Abstract 7154: Phenotypic screening from patient-derived organoids to predict therapeutic response

Abstract

Abstract Functional precision medicine (FPM) aims to personalize cancer treatment through ex vivo drug testing on patient tumor cells. Classical biochemical tests (like ATP assay) provide limited insight into how patient cells respond to treatment. Phenotypic markers based on imaging features called "radiomics" offer a promising approach to understanding tumor response. Combining FPM with phenotypic screening (PS), we evaluate the predictive power of biomarkers derived from brightfield time-lapse imaging of patient-derived organoids (PDO) from digestive cancer patients. We screened 24 patients, including 5 with colorectal (CRC) and 19 with pancreatic (PDAC) cancers, each with varying treatment histories (1 to 8 treatments prior to PDO biopsy and 12 different treatments). PDOs were exposed to the corresponding clinical treatments, and 96-hour post-treatment videos allowed for individual tracking of PDO responses. A total of 103 image features (18 intensity, 75 texture, 10 morphology) were extracted from each PDO. Predictive analysis was conducted at 24h, 48h, 72h, and 96h post-treatment, using single timepoints and time-averaging methods. PDO responses were assessed at 3 treatment doses, and area dose-response (ADR) was calculated using trapezoid estimation. Univariate and multivariate Cox models were applied to assess the association of image features with progression-free survival (PFS) and develop an image-based signature for PFS prediction. We compared image features extracted from entire movie sequences (49 frames), 24h-averaging (13 frames), 4h-averaging (3 frames), and single timepoints (24h, 48h, 72h, 96h). We found similar results from univariate Cox analyses and focused on single timepoint method, which identified 24 features significantly associated with PFS as early as 48 hours. These included intensity (7), texture (11), and shape (6) features (median C-index=0.63, 95% CI [0.60-0.67]). When combining each feature with clinical factors (disease and treatment), 7 features remained significantly associated with PFS. A subset of 3 non-correlated interpretable features was selected (one per category). A signature built from a texture feature measuring pixel intensity homogeneity, combined with disease and treatment best predicted PFS (C-index=0.72, HR=27.43 95% CI [1.89-399.28], p-value=0.046). Integrating image biomarkers with clinical variables enhances the predictive power of PDO-based functional drug screening. Extracting features from individual timepoints provides earlier insights than biochemical assays, which typically require 5 days post-treatment. This study encourages further exploration of how PS-based FPM could be implemented in a clinical setting. Future work will focus on validating the image-based signature in an independent cohort and across diverse experimental setups. Citation Format: Emilie Gontran, Leo Fillioux, Jerome Cartry, Ryme Bouyakoub, Sabrina Bedja, Alice Boileve, Ali Mouawia, Anna-Rose Gryspeert, Mia Cabantous, Gizem Altay, Maria Vakalopoulou, Stergios Christodoulidis, Jacques RR Mathieu, Paul-Henry Cournede, Fanny Jaulin. Phenotypic screening from patient-derived organoids to predict therapeutic response [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 7154.