Impacts on product quality attributes of monoclonal antibodies produced in CHO cell bioreactor cultures during intentional mycoplasma contamination events

Abstract A mycoplasma contamination event in a biomanufacturing facility can result in costly cleanups and potential drug shortages. Mycoplasma may survive in mammalian cell cultures with only subtle changes to the culture and penetrate the standard 0.2‐µm filters used in the clarification of harvested cell culture fluid. Previously, we reported a study regarding the ability of Mycoplasma arginini to persist in a single‐use, perfusion rocking bioreactor system containing a Chinese hamster ovary (CHO) DG44 cell line expressing a model monoclonal immunoglobulin G 1 (IgG1) antibody. Our previous work showed that M. arginini affects CHO cell growth profile, viability, nutrient consumption, oxygen use, and waste production at varying timepoints after M. arginini introduction to the culture. Careful evaluation of certain identified process parameters over time may be used to indicate mycoplasma contamination in CHO cell cultures in a bioreactor before detection from a traditional method. In this report, we studied the changes in the IgG1 product quality produced by CHO cells considered to be induced by the M. arginini contamination events. We observed changes in critical quality attributes correlated with the duration of contamination, including increased acidic charge variants and high mannose species, which were further modeled using principal component analysis to explore the relationships among M. arginini contamination, CHO cell growth and metabolites, and IgG1 product quality attributes. Finally, partial least square models using NIR spectral data were used to establish predictions of high levels (≥104 colony‐forming unit [CFU/ml]) of M. arginini contamination, but prediction of levels below 104 CFU/ml were not reliable. Contamination of CHO cells with M. arginini resulted in significant reduction of antibody product quality, highlighting the importance of rapid microbiological testing and mycoplasma testing during particularly long upstream bioprocesses to ensure product safety and quality.


