Atrial fibrillation: Current knowledge and future directions in epidemiology and genomics

Abstract

Atrial fibrillation (AF) is of public health importance and profoundly increases morbidity, mortality and health-related expenditures. Morbidities include the increased risks of cardiovascular outcomes such as heart failure and stroke, and the deleterious effects on quality of life, functional status and cognition. The clinical epidemiology of AF, its risk factors and outcomes, have been extensively investigated. Genetic advances over the last decade have facilitated the identification of mutations and common polymorphisms associated with AF. Metabolomics, proteomics and other “omics” technologies have only recently been applied to the study of AF, and have not yet been systematically investigated. Systems biology approaches, while still in their infancy, offer the promise of providing novel insights into pathways influencing AF risk. In the present review, we address the current state of the epidemiology and genomics of AF. We seek to emphasize how epidemiology and “omic” advances will contribute towards a systems biology approach that will help to unravel the pathogenesis, risk stratification, and novel targets for AF therapies. Our purpose is to articulate questions and challenges that hinge on integrating novel scientific advances in the epidemiology and genomics of AF. As a reference we have provided a glossary in the inset box.

Epidemiology

AF – increasing burden and burgeoning costs

In the US and Western Europe, the aging of the population and accompanying rise in the prevalence of AF have magnified its toll on morbidity and health care costs. The estimated US prevalence of 2.7 to 6.1 million is expected to increase to 5.6 to 12.1 million by the middle of the current century (Figure 1).^1,2 Cumulative lifetime risk estimates demonstrate that AF is largely a disease of aging. In US and European community-based cohort studies the lifetime risk of AF is 22–26% in men and 22–23% in women by age 80 years (Figure 2).^3,4 AF risk doubles with each progressive decade of age; <1% in individuals 50–59 years are affected, whereas about 10% of those 80–84 years and 11–18% of those ≥85 years have AF.^1,2,5

Figure 1.

The estimated U.S. prevalence of atrial fibrillation (AF) in the year 2050 ranges from 5.6 million to as high as 15.9 million individuals. The blue line shows estimates derived from the AnTicoagulation and Risk Factors in Atrial Fibrillation Study, comprised of cross-sectional data from a large California health maintenance organization.¹ The orange line shows data estimating the number of individuals with AF in the US through 2050. The investigators identified cases of ECG-confirmed incident AF in Olmsted County Minnesota, from 1980–2000, and employed U.S. Census Bureau data to estimate that the number of individuals with AF will reach 12.1 million by 2050. However, an increase in age-adjusted incidence of AF in Olmsted County was identified from 1980–2000. The red line shows the projected number of individuals with AF in the U.S. should this increase in age-adjusted incidence of AF persist. Figure used with permission and modified from Miyasaka et al.²

Figure 2.

Lifetime risk for developing atrial fibrillation (AF) from the Framingham Heart Study. Men and women without AF at age 40 were determined to have a 26% and 23% likelihood of developing incident AF by age 80. Reproduced from Lloyd-Jones et al.³

The increasing prevalence of AF has resulted in greater health care utilization and costs. From 1985 through 1999, hospitalizations for AF increased nearly 3-fold.⁶ A recent analysis of medical costs associated with AF employed commercial and Medical databases representing approximately 38 million individuals in the U.S. The study employed propensity-matching techniques to determine individuals with AF had 73% higher medical costs compared to matched controls. The incremental cost was $8075 per individual with AF in the U.S., resulting in a total national incremental cost ranging from $6.0 to $26.0 billion in 2008 values.⁷ Data from Western Europe similarly have demonstrated escalating resources consumed by AF hospitalizations.^8–12

The increased health care costs reflect sequelae of AF, including heart failure, dementia, stroke, and mortality. Clinical trial registries have established a direct increase in the prevalence of AF with each successive New York Heart Association class.¹³ In population-based studies the incidence of heart failure among individuals with AF ranged from 33¹⁴ to 44¹⁵ per 1000 person-years. Concomitant heart failure with AF is associated with a particularly adverse prognosis. In Framingham Heart Study participants with AF, the onset of heart failure was associated with a tripling in mortality compared to those with AF alone.¹⁴ Similarly, in an adjusted meta-analysis of observational heart failure studies, AF was associated with increased mortality (odds ratio [OR], 1.14; 95% CI, 1.03 to 1.26; P<0.05).¹⁶

In longitudinal, community-based analyses, AF is associated with an increased risk of silent and clinically-evident stroke.¹⁷ With nonrheumatic AF the risk of stroke is increased about 5 fold. Compared to strokes without AF, strokes occurring in the setting of AF have greater severity, tend to be more disabling, and have an increased risk of mortality compared to strokes of other etiologies (30-day mortality OR, 1.84; 95% CI, 1.04 to 3.27).¹⁸ In addition, AF is associated with a 2-fold increased risk of silent cerebral infarction (OR, 2.16; 95% CI, 1.07 to 4.40).¹⁹ Medicare registries and clinical trials have demonstrated the protective effect of warfarin on reducing stroke events.^20,21 Furthermore, subtherapeutic anticoagulation results in 5-times the odds of an ischemic stroke.²² Multiple alternatives to warfarin, i.e. factor IIa and Xa antagonists, have been successfully developed for AF stroke prevention.²³ Ongoing challenges include long-term studies assessing adherence, cost effectiveness, quality of life and access to such agents.

