Potential biomarkers and immune characteristics for polycythemia vera-related atherosclerosis using bulk RNA and single-cell RNA datasets: a combined comprehensive bioinformatics and machine learning analysis.

Abstract

Polycythemia vera (PV) is a myeloproliferative disease characterized by significantly higher hemoglobin levels and positivity for JAK2 mutation. Thrombosis is the main risk event of this disease. Atherosclerosis (AS) can markedly increase the risk of arterial thrombosis in patients with PV. The objectives of our study were to identify potential biomarkers for PV-related AS and to explore the molecular biological association between PV and AS. We extracted microarray datasets from the Gene Expression Omnibus (GEO) dataset for PV and AS. Common differentially expressed genes (CGs) were identified by differential expression analysis. Functional enrichment and protein-protein interaction (PPI) networks were constructed from the CG by random forest models using LASSO regression to identify pathogenic genes and their underlying processes in PV-related AS. The expression of potential biomarkers was validated using an external dataset. A diagnostic nomogram was constructed based on potential biomarkers to predict PV-related AS, and its diagnostic performance was assessed using ROC, calibration, and decision curve analyses. Subsequently, we used single-cell gene set enrichment analysis (GSEA) to analyze the immune signaling pathways associated with potential biomarkers. We also performed immune infiltration analysis of AS with "CIBERSORT" and calculated Pearson's correlation coefficients for potential biomarkers and infiltrating immune cells. Finally, we observed the expression of potential biomarkers in immune cells based on the single-cell RNA dataset. Fifty-two CGs were identified based on the intersection between up-regulated and down-regulated genes in PV and AS. Most biological processes associated with CGs were cytokines and factors associated with chemotaxis of immune cells. The PPI analysis identified ten hub genes, and of these, CCR1 and MMP9 were selected as potential biomarkers with which to construct a diagnostic model using machine learning methods and external dataset validation. These biomarkers could regulate Toll-like signaling, NOD-like signaling, and chemokine signaling pathways associated with AS. Finally, we determined that these potential biomarkers had a strong correlation with macrophage M0 infiltration. Further, the potential biomarkers were highly expressed in macrophages from patients with AS. We identified two CGs (CCR1 and MMP9) as potential biomarkers for PV-related AS and established a diagnostic model based on them. These results may provide insight for future experimental studies for the diagnosis and treatment of PV-related AS.