Advanced Insights into Autologous Cell Therapy Potency Methods

2023 NCB Non-Clinical Biostatistics Conference Statistical approach for autologous cell therapy potency method comparability June 20, 2023 Sangwook Choi Director, CMC Biostatistics Ifrah Javed Senior Specialist, CMC Biostatistics Mia Teixeira - Manager, CMC Biostatistics Global Biometrics & Data Sciences 1

Introduction to Autologous CAR-T Manufacturing Process C l i n i c C l i n i c M a n u f a c t u r i n g , Q C Te s t i n g a n d R e l e a s e C e n t ra l i z e d M a n u f a c t u r i n g S i t e I n f u s i o n S i t e C o l l e c t i o n S i t e Drug Product Release Testing Identity Safety CAR T Cell Expansion Gene Transfer Selection and Activation Leukapheresis Infusion Purity Potency Ex. IFN , TNF Strength FDA approved February 2021 EMA approved January 2022 FDA approved March 2021 EMA approved August 2021 Chimeric Antigen Receptor (CAR) T Cell Global Biometrics & Data Sciences 2

Unique features of cell therapy analytical methods In contrast to allogenic product manufacturing, where a single product lot can potentially supply hundreds of patients, autologous therapies require a new lot of cells to be made for every patient. Factors contributing to the complex nature of cell therapy manufacturing process: Patient specific manufacturing: Variations in starting cellular material, reagents, processing methods, culture conditions and extensive manufacturing processes may result in a significant variation in the final product. Interferences and matrix effects are uncontrollable. Clinical effectiveness depend on multiple cellular functions. Range of sample results not fully observed during development. Analytical Procedures are complex, time consuming and labor intensive. Most cell therapy assays are novel: Lack of standardized operations, industrial practice or regulatory guidance. What to measure: Sterility, cell number, size, viability, cell growth, protein/gene expression & regulation, cell differentiation, cell apoptosis/necrosis, cell secretion, cytotoxicity, antigen presentation. How to measure: Cell based assays, flow cytometry, biochemical (PCR, ELISA), enzymatic & ligand binding assays, molecular biological tools. Global Biometrics & Data Sciences 3

Challenges associated with Potency testing of cell therapy drug product Generally, very high variability among lots. The reportable value is an absolute potency and not a relative potency due to impracticality of reference standard. Limited quantity of final product to test. Limited time to perform lot release testing. Limited stability of most cellular therapy products. Limited time for rigorous development of potency assays. limited time for rigorous development of for rigorous development of potency assays Limited availability of incurred samples reference standard. Global Biometrics & Data Sciences 4

Applications of Method Comparability Analytical Method Transfer Analytical Method Bridging A documented process that quantifies a laboratory (receiving laboratory) to use a validated analytical test procedure that originated in another laboratory (sending laboratory). A new method is introduced with the purpose of replacing the existing method. Demonstrate suitable performance of the new method relative to the one it is intended to replace. Ensures the receiving laboratory has the procedure knowledge and ability to perform the transferred procedure as intended. Performed when there is a significant change in a method. (version-bridging). USP<1224> Transfer of Analytical Procedures. A new method introduced is not linked to an existing data set generated by an older method that is being discontinued. Global Biometrics & Data Sciences 5

Analytical Method Transfer: Experimental Design Balanced multi-level design Sample Selection Experimental Design: Sample selection must be justified considering assay working range, release specification and clinical experience. Concurrent side by side testing. Assays are executed under intermediate precision condition (multiple operators over multiple days at each site). Multiple DP lots and multiple vials/lot. The intent is to demonstrate the suitability of the lab (including Instrument, Analysis Software, Reagents, etc.), not the people. Need to ensure representativeness, homogeneity, consistency and stability. If possible, the same types of instruments, reagent and analysis software should be used at sending and receiving site. Global Biometrics & Data Sciences 6

Analytical Method Transfer Equivalence Test for Mean Comparison Equivalence Acceptance Criteria is derived from the sending lab s intermediate precision (EAC = k SDIP, SL). Coefficient (k) is typically higher than usual considering the complexity and inherent variability of analytical methods for CAR T cell therapy products, and reproducibility (inter-site variation). Total number of Test Results = Number of distinct lots available Number of replicates/DP lot. Given 5 distinct Drug Product lots (chosen by SMEs), the number of replicates for each lot was determined based on the power analysis. The number of replicates must achieve a minimum of 85% Power assuming several true mean differences. Global Biometrics & Data Sciences 7

Case Study A- Potency Method Transfer Step 1: Visual Assessment o 5 samples cover a wide range o Heteroscedastic characteristic o The potency results were natural log transformed for normality and to ensure consistent variance across the range of antigen specific potency secretion Step 2: Checking Outliers o An assessment was performed for outliers. DP3 result from the Sending Lab, Analyst 2 was excluded from all analyses. o The outlier was further confirmed by checking assay execution and sample condition Global Biometrics & Data Sciences 8

Case Study A - Potency Method Transfer Step 3: Lineal Mixed Model Yijkm = + Si+ Mj + Si Mj + Ak(i)+ Ak(j) Mj + Dm(ik)+ Dm(ik) Mj + eijkm Yijkm :the measurement of Material jth by Analystkth on Day mthatSite ith : overall mean Si : the effect of the Site ith Mj: the effect of the Materialjth Si Mj: the effect of the interaction between Siteithand Material jth Ak(i) :the effect of the Analyst kth Dm(ik) : the effect of the Day mth Dm(ik) Mj: the effect of the interaction between the Day mth and Materialjth eijkm: the residual error Variance component estimates for the linear mixed model 2 2 2= ??(?,?) + ??2 ?? + ??.?(?,?) Global Biometrics & Data Sciences 9

