2.1. Overall study design and data sources

As shown in Fig 1, the overall study design is presented as a flow chart. The FDA Adverse Event Reporting System (FAERS) database, utilized in the present study and available at https://www.fda.gov/drugs/surveillance/fdas-adverse-event-reporting-system-faers, comprises adverse event reports, medication error reports, and product quality complaints submitted to the FDA. Adverse events and medication errors within this system are coded using terminology from the Medical Dictionary for Regulatory Activities (MedDRA). The GSE41177 dataset from GEO database, including 32 AF and 6 sinus rhythm (SR) atrial appendage samples, was served as a training group to identify hub genes when combining the result of the network toxicology analysis. The following two atrial tissue transcriptomic datasets functioned as validation sets in our study: (1) GSE115574 dataset, including 28 AF and 31 SR atrial samples from 15AF patients and 15 SR patients; (2) GSE79768 dataset, comprising 14 AF and 12 atrial specimens. The GSE197518 dataset from the GEO database was retrieved to work as the Single-cell dataset, consisting of 7 AF samples from left atrial tissues collected from 7 patients with persistent AF undergoing surgical mitral valve repair and 5 SR samples from 5 control subjects. The databases for network toxicology analysis were detailed at the section 2.4. This study did not require ethical approval or informed consent because it involved only the analysis of publicly available, de-identified data from three sources: FAERS, GEO, and publicly available databases for network pharmacology. No human subjects were directly enrolled, no animal experiments were performed, and no privacy-sensitive information was accessed.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. The overall design of the present study.

https://doi.org/10.1371/journal.pone.0353143.g001

2.2. Data processing and signal mining of FAERS database

We performed a pharmacovigilance analysis of adverse events associated with DOX using data from the FAERS. Reports involving DOX were extracted from the FAERS public dashboard for the period spanning Q1 2004 to Q3 2025. Duplicate entries were removed to enhance data reliability according to FDA-recommended data cleaning steps. Adverse event cases related to the DOX were identified using the Medical Dictionary for Regulatory Activities (MedDRA version 26.0) and were defined based on the preferred term named “ATRIAL FIBRILLATION”. To evaluate the strength of potential safety signals, we computed commonly used disproportionality measures, including the reporting odds ratio (ROR) and information component (IC). A potential safety signal was defined when the lower limit of the 95% confidence interval (CI) of the ROR exceeded 1 (ROR025 > 1) and the number of individual case safety reports (n) was ≥ 3. For the Information Component (IC), a positive signal was considered when the lower bound of the 95% credible interval of the IC exceeded 0 (IC025 > 0), which represents a statistically significant increase in reporting probability based on the Bayesian confidence propagation neural network (BCPNN) approach. Formulas and interpretations for these metrics in the previous study [14]. Statistically significant signals were further examined with chi-square tests. The entire data extraction and analytical workflow adhered to established pharmacovigilance methodology and standard operating procedures, as outlined in S1 Fig. Moreover, The FAERS database provides the Indication Preferred Term (INDI_PT) field, which records the primary indication for the drug reported in each adverse event case, coded using the MedDRA dictionary. For this study, we systematically classified and aggregated these INDI_PT terms to determine the distribution of primary tumor types among the included reports.

2.3. Differentially expressed gene analysis between AF and SR patients

In the present study, GSE41177 dataset from GEO database was utilized to obtain the differentially expressed genes (DEGs). The screening parameter for identification of DEGs was as follows: (1) adjusted P value<0.05; (2) |log₂ fold change (log₂ FC)| > 0.5. A volcano plot was utilized to visualize the DEGs. According to the order of their log₂ FC, the top 10 significant genes were also annotated with a heatmap displaying the expression patterns of the top 15 most upregulated and downregulated genes.