Individuals with AF have a 1.7-fold odds of impaired cognitive function (95% CI, 1.2 to 2.5) and a 2-fold increased odds for dementia even after adjustment for anticoagulation status (OR, 2.3 95% CI, 1.4 to 3.7).²⁴ Similarly, there is evidence that decreased time in the therapeutic range, while on anticoagulation therapy, is associated with impaired cognition.²⁵ In Olmsted County, the cumulative rate of dementia in individuals with AF exceeded 10% at 5-year follow-up and contributed to an almost 3-fold increased mortality risk (hazard ratio, 2.98; 95% CI, 2.5 to 3.3).²⁶

Increased symptoms, diminished functional performance, and decreased quality of life (QOL) are receiving more attention as metrics for gauging AF treatments. Symptoms have been incorporated into AF guidelines as therapeutic targets, yet there remains a lack of agreement on standardized symptom ascertainment. The significant adverse association of AF with impairments in physical, mental, cognitive and social QOL domains has been documented.²⁷ However, the relation between QOL metrics in individuals with AF and clinical end-points such as heart failure, stroke, and mortality has had limited investigation.²⁸

An analysis in the Canadian Trial of Atrial Fibrillation showed improvement in QOL with AF treatment. The study was limited by 12 months of follow-up, was small in size, and was primarily comprised of individuals with preserved cardiac function.²⁹ In longer follow-up, the Atrial Fibrillation Follow-up Investigation of Rhythm Management (AFFIRM) study identified no difference in QOL outcomes in the rate or rhythm control strategy arms.³⁰

AF epidemiology: future directions

There remains a paucity of epidemiologic data on AF from many parts of the world including Eastern Europe, Africa and South America. Recent epidemiologic investigations in China and Japan reported that the AF prevalence was less than in the US and European countries.^31,32 Further, the effects of industrialization and urbanization on AF incidence have not been ascertained. Marshaling resources in areas with social and economic challenges will remain an obstacle towards estimating the global burden of AF.

AF ascertainment, monitoring, and classification in ethnic and racial minorities also have been limited. Much of the AF epidemiology described here was developed in cohorts of European descent. Classification schemes and treatment guidelines have been developed from studies limited by racial and ethnic homogeneity. The longitudinal risk factors, outcomes, and quantification of the burden of AF in ethnic/racial minorities need further elucidation.

Fundamental questions remain concerning AF subsets and the life course of AF including the long-term assessment of the occurrence, recurrence and progression of AF from onset till death. Guidelines and classification schemes describe AF as first detected³³ (first diagnosed³⁴), paroxysmal, persistent or permanent.³³ However, the classifications have clinical and epidemiologic limitations, in that an individual’s classification may only be evident retrospectively, and the durations embedded in the classification system are somewhat arbitrary. Likewise, it is uncertain how AF in the setting of acute illness, decompensation, or trauma relates to risk of subsequent AF. In addition, lone AF is a well-recognized clinical entity, and its environmental milieu and triggers merit continued investigation.

QOL and symptoms metrics have only recently been introduced,³⁵ but have not become a standardized component for clinical trial, epidemiologic study, or clinical practice. How QOL measures are affected by comorbidity and burden of AF requires continued investigation. The intersections of QOL, aging, and AF have similarly not been well explored, and nor has the relation of AF to diminished functional status and frailty.

Clinical risk factors of AF

Standard AF risk factors may be summarized succinctly as older age, male sex, smoking, obesity, hypertension, diabetes, myocardial infarction, heart failure, valvular heart disease, and cardiac surgery.^36,36–39 Figure 3 depicts the population attributable risk for AF risk factors in a longitudinal community-based study.³⁶ The biracial Atherosclerosis Risk in Communities (ARIC) recently reported that 56.5% of the population attributable risk was explained by having at least one borderline or elevated risk factor with high blood pressure being the most important risk factor.⁴⁰ Although a prominent component of the population attributable AF risk remains unknown, the ARIC data suggest important opportunities for AF prevention through risk factor modification.

Figure 3.

The population attributable risk of atrial fibrillation in men and women determined from a community-based longitudinal study.³⁶ For both men and women a substantial portion of atrial fibrillation risk remains unexplained. MI, myocardial infarction; HTN, hypertension; HF, heart failure; VHD, valvular heart disease; DM, diabetes mellitus; LVH, electrocardiographic left ventricular hypertrophy.

