variants helps to circumvent problems with confounding or reverse causation.⁹⁶ The variants must fit three assumptions — they must be relevant (that is, they must associate with the outcome of interest); they must not be affected by unmeasured confounders; and they must affect the outcome only through their effect on the risk factor of interest (that is, they must be exclusive). ⁹⁶ The first assumption of the variant can be tested, whereas the other two are determined by sensitivity analyses. ¹⁰⁰

Various modifications have been made to SMR to facilitate the analysis of GWAS studies. Two-sample Mendelian randomization only requires GWAS summary statistics, which is useful if the exposure genetic variant or SNP and the outcome SNP are assessed in different datasets. ¹⁰¹ Genetic pleiotropy (FIG. 3c) can be addressed by using multivariate SMR analysis, which reduces bias by simultaneously estimating the effect of all exposures on the outcome.^102,103 A 2019 study reported the development of transcriptome-wide Mendelian randomization (TWMR), which is a multivariate Mendelian randomization method that uses gene expression levels as the exposure.⁹² This method was applied to summary data from a publicly available GWAS dataset combined with an eQTL meta-analysis performed in blood to investigate causal associations between gene expression and 43 complex traits.⁹² This method was very successful and detected 3,913 causal gene–trait associations, 36% of which did have not a nearby genome-wide SNP, and would have therefore been more difficult to detect.⁹² These findings suggest that the TWMR method might help to integrate GWAS and eQTL data, and detect associations that would have otherwise been missed owing to the lack of power in GWAS.⁹²

SMR has been used for example to examine the relationship between serum urate levels and CKD¹⁰⁴. Human observational data indicated that serum uric acid levels have a causal role in the development of CKD. These studies used SNPs that are known to influence serum urate levels (for example, rs734533) as the instrumental variable¹⁰⁵. Whereas the analysis of a single SNP associated with urate levels also showed an association with CKD progression in a small cohort study ¹⁰⁴, a similar analysis using a collection of SNPs in a cohort with 110,347 participants¹⁰⁷ found no causal associations between serum uric acid levels and eGFR and CKD. ²⁰ SMR can also identify epigenetic changes, such as changes in methylation¹⁰⁸ and chromatin accessibility¹⁰⁹ that mediate the effect of a genotype on gene expression.

Defining disease-causing cell types

The identification of causal cell types is strongly linked to the discovery of causal variants and causal genes. As previously discussed, causal variants have a disease-specific enrichment pattern ¹¹⁰, which is likely due to the fact that the variant is only active when localized to open chromatin regions and that patterns of chromatin accessibility are cell type-specific. ²⁰ An approach to determining the involvement of cell types present in different kidney tissues it to microdissect kidney tissue into glomerular and tubular compartments, which can then be used for sequencing analyses (FIG. 2); an alternative to these bulk sequencing approaches is the use of single-cell technologies.

Remarkable advances in single-cell analytical methods currently allow the analysis of gene expression levels, open chromatin, surface protein levels and spatial organization of the genome on an unprecedented scale. This methodological revolution occurred around 2015, and included advances in barcoding methods and the development of microfluidic devices for cellular encapsulation [G]. ¹¹¹ Currently, barcoded droplet-based RNA-seq costs ~US$0.01–0.02 per cell, which makes it an affordable approach for obtaining comprehensive information on the expression of >1,000 genes.¹¹² These new methods enable transcript-based definition of cell types, which enables more in-depth cellular characterization than morphology-based annotation²¹, and have provided crucial information for the identification of disease-causing cell types and regulatory regions. ²¹

In addition to expression analysis, open chromatin annotation at the single-cell level through single-nucleus ATAC-seq (snATAC-seq) analysis will likely further revolutionize GWAS signal annotation. Bulk tissue approaches usually only detect signals from predominant cell types but snATAC-seq enables the identification of rare cell types and subpopulations of common cell types. ^113,114 snATAC-seq analysis therefore allows the efficient mapping of GWAS regions into narrow cell type-specific open chromatin signals.¹¹³ Unfortunately, at present, combining snATAC-seq and snRNA-seq analyses is difficult, although several new methods have been developed.¹¹⁵ Single-cell annotation and analytical methods will be crucial for further GWAS annotation.

