Real‐time quantification and supplementation of bioreactor amino acids to prolong culture time and maintain antibody product quality

Abstract Real‐time monitoring of cell cultures in bioreactors can enable expedited responses necessary to correct potential batch failure perturbations which may normally go undiscovered until the completion of the batch and result in failure. Currently, analytical technologies are dedicated to real‐time monitoring of bioreactor parameters such as pH, dissolved oxygen, and temperature, nutrients such as glucose and glutamine, or metabolites such as lactate. Despite the importance of amino acids as the building blocks of therapeutic protein products, other than glutamine their concentrations are not commonly measured. Here, we present a study into amino acid monitoring, supplementation strategies, and how these techniques may impact the cell growth profiles and product quality. We used preliminary bioreactor runs to establish baselines by determining initial amino acid consumption patterns, the results of which were used to select a pool of amino acids which gets depleted in the bioreactor. These amino acids were combined into blends which were supplemented into bioreactors during a subsequent run, the concentrations of which were monitored using a mass spectrometry based at‐line method we developed to quickly assess amino acid concentrations from crude bioreactor media. We found that these blends could prolong culture life, reversing a viable cell density decrease that was leading to batch death. Additionally, we assessed how these strategies might impact protein product quality, such as the glycan profile. The amino acid consumption data were aligned with the final glycan profiles in principal component analysis to identify which amino acids are most closely associated with glycan outcomes.

preferred platform for the manufacturing of therapeutic proteins such as monoclonal antibodies (mAb). 1,2 As nutrients such as amino acids are consumed during the manufacturing process and metabolites are produced, analytical methods to determine how their changing concentrations impact critical quality attributes (CQA) such as product glycosylation are essential. 3 For this purpose, we developed a process analytical technology (PAT) enabling application that allows for rapid assessment of bioreactor amino acid levels. For the sake of speed and simplifying sample preparation, this chromatographic approach measures nonlabeled underivatized amino acids using a mass spectrometer. This allows us to be able to quantitate amino acid concentrations in the bioreactor media in near real time and support customized control of individual nutrient species. After the bioreactor run was completed, we performed glycan characterization and measured additional product quality characteristics such as charge variants and size variants, permitting us to associate specific amino acids with their estimated effect on the final antibody quality profile. Our results demonstrate the feasibility of PAT tools in a manufacturing setting to aid bioprocessing and support the development of advanced feeding strategies to control product quality.
Our research addressed the need to better understand how complex input variables within the biomanufacturing process affect product quality. CQAs are physical, chemical, biological, or microbiological properties that must be within specified ranges to ensure the desired product quality, and in the case of mAb products one particularly important CQA is the distribution of product glycoforms. [4][5][6][7] N-linked glycosylation may affect various therapeutic properties of the antibody, from effector function, immunogenicity, stability, and clearance rate. 8 IgG antibodies feature conserved N-glycosylation sites at Asn 297 in the constant region of the Fc heavy chain, which influence effector binding to downstream molecules such as Fcγ receptors that mediate antibody-dependent cellular cytotoxicity (ADCC), a critical mechanism for many biotherapeutics. 9 Even for antibodies where the mechanism of action is not through ADCC, an abnormal glycan profile can affect efficacy through alteration of the immunogenicity, stability, or clearance rate.
The mechanisms by which glycoforms can affect antibody product performance have been experimentally characterized, such as how fucose sterically inhibits interaction with the Fcγ receptor and reduces ADCC-mediated activity. 10 Conversely, how the cell growth and media conditions within the bioreactor impact the glycan profile are not well understood. Bioreactor culturing conditions have been established in their ability to change product quality outcomes: specifically temperature, pH, and agitation rate have been shown to have small effects on the final glycan profile. 11 However, the role of overlooked nutrients in the media such as sugars, amino acids, and metals have been only minimally characterized with respect to how they can affect protein production and efficacy. 12 Amino acids are crucial for antibody production since they are the building blocks of the primary protein sequence. 13 As such, supplying the cells with sufficient levels of amino acids is needed to avoid protein anomalies, such as amino acid misincorporation which changes the primary sequence of the protein. 14 Khetan et al discovered that when asparagine was depleted from the bioreactor media, serine was misincorporated in asparagine's place at rates as high as 3%. Amino acids also serve as potential energy sources for the cell, such as glutamate via glutaminolysis. Thus, it is possible that amino acid depletion could affect more than the primary sequence of the protein, such as the posttranslational modifications and the chemical structure of the molecule. Regarding these considerations, we developed an at-line amino acid quantification method that may be a useful tool to monitor upstream bioprocessing. Specifically, we are studying the possibility that replenishment of selected amino acids could help maintain the product quality of the bioreactor cell culture. Altered supplementation of amino acids in cell culture can modify the critical quality attributes of the antibody, including glycosylation. 15 These changes are not limited to antibody products: an increase of recombinant human erythropoietin with decreased sialylation has been attributed to the addition of amino acids as feeds during cell culture. 16 In a study of tissue plasminogen activator produced in ammonium-stressed CHO cells, the addition of amino acids leads to increased glycosylation site occupancy by mature glycoforms. 17 However, contemporary studies have not applied these supplementation strategies with real time quantification which would help to better target and understand specific relationships between the nutrient addition and product quality.
Current amino acid analysis is divided into two approaches: derivatized and underivatized. 18 Derivatized methods require chemical modification of the amino acids so they can be detected via spectroscopy, typically with an ultraviolet (UV) or fluorescence detector. Underivatized amino acid analysis does not require a chemical reaction, lending itself to be more easily implemented in PAT development due to simpler sample preparation.
Underivatized analysis requires mass spectrometry since native amino acids cannot currently be differentiated directly with spectrometric methods. We present here an underivatized at-line method using mass spectrometry for quantification of amino acids that allowed us to assess depleted amino acids in near real time, using this information to determine which amino acids are consumed most quickly in the bioreactor and the relative importance of each amino acid toward the resulting glycan profile. before being added to the bioreactors. During the process, when glucose was less than 1 g/L and glutamine was less than 1 mM, glucose and L-glutamine were supplemented as a bolus to bring back the levels to 5 g/L and 8 mM, respectively.

