Genetics of Atrial Fibrillation in 2020 – GWAS, genome sequencing, polygenic risk and beyond

Abstract

Atrial fibrillation (AF) is a common heart rhythm disorder that leads to an increased risk for stroke and heart failure. AF is a complex disease with both environmental and genetic risk factors that contribute to the arrhythmia. Over the last decade, rapid progress has been made in identifying the genetic basis for this common condition. In this review we provide an overview of the primary types of genetic analyses performed for AF including linkage studies, genome-wide association studies (GWAS) and studies of rare coding variation. With these results in mind, we aim to highlighting the existing knowledge gaps and future directions for AF genetics research.

Introduction and Background

Atrial fibrillation (AF) is a common heart rhythm disorder with an estimated 33 million people affected worldwide[1]. Reported risk factors for AF include advancing age, obesity, hypertension, diabetes, and cardiovascular diseases[2]. Studies have also shown an increased risk for men to develop AF, compared to women[3]. As discussed in the accompanying article in this Compendium (formal reference provided by the journal), AF can lead to a number of serious medical consequences including stroke, heart failure, cognitive impairment, and increased mortality.

The treatment of AF remains challenging. While there are effective medications for anticoagulation to reduce the risk of stroke, antiarrhythmic medications are limited by lack of efficacy to reduce symptoms and have potential side effects. Alternatively, catheter ablation procedures can be effective in reducing the burden of AF, but these procedures are invasive, can be associated with complications and may require a repeat procedure for the long-term management of AF. Thus, there is a pressing need to develop new therapies for AF.

Similar to other common cardiovascular diseases such as hypertension and myocardial infarction, AF is a complex disease with shared environmental and genetic factors that contribute to disease pathogenesis. Over the last decade, multiple studies have observed familial aggregation of individuals with lone AF[4]. Similarly, the heritability of AF has been elegantly demonstrated in the Icelandic population[5]. Based on a study on monozygotic twins the heritability of AF has been estimated as high as 62%, indicating a strong genetic component[6]. In aggregate these studies have consistently observed an increased risk of AF particularly when a first-degree family member is affected, and among individuals with early-onset forms of the arrhythmia[7]. In more recent work, Weng and colleagues used genetic data to estimate that the heritability of AF based on common genetic variants in individuals of European ancestry is approximately 22%[8].

In the current review, we will provide an overview of multiple approaches used to examine the genetic basis of AF. We will present the most relevant results from these analyses and discuss emerging technological advances that could be leveraged to expand our understanding of the field. We discuss three broad genetic approaches applied to AF including: 1) linkage analysis using families with Mendelian forms of AF, 2) genome-wide association studies (GWAS) studies using genotyping array data, and 3) coding variation from genome sequence data (Figure 1). These approaches are not mutually exclusive but are helpful as a framework to consider when reviewing the genetic studies of AF published to date. We will subsequently describe the application of GWAS data to clinical risk prediction. Finally, we will discuss the knowledge gaps in the field of AF genetics and describe emerging technologies that may shape the future of the field.

Figure 1. Three primary types of genetic analyses for AF.

Linkage analyses primarily focus on large families with hereditary forms of AF. The disease associated linked regions can include multiple candidate genes one of which will contain a disease causing mutation. GWAS analyses are based on genotype array data that consists largely of non-coding variants that are presumed to regulate genes in the region or locus. Analyses of coding variation are derived from whole-exome or whole-genome sequencing data. Rare coding or loss-of-function variants are grouped and jointly tested in AF cases versus controls to identify specific disease-causing genes. Please note that these approaches are not mutually exclusive and are often combined depending upon the study design.

Abbreviations: AF, atrial fibrillation; GWAS, genome-wide association study; LOF, loss-of-function; SNP, single-nucleotide polymorphism; STR, short tandem repeat.

Three Broad Genetic Approaches Applied to AF

Familial AF and Linkage Analysis

Linkage analysis is typically performed in families with many affected individuals and a clear hereditary pattern. The analytic approach leverages genetic linkage or the tendency for a genetic marker near a disease-causing gene to be inherited together. The earliest application of this approach used a few hundred highly informative markers scattered throughout the genome. By matching transmission of the genetic marker with disease status in the family, a disease-causing region or locus can be linked to a given marker. In turn, the genes in this region can be sequenced to identify mutations that associate with disease. This approach has been elegantly used to identify many of the early genes implicated in hypertrophic cardiomyopathy and long QT syndrome. As summarized in Figure 2, a number of causative mutations for AF have been identified in large families or populations. Specific examples include the ion channel KCNQ1, the cardiac peptide NPPA, the transcription factor TBX5, and a motor protein MYL4.

