For the AMI cohort, two independent datasets were analyzed on the GPL6244 platform, namely, GSE59867 (111 AMI patients and 46 stable CAD patients at admission) and GSE62646 (28 AMI patients and 14 stable CAD patients at admission) (Pan et al., 2023). The peripheral blood cohort of OC was obtained from GSE31682, which comprises 20 healthy controls and 48 OC patients. After excluding patients with incomplete follow-up information (Feng et al., 2022), we obtained RNA sequencing (RNA-seq) data from both the Cancer Genome Atlas (TCGA) database (Blum et al., 2018) and the International Cancer Genome Consortium (ICGC) database. Additionally, data from the Gene Expression Omnibus (GEO) database (Barrett et al., 2013) was obtained for the GPL570 platform (n = 597), which included GSE19829, GSE18520, GSE9891, GSE26193, GSE30161, and GSE63885. To integrate the ICGA and TCGA data and define the meta-RNA-seq dataset, the meta-microarray dataset was defined using the GPL570 platform. Finally, the “sva” package was utilized to effectively address and eliminate batch effects across the different datasets.

Considering the limited availability of human AMI single-cell RNA (scRNA) sequencing datasets, we employed a mouse single-cell sequencing dataset (GSE135310) as an alternative (Pan et al., 2023). This dataset includes single-cell RNA sequencing files for cardiac CD45⁺ total leukocytes, isolated from mice subjected to AMI or sham surgery at various time points. As we lacked peripheral blood single-cell data from healthy individuals within the same batch, we focused on analyzing datasets obtained before and after chemotherapy from the same batch. In this regard, we obtained the GSE213243 dataset, comprising 2 peripheral blood samples from OC patients. In summary, we performed a series of data filtering steps to ensure the quality of our scRNA data. We retained cells with an expression of RNA features ranging from 200 to 2500 while keeping the percentage of mitochondrial RNA content below 10%. Additionally, we employed the Harmony algorithm to mitigate batch effects in our analysis. To annotate all clusters, we utilized the “SingleR” package for comprehensive annotation.

As consistent with our previous publication (Pan et al., 2023), we performed mRNA validation of peripheral blood samples using the same cohort of patients. In brief, ten early AMI patients and ten CAD patients were recruited from our hospital between January 2023 and March 2023. Blood samples were collected from the patients shortly after experiencing chest pain, before the administration of antiplatelet or anticoagulant drugs. Peripheral blood mononuclear cells (PBMCs) were isolated from the collected blood samples using established techniques (Boyum, , 1968).

Immunohistochemistry (IHC) staining involves the use of antibodies that are specifically designed to recognize and bind to target antigens of interest. The antibodies are labeled with a chromogenic or fluorescent dye, enabling the visualization and localization of the target molecules under a microscope. The IHC sections utilized in this study were obtained from the Human Protein Atlas (HPA) database. To validate the mRNA expression levels of AMI, PBMCs from patients were utilized. As for OC, cell lines were employed for the validation of mRNA expression levels. IOSE-80 (CP-H055), and SKOV3 (CL-0215) were purchased from Procell Life Science and Technology Co. Ltd. They were cultured in RPMI 1640 medium supplemented with 10% FBS and 1% penicillin/streptomycin at 37°C in a humidified incubator with 5% CO₂. In short, total RNA was extracted from samples using the FastPure Cell/Tissue Total RNA Isolation Kit V2 (Vazyme, Nanjing, China). Subsequently, RT-qPCR was performed on a LightCycler 480 II Real-time PCR instrument with the HiScript III All-in-one RT SuperMix Perfect for qPCR (Vazyme, Nanjing, China) and ChamQ universal SYBR qPCR Master Mix (Vazyme, Nanjing, China). We used the 2^{^-∆∆Ct} method to calculate gene expression levels. The primer sequences used were designed based on previously published references and PrimerBank database (Wang et al., 2012), including WEE1 (Ma et al., 2022), PYHIN1 (Lee et al., 2019), SEC61A2, and HAL (Kozaczek et al., 2019).

In our study, we utilized the Weighted Gene Co-expression Network Analysis (WGCNA) (Langfelder and Horvath, 2008) in order to investigate gene expression patterns and identify gene modules with similar expression profiles. WGCNA is a powerful tool that allows us to construct a co-expression network based on correlations between genes across different samples. By analyzing the module and module-trait (monocyte score from CIBERSORT) relationships, we can uncover biologically relevant modules associated with specific traits or phenotypes of interest. We established the optimal soft thresholding powers (β) for OC and AMI samples to be β = 7 and β = 9, respectively. Additionally, we ensured that each module consisted of no fewer than 50 genes.

