Real-time characterization of mammalian cell culture bioprocesses by magnetic sector MS

Mammalian cell culture processes were characterized upon the analysis of the exhaust-gas composition achieved through the on-line integration of a magnetic sector MS analyser with benchtop bioreactors. The non-invasive configuration of the magnetic sector MS provided continuous evaluation of the bioreactor's exhaust gas filter integrity and facilitated the accurate quantification of O2 and CO2 levels in the off-gas stream which ensured preserved bioreactor sterility prior to cell inoculation and provided evidence of the ongoing cellular respiratory activity throughout the cultures. Real-time determination of process parameters such as the Respiratory Quotient (RQ) allowed for precise pin-pointing of the occurrence of shifts in cellular metabolism which were correlated to depletion of key nutrients in the growth medium, demonstrating the suitability of this technology for tracking cell culture process performance.


Introduction
The complexity of mammalian cell culture processes for biopharmaceutical manufacturing has resulted in the rapid development of Process Analytical Technology (PAT) tools aimed at improving batch-to-batch reproducibility through the implementation of real-time process monitoring techniques. [1][2][3] In particular, on-line platforms can provide rapid access to key performance indicators (KPIs) and critical process parameters (CPPs) which can assist the operator's decision-making process and facilitate the timely implementation of required corrective actions. 4,5 Cells have the ability to adjust their metabolic ux depending on specic culture conditions 6,7 and information on the consumed energy source can be obtained by monitoring shis in respiratory proles. [8][9][10][11] On-line measurement of key respiratory parameters, including the Carbon Dioxide Evolution Rate (CER), Oxygen Uptake Rate (OUR) and the ratio of these two parameters, the Respiratory Quotient (RQ), facilitated the implementation of adaptive feeding control strategies aimed at increasing yields in fungal [12][13][14] and bacterial 15 fermentation processes. However, the implementation of RQ as a practical process control tool in mammalian systems has been limited by the difficulties associated with its accurate measurement, resulting in a broad range of reported values which varied depending on the cell line type and cultivation conditions. 11 The high solubility of CO 2 and the presence of bicarbonate buffer in the culture medium can introduce errors in the CER measurements. 16,17 Furthermore, headspace dilution effects may also introduce a delayed response in the exhaust gas signal which can potentially result in erroneous CER, OUR and RQ measurements. 18,19 Among all measurable respiratory parameters, OUR values have been preferentially implemented for the development of adaptive feeding control strategies in mammalian cell culture processes 20-23 since they can be accurately determined through global mass balance, liquid phase balance and dynamic method approaches as a result of poor O 2 solubility 24,25 while mathematical models have been reported to improve the accuracy of CER measurements by accounting for the liquid-gas transfer of CO 2 from bicarbonate as a function of pH variations in the culture medium. 19,26 However, the main bottleneck for accurate off-gas characterization in mammalian cultures is related to the low cell densities, resulting in relative small changes in O 2 and CO 2 concentrations which are difficult to quantify. 26 Electronic noses based on metal oxide sensors have been historically adopted for on-line characterization of mammalian culture off-gas as exposure to volatile species produced a variation in the sensor's electrical response. 8 The low sensitivity, lack of selectivity and the need for chemometric interpretation of data rendered this technology inefficient as a PAT tool. Paramagnetic or electrochemical detectors were further implemented for on-line O 2 quantication in mammalian cell cultures while CO 2 measurements were obtained through potentiometric or non-dispersive IR (NDIR) detectors 17,19,26 which can be congured to monitor a single gas sample stream, either the inlet or outlet gas line from the bioreactor, or can perform analysis of alternated inlet and outlet streams through line multiplexing. Simultaneous monitoring of multiple bioreactors required during process development operations would therefore require an unpractical equivalent number of standalone detectors. Despite showing promising performance for real-time mammalian bioprocess monitoring, paramagnetic and NDIR detectors are limited by slow sampling speed, narrow operating range and low accuracy, which is typically AE0.2% (v/v) for O 2 and AE2% (v/v) for CO 2 . 17,19 Mass spectrometers can provide improved sensitivity for offgas analysis which facilitates the detection of physiological changes in animal cell bioprocesses characterized by low respiratory activity. 27 Quadrupole analysers have been implemented for real-time CER-OUR determination in CHO cell cultures 20 and RQ estimations in hybridoma cultivations. 28 In this study, we evaluated a magnetic sector MS gas analyser as a tool for real-time monitoring of CHO cell cultivation processes. Despite the previous integration of magnetic sector MS for monitoring hybridoma 29 and CHO cell cultures, 27 recent developments have occurred particularly with regard to sampling methodologies which facilitate the analysis of multiple gas streams within a matter of minutes using a single MS analyser, thereby potentially allowing simultaneous monitoring of multiple bioreactors which traditionally limited the integration of paramagnetic and NDIR exhaust-gas analysers during process development stages. Furthermore, magnetic sector analysers provide improved stability relative to quadrupole counterparts, 30 featuring characteristic at top peaks which do not yield variations in signal intensities upon occurrence of small mass dris, therefore being particular advantageous for long-term process monitoring applications. The performance of the MS gas analyser was assessed throughout multiple bioprocessing stages, from pre-inoculation of the bioreactor to culture harvest, by evaluating batch and fed-batch mammalian cell cultures performed in benchtop bioreactors which were subjected to alteration in processing parameters aimed at establishing the suitability of magnetic sector MS as an effective real-time process monitoring technology.

