Systems biology analysis reveals miRNA–mRNA networks associated with cardiomyocyte responses to SARS-CoV-2 infection

Abstract

Cardiac complications are common and clinically significant in COVID-19, yet their underlying molecular drivers remain poorly understood. Here, we integrate post-mortem histopathology with transcriptomic and microRNA (miRNA) analyses to delineate the regulatory architecture of SARS-CoV-2-induced myocardial injury. Histological examination of cardiac tissue from 29 deceased COVID-19 patients revealed pronounced immune infiltration, cardiomyocyte necrosis, and fibrotic remodeling. To gain new insights into these pathological processes, we first analyzed 42 transcriptomes from SARS-CoV-2-infected cardiomyocytes, including primary human, iPSC, and hESC-derived cells. We identified 871 differentially expressed genes (DEGs) in infected cardiomyocytes associated with immune activation, extracellular matrix (ECM) remodeling, and impaired contractile function. Based on these dysregulated genes, we then inferred miRNA–mRNA regulatory networks and, through miRTarBase analysis, we uncovered 331 miRNAs as putative regulators of these DEGs, including miR-29a, miR-145, and miR-199a, known to modulate fibrosis, ECM composition, and cardiomyocyte survival. Finally, selected candidates were evaluated in blood samples from COVID-19 patients. We profiled circulating miRNAs in plasma from COVID-19 patients and detected 32 dysregulated miRNAs, 11 of which overlapped with the predicted set. Notably, downregulation of miR-29a correlated with profibrotic signatures, while immune-regulatory miR-21 was upregulated in mild disease. Together, our multi-modal analysis reveals a miRNA-mRNA regulatory program orchestrating inflammation, fibrosis, and contractile dysfunction in the SARS-CoV-2-infected heart. These findings highlight molecular candidates for risk stratification and therapeutic targeting in COVID-19-associated cardiomyopathy.

Introduction

The Coronavirus Disease 2019 (COVID-19) pandemic, caused by SARS-CoV-2, rapidly evolved into a global public health crisis, producing a broad spectrum of clinical outcomes ranging from asymptomatic infection to severe viral pneumonia and acute respiratory distress syndrome (ARDS)^1,2. Beyond the lungs, SARS-CoV-2 exerts a profound impact on the cardiovascular system. Myocarditis, myocardial infarction, arrhythmias, pericarditis, thromboembolic events, and other cardiac complications have been widely reported, many of which are associated with worse clinical outcomes^3–5. These manifestations may stem from direct viral cardiotoxicity and endothelial dysfunction or arise indirectly from hyperinflammatory states such as cytokine storms, leading to long-term myocardial damage. Notably, myocardial injury, as indicated by elevated cardiac troponin levels above the 99th percentile, is the most prevalent cardiovascular abnormality among COVID-19 patients and is strongly associated with increased mortality^4,6. Furthermore, children with COVID-19 may develop a hyperinflammatory syndrome resembling Kawasaki disease, involving cardiac dysfunction and vascular inflammation⁷. Despite extensive clinical data, the molecular mechanisms linking SARS-CoV-2 infection to cardiac injury remain incompletely understood.

A growing body of evidence highlights the maladaptive interactions between cardiac stromal cells and infiltrating immune cells as a key driver of unresolved inflammation and fibrotic remodeling in the myocardium^8–10. Under physiological conditions, the crosstalk between these cell types facilitates inflammation resolution and tissue repair^9,11. However, during SARS-CoV-2 infection, dysregulated immune–stromal communication can lead to excessive extracellular matrix (ECM) deposition and fibrosis. This pathological remodeling impairs myocardial contractility and underlies the progression to chronic cardiac dysfunction^4,12. Thus, deciphering the molecular signals governing this interface is critical to uncovering new therapeutic strategies aimed at mitigating cardiac sequelae in COVID-19.

Among the most promising molecular regulators of cardiac and immune responses are microRNAs (miRNAs), small non-coding RNAs that exert post-transcriptional control over gene expression. In the past two decades, miRNAs have emerged as essential regulators of cardiovascular development, homeostasis, and disease^13–16. They orchestrate key biological processes implicated in the pathophysiology of cardiovascular disease (CVD) including inflammation, apoptosis, angiogenesis, and ECM remodeling. Moreover, miRNAs are increasingly recognized for their roles in infectious diseases, modulating host antiviral responses, cytokine production, and viral replication^17,18. These versatile molecules not only act as biomarkers of disease severity but also hold therapeutic potential due to their capacity to fine-tune cellular programs involved in tissue damage and repair.

Despite their relevance, the landscape of miRNA-mediated regulation in cardiomyocytes following SARS-CoV-2 infection remains poorly characterized. Here, we sought to address this gap by integrating transcriptomic analyses and miRNA profiling to identify cardiomyocyte-intrinsic regulatory networks associated with immune signaling, sarcomeric gene expression, and ECM-related responses following SARS-CoV-2 exposure. Our findings reveal novel miRNA–mRNA interactions with potential functional relevance to COVID-19-associated cardiac injury and provide a foundation for miRNA-based diagnostic and therapeutic strategies targeting cardiovascular complications of SARS-CoV-2.