| Seed train
This experiment used a previously described recombinant CHO DG44 cell line that expresses a model chimeric IgG1. 19 For all the reactors subjected to feed strategies 1 through 3, 5, and 6, the seed train was started by thawing 1 vial (1 ml) of banked cells (7 × 10 7 viable cells/ml) into 200 ml of 37 C CD OptiCHO media (Life Technologies, A11222) supplemented with 1X Penicillin/Streptomycin (Corning, 30-001-Cl) and 8 mM L-glutamine (Corning, 25-005-CV) in a 1 L spinner flask (Corning, 4500-1L). Incubator CO 2 was set at 5%, temperature was set at 37 C and stir speed was kept at 65 RPM. Scale-up procedures were used until 4 1 L spinner flask cultures with >2 × 10 6 cells/ml were created.
Alternatively, for the reactors subjected to feed strategies 4 and 7, the seed train was started by thawing a 1 ml vial of banked cells (7 × 10 7 viable cells/ml) into 50 ml of 37 C media contained in a 125 ml disposable shake flask (Corning, PBV12-5). Upstream scale-up procedures were followed until the cells were inoculated into a 2 L cell bag (GE Healthcare, CB0002L10-31) on a GE ReadyToProcess WAVE™ 25 rocker system at 1 × 10 6 cells/ml in 750 ml. Gas mix flow rate was set to 0.3 L/min, pH was set to 7.1 (CO2/base-controlled), DO was set to 50% air saturation, and rocking speed and angle were 20 rpm and 6 , respectively. After 1 week, the cells reached 5.7 × 10 6 cells/ml and were prepared for inoculation.
On the day of inoculation, cell culture fluid from either the spinner flasks or WAVE bag was transferred to disposable, sterile 250 ml conical tubes and centrifuged at 800 rpm for 10 min. Supernatant was slowly decanted and cell pellets were resuspended in 5 ml of fresh, prewarmed media and pooled for bioreactor inoculation.

| Bioreactor processing conditions
A BIOSTAT B-DCU II (Sartorius Stedim Biotech, Goettingen, Germany) bioreactor system with two to four 5 L vessels was run in batch or fed-batch mode for 5-10 days (120-240 hr). Set points for the culture processes (Table 1) were maintained automatically by the controller.
Culture foaming was reduced when foam reached foam sensors on the head plate of the reactor and additionally by manual addition as needed using a 3% EX-CELL gamma-irradiated antifoam emulsion (Sigma-Aldrich, 59920C-1B). Each reactor was equipped with a FUTURA 12 mm biomass probe with FUTURA head amplifier (ABER Instruments, 2330-00) to obtain viable cell density measurements in real-time. Bolus feeds of glucose, glutamine, and amino acid blends were supplemented as shown in Table 2.

