Abstract A015: Tumor landscape analysis: An ecologically informed framework to understand tumor microenvironments

Abstract Background: The spatial organization of the tumor microenvironment (TME) encodes the selective pressures that drive cancer evolution, but many tools focus on local cell contacts or single-modality readouts. We developed Tumor Landscape Analysis (TLA), a modular framework that borrows from landscape ecology and spatial statistics to quantify heterogeneity, fragmentation, and diversity across scales, from single cells to whole-slide ecosystems. Methods: TLA ingests whole-cell or nuclear centroids and region-level segmentations to compute (i) point statistics such as nearest-neighbor distance, Ripley’s H, and Morisita-Horn colocalization; (ii) kernel-smoothed local spatial profiles including Getis-Ord Gi* hot and cold spots; and (iii) fragmentation, shape, and diversity metrics such as patch density, contagion, shape index, and Shannon entropy. We also introduce Local Microenvironments (LMEs), unsupervised ecological niches defined by joint cell-type abundance and spatial mixing. Pilot analyses were performed in ductal carcinoma in situ (DCIS; n=353 samples, 175 patients) with cell-level maps of epithelial, lymphocyte, and fibroblast populations, and in pancreatic ductal adenocarcinoma (PDAC; n=203 resected cases after neoadjuvant therapy) using cancer and stroma segmentations. Results: In DCIS, spatial relationships rather than raw abundance associated with recurrence risk. Shorter lymphocyte-to-fibroblast minimum nearest-neighbor distance and lower Ripley’s H, indicating greater intermixing, correlated with earlier recurrence independent of cell counts. This highlights immune–stromal proximity as a candidate eco-evolutionary driver of persistence. In PDAC, tissue-level fragmentation captured prognostic architecture after therapy. Two shape-based signatures emerged: (1) stromal uniformity, where lower standard deviation of stromal patch shape index associated with improved disease-free survival, consistent with a more constrained, less reactive stroma; and (2) tumor geometric complexity, where higher mean cancer patch shape index, reflecting more irregular or dendritic nests, associated with lower recurrence risk, whereas compact nests predicted worse outcomes. Across datasets, LME maps localized niche phenotypes, such as mixed-immune or segmented tumor habitats, that align with hypothesized selection landscapes and provide inputs for fragmentation analysis. Conclusions: TLA operationalizes an eco-evolutionary view of cancer by unifying interpretable, multiscale spatial metrics in a single workflow that works with any rasterized data input. By quantifying how habitats, edges, and mixing patterns are organized, TLA reveals niche structures and architectural motifs linked to clinical outcomes and complements foundation models and modality-specific pipelines with theory-grounded, clinically translatable descriptors. Ongoing work integrates TLA features with molecular data and outcomes to derive robust biomarkers and to inform adaptive, niche-targeted strategies in precision oncology. Citation Format: Ryan M. Carr, Luis Cisneros, Merih Deniz. Toruner, Martin E. Fernandez-Zapico, Carlo Maley. Tumor landscape analysis: An ecologically informed framework to understand tumor microenvironments 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 (23Suppl): Abstract nr A015.