Results

Histological Evidence of Cardiac Inflammation and Fibrotic Remodeling in Fatal COVID-19

To explore the histopathological basis of cardiac dysfunction observed in severe cases, we analyzed myocardial tissue from 29 individuals who died from COVID-19 during the first pandemic wave in Brazil (2020). The cohort had a mean age of 59.6 ± 17.3 years, with a mean hospital stay of 9.3 ± 5.5 days prior to death, reflecting severe clinical progression. All patients required hospital-level care, including 53% who underwent mechanical ventilation and 47% who received supplemental oxygen by nasal cannula (Supplementary Data 1). The majority presented with significant comorbidities, such as hypertension (58%), obesity (48%), and diabetes (44%) were the most common conditions, followed by a history of smoking (31%), heart disease (27%), lung disease (27%), kidney disease (17%), history of stroke (10%), and cancer (6%) (Supplementary Fig. 1). These data reflect the high burden of chronic illness in this cohort and emphasize the complexity of the clinical context contributing to COVID-19 severity and cardiac involvement. Despite diverse clinical backgrounds, a convergent histological pattern emerged, suggestive of a common pathogenic process driven by viral-induced inflammation. Hematoxylin and eosin-stained slides revealed cytotoxic lymphocyte-mediated cardiomyocyte necrosis accompanied by dense immune infiltrates (Fig. 1), alongside architectural disarray of muscle fibers, perinuclear vacuolization and perinuclear edema, nuclear hyperchromasia, and extensive interstitial fibrosis. These alterations reflect a maladaptive tissue response in which excessive immune activation disrupts myocardial integrity and promotes fibrotic remodeling.

Fig. 1. Histopathological Features of Myocardial Injury in COVID-19.

Representative H&E-stained sections of cardiac tissue from COVID-19 patients demonstrating hallmark lesions. The squares are emphasized and shown at a higher magnification in the right panel: (1) Necrosis due to attack by cytotoxic lymphocytes; (2) Immune infiltrate/ lymphocyte attacking; (3) Fibers of different sizes (thick, wide, irregular); (4) Degenerative perinuclear edema; (5) Lymphocyte attacking/ hyperchromatic nucleus; (*) Myocardial interstitial fibrosis with dense ECM accumulation.

Of note, elevated systemic inflammatory markers, such as D-dimer (mean 3.6 µg/mL) and CRP (mean 13.0 mg/dL), support the presence of widespread immune activation (Supplementary Data 1). These findings reinforce the notion that myocardial injury in COVID-19 may result not only from direct viral effects but also from sustained immune dysregulation, ultimately impairing cardiac function through persistent extracellular matrix (ECM) deposition and scar formation.

Immune Activation and Fibrotic Signatures in SARS-CoV-2-Infected Cardiomyocytes

To gain new insights into cardiomyocyte transcriptional programs potentially associated with the cardiac pathology observed in fatal COVID-19, we performed an integrative transcriptomic analysis of SARS-CoV-2-infected cardiomyocytes (CMs). We curated and analyzed RNA-seq data from four independent datasets (GSE150392, GSE151879, GSE156754, GSE162736), including primary human cardiomyocytes, human embryonic stem cell-derived CMs (hESC-CMs), and induced pluripotent stem cell-derived CMs (iPSC-CMs) subjected to different viral loads and durations of exposure (Fig. 2A, Supplementary Data 2). The number of differentially expressed genes (DEGs) varied across experimental conditions, likely reflecting differences in cell models, viral exposure parameters, and transcriptomic responses across datasets.

Fig. 2. Transcriptomic Landscape of SARS-CoV-2-Infected Cardiomyocytes.

A Summary of RNA-seq datasets, showing cell types, viral load (MOI), exposure duration, and total number of DEGs. B Heatmap of enriched Gene Ontology (GO) terms across conditions. Only GO terms with log₂(adjusted p-value) > 2 in at least four models are displayed. Hierarchical clustering was performed using Euclidean distance. C Bar plot quantifying DEGs involved in immune response, muscle changes, ECM remodeling, and coagulation per dataset. D Circos plot linking overlapping genes across functional GO categories; ribbon thickness denotes gene overlap. E Density plots of log₂(fold change) values for DEGs per GO category, visualized using cubic spline regression.

Through functional enrichment and Gene Ontology annotation, we identified 1188 DEGs across four biologically relevant categories: immune response, extracellular matrix (ECM) remodeling, muscle contraction, and coagulation (Fig. 2B, Supplementary Data 3). Immune-related pathways, particularly those involving interleukin and interferon signaling, and toll-like receptors (TLR7/8 and TLR9), were broadly upregulated, consistent with an antiviral and inflammatory response (Supplementary Fig. 2, 3). This was mirrored by upregulation of platelet activation and fibrin clot formation pathways, implicating thromboinflammatory mechanisms in the cardiac response to viral insult. Moreover, ECM-related genes exhibited signatures of collagen deposition, matrix degradation, and integrin signaling, hallmarks of fibrosis and tissue remodeling. Concomitantly, genes governing muscle contraction and calcium handling were largely downregulated, pointing to structural and functional cardiomyocyte impairment (Fig. 2C–E, Supplementary Fig. 4, 5). Circos plot analysis highlighted the convergence of immune, matrix, and coagulation pathways, revealing shared molecular players likely orchestrating this maladaptive remodeling (Fig. 2D). Density plots further confirmed a global upregulation of immune transcripts and suppression of contractile genes (Fig. 2E), suggesting that SARS-CoV-2 alters cardiomyocyte homeostasis toward inflammatory and matrix-related signaling states, alongside reduced contractile programs. Although derived from in vitro models, these findings converge with the post-mortem histopathological evidence, marked by immune cell infiltration, myocyte necrosis, and interstitial fibrosis, highlighting cardiomyocyte transcriptomic dysregulation as a potential contributing component of SARS-CoV-2-associated cardiac injury.

