New biomarkers from multiomics approaches: improving risk prediction of atrial fibrillation

Abstract

Atrial fibrillation (AF) is a common cardiac arrhythmia leading to many adverse outcomes and increased mortality. Yet the molecular mechanisms underlying AF remain largely unknown. Recent advances in high-throughput technologies make large-scale molecular profiling possible. In the past decade, multiomics studies of AF have identified a number of potential biomarkers of AF. In this review, we focus on the studies of multiomics profiles with AF risk. We summarize recent advances in the discovery of novel biomarkers for AF through multiomics studies. We also discuss limitations and future directions in risk assessment and discovery of therapeutic targets for AF.

1. Introduction

Atrial fibrillation (AF) is the most common cardiac arrhythmia with increasing prevalence and incidence globally.^1–3 The lifetime risk of AF in individuals older than 55 years is ∼37%.^4,5 It is estimated that ∼59.7 million people were affected by AF worldwide in 2019, about double the number in 1990. An additional five million individuals are diagnosed each year.^1,6 The number of AF cases is estimated to reach 12.1 million by 2030 in USA alone.⁷ AF is associated with increased risk of many comorbidities, such as stroke,⁸ heart failure,⁹ myocardial infarction,¹⁰ dementia,¹¹ as well as increased mortality.^9,12,13 The adjusted annual incremental cost for individuals with AF is $18 601,¹⁴ resulting in an increase in the US health care costs of $28.4 billion (estimated for 2016).¹⁵ Therefore, it is important to identify new strategies to prevent AF.¹⁶

Many risk factors have been identified for AF,^17–21 including advancing age, smoking, alcohol consumption, obesity, diabetes, elevated blood pressure, heart failure, and myocardial infarction. Risk scores, based on the epidemiologic cohorts’ data, have been developed to predict the 5- and 10-year risk of AF.^22,23 However, the molecular mechanisms underlying AF pathogenesis are not yet fully understood, and therapies for AF are only partially effective with substantial morbidity.^24,25

Recent advances in high-throughput technologies make large-scale molecular profiling possible. In the past decade, multiomics studies of AF have identified hundreds of potential biomarkers of AF. In this review, we focus on the studies of multiomics profiles with AF risk. In addition, we summarize recent advances in the discovery of novel biomarkers for AF through multiomics studies. We also discuss limitations and future directions in risk assessment and discovery of therapeutic targets for AF.

1.1 Search strategy

An electronic search for relevant publications was performed in the PubMed database. The major search terms included ‘atrial fibrillation OR AF’ AND ‘incident’ AND ‘biomarker’. Specific search terms included ‘genomic’ OR ‘transcriptomic’ OR ‘proteomic’ OR ‘metabolomic’. Two authors (J.K. and H.L) independently screened all retrieved records by titles and abstracts and then full-text articles. A detailed search strategy could be found in Supplementary material online, Table S1.

2. Multiomics study of AF

2.1 Genomics

2.1.1 Heritability

The familial nature of AF was first reported in 1936 with a case series noting in a footnote that three brothers were affected, all before the age of 30 years.²⁶ Familial aggregation of AF was later validated in several other studies, suggesting that familial AF comprised up to 15% of lone AF cases.^27,28 Fox et al.²⁹ observed that approximately one-third of AF patients had at least one affected parent. Marcus et al.³⁰ found that first-degree family history was associated with increased risk of lone AF. Christophersen et al.³¹ reported the importance of genetic heritability in same-sex twins and found that the co-twin of an AF-affected twin might be considered high risk for the development of AF. Lubitz et al.³² further validated the familial nature of AF and reported that having one first-degree family member was associated with a 40% increased risk. In addition, there was a 24% increased risk in AF per additional affected first-degree family member. A younger age of onset in familial members was also associated with increased risk of AF.^32,33 Therefore, a better understanding of the genetic architecture of AF may pave the road for greater insight into AF pathogenesis.³⁴

2.1.2 Candidate genes

The search for causal genes in AF started in the late 1990s.^35,36 Since then, over 35 genes have been identified to associate with AF in familial studies.^36–38 One example is SCN5A, which encodes a cardiac sodium channel essential to the cardiac action potential. SCN5A mutations were observed in up to 6% of AF patients, including lone AF and those with underlying cardiac co-morbidities.³⁹ Makiyama et al.⁴⁰ described a novel gain-of-function SCN5A mutation in familial AF.

