The horse cardiac transcriptome: Moving towards a molecular understanding of atrial fibrillation

Abstract

Background

High recurrence rates after atrial fibrillation (AF) treatment may be driven by myocardial changes induced by the arrhythmia itself. Understanding the molecular mechanisms behind these changes is crucial for developing targeted therapies and improving outcomes.

Objectives

To characterise the cardiac transcriptome of healthy horses and explore transcriptional changes associated with persistent AF (naturally occurring and tachypacing‐induced).

Study Design

Case–control study.

Methods

RNA‐sequencing was performed on atrial and ventricular tissue samples collected from six horses with naturally occurring persistent AF (lasting 2–12 weeks) and six healthy controls. Differential gene expression and pathway enrichment analyses were conducted to identify chamber‐specific differences and molecular pathways associated with AF. Findings were integrated with proteomic data and compared with transcriptional changes in a separate cohort of 10 horses with tachypacing‐induced AF. Atrial metabolic remodelling was further investigated by evaluating the activity of AMP‐activated protein kinase (AMPK), a central metabolic regulator and measuring local glycogen content.

Results

The transcriptomes of the four heart chambers had distinct molecular identities. Expression of ion channels and genes encoding calcium handling proteins was largely similar to humans, despite important differences in the ventricular expression of repolarising potassium channels. Persistent AF was associated with minimal ion channel changes but significant upregulation of metabolic, fibrotic and myofibrillar pathways. Metabolic remodeling included upregulation of fatty acid and glycolytic pathways, increased glycogen content in the left atrium and preserved AMPK activity in the right atrium. Transcriptomic profiles of naturally occurring persistent AF correlated well with those of tachypacing‐induced AF.

Main Limitations

The study cannot distinguish changes predisposing to AF from those caused by it.

Conclusion

Persistent AF was associated with changes in metabolic and fibrotic pathways in the atria, with minimal ion channel remodeling. Targeting these pathways, rather than focusing solely on the electrical disturbance, may improve treatment outcomes in equine AF.

1. INTRODUCTION

Atrial fibrillation (AF) is the most common performance‐limiting arrhythmia in horses, affecting up to 2.5%–6% of the general population. ¹ , ² While initial pharmacological and electrical cardioversion is highly effective, ³ up to 40% of treated horses experience recurrence. ⁴ , ⁵ In other species, disease progression and recurrence are often attributed to AF‐induced arrhythmogenic remodelling, involving structural, metabolic and electrical changes in the atrial myocardium (the ‘atrial substrate’) that sustain arrhythmia. ⁶ In horses, AF similarly leads to structural changes such as left atrial (LA) enlargement and fibrosis, ⁷ as well as electrical remodelling, including shortened electrical refractoriness and increased atrial fibrillatory rate. ⁷ , ⁸ While local metabolic changes in the atria are increasingly recognised as important contributors to AF in humans, ⁹ , ¹⁰ they remain unexplored in horses. Furthermore, the molecular pathways underlying atrial remodelling in equine AF have yet to be investigated.

Understanding the mechanisms underlying AF in horses could be essential for guiding the development of new equine therapies, similar to advancements made in human medicine. ⁶ , ¹¹ Two principal approaches emerge: targeting the electrical disturbance (e.g., through pulmonary vein ablation or atrial‐specific ion channel blockers) or addressing the atrial substrate. ¹² Both strategies can benefit from insights provided by omics technologies, which offer high‐throughput tools for studying global sets of biological molecules. ¹³ For instance, transcriptomics (e.g., through RNA‐sequencing technology) enables quantitative analysis of gene expression, while proteomics (e.g., through mass spectrometry) identifies proteins and their relative abundances. RNA‐sequencing is particularly powerful, as it captures the expression of thousands of genes from a single sample. ¹⁴ In the context of AF, it can therefore reveal cardiac ion channels with atrial‐selective expression, which could be potential targets for selective pharmacological blockade. ¹⁵ Beyond ion channels, transcriptomics can also provide insights into molecular pathways that underlie atrial substrate remodelling and thereby support the persistence and progression of AF. ¹⁶ A critical first step towards understanding AF‐induced remodelling is defining the molecular profile of the healthy horse heart. This fundamental knowledge is essential for distinguishing normal cardiac physiology from disease‐related changes associated with AF or other cardiac diseases.

It is also crucial to consider how different equine models of AF contribute to our understanding of arrhythmogenesis. While naturally occurring AF can offer insights into early disease events, including potential triggers, most research on atrial remodelling in horses has relied on experimental models of tachypacing‐induced AF. ⁷ , ⁸ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ This experimental model replicates important aspects of atrial remodelling in human AF, ⁷ but it remains uncertain how accurately it reflects naturally occurring AF in horses, potentially limiting the clinical relevance of findings to equine patients. To address these gaps, we used RNA‐sequencing to perform an exploratory profiling of the equine four‐chamber cardiac transcriptome in healthy horses in sinus rhythm and in horses with naturally occurring persistent AF. Our findings were then compared with transcriptomic changes previously observed in horses with tachypacing‐induced AF. We aimed to characterise ion channel expression in health and disease and investigate molecular changes associated with equine AF, with the ultimate goal of identifying potential therapeutic targets for future research.

2. MATERIALS AND METHODS

2.1. Inclusion of horses

The studies were conducted at the Large Animal Teaching Hospital, Department of Veterinary Clinical Sciences, University of Copenhagen, Taastrup, Denmark. Six horses diagnosed with persistent AF based on resting ECG examinations were included in this study (Table S1). These horses were donated for research after their owners decided not to pursue treatment. The AF duration was estimated to be 2–12 weeks based on owner‐reported history, which included signs such as reduced performance, general weakness and detection of an irregular and disproportionately high pulse. The estimated mean duration was 38 ± 28 days. General clinical and biochemical examinations revealed no abnormalities beyond their AF diagnosis. Following euthanasia, full‐thickness myocardial biopsies were collected from the caudal walls of the right atrium (RA) and left atrium (LA), and free walls of the right ventricle (RV) and left ventricle (LV) using a scalpel. Each biopsy weighed ~2 g. Similar tissue samples were obtained from six healthy control horses in sinus rhythm, previously included in a separate study. ²¹ The healthy horses had no history of cardiac disease and showed no significant findings on clinical, echocardiographic, electrocardiographic or biochemical examinations. All tissue samples were snap‐frozen in liquid nitrogen immediately after collection and stored at −80°C.

In addition to naturally occurring AF cases, we also included transcriptomic data from horses with tachypacing‐induced AF (4 months) for comparative analysis. This dataset consists of 10 horses with tachypacing‐induced AF and four sham‐operated controls (Table S2) from an ongoing study. AF was induced through 2 weeks of right atrial tachypacing at 10 Hz using a neurostimulator (Itrel4, Medtronic), following a previously described protocol. ⁷ Horses were euthanised after 4 months, and tissue samples were collected using the same protocol as for horses with natural AF. The RNA‐sequencing data from these horses is publicly available in the NCBI Sequence Read Archive (PRJNA1209939).

2.2. Transcriptome analysis

The bioinformatics workflow, including scripts, outputs and instructions, is available on GitHub at https://github.com/simonhaugaard/Natural_AF.

