Multi-omics profiling of circulating proteins and metabolites in persistent atrial fibrillation: VDAC1 identified as a candidate circulating protein Authors: Jun Wang, Xiaozhen Chen, Zilan Zhang, Chun Gui Published: 2026 ABSTRACT BackgroundAtrial fibrillation (AF) is the most common heart rhythm disorder worldwide. Changes in metabolism and damage to mitochondria affect left atrial structure and electrical function. Few studies have applied multi-omics to plasma from AF patients.MethodsWe enrolled 37 patients with persistent AF and 36 with supraventricular tachycardia (SVT). Blood samples were drawn from the coronary sinus in all patients. Ten subjects from each group were randomly selected for 4D-DIA proteomics and untargeted metabolomics. ELISA validation was done on 35 patients per group. Four GEO atrial tissue datasets and one single-nucleus RNA-seq dataset were used for external support. Differentially expressed proteins (DEPs) were defined by |fold change| ≥ 1.5 and Benjamini-Hochberg adjusted P < 0.05. Protein-protein interaction networks were built with STRING and Cytoscape.ResultsWe quantified 3268 plasma proteins and identified 217 DEPs in AF. Network analysis showed voltage-dependent anion channel 1 (VDAC1) as a central mitochondrial hub. It connected respiratory chain subunits and autophagy regulators. ELISA confirmed higher plasma VDAC1, citrate synthase(CS) and secreted frizzled-related protein 2(SFRP2) in AF patients (all P < 0.001). Metabolomics identified 505 differentially abundant compounds. Glycerophospholipid and fatty acid pathways were disrupted. Integrated analysis revealed eight shared dysregulated pathways linked to energy metabolism. Four GEO datasets showed higher atrial VDAC1 transcripts in AF (P < 0.05). Single-nucleus RNA-seq localized VDAC1 to cardiomyocytes.ConclusionsPlasma protein and metabolite profiles differ between persistent AF and SVT. Mitochondrial and lipid pathways are both affected. VDAC1 is elevated in plasma, atrial tissue. These findings are supported across multiple cohorts. Introduction Atrial fibrillation affects more than 30 million patients worldwide [1]. AF triples the risk of stroke and doubles mortality in patients with heart failure [2]. Despite treatment with catheter ablation and antiarrhythmic drugs, about one-third of patients experience recurrence within one year [3]. The molecular mechanisms underlying AF remain incompletely understood, limiting the development of new therapies. Cardiomyocytes rely on mitochondria for ATP production. Prolonged mitochondrial damage depletes ATP, increases reactive oxygen species and calcium homeostasis. All these changes impair cardiac electrical stability [4]. Mitophagy clears damaged mitochondria, thereby maintains organelle function. Previous studies in human atrial tissue have shown reduced autophagic activity in AF patients, leading to the accumulation of dysfunctional mitochondria and exacerbation of atrial pathological changes [5], [6]. Multi-omics profiling can help characterize the molecular mechanisms of AF. Integrative transcriptomics, proteomics, and metabolomics analyses of myocardial biopsies have revealed decreased fatty acid oxidation and increased oxidative stress in AF patients [7]. Tissue-based studies are not suitable for population-level screening, whereas plasma testing is less invasive. Single-nucleus transcriptomics can capture cell type-specific changes that may be missed in bulk tissue analysis. Dataset GSE255612 contains 18 AF patients and 16 controls, and it identified 15 cardiac cell subtypes. Cardiomyocytes and macrophages showed the most marked transcriptional changes during arrhythmia [8]. We conducted paired proteomic and metabolomic analyses on plasma from 10 AF and 10 SVT patients. Patients with SVT and no history of AF were recruited as a tachycardia-matched control group. This minimized heart rate as a confounding factor. An additional 25 subjects per group were included for ELISA validation. Total was 35 per group. Based on protein-protein interaction network ranking and functional enrichment, we focused on VDAC1. VDAC1 is an outer mitochondrial membrane protein. It is involved in energy metabolism and autophagy regulation. We analyzed four AF transcriptome datasets from GEO. We wanted to examine VDAC1 expression in atrial tissue. VDAC1 is located on the outer mitochondrial membrane. It participates in calcium and metabolite transport. It also plays roles in apoptosis and cellular stress responses [9]. Previous studies have shown that VDAC1 overexpression leads to intracellular calcium overload [9]. Combining our data with public transcriptional evidence, we propose a hypothesis. Atrial cardiomyocytes under stress may upregulate VDAC1. They may release it into the circulation via extracellular vesicles (EVs). Coronary sinus blood partly reflects myocardial metabolic activity. Fig. 1 shows the study workflow. Plasma from 37 AF and 36 SVT patients was divided into a discovery cohort (n = 10 per group) for proteomics and