Integrative transcriptomic analysis identifies an oxidative stress-associated diagnostic signature for myocardial infarction

Abstract

Reactive oxygen species (ROS) play a pivotal role in myocardial infarction (MI), contributing to oxidative stress, inflammation, and tissue remodeling. However, ROS-related gene signatures with diagnostic and mechanistic relevance in MI remain insufficiently defined. Transcriptomic data from six MI cohorts were integrated, with GSE66360 as the training set and five datasets as external validation. Batch correction was performed using ComBat. ROS pathway activity was assessed by single-sample gene set enrichment analysis (ssGSEA). Differentially expressed genes (DEGs) shared across datasets were intersected with ROS-related genes to construct an elastic net logistic regression model. Model performance was evaluated using ROC, calibration, and decision curve analysis. SHAP analysis was conducted for interpretability. Upstream transcription factor and miRNA interactions were predicted, and single-gene GSEA was used to explore biological pathways. Immune infiltration and checkpoint expression were analyzed using multiple deconvolution algorithms. Human AC16 cardiomyocytes were used to explore the functions of ADAM9 in MI. To validate our findings in vivo, we established the mice MI model and performed histology, immunostaining, and qRT-PCR to examine the six signature genes. ROS pathway activity was consistently elevated in MI samples across all cohorts. A six-gene signature (MMP9, ADAM9, BST1, TLR4, CLEC7A, CYP1B1) showed strong diagnostic performance. SHAP analysis identified MMP9 as the top contributor. Regulatory network analysis highlighted NFKB1, STAT1, and miR-21-5p as upstream regulators. Functional enrichment revealed an association with inflammatory and immune pathways. Software algorithm predictions from patient blood cell samples showed that the MI sample exhibited increased infiltration of macrophages, dendritic cells, and fibroblasts, along with upregulation of immune and inflammatory genes. Several model genes correlated positively with endothelial cell infiltration. Function studies in human AC16 cardiomyocytes suggested that ADAM9 inhibited cardiomyocyte survival and enhanced oxidative stress. Experimental validation confirmed that all six genes were significantly upregulated at both mRNA and protein levels in infarcted mouse hearts. We identified a ROS-related six-gene diagnostic signature for MI, with strong performance and mechanistic links to immune activation and vascular remodeling. This model may aid early diagnosis and provide insight into redox-immune interplay in MI.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11239-025-03205-z.

Introduction

Myocardial infarction (MI) continues to represent one of the leading causes of death globally, despite significant advances in early reperfusion strategies and pharmacological management [1].

A critical challenge in managing MI lies in minimizing reperfusion injury and adverse cardiac remodeling, processes in which reactive oxygen species (ROS) play a central role [2]. During ischemia–reperfusion, an abrupt surge in mitochondrial and NADPH oxidase–derived ROS leads to oxidative damage of lipids, proteins, and DNA, triggering cardiomyocyte death through pathways such as apoptosis, necroptosis, ferroptosis, and autophagy [3, 4].

ROS-driven oxidative stress also activates pro-inflammatory signaling via redox-sensitive transcription factors like NF‑κB and TLR4, amplifying leukocyte recruitment and exacerbating tissue injury [5, 6]. Additionally, matrix metalloproteinase-9 (MMP9), a well-characterized ROS-responsive enzyme, directly degrades extracellular matrix proteins and disrupts cardiac structure and vasculature after MI [7, 8]. Notably, elevated serum MMP9 levels have been linked to greater infarct size and adverse remodeling clinically, underscoring its potential as both a biomarker and therapeutic target.

Given the centrality of oxidative stress and immune activation in MI pathophysiology, integrating ROS-related gene expression into diagnostic or prognostic models has garnered growing interest [9]. Recent transcriptomic studies have developed gene signatures that reflect both redox burden and inflammatory activity, demonstrating promising diagnostic accuracy. However, many such efforts lack interpretability and comprehensive validation across diverse cohorts.

Here, we systematically integrate multiple MI cohorts to derive and validate a biologically interpretable, ROS-associated gene signature. Using single-sample gene set enrichment analysis (ssGSEA), differential expression profiling, elastic net modeling, and SHAP interpretability, we identified a robust six-gene panel-MMP9, ADAM9, BST1, TLR4, CLEC7A, and CYP1B1-that discriminates MI with high accuracy. We further investigate upstream transcriptional and post-transcriptional regulation, immune microenvironment interactions, and pathway enrichment, offering mechanistic insight into how oxidative stress-immune interplay underpins MI. We hypothesized that the ROS-related signature could be used for the diagnosis of MI, and ROS-related genes can promote the progression of myocardial infarction by facilitating the accumulation of ROS. This work advances the translational potential of ROS-biomarker-based precision diagnostics and suggests new molecular targets for therapeutic exploration.

