Genetic and molecular underpinnings of atrial fibrillation

In atrial fibrillation (AF), the most common sustained arrhythmia¹, chaotic atrial electrical activity sustained by re-entrant circuits, replaces normal rhythmic atrial depolarization and contraction initiated by the sinus node. The chaotic electrical activity causes uncoordinated atrial contractions and the ineffective movement of blood through the atria. In addition, fibrillating atria often exhibit morphological remodeling, including dilatation and fibrosis^2,3.

Patients with AF have elevated risk for stroke (5-fold)^4,5 heart failure (HF; 1.7-fold) and death (3.7-fold)^6,7,8. The worldwide number of individuals with AF was estimated to be 46.3 million in 2016⁹, and the prevalence will increase with increasing mean population age¹⁰. The lifetime risk for the US population is estimated to be 1 in 3 for individuals of European ancestry¹¹. The disease represents an ever-increasing burden on the healthcare system, adding nearly $25,000 to the annual cost of care per person¹². Despite its high socioeconomic cost, therapies to treat AF remain insufficient. Currently, available treatment options attempt to control heart rhythm or rate but have variable efficacy and significant risk for adverse effects.

We are entering into a new era for the treatment of myopathies and arrhythmogenic conditions, with novel treatments being developed to target underlying molecular and genetic mechanisms. For example, a treatment approach for hypertrophic cardiomyopathy through inhibiting (β)-myosin heavy chain ATPase activity recently gained FDA approval¹³, and gene therapies are currently being developed to treat the genetic lesions underlying HCM^14,15. Similarly, elucidating the molecular basis for other muscle diseases and proarrhythmic conditions including Duchenne muscular dystrophy, Barth syndrome, catecholaminergic polymorphic ventricular tachycardia (CPVT), and arrhythmogenic cardiomyopathy has resulted in the development of gene therapies that have shown efficacy in preclinical or clinical trials^{16,17,18,19,20}. To make AF similarly targetable, we must continue to advance our understanding of AF pathophysiology at the molecular and genetic levels.

In AF, one barrier to developing targeted therapies is an incomplete understanding of the aCM-specific gene regulatory network, the regulatory circuitry that aCMs use to express genes at the appropriate levels to maintain normal atrial contraction and rhythm. This network includes transcription factors, chromatin epigenetic modifications, and three-dimensional chromatin architecture. aCMs are phenotypically distinct from ventricular cardiomyocytes (vCMs)^21,22,23, and these phenotypic differences are driven by the differential expression of thousands of genes^21,24. Since aCM characteristics have been tuned to maintain atrial rhythm, comparing aCMs with vCMs can identify aCM properties, genes, and regulatory modules critical for maintenance of atrial rhythm.

This review will examine differences between aCMs and vCMs, including their unique developmental trajectories, ultrastructures, and gene expression profiles in mouse and human tissues, in order to identify features of aCMs that might be important for atrial rhythm homeostasis. Genes linked to AF by genome-wide association studies (GWAS), studies of AF families, and targeted sequencing of AF patients are then examined in the context of the aCM and vCM gene regulation and AF pathophysiology.

Although aCMs and vCMs are both cardiomyocytes, each cell type has specific distinguishing features. Because aCM- and vCM-selective features are likely critical for proper atrial and ventricular function, understanding them and the regulatory mechanisms that maintain them could provide important insights into their dysfunction in disease and potentially lead to future therapies. We consider differences in physiological characteristics and developmental programs and examine genes differentially expressed between aCMs and vCMs in mice and humans.

Chamber-specific differences in cardiomyocyte organization and function

aCMs and vCMs differ in size, ultrastructural organization, metabolism, and electrical function^21,22,23. The most obvious difference is their physical size; compared to vCMs, aCMs are both shorter and thinner (Fig. 1a)²¹. Perhaps related to their smaller size, aCMs have a lower density of T-tubules, invaginations of the plasma membrane that coordinate excitation–contraction coupling, than vCMs. Another contrasting feature of aCMs and vCMs is their metabolism. vCMs contain a much greater density of mitochondria²¹. In keeping with this observation, vCM-selective genes are enriched for functional terms related to oxidative phosphorylation and mitochondrial function²¹.

a Micrographs of isolated adult mouse ventricular and atrial cardiomyocytes. b Volcano plot obtained by pseudobulk RNA-seq analysis obtained from the human heart atlas. aCMs and vCMs were extracted from samples with >500 aCM or vCMs, pseudobulked, and compared using DEseq2. Genes with chamber selectivity were considered with Log2FC > |1| and pval adj. < 0.05. c Human and mouse data were plotted against each other to examine species-specific biases in chamber selectivity. The Pearson correlation coefficient (r = 0.4, p = 1.78E − 40) was determined for genes that were differentially expressed in both datasets. d Intersection between genes selectively expressed in human or mouse atria or ventricles.

Full size image

Electrically, aCMs exhibit less polarized resting membrane potentials and faster action potential kinetics than vCMs. More rapid membrane repolarization occurs in aCMs due to inward-flowing potassium currents found selectively in aCMs, including the atrial-specific ultra-rapid repolarization current (I_Kur) mediated by the KCNA5 channel²⁵ and the acetylcholine-activated K⁺ current (I_KACh), mediated by the KCNJ3/5 channels²⁶. These ion channels are expressed at greater levels in aCMs compared to vCMs, resulting in the characteristic action potential shape of each cardiomyocyte subtype.

aCM and vCM-selective gene expression

Gene expression profiles differ between aCMs and vCMs. These chamber-selective expression profiles include transcription factors important for the maintenance of aCM or vCM identity (discussed in greater detail below), and altering the expression of these key transcription factors can blur the distinct phenotypes of aCMs and vCMs. For example, inactivating Nr2f2 in early development resulted in a denser and more organized T tubule network in cardiomyocytes of the atrial chamber²⁷, a vCM-like feature, while Tbx5 inactivation in aCMs increased action potential duration²⁸, making them more similar to vCMs.