metabolomics. An expanded cohort (n = 25 additional per group; 35 total per group) was used for ELISA. External support used one RNA-seq dataset (GSE282504), three microarray datasets, and single-nucleus RNA-seq data (GSE255612). Integrated multi-omic analysis identified VDAC1 as a candidate marker of AF-related mitochondrial injury.Fig. 1Study workflow. Plasma from 37 AF and 36 SVT patients was divided into a discovery cohort (n = 10 per group) for proteomics and metabolomics. An expanded cohort (n = 25 additional per group; 35 total per group) was used for ELISA. External support used one RNA-seq dataset (GSE282504), three microarray datasets, and single-nucleus RNA-seq data (GSE255612). Integrated multi-omic analysis identified VDAC1 as a candidate marker of AF-related mitochondrial injury.Fig. 1 Study workflow. Plasma from 37 AF and 36 SVT patients was divided into a discovery cohort (n = 10 per group) for proteomics and metabolomics. An expanded cohort (n = 25 additional per group; 35 total per group) was used for ELISA. External support used one RNA-seq dataset (GSE282504), three microarray datasets, and single-nucleus RNA-seq data (GSE255612). Integrated multi-omic analysis identified VDAC1 as a candidate marker of AF-related mitochondrial injury. Methods 1. Study population 1. Study population Study population We enrolled 37 patients with persistent AF. We also enrolled 36 patients with SVT. SVT types included atrioventricular nodal reentrant tachycardia and accessory pathway-mediated tachycardia. All underwent elective electrophysiological study or cardiac surgery at our center. Persistent AF was diagnosed according to the 2023 ACC/AHA/ACCP/HRS and 2024 ESC guidelines [10], [11]. Diagnosis required 12‑lead ECG showing absent P waves, fibrillatory baseline, and irregular RR intervals. SVT was diagnosed per the 2019 ESC guideline [12]. All SVT patients were in sinus rhythm at enrollment. None had a history of AF. Exclusion criteria were acute severe infection, malignancy, severe hepatic or renal dysfunction, and myocardial infarction within three months before blood collection. All participants provided written informed consent. The institutional review board approved the study (approval number: 2025-E0226). Plasma samples were stored at −80 °C. Ten participants per group were randomly selected for discovery-phase multi-omic testing. ELISA was performed on all 35 AF and 35 SVT samples, which included the original 10 discovery samples plus 25 additional per group. SVT patients were selected to reduce heart rate-related confounding. Coronary sinus blood was collected during interventional procedures. Because this approach is invasive, we did not include healthy subjects. The limitations of this strategy are addressed in the Discussion.2.Sample collection 2.Sample collection Sample collection Coronary sinus blood was collected into EDTA-containing tubes at the start of the procedure. This was done before ablation or electrical stimulation. Plasma was separated within two hours by centrifugation at 3000 ×g for 15 min at 4 °C. Samples were aliquoted and stored at −80 °C.3.Proteomics 3.Proteomics Proteomics Plasma from 10 AF and 10 SVT patients was processed following published diaPASEF protocols [13], [14]. High-abundance proteins were removed using ProteoMiner™ kits (Bio-Rad). Protein concentration was measured by BCA assay. Samples were digested with trypsin overnight at 37 °C. The enzyme:protein ratio was 1:50. Peptides were desalted on C18 columns. Peptides were separated on a nanoElute UHPLC system. This was coupled to a timsTOF Pro2 mass spectrometer. An IonOpticks C18 column (25 cm × 75 μm, 1.6 μm) was used at 50 °C. The mobile phases consisted of 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B). The gradient ran for 40 min at 300 nL/min. Data were acquired in diaPASEF positive ion mode with 48 isolation windows. The m/z range was 100–1700. Ion mobility was 0.85–1.3 V·s/cm2. Raw data were processed with DIA-NN v1.8.1. The search was against the UniProt human database UP000005640 [13]. Precursor and protein FDR were set at 1%. Relative quantification used the MaxLFQ algorithm. DEPs were defined by |fold change| ≥ 1.5 and Benjamini-Hochberg adjusted P < 0.05.4.Metabolomics 4.Metabolomics Metabolomics The same 20 discovery samples were extracted with methanol-acetonitrile. Internal standards were included. Samples were vortexed for 3 min. They were centrifuged at 12,000 rpm for 10 min at 4 °C. Supernatants were stored at −20 °C for 30 min. They were re-centrifuged before injection. Metabolites were separated on a Shimadzu LC-30 A UHPLC system. This was coupled to a SCIEX TripleTOF 6600+ mass spectrometer. A Waters HSS T3 column (2.1 × 100 mm, 1.8 μm) was used at 40 °C. Flow rate was 0.4 mL/min. Injection volume was 4 μL. Both positive and negative ion modes were used. Data were processed with Compound Discoverer 3.2. Metabolites were identified against HMDB and mzCloud libraries. PCA and OPLS-DA were performed using MetaboAnalyst 5.0 [15]. Differential