Systemic transcriptional alterations and tissue remodeling associated with SARS-CoV-2 infection in cardiomyocytes

To define the transcriptional landscape underlying subsequent miRNA–mRNA network analyses, we investigated molecular functions enriched among genes dysregulated in SARS-CoV-2-infected cardiomyocytes. Functional enrichment analysis identified recurrent alterations involving immune signaling, extracellular matrix (ECM) organization, coagulation, and cardiac muscle-related processes (Fig. 3A and Supplementary Fig. 6). Immune-associated pathways showed extensive overlap with ECM remodeling and coagulation, suggesting coordinated activation of inflammatory and reparative transcriptional programs within cardiomyocytes. In contrast, genes associated with cardiac muscle function formed a more distinct module, with CAV1 and CXCR4 representing potential molecular links between inflammatory signaling and cardiac remodeling (Fig. 3B). Among the dysregulated pathways, ECM-related genes were particularly prominent. The transcriptomic profile revealed increased expression of genes associated with matrix organization and remodeling, including COL1A1, COL8A1, COL11A1, TGFB2, LOX, and POSTN, supporting the activation of pro-fibrotic transcriptional programs in cardiomyocytes (Fig. 3C and Supplementary Fig. 7). Although ECM deposition in vivo is primarily mediated by fibroblasts, these findings suggest that infected cardiomyocytes may contribute to tissue remodeling through regulatory and paracrine signaling mechanisms following viral perturbation. To relate these molecular alterations to tissue architecture, myocardial sections from COVID-19 patients were analyzed by fractal dimension analysis. Cardiac sections showed increased extracellular matrix accumulation and marked tissue disorganization, reflected by an elevated fractal dimension of the ECM network, whereas nuclear organization exhibited comparatively smaller changes (Fig. 3D). Together, these findings indicate that SARS-CoV-2 infection is associated with coordinated transcriptional changes in cardiomyocytes that may contribute to systemic cardiac remodeling through intercellular communication and extracellular matrix-associated signaling.

Fig. 3. Cardiomyocyte Transcriptional Programs Associated with Extracellular Matrix Remodeling and Myocardial Disorganization.

A Dot plot of GO enrichment for immune, ECM, muscle, and coagulation-related terms; dot size = gene count, color = adjusted p-value. B Gene interaction network showing selected genes expressed in cardiomyocytes and their connectivity across functional categories (e.g., CAV1, CXCR4). C GO enrichment for each cluster: Cluster 1 = collagen organization; Cluster 2 = matrix degradation and integrin pathways. D H&E-stained section and binarized ECM/nuclei images used for fractal analysis. ECM (D = 1.1992) and nuclear (D = 1.5035) patterns indicate structural disarray and cellular disorganization, respectively. The ECM-related signatures identified in cardiomyocytes are interpreted as transcriptional activation of pro-fibrotic signaling programs, which may indirectly contribute to myocardial fibrosis through intercellular communication.

Post-transcriptional microRNA networks link inflammation, fibrosis, and contractile remodeling in SARS-CoV-2-infected cardiomyocytes

After identifying the DEGs associated with myocardial remodeling in SARS-CoV-2-infected cardiomyocytes, we next sought to investigate how non-coding RNAs might orchestrate these transcriptional programs. We focused on microRNAs (miRNAs), which post-transcriptionally fine-tune gene expression in both health and disease.

From the DEG pool, we predicted miRNAs likely to regulate immune, ECM, and cardiac contractile genes, selecting the top 40 candidates based on combined enrichment scores across all experimental models (Fig. 4A). Hierarchical clustering (Euclidean distance) revealed discrete groups of miRNAs with overlapping regulatory patterns and functional convergence across biological processes related to inflammation, fibrosis, and myocardial architecture. Network modeling of immune-related gene regulation revealed miR-21-5p, miR-34a-5p, and miR-145-5p as prominent regulatory hubs (Fig. 4B). These miRNAs are implicated in key inflammatory pathways, including cytokine signaling, immune cell recruitment, and leukocyte activation. In the ECM cluster, the miR-29 family (miR-29a-3p, miR-29b-3p, miR-29c-3p) emerged as dominant regulators of fibrosis (Fig. 4C), consistent with their established roles in suppressing collagen synthesis and promoting ECM turnover¹⁹. These miRNAs are also known to act within the circ-sh3rf3/GATA-4/miR-29a axis to regulate fibroblast-to-myofibroblast transition and fibrotic remodeling²⁰. In parallel, miR-145-5p, a key modulator of myofibroblast differentiation and collagen deposition, was enriched in our network²¹, suggesting a synergistic pro-fibrotic miRNA profile.

Fig. 4. MicroRNA Regulatory Networks in SARS-CoV-2-Infected Cardiomyocytes.