Several sarcomeric genes are differentially expressed between aCMs and vCMs, including the aCM-specific myosin light chains 4 and 7 (Myl4 and Myl7) and the vCM-specific Myl2 and Myl3. There are other well-known examples, including Nppa, which is initially expressed in all cardiomyocytes, becomes restricted to postnatal aCMs, and is re-expressed in postnatal vCMs under pathological stress²⁹. While these markers represent some of the most highly differentially expressed aCM and vCM-selective genes, many other genes are differentially expressed to a lesser degree. Early comparison of gene expression differences between atria and ventricles employed whole tissues^30,31, which obfuscated the comparison between cardiomyocyte subtypes. Recently, we used bulk RNA sequencing to measure gene expression differences between isolated neonatal murine aCMs and vCMs²¹. We detected 1126 genes with aCM-selective gene expression and 872 genes with vCM-selective gene expression (|Log₂FC| > 1.5, Padj < 0.05).

Although mice and humans share many aspects of cardiac biology, there are also notable differences, including differences in heart rate and a subset of sarcomere and ion channel genes¹⁴. To compare differences in gene expression between human aCMs and vCMs, we analyzed publicly available snRNAseq datasets of non-diseased adult human atria and ventricles^32,33. Pseudobulk RNA sequencing analysis of cardiomyocyte nuclei from these tissues³³ identified human aCM- and vCM-selective genes (|Log₂FC| > 1 and Padj < 0.05) (Fig. 1b). Like mice, human aCMs had more chamber-selective genes (1976) than vCMs (793, Supplementary Table 1). We compared aCM vs. vCM expression differences between species (Fig. 1c). Despite the numerous technical and experimental differences between these analyses, twice as many genes were concordant between species (quadrants I and III) than discordant (quadrants II and IV), and genes with chamber-selective expression in both species were moderately but highly significantly correlated (Pearson r = 0.4, p = 1.78E-40). This analysis demonstrates the cross-species conservation of many chamber-specific characteristics. However, many genes exhibited selective expression in only one species. Additional experiments will be required to determine if these divergent genes are biological or attributable to technical differences between these studies. These data support the use of mice to study chamber-specific gene expression and diseases but suggest species-specific differences might complicate the cross-species interpretation of some results.

To circumvent the limitations of mice to model human disease, protocols have been developed to selectively differentiate human pluripotent stem cells (PSCs) into aCMs or vCMs^34,35. These protocols, which largely depend on the addition of retinoic acid during differentiation to promote aCM as opposed to vCM specification, were recently reviewed³⁶. While largely successful in producing PSC-aCMs with atrial-like action potential kinetics and gene expression profiles, some limitations include high spontaneous beating frequency, an immature phenotype, and the expression of SHOX2, a marker of the sinoatrial node lineage. More recent studies combined the co-culture of PSC-aCMs with primary human adult atrial cardiac fibroblasts and soft lithographic patterning of ECM proteins in order to obtain more mature aCMs³⁷. However, this process is laborious and technically demanding, creating the need for more streamlined approaches.

In spite of their limitations, PSC-aCMs have been successfully applied to study several AF-causing variants^38,39,40. Ghazizadeh et. al. studied PSC-aCMs harboring a variant in the atrial-selective gene MYL4 that segregated with familial AF in an Icelandic family⁴¹. This approach revealed multiple cellular mechanisms promoting AF, including increased retinoic acid production and actin disorganization. Increased protein kinase C activity, downstream of increased retinoic acid signaling, elevated phosphorylation, and channel permeability of the gap junction protein CX43. A drug screen in the PSC-aCMs and a zebrafish model harboring the same variant showed that the small molecule carbenoxolone, a CX43 inhibitor, reversed these phenotypes. This work highlights the use of in vitro modeling approaches to dissect AF mechanisms and identify potential therapeutic approaches. It also illustrates the value of approaches to produce and characterize PSC-aCMs.

Development and maintenance of aCMs and vCMs

During heart development, aCMs and right ventricular vCMs arise from second heart field progenitors, whereas left ventricular vCMs arise primarily from first heart field progenitors. aCMs and vCMs proceed through specialized developmental programs resulting in the two distinct but related cell types^{27,42,43,44,45,46,47,48}. Following heart looping, the primitive atria and ventricles are demarcated by the selective expression of transcription factors, signaling pathway proteins, and cardiomyocyte functional components specific to aCMs and vCMs. One of the earliest identified chamber-selective transcription factors, IRX4, is restricted to the developing ventricles, and its ectopic overexpression in embryonic chick atria induced the expression of a vCM-specific myosin⁴⁴. Since this initial discovery, multiple transcription factors and signaling pathways contributing to the vCM developmental program have been identified, including FGF signaling and the transcription factors NKX2-5/2-7 and HEY2^43,46,47.