Determination of respiratory parameters by magnetic sector MS
A Prima BT magnetic sector MS benchtop system (Thermo Fisher Scientic, Winsford, UK) operated through the GasWorks 3.0 soware package was implemented for direct sampling of headspace off-gas from mammalian CHO cell culture processes. The initial set up of the MS analyser involved running a 100% (v/v) He stream (BOC Gases Ireland, p/n 379-V) to rule out the presence of leaks in the transfer lines which could result in MS contamination, followed by 20% (v/v) CO 2 in Ar (BOC Gases Ireland, p/n 225765-V-C) and a standard calibration air mixture comprising of 1% (v/v) Ar, 10% (v/v) CO 2 , 22% (v/v) O 2 and 67% (v/v) N 2 (BOC Gases Ireland, p/n 149624-AV-G). A single MS calibration was performed due to the high signal stability provided by the instrument which provided high measurement precision, determined for O 2 to be 0.002-0.003% (v/v) over a 7 day monitoring period. Gas streams were introduced into the MS through the 16-port Rapid Multi-Stream Sampler (RMS), an optically-encoded rotating stream selector and analysed for CO 2 , O 2 , N 2 and Ar with sample measurements taken every 30 seconds for each stream. Gas ow-rates reaching the detector were obtained through a sensor positioned within the RMS which provided measurements with an accuracy of AE12 mL min À1 . The RMS was maintained at 60 C during operation. The GasWorks soware calculated the CER, OUR and the RQ values determined from their ratio (not reported) as dened in eqn (1)-(3) respectively, upon characterization of the gas supplied to the bioreactor (inlet) and of the exhaust-gas released from the vessel (outlet), as illustrated in Fig. 1.
The measurement of N 2 which is inert relative to the cellular respiration process since it is not consumed or produced, facilitated the denition of the balance shown in eqn (4).
The limited accuracy of ow-meters is known to introduce errors in the calculation for RQ. Hence, eqn (4) was resolved for ow-out, yielding eqn (5), which facilitated the rearrangement of eqn (1) and (2) to allow the determination of the reported ow-corrected CER-OUR values, dened by GasWorks as CDC and OXC as shown in eqn (6) and (7) respectively, and of the RQ values as illustrated in eqn (8) which were reported for the experiments performed.
The respiratory proles based on the CDC, OXC and RQ values for the batch culture identied as B-1 were reported upon averaging measurements within 30 minute sampling intervals. For the remaining cultures described, averaged values from 6 hour sampling intervals were reported instead. Colour coded error bars were included in the graphical representations which were representative of the obtained standard deviation.

On-line integration of magnetic sector MS analyser with benchtop bioreactors
Two distinct congurations were evaluated for the on-line integration of the magnetic sector MS analyser with 3 L and 7 L total volume Thermo Fisher HyPerforma glass bioreactors

Anal. Methods
This journal is © The Royal Society of Chemistry 2020 Analytical Methods Paper The second MS-bioreactor conguration, dened as V2 and shown in Fig. 1(B), involved the analogue transfer of individual gas ow-rate values from the Thermo Fisher G3Lab system auxiliary ports to GasWorks which calculated in real-time the composition of the inlet gas stream supplied from the ratio of these parameters and from the composition of air which was determined by MS upon running a gas transfer line directly from the air supply to the detector. The ow path for exhaustgas analysis remained unaltered in the V2 conguration.

