Modeling, estimation, and control of glycosylation in monoclonal antibodies produced in CHO cells

Date
2016
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Monoclonal antibodies (mAbs) are a class of commercially valuable biopharmaceuticals that are used for treating diseases such as psoriasis, rheumatoid arthritis, and multiple types of cancer. A vast majority of these biotherapeutics are expressed in mammalian cell lines such as Chinese Hamster Ovary (CHO) cells to enable post-translational modifications that generate human-like protein structures. One such post-translational modification that results in structural and pharmacological changes in the protein is N-linked glycosylation, involving the addition and subsequent modification of an oligosaccharide to the protein backbone. The non-template driven, enzymatic modification of the attached oligosaccharide yields a heterogeneous distribution of glycan isoforms, altering the immunogenicity, stability and half-life of the mAb, and hence the final drug product quality. Maintaining the desired product quality of mAbs in the presence of process variations during manufacturing has been difficult for a variety of reasons, including: (i) a lack of quantitative understanding of the effect of input factors on product quality attributes; (ii) the absence of on-line or real-time measurements of quality attributes as these are monitored infrequently or using time-consuming assays; (iii) the lack of effective control strategies that incorporate these infrequent measurements (as and when they become available) to regulate product quality. To ensure product safety and therapeutic efficacy, regulatory agencies are encouraging manufacturers to monitor and control the drug product quality, specifically maintaining the glycan distribution within an acceptable range. The overall goal of this dissertation, therefore, is to develop a rational framework to model the effect of different input factors on the glycosylation profile, estimate the glycan distribution using a dynamic mathematical model supplemented with infrequent measurements, and control the final glycosylation profile in monoclonal antibodies produced in CHO cells. ☐ As the glycosylation profile in mAbs is influenced by several process variables spanning multiple scales—from operating conditions at the bioreactor (macro) scale, to factors at cellular (meso) scale and organelle (micro) scale—we developed an integrated multi-scale model of glycosylation and validated the model predictions using experimental results obtained with an in-house cell line. The model serves as a useful link between nutrient concentrations and cell growth at the macro-scale and the glycosylation profile at the micro-scale. ☐ In parallel, we used a holistic approach that combined factorial design of experiments and a novel computational technique to identify the various combinations of glycan species that are affected by dynamic media supplementation and to quantify mathematically how they are affected. Our experiments demonstrated the importance of taking into consideration the time of addition of trace media supplements, not just their concentrations, and the corresponding mathematical analysis provided insight into what supplements to add, when, and how much, in order to induce specific changes in the glycosylation profile. ☐ We developed a two-step framework to control the glycosylation profile by first generating quantitative input-output relationships using the previously described holistic approach and then designing proportional (P) and proportional integral (PI) controllers based on this quantitative input-output relationship. The set-point tracking performance of these P and PI controllers was evaluated via simulations under nominal conditions (i.e. when the model is assumed to be representative of the actual ‘plant’ or process) and model-plant mismatch conditions. Our results demonstrated that the developed framework can be implemented to design glycosylation controllers to achieve a desired target glycosylation profile under different conditions. ☐ The P and PI controllers that we have developed are suited for batch-to-batch control as they depend on the final glycosylation profile. To achieve real-time control of glycosylation we require real-time information of the glycan distribution obtained from glycan assays; however, current glycan assays are infrequent and characterized by long analysis times. We address this limitation in glycosylation analysis using two approaches: (i) by formulating a rational framework based on observability analysis to guide the development of novel assays that can simplify glycan analysis or reduce analysis time; and (ii) by designing a state estimator to predict the glycan distribution profile in the absence of measurements using the previously developed multi-scale model and updating those predictions as and when measurements become available. ☐ The framework developed in this dissertation will form the basis of an online control scheme to control the final glycosylation profile in the product, thereby achieving consistent product quality.
Description
Keywords
Citation