If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
A. I. Virtanen Institute for Molecular Sciences, University of Eastern Finland, Kuopio, 70211, FinlandDepartment of Pathology and Laboratory Medicine, University of California, UCLA, Los Angeles, USA
Genome-wide association studies (GWAS) have identified over 300 risk loci for coronary artery disease (CAD).
•
The majority of the variants are located within the non-coding regions of the genome with no known function.
•
The enrichment of GWAS variants within cis-regulatory elements of specific cell types suggests the variants exert their effects by regulating gene expression in defined tissue contexts.
•
Functional genomics approaches translate GWAS findings into mechanistic knowledge with clinical and therapeutic potential.
Abstract
Genome-wide Association Studies (GWAS) have identified more than 300 loci associated with coronary artery disease (CAD), defining the genetic risk map of the disease. However, the translation of the association signals into biological-pathophysiological mechanisms constitute a major challenge. Through a group of examples of studies focused on CAD, we discuss the rationale, basic principles and outcomes of the main methodologies implemented to prioritize and characterize causal variants and their target genes. Additionally, we highlight the strategies as well as the current methods that integrate association and functional genomics data to dissect the cellular specificity underlying the complexity of disease mechanisms. Despite the limitations of existing approaches, the increasing knowledge generated through functional studies helps interpret GWAS maps and opens novel avenues for the clinical usability of association data.
]. Over the past two decades, the field has moved from single candidate gene studies to large genome wide association studies (GWAS). This has been made possible via considerable increases in sample sizes, which in the latest studies surpass a million participants, and by the development of relatively cheap genome-wide genotyping methods. Today, over 300 CAD GWAS loci have been discovered, which explains ∼30–40% of the disease heritability [
Krishna G Aragam, Tao Jiang, Anuj Goel, Stavroula Kanoni, Brooke N Wolford, Deepak S Atri, Elle M Weeks, Minxian Wang, George Hindy, Wei Zhou, et al. Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants. Nat Genet . 52 (12) (2022 Dec) 1803-1815, doi: 10.1038/s41588-022-01233-6.
]. These associations could reveal the molecular mechanisms altered during the disease process and inform novel drug targets, biomarkers, or risk stratification. Despite the steady growth in the number of studies, the interpretation of CAD GWAS findings has remained challenging, as the majority of the risk variants are located within non-coding parts of the genome [
]. This has spurred a lot of research in the field of functional genomics aiming to establish the causality of GWAS variants and genes.
In this review, we summarize the progress in the field over the last decade with a particular focus on CAD GWAS loci. Firstly, we discuss examples by which the non-coding variants associated with CAD have been shown to exert their effect on nearby gene expression and cellular/organismal phenotype. Secondly, we highlight the potential of integrating GWAS results to functional genomics output to inform cell-type specific disease mechanisms. Finally, we discuss the methods that are used to explore the causality of the variants and identify their target genes that ultimately are expected to generate mechanistic knowledge about how disease susceptibility is encoded in the non-coding portion of the genome.
2. Insights and implications of GWASs of CAD
Despite significant advances in the catalogues linking genetic variants to CAD, we are limited in our ability to establish the functional relationship between the genotype and phenotype. Since most variants are located within the non-coding parts of the genome, there are several challenges that have limited our ability to interpret the GWAS findings to functional mechanisms. First, association signals at a given GWAS locus do not specify which variant(s) are causal, as most risk loci encompass tens to hundreds of variants in a high linkage disequilibrium (LD). Secondly, we face the limitation of conclusively linking a variant containing regulatory elements to their target genes. Therefore, a general strategy to overcome this challenge comprises the prioritization and validation of causal variants and target genes using a variety of computational and experimental methods, which then lead the way to pathway identification (Fig. 1). In the future, these methods are expected to continue shedding light into the mechanisms of action for each CAD locus.
Fig. 1General strategy to prioritize causal variants and targets genes from GWAS data.
(A) GWASs map trait-associated loci at a genome-wide scale. The Manhattan plot presented in the upper right panel was retrieved from the Common Diabetes Knowledge Portal by querying “Coronary Artery Disease” as phenotype (https://md.hugeamp.org/). (B and C) The most plausible causal variant is identified by integrating refined association data (through deep sequencing or whole genome sequencing) with functional genomics data (functional fine mapping defining chromatin regulatory features such as DNA accessibility, enhancer histone marks, TF binding and allele specific transcriptional activity). (D) For linking variants to target genes, cis-eQTL, splice-QTL, protein-QTLs and chromosomal architecture (chromatin looping and conformation) are employed.
Considerable efforts have been undertaken to provide systematic fine mapping and gene prioritization across CAD GWAS loci. In Fig. 2, we summarize the information from several state-of-the-art gene prioritization tools benchmarked using curated sets of gold-standard genes and expert curation [
]. These genes are particularly implicated in biological processes related to vascular development, cellular proliferation and migration, extracellular matrix organization and angiogenesis based on Enrichr analysis [
] (Fig. 2). In addition, another significant category is represented by genes associated with hepatic lipid metabolism including cholesterol, lipoprotein and triglyceride metabolism, in line with the findings that ∼30% of the CAD risk is mediated through alterations in lipid levels [
]. Regulation of transcription, inflammation and renal functions are also enriched, whereas a large group of genes have not been characterized or their functions are, in principle, not previously linked to CAD pathophysiology. Overall, most of GWAS loci are common single nucleotide polymorphisms (SNPs) with relatively small association effect size, suggesting the contribution of several interacting mechanisms involving various genes, pathways and cell types which together translate into a significant pathogenic effect.