2.2.1. RNA‐isolation

Heart tissue samples were homogenised in lysis buffer using a TissueRuptor (QIAGEN). RNA was extracted using the miRNeasy Mini Kit (QIAGEN, Cat. No. 217004), with on‐column DNA digestion performed using the RNase‐Free DNase Set (QIAGEN, Cat. No. 79254), following the manufacturer's protocols. RNA integrity was assessed using an Agilent RNA 600 Nano Kit on a Bioanalyser instrument (Agilent Technologies). An RNA‐integrity (RIN) value above 6 was considered acceptable.

2.2.2. Library preparation and sequencing

Poly(A) RNA was isolated from 1 μg of total RNA using the NEBNext Poly(A) mRNA Magnetic Isolation Module (NEB, Cat. No. E7490S). Library preparation was performed with the NEBNext Ultra™ II Directional RNA Library Prep Kit for Illumina (NEB, Cat. No. E7760S) with NEBNext Multiplex Oligos for Illumina (96 Unique Dual Index Primer Pairs, NEB, Cat. No. E6440S). The poly‐A RNA was fragmented, reverse‐transcribed using random primers and indexed during PCR amplification. Library concentration was quantified using the Qubit dsDNA Assay (Thermo Fisher, Cat. No. Q33230) and size distribution was analysed using the Agilent High Sensitivity DNA Kit (Agilent, Cat. No. 5067‐4626) on a Bioanalyser. Equimolar pooling of the libraries was followed by sequencing on an Illumina NovaSeq X Plus 10B flow cell using a 150‐cycle paired‐end protocol. The raw sequencing data from this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject number PRJNA1198487.

2.2.3. Bioinformatics processing and differential expression analysis

Sequencing reads from FASTQ files were assessed for quality using FastQC (v0.12.1, https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Pre‐processing was conducted with FastP (v0.23.4), ²² yielding an average of 25.7 million cleaned reads per sample. The cleaned reads were aligned to the Horse Ensembl genome (EquCab 3.0.111) using STAR aligner (v2.7.11b) ²³ resulting in an average alignment of 23.2 million reads per sample. FeatureCounts from the Subread package (v2.0.6) ²⁴ assigned an average of 15.3 million reads per sample.

The resulting gene expression count matrix was processed in R (v.4.3.2), where low‐expression genes were filtered using the filterByExpr function from the edgeR package (v4.0.16). ²⁵ Trimmed Mean of M‐values (TMM) normalisation was applied to account for differences in library size using calcNormFactors. Differential expression analysis was performed by fitting a generalised linear model using the voomLmFit function in limma, ²⁶ with the formula ~0 + Condition, where Condition was defined by both anatomical region (LA, RA, LV or RV) and group (AF or control). To account for batch effects arising from each horse contributing multiple samples (one sample per chamber), horse identity was included as a blocking factor in the differential expression analysis. Genes with a false discovery rate (FDR) <0.05 after Benjamini–Hochberg correction were considered differentially expressed. Chamber‐specific contrasts were analysed separately (e.g., RA AF vs. RA control), except for the atrial vs. ventricular comparison in healthy horses, where LA and RA were averaged and compared with averaged LV and RV samples to capture atrioventricular differences.

2.2.4. Dimension reduction and pathway enrichment

Principal component analysis (PCA) was performed on log‐transformed Counts Per Million (logCPM) values using prcomp (v3.6.1) to explore variance across samples. Batch effects attributable to horse‐specific differences (each horse contributing to four heart samples) were removed using the removeBatchEffect function in limma. ²⁷ Batch‐corrected PCA plots were generated to display clustering of samples and gene loadings contributing to variance. Gene loadings for the top principal components were extracted and ranked by their absolute values to identify key genes contributing most to the observed variance. Specific top loadings of interest were highlighted on the plots based on their established roles in cardiac electrophysiology and structure.

To identify enriched pathways, over‐representation analysis (ORA) and gene set enrichment analysis (GSEA) were performed using clusterProfiler (v4.10.1) ²⁸ with equine‐specific Gene Ontology (GO) databases. The ORA was performed separately on up‐ and downregulated genes (FDR <0.05). For GSEA, log2 fold changes from the differential expression analysis were ranked, and 1 million permutations were conducted.

2.2.5. Ion channel and calcium handling genes

For assessing the gene expression of ion channel and calcium handling proteins, raw sequencing counts were normalised using TMM (Trimmed Mean of M‐values) and log‐transformed into Counts Per Million (logCPM). Ion channel and calcium handling genes were identified from a curated, predefined list of relevant genes based on previous human literature. ²⁹ , ³⁰

2.2.6. Regional differences and overlap analysis

Region‐specific gene expression in healthy horses was assessed using the BgeeCall ³¹ package (v1.22.0), which classifies genes as present or absent based on reference intergenic sequences. A gene was considered expressed in a given region if it was detected as ‘present’ in at least 75% of healthy horse samples from that region. To visualise the overlap of detected genes across regions, an UpSet plot was generated using the ComplexUpset (v1.3.3) ³² package. Atrial‐specific genes were defined as those expressed in both atria but absent in the ventricles.

2.2.7. Overlap with human genes with genetic association to atrial fibrillation

Genes identified as differentially expressed between AF horses and controls in the LA and RA were compared with known human AF‐associated genes from GWAS studies, ³³ to the curated ion channel list, and to atrial‐selective genes identified from the UpSet plot analysis described above. Overlaps were visualised using a Venn diagram generated with ggvenn (v0.1.10).

2.2.8. Mitochondrial pathway analysis

Mitochondrial dysregulation plays a central role in human AF. ¹⁰ To investigate its involvement in equine AF, LA and RA differentially expressed genes (DEGs) between AF horses and controls were analysed using MitoXplorer 2.0, ³⁴ a bioinformatics tool that maps genes to 35 curated mitochondrial functions. The results were imported into R to visualise the top 10 mitochondrial categories, defined as those with the highest number of differentially regulated genes.

2.3. Proteomic integration

We integrated transcriptomic and proteomic data from horses in AF by matching DEGs with differentially expressed proteins from the LA and RA. The proteomic data was obtained from mass‐spectrometry analyses conducted on samples from the same cohort of horses as part of a different study. ³⁵ Only protein‐transcript pairs detected across both omics layers were included. Pearson correlation was used to assess the linear relationship between RNA and protein log2 fold changes, and a linear regression model with a t‐test was applied to evaluate whether the intercept deviated from zero.

2.4. Correlation between natural and induced atrial fibrillation

To assess overlap between the two AF models, we compared DEGs from horses with naturally occurring AF (vs. healthy controls) with those from tachypacing‐induced AF (vs. sham controls), focusing on the LA and RA. Sample processing and bioinformatics workflows were consistent across studies. However, the samples from the horses with tachypacing‐induced AF were sequenced in a separate run, preventing direct one‐to‐one comparison.

The genes detected in both datasets were compared separately for each atrium, retaining only genes differentially expressed in at least one of the AF models. The correlation of log2 fold changes between the two AF models was assessed using Pearson correlation, and a linear regression model with a t‐test was applied to evaluate whether the intercept deviated from zero. Genes showing opposing directions of regulation between natural and induced AF were identified separately for each atrium and analysed for GO term enrichment using ORA.