Bioreactor sterility assessment
A 7 L total volume bioreactor vessel was subjected to porous steam-sterilization at 121 C followed by aseptic lling with 3 L of liquid BalanCD growth A medium (Irvine Scientic, CA, USA) containing 2.2 g L À1 sodium bicarbonate. An agitation rate of 188 rpm was set and a steady gas stream comprising of 0.3 L min À1 air and 0.020 L min À1 CO 2 was delivered from the G3Lab MFC to the vessel through a ring macro-sparger. The composition of the exhaust gas was determined by the Prima BT MS system which was interfaced with the 7 L vessel through the V1 conguration. An increase in temperature from 25 C to 37 C aer 110 hours was introduced from the start of the experiment. Baseline measurements were determined between 150-200 hours by monitoring signal stability for O 2 and CO 2 from off-gas measurements acquired by MS and in situ pH-% dO 2 measurements determined every 30 seconds through a TrupH and a TruDO probes (Thermo Fisher Scientic, MA, USA). Bioreactor contamination was deliberately introduced at 204 hours through the removal of the TruDO probe for 30 seconds which was subsequently re-inserted into the vessel. Baseline variations in signal responses for O 2 and CO 2 greater than 3Â times the corresponding standard deviation values from each measurement technique were considered to be clear signs of contamination as visualized aer 204 hours.

Mammalian cell culture process conditions
A CHO-K1 cell line producing a humanized IgG1 monoclonal antibody was initially cultured in shake-asks within a humid-ied incubator kept at 37 C with 5% (v/v) CO 2 and an agitation rate of 120 rpm using chemically dened BalanCD growth A liquid medium which was further supplemented with 1.2-1.7 g L À1 L-glutamine (Sigma-Aldrich, Dublin, Ireland). The cell seeding density upon vial thawing was 0.3 Â 10 6 cells per mL À1 in 250 mL total volume Erlenmeyer polycarbonate asks (Corning, NY, USA). A cell seeding density of 0.8 Â 10 6 cells per mL À1 was maintained through regular passaging every 2 days once densities of 3 to 4 Â 10 6 cells per mL À1 were reached. The cells were transferred aer two weeks of continuous passaging into four 1 L total volume polycarbonate shake-asks which were used as inoculum for bioreactor cultivations. Batch and fed-batch cultures, identied as B-x and FB-x cultures This journal is © The Royal Society of Chemistry 2020 Anal. Methods
respectively, were performed in the Thermo Fisher HyPerforma vessels integrated with the MS analyser through both V1 and V2 on-line congurations as summarized in Table 1. Bioreactors were inoculated at densities of 0.8 Â 10 6 cells per mL À1 and maintained at 37 C. Mixing was performed through single 3blade marine type impellers at stirring rates of 200 rpm for 3 L scale, 2 L medium working volume, and 188 rpm for 7 L scale, 5 L medium working volume, achieving k La value of 4 and 3 h À1 , respectively, as determined by the dynamic gassing out method performed with 0.5 L min À1 air ow-rates. A prototype liquid feed supplied by Irvine Scientic was supplemented in fedbatch cultures performed at 7 L scale only, involving 6% (v/v) additions of the initial 3.5 L culture volume aer 48, 72, 96, 168 and 216 culture hours. Individual streams of air, O 2 and CO 2 were supplied from pressurized cylinders at 2 bar to the G3Lab MFC which delivered a combined gas stream to the liquid medium through a L-pipe sparger at 3 L scale and a ring sparger at 7 L scale. A number of culture process parameters were varied during fed-batch processes to evaluate the performance of the magnetic sector MS analyser including: the implementation of a temperature drop from 37 C to 32 C at 144 culture hours in the culture FB-2, bolus glucose additions at 120 and 144 culture hours in the culture FB-3 to maintain levels at approximately 5 g L À1 and the introduction of rapid uctuations in O 2 ow-rates supplied to culture FB-4 aimed at maintaining % dO 2 saturation levels between 40-50%. Neutralization of culture pH was achieved in fed-batch cultures by automatic additions of CO 2 and 1 M NaOH while foaming was controlled with 2% (v/v) Antifoam C emulsion (Sigma-Aldrich, Ireland) which was added as needed.