Fig. 2Gene ontology analysis of CAD associated genes.
Krishna G Aragam, Tao Jiang, Anuj Goel, Stavroula Kanoni, Brooke N Wolford, Deepak S Atri, Elle M Weeks, Minxian Wang, George Hindy, Wei Zhou, et al. Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants. Nat Genet . 52 (12) (2022 Dec) 1803-1815, doi: 10.1038/s41588-022-01233-6.
]. The first 20 non-redundant categories were selected for the lollipop graph depicting gene count (size of the circle) and Log10 (p-adj) (colour scale) from EnrichR for each GO category. The first, most significant process was selected among the redundant categories (i.e. for several cholesterol related processes only the most significant one was used for the graph). It should be noted that the genes presented in the list (lower panel) are not necessarily validated for their causal role in CAD, but they represent candidate genes linked to CAD variants.
3. A decade of evidence of the role of cis-regulatory elements mediating the effects of risk variants
Understanding the biological mechanisms underlying the genetic risk of CAD is crucial to improve diagnostic and patient management and to develop more targeted therapies. However, the functional translation of association data has advanced at a significantly slower pace, limiting the clinical impact and practical usability of GWAS knowledge. In recent years, increasing efforts have been directed to overcome this asymmetry. Integration of GWAS with multidimensional genomic data generated by large consortiums such as ENCODE and Roadmap Epigenomic Project have revealed that a majority of lead association SNPs map to cell type specific cis-regulatory elements (CREs), e.g. enhancer and promoters, indicating that GWAS loci act largely through the regulation of transcription. In fact, CAD SNPs are highly enriched at cell specific enhancers pointing to enhancer-mediated regulation of transcription as a major mechanism for the genetic risk [
The molecular basis for any functionally meaningful associations is the genotype imbalance. In the specific case of SNPs within CREs, this means that a single-base mutation may disrupt transcription factor (TF) binding sites, chromatin structure and overall DNA accessibility to coregulators, thereby inducing genotype-specific bias in gene expression. One of the first post-GWAS CAD locus validation demonstrated that the SNP rs12740374 facilitates the binding of C/EBP to a regulatory site and alters the expression of nearby genes in hepatic cells. Interestingly, although the causal SNP appears to affect the expression of nearby genes, the authors of the study showed that SORT1, and not the closest genes CELSR2 and PSRC1, mediates the biological mechanisms that increase CAD risk [
PSRC1 may affect coronary artery disease risk by altering CELSR2, PSRC1, and SORT1 gene expression and circulating granulin and apolipoprotein B protein levels.
]. Similarly, noncoding variant rs9349379 of the PHACTR1 locus located within a vascular-specific enhancer was shown to regulate the expression of EDN1, a gene located 600 kb upstream [
]. Other targeted validation studies have comprehensively demonstrated enhancer-driven effect of CAD GWAS SNPs at 1p32.2, HDAC9, LMOD1 and SMAD3 loci [
A prototypical example of the challenges in decoding CAD association signals is the Chr9p21 locus, the most common and most strongly associated CAD locus. The genetic variants associated with CAD have been shown to be associated with allele specific methylation of the long noncoding RNA (lncRNA) antisense non-coding RNA in the INK4 locus (ANRIL) promoter in blood cells [
]. Notably, studies in CAD patients investigating the effect of Chr9p21.3 genotype in gene expression found that linear atherogenic RNA forms of ANRIL were upregulated in the carriers of the risk genotype, whereas the antiatherogenic circular forms were downregulated [reviewed in 32]. However, Chr9p21.3 genotype associates not only with ANRIL expression, but in specific cells or conditions, also with the CDKN2A/B tumor suppressors encoded in the locus, making it hard to disentangle the cis-regulatory mechanisms [
]. In line with this, ANRIL has been associated with dysfunction and proliferation of vascular endothelial cells and with metabolism of vascular smooth muscle cells (SMCs) and monocytes [
], suggesting that different mechanisms could act in different cell types. This example highlights the challenge of identifying which cell types are causal to the disease. On the other hand, disease-associated loci often contain multiple genes, making it challenging to distinguish those cells affected.
Altogether, these examples have defined specific functional variants acting as central mediators of the genetic risk of CAD by regulating the transcription of relevant genes. However, follow-up studies are necessary to interpret GWAS results and to infer the exact disease-causal variants, the genes they regulate, and the cell types in which they act. Moreover, a more complete understanding of CAD genetic risk demands the validation of additional hundreds of susceptibility loci often requiring high-throughput based methods. In the following section, we discuss some of the most commonly used approaches for defining and validating causal variants and target genes while presenting practical examples of studies investigating notable CAD GWAS loci.