2.5. Metabolic characterisation

2.5.1. AMPKα2 immunoblotting

AMP‐activated protein kinase (AMPK) is a master regulator of cellular energy homeostasis and has been implicated in metabolic remodelling in human AF. ³⁶ To explore its role in equine AF, we first quantified the α2‐subunit of AMPK (AMPKα2) in RA biopsies to normalise subsequent AMPK activity measurement. Samples included RA biopsies from all horses in this study (healthy and naturally occurring AF) and 10 horses with tachypacing‐induced AF from an ongoing study.

Biopsies were homogenised (2 × 60 s at 30 Hz) in MG‐buffer/ice‐cold lysis buffer (10% glycerol, 20 mmol/L sodium pyrophosphate, 1% NP‐40, 2 mmol/L phenylmethylsulfonyl fluoride [PMSF], 150 mmol/L sodium chloride, 50 mmol/L HEPES, 20 mmol/L β‐glycerophosphate, 10 mmol/L sodium fluoride, 1 mmol/L EDTA, 1 mmol/L EGTA, 10 μg/mL aprotinin, 3 mmol/L benzamidine, 10 μg/mL leupeptin and 2 mmol/L sodium orthovanadate, pH 7.5) using steel beads in a TissueLyser II (QIAGEN, Hilden). Homogenates were rotated end‐over‐end for 30 min at 4°C and centrifuged at 18,000g for 20 min at 4°C to isolate tissue lysates. Lysates were frozen in liquid nitrogen and stored at −70°C. Total protein concentrations were measured using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, Cat. No. 23225), and diluted to equal protein concentration with Milli‐Q water. Proteins were denatured by mixing lysates with Laemmli sample buffer and resolved by SDS‐PAGE on self‐cast gels. Proteins were then transferred to PVDF membranes (Immobilon‐P, Millipore) via semi‐dry transfer. Membranes were blocked for 5 minutes in TBS‐T containing 2% non‐fat milk and incubated overnight at 4°C with an AMPKα2‐specific primary antibody (Dundee, Cat. No. DA015) at 0.2 μg/mL. After washing in TBS‐T (5 min), membranes were incubated for 45 min at room temperature with an HRP‐conjugated secondary antibody (anti‐sheep, Jackson Immuno Research, 713‐035‐147) at 0.16 μg/mL. Membranes were washed again in TBS‐T for 30 min, and chemiluminescence detection was performed using a ChemiDoc MP imaging system (BioRad). Protein bands were quantified using Image Lab software, with molecular weights verified against Protein Plus Precision All Blue standards (BioRad). Linearity of band intensities was confirmed to ensure accurate quantification.

2.5.2. AMPKα2 activity assay

The AMPKα2 activity was determined through immunoprecipitation as previously described. ³⁷ Tissue lysates (200 μg) were incubated overnight at 4°C with 5 μg of AMPKα2‐specific antibody (Dundee, Cat. No. DA015), G‐Agarose beads (Millipore) and immunoprecipitation buffer (50 mmol/L NaCl, 1% Triton X‐100, 50 mmol/L sodium fluoride, 5 mmol/L sodium pyrophosphate, 20 mmol/L Tris, pH 7.5, 500 mmol/L PMSF, 2 mmol/L dithiothreitol, 4 mg/mL leupeptin, 50 mg/mL soybean trypsin inhibitor, 6 mmol/L benzamidine and 250 mmol/L sucrose). Immunoprecipitates were collected by centrifugation at 500g for 1 min and washed sequentially, once with IP buffer, once with 6× assay buffer (240 mmol/L HEPES, 480 mmol/L NaCl, pH 7.0) and twice with 3x assay buffer (120 mmol/L HEPES, 240 mmol/L NaCl, pH 7.0).

Kinase activity was assessed using a reaction containing 74 kBq (2.0 μCi) of [33P]‐γ‐ATP, 200 μM AMP and 100 μM AMARA‐peptide (Schafer‐N, Denmark) as substrate. The reaction was terminated by adding 10 μL of 1% phosphoric acid and 20 μL of the reaction mixture was spotted onto P81 filter paper (Saint Vincent's Institute, AUS). Filter papers were washed three times with 1% phosphoric acid for 15 minutes per wash and analysed for radioactivity using a Typhoon FLA 7000 IP (GE Healthcare). The RA AMPKα2 activity was normalised to total AMPKα2 protein levels as determined by the immunoblotting of tissue lysates.

2.5.3. Atrial glycogen measurement

Atrial glycogen content was quantified in RA and LA samples using a fluorometric method to measure glycosyl units after acid hydrolysis. ³⁸ Tissue lysates (100 μg) were boiled in 200 μL of 2 N HCl for 2 h, cooled and analysed on a Pentra 400 clinical chemistry analyser (Horiba ABX Diagnostics, Montpellier, France).

2.6. Data analysis

Bioinformatics analysis of RNA‐sequencing data is detailed above. The following statistical methods apply only to biochemical measurements (glycogen and AMPK), which were performed on atrial samples. Normality was assessed through visual inspection of residual plots. Statistical comparisons for data from a single atrium were performed using one‐way ANOVA, while paired data from both atria were analysed with a mixed‐effects model. Holm–Šídák correction was applied for post‐hoc multiple comparisons. These statistical analyses were conducted using Prism 10 (GraphPad Software). A p‐value of <0.05 was considered statistically significant.

3. RESULTS

3.1. The cardiac transcriptome of healthy horses

We detected a total of 38,849 genes across all samples, and after filtering out low‐expressed genes, 14,510 were retained for further analyses (Data S1).

3.1.1. Dimension reduction and differential expression analysis

We first characterised the cardiac transcriptome of healthy horses, using only samples from the control horses. Principal component analysis showed distinct clustering of atrial and ventricular samples along the first principal component (PC1), which accounted for 56.9% of the total variance (Figure 1A). Potassium channel genes (KCNJ5, KCNK3), myosin light chain genes (MYL2, MYL7) and atrial natriuretic peptides (NPPA, NPPB) were among the key genes contributing to this separation. The second principal component (PC2) distinguished left from right ventricular samples (Figure S1A), while the third principal component (PC3) separated left and right atrial samples.

FIGURE 1.

Transcriptomic profile of the four chambers in the healthy horse heart. (A) Principal Component Analysis (PCA) of transcriptomes from all four cardiac chambers of healthy horses, showing sample distribution along principal components (PC) 1 and 3. Each point represents a sample, colour‐coded by cardiac chamber (left atrium (LA), right atrium (RA), left ventricle (LV) and right ventricle (RV)). Arrows indicate the direction and strength of selected top loadings. (B) Volcano plots displaying differentially expressed genes (DEGs) between the atria and ventricles (top) and between the right and left atria (bottom). Upregulated genes with a false discovery rate (FDR) <0.05 are shown in red, downregulated genes with a FDR <0.05 are shown in blue. Top DEGs are labelled for each comparison. (C) Chord diagram illustrating selected enriched pathways (q‐value <0.05) and relevant associated genes (FDR <0.05) from over‐representation analysis (ORA) comparing atrial and ventricular transcriptomes. Pathways are represented as segments on the outer ring to the right, and connecting lines indicate the top genes associated with each pathway. (D) Relative expression (mean log2 Trimmed Mean of M‐values (TMM)‐normalised counts per million (CPM)) of selected ion channel genes across the four heart chambers, grouped by function. Gene names in bold indicate significant differences between the atria and ventricles, while asterisks (*) denote significant differences between the left and right atria (FDR <0.05).

