(567f) Characterizing the Freezing Process of Biopharmaceuticals in Vials at Commercial Scale: COVID-19 Vaccine Case Study and Mechanistic Modeling

The role of freezing in the pharmaceutical industry has recently garnered substantial public interest in the rise of the COVID-19 pandemic. Storage conditions of COVID-19 vaccines have received global news coverage, with low temperature cold chains extremely costly or unavailable depending on the market. Although the stabilizing effect of freezing has been studied extensively in the past, its quantification and prediction remain challenging until today.

The stability of a frozen drug product depends not only on its physicochemical properties, but also on the way it is frozen, i.e. on how it nucleates and solidifies. Ice nucleation, the first step of freezing, is a stochastic process whose effect is of particular importance for small volumes such as the ones used in filled vials. It induces variability in both nucleation and solidification properties among frozen products, with potential implications on stability and processability. Predicting this heterogeneity among frozen products therefore is of great interest for freezing process design and optimization. Unfortunately, no model available in literature is capable of such prediction, likely due to the complexities of the production scale setups used in industry: Typically, tens of thousands of vials are densely packed together in a pallet and frozen slowly in a cold storage room. Such configuration results in enhanced variability due to thermal interaction among the vials, as well as due to spatial differences in heat transfer across the pallet.

To tackle this challenge, we developed a mechanistic modeling framework that computes the freezing behavior of all vials in an arbitrarily sized batch. It takes into account the stochasticity of ice nucleation via a Monte Carlo approach. We compare the simulation findings with experimental data obtained from engineering runs of the Janssen COVID-19 vaccine, for which the model qualitatively reproduced all observed trends. To promote the use and further development of the model in the wider community, we provide open source access to it in the form of a python package (available at https://pypi.org/project/ethz-snow/ ).

Acknowledgement

The authors thank the Janssen Pharmaceutical Companies of Johnson & Johnson for the support in the course of the project and the financial funding. We especially thank Juan Carlos Araque and Ryan Wall.