Machine learning-based model predictive controller design for cell culture processes

The biopharmaceutical industry continuously seeks to optimize the critical quality attributes to maintain the reliability and cost-effectiveness of its products. Such optimization demands a scalable and optimal control strategy to meet the process constraints and objectives. This work uses a model predictive controller (MPC) to compute an optimal feeding strategy leading to maximized cell growth and metabolite production in fed-batch cell culture processes. The lack of high-fidelity physics-based models and the high complexity of cell culture processes motivated us to use machine learning algorithms in the forecast model to aid our development. We took advantage of linear regression, the Gaussian process and neural network models in the MPC design to maximize the daily protein production for each batch. The control scheme of the cell culture process solves an optimization problem while maintaining all metabolites and cell culture process variables within the specification. The linear and nonlinear models are developed based on real cell culture process data, and the performance of the designed controllers is evaluated by running several real-time experiments.     

Two-stage depth filter perfusion culture for recombinant antibody production by recombinant Chinese hamster ovary cell

A rCHO cell line of DUKX origin 26*-320, producing recombinant antibody against the human platelet, was cultivated in a two-stage depth filter perfusion system (DFPS) for 20 days in order to attain high recombinant antibody concentration. The productivity of the first stage DFPS bioreactor reached 53 times that of the batch culture in a controlled stirred tank reactor and was showed 12.1 mg/L antibody concentration at a perfusion rate of 6.0 d⁻¹. Glucose concentration in the first DFPS was maintained at 1.5 g/L to avoid cell damage in the perfusion culture. A second stage DFPS system was attached to the first DFPS, which resulted in a low glucose concentration of 0.02 g/L and a high antibody concentration of 23.9 mg/L. The two-stage depth filter perfusion culture yielded 60% higher product concentration than the batch and 49-fold higher productivity of 69.3 mg/L/d in comparison with that (1.4 mg/L/d) in a batch system. Furthermore, antibody concentration of the second stage was 97% higher than that of the first stage, and the antibody productivities were comparable to that of the first stage. This two-stage DFPS system also showed potential for higher titer production of recombinant antibody and high volumetric productivity for long-term culture of bio-pharmaceutical substances.     

Forecasting and control of lactate bifurcation in Chinese hamster ovary cell culture processes

Biomanufacturing exhibits inherent variability that can lead to variation in performance attributes and batch failure. To help ensure process consistency and product quality the development of predictive models and integrated control strategies is a promising approach. In this study, a feedback controller was developed to limit excessive lactate production, a widespread metabolic phenomenon that is negatively associated with culture performance and product quality. The controller was developed by applying machine learning strategies to historical process development data, resulting in a forecast model that could identify whether a run would result in lactate consumption or accumulation. In addition, this exercise identified a correlation between increased amino acid consumption and low observed lactate production leading to the mechanistic hypothesis that there is a deficiency in the link between glycolysis and the tricarboxylic acid cycle. Using the correlative process parameters to build mechanistic insight and applying this to predictive models of lactate concentration, a dynamic model predictive controller (MPC) for lactate was designed. This MPC was implemented experimentally on a process known to exhibit high lactate accumulation and successfully drove the cell cultures towards a lactate consuming state. In addition, an increase in specific titer productivity was observed when compared with non-MPC controlled reactors.     

Model-based optimization of antibody galactosylation in CHO cell culture

Exerting control over the glycan moieties of antibody therapeutics is highly desirable from a product safety and batch-to-batch consistency perspective. Strategies to improve antibody productivity may compromise quality, while interventions for improving glycoform distribution can adversely affect cell growth and productivity. Process design therefore needs to consider the trade-off between preserving cellular health and productivity while enhancing antibody quality. In this work, we present a modeling platform that quantifies the impact of glycosylation precursor feeding – specifically that of galactose and uridine – on cellular growth, metabolism as well as antibody productivity and glycoform distribution. The platform has been parameterized using an initial training data set yielding an accuracy of ±5% with respect to glycoform distribution. It was then used to design an optimized feeding strategy that enhances the final concentration of galactosylated antibody in the supernatant by over 90% compared with the control without compromising the integral of viable cell density or final antibody titer. This work supports the implementation of Quality by Design towards higher-performing bioprocesses.     