Differential expression analysis detected 7889 DEGs between atrial and ventricular samples at a 5% FDR threshold (Figure 1B). When comparing the right and left atria, 520 genes were differentially expressed, with genes such as HCN4 and BBOX1 showing higher expression in the RA. Comparing the left and right ventricle, 1299 genes were differentially expressed (Figure S1B).

3.1.2. Pathway enrichment analyses

To investigate the biological pathways associated with regional transcriptomic differences, we performed enrichment analyses. The ORA revealed that genes highly expressed in the atria were enriched for pathways related to the collagen‐containing extracellular matrix (ECM) and glutamatergic synapses (Figure 1C). In contrast, genes highly expressed in the ventricles were enriched for mitochondrial inner membrane components and mitochondrial fusion. GSEA further showed that the LA was enriched for mitochondrial and ribosomal pathways relative to the RA, while the LV was enriched for the ECM‐related pathways compared with the RV (Figure S1C).

3.1.3. Regional differences and ion channel expression

We next aimed to identify genes with region‐specific expression patterns by performing an overlap analysis based on the presence/absence function in Bgee. This revealed that 12,818 genes were expressed across all four chambers, while 308 genes were atrial specific and 7 were ventricular specific (Figure S1D). Notably, atrial specific genes included several potassium channel subunits (KCNJ5 (K_ir3.4), KCNN3 (K_Ca2.3), KCNA4 (K_V1.4) and KCNK17 (K_2P17.1)).

To characterise the expression of cardiac ion channels in horses, we analysed their relative expression using a curated list of ion channel genes derived from human studies. ²⁸ , ²⁹ The predominant sodium‐ and calcium channel genes were SCN5A (encoding Na_v1.5) and CACNA1C (encoding Ca_v1.2) across all four heart chambers (Figure 1D). Several voltage‐gated potassium channels were expressed in all chambers, including KCND3 (K_v4.3), KCNA5 (K_v1.5), KCNH2 (K_v11.1) and KCNQ1 (K_v7.1). Among the inward‐rectifier potassium channels, KCNJ5 (K_ir3.4) was almost exclusively expressed in the atria, whereas KCNJ2 (K_ir2.1) and KCNJ12 (K_ir2.2) showed higher expression in the ventricles. KCNN2 (K_Ca2.2) and KCNK3 (TASK‐1) were more highly expressed in the LA, while HCN4 expression was largely confined to the RA. KCND1 (K_v4.1), KCND2 (K_v4.2) and KCNJ8 (K_ir6.1) were either not expressed or expressed at levels below the filtering threshold. The cardiac expression of calcium handling genes was dominated by ATP2A2 (SERCA2), CASQ2 (Calsequestrin‐2), CALM1 (Calmodulin 1) and RYR2 (Ryanodine receptor 2) (Figure S1E).

3.2. Transcriptional changes in horses with natural atrial fibrillation

3.2.1. Dimension reduction and differential expression analysis

We next analysed the changes associated with AF by comparing the transcriptomes of each chamber between controls and horses with naturally occurring AF. The PCA revealed a clear separation between atrial and ventricular samples along the first component, accounting for 35% of the total variance, while the second component separated AF horses from controls, explaining 29% of the variance (Figure S2A). Differential expression analysis identified 1330 DEGs in the RA, 714 in the LA, 268 in the RV and 60 in the LV (FDR <0.05) (Figures 2A and S2B). Among the top downregulated genes in AF across both atria were OTUD1, ATF3 and NR4A3.

FIGURE 2.

Left‐ and right atrial changes in horses with atrial fibrillation. (A) Volcano plots displaying differentially expressed genes (DEGs) in horses with natural AF compared with controls in the left atrium (LA) (to the left) and right atrium (RA) (to the right). Upregulated genes with a false discovery rate (FDR) < 0.05 are shown in red, downregulated genes with a FDR <0.05 are shown in blue. The top DEGs are labelled for each comparison. (B) Gene set enrichment analysis (GSEA) results from the LA, depicting the enriched up‐ and downregulated pathways (q‐value <0.05), ordered by normalised enrichment score (NES). (C) Sankey diagram illustrating key pathways that were significantly upregulated (q‐value <0.05) and enriched in the RA, along with the specific genes driving these pathways (FDR <0.05), based on GSEA. * = acyl‐CoA metabolic process. (D) Venn diagram showing the overlap between DEGs in the LA and RA (FDR <0.05) and the curated list of ion channel genes from Figure 1D. Ion channel genes identified as differentially expressed are highlighted in the relevant overlaps, with upregulated genes in red and downregulated genes in blue. (E) Correlation plot comparing RA transcriptomic data from this study with proteomics data from the same cohort of horses as part of a different study. ³⁵ Each point represents a gene/protein pair differentially expressed in both datasets. Pairs upregulated in both omics layers are shown in red, downregulated genes in blue and genes regulated in opposite directions in grey. The black line represents the linear trend.

3.2.2. Pathway enrichment analyses

On pathway enrichment analyses, the DNA‐binding transcription factor activity pathway was downregulated in AF across both atria (Figure S3B). This was driven by the biatrial downregulation of transcription factors such as NR4A1, NR4A3 and TBX3. In the LA, the ventricular cardiac muscle morphogenesis and collagen fibril organisation pathways were upregulated in AF (Figure 2B). The morphogenesis pathway was driven by higher expression of sarcomeric genes, such as MYL2 and MYH7. The inflammatory pathway was downregulated in the LA. In the RA, AF was associated with an upregulation of pathways related to the extracellular region and collagen fibril organisation (Figure 2C).

In the LV, GSEA showed an upregulation of the mitochondrial inner membrane pathway (Figure S2C). In the RV, genes related to the extracellular region and growth factor activity pathway were upregulated in AF.

3.2.3. Changes in atrial specific and ion channel genes in atrial fibrillation

To identify potential atrial‐specific therapeutic targets for AF in horses, we examined the overlap between the genes identified as atrial‐specific and the DEGs in the LA and RA. This analysis revealed four atrial‐specific genes with altered expression in AF: VSNL1 and HTR1D were upregulated, while RBM11 and SCN1A were downregulated (Figure S3C). Next, we assessed the overlap between DEGs and cardiac ion channel genes. In the LA, KCNJ4 (K_ir2.3) was downregulated in AF (Figure 2D). Across both atria, SCN1A (Na_v1.1) was downregulated in AF, while GJA5 (Connexin40) and KCNE1 (MinK) were upregulated in the RA. CACNA1C (Ca_v1.2) expression and SCN8A (Na_v1.6) were downregulated in the RA.

3.2.4. Overlap with human genes associated with AF

We identified 12 differentially regulated genes in horse AF that have been associated with AF in human GWAS. ³² Four genes were overlapping across both atria, including the two transcription factors TBX3 (downregulated in horse AF), TBX5 (upregulated in horse AF) (Figure S3D).

