Development of a miniature bioreactor model to study the impact of pH and DOT fluctuations on CHO cell culture performance as a tool to understanding heterogeneity effects at large‐scale

Abstract Understanding the impact of spatial heterogeneities that are known to occur in large‐scale cell culture bioreactors remains a significant challenge. This work presents a novel methodology for mimicking the effects of pH and dissolved oxygen heterogeneities on Chinese hamster ovary (CHO) cell culture performance and antibody quality characteristics, using an automated miniature bioreactor system. Cultures of 4 different cell lines, expressing 3 IgG molecules and one fusion protein, were exposed to repeated pH and dissolved oxygen tension (DOT) fluctuations between pH 7.0–7.5 and DOT 10%–30%, respectively, for durations of 15, 30, and 60 min. Fluctuations in pH had a minimal impact on growth, productivity, and product quality although some changes in lactate metabolism were observed. DOT fluctuations were found to have a more significant impact; a 35% decrease in cell growth and product titre was observed in the fastest growing cell line tested, while all cell lines exhibited a significant increase in lactate accumulation. Product quality analysis yielded varied results; two cell lines showed an increase in the G0F glycan and decrease in G1F, G2F, and Man5; however, another line showed the opposite trend. The study suggests that the response of CHO cells to the effects of fluctuating culture conditions is cell line specific and that higher growing cell lines are most impacted. The miniature bioreactor system described in this work therefore provides a platform for use during early stage cell culture process development to identify cell lines that may be adversely impacted by the pH and DOT heterogeneities encountered on scale‐up. This experimental data can be combined with computational modeling approaches to predict overall cell culture performance in large‐scale bioreactors.