Materials and methods

Data acquisition and preprocessing

Microarray gene expression data for MI and control samples were obtained from the Gene Expression Omnibus (GEO) database. The dataset GSE66360 was used as the training cohort, while GSE61144, GSE29111, GSE60993, GSE48060, and GSE166780 were combined to form the external validation cohort. Raw expression matrices were normalized using R (version 4.2.2), and batch effects across multiple datasets were corrected using the ComBat function from the “sva” package to ensure comparability across samples [10]. Samples of GSE66360 were circulating endothelial cells from the whole blood of MI patients and controls. Samples of GSE61144, GSE29111, GSE60993, and GSE48060 were the whole blood of MI patients and controls. Samples of GSE166780 were the peripheral blood monocytes of MI patients and controls.

Identification of ROS-Related pathways

Gene sets relevant to reactive oxygen species (ROS) biology were curated from the Molecular Signatures Database (MSigDB) [11], including five representative gene sets: (1) HALLMARK REACTIVE OXYGEN SPECIES PATHWAY; (2) GOBP CELLULAR RESPONSE TO REACTIVE OXYGEN SPECIES; (3) GOBP REACTIVE OXYGEN SPECIES BIOSYNTHETIC PROCESS; (4) GOBP REACTIVE OXYGEN SPECIES METABOLIC PROCESS; (5) GOBP RESPONSE TO REACTIVE OXYGEN SPECIES.

Pathway activity scoring and differential expression analysis

The ssGSEA was applied via the “GSVA” package to quantify the activation of ROS-related pathways across individual samples. Differences in pathway activity between MI and normal samples were compared using the Wilcoxon rank-sum test. Differentially expressed genes (DEGs) between the two groups were identified in both training and validation cohorts using the “limma” package.

Functional enrichment and intersecting gene analysis

GO enrichment (including biological process [BP], cellular component [CC], and molecular function [MF]) and KEGG pathway analyses were performed on the intersecting DEGs using the “clusterProfiler” package. Overlapping upregulated and downregulated genes across the training and validation datasets were visualized using Venn diagrams.

Construction of diagnostic model

Candidate ROS-related genes intersecting with DEGs were evaluated for expression correlations in both cohorts. An elastic net logistic regression model was constructed using the “glmnet” package with 10-fold cross-validation to optimize the lambda penalty parameter. Genes with non-zero coefficients at the optimal lambda were retained as diagnostic markers, and the model was used to compute individual risk scores.

Model performance assessment

The diagnostic performance of the ROS-based model was assessed by receiver operating characteristic (ROC) curves and area under the curve (AUC) metrics. Model calibration was visualized using calibration curves, while precision-recall (PR) curves were also plotted. Clinical net benefit was quantified via decision curve analysis (DCA).

Model explainability via SHAP analysis

To further interpret the contribution of individual genes in the ROS-related diagnostic model for MI, we employed SHapley Additive exPlanations (SHAP), a unified framework for model interpretability. SHAP feature importance was quantified by the mean absolute SHAP value for each gene, indicating its overall influence on model output. SHAP summary plots were generated to visualize the distribution of SHAP values for each feature across all samples, stratified by feature expression level. In addition, SHAP dependence plots were used to explore the relationship between gene expression (Z-score standardized) and SHAP value, thereby characterizing the functional response of the model to changes in gene expression for each predictor.

Construction of upstream regulatory networks and functional enrichment

To investigate potential upstream regulatory mechanisms for the key genes in the ROS-based diagnostic model, we constructed transcription factor (TF)-gene and microRNA-gene interaction networks. Transcription factors targeting model genes were identified using the TRRUST and JASPAR databases and visualized using Cytoscape. Similarly, high-confidence miRNA-mRNA interactions were obtained through the miRNet platform. In addition, single-gene GSEA was performed for each of the six model genes (MMP9, ADAM9, BST1, CLEC7A, TLR4, CYP1B1) using the clusterProfiler R package. Samples were ranked based on each gene’s expression level, and curated KEGG pathways were used as the reference gene set.

Immune cell infiltration estimation and immunogenomic correlation analysis

To assess the immune microenvironment characteristics associated with MI and the ROS-based diagnostic gene signature, we applied four complementary computational deconvolution algorithms: MCPcounter [12], quanTIseq [13], xCell [14], and EPIC [15]. These tools were used to estimate the relative abundance of various immune and stromal cell types from bulk transcriptomic data. Differences in cell infiltration levels between MI and control samples were evaluated using the Wilcoxon rank-sum test. In parallel, expression profiles of a curated panel of immune and inflammatory genes-including both inhibitory and stimulatory markers-were compared between groups and visualized using heatmaps and density plots. Lastly, Spearman correlation analysis was performed to investigate associations between the expression of the six model genes (BST1, ADAM9, CLEC7A, MMP9, CYP1B1, TLR4) and endothelial cell infiltration levels estimated by both xCell and EPIC.

