A Prospective Feasibility Trial to Challenge Patient–Derived Pancreatic Cancer Organoids in Predicting Treatment Response

Abstract

Simple Summary Pancreatic cancer (PC) is characterized by an exceptionally aggressive tumor biology, high inter- and intratumor heterogeneity, and resistance to conventional chemotherapy, targeted agents, and immunotherapy. With its rising incidence and dismal prognosis, PC is projected to become the second-leading cause of cancer-related death worldwide in 2030. Tumor heterogeneity induces a considerable variation in responses to antitumor therapies, yet reliable models or biomarkers to predict the effectiveness of treatment strategies for eligible subgroups are not established. Current combination chemotherapeutic regimens are often ineffective and frequently exhibit substantial systemic toxicity impeding longer-term treatment. Patient-derived pancreatic cancer organoids (PDOs) exhibit features of the parental tumor and may thereby represent a powerful preclinical tool to predict drug response. Ex vivo pharmacotyping may enable therapy response prediction and harness personalized treatment in PC patients. In clinical practice, a PDO-guided selection of effective drugs may provide substantial benefit and improve survival outcomes in this heterogeneous disease. Abstract Real-time isolation, propagation, and pharmacotyping of patient-derived pancreatic cancer organoids (PDOs) may enable treatment response prediction and personalization of pancreatic cancer (PC) therapy. In our methodology, PDOs are isolated from 54 patients with suspected or confirmed PC in the framework of a prospective feasibility trial. The drug response of single agents is determined by a viability assay. Areas under the curves (AUC) are clustered for each drug, and a prediction score is developed for combined regimens. Pharmacotyping profiles are obtained from 28 PDOs (efficacy 63.6%) after a median of 53 days (range 21–126 days). PDOs exhibit heterogeneous responses to the standard-of-care drugs, and are classified into high, intermediate, or low responder categories. Our developed prediction model allows a successful response prediction in treatment-naïve patients with an accuracy of 91.1% for first-line and 80.0% for second-line regimens, respectively. The power of prediction declines in pretreated patients (accuracy 40.0%), particularly with more than one prior line of chemotherapy. Progression-free survival (PFS) is significantly longer in previously treatment-naïve patients receiving a predicted tumor sensitive compared to a predicted tumor resistant regimen (mPFS 141 vs. 46 days; p = 0.0048). In conclusion, generation and pharmacotyping of PDOs is feasible in clinical routine and may provide substantial benefit.