Advanced Process Monitoring and Feedback Control to Enhance Cell Culture Process
Production and Robustness
An Zhang,1 Valerie Liu Tsang,1 Brandon Moore,1 Vivian Shen,1 Yao-Ming Huang,1 Rashmi Kshirsagar,2 Thomas Ryll2
1Cell Culture Development, Biogen Idec, Inc., Research Triangle Park, 5000 Davis Drive,North Carolina 27709; telephone: þ1-919-993-1678; fax: þ1-888-727-9627; e-mail: firstname.lastname@example.org
2Cell Culture Development, Biogen Idec, Inc., Cambridge, Massachusetts
ABSTRACT: It is a common practice in biotherapeutic manufactur- ing to deﬁne a ﬁxed-volume feed strategy for nutrient feeds, based on historical cell demand. However, once the feed volumes are deﬁned, they are inﬂexible to batch-to-batch variations in cell growth and physiology and can lead to inconsistent productivity and product quality. In aneffort to control critical quality attributes and to apply process analytical technology (PAT), a fully automated cell culture feedback control system has been explored in three different applications. The ﬁrst study illustrates that frequent monitoring and automatically controlling the complex feed based on a surrogate (glutamate) level improved protein production. More importantly, the resulting feed strategy was translated into a manufacturing- friendly manual feed strategy without impact on product quality. The second study demonstrates the improved process robustness of an automated feed strategy based on online bio-capacitance measure- ments for cell growth. In the third study, glucose and lactate concentrations were measured online and were used to automatically control the glucose feed,which in turn changed lactate metabolism. These studies suggest that the auto-feedback control system has the potential to signiﬁcantly increase productivity and improve robust- ness in manufacturing, with the goal of ensuring process performance and product quality consistency.
Biotechnol. Bioeng. 2015;9999: 1–10.
。 2015 Wiley Periodicals, Inc.
KEYWORDS: Chinese hamster ovary (CHO) cellculture; human embryonic kidney (HEK) cell culture; fed-batch; auto-feedback control; capacitance; glucose/lactate control
Fed-batch mammalian cell culture is widely used in industrial manufacture of recombinant therapeutic proteins because of its
Conflict of interest: The authors declare no conflict of interest. Correspondence to: A. Zhang
Received 10 March 2015; Revision received 12 May 2015; Accepted 18 June 2015 Accepted manuscript online xx Month 2015;
Article first published online in Wiley Online Library
(wileyonlinelibrary.com). DOI 10.1002/bit.25684
ease of operation, reliability, ﬂexibility, and productivity. Param- eters, such as agitation, pH, temperature, and dissolved oxygen (DO), are monitored in real-time and controlled at set points by process control systems. Ofﬂine measurements of viable cell density (VCD), osmolality, and metabolites like glucose and lactate are performed periodically via manual sampling. Complex feeds are added to the bioreactor during the course of the batch to replenish nutrients. To develop and optimize a fed batch process, the feed strategy (feed media composition, amount, and timing) is typically adjusted, based on cell growth, nutrient consumption (amino acid and glucose concentration), and byproduct accumulation (lactate and ammonia) to increase biomass, maintain high cell viability and cell speciﬁc productivity, and prolong culture duration. Among those approaches, reducing lactate and ammonia accumulation is often desired to mitigate any potential negative impact on cell physiology (Gawlitzek et al., 1999,2000; Ryll et al., 1994). Maintaining low levels of glucose and/or glutamine in culture is one of the most commonly used strategies, as it is relatively easy to implement (Chee Furng Wong et al., 2005; Cruz et al., 1999; Glacken et al., 1986; Gong et al., 2006; Kuwae et al., 2005; Li et al., 2005; Ljunggren and Haggstrom, 1994; Sauer et al., 2000). Gagnon et al. (2011) also controlled high-end pH to limit the delivery of glucose and then effectively suppressed lactate accumulation in Chinese hamster ovary (CHO) fed-batch cultures. In addition, biomass is often used to determine the nutrient feed amount in cell culture processes, asit is closely related to the physiological and metabolic status of the cells, as well as productivity of the therapeutic proteins. Huang et al. (2010) and Yu et al. (2011) reported a 10 g/L fed-batch process using an optimized feed media and biomass-based dynamic feed strategy, respectively. However, as these parameters are not monitored in real-time, feedback-based development is limited by the frequency of ofﬂine sampling. In addition, off-line monitoring of such cultivations via manual sampling is often labor-intensive and
can introduce operator-dependent error into the process.