Cell culture and transfection

Human AC16 cardiomyocytes acquired from Procell (CL-0790, Wuhan, China), and cultured in DMEM/F12 medium supplemented with 10% FBS, 100 U/mL penicillin, and 100 µg/mL streptomycin. shRNAs and complementary DNA (cDNA) plasmids of ADAM9 were obtained from Addgene (Beijing Zhongyuan Co., China). The sequences of shRNAs and cDNA plasmid of ADAM9 were provided in Table S1. Then, AC16 cells at 70% confluency were transduced with shRNAs and cDNA plasmids using lentiviral vectors. Knockdown and overexpression of ADAM9 were confirmed by quantitative reverse transcription polymerase chain reaction (qRT-PCR).

Functional validation of ADAM9 in human cardiomyocytes

Cell viability, proliferation, and apoptosis were assessed using CCK-8, EdU, and Annexin V/PI staining assays, respectively. For the CCK-8 assay, transfected cells were seeded in 96-well plates at 2,000 cells/well. At 0, 1, 2, and 3 day, 10 µL of CCK-8 reagent was added to each well, followed by a 2-hour incubation at 37 °C, and the absorbance was measured at 450 nm. To evaluate proliferation, an EdU assay was performed by incubating cells with 50µL EdU (RiboBio, Guangzhou, China) for 2 h, after which they were fixed, permeabilized, and stained for fluorescence microscopy imaging. Apoptosis was analyzed using an Annexin V-FITC/PI apoptosis detection kit (KeyGen Biotech, China). Briefly, the resuspended cells were stained with Annexin V-FITC and PI for 15 min in the dark and then analyzed by flow cytometry. To further investigate the oxidative stress and cell death pathways, intracellular ROS levels and apoptosis were detected using DCFH-DA fluorescent probe (Abcam, USA) and TUNEL staining (Abcam, USA), respectively. For ROS measurement, cells were incubated with DCFH-DA, which is oxidized by ROS in live cells to produce a fluorescent DCF signal. The fluorescence intensity was measured to quantify the ROS levels. Apoptosis was additionally confirmed by the TUNEL assay, which enzymatically labels DNA strand breaks, a hallmark of late-stage apoptosis. The TUNEL-positive cells were visualized and counted under a fluorescence microscope.

In vivo validation of the Six-Gene ROS signature in a mice MI model

To validate our ROS-related six-gene signature, we faithfully reproduced the mice MI model described by Ke et al. [16]. A total of 20 C57BL/6 mice were used in this study (10 for the MI group and 10 for the Sham group). Mice were randomly assigned to either the MI or Sham surgery group using a computer-generated random number sequence after acclimatization. The researchers performing the surgeries, post-operative care, and functional assessments were blinded to the group allocations throughout the experiment. Furthermore, all histological and immunofluorescence analyses were performed by investigators blinded to the experimental groups. Pre-defined criteria were applied. Mice were included only if they survived the full post-operative period. Mice were excluded from the analysis if they showed signs of severe distress, infection, or if the surgical procedure was technically unsuccessful. Adult male C57BL/6J mice (8–10 weeks) underwent permanent ligation of the LAD artery, while sham controls received an identical thoracotomy without ligation. Hearts were harvested 24 h later, and paraffin-embedded sections were stained with hematoxylin and eosin (HE) to confirm infarct-associated necrosis, inflammation, and tissue remodelling. Tissue sections were subjected to antigen retrieval by heating in citrate buffer (pH 6.0) for 20 min. CLEC7A expression was assessed by immunohistochemistry, and MMP9, ADAM9, BST1, TLR4, and CYP1B1 by immunofluorescence, with primary antibodies incubated overnight at 4 °C and corresponding secondary antibodies for 1 h at room temperature; nuclei were counterstained with DAPI. Images were captured using confocal or light microscopy. The primary antibodies used in this study were described as follows: MMP9 (A0289, 1:50, Abclonal), ADAM9 (A22058, 1:200, Abclonal), BST1 (A9900, 1:50, Abclonal), TLR4 (A11226, 1:50, Abclonal), CYP1B1 (A1377, 1:50, Abclonal), CLEC7A (CSB-PA126836, 1:10, CUSABIO), α-actinin (ab68194, 1:200, Abcam). The qRT-PCR of cardiac RNA normalized to GAPDH quantified mRNA levels of the six genes. The primer sequences utilized in the qRT-PCR assays are provided in Table S2.

Statistical analysis

All quantitative data are presented as mean ± SEM. Comparisons between two groups were performed using two-tailed unpaired Student’s t-tests, while differences among multiple groups were examined by one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test for multiple comparisons. ROC curves and AUC values were calculated to assess diagnostic performance. Calibration was evaluated by Hosmer–Lemeshow goodness-of-fit test, and DCA determined clinical net benefit. All tests were two-sided, with P < 0.05 considered statistically significant. Statistical analyses were conducted using GraphPad Prism 9.0 and R 4.2.0.

