Monoclonal antibodies (mAbs) continue to dominate the biopharmaceutical landscape, representing one of the most commercially valuable and clinically impactful classes of therapeutics across oncology, autoimmune disease, and infectious disease indications. A defining feature of therapeutic mAbs is N-linked glycosylation — a post-translational modification that subtly but profoundly influences effector function, half-life, and immunogenicity. Yet achieving consistent glycan profiles through mammalian cell culture bioprocesses remains one of the field’s most intractable challenges. 

In a groundbreaking preprint on arXiv, Ma and colleagues from MIT and Polytechnique Montréal introduce a new Adaptive Nonlinear Model Predictive Control (ANMPC) framework designed to address glycosylation control in fed-batch Chinese hamster ovary (CHO) cell cultures in silico. 

Why Glycosylation Matters

Glycosylation heterogeneity arises from the elaborate enzymatic network of the endoplasmic reticulum and Golgi apparatus, producing diverse glycoforms attached to the Fc region of mAbs. These glycoforms significantly modulate key quality attributes:

• Antibody-dependent cellular cytotoxicity (ADCC) and complement-dependent cytotoxicity (CDC) through Fcγ receptor interactions.

• Pharmacokinetics and immunogenicity via effects on serum half-life and immune recognition. 

• Regulatory classification of glycosylation as a critical quality attribute (CQA) that must remain within strict specifications.

Despite decades of advances in cell line engineering and media optimization, real-time control of glycosylation during production has been limited by measurement constraints and dynamic biological complexity. 

The ANMPC Framework

The core innovation of the study lies in coupling multiscale mechanistic modeling with adaptive control theory:

• Multiscale model: Integrates extracellular nutrient and metabolite dynamics with intracellular nucleotide sugar donor synthesis and Golgi enzymatic reactions to predict glycan distributions. 

• Adaptive control: Model parameters are updated online using incremental measurements, mitigating model–plant mismatch — a persistent problem in traditional open-loop optimization. 

• Shrinking horizon optimization: At every control interval, an optimization problem predicts future process behavior and computes control inputs (e.g., feed rates, nutrient pulses), from which only the immediate action is implemented. 

The authors report that, in silico, ANMPC outperforms both static open-loop optimization and conventional state nonlinear MPC by up to 130% and 96%, respectively, in controlling targeted glycoforms. Their simulations also demonstrate robustness to partial measurement availability and delayed actuation — conditions closer to industrial reality. 

Context in Bioprocess Control

Model predictive control (MPC) has a rich heritage in chemical and process industries for managing multivariable systems with constraints. Its translation to bioprocessing, particularly in mammalian cell culture, has been anticipated for years as part of the quality by design (QbD) and process analytical technology (PAT) paradigms endorsed by regulators. 

Recent industrial research efforts combining digital twins, soft sensors, and data-driven models point toward the same strategic objective: real-time management of critical process parameters and CQAs like glycosylation. 

However, practical implementation is constrained by the cost and latency of glycan analytics — unlike simpler state variables such as pH or dissolved oxygen, glycan profiling traditionally relies on offline assays. Innovative hybrid approaches leveraging soft sensors and inferential models are emerging to fill this gap. 

Implications and Future Directions

The ANMPC framework is a conceptual milestone toward closed-loop glycosylation control. If integrated with next-generation real-time analytical platforms and scalable digital twins, such strategies could:

• Enhance batch-to-batch consistency of glycan profiles to meet regulatory CQAs reliably.

• Reduce development timelines for biosimilars by better matching reference product glycoforms.

• Improve process robustness against biological variability.

Intensive validation and industrial piloting remain crucial next steps.

References & Further Reading

Primary Article

• Adaptive Nonlinear Model Predictive Control of Monoclonal Antibody Glycosylation in CHO Cell Culture – arXiv.org, Oct 2025. Adaptive Nonlinear Model Predictive Control of mAb Glycosylation (arXiv:2510.12333)

Glycosylation and Bioprocess Context

• Edwards E. Strategies to control therapeutic antibody glycosylation patterns. PMC – Biotechnology and Bioengineering, 2022. Strategies to Control Therapeutic Antibody Glycosylation (PMC9310845)

• Luo Y. Model-based control of titer and glycosylation in fed-batch mAb production. AIChE Journal, 2023. Model‑Based Control of mAb Production (AIChE Journal)

• Zhu H, et al. On the achievable consistency of glycan distribution in biomanufacturing of therapeutic mAbs. npj Advanced Manufacturing, Jan 2026. Consistency of Glycan Distribution in mAb Biomanufacturing (npj Advanced Manufacturing)

• Sommeregger W. Towards model predictive control of mammalian cell culture processes. Wiley – Biotechnology Journal, 2017. Towards MPC of Mammalian Cell Culture (Biotechnology Journal)

• Lonza. Understanding antibody glycosylation and its impact. Lonza Knowledge Centre, 2024. Understanding Glycosylation in Antibodies (Lonza)