Figure 2. Major AF-associated genes and lines of evidence.

The figure illustrates AF-associated genes that were discovered through family based or gene-based studies. For each gene the lines of evidence are listed. The table includes the evidence from familial AF genetic analysis, whether coding variants in the gene are associated with AF, if the gene lies within an AF GWAS locus, whether loss-of-function variation is associated with AF, and functional evidence that has been reported for the gene in the context of AF.

Abbreviations: AF, atrial fibrillation; GWAS, genome-wide association study.

The first mutation linked to familial AF was found in the ion channel KCNQ1, a gene that encodes the alpha subunit of the I_Ks current. The AF related mutations result in a gain of channel function and likely shortening of the atrial refractory period, a finding that would make it easier for reentry to continue and increase the susceptibility to AF[9]. In a distinct family, a mutation was identified in NPPA, the gene encoding the atrial natriuretic peptide (ANP), a protein that is highly expressed in the heart[10]. A frame-shift mutation was found to remove a stop codon and lead to an extended mutant protein that had appeared to be protected from degradation resulting in greater circulating levels and increased activity[11]. In vivo experiments from the same study have shown that mutations in NPPA lead to a shortening of the monophasic action potential duration as well as the effective refractory period. These changes on atrial electrophysiology could increase susceptibility to AF.

A gain-of-function mutation in the transcription factor, TBX5, was associated with familial AF in the setting of Holt-Oram syndrome[12], a developmental disorder that leads to heart and limb malformations[13]. In vitro studies demonstrated that the mutated TBX5 had enhanced binding to DNA and could lead to up-regulation of downstream targets such as NPPA and CX40[12,14]. In a large-scale study of Icelanders, autosomal recessive mutations in MYL4 were identified among individuals with early-onset AF[15]. This atrial-specific myosin light chain was found to have a mutation in an F-actin binding region, and modeling the mutation in zebrafish resulted in disruption of the sarcomere and atrial enlargement.

Thus, individual families with hereditary forms of AF can be incredibly helpful in informing disease biology. However, the mutations identified to date are by nature rare and therefore have a small impact on the overall scope of this common arrhythmia.

Genome-Wide Association Studies for AF

In contrast to studies in families, GWAS permit the analysis of entire populations by comparing individuals with and without AF at a large scale. Each individual is genetically “fingerprinted” using a low cost, high throughput genotyping array. These arrays are used to determine the status of hundreds of thousands of genetic variants or single nucleotide polymorphisms (SNPs) throughout the genome. While individual SNPs contain relatively little information, in aggregate, the use of hundreds of thousands of markers can capture the majority of the genetic diversity between individuals. In order to compare the data from one genotyping platform to another, the data is imputed or harmonized to a common reference panel consisting of millions of genetic markers[16]. Comparisons of all the genetic markers are then made between cases and controls to identify regions associated with disease (Figure 1, middle). Importantly, unlike the analysis in families in which a single causative mutation is identified, in GWAS a region or locus is linked to disease. This region may or may not contain any genes. Many causative variants identified by GWAS are in non-coding regions of the genome and have an effect on the regulation of a nearby gene. In recent years, the cost of genotyping arrays has fallen to less than $50 per individual so it is now feasible undertake large-scale studies of common diseases and studies with more than 100,000 cases are increasingly common. Large sample sizes help to both ensure the validity of the results and to define the full extent of the genetic basis of the disease.

The first GWAS for AF was reported in 2007 and remarkably started with only a few hundred AF cases in the initial discovery[17]. The small number of individuals needed in this first report was a reflection of the strength of the genetic association which is considerably greater than that observed for most other genome wide studies. Carrying a single variant at this 4q25 locus near the gene PITX2 conferred over a 60% increased risk of disease in the general population and an even greater risk in younger individuals. As with many disease-associated regions identified by GWAS, the association with AF at the PITX2 locus is in a non-coding region of the genome. As reviewed in other manuscripts in this issue (formal reference to be provided by the journal), the AF risk variants modulate the expression of PITX2 and loss of this region or the PITX2 gene itself can results in AF[18,19].