Results

ROS pathway activity is consistently upregulated in MI samples

To examine the role of ROS in MI, we first assessed gene expression consistency across the validation datasets. Principal component analysis (PCA) of the validation cohort demonstrated marked inter-dataset variation prior to normalization, with samples clustering predominantly by source dataset rather than biological condition (Fig. 1A). Application of the ComBat algorithm effectively reduced these batch effects, resulting in improved sample clustering according to disease status (Fig. 1B), confirming successful data harmonization. Using ssGSEA, we quantified the activity of five curated ROS-related pathways in each sample. Heatmaps revealed distinct expression profiles between MI and control groups across cohorts (Fig. 1C), and boxplots showed that four out of five pathways (Cellular_response_to ROS, ROS biosynthetic process, ROS_metabolic process, Response_to ROS) were significantly upregulated in MI samples compared to controls (Fig. 1D-E). These findings highlight consistent activation of oxidative stress–related molecular programs in the infarcted myocardium.

Fig. 1.

Batch correction and enrichment of ROS-related pathways in MI (A) PCA of the validation cohort before batch correction showing pronounced inter-dataset variation and clustering by dataset origin. (B) PCA after batch correction using the ComBat algorithm, indicating effective removal of batch effects. (C) Heatmap of single-sample gene set enrichment analysis (ssGSEA) scores for five curated ROS-related pathways across training and validation cohorts. (D-E) Boxplots comparing ROS pathway activity between MI and control samples, showing significant upregulation of four pathways in MI tissues. **p < 0.01, ***p < 0.001

Differentially expressed genes indicate immune activation and translational suppression in MI

To identify robust MI-related molecular signatures, differential expression analysis was performed in both training and validation cohorts. Volcano plots illustrated widespread transcriptomic alterations, with numerous genes significantly upregulated or downregulated in MI samples (Fig. 2A-B). A total of 176 upregulated and 98 downregulated genes were found to be consistently shared across cohorts (Fig. 2C-D). Gene Ontology (GO) enrichment analysis demonstrated a clear dichotomy in the functional roles of differentially expressed genes. The upregulated genes were predominantly associated with immune activation processes, showing significant enrichment (FDR < 0.05) in terms including leukocyte chemotaxis, pattern recognition receptor signaling pathway, and cytokine receptor activity. Conversely, the downregulated genes were fundamentally involved in translational machinery and protein synthesis, being strongly enriched in ribosome assembly, translation initiation, and structural constituent of ribosome (Fig. 2E-F). KEGG pathway analysis further corroborated this distinct functional partitioning and provided higher-level pathway context. The upregulated genes were significantly enriched in critical inflammatory and immune signaling pathways, most notably the NF-kappa B signaling pathway, Toll-like receptor signaling pathway, and NOD-like receptor signaling pathway, along with complement and coagulation cascades. In stark contrast, the downregulated genes were primarily mapped to metabolic and biosynthetic pathways, including ribosome, oxidative phosphorylation, and aminoacyl-tRNA biosynthesis (Fig. 2G-H). These comprehensive enrichment results collectively indicate a profound transcriptomic shift in MI, characterized by robust activation of innate immune responses coupled with a concurrent suppression of fundamental cellular metabolic and translational processes., reflecting a dual signature of immune activation and biosynthetic suppression in MI pathogenesis.

Fig. 2.

Differentially expressed genes (DEGs) and functional enrichment in MI. (A-B) Volcano plots displaying DEGs in training (A) and validation (B) cohorts. Red and green dots indicate significantly upregulated and downregulated genes, respectively. (C-D) Venn diagrams showing the overlap of upregulated (C) and downregulated (D) genes shared between training and validation cohorts. (E-F) Gene Ontology (GO) enrichment analysis of intersecting DEGs. Upregulated genes (E) are enriched in immune-related processes, while downregulated genes (F) are involved in translation and ribosomal pathways. (G-H) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment showing top enriched pathways for upregulated (G) and downregulated (H) genes

A ROS-Related gene signature derived from elastic net regression effectively distinguishes MI

Based on the intersection of differentially expressed genes and ROS-related gene sets, candidate genes were selected for diagnostic modeling. Correlation analysis demonstrated that these genes displayed coordinated expression patterns across both training and validation datasets (Fig. 3A). An elastic net logistic regression model was constructed using ten-fold cross-validation, which yielded an optimal λ value and retained six genes with non-zero coefficients: MMP9, ADAM9, BST1, TLR4, CLEC7A, and CYP1B1 (Fig. 3B-C). The resulting model was applied to both cohorts, and the distribution of predicted probabilities showed clear separation between MI and control samples, suggesting that this ROS-related gene signature provides effective diagnostic discrimination (Fig. 3D).

Fig. 3.