Off-line measurement of cell culture parameters
Bioreactor sampling preceded bioreactor feeding and involved daily collection of 5 mL volume samples which were analysed off-line using a BioProle FLEX 2 analyzer (Nova Biomedical, MA, USA) providing concentrations for glucose, lactate, glutamine, % cell viability and viable cell density. Antibody titer was determined by affinity chromatography on an Ultimate 3000 LC system (Thermo Fisher Scientic, MA, USA) with UV detection at 280 nm involving 20 mL injections of cell culture supernatant on a 4 Â 35 mm MabPac protein A column (p/n 082539, Thermo Fisher Scientic, MA, USA) maintained at 25 C. Gradient conditions involved pumping 100% 1Â PBS (p/n 12821680, Fischer Scientic, Dublin, Ireland) at pH 7.0 until 0.9 minutes at a ow-rate of 0.6 mL min À1 followed by a switch to 100% 1Â PBS with 30 mM HCl at pH 2.0.

Off-gas analysis for the determination of bacterial contamination
Contamination events result in considerable productivity losses for biopharmaceutical companies as affected batches become essentially unusable. Sterility must be preserved through all bioprocessing stages to ensure product purity as microbial species can replicate within the same nutrient-rich temperature controlled environments. The potential sources of process contamination are multiple, ranging from inefficient sterilization conditions to inadequate aseptic techniques implemented by operators. Furthermore, contaminants can access through cracks formed within welded sample lines or seals which might not be always readily detectable or preventable. Unjustied variations in medium pH and % dO 2 measurements are considered to be the standard approach for establishing the presence of bacterial contaminants 31 however, analysis of the exhaust gas can be equally effective for detecting the presence of undesired respiratory activity while being minimally invasive. For this purpose, proles for O 2 and CO 2 were determined in real-time from the off-gas released from a 7 L benchtop bioreactor lled aseptically with culture medium. Subsequently, deliberate alteration of the sterile barrier was performed in effort to establish the efficacy of MS analysis for rapid detection of contamination. The integration of the bioreactor with the MS analyser was achieved through an on-line conguration, dened as V1 and shown in Fig. 1(A), which involved diverting a portion of the inlet gas supplied to the bioreactor towards the RMS for compositional characterization. Off-gas measurements were correlated with pH and % dO 2 values obtained from probes displaced in situ within the medium while preserving measurement consistency among all the techniques evaluated upon adjustment of the read-out frequency to 30 seconds. The rapid multi-stream sampling (RMS) conguration of the magnetic sector MS analyser facilitated automatic analysis of up to 16 different gas streams, therefore potentially allowing simultaneous monitoring of 8 bioreactors based on the implemented conguration, upon correlating the composition of the gas supplied to individual cultures with a corresponding off-gas prole. However, only two streams were monitored in this study involving the inlet and outlet gas streams of a single bioreactor to illustrate proof of concept. The cycling time, which is the time required for the RMS to sequentially sample all the congured gas streams and return to the original sampling point, was 1 minute but could be decreased further to 20  34 However, contamination events due to seal damage are challenging to reproduce and further conditions can contribute to successful migration of micro-organisms into the vessels such as unfavourable differential pressures between the inner vessel and surrounding unsterile environment. The dO 2 probe was deliberately manipulated for the purpose of this study at 204 hours by creating an opening through temporary detachment, approximately 30 seconds, from the headplate and exposing its outer surface to the unsterile outer environment to force medium contamination upon its reinsertion into the vessel. Signal stability was subsequently monitored and consistent shis in both in-line and off-gas on-line measurements from the corresponding signal baseline greater than 3 times standard deviation values were considered to be clear indication of contamination. Good correlation for the determination of microbial contamination based on MS detection and % dO 2 values determined in situ was established since both approaches conrmed the presence of micro-organism induced respiratory activity within a $1 hour time frame difference as summarized in Table 2. A rapid decrease in %O 2 values from the off-gas induced by microorganism respiration was visualised aer 22.4 hours from the removal of the dO 2 probe, with corresponding increases in offgas %CO 2 values observed at 23.7 hours, and from the equally rapid decrease in % dO 2 determined in situ aer 22.6 hours followed by increased medium turbidity. Bacteria growth produced a characteristic drop in pH 35 which however occurred at later intervals, aer 25.8 hours. A 5.70% (v/v) increase in CO 2 levels was observed at 235 hours from the analysis of the off-gas composition which was considerably higher than a corresponding decrease of 2.50% (v/v) determined for O 2 , therefore unlikely to be related exclusively to bacteria cell respiratory activity. The rapid drop in pH induced by bacteria growth shied the carbonate system equilibrium, which was converted almost entirely into CO 2 , therefore presumably resulting in excess quantities of CO 2 released into the off-gas. The subsequent decrease in off-gas CO 2 levels occurred as anticipated upon pH stabilization 36 since the carbonate equilibrium shied CO 2 back to the HCO 3 À form, producing a measured baseline increase for CO 2 of 1.40% (v/v), and a corresponding 1.30% (v/v) decrease in O 2 levels at 260 hours which was clearly more representative of micro-organism respiratory activity. Off-gas measurements ultimately facilitated the visualization of bacteria death at 340 hours as the composition of the exhaust gas returned to baseline levels with the simultaneous rapid increase in pH and % dO 2 saturation.

