650. Development of a Manufacturing Process Using Monte Carlo Simulations to Support KTE-C19 (Anti-CD19 CAR T Cells) Studies in Leukemia

Abstract

Introduction: ZUMA-3 and -4 (NCT02614066 and NCT02625480) are ongoing phase 1-2 multicenter trials evaluating KTE-C19 in subjects with relapsed/refractory acute lymphoblastic leukemia (r/r ALL). We designed a process suitable for CAR-T manufacturing in the presence of circulating leukemic cells using risk-based Monte Carlo simulations. The result is a robust process capable of generating the target subject dose with high certainty, and minimal operational complexity and variability.Methods: Process design was initiated with the development and classification of unit operations within a sequential model. Rather than defining each process step with a large multivariate array of parameters, probability density functions (PDF) were utilized to represent all potential operational states. Best-fit values for the scale and shape of each PDF, with most units being Weibull and Johnson SU distributions, were computed from empirical data. The values were subsequently utilized to perform a Monte Carlo simulation (n=10,000 iterations) for the generation of a cumulative distribution function for each unit operation, thus characterizing ranges of operational confidence as a function of varying process inputs. The manufacturing success rate for the model was evaluated based on random and independent modifications of key process parameters, and each value was graded based on their relative impact on the process. This method of identifying high-risk conditions throughout the manufacturing design provided a basis for process optimization activities and experimental design. Furthermore, the number of T cells required to initiate the manufacturing process was minimized by performing an inverse mass balance of the simulation. Once optimal ranges were identified for each unit operation, a master protocol was prepared for testing the entire process at full-scale.Results: In preparation for clinical manufacturing, healthy-donor GMP runs (n=8) initiated with no more than 4 × 108 T cells have successfully manufactured the target cell dose of 2 × 106 CAR+ T cells/kg body weight within ≤7 process days. The upfront T cell isolation process from leukapheresis material achieved a target cell recovery with nominal variance (median 54%, range 46%-65%), and %CD3+CD45+CD56− cell enrichment (median 87%, range 78%-94%). Transduction efficiency was measured using a direct identity assay of anti-CD19 surface expression (median 54%, range 38%-66% CAR+ T cells), and stable rates of fold expansion (FE) and doubling time (Td) from day 2 to 6 were achieved (median 9.0 FE, range 6.6-11 FE), (median 27 hours, range 26-35 hours Td). The products contained predominantly naive (median 52%, range 18%-92%) and central memory (median 46%, range 2.6%-64%) T cells, with the remaining population containing effector memory (median 4.0%, range 1.0%-17%) and effector phenotypes (median 0.9%, range 0.3%-5.1%).Conclusion: Using principles of risk-based Monte Carlo simulations, a robust and optimized process has been developed to address the potential challenges of manufacturing an autologous T cell product for leukemic subjects. Performance data generated from healthy-donor material provide a range of confidence for work with patient apheresis material, and this process which was designed to minimize the manufacturing failure rate will be utilized in ZUMA-3 and -4. The process model can be reassessed with data from relevant subject populations to further improve modeling activities.