Equal mixing time enables scale‐down and optimization of a CHO cell culture process using a shaken microbioreactor system

The advancement of microbioreactor technology in recent years has transformed early‐ and mid‐stage process development. The monitoring and control capabilities of microbioreactors not only promote the quick accumulation of process knowledge but has also led to an increased scalability when compared to traditionally used systems such as shake flasks and microtitre plates. This study seeks to establish a framework for the micro‐Matrix microbioreactor (Applikon‐Biotechnology BV) as process development tool. Using the Dual Indicator System for Mixing Time, the system was initially characterized for mixing properties at varying operating conditions, which was found to yield mixing times between 0.9 and 41.8 s. A matched mixing time was proposed as scale‐down criterion for an IgG4 producing GS‐CHO fed‐batch process between a 5 L stirred tank reactor (STR) and the micro‐Matrix microbioreactor. Growth trends, maximum viable cell concentrations, final titre, and glycoprofiles were nearly identical at both scales. The scale‐down model was then employed to optimize a bolus feeding regime using response surface methodology, which led to a 25.4% increase of the space‐time yield and a 25% increase of the final titre. The optimized feeding strategy was validated at the small‐scale and successfully scaled up to the 5 L STR. This work for the first time provides a framework of how the micro‐Matrix microbioreactor can be implemented in a bioprocess development workflow and demonstrates scalability of growth and production kinetics as well as IgG4 glycosylation between the micro‐Matrix and a benchtop‐scale STR system.


INTRODUCTION
Process development within today's bioprocessing industry increasingly employs microbioreactor systems for screening and optimization. [1] The smaller footprint of such instruments allows for a high degree of parallelization and the small working volumes dramatically reduce the running costs compared to conventional benchtop bioreactors. Additionally, the application of disposables further increases throughput by eliminating time-consuming setup and cleaning steps. [2] As a result of this high throughput, large amounts of data can be generated quickly, which can then be used for the analysis with advanced statistical tools such as Design of Experiment (DoE) and multivariate data analysis (MVDA) to rapidly understand and optimize a cultivation process. [3,4] To leverage the findings made in the high throughput systems, scalability to the larger scale is pivotal. An important step in that direction was brought about by the advancements in sensor technologies. Disposable, precalibrated dissolved oxygen (DO) and pH sensors are now regularly employed in microbioreactors to enable monitoring and control of these parameters at the millilitre scale. [5] In addition to the standard culture parameters, hydrodynamic and mass transfer conditions should be similar across scales to guarantee an analogous process performance. To do so, a scaling criterion is defined and set to remain constant between different cultivation systems. [6] The choice of scaling criterion is primarily dependent on the cultivation system and cell type in use. Microbial growth is more likely to be limited by the transfer of oxygen into the growth medium, which renders the volumetric mass transfer coefficient (k L a) a suitable scale criterion for microbial fermentation processes. As mammalian cells require considerably less oxygen, the k L a is often not the limiting factor [7] and an alternative scale criterion might be more suitable.
A matched power per unit volume (P/V) is widely used for the scale translation of microbial and mammalian processes [8][9][10] as the P/V combines mass transfer and hydrodynamic conditions. [11] However, to experimentally determine P/V in shaken systems is associated with considerable experimental effort. Although several studies have focused on the characterization of the power consumption in shake flasks and microtitre plates, [11][12][13][14] many of the more specific microtitre plate versions have not yet been considered in the literature.
Finally, a constant mixing time can be used as an engineering basis for the process translation between small-scale and benchtop-scale cell culture systems. This parameter describes the time it takes for a system to reach a state of specified homogeneity after a perturbation has been introduced. The advantage of this scaling criterion is that a measure of the mixing time can be established with relatively little experimental effort irrespective of the mode of agitation. A constant mixing time has been shown to result in comparable growth and production kinetics between the μ24 microbioreactor and shake flasks [15] as well as microtitre plates and stirred tank bioreactors . [16] As indicated, the fundamental differences between shaken and stirred bioreactors complicate a scale translation between such sys- Prior to inoculation, the lid was autoclaved and left in the laminar flow cabinet to dry and cool down for at least 2 h. 0.8 mL of cell suspension was then transferred to each well of the 24 SRW. The lid was placed on top of the 24 SRW and a Micro-Flask cover clamp (Applikon-Biotechnology BV, The Netherlands) was used to fixate the assembled plate on an orbital shaker with 25 mm throw (Sartorius, UK). The fedbatch cultivation was performed at 220 rpm, 37 • C, and 5% CO 2 . Sampling was done sacrificially every 2-3 days. As a result of the irregular evaporation across microtitre plates, [17] corner wells were sampled first, then wells positioned on the side of the 24 SRW, and lastly wells with a central location on the plate. The manufacturer of the Micro-Flask lid states an average evaporation rate of 30 μL d -1 per well. This liquid loss was counteracted through daily bolus additions of 30 μL deionized (DI) water.

