Studying the mechanisms and predicting patient responses of B-cell acute lymphoblastic leukemia under CAR T-cell immunotherapy with a computational tool

Introduction:: Adaptive CD19-targeted CAR (Chimeric Antigen Receptor) T-cell transfer has become a promising treatment for B-cell acute lymphoblastic leukemia (B-ALL). Though patient responses vary across different clinical trials, there currently lacks reliable methods to dissect and predict patient responses to such novel therapies. Collective analysis and effective interpretation of clinical trial data from different clinical studies to reveal the key cellular mechanisms underlying the heterogeneous patient outcomes observed in different clinical trials are needed. Besides, a clinical data-based prognostic model contributing to the prediction of patient responses to CAR T-cell treatment at early stage is largely absent. We herein established a computational model of CAR T-cell therapy to recapitulate key cellular mechanisms and dynamics during treatment based on clinical data from different CAR T-cell trials and predicted responses with input of early-stage CAR T-cell dynamics.

Materials and Methods:: We reproduced the pathophysiological processes and immunological interactions under CAR T-cell therapy with differential equations. Real-time data of CAR T-cell and tumor burden were collected from clinical studies. For clinical trials that did not disclose raw data, the clinical data were digitized by WebPlotDigitizer (https://automeris.io/WebPlotDigitizer) from published figures of these clinical studies. Parameter estimation (on individual and population level) were conducted using the stochastic approximation expectation maximization (SAEM) algorithm for nonlinear mixed-effect modeling (NLME), as implemented in Monolix version 2020R1 (Lixoft, France).

Results, Conclusions, and Discussions:: We framed a matrix of differential equations to establish a computational framework of CAR T-cell therapy to recapitulate key cellular mechanisms and dynamics during treatment with responses of continuous remission (CR), non-response (NR), and CD19-positive (CD19+) and CD19-negative (CD19−) relapse (Figure 1A). We searched and collected clinical data of 209 B-ALL patients with different responses from 10 clinical trials of anti-CD19 CAR T-cell therapies and sampled into 32 groups to calibrate the model. CAR T-cell function factor FT describing the overall CAR T-cell functionality consisting of growth, cytotoxicity and persistence was defined to reveal distinct CAR T-cell patterns in different responses (Figure 1B). Besides, the negative relapse factor FNegR, was introduced to dissect the CD19- relapse pattern integrating the genetic mutation causing loss of surface expression of CD19 and the bystander killing of CAR T-cell (Figure 1C). Combining the peak and AUC7 index of CAR-T cell, we defined the prediction factor FP, to predict the treatment response (Figure 1D-E). Further, we sought to predict patient outcome with input of the first 7-day clinical data of CAR T-cell dynamics and the initial tumor burden of a given patient. We first tested representative patients for different response groups and generated clinical-derived virtual patient cohorts for a clinical trial simulation (Figure 1F-K). Overall, our prediction method reached an accuracy of 74.38% in total of 1,600 virtual patients with respective prediction accuracy of 65.63% for CR&CD19- relapse, 75.50% for CD19+ relapse, and 90.75% for NR (Figure 1L). Next, we chose the initial CD19- tumor burden as an index to differentiate the occurrence of CD19- relapse from CR (Figure 1M). In conclusion, we constructed a model of the leukemia-immune interactions with clinical dataset, determined key factors effecting treatment efficacy, and predicted patient response to CAR T-cell therapy with early-stage CAR T-cell dynamics data. We hope that this platform can aid clinicians with quantitative and reliable prognosis in early-stage of CAR T-cell therapy to customize following treatment and improve the clinical outcomes.

Acknowledgements (Optional): :

References (Optional): : 1. Hay, K. A. et al. Blood. 2019; 133: 1652-1663.
2. Lee, D. W. et al. The Lancet. 2015, 385: 517-528.