Variant-to-function analyses for kidney related traits

As previously discussed, GWAS studies have defined the genetic architecture of a variety of kidney-related traits.^8–10,116 These studies include hypertension GWAS performed in a variety of large cohorts¹⁴, eGFR GWAS in ~1 million Europeans and multi-ethnic cohorts, ^8–10 as well as electrolytes and protein levels in urine, and of serum and urinary metabolite concentrations.^{8–10,116,117} The large sample size in these studies, up to 1 million participants, ^8–10 has enabled the identification of hundreds of DNA regions with very strong association with a variety of renal traits, such as eGFR, serum urea level, blood pressure and albuminuria. Some genetic loci have been identified in multiple GWAS studies in which different traits were mapped such as blood pressure, eGFR, ^8,10,116, and albuminuria. ^13,118

Unfortunately, very few kidney specific epigenome datasets are currently available for the interrogation of genetic loci (Fig2). The Encyclopedia of DNA elements (ENCODE), the Roadmap epigenomics project and the International Human Epigenome Consortium (IHEC), which characterized functional (gene regulatory) elements in multiple tissue and cell types, contain a relatively small number of human kidney tissue samples for epigenomic analysis. These datasets include some DHS analysis of fetal human kidney samples and genome-wide histone ChIP-Seq data for adult human kidney tissue samples. The human kidney is also very sparsely represented in the Genotype–tissue expression (GTEx) database — the 2019 version (V8) of this dataset includes data from 73 whole human kidney samples.

Below we discuss various kidney studies that have created novel resources for variant-to-function analyses in nephrology; some of these studies have also identified causal genes and the potential underlying mechanisms that link these genes to kidney disease.

In a 2019 study, researchers generated DHS, chromatin conformation and gene expression profiles from cultures of primary human kidney glomerular and cortical cells¹¹⁹. These new datasets allowed the prioritization of GWAS variants. This approach identified 42 kidney GWAS loci, which were physically and functionally connected to 46 potential target genes. Although this is an important initial dataset, its use of cultured cells instead of human tissue remains a key limitation as cell culture is likely to alter the gene expression pattern and the phenotype of the cells.

In 2017, we reported one of the first human kidney eQTL datasets, in which we analysed ~100 whole human kidney cortex samples from European donors from The Cancer Genome Atlas (TCGA) database¹²⁰. In this study, we identified 1,886 genes whose expression levels correlated with sequence variants, also known as eGenes¹²⁰. To identify potential genes involved in CKD, we sought to identify genetic variants associated with kidney function that uniquely influenced gene expression in the kidney by performing a colocalization analysis using our kidney eQTL data and data from GWAS studies. This approach identified lysosomal beta A mannosidase (MANBA) as a potential gene linked to CKD development. We found statistically significant colocalization between genetic variants associated with MANBA expression in the kidney and CKD GWAS variants, confirming the potential role of MANBA as a CKD target gene. ¹²⁰

In a 2018 study, researchers analyzed gene expression data from 280 whole kidney tissue samples — 180 samples from the Transcriptome of renal human tissue (TRANSLATE) study¹²¹ and 100 TCGA samples.¹²⁰ This study reported 3,786 eGenes (approximately 17.2% of all kidney genes); 35 of these genes had an eQTL that overlapped directly with CKD GWAS loci. However, as discussed earlier, the direct overlap method has substantial limitation, as the causal variant on multiple loci remains unknown. Mendelian randomization analysis supported a causal role of NAT8B, CASP9 and MUC1 in the development of kidney disease.¹²¹