2.4. Identification of molecular targets in DOX-induced AF

The molecular structure and simplified molecular input line entry system (SMILES) notation of DOX were obtained from ChemSpider (https://www.chemspider.com/) [15]. To identify potential molecular targets of DOX, we conducted comprehensive target prediction analyses using multiple databases. Firstly, putative DOX-related targets were predicted using the ChEMBL (https://www.ebi.ac.uk/chembl/) [16], SwissTargetPrediction (http://www.swisstargetprediction.ch/) [17], and Similarity Ensemble Approach (SEA, https://sea.bkslab.org/) [18] databases using SMILES or “doxorubicin” as the key word. To identify therapeutic molecular targets of AF, we consulted GeneCards (https://www.genecards.org) [19] and Online Mendelian Inheritance in Man (OMIM, http://www.omim.org/) [20] databases to extract them with the keyword “atrial fibrilation”. Accordingly, target protein names were standardized according to UniProt nomenclature (https://www.uniprot.org/) [21]. Through intersection analysis of DOX-related targets, AF-associated genes, and DEGs of AF in the section 2.2, we identified candidate mechanistic targets involved in DOX-induced AF after removing redundant entries of target genes.

2.5. Functional annotation and pathway enrichment analysis and hub genes identification

To explore the biological functions and signaling pathways of the candidate genes selected in Section 2.3, we performed Gene Ontology (GO) [22] and KEGG [23] pathway enrichment analyses with a significance threshold of both p-value and adjusted p-value < 0.05. To delineate the molecular interaction landscape, we constructed a protein-protein interaction (PPI) network using the STRING database (https://cn.string-db.org/) [24] with the identified candidate targets. A minimum interaction confidence score threshold of 0.15 was selected to capture enough interactions while maintaining network comprehensiveness. Cytoscape (version 3.7.1) was then utilized for visualizing the analysis result of the DOX-induced AF network. Assisting by CytoHubba plugin [25], Hub genes were identified with the Maximal Clique Centrality (MCC) algorithm.

2.6. Prediction performance of the hub genes to predict AF

Hub genes, including CCR2, PDE5A, CXCR2, HSP90AB1, and KDR, were utilized to construct 31 predictive combinations from single-gene to penta-gene forms. The GSE41177 dataset served as the training set, with GSE79768 and GSE115574 as external validation sets to evaluate prediction performance. AUC (Area Under the ROC Curve) was calculated for each combination across all datasets, and the mean of AUCs was used to assess overall prediction accuracy. Nomograms were built to visualize the correlation between gene expression-derived points and AF mortality risk, while Decision Curve Analysis (DCA) was performed to determine clinical net benefits at different threshold probabilities. A calibration plot was generated to compare the observed AF risk with the predicted probability, verifying the consistency of the prediction model. Additionally, gene expression levels were compared between control and disease groups in the three datasets to validate differential expression of the genes in the selected model.

2.7. Gene set enrichment analysis (GSEA) and Immunoinfiltration analysis

To elucidate the underlying biological functions and pathways, we performed Spearman correlation analysis between CCR2/PDE5A and all other genes in the GSE41177 dataset. The resulting correlation coefficients were ranked before being used for GSEA with the following significance thresholds: P < 0.05, ∣NES∣ > 1, and FDR < 0.25. To assess immune cell infiltration, we estimated the relative abundances of multiple immune cell types using the CIBERSORT R package and compared their profiles between AF and SR samples (GSE115574). Immune cell subsets showing statistically significant differences between the AF and SR groups (p < 0.05) were also identified. Finally, Pearson correlation analysis was applied to evaluate associations between these differentially infiltrated immune cell subsets respectively.

2.8. Molecular docking analysis

To validate whether CCR2 and PDE5A could serve as potential targets of DOX to induce AF, molecular docking analyses were performed. Protein structures corresponding to genes were collected from the RCSB Protein Data Bank (PDB, https://www.rcsb.org/) [26]. The 3D structure of DOX was downloaded from PubChem (https://pubchem.ncbi.nlm.nih.gov) [27]. Following format conversion to PDB, molecular docking was carried out using AutoDock Vina to screen binding poses and evaluate affinity. The most stable conformation, determined by minimal binding energy, was further examined in PyMOL (version 3.0.3) and LigPlus (version 2.2.4) to characterize interaction patterns. Specifically, binding affinities were classified as weak (−4.25 to −5.00 kcal/mol), moderate (−5.00 to −7.00 kcal/mol), or strong (below −7.00 kcal/mol) [28].

