Rationale: Esophageal squamous cell carcinoma (ESCC) is a highly aggressive malignancy. The metastasis and poor prognosis of ESCC are closely associated with tumor microenvironment (TME) heterogeneity, which is driven by epithelial-mesenchymal transition (EMT). Clinically, how to diagnose and target EMT progression remains a key challenge for ESCC. Methods: Integration of pathological images and bulk RNA sequencing profiles identified a high-risk subtype exhibiting EMT enrichment and immunosuppression. Single-cell and spatial transcriptomics revealed EMT macrostates and their spatial distribution. The role of CACNA1C in programming malignant phenotype was tested in vitro. A pathological image-based deep learning model successfully predicted the spatial expression distribution of CACNA1C, indicating possible clinical utility. Results: EMT progression comprised three macrostates: the early state (high epithelial and metastatic potential), the stable state (hybrid E/M phenotype and high stemness), and the late state (high mesenchymal and invasive propensity). ITGA3 and ITGB4 antagonistically regulate malignant phenotype in the early state. Notably, suppression of CACNA1C induced transdifferentiation from stable/late-state cells to normal epithelium-like cells. Conclusions: This study provides novel insights into the EMT mechanism in ESCC, proposes an intervention strategy, and emphasizes the promising clinical application of pathological images in EMT assessment.
Keywords: bulk sequencing; deep learning; epithelial mesenchymal transition; esophageal squamous carcinoma; histopathology; single-cell sequencing; spatial transcriptomes.
© The author(s).