4. Contribution of different tissues and cell types to the genetic risk of CAD
CREs, especially enhancers, are frequently cell type and state specific. Therefore, identification of CREs, operational cell types and tissues remains a challenge toward the functional understanding of risk loci (Fig. 3). Initial efforts to tackle this question have relied on integrating large-scale epigenomic datasets from Roadmap Epigenomics, ENCODE, and BLUEPRINT consortia [
] with disease-associated GWAS variants. Such studies have provided the first evidence that several cardiac traits associated variants are found enriched in heart tissue enhancers, including the PR heart repolarization interval, QRS duration, blood pressure and aortic root size [
]. As more data has been accumulated, the resolution of predictions has also increased, which is exemplified by the EpiMap, a compendium comprising 10,000 epigenomic maps across 800 samples where the epigenomic annotations are used to investigate functional and disease enrichments for enhancer-overlapping SNPs in each enriched tissue [
]. This study demonstrates that the CAD-associated SNPs are divided into heart enhancer SNPs enriched in vessel morphogenesis, endocrine enhancer SNPs in lipid homeostasis, liver enhancer SNPs in cholesterol and lipid metabolism and transport, adipose enhancer SNPs in axon guidance and focal adhesion, and embryonic stem cell-derived–muscle enhancer SNPs, enriched in aorta development. Interestingly, many of the tissue enhancers exhibited shared components, as exemplified by co-association of heart, muscle and endothelial enhancer CAD SNPs with high blood pressure and atrial fibrillation and enhancer overlaps with the top SNPs for LDLR, APOE and COL4A1 in multiple tissues, suggesting multiple mechanisms of action even at the single-locus level [
(A) Example of a GWAS locus illustrating the association between genetic variants and a trait. The variant above the dotted line represents genome-wide significant associations. (B) Overlaying SNP information with functional genomics data such as chromatin annotations can help to identify cell type through which variants act. (C) Overview of methods used to identify causal variants with a functional effect on transcription factor binding or epigenetic mark deposition (EMSA and ChIP), allele specific enhancer activity by MPRA or target gene expression by CRISPR. Each of the methods need specific input material to visualize the functional effect of the causal variants i.e. EMSA makes use of synthetic oligos, MPRA is based on episomal naked DNA plasmids or virally integrated vectors, ChIP-Seq uses fixed or native chromatin and CRISPR targets native chromatin of cells.
Despite the insight provided by the large epigenomic datasets, these catalogues often lack cell type resolution of regulatory sequences that would reflect the native tissue context. Recently, this has been addressed by the single cell omics technologies, which have enabled the profiling of the epigenome at single-cell resolution. To this end, we have recently profiled the chromatin accessibility of ∼7000 cells derived from human samples of endarterectomy using single cell (sc)ATAC-Seq (assay for transposase-accessible chromatin with high-throughput sequencing) [
]. Our results demonstrated that the open chromatin CREs of smooth muscle cells (SMCs) and endothelial cells (ECs) show the highest enrichment of GWAS SNPs for CAD [
]. These findings were largely corroborated in a study by Turner et al., where the authors profile over 28,000 nuclei and demonstrated that the majority of CAD GWAS SNPs resided within SMC CREs followed by macrophage, fibroblast and EC CREs [
]. These cell type associations have been also supported by the cell type specific expression of nearby candidate genes in recent scRNA-seq (single cell RNA sequencing) based studies [
Intersecting single-cell transcriptomics and genome-wide association studies identifies crucial cell populations and candidate genes for atherosclerosis.
]. Hocker et al. employed scATAC-Seq to profile the cell type specific landscape of CREs in the four chambers of the human heart demonstrating enrichment of CAD variants within SMC, fibroblast and EC CREs, enrichment of atrial fibrillation variants within cardiomyocyte CREs, and enrichment of varicose vein variants with EC CREs [
Similar approaches have now been extended across tissues to generate an atlas of scATAC-Seq in the tissue of 30 adult humans, which enabled systematic analysis of noncoding variants associated with complex human traits and diseases [
]. For example, hypertension is associated with similar cell types as CAD, whereas cholesterol traits associate with hepatocytes, type 2 diabetes with pancreatic cells, body mass index (BMI) with neurons, and waist-to-hip adjusted BMI with adipocytes and fibroblasts. Altogether, these recent resources have laid the foundation for the analysis of the genetic basis of CAD across tissues and cell types while prioritizing candidates for further laboratory experiments to determine their functions.