3.2.5. Mitochondrial phenotyping

The RA and LA DEGs were annotated with MitoXplorer2.0 categories ³³ to evaluate the mitochondrial response to AF. Overall, most mitochondrial‐related genes were upregulated in AF, particularly those involved in mitochondrial translation and oxidative phosphorylation (Figure S4A). However, several nuclear‐encoded mitochondrial regulatory genes, including PPARGC1A (PGC‐1α), were downregulated in the RA.

3.2.6. Integrative transcriptomic and proteomic analysis

We performed a multi‐omics integration to identify molecular changes consistently reflected at both the transcriptomic and proteomic levels in the atria of AF horses. The DEGs identified in this study were compared with differentially expressed proteins from the same cohort of horses as part of a different study ³⁵ In the RA, we identified 28 significantly differentially expressed gene‐protein pairs present in both datasets, with 13 upregulated and eight downregulated across both datasets (Figure 2E). The linear correlation between transcriptomic and proteomic log2 fold changes was moderate (Pearson's r = 0.49, p = 0.008). In the LA, we identified 18 differentially regulated gene‐protein pairs, with nine upregulated and three downregulated across both datasets (Figure S4B). The correlation was weak and non‐significant (Pearson's r = 0.3, p = 0.09). Notably, NT5DC1, PFKP and PYGB were consistently upregulated in both atria across both omics layers (Figure S4C).

3.3. Comparison between natural and tachypacing‐induced atrial fibrillation

3.3.1. Correlation between atrial changes in natural and induced atrial fibrillation

To compare transcriptomic responses, log2 fold changes from natural AF (current study) were correlated with those from horses with 4 months of tachypacing‐induced AF (ongoing study).

In the LA, 1092 genes were differentially expressed in either natural or induced AF, with a strong linear correlation in log2 fold changes between the two conditions (Pearson's r = 0.66, p < 0.001) (Figure 3A). The majority of genes (85%) were regulated in the same direction in both conditions. Among the remaining genes, 87 genes were upregulated in natural AF but downregulated in induced AF, including ECM‐associated genes (e.g., COL1A1 and COL1A2). This subset of genes was enriched for the canonical Wnt‐signalling pathway. Conversely, 74 genes were downregulated in natural AF but upregulated in induced AF, including mitochondrial genes like SUCLG2 and PDK4, and this set of genes was enriched for the mitochondrion GO term.

FIGURE 3.

Comparison between natural and tachypacing‐induced atrial fibrillation and associated metabolic changes. (A) Correlation plot showing the relationship between log2FC of differentially expressed genes (DEGs) in the left atrium (LA) (top) and right atrium (RA) (bottom) between naturally occurring and induced atrial fibrillation (AF). Each point represents a gene, with blue indicating genes downregulated in both natural and induced AF, red indicating genes upregulated in both, and grey indicating genes with opposite regulation trends. Black lines indicate linear trends. Insets in the corners show the top enriched pathways (q‐value <0.05) from over‐representation analysis (ORA), with green dots for pathways upregulated in induced AF but downregulated in natural AF, and yellow dots for the reverse trend. (B) AMPKα2 activity normalised to total AMPKα2 protein in horses with natural AF (red), induced AF (blue) and controls (grey). Statistical comparisons were performed using one‐way ANOVA with Holm–Šidák post hoc correction. (C) Atrial glycogen content in the LA and RA, analysed using a mixed‐effects model with Holm–Šidák post hoc correction. Bars represent mean ± standard deviation. Asterisks indicate significance levels (*p < 0.05, **p < 0.01).

In the RA, 2626 genes were differentially expressed in either natural or induced AF, with a moderate correlation between log2 fold changes (Pearson's r = 0.49, p < 0.001). Most genes (78%) were regulated in the same direction in both conditions. A total of 453 genes were upregulated in natural AF and downregulated in induced AF, including ECM genes (e.g., COL1A1, LOXL1) and natriuretic peptide hormone genes (NPPA, NPPB). This subset was enriched for GTP‐binding. Conversely, 124 genes were downregulated in natural AF but upregulated in induced AF, including mitochondrial (e.g., NDUFAF5, SUCLG2) and sarcomeric genes (e.g., MYL3 and MYOM3). This subset of genes was enriched for mitochondrial pathways.

3.3.2. AMP‐activated kinase activity and atrial glycogen content

To further explore the metabolic differences between natural and induced AF, we measured atrial AMPK activity and glycogen content. In the RA, the activity of the AMPKα2 subunit (mean ± SD) was significantly lower in horses with induced AF (14.35 ± 4.87 pmol/μg protein) compared with both naturally occurring AF (29.73 ± 16.02, p = 0.04) and to healthy controls (30.68 ± 13.59, p = 0.03) (Figure 3B). However, RA glycogen content did not differ significantly among the groups, with values of 1124 ± 166 pmol/μg protein in controls, 1118 ± 362 in naturally occurring AF, and 1162 ± 244 in induced AF (Figure 3C).

In the LA, glycogen content was significantly higher in both naturally occurring AF (1262 ± 269 pmol/μg protein, p = 0.03) and induced AF (1367 ± 255 pmol/μg protein, p = 0.006) compared with controls (930 ± 286 pmol/μg protein).

4. DISCUSSION

Our study is the first to profile the transcriptome of all four heart chambers in healthy horses and horses with naturally occurring persistent AF. We show that the ion channel expression in horses is largely similar to humans, despite a few key differences in the expression of potassium channels responsible for early ventricular repolarisation. In horses with persistent AF, we observed significant transcriptional changes in pathways related to ECM, myofibrils and metabolism, while ion channel changes were minimal. Glycolytic activity was elevated in both atria, accompanied by increased glycogen levels in the LA. The transcriptional profile of horses with experimentally induced AF correlated well with that of horses with naturally occurring disease, particularly in the LA. Together, these findings provide a comprehensive molecular characterisation of AF in horses and identify potential targets that could be relevant for future AF treatments.

4.1. Novel transcriptional characterisation of the horse heart chambers

The four heart chambers had distinct transcriptional identities, which likely reflect their specialised physiological roles: the mitochondrial enrichment in the ventricles, especially the LV, indicates greater oxidative capacity supporting pump function, ³⁹ while ECM enrichment in the atria aligns with higher atrial collagen content observed in healthy humans ⁴⁰ and horses. ⁷ The chamber‐specific gene expression in horses resembled that in humans, with higher atrial expression of NPPA and MYL4, and higher ventricular expression of IRX5, MYH7 and HEY2. ⁴¹ Myosin isoform expression also followed expected patterns, with high atrial expression of MYH6 (α‐myosin heavy chain) supporting fast‐twitch contractions for atrial systole, and high ventricular expression of MYH7 (β‐myosin heavy chain) reflecting the greater contractile demand of the ventricles. ⁴²