Several factors have been shown to promote aCM differentiation. Retinoic acid (RA) signaling is critical for determining the posterior second heart field (pSHF), which gives rise to most of the atrial compartment³⁶. In addition to retinoic acid signaling, the transcription factor Nr2f2 promotes atrial differentiation²⁷. Nr2f2 inactivation at E12.5 resulted in the downregulation of Myl7, an atrial-specific myosin light chain, and upregulation of Myl2, its ventricle-specific counterpart. However, the inactivation of Nr2f2 later in development failed to alter aCM identity, suggesting distinct requirements for aCM differentiation vs. maintenance, with Nr2f2 required for the former and not the latter.

Cell identity is maintained through the regulation of gene expression by tissue-specific enhancer–promoter networks. Recent studies by our group examined transcription factors important for the postnatal maintenance of either aCM or vCM-selective gene expression^21,24. We used a massively parallel reporter assay to identify enhancer elements with selective activity in aCMs and vCMs. Motif analysis suggested that the estrogen-related receptor (ERR) and TBX5 were important vCM-selective and aCM-selective enhancer activity, respectively. Inactivating ERRα and ERRγ in postnatal mouse hearts resulted in the downregulation of many vCM-selective genes and upregulation of many aCM-selective genes, confirming that ERRα/γ are essential to maintain vCM gene expression.

TBX5 is a transcription factor that is initially expressed in the cardiac crescent⁴⁹. Expression soon becomes restricted to the posterior portion of the heart tube, which will develop into the atria. As the heart loops and septates, this graded expression is maintained, with highest expression in atria, moderate expression in the left ventricle, and little to no expression in the right ventricle and outflow tract. TBX5 is also expressed in the AV node and cardiac conduction system, where it is essential for cardiac conduction system specification and gene expression⁵⁰. In the mature heart, Tbx5 expression is ~10-fold higher in aCMs than left ventricular vCMs, and postnatal inactivation of Tbx5 in aCMs caused AF^23,24,28,51, which was accompanied by downregulation of aCM-selective genes and upregulation of vCM-selective genes²⁴. Upon Tbx5 inactivation, regions near aCM-selective genes that are normally bound by TBX5 became less accessible, had diminished H3K27ac, an active enhancer mark, and lost contacts with promoters²⁴. Increased TBX5 activity also causes AF, as a TBX5 missense gain of function mutation caused familial AF, and this was recapitulated in mouse models^52,53. Together, these data demonstrate that TBX5 is required to maintain aCM identity and atrial rhythm.

Notably, altering the chamber-selective expression programs of either aCMs or vCMs is detrimental to heart function. Mice lacking cardiomyocyte ERRα and ERRγ develop bradycardia and lethal cardiomyopathy⁵⁴, while mice lacking TBX5 in aCMs develop permanent AF^24,28, atrial cardiomyopathy, and at later stages, fibrotic remodeling restricted to the atrial chambers (ref. ²⁴ and our unpublished observations). These observations suggest that maintaining chamber-selective cardiomyocyte identity is critical for healthy heart function.

AF demonstrates high heritability, pointing to a strong genetic component. A previous study of Danish monozygotic and dizygotic twins suggested that AF heritability was as high as 62%⁵⁵. However, family-based studies may overestimate heritability, and a more recent study estimated that AF heritability is 22.1%, with the large majority attributable to the additive effects of common genetic variants (minor allele frequency > 5%)⁵⁶. In addition to common variants with smaller effect sizes, rare variants with larger effect sizes are also observed. Lone AF (AF in the absence of structural heart disease) aggregates within families⁵⁷. In the Framingham Heart Study, 30% of offspring participants that developed new-onset AF had at least one parent with AF, and an AF diagnosis in at least one parent was positively associated with developing AF (OR, 1.85, p = 0.02)⁵⁸. Another study in the Icelandic population demonstrated that an AF diagnosis in individuals under 60 years of age was more than five times as likely if they were a first-degree relative of a family member diagnosed with AF prior to the age of 60⁵⁹. Together, these studies highlight the substantial genetic components underlying AF susceptibility.

Identifying genes associated with AF pathogenesis through genome-wide association studies

GWAS, linkage analysis, and coding variation have been used to determine the genetic underpinnings of AF. The first AF GWAS, reported in 2007, examined a cohort of 550 patients vs. 4476 controls. This study identified two single nucleotide polymorphisms (SNPs) in the 4q25 locus near the transcription factor PITX2⁶⁰. Since then, GWAS studies have expanded dramatically in study population size and hence statistical power^{61,62,63,64,65,66,67,68,69}, with the most recent study⁶⁹ comparing 77,690 cases from Japan and Europe against 1,167,040 controls. From this analysis, 150 loci with genome-wide significance (log₁₀ Bayes factor (BF) > 6) were identified, including 33 novel loci. Among the 3637 genetic variants in linkage disequilibrium (r² > 0.8) with the lead SNP of each locus, only 19 are missense variants in protein-coding genes, demonstrating that the majority of AF variants are found in the non-coding genome. Presumably, these non-coding variants alter the transcriptional regulation of their target genes.

To translate the discovery of AF risk variants into new therapies, it is essential to decipher the gene(s) that they regulate. This process is complicated by the fact that cis-regulatory regions do not necessarily regulate the closest gene or even a directly neighboring gene. Historically, genes have been associated with SNPs through expression quantitative trait loci (eQTL) analysis⁷⁰. This method has now been augmented with molecular biological approaches that more directly implicate non-coding variants with target gene expression, including mapping enhancer–promoter contacts and epigenetically modulating candidate loci using CRISPR-i/a^71,72,73 and CRISPR genome editing⁷⁴.

