Control of IgG glycosylation by in situ and real‐time estimation of specific growth rate of CHO cells cultured in bioreactor

The cell‐specific growth rate (µ) is a critical process parameter for antibody production processes performed by animal cell cultures, as it describes the cell growth and reflects the cell physiological state. When there are changes in these parameters, which are indicated by variations of µ, the synthesis and the quality of antibodies are often affected. Therefore, it is essential to monitor and control the variations of µto assure the antibody production and achieve high product quality. In this study, a novel approach for on‐line estimation of µ was developed based on the process analytical technology initiative by using an in situ dielectric spectroscopy. Critical moments, such as significant µ decreases, were successfully detected by this method, in association with changes in cell physiology as well as with an accumulation of nonglycosylated antibodies. Thus, this method was used to perform medium renewals at the appropriate time points, maintaining the values of µ close to its maximum. Using this method, we demonstrated that the physiological state of cells remained stable, the quantity and the glycosylation quality of antibodies were assured at the same time, leading to better process performances compared with the reference feed‐harvest cell cultures carried out by using off‐line nutrient measurements.


| INTRODUCTION
The need for biopharmaceutical protein products, in particular recombinant monoclonal antibodies (mAbs), has been constantly increasing in the last decade (Walsh, 2014). Chinese hamster ovary (CHO) cells are the most commonly used cell lines for mAb production due to their capability of performing human-like posttranslational modifications (PTMs; Cole, Demont, & Marison, 2015). Glycosylation is one of the most important PTMs for mAbs, as it impacts effector functions, pharmacokinetics, antigenicity, safety, stability, and solubility of the mAbs produced (Tharmalingam, Wu, Callahan, & T. Goudar, 2015). Therefore, glycosylation is a critical quality attribute (CQA) for mAb production processes that should be taken into consideration in all manufacturing steps, as outlined in the quality by design (QbD) approach initiated by the Food and Drug Administration (FDA) in 2004 (ICH, 2017). The QbD initiative emphasizes the importance of using process analytical technologies (PAT) to monitor and control in real-time the critical process parameters (CPPs) which may influence the product CQAs in biopharmaceutical production processes.
Among various CPPs of mAb production bioprocesses, viable cell density (VCD) is undoubtedly one of the most important, as it is often closely related to the cell growth, the mAb production, and overall process performances (Lee, Carvell, Brorson, & Yoon, 2015). With Biotechnology and Bioengineering. 2019;116:985-993. wileyonlinelibrary.com/journal/bit increasing production potential of these bioprocesses, reliable in-line and real-time measurements of VCD are more and more required.
Various PAT tools have been proposed to monitor VCD in real-time, including dielectric spectroscopy (Ansorge, Esteban, & Schmid, 2007;Cannizzaro, Gügerli, Marison, & von Stockar, 2003;Courtès, Ebel, Guédon, & Marc, 2016;Opel, Li, & Amanullah, 2010;Yardley, Kell, Barrett, & Davey, 2000), near-infrared and Raman spectroscopies (Abu-Absi et al., 2011;Cervera, Petersen, Lantz, Larsen, & Gernaey, 2009), in situ microscopy (Guez, Cassar, Wartelle, Dhulster, & Suhr, 2004), acoustic resonance densitometry (Kilburn, Fitzpatrick, Blake-Coleman, Clarke, & Griffiths, 1989), and soft-sensor-based approaches (Kiviharju, Salonen, Moilanen, & Eerikäinen, 2008). Among these technologies, dielectric spectroscopy is probably one of the most reliable methodologies to monitor VCD due to its simplicity, robustness, and its capability of in situ fast signal acquisition which is noninvasive and nondestructive for cell cultures. The dielectric spectroscopy is also insensible to air bubbles and cell debris which often cause problems for other methods (Justice et al., 2011). Dielectric spectroscopy is based on measurements of the ability of viable cells to store electrical charge as a function of the frequency of the applied electrical field. The basic theoretical background of dielectric properties of biological cells was described elsewhere (Yardley et al., 2000). However, on-line monitoring of VCD alone gives an incomplete vision of cell growth and cell physiological state. To date, it remains challenging to monitor in real-time the cellular physiological state due to the complexity of animal cell system (Henry, Kamen, & Perrier, 2007).
The specific cell growth rate (µ) is one of the most direct indicators of cell physiological state. When cells are in their active growth stage (µ close to its maximum, µ max ), they are able to produce enough energy for energy-consuming actions such as cell division and recombinant protein synthesis. On the contrary, under nutrients depletion, toxins accumulation, or when other physicochemical stresses appear, energy production of cells may be negatively influenced, resulting in a decrease in µ (Kondo et al., 2000). In this case, not only the quantity but also the quality of the recombinant proteins produced could be affected, since noncorrectly performed PTMs associated to mAb production could appear during the process (Kochanowski et al., 2008). Therefore, µ appears as an essential parameter for mAb production bioprocesses, which should be carefully monitored and controlled throughout the process to ensure the quantity and the quality of the mAb synthesized. Classically, µ is often calculated indirectly from off-line measurements of cell densities, or by macroscopic kinetic modeling approaches, which could be inaccurate or provide only estimated values, resulting in delayed information which is not adequate for control strategies (Xiong, Guo, Chu, Zhuang, & Zhang, 2015).
In the present work, a novel approach of real-time estimation of µ was developed based on on-line monitoring of VCD by using in situ dielectric spectroscopy and mass balance equations. This approach was then applied on a recombinant immunoglobulin G (IgG) production process performed by CHO cells cultured in a bioreactor in feed-harvest mode to investigate its potential for the improvement of the process control. Quality of the IgG produced was assessed by analyzing the macroheterogeneity of IgG glycosylation which is the glycosylation site occupancy. Overall, the results of this study demonstrate that µ is an important process parameter allowing to detect early cell physiological state changes. On-line estimation of µ by using dielectric spectroscopy leads to better process control for the feed-harvest cultures, and as a result, the antibody quantity and the quality concerning the glycosylation were assured.