| INTRODUCTION
Mycoplasmas are small (0.1-0.3 µm diameter) bacteria without cell walls, allowing for penetration of the typical 0.1-0.22 µm sterilizinggrade filters used in scientific research and biomanufacturing (Razin, 2006). Mycoplasma contamination risk in biomanufacturing is lower than the risk in academic research laboratories where it is still commonplace to use media supplemented with animal-derived serum components (Drexler & Uphoff, 2002), but even with the use of chemically derived media as the norm (Fletcher & Harris, 2016), contamination risk is still present via other medium additives and manual manipulations of cell lines. Mycoplasma incidence rates are estimated to be 15-35% in banked cell lines and 0.44-6.70% within the biopharmaceutical industry, depending on the detection assay used (Armstrong, Mariano, & Lundin, 2010;Chandler, Volokhov, & Chizhikov, 2011;Drexler & Uphoff, 2002). Of these contamination events, 95% of the mycoplasma species identified were one of the following: Mycoplasma orale, Mycoplasma arginini, Mycoplasma hyorhinis, Mycoplasma fermentans, Mycoplasma hominis, or Acholeplasma laidlawii (Drexler & Uphoff, 2002;Kljavin, 2007). Despite comprehensive control strategies (Guilfoyle et al., 2013;Quality Risk Management Q9, 2005;Rosenberg et al., 2011), mycoplasma can potentially contaminate a bioprocessing scheme through other cell culture medium components (Drexler & Uphoff, 2002;Kljavin, 2007Kljavin, , 2011Windsor, Windsor, & Noordergraaf, 2010) or during manual manipulation of cell lines such as cell banking and cell line development (Nikfarjam & Farzaneh, 2012). Some species of mycoplasma such as Acholeplasma laidlawii (Windsor et al., 2010) cause occult contaminations of cell cultures with minimal visible changes to cell health or cell culture performance, while other mycoplasma contaminations can subtly alter the culture performance through potential mechanisms including competition for culture nutrients and alteration of the host cell expression profile (Rottem, 2003). Although there are differences among species, mycoplasma tends to grow significantly slower than other bacteria, with doubling times ranging anywhere from 1 to 9 hr (Drexler & Uphoff, 2002). The small size, slow growth, and sometimes subtle effects of mycoplasma make these bacteria particularly ideal for adopting early detection strategies in biomanufacturing.
The ability of mycoplasma to evade detection and grow in mammalian cell cultures creates risk in some ways more like viral contamination than typical bacterial contamination. Mycoplasma contamination, much like viral contamination in the biotechnology industry, although rare, can lead to millions of dollars spent for investigation and decontamination measures, drug shortages, and damage to the public's confidence in the manufacturer (Barone et al., 2020). Usually when biopharmaceutical manufacturers find mycoplasma in their cultures, they immediately decontaminate after taking a limited number of culture samples for specification and test raw material samples to trace the source of the contamination (Angart, Kohnhorst, Chiang, & Arden, 2018). To understand how a mycoplasma contamination event may appear in a typical biomanufacturing scheme and provide knowledge on how process monitoring may identify a mycoplasma contamination event in the manufacturing environment, our group developed models of early-stage process and late-stage process bioreactor culture contamination events using Mycoplasma arginini (M. arginini) and Chinese hamster ovary (CHO) cells grown in serum-free medium expressing a model immunoglobulin 1 (IgG1) product. We examined the growth kinetics of M. arginini, a species selected due to its perceptible effects on CHO cells in pilot coculture shake flask studies Wang et al., 2017), in a controlled bioreactor environment and the effects on CHO cell culture performance and process parameters (Fratz-Berilla et al., 2019). We investigated the effects of mycoplasma presence on CHO cell health, productivity, culture metabolism, and process parameters compared to uninfected controls and observed effects 2-6 days after M. arginini introduction to the culture, depending on the viable cell density and perfusion conditions at the time of contamination (Fratz-Berilla et al., 2019). In this report, we investigate the changes in IgG1 product quality produced by CHO cells induced by the M. arginini contamination events. We observed changes in charge variant profiles, purity, and glycan patterns that were further modeled using principal component analysis (PCA) to explore the relationships among M. arginini contamination, CHO cell growth and metabolites, and IgG1 product quality attributes. Finally, partial least square (PLS) models using NIR spectra collected every 2 min from the bioreactor cultures were used to establish predictions of high levels (>10 4 colonyforming unit [CFU/ml]) of M. arginini contamination. | 2803 inoculated a recombinant CHO DG44 cell line that expresses a model chimeric IgG1 in a ReadyToProcess WAVE™ 25 rocker (GE, 28988000) operated in dual mode with two 2-L single-use cell bags (GE, CB0002L10-3; 1 L maximum operating volume) containing porous polyethylene-based perfusion filters. M. arginini was spiked into bioreactors either early in culture, before the start of perfusion when CHO cells were~2 × 10 6 cells/ml, or late in culture, during perfusion of 2 L/day when CHO cells were 9-12 × 10 6 cells/ml and compared to control uncontaminated bioreactors (Fratz-Berilla et al., 2019). The mycoplasma spike for the bioreactor was prepared from a frozen stock of M. arginini in 45% glycerol thawed at room temperature in a laminar flow hood using a sterile syringe and disposable pipette basin. The mycoplasma stock was mixed with 3 ml media (OptiCHO/ L-Glutamine + soy hydrolysate) warmed to 37°C to seed a final target concentration of 10 1 (low spike) or 10 2 -10 3 (high spike) CFU/ml in 1 L bioreactor volume. A negative control containing an identical volume of media and supplements only was also prepared for the uncontaminated bioreactor and control bioreactor cultures were confirmed to be uncontaminated during the entirety of the run. Each reactor was connected to a flow cell of a near-infrared (NIR) spectrometer (Sartorius Stedim) via the perfusion filter line in which culture material was continuously pumped either back into the bioreactor (batch mode) or into the perfusate collection carboy (perfusion mode). Three 14-19 day runs of two bioreactor cell bags, as described in Table 1, were completed.

