(M1430-09-56) A Model-Based and Quality-by-Design Compatible Lyophilization Cycle Development Strategy for Biopharmaceuticals

Purpose: Lyophilization cycle development, emerges from a trial-and-error based approach, towards a holistic Quality-by-Design compatible strategy, integrating computer-aided modelling to generate better process understanding and a process parameters design space. Our purpose was to develop a modeling-based approach to design robust lyophilization protocols, that in addition achieves more aggressive drying conditions by optimizing a two-stage shelf temperature regime during primary drying.

Methods: The mechanistic model of the primary drying phase of a lyophilization process was constructed in Python v3.7.0 (Beaverton, OR, USA). The model is a heat-mass balance that consists of a set of mathematical equations that were solved simultaneously [1,2]. For the input parameters, a heat transfer coefficient (K_v) library for various glass type I vials and a pilot-scale freeze-dryer (M.Christ E2-12D) was built up by using a gravimetric approach in the chamber pressure range from 5-25 Pa. For the poster we will focus on data from 6R glass type I vials. A software tool was developed to process the generated data from freeze-drying cycles and perform the predictions. Two formulations with different collapse temperature (Tc) (‘high’ Tc formulation consisted of 2 mM histidine, 10 mg/ml lysozyme, 5% (w/v) 2-hydroxypropyl-β-cyclodextrin and 0.03% (w/v) polysorbate 80 at a pH of 6.5 and ‘low’ Tc formulation used for the conservative processes contained 2 mM histidine, 1 mg/ml lysozyme, 10% (w/v) sucrose and 0.03% (w/v) polysorbate 80 at a pH of 6.5) were designed by adapting the protein concentration and concentration/type of lyoprotectant to challenge the process optimization approach under conservative and aggressive lyophilization conditions. During the first step for the model-based lyophilization cycle development, a dried layer resistance (Rp) profile was extracted from data of a conservative cycle for both formulations. Next, an interactive design space was constructed by applying the K_v library and the extracted Rp values. A robust protocol from the design space was verified by applying an uncertainty analysis of the classical 1-D heat-mass balance describing the primary drying. Variabilities of all critical input parameters were carefully characterized in advance, and applied in the uncertainty analysis of a two-stage primary drying protocol to estimate the risk of failure. Only 0.1% of the samples were allowed to have a product temperature surpassing their collapse temperature, to yield a robust protocol. Finally, process verification has been performed, by running two lyophilization cycles based on the parameters predicted by the model, and characterization of the subsequent lyophilized products (residual moisture and cake appearance).

Results: Freeze-drying microscopy resulted in a Tc of -12 °C for the ‘high’ Tc formulation and of -33 °C for the ‘low’ Tc formulation. The Kv and its variability across the whole shelf was dependent on the chamber pressure setting, as shown for 6R glass type I vials in Figure 1. The average Kv value increased while the Kv variability, expressed as the relative standard deviation, decreased with higher chamber pressures. This is most likely due to a lower importance of radiation and a higher contribution of gas conduction to the total heat transfer at higher pressure. The regressed Rp parameters were formulation dependent but showed a similar trend with a tapering off towards higher dried product layer lengths (Figure 2). These observations are consistent with the current freeze-drying literature. The lyophilization processes based on the generated model were shortened with by a factor of 16% in comparison with a traditional one-stage primary drying phase. Moreover, the product temperature data of the lyophilization cycles that has been predicted by the generated model were located within the 99.9% prediction intervals. The protocols derived from modelling were experimentally verified for both the ‘high’ and ‘low’ Tc formulation (Figure 3). Both, product temperature sensors and comparative pressure measurement indicated the end of primary drying close to the predicted drying time for both the ‘high’ and the ‘low’ Tc formulation. The resulting lyophilization processes resulted in lyophilized products with suitable cake appearance, and residual moisture values of 0.35 ± 0.05% and 1.70 ± 0.29% for the ‘high’ and ‘low’ Tc formulation, respectively.

Conclusion: Integration of modelling into lyophilization process, can help to generate a better process understanding and provide justifications for regulatory filings how critical process parameters, such as shelf temperature and chamber pressure, were established.

References: [1] Vanbillemont, Brecht, Niels Nicolaï, Laurens Leys, and Thomas De Beer Model-Based Optimisation and Control Strategy for the Primary Drying Phase of a Lyophilisation Process. Pharmaceutics 12.2 (2020): 181
[2] Vanbillemont, Brecht, et al. "A model-based optimization strategy to achieve fast and robust freeze-drying cycles." International Journal of Pharmaceutics: X 5 (2023): 100180.

Acknowledgements: The authors thank Anna-Lena Greiner and Vanessa Ehrl for the experimental support.

Figure 1: Graphical respresentation of the Kv experimental data and regressions for 6R glass type I vials. A: Experimental Kv distribution per pressure level is plotted as a boxplots with empty dots as outliers. The red line represents the weighted non-linear regression of the average. B: RSDkv at each pressure level is plotted with a marking. The red line represents the linear regression. C: Empirical distribution of normalized Kv data.

Figure 2: Graphical respresentation of the Rp extracted from conservative lyophilization cycles. Experimental data is represented in boxplots with empty dots as outliers, blue line plot represents the Rp regression which is enlarged by ±2 σRp using a blue shaded area. A: ‘high’ Tc formulation and B: ‘low’ Tc formulation.

Figure 3: Graphical overview of the verification experiments with process data overlayed with the predictions (blue area) for the ‘high’ (A) and the ‘low’ (B) Tc formulation. Cond represents the condenser temperature.