Construction of a ROS-related diagnostic gene signature for MI using elastic net regression. (A) Correlation matrix showing pairwise gene expression relationships among candidate ROS-related DEGs in training and validation cohorts. (B) Ten-fold cross-validation curve identifying the optimal regularization parameter (lambda) for elastic net regression. (C) Coefficient plot showing the contribution of six retained genes (MMP9, ADAM9, BST1, TLR4, CLEC7A, CYP1B1) to the model. (D) Predicted MI probabilities for each sample in the training and validation cohorts based on the constructed risk score, demonstrating clear separation between MI and control groups

The ROS-Based risk score shows strong diagnostic accuracy and clinical utility

To evaluate the diagnostic performance of the ROS-derived risk score, receiver operating characteristic (ROC) curves were generated for both the training and validation datasets. The composite score, integrating the expression profiles of the six-gene signature, demonstrated significantly superior diagnostic performance compared to any individual gene component. In the training cohort, the score achieved an area under the curve (AUC) of 0.907, which was significantly higher than the average AUC of individual genes. This robust performance was consistently maintained in the independent external validation cohort, where the composite score yielded an AUC of 0.755, confirming its generalizability across different patient populations and experimental platforms (Fig. 4A-B). The maintained discriminatory power in the validation set, despite expected performance attenuation common in external validation, underscores the signature’s robustness and potential clinical utility as a stable biomarker panel. Calibration plots confirmed good agreement between predicted probabilities and observed outcomes, indicating reliable model fit (Fig. 4C). Additionally, decision curve analysis showed that the risk score provided greater net clinical benefit across a wide range of decision thresholds relative to single-gene models (Fig. 4D), highlighting its potential applicability in clinical risk stratification for MI.

Fig. 4.

Diagnostic performance and clinical utility of the ROS-based gene signature. (A-B) Receiver operating characteristic (ROC) curves for individual model genes and the composite risk score in the training (A) and validation (B) cohorts, with area under the curve (AUC) values reported. (C) Calibration plots comparing predicted versus observed probabilities for the ROS-based risk model in both cohorts, indicating good model fit. (D) Decision curve analysis (DCA) showing the net clinical benefit of the composite risk score across a range of threshold probabilities, outperforming single-gene models in both cohorts

SHAP analysis reveals Gene-Level contributions to the MI risk score

To gain insight into how individual genes influenced the output of the MI diagnostic model, we performed SHAP-based model interpretation. The feature importance plot showed that MMP9 contributed the most to the risk prediction, followed by ADAM9, BST1, CLEC7A, TLR4, and CYP1B1 (Fig. 5A). The SHAP summary plot illustrated the distribution of SHAP values across all samples, where higher expression of MMP9, ADAM9, and BST1 corresponded to greater positive contributions to MI prediction, while lower expression values tended to reduce the predicted risk (Fig. 5B). SHAP dependence analysis provided a detailed characterization of the relationship between gene expression levels and their contribution to the model’s prediction. The dependence plots revealed predominantly monotonically increasing relationships for most signature genes, where higher expression levels corresponded to stronger positive SHAP values, indicating their directionally consistent contribution to MI classification probability (Fig. 5C). Notably, MMP9 and BST1 demonstrated distinct nonlinear threshold effects, characterized by substantially increased positive SHAP values only after exceeding specific expression thresholds. This threshold behavior suggests that the diagnostic influence of these genes becomes particularly pronounced only when their expression reaches biologically relevant levels, potentially reflecting activation thresholds in the underlying pathological processes. These findings provide mechanistic insight into how the model integrates ROS-related gene expression to infer MI risk, supporting the biological plausibility and interpretability of the predictive signature.

Fig. 5.

SHAP-based interpretation of the ROS-related diagnostic model for MI. (A) SHAP feature importance plot ranking the six genes by their mean absolute SHAP values, indicating their relative contribution to MI prediction. (B) SHAP summary plot showing the distribution of SHAP values for each gene across all samples. Color represents gene expression level (Z-score), and position on the x-axis reflects the gene’s effect (positive or negative) on the MI risk score. (C) SHAP dependence plots illustrating the relationship between gene expression and SHAP value for each of the six genes in the model. Each dot represents a sample; color scale reflects expression value. The plots demonstrate gene-specific response patterns, highlighting key contributors such as MMP9 and BST1 with nonlinear or threshold effects

Integrated regulatory and pathway analysis of key ROS-Related genes

To elucidate the regulatory landscape of the ROS-based diagnostic genes, we constructed TF and miRNA interaction networks. The TF-gene network revealed that key genes such as MMP9, TLR4, CYP1B1, and BST1 are potentially regulated by inflammation-related transcription factors including NFKB1, STAT1, SP1, and RELA (Fig. 6A). In parallel, miRNA–mRNA network analysis identified several miRNAs, such as hsa-miR-21-5p, miR-146a-5p, and miR-124-3p, as post-transcriptional regulators of multiple model genes including ADAM9 and TLR4 (Fig. 6B). To further explore their functional relevance, single-gene GSEA was performed for all six diagnostic genes. Results showed consistent enrichment of immune and inflammatory pathways-such as Toll-like receptor signaling, NOD-like receptor signaling, and hematopoietic cell lineage-associated with high expression of MMP9, CLEC7A, TLR4, and BST1 (Fig. 6C), highlighting their potential roles in innate immune activation and MI pathogenesis.