Another example is GJA5, which encodes the connexin-40 responsible for atrial conduction. GJA5-knockout mice showed increased vulnerability to atrial arrhythmias.^41,42 Loss-of-function variants in GJA5 inhibit cell–cell electrical coupling or gap-junction congregation at the surface.⁴³ Chimeric mice with inconsistent expression of connexin-43 had conduction delays in their atrial myocardium, suggesting that mosaicism in the connexin tissue might be involved in arrhythmia initiation.⁴⁴ However, most of the variants in candidate studies are rare, observed only in affected families, and replication efforts mostly failed.⁴⁵ Hence, researchers have turned to comprehensive genome-wide screening of variants in large populations.

2.1.3 Genome-wide association studies

The most popular method to identify genetic variants for common diseases is the genome-wide association study (GWAS), which allows the comparison of allele frequency between disease cases and referents for each genetic variant [e.g. single-nucleotide polymorphisms (SNPs)]. The first GWAS of AF was performed in 2007 based on participants from Iceland, which identified the first common locus 4q25 for AF.⁴⁶ Lubitz et al.⁴⁷ later showed that at least four distinct AF susceptibility signals at the locus were associated with prevalent and incident AF. The closest gene at this locus is PITX2, which plays an important role defining right-left cardiac symmetry, developing pulmonary vein myocardium, and inhibiting sinus node formation in the left atrium.^48,49 In animal models, PITX2c knockout-mice had shorter atrial refractory periods and were more predisposed to atrial arrhythmias.⁵⁰ Using human left atrial samples, it had been shown that patients with AF have lower expression of the PITX2c isoform compared to non-AF individuals.⁵¹ In order to identify additional AF-related loci and since most genetic loci have small effects, the AFGen consortium was organized as an international effort for AF GWAS. The Consortium has led the field of AF genetics, describing the vast majority of genetic loci associated with AF in the past decade.^47,52–56 The latest AF GWAS included over 65 000 AF cases from more than 50 cohorts, and identified 97 AF-related loci.⁵⁷ Large biobanks, such as the UK BioBank, are also being utilized to identify additional genetic loci.⁵⁸ Many of the genetic loci identified through GWAS implicated genes related to cardiopulmonary development, cardiac-expressed ion channels, channelopathies, and signalling molecules, emphasizing the polygenic nature of AF.⁵⁹

Based on the GWAS results, a polygenic risk score (PRS) may be constructed that allows individualized projections of AF risk. PRSs represent the combined effect of multiple genetic variants on disease risk. Whereas individual SNPs carry relatively little effect (ranging from 0.07 to 0.37⁵⁶), pooled SNPs register observable differences among individuals. By pooling SNPs to create the PRS, Weng et al.⁴ reported that individuals with the highest tertile PRS had 1.82-fold lifetime risk of those with lowest tertile PRS. Moreover, individuals with high PRS but low clinical risk were associated with a lifetime AF risk of 43.6%, which was comparable to those with high clinical risk.

Despite the remarkable success of GWAS, several challenges remain. Genetic variation accounted for 22.1% of variance in AF risk,⁶⁰ but known AF-related loci may explain only a relatively small proportion of AF heritability, suggesting many more are yet to be identified. One possible reason for the missing heritability is unidentified rare variants with large effects compared to common genetic traits.^61,62 Most of the GWAS analyses were conducted using microarray-based genotyping platforms, which have limited resolution to identify rare variants. In addition, the top variants from the GWAS are considered to tag, or serve as proxies, for the underlying functional variants. The functions of most GWAS loci are yet to be determined.⁶³

2.1.4 Targeted sequencing and exome sequencing

The advance of next-generation sequencing allows the identification of rare variants in the general population. Lin et al.⁶⁴ performed targeted sequencing on 77 GWAS loci and studied the association of both common and rare variants with AF. Rare damaging variants within PRRX1 were found to associate with AF. In addition, a common variant rs11265611 was found to associate with AF. This SNP is located in the first intron of IL6R, which codes the interleukin-6 receptor, a pro-inflammatory marker triggering acute-phase proteins.^65,66 Inflammation is known as one of the causal pathways related to AF initiation and maintenance.⁶⁷

Similar to targeted sequencing, exome sequencing aims to perform sequencing on the entire exome. Lubitz et al.⁶⁸ performed exome sequencing on more than 1700 individuals with AF and 9000 controls from three cohorts. None of the rare variant regions were significant after adjusting for multiple testing, suggesting that rare coding variants may not be the predominant mechanism for AF in the community.