Recent studies utilized these approaches to identify genetic drivers of AF^75,76,77,78. van Ouwerker et al.⁷⁵ identified candidate genes located within 1.9 mb of 104 genomic loci implicated in AF by (1) their presence within the same topologically associated domain; (2) contact of the implicated region with the gene promoter, as determined by promoter-capture Hi-C; (3) expression level in adult and fetal left and right whole atrial tissue and aCMs from adult left atria; and (4) eQTL analysis. If a gene was expressed, it received a score of at least 10, and a score of 11 or greater was considered a causative gene. This approach identified 264 potentially causative genes. However, a weakness of this study is that the scoring system lacked a strong statistical foundation⁷⁸.

Another recent study⁷⁷ used single nucleus RNA and ATAC sequencing (snRNAseq and snATACseq) to identify regulatory elements specific for each cardiac cell type. Region/gene associations were determined through co-accessibility analysis of the putative enhancers and target gene promoters⁷⁹. Next, candidate cis-regulatory elements (CREs) specific for aCMs and vCMs were intersected with AF SNPs determined to be causative by Bayesian fine mapping (posterior probability of association >10%), resulting in the identification of 40 fine-mapped variants within 38 cardiac CREs and linked to 38 putative genes. The authors went on to validate one variant predicted to regulate KCNH2. The variant allele decreased enhancer activity in a luciferase assay, and CRISPR-mediated genomic deletion of the enhancer containing the variant reduced KCNH2 mRNA levels and altered electrophysiological properties. This approach was successful but only prioritized genes at a small number of AF-associated loci.

Selewa et al. identified genes regulated by AF SNPs by using cardiomyocyte chromatin features (open chromatin, H3K27ac, and promoter-capture Hi-C contacts) to nominate associated cardiomyocyte enhancers⁷⁸. These were linked to genes using a pipeline called ‘Mapgen’ to determine a Posterior Inclusion Probability (PIP) score for each gene surrounding each fine-mapped SNP based on the likelihood that the SNP regulated the gene. This likelihood was assigned a maximal value if it met any of the following criteria: (1) the SNP contacted the gene promoter, based on iPSC-CM promoter-capture Hi-C; (2) the SNP was located within an open chromatin region and within 20 kb of the gene; or (3) the SNP was located within the gene’s untranslated regions. If a SNP did not meet any of these criteria, the likelihood was determined by the distance from the SNP to the gene transcriptional start site, which decayed exponentially. Next, a ‘gene PIP score’ was determined by considering individual PIP scores for each of the SNPs predicted to target a given gene. Thus, if multiple SNPs targeted a single gene with high PIP scores, the gene was considered likely to drive AF and received a high gene PIP score. This approach resulted in the identification of SNP-gene pairs that captured >80% of the causal signal for each of the 122 loci examined, pinpointing a single gene at 42 loci and two genes at 34 loci. A total of 48 genes were considered highly likely (PIP > 0.8) by this approach. For several genes, high gene PIP scores were driven by aggregating multiple SNPs with moderate PIP scores. To validate their model, the authors demonstrated that several SNPs altered enhancer activity in luciferase assays in aCM-like HL-1 cells. While this study describes a powerful new approach for identifying potentially causal genes, it was based on vCM chromatin feature maps even though these features differ between aCMs and vCMs, especially near genes that are selectively expressed between the two cell types⁷⁷. Scoring likely would be improved by using aCM-specific chromatin feature maps.

Experimental validation of informatic predictions remains a major bottleneck, both for validating the impact of SNPs on candidate gene expression, and for validating the significance of gene expression changes on AF pathogenesis.

Features of genes implicated in AF

We summarized the chamber-selective expression of the genes linked to AF by their proximity to AF GWAS loci or by Mapgen (Table 1). AF-linked genes had a higher propensity to be aCM-selective (Fig. 2a). In total, 41 (19%) AF genes were aCM-selective: 13 (6%) in humans, 18 (13%) in mice, and 10 (5%) in both. On the other hand, 25 (12%) genes showed vCM selectivity in one or both species, with 7 (3%) vCM-selective in humans, 13 (6%) in mice, and 5 (3%) in both. Although these genes were more highly expressed in vCMs, they were also expressed in aCMs, where their expression level is likely important for normal atrial rhythm.

a Chamber-selective expression of atrial fibrillation genes. b Functional class of atrial fibrillation genes identified by GWAS. c Functional class of atrial fibrillation genes identified in pedigree and targeted sequencing studies.

Full size image

We annotated the main functional classes of AF-associated genes and their chamber-selective expression (Fig. 2b). Ion channel genes were significantly enriched for aCM-selective expression, while Ca²⁺-handling genes frequently exhibited vCM-selective expression. These results highlight the importance of ion channels to regulate the atrial action potential and maintain rhythm. While AF genes were more often aCM-selective than vCM-selective, most AF genes were not aCM-selective (Fig. 2a; Table 1).