Increasing batch-to-batch reproducibility of CHO-cell cultures using a model predictive control approach

By means of a model predictive control strategy it was possible to ensure a high batch-to-batch reproducibility in animal cell (CHO-cell) suspensions cultured for a recombinant therapeutic protein (EPO) production. The general control objective was derived by identifying an optimal specific growth rate taking productivity, protein quality and process controllability into account. This goal was approached indirectly by controlling the oxygen mass consumed by the cells which is related to specific biomass growth rate and cell concentration profile by manipulating the glutamine feed rate. Process knowledge represented by a classical model was incorporated into the model predictive control algorithm. The controller was employed in several cultivation experiments. During these cultivations, the model parameters were adapted after each sampling event to cope with changes in the process’ dynamics. The ability to predict the state variables, particularly for the oxygen consumption, led to only moderate changes in the desired optimal operational trajectories. Hence, nearly identical oxygen consumption profiles, cell and protein titers as well as sialylation patterns were obtained for all cultivation runs.     

Process design and development of a mammalian cell perfusion culture in shake-tube and benchtop bioreactors

The development of mammalian cell perfusion cultures is still laborious and complex to perform due to the limited availability of scale-down models and limited knowledge of time- and cost-effective procedures. The maximum achievable viable cell density (VCD_max), minimum cell-specific perfusion rate (CSPR_min), cellular growth characteristics, and resulting bleed rate at steady-state operation are key variables for the effective development of perfusion cultures. In this study, we developed a stepwise procedure to use shake tubes (ST) in combination with benchtop (BR) bioreactors for the design of a mammalian cell perfusion culture at high productivity (23 pg·cell⁻¹·day⁻¹) and low product loss in the bleed (around 10%) for a given expression system. In a first experiment, we investigated peak VCDs in STs by the daily discontinuous medium exchange of 1 reactor volume (RV) without additional bleeding. Based on this knowledge, we performed steady-state cultures in the ST system using a working volume of 10 ml. The evaluation of the steady-state cultures allowed performing a perfusion bioreactor run at 20 × 10⁶ cells/ml at a perfusion rate of 1 RV/day. Constant cellular environment and metabolism resulted in stable product quality patterns. This study presents a promising strategy for the effective design and development of perfusion cultures for a given expression system and underlines the potential of the ST system as a valuable scale-down tool for perfusion cultures.     

Dynamics of immature mAb glycoform secretion during CHO cell culture: An integrated modelling framework

Ensuring consistent glycosylation‐associated quality of therapeutic monoclonal antibodies (mAbs) has become a priority in pharmaceutical bioprocessing given that the distribution and composition of the carbohydrates (glycans) bound to these molecules determines their therapeutic efficacy and immunogenicity. However, the interaction between bioprocess conditions, cellular metabolism and the intracellular process of glycosylation remains to be fully understood. To gain further insight into these interactions, we present a novel integrated modelling platform that links dynamic variations in mAb glycosylation with cellular secretory capacity. Two alternative mechanistic representations of how mAb specific productivity (q_p) influences glycosylation are compared. In the first, mAb glycosylation is modulated by the linear velocity with which secretory cargo traverses the Golgi apparatus. In the second, glycosylation is influenced by variations in Golgi volume. Within our modelling framework, both mechanisms accurately reproduce experimentally‐observed dynamic changes in mAb glycosylation. In addition, an optimisation‐based strategy has been developed to estimate the concentration of glycosylation enzymes required to minimise mAb glycoform variability. Our results suggest that the availability of glycosylation machinery relative to cellular secretory capacity may play a crucial role in mAb glycosylation. In the future, the modelling framework presented here may aid in selecting and engineering cell lines that ensure consistent mAb glycosylatio.     