A Bubble heatmap of top 40 predicted miRNAs targeting DEGs; size = significance (–log10 p-value), color = number of targets. Hierarchical clustering was performed using Euclidean distance considering rows and columns. B–D miRNA–target interaction networks for: B immune regulation; C ECM remodeling; D muscle contraction. Notable hubs: miR-21-5p, miR-145-5p (immune); miR-29 family (ECM); miR-1-3p, miR-199a-5p (muscle). Networks illustrate post-transcriptional regulation of cardiac pathophysiology. E Molecular network illustrating miRNAs (grey triangles) predicted to regulate genes (colored circles) implicated in cardiomyocyte alterations. Node size reflects the degree of connectivity (number of edges). Edges represent documented or predicted miRNA–mRNA interactions. Gene nodes are color-coded according to functional categories: immune response (purple), muscle contraction (pink), ECM remodeling (blue), and coagulation (green).

Cardiac muscle–specific miRNAs formed a distinct regulatory circuit (Fig. 4D). Downregulation of miR-1-3p and miR-133a-3p, critical regulators of sarcomere integrity and cardiac contractility, has previously been associated with maladaptive remodeling and dysfunction²². We also identified miR-199a-5p, whose overexpression induces cardiomyocyte elongation and eccentric hypertrophy resembling dilated cardiomyopathy²³. The integrative network analysis (Fig. 4E) further revealed a dense regulatory landscape, with miRNAs targeting multiple genes within overlapping functional categories. Altogether, these findings highlight a coordinated network of miRNAs linking immune responses, extracellular matrix remodeling, and myocardial contractility in SARS-CoV-2-exposed cardiomyocytes, shedding light on the post-transcriptional mechanisms underlying COVID-19-associated cardiac injury.

Circulating miRNA signatures validate key regulatory candidates linked to COVID-19 cardiac damage

To experimentally validate our transcriptome-derived miRNA predictions in a clinically accessible sample, we profiled circulating miRNAs in whole blood from patients with COVID-19 with varying disease severity. Recognizing the inherent limitation of post-mortem cardiac tissue analyses, namely, RNA degradation and postmortem transcriptional changes, we instead focused on a liquid biopsy approach that enables in vivo insights into host regulatory responses. A total of 12 blood samples were analyzed, comprising individuals without infection, patients with COVID-19 not requiring intensive care, and patients with COVID-19 admitted to the ICU. Patients hospitalized at Hospital Beneficência Portuguesa were classified as having moderate (non-ICU) or severe (ICU) disease according to supplemental oxygen requirements and intensive care unit admission. Basic demographic information stratified by disease severity is provided in Supplementary Data 4. Blood samples were collected from healthy donors (n = 4), COVID-19 patients not requiring intensive care (non-ICU, n = 4), and patients admitted to the ICU with severe COVID-19 (n = 4), followed by profiling of 694 miRNAs using TaqMan Low-Density Arrays (TLDA) (Fig. 5A; patient details in Supplementary Data 4). Principal component analysis revealed distinct transcriptional signatures between groups, with ICU patients forming a separate cluster, indicating a unique miRNA expression profile associated with severe disease (Fig. 5B). Volcano plots comparing groups revealed significant differential expression of miRNAs between ICU and control, non-ICU and control, and non-ICU and ICU cohorts (Fig. 5C). Specifically, 13 miRNAs were differentially expressed in ICU vs. control, 11 in non-ICU vs. control, and 14 in ICU vs. non-ICU patients, supporting the hypothesis that disease severity modulates miRNA signatures. Of the Top 40 predicted candidate miRNAs, 34 were represented in the 694-miRNA TLDA panel, whereas 6 were not included in the assay. Among the 34 evaluated miRNAs, 11 (32.4%) were significantly dysregulated in plasma samples from COVID-19 patients (Fig. 5D), while the remaining candidates represented in the panel did not reach statistical significance in our cohort. This convergence reinforces the biological plausibility of these miRNAs as systemic indicators of COVID-19-associated cardiac manifestations. Among these, miR-29a-3p, miR-145-5p, and miR-199a-5p emerged as central regulators, collectively targeting a broad spectrum of genes implicated in immune signaling, extracellular matrix remodeling, and cardiac contractility (Fig. 5E). The translational relevance of these miRNAs is underscored by their potential utility as circulating biomarkers and therapeutic targets in COVID-19-induced cardiac injury.

Fig. 5. Circulating miRNA Profiles Reveal COVID-19 Severity Signatures.

A Workflow of circulating miRNA profiling by TLDA in plasma samples from healthy donors (n = 4), non-ICU COVID-19 patients (n = 4), and ICU COVID-19 patients (n = 4). B Principal component analysis (PCA) showing separation among the three groups, with ICU samples forming a distinct cluster. C Volcano plots showing differentially expressed circulating miRNAs between the study groups. D Venn diagram illustrating the overlap between the Top 40 candidate miRNAs predicted from cardiomyocyte transcriptomic meta-analysis and the 694-miRNA TLDA panel. Of the 40 predicted miRNAs, 34 were represented in the TLDA array, and 11 (32.4%) were significantly dysregulated in plasma samples from COVID-19 patients. The remaining predicted miRNAs were either not significantly altered in plasma or were not represented in the TLDA panel. E Functional classification of target genes regulated by the 11 shared miRNAs, grouped according to immune response, extracellular matrix remodeling, coagulation, and cardiac muscle-related processes.

