(563d) Towards Quality Control of Biotherapeutic Products through Soft Sensing of Intracellular States

The Quality by Design (QbD) paradigm supported by the US Food and Drug Administration (US FDA) offers a proactive approach to pharmaceutical process development, providing a strong incentive for industry to undergo a digital transformation (Narayanan et al., 2020). To achieve product quality and productivity goals during manufacturing, accurate dynamic information on the cell culture is essential to meet the demand of critical quality attributes (CQA). Glycosylation is considered as one of the most important CQA and is linked to the cell metabolism through an intracellular state of the cell, namely the nucleotide sugar donors (NSD). As NSDs are synthesised metabolically and act as co-substrates for the glycosylation process, they offer invaluable insights in designing the feeding strategy and predicting the product quality profile (Kotidis et al., 2019). However, compared to extracellular states such as cell density, titre and metabolites, intracellular NSD concentrations are challenging to obtain experimentally, therefore it is not a common industry practice to measure them.

To better understand the interaction between metabolism and the intracellular glycosylation process, several predictive model frameworks have been published to link extracellular metabolites with intracellular states through cell metabolism and, specifically, NSD synthesis pathways. This is followed by the glycosylation model that predicts the glycan distribution using NSD as inputs (Jimenez del Val et al., 2011, Jedrzejewski et al., 2014, Villiger et al., 2016, Sou et al., 2017, Kotidis et al., 2019). Whilst these kinetic models are fairly accurate, they are highly specific to the system of study, complex and require new experimental measurements to reparametrize the model for adaptation to new experimental conditions, cell line or product.

As an alternative, data driven approaches and hybrid models have been applied to mitigate the challenge of incomplete system understanding and model specificity (Tsopanoglou and Jiménez del Val, 2021). Herein, we propose a framework consisting of two stages. The first part employs a data assimilation technique, Ensemble Kalman Filter (EnKF), to estimate the NSD concentrations without experimentally measuring them. The EnKF integrates the mechanistic backbone of the physical model and the hard sensor observations of the measurable states, enabling unmeasurable intracellular states to be estimated through a model inference. Our results show that the EnKF outperforms the predictive mechanistic model and has a corrective action on both the extracellular and intracellular states. The estimated NSD concentrations are then used as inputs for training a Recurrent Neural Network (RNN) to predict the glycosylation profile of the recombinant protein product. Compared to a ‘static’ feed-forward neural network that has previously been published (Kotidis and Kontoravdi, 2020), the RNN takes into account the sequential process of glycosylation and successfully predicts the dynamic variation of the glycan profile based on historical information. The framework combining the EnKF and the RNN supports decision making on designing the feeding strategy of the cell culture, demonstrating great potential towards online monitoring and control of key product quality attributes such as glycan distribution.

References:

JEDRZEJEWSKI, P. M., DEL VAL, I. J., CONSTANTINOU, A., DELL, A., HASLAM, S. M., POLIZZI, K. M. & KONTORAVDI, C. 2014. Towards controlling the glycoform: a model framework linking extracellular metabolites to antibody glycosylation. Int J Mol Sci, 15, 4492-522.

JIMENEZ DEL VAL, I., NAGY, J. M. & KONTORAVDI, C. 2011. A dynamic mathematical model for monoclonal antibody N-linked glycosylation and nucleotide sugar donor transport within a maturing Golgi apparatus. Biotechnol Prog, 27, 1730-43.

KOTIDIS, P., JEDRZEJEWSKI, P., SOU, S. N., SELLICK, C., POLIZZI, K., DEL VAL, I. J. & KONTORAVDI, C. 2019. Model-based optimization of antibody galactosylation in CHO cell culture. Biotechnol Bioeng, 116, 1612-1626.

KOTIDIS, P. & KONTORAVDI, C. 2020. Harnessing the potential of artificial neural networks for predicting protein glycosylation. Metab Eng Commun, 10, e00131.

NARAYANAN, H., LUNA, M. F., VON STOSCH, M., CRUZ BOURNAZOU, M. N., POLOTTI, G., MORBIDELLI, M., BUTTE, A. & SOKOLOV, M. 2020. Bioprocessing in the Digital Age: The Role of Process Models. Biotechnol J, 15, e1900172.

SOU, S. N., JEDRZEJEWSKI, P. M., LEE, K., SELLICK, C., POLIZZI, K. M. & KONTORAVDI, C. 2017. Model-based investigation of intracellular processes determining antibody Fc-glycosylation under mild hypothermia. Biotechnol Bioeng, 114, 1570-1582.

TSOPANOGLOU, A. & JIMÉNEZ DEL VAL, I. 2021. Moving towards an era of hybrid modelling: advantages and challenges of coupling mechanistic and data-driven models for upstream pharmaceutical bioprocesses. Current Opinion in Chemical Engineering, 32.

VILLIGER, T. K., SCIBONA, E., STETTLER, M., BROLY, H., MORBIDELLI, M. & SOOS, M. 2016. Controlling the time evolution of mAb N-linked glycosylation - Part II: Model-based predictions. Biotechnology Progress, 32, 1135-1148.