Epigenetic and Transcriptional Networks Underlying Atrial Fibrillation

Abstract

Genome-wide association studies (GWAS) have uncovered over a hundred genetic loci associated with atrial fibrillation (AF), the most common arrhythmia. Many of the top AF associated loci harbor key cardiac transcription factors, including PITX2, TBX5, PRRX1 and ZFHX3. Moreover, the vast majority of the AF-associated variants lie within non-coding regions of the genome where causal variants affect gene expression by altering the activity of transcription factors and the epigenetic state of chromatin. In this review, we discuss a transcriptional regulatory network model for AF defined by effector genes in GWAS loci. We describe the current state of the field regarding the identification and function of AF-relevant gene regulatory networks, including variant regulatory elements, dose-sensitive transcription factor functionality, target genes and epigenetic states. We illustrate how altered transcriptional networks may impact cardiomyocyte function and ionic currents that impact AF risk. Lastly, we identify the need for improved tools to identify and functionally test transcriptional components in order to define the links between genetic variation, epigenetic gene regulation and atrial function.

Introduction

While historically considered an acquired disease, atrial fibrillation (AF) risk is now understood to include a major heritable component, with evidence including rare mutations in families and association with common variants in the population at large.^1–3 Recent work by large consortia have identified over 100 genetic loci associated with AF by genome-wide association studies (GWAS).^4,5 Together with epidemiological studies, the current model suggests that AF is a “complex trait”, influenced by a combination of multiple genetic and environmental risk factors that contribute cumulatively to disease predisposition.^6,7,8. Genome-wide polygenic risk scores of common variants have recently been developed that identify individuals at greater than threefold increased risk of AF in the general population.⁹ In the near future, use of such approaches may identify high risk patients with a combination of inherited susceptibilities and potentially provide preventive treatment or lifestyle adjustments to decrease AF risk. However, the polygenic risk score currently does not help to determine a targeted treatment, as it does not give any specific information on the affected molecular pathways, such as ion flux deficiencies affecting conduction velocity or myocardial automaticity, or ancillary pathologies such as fibrosis.

Most AF-associated GWAS variants reside in the non-coding genome. The majority of disease-associated variants, as well as variants in high linkage disequilibrium, are enriched in predicted transcriptional regulatory elements, implying that the identified GWAS variants alter transcriptional regulation.^10–12

A number of loci harboring TFs, including PITX2, ZFHX3, PRRX1, TBX5, NKX2–5 and HAND2, have been linked to AF susceptibility (Table 1). While information regarding the actual target genes of the noncoding variation (i.e. variant regulatory elements) in most of these loci is still lacking, experimental evidence from animal and cell models with modified transcription factor genes has implicated these transcription factors in normal atrial rhythm control and the pathology of AF (Table 1).^13–25 These transcription factors target thousands of regulatory elements for genes involved in atrial function, including ion channels, gap junctions, structural/cytoskeletal components, and others.^26–30 Thus, growing evidence supports a transcriptional model for the mechanisms underlying GWAS in AF: noncoding variation in regulatory sequences affects expression of genes encoding transcriptional regulators, and/or function of regulatory elements of given target genes, altering gene expression and conferring disease susceptibility (Fig. 1). Therefore, resolution of the individual components of the GWAS-associated transcriptional networks influenced by genetic variation is a fundamental goal of current research.

Table 1.

Overview of TFs near AF associated loci and their phenotypes

Locus	(Cardiac) phenotype	eQTL (direction)	Refs.
PITX2	At least 6 independent common genetic variants upstream of the PITX2 gene are strongly associated with AF in populations. PITX2 regulates left-right differentiation (including sinoatrial node restriction to the right atrium), and deficiency leads to structural and electrical remodeling of the heart, which may lead to AF through different mechanisms.	-	4,45,59–62
ZFHX3	Gene involved in tumorigenesis. Heterozygous germline deletion results in preweaning mortality and increased cell proliferation. Also involved in circadian rhythm.	-	63–66
TBX5	Heterozygous mutations cause limb and cardiac defects (Holt-Oram syndrome). TBX5 is essential for septal and atrioventricular conduction system development, and atrial ventricular bundle maintenance. TBX5 point mutations are known to cause AF.	TBX5 * (risk***)	52,53,67
PRRX1	Involved in skeletogenesis, limb-, elastic artery-, ductus arteriosus and pulmonary vascular development.	PRRX1 * (protective***)	68–71
NKX2–5	Essential for cardiogenesis and for the development of the conduction system. Mutations in this gene have been found to lead to congenital heart disease, conduction defects as well as atrial fibrillation.	-	72,73–75
BMP2	Involved atrioventricular canal and in valve formation, cardiomyocyte differentiation, cushion formation and septation.	-	76–80
CREB5	Together with ATF-2 responsible for adipocyte differentiation, and involved in the regulation of colorectal cancer metastasis.	-	81,82
HAND2	KO of HAND2 leads to embryonic lethality between E9.5-E10.5, due to anterior heart tube hypoplasia and vascular development defects. OFT morphology is dependent on Hand2, and cardiac morphological development is sensitivity to Hand2 dosage.	HAND2-AS1** (risk***)	83–86
TBX3	Heterozygous mutations cause defects of limbs, mammary glands, teeth, and additional tissues (ulnar-mammary syndrome). In heart required for specification of the sinoatrial node and pacemaker function, atrioventricular conduction system, and outflow tract development.	-	87–89

^*MetaXcan heart,

^**eQTL GTEx heart.

^***Risk: higher expression associated with increased risk for AF. Protective: higher expression associated with decreased risk for AF

Figure 1.