2.1.3
Micro-Matrix cultivations The micro-Matrix feeding module was autoclaved and then fitted with single-use filter bars (Applikon-Biotechnology BV, The Netherlands). To increase the CO 2 -fraction of the inflowing gas, the same gas blends were used as described for the micro-Matrix. Bolus additions of 10 mL 1% Antifoam C Emulsion (Sigma-Aldrich, UK) were added on a daily basis to the culture.

Fed-batch strategy
In all cultivation platforms, CD-CHO without MSX was used as basal medium. The ratio of sample volume to working volume was higher in the micro-Matrix compared to the 5 L STR and the 24 SRW. To prevent overfeeding in the micro-Matrix cultures, the feed volume was based on the current working volume rather than the initial working volume. Bolus feeding of Efficient Feed B (Life Technologies, UK) commenced after 3 days of cultivation (10% of the working volume) and was repeated on day 5 (9.1%), day 7 (8.3%), day 9 (7.7%), and day 11 (7.1%). Feeding regimes of the optimization experiment were translated between the scales accordingly.

Mixing time measurements
All measurements were performed using a custom-built deep-square well, made from acrylic plastic to allow for visual observations inside the well. The single-well mimic was 16 mm long, 16 mm wide, and 40 mm high, which is dimensionally equivalent to the wells of the micro-Matrix cassette. The well was fixed onto an orbital shaker with 25 mm throw (Sartorius, UK). The color change of a Dual Indictor System for Mixing Time (DISMT) solution was used to determine mixing times inside the well. [18,19] Depending on the pH, the DISMT solution assumes either a red (pH < 6.

Sampling and in-process analytics
Cell counts were performed immediately after sampling using a ViCELL

N-glycan release
The N-glycans from the IgG4 samples were released using PNGaseF, procainamide labelled and analyzed on a HILIC-FLD-MS platform as previously described. [20,21] Reagents used for N-glycan release were obtained from the LudgerZyme PNGase F Release Kit (LZ-rPNGaseFkit, Ludger). Briefly, cell-expressed IgG4 glycoproteins obtained using the different culture methods were dried down. The IgG4 glycoproteins were denatured by adding water (18 μL) and 10x denaturation buffer (2 μL) to the samples, capping the vials tightly and incubating at 99 • C for 10 min. Following the denaturation, the samples were allowed to cool to room temperature before adding the following: water (12 μL   backwards elimination with a cut-off p-value of 0.1. [22] Model analysis with ANOVA ensured model significance, non-significant lack of fit, sufficiently high effect-to-noise ratio, and a difference between adj. R 2 and pred. R 2 no larger than 0.2. Factors and levels are summarized in Table S1.

Design of experiment
The bolus feeding regime was optimized for the space-time yield (STY). The STY was calculated at the end of the process using equation 1. [23] Where Y is the product yield (mg) at the end of the process, V is the final working volume (L), and t − t 0 represents the process duration (d).

Mixing time behavior in the micro-Matrix is similar to comparable cell culture systems
An exemplary mixing time experiment within the micro-Matrix well mimic is shown in Figure 1A At lower shaking speeds, no deformation of the liquid surface could be observed. This behavior was previously described by. [24,25] Once a critical shaking speed is reached, the surface tension of the liquid is overcome, and the surface begins to deform and move in unison with the shaking motion. This critical shaking speed was exceeded for all working volumes at a shaking speed of 200 rpm. Yet, for shaking speeds of 200-220 rpm, mixing was inefficient at the bottom of the well. To prevent concentration gradients and sedimentation of the cells, it was therefore deemed critical to perform cell cultivations at shaking speeds above 220 rpm.
The mixing time was found to range between 0.9 and 41.8 s ( Figure 1B). An increased working volume generally led to longer mixing times, whereas increased shaking speeds shortened mixing times. Particularly for a working volume of 5 mL, and contrary to expectations, the mixing time appeared to increase at higher shaking speeds, an effect that has previously been documented for shaken 24 deep-square well plates. [19] The obtained mixing times fall within the range of comparable microbioreactor and benchtop bioreactor systems. A characterization of the μ24 bioreactor established mixing times of 1-13 s with working volumes of 5 and 7 mL, [15] whereas [26] reported mixing times ranging from 1.7-12,900 s for the 24 SRW format and 10-100 s for a 5 L STR. The ambr 15 was reported to reach mixing times as low as 5 s for a working volume of 13 mL. [27]