Fig. 6.

Upstream regulatory interactions and functional pathways associated with the ROS-based diagnostic genes in MI. (A) Transcription factor (TF)-gene interaction network showing predicted upstream TFs regulating MMP9, TLR4, CYP1B1, and BST1. (B) miRNA-mRNA regulatory network illustrating predicted miRNAs targeting the model genes, based on training databases. (C) Single-gene GSEA plots for each model gene. Key inflammatory and immune signaling pathways were found to be significantly enriched in high-expression groups, including TOLL-LIKE RECEPTOR SIGNALING, NOD-LIKE RECEPTOR SIGNALING, and HEMATOPOIETIC CELL LINEAGE

ROS-Related genes are linked to immune microenvironment remodeling in MI

To explore the immunological relevance of the diagnostic gene signature, we analyzed the immune cell composition of MI and normal samples using four independent deconvolution frameworks. Heatmaps revealed significantly elevated infiltration of multiple immune and stromal cell types in MI tissues, including regulatory T cells, M1 and M2 macrophages, dendritic cells, neutrophils, and fibroblasts (Fig. 7A). These findings were consistently observed across MCPcounter, quanTIseq, xCell, and EPIC, supporting the presence of a highly remodeled immune microenvironment in infarcted myocardium. In addition, we profiled the expression of 30 immune and inflammatory genes, including 15 inhibitory and 15 stimulatory markers. Several inhibitory checkpoint genes-such as IL12A, ICAM1, IL1B, and TNF-and stimulatory checkpoints including ARG1, IL10, LILRB1, and LILRB2 were significantly upregulated in MI samples (Fig. 7B), suggesting concurrent immune activation and suppressive signaling. Furthermore, correlation analysis demonstrated that the expression of CLEC7A, MMP9, and CYP1B1 was positively associated with endothelial cell infiltration, while TLR4 and ADAM9 showed weaker or nonsignificant relationships (Fig. 7C). These results highlight a potential link between the ROS-associated diagnostic genes and immune cell dynamics, particularly in regulating vascular and stromal components within the MI microenvironment.

Fig. 7.

Immune infiltration and immune and inflammatory genes landscape associated with ROS-related diagnostic genes in myocardial infarction. (A) Heatmaps of immune and stromal cell infiltration estimated by MCPcounter, quanTIseq, xCell, and EPIC algorithms. (B) Expression profiling of 30 immune and inflammatory genes (15 inhibitory and 15 stimulatory) between MI and control groups. (C) Correlation analysis between ROS-related model gene expression and endothelial cell infiltration as estimated by EPIC and xCell. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

Functional validation of ADAM9 in human cardiomyocytes

To investigate the functional role of ADAM9 in myocardial infarction pathology, we performed gain-of-function and loss-of-function studies in human AC16 cardiomyocytes. The efficiency of ADAM9 knockdown and overexpression was first confirmed by qRT-PCR (Fig. 8A). Modulation of ADAM9 expression significantly altered cardiomyocyte viability and survival. CCK-8 assays demonstrated that ADAM9 overexpression substantially reduced cell viability, whereas ADAM9 knockdown enhanced it compared to controls (Fig. 8B). This finding was further supported by EdU incorporation assays, which revealed that ADAM9 overexpression inhibited cardiomyocyte proliferation, while its knockdown promoted proliferative activity (Fig. 8C). We next examined the impact of ADAM9 on cell death pathways. Flow cytometric analysis of apoptosis showed that ADAM9 overexpression increased the percentage of apoptotic cells, whereas ADAM9 knockdown exerted a protective effect against apoptosis (Fig. 8D). Consistent with this observation, TUNEL staining confirmed that ADAM9 overexpression elevated cardiomyocyte mortality, while its reduction had the opposite effect (Fig. 8E). Given the central role of oxidative stress in MI pathogenesis, we measured intracellular ROS levels using DCFH-DA staining. Notably, ADAM9 overexpression significantly increased ROS production, while ADAM9 knockdown effectively attenuated oxidative stress in cardiomyocytes (Fig. 8F). Collectively, these results demonstrate that ADAM9 functions as a critical regulator of cardiomyocyte survival and oxidative stress, where its overexpression exacerbates cellular injury while its suppression confers protection, suggesting its potential as a therapeutic target in myocardial infarction.

Fig. 8.