The Nephrotic Syndrome Study Network (NEPTUNE) analyzed the correlation between genotype and gene expression in ~150 samples from individuals with nephrotic syndrome, using microdissected kidney tissue¹²². The use of microdissection enabled the authors to distinguish between genes regulated in tubular or glomerular compartments. However, the analysis of gene expression was ascertained from microarrays and the analysed cohort included individuals with mixed ethnicities and varying degree of disease severity.¹²² This analysis only reported 894 glomerular eGenes and 1,767 eGenes in the microdissected tubulointerstitium.¹²² This is likely due to the heterogeneity of samples used for the analysis. Furthermore only 53% of glomerular eQTLs and 66% of tubular QTLs could be replicated in the GTEx database. ¹²² Key differences between the GTEx and NEPTUNE studies respectively include — the use of samples from healthy individuals versus individuals with nephrotic syndrome, the use of whole kidney samples versus microdissected tubules and glomeruli, and the use of RNA-seq versus gene microarrays to quantify gene expression. ⁷⁸

In 2018 we reported an eQTL dataset generated from 120 microdissected glomerular and tubule samples obtained from individuals of European descent with normal kidney function (eGFR> 60ml/min/1.73m2) and structure (such as <10% fibrosis and sclerosis on histological analysis)²². Furthermore, using an in silico cellular deconvolution [G] method, we only included samples with similar cellular composition, to minimize confounding due to cellular heterogeneity.. This dataset identified more than 4,000 glomerular and tubular eGenes with 70–80% replication rate in other tissue samples (using the GTEx database). The high replication rate validates the identified eGenes as true eQTLs, as it has been shown that most cis-eQTLs are shared across multiple tissues⁷⁹. Computational integration of the GWAS and eQTL datasets using the coloc prioritized 27 genes, for which association with kidney function and changes in gene expression originated from the same genetic variants. These genes are likely to be causal genes for CKD. ²²

To understand where these putative causal genes are expressed we integrated the GWAS target genes with the mouse single-cell transcriptome maps. We found that 10 of the CKD GWAS target genes showed almost exclusive proximal tubule specific expression pattern ²² We further analyzed the region in chromosome 5 that contains complement C9 (C9) and DAB adaptor protein 2 (DAB2), as this area showed significant association with eGFR in GWAS studies and the same area showed association with DAB2 expression in microdissected human kidney samples ²². Bayesian colocalization indicated significant colocalization between tubule-specific DAB2 levels and SNPs identified in an eGFR GWAS¹². Part of the GWAS locus also overlapped with an enhancer region, identified by presence of H3K27ac as detected by ChIP-seq²². This genetic region showed markers of gene regulatory elements, but not in any other tissue samples.²² Finally, we generated mice with tubule-specific heterozygous deletion of DAB2 to show the functional role of DAB2 in kidney tubules. The risk allele was associated with higher DAB2 levels and accordingly, animals with only one functional copy of tubular DAB2 were protected from kidney disease induced by folic acid.¹²³

In addition to DAB2 and MANBA, earlier studies also suggested role for uromodulin (UMOD) in CKD^12,124,125 Although UMOD has not been identified as an eGene in the above mentioned eQTL studies, ¹⁰ direct analysis of risk genotype and UMOD expression indicates that UMOD levels are higher in individuals with a CKD risk variant than in those that do not carry a risk variant.¹²⁵ Transgenic mice that expressed increased levels of tagged UMOD, developed salt-sensitive hypertension and kidney lesions¹²⁶, features that are also observed in individuals who are homozygous for UMOD promoter CKD risk variants.¹²⁵ Increased UMOD expression resulted in the activation of the sodium cotransporter solute carrier family 12 member 1 (encoded by Slc12a1, also known as NKCC2).¹²⁵ Moreover, inhibition of this sodium cotransporter with furosemide lowered blood pressure most effectively in patients with hypertension who were homozygous for UMOD promoter risk variants.¹²⁵

GWAS studies also identified a strong association between kidney function and genetic variants in the region that encodes SHROOM3....^12,127 Although SHROOM3 was also not identified by GWAS–eQTL integration, studies with mouse models have shown that disruptions in this gene can affect glomerular function. ^127,128SHROOM3 is expressed in podocytes and is necessary for the development and/or maintenance of the complex podocyte architecture.¹²⁹ Shroom3^−/− mice developed marked glomerular abnormalities, including cystic and collapsing and/or degenerating glomeruli, as well as marked disruptions in podocyte arrangement and morphology.¹²⁸ These podocyte-specific abnormalities might result from altered Rho–kinase–myosin II signaling and the loss of apically-distributed actin.¹²⁸

