The active ingredients of four herbs in DYX were obtained from Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP; developed by College of Life Science, Northwest A&F University, available at https://tcmspw.com/tcmsp.php) and Traditional Chinese Medicine Integrated Database (TCMID; developed by Shanghai Key Laboratory of Regulatory Biology, accessible at https://megabionet.org/tcmid/) using oral bioavailability (OB) ≥30% and drug-likeness (DL) ≥ 0.18 as screening criteria (37, 38). The targets of active ingredients in DYX were obtained from the TCMSP database. The protein sequences of the targets were converted into official gene symbols by the UniProt database (a comprehensive resource maintained by the European Bioinformatics Institute, the Swiss Institute of Bioinformatics, and the Protein Information Resource, available at https://www.uniprot.org/). The STRING (developed by the European Molecular Biology Laboratory, accessible at https://www.string-db.org/) database was applied to analyze the interaction between active ingredients and targets. The Cytoscape software was used to construct the “active ingredient–target” network. The protein interactions were searched by the SRTING database. The targets of DYX were entered into the database, and the scoring condition was set to >0.40 to obtain the protein–protein interaction (PPI) network of DYX. The network was visualized and topologically analyzed in Cytoscape, and nodes with a degree higher than the median were filtered multiple times to obtain core targets.

Datasets were derived from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) with the following keywords: (Myocardial Infarction) AND “Homo sapiens” AND “Expression profiling by array” AND “Series”. The screening standards included the following: the microarray datasets referred to profiles of gene expressions of whole blood; the microarray datasets contained samples from MI and samples from a healthy state; all included samples were not treated with drugs. Eventually, datasets GSE66360 (39) and GSE60993 (40) were selected for the next analysis. The clinical information for the samples of MI patients and healthy controls was provided in Supplementary Table S1.

The LIMMA package in R software was utilized to identify differentially expressed genes (DEGs) through the comparison of the expression dataset in GSE66360. The volcano plot and heatmap were drawn to present the expression of DEGs. DEGs with adjusted p < 0.05 and |log₂FC| > 2 were considered statistically significant.

For the exploration of the function and pathway of the DEGs, our study conducted the Metascape (http://metascape.org/gp/index.html) analysis, choosing species as “H. sapiens.” Meanwhile, gene ontology (GO) enrichment analysis was performed with the use of the clusterProfiler package, and adjusted p < 0.05 was considered statistically significant. Disease ontology (DO) analysis in the clusterProfiler package was also employed to analyze the relationship between disease and DEGs. To more intuitively clarify the gene expression level of significantly enriched functional pathways, gene set enrichment analysis (GSEA) was performed in R software.

After excluding batch effects, median absolute deviation (MAD) was calculated for gene expression, and the top 10% of genes with the smallest MAD values were removed. Outlier genes and samples were culled using the WGCNA package in R software, and a scale-free co-expression network was constructed by soft thresholding. The optimal soft threshold was chosen to cluster this network into different functional modules, each containing a minimum of 30 genes. Modules were hierarchically clustered by calculating the eigengene, and similar modules were merged. Pearson’s correlation analysis was used to determine the correlation between modules and clinical features by combining gene significance (GS) and module membership (MM). Clinical modules with high correlation with MI samples and normal samples were selected, and the core genes were finally screened out from the target modules.

In this study, a Venn diagram was used to obtain the intersection of DEGs, core targets in the PPI network, and core genes in the WGCNA for further analysis.

For the in-depth test of the efficacy of key biomarkers, the GSE60993 dataset served as the validation set. It was assessed based on the receiver operating characteristic (ROC) with the pROC package, and the area under the curve (AUC) was calculated to evaluate the predictive effect achieved by the model. A two-sided p < 0.05 showed statistical significance. Meanwhile, the different expression level of candidate genes between healthy people and MI patients was presented with the ggpubr package, p < 0.05 showed statistical significance.