Simplified scheme depicting the relation between noncoding variants clustered in variant regions, variant regulatory elements, transcription factor (TF) dose and target genes. A) An AF-associated region is depicted with neighboring genes. The major allele of an AF-associated variant lies in a regulatory element, which interacts with transcription factors and cofactors leading to physiological levels of transcription of (a) target gene(s) through physical interaction. On the right, two situations are depicted that could result from the presence of a minor (risk) allele in the regulatory element. The risk allele could interfere with binding of the correct TF, leading to diminished expression of target genes. The risk allele could cause altered binding affinity favoring another TF, causing (tissue-specific) gain/loss of expression of target gene(s), that results in atrial fibrillation predisposition. B) Variation in a regulatory element (RE) can lead to the altered expression level of a gene for a TF (yellow) or target effector gene. The changed dose of the TF influences the expression of many effector genes, possibly including its own gene. (Illustration Credit: Ben Smith).

In this review, we summarize the progress of the field in linking transcription factor networks to AF, and discuss currently employed approaches, progress and difficulties in identifying the variants and target genes contributing to AF development. We describe the potential role and current knowledge of epigenetics (i.e. gene regulation by transcription factors, chromatin states, and conformation) in modulating the effect of variant regulatory elements. Moreover, we discuss the combinatorial effects of multiple common variants that could increase predisposition to AF. Finally, we discuss AF variants in the context of development and aging.

Transcription factors, atrial rhythm and AF risk

The identification of non-coding regulatory elements harboring AF risk implies the importance of transcriptional regulators in AF, borne out by the identification of AF risk signals at many transcription factor loci. There are hundreds of validated and suspected transcription factors, some of which have tissue- or condition- specific expression to regulate cell differentiation and developmental programs (reviewed in³¹). The binding of accessible regulatory elements is dependent on both availability and activity (including binding affinity) of transcription factors. Futhermore, transcription factor dose is crucial in the regulation of transcription. For example, mutations or haploinsufficiency of developmentally expressed transcription factors such as TBX5, NKX2–5, PRRX1, HAND2, and PITX2 result in genetic syndromes that have many symptoms including congenital defects of the heart and other tissues, and conduction abnormalities (Table 1).^32–34 Many common variants associated with AF and cardiac conduction are near these and other transcription factor-encoding genes, suggesting that variation in their regulation may impact AF predisposition.

An example of the importance of transcription factor dosage in AF is found in the PITX2 locus. PITX2 expresses three isoforms, PITX2A, PITX2B, and PITX2C, the latter being expressed from an independent promoter and representing the major cardiac isoform.^35–38 Mouse studies revealed that Pitx2c is required for asymmetric heart development, imposing left-sided identity on the left atrium and sinus venosus (including suppression of sinoatrial node development at the left side), and for proper development of the pulmonary vein myocardium.^39–43 In addition, cardiac expression of PITX2 is almost exclusively restricted to the left atrium in the normal adult heart,³⁶ and its expression increases after myocardial injury in the nenonatal left ventricle.⁴⁴ Both the left atrium and the sleeves of the pulmonary vein are potential substrates for development of AF, and the pulmonary vein is often the origin of ectopic foci in AF patients.⁴⁵ While homozygous Pitx2c mutant mice display embryonic lethality, reduced expression of Pitx2c during development results in AF inducibility upon stimulation, which implicates a developmental role of this gene in AF susceptibility.^35–37 One of the key findings of the gene expression studies is that PITX2 is one of the most differentially expressed left atrial genes in healthy adult heart. In addition, genes associated with a right atrial phenotype such as BMP10 (bone morphogenic protein 10) appear to be repressed by PITX2, thus linking PITX2 to left-sided identity of the left atrium.^17,36,44 Various possible mechanistic explanations of PITX2 deficiency resulting in AF have been proposed. It has been observed that reduction in Pitx2 resulted in potassium and calcium channel gene dysregulation, ultimately leading to a shortening of the atrial action potential, a depolarized resting membrane potential, and predisposition to AF (Fig. 2).^21,45 Pitx2 may also regulate miR-17–92 and miR-106b-25; when deleted from the mouse genome, AF susceptibility occurred.⁴⁶ Analysis of gene expression in an atrial-specific knockout of Pitx2 demonstrated similar transcriptional alterations in calcium homeostasis and AF-related microRNAs (Fig. 2).⁴⁷ In addition, through the use of zebrafish and murine models of Pitx2 deficiency, changes in sarcomere length and metabolic dysfunction in cardiomyocytes were linked to AF.^17,20,48,49 These findings suggest that, although Pitx2-deficient mice with a predisposition to AF apparently have morphologically normal left atria,^36,44,45 there are critical Pitx2 targets beyond ion channels that have important implications for electrophysiology. The postnatal function of Pitx2 was investigated using conditional deletion techniques, resulting in altered expression of genes encoding ion channels, cell junction proteins, calcium handling genes and transcription factors, several of which were previously implicated by GWAS.^17,21,36,44 Importantly, these mice show ECG abnormalities indicative of sinus node dysfunction, but surprisingly are not inducible for AF. These conditional deletion studies in mice appear to indicate that Pitx2 also has an important role in the postnatal left atria necessary for maintaining normal sinus rhythm, in addition to its more established developmental functions. Continued studies of the diverse roles of Pitx2 in both the developing and postnatal heart will deepen our understanding of this critical cardiac transcription factor. The multiple roles of Pitx2 are depicted in Fig. 2.

Figure 2.

Schematic of PITX2 functions within a left atrial cardiomyocyte. It has been established that TBX5 is a transcriptional activator of PITX2. PITX2 is then capable of regulating gene expression throughout the genome, having both activating and repressing capabilities. Direct PITX2 targets, validated largely by luciferase assays and ChIP-seq, are listed by cellular function. While direct ion channel and gap junction genes were some of the first PITX2 targets discovered, there is now evidence in multiple model organisms that PITX2 is also capable of regulating cardiomyocyte metabolism and the antioxidant response to stress. Furthermore, PITX2 has been shown to directly alter microRNA transcription, which in turn can globally regulate gene expression in the cardiomyocyte. It is important to note that there are additional putative PITX2 targets not listed here that have been discovered through RNA sequencing experiments of PITX2 deficient murine models; however, these transcriptional alterations need further validation to distinguish direct targets of PITX2 from secondary effects. Overall, these transcriptional targets of PITX2 are able to determine critical physiological outputs of the cardiomyocyte, such as regulating cardiac rhythm and sarcomere structure. (Illustration Credit: Ben Smith).