The expression of ion channels involved in depolarising inward currents and calcium handling genes in healthy horses was largely similar to that of humans. ⁴³ There were, however, a few important differences in the ventricular expression of ion channels involved in early repolarisation. Ventricular KCNA4 (K_V1.4) expression, the molecular correlate of the I _to,slow current, was lower compared with humans. ⁴³ The I _to,slow current is responsible for the transient repolarisation immediately following depolarisation and shows strong regional variation within the ventricles of other species. ⁴⁴ , ⁴⁵ Whether the low expression in horses reflects regional variation or a species‐specific property of the equine action potential remains unclear. On the other hand, horses had a high ventricular expression of KCNA5 (K_v1.5), consistent with previous results from our group. ¹⁸ In humans, KCNA5 is atrial‐specific, where it encodes the I _Kur current that contributes to atrial repolarisation. ⁴³ Its ventricular expression in horses indicates a potential role of K_v1.5 in early ventricular repolarisation, aligning with previous observations. ⁴⁶ GJA5, encoding connexin 40 (Cx40), also showed a high expression in the ventricular myocardium. While Cx40 is restricted to the atria and conduction system after embryogenesis in other species, ⁴⁷ its high ventricular expression in horses may reflect the specialised equine conduction system, including a dense intramural network of Purkinje fibres. ⁴⁸ Despite these differences in early ventricular repolarisation and gap junction genes, the expression of late repolarisation channels (e.g., KCNH2 (I _Kr) and KCNQ1 (I _Ks)), closely resembled that of humans. In humans, dysfunction of these channels often underlies long QT syndrome, a disorder marked by delayed ventricular repolarisation and increased risk of ventricular arrythmias and sudden cardiac death (SCD). ⁴⁹ The shared expression levels support the notion that horses possess the molecular basis for long QT syndrome. ⁵⁰ , ⁵¹ The ion channel profile presented here establishes a critical baseline for future research into SCD in racehorses.

4.2. Potential atrial specific targets for atrial fibrillation

Atrial‐selective potassium channel blockers are actively investigated for use in the treatment of AF, as they are expected to prolong the atrial action potential without the risk of pro‐arrhythmic ventricular side effects. ⁵² Here, we identified KCNJ5 (K_ir3.4, I _K,ACh) and KCNN3 (K_Ca2.3, SK3 channel) as truly atrial specific in horses. KCNJ5, encoding the I _K,ACh current, becomes constitutively active during AF in humans, contributing to pro‐arrhythmic action potential shortening. ⁵³ Our previous work has shown that selective I _K,ACh blockade effectively terminates short‐term AF in horses, ⁵⁴ and the atrial selectivity shown here supports its suitability as a therapeutic target. The expression of small conductance calcium‐activated potassium channels (K_Ca2.x, SK‐current) in horses differs from that in humans. SK1 was the dominant subtype in both atria and ventricles, whereas it is expressed at low levels in humans. ¹⁵ SK2 channels showed a left‐sided predominance in the horse heart (fourfold higher expression in the LA than RA), while SK3 was atrial specific, but the least expressed SK subtype. Although SK channel gene expression does not correlate well with current function in humans, ⁵⁵ we have previously shown that a general SK‐channel blocker can terminate AF in horses without affecting ventricular repolarisation, ¹⁸ indicating functional atrial specificity in the horse.

Although KCNK3 (K_2P3.1, TASK‐1) did not meet the strict threshold for atrial selectivity in our dataset, its ventricular expression was 200‐fold lower than in the atria. This steep atrial–ventricular gradient could suggest potential for atrial‐specific targeting. Studies in humans and pigs have shown promise for TASK‐1 channel blockade in AF, ⁵⁶ , ⁵⁷ and their efficacy in horses is currently under investigation. ⁵⁸ In contrast, another K_2P‐channel, KCNK17 (K_2P17.1, TASK‐4), showed true atrial‐specific expression in our dataset, consistent with reports in humans, ⁵⁹ where it has also recently gained attention as a potential AF drug target. ⁶⁰

4.3. Atrial fibrillation induced atrial remodelling

4.3.1. Ion channel remodelling

We observed few changes in the expression of atrial ion channels in horses with naturally occurring AF (Figure 4), indicating that ion channel remodelling at the transcriptional level may play a limited role in the early pathophysiology of AF in horses. We did, however, observe signs of connexin remodelling, with upregulation of GJA5 (Cx40) in the RA. Given that both up‐ and downregulation of connexins have been reported in AF, ⁶¹ and subcellular distribution may be more important than expression levels, ⁶² the functional consequence of these changes remains unclear.

FIGURE 4.

Cardiac action potentials and the expected underlying currents based on human data with atrial expression changes in equine atrial fibrillation (AF). The atrial and ventricular action potentials are composed of four phases: Depolarisation (0), early repolarisation (1), plateau (2) and late repolarisation (3) before resting (4). Inward depolarising currents that drive these phases in humans are depicted in orange, while outward repolarising currents are shown in blue. The corresponding α‐subunits and genes responsible for each current are based on data from humans ⁶⁸ and horse. ⁴⁶ Arrows indicate the changes in mRNA expression of ion channels associated with the action potential in horses with naturally occurring (2–12 weeks) and tachypacing‐induced AF (4 months). The cross (^†) indicates that CACNA1C was only downregulated in the right atrium. The asterisk (*) denotes that KCNJ4 was downregulated in the left atrium in natural AF, and KCNJ2 was downregulated in the right atrium in induced AF. The dotted line indicates absence of channel function.

KCNE1 (MinK), the β‐subunit for KCNQ1 (K_v7.1), ⁶¹ which also interacts with KCNH2 (K_v11.1), ⁶³ was upregulated in the RA during natural AF and in both atria during induced AF. In humans, gain‐of‐function mutations in KCNE1 have been linked to AF, ⁶⁴ and its upregulation in horse AF could contribute to increased delayed rectifier current, promoting action potential shortening and AF perpetuation. In parallel, we observed downregulation of CACNA1C (Ca_v1.2) in the RA, a hallmark of AF‐induced ion channel remodelling in humans, where reduced L‐type calcium current similarly contributes to shortening of the action potential. ⁶⁵ We did not assess the functional properties of these ion channels in horses. Future studies employing techniques such as whole‐cell patch‐clamping are needed to elucidate current properties.

Given the limited ion channel remodelling, we further assessed the DEGs for atrial specificity and identified VSNL1 as both atrial specific and biatrially upregulated in AF. VSNL1 is a sinus node marker in mice that regulates calcium signalling and heart rate. ⁶⁶ It has also been identified as a hub gene for AF in humans, ⁶⁷ suggesting a conserved role in atrial automaticity with possible relevance to AF pathophysiology across species.

4.3.2. Transcription factor remodelling

Cardiac transcription factors (TFs) can also regulate electrophysiology through altered expression or function of ion channels. ⁶⁷ In horses with AF, several cardiac‐specific TFs were differentially regulated. The T‐box TFs TBX3 and TBX5 have been genetically linked to AF in humans, ³² and interact with numerous ion channels in other species. ⁶⁹ In this study, TBX5 was upregulated and TBX3 was downregulated across both atria in horses with AF, which would be expected to affect cardiac conduction velocity ⁷⁰ and thus potentially contribute to AF susceptibility.

