Cross-ancestry genome-wide analysis of atrial fibrillation unveils disease biology and enables cardioembolic risk prediction

Atrial fibrillation (AF) is the most common cardiac arrhythmia, affecting about 46.3 million people worldwide, with rising prevalence due to aging and improved detection of subclinical AF. AF is linked to serious complications such as stroke and heart failure, imposing substantial healthcare burden. Besides clinical risk factors (aging, obesity, hypertension, heart failure), AF has a strong genetic component. Prior GWASs, largely in European populations, have identified over 100 loci implicating pathways in cardiac development, electrophysiology, contractility, and structure. However, genetic architecture in non-European populations remains underexplored, limiting biological insights and the transferability of polygenic risk scores (PRS) across ancestries. This study aimed to characterize AF genetics in a non-European (Japanese) population, enhance discovery power via cross-ancestry meta-analysis, elucidate biology through integrative functional analyses, and evaluate clinical utility of cross-ancestry-derived PRS for AF and related outcomes.

Previous AF GWASs predominantly in European cohorts have identified numerous loci implicating cardiac developmental and electrophysiological pathways, yet left gaps in understanding in non-European groups. PRS derived from European-centric GWAS often underperform when applied to other ancestries due to differences in allele frequencies and linkage disequilibrium, raising concerns about cross-ancestry portability and potential exacerbation of health disparities. Inflammation has been hypothesized to play a role in AF, with suggestive associations for IL6R previously reported, but robust genetic evidence has been limited. Transcriptional regulators such as ERRγ have been linked to cardiac maturation and function, though their specific contribution to AF risk and mechanism was previously unclear.

Study design included: (1) a Japanese GWAS (BioBank Japan, BBJ) with 9,826 AF cases and 140,446 controls; (2) replication in an independent Japanese cohort (4,602 cases, 44,075 controls); (3) cross-ancestry meta-analysis combining BBJ with large European GWAS meta-analysis (AFGen consortium, deCODE, UK Biobank, DiscovEHR, MGI) and FinnGen; and (4) downstream functional and predictive analyses.

Participants and genotyping/QC: All participants were Japanese from BBJ. QC excluded samples with call rate <0.98, relatedness (PI_HAT >0.2), heterozygosity outliers (>+4 s.d.), and PCA outliers from the Japanese cluster. Cases were AF or atrial flutter diagnosed by physician or ECG. Genotyping used Illumina HumanOmniExpress (and HumanExome) arrays; variant QC required SNP call rate ≥0.99, MAF ≥0.01, HWE P ≥1×10⁻⁶. Prephasing used EAGLE; imputation via minimac3 with 1000 Genomes Phase 3 reference plus Japanese in-house reference (1,037 samples). X chromosome was imputed separately for males and females; male dosages scaled 0–2. Replication cohort used Illumina Asian Screening Array with analogous QC; imputation with SHAPEIT2/minimac4 and larger in-house reference (3,256 Japanese).

GWAS analysis: Logistic regression (additive model) adjusted for age, age², sex, and top 20 PCs using PLINK 2.0. Inclusion criteria: imputation quality >0.3; MAF ≥0.001. X-chromosome association separately in males/females combined by inverse-variance fixed-effects (METASOFT). Heterogeneity assessed by Cochran’s Q; variants with P_het <1×10⁻⁶ excluded. Genome-wide significance thresholds: P<5×10⁻⁸ for MAF ≥1%; P<5.71×10⁻⁹ for MAF <1% (Bonferroni over 8,753,038 variants). Genomic inflation λGC = 1.12; LD score regression intercept = 1.02 indicated polygenicity. Loci defined by merging significant variants within ±500 kb; loci overlapping previously reported coordinates were considered known. Stepwise conditional analysis identified additional independent signals (locus-wide significance P<1×10⁻⁵), iterating until no further signals.

