Accurate definition of control strategies using cross validated stepwise regression and Monte Carlo simulation.

Drug manufacturing processes must consistently deliver safe and effective product. A key part of achieving this is process validation utilizing Quality by Design (QbD) principles. To meet process validation requirements, process characterization (PC) studies are often performed to expand process understanding and establish an appropriate control strategy that enables the manufacturing process to consistently deliver a target product profile. Two key elements of the control strategy resulting from PC work are a list of critical process parameters (CPPs) and defined operating ranges (ORs). These are frequently derived based on mathematical models describing the relationship between process parameters and critical quality attributes (CQAs). Risk assessment and design of experiments (DOE) techniques are effectively deployed in the industry to identify parameters to study and build process understanding. However, traditional data analysis techniques do not fully utilize the data produced by these studies. In particular, stepwise regression algorithms based on p-values are prone to generate false positives and overfit data, potentially leading to unnecessarily complex control strategies. Many of the deficiencies of traditional stepwise regression can be alleviated by applying cross validation to stepwise regression algorithms, as well as Monte Carlo simulations to estimate model accuracy and predict CQA distributions. These methods can greatly enhance process understanding and assist in the selection of CPPs. A series of PC studies were performed in bioreactors to evaluate a process to produce a recombinant monoclonal antibody. The studies examined process parameters such as dissolved oxygen, pH, temperature, inoculation density, as well as cell density at two key process steps. The resulting data were analyzed using several Monte Carlo based methods. First, cross validation was used to determine model size and select parameters to be included in the model. Next, Monte Carlo cross validation was used to compare the accuracy of different models. Finally, simulated CQA profiles were generated to validate proposed ORs. This workflow provides greater process understanding based on a given PC data set and provides higher statistical confidence in both CPP selection and establishment of a control strategy.

Red colored IgG4 caused by vitamin B12 from cell culture media combined with disulfide reduction at harvest.

While many antibody therapeutics are formulated at low concentration (~10-20 mg/mL) for intravenous administration, high concentration (> 100 mg/mL) formulations may be required for subcutaneous delivery in certain clinical indications. For such high concentration formulations, product color is more apparent due to the higher molecular density across a given path-length. Color is therefore a product quality attribute that must be well-understood and controlled, to demonstrate process consistency and enable clinical trial blinding. Upon concentration of an IgG4 product at the 2000 L manufacturing scale, variability in product color, ranging from yellow to red, was observed. A small-scale experimental model was developed to assess the effect of processing conditions (medium composition and harvest conditions) on final bulk drug substance (BDS) color. The model was used to demonstrate that, for two distinct IgG4 products, red coloration occurred only in the presence of disulfide reduction-mediated antibody dissociation. The red color-causing component was identified as vitamin B 12, in the hydroxocobalamin form, and the extent of red color was correlated with the cobalt (vitamin B 12) concentration in the final pools. The intensity of redness in the final BDS was modulated by changing the concentration of vitamin B 12 in the cell culture media.

Cell culture monitoring via an auto-sampler and an integrated multi-functional off-line analyzer.

Mammalian cell-based bioprocesses are used extensively for production of therapeutic proteins. Off-line monitoring of such cultivations via manual sampling is often labor-intensive and can introduce operator-dependent error into the process. An integrated multi-functional off-line analyzer, the BioProfile FLEX (NOVA Biomedical, Waltham MA) has been developed, which combines the functionality of three off-line analyzers (a cell counter, an osmometer, and a gas/electrolyte & nutrient/metabolite bio-profile analyzer) into one device. In addition, a novel automated sampling system has also been developed that allows the BioProfile FLEX to automatically analyze the culture conditions in as many as ten bioreactors. This is the first report on the development and function of this integrated analyzer and an auto-sampler prototype for monitoring of mammalian cell cultures. Evaluation of the BioProfile FLEX was conducted in two separate laboratories and involved two BioProfile FLEX analyzers and two sets of reference analyzers (Nova BioProfile 400, Beckman-Coulter Vi-Cell AS, and Advanced Instruments Osmometer 3900), 13 CHO cell lines and over 20 operators. In general, BioProfile FLEX measurements were equivalent to those obtained using reference analyzers, and the auto-sampler did not alter the samples it provided to the BioProfile FLEX. These results suggest that the system has the potential to dramatically reduce the manual labor involved in monitoring mammalian cell bioprocesses without altering the quality of the data obtained, and integration with a bioreactor control system will allow feedback control of parameters previously available only for off-line monitoring.