2.1.5 Whole-genome sequencing

As sequencing cost continues to decline, whole-genome sequencing is used with increasing frequency to identify genetic variants associated with complex disease. Several large-scale whole-genome sequencing projects have been initiated, including the Trans-Omics for Precision Medicine (TOPMed) project (https://www.nhlbiwgs.org/) that has generated variant calling from more than 200 K whole-genome sequencing samples. Whole-genome sequencing offers several advantages compared to target sequencing. First, whole-genome sequencing is able to cover the entire genome, which is particularly useful in identifying genetic loci that are located in non-coding regions. Second, since there is no requirement for the target region enrichment, whole-genome sequencing is capable of providing more consistent coverage. Moreover, whole-genome sequencing can also identify structural variants, which remain a challenge for targeted sequencing or exome sequencing.

Based on the TOPMed project, Choi et al.⁶⁹ reported that loss-of-function variants in titin (TTN) were associated with early AF onset. The study included 2781 early-onset AF cases (<66 years) and 4959 referents. Titin is essential in myocardial function as a passive stabilizer of the sarcomere, suggesting that structural sarcomeric abnormalities might play a role in AF pathogenesis. Furthermore, the association with hypertrophic and dilated cardiomyopathies may explain the co-occurrence of AF in these patients.

2.2 Epigenomics

DNA methylation represents an important type of epigenetic modification without changing underlying DNA sequences. The methylation process adds a methyl group to the cytosine of cytosine-phosphate-guanine dinucleotides (CpG), and thus, modifies the conformational structure of chromatin and the binding of transcription factors. Different tissue types have different methylation states, and the methylation status may be modified by both genetic and environmental factors.^70,71 Many AF risk factors such as obesity,⁷² smoking,⁷³ inflammation,⁷⁴ and alcohol consumption⁷⁵ also alter methylation profiles. The first epigenome-wide association study of AF in the community included 183 prevalent AF cases, 220 incident AF cases, and 2236 referents from the Framingham Heart Study.⁷⁶ Two CpG sites were significantly associated with prevalent AF, and five different CpG sites were associated with incident AF. One of the significant CpG sites was cg13639451, located upstream of WFIKKN2, which is known to be involved in the muscle fibre development and cardiac excitation-contraction coupling.^77,78 Cg07191189 was another CpG site for prevalent AF. It is located near STRN that encodes striatin, which is able to bind with caveolin-134 and is known to be involved in cardiac development.⁷⁹ Shen et al.⁸⁰ studied the global DNA methylation in the right atrial myocardial tissue obtained from rheumatic valvular patients. From 10 AF patients and 10 referents, it was observed that AF patients tended to have higher global DNA methylation levels than referents (P < 0.05).

2.3 Transcriptomics

Gene expression is considered an intermediate phenotype between genetic variation and disease traits.⁸¹ Both animal models and human studies suggest that AF pathogenesis is accompanied by alterations in gene expression.^82–84 Using whole blood transcriptomic profiles of individuals from the Framingham Heart Study, seven transcriptomic signatures were found to associate with prevalent AF.⁸⁵ The most significant gene was PBX1 that has an important role in great-artery patterning, septation of outflow tract,⁸⁶ and development of persistent truncus arteriosus.⁸⁷  SLC7A1 was another significant gene, which has been related to endothelial dysfunction and hypertension.⁸⁸ An AF-specific interaction network also was built and it was enriched with genes involved in multiple cardiovascular signalling pathways, such as the hypoxia signalling pathway responsible for oxygen deficit in cardiovascular organs, and the antiproliferative signalling pathway involved in the remodelling of cardiac myocardium.^89,90 Of note, none of these genes were associated with incident AF, suggesting that different signalling pathways are responsible for AF onset and maintenance.