Real-time monitoring of batch CHO cell culture processes
The implementation of PAT tools during upstream bioprocessing stages must result in the real-time visualization of process related events for rapid detection of deviations. The  simplest approach for off-gas characterization involves correlating experimentally obtained off-gas measurements with performance indicators and process parameters, specically viable cell density values and key nutrients or metabolites known to impact cell growth. The preliminary evaluation of the magnetic sector MS as a tool for real-time mammalian cell culture monitoring involved proling the off-gas composition from a batch culture based on an antibody producing CHO-K1 cell line which was performed in a 3 L total volume bioreactor integrated on-line with the magnetic sector MS analyser through the previously described V1 conguration shown in Fig. 1(A). Steady state aeration conditions were implemented throughout the process without maintaining set pH and % dO 2 values which would otherwise require the introduction of variable process inputs with a potential impact on off-gas measurements. A medium hold study was initiated 24 hours prior to cellular inoculation to ensure the preservation of a sterile environment within the bioreactor, conrmed upon the visualization of inlet and off-gas proles at the outlet with identical composition as shown in Fig. 3(A). Clear visualization of cellular respiratory activity during the lag phase was detectable from off-gas proles as soon as 8 hours post CHO cell inoculation as a result of the high sensitivity provided by the magnetic sector MS analyser. A viable density of 1.2 Â 10 6 cells per mL À1 was determined at this time point from off-line cell count measurements which correlated with a 0.03% (v/v) increase in CO 2 levels and a corresponding decrease of 0.04% (v/v) in O 2 levels detected from off-gas measurements. The observed changes in the off-gas composition were attributed to cellular respiration as a result of the implemented stable aeration conditions and of the minimal variations in pH values which remained close to neutrality at this time point. A considerable increase in the levels of CO 2 released into the offgas was observed during the exponential cell growth phase which occurred between 24-120 culture hours, with a corresponding drop in O 2 levels, facilitating the clear visualization of the cellular transition state from the logarithmic growth phase to the stationary phase at 100 culture hours which ultimately yielded the greatest detected variations for O 2 and CO 2 from offgas measurements. Stoichiometric balance between the levels of O 2 consumed and CO 2 produced was observed at this time point, which varied by 0.65% (v/v) for each species, conrming that any potential CO 2 contributions from the bicarbonate medium component were unlikely to be signicant at the measured pH of 6.6. A rapid decline of measured CO 2 levels in the off-gas was observed as cell death occurred aer 200 culture hours with a consequent increase in O 2 values since this was no longer consumed for respiration. An equally signicant rise in % dO 2 and in pH values was observed at this time point due to the corresponding increase in O 2 medium saturation and cellinduced consumption of accumulated lactate. Ultimately, the observed CO 2 prole determined from off-gas analysis closely resembled the cell growth curve derived from traditional off-line cell counting techniques while an inverse trend was observed for O 2 once consumed as a consequence of cellular respiration. Furthermore, the real-time estimation of the respiratory parameters CDC, OXC and the ratio of these, the RQ values, provided insight on variations in substrate consumption trends as illustrated in Fig. 3(B). A predominantly dominant carbohydrate-induced cell metabolism was observed from the estimation of RQ values close to the unity between 25-100 culture hours, matching a previously reported prole observed in DG44 CHO cell processes. 19 However, a clear divergence between CDC and OXC values was observed from 125 culture hours onwards indicative of a metabolic switch which produced a decreased RQ value of 0.8 at this time point. Off-line metabolite measurements facilitated correlation of the observed decrease in RQ value with the depletion of two major substrates required for mitochondrial energy production activity, specically glutamine and glucose, 37 and a switch in lactate metabolism which was consumed rapidly until 300 hours when a drop in cell viability occurred as conrmed by the measured RQ value of 0 at this point indicative of the absence of cellular respiration.

