Functional assessment of patient drug sensitivity using AI-powered image analysis on patient-derived organoids and micoorganoid spheres

Lately, there has been a push for new laboratory models that more accurately represent human biology than traditional models. These include miniature tumor models such as, patient-derived organoids (PDOs) and MicorOrganoidSpheres (MOS), which act as functional avatars of patient tumors and treatment response 1,2 . While recent studies provided first evidence that patient responses to standard-of-care therapies could be recapitulated, these studies were only able to predict clinical responses in a subset of patients 3 . This limitation largely stems from the analytical methods used, namely CellTiter-Glo 3D 4 , which rely on a bulk, endpoint analysis and only extracts a fraction of clinically relevant insights that PDOs provide 5 . Therefore, we hypothesized that using kinetic, higher-dimensional analysis methods, further improves the predictive performance of PDOs. We combined live-imaging techniques with AI-driven analysis to capture dynamic drug responses. Using a fully characterized a PDO panel (n=8) from patients with pancreatic ductal adenocarcinoma (PDAC) and a fully characterized MOS panel (n=43) from patients with colorectal cancer, we matched our multiparametric analysis with retrospective clinical patient response to standard of care therapies (e.g. gemcitabine-paclitaxel, FOLFIRINOX, oxaliplatin). Our PDO analysis quantified resistant and sensitive PDO clones within the patient, and identified patient-specific sensitives to therapy that were in-line with progression-free survival of matched patients (R=0.97) 6 . This was a significant improvement to the relative viability readouts from CellTiter-Glo3D (R 2 =0.26). Our MOS analysis correlated with patient sensitivity and resistance to oxaliplatin. Taken together, our work highlights the importance of using sophisticated analysis methods to measure the complexity new laboratory models, such as MOS and PDOs. Our ongoing work include developing more robust predictive models, using the multiparametric readouts of our analysis platform. 1. Hadj Bachir, E., et al. Biol Cell 114 (2021) 2. Ding, S., et al. Cell Stem Cell 29 (2022) 3. Driehuis, E., et al. Proc Natl Acad Sci 116 (2019) 4. Sachs, N., et al. Cell 172 (2018) 5. Phan, N., et al. Commun Biol 2 (2019) 6. Le Compte, M., et al. npj Precis Oncol (2023)