Since TBX5 is a master regulator of aCM identity^21,24, we also examined if genes implicated in AF are downstream of TBX5 based on our recently published single-cell multiome dataset of TBX5 knockout and control atria²⁴. Almost a third of GWAS-implicated genes (64, 29.6%) were significantly differentially expressed in TBX5 knockout aCMs. This striking number included seven of the ten genes that showed aCM-selectivity in both mice and humans. Amongst scored genes, the PIP score for TBX5 targets was significantly higher than non-targets (0.68 vs. 0.56). However, the proportion of genes that did and did not receive a score did not significantly differ between TBX5 targets and non-targets. A greater number of genes regulated by TBX5 were downregulated upon TBX5 inactivation (39 of 64, 61%) than upregulated (25 of 64, 39%), consistent with TBX5’s general function as a transcriptional activator. Overall, these data highlight the important role of TBX5 in regulating atrial gene expression and maintaining normal atrial rhythm.

Rare, large-effect AF variants

While GWAS studies focus on common genetic variants, often with small effect sizes, rare variants causing AF have been described, often in AF pedigrees. The genetics of familial AF were recently reviewed⁸⁰. Overall, familial AF studies have implicated 38 genes; 13 of these genes were also implicated by GWAS, whereas 25 were uniquely identified through pedigree studies (Tables 1 and 2)⁸⁰. In contrast to SNPs identified by GWAS, which are mostly noncoding and affect a variety of different gene categories (Fig. 2b), familial AF variants typically occur in protein-coding regions and predominantly affected genes encoding ion channels (Table 2). Many affected genes were selectively expressed in aCMs (nine genes) or vCMs (eight genes) in either mice or humans (Table 2), and chamber-selective genes were not enriched in specific functional categories (Fig. 2c). A smaller percentage of these genes (6, 16%) were downstream targets of TBX5²⁴ compared to the nearly 30% of genes identified by GWAS.

Mechanisms of AF pathophysiology have been intensely investigated, and yet remain incompletely understood^81,82,83. In properly functioning atria, aCMs contract coordinately and only in response to electrical impulses from the sinoatrial node. In AF, re-entrant circuits within the atria permit sustained and chaotic electrical activity. This requires a substrate vulnerable to re-entry plus frequent initiating signals. Factors that create a vulnerable substrate include heterogeneity of conduction velocity or repolarization⁸⁴; unidirectional conduction block; and reduction of minimum path length for re-entry⁸⁵, which is the product of effective refractory period and conduction velocity. Initiating signals include ectopic foci and excessive Ca²⁺ release leading to triggered activity⁸⁵. In this section, we link altered gene expression and AF genetic lesions to these fundamental fibrillogenic mechanisms.

Gene regulation

As previously discussed, a genetic program maintains aCM identity. While TBX5 has been the only TF reported to be critical for aCM identity maintenance²⁴ thus far, many other TFs are expressed in aCMs⁸⁶, and TFs were the most common gene type associated with AF GWAS loci (Table 1). Mutations in several TF genes result in autosomal dominant AF inheritance, including GATA4, GATA5, GATA6, ZFHX3, NKX2-5, NKX2-6, and PITX2c⁸⁰. TBX5 and SHOX2 mutations also drive familial AF^52,53,87.

Study of mouse models with perturbed expression of TF genes in aCMs enables identification of key aCM genes that participate in AF pathogenesis, since one can identify the genes directly regulated by mutated TFs. For example, Tbx5 inactivation in mice causes spontaneous AF and altered the expression of genes encoding ion channels and Ca²⁺ handling, creating both the trigger and substrate for AF^23,28. Tbx5 knockout aCMs had markedly prolonged action potentials and increased afterdepolarizations, which favor ectopic and triggered activity. These changes in action potential characteristics were linked to reduced expression of TBX5 target genes SERCA2A and RYR2, which govern Ca²⁺ handling, and multiple potassium channels, which promote repolarization. Tbx5 knockout aCMs also expressed less SCN5A and CX43, which would facilitate re-entry by reducing conduction velocity. Pitx2c inactivation made mice susceptible to pacing-induced AF by altering the expression of similar target genes⁸⁸. However, since PITX2c is a transcriptional repressor, its inactivation increased expression of many of the same genes, including SERCA2A, RYR2, and CX43^28,88. Indeed, heterozygosity for either Tbx5 or Pitx2c resulted in AF inducibility that was reduced in Tbx5 and Pitx2c double heterozygous mice²⁸. These results indicate that distinct and even opposite perturbations of atrial gene expression can both cause AF.

While mouse models lacking Pitx2c consistently demonstrated ECG changes resembling AF or higher AF vulnerability and were recently reviewed⁸¹, the mechanisms by which Pitx2c inactivation predisposes to AF remain uncertain due to inconsistencies between studies and between models. In humans, a damaging PITX2c variant was linked to lone AF in a pedigree⁸⁹. GWAS strongly linked a region near PITX2 to AF, yet whether this region acts by altering PITX2 protein levels in aCMs remains to be demonstrated⁹⁰. Recent work profiling human atrial development using single nucleus multiomics and spatial transcriptomics demonstrates that PITX2 is highly expressed in aCMs during embryonic development, and downregulated in mature aCMs, suggesting the strong GWAS signal associated with PITX2 might be associated with developmental abnormalities that enhance AF risk⁹¹.