To further investigate the relevance of the predicted miRNA–mRNA networks, we re-analyzed the sequencing datasets used in this study to identify miRNAs detected in SARS-CoV-2-exposed cardiomyocyte models. Although these datasets were not specifically generated for small RNA profiling, a total of 345 miRNAs showed sufficient abundance across the four datasets. Among these, 14 miRNAs overlapped with the Top 40 candidate miRNAs predicted from the transcriptome-based meta-analysis, while 15 miRNAs overlapped with dysregulated circulating miRNAs identified in plasma samples from COVID-19 patients (Supplementary Fig. 8). Notably, muscle-enriched miRNAs such as miR-1 and miR-133 were detected in cardiomyocyte datasets but not in plasma, consistent with their tissue-enriched expression profile. In contrast, miR-199a and miR-454 were detected both in cardiomyocytes and patient plasma, supporting their potential relevance as cardiac-associated regulators and circulating biomarkers.

Discussion

Through integrative multi-omic profiling and histopathological validation, we propose a mechanistic model (Fig. 6) illustrating how SARS-CoV-2 infection orchestrates a sequence of cellular and molecular events culminating in myocardial fibrosis and contractile dysfunction. Our findings support a model in which SARS-CoV-2 exposure induces cardiomyocyte-intrinsic transcriptional response linked to immune cell recruitment and cytokine release, processes modulated by microRNAs such as miR-145-5p and miR-126-3p. These mediators contribute to cellular damage, while concurrently, fibroblasts differentiate into myofibroblasts under the control of miR-199a-5p and the miR-29 family, promoting ECM deposition. As fibrosis advances, architectural disarray and contractile loss consolidate a chronic pathological state, as supported by prior reports^4,8,12,24

Fig. 6. Model of COVID-19-Induced Myocardial Fibrosis.

Illustrative model of five sequential stages: A Homeostasis: Healthy myocardium. B Inflammation: Immune infiltration and cytokine storm; regulated by miR-145-5p, miR-126-3p, miR-21-5p, miR-454. C ECM Remodeling: Myofibroblast differentiation and ECM deposition; regulated by miR-199a-5p and miR-29 family. D Interstitial Fibrosis: ECM excess and apoptosis. E Fibrotic Scar: Persistent myofibroblast activity leads to scar formation and loss of elasticity; ECM-regulatory miRNAs remain elevated. Original artwork created for this study by the authors.

Comparative transcriptomics of primary, iPSC-derived, and hESC-derived cardiomyocytes revealed distinct immune-related gene expression profiles. iPSC-derived cardiomyocytes demonstrated broader transcriptional reprogramming, possibly due to differences in epigenetic memory or regulatory architecture when compared to hESC-derived counterparts²⁵. Ethical limitations and restricted accessibility of hESCs in many countries further constrain their use²⁶. Although primary cardiomyocytes offer biological fidelity, their limited proliferative capacity and accessibility pose practical challenges^27,28. Our synthesis highlights the relative merits and limitations of each model in COVID-19 cardiac research.

Despite heterogeneity in DEGs, enrichment analyses showed convergence across inflammatory, ECM, coagulative, and contractile processes. To refine our understanding, we explored the role of microRNAs, key post-transcriptional regulators of gene expression²⁹. Among these, miR-29a and miR-145 were prominently downregulated in infected patients. The downregulation of miR-29a is in line with its known role as an anti-fibrotic regulator targeting multiple ECM components (COL1A1, COL3A1, ELN, FBN1)^30–32 and acting through the Circ-sh3rf3/GATA-4/miR-29a axis to inhibit myofibroblast differentiation²⁰. Similarly, reduced miR-145-5p expression in ICU samples is associated with upregulation of alpha-smooth muscle actin (α-SMA) and collagen Vα1, favoring myofibroblast persistence and maladaptive fibrosis^33,34.

Moreover, miR-29a has been identified as a key player in the Circ-sh3rf3/GATA-4/miR-29a regulatory cascade, which influences fibroblast-to-myofibroblast differentiation and myocardial fibrosis²⁰. Myofibroblasts are key players in fibrosis and scar formation, as they produce excessive ECM components like collagen. They are characterized by the expression of α-SMA and are involved in the production of extracellular matrix components. In this context, miR-145-5p is also involved in promoting myofibroblast differentiation and collagen deposition, enhancing the fibrotic response^33,34. In our study, we found a downregulation of this miRNA in the blood of ICU patients. In heart tissue, it was demonstrated that when miR-145-5p is inhibited (downregulated), it leads to the upregulation of collagen V α1 and α-SMA, both of which are markers of myofibroblast differentiation²¹. The upregulation of collagen V α1 and α-SMA suggests that miR-145-5p inhibition promotes a pro-fibrotic environment that restricts scar resolution and impairs cardiac repair²¹. Importantly, pathways related to ECM remodeling and immune signaling identified here should be interpreted as cardiomyocyte-intrinsic transcriptional responses with potential paracrine relevance, rather than direct evidence of tissue-level fibrosis or immune infiltration. We also identified reduced miR-199a levels in both non-ICU and ICU patients. Looking in the literature, we found that this miR-199a has dual roles in myocardial injury, acting both as a regulator of a hypoxia-triggered pathway³⁵ and a potential inducer of regeneration³⁶. Inhibiting miR-199a early after injury may reduce apoptosis and oxidative stress, improving cardiomyocyte survival³⁵. Conversely, controlled overexpression of miR-199a at later stages might promote cardiomyocyte proliferation and regeneration, aiding in tissue repair³⁶.

