Data-driven computational modeling of CAR-T cell function

Introduction The complex dynamics of chimeric antigen receptor T-cell (CAR-T cell) cytotoxicity and proliferation are potential factors that influence the clinical response to CAR-T therapy. The patient-specific functionality of CAR-T products play a role in these dynamics. CAR-T products comprise phenotypically and functionally distinct populations of cells that impact therapy response in different ways. We hypothesized that product-specific parameters exist that predict individual patient responses to therapy and that these can be elucidated by simulating the interactions of CAR-T products and tumor cells using an in vitro assay-based model. Methods We use an ordinary differential equation (ODE)-based pharmacokinetic (PK) and pharmacodynamic (PD) model to characterize key CAR-T cell functional parameters. Parameters for the model developed using our method are product-specific and derived from in vitro assays performed on individual patient CAR-T products from clinical trial NCT04186520. Results Our results demonstrate that while considerable variability is present in in vitro cytotoxicity kinetics and subsequently estimated model parameters between each product, these differences do not predict early (28 days) or late responses (90 days) after treatment across the total cohort of patients investigated. However, we show that differences in an estimated model parameter for increased CAR-T cell responsiveness to tumor cytotoxicity are correlated with durable therapy responses (no relapse through 180 days). Additionally, in a cohort of diffuse large B-cell lymphoma (DLBCL) patients, we demonstrate that a model parameter estimating cooperativity between CAR-T cells is also correlated with durable therapy responses and that may be related to differences in CD4:CD8 ratios in the CAR-T cell product. Conclusions Overall, our work demonstrates that while pre-treatment CAR-T cell functional parameters vary on a patient and product basis, these parameters do not predict initial therapeutic responses. We find that initial therapeutic responses are possible across a range of initial product kinetic parameters. However, we observed that their potentially exist unique kinetic properties associated with the initial product that is predictive of disease relapse.