In the ensuing years, many other GWAS for AF have been completed in predominantly European ancestry participants[20–24]. Initially novel loci were identified at 16q22 close to the gene ZFHX3 and 1q21 close to the KCNN3[25–27]. In two recent studies that both included over 60,000 cases[28,29] nearly 140 AF loci have been identified to date (Table 1). While rapid progress has been made in AF genetics, it is important to note that the vast majority of participants are of European descent (Figure 3). The two largest non-European GWAS were published in 2017 and report results from Korean[30] and Japanese[31] cohorts. At least three variants for AF show significant heterogeneity across different ancestries (rs2129977 at PITX2, rs11598047 at NEURL and rs2359171 at ZFHX3). Additionally, while a signal at the PITX2 locus can be found across ancestries, the top variant is not always the same nor strongly correlated[28]. Two AF loci from a GWAS in Japanese ancestry, at the genes NEBL and SH3PXD2A, do not replicate in European ancestry and may be ancestry specific[31].”

Table 1.

To date there are at least 138 AF loci identified in single variant testing with P-value < 5 × 10⁻⁸. The table below includes the most significant variant at each locus from the latest AF GWAS publications by Roselli et al.[28] or Nielsen et al.[29]. The listed genes are either the nearest gene(s) or the eGene. The eGene is defined as a gene with an eQTL to the variant at an AF GWAS locus and is highlighted in bold. The eGene is listed when it is the only eGene at that locus. Notably, only 18 out of 138 variants are associated with only one eGene.

Rsid	Nearest Gene(s) or eGene	Rsid	Nearest Gene(s) or eGene	Rsid	Nearest Gene(s) or eGene
rs187585530	UBE4B	rs716845	KCNN2	rs1822273	NAV2
rs880315	CASZ1	rs2012809	FBN2, SLC27A6	rs949078	SORL1, MIR100HG
rs7529220	HSPG2, CELA3B	rs34750263	WNT8A, NME5	rs76097649	KCNJ5
rs2885697	SCMH1	rs174048	ARHGAP26, NR3C1	rs6490029	CUX2
rs11590635	AGBL4	rs12188351	SLIT3	rs10842383	LINC00477, BCAT1
rs56202902	FAF1	rs6882776	NKX2–5	rs113819537	SSPN
rs146518726	C1orf185	rs73366713	ATXN1	rs12809354	PKP2
rs12044963	KCND3	rs34969716	KDM1B	rs7978685	NACA
rs4484922	CASQ2	rs1307274	C6orf1, NUDT3	rs35349325	BEST3
rs79187193	GJA5	rs3176326	CDKN1A	rs11180703	KRR1, PHLDA1
rs11264280	KCNN3, PMVK	rs6907805	CGA, ZNF292	rs883079	TBX5
rs72700114	METTL11B, LINC01142	rs210632	GOPC	rs12810346	TBX5-AS1, TBX3
rs608930	GORAB, PRRX1	rs17079881	SLC35F1	rs10773657	HIP1R
rs10753933	PPFIA4	rs13191450	GJA1, HSF2	rs12298484	DNAH10
rs4951261	NUCKS1	rs12208899	LINC00326, EYA4	rs6560886	FBRSL1
rs6546620	KIF3C	rs117984853	UST	rs9580438	LINC00540, BASP1P1
rs6742276	XPO1	rs11768850	SUN1	rs35569628	CUL4A
rs2540949	CEP68	rs55734480	DGKB	rs28631169	MYH7
rs10165883	SNRNP27	rs6462078	CREB5	rs2145587	AKAP6
rs72926475	REEP1, KDM3A	rs74910854	PMS2P2	rs73241997	SNX6,CFL2
rs28387148	GYPC	rs11773884	CDK6	rs2738413	SYNE2
rs67969609	TEX41	rs62483627	COG5	rs74884082	DPF3
rs12992412	MBD5	rs11773845	CAV1	rs10873299	LRRC74, IRF2BPL
rs56181519	WIPF1	rs55985730	OPN1SW	rs147301839	MYZAP
rs35504893	FKBP7	rs7789146	KCNH2	rs62011291	USP3
rs295114	SPATS2L	rs35620480	LINC00208, GATA4	rs12591736	TLE3, UACA
rs35544454	ERBB4	rs7508	ASAH1	rs74022964	HCN4, REC114
rs6810325	MKRN2	rs7846485	XPO7	rs12908004	LINC00927, ARNT2
rs73032363	THRB	rs62521286	FBXO32	rs12908437	IGF1R
rs6790396	SCN10A	rs35006907	MTSS1, LINC00964	rs2286466	RPL3L
rs2306272	SLC25A26	rs7460121	MIR30B	rs2359171	ZFHX3
rs17005647	FRMD4B	rs6993266	PTK2	rs7225165	YWHAE, CRK
rs7632427	EPHA3	rs4977397	SLC24A2, MLLT3	rs8073937	POLR2A, TNFSF12
rs17490701	PHLDB2	rs4385527	C9orf3	rs72811294	MYOCD
rs1278493	PPP2R3A	rs4743034	ZNF462	rs11658278	ZPBP2
rs4855075	GNB4	rs10760361	PSMB7	rs242557	MAPT
rs60902112	XXYLT1	rs2274115	LHX3	rs76774446	GOSR2
rs9872035	PAK2	rs2296610	NEBL	rs7219869	KCNJ2, CASC17
rs3822259	WDR1	rs7919685	NRBF2	rs12604076	CYTH1
rs1458038	PRDM8, FGF5	rs7096385	SIRT1	rs9953366	SMAD7
rs3960788	UBE2D3	rs60212594	SYNPO2L	rs8088085	MEX3C
rs2129977	PITX2, C4orf32	rs11001667	C10orf11	rs2145274	CASC20, BMP2
rs55754224	CAMK2D	rs1044258	C10orf76	rs7269123	C20orf166
rs10213171	ARHGAP10	rs11598047	NEURL	rs2834618	LOC100506385
rs10520260	HAND2-AS1	rs2047036	SH3PXD2A	rs465276	TUBA8
rs6596717	LOC102467213, EFNA5	rs10749053	RBM20	rs133902	MYO18B