In our network analysis and validation experiments, we identified miR-21 as a regulator of 131 immune response-related targets. We also observed a subtle overexpression of this miRNA in non-ICU patients, with a more prominent increase in 2 out of 4 patients from this group. It is important to mention that miR-21 plays a pivotal role in shaping immune responses by promoting Th2 immunity and creating an anti-inflammatory environment, primarily by suppressing Th1-associated cytokines in T cells and phagocytes³⁷. Its upregulation in immune cells contributes to immune modulation in inflammatory diseases and infections, while its inhibition enhances pro-inflammatory Th1 responses and M1 macrophage polarization, underscoring its critical role in maintaining immune homeostasis^38,39. Furthermore, our interpretation regarding immune- and ECM-related pathways are based on transcriptomic analyses. Although these findings reveal activation of regulatory programs in cardiomyocytes, protein-level validation of cytokines, inflammatory mediators, and ECM-associated molecules would provide important functional support for these observations.

The miRNAs miR-193-5p, miR-148, miR-451, and miR-369, previously described and highlighted in recent studies investigating circulating miRNAs in COVID-19 patients^40–42, were not identified in our validation analysis. These discrepancies between our findings and previous reports can be attributed to different populations and several methodological factors. Firstly, population factors such as genetic ancestry, environment, cultural habits, and socioeconomic conditions can significantly influence miRNA expression levels in response to the same pathological condition. Those genetic differences between populations affect miRNA regulation, modulating immune response and susceptibility to cardiac complications, which can explain the inter-individual variation. Additionally, the differences in the technological platforms used for miRNA analysis and the number and characteristics of the samples analyzed, such as disease stage, clinical severity, and methods of sample collection and processing, can substantially influence the results obtained. These methodological variations underscore the complexity of direct comparisons between different studies involving circulating miRNAs, highlighting the need for standardization of experimental protocols and subsequent analyses to achieve more consistent and comparable results in the scientific literature.

We acknowledge that the absence of control histology images represents a limitation of the study. However, the minimally invasive autopsy protocol was approved by the ethics committee specifically for deceased COVID-19 patients, and did not include the collection or analysis of postmortem tissues from non-COVID-19 control individuals. Therefore, matched control histological images were not available for inclusion in this study. Moreover, it is important to acknowledge that our validation analysis was conducted on a limited sample size, comprising only four individuals per group, for a total of 12 participants. This small cohort inherently constrains the statistical power and generalizability of our findings, underscoring the need for future studies with larger, more diverse populations to confirm and extend these results. Additionally, our miRNA analysis relied on target prediction derived from cardiomyocyte transcriptomic data, combined with validation in circulating samples from COVID-19 patients. While this integrative approach provides indirect support for the involvement of these miRNAs, it does not establish a direct causal link between SARS-CoV-2 infection and miRNA dysregulation in cardiomyocytes, representing a limitation of the study.

In conclusion, this study offers a multi-omic, integrative view of the transcriptional and post-transcriptional changes induced by SARS-CoV-2 infection in cardiac tissue. It uncovers novel regulatory miRNA–mRNA networks associated with immune responses and fibrotic remodeling. These findings contribute to a mechanistic understanding of COVID-19-associated cardiac injury, and provide insights for the development of RNA-based diagnostics and therapeutics for preventing post-viral cardiomyopathy.

Methods

Fractal Dimension Analysis of Cardiac Muscle Histology

To quantitatively analyze the nuclear disorganization in cardiac muscle tissue using fractal dimension analysis, we examined histological images from 29 individuals who died from COVID-19. This study was approved by the Ethics Committee of Faculdade de Medicina de Ribeirão Preto, CAAE: 32475220.5.000.5440. All the patients provided written informed consent. Tissue samples were obtained from deceased COVID-19 patients through ultrasound-guided minimally invasive autopsy using biopsy needles and/or a dedicated sampling device. After collection, the samples were fixed overnight in 10% neutral buffered formalin. Subsequently, tissues were processed through standard dehydration and clearing steps using graded alcohol and xylene baths, embedded in paraffin, and sectioned at 5 μm thickness. Hematoxylin and eosin staining was performed, and histological sections were analyzed by bright-field microscopy using a Leica DMI6000B microscope. The study aimed to determine whether the hearts exhibited fibrotic scars and assess the feasibility of quantifying this using fractal analysis. Photographed histological slides were binarized with ImageJ, applying the X thresholding method. The fractal dimension (FD) was determined using the box-counting method. This approach estimates the relationship between resolution and scale, providing a quantitative measure of the fractal complexity of the object. The FD was computed as FD = $\frac{\log N_r}{\log r-1}$ where Nr is the number of boxes required to cover the object, and r is the scale. Values range from 0 to 2, with higher FD indicating greater disorganization of nuclei. This approach enabled us to provide a quantitative measure of histological disruption consistent with fibrotic remodeling in the cardiac tissues of COVID-19 victims.