In a different study¹²⁷, the risk allele A of the SNP rs17319721 correlated with increased SHROOM3 expression in kidney allografts. Analysis of the SNP indicated that the intronic region of SHROOM3 contained a transcription factor 7–like 2 (TCF7L2)-dependent enhancer element that increased SHROOM3 transcription.¹²⁷ Unlike in the previous study, which focused on podocytes¹²⁹, SHROOM3 was shown to localize to kidney tubules¹²⁷ and treatment of renal tubular cells with TGFβ1 upregulated SHROOM3 expression in a β-catenin and TCF7L2–mediated manner. SHROOM3 overexpression facilitated canonical TGFβ1 signaling and increased expression of α1 collagen (COL1A1)¹²⁷. Inducible and tubular cell–specific knockdown of Shroom3 markedly abrogated interstitial fibrosis in mice with unilateral ureteric obstruction.¹²⁷

In summary, several key datasets have been generated for the functionalization of GWAS variants, including epigenome and molQTL datasets. These datasets have been instrumental in functionalizing a few genetic loci, including UMOD, MANBA, DAB2 and SHROOM3. A key bottleneck remains in the experimental validation of the prioritized genes and variants.

Conclusions

Human genetic studies have led to paradigm-shifting observations in relatively rare monogenic forms of CKD, including focal segmental glomerulosclerosis (FSGS), and in polygenic forms of kidney disease such as membranous nephropathy.¹³⁰ In case the of FSGS, this work determined that the podocyte was a key causal cell and identified a limited set of pathogenic pathways for therapeutics development.¹³¹ In the case of membranous nephropathy, these studies enabled precision medicine approaches and the discovery of novel diagnostic, prognostic and therapeutic markers.¹³²

Worldwide, hypertensive and diabetic kidney disease accounts for >75% of cases of CKD and end-stage kidney disease.¹³³ understanding the pathogenesis of these conditions is fundamental to improve treatment strategies and patient outcomes. Although some exome sequencing studies indicate that coding region mutations might contribute to ~9 % of CKD risk,¹³⁴ most studies consistently indicate that majority of genetic variation that explains heritability of common diseases occurs in non-coding regions of the genome. ^15,16 Therefore, identifying kidney disease causal genes, cell types and pathways by defining the biological pathways highlighted by GWAS is crucial. These studies will not only improve our understanding of the pathogenesis of kidney disease but might also identify key targetable pathways.

GWAS causal variants influence the regulation of target genes and are thus usually enriched in gene regulatory regions of open chromatin. These causal variants often alter transcription factor binding strength and quantitatively change the expression of target genes in a cell type-specific manner —we and others have catalogued genotype-driven variation in gene expression in the glomerular and tubule compartments of human kidneys²². ¹³⁵ Integrating various data sets such as GWAS, molQTLs, single-cell gene expression and epigenome and using robust computational methods for analysis, which are continuously being developed, is of crucial importance and has been successful in identifying putative disease-causing genes. We strongly believe that such target genes must also be validated in appropriate model systems. Moreover, the use of kidney single-cell gene expression analyses, for example, enables the identification of the kidney compartment in which these disease-associated genes are enriched.²¹ Genetically modified mouse models, such as those used for Umod¹²³, Dab2²² and Shroom3^126,127, can then be used to study the effect of these disease-associated genes.

Genetics is not the only factor that explains CKD development but an analysis that involved the pharmaceutical industry, indicated that the likelihood of successful drug development is substantially increased when a causal gene and/or its pathway are targeted.¹³⁶ This observation is further supported by studies in cardiology, where genetic studies have identified key disease-driving genes such as PCSK9, which led to the development a new FDA-approved drug for lowering cholesterol levels.¹¹³ Defining the target genes, cells types and underlying mechanisms involved in kidney disease will be crucial for the development of new translational approaches and drugs in nephrology, which are desperately needed.