To predict the binding of candidate gene proteins to small molecules of components of DYX, we performed molecular docking after optimizing the molecular structure and the free energy of small ligand molecules and acceptor proteins. First, the PDB files of the target protein were downloaded from the PDB database (http://www.rcsb.org/). The proteins were saved in PDBQT format as the docking acceptor after dehydration and hydrogenation. Next, the small-molecule 2D structures were downloaded from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/). The minimum free energy was calculated, and the structure was optimized and saved as a mol2 format file. Finally, molecular docking was performed using AutoDock Vina to estimate the binding activity of small molecules to proteins mainly by the binding free energy. The lower the binding energy value, the greater the likelihood that the ligand will bind to the receptor.

The CIBERSORT algorithm was adopted to identify 22 immune cells in the samples. CIBERSORT p < 0.05 showed statistical significance. The reshape2, ggpubr, and ggExtra packages were applied to demonstrate the different distributions of immune cells in healthy controls and patients with MI. Based on Spearman correlation analysis, correlations were calculated for 22 immune cells. The results of the correlations were visualized by a heatmap using the corrplot package. The difference between groups of 22 immune cells was analyzed based on the t-test, and adjusted p < 0.05 was statistically different. The results were visualized using the ggpubr package.

The correlation of hub genes with 22 immune cells was determined using the psych package. Results were visualized using the ggpubr and reshape2 packages.

A single-cell RNA sequencing (scRNA-seq) dataset [GSE184073 (41)] of immune cells in coronary plaques was obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). Two coronary plaque samples were included in the dataset, one from a patient with acute coronary syndrome (ACS) and the other from a patient with stable angina pectoris (SAP).

The “Seurat” (42) package in R software was applied to read the dataset. Low-quality cells were excluded using the following criteria: (1) features < 200 and >5,000 and (2) mitochondrial genes >25%. First, the expression data of the screened cells were normalized. Then, 2,000 highly variable genes were obtained for principal component analysis (PCA). After integrating the two samples using the “harmony” (43) package to remove batch effects, the first 30 principal components in PCA were used for subsequent cell clustering. Finally, dimensionality reduction was performed using the uniform manifold approximation and projection (UMAP) algorithm.

The “FindAllMarkers” function was employed to identify DEGs between cell clusters as markers of cells. CellMarker (http://bio-bigdata.hrbmu.edu.cn/CellMarker) (44) is a database that collects cell markers identified in previous articles. Cell clusters were manually annotated by reference to CellMarker.

The “AddModuleScore” function was utilized to assess the expression of FOS in different cell clusters. Violin plots and feature plots were used to demonstrate FOS expression in different samples or different cell clusters.

FOS expression was not consistent across T cells. All T cells were categorized into a high FOS group and a low FOS group. The “FindMakers” function was used to identify DEGs between the two groups (adjusted p < 0.05 and |log₂FC| > 0.5). To further explore the differences in the involvement of T cells in biological processes, Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis was performed on DEGs.

The TCMSP database and TCMID were searched for the active ingredients of DYX, and a total of 31 active ingredients were obtained after screening. A total of 198 non-duplicated targets were obtained by searching using these active ingredients individually. Figure 2 illustrates the “active ingredient–target” network of DYX. The outermost dots are the active ingredients of DYX, and the inner red dots are the targets. After removing the 26 non-interacting targets, the remaining 172 targets constituted the PPI network (Figure 3A). There are 172 nodes and 1,042 edges in the network. The core targets were then screened multiple times using the median of the degree. Firstly, targets with a degree value of >8 were screened, leaving 82 nodes and 717 edges (Figure 3B); next, targets with a degree value of >15 were screened, with 38 nodes and 313 edges remaining (Figure 3C); and finally, targets with a degree value of >17 were screened, resulting in a network consisting of 15 nodes and 90 edges (Figure 3D). A total of 15 selected core potential therapeutic targets were obtained: FOS, STAT1, EGFR, RELA, MYC, HIF1A, AR, EGF, CCND1, CASP3, ESR1, MAPK1, AKT1, JUN, and TP53.