5. In search of causal variants through computational fine-mapping
The biological underpinning of GWAS variants is extremely challenging (Fig. 1). This is due not only because a given trait is often associated with several loci, but also because many variants within the same association locus are co-inherited, generating a phenomenon of non-random association between variants called linkage disequilibrium (LD) [
]. In the context of GWAS, an important practical implication of LD is that it makes it difficult or impossible to define causal variants based on association metrics. Therefore, in a first post-GWAS stage, the most plausible variants within the GWAS signals should be defined and prioritized (Fig. 1). Targeted deep sequencing or full genome sequencing complemented by genotype imputation allows for complete genotyping of target regions. Further conditional statistical analyses then use this fine-genotyping information in the refinement of the association signal within the GWAS locus (Fig. 1B). Such conditional analyses constitute iterative procedures that look for SNPs with the lowest p value of association until no additional SNP reaches the pre-assigned p value threshold [
]. Often, these approaches lead to the identification of alternative SNPs within LD that exhibit more significant association and larger effect sizes compared to the initial GWAS lead SNP, as exemplified by validation studies of several CAD loci (PHACTR1, [
]). However, while this is informative about the number of association signals within the GWAS locus, it fails to provide measures of causality for variants. To overcome this problem, many recent fine-mapping methods have adopted a Bayesian framework to assign posterior probabilities and credible sets of SNPs that refine the association signals of GWAS [
In a further prioritization step, refined loci are probed for regulatory potential through SNP enrichment analyses (Fig. 1C; Fig. 3A and B). In simple terms, this looks for the co-occurrence of association with chromatin regulatory features at a given locus. Transcriptional factor binding, DNaseI hypersensitivity regions [
] have been largely used as co-localization criteria, as they define chromatin regions harbouring cis-regulatory elements such as enhancers and promoters. In one of the pioneering studies in the field of CAD, Miller et al. overlapped CAD GWAS information with the epigenomic profiles in primary cultured human coronary artery smooth muscle cells (HCASMCs) [
]. Based on this intersection, the authors prioritized 64 candidate variants and performed allele-specific binding and expression analyses at seven top candidate loci: 9p21.3, SMAD3, PDGFD, IL6R, BMP1, CCDC97/TGFB1 and LMOD1 [
Incorporating functional annotations into computational fine-mapping approaches has been shown to improve identification of likely causal variants and several methods have emerged around this concept [
]. Pickrell et al. applied a functionally informed fine-mapping method (FGWAS) that uses the chromatin state enrichment information to reweight GWAS summary statistics and compute variant-specific association probability [
]. The authors demonstrated that 13 CAD risk loci could be fine-mapped to just a single variant, including the known missense variants in PCSK9, ANGPTL4 and APOE, PHACTR1/EDN1 and HDAC9/TWIST [
Krishna G Aragam, Tao Jiang, Anuj Goel, Stavroula Kanoni, Brooke N Wolford, Deepak S Atri, Elle M Weeks, Minxian Wang, George Hindy, Wei Zhou, et al. Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants. Nat Genet . 52 (12) (2022 Dec) 1803-1815, doi: 10.1038/s41588-022-01233-6.
Several integrative computational resources are available for analysing the effect of genetic variation on genomic functions, providing both analytical and predictive outcome for scientists working in the field and generating their own functional data. For example, BaalChIP [
] and RegulomeDB (https://regulomedb.org/) integrate multiple TF binding, chromatin accessibility, histone marks and other functional data to predict functional effects of genetic variation on epigenetic landscape (Fig. 3C). Open Target Genetics portal [
] are also excellent resources for variant prioritization that inform not only about epigenetic features, but also provide association data integrated from large number of subjects from various demographics. Even when the outcome of the methods and the information summarized by these portals may be predictive, the fact that they use experimental data helps researchers to expand the number of variants or loci that may be subject to targeted validation (Fig. 1).
6. Experimental strategies to identify causal variants
As valuable as the computational methods are in prioritizing and fine-mapping CAD relevant variants, experimental strategies are often encouraged to complement the predictions with functional validations (Table 1). Given their impact on transcriptional processes, a majority of CAD functional risk variants are thought to act by disrupting TF binding to the DNA (Fig. 3C). One of the oldest methods to measure allelic differences in TF binding to DNA is the electrophoresis mobility shift assay (EMSA) [
]. In this method, allele-specific synthetic DNA oligonucleotides are probed with TF or protein extracts from relevant cell types or tissue; if one of the alleles disrupts binding, this is detected by a shift in migration of the DNA-protein complex in an electrophoresis gel [
], the fact that it provides a fast and direct interaction readout has made it a popular method even today. For example, as part of our analysis of liver-specific CRE variants associated with CAD risk [
], we used EMSA to detect the effect of rs9591145 SNP predicted to affect CEBP binding, allowing for a direct measure of allelic effect on TF binding and robust evidence of allelic imbalance in the N4BP2L2 locus. Similarly, EMSA was recently used to demonstrate that one of two regulatory variants associated with allele specific expression of FURIN and FES in monocytes and vascular SMCs, rs17514846 and rs1894401, interacted with monocyte nuclear proteins in an allele specific manner [
]. The authors further performed a super-shift assay to demonstrate that PRDM1 TF preferentially interacted with the rs1894401 risk allele. Leveraging on the same principle of DNA-protein interaction, Prestel et al. used proteome wide analysis of SNPs to identify metabolically labelled nuclear factors interacting with synthetic oligonucleotides followed by mass spectrometry [
]. This allowed identification of six transcription factors of which E2F3 was demonstrated to result in allele-specific binding at HDAC9 locus harbouring rs2107595 regulatory variant.
Table 1Non-exhaustive list of the functional characterization studies involving experimental validation of CAD loci.
Locus
Target Gene(s)
Experimental Method
Summary Functional Mechanism
Ref.