Table 1.

Techniques applied in variant-to-function annotation

Research objective	Techniques	Ref
Identification of areas of open chromatin	DNase hypersensitivity analysis ATAC-seq	144,145 31
Assessment of chromatin conformation	Hi-Ca ChiA-PET HiChIPb	52 53 149
Annotation of histone tail modifications	ChIP-seq	150
Assessment of DNA methylation	Methylation arrays (Illumina 450K and 850K) Whole-genome bisulfate sequencing Reduced representation bisulfate sequencing Isoschizomer-based analysis (HELP)	151,152 153 154 155
Measurement of enhancer activity	Massively parallel reporter assays STARR sequencing	156 58,59
Single-cell analysis of molecular traits	scRNA-seq scATAC-seq scM&T seq scNMT seq CITE-seq. REAP-seq	21 113 157 158 159 160
Molecular QTL studies	eQTL caQTL sQTL hQTL meQTL pQTL mQTL	71 87 86 161 162 163 164

^aHi-C is a chromatin conformation capture method that includes the use of high-throughput sequencing.

^bHiChIP is another method of chromatin conformation capture, in which contacts between different DNA sequences are formed before lysis and chromatin immunoprecipitation is subsequently performed on these contacts. ¹⁴⁹

^cIn this method, chromatin immunoprecipitation is performed with an antibody that recognizes 5-methyl cytosine (MeDIP). ATAC-seq, assay for transposase-accessible chromatin using sequencing; ChIA-PET, chromatin interaction analysis by paired-end tag sequencing; ChIP-seq, chromatin immunoprecipitation followed by sequencing; caQTL, chromatin accessibility QTL; CITE-seq, cellular indexing of transcriptomes and epitopes by sequencing; eQTL, expression quantitative trait loci; HELP, HpaII tiny fragment enrichment by ligation-mediated PCR; hQTL, histone QTL; MBD-ChIP, methylcytosine DNA-binding domain ChIP; mQTL, metabolic QTL; meQTL, methylation QTL; pQTL, protein QTL; REAP-seq, RNA expression and protein sequencing assay; scM&T-seq, single-cell methylome and transcriptome sequencing; scNMT-seq, single-cell nucleosome, methylome and transcriptome sequencing; scRNA-seq, single-cell RNA sequencing; sQTL, splicing QTL; STARR-seq, self-transcribing active regulatory region sequencing.

Table 2.

Techniques combinations used in human kidney studies to delineate GWAS findings

Techniques	Human Studies	Refs
Epigenome Analysis — DNA hypersensitivity analysis, RNA seq,Isoschizomer-based analysis (HELP), Illumina 450K,	Fetal Kidney Adult Kidney (14 healthy controls and 12 CKD patients) 165 Adult Blood Samples (181 Pima Indian subjects)166	ENCODE accession number GSM530655 165 166
DNase-seq, RNAse-seq, PolyA-seq, Chip-seq, RRBS, FAIRE-seq	68 human embryonic kidney samples (encodeproject.org)	167
GWAS, eQTL, scRNA-seq	Susztak lab, 151 human kidney samples microdissected into 121 tubules and 119 glomeruli GTEx database, 73 whole human kidney samples NEPH QTL, http://nephqtl.org 181 microdissected human kidney samples from patients with nephrotic syndrome () TRANSLATE study, 280 human kidney samples from unilateral non-invasive renal cancer.	22 122 121

ChIP-seq, chromatin immunoprecipitation followed by sequencing; eQTL, expression quantitative trait loci; FAIRE–seq, formaldehyde-assisted isolation of regulatory elements followed by high-throughput sequencing; GTEx, genotype–tissue expression; GWAS, genome-wide association studies; H3K4me1, monomethylated H3K4; H3K9ac, acetylated H3K9; H3K27me3, trimethylated histone 3 lysine 27; RRBS, reduced representation bisulfite sequencing; scRNA-seq, single-cell RNA sequencing; TRANSLATE, Transcriptome of renal human tissue.

