(458a) Glycopy: A Multiscale Model-Based Simulation, Optimization, and Optimal Control Python Package for Monoclonal Antibody Glycosylation in Cell Culture

Glycosylation – the controlled enzymatic modification of an organic molecule by addition of a sugar molecule – plays an important role in the clinical efficacy of monoclonal antibodies (mAb), the dominant products in the biopharmaceutical market [1]. As such, the glycosylation pattern is a critical quality attribute (CQA) that should be controlled at specific range to ensure product quality [2]. It is desirable to develop mathematical models to predict glycosylation profiles [3], which would then be used for analysis, design, and control as per the principles of quality-by-design (QbD) [4].

Several glycosylation models have been proposed in the literature [5–7]. The potential value of using such models in optimizing glycosylation has been demonstrated in a recent study [7], in which a model was used to increase the concentration of a galactosylated mAb by 93%. The objectives of our work are to further advance the optimization and control of glycosylation by (1) developing a multi-scale model for glycosylation that produces more accurate predictions, and (2) providing an open-source, flexible, and extendable package in Python [8] for the modeling, optimization, and optimal control of glycosylation, which we denote as GlycoPy.

Our multiscale model [7] consists of a cell culture model to predict mAb production rate, a nucleotide sugar donor (NSD) synthesis model to predict intracellular NSD concentrations, and golgi glycosylation (both N-linked and O-linked, steady-state and dynamic, simplified and original) models to predict the glycan profiles. The models are organized modularly so that the users can adapt and extend for their specific cells/cell lines conveniently and can include new mechanisms/submodels easily. Specifically, a reaction network module was defined to manage reaction models. Enzymes and transport proteins commonly used in different reactions are also submodels that can be defined independently. The package was built upon an automatic differentiation framework, so accurate derivatives can be generated efficiently, facilitating efficient and reliable model-based optimization. Since all the model variables and equations are symbols that are finally sent to integrators written in C language for dynamic simulation, the users do not need to worry about the inefficiency of Python language for solution. Based on the modeling platform, more functionalities including parameter estimation, dynamic optimization, model predictive control (MPC) (linear, nonlinear, and adaptive formulations), and optimal design of experiments (including batch and receding horizon methods) are developed with both single- and multiple-shooting algorithms implemented for solution.

The production of mAb from Chinese hamster ovary (CHO) cells was used as an example to demonstrate the functionalities of the package. The parameter estimation in this package uses automatic differentiation to increase the robustness and increase the speed by 1 to 2 orders of magnitude. Through applying the receding horizon method to design experiments online, we can estimate the model parameters (around 100 parameters) much more accurately than the best of the existing literature reports [e.g., Ref. 7]. Finally, adaptive MPC is shown to increase the mAb productivity with glycan profiles controlled within a desired range and outperform nonadaptive predictive controllers. The results demonstrate the efficiency, versatility, flexibility, and extendability of GlycoPy as an open-source package for modeling, optimization, optimal design of experiments, and optimal control of mAb glycosylation.

(1) Walsh, G. Biopharmaceutical Benchmarks 2018. Nat. Biotechnol. 2018, 36 (12), 1136–1145. https://doi.org/10.1038/nbt.4305.

(2) Batra, J.; Rathore, A. S. Glycosylation of Monoclonal Antibody Products: Current Status and Future Prospects. Biotechnol. Prog. 2016, 32 (5), 1091–1102. https://doi.org/10.1002/btpr.2366.

(3) Majewska, N. I.; Tejada, M. L.; Betenbaugh, M. J.; Agarwal, N. N-Glycosylation of IgG and IgG-Like Recombinant Therapeutic Proteins: Why Is It Important and How Can We Control It? Annu. Rev. Chem. Biomol. Eng. 2020, 11 (1), 311–338. https://doi.org/10.1146/annurev-chembioeng-102419-010001.

(4) Rathore, A. S.; Winkle, H. Quality by Design for Biopharmaceuticals. Nat. Biotechnol. 2009, 27 (1), 26–34. https://doi.org/10.1038/nbt0109-26.

(5) Shelikoff, M.; Sinskey, A. J.; Stephanopoulos, G. A Modeling Framework for the Study of Protein Glycosylation. Biotechnol. Bioeng. 1996, 50 (1), 73–90. https://doi.org/10.1002/(SICI)1097-0290(19960405)50:1<73::AID-BIT9>3.0.CO;2-Z.

(6) Ehsani, A.; Niedenfuehr, S.; Eissing, T.; Behnken, S.; Schuppert, A. How to Use Mechanistic Metabolic Modeling to Ensure High Quality Glycoprotein Production. In Computer Aided Chemical Engineering; Espuña, A., Graells, M., Puigjaner, L., Eds.; Elsevier, 2017; Vol. 40, pp 2839–2844. https://doi.org/10.1016/B978-0-444-63965-3.50475-X.

(7) Kotidis, P.; Jedrzejewski, P.; Sou, S. N.; Sellick, C.; Polizzi, K.; del Val, I. J.; Kontoravdi, C. Model-Based Optimization of Antibody Galactosylation in CHO Cell Culture. Biotechnol. Bioeng. 2019, 116 (7), 1612–1626. https://doi.org/10.1002/bit.26960.

(8) Van Rossum, G.; Drake, F. L. Python 3 Reference Manual; CreateSpace: Scotts Valley, CA, 2009.