A quality-by-design approach to improve process understanding and optimise the production and quality of CAR-T cells in automated stirred-tank bioreactors

Abstract

Ex vivo genetically-modified cellular immunotherapies, such as chimeric antigen receptor T cell (CAR-T) therapies, have generated significant clinical and commercial outcomes due to their unparalleled response rates against relapsed and refractory blood cancers. However, the development and scalable manufacture of these novel therapies remains challenging and further process understanding and optimisation is required to improve product quality and yield. In this study, we employ a quality-by-design (QbD) approach to systematically investigate the impact of critical process parameters (CPPs) during the expansion step on the critical quality attributes (CQAs) of CAR-T cells. Utilising the design of experiments (DOE) methodology, we investigated the impact of multiple CPPs, such as number of activations, culture seeding density, seed train time, and IL-2 concentration, on CAR-T CQAs including, cell yield, viability, metabolism, immunophenotype, T cell differentiation, exhaustion and CAR expression. Initial studies undertaken in G-Rex^® 24 multi-well plates demonstrated that the combination of a single activation step and a shorter, 3-day, seed train resulted in significant CAR-T yield and quality improvements, specifically a 3-fold increase in cell yield, a 30% reduction in exhaustion marker expression and more efficient metabolism when compared to a process involving 2 activation steps and a 7-day seed train. Similar findings were observed when the CPPs identified in the G-Rex^® multi-well plates studies were translated to a larger-scale automated, controlled stirred-tank bioreactor (Ambr^® 250 High Throughput) process. The single activation step and reduced seed train time resulted in a similar, significant improvement in CAR-T CQAs including cell yield, quality and metabolism in the Ambr^® 250 High Throughput bioreactor, thereby validating the findings of the small-scale studies and resulting in significant process understanding and improvements. This study provides a methodology for the systematic investigation of CAR-T CPPs and the findings demonstrate the scope and impact of enhanced process understanding for improved CAR-T production.

Keywords: CAR-T; T cells; immunotherapy; process optimisation; process understanding; quality-by-design; stirred-tank bioreactor.

Figure 1

Experimental CAR-T expansion G-Rex^® DOE overview. Cells thawed from one donor were activated a day later using Dynabeads^® and the experimental IL-2 concentration (30, 65 or 100 IU/mL). The following day, all conditions were transduced with anti-CD19 CAR lentivirus and pre-expanded for the experimental seed train lengths (3, 5 or 7 days). Pre-expanded cells were then seeded into G-Rex^® 24-well plates at varied seeding densities (0.25, 0.5 or 0.75 x 10⁶ cells/mL) and expanded for 7 days with monitored cell growth and quality. For conditions with a second activation step, Dynabeads^® were added before G-Rex^® seeding.

Figure 2

Repeated activation, higher seeding density, and longer seed train time negatively impact CAR-T cell yield. Regression coefficients included in the modelled response for population doublings (PD) on day 7 of G-Rex culture (A). Modelled factor-interaction between seed-train time and number of activations (B). Modelled effects of seeding density and seed-train time on PD, assuming 1 activation and 30 IU/mL of IL2 (C). Effect of number of activations on cell fold (D), and PD (E). Error bars represent standard deviation. Statistical pooled t-test analyses completed where ** is p<0.01, *** is p<0.001.

Figure 3

Cell quality characteristics impacted by repeated activation and longer seed train time. The effects of the experimental factors on CD3+CD69+ and CD3+PD1+LAG3+ expression on day 7 of G-Rex^® culture were modelled: significant regression coefficients plot (A, E), factor-factor interaction plot (B, F), and contour plot of modelled response (C, G). The effect of number of activations on mean CD3+CD69+ and CD3+PD1+LAG3+ expression (D, H). Statistical pooled t-test analyses completed where ** is p<0.01, *** is p<0.001.

Figure 4

Effect of activations on the average glucose consumption (A), lactate production (B), and dlac/dgluc (C). Error bars represent standard deviation. Statistical pooled t-test analyses completed where *** is p<0.001. Dashed lines represent the confidence interval of the model. All model graphs shown represent the model when IL-2 = 30 IU/mL and seeding density = 2.5x10⁶ cells/cm².

Figure 5

Effects of process parameters translated to flasks and multiple donors. Cell fold in flask culture by day (A), and on day 7 (B). T-cell differentiation (C), exhaustion (D), and activation (E) on day 7 of flask culture. Average glucose consumption (F), lactate production (G), and change in lactate divided by change in glucose (dLac/dGluc) (H). Day 0 represents the day flasks were seeded post seed train (ST). Points represent each donor (square = HD1, triangle = HD2, circle = HD3). Error bars represent standard deviation. Statistical pooled t-test analyses completed where * is p<0.05, ** is p<0.01, *** is p<0.001.

Figure 6

CAR-T cells effectively kill target cells and secrete cytokines post-expansion. Following expansion in flasks, cells were co-cultured with target GFP-positive NALM6 cells. Relative number of NALM6 cells during the 2-day co-culture (A). Final relative number of NALM6 cells (B) and representative images of each culture type (20X magnification) (C) at the end of the 2-day co-culture. Cytokine levels in the medium for IFN-γ (D) and TNF-α (E) at the end of the 2-day killing assay co-culture. Each condition completed in quadruplicate (n=4). Points represent each donor (square = HD1, triangle = HD2, circle = HD3). Error bars represent standard deviation. One-way ANOVA and Dunn’s test completed where * is p<0.05, ** is p<0.01, *** is p<0.001.

Figure 7

Optimised process increased cell expansion in the stirred-tank bioreactor (STR). pH (A) and dissolved oxygen (DO) (B) over time in the STR cultures. Dashed lines represent HD1, dotted lines represent HD2, and solid lines represent HD3. Horizontal black dashed lines represent the parameter setpoints. Cell fold by day in the STR (C). Cell fold (D), cell density (E), T-cell exhaustion (F) and activation (G) on day 7 of STR culture. Average glucose consumption (H), lactate production (I), and change in lactate divided by change in glucose (dLac/dGluc) (J). Day 0 represents the day STRs were seeded post seed train (ST). Points represent each donor (square = HD1, triangle = HD2, circle = HD3). Error bars represent standard deviation. Statistical pooled t-test analyses completed where * is p<0.05, *** is p<0.001.