Figure 3. Ancestry of the cases in genome-wide association studies for AF.

European ancestry sample is plotted towards the left in white and non-European ancestry is plotted towards the right highlighted in different colors. Plotted is the number of cases included in each published AF GWAS study or meta-analysis. 10 out of 12 studies include predominantly European ancestry samples, shown in white. Two studies are Japanese only and Korean only. Within the multi-ancestry meta-analyses Roselli et al.[28] included the largest proportion of non-European cases including Japanese, Brazilian, African American and Hispanic samples.

Abbreviations: AF, atrial fibrillation, AFR, African-American, BRA, Brazilian, EUR, European, HISP, Hispanic, JAP, Japanese, KOR, Korean.

Identifying genes at AF loci

Genetic variants identified in GWAS studies are largely located in non-coding regions of the genome. These non-coding variants are presumed to alter the activity of a transcriptional regulatory element such as an enhancer or repressor that in turn results in modifying the transcription of a nearby gene. Importantly, for most GWAS variants there is usually no straightforward path from an association by GWAS to a gene and a disease mechanism. A recent study used STARR-seq to identify regulatory elements and their target genes at multiple GWAS loci for AF. Further, they found that the loss of a regulatory element at the HCN4 locus led to reduced gene expression[32]. A myriad of analyses can follow a large scale GWAS ranging from computational analyses to the derivation of polygenic risk scores for AF risk prediction and the incorporation of GWAS data into other analyses (Table 2). In the subsequent sections we will touch on a few helpful approaches with respect to AF, but for other potential directions please also see these recent reviews[33–35]. One straightforward application of GWAS data is to perform a pathway analysis to evaluate the collective effect of the genetic association on different biological functions. Globally the AF associated genes represent distinct functional groups including those underlying cardiac development, cellular electrophysiology, cardiomyocyte contractility and structure[24,28,29]. A similar approach was taken in a GWAS of the Japanese population and implicated suppression of the mTOR signaling pathway in AF[36].

Table 2.

Overview of studies that integrated AF GWAS data with other clinical, epigenetic or genetic datasets.

Study type	Description and reference
Mendelian Randomization	Obesity [72]
	Thyroid Function [73]
	Serum albumin [74]
	Body mass index/body composition and cardiovascular conditions including AF [75]
	Body composition [76]
Methylation	Genome-wide DNA methylation analysis [77]
	Methylome-wide Association Study [78]
Heritability	Heritability of AF [8]
Ancestry specific analyses	European ancestry as a risk factor for AF [79]
AF recurrence after ablation	Common variants predict AF recurrence [80]
Genetic Interaction	Gene-gene interactions [81]
	Genetic interactions with risk factors [82]
Associations with Mitochondrial DNA	Mitochondrial DNA and Incident AF [83]