2.9. Single-cell data analysis

scRNA-seq data (GSE197518) from the GEO database were pre-processed using the Seurat v5 R package. The steps are as follows: (1) Quality control: Gene count between 200 and 5000 per cell, and mitochondrial ratio under 20%. (2) Dimensionality reduction: Clustering was performed with the “FindClusters” function at 10 PCs and a resolution of 0.4. UMAP algorithm was applied to simplify the data structure and reveal the underlying cell population characteristics. (3) Cell annotation: Cell types were identified and annotated based on marker gene information from relevant literatures [29,30]. (4) Differences illustration of proportions of cell sub-types between the AF and SR groups. (5) Identification of key cell types: The differences of expression level of CCR2 and PDE5A in cell sub-types and key cell types were identified between the AF and SR groups by performing the Wilcoxon test.

3.1. FAERS database analysis and Identification of DEGs and potential target genes for DOX-induced AF

Regarding the clinical relevance of DOX in AF, it showed a statistically significant signal for AF-AEs (ROR = 2.04, 95% CI: 1.50–2.77), supported by other disproportionality metrics (IC025 = 0.54) (S1 Fig). Moreover, the distribution of primary tumor types, classified by INDI_PT in the FAERS database, was dominated by lymphoma (34%), followed by missing/unknown indications (17%), breast cancer (12%), sarcoma (11%), ovarian cancer (7%), other rare tumor types (7%), myeloma (5%), leukemia (5%), and liver cancer (2%). A total of 4226 DEGs were identified through threshold-based screening, including 2178 up-regulated genes and 2048 down-regulated genes in the disease group (Fig 2A and Fig 2B). The SMILES notation for DOX is “COc1cccc2c1C(=O)c1c(O)c3c(c(O)c1C2=O)C[C@@](O)(C(=O)CO)C[C@@H]3O[C@H]1C[C@H](N)[C@H](O)[C@H](C)O1”. The ChEMBL, SwissTargetPrediction, and SEA databases identified 249, 80, and 16 potential targets, respectively. After removing duplicated targets, 324 toxicity targets were identified (Fig 2C and S1 Table). The Genecards and OMIM databases contained 6049 and 65 AF-related target genes (Fig 2D), respectively. Subsequently, 6096 AF-related target genes were obtained by removing duplicated genes (S1 Table). Through intersection analysis of DOX targets, AF-associated genes, and DEGs, after redundant entries were removed, candidate mechanistic targets involved in DOX-induced AF were successfully obtained (Fig 3A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Identification of differentially expressed genes (DEGs) and potential targets for doxorubicin (DOX) and atrial fibrillation (AF).

(A) Volcano plot visualizing DEGs between AF and sinus rhythm (SR) atrial tissues in the GSE41177 dataset. (B) Heatmap showing the top 15 most upregulated and downregulated DEGs. (C) Venn diagram illustrating the overlap of DOX-related targets predicted by three databases: ChEMBL, SwissTargetPrediction, and SEA. (D) Venn diagram illustrating the overlap of AF-related genes retrieved from GeneCards and OMIM databases.

https://doi.org/10.1371/journal.pone.0353143.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Functional annotation, pathway enrichment, and hub gene identification of candidate targets.

(A) Venn diagram showing the intersection of DOX-related targets, AF-related genes, and DEGs. (B–C) Gene Ontology (GO) enrichment analysis of candidate targets. (D–E) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of candidate targets. (F) Protein-protein interaction (PPI) network of candidate targets visualized by Cytoscape. Hub genes (CCR2, PDE5A, CXCR2, HSP90AB1, KDR) identified by the Maximal Clique Centrality (MCC) algorithm are in the center.