In the current industrial cell culture manufacturing processes,
key parameters, such as DO, pH, and temperature, are controlled online, so it would be the natural progression to extend online control to feeding to improve process control and reduce operator
。 2015 Wiley Periodicals, Inc. Biotechnology and Bioengineering, Vol. 9999, No. xxx, 2015 1
error. Various process analytical technology (PAT) (Food and Drug Administration, 2004) tools have been developed in the last decade to monitor metabolites and VCD in a rapid but efﬁcient manner during the cultivation of mammalian cells. Larson et al. (2002) reported the use of an online high-pressure liquid chromatography (HPLC) system capable of measuring amino acids and carbohy- drates to study metabolism in mammalian cell culture systems. In addition, the NOVA FLEX online auto-sampler system, composed of an automated sampling system and a multi-functional analyzer, is able to automatically measure metabolites and perform feedback functions at scheduled intervals (Derfus et al., 2010). However, the sampling frequency at bench scale is limited due to the relatively large sampling volume, which constrains its application in all
processes. The new BioPAT1 Trace system is a real-time
monitoring system that can be used to monitor and control glucose and lactate at high frequency without culture volume reduction. Both technologies were employed in this work.
Many techniques have also been developed to monitor VCD in real time. These examples include optical cell density, dielectric spectroscopy, near-infrared spectroscopy, Raman spectroscopy, in- situ microscopy, acoustic resonance densitometry, andsoft-sensor- based approaches among others (Abu-Absi et al., 2010; Bittner et al., 1998; Crowley et al., 2005; Dorresteijn et al., 1996; Finn et al., 2006; Guez et al., 2004; Harris et al., 1987; Lu et al., 2013; Wu et al., 1995). Among these methodologies, dielectric spectroscopyis one of the most reliable methods to monitor viable cells in a selective manner, whereas others may suffer from difﬁculties in differ- entiating viable cells from dead cells. In dielectric spectroscopy, the permittivity (capacitance) of the culture broth is measured in real- time, thus enabling fast, simple, and efﬁcient process monitoringand control. In addition, Braasch et al. (2013) developed a novel device, a dielectrophoretic (DEP) cytometer, which provides a potential application for earlier detection of apoptosis, thus allowing for more timely feed adjustments. In this article, we introduce enhanced process control strategies to demonstrate fully automated online monitoring of cell culture metabolites and real- time biomass integrated with a computer-controlled feedback loop to improve fed-batch process production and robustness.
Based on our past experience (Tsang et al, ACS National Meeting 2008), certain metabolites can act as surrogate markers for other metabolites that are not readily measured (amino acids, vitamins,
etc.). In our ﬁrst study, we illustrate that monitoring and auto- feedback control based onglutamate level could dynamically adjust the complex feed amount during the cell culture process and improve production. To our knowledge, this is the ﬁrst reported fed-batch process utilizing online amino acid measurements to trigger automatic feedback control delivering complex nutrient feed. Furthermore, the feed strategy established by the feedback control loop was translated into a manufacturing-friendly manual feed process without impact on product quality. In the secondstudy, two individual dosing functions were applied to develop a truly automated feedback control process using a bio-capacitance (BC) probe to dynamically determine complex nutrient feed amounts while using the NOVA Flex autosampler to dynamically determine glucose feed. The beneﬁt of the real-time adaptive feeding to process robustness was demonstrated with low-seed density conditions to mimic under-seeding and/or low growth conditions that can potentially occur in manufacturing. Finally, we also demonstrated alactate-based glucose feed strategy, which resulted in signiﬁcant increase in monoclonal antibody production. Compared with previous pH-based strategies (Gragnon et al., 2011), our approach more accurately and quickly measures and controls glucose and
lactate levels simultaneously using the BioPAT1 Trace analyzer
system instead of using pH as an indicator because the medium buffer system could delay the pH response to lactate change in medium and make it difﬁcult to control lactate at a speciﬁed level.
Materials and Methods
Cell Lines and Media
This study used two recombinant CHO cell lines (cell lines A and B) secreting monoclonal antibodies and one human embryonic kidney (HEK) cell line (cell line C) secreting a fusion protein. Biogen Idec proprietary basal and feed media or commercial media were used for the different cell lines (Table I) (Huang et al., 2010).
Cell Culture Expansion and Fed-Batch Process
For culture maintenance and expansion, the incubator was controlled at 36–37 oC and 5–8% CO2 with platform agitation at 75 rpm. The cells were passaged every 3–4 days with seeding
Table I. Media, process parameters, feedback control strategy, and results for the three processes.