The increasing prevalence of many AF risk factors including the aging of the population, obesity, and diabetes coincide with the rising prevalence of AF. Age is a foremost risk factor for AF,⁴¹ from 2010 to 2050, the U.S. population ≥65 years of age is expected to increase from 13.0% to 20.2% and the population ≥85 years from 1.9% to 4.3%.⁴² Similarly, the prevalence of obesity, now up to one-third of the US population,⁴³ may contribute to an increase in the prevalence of AF. Obesity has been associated with an increase in AF risk from 1.5 to 2.3-fold in community-based cohorts with follow-up ranging from 5 to 14 years.^37,38,44 Obesity (HR, 1.54; 95% CI, 1.2 to 2.0) and severe obesity (HR 1.87, 95% CI 1.4 to 2.5) are associated with progression from paroxysmal to permanent AF over a median 5-year follow-up.⁴⁵ Obstructive sleep apnea, rising in prevalence with obesity, also is associated with increased AF risk (OR, 2.19; 95% CI, 1.40 to 3.42).⁴⁶

Miscellaneous risk factors for AF that have been reported include hyperthyroidism,^47,48 alcohol use,^49–51 and exercise.^52–56 Of interest, for both alcohol and physical activity, a threshold effect has been reported wherein increased risk of AF was only observed in individuals with heavy alcohol consumption⁵¹ or frequent vigorous exercise.⁵⁶

Indices of AF risk: biomarkers and intermediate phenotypes

Multiple biomarkers have been assessed for their association with AF. N-terminal pro-B-type natriuretic peptide (BNP) has been associated with a multivariable-adjusted, 4-fold increased risk of AF in 10 year median follow-up (HR 4.0, 95% CI, 3.2 to 5.0) when comparing the highest to lowest quartiles in a longitudinal, community-based study.⁵⁷ Inclusion of BNP likewise significantly strengthened an AF risk score derived in a community-based cohort.⁵⁸ Interleukin-6 has been elevated in participants with AF in diverse cohorts including individuals with chronic AF and coronary artery disease.^59,60 Other biomarkers associated with incident or recurrent AF include osteoprotegerin,⁶¹ troponin,^62,63 endothelin,⁶³ and plasminogen activator inhibitor-1.⁶⁴

P wave indices, electrocardiographic markers of intra- and interatrial conduction, have been proposed as non-invasive markers of electrical remodeling. The vectorcardiographic measurements of P wave duration, area, terminal force and dispersion have been used in AF risk prediction.^65,66 Associated with age, sex, diabetes, body mass index and waist circumference,⁶⁷ P wave indices serve as intermediate phenotypes of AF risk and reflect the delayed atrial conduction described in the setting of atrial inflammation, structural remodeling, and progressive atrial fibrosis.^68–70 Left ventricular hypertrophy also has been reported to predict new-onset AF in prospective analysis.⁷¹ Echocardiographic predictors of AF have effect size estimates similar to ECG predictors and include increasing left atrial diameter and left ventricular wall thickness and decreasing left ventricular systolic function.⁷²

Risk score development has received increasing attention in AF epidemiology. Risk scores serve multiple purposes, including providing benchmarks against which to test novel risk markers; assisting as tools for risk estimation and communication with patients; and identifying low or high risk individuals in decisions regarding clinical management or clinical trial enrollment. The Framingham Heart Study’s AF risk score identified significant predictors for AF risk including age, male sex, treatment for hypertension, significant cardiac murmur, history of heart failure, and increasing body mass index, blood pressure and PR interval.⁷³ The Framingham model’s c statistic was 0.78 (95% CI 0.76 to 0.80).⁷³ In subsequent models developed in the ARIC Study,⁷¹ as well as the Age, Gene/Environment Susceptibility-Reykjavik Study and the Cardiovascular Heart Study,⁷⁴ similar predictive abilities were achieved. The ARIC Study AF risk score validated the Framingham model, achieving a c-statistic of 0.68 in the ARIC cohort, and developed a novel study-specific risk score, which adjusted for race and achieved a c-statistic of 0.78.⁷¹ C-reactive protein and natriuretic peptides significantly enhance reclassification for risk of AF.⁵⁸

AF risk factors: future directions

Racial differences in AF remain poorly understood. Overall, blacks have a higher prevalence of multiple AF risk factors (obesity, diabetes, hypertension and heart failure),⁷⁵ yet lower incidence of AF.^1,76 In the ARIC Study, the cumulative risk of AF at age 80 years reached 21% in white men and 17% in white women, but was only 11% in black men and women.⁷⁷ The contrasting burden of risk factors with decreased AF incidence has been termed a “racial paradox.”⁷⁸ There is limited understanding of the biological and pathophysiologic differences modulating the racial differences in AF. An analysis of the Heart and Soul Study identified racial differences in left atrial dimensions.⁷⁹ Racial variation in P wave indices has been identified in both the ARIC study and National Health and Nutrition Examination Survey (NHANES III).^65,80 Some of the racial paradox of AF may be mediated by genetics. In a biracial meta-analysis of the Cardiovascular Health Study and ARIC study using ancestry informative genetic markers, European ancestry was identified as a risk factor for AF (estimated hazard ratio per 10% increase in European ancestry of 1.17, 95% CI 1.07 to1.29).⁸¹

Biomarkers represent important opportunities for their role in AF risk assessment, and criteria to evaluate new markers have been articulated.⁸² Numerous candidate biomarkers representing inflammatory, hemostatic, oxidative stress, matrix remodeling, and neurohormonal pathways may be tested for their associations with AF risk. We advocate a rigorous methodology for examining the value of putative novel biomarkers (or subclinical testing measure) in risk prediction. Biomarker assessment requires metrics of discrimination, calibration and risk reclassification. Providing independent replication of the relation between AF and one or clusters of biomarker(s) will further strengthen the credibility and utility of biomarkers in assessing AF risk. Finally, putative biomarkers should add incremental information over and above established AF predictors in order to demonstrate their value.