Effect of culture temperature on erythropoietin production and glycosylation in a perfusion culture of recombinant CHO cells

To investigate the effect of culture temperature on erythropoietin (EPO) production and glycosylation in recombinant Chinese hamster ovary (CHO) cells, we cultivated CHO cells using a perfusion bioreactor. Cells were cultivated at 37 degrees C until viable cell concentration reached 1 x 10(7) cells/mL, and then culture temperature was shifted to 25 degrees C, 28 degrees C, 30 degrees C, 32 degrees C, 37 degrees C (control), respectively. Lowering culture temperature suppressed cell growth but was beneficial to maintain high cell viability for a longer period. In a control culture at 37 degrees C, cell viability gradually decreased and fell below 80% on day 18 while it remained over 90% throughout the culture at low culture temperature. The cumulative EPO production and specific EPO productivity, q(EPO), increased at low culture temperature and were the highest at 32 degrees C and 30 degrees C, respectively. Interestingly, the cumulative EPO production at culture temperature below 32 degrees C was not as high as the cumulative EPO production at 32 degrees C although the q(EPO) at culture temperature below 32 degrees C was comparable or even higher than the q(EPO) at 32 degrees C. This implies that the beneficial effect of lowering culture temperature below 32 degrees C on q(EPO) is outweighed by its detrimental effect on the integral of viable cells. The glycosylation of EPO was evaluated by isoelectric focusing, normal phase HPLC and anion exchange chromatography analyses. The quality of EPO at 32 degrees C in regard to acidic isoforms, antennary structures and sialylated N-linked glycans was comparable to that at 37 degrees C. However, at culture temperatures below 32 degrees C, the proportions of acidic isoforms, tetra-antennary structures and tetra-sialylated N-linked glycans were further reduced, suggesting that lowering culture temperature below 32 degrees C negatively affect the quality of EPO. Thus, taken together, cell culture at 32 degrees C turned out to be the most satisfactory since it showed the highest cumulative EPO production, and moreover, EPO quality at 32 degrees C was not deteriorated as obtained at 37 degrees C.     

Hydrocyclones as cell retention device for CHO perfusion processes in single-use bioreactors

In this study, a hydrocyclone (HC) especially designed for mammalian cell separation was applied for the separation of Chinese hamster ovary cells. The effect of key features on the separation efficiency, such as type of pumphead in the peristaltic feed pump, use of an auxiliary pump to control the perfusate flow rate, and tubing size in the recirculation loop were evaluated in batch separation tests. Based on these preliminary batch tests, the HC was then integrated to 50-L disposable bioreactor bags. Three perfusion runs were performed, including one where perfusion was started from a low-viability late fed-batch culture, and viability was restored. The successive runs allowed optimization of the HC-bag configuration, and cultivations with 20–25 days duration at cell concentrations up to 50 × 10⁶ cells/ml were performed. Separation efficiencies up to 96% were achieved at pressure drops up to 2.5 bar, with no issues of product retention. To our knowledge, this is the first report in literature of high cell densities obtained with a HC integrated to a disposable perfusion bioreactor.     

Fermentanomics: Relating quality attributes of a monoclonal antibody to cell culture process variables and raw materials using multivariate data analysis

Fermentanomics is an emerging field of research and involves understanding the underlying controlled process variables and their effect on process yield and product quality. Although major advancements have occurred in process analytics over the past two decades, accurate real‐time measurement of significant quality attributes for a biotech product during production culture is still not feasible. Researchers have used an amalgam of process models and analytical measurements for monitoring and process control during production. This article focuses on using multivariate data analysis as a tool for monitoring the internal bioreactor dynamics, the metabolic state of the cell, and interactions among them during culture. Quality attributes of the monoclonal antibody product that were monitored include glycosylation profile of the final product along with process attributes, such as viable cell density and level of antibody expression. These were related to process variables, raw materials components of the chemically defined hybridoma media, concentration of metabolites formed during the course of the culture, aeration‐related parameters, and supplemented raw materials such as glucose, methionine, threonine, tryptophan, and tyrosine. This article demonstrates the utility of multivariate data analysis for correlating the product quality attributes (especially glycosylation) to process variables and raw materials (especially amino acid supplements in cell culture media). The proposed approach can be applied for process optimization to increase product expression, improve consistency of product quality, and target the desired quality attribute profile. © 2015 American Institute of Chemical Engineers Biotechnol. Prog., 31:1586–1599, 2015     