Rational selection of a scale-down strategy based on matched mixing time and matched CO 2 addition profile
To efficiently employ a small-scale cultivation system for process optimization, a scaling strategy is first devised and then validated by demonstrating that the process outcome is independent of the scale.
Here, a benchmark fed-batch process was scaled down from a 5 L STR to the micro-Matrix as well as a 24 SRW microtitre plate format and compared for growth kinetics, productivity, and IgG glycoprofile between scales.
As mammalian cells typically show relatively low requirements for oxygen compared to microbial cells, oxygen mass transfer is generally not a limiting factor. [28] Instead, the scalability of cell culture processes relies more heavily on similar hydrodynamic conditions, which the mixing time can be a useful proxy for. In this study, a matched mixing time of 6 s was chosen as criterion for the scale translation from shaken small-scale systems to the benchtop-scale. It should, however, be noted that mixing at the investigated scales is sufficiently effective at relatively low shaking and stirrer speeds. With an increasing scale, a matched mixing time approach would lead to a drastic increase of the power input in conventional STRs, [29] which could result in unsustainable shear rates. A different scaling parameter should therefore be considered for further scale-up from benchtop STR to pilot-and manufacturing-scale. [30] Initial experiments (data not shown) indicated that the growth of GS-CHO cells is affected considerably by the percentage of CO 2 in the inflowing gas. Therefore, it was considered critical to maintain a minimum CO 2 -fraction of 5% in the inflowing gas between scales. The operating conditions for all scales are summarized in Table S2.

Successful scale-down of growth and production kinetics between the micro-Matrix and a 5 L STR
The growth trend was comparable between the 5 L STR and the micro-Matrix (Figure 2A). Furthermore, the maximum VCC did not show significant differences (p > 0.05) for these systems. Although the growth of cells cultivated in the 24 SRW was slightly slower, the trend was comparable to the micro-Matrix and 5 L STR system. Similarly, the maximum VCC in the 24 SRW was reached at a later stage of the cultivation but was not significantly different to either micro-Matrix or 5 L STR (p > 0.05). The viability ( Figure 2B
The growth and production parameters summarized in Table 1 further illustrate reasonable agreement between the controlled systems.
Notably, the micro-Matrix cultivation resulted in a marginally lower final titre and specific productivity, which was likely caused by the premature onset of cell death compared to the 5 L STR cultivation.
Although the 24 SRW format achieved the highest specific productivity, as a result of the comparatively low cumulative integral viable cell concentration (cIVCC), the final titre was lower compared to the other cultivation systems.

Glycosylation is comparable between controlled cell culture systems
In order to ascertain whether the N-linked glycosylation of the IgG4 was comparable, as the process was scaled, the distribution of key glycosylation features within the glycoprofiles were compared ( Figure 3A-D). The percentage of fucosylated glycans was above 95% at all scales, which was similar to earlier studies of the same cell line and product. [31,32] Only a small fraction of antibodies (1.3%-3.0%) carried glycans with a bisecting N-acetylglucosamine moiety. Fluctuations between scales were negligible for these glycoprofiles. The distribution of sialylated antibodies was similar for the controlled systems (4.2%-5.7%), but slightly reduced for the 24 SRW format (1.9%). Similarly, galactosylation was comparable for all conditions run in the micro-Matrix and 5 L STR (36.1%-40.6%), but considerably reduced for antibodies that were produced in the 24 SRW (21.4%).
A more detailed representation of the glycoprofiles ( Figure 3E and This difference between the glycoprofiles of the IgG4 samples under the controlled and uncontrolled culture conditions was likely caused by a difference in the pH profile. Cultures in the 24 SRW were not pH controlled and were therefore subjected to a changing pH environment as illustrated by previous studies that employed pH monitoring in microwell cultivations of CHO cells. [33,34] An effect of the culture pH on the glycoprofile has been reported frequently in the literature but proved highly dependent on the cell line and the investigated glycoprotein. For instance, [35] observed increased galactosylation of an IgG3 produced by hybridoma cells with increasing culture pH, F I G U R E 4 Contour plot of the space-time yield (STY) response model. The feed start was set to the predicted optimum of 5.2 days whereas [36] and [37] reported the opposite effect for a human antibody produced in a human cell line and an IgG1 produced in hybridoma cells, respectively.
The results show that a matched mixing time proved to be a suitable scaling criterion for the process translation from a conventional benchtop-scale stirred bioreactor system to the shaken micro-Matrix microbioreactor. Using the scale-down model then allowed for highthroughput optimization of the feeding regime.

