Deciphering Vascular Smooth Muscle Cell Phenotype Switching via Integrated Single-Cell Transcriptomics and Machine Learning: CX3CR1 and COL1A1 as Novel Diagnostic Biomarkers for Atherosclerosis

Background: Atherosclerosis is a chronic inflammatory disease of the arterial wall in which immune-cell accumulation and vascular stromal remodeling jointly shape plaque progression and stability. Because bulk-tissue transcriptomes combine signals from multiple cell types, the cellular origin of candidate expression markers must be resolved before they can be interpreted as vascular smooth muscle cell (VSMC)-intrinsic features. Methods: We integrated the GSE309462 human carotid-plaque single-cell RNA-sequencing dataset with the bulk arterial-tissue datasets GSE100927 and GSE57691. Intersection-derived genes were prioritized using LASSO, random forest, and SVM-RFE. Discriminatory direction was standardized for receiver operating characteristic analysis, the five consensus genes were evaluated in the independent plaque-stage dataset GSE28829, and their cellular origins were audited across 99,027 cells from 12 donors. Results: Forty-nine genes were shared between the SMC-cluster-associated gene set and the bulk arterial-tissue differential-expression results. Five genes (MPEG1, CX3CR1, SERPINA3, LMCD1, and COL1A1) were selected by all three machine-learning procedures. In the training data, direction-standardized AUCs ranged from 0.859 to 0.922. In GSE28829, MPEG1, CX3CR1, and COL1A1 showed higher expression in advanced plaques and AUCs of 0.904, 0.827, and 0.928, respectively; LMCD1 was higher in early plaques (AUC 0.870), whereas SERPINA3 showed limited discrimination (AUC 0.442). Single-cell localization showed that MPEG1 and CX3CR1 were predominantly expressed in macrophages, whereas LMCD1 and COL1A1 were enriched in VSMCs and other vascular stromal cells. Conclusion: The expression signature captures compartmentally distinct processes, particularly myeloid-cell infiltration and extracellular-matrix remodeling. These genes should therefore be interpreted as plaque-associated tissue markers and mechanistic candidates rather than validated non-invasive diagnostic biomarkers or definitive VSMC-lineage markers.