Selection of Transcriptomic Datasets for Cardiomyocytes

High-throughput sequencing expression profiles (published up to March 9, 2021) were retrieved from the GEO database (http://www.ncbi.nlm.nih.gov/geo/) using the keywords “SARS-CoV-2” and “cardiomyocytes”. This search yielded four datasets, comprising experiments conducted on primary cardiomyocytes (CMs), induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs), and human embryonic stem cell-derived cardiomyocytes (hESC-CMs). The datasets included in the study were GSE150392⁴³, GSE151879⁴⁴, GSE156754⁴⁵ and GSE162736⁴⁶. A total of 42 SARS-CoV-2-treated samples and 36 control samples were selected for analysis. The dataset descriptions are available in Supplementary Data 2. This curated selection provides a robust foundation for understanding the transcriptomic effects of SARS-CoV-2 infection on cardiomyocytes derived from different sources.

Differential Expression Analysis and Selection of Analyzed Genes

We used the NetworkAnalyst 3.0 tool⁴⁷ to identify differentially expressed genes (DEGs) between SARS-CoV-2-infected cardiomyocytes and the control group. Each dataset was log2-transformed, and the DESeq2 method⁴⁸ was employed for differential expression analysis of count data. A cutoff of fold change (FC) ≥ 2 and adjusted p-value < 0.05 was applied. This analysis was performed separately for each dataset. Data were normalized within each dataset, and batch-related variables were included using the “This is a blocking factor” option to control for technical variability during statistical analysis. We did not merge because they differed in cell type, MOI, exposure time, and sequencing methodology. Instead, each dataset was analyzed independently, and results were integrated by identifying shared differentially expressed genes and recurrent enriched pathways across studies.

To investigate genes associated with immune response, cardiac muscle contraction, extracellular matrix remodeling, and coagulation, we selected ontological terms through the QUICKGO platform (https://www.ebi.ac.uk/QuickGO/). The genes included in the analysis corresponded to the following terms: (1) Immune response-related terms: Inflammatory response (GO:0006954), Cytokine activation (GO:0005125), Cytokine binding (GO:0019955), Cytokine receptor interaction (GO:0090722), Viral process (GO:0016032). After removing duplicates; (2) Cardiac muscle contraction-related terms: Smooth muscle contraction (GO:0006939), Cardiac conduction (GO:0061337), Muscle contraction (GO:0006936), Striated muscle contraction (GO:0006941); (3) Extracellular matrix remodeling-related terms: Extracellular matrix organization (GO:0030198), Integrin activation (GO:0033622); and (4) Coagulation-related terms: Fibrinogen complex (GO:0005577), Platelet activation (GO:0030168), Blood coagulation (GO:0007596), Platelet aggregation (GO:0070527). The list of genes from each category is presented in Supplementary Data 3. This classification allowed us to systematically analyze DEGs within biologically relevant contexts and to identify genes potentially contributing to the pathological mechanisms of SARS-CoV-2 infection.

Functional Enrichment Analysis

The differentially expressed genes (DEGs) from each category were analyzed using the Enrichr tool (https://maayanlab.cloud/Enrichr/)⁴⁹. Terms with a p-value < 0.05 were considered significant for representation. Additionally, the WebGestalt tool was utilized, where terms were considered enriched if they had a p-value < 0.05 and a combined score > 10. Data generated by WebGestalt and the enrichment results from the Reactome Pathway Database were visualized using a volcano plot. For this, the R package clusterProfiler⁵⁰ was employed to create dot plots of enriched terms associated with the genes. This approach enabled a comprehensive visualization of significant pathways and functional categories linked to the DEGs.

Cubic Spline Analysis

To model the relationship between the percentage of genes within each expression level range, a cubic spline was applied. This statistical technique enables a smooth representation of variable relationships, facilitating the identification of trends within the data. In this study, the analyzed data represent the gene expression levels, expressed as [log₂(fold change)], within the four categories (immune, cardiac, ECM, and coagulation). Initially, the percentage distribution of genes was calculated for each expression interval ranging from -6 to 6. Subsequently, a cubic spline was fitted with 200 automatically distributed knots to model the relationship between this percentage distribution and the observed values. The implementation was performed using a generalized additive model (GAM) via the smooth.spline function from the mgcv library in the R environment, with the “cs” base type specified for the smoothing function. This approach ensures a flexible and robust modeling of gene expression trends.

RNA isolation

The total RNA was extracted from blood of patients using the TempusTM Blood RNA Tube (Thermo Fisher Scientific, USA), using the protocol recommended by the manufacturer. The RNA concentration, purity and integrity were assessed using the QUBIT®2.0 fluorometer (Life Technologies, USA), NanoDrop® ND-1000 spectrophotometer (Thermo Fisher Scientific Inc., USA), and 2200 TapeStation® system (Agilent Technologies,USA), respectively. Only RNA samples with A260/280 ratio of 1.8-2.0, A260/230 ratio > 2.0, and RIN > 7 were used for subsequent analysis.

