(683b) Control of Glycosylation and Titer in Fed-Batch Monoclonal Antibody Production

Therapeutic monoclonal antibodies (mAbs) are currently produced commercially in fed-batch cultures using media recipes and protocols based primarily on heuristics. The primary production objectives are maximizing mAb titer (productivity requirement) and achieving desired glycan distributions (quality attribute) that result from glycosylation—an enzymatic, post-translational process that attaches sugar molecules (glycans) to antibodies. The distribution of glycans is a critical quality attribute of mAb manufacturing, because it can either promote or inhibit desirable therapeutic effects of mAb drugs. However, controlling glycosylation is challenging primarily because it is a complex, non-template driven process that takes place within the cell. Consequently, in addition to the apparent chaotic nature of glycan attachment, the variables available for influencing glycosylation are far fewer than the controlled variables. Furthermore, as a consequence of the multifaceted nature of glycosylation, the process is difficult to model, and on-line sensors do not currently exist for characterizing glycan distribution frequently enough for effective on-line control. Nevertheless, recent advances in both glycosylation modeling and glycan assays have improved the possibility of implementing model-based control of glycosylation in place of the current heuristics-based approach.

Here we present an overall model-based strategy for meeting productivity and product quality requirements in the manufacture of mAbs. The proposed multivariate cascade control system consists of two nested control loops. The outer loop, which is responsible for controlling product attributes such as mAb titer, cell density, and the relative abundance of dominant glycans, uses a model predictive controller (MPC) to determine the setpoints for the inner loop. The inner loop in turn employs multiple PI controllers to sustain the cell culture by maintaining the levels of nutrients, pH, temperature, etc., at desired values. Because measurements of glycan distribution are not available frequently enough, the outer loop incorporates a state observer which uses available sensor measurements and model predictions to reconstruct the complete system state that the MPC uses, along with the process model, to determine the desired control policies.

As a first step toward implementing on-line control of industrial fed-batch mAb manufacturing processes, we tested the complete control system in a MATLAB simulation, investigating a variety of industrially relevant process conditions under several operating sensor configurations—frequencies, qualities, and delays. The simulation results offer insight into optimal sensing strategy as well as the usefulness and limitations of the proposed control system.