Functional Validation of ADAM9 in Human Cardiomyocytes. (A) ADAM9 expression in AC16 cells transduced with three distinct sh-ADAM9 constructs or overexpressing plasmid. (B) CCK-8 assay assessing the proliferation of AC16 cells transduced with three distinct sh-ADAM9 constructs or overexpressing plasmid. (C) EdU assay assessing the proliferation of AC16 cells transduced with three distinct sh-ADAM9 constructs or overexpressing plasmid (magnification: ×200, scale bar = 100 μm). (D) Apoptosis analysis by flow cytometry in AC16 cells transduced with three distinct sh-ADAM9 constructs or overexpressing plasmid. (E) TUNEL staining assessing the apoptosis of AC16 cells transduced with three distinct sh-ADAM9 constructs or overexpressing plasmid (magnification: ×200, scale bar = 100 μm). (F) The DCFH-DA fluorescent probe staining used for detecting ROS of AC16 cells transduced with three distinct sh-ADAM9 constructs or overexpressing plasmid (magnification: ×200, scale bar = 100 μm). **p < 0.01, ***p < 0.001, ****p < 0.0001

ROS-Related Six‐Gene signature is significantly upregulated in a mice MI model

In the mice MI model, histological and molecular analyses consistently demonstrated upregulation of all six signature genes (Fig. 9). HE staining revealed that, unlike Sham hearts which displayed intact myocardial architecture, MI hearts exhibited widespread cardiomyocyte disarray, large necrotic areas, inflammatory cell infiltration, and interstitial edema, confirming successful infarction (Fig. 9A). Immunofluorescence analysis showed markedly increased protein expression of MMP9, ADAM9, BST1, TLR4, and CYP1B1 in MI (Fig. 9B-F). Immunohistochemistry for CLEC7A further corroborated a robust increase in protein levels in MI versus Sham tissue (Fig. 9G). qRT-PCR confirmed that mRNA levels of all six genes were significantly elevated in MI hearts compared to Sham controls (Fig. 9H). Co-localization immunohistochemistry of ADAM9 in cardiomyocytes showed that ADAM9 protein expression is significantly upregulated in the myocardium post-MI (Fig. 9I). Also, ROS levels were increased in the infarcted heart (Fig. 9J). Together, these data validate the bioinformatically derived ROS-related signature in vivo and underscore its relevance to myocardial ischemic injury.

Fig. 9.

Experimental validation of the ROSrelated sixgene signature in a mice MI model. (A) HE staining: Sham hearts showed normal architecture; MI hearts displayed necrosis, inflammatory infiltration, and edema. (B-F) Immunofluorescence: MMP9, ADAM9, BST1, TLR4, and CYP1B1 were markedly upregulated in MI border zones (magnification: ×200, scale bar = 100 μm). (G) IHC: CLEC7A expression was increased in MI hearts. (H) qPCR: All six genes showed significant mRNA upregulation in MI versus Sham. (I) Immunofluorescence colocalization for ADAM9 and α-actinin (magnification: ×200, scale bar = 100 μm). (J) The DCFH-DA fluorescent probe staining used for detecting ROS (magnification: ×200, scale bar = 100 μm). **p < 0.01, ***p < 0.001, ****p < 0.0001

Discussion

In this study, we integrated multiple transcriptomic datasets and identified a robust six-gene ROS-related signature (MMP9, ADAM9, BST1, TLR4, CLEC7A, CYP1B1) with high diagnostic accuracy for MI. This model not only demonstrated consistent performance across cohorts, but also revealed biologically meaningful associations with immune activation and vascular remodeling in the infarcted myocardium.

Consistent with previous studies emphasizing the role of oxidative stress in ischemia-reperfusion injury, our ssGSEA analysis confirmed broad upregulation of ROS-associated pathways in MI tissues [17]. In particular, MMP9 and TLR4, two top-ranking genes in our model, have been well documented as central mediators of inflammation and tissue degradation during cardiac injury [18–20]. The other components-such as CLEC7A and CYP1B1-though less studied in MI, have been implicated in innate immune sensing and redox metabolism, indicating novel mechanistic angles.

Our functional enrichment results revealed a dual pattern of immune activation and translational suppression in MI. Upregulated DEGs were enriched in leukocyte chemotaxis, cytokine signaling, and Toll-like receptor pathways, aligning with the classical understanding of post-MI inflammation [21, 22]. Meanwhile, downregulation of ribosomal and biosynthetic pathways reflects an energy-conserving stress response in injured cardiomyocytes.

Importantly, this study adds an immunogenomic dimension to ROS profiling. Using multiple deconvolution algorithms, we observed elevated infiltration of endothelial cell, dendritic cells, and fibroblasts in MI samples-cell types known to modulate inflammation and fibrosis. The expression of several diagnostic genes correlated positively with endothelial cell infiltration, suggesting links to vascular injury or remodeling. Additionally, the upregulation of inhibitory and stimulatory immune and inflammatory genes highlights a state of immune imbalance that may represent both activation and exhaustion. Our in vivo experiments further validate that this six-gene ROS signature is markedly upregulated at both mRNA and protein levels in infarct border zones enriched with inflammatory cells, reinforcing its dual utility as a diagnostic marker and mechanistic probe of oxidative-immune crosstalk in MI.