Cellular protein levels do not always closely correlate with levels of the mRNA transcript that encodes them⁹², and post-translational regulatory mechanisms can further alter protein activity. Therefore, another critical question is how AF alters the aCM proteome. A recent study used mass spectrometry to compare a mouse model of physiological hypertrophy (IGF1R overexpression), which did not exhibit atrial dysfunction, to a model of pathological hypertrophy (overexpression of Mst1 and dominant negative PI3K), which developed dilated cardiomyopathy with fibrotic remodeling of the atria and impaired atrial contractility⁹³. Differential analysis between these models identified the downregulation of proteins associated with mitochondria, metabolism, and heart contraction in atria of the pathological hypertrophy model. Comparative analysis of the pathological model to proteins differentially identified in human AF showed conservation of changes in fatty acid metabolism, heart contraction, and mitochondrial organization.

A mass spectrometry analysis of atrial protein downregulated in a canine tachypacing model of AF similarly showed downregulation of proteins involved in contractility, with corresponding reduced force of contraction, decreased resting tension, and increased calcium sensitivity of AF aCMs⁹⁴. AF samples demonstrated increased degradation of myofilament proteins Titin, MYBPC3, and ACTN2 in a pattern consistent with calpain cleavage. In settings of HF, activation of the Ca²⁺-sensitive calpain protease system results in cleavage of sarcomeric proteins, some of which are known to play cardioprotective roles in the ventricle⁹⁵. Although protein levels of calpain and its endogenous inhibitor calpastatin were unaltered in the AF model, calpain activity was increased, possibly due to elevated intracellular Ca²⁺ in AF⁹⁴.

Together, these studies show that the aCM cell state is profoundly altered in AF at both the transcriptomic and proteomic levels. More work is needed to determine if these alterations are reversible, and how each of the many changes contributes toward the progressive nature of AF.

Electrical function

aCMs express ion channels that shape their action potential and mediate excitation–contraction coupling. GWAS and coding variants have implicated several potassium channel genes in AF pathogenesis. Potassium channels mediate cardiac repolarization. With a few exceptions, AF-associated potassium channel variants increase potassium currents and therefore shorten action potential duration and effective refractory period, which create a vulnerable substrate by allowing the shorter re-entrant circuits responsible for AF. For example, KCNJ5 interacts with KCNJ3 to form the acetylcholine-activated potassium channel (I_KACh), which is specifically expressed in aCMs and was shown to be constitutively active in patients with chronic AF⁹⁶. A KCNJ3-N83H missense variant that increased I_KACh caused familial bradycardia and AF⁹⁷. Zebrafish expressing KCNJ3-N83H exhibited bradycardia that was rescued by the addition of a selective K⁺ channel inhibitor, NIP-151. In mice, pharmacological stimulation of I_KACh enhanced AF inducibility, whereas ablation of KCNJ5 prevented AF⁹⁸. A strong GWAS signal was also observed near KCNJ5, suggesting that common variants controlling its expression influence AF susceptibility^67,68,69. A pedigree with a missense variant in KCNQ1, which encodes the pore-forming subunit of the potassium channel I_Ks⁹⁹, likewise implicated this channel in AF. In this four-generation pedigree, a KCNQ1-S140G missense variant that increased channel activity caused autosomal dominant AF¹⁰⁰. In addition to these two examples, coding variants have been described as AF drivers for the potassium channels genes KCNE1, KCNE2, KCNE3, KCNE5, KCNJ8, KCNA5, and KCND3^{101,102,103,104,105,106,107}, but the causative role of these variants was less well substantiated due to limitations such as small family sizes or low numbers of affected patients.

Altered expression or activity of other types of channels are also implicated in AF. Variants in both the pore-forming alpha subunit and regulatory beta subunits of the major cardiac sodium channel, responsible for the depolarizing current I_Na, have also been associated with AF^{108,109,110,111,112}. SCN5A and SCN4B have been most directly implicated. One study¹⁰⁸ directly sequenced the coding region and splice junctions of the entire SCN5A gene in 375 patients with AF (118 patients with lone AF), resulting in the identification of 8 novel variants in 10 probands¹¹³. The identified AF-associated variants were in highly conserved residues and altered SCN5A activity. Six probands reported AF in family members, and pedigree analysis indicated that these variants co-segregated with AF with a dominant inheritance pattern in all six families. A family that exhibited sick sinus syndrome, atrial flutter, and AF was found to have a truncating SCN5A variant, R1860Gfs*12, which resulted in a 70% decrease in channel current density when expressed in HEK293 cells¹¹⁴. Loss of I_Na is predicted to slow atrial conduction velocity, reducing the path length required for re-entry and creating a favorable environment for multiple-circuit re-entry¹¹⁵. These data are consistent with AF GWAS studies, which implicated a locus containing SCN5A and the neighboring gene SCN10A. The implicated region functioned as an enhancer element that was activated by TBX5, and the SNP reduced the enhancer’s activity¹¹⁶.

Two out of the three small-conductance Ca²⁺-activated K⁺ (SK) channels, encoded by KCNN2 and KCNN3, were implicated in AF by human genetic studies^61,65 and have been shown to have enhanced channel function in humans with chronic AF¹¹⁷. Although protein levels of the channels were not different in RA cardiomyocytes between control and chronic AF patients, SK2 membrane localization was enhanced in AF patients and channel activity was increased due to decreased calmodulin phosphorylation mediated by PP2A. Recently, an SK2 channel inhibitor AP30663 met the primary endpoint of cardioversion occurring within 1 h of administration in a phase 2 clinical trial, without serious treatment-related adverse events¹¹⁷. These exciting results point to the potential efficacy of SK channel inhibition as a novel antiarrhythmic approach for AF.