A total of 43 DEGs were identified between healthy people and patients with MI in GSE66360, and the heatmap shows the expression of 43 DEGs between the two groups (Figure 4A). As shown in Figure 4B, red dots indicated upregulated genes (41 genes), green dots indicated downregulated genes (2 genes), and black dots indicated no statistical significance. The expression of most DEGs was downregulated in healthy individuals and upregulated in patients with MI.

GO enrichment analysis (Figure 4C) showed that DEGs were associated with migration of neutrophils and myeloid leukocytes, suggesting immune cell infiltration after MI. Tertiary granules, a structure of leukocytes, were also shown to be associated with DEGs. The molecular function section showed that these DEGs were also involved in immune receptor activity and pattern recognition receptor activity. DO analysis indicated that DEGs were associated with arteriosclerosis and bacterial infectious diseases (Figure 4D). This may be associated with the involvement of these genes in plaque formation and the promotion of inflammation.

GSEA revealed distinct pathway enrichments in healthy individuals and MI patients. In controls, downregulated DEGs were linked to “primary immunodeficiency,” “purine metabolism,” “homologous recombination,” “nucleotide excision repair,” and “peroxisomal enzymes” (Figure 4E). These pathways are crucial for immune regulation and cellular homeostasis, and their inhibition suggests a stable immune state in healthy individuals. In MI patients, upregulated DEGs were associated with “complement and coagulation cascades,” “cytokine–cytokine receptor interactions,” “leishmania infection,” “MAPK signaling pathway,” and “Toll-like receptor signaling pathway” (Figure 4F). The “complement and coagulation cascades” pathway indicates an immune response to tissue damage, while “cytokine–cytokine receptor interactions” suggests amplified inflammatory signaling. The “MAPK” and “Toll-like receptor signaling pathways” are vital for initiating immune and inflammatory responses. These findings imply that post-MI, immune cells, particularly leukocytes, infiltrate the affected area as part of the inflammatory response to myocardial damage. This analysis underscores the immune system's significant involvement in MI's pathophysiology.

In this study, we set the soft threshold to 13 (R² = 0.98) (Figure 5A). Then, 18 modules were obtained based on average hierarchical clustering and dynamic tree shearing (Figure 5B). Among them, the green module was highly correlated with MI and was therefore selected for further analysis (Figure 5C). Correlation analysis between MM in the green module and gene significance for MI also showed that the green module containing 828 genes was relevant to MI (Figure 5D, r = 0.68, p = 9.2 × 10⁻¹¹³).

A Venn diagram was employed to show the intersection of the results of DEGs, PPI network, and WGCNA (Figure 6A). FOS was identified as the sole hub marker (Supplementary Table S2). The ROC curves showed an AUC of 0.833 for FOS in the dataset GSE66360 (Figure 6B) and an AUC of 0.845 for FOS in the validation dataset GSE60993 (Figure 6C), which indicated good efficacy of FOS in the diagnosis of MI. Moreover, the trend of FOS expression was consistent in both datasets (Figures 6D,E). In both, the expression level of FOS in MI patients was higher than that in the control group, and the differences between the groups were statistically significant (p < 0.001 in GSE66360, p = 0.01 in GSE60993).

As shown in Figure 7, quercetin demonstrated the strongest binding affinity (ΔG = −10.4 kcal/mol), forming hydrogen bonds with DA-9, DG-10, and DA-31 (Figure 7A). This triple-point interaction suggests direct competition with the FOS/JUN dimer for DNA minor groove occupancy. Baicalein exhibited distinct binding at the protein–DNA interface (ΔG = −9.6 kcal/mol), engaging Arg279 of JUN and Arg155 of FOS (Figure 7B). Notably, its flavone scaffold simultaneously intercalates with DA-9, creating a “bridge” between protein and DNA domains (Figure 7B).