https://doi.org/10.1371/journal.pone.0353143.g003

3.2. Functional annotation, pathway enrichment analysis, and hub genes identification

GO and KEGG pathway enrichment analyses were performed for the candidate genes, with a significance threshold of P value and adjusted P value < 0.05. The results showed that these candidate genes were significantly enriched in biological processes such as calcium ion homeostasis, intracellular calcium ion homeostasis, positive regulation of leukocyte tethering or rolling, and chemotaxis. For molecular functions, enrichment was observed in protein phosphatase activator activity, 3’,5’-cyclic-AMP phosphodiesterase activity, 3’,5’-cyclic-GMP phosphodiesterase activity, 3’,5’-cyclic-nucleotide phosphodiesterase activity, and C-C chemokine receptor activity (Fig 3B, Fig 3C, and S2 Table). KEGG pathway enrichment indicated involvement in the cGMP-PKG signaling pathway, Renin secretion, cAMP signaling pathway, Calcium signaling pathway, VEGF signaling pathway, and Growth hormone synthesis, secretion and action, among others (Fig 3D and Fig 3E, S2 Table). The PPI network of candidate targets was constructed using the STRING database and visualized with Cytoscape 3.7.1. Using the CytoHubba plugin and MCC algorithm, hub genes, including CCR2, PDE5A, CXCR2, HSP90AB1, and KDR, were identified (Fig 3F, S3 Table).

3.3. Prediction performance of hub genes for AF

A total of 31 predictive combinations (single-gene to penta-gene) were constructed using the identified hub genes. The CCR2 + PDE5A+CXCR2 combination achieved the highest mean AUC value of 0.87 by considering the comprehensive performance in both training and validation sets (Fig 4A and Fig 4C). A nomogram was constructed to visualize the AF risk model and a calibration plot showed good consistency between the observed AF risk and the predicted probability, validating the reliability of the prediction model (Fig 4B and Fig 4D). DCA demonstrated that the predictive models offered significant clinical net benefits across different threshold probabilities (Fig 4E). Furthermore, comparison of gene expression levels between control and AF groups in both training and validation sets further confirmed the significance of genes in the best-performed model, supporting the potential of CCR2 and PDE5A as predictive biomarkers (Fig 4F and Fig 4G). As for CXCR2, not all gene expression levels between control and AF groups in both training and validation sets were observed significantly different (Fig 4H). Though not being selected into the best-performed model, HSP90AB1 also owned a significantly high expression level in both 3 datasets (S2 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Predictive performance of hub gene combinations for Atrial Fibrillation (AF).

(A) Bar plot showing the mean Area Under the Curve (AUC) of 5 hub gene combinations (single-gene to penta-gene) across training (GSE41177) and validation (GSE79768, GSE115574) datasets. The CCR2 + PDE5A+CXCR2 combination achieved the highest mean AUC (0.87). (B) Nomogram for predicting AF risk based on the CCR2 + PDE5A+CXCR2 combination. (C-E) ROC curves, calibration plot and decision curve analysis (DCA) of the CCR2 + PDE5A+CXCR2 model. (F–G) Box plots comparing CCR2 (F), PDE5A (G) and CXCR2 (H) expression levels between AF and Sinus Rhythm (SR) groups in training and validation datasets. ns = non-significant, **P < 0.01, ***P < 0.001.

https://doi.org/10.1371/journal.pone.0353143.g004

3.4. Gene set enrichment analysis (GSEA) and Immunoinfiltration analysis

CCR2 is primarily involved in notable pathways, which include antigen processing and presentation, hematopoietic cell lineage, leishmaniasis, staphylococcus aureus infection, and Th17 cell differentiation pathways (Fig 5A). Moreover, PDE5A is primarily involved in notable pathways, which include leukocyte transendothelial migration, measles, osteoclast differentiation, regulation of actin cytoskeleton, Yersinia infection pathways (Fig 5B). Following the algorithm’s confidence evaluation criteria, valid immune cell content data were retained from all samples (Fig 5D). After rigorous statistical testing, 6 immune cell types with significant differences were identified: B cells memory, eosinophils, M2 macrophages, monocytes, neutrophils, and T cells regulatory (Tregs) cells (Fig 5C). The correlation matrix shows correlation coefficients between different immune cell types (Fig 5E).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Gene Set Enrichment Analysis (GSEA) and immune infiltration profiling.