We used cross-validation (10-flod) to determine the optimal value for λ. In short, our objective was to determine the optimal model λ value by constructing a penalty function with the occurrence of AMI as the endpoint event and the variation in gene expression as the variable for each sample. The Least absolute shrinkage and selection operator (LASSO) regression model is fit on the training set for each λ value (Cheng et al., 2022), and the performance is evaluated on the validation set using a chosen metric, such as mean squared error or area under the curve. The λ value that minimizes the error on the validation set is considered the optimal choice. Hence, we conducted a more comprehensive screening of potential diagnostic genes from the pool of monocyte-associated genes.

We utilized a nomogram based on Logistic regression to facilitate predictive modeling and risk assessment (Zhang et al., 2022). By fitting the logistic regression equation, we estimated the coefficients of the four variables (WEE1, PYHIN1, SEC61A2, and HAL) and captured their contributions to the probability of the outcome. Each predictor variable was assigned a corresponding point value based on its coefficient. The total points were summed up to determine the individual’s predicted probability of the outcome. Moreover, we utilized calibration curves, decision curve analysis (DCA) curves, and clinical impact curves to validate the performance of the nomogram in GSE62646 and GSE59867.

The Cox regression model, also referred to as the proportional hazards model, is a widely employed statistical approach utilized in survival analysis to analyze time-to-event data. To construct the Cox regression model, we initially selected a group of potential predictor variables, namely, WEE1, PYHIN1, SEC61A2, and HAL. Subsequently, we performed model training by fitting the Cox regression equation to the meta-RNA-seq cohorts, thereby estimating the coefficients for each gene. The risk score was then calculated using the formula: risk score = Σ (Exp_i * coef_i), where coef and Exp represent the coefficient and expression of each gene, respectively. After establishing the Cox regression model, we proceeded to validate its performance utilizing meta-microarray cohorts. We employed calibration plots, receiver operator characteristic (ROC) curves, and log-rank tests to assess the model’s efficacy in distinguishing between high-risk and low-risk individuals.

We employed hallmark gene sets (h.all.v7.5.1.symbols.gmt) (Liberzon et al., 2015), which are collections of genes representing key biological processes and signaling pathways. These gene sets cover a wide range of fundamental cellular activities, such as cell cycle regulation, DNA repair, immune response, and metabolism. By applying Gene Set Variation Analysis (GSVA) (Hanzelmann et al., 2013), we transformed our gene expression data into pathway enrichment scores for each sample. This was achieved by comparing the expression levels of genes within each hallmark gene set to the background distribution in our dataset. The resulting enrichment scores provided quantitative measures of the activity levels of these pathways in individual samples.

Statistical analysis was performed using R software (v4.1.2). To evaluate differences, the significance of most cases was assessed using the Wilcoxon rank-sum test. Statistical significance was defined as a p-value below 0.05 and indicated as *p < 0.05, **p < 0.01, or ***p < 0.001.

Coronary artery disease (CAD) is a principal cause of morbidity and mortality worldwide. Patients with stable CAD are still at risk of AMI, which is a severe complication of CAD. Thus, in clinical settings, patients with CAD are commonly used as control groups to investigate changes in blood indicators among AMI patients. In our study, we first compared differences in cell proportions using the CIBERSORT algorithm on a peripheral blood microarray dataset consisting of AMI and CAD patients. AMI patients exhibit decreased levels of CD8⁺ T cells, memory CD4⁺ T cells, resting NK cells, and M2-type macrophages. Conversely, they have increased levels of Tregs, resting mast cells, neutrophils, and monocytes (Figure 1A). Subsequently, we further investigated the changes in cell content in the peripheral blood of OC patients compared to normal individuals. Interestingly, only the trend of monocytes cell changes was consistent with that of AMI patients. This suggests a potential common role of monocytes in both conditions (Figure 1B). To further validate the robustness of our results, we conducted dimensionality reduction analysis on single-cell data from both the sham and AMI patients, resulting in six distinct clusters that could be clearly distinguished: monocytes, endothelial cells, macrophages, granulocytes, NK cells, and B cells (Figure 1C). Of note, after AMI occurred, the proportion of monocytes increased, consistent with the results of bulk transcriptome analysis. Due to the lack of peripheral blood single-cell data from healthy individuals from the same batch, we analyzed the datasets obtained before and after chemotherapy from the same batch. Similarly, the single-cell data from patients were sorted into four clusters (Figure 1D), with a significant decrease in the proportion of monocytes observed after chemotherapy. Therefore, the proportion of monocytes in the peripheral blood of patients decreased when the tumor-load was reduced by chemotherapy. Conversely, monocytes proportion significantly increased when the tumor occurred (Figure 1E). Given that pathology is often considered the gold standard for cancer diagnosis, we re-assessed the prognostic value of monocytes (CD14 is a typical marker for monocytes) in the TCGA-OV cohort (cancer tissues). Our results also demonstrate that monocytes are a risk factor for OC prognosis. Specifically, as the expression of CD14 increases in bulk tissues, the prognosis of patients worsens (Supplementary Figure S1).