Anal. Methods
This journal is © The Royal Society of Chemistry 2020

Impact of feed additions on off-gas proles
The performance of the magnetic sector MS was subsequently evaluated upon the implementation of process variables in fedbatch cultures performed at 7 L scale involving the previously evaluated CHO-K1 cell line. The addition of nutrients performed at dened feeding intervals, which impacted the medium's ionic strength/osmolality, and the implementation of shis in temperature during the stationary cell growth phase were likely to affect gas solubility in the medium and affect the off-gas proles obtained. Furthermore, rapid variations of the inlet gas composition supplied to the cultures were implemented, as shown from the G3Lab proles in the ESI Fig. S1, † producing variations in gas mass transfer which were likely to complicate accurate off-gas characterization, particularly for CO 2 as a result of its high solubility. 17 The supplementation of O 2 was performed to sustain % dO 2 values in the medium and preserve aerobic conditions desirable for extending cell viability 38 while additions of base titrant and CO 2 were introduced to maintain physiological pH values. 39 It was evident from the fed-batch culture FB-1 shown in Fig. 4(A) that O 2 supplementation, which increased its concentration from 19% (v/v) to 34% (v/v) during the initial 87 hours of the culture, rendered the visualization of respiratory activity from off-gas proles challenging, particularly during the lag and logarithmic growth phases. Here, cell growth was preferentially tracked by monitoring variations in CO 2 levels since here the observed concentration shis were considerably lower than what observed for O 2 , decreasing from 8.40% (v/v) to 1.40% (v/v) in the gas supplied to the bioreactor during the initial 48 culture hours. The observed drop in CO 2 levels occurred as a consequence of cell-induced lactate accumulation, conrmed by offline measurements, which decreased the pH and minimized further CO 2 supplementation requirements. A 0.4% (v/v) variation in CO 2 between inlet and outlet proles was detected at 48 culture hours however, it was not possible to attribute this variation entirely to cellular respiration as potential contributions induced by mass transfer variations could have also contributed as a consequence of the variable gas supply introduced into the culture. Therefore, the composition of the inlet gas supplied to the culture was subsequently stabilized by limiting the supplemented O 2 ow-rate to 0.1 L min À1 from 87 to 120 culture hours, corresponding to the stationary phase of the cell growth curve, to facilitate the clear visualization of cellular respiratory activity during this period as conrmed by a 0.6% (v/v) concentration difference between inlet and outlet signals for both O 2 and CO 2 . Limitations in O 2 supply however, undesirably restricted the levels of % dO 2 present in the medium to <5%, creating an almost anaerobic environment which stimulated the accumulation of high lactate levels reaching 6 g L À1 at 120 culture hours as shown in ESI Fig. S3. † Evident signs of cellular respiration were further detected until culture harvest at 288 culture hours as a consequence of minimal inlet gas ow-rate variations whereby variations of 0.3% (v/v) for O 2 and 0.4% (v/v) for CO 2 were measured between inlet and outlet signals, suggesting that cell viability was extended possibly as a result of increased nutrient supplementation throughout the culture. The variable composition of the gas supplied to the culture rendered the accurate proling of CDC, OXC and RQ trends challenging due to an increase in signal noise, which was particularly evident for CDC values as a result of variations in CO 2 mass transfer, thereby requiring averaging of the MS-generated respiratory data at 6 hour time intervals to smoothen signals for CDC, OXC and RQ as illustrated in Fig. 4(B). Stabilization of the gas composition between 87 and 120 culture hours facilitated the visualization of a clear drop in RQ value from 1.1 to 0.7 at 120, 200, and 250 hours indicative of a cellular metabolic switch since it concurred with the depletion of glucose as illustrated in ESI Fig. S3, † similarly to what had been observed in the previously described batch culture B-1 illustrated in Fig. 3(B). A correlation between increases in RQ values and feed additions performed at 96, 168 and 216 culture hours was further visualised. The feed added contained a proprietary mixture of nutrients which included glucose, as determined from off-line analysis, thereby contributing to the increases in RQ values which returned to $1 aer feeding.