The CIBERSORT algorithm was used to analyze the immune cell infiltration. The proportion of 22 immune cells was shown in a bar plot (Figure 8A; Supplementary Table S3). The proportions of immune cells from the whole blood of healthy people and MI patients displayed distinct and group-bias clustering. Resting CD4⁺ memory T cells showed the largest difference in proportion between MI and the control group. Activated CD4⁺ memory T cells had the strongest positive correlation with resting dendritic cells (r = 0.76). Meanwhile, resting CD4⁺ memory T cells had the strongest negative correlation with resting mast cells (r = −0.65), as shown in Figure 8B. Thirteen types of immune cells were differentially expressed (Figure 8C). Compared with the control group, MI patients generally contained a higher proportion of follicular helper T cells (p < 0.001), regulatory T cells (Tregs) (p = 0.002), gamma delta T cells (p < 0.001), resting natural killer (NK) cells (p = 0.008), activated NK cells (p < 0.001), monocytes (p < 0.001), activated dendritic cells (p = 0.002), resting mast cells (p < 0.001), activated mast cells (p < 0.001), and neutrophils (p < 0.001), whereas the proportions of CD8⁺ T cells (p = 0.007), resting CD4⁺ memory T cells (p < 0.001), and M0 macrophages (p = 0.014) were relatively lower.

Based on the results of correlation analysis, FOS displayed a positive correlation with neutrophils (p < 0.001), monocytes (p < 0.001), activated mast cells (p < 0.001), activated NK cells (p < 0.001), dendritic cells (activated) (p = 0.010), resting NK cells (p = 0.014), and eosinophils (p = 0.016) and showed a negative correlation with M0 macrophages (p = 0.016), plasma calls (p = 0.010), resting mast cells (p = 0.007), gamma delta T cells (p = 0.004), and resting CD4⁺ memory T cells (p < 0.001) (Figures 9A,B). These correlations indicate that FOS may play a significant role in the immune response following MI, influencing the activity and abundance of various immune cell types. The positive correlation with proinflammatory cells such as neutrophils and monocytes suggests that FOS may contribute to their recruitment and activation during the early inflammatory phase post-MI. The negative correlation with certain anti-inflammatory or immune-regulatory cells, such as M0 macrophages and resting mast cells, implies that changes in FOS levels might be associated with the resolution of inflammation and tissue repair.

A total of 2,363 cells were obtained after screening. After dimensionality reduction and clustering, these cells were classified into 12 cell clusters (Figure 10A). Cell clusters were manually annotated to eight cell types, including T cells, dendritic cells, monocytes, macrophages, B cells, natural killer T (NKT) cells, mast cells, and progenitors (Figure 10B). Referring to CellMarker, IL7R, CD8A, and CD8B were used as markers for T cells; FCN1, MNDA, and VCAN as markers for dendritic cells; SGK1, CXCL3, and TIMP1 as markers for monocytes; C1QA, C1QC, and APOE as markers for macrophages; IGHM, CD79A, and MS4A1 as markers for B cells; KLRD1, GNLY, and GZMB as markers for NKT cells; CPA3, TPSB2, and IL1RL1 as markers for mast cells; TUBB1, GNG11, and GP1BB as markers for progenitors. The dot plot shows that the markers were highly expressed in the corresponding cell clusters (Figure 10C).

There was no statistical difference in the expression of FOS in stable and unstable plaque samples (Figure 10D). This suggests that FOS is not related to the stability of the plaques. However, given the relatively small sample size of dataset, this remains to be explored in more studies. Among the cell clusters, mast cells had the highest levels of FOS expression, followed by dendritic cells and monocytes (Figure 10E).

In addition, some T cells also showed high levels of FOS expression (Figure 10F). T cells were categorized into high FOS and low FOS groups based on the level of FOS expression. The circle heatmap shows 36 DEGs between groups (Figure 10G). These DEGs were mainly associated with inflammatory signaling pathways such as the MAPK signaling pathway, IL-17 signaling pathway, TNF signaling pathway, and NF-κB signaling pathway (Figure 10H). It suggested that there were subpopulations of T cells that highly expressed FOS and were involved in multiple inflammatory responses in coronary plaques.