As noted above, a major challenge with GWAS is that the analyses usually identify a region of interest rather than a specific causative gene. Bridging the gap from variants to genes remains a major challenge in disease genetics, particularly for complex polygenic traits such as AF. One common and helpful approach is to use expression quantitative trait loci (eQTL) mapping. An eQTL analysis links the genotype of a SNP at an AF locus to the expression of genes in the region. If an AF associated SNP is strongly linked to the expression of a single gene, that gene is likely to be the causative gene at the locus. While this approach is simple in theory, practically there are two primary limitations. First, eQTL analyses are often tissue specific. Although there are terrific publicly available resources such as GTEx or the Genotype-Tissue Expression project (https://gtexportal.org) that includes gene expression profiles from many tissues, unfortunately the cardiac analyses were limited to the left ventricle and right atrial appendage[10]. Recent work in the Cleveland Clinic Atrial Tissue Bank[22,37] and the Myocardial Applied Genomics Network[28] have addressed this limitation by investigating the gene expression profiles of left atrial tissue. A second limitation of using eQTLs is that it requires large sample sizes and this is not always possible when tissues are hard to obtain such as the left atrium or the pulmonary veins in the case of AF. Finally, while eQTLs are powerful, they only explain a fraction of the disease loci. For example, in the latest two GWAS of AF only 13% of the variants at AF loci could be linked to the expression of a single gene, and at 22% of AF loci the variants linked to one or more genes. These results are summarized in Table 1.

A complimentary approach to eQTL analyses is to use the three-dimensional architecture of the genome to identify causative AF genes. Since many AF SNPs are in non-coding regions of the genome, they are presumed to alter regulatory elements such as enhancers or repressors that in turn bind to the promoter of a nearby gene to regulate its expression. The contact points between AF associated regulatory elements and gene promoters can be assessed genome-wide using chromosome conformation capture technologies such as HiC. Similar to eQTL analyses, enhancer-promoter contacts can be tissue specific and a recent HiC study from the human heart roughly doubled the number of AF associated genes derived from GWAS data[38]. Other analyses of AF GWAS data such as the incorporation of multi-omic datasets[39] or epigenetic analyses including STARR-seq[40] are described in detail in the accompanying review by XX and colleagues (formal reference to be provided by the journal). Ultimately, though any gene implicated by these methods will require further validation in vitro and in vivo.

Assessing polygenic risk from GWAS

Since we now have very dense GWAS datasets for AF, it is natural to wonder whether this data could be used to in a clinical setting to identify high risk individuals, stratify screening efforts or look for differential treatment outcomes. The polygenic nature of AF as captured in GWAS can be transformed into a genetic risk score for each individual. An AF polygenic risk score (PRS) summarizes the cumulative genetic risk and can be computed using anywhere from just a few variants at the top loci or the data from millions of SNPs across the entire genome (Figure 4).

Figure 4. Overview of polygenic risk scores (PRS) for AF.

A polygenic risk score is calculated for each individual as a sum of the product of genetic dosage and a weight. The weights are derived from the effect estimates of a genome-wide association study. The PRS of individuals in a population follows a Gaussian distribution. Individuals in the highest percentile of the distributions show an increased risk for AF versus the remaining population. Potential applications of an AF PRS can include improving risk prediction, prioritizing high risk individuals for screening, and examining differential outcomes of AF.

Abbreviations: AF, atrial fibrillation, PRS, polygenic risk score.

An initial AF PRS scores used the top 12 genome-wide significant genetic variants[41], while more contemporary iterations that incorporate over 6.6 million variants[42]. In the later approach, the top 1.5% individuals with a high PRS had a more than a 4-fold increased risk for AF. The variants and weights included in the latest AF PRS are publicly available (http://broadcvdi.org). In one interesting application of an AF PRS, Weng and colleagues found that individuals in the highest tertile of polygenic risk had a higher lifetime risk for AF (47%) compared to the individuals in the lowest tertile (26%)[43]. The combination of clinical and genetic data permitted additional refinement of risk. For example, individuals with a low clinical risk but a high genetic risk had an overall lifetime risk for AF of 44%. A complementary review in this Compendium focuses on the potential clinical applications of AF genetics and polygenic risk scores (formal reference to be provided by the journal).

Exome and genome sequencing to identify AF genes

While GWAS has been deployed at great scale, a major limitation of the method is that it is only able to capture known variants in the genome. Yet many diseases arise from unique mutations in the coding region of genes that arise spontaneously in an individual of family. While these mutations can be identified using a targeted approach of a single or small number of genes, the availability of genome wide approaches has largely replaced these earlier studies. Currently, it is possible to sequence the entire protein-coding region of the genome for about $200 or the entire genome for less than $800, and prices are continuing to fall. These technological advancements and price reductions have started to enable the application of large scale sequencing studies for AF.