Cross-ancestry meta-analysis: Combined BBJ with European meta-GWAS (EUR) and FinnGen (FIN). Calibration checked via LDSC intercepts (EUR=1.052, FIN=1.033) and genetic correlation (EUR–FIN r=0.918). MANTRA algorithm was used to account for ancestry heterogeneity. Variants with MAF ≥1% in both Japanese and European datasets were tested; significance at log10 Bayes Factor (BF) > 6.

Transcriptome-wide association study (TWAS): Used MetaXcan v0.3.512 with GTEx v8 models for atrial appendage (n=10,414 genes) and left ventricle (n=9,702). Bonferroni thresholds: P<4.8×10⁻⁵ (atrial appendage) and P<5.2×10⁻⁵ (left ventricle). For each TWAS-significant gene, the AF-associated variant with the lowest P among those in the prediction model was selected, and distances to canonical TSS were computed. Gene Ontology enrichment performed using FUMA v1.3.7 with FDR correction.

Transcription factor enrichment: Defined AF-associated loci as ±500 kb around lead variants or high-LD proxies (r²>0.8, 1KG EUR). Overlap of ChIP-Atlas peak calls (15,109 experiments, 1,028 TFs) within AF loci versus permuted control regions assessed by two-tailed Fisher’s exact test with Bonferroni correction.

Functional assays in iPSC-derived cardiomyocytes (iPSCMs): Human iPSCs were differentiated to cardiomyocytes; ERRγ inverse agonist GSK5182 (10 µM, 4 days) administered. qPCR quantified expression of ion channel and sarcomere genes (normalized to RPS28). Motion analysis used S18000 Cell Motion Imaging System to measure spontaneous beating rate and contraction duration. Calcium transients recorded with Cal520AM and FDSS/uCell before/after isoproterenol (100 nM); measured peak counts (rate) and PWD80. Statistics: two-sided Student’s t-test; mean ± s.e.m.

Polygenic risk score (PRS): Dataset split into discovery/validation (6,890 cases, 49,451 controls), test (2,953 cases, 21,194 controls), and survival (70,645 controls). Ten-fold cross-validation within discovery to select best parameters. Summary statistic combinations: BBJ, EUR, FIN, BBJ+EUR, BBJ+FIN, EUR+FIN, BBJ+EUR+FIN; meta-analyses via METASOFT fixed/random models. PRS built via pruning-and-thresholding (P thresholds: 0.5, 5×10⁻², 5×10⁻⁴, 5×10⁻⁶, 5×10⁻⁸; r² thresholds 0.8, 0.5, 0.2) and LDpred2-grid (HapMap3 SNPs, varying p, sparse option). LD references: 1KG EAS for BBJ; 1KG EUR for EUR/FIN/EUR+FIN; both for multi-ancestry models. Performance evaluated in validation and then in independent test cohort using bootstrapping (5×10⁴ resamples); metrics: Nagelkerke’s pseudo R² and AUC from models including age, sex, top 20 PCs, and normalized PRS. Pairwise model comparisons via bootstrap ΔR² with Bonferroni (P<2.3×10⁻³).

Association with phenotypes and survival: Age at AF onset analyzed by linear regression comparing high-PRS groups (top 1%, 5%, 10%, 20%) versus others, adjusted for sex and PCs. In AF-free controls (n=121,351 with antithrombotic data), logistic regression tested associations between AF-PRS and stroke subtypes adjusting for age, sex, PCs, and antithrombotic use; Bonferroni for six outcomes. Survival analysis (n=132,737; median follow-up 8.4 years) used Cox models for all-cause, cardiovascular, non-cardiovascular, and cause-specific deaths (HF, IHD, stroke), adjusted for sex, age, PCs, disease status.

Mendelian randomization (MR): Two-sample MR using UK Biobank summary statistics, with AF variants from BBJ+FIN cross-ancestry meta-analysis as instruments (independence via pruning r²<0.01, 10,000 kb window). Pleiotropic variants associated with cardiovascular risk factors/diseases identified via PhenoScanner and excluded. Causal estimates via inverse-variance weighting (IVW); MR-PRESSO to detect outliers; Cochran’s Q and MR-Egger intercept for heterogeneity/pleiotropy. Quantitative trait exposures (e.g., height, BMI, blood pressure) used trait-associated instruments; AF as outcome.

