Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal malignancy with a five-year survival rate below 10%. Cancer-associated fibroblasts (CAFs) promote epithelial–mesenchymal transition (EMT) and metastasis, yet the specific CAF subtypes and molecular axes driving PDAC progression remain incompletely understood. Here, using multi-omics data from PDAC samples, we identified a malignant ductal subpopulation, termed Ductal-T0, characterized by the highest EMT activity and prominent acquisition of myofibroblastic CAF (myCAF)-like transcriptional programs. Computationally, we predicted that myCAF-secreted THBS2 and FN1 engage the ITGA3/ITGB1/SDC1/SDC4 receptor axes in Ductal-T0 cells, which could activate TNF, NF-κB, TGF-β, and PI3K-AKT-signaling pathways to promote EMT. Pseudotime trajectory and velocity analyses suggested that Ductal-T0 cells exhibited the highest propensity to acquire myCAF-like features among all ductal subpopulations. Survival analysis revealed that an increased proportion of Ductal-T0 cells and elevated abundance of THBS2-ITGA3/ITGB1 and THBS2-SDC1 ligand–receptor pairs were significantly associated with poor prognosis. Spatial transcriptomics further revealed that myCAFs and Ductal-T0 cells co-localized at the tumor margin, which may contribute to reduced immune cell presence via dense extracellular matrix (ECM) barrier formation—a computationally inferred model of EMT-associated immune exclusion and metastatic progression—and identify THBS2 as a promising candidate for future therapeutic investigation to disrupt CAF–tumor crosstalk in PDAC.