Once a population of cases and controls has been sequenced, common genetic variants are analyzed using the approach previously described for GWAS. In contrast, rare coding genetic markers, often defined as present in less than 1% of the population, are analyzed in gene-based tests. For a gene-based test, coding variants are analyzed jointly across a gene unit for an association with disease, as depicted in Figure 1. The most commonly used approach restricts the analysis to variants predicted to lead to a loss-of-function of the encoded protein. A considerable advantage of testing rare loss-of-function variants over GWAS is establishing a direct link from gene function to disease. In addition, this analysis provides a clear direction of effect between loss-of-function of the encoded protein and disease. While many sequencing analyses focus on loss-of-function variants, it is important to note that gain-of-function mutations such as those identified in families with TBX5 or KCNQ1 would not be identified with this approach.

In 2017 Thorolfsdottir and colleagues analyzed whole-genome sequencing data from the Icelandic population including 14,255 AF cases and 374,939 controls[44]. The Icelandic population provides a unique resource for genetic discovery, because it is a relatively homogeneous and genetically isolated population that is enriched with rare loss-of-function coding variants. The study identified a low-frequency missense mutation in the gene PLEC that encodes the cytoskeletal protein plectin. The missense mutation in PLEC is associated with increased risk for AF. Additionally, a missense mutation in the myosin gene MYH6 was significantly associated with AF, a gene that has previously been associated with sick sinus syndrome[45].

Within the last two years, at least four studies have identified loss-of-function mutations in TTN among individuals with AF. Ahlberg and colleagues found an enrichment of loss-of-function mutations in TTN among families and individuals in Denmark with early-onset AF[46]. Shortly thereafter, Choi and colleagues observed a similar finding among unrelated individuals with early-onset AF. Using nearly 2,800 AF cases and 5,000 controls, they found that TTN loss of function mutations were present in ~2% of individuals with an onset of AF before 65 years of age[47]. With younger ages of onset, the frequency of TTN mutations rose to a high of over 7% of individuals younger than 30 years of age. As with dilated cardiomyopathy, the association between TTN and AF had a stronger effect when the analysis was restricted to TTN exons that were highly expressed in cardiac tissue.

While these studies pointed to an increased frequency of TTN mutations in individuals with early-onset AF, the question then arose whether TTN mutations were detectable in the general population with AF. To address this, Choi and colleagues used exome sequencing data in 1,400 AF cases and more than 40,000 controls from the UK Biobank. They observed a similar strong association between loss-of-function variation in TTN and AF[48]. Further, the penetrance of AF among TTN mutation carriers was markedly higher among individuals with an increased AF polygenic risk.

Future Directions in AF Genetics

While rapid progress has been made in our understanding of the genetic basis of AF over the last decade, it is important to realize that we are currently in the midst of an explosive growth in the scale of genetic data available worldwide. In the following sections we have sought to put these emerging resources in the context of future potential studies in AF genetic research (Figure 5). While there are many other potential avenues of exploration, hopefully this will serve as a broad framework for the reader.

Figure 5: Future directions in AF genetics.

Overview of emerging technologies and analyses that could shape the field of AF genetics for the next decade. Large-scale rare coding sequence data: With dropping sequencing costs large-scale exome sequencing data sets will become available and accelerate the detection of rare and ultra rare coding variation that associate with AF. Structural genetic variation: whole-genome sequencing data allows the detection of structural variation such as inversions, translocations and large insertions and deletions. Methods to detect structural variation are improving and could lead to uncovering novel structural variant contributions to AF. Polygenic risk for diverse ethnicities: Increasing the contribution of non-European samples in AF GWAS will improve the polygenic risk prediction for diverse ethnicities. Functional cellular knockout assays: Gene knockout studies in relevant cell types, such as atrial cardiomyocytes, will enable the evaluation of AF candidate genes from GWAS loci in the context of functionally relevant readouts. Single cell RNA-sequencing: Next generation sequencing technologies such as the transcriptional profiling of individual cells from cardiac tissue will transform AF genetics and increase the resolution of gene expression profiles to a cell type specific level. Cell type specific expression quantitative trait loci could resolve the causal gene at AF GWAS loci. Machine learning on big data: Machine learning can facilitate the integration of big data sources such as gene expression profiles, proteomics data, protein-protein interaction networks, methylation data, regulatory regions and spatial organization of the DNA. Machine learning algorithms will support the goal to identify causal genes for AF, resolve regulatory mechanisms at AF GWAS loci and uncover patterns that imply disease mechanisms of AF.

Abbreviations: AF, atrial fibrillation, LOF, loss-of-function, RNA, ribonucleic acid.