TBX5 is another well-studied example of a dose dependent transcription factor, which is at the center of the cardiac transcription factor network. TBX5 mutations resulting in haploinsufficiency cause Holt-Oram syndrome of congenital upper limb (hand) and heart defects.^50–52 In contrast, a “gain-of-function” mutation in a large Dutch family causes atypical Holt-Oram syndrome, without obvious cardiac structural defects, but with over 65% of affected family members displaying early onset AF.⁵³ Regions near TBX5 have been associated with prolongation of the PR-interval in addition to AF risk.^4,5,54 Murine models of spontaneous AF are rare, however, conditional deletion of Tbx5 in the adult heart resulted in a model of primary, spontaneous AF.^14,55 This model identified Tbx5 as direct activator of Pitx2, with Tbx5 and Pitx2 antagonistically regulating the expression of key ion channel protein-encoding genes such as Scn5a, Gja1, Ryr2 and Atp2a2 (Fig. 2, 3). AF in Tbx5 deficient mice was mechanistically connected to calcium transport mechanisms and expression of Atp2a2, which encodes the main cardiac sarco/endoplasmic reticulum Ca²⁺-ATPase (SERCA2).^16,18 Genetic deletion of the SERCA2 inhibitor phospholamban (Pln) normalized SERCA function and rescued Tbx5 deficiency-associated AF. Interestingly, variants in the vicinity of PLN have also been associated with AF.⁴ Moreover, conditional haploinsufficiency of Tbx5 in adult mice caused reduced expression of predicted target genes and a predisposition to AF; additional Pitx2 haploinsufficiency rescued these phenotypes, supporting the notion of antagonistic gene regulation in vivo. These data resulted in a model describing an incoherent feed forward loop driven by TBX5 and modulated by PITX2 (Fig. 2). This Tbx5 driven regulatory network was further interrogated using the sequencing of non-coding RNAs (ncRNAs) to identify regulatory elements crucial for cardiac rhythm homeostasis. This differential deep sequencing approach identified lncRNA-associated regulatory elements which controlled the expression of calcium handling genes, including Ryr2 and Atp2a2.⁵⁶ These data elucidated important aspects of the regulatory network driving calcium handling physiology, and also identified a lncRNA required for Ryr2 expression.⁵⁶ This lncRNA was also tightly associated with chromatin and required for stabilization of RNA PolII at the Ryr2 promoter, all of which suggested it had a functional role in gene expression.⁵⁶ While non-coding RNA transcripts in AF will not be generally discussed in this review, the investigation of functional lncRNAs in future efforts could provide instrumental insight into the maintenance of gene expression for cardiac rhythm.

Figure 3.

Transcription factor networks control the expression of effector genes defining properties relevant for atrial structure, function and rhythm. Tbx5 and Pitx2 interact antagonistically as transcriptional activator and repressor, respectively, to co-regulate a gene regulatory network that governs effector genes involved in calcium cycling, sodium currents, potassium currents, and cell-cell conduction. Other transcription factors such as Gata4 and Nkx2–5 are also known to interact with Tbx5 and Pitx2 to co-regulate genes involved in atrial structural development and electrophysiological properties. The interactions of other transcription factors (such as Prrx1) in AF physiology and their role in atrial rhythm remain to be determined.

The cooperative interactions between the transcription factors TBX5, GATA4 and NKX2–5 is key for proper binding at their respective binding sites, and cooperative regulation of the expression of the downstream target genes and of cardiac development.^57,58 In contrast, Tbx5 and Gata4 were demonstrated to interact antagonistically for regulation of atrial rhythm control (Fig. 3).¹⁸ The adult-specific Tbx5 haploinsufficiency phenotype of arrhythmia susceptibility and prolonged action potential duration¹⁴ was rescued by Gata4 haploinsufficiency, but not by Nkx2–5 haploinsufficiency.¹⁸ Because Gata4 haploinsufficiency normalized the reduced expression of calcium channel genes Ryr2 and Atp2a2 in the Tbx5 haploinsufficient mice, it was hypothesized that the beneficial effect was mediated by rescue of calcium homeostasis (Fig. 3). In the healthy cardiomyocyte, the inhibiting effect of Gata4, coupled with the stimulating effect of Tbx5 results in balanced Atp2a2 expression. However, in Tbx5 haploinsufficient mice, Gata4 overtakes the balance causing over-repression of Atp2a2 expression. Decreased Atp2a2 expression causes Ca²⁺ cycling dysfunction, resulting in reduced Ca²⁺-influx into the SR. Consequently, there is excess cytosolic Ca²⁺, leading to prolonged action potential duration and arrhythmia susceptibility. Gata4 haploinsufficiency in Tbx5 haploinsufficient mice restored the transcription factor dosage balance and sinus rhythm. Pln haploinsufficiency also had a similar rescuing effect on Tbx5 haploinsufficiency, validating the involvement of SERCA. This study illustrates the potential collaboration of transcription factors in a network, contributing to AF (Fig. 3).

For a number of AF-associated genes, including TBX5, eQTL analysis predicts increased, not decreased expression will result in increased AF susceptibility.⁴ In line with this, the familial “gain-of-function” mutation in TBX5 caused early onset AF development in humans with atypical Holt-Oram syndrome.⁵³ Consistently, when the mouse orthologs of PR interval or AF associated regions were removed from the Tbx3 and Tbx5 loci, respectively, moderate (<2 fold) increases in expression of the respective genes were observed, resulting in target gene deregulation and conduction disorders (VMC, unpublished data). These data indicate expression levels of these transcription factors are tightly balanced in sinus rhythm, and that similar to haploinsufficiency, mildly increased transcription factor expression can have physiologically relevant consequences. To date, loss of function models have been used to explore the role of particular genes in AF, however the above findings suggest that physiologic overexpression models need to be considered to appropriately recapitulate AF risk-associated changes in gene expression and accurately model AF molecular and clinical features.