The nuclear receptor subfamily TFs NR4A1 and NR4A3 were also markedly downregulated in both atria of horses with AF. NR4A1 is known to regulate glycolysis and fibrotic remodelling. ⁷¹ , ⁷² NR4A3, although less studied, showed a 30‐fold downregulation in both atria of horses with AF, aligning with the substantial downregulation observed in human AF patients ⁷³ and in rapid atrial pacing pigs. ⁷⁴ In mice, NR4A3 knockdown induces atrial hypertrophy, fibrosis and increases AF susceptibility, potentially by modulating atrial energy metabolism. ⁷⁵ With its strong atrial specificity in horses (90‐fold higher expression in atria compared with ventricles), downregulation in both natural and induced AF and mechanistic link to atrial remodelling in other animals, the NR4A3 pathway could be an atrial‐selective target for preventing or reversing AF‐induced remodelling.

4.3.3. Extracellular matrix remodelling

Beyond the electrical disturbances, AF progression in humans is also perpetuated by structural atrial remodelling, specifically through changes in the ECM. ⁶ Consistent with this, AF horses showed biatrial enrichment of ECM‐related pathways and upregulation of key cardiac collagen genes such as COL1A1, although not all collagen transcripts were elevated. Interestingly, inflammation‐related pathways, known to drive fibrogenesis and play a role in human AF arrhythmogenesis, ⁷⁶ were downregulated in horses with AF. This finding may reflect differences in disease aetiology, as horses develop AF in the absence of the comorbidities and metabolic dysfunction common in human patients, ¹⁰ which are often associated with chronic low‐grade inflammation.

4.3.4. Metabolic remodelling

Metabolic dysfunction is a well‐established risk factor for AF in humans, ⁷⁷ but AF‐induced metabolic remodelling has only recently been recognised as a contributor to arrhythmogenesis. ¹⁰ In the healthy adult myocardium, fatty acid oxidation is the primary energy source accounting for ~40%–60% of mitochondrial ATP production. ⁷⁸ Human studies suggest that AF promotes glycolysis while suppressing fatty acid oxidation. ⁹ , ⁷⁹ In contrast, horses with naturally occurring AF showed upregulation of genes involved in fatty acid metabolism and oxidative phosphorylation in both atria. However, the glycolytic genes PFKP and PYGB were also consistently upregulated at both transcript and protein levels in both atria. PFKP, a stress‐induced glycolytic enzyme, is linked to pathological hypertrophy in failing cardiomyocytes, ⁸⁰ while PYGB is a rate‐limiting glycolytic enzyme that has also been associated with human AF. ⁸¹ These signs of increased glycolysis could reflect a reprogramming of atrial cardiomyocytes towards a foetal‐like phenotype, which relies more heavily on glucose as an energy substrate. ⁷⁸ Similar metabolic shifts have been observed in human AF, ⁷⁹ where it is often accompanied by local glycogen accumulation. ⁸² The elevated LA glycogen levels observed in horses with AF in this study further support foetal reprogramming as a component of the atrial metabolic response in equine AF.

4.4. Comparison to natural atrial fibrillation

The metabolic responses differed slightly between experimentally induced and natural AF. In induced AF, there was an even more pronounced upregulation of mitochondrial and glycolytic pathways, possibly reflecting a longer duration of disease. In human AF, the metabolic response evolves over time. Initially, there is a compensatory phase, characterised by increased activity of the master metabolic regulator, AMPK and enhanced ATP production. Acting as a cellular fuel gauge, AMPK detects energy depletion from the high atrial rates and restores metabolic homeostasis by activating ATP‐generating catabolic pathways, while inhibiting ATP‐consuming anabolic pathways. ⁹ Over time, this metabolic response decompensates with declining AMPK activity and metabolic stress from an energy demand–supply mismatch. ⁹ Consistent with this progression, AMPK activity was also preserved in horses with short‐term natural AF but diminished in long‐term induced AF. Ongoing research in our group will clarify whether AMPK activation can protect against atrial remodelling in horses, as it has been suggested in humans. ⁹ , ⁸³

The difference in disease duration between the two models offers clues into the cause and consequence of AF. Natural AF includes both the initial trigger and early remodelling, while induced AF isolates the remodelling driven by long‐term high atrial rates. For instance, collagen 1a1 (COL1A1), one of the principal cardiac fibrillar collagens, ⁸⁴ was upregulated in both atria during natural AF but not in induced AF. This raises the question of whether atrial fibrosis in equine AF is solely a consequence of high atrial rates or rather a causative factor contributing to arrhythmia development—a question that is also being debated in human cardiology. ⁸⁵

Despite metabolic and fibrotic differences, the strong transcriptional correlation between the two AF types indicates that the experimental model of AF replicates many molecular aspects of atrial remodelling in natural AF, supporting its overall clinical relevance and applicability to the general equine AF population.

5. LIMITATIONS

While this study is the first to perform RNA‐sequencing on all four chambers of the equine heart, it has several important limitations. The relatively small sample size and variability in sex and breed composition between groups may limit the generalisability of the findings. Additionally, the presence of other conditions, such as mitral valve regurgitation and LA enlargement in some AF horses, raises the possibility that the observed remodelling may not be solely attributable to AF. The use of single time point analysis restricts our ability to capture the dynamic nature of AF changes during its progression, potentially missing molecular changes occurring in earlier or later stages of disease. Moreover, this study cannot directly distinguish gene expression changes predisposing to AF from those resulting from it and therefore cannot establish causality.

While this study provides important and novel insights into the molecular mechanisms of AF in horses, future research should validate the targets suggested here, ideally using a longitudinal approach to better capture the dynamics of disease progression. Integrating transcriptomic analyses with protein expression and functional data would also provide a more comprehensive understanding of the molecular mechanisms underlying AF.

6. CONCLUSION

Although AF is an electrical disturbance, our transcriptomic analysis revealed limited changes in the expression of ion channels but substantial, previously undocumented upregulation of fibrotic, metabolic and myofibrillar pathways. Current treatments target the electrical mechanisms of the disease, but addressing these other aspects of atrial substrate remodelling could be essential for improving treatment outcomes and reducing AF recurrence.

FUNDING INFORMATION

S.L.H. was funded by a research grant from the Danish Cardiovascular Academy, which is funded by the Novo Nordisk Foundation (Grant number NNF20SA0067242) and the Danish Heart Foundation. S.D.N. was funded by the Independent Research Fund Denmark (Grant number 1032‐00053B). M.J.S. was funded by a research grant from the Danish Cardiovascular Academy, which is funded by the Novo Nordisk Foundation (Grant number PhD 2022002‐HF) and the Danish Heart Foundation.

CONFLICT OF INTEREST STATEMENT

K. M. H. is an employee and shareholder of Novo Nordisk A/S.

AUTHOR CONTRIBUTIONS

Simon Libak Haugaard: Conceptualization; investigation; writing – original draft; methodology; visualization; writing – review and editing; software; formal analysis; project administration. Sarah Dalgas Nissen: Conceptualization; investigation; funding acquisition; writing – review and editing; methodology; supervision. Mélodie Jil Schneider: Conceptualization; writing – review and editing; methodology. Jesper Bratz Birk: Investigation; writing – review and editing; methodology. Helena Carstensen: Conceptualization; investigation; writing – review and editing; methodology. Charlotte Hopster‐Iversen: Conceptualization; investigation; writing – review and editing; methodology. Ali Altıntaş: Methodology; writing – review and editing; visualization; software; formal analysis; data curation; supervision. Romain Barrès: Resources; supervision; writing – review and editing; formal analysis. Rasmus Kjøbsted: Investigation; methodology; writing – review and editing; formal analysis. Jørgen F. P. Wojtaszewski: Investigation; writing – review and editing; methodology; formal analysis; resources. Kate M. Herum: Conceptualization; investigation; methodology; writing – review and editing; visualization; supervision. Thomas Jespersen: Supervision; resources; methodology; writing – review and editing; investigation; conceptualization. Rikke Buhl: Conceptualization; investigation; funding acquisition; methodology; writing – review and editing; project administration; supervision; resources.