| INTRODUCTION
Scaling up bioprocesses from bench to production-scale is crucial in the development of biopharmaceutical products. Small-scale bioreactor models allow for time and cost-efficient development and optimization of cell culture processes. During initial cell line selection, miniature bioreactors within the microliter and milliliter scales are used for high-throughput experimentation to reduce development time and costs. 1 Cultures are then typically scaled up to bioreactors with volumes between 1 and 50 L. Once a robust operating window has been established the process is finally scaled up to productionscale vessels. These can range from single-use bioreactors, up to 3000 L, to larger stainless steel vessels up to 25,000 L. 2 There are a variety of scale-up approaches [3][4][5] which become increasingly complex and challenging to implement with scale; larger vessels are known to be less homogeneous due to longer mixing times. 6,7 The lower agitation and aeration rates used in large-scale cell culture vessels causes fluctuations, or gradients, in temperature, dissolved oxygen concentration, and pH. 8,9 In the case of microbial fermentations, a number of publications have described the effects of heterogeneous conditions caused by poor mixing. [10][11][12][13] Similar studies on mammalian cell cultures are rare.
As cell densities increase in industrial cell culture processes, heterogeneities are considered more likely and a number of pioneering studies have begun to address this issue. 14,15 One challenge in studying heterogeneities in large-scale vessels is identifying and characterizing them experimentally. Most production-scale vessels are located in commercial good manufacturing practice (GMP) facilities and hence experimentation is difficult due to the associated costs, availability, as well as the limited numbers of probes installed along the length of the bioreactor. Of the few studies available, Xing et al 16  Others have shown that in vessels up to 8000 L poor mixing can lead to high pH excursions after base addition. 17 Recently, a study using a transparent 15,000 L bioreactor has shown that correlations assuming a constant dimensionless mixing time are not valid at that scale; deviations up to 20% during single phase operation were reported. For two-phase, gas-liquid, operation the authors were able to visualize temporal and spatial heterogeneities and showed that the dispersed gas phase had a strong influence on liquid mixing time. 18 Despite the few examples of large-scale bioreactor studies in literature, detailed and reliable data is still missing. 19 This leaves gaps in knowledge around the true nature of mixing regimes within these large vessels. An alternative to experimental studies is the use of computational fluid dynamic (CFD) models to identify phenomena that impact scale-up. Detecting small perturbations in pH or DOT can be difficult with physical probes but is relatively easy using computational methods. CFD has been used to identify and characterize heterogeneities and model their effects on the performance of microbial cultures through coupled kinetic or metabolic models. [20][21][22][23][24][25] It remains a challenge to validate the impact of these heterogeneities and very little work on modeling large scale cell culture performance is available.
A number of scale-down approaches have been developed to study the effects of heterogeneities on mammalian cell culture experimentally. In cell culture the addition of base, typically sodium hydroxide or sodium bicarbonate, is used for pH control and these additions can give rise to regions of high pH around the "addition zone." 17 Experimental scale-down models have been designed to mimic this and are typically characterized as single or multicompartment models. In a single compartment model, a parameter (for example pH) is fluctuated throughout the whole vessel to simulate heterogeneities at large scale. In a multi-compartment model the fluctuation is introduced in a smaller bioreactor, either a STR or plug flow reactor (PFR), which represents the "addition zone." This reactor is connected to a larger STR which represents the "bulk" or wellmixed fraction of the production-scale bioreactor and the culture fluid is constantly circulated through both vessels. This is considered Several groups have developed scale-down compartment models to investigate heterogeneity effects. [28][29][30][31][32][33] Their approaches have two key differences: (i) whether the perturbation is a single shift or multiple shifts (i.e., the frequency of perturbation), and (ii) the volume of culture exposed to the fluctuation. To address perturbation frequency  27 The effects may be cell line specific as other studies have found little effect of pH fluctuations in these compartment models. 35 The issue of "percentage volume exposure" is difficult to address because there is very little data on the size of the base addition zone. Namdev et al. 36 suggests that the smallest zone of heterogeneity was 0.5% of total bioreactor volume in a study of Saccharomyces cerevisiae fermentation in a 300 m 3 STR. Others suggest this to be higher, around 1% 8 or 5%. 9,17 The compartment model used by Osman et al. exposed 17% of the culture to the perturbations.
Studies on the effect of DOT fluctuations are rarer but there is some evidence that DOT fluctuations affect N-linked glycosylation but not productivity of a hybridoma cell line. 37 Traditional compartment models using STRs and PFRs can be complex to set up and are not amenable to modern high throughput experimentation. This study presents a new approach to creation of a single-compartment model using the ambr ® 15 automated miniature bioreactor system. The aim is to mimic the conditions within a small zone of a large-scale vessel where the cells would be exposed to heterogeneities. The utility of this approach is that it will enable rapid identification of cell lines which might be vulnerable to fluctuations in pH and/or DOT. This study describes the development of a fluctuation methodology in ambr ® 15 bioreactors and initial evaluation of a number of cell lines with different growth characteristics. The impact of pH and DOT fluctuations on growth and productivity as well as product quality was assessed. The data obtained from these scaled-down studies could be fed into computational population balance models which could then predict overall behavior of the cell lines at large-scale as has been demonstrated with microbial fermentations. [38][39][40][41] 2 | MATERIALS AND METHODS

| Bioreactor setup and operation
The ambr ® 15 system (Sartorius Stedim Biotech, Royston, UK) is widely used in industry as a tool for high-throughput process development and is described in detail elsewhere. [42][43][44] Each ambr ® 15 vessel is equipped with one pitched blade impeller, a sparge tube, and sensors to monitor and control pH, dissolved oxygen tension (DOT) and temperature. Set points can be changed independently for each bioreactor and are maintained via PI control. The ambr ® 15 vessels were inoculated at a seeding density of 0.7 Â 10 6 cells ml À1 in an initial volume of 13.5 ml. The temperature of each vessel was controlled at 35.5 ± 0.5 C. The pH was controlled to a set point of 7.0 ± 0.1 using CO 2 (max flowrate 1.23 ml min À1 ) and 1 M sodium bicarbonate. DOT was controlled at 50% of air saturation using oxygen. These set points were used for all experiments unless otherwise stated, that is, when conducting cultures with fluctuating pH or DOT conditions as discussed later. Pure nitrogen was sparged at a constant rate of 0.15 or 1.00 ml min À1 . A solution of 0.5% vol/vol Antifoam C (Sigma Aldrich, Germany) was added to each vessel at an average rate of 20 μl per day. Proprietary media and feeds were used. Nutrient feeds were added periodically and glucose was maintained with daily additions of 500 g L À1 glucose stock solution.