Transcription is controlled by a combination of active and repressive transcription factors in concert with epigenetics, i.e. “molecules and mechanisms that can perpetuate alternative gene activity states in the context of the same DNA sequence”.⁹⁰ This definition includes DNA methylation, histone modifications, chromatin conformation and (nuclear/noncoding) RNAs, all of which are involved in regulation of gene transcription. For summaries of the literature on the role of noncoding RNAs, mostly microRNAs, in AF we refer to others.^91–96

Causative variant regions: identification of regulatory elements and single nucleotide polymorphisms that alter their activity

The vast majority of trait-associated variants are located in noncoding regions of the genome. These regions are posited to contain elements acting as regulatory chromatin. Therefore, identification of those elements harboring functional phenotype-associated genetic variation is required to understand how transcriptional regulation contributes to AF. Genetic variation may alter the function of regulatory elements by disrupting the binding of transcription factors to the associated binding motif in the regulatory element (Fig. 1) even if the variant does not directly disrupt the motif itself (reviewed in¹²).

Several factors complicate the search for transcription-modulating variants and their target genes. Multiple regulatory elements could be present in a single disease-associated linkage disequilibrium block, where the function could be either redundant, additive, synergistic or modulating.^97,98 These properties are of critical physiological importance but are not captured in current functional regulatory element assays (Table 2). Furthermore, while the mouse epigenome is more thoroughly characterized than that of human, conservation of mouse-human orthologous regulatory elements (based on H3K27ac ChIP-seq) conservation is limited,⁹⁹ and species-specific acquisition of new TF-binding sites within enhancers occurs frequently.¹⁰⁰ Therefore, mouse (or other non-human) data sets used to predict regulatory elements are very useful but should be interpreted with care. Currently, our ability to predict the function of regulatory elements from epigenetic data is poor. Additionally, regulatory element in vivo function (e.g. repression, modulating chromatin conformation) is poorly captured by most assays. Limited anecdotal examples of in vivo characterization of the physiological role of regulatory elements in atrial biology and AF has revealed the vast complexity of the gene regulatory mechanisms which control and establish normal cardiac rhythms.^15,18,19,101

Table 2:

Techniques and characteristics

Technique	Output	Advantages	Disadvantages	Refs
STARR-seq, Cap-STARR-seq, ChIP-STARR	Quantifies enhancer strength in complex candidate libraries	High throughput and size of test element	Elements tested episomally, in combination with a single selected gene promoter. Difficult to perform in physiologically relevant cell type.	123–125
SuRE, Variant MPRA	Survey of regulatory elements, Quantifies allele-specific enhancer strength	Allele-specific, high sensitivity and throughput	Elements tested episomally. Difficult to perform in physiologically relevant cell type.	117,126
MPRA, CRE-seq	Multiplexed measurement of transcriptional regulatory activity via synthetic reporter gene	High throughput	Restricted size of test element. Elements tested episomally. Difficult to perform in physiologically relevant cell type.	127–129
Cellular genetic screens (CRISPR/Cas9-based)	Cellular CRISPRi perturbations followed by single cell RNA-seq	Enhancer-target gene pairing	Perturbation not equivalent to natural situation	121,122
ATAC-seq	Assay for transposase accessible chromatin	Low cell number input, fast protocol	No functional information	130
ChIP-seq	Sequence reads obtained by antibody-captured protein-DNA interactions	Genome-wide	Dependent on crosslinking (PFA) and specificity of antibody	131,132
ncRNA-seq	Sequencing of RNA to identify active regulatory elements	Endogenous, in vivo, requires small input, genome-wide	Dependent on accessibility data, on level of expressed RNA and its stability.	56,103,105,133
CUT&RUN	Antibody-targeted controlled cleavage by nuclease	Low background signal, suitable for single cell analysis	Likelihood of over-digestion, dependence on antibody quality	134
3C	Chromatin architecture analysis “one-versus-one”	High resolution	Low throughput, use of primers, questionable quantification	112,135
4C	High-throughput chromatin architecture analysis “one-versus-all”	High-throughput	Dependence on restriction enzyme, missed close-range interactions, semi-quantitative (UMI-4C is quantitative)	136–138
5C	Medium-throughput chromatin architecture analysis “many-versus-many”	Medium-throughput	Limited in terms of size of genomic region assayed, dependence on restriction enzyme, missed close-range interactions	135,139–141
Hi-C, SPRITE	High-throughput chromatin architecture analysis “all-versus-all”	High-throughput, genome-wide	Dependence on restriction enzyme, missed close-range interactions	142,143
PC Hi-C	High-throughput chromatin architecture analysis “one-versus-all”	Enrichment for promoter interactions	Dependence on restriction enzyme, missed close-range interactions	144–146

Several epigenomic features can define regulatory elements: occupancy by specific transcription factors, association with activating histone post-translational modifications (such as H3K27ac), localization within DNase-hypersensitivity or ATAC-seq identified region of open chromatin, or the production of non-coding RNAs. For cell type or organ-specific occupation by a transcription factor or histones with an activating post-translational modification, various assays have been developed over the decades (Table 2). These epigenetic signatures have provided semi-comprehensive lists of candidate cell type specific regulatory elements and insight into their behavior.¹⁰² A significant body of literature indicates that active regulatory elements are transcribed, producing non-coding RNAs (ncRNAs).^56,103–106 The production of ncRNA from regulatory elements has been used to quantitatively define regulatory element activity,^104,107 including from native regulatory elements from the mouse atria in vivo.⁵⁶ The identification of regulatory elements based on ncRNA transcription has been applied to identify functionally active context dependent enhancers and has allowed identification of functional transcription factor-dependent regulatory elements.⁵⁶ In addition to their limited conservation across species, regulatory elements are often specific to cell-type and condition (developmental stage, stress).^108,109 Therefore, epigenetic datasets derived from human atrial tissues, stages of development and (disease) conditions are required to accurately identify AF-relevant regulatory elements.

