Abstract B027: Ecological landscape analysis of the tumor microenvironment predicts recurrence risk in pancreatic ductal adenocarcinoma

Abstract Pancreatic ductal adenocarcinoma (PDAC) remains among the most aggressive cancers, due to its limited therapeutic options and poor prognosis. Even with the adoption of total neoadjuvant therapy (TNT) and potentially curative surgical resection, PDAC relapse rates remain unacceptably high. Current systems for evaluating tumor response are valuable but limited, as they fail to capture the ecological complexity and broader architecture of the tumor microenvironment (TME). This results in uncertainty regarding the prognosis and recurrence risk of patients. To address this and to capture the complexity of the PDAC TME, we developed the Tumor Landscape Analysis (TLA). Using digitized H&E slides from 203 PDAC patients who received TNT and surgical resection at our institution, we applied a machine-learning pipeline to segment tumor and stromal compartments. We then quantified the TME architecture using ecological frameworks to assess the disease in multiple scales: individual patches, whole-slide landscapes, and aggregated patient-level summaries. Our analysis revealed that higher-order spatial features are independently prognostic of disease-free survival (DFS). Two statistical models showed different but complementary spatial characteristics that predict disease risk. In the first model, patients with more irregular cancer patch shapes and lower variability in stromal shape had significantly better DFS, with high-risk patients relapsing in a median of 7.2 months compared to 11.6 months in the low-risk group (HR=1.82, p=0.0003). In the second model, smaller stromal patch areas combined with greater tumor-stroma edge density identified patients at elevated risk, with high-risk individuals showing a median DFS of 8.0 months versus 15.7 months in low-risk patients (HR=1.73, p=0.0007). Importantly, these associations persisted even after adjustment for the T and N stage, showing that spatial features provide unique information beyond traditional pathology measures. These models help visualize PDAC as a comprehensive ecosystem that carries clinical significance. By quantifying tumor conformation and configuration, our models reveal prognostic information not available through CAP grading, percent tumor burden, or routine cell counts. This study suggests that tumor-stroma interactions and their spatial organizations are potentially biologically important for recurrence. We present a new, ecologically inspired approach to digital pathology for PDAC that not only improves our ability to predict recurrence and survival in PDAC but also sheds light on cancer as a complex, evolving ecosystem. Citation Format: Merih D. Toruner, Luis H. Cisneros, Miranda Lin, Andrea Maraone, Carlo Maley, Paul Dizona, Qian Shi, Martin E. Fernandez-Zapico, Ryan M. Carr. Ecological landscape analysis of the tumor microenvironment predicts recurrence risk in pancreatic ductal adenocarcinoma [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Cancer Evolution: The Dynamics of Progression and Persistence; 2025 Dec 4-6; Albuquerque, NM. Philadelphia (PA): AACR; Cancer Res 2025;85(23_Suppl):Abstract nr B027.