(322c) Real Time Dynamic Optimisation for Enhanced Product Quality in Bioprocesses

Biopharmaceutical processes are prone to many challenges including highly nonlinear dynamics, batch-to-batch variability and high sensitivity to changes in operating parameters. Such complexities can introduce heterogeneity in biopharmaceutical products. At the same time, consistent product quality remains a top priority for this industry to ensure patient safety. Antibody-producing platforms consist of two main operations: upstream (USP) and downstream (DSP) production. In USP cells are inoculated into a series of progressively larger bioreactors for the growth of mammalian cells, typically Chinese hamster ovary (CHO) cells in the case of monoclonal antibodies (mAbs). These are then transferred to the production bioreactor for protein expression. At the end of this step DSP begins with a series of filtration and purification stages to remove impurities and other unwanted components from the final drug formulation.

The performance of such processes is assessed both for productivity and product quality. The latter is a broad concept that includes various aspects, amongst which product functionality. In mAb production, glycosylation percentage is one of the key indicators to measure cell culture quality performance. Glycosylation is a post-translational modification of mAbs that affects their stability and efficacy as therapeutics. As such, it is considered a Critical Quality Attribute (CQAs) under the Quality by Design (QbD) framework and needs to be controlled within a pre-specified range [1]. There have been various studies on the improvement of mAb glycosylation intervening both at process- [2], [3] and genetic- level [4], [5]. However, the complexity of the system and its underlying interactions make process optimisation difficult, with optimality unlikely to be achieved solely by experimental studies. This is because strategies to improve mAb glycosylation can enhance mAb glycosylation but, at the same time, reduce cell growth and, thus, mAb concentration at harvest.

In this work we use an Ordinary Differential Equation (ODE) model as digital-twin for an IgG-producing CHO cell culture system [6]. The model [7] comprises three main components; namely an unstructured model to describe cell growth, a reduced kinetics model for the intracellular nucleotide sugar donor (NSD) synthesis [8]and a mAb glycosylation model. The mAb glycosylation model as considered in this work, comprises four continuous stirred tank reactors (CSTRs) using the reaction network presented in del Val et al.[9]. We formulate and solve an exhaustive dynamic optimisation experiment to identify feeding regimes for galactose and uridine under which the amount of galactosylated antibody is maximised. The performance of every optimisation formulation is assessed based on: (a) achieved galactosylation percentage, (b) productivity, (c) culture viability and (d) experimental applicability. Towards an effort to enhance USP automation, measured process disturbances are introduced in the problem formulation. In this respect, the most successful scenarios are used as a basis for the design of a real time dynamic optimisation strategy as a means to achieve the optimal operational policy [10]. The system is re-optimised online and based on the effect of the disturbances to the process, adapts and calculates the optimal input trajectories. In this work these are validated in closed-loop against the process model.

References

[1] A. Eon-Duval, H. Broly, and R. Gleixner, “Quality attributes of recombinant therapeutic proteins: An assessment of impact on safety and efficacy as part of a quality by design development approach,” Biotechnol. Prog., vol. 28, no. 3, pp. 608–622, May 2012.

[2] R. K. Grainger and D. C. James, “CHO cell line specific prediction and control of recombinant monoclonal antibody N -glycosylation,” Biotechnol. Bioeng., vol. 110, no. 11, pp. 2970–2983, Nov. 2013.

[3] M. J. Gramer et al., “Modulation of antibody galactosylation through feeding of uridine, manganese chloride, and galactose,” Biotechnol. Bioeng., vol. 108, no. 7, pp. 1591–1602, Jul. 2011.

[4] C. Chung, Q. Wang, S. Yang, B. Yin, H. Zhang, and M. Betenbaugh, “Integrated Genome and Protein Editing Swaps α -2,6 Sialylation for α -2,3 Sialic Acid on Recombinant Antibodies from CHO,” Biotechnol. J., vol. 12, no. 2, p. 1600502, Feb. 2017.

[5] N. Yamane-Ohnuki et al., “Establishment ofFUT8 knockout Chinese hamster ovary cells: An ideal host cell line for producing completely defucosylated antibodies with enhanced antibody-dependent cellular cytotoxicity,” Biotechnol. Bioeng., vol. 87, no. 5, pp. 614–622, Sep. 2004.

[6] P. Kotidis et al., “Constrained global sensitivity analysis for bioprocess design space identification,” Comput. Chem. Eng., vol. 125, pp. 558–568, Jun. 2019.

[7] P. Kotidis et al., “Model‐based optimization of antibody galactosylation in CHO cell culture,” Biotechnol. Bioeng., vol. 116, no. 7, pp. 1612–1626, Jul. 2019.

[8] S. N. Sou, P. M. Jedrzejewski, K. Lee, C. Sellick, K. M. Polizzi, and C. Kontoravdi, “Model-based investigation of intracellular processes determining antibody Fc-glycosylation under mild hypothermia,” Biotechnol. Bioeng., vol. 114, no. 7, pp. 1570–1582, Jul. 2017.

[9] I. Jimenez del Val, J. M. Nagy, and C. Kontoravdi, “A dynamic mathematical model for monoclonal antibody N-linked glycosylation and nucleotide sugar donor transport within a maturing Golgi apparatus,” Biotechnol. Prog., vol. 27, no. 6, pp. 1730–1743, Nov. 2011.

[10] J. V. Kadam, M. Schlegel, B. Srinivasan, D. Bonvin, and W. Marquardt, “Dynamic optimization in the presence of uncertainty: From off-line nominal solution to measurement-based implementation,” J. Process Control, vol. 17, no. 5 SPEC. ISS., pp. 389–398, Jun. 2007.