While regulatory elements have been identified over one megabase (Mb) away from the target promoter(s), they are thought to act in close proximity, by the three-dimensional conformation of the DNA.^110,111 Because regulatory elements need to be in close physical proximity to their target promoters to directly regulate transcriptional activity,^112–114 knowledge of the conformation of the DNA is necessary in the search for target genes of variant regulatory elements.

More than a million candidate regulatory elements have been identified across the human genome, many of which are cell type- or condition-specific and temporally restricted.^115,116 Nevertheless, the majority of regulatory elements lack functional validation and their target genes are largely unknown. Therefore, the invention and application of new high throughput methods for functional evaluation of regulatory elements is a current focus of functional genomics research. Currently, high-throughput enhancer assays such as STARR-seq and MPRA are utilized for the large scale discovery of regulatory elements (Table 2), and have identified disease-relevant variant regulatory elements and expression modulating variants.^97,117–120 While these approaches enable the functional screening of large regions of the genome, a major drawback is that they are extra/chromosomal and are thus not subject to chromatin packaging (Table 2). Moreover, for practical reasons, most studies thus far have identified regulatory elements with an enhancing effect on transcription, while active repression of transcription is also physiologically relevant. New methods of identification and characterization of repressive regulatory elements is therefore essential for a comprehensive understanding of the relevant regulatory networks. Recently developed approaches consider regulatory element activity, and regulatory element-promoter contacts within native genomic context in model cell lines.^120–122 Application of such assays in AF-relevant cell types, such as atrial cardiomyocytes or fibroblasts derived from human with or without AF or from animals modeling AF, will be technically challenging, but may uncover functional variant regulatory elements at scale.

Several efforts have been made to identify human cardiac enhancers.^147,148 One study used >35 epigenomic data sets from mouse and human pre- and postnatal hearts to generate a reference of >80,000 putative human heart enhancers.¹⁴⁹ Recently, the enhancer prediction tool EMERGE¹⁵⁰ was trained using 70 human cardiac epigenetic data sets and a set of confirmed cardiac enhancers¹⁵¹ to predict human cardiac enhancers, 1750 of which are located in over 100 AF loci.¹⁵ The accessible chromatin in non-diseased human left atrial cardiomyocytes was identified using ATAC-seq.^15,19 These regions represent >83,000 putative regulatory elements in human atrial cardiomyocytes. An example of the output of such assays of the locus PRRX1 is shown in Fig. 4. Cross-referencing both this data set and the EMERGE predicted regulatory element set with AF-associated SNPs⁴ revealed 876 putative variant regulatory elements genome-wide.¹⁵ Several candidates were validated by luciferase assay, showing the potential of such a strategy for identification of disease-specific variant regulatory elements. Future efforts should provide epigenetic data from AF-prone heart tissue including pulmonary veins and from other cell types in the heart, such as fibroblasts, to identify tissue-specific AF-relevant variant regulatory elements.

Figure 4.

UCSC track showing the locus containing PRRX1 associated with AF and different datasets used for regulatory element and target gene identification. USCS browser view with lead variants, topologically associated domains (Juicebox TADs)(Durand et al., 2016;Robinson et al., 2018), identified enhancers (Tucker et al., 2017), SNPs associated with AF (p<10–4) (Roselli et al., 2018), annotated genes, promoter capture Hi-C (PCHi-C) data (Montefiori et al., 2018), ATAC-seq representing accessible chromatin in cardiomyocytes of left atria (Hill et al., 2019; van Ouwerkerk et al., 2019), EMERGE enhancer prediction (van Duijvenboden et al., 2015;van Ouwerkerk et al., 2019) and expression of left atria whole tissue (van Ouwerkerk et al., 2019).

Identification of the relationship between variant SNPs and biological effect

A second challenge is to link the variant regulatory elements to target genes (Fig. 1). Regulatory element-promoter targeting is selective and regulated.^97,114,152 The three principal approaches to define specific target genes of variant regulatory elements in any complex disease are 1) identification of the variant regulatory elements as described in the previous section; 2) identification of which genes at a locus are affected by associated variants (Fig. 5); and 3) the variant regulatory element as well as the directional effect on transcription should be validated in vitro or in vivo, by e.g. genome editing of cultured human cells or mice, to enable evaluation of the biological effect of different components on transcriptional networks and (electro)physiological function (Fig. 5).

Figure 5.

Commonly employed approaches and challenges to study the mechanistic link between genetic variation and AF predisposition. Technical challenges and unmet needs include: 1) Relevant cell type- and condition-specific (hypertension, aging etc.) conformation and transcriptome data are required to define candidate AF associated genes. 2) Relevant cell type- and condition-specific epigenetic data are required to identify physiologically relevant variant regulatory elements. 3) Assays to define physiologically relevant functions of regulatory elements (activation, repression, conformation, combinatorial) are required. 4) Sensitive models to recapitulate the effect of variants are needed.

A common and productive approach to resolve which genes within a locus are targeted by the identified GWAS variants is to link the signals to levels of gene expression. Such expression quantitative trait locus (eQTL) analysis involves the correlation of risk variants with gene expression levels in a specific tissue relevant to the disease studied.^4,5 However, only a fraction of the AF-associated sentinel SNPs could be linked to one or multiple candidate target genes. Due to the limited availability of relevant genotyped tissue or cell type (e.g. left atrial cardiomyocytes, pulmonary vein wall, atrial fibroblasts) for expression analysis, lack of knowledge of relevant tissue for particular genes, and the inability to detect small but biological relevant differences in expression, we currently lack sufficient power to identify target genes for all AF-associated loci.¹⁵³