Together, these data show that proper control of electrical signaling in aCMs is functionally required for rhythm maintenance. Altered ion channel functionality affects atrial conduction velocity and the effective refractory period of aCMs.

Communication between neighboring aCMs

An important ultrastructural feature of aCMs, as well as vCMs, is the intercalated disc (ICD), which is located on the ends of CMs at their junction with neighboring CMs. The ICD contains desmosome and cadherin complexes that physically couple neighboring CMs, and gap junction channels that support the propagation of action potentials between neighboring CMs¹¹⁸. There are two major gap junction proteins expressed in the aCMs: GJA1, which is expressed in both aCMs and vCMs, and GJA5, which is aCM-selective¹¹⁹. GJA5 neighbors a GWAS locus associated with AF risk⁶⁷. Loss-of-function somatic mutations in both GJA1 and GJA5 have further implicated these genes in AF^119,120. For example, genomic DNA sequencing of atrial tissue and peripheral blood lymphocytes of 15 AF patients revealed GJA5 variants in the heart tissue of 4 patients. The variants were absent in the blood of three of the four patients, consistent with somatic mutation. Interestingly, both GJA1 and GJA5 are downregulated in mice with cardiomyocyte inactivation of Tbx5 or Lkb1, two genetic mouse models with spontaneous AF^{24,28,121,122}. These data underscore the importance of these gap junctions for supporting normal atrial rhythm, probably through supporting inter-aCM electrical coupling. Disrupting aCM electrical coupling could predispose to re-entry by reducing conduction velocity and increasing conduction velocity heterogeneity within the atrial myocardium.

Excitation–contraction coupling

Coordinated atrial contractions promote the flow of blood through the atria and into the ventricles. Fibrillating atria fail to contract uniformly, and individual aCMs have decreased contractile force, resulting in blood pooling and increased risk of thrombosis and stroke. During excitation–contraction coupling, sarcolemmal depolarization opens voltage-gated L-type calcium channels. The influx of Ca²⁺ signals to the ryanodine receptor, RYR2, to release Ca²⁺ stored in the sarcoplasmic reticulum, producing a Ca²⁺ transient that interacts with sarcomeric proteins and stimulates contraction. During diastole, SERCA2A pumps cytoplasmic Ca²⁺ back into the sarcoplasmic reticulum, and NCX1, a sodium–calcium exchanger on the sarcolemma, returns Ca²⁺ to the extracellular space. Many proteins involved in this intricate process are genetically implicated in AF, including RYR2¹²³, JPH2¹²⁴, PLN⁶⁵, CASQ2⁶⁷, and CAMK2D⁶⁷ (Tables 1 and 2).

Mutations in RYR2 are most commonly associated with CPVT. In addition to hallmark exercise-induced ventricular tachycardia, some patients have atrial arrhythmias including AF¹²⁵. This is modeled in mice with CPVT-causing RYR2 variants, which have increased AF vulnerability^126,127. These experimental and clinical observations clearly implicate Ca²⁺ handling as an important mechanism in AF. Leaky RYR2 channels result in increased diastolic Ca²⁺ leak¹²⁸, which activates Ca²⁺-calmodulin dependent kinase 2 (CAMK2), a key mediator of the fight-or-flight response whose chronic activation contributes to pathological cardiac remodeling and arrhythmias¹²⁹. Activated CAMK2 phosphorylates several Ca²⁺ handling proteins including RYR2, creating a positive feedback loop that increases aberrant RYR2 Ca²⁺ release. Elevated diastolic Ca²⁺ promotes electrogenic Na⁺–Ca²⁺ exchange by NCX1, leading to delayed afterdepolarizations and triggered activity¹³⁰. These afterdepolarizations also cause dispersion of repolarization and refractoriness, increasing susceptibility to re-entry¹³¹. The importance of CAMK2-mediated phosphorylation of RYR2 is highlighted by studies of mice in which the CAMK2 phosphorylation site on RYR2 (serine 2814) has been ablated. These RYR2-S2814A mice were protected from AF triggered by genetic or environmental insults^132,133. In addition to RYR2, CAMK2 also phosphorylates PLN to relieve PLN-mediated SERCA2 inhibition¹³⁴, resulting in increased SR calcium loading that is further enhanced by the aCM depolarization frequency in AF¹³⁵. HF patients had reduced RYR2 protein in atrial myocardium, and relative Ser2814-RyR2 phosphorylation was greater in the subset with chronic AF¹³⁶. Increased CAMK2-mediated RYR2 phosphorylation was also shown to occur in patients experiencing AF following open heart surgery, a complication mediated by inflammatory signaling (see below) that affects ~30% of adult cardiac surgery patients¹³⁷. Together, these studies indicate that excessive diastolic RYR2 activity creates favorable conditions for AF¹³⁸.