MicroRNAs (miRNAs) are a class of short non-coding RNAs that regulate mRNA expression. They increasingly have been recognized as potential biomarkers for cardiovascular disease, in part, due to their stability in peripheral circulation.^91,92 Dawson et al.⁹³ reported MiR-29 was a potential biomarker for AF. Liu et al.⁹⁴ studied the association of plasma miRNAs with AF in a Chinese hospital-based cohort. Compared to referents, the expression of miRNA-150 was significantly down-regulated in paroxysmal AF and persistent AF patients. MiRNA-150 is involved in the regulation of cardiac fibrosis and cellular proliferation in myocardial infarction and heart failure.⁹⁵ McManus et al.⁹⁶ performed another study that included 153 prevalent AF cases, 107 incident AF cases, and 2185 referents from the Framingham Heart Study. The only miRNA associated with prevalent AF was miRNA-328. MiRNA-328 is known as an important gene regulator secreted by cardiomyocytes under stress.⁹⁷ In an experimental model, the expression of miRNA-328 was up-regulated in the left atrial samples in animals with AF.⁹⁸ As some circulating miRNAs are platelet-derived, anti-platelet medications might directly affect miRNA levels, and thus be important for AF patients with concurrent cardiovascular or cerebrovascular diseases with indication for anti-platelet treatment.⁹⁹ The major part of miRNAs in plasma is localized in microparticles and originates in up to 45% from platelets.¹⁰⁰ Furthermore, previous studies indicated an association between altered expression of miR-146b and the P38MAPK/COX-2 pathway, making a causal relationship between antiplatelet medication and Cox inhibition possible.¹⁰¹

In another study based on clinical samples, McManus et al.¹⁰² observed that circulating plasma expression of miRNA-21 and miRNA-150 were significantly associated with AF. Patients with persistent AF had lower miRNA-21 expression compared with paroxysmal AF, suggesting that higher AF burden could affect plasma miRNA expression. The miRNA-21 expression in the right atrium was down-regulated in patients undergoing cardiovascular surgery. In addition, a three-fold increase of miRNA-21 and miRNA-150 was observed after rhythm restoration in patients undergoing AF catheter ablation, suggesting significant dynamic changes after sinus rhythm restoration. Despite a relatively small study sample, this was the first study describing a strong association between circulating and tissue miRNAs linked to AF and significant improvement of miRNA levels after sinus rhythm restoration.¹⁰²

2.4 Proteomics

Proteomics refers to simultaneous screening of large numbers of proteins as potential biomarkers for different diseases. Lind et al.¹⁰³ used a custom-made proteomics chip to profile plasma proteins in participants from the Swedish study. Among the 92 screened proteins, 7 were significantly associated with incident AF after adjustment for age and sex. Two proteins, IL-6 and NT-proBNP, remained significant after adjusting for additional AF risk factors.¹⁰³ Willeit et al.¹⁰⁴ performed another proteomic study based on 880 participants from the Bruneck Study who were free of AF at the baseline. One hundred and seventeen participants developed AF during 20-year follow-up. Among the 13 inflammation markers measured at the baseline, sVCAM-1 was found to associate with incident AF after adjusting for age and sex. Ko et al.¹⁰⁵ screened 1373 proteins by using the SOMAScan assay in the Framingham Heart Study and identified eight proteins associated with incident AF. Two of them, NT-proBNP and ADAMTS13 remained significant after adjusting for known risk factors for AF. NT-proBNP is a marker of myocardial stress and ventricular remodelling known to be associated with AF.¹⁰⁶ ADAMTS13 is a protease of von Willebrand factor involved in left atrial remodelling,¹⁰⁷ and it has been used as a biomarker of rhythm outcomes after cardioversion.¹⁰⁸ In a recent study, Staerk et al.¹⁰⁹ tested the association of 85 proteins with incident AF, which were measured using the Luminex xMAP platform. Higher levels of NT-proBNP and insulin-like growth factor-binding protein 1 as well as lower levels of insulin-like growth factor 1 were associated with increased risk of incident AF after adjusting for AF risk factors.