Regulatory elements and target genes usually share the same topologically associated domain (TAD), a self-interacting genomic region in which DNA sequences physically interact with each other more frequently than with sequences outside the TAD.^154,155 TAD structures and loops within TADs are mediated by cohesin and CCCTC-binding factor (CTCF).^{152,156–159} Once regulatory elements have been identified, the potential target genes can be prioritized by interrogating their three-dimensional chromatin contacts in the relevant tissue. Recent Hi-C analyses of human atrial and ventricular tissue combined with improved statistical analyses of interactions have provided a highly relevant contact map facilitating the identification of target genes.¹⁶⁰ While regulatory element-promoter interactions are required for transcriptional regulation, close proximity observed by 3C-derived technologies does not necessarily imply a functional interaction.^111,126 Moreover, Hi-C and promoter-capture (PC)-Hi-C technologies do not identify all possible interactions, as contacts within 30kb are technically challenging to capture. Therefore, interactions between regulatory element and promoters, or a lack thereof, should be interpreted with care and require functional validation. In an effort to map cardiovascular disease risk loci interactions, two studies used PCHi-C in human iPSC-derived cardiomyocytes to map the promoter interactome, identifying interactions (physical proximity) between promoters and disease-associated regions with regulatory elements.^161,162 Indeed, both studies found that interactions involving cardiovascular disease and heart rhythm GWAS SNPs were enriched in promoter interactions with genes relevant for heart development and disease. Similarly, Hi-C of primary cardiac tissue combined with a new more sensitive and quantitative loop calling has been used to link the heart regulatory interactome and disease variants.¹⁶⁰ These genome-wide interaction maps of genome organization have also been useful to identify candidate target genes of AF variant regulatory elements, as PCHi-C datasets identify interactions between potential regulatory elements and target gene promoters (Table 2, Fig. 4).¹⁵

While the variant regulatory elements and their target genes have been predicted based on features such as epigenetic signatures, expression, contact, and eQTL,¹⁵ a few functional variant regulatory elements and their target genes have been experimentally identified (Table 3). The strongest AF risk locus -with associations orders of magnitude more significant than the subsequent loci- is found 150 kb upstream of the homeobox transcription factor PITX2 located on chromosome 4q25.^4,5 One study found a risk variant in high LD with the sentinel SNP that confers differential expression mediated by a transcription factor in an intron of PITX2a/b, upstream of the promoter of the cardiac specific isoform PITX2c.¹⁶³ The region containing the variant of interest (rs2595104) showed enhancer activity in zebrafish, and the risk allele, which is known to predispose to AF, demonstrated reduced enhancer activity in vitro. Subsequent CRISPR/Cas9-mediated deletion of the rs2595104 region in human stem cell-derived cardiomyocytes showed reduced PITX2c expression compared to the unedited allele.¹⁶³ Moreover, rs2595104 lies in a TFAP2a binding site, a ubiquitously expressed transcription factor that was previously shown to be important for PITX2 expression.^164,165 The authors verified that TFAP2a binds preferentially to the non-risk allele, and that TFAP2a knock down mediated the reduction of PITX2c expression.

Table 3.

Currently identified AF-associated (variant) regulatory elements and target genes

Locus	Regulatory element/variant	Models used	Target gene identified	Refs
PITX2	Intron of PITX2a/b / rs2595104	Enhancer reporter assays in zebrafish	PITX2c expression down after deletion of the rs2595104 region in human stem cell derived cardiomyocytes	163
PITX2	20 kb region upstream of Pitx2	Orthologues region deleted from mouse genome	Pitx2c expression down (males) Enpep expression up (females)	19
PRRX1	Upstream of Prrx1 / rs577676	Enhancer reporter assays in embryonic zebrafish / HL-1 cells	-	13
GJA1	0.7 Mbp downstream of GJA1	Orthologues region deleted from mouse genome	Gja1 expression decreased in atrial tissue	15
KCNN3	First exon and promoter of Kcnn3	Orthologues region deleted from mouse genome	Reduced expression of several nearby genes	15
SCN5a	Downstream of Scn5a	Orthologues region deleted from mouse genome	Scn5a expression decreased in atrial tissue	166
SCN10a	Intronic of Scn10a/ rs6801957	Orthologues region deleted from mouse genome	Scn5a expression decreased in cardiac tissue	167,168

Many more (876) putative regulatory elements have been identified in AF associated regions using EMERGE and ATAC-seq, and although a handful of these showed allele-specific activity in luciferase assays, the vast majority require in vivo validation as well as target gene identification.¹⁵

It was hypothesized that the highest risk-associated region near 4q25 was regulated by TBX5.¹⁴ Using a combination of ATAC-seq, 4C, and ChIP-qPCR (Table 2), TBX5 was confirmed to occupy the regulatory element that is in contact with the PITX2 promoter in human LA. Furthermore, this RE regulatory element harbors an AF risk variant rs1906595, the major allele of which abolishes regulatory element activity.

In the mouse orthologue of the AF-associated variant region 170 kb upstream of Pitx2, a non-tissue-specific regulatory element specifically contacts the promoters of Pitx2c as well as Enpep, a gene adjacent to Pitx2.²² Further analyses are required to establish whether both Pitx2c and Enpep are targets of variant regulatory elements in the upstream gene desert, and what could be mechanisms underlying the change in target gene expression.

In a recent study the mouse orthologues of two variant noncoding regions in the gene desert upstream of Pitx2 were deleted from the mouse genome using CRISPR/Cas9 genome editing.¹⁹ One of these regions, a 20 kb region upstream of Pitx2 with regulatory activity, is in contact with the promoter of Pitx2 as shown by 4C-seq (Table 2). Deletion of the regulatory element reduced expression of Pitx2c specifically in male mice, and led to inducible AF susceptibility. Of note, regulatory element deletion caused increased Enpep expression only in females. Moreover, the regulatory element region was essential for and specific to maintaining the active chromatin state of the Pitx2c promoter specifically. Disruption of the Pitx2 intronic CTCF binding site caused reduced Pitx2 expression, AF predisposition and reduced active chromatin marks on Pitx2 indicating that long-range looping was mediated by CTCF. However, the separate deletion of the two enhancer regions did not lead to major phenotypic abnormalities without stress, and the authors speculate that a pair-wise deletion could uncover that these two enhancers confer phenotypic robustness.¹⁹ Further research should uncover whether these two identified enhancers are both needed in collaboration to reduce Pitx2 expression (or other downstream mechanisms) to a degree sufficient for an arrhythmogenic phenotype. Despite these efforts, AF relevant target genes of the highest risk-associated region near 4q25 are still uncertain. Although there is no correlation between PITX2 expression changes and AF, most models suggest that regulatory elements present in AF risk variant loci are linked to PITX2 regulation.