(A–B) GSEA plots showing pathways enriched in genes positively correlated with CCR2 (A) and PDE5A (B) in the GSE41177 dataset. (C) Box plots comparing the relative abundance of infiltrated immune cell subsets between atrial fibrillation (AF) (disease) and sinus rhythm (SR) (control) groups. (D) Bar plot showing the relative proportion of all immune cell types in AF and SR samples. (E) Correlation matrix of differentially infiltrated immune cell subsets. ns = non-significant, **P < 0.01, ***P < 0.001.

https://doi.org/10.1371/journal.pone.0353143.g005

3.5. Results of Molecular docking validation between DOX and core targets

To verify the potential binding interactions between DOX and the identified core genes (CCR2 and PDE5A), molecular docking assays were conducted. The outcomes revealed robust binding affinities between DOX and these two target proteins, with the minimum binding energies persistently falling below −7 kcal/mol. This finding suggests that the molecular interactions between DOX and the target proteins are stable and occur spontaneously (Fig 6).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Molecular docking validation of DOX binding to CCR2 and PDE5A.

(A) Three-dimensional (3D) structure of the DOX-CCR2 complex. (B) 3D structure of the DOX-PDE5A complex.

https://doi.org/10.1371/journal.pone.0353143.g006

3.6. The findings of single-cell analysis

Eight major cell types (endothelial cells, fibroblasts, lymphocytes, macrophages, monocytes, neutrophils, pericytes, and smooth muscle cells) were identified based on annotation for each cell subpopulation (Fig 7A). CCR2 was found to present a higher expression level mainly in macrophages and lymphocytes, while PDE5A in fibroblasts (Fig 7B). To characterize the cell-type-specific expression patterns of the target genes, we performed single-cell transcriptomic profiling across 8 distinct cell populations, and visualized the results as a dot plot (Fig 7C). Differences in the proportion of all cell types were identified between the AF and SR groups, especially for an increase in macrophages (Fig 7D). Additionally, the expression of biomarkers CCR2 and PDE5A in different cell types between the two groups was examined subsequently. As Fig 7E and Fig 7F display, significant differential expressions in macrophages and lymphocytes of CCR2 can be observed. After further clustering the lymphocytes, significant differential expressions in T cells and B cells of CCR2 can also be observed. Moreover, a significant differential expression in fibroblasts of PDE5A was also identified (S3 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Single-cell RNA sequencing (scRNA-seq) analysis of CCR2 and PDE5A expression patterns.

(A) Uniform Manifold Approximation and Projection (UMAP) plot of 8 major cell types in atrial tissues. (B) UMAP plots showing the expression distribution of CCR2 (left) and PDE5A (right) across all cell types. (C) Dot plot showing the expression of annotated genes across 8 cell types. (D) Bar plot showing the proportion of all cell types between atrial fibrillation (AF) and sinus rhythm (SR) groups. (E–F) CCR2 expression in macrophages and lymphocytes between AF (right) and SR groups (left). ***P < 0.001.

https://doi.org/10.1371/journal.pone.0353143.g007

4.1. Overview of the present study

AF represents a prevalent and debilitating cardiac arrhythmia that severely compromises quality of life and elevates the risk of life-threatening complications such as stroke and heart failure [3]. For cancer patients receiving DOX-based chemotherapy, the incidence of DOX-induced AF has emerged as a critical clinical concern [31]. As for clinical practice, Toshio Kinoshita et al. [32] identified early electrocardiographic indices for predicting chronic doxorubicin-induced cardiotoxicity. However, the specific molecular mechanisms underlying DOX-induced AF remain poorly elucidated. In the present study, a multi-omics strategy integrating transcriptomics, network toxicology, single-cell analysis, and molecular docking was applied to identify key hub genes, including CCR2 and PDE5A, and characterize their functions, offering new perspectives for therapeutic intervention. Molecular docking confirmed stable binding of DOX to the predicted targets CCR2 and PDE5A. Finally, single-cell analysis was leveraged to resolve the cell-type-specific expression patterns of the key hub genes within the atrial tissue microenvironment. Collectively, this framework identified CCR2 and PDE5A as potential therapeutic targets for DOX-induced AF and clarified their roles in specific atrial cell populations.