2.5 Metabolomics

Metabolomics are studies to perform systematic analyses of metabolites with diseases. Mayr et al.¹¹⁰ observed patients with persistent AF had elevated ketone body metabolism in the atria. De Souza et al.¹¹¹ reported increased metabolic stress was associated with impaired energy utilization and increased ketoacid metabolism in a canine AF model. Alonso et al. performed the metabolomics profiling of blood samples from ∼1900 African American participants in the Atherosclerosis Risk in Communities (ARIC) study.¹¹² The participants were followed over 20 years. Two bile acids (glycolithocholate sulfate and glycocholenate sulfate) were significantly associated with incident AF, suggesting a connection between liver function and AF, in agreement with previous studies.^113,114 Potential arrhythmogenic effects of bile acids had been reported in humans,¹¹⁵ and elevated maternal bile acid levels during pregnancy were associated with cardiac arrhythmias in fetuses.¹¹⁶ In another study, Ko et al.¹¹⁷ analyzed liquid chromatography-tandem mass spectrometry metabolomics profiles in over 2000 individuals of European ancestry from the Framingham Heart Study; they did not reveal metabolites significantly associated with incident AF. In addition, neither bile acids identified in the ARIC study were replicated, possibly due to different ancestries, tissue types, or metabolic profiling platforms.¹¹⁸

3. Tissue consideration for multiomics

Unlike genetic profiles, other omics profiles, such as transcriptomic and epigenomic profiles, can change from one tissue to another. The most relevant tissue for AF is the left atrium. However, it is not feasible to perform invasive specimen collection on a large scale, especially in community-based cohorts. Most multiomics studies have measured molecular profiles from blood samples, which may be different from the heart samples. Furthermore, the left and right atria can have quite distinct expression patterns for some genes, resulting in different structure and function between these two chambers.^119,120

One useful resource is the Genotype-Tissue Expression (GTEx) project, which quantified gene expression across 54 tissue types collected from 948 donors.¹²¹ Computational methods, such as PrediXcan,¹²² MetaXcan,¹²³ RIVER,¹²⁴ and ExPecto,¹²⁵ are being developed to predict tissue-specific transcription profiles from genetic variants. For example, Roselli et al.⁵⁷ performed a transcriptome-wide association study of AF using MetaXcan based on the GWAS summary association statistics. The pre-computed models were generated from the left and right atria from the GTEx project.¹²¹ The models predicted expression of 57 genes associated with AF, 42 of which were located in a single AF GWAS loci.

4. Data integration

As multiomics data become increasingly available, future studies might integrate different omics to obtain a more comprehensive picture of AF pathogenesis. Quantitative trait loci (QTL) analyses study the associations between genetic variants and other omics data.⁸¹ One study investigated the association between genetic variants and gene expression in the left and right atria.¹²⁶ It identified 187 eQTLs from 53 left atrial samples and 259 eQTLs from 52 right atrial samples. Many of the eQTLs were shared between the left and right atrial samples, including rs3740293, which was one of the top AF-related variants. The SNP is located at the 3′-UTR of SYNPO2L, about 5 kb upstream of MYOZ1. It was significantly associated with the expression of MYOZ1 in both left and right atria, but not SYNPO2L, suggesting that the functional gene at this locus is likely to be MYOZ1 instead of SYNPO2L.¹²⁶ The association was further validated using 70 additional samples collected from patients at Massachusetts General Hospital. In another study, the mQTL analyses were performed to investigate the association of AF-related variants with nearby DNA methylation.⁷⁶ The most significant association was observed between rs6490029 at the CUX2 locus and cg10833066. The methylation of cg10833066 increased with increasing copies of the ‘A’ allele of rs6490029, which also was associated with higher AF risk as found in the previous GWAS.⁵⁴ Further studies are needed to validate these results and apply QTL mapping to other AF-related electrophysiological traits like PR and RR intervals.^127,128

Another approach to data integration is to perform multiomics modelling, which integrates the results from different omics association studies. Wang et al.¹²⁹ recently developed a strategy to integrate the GWAS, the epigenome-wide association study, and the transcriptome-wide association study results for AF. The summary statistics from different omics data were meta-analyzed and weighted by the sample size of each omics data type. A tissue-specific network was then built¹³⁰ and used to predict potential AF-related genes, which increased the proportion of heritability from 3.5% (by GWAS alone) to 10.4%. Importantly, a few potential drug targets were identified, including ADORA1, ATP1A3, ATP1B2, CACNA1D, KCNQ4, NR3C2, and THRA. Similarly, van Ouwerkerk et al.¹³¹ developed another approach to prioritize potentially functional variants by integrating transcriptomic and epigenomic data together with chromatin conformation information. A list of potentially target genes was identified from AF-associated variants, which could be further investigated for future studies.