1p13/rs12740374
SORT1 (Sortilin 1)
EMSA, Reporter assay, ChIP-qPCR
Variant creates of a C/EBP binding site and alters the hepatic expression of the SORT1 gene, which regulates plasma levels of LDL and VLDL.
PHACTR1 (phosphatase and actin regulator 1) /rs9349379
EDN1 (Endothelin 1)
Reporter assay, 4C-Seq, CRISPR deletion and variant editing
The SNP resides in an aorta specific enhancer. Enhancer deletion and variant editing leads to increased expression of EDN1 and Big ET-1 protein, through mechanisms that does not involve long-range chromatin loops.
Reporter assay, CRISPRi and deletion, ATAC-seq (caQTL), eQTL
The risk variant creates KLF2 binding site which increases enhancer activity under unidirectional flow to promote PLPP3 expression (mechanosensing role).
The risk allele disrupts the binding of Forkhead box O3 (FOXO3), leading to downregulation of LMOD1, which results in increased proliferation and migration and decreased cell contraction of HCASMCs.
CAD risk variants located in hepatocyte specific CREs located within 3D chromatin interaction hubs that regulate expression of many genes (only top one shown) in the liver.
Two CRE variants reduce FES expression in monocytes which promotes their migration. FES depletion results in larger plaques with more monocyte/macrophages and SMCs.
] have on the other hand used chromatin immunoprecipitation followed by qPCR in their respective studies of 9p21, SMAD3, LMOD1, PDGFD, IL6R and GUCY1A3 CAD loci to demonstrate allele specific TF activity. In this method, the binding of the TF can be detected in an allele-specific manner using TaqMan SNP genotyping assays (Fig. 3C). By combining allele-specific probes labelled with different fluorescent dyes and PCR primers for the desired region, these assays allow the multiplexed detection of two signals from the same qPCR reaction mix. This is particularly interesting, and useful, because the allelic differences are measured within the sample, somewhat minimizing experimental errors while increasing consistency and sensitivity [
Identification and characterization of a FOXA2-regulated transcriptional enhancer at a type 2 diabetes intronic locus that controls GCKR expression in liver cells.
]. However, the premise of “both alleles in the same reaction” conditions the assay to be used mainly in heterozygous cells and the probes used to detect the allelic imbalance may have inherent bias towards one allele and thus require calibration before conducting the experiment, as done previously [
]. Interestingly, these studies have further validated the TF effect on reporter and/or endogenous gene expression using TF knockdown as exemplified by FOXO3-mediated regulation of LMOD1 [
High-throughput methods to analyze allele specific CRE activity have also been applied to identify the causal variants, including genome-wide ChIP-Seq or ATAC-Seq analysis across many donor samples (Fig. 3C). Here, molecular quantitative trait loci (molQTL) analysis is used to identify statistically significant associations between the epigenetic phenotypes and the genotype. To this end, in a study by Stolze et al., human aortic endothelial (HAEC) donors were used to investigate the associations between genotype and the quantitative abundance of ATAC-Seq or H3K27ac ChIP-Seq signal as well as binding of ERG and NF-κB transcription factors [
]. Of important note, we also demonstrated that proinflammatory stimulus of HAECs resulted in a large set of novel molQTLs, in line with a context specific effect of functional variants. In addition, this study added to the growing evidence indicating that variants are more likely to influence the binding of cell type specific transcription factors, which then indirectly alter the binding of stimulus-specific transcription factors, such as NF-κB [
Single cell technologies are also highly useful for molQTL discovery. A pioneering study in the field by Turner et al. recently calculated chromatin accessibility (ca)QTLs across 41 patients profiled using scATAC-Seq [
]. The authors identified caQTLs in four major coronary cell types, including SMCs (1984 caQTLs), macrophages (1210), fibroblasts (510) and endothelial cells (49), of which over 90% were specific to one cell type. This highlights the importance of investigating the molQTLs in the disease-relevant cell types. Unfortunately, molQTL discovery from bulk or single cell data is currently restricted by the limited study sample sizes which demand a higher threshold of minor allele frequency and thereby causes loss of associations especially in rare cell types. In the future, larger sample sizes across a spectrum of disease stages are needed to increase the power of discovery and to identify context-specific regulatory mechanisms.