Assessment of sterile barrier integrity by MS
The on-line integration of highly sensitive magnetic sector MS instrumentation with bioreactors provided real-time information on the integrity of essential sterile barriers. The removal of the dO 2 probe, previously described in Section 3.1, which was performed to deliberately contaminate the vessel is unlikely to be the cause of contamination in real process scenarios however, it provided evidence that the occurrence of gaps within the sterile barrier, which can potentially occur due to crack formation or imperfect welds among transfer lines, can be identied by MS and correlated in real-time to process events as they provide an escape route for the exhaust gas which disrupts the stability of the stream directed towards the MS. The formation of an opening on the headplate yielded a ow disruption of the exhaust-gas stream towards the RMS which was detectable in real-time as it resulted in a small but signicant drop in the CO 2 values measured from the off-gas at 204 hours, decreasing from 4.90% (v/v) to 4.80% (v/v) as illustrated in Fig. 2. The off-gas MS analyzer was therefore found to be suitable for the determination of undesirable gas escape routes formed if sterile boundaries become compromised which could potentially occur for example due to the formation of cracks along sampling or feeding lines.
Exhaust lter clogging can also occur during cultures as a result of condensate build-up which can yield undesirable pressure increases within the vessels. Typically, the structural integrity of lters would only be veried during bioreactor build-up phases upon pressurizing the lter housing by pumping in compressed air and ensuring that the achieved pressures match nominal values. However, such test does not account for potential issues which could occur during the process. Indeed, partial obstructions on the off-gas transfer line were oen visualized as a consequence of condensate build-up on the exhaust lter positioned aer the condensate catch-pot, producing abnormal spikes in the exhaust-gas ow-rate measurement which were not attributable to corresponding variations of the inlet gas ow rates as shown in Fig. 5 at 168 culture hours. The geometry of the lter and corresponding surface area were found to inuence the propensity to clog which was particularly problematic when lters of 37 mm in diameter were implemented while disruption effects were not observed when replaced by lters of 50 mm diameters.
Further valuable process related information was obtained upon monitoring the ow-rate measurements of the inlet and off-gas streams as determined by the RMS. A decrease in the exhaust-gas ow-rates was visualised in culture FB-1 between 50-60 culture hours, as shown in Fig. 5, with a corresponding increase of the inlet gas ow-rate reaching the MS, therefore suggesting that at this time point the gas supply was primarily directed towards the RMS rather than the bioreactor. A possible explanation was attributed to the increase in cellular biomass within the bioreactor over time which led to a higher resistance to ow for the supplied gas. Frequent adjustments of the owmeter back-pressure regulation valves were subsequently required to re-direct the supplied gas stream towards the culture, shown in Fig. 5, producing clear disturbances in the measured O 2 and CO 2 proles observable at 68, 213 and 262 culture hours in Fig. 4(A).
The clear limitations observed regarding the gas delivery methods in the V1 MS-bioreactor conguration were counteracted by removing the ow diversion mechanism required for directing a portion of the bioreactor's inlet gas towards the RMS. Instead, an analogue signal receiver was integrated as part of a subsequent (V2) conguration, shown in Fig. 1(B), whereby the composition of the inlet gas supplied to the culture was calculated by the GasWorks soware automatically upon accurate characterization of the air composition introduced into the culture, which was determined by running a separate line from the air supply directly to the MS analyzer, and analogue transfer of ow-rate information for each individual gas from the G3Lab setup, resulting in the improved stability of the exhaust-gas ow-rate as illustrated in ESI Fig. S2. † A further benet of the V2 conguration involved the increased amount of available sampling ports on the MS which would be now capable of potentially monitoring 15 different bioreactors simultaneously assuming that they all shared the same air supply.

