Abstract

Pancreatic ductal adenocarcinoma (PDAC) displays striking spatial heterogeneity, particularly in the organization and interactions between cancer cells and cancer-associated fibroblasts (CAFs). Yet, the mechanisms that shape this spatial complexity remain poorly understood. Here, we integrate high-resolution spatial transcriptomic data from human PDAC samples with mechanistic computational modelling to uncover the principles governing tumour microenvironment (TME) organization. Spatial proximity and neighbourhood analyses reveal microenvironmental niches enriched for distinct CAF subtypes (iCAFs and myCAFs), as well as characteristic clustering patterns between cancer cells and CAFs. Receptor–ligand interaction profiling further highlights candidate signalling axes mediating cancer–CAF crosstalk. To mechanistically probe these observations, we develop a spatially explicit agent-based model (ABM) that incorporates cytokine diffusion and phenotype transitions to simulate local cell–cell interactions. The model tests whether simple, rule-based interactions are sufficient to reproduce the emergent spatial architectures and phenotypic heterogeneity observed in the transcriptomic data. Together, this integrative framework bridges spatial omics with mechanistic modelling, offering new insights into how cellular interactions shape PDAC tissue architecture and contribute to TME organization.