• Japanese GWAS (BBJ; 9,826 cases, 140,446 controls) identified 31 AF-associated loci at genome-wide significance, including five previously unreported loci. SNP-heritability on liability scale was 11.7% (s.e.m. 2.6%).
• Replication in an independent Japanese cohort (4,602 cases, 44,075 controls) confirmed all five new loci with nominal P<0.05 and consistent directions.
• East Asian-specific rare variants were discovered: rs202030113 (MAF≈1.2%) in SYNE1 near a splice site (spliceAI score 0.33), and rs778479352 (MAF≈0.25%) intronic in FGF13 with strong association (OR=2.00, 95% CI 1.73–2.31, P=1.6×10⁻²⁰), located in an active chromatin region (H3K4me3/H3K27ac), suggesting a regulatory element.
• Stepwise conditional analysis revealed 18 additional independent signals (locus-wide P<5×10⁻⁸), totaling 49 signals. Ten loci had multiple independent signals; PITX2–C4orf32 had six.
• Cross-ancestry meta-analysis (BBJ, EUR, FIN; 77,690 cases, 1,167,040 controls; 5,158,449 variants with MAF ≥1%) identified 150 genome-wide significant loci (log10 BF >6), including 33 novel loci; combined with Japanese GWAS yielded 35 new loci overall.
• Among variants in high LD with lead SNPs, 19 missense variants were observed. Notable new-locus coding variants included rs848208 (p.Ala970Val) in SPEN (linked to conduction defects in model systems) and rs3746471 (p.Arg1045Trp) in KIAA1755 (previously associated with heart rate/variability).
• TWAS with GTEx v8 prioritized 132 genes (atrial appendage) and 127 genes (left ventricle) associated with AF. IL6R was significant in atrial appendage (β=0.221, P=2.147×10⁻⁶), implicating inflammatory signaling. Only 34–35 genes overlapped with nearest-by-genome position genes; median variant–TSS distances were 2.25 kb (atrial appendage) and 1.14 kb (left ventricle). Gene Ontology analysis highlighted pathways in cardiac development, conduction, and contractile/structural functions.
• Transcription factor enrichment analysis (ChIP-Atlas) showed significant enrichment of ERRγ binding at AF-associated loci after Bonferroni correction (P=3.3×10⁻⁶). ERRγ ChIP-seq peaks overlapped loci near ion channel genes (CAMK2D, KCNJ5, KCNH2, HCN4).
• Functional assays in human iPSC-derived cardiomyocytes: ERRγ inverse agonist GSK5182 reduced expression of ion channel and sarcomere genes; decreased and irregular spontaneous beating rate; prolonged contraction and calcium transient duration (PWD80); attenuated isoproterenol-induced rate increase, recapitulating AF-like electrophysiological features.
• PRS performance in Japanese test cohort (2,953 cases, 21,194 controls): single-population PRS showed similar performance for BBJ and EUR (pseudo R² 0.124 vs 0.122; P=0.681), FIN lower (0.102). Cross-ancestry PRS from BBJ+EUR outperformed EUR+FIN (0.144 vs 0.131; P<4×10⁻⁴). Best performance from BBJ+EUR+FIN (pseudo R²=0.146; 95% CI 0.115–0.170; AUC=0.738; 95% CI 0.726–0.745).
• Phenotype associations: Higher AF-PRS associated with earlier AF onset; top 1% PRS had ~4 years younger onset versus others. In AF-free controls, PRS associated with cerebral infarction (OR=1.04, 95% CI 1.02–1.07; P=4.0×10⁻¹⁰) and cardioembolic stroke (OR=1.35, 95% CI 1.13–1.63; P=1.3×10⁻⁴), strongest for cardioembolic stroke.
• Mortality: AF-PRS was associated with cardiovascular death (HR=1.06, 95% CI 1.02–1.11; P=4.4×10⁻⁴) and stroke death (HR=1.13, 95% CI 1.04–1.22; P=2.7×10⁻⁵), but not with all-cause (HR=1.02; P=0.13), non-cardiovascular (HR=1.00; P=0.50), heart failure (HR=1.05; P=0.37), or ischemic heart disease death (HR=1.04; P=0.26).
• MR analyses indicated genetic liability of AF to heart failure, cardiomyopathy, stroke, transient ischemic attack, and valvular diseases (e.g., rheumatic valve disease OR=1.139 (95% CI 1.133–1.630), P=9.4×10⁻¹⁰; valvular heart disease OR=1.183 (95% CI 1.112–1.258), P=1.1×10⁻⁷). Quantitative traits with causal effects on AF included height (OR=1.398, P=3.3×10⁻⁶), BMI (OR=1.133, P=1.8×10⁻⁴), and blood pressure (SBP OR=1.400, P=1.2×10⁻¹¹; DBP OR=1.455, P=2.1×10⁻¹⁶; pulse pressure OR=1.267, P=9.2×10⁻⁸).