A region near PRRX1, a transcriptional co-activator important for development of the heart,⁶⁸ associated with Agnathia-Otocephaly and Dysgnathia Complex,¹⁶⁹ also harbors a variant regulatory element. Putative enhancers were selected in the AF-associated region on 1q24 upstream of PRRX1 using conservation and genomic markers of transcriptional enhancers.¹³ Two of these putative enhancers showed strong eGFP expression in cardiac and skeletal muscle tissue when tested in embryonic zebrafish (Fig. 4). Twenty-one common SNPs within these regulatory elements were tested for enhancer activity of the risk versus non-risk allele in HL1 cells, demonstrating that only SNP rs577676 led to decreased regulatory element activity. Possible target genes in the variant region were identified using Hi-C and 3C, suggesting physical contact between the identified regulatory element and the promoter of PRRX1 in cardiac fibroblasts compared to human embryonic stem cells. This supports the idea that the risk variant negatively regulates the expression of PRRX1, providing a direct mechanism linking the risk variant and target gene for this locus. Additionally, optical mapping of cardiomyocytes differentiated from PRRX1 knockout human embryonic stem cells and Prrx1a morpholino-treated zebrafish embryos revealed the action potential duration was significantly shortened in PRRX1-knockout cardiomyocytes, a phenotype often seen in AF patients.

The mouse orthologue of another AF-associated regulatory element was found regulating expression of the AF-relevant gene Gja1 in the heart.¹⁵ Gja1 encodes the major gap junctional protein with a role in impulse propagation and cell-cell coupling.^170,171 Using CRISPR/Cas9 genome editing a region containing the mouse homologue of a 28 kb AF associated locus 0.7 Mb downstream of GJA1 was deleted. Mice homozygous for the deletion showed atrial-specific decreased expression of Gja1. The same study found that deletion of the mouse orthologue of the variant region close to the potassium-calcium activated channel KCNN3 caused the reduced expression of several nearby genes, whereas deletion of the 33 kb orthologue of the variant region in the first intron of the zinc-finger transcription factor, ZFHX3 did not alter cardiac expression of any gene within the TAD in vivo.

Multiple regulatory elements were found around the SCN5A-SCN10A locus, which has frequently been associated with cardiac rhythm and conduction as identified through GWAS and with ECG parameters being linked to AF^{4,5,168,172–176} and Brugada syndrome.¹⁷⁷ One of the regulatory elements of SCN5A is positioned in an intron of SCN10A,¹⁶⁸ and contains a common variant that is associated with PR interval.⁵⁴ The variant disrupts a T-box binding site in the regulatory element, and is associated with decreased SCN5A expression.^167,168 Two other regulatory elements are located in introns and downstream of SCN5A, the latter of which also contains variants associated with PR interval and QRS duration.^54,178–180 The regulatory element cluster (“super enhancer”) downstream of SCN5A is evolutionary conserved, and drives gene expression in the heart in a pattern resembling that of Scn5a. When deleted from the mouse genome, the cluster, which harbors conduction parameter-associated SNPs, was found to be essential for Scn5a expression and for bringing the other regulatory elements and promoters in the locus physically together to regulate Scn5a.¹⁰¹

Epigenetic states link environment and gene regulation

There are several risk factors for cardiovascular disease, the most important of which is age. Interestingly, epigenetic alterations have been proposed to underlie age-related transcriptional alterations.¹⁸¹ Indeed, changes in epigenetic state induced by aging (or other risk factors such as hypertension and metabolic disease) could provide the basis for the “uncovering” of genetic risk factors at AF-associated loci. Moreover, combined GWAS loci only account for a portion of heritability, suggesting that a part of the so-called missing heritability might be found in non-sequence related epigenetics.¹⁸² An example of the importance of changes in epigenetic state is the epigenetic clock. Chronological age is highly correlated with epigenetic alterations of the methylome,¹⁸³ and deviation of this methylation state could be used to predict healthspan¹⁸⁴ and is correlated with cardiovascular disease incidence.¹⁸⁵ Altered DNA methylation was found near or on the promoters of poised genes, which contributed to transcriptional heterogeneity of these genes.¹⁸⁶ A methylome-wide association study of an AF cohort revealed 14 AF-associated SNPs to be associated with a CpG site in blood-derived genomic DNA.¹⁸⁷ In a smaller study, a CpG islet proximal to PITX2 was found to be hypermethylated in AF patients compared to controls, along with decreased PITX2 expression in AF patients.¹⁸⁸ Furthermore, hypermethylation in the AF susceptible loci containing PITX2 and CCDC141, as well as hypomethylation in the locus containing CACNA1C locus, were identified in all 7 patients with permanent AF.¹⁸⁹ These data suggest that changes in DNA methylation could be involved in AF.

Not only methylation, but also the alteration of other epigenetic mechanisms such as histone modification and chromatin remodeling could contribute to disease risk.¹⁹⁰ Transcription factors regulate or maintain cell fate by activating certain and repressing other target genes in cooperativity with chromatin remodelers and histone modifying enzymes. Chromatin remodelers respond to modifications to histone post-translational modifications including methylation, phosphorylation, acetylation, ubiquitylation, and sumoylation.^191,192 Recognizing these modifications, chromatin remodelers make distant regulatory elements as well as promoters accessible (with tissue-specificity) by regulating nucleosome dynamics.¹⁹³ There is extensive cross-talk between histone modifications and chromatin remodelers, the balance of which determines the recruitment of transcription factors.¹⁹⁴ Evidence of a role for both histone modifications and chromatin remodelers in “complex trait” diseases such as AF is emerging. For example, the interplay between transcription factors and chromatin remodeling complex dosage of Brg1 were shown to be important during development,¹⁹⁵ which suggests a mechanism of interdependence between transcription factors and remodeling proteins in the heart. Additionally, histone deacetylases, were found to modulate pathogenic gene expression in many diseases including AF progression.¹⁹⁶ Moreover, there is evidence that increased expression of histone modifier enhancer of zeste homolog 2 (EZH2), which catalyzes the deposition of methylation to histone 3 at lysine 27 (H3K27me3) is associated with fibrosis in AF patients.¹⁹⁷

