Model-Based Cellular Kinetic Analysis of Chimeric Antigen Receptor-T Cells in Humans

Abstract

Chimeric antigen receptor (CAR)-T cell therapy has achieved considerable success in treating B-cell hematologic malignancies. However, the challenges of extending CAR-T therapy to other tumor types, particularly solid tumors, remain appreciable. There are substantial variabilities in CAR-T cellular kinetics across CAR-designs, CAR-T products, dosing regimens, patient responses, disease types, tumor burdens, and lymphodepletion conditions. As a "living drug," CAR-T cellular kinetics typically exhibit four distinct phases: distribution, expansion, contraction, and persistence. The cellular kinetics of CAR-T may correlate with patient responses, but which factors determine CAR-T cellular kinetics remain poorly defined. Herein, we developed a cellular kinetic model to retrospectively characterize CAR-T kinetics in 217 patients from 7 trials and compared CAR-T kinetics across response status, patient populations, and tumor types. Based on our analysis results, CAR-T cells exhibited a significantly higher cell proliferation rate and capacity but a lower contraction rate in patients who responded to treatment. CAR-T cells proliferate to a higher degree in hematologic malignancies than in solid tumors. Within the assessed dose ranges (10⁷ -10⁹ cells), CAR-T doses were weakly correlated with CAR-T cellular kinetics and patient response status. In conclusion, the developed CAR-T cellular kinetic model adequately characterized the multiphasic CAR-T cellular kinetics and supported systematic evaluations of the potential influencing factors, which can have significant implications for the development of more effective CAR-T therapies.

Figure 1.

(A) A typical multi-phase kinetics of CAR-T cell: distribution, expansion, contraction and persistence phase; (B) Model structure for CAR-T kinetics; (C) Schematic diagram of three-step workflow for modeling and analysis.

Figure 2.

Model fitting and validation plots in the trial of DLBCL. (A) Individual fitting plots of representative patients; (B) The goodness-of-fit plot; (C) Visual predictive check (VPC) plot

Figure 3.

Individual model-estimated parameters for responders (CR/PR) and non-responders (PD/NR). Data from responders and non-responders are shown separately and compared for each trial. Two ALL trials were also compared, although response information was not available in the trial of ALL (Pediatric). Data distribution were described by boxplot (minimum, first quartile, median, third quartile, and maximum). p < 0.05 (Wilcoxon test) is considered as significant difference.

Figure 4.

Individual model-estimated parameters of three tumor types (ALL/CLL/MM, lymphoma and solid tumors). Data from corresponding trials were pooled into three main categories. Data distribution were described by boxplot (minimum, first quartile, median, third quartile, and maximum). p < 0.05 (Kruskal–Wallis test) is considered a significant difference.

Figure 5.

Patient covariates for responders (CR/PR) and non-responders (PD/NR). (A) A comparison for individual dosing by the response in the trials of CLL, GBM, MM, and NSCLC; (B) A comparison for pre-treatment tumor burden by the response in the trials of CLL and MM. Tumor burden values were not comparable across trials, as they were measured differently, indicated as “BCMA MFI” and “total IgH reads”, respectively; (C) CD4: CD8 ratio was compared by the response in the trials of MM and NSCLC; (D) Patient age was compared by the response in the trials of MM, GBM and NSCLC.

Figure 6.

The simulated kinetics of effector CAR-T (green), memory-phenotypic CAR-T (red), and total CAR-T (blue) for each clinical trial. Simulations were performed based on the population typical values of parameters. CAR-T cell kinetics were simulated until 10 months. Dotted lines represent the duration beyond the period of the clinical trials.