4.2. The role of CCR2 in the DOX-induced AF

CCR2, a member of the C-C chemokine receptor family, serves as the main chemokine receptor of CCL2 and plays a pivotal role in regulating immune cell recruitment and inflammation [12]. Our GSEA results revealed that CCR2-related genes were enriched in pathways such as antigen processing and presentation, hematopoietic cell lineage, and Th17 cell differentiation. APCs such as macrophages, may initiate and amplify an adaptive immune response. CCR2 likely facilitates this cellular foundation by recruiting more monocytes/macrophages to the heart, thereby directly promoting a chronic inflammatory state and immune-mediated tissue remodeling. Moreover, Yang et al. demonstrated Th17/Treg ratio in serum predicts onset of postoperative AF [33].

Atrial macrophages and their monocyte-derived CCR2 + subset expanded most among patients with AF and mitral regurgitation were also observed by Hulsmans et al. [29], which was further validated in Ccr2 − / − mice. Corresponding with the present study, Miyosawa et al. [34] demonstrated the number of CCR2 + monocytes/macrophages was high in left atrial appendages resected from patients with enlarged left atrial. In another study, Wu et al. [35] elucidated CCR2 was differentially methylated and differentially transcribed in valvular atrial fibrillation (VAF) left atrial tissue.

4.3. The role of PDE5A in the DOX-induced AF

PDE5A, a phosphodiesterase that hydrolyzes cyclic guanosine monophosphate (cGMP), is a well-characterized regulator of cardiac contractility and vascular tone [36]. Increased expression/activity of PDE5A was detected in human patients with cardiac diseases [37]. Moreover, PDE5 inhibition was aproved to attenuate cardiomyocyte apoptosis and left ventricular dysfunction in a chronic model of DOX-induced cardiotoxicity [38] and diminish TGF-beta1 induced collagen synthesis of right ventricular cardiac fibroblasts in vitro induced by pulmonary artery banding [39]. PDE5A was found up-regulated both globally and in association with serine/threonine protein phosphatase type-1c (PP1c), which may point to a compensatory mechanism early in AF pathogenesis. A mendelian randomization study also elucidated genetically predicted PDE5 inhibition was significantly associated with a lower risk of AF [40].

In our study, the KEGG enrichment analysis showed that PDE5A was involved in the cGMP-PKG signaling pathway. Regarding this pathway, Huang et al. demonstrated silencing RGS7 attenuates AF progression by activating the cGMP-PKG signaling pathway [41]. In terms of GSEA analysis, leukocyte transendothelial migration was also identified as a mainly enriched in pathway in AF patients [42]. Similarly, in Zhao’s study [43], GO analysis observed the enrichment of regulating actin cytoskeleton organization in CREM-IbdeltaC-X mice with persistent atrial fibrillation. Apart from this, our study also observed that PDE5A is differentially expressed in fibroblasts between AF and SR patients, which was involved in atrial structural remodeling—including fibrosis and extracellular matrix (ECM) deposition [44,45]. In addition, PDE5 inhibitors have been shown to improve cardiac function and reduce arrhythmia risk in preclinical models [36], highlighting the therapeutic potential of targeting PDE5A in DOX-induced AF.

4.4. Cellular-level Alterations in DOX-induced AF

According to the significant results of immunoinfiltration analysis, all cell types have been elucidated associated with AF [46–48]. Single-cell analysis confirmed such change of macrophage-proportion in atria and revealed CCR2 was differentially expression in AF patients compared to SR controls. He et al. demonstrated that the proportions of embryonic fibroblasts and actively proliferating fibroblasts were increased using the single-cell dataset (GSE197518), which comprises freshly isolated atrial cardiofibroblasts [49]. However, our study observed a decreased proportion of fibroblasts in AF patients in the Single-cell analysis. It may partly be explained that intense immune cell infiltration diluted the cellular landscape, causing fibroblasts to appear proportionally reduced, or it may reflect genuine regional differences in tissue cellularity.