A systematic approach for scale-down model development and characterization of commercial cell culture processes

The objective of process characterization is to demonstrate robustness of manufacturing processes by understanding the relationship between key operating parameters and final performance. Technical information from the characterization study is important for subsequent process validation, and this has become a regulatory expectation in recent years. Since performing the study at the manufacturing scale is not practically feasible, development of scale-down models that represent the performance of the commercial process is essential to achieve reliable process characterization. In this study, we describe a systematic approach to develop a bioreactor scale-down model and to characterize a cell culture process for recombinant protein production in CHO cells. First, a scale-down model using 2-L bioreactors was developed on the basis of the 2000-L commercial scale process. Profiles of cell growth, productivity, product quality, culture environments (pH, DO, pCO2), and level of metabolites (glucose, glutamine, lactate, ammonia) were compared between the two scales to qualify the scale-down model. The key operating parameters were then characterized in single-parameter ranging studies and an interaction study using this scale-down model. Appropriate operation ranges and acceptance criteria for certain key parameters were determined to ensure the success of process validation and the process performance consistency. The process worst-case condition was also identified through the interaction study.     

A direct computer control concept for mammalian cell fermentation processes

In the last 10 years, new assignments and the special demands of mammalian cells to the culture conditions caused the develoepment of complex small scale fermentation setups. The use of continuous fermentation and cell retention devices requires appropriate process control systems. An arrangement for control and data-acquisition of complex laboratory-scale bioreactors is presented. The fundamental idea was the usage of a standard personal computer, which is connected to pumps, valves and sensors via ADA-transformation. The possibility of free programming allowed the development of user-oriented software, especially designed for the far-reaching requirements of a university laboratory in the field of animal cell culture. Control of aeration, pumps, data-acquisition and data-storage are combined within one program, which allows the automation of standard operations like measurement of k_La- or OTR-values. Pump control algorithms for all common fermentation strategies (batch, fed batch, chemostat, perfusion) are included and can be selected any time during cultivation. Oxygen partial pressure and pH are controlled via direct digital control (ddc), providing simple adaption of control parameters and set points to current fermentation conditions.     

A compact computational model for cell construct development in perfusion culture

A problem nowadays tissue engineers encounter in developing sizable tissue implants is the nonuniform spread of cells and/or extracellular matrices. Research shows such a nutrients transport restriction may be improved by employing hydrodynamic culture systems. We propose a compact model for the simulation of cell growth in a porous construct under direct perfusion. Unlike the previous model proposed in the literature, which composes a cellular scaffold sandwiched between two culture media layers, the current model includes only the scaffold layer to simplify the mathematical and computational complex. Results show the present single-layer model can predict cell spreads and the nutrient and metabolic waste distribution as accurately as does the three-layer model. Only if the hydrodynamic aspects such as the pressure and viscous stress are prominent to know, should the more sophisticated analyses with the three-layer model be employed. The compact model provides comparable investigations for the tissue-engineering construct developments.     

The molecular weight and concentration of dextran sulfate affect cell growth and antibody production in CHO cell cultures

