Tumors are complex ecosystems composed of diverse cell types that interact spatially and functionally to drive disease progression. While the conceptual framework of cancer hallmarks has provided a unifying lens to interpret oncogenic processes, how these hallmarks are spatially organized across tumor compartments remains largely unexplored. Here, we leverage spatial transcriptomics data from 63 untreated tumors spanning 10 cancer types to systematically map the spatial distribution of 13 canonical cancer hallmarks. We show that hallmark activity is spatially patterned and compartmentalized: seven hallmarks, such as genomic instability, apoptosis resistance, and epigenetic reprogramming are predominantly active in cancer cells, while the remaining six, including immune evasion, angiogenesis, and invasion, are enriched in the tumor microenvironment (TME).
By integrating spatial transcriptomics with inferred copy number variation, we reveal that subclonal genetic divergence is associated with hallmark specialization, suggesting that clonal populations may evolve to occupy distinct ecological niches within the tumor. We further uncover structured spatial interdependencies between neoplastic and TME hallmarks using machine learning models, showing that hallmark activity in one compartment often predicts activity in the other. These spatial dependencies form layered architectures across tumors, with hallmark-defined ecological zones conserved across cancer types.
To test clinical relevance, we applied our approach to 33 pre-treatment muscle-invasive bladder cancer samples from the DUTRENEO clinical trial. Spatial hallmark dependencies, particularly those involving angiogenesis, immune evasion, and resistance to cell death correlated with response to neoadjuvant chemotherapy and immunotherapy. For instance, angiogenesis tightly coupled to apoptosis resistance predicted chemoresistance in immune-cold tumors, while spatial decoupling of immune evasion from cancer cell survival predicted immunotherapy response.
Our findings establish the spatial organization of cancer hallmarks as a defining feature of tumor ecosystems, offering a novel framework to interpret functional heterogeneity and predict therapeutic vulnerability.