This study expands AF genetic architecture beyond European populations by conducting a large Japanese GWAS and integrating it with European datasets via cross-ancestry meta-analysis. The discovery of East Asian-specific rare variants (e.g., in SYNE1 and FGF13) underscores ancestry-specific risk architecture and implicates mechanisms involving the nuclear envelope and sodium channel regulation in atrial conduction. Cross-ancestry meta-analysis substantially increased locus discovery, highlighting both shared and ancestry-informed allelic effects.

Integrative analyses linked GWAS loci to functional biology. TWAS implicated IL6R, providing genetic support for inflammatory signaling in AF pathogenesis. Transcription factor enrichment and functional validation identified ERRγ as a key regulator orchestrating ion channel and contractile gene expression in cardiomyocytes, with perturbation leading to AF-like electrophysiological phenotypes. These findings connect genetic associations to plausible molecular and cellular mechanisms underlying AF.

Clinically, a cross-ancestry-derived PRS improved predictive performance in the Japanese population compared with single-ancestry models and was associated with earlier AF onset, increased risk of cardioembolic stroke among individuals without diagnosed AF, and higher cardiovascular and stroke mortality. This suggests potential utility of AF-PRS for risk stratification, identifying individuals at risk for subclinical AF or cardioembolic stroke, and informing precision medicine approaches. MR results reinforced AF’s causal role in multiple cardiovascular outcomes and identified blood pressure, height, and BMI as causal risk factors for AF, emphasizing modifiable targets such as hypertension control.

By combining a large Japanese GWAS with cross-ancestry meta-analysis, this study identified 35 new AF susceptibility loci (150 total significant loci), including East Asian-specific rare variants, and connected these loci to biological mechanisms involving IL6R-mediated inflammation and ERRγ-driven transcriptional regulation of ion channels and contractile genes. A cross-ancestry-derived PRS achieved the best predictive performance in the Japanese cohort, associated with earlier AF onset, higher cardioembolic stroke risk in undiagnosed individuals, and increased cardiovascular and stroke mortality, highlighting translational potential. Future work should refine causal gene and variant mechanisms through functional genomics, expand multi-ancestry cohorts to enhance PRS portability and equity, and evaluate PRS-guided clinical interventions (e.g., AF screening, anticoagulation strategies) in prospective trials.

• Cause-specific mortality analyses had limited power for certain endpoints (e.g., heart failure death) due to fewer events, yielding wide confidence intervals and non-significant associations.
• While MANTRA addressed ancestry heterogeneity, the cross-ancestry meta-analysis was restricted to Japanese and European-derived datasets; broader inclusion of additional ancestries may reveal further loci and improve PRS portability.
• TWAS and TF enrichment infer regulatory links but remain correlative; although ERRγ was functionally assessed in iPSCMs, comprehensive in vivo validation and causal dissection for many loci/genes are still needed.
• PRS performance, while improved with multi-ancestry data, remains modest and was evaluated within Japanese cohorts; external validation across diverse populations and clinical settings is necessary.