Dietary risk factors in AF epidemiology have had limited exploration, likely from challenges from longitudinal tracking of accurate exposures, as recently reviewed.⁸³ There have been contrasting findings on the protective effect of specific kinds of fish on AF prevention.^84,85 Long-term, highly accurate monitoring of nutritional intake, or development of nutritional biomarkers, is essential for clarifying dietary contributions to AF risk. The clinical and genetic interactions augmenting such risk also have not been explored. Finally, environmental exposures such as carbon monoxide and air pollution and their associations with AF risk have not been examined in community- or population-based studies.

Genetics of AF

Familial AF

The familial nature of AF was reported as early as 1942.⁸⁶ In the ensuing 50 years, there were infrequent reports of similar families with an apparent Mendelian inheritance of AF. In 1996, Brugada and colleagues described the first genetic locus for AF;⁸⁷ however, the causative gene at the locus remains elusive. In 2003, Chen and colleagues described the first mutation in a Chinese family with early-onset AF. The causative mutation resulted in a gain of function in KCNQ1 (the alpha subunit of I_Ks).⁸⁸ In subsequent years, multiple mutations have been identified in potassium^88–93 and sodium^94–96 channels, gap junction proteins,⁹⁷ and signaling molecules.⁹⁸ However, screening each of the identified genes has revealed few additional mutations in families with AF.^99,100 Thus, unlike other familial monogenic cardiac disorders such as hypertrophic cardiomyopathy and long QT syndrome, wherein a finite set of genes are responsible for the vast majority of the cases, AF is more genetically heterogenous.^99,100

Although familial AF has long been recognized, it has only recently been appreciated that AF occurring in the general population is heritable. In the Framingham Heart Study a parental history of AF was shown to nearly double AF risk in offspring.¹⁰¹ More recently, familial AF and premature familial AF improved discrimination and category-less reclassification of AF risk.¹⁰² Similarly, investigators from Iceland found that the risk of AF approached nearly 2-fold in first degree relatives of those with AF (risk ratio, 1.77; 95% CI, 1.67 to 1.88) and increased even further with an earlier age of AF onset.¹⁰³ Finally, in a study of Danish twins, monozygotic twins were significantly more likely to be concordant for AF than dizygotic twins.¹⁰⁴

Genome-wide association studies have identified 3 loci for AF

The application of genome-wide association studies (GWAS) to AF has the potential to transform our understanding of molecular mechanisms underlying the arrhythmia. Employing large cohorts, GWAS utilize single nucleotide polymorphisms, or SNPs, as markers of genetic variation between affected and unaffected individuals. High-density genotyping arrays now combine the power of association studies with systematic genome-wide research. In contrast to candidate gene studies GWAS arrays are unconstrained by pre-existing knowledge of which genes contribute to disease. To date, AF GWAS have been successful in identifying 3 novel genetic loci.^105–109

The first locus, on chromosome 4q25, is upstream of the gene PITX2.^107,109 PITX2 encodes for the paired-like homeodomain transcription factor 2 that is a plausible candidate gene for AF, albeit unrecognized as such prior to the first AF GWAS. Mommersteeg et al. demonstrated that PITX2c-deficient mice fail to form embryonic pulmonary myocardial cells and as a result lack pulmonary myocardial sleeves. The PITX2c isoform suppresses the default formation of a sinus node in the left atrium.^110,111 In the last year, two additional studies have advanced our understanding of the role of PITX2c in AF. Wang et al. have demonstrated that PITX2c is expressed in post-natal mouse left atria, pulmonary veins, and right ventricles, and that PITX2c −/− predisposes mice to atrial arrhythmias.¹¹² Further, microarray analysis revealed that in PITX2c −/− mice, gene expression in sinus node tissue was up-regulated. Up-regulated genes included SHOX2 and TBX3, in addition to multiple channel genes (including KCNQ1).¹¹² Similarly, in human atrial tissue, Kirchhof et al. found that PITX2 expression levels were approximately two orders of magnitude higher in the left atrium compared to the right atrium or the ventricles.¹¹³ PITX2c heterozygote mice had shorter atrial action potential durations compared to the wild type, and were susceptible to AF induced by pacing, whereas no differences in cardiac morphology, including interstitial fibrosis and function, were observed.¹¹³ However, while PITX2 is an attractive candidate gene for AF at this locus, the precise mechanism through which PITX2, or other genes at this locus, result in AF remains unknown.

The second genetic locus for AF is located at the transcription factor ZFHX3 on chromosome 16q22. ZFHX3 (also known as ATBF1) encodes the transcription factor zinc finger homeobox protein 3, and is expressed primarily in the brain, but also has been isolated in myocardium.¹¹⁴ The mechanism through which genetic variation at this locus relates to AF is unclear.