metabolites were defined by VIP > 1 and FDR-adjusted P < 0.05. Quality control(QC) samples were injected every 10 runs. Overlaid TIC chromatograms showed good peak alignment. QC sample correlations were high (r > 0.99). CV distribution showed >75% of metabolites with CV < 0.3 (Supplementary Fig. S1).These results indicate that the metabolomics data were of high quality and reproducible.5.ELISA 5.ELISA ELISA ELISA was performed on plasma from 35 AF and 35 SVT patients. This included 10 discovery plus 25 additional per group. Two AF samples and one SVT sample were excluded due to insufficient volume. Commercial ELISA kits were used according to manufacturer instructions: VDAC1 (USCN Life Science, SEG901Hu, detection range 0.312–20 ng/mL), CS (Assay Genie, AEKE08525, detection range 6.25–400 ng/mL), and SFRP2(Assay Genie, AEKE07749, detection range 1.25–80 ng/mL). Each sample was run in duplicate. Intra- and inter-assay coefficients of variation (CV) were below 10% for all kits.6.GEO data reanalysis 6.GEO data reanalysis GEO data reanalysis Four datasets were downloaded from NCBI GEO: GSE282504 (RNA-seq): left atrial appendage from 20 AF and 10 Sinus Rhythm(SR) patients [16] GSE41177 (Affymetrix): left atrial tissue from 32 AF and 6 SR patients [17]. GSE79768 (Affymetrix): left atrial tissue from 14 AF and 12 SR patients [18]. GSE2240 (Affymetrix): right atrial appendage from 10 AF and 20 SR patients [19]. RNA-seq counts were normalized with DESeq2 [20]. Microarray data were processed with limma [21]. VDAC1 expression was compared between AF and SR using Student's t-test. Each dataset was analyzed separately.7.Single-nucleus RNA-seq 7.Single-nucleus RNA-seq Single-nucleus RNA-seq Dataset GSE255612 was re-analyzed. It contains more than 175,000 left atrial nuclei from 18 AF and 16 control patients. We used the published Seurat workflow for quality filtering, clustering, and cell-type annotation [8]. VDAC1 expression was visualized with UMAP.8.Bioinformatics 8.Bioinformatics Bioinformatics Functional enrichment analysis used METASCAPE. We used GO, KEGG, and Reactome databases. PPI networks were built with STRING v11.5. They were visualized with Cytoscape v3.9.1 [22]. Hub proteins were ranked by degree centrality using cytoHubba [23]. Cross-omics pathway overlap was analyzed with MetaboAnalyst 5.0 [15].9.Statistics 9.Statistics Statistics Normality was assessed using the Shapiro-Wilk test. Normally distributed continuous variables are reported as mean ± SD. They were compared using Student's t-test. Non-normally distributed variables are reported as median (IQR). ELISA data are shown as median (IQR) and were compared by Mann-Whitney U test. Categorical variables were compared with chi-square tests. Benjamini-Hochberg correction was applied for multiple comparisons in omics analyses. For multivariable adjustment, VDAC1 was log-transformed and fitted into a general linear model. Covariates were chosen based on baseline differences between groups (P < 0.10) and biological relevance. The model included AF status, age, sex, BMI, SBP, DBP, creatinine, and amiodarone use. Relationships between VDAC1 and clinical variables were assessed using Pearson correlation. All tests were two-sided. P < 0.05 was considered statistically significant. Analyses used SPSS 26.0 and R 4.2.0. Results 1. Patient characteristics 1. Patient characteristics Patient characteristics Table 1 summarizes baseline data for 37 AF and 36 SVT patients. The AF group had more males (67.6% vs 41.7%, P = 0.019). They were older (59.41 ± 8.58 vs 50.43 ± 13.12 years, P < 0.001). BMI was higher (24.65 ± 2.56 vs 23.03 ± 2.50 kg/m2, P = 0.008). Hypertension was more common in AF (51.4% vs 25.7%, P = 0.026). Amiodarone use was 62.2% in the AF group. It was 0% in the SVT group (P < 0.001).Table 1Baseline clinical characteristics of study participants.CharacteristicSVT group (n = 36)AF group (n = 37)P valueMale, n (%)14 (40.0)25 (67.6)*0.019Age, years50.43 ± 13.1259.41 ± 8.58*0.001SBP, mmHg126.94 ± 17.01129.86 ± 18.030.482DBP, mmHg77.89 ± 9.3978.70 ± 14.230.776Body mass index, kg/m223.03 ± 2.5024.65 ± 2.56*0.008Smoking history, n (%)4 (11.4)7 (18.9)0.377Hypertension, n (%)9 (25.7)19 (51.4)*0.026Coronary artery disease, n (%)3 (8.6)6 (16.2)0.327Diabetes mellitus, n (%)5 (14.3)2 (5.4)0.204Amiodarone use, n (%)0 (0.0)23 (62.2)*< 0.001β-blocker use, n (%)2 (5.7)6 (16.2)0.156Statin use, n (%)9 (25.7)17 (45.9)0.074CHA₂DS₂-VASc score1.00 ± 0.001.81 ± 1.37*0.001NT-proBNP, pg/mL73.66 ± 103.43540.31 ± 847.52*0.002hsTnT, ng/mL4.64 ± 2.106.85 ± 4.78*0.015Creatinine, μmol/L68.46 ± 12.0983.86 ± 20.09*< 0.001Blood urea nitrogen, mmol/L5.12 ± 1.285.91 ± 2.080.060Cystatin C, mg/L0.75 ± 0.111.00 ± 0.30*< 0.001Neutrophil-to-lymphocyte ratio1.93 ± 0.622.72 ± 1.46*0.004Left atrial diameter, mm30.40 ± 3.4639.84 ± 7.48*< 0.001Right atrial diameter, mm33.49 ± 3.1740.08 ± 6.73*< 0.001Left ventricular diastolic diameter, mm46.40 ± 2.8150.97 ± 6.83*< 0.001Left ventricular ejection fraction, %69.49 ± 4.8764.12 ± 9.90*0.005Continuous