(30e) Use of Statistical Modeling Tools to Characterize the Design Space for a Late-Stage Mammalian Cell Culture Process

We describe here the design space characterization for the production of a recombinant human protein from a mammalian cell culture process. These activities included (1) the establishment of a 3L scale-down model representative of the 2000 L manufacturing process, (2) a fractional factorial parameter screening study, (3) a response-surface modeling study, which examined pH, seed density, harvest day, and temperature and their impact on growth, productivity, and product quality attributes, and (4) model verification. We found that several response parameters were well modeled and showed a high degree of predictability, while others had poor-fitting models. We made a number of observations based on these individual models, including: (1) that temperature set point was optimized for titer; (2) that decreases in pH set point or extension of harvest day led to decreased levels of Glycan Attribute 1; and (3) that decreases in temperature or harvest day led to increased levels of Glycan Attribute 2. The various mathematical models that were generated from this work allowed for the identification of process optima, the setting of action limits for the Process Validation campaign, and the identification of critical, key, and non-key parameters.  Lessons learned from these studies will be reviewed and recommendations made to streamline the workflow for cell culture process characterization and optimization using multivariate, factorial design of experiments.