| Cell counts and nutrient analysis
Samples were run on the BioProfile FLEX analyzer (Nova Biomedical) either automatically using the BioProfile FLEX on-line autosampler or taken manually to measure viable cell density (VCD), pH, glutamine, glucose, lactate, glutamate, and ammonium. Additionally, cell count was monitored using a biomass probe as mentioned previously. Samples were taken anywhere from every 4 hr to once daily, depending on the run. Additional sample volume obtained for amino acid analysis was clarified by centrifugation at 300g for 5 min at 4 C and sterile filtered using 0.22 μm PVDF filters. Cell-free samples were frozen and stored at −20 C until future analysis.

| Downstream mAb purification
The methods used to purify and concentrate the antibody produced by the bioreactors were described previously. 20 T A B L E 1 Summary of culture process set points and controls

| Glycan characterization
The GlycoWorks method used to label and quantitate glycans isolated from the harvested antibodies was described previously. 20 The glycan profiles were only evaluated for the final harvest product for each bioreactor that was run. In principal component analysis, the model represents the

| RESULTS AND DISCUSSION
Initially, we ran multiple bioreactors to serve as test runs and establish baseline conditions prior to experimentation with amino acid supplementation strategies ( Table 2). One bioreactor was run in batch mode  A representative example of the nonderivatized amino acid concentrations measured by mass spectroscopy is displayed in Figure 3.
As shown in Figure 3a  blends, can be found in Table 2.
As shown in Figure 2c  more surprisingly when amino acid blend A was added in feed strategy 6 ( Figure 2e) and the VCD was already decreasing, the amino acid addition was able to reverse this loss. This indicates that the blend containing Tyr, Cys, Pro, and Asn has the beneficial effect of prolonging bioreactor production time and increasing the VCD. This combination of amino acids, however, did not have any effect when it was added as a second supplementation event (amino acid blend A + B) that occurred 50+ hours after the first amino acid supplementation.
Since past studies have shown that changing the growth conditions of the cells can affect product quality, the next step was to characterize the protein product and determine if any significant changes had occurred. To address this, the final antibody product from these cultures was collected, purified, and analyzed to determine the charge variant profile, size variant profile, and glycan profile.
Charge variant and size variant analysis using microfluidic electrophoretic approaches performed on the product antibody revealed small differences between feeding strategies 4 and 6 (which used amino acid blends A and A + B) and feeding strategies 5 and 7 (which used amino acid blends B, C, and D). Amino acid blend A, which was responsible for the boost in VCD in the bioreactors, also produced a small decrease in basic species (Table 3). Another important distinction for the antibody products produced between by the amino acid blends was the formation of a second peak (labeled as Peak 2 in Tables 4 and 5) during purification that was only observed in cultures supplemented by amino acid blend A. This secondary peak was isolated and analyzed alongside the primary peak for size analysis and glycan analysis. Size variant analysis of the intact mAb product revealed that the products were comparable (Peak 1), while the secondary peak was likely containing some level of contamination (Table 4) as size variant analysis is typically used as a measure of purity.
Reduced size variant analysis was performed to get a better understanding of how the light and heavy chains constituting the mAb might have been affected by different amino acid blends during upstream processing ( Table 5). The main peak showed no differences due to the amino acid blends as the percentages of light and heavy chain were not changed between feeding strategies. However, the secondary peak for feed strategies 4 and 6 affirmed that the species in this peak were different, with a significantly lower amount of heavy chain than in the main peak and a large amount of unknowns.
Altogether, the charge and size variant analyses supported that there were small differences in the charge variant profile that likely would not have a significant effect on the therapeutic properties.
Likewise, the amino acid supplementation strategies had small, but statistically significant, effects on product quality from the standpoint of the glycan profile ( Figure 4). The amino acid feeds that resulted in increased VCD and longer batch age performance (Feed strategies 4 and 6) also resulted in higher amounts of high mannose species production and lower amounts of terminal galactosylation (G1F and G2F).
The main peaks and secondary peaks had highly similar glycan profiles, so only the main peaks are shown in Figure 4. Collectively, our protein structural analysis illustrates the importance of understanding how process parameters and bioreactor nutrients can affect product T A B L E 3 Charge distribution of purified antibody from the amino acid supplemented bioreactor run