To enhance gene filtration from monocytes, we integrated the entire gene expression profile into WGCNA. Additionally, we utilized the score of monocyte expression in the CIBERSORT algorithm as a clinical feature to identify key modules most pertinent to monocyte expression. During the construction of the co-expression network, we observed optimal soft threshold powers of β = 7 for OC (Figure 2A) and β = 9 for AMI samples (Figure 2B). Through careful examination of correlation coefficients and p-values (Figures 2C, D), we determined that the purple and brown modules exhibited the strongest absolute correlation with the monocyte score in OC (Figure 2E), while in AMI, the purple and yellow modules displayed the strongest absolute correlation (Figure 2F). As a result, we designated these four modules as key modules and subsequently identified 23 overlapping genes within them (Figure 2G). Subsequently, we conducted further screening of the aforementioned 23 genes in the AMI dataset using the LASSO algorithm. Our objective was to determine the optimal model λ value by constructing a penalty function with the occurrence of AMI as the endpoint event and the variation in gene expression as the variable for each sample (Figures 2H, I). Consequently, we identified seven genes: HRH4, LTBR, WEE1, HAL, PYHIN1, S100A12, and SEC61A2. Of particular significance, based on these seven genes, we proceeded with prognostic validation utilizing the RNA-seq cohort of OC. Our findings indicated that WEE1, PYHIN1, and SEC61A2 served as risk factors impacting the prognosis of ovarian cancer, while HAL demonstrated a protective effect (Supplementary Figure S2).

In summary, our findings suggest that WEE1, PYHIN1, SEC61A2, and HAL derived from monocytes may serve as a potential predictor for AMI and cancer prognosis.

In the AMI dataset, we made noteworthy observations. Firstly, WEE1 (p = 2.4e-11) exhibited significant downregulation and displayed superior diagnostic potential, as evidenced by an AUC of 0.839 (95% CI: 0.768–0.899), as shown in Figure 3A. Similarly, PYHIN1 (Figure 3B) and SEC61A2 (Figure 3C) were markedly downregulated in AMI samples and demonstrated favorable diagnostic performance, with AUCs of 0.782 and 0.747, respectively. It is worth mentioning that HAL exhibited significant upregulation in AMI samples, yielding an AUC of 0.766 (Figure 3D). To further enhance the diagnostic accuracy of the model and facilitate its clinical application, we developed a nomogram utilizing the aforementioned four genes as predictors for AMI (Figure 3E). The calibration curve demonstrated the ability of the nomogram to accurately and reliably diagnose AMI among CAD patients (Figure 3F). Moreover, the DCA curve (Figure 3G) and decision curve analysis (Figure 3H) provided further evidence of its robust diagnostic capacity. Importantly, in the validation dataset (GSE62646), the calibration curve (Figure 3I), DCA curve (Figure 3J), and clinical impact curve (Figure 3K) consistently underscored the excellent external validation capability of the nomogram.

Considering that cancer treatment primarily involves surgery and tissue samples are readily available, we conducted a comprehensive investigation into the expression of the aforementioned four genes across various tissue samples by combining the GTEx database and the HPA database. At both the protein and mRNA levels, WEE1 exhibited significantly higher expression in tumor samples (Figure 4A). Similarly, PYHIN1 displayed significantly higher mRNA expression in tumor samples, although no significant protein staining (Figure 4B). In contrast, HAL did not exhibit significant differences in mRNA expression between the two sample types, but protein upregulation was evident in tumor samples (Figure 4C). Unfortunately, an IHC antibody for SEC61A2 was unavailable. However, we characterized the protein’s structure and confirmed its significant upregulation at the mRNA level in normal samples (Figure 4D). Subsequently, to quantify the survival risk for each ovarian cancer patient, we developed a risk model utilizing a multi-factorial Cox formula (Figure 4D) based on the four aforementioned genes (WEE1, PYHIN1, SEC61A2, and HAL). The risk score for each OC patient was calculated using the equation: Risk score = (−0.195 × WEE1 expression) + (−0.335 × PYHIN1 expression) + (−0.191 × SEC61A2 expression) + (0.137 × HAL expression). Subsequently, survival curves were generated for each gene in the model, revealing intriguing findings. PYHIN1, SEC61A2, and WEE1 emerged as protective genes, suggesting that reduced expression of these genes may contribute to prolonged overall survival in patients (Figure 4F). Conversely, HAL was identified as a high-risk gene, implying that increased expression of HAL may be associated with decreased patient survival. Notably, our expression level assessment demonstrated predominantly negative correlations among the genes (Figure 4G). Specifically, HAL displayed negative correlations with WEE1 and PYHIN1, while exhibiting a positive correlation with SEC61A2. Regarding risk scores, a noteworthy positive correlation was solely observed with HAL (R = 0.442, p < 0.001). The formula described above was applied to both the meta-RNA-seq cohort and the meta-microarray cohort, resulting in the generation of a new risk score based on the expression profile. Patients were then classified into high-risk and low-risk categories using the median risk score. Kaplan-Meier survival analysis conducted in the meta-RNA-seq cohort revealed that high-risk patients exhibited significantly worse overall survival compared to low-risk patients (Figure 5A). Similarly, in the meta-microarray cohort, the high-risk group had a lower likelihood of survival (Figure 5B). To assess the prognostic accuracy of the prognostic features, ROC curves for 1, 3, 5, and 10-year OS were analyzed. In the meta-RNA-seq training cohort, the AUC values for these time points were 0.728, 0.692, 0.673, and 0.708, respectively, (Figure 5C). Similarly, the meta-microarray validation set showed superior AUC values of 0.595, 0.578, 0.625, and 0.697 for the 1, 3, 5, and 10-year AUCs, respectively, (Figure 5D).

