Abstract P19: CellTopo quantifies tissue structure and organization in solid tumors

Abstract

Highly structured tissues in human gut, breast, and lymphoid organs maintain precise cellular organization that enables their functions. In cancer, tissue structure becomes progressively dysregulated which promotes tumour progression and therapeutic resistance. Despite current AI applications in pathology, cellular organization cannot be easily quantified to discern tumour subtype, stage and treatment decisions without extensive manually annotated data. Here, we present CellTopo, an unsupervised topology-based framework that quantifies tissue structure and organization using spatial coordinates of single cells to model tumour progression, immune infiltration, and treatment resistance. CellTopo constructs Vietoris–Rips complexes from coordinates to identify loop-like structures and assigns structureTopo scores to cells based on participation in structures. CellTopo then quantifies the structures’ circularity for irregularTopo scores and measures marker dissimilarity within structures to capture cell type mixing. Together, CellTopo first identifies biologically relevant structures, such as gut crypts, breast ducts, and tertiary lymphoid structures (TLS) more accurately than marker-based watershed methods and then analyses structures for shape irregularity and cell type mixing. When applied to colorectal cancer spatial proteomics datasets, CellTopo’s structureTopo score, distinguished two immune-associated CRC structures; Crohn’s-like reaction (CLR) with high TLS, and diffuse inflammatory infiltration (DII), which lacks TLS. CLR exhibited higher structureTopo scores, indicating lower structural dysregulation and more favourable prognosis. In contrast, when applied to breast cancer, CellTopo identified invasive lobular carcinoma (ILC) as exhibiting greater topological complexity when compared to ductal carcinoma in situ (DCIS), highlighting architecture differences arising from distinct sites-of-origin. Furthermore, in immunotherapy-treated breast cancer, CellTopo distinguished baseline, on and post-treatment samples, and predicted responders at baseline. By modelling local simplex-based structures around cells in tissue imaging data, CellTopo offers a robust, training-free method to identify spatial features linked to tumour progression, stage, and treatment response to improve patient stratification.