variables are expressed as mean ± standard deviation; categorical variables as n (%). Group comparisons used Student's t-test or Mann-Whitney U test for continuous variables and chi-square test for categorical variables. AF, atrial fibrillation; SVT, supraventricular tachycardia; SBP, systolic blood pressure; DBP, diastolic blood pressure; NT-proBNP, N-terminal pro-brain natriuretic peptide; hsTnT, high-sensitivity troponin T; CHA₂DS₂-VASc, congestive heart failure, hypertension, age ≥ 75 years, diabetes mellitus, stroke, vascular disease, age 65–74 years, sex category. Table 1Baseline clinical characteristics of study participants.CharacteristicSVT group (n = 36)AF group (n = 37)P valueMale, n (%)14 (40.0)25 (67.6)*0.019Age, years50.43 ± 13.1259.41 ± 8.58*0.001SBP, mmHg126.94 ± 17.01129.86 ± 18.030.482DBP, mmHg77.89 ± 9.3978.70 ± 14.230.776Body mass index, kg/m223.03 ± 2.5024.65 ± 2.56*0.008Smoking history, n (%)4 (11.4)7 (18.9)0.377Hypertension, n (%)9 (25.7)19 (51.4)*0.026Coronary artery disease, n (%)3 (8.6)6 (16.2)0.327Diabetes mellitus, n (%)5 (14.3)2 (5.4)0.204Amiodarone use, n (%)0 (0.0)23 (62.2)*< 0.001β-blocker use, n (%)2 (5.7)6 (16.2)0.156Statin use, n (%)9 (25.7)17 (45.9)0.074CHA₂DS₂-VASc score1.00 ± 0.001.81 ± 1.37*0.001NT-proBNP, pg/mL73.66 ± 103.43540.31 ± 847.52*0.002hsTnT, ng/mL4.64 ± 2.106.85 ± 4.78*0.015Creatinine, μmol/L68.46 ± 12.0983.86 ± 20.09*< 0.001Blood urea nitrogen, mmol/L5.12 ± 1.285.91 ± 2.080.060Cystatin C, mg/L0.75 ± 0.111.00 ± 0.30*< 0.001Neutrophil-to-lymphocyte ratio1.93 ± 0.622.72 ± 1.46*0.004Left atrial diameter, mm30.40 ± 3.4639.84 ± 7.48*< 0.001Right atrial diameter, mm33.49 ± 3.1740.08 ± 6.73*< 0.001Left ventricular diastolic diameter, mm46.40 ± 2.8150.97 ± 6.83*< 0.001Left ventricular ejection fraction, %69.49 ± 4.8764.12 ± 9.90*0.005Continuous variables are expressed as mean ± standard deviation; categorical variables as n (%). Group comparisons used Student's t-test or Mann-Whitney U test for continuous variables and chi-square test for categorical variables. AF, atrial fibrillation; SVT, supraventricular tachycardia; SBP, systolic blood pressure; DBP, diastolic blood pressure; NT-proBNP, N-terminal pro-brain natriuretic peptide; hsTnT, high-sensitivity troponin T; CHA₂DS₂-VASc, congestive heart failure, hypertension, age ≥ 75 years, diabetes mellitus, stroke, vascular disease, age 65–74 years, sex category. Baseline clinical characteristics of study participants. Continuous variables are expressed as mean ± standard deviation; categorical variables as n (%). Group comparisons used Student's t-test or Mann-Whitney U test for continuous variables and chi-square test for categorical variables. AF, atrial fibrillation; SVT, supraventricular tachycardia; SBP, systolic blood pressure; DBP, diastolic blood pressure; NT-proBNP, N-terminal pro-brain natriuretic peptide; hsTnT, high-sensitivity troponin T; CHA₂DS₂-VASc, congestive heart failure, hypertension, age ≥ 75 years, diabetes mellitus, stroke, vascular disease, age 65–74 years, sex category. Echocardiography showed enlarged atria and ventricles in AF. Left atrial diameter was 39.84 ± 7.48 vs 30.40 ± 3.46 mm (P < 0.001). Right atrial diameter was 40.08 ± 6.73 vs 33.49 ± 3.17 mm (P < 0.001). Left ventricular end-diastolic diameter was 50.97 ± 6.83 vs 46.40 ± 2.81 mm (P < 0.001). LVEF was lower in AF (64.12 ± 9.90% vs 69.49 ± 4.87%, P = 0.005). NT-proBNP, hsTnT, creatinine, cystatin C, and neutrophil-to-lymphocyte ratio were higher in AF (all P < 0.01). CHA₂DS₂-VASc scores were higher in AF (1.81 ± 1.37 vs 1.00 ± 0.00, P = 0.001). No differences were observed in blood pressure, smoking, diabetes, coronary disease, or statin use.2.Proteomic profiling identifies VDAC1 as a mitochondrial hub 2.Proteomic profiling identifies VDAC1 as a mitochondrial hub Proteomic profiling identifies VDAC1 as a mitochondrial hub We quantified 3268 plasma proteins. More than 75.8% matched two or more peptides. Filtering identified 217 DEPs. Of these, 163 were upregulated and 54 were downregulated(Fig. 2A-B).Fig. 2Proteomic analysis of plasma DEPs (n = 10 per group). (A) Volcano plot (red = upregulated, green = downregulated, gray = non-significant). (B) METASCAPE enrichment heatmap for upregulated DEPs. (C) METASCAPE enrichment heatmap for downregulated DEPs. (D) KEGG enrichment bubble plot. (E) VDAC1-centered PPI subnetwork.Fig. 2 Proteomic analysis of plasma DEPs (n = 10 per group). (A) Volcano plot (red = upregulated, green = downregulated, gray = non-significant). (B) METASCAPE enrichment heatmap for upregulated DEPs. (C) METASCAPE enrichment heatmap for downregulated DEPs. (D) KEGG enrichment bubble plot. (E) VDAC1-centered PPI subnetwork. Functional enrichment used METASCAPE. We integrated GO, KEGG, and Reactome. Upregulated proteins were enriched in carboxylic acid metabolism (GO:0019752). They were also enriched in carbohydrate metabolism (GO:0005975). Responses to oxygen-containing compounds (GO:1901700) were also seen. Carbon metabolism and TCA cycle pathways were enriched. This