4.5. Clinical implications and translational potential

The core therapeutic paradigm emerging from our findings involves a dual-target strategy to disrupt the pathogenic cycle of DOX-induced AF: inhibiting CCR2 and modulating PDE5A activity. This approach aims to halt the progression from acute cardiotoxicity to chronic arrhythmogenic substrate by simultaneously targeting the key cellular effectors, such as macrophages and fibroblasts, within the atrial microenvironment. Translational candidates can be stratified based on their primary molecular target: (1) CCR2/CCL2 axis inhibitors. Pharmacologic agents such as Cenicriviroc, a dual CCR2/CCR5 antagonist, have demonstrated safety in clinical trials for inflammatory diseases [50]. Our data suggest that by blocking the CCL2-CCR2 chemotactic axis, these compounds could specifically reduce the recruitment and pro-inflammatory activation of monocyte-derived macrophages in the atria, thereby mitigating a primary driver of atrial inflammation and subsequent structural remodeling. (2) PDE5 Inhibitors. Repurposing existing PDE5 inhibitors (e.g., Sildenafil, Tadalafil) generates a testable hypothesis that warrants evaluation in future preclinical and clinical studies, rather than representing an immediately actionable clinical strategy. Preclinical studies have shown that PDE5 inhibition attenuates DOX-induced cardiomyocyte apoptosis and left ventricular dysfunction [38]. Our findings extend this rationale to the atria, suggesting that by enhancing cGMP-PKG signaling, PDE5 inhibitors could counteract the pro-fibrotic phenotype of cardiac fibroblasts and improve endothelial function, addressing both fibrotic and vascular components of AF substrate.

Future cell-type-specific drug delivery systems to maximize on-target efficacy while minimizing systemic effects are still needed, especially crucial for immunocompromised cancer patients. Additionally, biomarker-guided therapy—using the identified hub gene panel for patient stratification could pave the way for precision medicine in the management of DOX-induced AF. Clinically, patients with elevated CCR2 or PDE5A expression would be flagged as high-risk for developing AF, who might benefit most from the above findings. This would prompt clinicians to increase the frequency of ECG monitoring (e.g., weekly or bi-weekly) during DOX infusion, consider lower cumulative DOX doses or alternative anthracycline regimens, and potentially initiate targeted therapy in future clinical trials. Patients with high CCR2 expression could be considered for use of adjunctive anti-inflammatory therapy (e.g., colchicine or CCR2 antagonists) to reduce AF risk, while PDE5 inhibitors (e.g., sildenafil, tadalafil) for those with high PDE5A expression. However, we emphasize that these interpretations are hypothesis-generating and require validation in future in vitro, in vivo, and clinical studies before any clinical application can be considered.

4.6. Limitations

Despite the novel findings of this study, several limitations should be acknowledged. Firstly, the transcriptomic and single-cell datasets used were obtained from public databases, which may introduce biases related to patient selection, sample processing, and clinical heterogeneity. Future studies should validate the findings in prospective cohorts of DOX-treated cancer patients to confirm the predictive value of the hub genes and the efficacy of targeted interventions. Secondly, FAERS is a passive spontaneous reporting system, which introduces inherent reporting biases, including underreporting, incomplete indication coding, and overrepresentation of certain cancer types treated with commonly reported drugs. Reports are submitted voluntarily by healthcare providers, patients, and manufacturers, and therefore the database does not capture all adverse events that occur. Moreover, this database lacks standardized and detailed clinical data, such as tumor stage, disease progression, comorbidities and detailed treatment information, which may interfere the results. Moreover, the molecular docking analysis results are needed to be verified in vitro and in vivo experiments. Thirdly, while we identified immune cell infiltration as a key feature of DOX-induced AF, the specific mechanisms by which CCR2 + macrophages and PDE5A+ fibroblasts interact to promote atrial remodeling remain unclear. Future studies should use cell co-culture models and spatial transcriptomics to dissect these cellular crosstalk pathways. Nevertheless, this study provides novel evidence and mechanistic insights into how doxorubicin (DOX) initiates and perpetuates atrial fibrillation (AF), highlighting CCR2 and PDE5A as central mediators within specific cardiac cell populations, which may inform the development of precision cardioprotective strategies for cancer patients undergoing anthracycline-based therapies.

Use of Large Language Models, AI and Machine Learning Tools

No large language models or AI tools were used in the creation of this work.