5. Future directions

The summary of omics findings related to AF is presented in Tables 1 and 2 and Figure 1. Despite the success of omics study in AF, several aspects are important to consider for future studies.

Table 1.

Genomics association studies of AF

Study	Population	Genes	Top variants
Gudbjartsson et al.46	Iceland    AF n = 2.251    No AF n = 13.238   UK/US European ancestry    AF n = 779    No AF n = 1.542   Chinese cohort    AF n = 333    No AF n = 2.836	PITX2	rs2200733   rs10033464
Benjamin et al.52	European ancestry    Prevalent AF n = 896    Referents n = 15.768    Incident AF n = 2.517    Referents n = 21.337   Replication in German AFNET    AF n = 2.145    Referents n = 4.073	MTHFR   PITX2   ZFHX3	rs17375901   rs17042171   rs2106261
Ellinor et al.54	European ancestry    AF n = 6.707    Referents n = 52.426   Replication    AF n = 5.381    Referents n = 10.030   Japanese cohort    AF n = 843    Prevalent AF n = 3.350	KCNN3-PMVK   PRRX1   PITX2   WNT8A   CAV1   C9orf3   SYNPO2L   SYNE2   HCN4   ZFHX3	rs6666258   rs3903239   rs6817105   rs2040862   rs3807989   rs10821415   rs10824026   rs1152591   rs7164883   rs2106261
Lubitz et al.47	European ancestry n = 64 683    Prevalent AF n = 3.302    Incident AF n = 3.869   Japanese cohort n = 11 309    Prevalent AF n = 7916	KCNN3-PMVK   PRRX1   PITX2   PITX2   PITX2   PITX2   CAV1   C9orf3   SYNPO2L   SYNE2   HCN4   ZFHX3	rs6666258   rs3903239   rs1448818   rs6817105   rs4400058   rs6838973   rs3807989   rs10821415   rs10824026   rs1152591   rs7164883   rs2106261
Sinner et al.55	European ancestry    No AF n = 52.426    AF n = 6.707   Replication    No AF n = 17.144    AF n = 6.691   Japanese cohort    No AF n = 3.350    AF n = 843   Replication    No AF n = 17.190    AF n = 7.530	NEURL   TBX5   CAND2   GJA1   NEURL   CUX2	rs12415501   rs10507248   rs4642101   rs13216675   rs6584555   rs6490029
Christophersen et al.56	Multi-ancestry cohort   GWAS    AF n = 17 931    No AF n = 115 142   ExWAS and RVAS    AF n = 22 346    No AF n = 132 086		rs72700118   rs3771537   rs2540949   rs2288327   rs337711   rs2967791   rs4946333   rs7508   rs35176054   rs75190942   rs6800541   rs89107   rs11047543
Low et al.140	Japanese cohort    AF n = 8180    Controls n = 28 612   Replication cohort:    AF n = 3120    Controls n = 125 064	KCND3   PPFIA4   SLC1A4–CEP68   HAND2   HAND2   NEBL   SH3PXD2A	rs12044963   rs17461925   rs2540953   rs17059534   rs7698692   rs2296610   rs2047036
Lee et al.141	Korean cohort    AF n = 672    Controls n = 3700   Replication study    AF n = 200    Controls n = 1812	PRRX1   PITX2   NEURL1   TBX5   ZFHX3	rs3903239   rs6817105   rs6584555   rs10507248   rs2106261
Roselli et al.57	Multi-ancestry cohort    AF n = 22 346   No AF n = 132 086		67 novel loci
Nielsen et al.58	European ancestry    AF n = 60 620   No AF n = 970 216		63 novel loci
Choi et al.69	Lone AF n = 2781   No AF n = 4959	TTN	OR (95% CI), P value   2.16 (1.34–3.48); 1.55×10−3

GWAS, genome-wide association study; ExWAS, exome-wide association study; RVAS, rare variant association study.