pattern suggests increased catabolic activity. Upregulated proteins were enriched in carboxylic acid metabolism, carbohydrate metabolism, and responses to oxygen-containing compounds, as well as carbon metabolism and TCA cycle pathways (Fig. 2C). Downregulated proteins showed distinct enrichment in extracellular matrix organization, complement and coagulation cascades, immune effector processes, proteolysis, ferroptosis, and autophagy (Fig. 2D). This signature suggests reduced structural integrity and immune surveillance. KEGG analysis further highlighted disturbances in TCA cycle, carbon metabolism, and amino acid biosynthesis (Fig. 2E). The PPI network contained 187 nodes and 624 edges. The top 10 hub proteins by degree centrality were ATP5F1A, CS, VDAC1, ATP5F1B, SDHA, MDH2, UQCRC2, ACO2, COX5A, and SLC25A4. VDAC1 showed extensive connections with mitochondrial respiratory chain proteins. It also connected with autophagy-related proteins (Fig. 2E).3.Metabolomics reveals lipid pathway alterations 3.Metabolomics reveals lipid pathway alterations Metabolomics reveals lipid pathway alterations PCA separated AF from SVT samples. PC1 explained 21.96% of variance. PC2 explained 15.65% (Supplementary Fig. S2A). OPLS-DA further distinguished groups. Permutation testing confirmed model reliability (R2Y = 0.984, Q2 = 0.773, P < 0.005) (Supplementary Fig. S2B). We identified 505 differential metabolites. Of these, 360 were upregulated in AF. 145 were downregulated (Fig. 3A). Heatmap analysis separated the two groups (Fig. 3B). The top 20 VIP metabolites (Fig. 3C) included lipid components. These included linoleic acid, glycerophospholipids, acylcarnitine C17:1, and MG (16:0). Exogenous compounds were also found. These included triterpenoids, hopanes, and dibutyl phthalate. These exogenous compounds may reflect diet, environmental exposure, or analytical contamination. This point is addressed in the Discussion.Fig. 3Metabolomic analysis (n = 10 per group). (A) Volcano plot. (B) Hierarchical clustering heatmap of metabolites. (C) Top 20 VIP metabolites. (D) KEGG enrichment bubble chart.Fig. 3 Metabolomic analysis (n = 10 per group). (A) Volcano plot. (B) Hierarchical clustering heatmap of metabolites. (C) Top 20 VIP metabolites. (D) KEGG enrichment bubble chart. Three metabolites were associated with mitochondrial metabolism. Nonenyl-4-hydroxybenzoic acid is a coenzyme Q precursor. (S)-3-hydroxybutyric acid is a ketone body from fatty acid oxidation. Carnitine C17:1 is an acyl carnitine involved in β-oxidation. The predominance of lipid metabolites was consistent with KEGG enrichment (Fig. 3D). Glycerophospholipid, linoleic acid, and arachidonic acid metabolism were the most significantly altered pathways. Mitochondrial-related pathways also reached statistical significance. These included porphyrin and glutathione metabolism, autophagy, and ferroptosis. Enrichment analysis revealed four mitochondrial functional themes. These were energy substrate oxidation, electron transport chain stability, redox balance, and clearance of damaged organelles.4.Integrated pathway analysis 4.Integrated pathway analysis Integrated pathway analysis We intersected KEGG enrichment results from proteomics and metabolomics. Eight pathways reached P < 0.01 in both data layers. They mainly involved lipid and carbon metabolism (Fig. 4). This suggests that systemic mitochondrial-lipid remodeling is a core feature associated with AF.5.ELISA validationFig. 4Common KEGG pathways enriched in both proteomic and metabolomic data. Eight pathways met P < 0.01 in both datasets.Fig. 4 5.ELISA validation ELISA validation Common KEGG pathways enriched in both proteomic and metabolomic data. Eight pathways met P < 0.01 in both datasets. Based on PPI and enrichment results, we selected VDAC1, CS, and SFRP2 for ELISA. VDAC1 and CS are mitochondrial proteins. SFRP2 is related to myocardial fibrosis [24]. All three proteins were higher in AF plasma (P < 0.001 for each). In the validation cohort (35 AF, 35 SVT), median plasma VDAC1 was 758.2 pg/mL (IQR 552.9–963.5) in AF. It was 573.6 pg/mL (IQR 440.5–706.8) in SVT (P < 0.001). For CS, median values were 442.2 pg/mL (IQR 328.1–556.3) vs 311.3 pg/mL (IQR 228.0–394.6) (P < 0.001). For SFRP2, median values were 19.1 ng/mL (IQR 14.7–23.5) vs 15.5 ng/mL (IQR 12.2–18.8) (P < 0.001)(Fig. 5A).Fig. 5ELISA validation of plasma VDAC1, CS, and SFRP2. (A) Full cohort (35 AF, 35 SVT). (B) Newly recruited patients only (25 AF, 25 SVT; discovery samples excluded). ***P < 0.001. Data are shown as median with interquartile range.Fig. 5 ELISA validation of plasma VDAC1, CS, and SFRP2. (A) Full cohort (35 AF, 35 SVT). (B) Newly recruited patients only (25 AF, 25 SVT; discovery samples excluded). ***P < 0.001. Data are shown as median with interquartile range. To assess whether these results were driven by the discovery samples, we performed a sensitivity analysis. We restricted it to the 25 newly recruited patients per group. All three proteins remained significantly elevated