Quantification of CAR T cell performance against acute myeloid leukemia using Bayesian inference

Chimeric Antigen Receptor (CAR) T cell therapy offers promising avenues for cancer treatment. Insights into CAR T cell kinetics and cellular dynamics may help identify better dosing and targeting regimens. Mathematical models of cancer and immune cell interactions are valuable tools that integrate existing knowledge with predictive capabilities, thereby narrowing the experimental search space. We formulated a mathematical model with a general T cell expansion functional form by drawing a parallel between predator-prey and immune-tumor interactions. We then compared the abilities of different T cell expansion candidate models to recapitulate a novel in vitro data set of CAR T cells targeting various myeloid antigens on leukemic target cells with different TP53 genotypes. We used Bayesian parameter inference for each candidate model based on the in vitro assay. This approach enabled us to statistically compare candidate models with competing assumptions and select a model that best described the in vitro cytolytic assay longitudinal dynamics. The best-performing CAR T cell expansion model accounts for the detrimental effects of a T cells average time to eliminate a leukemia cell and for effector T cell self-interference. We validated this model on unseen data and used it to predict the expected long-term outcomes of single- and multi-dose CAR T cell therapy against acute myeloid leukemia. Our work demonstrates the utility of predator-prey-like mathematical models and Bayesian inference to investigate and assess the performance of novel CAR T cell constructs, helping to guide the translation to clinically relevant and feasible dosing strategies.