AF was originally described as predominantly a disease affecting the electrical function of the heart. Yet family studies and GWAS both link structural genes to AF, including the alpha and beta myosin heavy chain isoforms MYH6¹³⁹ and MYH7¹⁴⁰, and TTN, which connects the Z-disk to the M-line in sarcomeres and functions as a molecular spring to imbue the sarcomere with elasticity^67,68,69. Mutations in these genes commonly result in inherited cardiomyopathy^141,142. The clinical connection between AF and cardiomyopathy is well appreciated¹⁴³, and recent experimental evidence has begun to elucidate the molecular mechanisms linking the conditions at the transcriptional level¹⁴⁴. Intriguingly, mutations in genes commonly associated with HF have been linked with early onset, lone AF. A recent study¹⁴⁵ examining a cohort of 1293 patients with lone AF diagnosed prior to age 66 found that pathogenic variants in genes associated with inherited cardiomyopathy syndromes were more common than variants in genes associated with inherited arrhythmias and were observed at greater frequencies in younger lone AF patients. Other recent targeted approaches have shown that loss-of-function mutations in TTN are highly prevalent in individuals with early AF diagnoses^146,147,148. One study¹⁴⁶ performed whole exome sequencing in families with at least three individuals affected with AF and observed an enrichment for TTN truncating mutations (P = 1.76 × 10⁻⁶). This finding was replicated in an additional cohort of 399 lone AF patients (odds ratio = 36.8; P = 4.13 × 10⁻⁶). The causal link between titin truncating variants and AF was further supported by the study of heterozygous ttn N-terminal truncation in zebrafish, which caused sarcomeric defects, irregularly irregular heart rhythm, and increased atrial fibrosis.

These data support an evolving view of AF as the result of atrial myopathy as well as electrical dysfunction.

Inflammatory signaling

Recent work has demonstrated inflammatory signaling as an important mechanism promoting AF¹⁴⁹. At least two different inflammatory mechanisms have been identified, including activation of the NLRP3 (NACHt-, LRR-, and pyrin domain-containing 3) inflammasome within aCMs¹⁵⁰, and the infiltration of CCR2⁺ proinflammatory macrophages into diseased atrial myocardium, which promotes pathological atrial remodeling¹⁵¹.

The NLRP3 inflammasome controls the production of inflammatory cytokines within cells through the cleavage of pro-IL-1β and pro-IL-18, which are released into the extracellular environment through membrane pores formed by the N terminus of gasdermin D (GSDMD-NT), which is also produced by inflammasome cleavage¹⁵². Recent work¹⁵⁰ examining protein levels of NLRP3, the inflammasome component ASC, and the active form of its effector protein pro-caspase1, revealed their upregulation in aCMs of chronic AF patients. Furthermore, overexpression of constitutively activated NLRP3 in cardiomyocytes increased AF vulnerability. The enhanced vulnerability was associated with an abbreviated atrial effective refractory period and fibrotic atrial remodeling. This mechanism was also shown to participate in post-operative AF by promoting CAMK2-mediated RYR2 phosphorylation of Ser2814¹³⁷, as mentioned previously.

Inflammatory signaling in the malfunctioning atria results in the recruitment of CCR2⁺ SPP1⁺ macrophages in human AF patients¹⁵¹. Expansion of this proinflammatory macrophage population was similarly observed in a mouse model of enhanced AF vulnerability produced by combining hypertension, obesity and mitral valve regurgitation (HOMER mice), risk factors driving AF in humans. SPP1, also known as osteopontin, is a matricellular protein that stabilizes collagen¹⁵³ and was shown to promote fibrosis in hypertensive mice¹⁵⁴. Transplanting Spp1^−/− bone marrow into wild-type mice that were then subjected to HOMER reduced AF inducibility and fibrotic remodeling. Examining crosstalk between SPP1⁺ macrophages and other cell types using single-cell transcriptomic data revealed signaling between these macrophages and other atrial immune cells and stromal cells expressing SPP1 receptors. In a subsequent study, the authors used an antibody-siRNA conjugate to silence Spp1 in cardiac macrophages, leading to reduced AF inducibility and atrial fibrosis¹⁵⁵.

These data reveal that inflammatory signaling both alters the electrical properties of aCMs and promotes remodeling of the atrial substrate to support AF.

Although developmental processes driving chamber-selective gene expression programs have been appreciated for a number of years, advances in functional genomics have only recently begun to illuminate the molecular mechanisms by which aCMs and vCMs maintain their differences, revealing ESRR and TBX5 as critical network components that facilitate maintenance of vCM and aCM identity, respectively. aCMs and vCMs represent the two major cardiomyocyte subtypes, but even within the same chamber there are multiple cardiomyocyte states with non-identical gene expression profiles¹⁵⁶. More work will be required to decipher the functional importance of these cardiomyocyte states in cardiac homeostasis and disease.

Large GWAS studies have identified 150 genomic loci that are significantly associated with AF, and continued expansion of these studies to better represent diverse ethnic groups, will likely further increase this number. These GWAS variants are overwhelmingly in non-coding regions that likely function as CREs. Linking these non-coding variants to specific genes is currently an active area of research. About 20% of candidate genes neighboring these GWAS loci are selectively expressed at higher levels in aCMs than vCMs, and the expression of about 30% is regulated by TBX5, a master regulator of aCM identity. We combined this information with scoring systems aimed at identifying causative AF genes to create an encyclopedia of genes with probable roles in AF (Tables 1 and 2). Additional work examining the function of these genes in animal and human PSC-derived models will undoubtedly enhance our understanding of the molecular mechanisms underlying AF.

AF results from genetic lesions that directly affect the electrical and Ca²⁺ handling properties of aCMs. In addition, genetic variants in cardiomyocyte structural components also cause AF. Whether these classes of perturbations cause AF through distinct mechanisms or through a common shared AF pathway is an important unresolved question. If there are distinct mechanisms that lead to AF, how can the primary mechanisms be discerned in individual patients to enable precise, targeted therapy? These types of questions will drive future efforts to understand AF pathophysiology and ultimately improve our ability to control AF in the clinic.