Table 2.

Multiomics association studies of AF

Omics	Study	Population	Tissue type	Platform	Significant signatures
Epigenomics—Methylation	Lin et al.76	European ancestry   FHS Offspring cohort n = 2639   Prevalent AF n = 183   Incident AF n = 220   Follow-up 9 years.	Whole blood	lllumina Infinium Human Methylation 450 BeadChip	Prevalent AF   cg13639451   cg07191189   Incident AF   cg26602477   cg15440392   cg04064828   cg27529934   cg06725760
Transcriptomics—mRNA expression	Lin et al.85	European ancestry   FHS Offspring cohort n = 2446   Prevalent AF n = 177   Incident AF n = 143   No AF n = 2126   Follow-up 7 years.	Whole blood	Affymetrix Human Exon 1.0ST Array	Prevalent AF   PBX1   C17orf39   PNP   C18orf10   SLC7A1   SPTB   ANKH
Transcriptomics—MicroRNA expression	Liu et al.94	Chinese ancestry   5 healthy controls, 5 patients with paroxysmal atrial fibrillation (PAF) alone, and 5 patients with persistent atrial fibrillation (PersAF)	Plasma	MirVana PARIS kit (Invitrogen) and massively parallel signature sequencing	16 miRNAs for PAF and 11 miRNAs for PersAF
Transcriptomics—MicroRNA expression	McManus et al.96	European ancestry   FHS offspring cohort n = 2445   Prevalent AF n = 153   Incident AF n = 107   Follow-up 5.4 years	Whole blood	TaqMan (PAXgene, Applied Biosystems)	Prevalent AF   miR-328
Transcriptomics—MicroRNA expression	McManus et al.102	AF n = 112   European ancestry n = 108   No AF n = 99   European ancestry n = 67	Right atrial tissue   Plasma	Quantitative reverse transcriptase–polymerase chain reaction Qiagen (Valencia, CA) miScriptAssays   BioMark System (Fluidigm)	miRs-21   miRs-150
Proteomics	Lind et al.103	European ancestry   PIVUS study   No AF n = 830   AF n = 148   Follow-up 10.0 years   European ancestry   ULSAM study   No AF N = 602   Incident AF = 123   Follow-up 7.9 years	Plasma	A proximity extension assay (PEA) chip	IL-6   NT-proBNP
Proteomics	Willeit et al.104	European ancestry   No AF n = 763   AF n = 117   Follow-up 20 years	Venous blood	A commercially available enzyme-linked immunosorbent assay kit (Bender MedSystems)	sVCAM-1
Proteomics	Ko et al.105	European ancestry   FHS Offspring cohort n = 1885   Incident AF n = 349   Follow-up 18.3 years.	Citrate-plasma	SOMAscan proteomic profiling platform	Analyzed plasma proteins (n = 1373)   NCAM-120   WFIKKN2   Ntrk3   EGFR   ADAMTS13   Angiopoietin-2   NT-proBNP   BMPR1A
Proteomics	Staerk et al.109	European ancestry   FHS Offspring and Third Generation cohorts (n = 3378)   Incident AF n = 401   Follow-up 12.3 years.	Plasma	Luminex xMAP platform	Analyzed plasma proteins (n = 85)   IGF1   IGFBP1   NT-proBNP
Metabolomics	Alonso et al.112	African ancestry   ARIC study (n = 1919)   Incident AF n = 183   Follow-up 22 years.	Serum	Untargeted, gas chromatography/mass spectrometry and liquid chromatography/mass spectrometry–based metabolomic quantification (Metabolon, Inc)	Bile acids glycolithocholate sulfate and glycocholenate sulfate
Metabolomics	Ko et al.117	European ancestry   FHS Offspring (n = 2458)   Incident AF n = 156   Follow-up 10 years.	Plasma	Targeted liquid chromatography-tandem mass spectrometry (AB SCIEX 4000 QTRAP triple quadrupole mass spectrometer for positively charged polar compounds and lipids) and an AB SCIEX 5500 QTRAP triple quadrupole mass spectrometer for negatively charged polar compounds	Analyzed 217 metabolites (54 positively charged, 59 negatively charged, 104 lipids)   None

IGF1, insulin-like growth factor 1; IGFBP1, insulin-like growth factor-binding protein 1; NT-proBNP, N-terminal prohormone of brain natriuretic peptide.