| In situ dielectric spectroscopy
The cell culture permittivity was measured in real-time using a sterilizable capacitance probe connected to an Evo 200 iBiomass system (Hamilton, Bonaduz, Switzerland) at a working frequency of 1,000 kHz. Measurements were performed every 12 min and the baseline was set by recording the permittivity of the cell-free medium, before cell seeding. The permittivity can be related to VCD using a linear relationship (Courtès et al., 2016): In this study α equals to 13 and β equals to 3.

Based on VCD on-line estimation and mass balance equation, µ
can be obtained in real-time by calculating the slope of a moving window linear regression on the logarithmic values of VCD in an interval of time (Equation (2)).
To reduce the noise of data collected from dielectric spectroscopy, moving average filters were used to have a more stable on-line µ estimation. Meanwhile, the distance residual (R) was calculated in real-time to evaluate the variations of the µ values with respect to the value of µ max (Equation (3)).
Here µ max is a predefined value of the maximum value of µ at the exponential growth phase obtained from previous cultures, and the µ t is the on-line estimation of µ at the instant t.
Moreover, using multifrequency dielectric measurements and off-line cell diameter measurements, a cell-specific dielectric property parameter, the intracellular conductivity (σ i ) can be calculated with Equations (4-6):

| Statistical analysis
The accuracy of the developed method was evaluated by the root mean square error (RMSE), calculated as follows: where y i is the off-line measured values (VCD or µ),ŷ i is the on-line predicted values (VCD or µ), and n the number of samples under consideration.

| RESULTS AND DISCUSSION
3.1 | Cell growth, metabolism, and IgG production kinetics in batch culture To assess the influence of cell growth on IgG glycosylation during batch cultures, CHO cells were cultured in a 2-L bioreactor, and cell density, viability, nutrients and by-products concentration, IgG production, and IgG glycosylation were monitored off-line. In addition, the off-line values of µ were calculated at the end of the culture using VCD values measured off-line. Although cells displayed excellent viabilities (>95%) until 127 hr, the value of µ started to decrease after about 100 hr of culture ( Figure 1a). This decrease of µ is probably due to the glutamine depletion (Figure 1b), since glutamine, in addition to its energetic substrate function, is proved to be one of the most important key-precursors in several biosynthetic pathways of CHO cells. Its availability can affect the purine and pyrimidine synthesis rate (Hayter et al., 1991). Alternatively, it is possible that the accumulation of lactate and ammonia inhibits the growth of cells according to other authors (Glacken, Fleischaker, & Sinskey, 1986;Hassell, Gleave, & Butler, 1991;Yang & Butler, 2000). However, in our case, the concentration of ammonium ions and lactate were relatively low (<5 and <20 mM, respectively; Figure 1b), which were shown to have little or negligible effect on the growth of cells (Hayter et al., 1991;Lao & Toth, 1997). The intracellular conductivity (σ i ), which measures the ability of the cell's intracellular environment to conduct electrical current, was relatively stable during the first 95 hr of culture, followed by a light decrease until 103 hr before a significant increase after this period. This observation indicates potential changes in cell physiological state, which is demonstrated by the variations of σ i (Figure 1c). As suggested by some authors, the decrease of this parameter often corresponds to nutrients limitations and depletions (Ansorge, Esteban & Schmid, 2009;Opel et al., 2010). In our case, the decrease of σ i is associated with glutamine exhaustion. In addition, the final increase of σ i was reported to be linked to cell death since alteration in cell membrane integrity could result in the entry of medium ions into cells, leading to significant changes in the intracellular environment (Opel et al., 2010). As for IgG production, Figure 1d shows that almost all the LI ET AL.