The third locus for AF is intronic to the calcium-activated potassium channel KCNN3 on chromosome 1q21.¹⁰⁶ The KCNN3 (or SK3) channel is part of a family of Ca²⁺-activated potassium channels gated by changes in intracellular Ca²⁺ concentration.¹¹⁵ KCNN1 (SK1), and KCNN2 (SK2) channels have predominantly atrial expression. In contrast, KCNN3 channels are equally distributed in atria and ventricles. In human and mouse cardiac repolarization models, KCNN3 channels have a prominent role during the late phase of the cardiac action potential.^116–118 Li et al. observed action potential duration prolongation and an increased number of early after depolarizations in the atrial myocytes of KCNN2 ^−/− knockout mice.¹¹⁹ Atrial arrhythmias, mainly pacing-induced AF, were more frequent in both heterozygous and homozygous KCNN2 mice.¹¹⁹ Interestingly, the familial occurrence of AF remains a risk factor for AF even after the adjustment for the 3 genetic variants identified by GWAS.¹⁰²

Figure 4 displays the 3 loci on a Manhattan plot, providing a means of illustrating their chromosomal position and statistical significance in meta-analyses. In addition, multiple genetic loci related to AF or established risk factors for AF have been identified throughout the genome. The multiple loci are illustrated in Figure 5, which is intended to indicate the number of loci, their proximity, and their relations.¹²⁰ As discussed below, such a map may inform development of a genetic systems-based approach towards AF.

Figure 4.

The figure displays the chromosomal positions for each of the three loci associated with atrial fibrillation (AF) in meta-analyses. Such a representation is designated a “Manhattan plot,” as it suggests the New York City skyline. While the x-axis displays the chromosomal position, the y-axis indicates the statistical significance for the association between the single-nucleotide polymorphism (SNP) and the phenotype of interest. The dotted line shows the threshold for statistical significance, adjusted for multiple testing to P<10⁻⁸. SNPs with P >0.05 are not displayed on the plot. The initial SNPs identified were on chromosome 4q25 located adjacent to PITX2 and had an effect size as high as 2-fold (Odds ratio [OR] 2.12; 95% confidence interval [CI], 1.77 to 2.54, P=6.3×10⁻¹⁸) in Icelandic individuals diagnosed with AF at ≤60 years.¹⁰⁸ The second locus, associated with ZFHX3, identified 3 SNPs with relative risk ranging from 1.26–1.65.¹⁰⁵ The third and most recently published SNP, identified on the 4q25 locus with OR 2.03, 95% CI, 1.79–2.30, P=2.5×10^−28.106 Figure modified with permission from Ellinor PT, et al.¹⁰⁶

Figure 5.

Circos¹²⁰ plot representing the genetic variants found by genome-wide association study for atrial fibrillation (AF) and AF risk factors. The outer ring represents the chromosomes and the inner rings detail the location of different single nucleotide polymorphisms related to AF and AF risk factors. The colors represent each phenotype or AF risk factor. AF, atrial fibrillation; BMI, body mass index; CHD, coronary heart disease; CRP, C-reactive protein; CSF, cardiac structure and function; DBP, diastolic blood pressure; DM, diabetes mellitus; ECG, electrocardiographic traits; echo, echocardiographic traits; HF, heart failure; HTN, hypertension; MI, myocardial infarction; PR, PR interval; SBP, systolic blood pressure. See electronic supplement for methods.

Genome-wide association studies of AF: future directions

Whereas recent advances with GWAS have identified potential new pathways for AF, the reported genomic loci contribute minimally towards risk of AF in the population. So far, studies of familial AF have discovered rare, highly penetrant genetic variants of AF. In contrast candidate gene studies and GWAS have revealed more common genetic variants with smaller effect sizes in AF cohorts and the general population (Figure 6). However, a large proportion of the heritability of AF cannot be explained by genetic variants reported to date. It has been postulated that the missing heritability of most common diseases, such as AF, partially may be explained by low-frequency variants with intermediate effect sizes (Figure 6).^121,122 Advances in sequencing technologies may make it possible to rapidly and reliably detect these low-frequency genetic variants. Whether exome sequencing and potentially whole genome sequencing studies will be able to uncover a substantial portion of the missing heritability of AF remains to be determined.

Figure 6.

The figure summarizes the identified genetic associations of atrial fibrillation (AF) and future areas of genetic study. Investigators have identified Mendelian families with AF genes having large effect size but rare frequency. Both candidate gene studies and genome-wide association studies (GWAS) have identified variants and loci with low-to-modest effect sizes. Among the two study types, candidate gene studies have tended to report larger effect sizes. Such an overestimation of effect size has been described in a phrase borrowed from economics as the “winner’s curse.” In contrast GWAS have identified three distinct loci with low-to-modest effect sizes. Sequencing efforts may fill a substantive genomic gap by identifying unrecognized low-frequency variants with moderate effect sizes. As a prominent component of the population attributable risk (see Figure 3) remains unknown, further genetic study may contribute additional insights into the etiology of AF. Figure modified from Manolio et al.¹²¹

Lastly, the contributions of structural variation¹²³ towards AF risk require investigation. Structural variation refers to the array of genomic microscopic and submicroscopic diversity within the human genome. Specific examples include copy-number variants, deletions, insertions, and other components introducing genomic heterogeneity.¹²⁴ Unidentified structural variants may contribute towards the heritability of AF, augment AF risk, or indirectly mediate AF risk through other clinical pathways. Furthermore, exploration of structural variants may yield insights into gene-gene and gene-environment pathways for AF risk, which have had only limited investigation.