in AF in this subset (all P < 0.001) (Fig. 5B). Fold changes in the validation cohort were consistent with the discovery data.6.Multivariable analysis of confounding factors 6.Multivariable analysis of confounding factors Multivariable analysis of confounding factors To test whether the VDAC1-AF link held up after accounting for baseline differences, we ran a multivariable general linear model. VDAC1 was entered as the dependent variable. The model adjusted for age, sex, BMI, creatinine, amiodarone use, and hypertension. After adjustment, AF status remained significantly tied to VDAC1 levels (B = −181.741, P = 0.001; partial η2 = 0.158). Amiodarone use also came out significant (B = −148.112, P = 0.039; partial η2 = 0.064). The other covariates—age, sex, BMI, creatinine, and hypertension—did not reach statistical significance (all P > 0.05) (Table 2).7.ROC analysisTable 2Multivariable regression analysis of factors associated with plasma VDAC1 levels.VariableFP valuePartial η2AF status−181.7410.0010.158Amiodarone use−148.1120.0390.064Sex57.9170.2060.025Age0.4980.8000.001BMI−8.6470.3120.016Creatinine4.4310.7930.001Hypertension13.0950.7730.001Dependent variable: VDAC1 (pg/mL). AF status coded as 0 = SVT, 1 = AF. Partial η2 ≥ 0.14 indicates a large effect size. 7.ROC analysis ROC analysis Table 2Multivariable regression analysis of factors associated with plasma VDAC1 levels.VariableFP valuePartial η2AF status−181.7410.0010.158Amiodarone use−148.1120.0390.064Sex57.9170.2060.025Age0.4980.8000.001BMI−8.6470.3120.016Creatinine4.4310.7930.001Hypertension13.0950.7730.001Dependent variable: VDAC1 (pg/mL). AF status coded as 0 = SVT, 1 = AF. Partial η2 ≥ 0.14 indicates a large effect size. Multivariable regression analysis of factors associated with plasma VDAC1 levels. Dependent variable: VDAC1 (pg/mL). AF status coded as 0 = SVT, 1 = AF. Partial η2 ≥ 0.14 indicates a large effect size. To evaluate the discriminative ability of plasma VDAC1, we performed ROC analysis. The area under the curve (AUC) was 0.816 (95% CI: 0.715–0.917; P < 0.001). This indicates good performance in distinguishing AF from SVT in this cohort (Supplementary Fig. S3).8.Correlation analysis 8.Correlation analysis Correlation analysis We examined the relationships between plasma VDAC1 and clinical parameters using Pearson correlation (Table 3). VDAC1 correlated positively with AF status (r = 0.475, P < 0.001) and left atrial diameter (r = 0.342, P = 0.003). A weak negative correlation was observed with LVEF (r = −0.306, P = 0.008). No significant correlation was found between VDAC1 and NT-proBNP (r = −0.098, P = 0.409).9.GEO datasets show VDAC1 upregulation in AFTable 3Correlation between plasma VDAC1 and clinical parameters.VariablerP valueAF status0.475<0.001Left atrial diameter0.3420.003LVEF−0.3060.008NT-proBNP−0.0980.409Pearson correlation was used. AF, atrial fibrillation; LVEF, left ventricular ejection fraction; NT-proBNP, N-terminal pro-brain natriuretic peptide. 9.GEO datasets show VDAC1 upregulation in AF GEO datasets show VDAC1 upregulation in AF Table 3Correlation between plasma VDAC1 and clinical parameters.VariablerP valueAF status0.475<0.001Left atrial diameter0.3420.003LVEF−0.3060.008NT-proBNP−0.0980.409Pearson correlation was used. AF, atrial fibrillation; LVEF, left ventricular ejection fraction; NT-proBNP, N-terminal pro-brain natriuretic peptide. Correlation between plasma VDAC1 and clinical parameters. Pearson correlation was used. AF, atrial fibrillation; LVEF, left ventricular ejection fraction; NT-proBNP, N-terminal pro-brain natriuretic peptide. VDAC1 mRNA was higher in AF across all four GEO datasets. GSE282504 (P = 0.018), GSE41177 (P = 0.021), GSE79768 (P = 0.035), GSE2240 (P = 0.042). Consistent upregulation across platforms and tissues supports an association between VDAC1 and AF (Fig. 6).10.Single-nucleus RNA-seq localizes VDAC1 to cardiomyocytesFig. 6VDAC1 expression in four GEO datasets. (A) GSE282504 (RNA-seq, 20 AF vs 10 SR). (B) GSE2240 (microarray, 10 AF vs 20 SR). (C) GSE41177 (microarray, 32 AF vs 6 SR). (D) GSE79768 (microarray, 14 AF vs 12 SR). *P < 0.05.Fig. 6 10.Single-nucleus RNA-seq localizes VDAC1 to cardiomyocytes Single-nucleus RNA-seq localizes VDAC1 to cardiomyocytes VDAC1 expression in four GEO datasets. (A) GSE282504 (RNA-seq, 20 AF vs 10 SR). (B) GSE2240 (microarray, 10 AF vs 20 SR). (C) GSE41177 (microarray, 32 AF vs 6 SR). (D) GSE79768 (microarray, 14 AF vs 12 SR). *P < 0.05. UMAP clustering of GSE255612 resolved 15 cardiac cell subtypes (Fig. 7A–B). VDAC1 transcripts were concentrated in cardiomyocytes. Signals were stronger in AF-derived cells (Fig. 7C). This is consistent with cardiomyocytes as a potential source of circulating VDAC1. However, it does not directly demonstrate protein release into plasma.Fig. 7Single-nucleus RNA-seq UMAP (GSE255612). (A) Cell subtypes. (B) UMAP of 15 cell types. (C) VDAC1 expression in AF and control cardiomyocytes.Fig. 7 Single-nucleus RNA-seq UMAP (GSE255612). (A) Cell subtypes. (B) UMAP of 15 