| Cell culture analytics
Daily measurements of metabolite concentrations (glucose and lactate) were made using a YSI 2900 analyzer (YSI Life Sciences, Yellow Springs, USA). All other measurements were made every 2 days. Viable cell density and viability measurements were made using a ViCell XR (Beckman Coulter, High Wycombe, UK) which uses the trypan blue dye exclusion method. Offline pH, pO 2 and pCO 2 measurements were also taken every 2 days using an ABL90 Flex Plus (Radiometer Ltd, Crawley, UK) blood gas analyzer. Beginning on day 2, samples were also taken for antibody titre measurements. Supernatant was analyzed using the Octet (ForteBio, California, USA) which uses a protein A-based quantification method. Samples were reduced with dithiothreitol (DTT) and glycans released using N-Glycosidase F (PGNase F, VWR Cat#3261). Samples were injected onto an Acquity UPLC BEH300 column (Waters UK, Elstree, UK), from which the eluate was directly fed into the SYNAPT mass spectrometer (Waters UK) with electrospray ionization (ESI). The mass to charge raw data was deconvoluted to a relative mass; the largest peak in each chromatogram is always the G0F glycoform and all other glycoforms are inferred based on their mass relative to the G0F peak.

| Derived cell-specific parameters
Average cell-specific glucose consumption (q Gluc ), lactate production (q Lac ) and protein production (q p ) rates were determined by linear regression of cumulative production (or consumption) against the cumulative integral of viable cell density (cIVC), using the following equations: where IVC t is the cumulative integral of viable density at time t (cellsday ml À1 ), VCD is the viable cell density (cells ml À1 ), and t is time (days). The maximum specific growth rate was determined from the highest specific growth rate of the culture calculated between measured time points using the equation below: Glucose consumption per day, gluc cont , was calculated using the following equation: where [gluc t ] is the concentration of glucose in the culture at time t.
Cumulative values for product concentration and gluc cont were then determined and rates were calculated from data during the stationary and exponential phases of cell growth. Lactate production rates were calculated between two time periods: day 0-6 and day 6-10, F I G U R E 1 Examples of "ideal" and "real" fluctuation profiles. (a) Shows an illustration of the how the fluctuations are programmed; dotted lines represent step changes in set points (pH or DOT) and solid lines represent the "ideal" profile caused by the step change.
(b) and (c) Are 2 h samples of "real" pH and DOT profiles, respectively, from two separate ambr ® 15 vessels, where pH and DOT were programmed to fluctuate during a fed-batch culture of AZCL_3 between pH 7-7.5 and DOT 10%-30%, respectively. The arrows highlight where a change in set point was programmed. Both profiles (b) and (c) show fluctuations of 15 min duration and frequency. DOT, dissolved oxygen tension in order to account for the change in lactate profile due to consumption.