A complement to the discovery of the molecular functions of regulatory risk variants consists of exploring their effect on transcription. To this end, transcriptional reporter assay represents the most generalized technique, which has been employed in most CAD variant studies exemplified and discussed throughout the previous sections [
] (Fig. 1C). In its targeted version, allele-specific DNA fragments encompassing putative enhancer or other regulatory sequences are cloned into bacterial plasmid vectors expressing a reporter gene, usually firefly luciferase or green fluorescent protein. By transfecting these vectors in relevant cell types, researchers can answer i) whether the regulatory element harbouring the candidate variant is indeed functional and ii) whether alternative genotypes result in differential transcriptional activity. However, targeted strategies are not viable for working with a large number of loci, at least not in their classical setting. To this end, one approach to systematically identify causal variants in a high-throughput manner is to apply massive parallel reporter assays (MPRA) to measure the effects of synthetic DNA libraries representing both alleles on the expression of the reporter gene (Fig. 3C). We have recently used a particular type of MPRA, called self-transcribing active regulatory region sequencing (STARR-seq), to analyze the allele specific activity of thousands of CAD variants representing over 200 CAD loci in hepatocytes, endothelial cells and smooth muscle cells [
], allowing the prioritization of causal variants for over 50 CAD loci. By providing integration of STARR-Seq results with molQTL analysis for transcription factor binding, chromatin accessibility, and H3K27 acetylation in HAECs we also provide evidence that functional variants are more likely to be located within accessible/active chromatin, exhibit an allelic mutation in a TF binding motif for a factor with an important role in the cell type and/or cell state of interest and associate to a molQTL [
]. Notably, a large majority of allele specific variant effects were cell or context specific. However, these also include several cases where different variants within the same GWAS locus exhibit allele specific enhancer activity in different cell types or where multiple, tightly linked causal variants mediate a similar effect, challenging the view that a single causal variant would underlie the association [
] have been situated at the forefront of experimental tools for the deep characterization of DNA regulatory elements (Fig. 3C). The possibilities for applications are multiple. We recently used a CRISPR derived epigenetic activation system (CRISPR-dCas9-VPR [
]) to investigate the endogenous transcriptional function of 6 enhancer hubs harbouring CAD SNPs which were prioritized by allele specific TF binding evidence and STARR-seq allele specific activity in liver cells [
]. This approach enabled us to unambiguously identify APOB, LIPA, FAS, CLTCL1, UBE2Z, and SNF8 as target genes of these CAD enhancer hubs. In the same study, we further used gRNA CRISPR-Cas9 system to delete two enhancers defined by rs17680741 and rs2297787 CAD SNPs, identifying MAT1A, TSPAN14, FAM213A, AS3MT, NT5C2, SFXN2, ARL3, CNNM2 and CYP17A1 as target genes. These results have two remarkable implications, both at the biological and at the methodological levels. Firstly, although some lipid metabolism targets such as APOB, LIPA or FAS represent plausible CAD candidate genes, our analysis also identified other genes that could play a role in mediating variants effect through alternative pathways. Secondly, the fact that one variant may be functionally linked to several genes demonstrates that the regulation of gene expression by transcriptional enhancers is redundant and, in turn, single association signals operate through a network of genes [
]. In an iteration of these tools, we recently used a gRNA CRISPR-Cas9 system to delete an enhancer at rs17293632 locus, which resulted in reduced expression of SMAD3 in teloHAEC cells [
As with other classical methodologies, such as overexpression and RNA silencing, CRISPR derived tools can be coupled with functional assays to investigate pathway and integrative effects beyond gene expression. However, CRISPR may be considered an experimental “hinge” that has the added value of linking integrative phenotypes with molecular processes being executed at the transcriptional level and affected by human genetic variation. Two elegant studies exemplified this from-genotype-to-function usage of CRISPR tools. In the first of these, Krause et al. demonstrated that the deletion of rs17114036 SNP promotes endothelial quiescence under unidirectional shear stress by regulating PLPP3 gene in endothelial cells [
]. In the second study, Gacita et al. identified enhancer modifying variants for MYH6 cardiomyopathy gene that upon CRISPR-Cas9 mediated deletion resulted in a dose-dependent increase in MYH6 and faster contractile rate in engineered heart tissues [
]. Altogether, although the studies discussed above are examples of targeted use, we expect that CRISPR tools will continue to be employed at high-throughput resolution, allowing the functional assessment of hundreds of genomic loci in CAD relevant models and systems [
Genes, and ultimately proteins, are the major executors of cellular programs in health and diseases. Identification of causal genes is therefore crucial to understand the biological processes underlying genetic associations. The proximity between a GWAS signal and a gene has been used as the first criteria for causal gene prediction. However, while this approach may yield relatively high prediction quality for variants mapping to protein coding regions, experimental characterization has proven that gene prediction based on the “closest gene” approach fails for many non-coding variants. Nonetheless, the structure of the genome and the physical organization of the chromatin means that gene prediction is more feasible within certain proximity. The association of QTLs with proximal molecular features is commonly and increasingly used to identify target-causal genes (Fig. 1D). Cis-expression QTL (cis-eQTL) is the most commonly employed method that measures the association between genetic variants and gene expression. Even when proximity is still a constraint of the method, the fact that it uses expression levels enables one to discriminate target genes even in heavily gene dense regions. The causality is further strengthened if there is a significant colocalization between GWAS and eQTL signals in a locus (Fig. 4A). Today, eQTL has become a routine criterion in gene linking due to the public availability of large datasets representing a wide array of tissue types. In this context, GTEx consortium enables researchers to access gene expression and eQTL data from 49 tissues representing healthy and diseased individuals [
] (https://www.gtexportal.org/). Another valuable resource, specifically for CAD studies, is the “Stockholm-Tartu Atherosclerosis Reverse Network Engineering Task study” (STARNET) [
], as it provides RNA-seq data from patients with cardiovascular disease across disease relevant tissues, including liver, adipose tissue and aorta. Accordingly, STARNET arterial samples have been shown to contribute most to the identification of trait-associated genes for CAD [
]. In this particular study, the authors further generated directional coexpression networks anchored in cis-eQTL associations to model the integrative effects of cardiometabolic on gene expression levels. Their results are now available through the STARNET portal, allowing for eQTL mining as well as gene-gene network visualization (http://starnet.mssm.edu/).
