(596c) Model-Based Process Optimization for Upstream Bioreactor Production–Downstream Chromatographic Separation of CHO Cell Monoclonal Antibodies (mAb)

Biological drugs have become an indispensable part of modern medical treatments and biopharmacy the most contemporary and fast growing branch of pharmacy. Therefore, the pharmaceutical industry is faced with the development of ever more challenging products, where further challenges are associated with (i) cost reduction of product development, (ii) development of robust technological process and (iii) maintaining a high quality of the final product. All three challenges are tightly linked to the philosophy of product development with integrated quality, which represents a systematic approach to product development, where the quality of the product is already installed during development/production process. The key element in reaching this goal is mechanistic understanding of the biochemical and engineering process and effects of certain parameters on the final quality of the product and the robustness of the process.

Monoclonal antibodies (mAbs) offer novel therapy avenues for cancer, inflammatory diseases, infectious diseases, and autoimmune diseases, have had remarkable success in both regulatory approval and global sales and represent one of the most valuable products of the biopharmaceutical industry. Their therapeutic efficacy depends on the post-translational process of glycosylation, which is influenced by manufacturing process conditions. Commercial mAbs are mostly produced by mammalian Chinese hamster ovary (CHO) cells where they also undergo glycosylation. In this process, a carbohydrate chain is covalently attached to the amide nitrogen of a certain asparagine (N-linked glycans) in a growing polypeptide chain. This modification serves various functions like correct folding of a protein, stability, cell-cell adhesion, serum half-life, etc.

In this work, we have focused on glycosylation (upstream) and glycan isolation (affinity chromatography). Both steps are involved in mAb production by biopharmaceutical industry. The synthesis of N-linked glycan starts in the endoplasmic reticulum, continues in the Golgi apparatus and ends at the plasma membrane, where the N-linked glycoproteins either are secreted or becomes embedded in the plasma membrane. The maturation of glycan is largely dependent on factors such as expression, activity, and localization of glycosidase and glycosyltransferase enzymes, intracellular levels and availability of nucleotides and nucleotide sugars (GDP-Man, UDP-GlcNAc, UDP-Glc, and UDP-Gal, etc.) and accessibility of glycosylation sites on the protein.

The second process is affinity chromatography, which as a method is employed for isolation of the mAb from biochemical mixtures based on a highly specific interaction such as that between antigen and antibody. Due to the high price of affinity resin, which accounts for 50-80 % of total purification cost, affinity chromatography is the most expensive downstream processing step. Therefore, improving efficiency of resin utilization offers a great opportunity for reduction of production cost, and a lot of research effort is directed into it.

The upstream and downstream processes in mAb production were evaluated with mechanistically and statistically based models, which were validated by experimental data. The developed models, incorporating fundamental biological and pharmaceutical know-how with an engineering approach, could serve as an optimization tool for the selected processes, leading to smart planning of experiments. Consequently, the procedures will be smoother and the work more rationalized. Finally, through the mechanistic description and the validated mathematical models, the need for optimization by physical experiments, will further reduce the material costs of a new developmental process.

1. I. Upstream process - glycosylation

To study the dynamics of the glycan production inside the Golgi apparatus, an N-linked glycosylation model is used. Following Jimenez del Val et al. (2011), we employ the mAb Fc glycosylation reaction scheme to determine the ensemble of possible glycans and their linkages. The scheme consists of 8 different enzymes (ManI, ManII, GnTI, GnTII, GnTIII, FucT, GalT, SiaT) which act as catalysts, and five different monosaccharides (GlcNAc, Mannose, Fucose, Galactose, Sialic Acid), producing a total of 77 glycans and 95 possible reactions.

The incorporated dynamical model consists of 10 decision tree rules for enzyme operation, where each enzyme reaction rate is calculated for each step in the pathway. The reaction rate calculation is based on either Michaelis-Menten, sequential-order Bi-Bi, or random-order Bi-Bi kinetics, depending on the enzyme type. Each reaction rate is a function of various variables, most important being the concentration of glycans, enzymes and nucleotide sugar and enzyme kinetic parameters (e.g., turn-over rate and dissociation constants). Reaction rates also depend on the concentration of glycans which are catalyzed by the same enzyme, but not in the same reaction. Furthermore, to simulate glycosylation in the Golgi apparatus in a more realistic manner, the enzyme concentration is a function of relative location in the Golgi, thus imitating different Golgi compartments. Instead of location, the enzyme concentrations can be expressed in terms of the residence time of each glycan in the Golgi, if we assume constant travel velocity of glycans through Golgi. Many of the parameters that have been employed were from different sources and different cell lines. Although this could limit confidence in our calculations, the parameters values have been selected to be the closest possible to the main mAb producing cell line (Chinese hamster ovary) and are the best available from literature.

Once reaction rates are obtained for each possible reaction, Monte Carlo procedure is used to choose which step of glycosylation will be executed. After execution, the rates are updated, and the procedure repeats. The glycosylation can stop if either the enzyme concentration in the Golgi apparatus is too low at a given time, or if the glycan in question is a final glycan, meaning it can not be further glycosylated. When the simulation stops and all the starting glycans pass through and exit the Golgi apparatus, we can monitor the relative fractions of each glycan in this organelle. This can also be done during the simulation, to obtain the temporal behavior of glycan fractions. The idea is that once this model is established, we can implement it in a reverse manner, to statistically evaluate input parameter values (e.g., concentrations in the Golgi, enzyme kinetic parameters, etc.) by using the experimentally determined relative fractions of glycans in the Golgi apparatus.

1. II. Downstream process – affinity chromatography

The effectiveness of affinity resin utilization is best reflected in process productivity, which was selected as a target function for optimization in this study.

The batch affinity chromatography was operating in a cyclic mode. One cycle was composed of four basic steps: loading of mAb on equilibrated affinity sorbent contained in chromatography column (1), washing of column to remove impurities (2), elution to desorb mAb bound on sorbent (3) and equilibration/regeneration step (4) to prepare the column with affinity sorbent for the next cycle. Process productivity (P) can be defined as the mass of mAb captured in given cycle time:

where t_cycle is time duration of one cycle and is composed of loading (t_L), washing (t_W), elution (t_E) and regeneration (t_R) times. Only loading the step was optimized in this study. Times to wash, elute and regenerate column and their sum (t_WER) were therefore held constant.

An additional constraint was a maximum allowed column velocity as prescribed by affinity sorbent material manufacturers. Different loading strategies shown in Table 1 were optimized with a mathematical model of affinity chromatography.

Table 1. Column loading strategies optimized with the aim of improving productivity of affinity chromatography

Three different sums of washing, elution and regeneration times (t_WER) were used in our calculation: 85, 100 and 115 min. A search for a point with the highest value of productivity and corresponding residence time (flow rate) by sweeping amongst points 0.25 min apart on the interval of residence times was considered more time efficient than a search of best point by Matlab fminsearch function.

Affinity chromatography column was modeled using a general rate model with bulk liquid mass balance:

where ε_c is column void fraction, u interstitial fluid velocity, z axial coordinate, D_ax axial dispersion coefficient and r_p particle radius, and particle mass balance:

where ε_p is particle porosity, τ pore tortuosity factor, D_eff effective pore diffusivity, r radial coordinate and q concentration of bound antibody.

Acknowledgement

The provision of financial support for the conduct of the research by the project NextBiopharmDSP (FP7/2007-2013), RDI project BioPharm.si (co-financed by the Republic of Slovenia, Ministry of Education, Science and Sport and European Union under the European Regional Development Fund, 2016-2020) and Slovenian Research Agency (ARRS) (Program P2–0152) is gratefully acknowledged