A Patient-Derived Organoid-Based Radiosensitivity Model for the Prediction of Radiation Responses in Patients with Rectal Cancer.

Abstract

Simple SummaryPredicting the tumor regression grade of locally advanced rectal cancer after neoadjuvant chemoradiation is important for customized treatment strategies; however, there are no reliable prediction tools. A novel preclinical model based on patient-derived tumor organoids has shown promising features of the recapitulation of real tumors and their treatment response. We conducted a small co-clinical trial to determine the correlation between the irradiation response of individual patient-derived rectal cancer organoids and the results of actual radiotherapy. Among the quantitative experimental data, the survival fraction was best matched and correlated with the patients’ real treatment outcome. In the machine learning-based prediction model for radiotherapy results using the survival fraction data, the prediction accuracy was excellent at more than 89%. Enhanced machine learning with the accumulation of further new experimental data would help in creating a more reliable prediction model, and this new preclinical model can lead to more advanced precision medicine.Patient-derived tumor organoids closely resemble original patient tumors. We conducted this co-clinical trial with treatment-naive rectal cancer patients and matched patient-derived tumor organoids to determine whether a correlation exists between experimental results obtained after irradiation in patients and organoids. Between November 2017 and March 2020, we prospectively enrolled 33 patients who were diagnosed with mid-to-lower rectal adenocarcinoma based on endoscopic biopsy findings. We constructed a prediction model through a machine learning algorithm using clinical and experimental radioresponse data. Our data confirmed that patient-derived tumor organoids closely recapitulated original tumors, both pathophysiologically and genetically. Radiation responses in patients were positively correlated with those in patient-derived tumor organoids. Our machine learning-based prediction model showed excellent performance. In the prediction model for good responders trained using the random forest algorithm, the area under the curve, accuracy, and kappa value were 0.918, 81.5%, and 0.51, respectively. In the prediction model for poor responders, the area under the curve, accuracy, and kappa value were 0.971, 92.1%, and 0.75, respectively. Our patient-derived tumor organoid-based radiosensitivity model could lead to more advanced precision medicine for treating patients with rectal cancer.