Fig. 4Overview of methods used to identify non-coding variant target genes.
Functionally annotated SNPs are mapped to target genes based on (A) eQTL associations that are often supported by colocalization analysis which evaluates if both eQTL/molQTL and GWAS signals are driven by the same causal variant(s); (B) 3D chromatin interactions based in Hi-C methods; (C) Correlation in accessibility between enhancer variant and promoter across many single cells or experiments, or (D) Correlation between peak accessibility and gene expression.
As exemplified by several CAD loci studies, cis-eQTL are highly cell type specific, which may limit their detection for CAD variants in heterogeneous tissues such as the vascular wall [
]. To address this limitation, several cell type specific cohorts have been established in CAD relevant cell types. Among the first, a cohort of ∼150 HAECs was established by Romanoski et al. where over 10,000 cis-eQTLs were discovered in basal and oxPAPC stimulated conditions [
]. Importantly, half of the endothelial eQTLs were not detected in GTEX and thus would be lost when studied in the tissue context. The same resource has also been used by Ref. [
] to demonstrate that rs17114036 CAD-associated SNP was a cis-eQTL for PPAP2B in HAECs from human donors. Further, in this example, the authors did not find genotype-expression association in other tissues in publicly available datasets, indicating that eQTLs act through endothelial specific regulatory elements.
Similar analysis has been conducted in three cohorts of SMCs. First, Miller et al. identified cis-eQTL SNPs for CDKN2B, PDGFD, SMAD3 and LMOD1 using qPCR based expression analysis from 64 independent HCASMC donors [
]. This study was later extended to a genome-wide transcriptome analysis of 52 donor samples that led to the identification of 1220 cis-eQTLs, among which FES, SMAD3, TCF21, PDGFRA, and SIPA1 also demonstrated significant colocalization with CAD GWAS variants [
]. Recently, Aherrahrou et al. used a cohort of human aortic SMCs from ∼150 donors and discovered ∼4900 eQTLs, ∼4400 splice QTLs (sQTL) and 96 circular RNA QTL (circQTL), among which ∼3700 cis-eQTLs were not previously observed in the GTEX dataset [
R. Aherrahrou, D. Lue, M. Civelek, Genetic regulation of circular RNA expression in human aortic smooth muscle cells and vascular traits, HGG Adv. 4 (1) (2022 Nov 30) 100164. doi: 10.1016/j.xhgg.2022.100164.
]. Of these, 84 eQTL, 164 sQTLs and 3 circQTLs colocalized with CAD loci, highlighting especially the importance of genetic regulation of mRNA splicing as a molecular mechanism for CAD genetic risk. Interestingly, the most significantly associated CAD locus, 9p21, was now additionally demonstrated to represent a sQTL for ANRIL in proliferative SMCs. In another recent study, a large biobank of SMCs isolated from umbilical cords of 1499 donors extended the discovery of cis-eQTLs to 42,000 eQTLs and 12,000 sQTLs, of which 84 demonstrated significant colocalization with CAD GWAS signals, including SEMA5A, NBEAL1, CARF, CARM1 and SNF8 [
]. Altogether, this evidence has contributed immensely to the characterization of cell type specific loci by refining the link between CAD GWAS variants and target genes while providing excellent resources allowing dive into the biological basis of CAD.
Another functional milieu affected by disease-associated variants is the looping of chromatin, which facilitates the physical contact between the promoter of the gene and distal enhancers (Figs. 1 and 4B). Several conformation capture methods have been developed to assess the mechanisms that regulate these three-dimensional regulatory interactions. Prestel et al. used circular chromosome conformation capture to assess the effect of alternative alleles on enhancer-promoter binding at HDAC9 locus [
]. As for ChIP approaches, combining capture methods with sequencing (Hi-C) have enabled determination of chromosomal architecture at genome-wide scale. Mumbach et al. were among the first to study chromatin interactions between CAD GWAS variants and target gene promoters using H3K27 centric chromosome conformation capture, HiChIP, in HCASMCs [
]. They identified 1062 gene targets and showed that in 90% of the cases, the nearest gene was not the interacting target genes. Additionally, the authors also demonstrated that HCASMC chromatin interactions were enriched in CAD-associated SNPs compared to T-cells. Similarly, Åkerborg et al. generated a catalogue of 423 CAD causal genes based on the intersection of Hi-C maps, GWAS signals and enhancer histone marks in aortic endothelial, smooth muscle cells and macrophages [
]. Finally, Lalonde et al., performed Hi-C assay in teloHAECs and identified 991 combinations of open chromatin regions and gene promoters that mapped to 38 CAD and 92 blood pressure GWAS loci [
]. Together, these three studies experimentally demonstrated that GWAS variants often interact with the promoters of several genes, providing molecular basis for the multiple-gene, redundant function of enhancers.
