(498d) Modeling CHO Cell Glycosylation Process Using Dynamic Kriging

Monoclonal Antibodies (mAbs) are highly demanded therapeutic products that are used for the treatment of cancers, autoimmune diseases, infectious and microbial diseases[1]. It has been reported that mAb amounts for almost 37% of biologic drugs market[2]. For most mAbs, glycosylation process is critical to protein folding, secretion, signal induction, and it impacts the product quality and efficacy [3]. To understand and improve the glycoform distribution, first-principle models have been built to quantify the effects of operating conditions, including process feed-variables, temperatures, pH to protein production, and glycosylation[4-6]. First-principle models capture the complex biochemical pathways of cell metabolism and protein glycosylation and correspond to nonlinear dynamic set of equations that are usually computational expensive. To reduce the computational complexity, a dynamic surrogate model can be used to capture the input and output relations describing the bioreactor operation. Kriging is a nonparametric modeling technique that predicts the target point based on the weighted sum of the observed function values at sampling points around the predicted point [7, 8]. Among different data-driven model techniques, kriging in most cases, requires a limited number of samples while achieving better efficiency and accuracy. It also provides error estimation for its predicted value[8, 9].

In this study, we build a dynamic kriging model to replace the first-principle model, which is used to capture the dynamic behavior of cell growth and simulate protein glycosylation with high computational efficiency. First, an unstructured kinetic model is built to capture the CHO cell culture process in fed-batch bioreactor and predict viable cell, glucose, lactate, and protein concentration. This model is coupled to a structured single-cell glycosylation model to determine the secreted glycoprotein fractions. Then a series of simulations are performed under different operating conditions based on a full factorial design, which provides inputs and outputs under all the time points to build the dynamic kriging surrogate model. By varying pH and the concentration of media supplements (MnCl₂), the dynamic kriging model is able to capture time dependent responds of the system, including viable cell density, glucose, lactate concentrations and glycan fractions by solving the kriging iteratively at each time step [8, 10]. This model is used to find the optimal operating conditions for specific product specifications and integrate to a bioreactor model to achieve on-line product prediction and control.

Reference

1. Newswire, P.R., Monoclonal Antibodies (mAbs) Market Analysis by Source (Chimeric, Murine, Humanized, Human), by Type of Production, by Indication (Cancer, Autoimmune, Inflammatory, Infectious, Microbial, Viral Diseases), by End-use (Hospitals, Research, Academic Institutes, Clinics, Diagnostic Laboratories) and Segment Forecasts, 2018-2024. 2017: Market Research Report No. GVR-1-68038-280-8.
2. Mullin, R., Contract services for biologics evolve. Chemical & Engineering News, 2017. 95(33): p. 34-35.
3. Bieberich, E., Synthesis, Processing, and Function of N-glycans in N-glycoproteins. Advances in neurobiology, 2014. 9: p. 47-70.
4. Sou, S.N., et al., Model-based investigation of intracellular processes determining antibody Fc-glycosylation under mild hypothermia. Biotechnol Bioeng, 2017. 114(7): p. 1570-1582.
5. Kotidis, P., et al., Model-based optimization of antibody galactosylation in CHO cell culture. 2019. 116(7): p. 1612-1626.
6. Villiger, T.K., et al., Controlling the time evolution of mAb N-linked glycosylation - Part II: Model-based predictions. 2016. 32(5): p. 1135-1148.
7. Bhosekar, A. and M. Ierapetritou, Advances in surrogate based modeling, feasibility analysis, and optimization: A review. Computers & Chemical Engineering, 2018. 108: p. 250-267.
8. Boukouvala, F., F.J. Muzzio, and M.G. Ierapetritou, Dynamic Data-Driven Modeling of Pharmaceutical Processes. Industrial & Engineering Chemistry Research, 2011. 50(11): p. 6743-6754.
9. Lucifredi, A., C. Mazzieri, and M. Rossi, APPLICATION OF MULTIREGRESSIVE LINEAR MODELS, DYNAMIC KRIGING MODELS AND NEURAL NETWORK MODELS TO PREDICTIVE MAINTENANCE OF HYDROELECTRIC POWER SYSTEMS. Mechanical Systems and Signal Processing, 2000. 14(3): p. 471-494.
10. Shokry, A. and A. Espuña, Sequential Dynamic Optimization of Complex Nonlinear Processes based on Kriging Surrogate Models. Procedia Technology, 2014. 15: p. 376-387.