Abstract IA006: Reconstructing evolution with Stochastically Emergent Tumors (SETs) reveals in vivo vulnerabilities

Abstract Rising incidence of early-onset cancers demands experimental systems that model how normal tissues traverse premalignant states to form clinically relevant, therapy-responsive tumors, and how host immunity and ancestry-linked genetics shape that trajectory. A major obstacle has been the lack of tractable in vivo platforms that enable genome-wide discovery while preserving continuous tumor evolution without chromosomal instability. We developed Stochastically Emergent Tumors (SETs), an organoid-derived in vivo evolution engine that addresses this gap and reframes discovery for early-onset and understudied patient groups. In this system, mismatch repair deficiency (MMRD) is induced in nonmalignant human organoids, which are passaged to accumulate stochastic point mutations and transplanted into mice to permit malignancy to emerge under authentic selective pressures. SETs evolve primarily through high-resolution point mutations rather than broad copy number changes, producing bioinformatically tractable clonal dynamics ideally suited for whole-genome driver discovery and machine learning. Compared with conventional models, SETs exhibit increased intertumoral heterogeneity and reproducible recovery of sensitizing and resistance alleles under selective pressure. As proof of principle in prostate cancer, endocrine therapy applied to prostate SET pools recovered known determinants of sensitivity and revealed novel drivers. Loss of ZFHX3 function, typically obscured within a multi-gene tumor suppressor locus in bulk cohorts, promoted luminal histology and sensitized tumors to androgen pathway inhibition in vivo, while alterations in KMT2D and CIC mediated resistance. Consistent with model predictions, ZFHX3 loss in patients was associated with markedly increased survival, a novel finding comparable in magnitude to the longest-surviving molecular subtypes of advanced prostate cancer. These results identify selective vulnerabilities that arise from stochastic evolution. We also defined an oncogenic threshold by relating pregraft mutation burden to tumor incidence: in a Pten-null background, roughly 900 coding mutations accumulated over 208 days were sufficient for malignant transformation in half of grafts. This quantifies the median number of stochastic events required for premalignant-to-malignant transition and establishes a measurable axis to assess how genetics and microenvironmental pressures shift that threshold. Because SETs generate neoantigen-rich point mutation landscapes, they can be adapted to immunocompetent and premalignant paradigms to interrogate tumor–immune coevolution. Coupled with spatial and functional readouts, SETs enable mapping of early T-cell surveillance, macrophage-mediated immune exclusion, and myeloid checkpoints that permit immune escape. SETs therefore provide a scalable, evolution-aware platform to identify lineage- and ancestry-associated drivers, immunopreventive targets, and biomarkers of early-onset cancer. Citation Format: Ruhollah Moussavi-Baygi, Matthew Ryan, Woogwang Sim, Samuel Hoelscher, Valbona Luga, Arun Chandrakumar, Lore Hoes, Junghwa Cha, Young Sun Lee, Katelyn Herm, Ben Doron, Danika Bakke, Cornelia Ding, Bradley Stohr, Peng Jin, Tejasveeta Nadkarni, Xiangyi Fang, Melita Haryono, An Nguyen, Wouter Karthaus, Charles Sawyers, Felix Feng, Hani Goodarzi, Rohit Bose. Reconstructing evolution with Stochastically Emergent Tumors (SETs) reveals in vivo vulnerabilities [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 IA006.