Expression profiling of microRNAs

Global microRNA profiling was performed on blood samples from patients with COVID-19 not requiring intensive care (Non-ICU; n = 4), patients with COVID-19 admitted to the intensive care unit (ICU; n = 4), and non-infected controls (n = 4). Control individuals had no evidence of infection at the time of sample collection. Patients hospitalized at Hospital Beneficência Portuguesa were classified as having moderate (Non-ICU) or severe (ICU) disease based on supplemental oxygen requirements and ICU admission. This study was approved by the Ethics Committee from Real e Benemérita Associação Portuguesa de Beneficência/SP CAAE: 36730020.6.0000.5483 and Faculdade de Ciências Farmacêuticas da Universidade de São Paulo CAAE: 36730020.6.3001.0067; 36730020.6.3003.5504. All the patients provided written informed consent. MicroRNA expression was assessed using the TaqMan® Array Human MicroRNA Cards A & B v3.0 (Thermo Fisher Scientific, USA), covering 694 mature microRNAs, according to the manufacturer’s instructions. The RNA samples were reverse transcribed using the Megaplex RT Primers Pools A&B (Thermo Fisher Scientific, USA). We ran the samples on the QuantStudio™ 12 K Flex System (Thermo Fisher Scientific, USA) using the following cycle conditions: 95 °C for 10 min followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min. Small RNA MammU6 was used as a reference to normalize the microRNA data. This reference microRNA was selected based on geNorm calculations⁵¹. Finally, raw data from each card set was retrieved and imported into Expression Suite Software v1.0.3 (Thermo Fisher Scientific, USA). Relative quantification of microRNA expression was evaluated using the 2^-ΔΔCT method⁵². Cutoffs for significant changes were set at log2 fold-change >1 or < 1 and p-value ≤ 0.05.

Principal Component Analysis (PCA)

PCA was employed to assess the potential for stratification into three clinical groups, i.e., healthy controls, non-ICU COVID-19 patients, and ICU COVID-19 patients based on the expression of 694 miRNAs. The data were processed using R within the RStudio environment. Prior to PCA, the expression values were centered and scaled. Principal components were computed using the prcomp R function, and individual sample projections were visualized with the fviz_pca_ind function from the factoextra R package⁵³. Plots were generated using factoextra⁵³ and ggplot2⁵⁴. Confidence ellipses for each group were drawn based on 95% confidence intervals.

Computational Analysis of miRNAs and Their Targets

To explore the interactome between messenger RNA (mRNA) and miRNAs—small noncoding RNAs that regulate gene expression post-transcriptionally—miRNA target genes were identified using the miRTarBase database⁵⁵, considering only interactions validated by the Luciferase reporter assay method. We selected miRTarBase due to its diverse methods for validating miRNA-mRNA interactions, which range from strong to weak evidence. The reporter assay method was prioritized as it provides the strongest evidence for miRNA-target gene interactions.

Data Visualization and Molecular Networks

To represent the number of differentially expressed genes (DEGs), we used GraphPad Prism version 10.4.1. Bubble heatmaps for visualizing the DEGs were generated using the Morpheus tool (https://software.broadinstitute.org/morpheus/)⁵⁶. Shared genes across datasets, as well as genes within each of the four categories, were illustrated using a Circos plot (http://circos.ca/)⁵⁷. Additionally, shared genes across datasets were represented using an Upset plot, created with the Intervene tool (https://asntech.shinyapps.io/intervene/). The miRNA-miRNA and miRNA-target interaction networks were constructed using the Cytoscape bioinformatics tool, version 3.10.3. Additionally, to evaluate the miRNA interactome associated with GO terms, we used the cnetplot⁵⁸ function from the clusterProfiler⁵⁹ package in RStudio to construct a network of miRNAs linked to significantly enriched GO terms (p ≤ 0.05). Genes that were uniquely or commonly involved across immune, cardiac, extracellular matrix (ECM), and coagulation-related conditions were subjected to enrichment analysis for Biological Processes (BPs), allowing the exploration of miRNA–miRNA interaction networks. To further investigate the overlap of BP terms across these conditions, we applied the emapplot⁵⁸ function.

Declaration of generative ai technologies in the writing process

During the preparation of this work the author(s) used Grammarly and ChatGPT 4 in order to improve language and readability. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.

Author contributions

P.P.F. and O.C.M. - co-wrote the manuscript and provided scientific insights; P.P.F., D.L.M.F., I.S.F., D.G.A.F., F.Y.N. - performed bioinformatic analyses; P.P.F., L.O.L., M.I.H. – performed the miRNA experiments with human samples; P.P.F., F.P.V., S.S.B., A.T.F., F.Q. - responsible for histopathological analyses; F.Q., J.B.B., G.M.B., R.M.C., M.F.S., and M.H.H. - responsible for blood samples of patients with COVID-19; P.P.F., D.L.M.F., A.H.C.M. – Statistical analysis; P.P.F., L.F.S., M.R., R.F.C., H.N., H.O., N.O.S.C., V.L.G.C., O.C.M. - revised and edited the final manuscript; O.C.M., V.L.C., M.I.H., N.O.S.C., M.H.H. - infrastructure for analysis; M.H.H. and O.C.M. - supervised the project.

Data availability

The published bulk RNAseq datasets can be found in the GEO database (GSE150392, GSE151879, GSE156754 and GSE162736). The microRNA expression dataset generated using TLDA is available from the corresponding authors upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41540-026-00795-z.