“Omics” in AF: current status and future directions

To fill the gap between genetic variants and the development of AF new advances are being made in the field of epigenomics, transcriptomics, proteomics, and metabolomics. Epigenetic phenomena, such as DNA methylation or histone modifications may alter gene expression, thereby influencing the potential impact of changes in the DNA sequence.¹²⁵

Research in the expansive fields of “omics” is emerging rapidly. Although, such endeavors will elucidate novel pathways in the initiation and maintenance of AF, currently “omics” research is still in its infancy. Two recent advances demonstrate the utility of “omics” data in AF investigation, but also illustrate the challenges in interpretation. First, individuals with permanent AF were demonstrated to have up-regulation of transcripts involved in metabolic activities, specifically multiple glycolytic enzymes, thereby suggesting up-regulation of glucose metabolism during AF.¹²⁶ Second, transcriptomic and proteomic analysis in individuals with persistent AF identified a rise in ketone bodies, suggesting ketone bodies may contribute towards negative feedback on glucose metabolism.¹²⁷

Published transcriptomic and proteomic investigations need to be interpreted with some caution. The study samples were typically small and the findings have not had opportunity for replication. Analysis of atrial tissue is complicated by anatomic and physiologic variability.¹²⁶ AF is a highly heterogeneous disease, with multiple contributing mechanisms, predisposing risk factors and conditions. Patient dissimilarities by age, sex, race, risk factors, AF patterns, medication use, and underlying comorbid conditions may contribute to variability in gene expression.

Regulatory loops involving microRNAs are intermediate regulators and modifiers of genes. While microRNAs may serve as broad regulatory mechanisms of an entire pathway, most of their targets and precise elements of their regulation have not been clarified. MicroRNAs have been linked to fibrotic and apoptotic pathways which may contribute to AF susceptibility, via electrical or structural remodeling.¹²⁸ Several microRNAs have been targeted for their involvement in AF, based on their regulation or association with genes encoding cardiac ion channels or Ca2+-handling proteins. In both animal models and humans with AF, over 3-fold up-regulation of microRNAs specific to calcium channel subunits has been identified from atrial tissue.¹²⁹ In a cardiac surgery series (n=62, half of whom had AF), left atrial samples from individuals with AF demonstrated significantly reduced levels of a microRNA associated with potassium channel regulation.¹³⁰

Important challenges remain for the study of microRNAs in AF. There is redundancy in microRNAs targeting specific genes and likewise single microRNAs may have multiple genetic targets. How the multiplicity of effects contribute to AF and intermediate pathways for AF, such as cardiac hypertrophy and failure, renders the interpretation of microRNAs even more challenging.¹³¹ In addition, variable microRNA expression in specific patterns of AF (e.g. newly diagnosed, paroxysmal, persistent or permanent AF) or in the setting of clinical or genetic risk factors will require study. Prospective investigations such as those conducted by Lu, et al.¹²⁹ may provide insights into microRNA expression across the spectrum of AF.

Systems biology: application towards AF

AF epidemiology and genetic investigations stand at an intellectual and scientific crossroad. Similar to many medical fields,¹³² much of the prior work in AF has applied a reductionist approach, utilizing an exposure-disease paradigm for understanding AF and its risk factors. Most prior genetic epidemiology similarly has employed a two-dimensional framework for examining genome-wide associations.

Accumulating evidence indicates that complex diseases are usually influenced by the interactions of multiple genes, which are incompletely captured by single-gene based approaches.^133–135 Furthermore, biological systems are comprised of circuitries of interacting components, such as proteins, nucleic acids and other small molecules, which operate in concert to create complicated molecular networks.^136–139 Added to such myriad factors are gene-gene and gene-environment interactions, followed in turn by longitudinal epidemiologic risk factor modification.

The paradigm of systems biology provides novel and important opportunities for combining the classic epidemiology of exposure-disease association, genetic associations, “omics” and molecular investigations. A systems biology approach¹³² departs from the linear investigation of exposure-disease or gene-phenotype and seeks to investigate a network of interconnected hierarchies from the cellular to the population level. Informed by the preceding stage of investigation, systems biology networks will facilitate examination of the relations between genome, epigenome, transcriptome, proteome, metabolome and clinical (“environmental”) risk factors and their connection to the ‘phenome’ of interest. Systems biology models the complex interactions between the multiple epidemiologic, genomic, and ‘omic arenas, combining and distilling hierarchies to generate non-linear understanding of disease pathogenesis.^140,141 Figure 7 illustrates selected components which may be integrated in a systems biology approach towards AF. Ultimately, we anticipate systems biology will provide a mechanistic basis for discerning phenotypic differences among individuals with AF through development of the “interactome” of network-based approaches integrating “omic” and environmental factors.¹⁴¹ By identifying and characterizing the components of a biological system, systems biology can examine the interaction and hierarchical association between the distinct components.^142,143

Figure 7.