Expanding genetic studies in non-Europeans

As poignantly illustrated in Figure 4, the vast majority of genetic analyses for AF have been performed in individuals of European descent. The historical tendency to focus on European populations has led to a non-representative distribution of ancestries in genetic studies compared to the real-world diversity. As we move forward it will be critical to expand our genetic resources throughout the world, not only for AF, but for all common diseases. Expanding beyond Europeans has two primary advantages. First, it is clear that there are unique lessons that can be learned about common diseases among different races and ethnicities. For example, in a large Japanese GWAS for AF, only 85% of the top hits overlapped with the results from individuals of European ancestry. Second, as we think about applying genetic risk scores to clinical care, it will be important that we don’t blindly apply a score developed in Europeans to other races and ethnicities. Such an approach may further exacerbate health care disparities[49].

Multiple programs in the United States are trying to address some of the disparity in genetic research including the NHLBI TOPMed Program (www.nhlbiwgs.org), Million Veteran Program (www.mvp.va.gov), and All of Us (www.allofus.nih.gov). The All of Us project is an ongoing longitudinal collection of over 1 million individuals and is being intentionally structured to ensure ethnic and racial diversity. The expansion of biobanks throughout the world including in China (www.ckbiobank.org), India[50] and Africa (https://h3africa.org) will also be essential to this effort.

Sequencing, sequencing, and more sequencing

An exciting development in AF genetics has been the continued expansion of the availability of exome and genome sequencing. As discussed above, the major advantage of sequencing over GWAS is that sequencing directly implicates the gene as causally related to AF. There are multiple international efforts that will coalesce to enhance our understanding of common disease genetics. These include ongoing work in the UK Biobank, deCODE, Estonia, Finland, Japanese Biobank, Million Veterans Program, the NHGRI CCDG and NHLBI TOPMed Programs among many others. As a result, within the next two years we can anticipate having datasets for AF consisting of more than 50,000 cases with either exome or genome sequencing. The power of such large datasets was nicely illustrated for autism where many additional disease-causing genes were identified with the enhanced power of these large-scale studies[51].

The utility in harnessing sequence data for AF can perhaps be exemplified best by the investigators at deCODE genetics[52]. The Icelandic population is relatively homogenous and can be traced back to a small set of common ancestors, making it one of the few bottleneck populations present in the current day world. The resulting enrichment for rare mutations can manifest in naturally occurring genetic knockouts in humans[53]. The uniqueness of the population structure combined with extensive whole genome sequencing has already led to the identification of multiple new AF genes. Similarly, in the coming years, sequencing of other bottleneck populations such as Finland[54] and Sardinia[55] will be a valuable resource for studying common diseases including AF.

Structural variation from whole-genome sequencing

Beyond gene-based tests, it is also clear that current whole-genome sequencing datasets have not been utilized to their fullest potential. Structural variants such as inversions, duplications, translocations as well as large deletions and insertions may be associated with AF. It has been estimated that structural variants can have a larger than expected impact on the genomic differences between individuals[56]. Identification of these structural modifications from the raw sequencing data is more challenging than identifying single variants, short insertions or short deletions. It requires data re-processing and the application of specialized algorithms to the sequencing data. To date, no single method can identify every type of structural variant with high confidence[57]. This suggests that a combination of methods may generate the highest yield. Structural variants can have a large impact on the function of genes and have been associated with diseases such as cancer[58], and complex diseases including Crohn’s disease, rheumatoid arthritis, diabetes[59]. We can anticipate that the systematic assessment of structural variation will identify novel genetic mechanisms for AF.

Expanding gene expression and epigenetic analyses to tissues relevant to AF

To move from associations to mechanisms for the AF GWAS loci, it will be essential to expand our repertoire of disease relevant eQTL and epigenetic datasets. Currently there are modestly sized transcriptional datasets from the human left atrium and very limited data from pulmonary venous tissues. Expanding these data from the 100–200 available samples by an order of magnitude will dramatically increase the availability of eQTLs that link AF disease variants to causative genes.

It will also be essential to move beyond the analyses of bulk tissues and to focus on the analyses of single cells for transcriptional and epigenetic profiling[60], techniques which have only rarely been applied to AF to date. Recent work by Tucker and colleagues in which they performed single nucleus RNA sequencing of the healthy human heart provides an example of the benefit from this approach. In a study of nearly 280,000 single nuclei from the four chambers of the human heart, a number of findings emerged that are relevant to AF. They were able to identify at least 10 major cell types in the heart, chamber specific transcription in non-myocyte populations, and an cardiomyocyte enrichment of the genes at AF GWAS loci[61].