3.4
Response surface methodology enables rapid optimization of the feeding strategy The feeding regime of the bolus fed-batch strategy was optimized within the framework of a circumscribed CCD. The star points were used to cover a wider design space. The star points' circumscribed spacing ensured rotatable predictability. The feeding regime was optimized for the STY instead of the final titre to avoid a bias towards long process run times. Using the STY, the process can be optimized for its duration as well as productivity, which can facilitate the identification of processes with high annual productivity given a short turnover. [23] An overview of the tested parameter combinations and their corresponding STY responses is provided in the supporting material (Table S4). relationship. An explanation for this interaction could be that in earlier stages of the cultivation the requirements for nutrients were lower than later on in the process. Therefore, small volumes of feed were suf-F I G U R E 5 Growth (A) and production (B) kinetics of GS-CHO cells grown with the optimized feeding regime in the micro-Matrix (•) and the 5 L STR (■), in comparison to historic data of the cells grown with the original feeding regime in the 5 L STR (■). Data points represent the mean ± SD (5 L STR optimized: n = 1; micro-Matrix optimized: n = 11; 5 L STR historic: n = 2). ficient to sustain growth in the beginning, while a more rapid addition was required when feeding commenced late.
The contour plot shows that the optimal STY was captured within the investigated design space. The numerical optimization tool of Design Expert was used to predict the factor combination that yielded the highest STY. The optimal STY of 71 mg d -1 L -1 was predicted for a feed start after 5 days and a total feed volume of 64% that was divided into six bolus additions to be added on consecutive days.

3.5
Small-scale model validation and scale up of the optimized feeding strategy show a consistent productivity increase The predicted optimal bolus feeding regime was repeated in the micro-Matrix for validation and then scaled up to the benchtop scale using the previously established scaling strategy. A comparison to the original fed-batch protocol was made to assess the success of the optimization. Figure 5 demonstrates minimal differences of the growth kinetics between the scales for the optimized feeding regime. In comparison TA B L E 2 Growth and production parameters of GS-CHO cells grown under optimized and non-optimized feeding regimes at different scales

L STR (optimized) micro-Matrix (optimized) L STR (historic)
Max. VCC ( x 10 6 cells mL -1 d -1 ) 14. 3 14.5 ± 0.1 11.5 ± 1.9 Cumulative IVCC ( x 10 6 cells mL -1 d -1 ) 106. Data represent the mean ± SD (5 L STR optimized: n = 1; micro-Matrix optimized: n = 11; 5 L STR historic: n = 2) to the original feeding strategy, a moderate increase of the maximum VCC and a prolonged stationary phase were observed, which led to an increased cIVCC (Table 2). Similarly, the progression of the titre was nearly identical between the scales for the optimized feeding strategy, whereas the original feeding strategy showed a substantially reduced productivity. The final titre of the optimized feeding protocol was 25% higher compared to the original protocol. The optimized feeding strategy in the micro-Matrix and in the 5 L STR resulted in STYs of 67.7 and 69.6 mg d -1 L -1 , respectively. Through optimization, a 25.4% increase of the STY was achieved compared to the standard protocol.

CONCLUSIONS
This contribution provides a framework for the use of the micro-Matrix system as a cell culture process development tool. Initially, a characterization of the mixing time was performed to gain an understanding of the cultivation environment generated by the system and to further serve as the basis for scale translations. Mixing time was found to be comparable to other microbioreactor systems and benchtop-scale bioreactors and was therefore deemed a suitable scaling criterion.
Based on the information obtained from the mixing time characterization, the mixing time was selected to remain constant between 5 L STR, micro-Matrix and 24 SRW to ensure comparable hydrodynamic conditions. The exemplary scale translation showed similar growth and production kinetics as well as similar glycosylation profiles between micro-Matrix and 5 L STR. The growth profile and the relative % areas of glycan structures differed slightly for cultivations performed in the 24 SRW, which was attributed to the lack of pH control in this format.
The high throughput of the micro-Matrix was then leveraged to optimize the bolus feeding strategy within the framework of a circum-

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon request.