The figure shows the informational flow which comprises the systems biology approach. Genetic loci provide targets for epigenomic, transcriptomic, proteomic and metabolomic investigations. Epidemiologic investigations identify risk factors. In turn, systems biology translates epidemiologic and genomic pathways, providing insight into electrical and structural substrates. Systems biology in turn may yield novel hypotheses for risk factors, target intermediate endophenotypes for genetic dissection, or elucidate novel up-stream targets for therapeutic intervention. HTN, hypertension; HF, heart failure; SDB, sleep disordered breathing; LA, left atrial.

Systems biology has not had application to AF research and its potential has not yet been realized. As an iterative process, the systems biology approach will require repeated experimentation and refinement. No large studies of AF have conducted the comprehensive cell-to-cohort integration of systems networks which we envision. Future systems biology efforts will be most effective if they incorporate work across disciplines including cellular and clinical electrophysiology, in vitro and animal science, genomics, epidemiology, bioinformatics, statistics, etc. Finally, we are a full generation from a systems biology which integrates longitudinal gene-environment and risk exposures in developing networks for modeling AF risk. More realistically, we anticipate that in the future systems biology will inform basic and epidemiological investigations, reminding us that AF’s complex expression is not limited to isolated exposures or genetic loci.

Conclusion

AF has profound clinical and public health burdens, which have grown over the last several decades. Epidemiologic approaches have determined the major clinical risk factors but there are large areas of uncertainty. Robust biomarker assessment, racial and global variation, and dietary and environmental exposures are some persistent areas that require epidemiologic investigation and clarification. A better understanding of AF risk factors and risk stratification will hopefully accelerate efforts at prevention.

Advances by GWAS have identified new genomic loci associated with AF risk, but explain very little of the population attributable risk for AF. “Omics” studies, still in early development and confined to small experimental studies and clinical samples, provide new avenues for identifying cellular and developmental pathways for AF. We advocate the importance of developing a systems biology approach, which will integrate experimental and clinical data across multiple disciplines and “omics” studies. The Table summarizes gaps and research opportunities discussed in our review that are critical to address to accelerate discovery. We envision that systems biology, informed by population-based epidemiology, will identify phenotypic traits, integrate genetic loci, and utilize “omics” platforms to provide new insights into the pathogenesis, prevention, risk prediction, and therapeutic strategies for AF.

Table.

Knowledge Gaps and Areas Essential for Advancing Understanding of the Pathogenesis, Risk, Prevention and Treatment of Atrial Fibrillation.

Research Domain	Important Knowledge Gaps	Areas of Potential Discovery and Scientific Advancement
Epidemiology	Global burden of AF Longitudinal patterns of AF Longitudinal risk factors, outcomes, and temporal AF trends in ethnic/racial minorities	Identification of the clinical, subclinical, and genomic factors underlying variability in AF onset, life course, and complications in diverse populations and regions Discovery of strategies to prevent AF onset and progression
Genetics	Majority of AF heritability Mechanisms by which genetic variants contribute to AF risk Gene-gene and gene-environment interactions to modify AF risk	Exome and whole genome sequencing may uncover missing heritability of AF Uncovering genetic determinants of AF and AF endophenotypes may improve risk prediction, and identify new therapeutic targets
“Omics”	Role of epigenetic variation in AF susceptibility Gene regulatory mechanisms associated with AF Post-translational mechanisms associated with AF Relation of “omics” variation to AF pattern or risk factor burden	DNA methylation, histone modification and microRNA may influence AF gene expression and atrial remodeling Discovering epigenomic, transcriptomic, metabolomic and proteomic mechanisms underlying atrial remodeling and AF risk may lead to new targets for AF prevention and treatment
Systems Biology	Relations between clinical risk factors, “omics,” and environment	Systems biology may integrate across “omic” discoveries to identify novel non-linear, non-reductionist pathways and mechanisms influencing AF risk Integration across multiple disciplines (basic, epidemiological, clinical, bioinformatic) will accelerate understanding of complex pathways influencing AF risk, and opportunities for prevention and treatment

AF, atrial fibrillation.

Glossary

Allele

An allele is one or more versions of a gene. Different alleles may result in phenotypic differences. An individual gets one allele from his/her mother, and one from the father. In the context of common genetics, alleles typically refer to the two possible bases of a single nucleotide polymorphism, or SNP.

Bioinformatics

The application of computer science to the analysis and integration of biological data. Bioinformatics has had an increasing role in genetics, genomics and other areas of “omics” investigation, and systems biology.

Computational biology

A field of science that utilizes computer science techniques, applied mathematics and statistics to investigate and simulate problems motivated by biology.

Calibration

Risk prediction models in which the predicted risk of events approximates the actual observed risk of events are considered well-calibrated. Calibration of prediction models is typically measured by dividing the distribution of predicted risk into different categories (quantiles) such as deciles, and comparing the expected number of events to the observed number of events, as in the Hosmer-Lemeshow chi-squared calibration test statistic.¹ A significant difference between observed and expected numbers of events suggests a poorly calibrated model.