Note:
The average apparent size of the light chain is~29 kDa and the heavy chain is~67 kDa. Peak areas were determined using GXII reviewer and represent the mean of three technical replicates on a mass basis. Error bars represent ±1 SD of the mean.
quality, as in this case where a favorable increase in VCD performance results in a potentially less favorable glycan profile outcome (with less galactosylation and increased high mannose glycoform amounts).
Due to the impact of amino acid blend supplementation on the glycan profiles of the produced antibodies, we used modeling approaches to better understand how amino acid consumption patterns are correlated to glycan outcomes. Univariate methods alone would not allow us to discover trends in the data that associate amino acid consumption patterns with glycan outcomes, so we employed multivariate approaches. We used batch data from all seven feeding strategies that we performed in this study to increase our statistical  We used five preliminary bioreactor runs to establish a baseline for amino acid consumption, both in batch mode and fed-batch mode feeding strategies. Our rapid amino acid quantification method allowed us to analyze a large number of media samples from these runs and narrow down the amino acids of interest to eight: Tyr, Cys, Pro, Asn, Met, His, Trp, and Thr. We put these species together for multiple blend strategies and supplemented them into the bioreactors at critical points of the cell growth and nutrient consumption curve to observe their effects on the growth profile. We discovered that addition of Tyr, Cys, Pro, and Asn (blend A) resulted in a significant boost in VCD in the cultures, in one case even reversing a culture where the VCD was decreasing toward death based on historical trajectories.
The baseline cultures that we performed before the experimental amino acid supplementation bioreactors indicate that once the VCD begins to decrease, the general health of the cultures has suffered F I G U R E 6 Time dependence of the amino acid correlation with final glycan profile outcomes. The most significant amino acids were mapped over batch age time (hours) to reveal time-dependent trends for correlation between each constituent and glycan profile. While most of the amino acids increased with significance over time, tyrosine decreased over the final 48 hr when the first amino acid supplementation events occurred (indicated by the viability) and cannot recover; despite this, the culture with feed strategy 6 ( Figure 2e) was able to reach a new VCD maximum after amino acid blend A was administered. While it is unclear which of these four amino acids are responsible for the VCD effect, we are considering future studies to narrow down the amino acid responsible.
While amino acid supplementation may successfully prolong the runtime of a bioreactor, these strategies might have unknown effects on product quality. To address this, we analyzed the final mAb product from these bioreactor runs to determine their charge variant, size variant, and glycan profiles. We discovered that the amino acid blends did have small effects on the charge variant and glycan profiles. In the case of the glycan profile, we observed a small increase the percentage of high mannose glycoforms, which are generally undesirable. This is due to the faster clearance of these glycoforms as therapeutics, lowering their beneficial impact. 22 The increase of these species is the result of incomplete processing of the glycan in the Golgi and is found when the culture is stressed, such as from high media osmolality or long culture times. In the final part of our study, we used PCA to see if any amino acids could be linked to the glycan outcome based on their consumption patterns by using data from all nine bioreactor runs. By comparing the cultures where amino acids were depleted and not replenished against those where they were, we can identify specific amino acid time points that are correlated with the glycan profiles.
Using this approach in a time-course-oriented fashion, we established specific amino acids and time points which were most highly correlated. From all the amino acids analyzed, we narrowed down our pool of candidate species to Tyr, Pro, His, Trp, Arg, Lys, and Phe. Of these, four were among those that were actively supplemented in our blends (Tyr, Pro, His, and Trp). Our analysis revealed that the loadings for these amino acids changed over the course of the run, with most of them decreasing over the first 24 hr and increasing thereafter. Tyrosine was an interesting outlier from this trend, with a significantly lower loading over the final 40 hr of the culture life. We conclude that the timing of the amino acid supplementation between the 80-hr and 120-hr timepoints were a strong contributing factor to the trend lines for loadings correlations that we observed, such as the decrease discovered at the 120-hr timepoint for tyrosine and the increases in loadings for Arg, His, Lys, Trp, and Pro over the same period.
The media within the bioreactor is a complex and continuously changing environment that has substantial effects on the cultured cells and their produced proteins. Due to the high number of nutrient variables that are present in the bioreactor that can all potentially affect product quality, future studies that utilize controlled and scheduled feeds (glucose/glutamine and amino acids) along with multivariate analysis will be necessary to more definitively establish the effect that amino acids have on the final protein. Multivariate analysis using data from all of the PAT instruments we have available (culture variables like temperature, pH, titer, and media variables such as glucose, glutamine, and lactate) along with our newly developed capabilities in measuring near real-time amino acid concentrations will allow us to create a more comprehensive model and properly establish amino acid dependent effects on productivity and product quality.
With the increasing push toward process understanding and Administration.

DISCLAIMER
The article reflects the views of the authors and should not be construed to represent FDA views or policies.