However, instead of solely providing a measure of a direct functional interaction, Hi-C might also detect random interactions, bystander interactions and interactions due to sharing of the same nuclear structure [
]. To improve the prediction accuracy, new statistical models that consider the CRE activity have been developed. For example, the activity-by-contact (ABC) model computes a score for each CRE-gene pairs as the product of enhancer activity (epigenetic marks) and contact (Hi-C) [
]. Importantly, this was shown to significantly improve the precision of GWAS target gene prediction compared to Hi-C alone. Open target genetics platform has created a ‘locus to gene’ L2G model that further combines in silico pathogenicity prediction from VEP and PolyPhen, colocalization of molQTLs, gene distance to credible set variants weighted by their fine-mapping probabilities and chromatin interaction [
]. Here, chromatin interaction data was complemented with enhancer-promoter activity correlation based on TSS expression or DNase hypersensitivity based on the premise that physically interacting CREs are likely to correlate in activity [
]. Importantly, a similar principle can now be applied to scRNA- and scATAC-Seq data to identify target genes based on enhancer-to-promoter co-accessibility (scATAC data alone) or enhancer accessibility to mRNA expression correlation (scATAC and scRNA-Seq data) for which several computational tools have been recently developed (Fig. 4C–D) [
]. Application of such tools in recent studies profiling human atherosclerotic lesions using scATAC-Seq has allowed prioritization of hundreds of cell type specific target genes [
]. Further statistical and/or functional fine-mapping of selected to CAD loci have revealed a growing list of genes potentially acting through SMCs (e.g. MFGE8, TBX2, PRDM16, COL4A1, SEMA5A, SNHG18 and BMP1), ECs (PECAM1, CLCN6, PLPP3, BCAR1, LDAH, POU4F1, ETV1 and NOS3) and macrophages (GGCX, FCHO1, SREBF1, DAGLB and IL6R) [
]. We anticipate these and future larger datasets will provide a valuable resource to interrogate causal disease processes.
8. Conclusions and perspectives
The integration of functional genomics approaches, such as those exemplified and discussed in this review, are beginning to define the cell types and genes operating in CAD pathophysiology. Despite the limitations of existing approaches, integrating an increasing amount of functional knowledge helps define the mechanisms that link genetic variation to human biology, funding an ever-growing base for the translation of GWAS findings into clinically actionable gene sets. In this context, incorporating functional priors emerging from experimental approaches can be used to improve the polygenic risk prediction and population transferability of the polygenic risk score, strengthening its future clinical utility [
]. In parallel, we anticipate that the integration of GWAS with large single-cell datasets will improve our understanding of cell-type and context specific effects of genetic variation [
Intersecting single-cell transcriptomics and genome-wide association studies identifies crucial cell populations and candidate genes for atherosclerosis.
]. Ultimately, the functional dissection of genetic risk opens the door to strategies for the pharmacological manipulation of molecular and cellular processes implicated in human diseases, including CAD.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We thank Dr. Tiit Örd for his support in generating the list of ∼1500 candidate genes represented in Fig. 2 [
]. Fig. 3, Fig. 4 were created with BioRender.com. This study was funded by the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (Grant No. 802825 to M.U.K), Academy of Finland, the Sigrid Juselius Foundation, the Finnish Foundation for Cardiovascular Research, and the Aarne Koskelo Foundation. Personal grants to UTA and MLR were received from the Matti and Vappu Maukonen foundation (UTA), Ella och Georg Ehrnrooths Stiftelse (UTA), the Finnish Foundation for cardiovascular Research (UTA and MLR), Instrumentrium Foundation (MLR) and Sigird Jusélius Foundation (MLR).
References
Roth G.A.
Mensah G.A.
Johnson C.O.
Addolorato G.
Ammirati E.
Baddour L.M.
Barengo N.C.
Beaton A.Z.
Benjamin E.J.
Benziger C.P.
et al.
Global burden of cardiovascular diseases and risk factors, 1990-2019: update from the GBD 2019 study.
Krishna G Aragam, Tao Jiang, Anuj Goel, Stavroula Kanoni, Brooke N Wolford, Deepak S Atri, Elle M Weeks, Minxian Wang, George Hindy, Wei Zhou, et al. Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants. Nat Genet . 52 (12) (2022 Dec) 1803-1815, doi: 10.1038/s41588-022-01233-6.
PSRC1 may affect coronary artery disease risk by altering CELSR2, PSRC1, and SORT1 gene expression and circulating granulin and apolipoprotein B protein levels.
Intersecting single-cell transcriptomics and genome-wide association studies identifies crucial cell populations and candidate genes for atherosclerosis.
Identification and characterization of a FOXA2-regulated transcriptional enhancer at a type 2 diabetes intronic locus that controls GCKR expression in liver cells.
R. Aherrahrou, D. Lue, M. Civelek, Genetic regulation of circular RNA expression in human aortic smooth muscle cells and vascular traits, HGG Adv. 4 (1) (2022 Nov 30) 100164. doi: 10.1016/j.xhgg.2022.100164.
00046-1&id=ga1.jpg){kind=link}
00046-1&id=gr1.jpg){kind=link}
00046-1&id=gr2.jpg){kind=link}
00046-1&id=gr3.jpg){kind=link}
00046-1&id=gr4.jpg){kind=link}