cell types. (C) VDAC1 expression in AF and control cardiomyocytes. Discussion This study has three main findings. First, plasma proteomics and metabolomics identified VDAC1 as a consistently elevated protein in persistent AF. This was validated in an expanded cohort. Supportive evidence came from four public transcriptional datasets. Second, metabolomic disturbances in AF were predominantly in lipid pathways. This suggests systemic metabolic remodeling. Third, VDAC1 expression in atrial tissue was localized mainly to cardiomyocytes in single-nucleus RNA-seq data. This is consistent with a myocardial origin of the circulating protein. Taken together, these findings position VDAC1 as a candidate circulating marker of mitochondrial dysfunction in AF. Mitochondrial dysfunction and atrial remodeling Mitochondrial injury may contribute to atrial remodeling. Rapid heart rates increase the energy demand of cardiomyocytes, leading to ATP depletion and reactive oxygen species accumulation [[4], [25], [26]]. Mitochondria supply the majority of ATP in cardiac cells and also regulate calcium signals. Mitochondrial dysfunction compromises electrical stability [[11], [27]]. Our PPI data show that VDAC1 connects to several respiratory chain subunits. This finding links tissue-level observations to plasma markers that are easier to measure. Mitophagy and VDAC1 Mitophagy is responsible for the clearance of damaged mitochondria. In AF, this biological process has been shown to be impaired, leading to the accumulation of dysfunctional organelles and sustained oxidative stress [[6], [28]]. Our network analysis revealed a correlation between VDAC1 and autophagy-related proteins, suggesting a potential connection between cellular energy production and organelle quality control. Nevertheless, the current data are only correlational; whether VDAC1 actively regulates mitophagy remains undetermined, as it may merely reflect the presence of mitochondrial stress. Further functional experiments are required to clarify these possibilities. In the multivariable model analysis, amiodarone use correlated with lower VDAC1 levels (B = −148.1, P = 0.039). This finding was unexpected, yet not entirely surprising given that amiodarone is known to affect mitochondrial function and membrane properties. More importantly, even after accounting for amiodarone use, AF remained independently associated with VDAC1 elevation (B = −181.7, P = 0.001), with a large effect size (partial η2 = 0.158). This argues against the possibility that the VDAC1 increase observed in AF is merely a side effect of drug treatment. Correlation with clinical parameters The positive correlation between VDAC1 and left atrial diameter supports a link between this circulating protein and atrial structural remodeling. The weak inverse correlation with LVEF suggests that VDAC1 elevation may not be confined to the atria tissues. The absence of significant correlation between VDAC1 and NT-proBNP suggests that the two molecules reflect distinct pathological pathways, which may correspond to mitochondrial stress and myocardial wall stress, respectively. The EV hypothesis: a speculative model We propose a hypothetical model in Supplementary Fig. S4. AF-related stress may upregulate VDAC1 in cardiomyocytes, which may then be released via EVs. This hypothesis remains speculative, as we did not isolate EVs from AF patient plasma, nor did we demonstrate that EV-bound VDAC1 affects calcium handling in recipient cells. The blue dashed boxes in the figure indicate uncertain elements. Prior studies have documented VDAC1 in EVs in non-cardiac contexts [29]. The glycerophospholipid alterations identified in the present study are consistent with changes in membrane turnover; yet these findings merely represent indirect evidence. Further direct experimental validation is therefore required, which specifically encompasses EV isolation, quantitative detection of EV-bound VDAC1, and targeted functional assays. SVT control group: rationale and limitations SVT patients were used as a tachycardia-matched control group, which reduced heart rate-related confounding. SVT patients are generally younger and have less structural heart disease. However, this introduced baseline differences between the groups (Table 1).To address this, we performed multivariable adjustment. We also used four independent public datasets comparing AF with sinus rhythm. All confirmed VDAC1 upregulation in AF. A three-group design would be more persuasive. This would include AF vs SVT vs sinus rhythm. But sampling coronary sinus blood in healthy volunteers is not feasible. The procedure is invasive. The use of SVT patients as controls helped match heart rate but introduced baseline differences. A sinus rhythm control group would be preferable in future studies. Limitations Following limitations should be considered when interpreting our findings. First, the discovery cohort enrolled merely 10 patients in each group. Considering the high-dimensional nature of the omics data, statistical