DATA INTEGRITY STATEMENT

Simon Libak Haugaard takes full responsibility for the integrity of the data and the accuracy of the data analysis.

ETHICAL ANIMAL RESEARCH

The study was approved by the local ethical committee (approval number 2022–005). Authorisation was also granted by the Danish Animal Experiments Inspectorate (licence numbers 2016‐15‐0201‐00425 and 2016‐15‐0201‐01128).

INFORMED CONSENT

Informed consent was obtained from owners prior to the inclusion of client‐owned horses.

Supporting information

Data S1. Supporting Information.

Figure S1. Transcriptional profiling of the four cardiac chambers of healthy horses. (A) Principal Component Analysis (PCA) of transcriptomes from all four cardiac chambers of healthy horses, showing sample distribution along principal components (PC) 1 and 2 (top) and 2 and 3 (bottom). Each point represents a sample, colour‐coded by cardiac chamber: left atrium (LA), right atrium (RA), left ventricle (LV) and right ventricle (RV). Arrows indicate the direction and strength of selected top gene loadings contributing to variability in each component. (B) Volcano plot showing differentially expressed genes between the right and left ventricles. Upregulated genes with a false discovery rate (FDR) <0.05 are shown in red, downregulated genes with a FDR <0.05 are shown in blue. (C) Top 10 enriched Gene Ontology (GO) pathways from gene set enrichment analysis (GSEA), comparing right and left atria (top) and right and left ventricles (bottom). Pathways with a q‐value <0.05 were considered significantly enriched. (D) UpSet plot showing the number of robustly expressed genes overlapping between the four cardiac chambers of the heart. A detection threshold of present in at least 75% of samples per region was used to define region‐specific gene expression patterns. (E) Relative (mean log2 Trimmed Mean of M‐values (TMM)‐normalised counts per million (CPM)) of selected calcium handling genes across the four heart chambers. Gene names in bold indicate significant differences between the atria and ventricles, while asterisks (*) denote significant differences between the left and right atria (FDR <0.05).

Figure S2. Transcriptional changes in the ventricles of horses with natural atrial fibrillation. (A) Principal Component Analysis (PCA) of transcriptomes from all four cardiac chambers of healthy horses and horses with atrial fibrillation (AF), showing sample distribution along principal components (PC) 1 and 2 (top) and PCs 1 and 3 (bottom). Each point represents a sample, colour‐coded by cardiac chamber: left atrium (LA), right atrium (RA), left ventricle (LV) and right ventricle (RV). Shapes indicate heart rhythm status (triangles for AF, circles for controls in sinus rhythm). Arrows represent the direction and strength of top gene loadings contributing to variability in each component. (B) Volcano plot showing differentially expressed genes between AF horses and controls in the LV (top) and RV (bottom). Upregulated genes with a false discovery rate (FDR) <0.05 are shown in red, downregulated genes with a FDR <0.05 are shown in blue Top differentially expressed genes are labelled. (C) Top 10 enriched Gene Ontology (GO) pathways identified from gene set enrichment analysis (GSEA) comparing AF to control horses in the LV (left) and RV (right). Pathways with a q‐value <0.05 were considered significantly enriched.

Figure S3. Transcriptional changes in the atria of horses with natural atrial fibrillation. (A) Top 10 enriched Gene Ontology (GO) pathways identified from over‐representation analysis (ORA) comparing AF to control horses in the left atrium (LA) (top) and right atrium (RA) (bottom). Pathways with a q‐value <0.05 were considered significantly enriched. (B) Top 10 enriched Gene Ontology (GO) pathways identified from gene set enrichment analysis (GSEA) comparing AF to control horses in the RA. Pathways with a q‐value <0.05 were considered significantly enriched. (C) Venn diagram illustrating the overlap between differentially expressed genes in the LA and RA (FDR <0.05) and the atrial‐specific genes identified in Figure S1D. Atrial‐specific genes that were differentially expressed in both chambers are highlighted in the relevant overlap, with biatrially upregulated atrial specific genes shown in red and downregulated genes shown in blue. (D) Venn diagram illustrating the overlap between differentially expressed genes (horses, FDR <0.05) and AF‐associated genes identified from human genome‐wide association studies (GWAS). Human GWAS targets that were also differentially expressed in horses with AF are highlighted in the relevant overlaps, with upregulated horse genes shown in red and downregulated genes shown in blue.

Figure S4. Mitochondrial response to atrial fibrillation and proteomic integration. (A) Top 10 mitochondrial categories with the highest number of differentially expressed genes in the left atrium (LA) (left) and right atrium (RA) (right). Categories with upregulated genes are shown in full colour and extend to the right, while categories with downregulated genes are displayed in a transparent shade and extend to the left. The mitochondrial categories are based on the human MitoXplorer 2.0 gene annotation. (B) Correlation plot showing the relationship between LA RNA‐seq data from this study and proteomics data from a previous publication on horses. ³⁴ Each dot represents a gene/protein pair that was detected as differentially expressed in both datasets. Genes upregulated in both omics layers are shown in red, those downregulated in both layers are shown in blue and genes regulated in opposite directions are shown in grey. The black line represents the linear trend between log2FC values in RNA‐seq and proteomics (Pearson's r = 0.23, p = 0.27). (C) Venn diagram showing the overlap between LA and RA upregulated gene/protein pairs identified from the correlation analysis. The overlap indicates gene/protein pairs that were differentially expressed and upregulated at both the gene level in this study and at the protein level in a previous study.

Table S1. Baseline characteristics of horses included in the study. This table summarises the baseline characteristics of the horses used in the study, including their group (Control or AF), sex, breed, age, weight and estimated duration of atrial fibrillation (AF). AF duration was estimated based on owner observations. Findings from cardiac auscultation and echocardiographic examinations are included. Missing values are indicated with a dash (–). Comparisons between groups were performed using Fisher's exact test for categorical variables (sex, breed) and t‐tests for continuous variables (age, weight, LA_diameter).

Table S2. Baseline characteristics of tachypacing‐induced AF horses. This table summarises the baseline characteristics of the horses used for tachypacing induced comparison, including their group (sinus rhythm sham‐operated controls or tachypacing induced AF), sex, breed, age and weight. Comparisons between groups were performed using Fisher's exact test for categorical variables (sex) and t‐tests for continuous variables (age, weight).

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available in GitHub at https://github.com/simonhaugaard/Natural_AF and via https://dataview.ncbi.nlm.nih.gov/object/PRJNA1198487?reviewer=p7cr2bk6bpskipuek60b8tfmpt.