Figure 1.

Multiomics studies in AF.

• First, current omics studies have been based mostly on samples collected at a single examination, whereas longitudinal omics profiling is still very scarce. It is challenging to investigate the effects of longitudinal changes in omics profiles on AF pathogenesis since omics platforms and profiles change over time. In addition, the relationship between AF and omics profiles could be bidirectional and the causal effects remain unclear.¹³² Therefore, longitudinal omics methods should be added to the future investigation of AF pathogenesis.

• Second, it would be valuable to explore the association of additional types of omics with AF, such as in samples collected from the gut microbiome,¹³³ urine,¹³⁴ saliva,¹³⁵ and right and left atrial and left pulmonary vein tissues. These omics data might identify additional molecular signatures and biomarkers for AF. Given the limited availability of heart samples, the use of induced pluripotent stem (iPS) cells to derive heart tissue from fibroblasts or even peripheral blood cells may greatly expand research capabilities.^136,137

• Third, further development of data integration methodology is needed to combine different omics data to better understand the pathophysiologic pathways of AF. Such efforts would be important to the risk stratification and the identification of novel therapeutic targets for AF.^138,139

• Fourth, further clinical and observational studies are needed to define whether multiomics profiles are different in individuals with AF dependent on disease burden. As hundreds of potential biomarkers of AF have been reported in various studies, additional scrutiny will be necessary to understand any potential bias of omic biomarkers for risk profiling. Almost all studies were adjusted for age and sex; however, many studies did not take into account known clinical risk factors that might confound the results. In addition, some biomarkers reached only lenient significance; the associations will no longer be significant after adjusting for multiple testing.

• Fifth, the mechanisms relating biomarkers are uncertain. Are they directly related, confounders, related via intermediate mechanisms, or merely spurious associations yet to be determined? In addition, even if the associations are replicated, it remains uncertain how the biomarkers are involved in the pathogenesis of AF.

• Sixth, most of the existing omics studies were based on participants of European ancestry from Western Europe and North America; generalizability of these findings to other ancestries and regions is largely unknown. Therefore, it is imperative to include participants from diverse ancestries/ethnicities and regions for future omics studies, such as African American, Asian, Hispanic, and Indigenous individuals.^{94,112,140,141}

• Finally, from clinical perspective, the incorporation of omics profiling into clinical routine is yet to come. The risk prediction of AF using omics is limited mostly to genomics information, whereas omics information has not been used to improve prediction beyond clinical score. In addition, very few studies have analyzed C-statistics and reclassification metrics of predictive value after adjustment for biomarkers or genetic risk (Supplementary material online, Table S2). Most of them showed only modest improvement of predictive value.^142–147 Future studies are needed to analyze risk prediction and to identify effect sizes, population attributable risks, and risk reclassification using different omics profiles.

In summary, the past decade has witnessed enormous progress in the multiomics study of AF, which has identified hundreds of potential biomarkers or targets for future investigations. Recent efforts paved the way towards more advanced analyses elucidating pathophysiological complexity of AF underlying processes. The next studies are needed to highlight not separate biomarkers based on genomic, proteomic, or metabolomics studies, but rather using a multiomics approach in diverse multiracial populations.

Supplementary material

Supplementary material is available at Cardiovascular Research online.

Conflict of interest: none declared.

Funding

J.K. received funding from the Marie Sklodowska-Curie Actions under the European Union’s Horizon 2020 research and innovation programme (Agreement No. 838259). L.T. was supported by the American Heart Association (18SFRN34150007). D.K. was supported by American College of Cardiology Foundation/Merck Research Fellowship in Cardiovascular Diseases and Cardiometabolic Disorders. S.R.P. was supported by NIH grant 5R01HL128914. E.J.B. was supported by NIH 2R01 HL092577, 1R01 HL141434 01A1, 2U54HL120163, and 1R01AG066010; American Heart Association, 18SFRN34110082. H.L. was supported by the European Commission Grant (Agreement No. 847770).

 

This article is part of the Spotlight Issue on Atrial Fibrillation.