| Statistical analysis
To test significance between certain conditions a simple two-tailed  The amplitude of the fluctuations achieved at the three frequencies were lower than programmed and many cultures struggled to reach pH 7.5. Table 1   Preliminary studies using the micro24 ® bioreactor indicated that the effects of fluctuations are likely to be cell line dependent. 46 A key difference between cell lines likely to influence their response to fluctuations is their peak viable cell density. In this study two industrial cell lines were compared; one with a 'low' peak VCD ($21 x 10 6 cells ml À1 ) and the other with a "high" peak VCD ($60 x 10 6 cells ml À1 ).
This section describes the investigation into the "low" growing cell line (AZCL_1).
Previous studies investigating the effects of pH heterogeneity   Figure 3 compares the performance of control vessels and those exposed to pH and DOT fluctuations with respect to cell growth, antibody productivity, and lactate concentration. The left column in Figure 3 shows the performance of the three control cultures where no fluctuations were introduced. All of the control cultures performed similarly to previous AZCL_1 cultures run at 5 L scale achieving a peak VCD of around 19 x 10 6 cells ml À1 . Control 3 appears to grow 22% less than the other two controls with higher gas flowrate, peaking at only 15 x 10 6 cells ml À1 . This is also reflected in the lower lactate accumulation, suggesting that the lower N 2 flowrate may be the cause of lower growth and lactate production (q lac ) in AZCL_1 cultures. Interestingly however, the titre was not affected in the same way, meaning the q p of Control 3 was higher than the two other controls (Table 3).
Comparing across the rows in Figure 3 it is clear that there is little difference in viable cell density, titre and lactate levels between control and fluctuated vessels. Looking at the derived culture parameters in Table 3, no effect on glucose consumption (q gluc ) was observed. Fluctuated vessels did, however, have higher initial lactate production (q lac ), particularly in pH-fluctuated vessels which had lactate production rates twice as high as controls between day 0 and 6. Fluctuations appeared to also slightly increase the growth rate (μ max ), as seen in Table 3. This   Figure 4. This was expected with AZCL_3 due to its rapid growth; the cell culture quickly depletes the nutrients in the media and feeds; while lactate and other metabolite concentrations increase to detrimental levels.
As with the low-growing cell line (AZCL_1), pH fluctuations are seen to have little effect on growth, productivity or lactate levels in F I G U R E 3 The performance of AZCL_1 ("low" growing cell line) ambr ® 15 cultures subjected to pH (middle column) and DOT (right column) fluctuations at frequencies of 15, 30, and 60 min. Cell growth (row 1), antibody titre (row 2) and lactate concentration (row 3) data were gathered from duplicate or triplicate vessels and error bars represent 1 SD from the mean. All controls (left column) were set to 50% DOT and pH 7.1 but varied in PI settings and N 2 flow rates as described in Table 2. DOT, dissolved oxygen tension AZCL_3 cultures. The control and pH-fluctuated vessels had very similar growth profiles peaking around 62 Â 10 6 cells ml À1 , which again match the growth profiles seen in previous 5 L cultures. The 60 min pH fluctuations did appeared to have slightly lower growth and product titre on average, compared to the other fluctuation frequencies.
However, the 15 and 30 min pH-fluctuated vessels had higher peak product titre (5.5 gL À1 ) compared to the control vessels (4.9 gL À1 ).
This was a slightly surprising result although these differences were not statistically significant.  (Table 2). pH-fluctuated vessels were held at 50% DOT. Data was gathered in duplicate or triplicate and the error bars represent 1 SD from the mean. The hashed horizontal lines indicate the highest "peak" values achieved in each culture. Effects of pH and DOT fluctuations on product quality. DOT, dissolved oxygen tension 5; but had a much lower consumption rate than the other controls (Table 4). This profile more closely matched the lactate profiles of the pH-fluctuated vessels. This suggests that although pH fluctuations appear not to have a significant effect on growth and productivity there may be an effect on metabolic activity, which has been observed in other studies. 45,47 This is a relevant observation as it is known that increased lactate production can be an issue during scale up. 3,49 In contrast, cultures with fluctuating DOT conditions demonstrated significant differences to control cultures in the case of the "high" growing AZCL_3 cell line. DOT-fluctuated cultures showed a 35% decrease in growth compared to controls. This decrease impacts the productivity by the same degree; DOT-fluctuated cultures had peak antibody titres between 2.5 and 3.1 gl À1 ; significantly lower than controls (p value <0.05). The specific productivity, q p , was not affected however ( Table 4). The DOT-fluctuated vessels had higher initial lactate production rates (q lac [day 0À6] ) than controls, as seen in Table 4, and exhibited minimal lactate consumption. Cultures instead entered a "negative feedback loop," where lactate accumulation induced more base addition (than controls) and caused lactate runaway, as seen in Figure 4. It appears that this increase in the lactate concentration is responsible for the observed decease in growth and productivity. The lactate profile of control 2 matched the profiles of the DOT-fluctuated vessels, which suggests that the increased lactate accumulation may be caused by the overall lower (30%) DOT. Specific glucose consumption, q gluc , was also affected by DOT fluctuations; all DO-fluctuated vessels exhibited higher q gluc than controls, as seen in Table 4.
A previous study has found that fluctuating DOT in a hybridoma cell line caused higher lactate yield. 37 These results also match other studies where a cell line producing a conjugated IgG2 was shown to have decreased product titre, increased lactate accumulation as well as lower drug-to-antibody (DAR) ratio after being exposed to DOT fluctuations. 50 Other recent papers have shown that reducing the DOT, in order to mimic the DOT levels cells are expected to be exposed to at large-scale, caused an increase in peak lactate and reduced the lactate consumption rates. 51 They also showed that lower DOT cultures matched the profiles of 5000 L cultures and this negative impact was attributed to DOT heterogeneity, or more specifically hypoxic conditions. In addition to decreased viability the study found that lowering DOT also reduced ammonium levels, which was consistent with increased glycolysis and reduced amino acid catabolism observed in the T A B L E 2 Summary of control parameters used for AZCL_1 and AZCL_3 cultures with programmed fluctuations. T A B L E 3 Summary of specific production and consumption rates for AZCL_1 ambr ® 15 cultures with fluctuating pH and DOT profiles. Data was gathered from duplicate or triplicate vessels and error bars represent 1 SD from the mean. The coefficient of determination (R 2 value) for the linear regressions were all above 0.95, except those marked with an asterix, which were in the range of 0.55-0.95 Condition q gluc (pg cells À1 h À1 ) q lac (day 0-6) (pg cells À1 h À1 ) q lac (day 6-10) (pg cells À1 h À1 ) q p (pg cells À1 h À1 ) μ max (h À1 ) lower DOT cultures 51 The agreement between the findings of our small-scale studies (15 ml) and these larger scale studies (>3 L) supports the use of the miniature bioreactor platform for studying the response of different cell lines to culture heterogeneities.  T A B L E 4 Summary of specific production and consumption rates for AZCL_3 ambr ® 15 cultures with fluctuating pH and DOT profiles. Data was gathered from duplicate or triplicate vessels and error bars represent 1 SD from the mean. The coefficient of determination (R 2 value) for the linear regressions were all above 0.95, except those marked with an asterix, which were in the range of 0.60-0.95 While DOT fluctuations had a pronounced impact on cell growth ( Figure 4), they had less impact on antibody glycosylation ( Figure 5).

