Abstract
Pancreatic cancer is a gastrointestinal malignancy with a dismal prognosis. Cancer stem cells (CSCs) are considered key drivers of its aggressiveness, metastatic capacity, and resistance to treatment. However, the intratumoral heterogeneity of CSCs and their roles within the tumor microenvironment remain poorly characterized. We integrated single-cell transcriptomic data (GSE214295) with large-scale bulk RNA-seq datasets (TCGA-PAAD, GEO and CPTAC) for comprehensive analysis. Malignant epithelial subpopulations were identified and their stemness evaluated using Seurat, CopyKAT, and CytoTRACE. Monocle2, scMetabolism, decoupleR, and CellChat were applied to investigate differentiation trajectories, metabolic characteristics, transcription factor activities, and intercellular communication. Finally, CSC-like prognostic index (CSCLPI) was constructed and validated with multiple machine-learning approaches. The robustness and generalizability of the model were validated using the GEO and CPTAC dataset. Seven malignant epithelial subpopulations were identified. Among them, the C2 cluster displayed the highest stemness score, lowest metabolic activity, and prominent CSC-marker expression. Pseudotime analysis placed C2 at the origin of the differentiation trajectory. C2 also exhibited activation of key stemness-related transcription factors (e.g., SOX9, MYC) and robust cell-cell communication via WNT and TGF-β signaling. The CSCLPI model, derived from C2-specific genes, stratified patients into distinct risk groups. The CSCLPI score correlated significantly with tumor mutation burden, stemness indices, immune escape, and chemoresistance, and showed consistent prognostic accuracy across multiple cohorts. A nomogram incorporating CSCLPI and clinical parameters demonstrated strong prognostic value in pancreatic cancer. This study systematically characterized the heterogeneity and functional landscape of CSC-related subpopulations in pancreatic cancer. The stemness-based CSCLPI model offers a robust prognostic tool and potential therapeutic targets for personalized treatment and CSC-directed strategies.