Compared to prior diagnostic models that often lacked biological interpretability or external validation [23, 24], our model was derived from intersecting ROS-related DEGs and interpreted through SHAP analysis, enhancing both transparency and robustness. SHAP plots revealed that MMP9 and ADAM9 were the most influential genes in the model, aligning with their established roles in extracellular matrix remodeling and inflammation [3, 4]. The in vitro experiment suggested that ADAM9 inhibited cardiomyocyte survival and enhanced oxidative stress. Similar to our results, Ao et al. indicated that the overexpression of ADAM9 can inhibit the improvement effect of sevoflurane treatment on myocardial injury and myocardial autophagy in mice with ischemia-reperfusion [25].

The six-gene signature primarily captures the systemic immune and inflammatory response following MI, reflecting changes in the abundance and activation status of myeloid cells in the peripheral blood. While this signature does not directly measure cardiomyocyte-intrinsic ROS levels, it provides a clinically accessible measure of the ROS-related systemic inflammatory milieu associated with MI. The clinical value of our blood-based signature lies in its complementary role to established biomarkers like troponin. Whereas troponin reliably indicates cardiomyocyte necrosis, our signature provides mechanistic insight into the concomitant oxidative stress and inflammatory state, potentially enabling improved risk stratification for adverse post-MI outcomes [26]. Furthermore, our findings suggest potential synergistic interactions among signature genes: TLR4 and S100A8 may initiate inflammatory signaling, MMP9 drives extracellular matrix remodeling, and ADAM9 directly amplifies cardiomyocyte ROS production, collectively forming a vicious cycle of myocardial injury. This integrated model moves beyond individual gene associations and proposes a coordinated pathological mechanism, positioning our signature not as a replacement for existing diagnostics but as a tool for patient stratification and prognostic assessment that aligns with the non-invasive nature of blood-based testing.

However, several limitations must be acknowledged. First, all analyses were based on publicly available bulk transcriptomic datasets, which lack spatial resolution and cell-type specificity. Second, while the model was validated across multiple cohorts, prospective clinical validation is needed. Third, functional experiments are necessary to confirm the mechanistic roles of the model genes in ROS and immune regulation during MI. Finally, there are limitations in the spatial resolution of cardiac tissue because our method was designed for global assessment rather than for regional distinction. In the future, technologies such as laser capture microdissection combined with high-resolution spatial transcriptomics will be highly suitable for depicting the precise spatial distribution of these gene expression patterns in the remote, border, and infarcted regions.

In conclusion, our study provides a validated, interpretable six-gene diagnostic model grounded in ROS biology and immune remodeling. It offers not only diagnostic potential but also mechanistic insights into oxidative stress-driven pathways in MI, and may inform future precision therapeutic strategies.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ROS

Reactive oxygen species

MI

Myocardial infarction

ssGSEA

Single-sample gene set enrichment analysis

DEG

Differentially expressed gene

MMP9

Matrix metalloproteinase-9

GEO

Gene Expression Omnibus

MSigDB

Molecular Signatures Database

BP

Biological process

CC

Cellular component

MF

Molecular function

ROC

Receiver operating characteristic

AUC

Area under the curve

PR

Precision-recall

DCA

Decision curve analysis

SHAP

SHapley Additive exPlanations

TF

Transcription factor

PCA

Principal component analysis

cDNA

Complementary DNA

qRT-PCR

Quantitative reverse transcription polymerase chain reaction

HE

Hematoxylin and eosin

ANOVA

Analysis of variance

Author contributions

Guarantor of integrity of the entire study: Yanting Wang. Study concepts: Yanting Wang. Study design: Yuan Zhang, Yanting Wang. Literature research: Yuan Zhang, Tianyu Liang.Experimental studies: Yuan Zhang, Tianyu Liang.Data acquisition: Yuan Zhang, Tianyu Liang. Data analysis: Yuan Zhang, Tianyu Liang.Statistical analysis: Yuan Zhang, Yanting Wang. Manuscript preparation: Yuan Zhang, Yanting Wang.Manuscript review: Yuan Zhang, Tianyu Liang, Yanting Wang.

Funding

This work was supported by the Zhejiang Provincial Medical and Health Science and Technology Project (Grant No. 2022KY606), and Zhejiang Provincial Medical and Health Science and Technology Research Project (Grant No. 2023KY538).

Data availability

Data is provided within the manuscript or supplementary information files.

Declarations

Competing interests

The authors declare no competing interests.

Consent for publication

Not applicable.

Ethics approval and consent to participate

All experiments were approved by the Institutional Animal Care and Use Committee of the Hubei Provincial Center for Disease Control and Prevention.