| Effects of pH and DOT fluctuations on product quality
Man5 levels appeared to be slightly lower in DOT-fluctuated vessels compared to pH-fluctuated vessels and controls 1 and 3 (by approx. 0.4%). When comparing DO-fluctuated vessels to control 2 the differences are even smaller. This suggests that the effects seen in the DOfluctuated vessels may be due to the overall lower DOT that cells are exposed to in these cultures, rather than the fluctuation itself. in a Fc-fusion protein producing CHO cell line, to DOT heterogeneity during scale-up. The AZCL_3 line did not express any sialylated glycans; however, two additional cell lines were exposed to DOT fluctuations and one of these (AZCL_4) expressed low levels of sialylated glycans, which is described in the next section.  Overall, this work suggests that the impact of pH and DOT fluctuations are cell line and product specific and are more likely to affect F I G U R E 6 Effect of DOT fluctuations on glycosylation of antibodies produced in AZCL_4 and AZCL_5 cultures. Controls at 30% and 10% DOT were used. Fluctuation frequencies of 15 min (DO15) and 60 min (DO60) was tested. AZCL_5 cultures were run with the higher PI and N 2 flowrate settings and AZCL_4 was run with the higher PI settings and lower N 2 flowrate (0.15 ml min À1 ), as described in Table 2. Total fs refers to the totally percentage of sialylated glycans. Error bars represent 1 SD about the mean (n = 2 or 3). DOT, dissolved oxygen tension higher growing cell lines. The automated miniature bioreactor therefore provides a platform for rapid investigation of potential large-scale culture heterogeneities on candidate cell lines and antibody products.

| Effect of DOT fluctuations on other cell lines
Data from fluctuating and non-fluctuating bioreactor studies is currently being used to try and predict overall culture performance at large scale using population balance approaches.

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