| Amino acid characterization by liquid chromatography-mass spectrometry
The crude bioreactor medium was centrifuged and passed through a 0.22-μm polyethersulfone filter (MilliporeSigma, SLGP033RS). A perchloric acid cleanup was used to remove protein and particulate matter, which involved mixing filtered bioreactor medium with 0.4N HClO 4 (Sigma-Aldrich, 311421) at a 1:1 ratio and centrifuging at 1,962g for 5 min at RT. The clarified medium was collected to be analyzed by liquid chromatography-mass spectrometry (LC-MS). Model for late-stage high-level bioreactor contamination during perfusion operation Abbreviation: CFU, colony-forming unit.
a Perfusion of the bioreactor was stopped when CHO cell viability fell below 50%.

| Antibody purification
The perfusate obtained each day was purified using two AKTA Avant

| Protein aggregation (size exchange chromotography multiangle light scattering)
The aggregate analysis was performed on a UPLC (Agilent 1290 Infinity I, G7120A) connected to a Multiangle Light scattering detector (Wyatt μDawn) and refractive index detector (Optilab UT-rEX).
Size exclusion chromatography was performed with a TSKgel UP-SW3000 4.6 mm ID × 30 cm/L (Tosoh Biosciences, 003449) and 1 × PBS (Corning, 46-013-CM) as the mobile phase operating at a flow rate of 0.4 ml/min. IgG1 samples were diluted to a concentration of 2 mg/ml in Tris-acetate buffer and 5 μl of the sample was injected per run. Peak quantitation was performed in Astra using the unit variance (UV) absorbance signal from the Agilent DAD detector at 280 nm and average molecular weight determinations were calculated for the same peak areas, using a refractive index increment (dn/dc) of 0.185.

| Purity and fragmentation (micro-capillary electrophoresis-sodium dodecyl sulfate)
Reduced and nonreduced size analysis was performed with a Lab- Dataset observations were projected onto score and loadings plots by SIMCA 13.0 for analysis.