To investigate the effect of dextran sulfate (DS), a widely used anti‐aggregation agent, on cell growth and monoclonal antibody (mAb) production including the quality attributes, DS with the three different MWs (4,000 Da, 15,000 Da, and 40,000 Da) at various concentrations (up to 1 g/L) was added to suspension cultures of two different recombinant CHO (rCHO) cell lines producing mAb, SM‐0.025 and CS13‐1.00. For both cell lines, the addition of DS, regardless of the MW and concentration of DS used, improved cell growth and viability in the decline phase of growth. However, it increased mAb production only in the CS13‐1.00 cells. Among the three different MWs, 40,000 Da DS was most effective in attenuating cell aggregation during the cultures of CS13‐1.00 cells, and showed the highest maximum mAb concentration. For SM‐0.025 cells, it significantly decreased specific mAb productivity, particularly at a high concentration of DS. Overall, DS addition did not negatively affect the quality attributes of mAbs (aggregation, charge variation, and glycosylation), though its efficacy on mAb quality depended on the MW and concentration of DS and cell lines. For both cell lines, the addition of DS did not affect N‐glycosylation of mAbs and decreased basic charge variants in mAbs. For CS13‐1.00 cells, the mAb monomer increased with the addition of 40,000 Da DS at 0.3–1.0 g/L. Taken together, to maximize the beneficial effect of DS addition on mAb production, the optimal MW and concentration of DS should be determined for each specific rCHO cell line. © 2016 American Institute of Chemical Engineers Biotechnol. Prog., 32:1113–1122, 2016     

A novel approach for the simultaneous quantification of a therapeutic monoclonal antibody in serum produced from two distinct host cell lines

Therapeutic monoclonal antibodies (mAbs) possess a high degree of heterogeneity associated with the cell expression system employed in manufacturing, most notably glycosylation. Traditional immunoassay formats used to quantify therapeutic mAbs are unable to discriminate between different glycosylation patterns that may exist on the same protein amino acid sequence. Mass spectrometry provides a technique to distinguish specific glycosylation patterns of the therapeutic antibody within the same sample, thereby allowing for simultaneous quantification of the same mAb with different glycosylation patterns. Here we demonstrate a two-step approach to successfully differentiate and quantify serum mixtures of a recombinant therapeutic mAb produced in two different host cell lines (CHO vs. Sp2/0) with distinct glycosylation profiles. Glycosylation analysis of the therapeutic mAb, CNTO 328 (siltuximab), was accomplished through sample pretreatment consisting of immunoaffinity purification (IAP) and enrichment, followed by liquid chromatography (LC) and mass spectrometry (MS). LC-MS analysis was used to determine the percentage of CNTO 328 in the sample derived from either cell line based on the N-linked G1F oligosaccharide on the mAb. The relative amount of G1F derived from each cell line was compared with ratios of CNTO 328 reference standards prepared in buffer. Glycoform ratios were converted to concentrations using an immunoassay measuring total CNTO 328 that does not distinguish between the different glycoforms. Validation of the IAP/LC-MS method included intra-run and inter-run variability, method sensitivity and freeze-thaw stability. The method was accurate (%bias range = -7.30–13.68%) and reproducible (%CV range = 1.49–10.81%) with a LOQ of 2.5 μg/mL.     

Progress toward forecasting product quality and quantity of mammalian cell culture processes by performance‐based modeling

The production of biopharmaceuticals requires highly sophisticated, complex cell based processes. Once a process has been developed, acceptable ranges for various control parameters are typically defined based on process characterization studies often comprising several dozens of small scale bioreactor cultivations. A lot of data is generated during these studies and usually only the information needed to define acceptable ranges is processed in more detail. Making use of the wealth of information contained in such data sets, we present here a methodology that uses performance data (such as metabolite profiles) to forecast the product quality and quantity of mammalian cell culture processes based on a toolbox of advanced statistical methods. With this performance based modeling (PBM) the final product concentration and 12 quality attributes (QAs) for two different biopharmaceutical products were predicted in daily intervals throughout the main stage process. The best forecast was achieved for product concentration in a very early phase of the process. Furthermore, some glycan isoforms were predicted with good accuracy several days before the bioreactor was harvested. Overall, PBM clearly demonstrated its capability of early process endpoint prediction by only using commonly available data, even though it was not possible to predict all QAs with the desired accuracy. Knowing the product quality prior to the harvest allows the manufacturer to take counter measures in case the forecasted quality or quantity deviates from what is expected. This would be a big step towards real‐time release, an important element of the FDA's PAT initiative. © 2015 American Institute of Chemical Engineers Biotechnol. Prog., 31:1119–1127, 2015