Abstract 1229: 3D-bioprinted cancer models for prediction of drug response driven by H&E-based AI algorithm

Many drugs show promising results in laboratory research but eventually fail to show clinical benefit. One main reason for this translational gap is that the cancer models used are inadequate. Most models lack the complex tumor-stromal-immune cell interactions with their tumor microenvironment (TME) that are required for progression. Conventional 2D models, where cells grow on rigid plastic plates mainly as mono-cultures, fail to recapitulate these complex interactions. Hence, we developed two platforms of 3D cancer models that better mimic the TME: (i) a 3D tumoroid model consisting of all the cells from the patient’s tumor tissue, following cell suspension, and isolated peripheral blood mononuclear cells (PBMC) from the same patient’s blood, and (ii) a vascularized, hydrogel-based 3D-bioprinted tumor model consisting of patient-derived tumor, PBMCs and TME cells. Our 3D models are based on a library of hydrogels that we developed as scaffolds for different tumor types according to the mechanical properties of the tissue of origin which dictates the proteo-genomic signature and, hence, the response to different therapies. The patient-derived models consist of cells isolated from a freshly-resected tumor mass and include PBMCs that either surround the 3D tumoroids or flown through 3D-bioprinted functional blood vessels connected to a pump. Using these unique high throughput 3D models, we are currently validating their ability to mimic patient-specific tumors and TME, in order to serve as a tool to predict patient response to different treatments (e.g., chemotherapy, immunotherapy, and targeted therapies). To decide which therapies to test on our 3D platforms, we consider cancer specific standard-of-care (SoC), investigational therapies as well as therapies proposed by ENLIGHT-DP, an AI-based algorithm developed by Pangea Biomed that utilizes H&E slides to generate individual drug response scores for targeted and immune therapies. The ENLIGHT-DP first predicts gene expression from H&E slides using a deep-learning AI algorithm and then uses the predicted expression to generate response scores for a variety of therapies. To validate the ability of our models to predict patient responses, we are currently conducting an 80-patient clinical trial (IRB approved) testing different treatments on samples covering 7 different cancer types. An interesting case from this trial is a mucosal melanoma patient who failed SoC treatments correlated with failure to respond in our models. Interestingly, ENLIGHT-DP predicted that the patient would respond to regorafenib, a suggestion corroborated by our 3D model. Following this promising outcome, the patient received it as compassionate use, resulting in an immediate partial response lasting for 6 months so far. These unique 3D models have the potential to facilitate target discovery and drug development, as well as to serve as a reliable system for precision medicine. Citation Format: Ronit Satchi-Fainaro, Anne Krinsky, Anshika Katyal, Opal Avramoff, Yulia Liubomirski, Gal Dinstag, Omer Tirosh, Ranit Aharonov, Tuvik Beker, Iris Barshack, Shaked Lev-Ari, Shirly Grynberg, Ronnie Shapira-Frommer. 3D-bioprinted cancer models for prediction of drug response driven by H&E-based AI algorithm [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 1229.