Another possible mechanism linking epigenetics and AF are reactive oxygen species (ROS) levels. Increased levels of ROS are known to accelerate aging, and they are associated with AF pathogenesis.^198,199 It is known that ROS levels have an impact on the mechanisms which govern the epigenome, such as DNA methylation, histone modification and ncRNAs.²⁰⁰ Increased ROS levels can disrupt the epigenetic balance and thus the epigenetic state of a cell, causing potentially altered mechanisms (such as “uncovering” of the effect of risk variants) leading to increased AF predisposition. Moreover, one study found that oxidative stress promotes AF via increased intracellular Ca²⁺ release by oxidized RyR2.²⁰¹ In mice, ROS is increased in response to injury, but also during the natural transition from glycolytic to oxidative metabolism shortly after birth as well as during aging. This increase in ROS inhibits cardiomyocyte regeneration.²⁰² Indeed, Pitx2 is known to regulate the transcriptional response to oxidative stress (cooperatively with Yap).⁴⁹ In response to injury, Pitx2 expression is induced, which in turn activates ROS scavengers, protecting cells from injury. Future models should provide insight into the interplay between ROS, PITX2, TBX5 and AF, but as of yet the interaction between these factors and the effect on effectors has not been studied.

In addition to the contribution of aging via ROS, fetal and developmental origins of disease may also be mediated by epigenetic responses to environmental factors.²⁰³ In a generalized view, heritable epigenetic states of chromatin are subject to developmental changes, and are sensitive to internal variation (e.g. variable transcription factor activity) and external signals (e.g. metabolic state) during development. This leads to interindividual variability in regulatory element deployment, gene expression and cellular and tissue phenotype. Indeed, the same could hold for AF, as there is a high density of AF-risk variants near cardiac developmental transcription factor genes such as NKX2–5, PITX2, TBX5, and TBX3 as well as their target genes such as ion channels (HCN4, KCNN3) and gap junctions (GJA1). Two of the genes, PITX2 and NKX2–5, well-known players in AF, are responsible for the correct development of the left atrial myocardium as well as the myocardial sleeves of the pulmonary veins.³⁹ The myocardial sleeves of the pulmonary veins are composed of cardiomyocytes resembling working atrial myocardium, and it are the sites where ectopic foci that initiate AF are often found.^204,205 It has been speculated that a developmental change in expression of these genes combined with other structural or epigenetic remodeling could ultimately lead to the development of ectopic foci in the pulmonary vein myocardial sleeves.^206–208 Likewise, HCN4 encodes a potassium channel responsible for the generation of pacemaker current in the sinoatrial node during early development, as well as atrial rhythm.^209,210 However, the strict requirement of HCN4 function is confined to embryonic development^209–211 as it seems to be dispensable in adult heart rhythm.²¹² Indeed, AF is characterized by irregularly irregular heart rates, suggesting that developmental changes in expression of this gene could contribute to AF in later life.

Conclusion

AF is a collective name for an expressed disease phenotype with many distinct pathological mechanisms. GWAS and genetically modified animal models have indicated that AF predisposition likely involves many additive small expression changes induced by variant regulatory elements. While several parameters have been clearly associated with AF, including impulse conduction, action potential duration, intracellular calcium handling, and fibrosis, the origins of AF and the cell-types involved are still not fully known. An additional confounding factor is that although many genetic components have been associated with AF, some heritable components remain unexplained. This missing heritability could be caused by epigenetics not identified by sequence-based GWAS. Indeed, there are developmental and aging related epigenetic changes that could contribute to interindividual variability. This could result in differential regulatory element deployment and gene expression in a tissue- or developmental stage-specific manner.

Here we conclude that there is growing evidence that a transcription factor network could modulate mechanisms underlying AF. Essential cardiac developmental transcription factors that are situated close to AF associated variants have been linked to AF or factors that predispose to AF. Here we propose a threefold model that can link genetic risk variants and transcription factor networks to AF predisposition. First, AF-associated genetic variation in binding sites of transcription factors can lead to altered expression of target effector genes (e.g. ion channels). Second, genetic variants could interfere with transcriptional regulation of these AF relevant transcription factors themselves (Fig. 1B). Third, transcription factors are now also known to interact as cofactors of each other in the regulation of target gene expression (Fig. 3). Therefore, it is likely that a slight alteration in expression levels of any of these transcription factors –as a result of variant regulatory element activity- can alter the delicate balance in the transcriptional network, affecting its output, which results in altered expression of target effector genes and AF predisposition (Fig. 3).

To uncover the contribution of epigenetic state to AF risk in relation to genetic variation there is a pressing need for availability of datasets from relevant cell types and developmental stages. Additionally, we require knowledge of the effect of disease states such as high blood pressure, stress and aging on these epigenetic parameters. This would allow the identification of variants with a biologically relevant impact on the system in the context of complex disease. Specifically, assays are required that subsequently test variant regulatory element function and their effect on target gene expression. Moreover, we lack the means to test the impact of combinations of different variant regulatory elements on target gene expression and the transcriptional network. Most importantly, there is an unmet need for the possibility to perform such techniques in relevant cell or tissue types. With the likely diverse causes of AF in different patients, a valuable addition to AF treatment would be to make patient-specific classifications of the disease based on genetic and epigenetic parameters.

Non-standard abbreviations:

AF

Atrial fibrillation

GWAS

Genome-wide association studies

eQTL

expression quantitative trait locus

TAD

topologically associated domain

CTCF

CCCTC-binding factor

PCHi-C

Promoter capture Hi-C

iPSC

induced pluripotent stem cell

SNP

single nucleotide polymorphism

ChIP-seq

chromatin immunoprecipitation sequencing

ATAC-seq

assay for transposase accessible chromatin