power was limited. The differentially expressed proteins and metabolites screened in this study should be interpreted as exploratory results for hypothesis generation, rather than definitive conclusions. Future studies with larger-scale independent cohorts are needed to confirm these findings. Second, the AF group and the SVT group exhibited significant differences in multiple baseline clinical characteristics, including age, sex, BMI, hypertension, renal function, and amiodarone medication history. Although confounding adjustment for the above variables was performed in the multivariable regression model, residual confounding effects cannot be completely ruled out. The intergroup differences observed in this study may partially result from the underlying clinical imbalance between groups, rather than being specifically caused by the pathophysiological process of AF. Third, our plasma assays capture total circulating VDAC1. We did not distinguish between free protein and protein associated with EVs, nor did we isolate EVs to test the release hypothesis. This remains a speculative mechanism and requires targeted functional validation. Fourth, we selected only three proteins for ELISA validation: VDAC1, CS, and SFRP2. This selection was made after the omics analysis, which may have introduced selection bias. Additional candidates from the proteomic screen should be validated in future studies. Fifth, while coronary sinus sampling allows sensitive detection of cardiac venous effluent, it cannot confirm a net cardiac release gradient without paired arterial or peripheral samples, leaving it unclear whether elevated VDAC1 reflects cardiac release or higher systemic levels. Sixth, as our study is a cross-sectional investigation, we cannot establish temporal relationships, nor can we predict clinical outcomes. Prospective cohort studies are needed to test whether baseline VDAC1 levels are associated with AF recurrence, progression, or treatment response. Seventh, the clinical applicability of VDAC1 as a biomarker is not yet established. Its modest discriminatory performance in our cohort suggests that VDAC1 is unlikely to be useful as a standalone diagnostic test. However, it may contribute to a multi-marker panel and provide mechanistic insight into AF-related mitochondrial changes. Conclusion Plasma VDAC1 was higher in patients with persistent AF than in those with SVT. This finding was supported by ELISA and transcriptomic data. The metabolic alterations we observed are predominantly in lipid pathways. This suggests that AF is associated with systemic metabolic changes. These may reflect myocardial mitochondrial dysfunction. These findings are preliminary. The clinical utility of VDAC1 requires validation in prospective studies. If confirmed, VDAC1 could become a circulating marker for atrial mitochondrial dysfunction in patients with AF. The following are the supplementary data related to this article.Supplementary Figure S1Quality control assessment of metabolomics data.(A) Pearson correlation analysis among QC samples. All QC samples showed r > 0.99.(B) Coefficient of variation (CV) distribution of all detected metabolites. An empirical cumulative distribution function (ECDF) plot is shown. Over 75% of metabolites had CV < 0.3.Supplementary Figure S2Metabolomic analysis (n = 10 per group). (A) PCA score plot. (B) OPLS-DA score plot.Supplementary Figure S3ROC curve for plasma VDAC1 in discriminating AF from SVT. AUC = 0.816 (95% CI: 0.715–0.917; P < 0.001).Supplementary Figure S4Proposed mechanism diagram. Left: AF cardiomyocyte (red = supported by previous studies and our data; blue dashed = speculative EV pathway). Right: SR cardiomyocyte with stable calcium homeostasis. SR, sinus rhythm; EVs, extracellular vesicles; ROS, reactive oxygen species. Quality control assessment of metabolomics data.(A) Pearson correlation analysis among QC samples. All QC samples showed r > 0.99.(B) Coefficient of variation (CV) distribution of all detected metabolites. An empirical cumulative distribution function (ECDF) plot is shown. Over 75% of metabolites had CV < 0.3. Metabolomic analysis (n = 10 per group). (A) PCA score plot. (B) OPLS-DA score plot. ROC curve for plasma VDAC1 in discriminating AF from SVT. AUC = 0.816 (95% CI: 0.715–0.917; P < 0.001). Proposed mechanism diagram. Left: AF cardiomyocyte (red = supported by previous studies and our data; blue dashed = speculative EV pathway). Right: SR cardiomyocyte with stable calcium homeostasis. SR, sinus rhythm; EVs, extracellular vesicles; ROS, reactive oxygen species. CRediT authorship contribution statement Jun Wang: Validation, Resources, Methodology, Data curation. Xiaozhen Chen: Writing – original draft, Methodology, Investigation. Zilan Zhang: Project administration, Methodology, Investigation, Funding acquisition. Chun Gui: Visualization, Project administration. Funding This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Declaration of competing interest The authors declare no conflicts of interest.