A logical extension of this work will be to compare the single cell transcriptional and epigenetic profiles of left atrial and pulmonary vein samples from AF cases versus controls. These results would also enable the discovery of changes of the cellular compositions, transcription and cell type specific eQTLs in AF versus healthy individuals.

Developing large scale functional screens for AF genes

Even with additional expression and epigenetic datasets, it will be critical to expand our throughput for the functional assessment of the genes at the AF GWAS loci. The current state of our AF GWAS results can illustrate the scope of the problem. At present there are close to 140 genetic loci for AF and within these loci there are more than 1000 genes or transcripts. Of these many loci, only a minority have a single gene that can be convincingly linked by eQTL or HiC analyses. This disconnect between our expansive knowledge of disease associated variants and limited understanding of the mechanisms, is not unique to AF, but is present for essentially all common diseases.

To help address this challenge, the International Common Disease Alliance was founded in 2019. This partnership between academia, governments, pharmaceutical and technology companies is a collaborative initiative with the goal to improve diagnosis, prevention and treatment of disease through accelerating research that focuses on translating genetic findings into disease biology. While a lofty goal, the implementation of high throughput functional studies to elucidate the missing link between non-coding genetic variants, causal gene and gene function was identified as one of the key priorities by this effort (www.icda.bio).

How could such functional screens be implemented for AF? It will clearly be impossible to characterize more than a small number of mouse or even zebrafish models and as a result, we will have to turn to cell-based models as an intermediate step. While cell-based models have many potential limitations, the scale of screening is the primary strength of this approach. Combining stem cell derived cardiomyocytes and CRISPR-Cas9 technology will facilitate high throughput gene knockout studies for cellular assays. Potential cell readouts could include electrophysiological measurements of calcium signaling or action potential duration[62,63], structural assessments of sarcomere assembly[64], contractility[65] and transcription[66,67]. Given that current AF GWAS loci represent a cross section of transcription factors, ion channels and sarcomeric proteins, cell-based screening will likely require multiple readouts.

It will also be important that more consideration is given to the development of cell models that more fully recapitulate the diverse etiologies of AF. For example, there are at least 10 major cell types in the human atria and the study of myocytes will not be helpful if an AF gene is predominantly expressed in another cell type such as fibroblasts[68] and in a cell subtype arising from the conduction system or pulmonary veins.

Implementing machine learning to identify endophenotypes of AF

The rapid evolution in machine learning methods has already started to transform the medical field. The availability of multi-dimensional datasets ranging from electronic health records, imaging data, clinical measurements and genetics is providing the basis for algorithm-based clinical research that has the power to improve risk prediction, response to treatment and clinical diagnostics for AF. Particularly exciting recent work has applied machine learning to the electrocardiogram to identify subclinical markers or endophenotypes that predict the future development of AF[69,70]. In the future it will be interesting to study the genetics of these AF endophenotypes and to use machine learning to enhance the prediction of loss-of function impact of coding variants[71]. Further, it is clear that novel computational approaches will be required to integrate the existing and emerging data sources in AF genetics. Predicting the most likely causal genes at an AF GWAS locus or identify gene networks relevant for AF will require the integration of large data sources. The combination of GWAS results with gene expression profiles, proteomics data, protein-protein interaction networks, methylation data, regulatory regions and spatial organization of the DNA requires efficient computational solutions that can deal with multi-dimensional data. For a broader discussion on this topic please see the accompanying review in this Compendium (formal reference to be provided by the journal).

Conclusions

AF is a complex disease with a combination of both environmental and genetic factors that contribute to the pathogenesis of the arrhythmia. Rapid progress has been made in identifying many common variant loci in GWAS for AF, yet major challenges remain in moving from disease associations to specific mechanisms. Recent genome and exome-based sequencing studies have identified TTN as the gene most commonly associated with mutations in individuals with AF. Future studies will seek to explore the application of polygenic risk scores to clinical care, build out genetic studies in non-European populations, and further expand single cell sequencing and omic technologies in cells and tissues relevant to AF. Further refinement of the genetic basis of AF will ultimately facilitate the identification of new therapeutic targets and enable more precise risk stratification for this common arrhythmia.

Non-standard Abbreviations and Acronyms

AF

Atrial fibrillation

BRA

Brazilian

eQTL

Expression quantitative trait locus

EUR

European

GWAS

Genome-wide association study

HISP

Hispanic

JAP

Japanese

KOR

Korean

LOF

Loss-of-function

PRS

Polygenic risk score

RNA

Ribonucleic acid

SNP

Single nucleotide polymorphism

STR

Short tandem repeat