Key Points.

• Genome-wide association studies (GWAS) have identified large number of nucleotide variations that show strong and reproducible association with kidney-related traits such as serum and urine metabolites, estimated glomerular filtration rate, albuminuria and hypertension.

• More than 95% of disease-associated GWAS signals are located in non-coding regions of the genome, which can be identified by analysing chromatin accessibility and conformation, DNA methylation and transcription factor binding.

• Disease-associated genetic variants seem to be located in cell-type specific gene regulatory regions, where they modulate disease risk by quantitatively altering gene transcript levels.

• The use of advanced computational approaches to combine datasets orthogonal to GWAS data, including molecular quantitative trait studies, single cell transcriptomics and epigenetic information, is necessary for the prioritization of GWAS variants.

Box 1. Role of rare variants in common kidney disease.

Over the last few years, a large collection of whole exome sequencing datasets has been generated, initially to be used in family studies of disease inheritance and later for large biobanks.¹³⁷ These large-scale whole exome sequencing studies have identified novel genes that after experimental validation, could yield potential therapeutic targets¹³⁸, for example in nephronopthisis, RBM48, FAM186B, PIAS1, INCENP and RCOR1 have been identified, which are novel cillopathy genes ¹³⁹

However, the low number of identified targets also indicates that the contribution of rare genetic variants to the development of common diseases such as diabetes mellitus¹⁴⁰ and hyperlipidemia¹⁴¹ is relatively small. Unbiased exome sequencing studies indicate that the contribution of coding variants to the development of chronic kidney disease (CKD) might be larger than that previously observed for other traits¹³⁴; however, these studies must be replicated in larger cohorts. ¹⁴²

Nonetheless, these studies have demonstrated that a clear convergence of genes, cell types and underlying mechanisms can often be observed for a specific disease condition. ²¹ For example, we observed that monogenic coding variants associated with nephrotic syndrome are enriched for podocyte-specific expression.²¹ Similarly, a GWAS performed for albuminuria in the context of diabetic kidney disease showed enrichment for podocyte-specific signals.¹⁴³ By contrast, coding mutations that are associated with either low of high blood pressure were enriched in genes expressed in the distal convoluted tubule expression.²¹ Of note, although genes identified in a blood pressure GWAS study were different from those implicated in monogenic blood pressure syndrome, they were also enriched for expression in the distal convoluted tubule.¹¹⁶ These studies are encouraging and potentially indicate that the genes and genetic signals involved in the pathogenesis of a disease will coalesce around a limited number of molecular and cellular processes.

Glossary terms

Secondary chromatin structure

The structure formed by the folding of primary chromatin

Imputation

The inclusion of genotypes that are not directly measured by estimating the missing data based on reference genome datasets and SNP linkage

Fine-mapping

The process by which a variant is assigned to a complex trait

Haplotype

A set of genetic polymorphisms that are inherited together

Deep phenotyping

Promoter

The DNA sequences on which transcription is initiated

Enhancer

A DNA sequence that is bound by transcription factors to increase the transcription of a gene

Insulator

A DNA sequence that limits chromatin activation

Transposable elements

Cis-regulatory elements

Regions of non-coding DNA that affect the expression of nearby genes

Massively parallel reporter assays

High-throughput experiments, in which the transcriptional activities of many regulatory sequences can be obtained

Quantitative trait loci - (QTL).

genetic loci that are associated with variation in a phenotype

Alternative splicing

Process by which multiple transcripts are produced from a single gene with different functions

Mendelian randomization

Orthogonal datasets

Independent and uncorrelated data sets

Posterior probability

The probability of an event occurring based on information from a prior event

Instrumental variable

Cellular encapsulation

A technique whereby cells are entrapped in a spherical semi-permeable polymeric membrane

Cellular deconvolution

The estimation of gene expression data for individual cell types within a bulk expression dataset