Discrimination

Discrimination refers to the ability of a model to distinguish between two or more outcomes. In the case of risk prediction models, the degree to which a model assigns a higher predicted risk to those who develop events than to those who do not develop events may be quantified using the c-statistic. The c-statistic ranges from 0 to 1. Values of 0.5 represent a model that discriminates no better than chance alone, whereas higher values represent models with better discrimination.

Endophenotype

Phenotypes that serve as proxies for another phenotype of investigative interest may be referred to as endophenotypes. Endophenotypes typically are heritable, are associated with condition or illness of interest, and may exist as an intermediate manifestation prior to the development of a manifest phenotype.

Epigenomics

The field of genetics investigating pathways that regulate DNA modifications and expression, such as methylation or histone remodeling.

Functional genomics

An investigative discipline focused on the phenotypic manifestations of genetic variation, and dynamic interactions between the genetic code and other factors. Examples include transcription and translation and the interaction of the genetic code with other genes (gene-gene interaction) or environmental factors (gene-environment interaction).

Genome-wide association study (GWAS)

A common analytical approach to examine the relations between single nucleotide polymorphisms (SNPs) or other DNA sequence variants with a phenotype of interest. GWAS typically examine associations hundreds of thousands to millions of sequence variants and a trait. Due to the probability of false positive results with such multiple testing, GWAS generally employ conservative significance thresholds (e.g., P<5×10⁻⁸) and require replication of findings in order to report statistically robust associations (See Figure 4 and 6).

Heritability

In the context of genetics, heritability refers to a measure typically quantified as the proportion of variation in a phenotype that can be attributed to genetic variation. Heritability therefore can range from 0 to 1.

Integrated discrimination improvement

A statistical test which, when used for the assessment of novel biomarkers or risk factors, assesses the probability of altering risk classification.

Interaction

Biological Interaction refers to the interrelation between two molecules or biological elements. Statistical interaction (also referred to as effect modification) describes the additional measurable effect on an outcome when considering two or more variables jointly, beyond that which occurs when measuring each variable separately. Interaction is central in genetics, proteomics and other “omics”, highlighting the association between genes, pathways of expression, and the environmental factors which result or influence such expression.

Linkage analysis

A gene mapping technique that examines the degree to which genetic markers are transmitted in a pedigree in order to identify a locus associated with a disease. Genes close to a marker that is transmitted among affected individuals are more likely to be associated with disease; recombination is less likely to occur at shorter genetic distances.

Metabolomics

The field of study that investigates the composition and effects of small molecules (metabolites) that remain as intermediary and end products of cellular processes.

microRNA

MicroRNAs are short RNA sequences (22 base pairs on average) that are complementary to messenger RNA sequences. By binding to messenger RNAs, microRNAs are able to regulate gene function by up- or down-regulating messenger RNA translation.

Monogenic

Disorders caused by defects in a single gene. Although a phenotype may have variable expression due to incomplete penetrance, the effect size of a monogenic defect is usually substantive enough to explain the disease occurrence.

Mutation

A change in the genetic code of the DNA. Typically, a single base is modified, but other forms such as modifications of longer DNA stretches, insertions, deletions or duplications of genetic code are possible. Conceptually, mutations are rare events, typically with large effect sizes (see Figure 5), in contrast to polymorphisms, which are common.

Network

A network in the context of systems biology contains the extensive relations between its constituent components (see Supplemental figures).

“Omics”

Omics is a term that encompasses the study of genomic, epigenomic, transcriptomic, metabolomic, proteomic and phenomic data, and their interactions to understand complex processes involved in biological systems.

Phenotype

A disease, characteristic or quantifiable biological measure that may be studied.

Polygenic

Refers to diseases and conditions caused by multiple DNA alterations. Unlike mongenic conditions, the effect size of each single causal DNA sequence variant may be small. In current studies, polygenic conditions etiologies are implicated when single nucleotide polymorphisms (SNPs) at multiple genetic loci are associated with a condition (Figure 5).

Population attributable risk

The extent to which a single exposure may be identified as the cause of an outcome in a defined population. Alternatively, the reduction in disease incidence if a single exposure could be completely eliminated from a population (Figure 3).

Proteomics

The scientific field investigating proteins, their structure, functions and pathophysiologic implications.

Risk reclassification

In a risk prediction model, risk reclassification refers to the degree to which predicted risk is correctly increased or decreased beyond a referent set of predictors, after considering potentially new predictor variables.²

Single nucleotide polymorphism (SNP)

A common (≥5%) or low frequency² (≥0.5% to <5%) type of DNA sequence variant comprised of a single nucleotide substitution. SNPs are usually biallelic (see Figure 6).

Systems biology

The endeavor to move beyond the boundaries of individual fields, and integrate diverse disciplines such as epidemiology, genetics, proteomics, and metabolomics. The objective of such an approach is to understand disease processes as comprehensively as possible in order to map pathways from genetic and cellular expression to their phenotypic variation.

Transcription

The genetic process in which genetic code of the DNA is copied in the form of a messenger RNA. The messenger RNA will undergo various processing and regulation steps which ultimately result in protein expression.

Transcriptomics

The field of study investigating transcribed nucleic acid molecules such as mRNA, rRNA, tRNA, and others.