| PLS regression
PLS regression was conducted by SIMCA 13.0 to demonstrate the dependence of mycoplasma concentration on process spectra and to further elucidate the prediction capability of the acquired process spectra on mycoplasma presence. PLS regression has previously been applied to NIR prediction methods, and these strategies are well documented (Feng, Wu, & Zeng, 2015). To summarize, loadings matrices of both spectra X and mycoplasma Y are transposed and multiplied by each corresponding score, and then added to residual matrices of variance not described by the principal components.
Actual  3.3 | Contamination of CHO cell bioreactor cultures with M. arginini results in increased high mannose species, especially mannose 6 and higher Galactosylation varied from batch-to-batch and within each batch, with the largest differences being between the percentage of G0F and G1F species. G0F made up anywhere from 34% to 68% of the total glycans, and G1F varied from 15% to 49% during control operations (Figure 3). Supplementations of CHO cultures with high levels of iron and manganese are known to increase galactosylation (Crowell, Grampp, Rogers, Miller, & Scheinman, 2007; F I G U R E 2 Purity and aggregation in mAb purified from control perfusion bioreactors and those contaminated with Mycoplasma arginini by perfusion day. Data represent the mean of three technical replicates and error bars represent ±1 standard deviation of the mean. Statistical significance is defined by two-tailed paired t test between corresponding control bioreactor perfusion days (Day 2-High and Day 3-High) or perfusion day before mycoplasma spike (Day 9-Low and Day 12-High) (*p < .01) [Color figure can be viewed at wileyonlinelibrary.com] Gramer et al., 2011), and enhanced supplementation of copper and zinc have shown to have a significant impact on production and product quality as well (Graham, Bhatia, & Yoon, 2019;Graham et al., 2020;Yuk et al., 2015). However, we could not identify any overt trends between the levels of iron and manganese (Figures S1 and S2) and the amount of galactosylation. The high variability between and within batches makes it difficult to interpret the potential causes of change in galactosylation profiles.
The presence of high mannose species have been associated with periods of CHO cell stress (Fan et al., 2015;Powers et al., 2019;Villacres, Tayi, Lattova, Perreault, & Butler, 2015), increased specific productivity (Zalai et al., 2016), and high concentrations of manganese and limited glucose (Surve & Gadgil, 2015), and thus we ex- | 2809 for those with 6 or more mannose residues (Figures 4 and 5). Interestingly, levels of mannose 5 species did not apparently increase in Day 2-High and Day 3-High (Figure 4), but combined levels of mannose 6-8 species trend higher in contrast to the same perfusion days of Day 2-Control and Day 3-Control, respectively ( Figure 5). For Day 9-Low and Day 12-High, some increases in mannose 5 species are apparent (Figure 4). Increases in levels of mannose 6-9 species reach 16.2% and 17.4% of the total glycan profile for Day 9-Low and Day 12-High, respectively, by the final day of perfusion ( Figure 5).
These high mannose species containing higher than 5 mannose residues are produced early in the oligosaccharide modification scheme that occurs in the Golgi, indicating that mycoplasma contamination is causing significant enough intracellular disturbances to disrupt the glycan reaction pathways at the beginning stages. Moreover, the presence of ammonia accumulation has long been known to be a byproduct of some mycoplasma presence via arginine depletion (Matsuura, Seto, & Watanabe, 1990). Although 3.5 | NIR spectra PLS model indicates low and high extreme mycoplasma presence PLS models were constructed in SIMCA 13.0 and principal components were added until greater than 90% of variability was captured by each model (Table 2). In both Day 12-High and Day 9-Low models ( Figure 7), a cluster of points of no mycoplasma presence is within the predicted mycoplasma range from −2 to 2 log (CFU/ml). Moreover, both models predict saturated mycoplasma concentrations at a range of approximately 4 to 7.5 log (CFU/ml). To explore the effects of batch age on the mycoplasma models, separate batch age prediction models were constructed using the same NIR spectra. The more evident linearity and lack of polarized clustering on the batch age plots provide insight into the reliability of the saturated mycoplasma clusters in the prediction models.
Accurate prediction of mycoplasma concentrations by these two NIR models lacks feasibility based on Q 2 prediction and RMSEV values (Table 2). However, the two clusters representing both no mycoplasma presence and mycoplasma saturation do provide an avenue to utilize the NIR prediction models for low and high alarm signaling.
As both models dependably reveal the presence of saturated F I G U R E 6 PCA loadings plots of Day 9-Low and Day 12-High. Culture data included are concentration of mycoplasma (red), IgG titer (gray), nutrients (glucose and glutamine-green), arginine (yellow), and waste (ammonium, lactate, and glutamate-purple those with 6 or more mannose residues. We used MVDA to more thoroughly dissect and find relationships among our complex datasets containing information on CHO cell growth, mycoplasma growth, medium nutrients and waste, trace metals, CQAs, and NIR spectra. PCA allowed us to confirm some biologically relevant relationships but the considerable batch-to-batch variability diluted what may be more subtle correlations that would not be found using only univariate analysis on the individual factors. However, using PCA, we observed that mycoplasma presence did have a positive correlation with acidic charge variants and high mannose species along the first principal component, but the correlations were much stronger with Day 12-High than in Day 9-Low, where mycoplasma had a less significant impact along the first principal component, possibly indicating that the lower mycoplasma spike concentration may have influenced the culture dynamics more than could be previously deciphered by univariate analysis, or that the complexity of the culture was not fully captured in the datasets collected. Using PLS, we found that NIR spectra predicted mycoplasma contaminations at high (4-8 log [CFU/ml]) concentrations, but some false positives occur at low mycoplasma concentrations.
With the growing desire to use spectral data, such as Raman and NIR, along with MVDA and modeling to monitor and control upstream bioprocesses, it may be important to understand what a contamination event, especially an otherwise occult contamination F I G U R E 7 PLS models of NIR spectra to predicted mycoplasma contamination in Day 9-Low and Day 12-High. NIR spectra can predict mycoplasma contaminations at high (4-8 log [CFU/ml]) concentrations, but some false positives occur at low mycoplasma concentrations. The batch age predictions indicate a strong linear correlation and lack of polarized clustering between the predicted and observed batch age, which provides evidence of reliability of the saturated mycoplasma clusters in the prediction models. CFU, colony-forming unit; NIR, near-infrared; PLS, partial least square