Impact of process variables on off-gas proles
Fed-batch cultures performed subsequently to the implementation of the V2 MS-bioreactor conguration were aimed at assessing the performance of the MS analyser for mammalian cell culture monitoring upon alteration of the culture temperature conditions and glucose supplementation, implemented    Fig. 6(B), occurring simultaneously to glucose/glutamine depletion, followed by a drop in cell viability to 88% at 120 hours which decreased further to 55% at 144 hours. The implementation at this point of a culture temperature drop from 37 C to 32 C did not impact off-gas measurements contrary to expectations since a consequent increase in CO 2 solubility was anticipated to yield a decrease in measured CO 2 levels from the off-gas. The composition of both inlet and off-gas streams was identical from 144 to 288 hours and no respiratory activity could not be visualised as a result of the low cell viability present which therefore resulted in inaccurate RQ data during this period (not illustrated). However, the decreased temperature conditions prevented further drops in cell viability, which remained stable until the point of harvest at 288 hours, and produced an increase in protein yield commonly observed in CHO cell cultures 41 illustrated in the ESI Fig. S4. † A process change involving the supplementation of glucose was performed in culture FB-3 at 120 and 144 culture hours, as shown in Fig. 7(B), which prevented the rapid deterioration of culture conditions observed in culture FB-2 as conrmed by the stability of the RQ values which were maintained at 0.75 until 160 culture hours. The determined outcome was not surprising since cells could divert more energy to growth therefore resulting in a likely change of metabolism as conrmed by visualized disturbances in both the OXC and CDC proles at the two feeding intervals. However, the increase in glucose levels combined with the anaerobic conditions present due to O 2 ow-rate limitations resulted in the high accumulation of lactate which reached concentrations of 12 g L À1 in culture FB-3, almost double the levels observed in cultures FB-1 and FB-2, as shown in ESI Fig. S3. † A fourth fed-batch culture, identied as FB-4 and shown in Fig. 8, was evaluated to determine the effects of rapid changes in inlet gas composition on off-gas proles which were not stabilized as in the previously evaluated fed-bath cultures.
In culture FB-4, aeration rates were considerably decreased to 0.1 L min À1 to minimize shear stress on the cells, yielding considerably reduced levels of accumulated lactate which peaked 2.4 g L À1 at 120 hours as shown in ESI Fig. S3. † A shi in lactate metabolism was further visualized from 120 culture hours as it became a carbon nutrient source until 250 hours, revealing a positive effect on antibody production as observed in the ESI Fig. S4. † Variable quantities of O 2 and CO 2 were introduced throughout the culture as necessary to maintain dO 2 levels between 40-50% saturation and pH values at 7.0. However, the low initial gas ow-rates supplied to the culture prevented the satisfactory characterization of the off-gas composition due to the consequently low off-gas ow-rates reaching the MS which necessitates of values between 200-500 mL min À1 for accurate measurements. The rate of aeration was marginally increased to 0.15 L min À1 aer 24 culture hours in effort to consequently increase the ow-rates of the off-gas. However, satisfactory ow-rate conditions were achieved only aer 48 culture hours as air became enriched with O 2 . Clear signs of respiratory activity were detected from off-gas O 2 measurements shown in Fig. 8  O 2 inlet ow supply required for the preservation of % dO 2 levels between 40-50% were particularly signicant during 160-250 culture hours, as illustrated from the bioreactor gas owrate proles shown in the ESI Fig. S1, † and resulted in large off-gas measurement errors in the same time interval which impacted predominantly the determination of meaningful OXC values, thereby resulting in unstable RQ data as shown in Fig. 8(B), and prevented the visualization of further cellular respiratory activity.

Conclusions
Real-time monitoring of mammalian cell culture processes was performed through off-gas characterization upon the on-line integration of a magnetic sector MS analyser with benchtop bioreactors. Critical process-related information, including variations in culture medium temperature during medium hold stages and alterations within the sterile barriers were readily identied from disturbances in the off-gas prole signals which can be extremely benecial when root cause analysis investigations are performed to explain unaccounted variations in process performance. The rapid detection of contamination events by MS was demonstrated from the visualization of clear microbial respiratory activity signs, revealing similar performance to % dO 2 measurements obtained in situ. Respiratory activity from CHO cell processes was visualised from the lag phase depending on the experimental conguration implemented, involving relatively low VCDs of $1 Â 10 6 cells per mL À1 . A signicant drop in off-gas O 2 concentrations of 0.04% (v/v) was visualized at this time point which would be impossible to detect using less sensitive paramagnetic analyzers due to their inferior precision of typically AE0.2% (v/v). Clear shis in RQ values from mammalian cell culture processes were visualised upon determination of respiratory proles by MS and correlated with variations in substrate consumptions measured off-line. However, rapid changes in gas supply composition to the culture rendered the monitoring of mammalian cell respiratory activity challenging particularly during stationary phases when the cell density level and consequent O 2 demand were high. A possible solution for this issue would involve the implementation of a mass-ow controller prior to the MS to stabilize the off-gas ow however, the efficacy of this setup could not be yet veried. While the MS requires a signicant initial capital investment, it integrates multiple functionalities within a single platform which would otherwise require multiple standalone consumables and alternative equipment. Relative to in-line probes for example, the MS analyser is not subjected to stressful autoclaving conditions which affect the lifespan of probes, and can be congured to monitor up to 15 bioreactors simultaneously based on the non-invasive V2 conguration, therefore being particularly benecial for process development applications. The combination of these features with the high signal stability and infrequent calibration requirements of the magnetic sector MS translate into an attractive process monitoring solution particularly as biopharmaceutical manufacturing operations shi from traditional fed-batch to continuous modes.