| 987
IgGs produced were fully glycosylated until 100 hr of culture, then nonglycosylated IgGs began to accumulate in cell culture and reached about 30% of the total IgGs, resulting in an important product quality decrease.
It seems that the diminution of µ coincided with the accumulation of nonglycosylated IgG form, which could be explained by an insufficient energy production of CHO cells due to the glutamine exhaustion after about 100 hr, as suggested by Xie and Wang (1996).
This decrease of µ led to cell physiological state changes as showed by σ i calculations. Consequently, the recombinant protein production was also affected, resulting in a loss of the ability for cells to perform correctly the PTMs for the proteins produced because of less available energy (Hayter et al., 1993). In this study, we demonstrated that the global glycosylation of IgGs was influenced by the growth of cells, since the nonglycosylated IgG level increased abundantly after the decrease of µ, leading to a significant decrease in the final product quality at the end of the culture. Therefore, as a direct indicator of the cell physiological state, µ is a key parameter of cell culture processes and should be accurately monitored and controlled as rapidly as possible, to limit the physiological changes of cells, and thus make sure that the glycosylation of IgGs produced is not altered. To calculate the value of µ, VCD predicted was first converted to a logarithmic form (ln(VCD)), then the slope of the moving window linear regression of ln(VCD) as a function of culture time was calculated to obtain the value of µ, as illustrated by Equation (2). However, data collected from capacitance probe presented high level of noise, so smoothing filters were applied to make data more stable for µ estimation. In this study, two moving average filters were | 989 changes in protein glycosylation occurred at the end of the exponential cell growth phase (Castro, Ison, Hayter, & Bull, 1995;Hooker et al., 1995;Liu et al., 2014). They suggested that the depletion in nutrients or in glycosylation precursors in the medium could be responsible for the change in the glycosylation pattern.

|
However, another study suggested that protein glycosylation was not dependent on limitation in glucose, nor glutamine, but highly dependent on the energetic status of the cells (Kochanowski et al., 2008). Therefore, when there were alterations in cell physiological state as indicated by the decreases of µ and confirmed by the variations of σ i , the nonglycosylated antibody level increased, although this level was reduced slowly after medium renewals. Consistently, a constant value of σ i was maintained all along the cell culture before the final death phase (Figure 5c), indicating that cell physiological state remained stable, which led to better process performance regarding antibody quantity as well as antibody quality.

|
As demonstrated in Figure 4, the total cumulative production of IgG was 0.96 g in 310 hr, which was more important than that produced in the cell cultures using off-line (early) strategy, and similar to that produced in the cell culture using off-line (late) strategy. Most importantly, the quality of the produced IgG was maintained, as the level of nonglycosylated form of IgG produced was kept low (~5%) throughout the culture using this on-line strategy (Figure 6c). Using on-line estimation of µ, these results suggest that when the growth rate of cells is kept stable by medium renewals, correctly and timely performed, both the quantity and the quality of the antibody produced can be assured, leading to better process performances with minimum operator intervention.

| Statistical analysis
As for statistical analysis, the RMSE of VCD prediction for the three feed-harvest strategies (off-line late, off-line early, and on-line) was 3.4, 2.7, and 2.6 × 10 5 cell/ml, respectively. These low relative errors

| CONCLUSION
Cell-specific growth rate (µ) is a key parameter for antibodyproducing animal cell culture processes which reflects the cell physiological state. To optimize the antibody production, and especially, to ensure the quality of the final product, µ should be carefully monitored and controlled in real-time. In this study, a method of on-line estimation of µ was developed based on in situ dielectric spectroscopy measurements. As a result, the variations of µ were monitored in real-time using various mathematical methods to process these data, allowing rapid detection of the critical moment when µ decreased significantly. This on-line detection of the critical moment was successfully implemented in feed-harvest cell cultures, and µ was maintained around its maximum value by performing medium renewals at critical moments. This on-line strategy allowed to maintain an appropriate cell physiological state. Consequently, a better process performance was obtained with both the quantity and the quality (concerning the glycosylation) assured for the antibody produced. When harvested at a critical moment, the nonglycosylated antibody level was kept minimum, leading to the highest recovery of the antibody correctly glycosylated compared with the cell cultures performed with off-line strategies. The method of on-line estimation of µ shows great potentials for antibody production bioprocesses, and future efforts should be aimed at the implementation of this online method to fully automated feedback control scheme in animal cell culture processes for antibody quality control.