To ensure consistency with the AMI risk signature, we utilized a visual nomogram to evaluate the risk of OC patients by integrating FIGO staging and risk stratification (Figure 5E). The accuracy of nomogram was assessed and found to be superior in both the meta-RNA-seq cohort and the meta-microarray cohort (Figure 5F). Additionally, ROC analysis was performed, indicating that early survival prediction using the nomogram outperformed other clinical models and risk scores, whereas for long-term survival prediction (>5 years), utilizing the risk score alone yielded better results (Figure 5G).

We utilized the CIBERSORT method to analyze the immune cell composition of tissue samples, comparing the high-risk and low-risk groups, and associating them with model genes. Interestingly, we discovered that the high-risk group exhibited lower levels of CD8⁺ T cells and M1 macrophages (Figure 6A). CD8⁺ T cells, typically referred to as cytotoxic T lymphocytes, secrete various cytokines involved in immune responses (Mittrucker et al., 2014), while M1-type macrophages are capable of producing pro-inflammatory cytokines (Mills et al., 2016). The reduction of CD8⁺ T cells and M1 macrophages may indicate an “cold environment” in high-risk patients. Furthermore, we observed a significant positive correlation between the expression of the PYHIN1 gene and CD8⁺ T cells, suggesting that the PYHIN1 gene primarily influences changes in the tumor microenvironment by regulating the proliferation of cytotoxic T lymphocytes (Figure 6B). We demonstrated the specific distribution of risk scores and different immune cell types, revealing a negative correlation between CD8⁺ T cells and risk scores. Moreover, it appears that changes in risk scores also influence the proportions of M0/M1/M2 macrophages. Furthermore, we characterized the correlation between model genes and immune cells in AMI samples, which similarly showed a strong significant positive correlation between PYHIN1 gene expression and CD8⁺ T cells, potentially regulating the proliferation of resting NK cells (Figure 6C).

The analysis of hallmark pathway gene features using the GSVA method reveals distinct differences between high-risk and low-risk groups in OC. A direct comparison between these groups demonstrates specific enrichment features. In the high-risk group, the top five enriched features include Estrogen Response Early, Myogenesis, Notch Signaling, Bile Acid Metabolism, and Heme Metabolism (Figure 7A). Conversely, the low-risk group exhibits the top five enriched features: E2F Targets, G2M Checkpoint, PI3K/AKT/MTOR Signaling, MYC Targets V1, and Mitotic Spindle. Additionally, we investigated the significantly different pathway signals between patients with AMI and coronary artery disease. The findings reveal substantial activation of the P53 Pathway, Hypoxia, and Notch Signaling in AMI patients (Figure 7B). Of particular interest is the shared activation of the Notch Signaling pathway, which may provide insight into the common high-risk factors underlying both AMI and cancer.

To validate the reliability of the four prognostic genes identified, qRT-PCR testing was performed on both clinical samples and cell lines. In PBMC samples, we observed consistent differential expression patterns of the four genes with the results obtained from the microarray analysis (Figures 3A–D). Specifically, compared to CAD, WEE1, PYHIN1, and SEC61A2 were downregulated in AMI, while HAL exhibited upregulation (Supplementary Figure S3A). Furthermore, it is noteworthy that the validation performed at the cell lines also demonstrated consistent results with RNA-seq analysis. Specifically, when compared to IOSE-80 cells, SKOV3 cells exhibited upregulation in the expression of WEE1, PYHIN1, and HAL, while the expression of SEC61A2 was downregulated (Supplementary Figure S3B).