Case Study A - Potency Method Transfer Step 4: Comparison of Means To compare the mean (least square means of the linear mixed model) values obtained at Sending Lab (SL) and Receiving Lab (RL),the acceptance criterion for the mean difference, represented as EAC is set to 1.2xSDIP . Global Biometrics & Data Sciences 10

Case Study A - Potency Method Transfer Step 5: Comparison of Variances Variance Components are estimated using the Restricted Likelihood (REML) procedure. 1Variance and SD are calculated in the natural log scale When standard derivation (SD) is calculated on a log scale, %CV is calculated using anti-log re-transformation as 100 exp SD2 1. Global Biometrics & Data Sciences 11

Analytical Method Bridging Three-pronged approach: Equivalence Test (primary), Deming Regression and Lin s Concordance Correlation (secondary) Statistical Analysis Metric Description Paired two one-sided test (TOST). Comparison of Overall Means Equivalence Test The equivalence limit is determined to be 1.2 SDIP,SL A technique for fitting linear models where both variables, X and Y, are measured with error. Separate Assessment of Proportional and Constant Bias The ratio = Var(X)/Var(Y) is assumed to be constant. Deming Regression Point estimates for Slope and Y-Intercept and two-sided 95% Confidence Intervals are evaluated. Regression equation may be used as a calibration from old method scale to new method. Global Biometrics & Data Sciences 12

Analytical Method Bridging Statistical Analysis Metric Description Bias correction factor measures how far the best-fit line deviates from the 45 degree (X=Y line). The precision around the best fit (Pearson s correlation) is part of the Lin s Concordance Correlation. Linear Correlation considering both precision and bias The concordance correlation estimate and its lower one-sided 95% confidence bound are reported. Lin s Concordance correlation Fisher Transformation to calculate the Standard Error(??): ? = ??? 1( ?2) The sample concordance correlation coefficient (CCC): The lower-one sided 95% confidence bound for Z is ??= ? 1.6449? ? 2??? 2+ ( ? ?)2 ?2= 2+ ?? ?? By inverting the transformation equation, the 95% lower-confidence limit for ?? is ??2 = tanh(??) Global Biometrics & Data Sciences 13

Case Study B - Analytical Method Bridging Potency assay version 1 is being bridged to potency assay version 2. Matched reportable values (N=30 paired data, 15 CD8+ and 15 CD4+ DP Samples)were natural log transformed. Same target cells were used in both methods to minimize potential sources of variability. I. Primary Assessment: Equivalence Test (EAC = 1.2 x SDIP,SL) . Mean Difference 90% Lower CL 90% Upper CL EAC Lower EAC Upper Decision 0.13 0.04 0.21 -0.54 0.54 PASS Global Biometrics & Data Sciences 14

Case Study B - Analytical Method Bridging II. Secondary Assessment: A. Deming Regression Bivariate Fit of v2_fg By v1_fg Bivariate Fit of Ln_v2_fg By Ln_v1_fg Statistical Method Result Deming Regression Equation Converting Potency V1 to V2: ?? ?? = ?? ?? ?.??? + ?.??? Deming Regression (log scale) Slope (95% CI) 1.004 (0.914, 1.095) Y-Intercept (95% CI) 0.108 (-0.278, 0.495) B. Lin s concordance Correlation: Lin s concordance = 0.97 with a 95% one-sided lower confidence interval of 0.95. Global Biometrics & Data Sciences 15

Conclusion o Analytical methods for autologous CAR-T product release pose unique challenges with regards to their life-cycle management, for example, method transfer and bridging. o Statistical approaches for such analytical method comparability studies need to incorporate these unique features. o Our approaches for method transfer and bridging were described using case studies involving a Potency Method. o An equivalence test based on a linear mixed model was used to conduct the method transfer. o For method bridging, we adopted a three-pronged approach that compliments each other. Global Biometrics & Data Sciences 16

References Armstrong, Alison. "Advances in assay technologies for CAR T-cell therapies." BioPharm International 29.2 (2016): 32-+. Burdick, Richard K., et al. Statistical applications for chemistry, manufacturing and controls (CMC) in the pharmaceutical industry. Vol. 10. Cham, Switzerland: Springer, (2017). Clinical and Laboratory Standards Institute. "Measurement procedure comparison and bias estimation using patient samples. Approved guideline." CLSI document EP09-A3 (2013). Kassim, Sadik H. "Toward an integrated model of product characterization for CAR-T cell therapy drug development efforts." Cell Gene Therapy Insights 3.4 (2017): 227-23. Martin, Robert F. "General Deming regression for estimating systematic bias and its confidence interval in method- comparison studies." Clinical chemistry 46.1 (2000): 100-104. McBride, G. B. "A proposal for strength-of-agreement criteria for Lin s concordance correlation coefficient." NIWA client report: HAM2005-062 45 (2005): 307-310. Linnet, Kristian. "Estimation of the linear relationship between the measurements of two methods with proportional errors." Statistics in medicine 9.12 (1990): 1463-1473. Ritter, Nadine, et al. "Bridging Analytical Methods for Release and Stability Testing." BioProcess International 14 (2016): 2. Stroncek, David F., et al. "Potency analysis of cellular therapies: the role of molecular assays." Principles of Translational Science in Medicine. Academic Press, 2021. 49-70. Tsong, Yi, Xiaoyu Dong, and Meiyu Shen. "Development of statistical methods for analytical similarity assessment." Journal of biopharmaceutical statistics 27.2 (2017): 197-205. Wiwi, Christopher. "Analytical considerations for cellular therapy manufacturing. Cell Gene Therapy Insights 2016: 2(6), 651-661. 10.18609/cgti.2016.071 Global Biometrics & Data Sciences 17