Because the development of biologics is lengthy and costly, many groups are looking towards data science and mechanistic modeling to expedite process development by identifying sources of variability that may be difficult to measure. Models can include inputs from various measurements taken across different or replicate processes and be used to describe outputs and behaviors. The quality of the models depends significantly on data contributions, mainly the frequency (was the measurement taken every day or hour?), accuracy (how “true” is the data?), and modality (is the data a measure of cell viability, cell productivity, metabolic health, genetic variability, etc.?) of the variables. Once a model is generated, correlating the results to historical and future experiments may be challenging, but data visualization and analytics tools are being developed to assist researchers.

Recently, the Betenbaugh group at Johns Hopkins University provided an unorthodox approach to modeling metabolic differences of Chinese-hamster ovary (CHO) cells and applied these models to historical datasets and new processes of CHO cells grown in shaker flasks and fed different commercial cell media. The researchers focused on the effects of CHO essential amino acid (Arg, Cys, His, Ile, Leu, Lys, Met, Phe, Pro, Thr, Trp, Val) levels on culture growth by probing historical data and testing various media and conditions in new cultures. Reexamining a published study, the new models had predictive error rates for the essential amino acid uptake of approximately 26% when compared to measured values (assuming zero error in the measurements). Using these model-based predictions, the researchers concluded that the cells consumed near the minimum amount of essential amino acids for biomass and protein synthesis. In more perturbed systems, the models underestimated the essential amino acid uptake, which was explained by the idea that the cells would respond to physiological stress by increasing nutrient consumption.

To determine if this modeling approach could be replicated with different nutrient compositions, IgG1 producing CHO-K1 cells were batch cultured in shake flasks with two various commercial media. In addition to amino acid analysis, glucose, lactate, and viable cell densities were measured daily to contribute to the models. Despite missing three essential amino acids (Arg, Cys, Trp) in the analysis due to measurement limitations, there was a close correlation between the experimental uptake rates and predicted uptake rates for the other eight essential amino acids. Even though there were different initial nutrient levels between the two media, there was no correlation to the absolute amino acid levels and the uptake rates. Therefore, by implementing an essential nutrient minimization model, the growth rate could be used to estimate the consumption rates of the essential amino acids.

Additionally, the team used the same modeling method on historical data from three cell lines (CHO-K1, CHO-DG44, and CHO-S) to find that the DG44 cell line had a higher amino acid requirement to sustain growth and flux rates, which differentiated the cell line metabolically from the other cell lines. The team also used Dual Price analysis to assess the relative value of each nutrient towards improved growth from this experimental data. Of note from the Dual Price analysis was that methionine levels correlated to increased uptakes of Asn, Asp, glucose, Gln, and Ser across all three cell lines. Since methionine requires energy and other nutrients to be catabolized, the excess uptake of this amino acid during a culture process would be a metabolic burden on the cells.

The large number of nutrients required by CHO cells and the burdensome monitoring of these nutrients leads researchers to look towards modeling approaches that can improve manufacturing and process control. The utilization of the uptake rate objective functions used in this study suggests that it is a valuable approach to help in the selection of cell lines and clones by considering their nutrient uptakes and viability when grown in different media conditions. Dual Price analysis offered additional insight into how particular nutrient uptake had a growth-limiting effect on several cell lines. Predictions generated by these approaches may help groups monitor bioreactor processes in a more predictive way, and possibly circumvent some of the more difficult tests required during bioprocesses.


By, Glenn A. Harris