Abstract 3080: A quality-controlled patient-derived tumor organoid biobank facilitates applications such as an integrated database of chemotherapeutic drug response using deep learning-based imaging methods

Abstract

Abstract Patient-derived tumor organoids (PDTOs) have recently been used as an innovative preclinical model to predict patient-specific drug responses. However, it is critical to establish high quality, reproducible, and well-annotated PDTOs that closely recapitulate the original tumors from which they were derived. Here, we established a biobank of over 20 colorectal cancer (CRC) PDTOs representing a racially and ethnically diverse patient population that has been subjected to rigorous quality control (QC) measures. We demonstrate the utility of a well-characterized biobank across several applications including the development of an integrated chemotherapeutic drug response database and image-based phenotypic analyses using deep-learning neural networks (NNs). To ensure consistency across our biobank, PDTOs were subjected to a meticulous processing pipeline which includes the extensive collection of pertinent culturing metadata (i.e., multi-timepoint brightfield images pre- and post-passaging for consistent size and health of the PDTOs, short tandem repeat sequencing to establish a unique patient barcode, and routine mycoplasma testing results to minimize contamination) and DNA/RNA sequencing to verify the PDTOs reflect the genetic profile of the original tumor. A dashboard is used to manage culturing and characterization tasks across the Cell Line Team and samples are registered in a Lab Information Management System (LIMS) for tracking of data associated with the samples, including detailed patient information. Our high-quality biobank supports a diverse project portfolio including the creation of a database of PDTO drug responses to clinically-approved chemotherapy treatments 5-Fluorouracil and Irinotecan. This database compares PDTO-specific dose-response curves between ATP-based viability assays and image-based NN analysis and we found the IC50 values from the two analytical methods to be comparable. The NNs were designed to segment and classify organoids as live or dead from brightfield images. The label-free, non-destructive nature of this method enables dynamic imaging and analysis over multiple timepoints. Notably, our biobank and NN analysis work in concert where the biobank provides valuable, consistent datasets that are used to train and validate NNs and the NN is subsequently used to understand phenotypic drug responses of PDTOs. From these results we can show inter-patient heterogeneity in drug responses. In conclusion, we generated a reliable biobank to support numerous applications including the creation of a PDTO drug response database with deep learning validation. Such a biobank is only feasible through meticulous QC and can serve as a great resource to understand treatment outcomes and identify better therapeutic options for patients in the future. Citation Format: Scott Valena, Pratiksha Kshetri, Brandon Choi, Ah Young Yoon, Shohei Imamura, Seungil Kim, Michael E. Doche, Shannon M. Mumenthaler. A quality-controlled patient-derived tumor organoid biobank facilitates applications such as an integrated database of chemotherapeutic drug response using deep learning-based imaging methods [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 3080.