Cell line Cell line A Cell line B Cell line C Seed (e5 vc/mL) 4.0 10.0 5.0
Duration (day) 13 17 13
Basal media BIIB basal 1 þ plant hydrolysate BIIB basal 2 Commercial production medium
Feed media BIIB feed 1þ plant hydrolysate BIIB feed 2 Commercial feed medium
Online input Glutamate concentration Bio-capacitance, glucose concentration Glucose and lactate concentrations
Automated response Complex feed addition Complex feed and glucose stock solution additions
Glucose stock solution addition
Goal Improve productivity Improve manufacturing robustness Improve process robustness and productivity
Result Titer increased 35% within the sameculture duration (60% overall with extended duration)
Process is able to accommodate large differences in cell seeding density with no significant impact to titer and product quality
Lactate formation is controlled and viability is maintained, leading to 11% higher titer within the same culture duration and 32% higher due to viability improvement
densities of 3–5e5 cells/mL in 3 L shaker ﬂasks (Corning Life Science, Tewksbury, MA). All studies were conducted in 5-L glass vessels (Applikon Biotechnology, Foster City, CA) with a 2–2.5 L initial working volume and seeding densities of 4–10e5 cells/mL. The culture process was controlled by TruBio DV controllers (Finesse Solutions, San Jose, CA) (Table I). Details can be found elsewhere (Gilbert et al., 2013; Huang et al., 2010; Kshirsagar et al., 2012).
Cell line A and B historical fed-batch processes were conducted with ﬁxed percentage nutrient feeds based on culture volume at designated time points. Bolus glucose additions were fed as needed to maintain the glucose concentration above 2 g/L. Cell line C applied a continuous nutrient feed starting on day 3 with a predetermined feed rate. Glucose was also fed continuously starting on day 7.
Three customized dosing codes were developed and installed in
each DeltaV vessel station, which enabled the process to be simultaneously controlled with up to three different feeds. In addition, a customized comparison (<, ¼, >)/logic (and, or) module was also installed for multiple criteria control, which allows more complex control strategies.
In these studies, three different dynamic feed strategies were
implemented (Table I). The ﬁrst study applied a set point control strategy and was executed with cell line A. An amino acid, glutamate, was selected as a surrogate to represent the overall nutrient consumption because glutamate had been shown to be consumed in constant molar ratios compared to other key nutrients. A pre-determined concentration was set as a threshold for an automated response based on the online measurement. If glutamate concentrations fell below the set point, a feed pump was triggered to add complex nutrient feed containing glutamate to raiseculture glutamate concentration to the set point. The nutrient feed also contained other amino acids and nutrients, so all other nutrients were replenished simultaneously.
The second study employed a fully automated feedback control strategy and was executed with cell line B. The nutrient and glucose feeds were simultaneously controlled by two different dosing codes. The nutrient feed amount was determined by integrated biomass (Huang et al., 2010), which was calculated with BC. The dosing code automatically collects the present BCn value and stores the previous BCn-1 value at predetermined time intervals. The integrated bio- capacitance (IBC) was determined from the area under the BC curve and was estimated by using the trapezoid approximation across the predetermined time interval (24 h). The nutrient feed was administered every 24 h. The glucose concentration was maintained at the pre-determined set point using NOVA Flex autosampler with sampling every 4 h.
In the thirdstudy, in order to improvethe process for cellline C, glucose and lactate levels were controlled at the same time by monitoring and controlling the glucose stock solution feed using the BioPAT1 Trace with dialysis probes (Sartorius, Bohemia, NY) and a comparison/logic module. Brieﬂy, BioPAT1 Trace is a fully automated and self-calibrating device, which can measure glucose and lactate simultaneously in real-time with fast response time(up to 30 measurements/h) and without culture volume reduction. In this study, glucose and lactate concentrations were automatically measured every 15 min. A glucose stock solution was added when both the glucose and lactate levels were lower than their predetermined set points.
Cell density and viability were measured by CEDEX (Innovatis AG, Bielefeld, Germany) using trypan blue exclusion. Glucose, glutamine, lactate, ammonium, potassium, and sodium were measured using a BioProﬁle 400 or FLEX (NOVA Biomedical, Waltham, MA) or Cedex BioHT (Roche Diagnostics GmbH, Mannheim, Germany). pH, pCO2, and pO2 were measured using a Bioproﬁle pHOX analyzer (NOVA Biomedical). Osmolality was measured using an Auto Osmometer Model 3900 (Advanced Instruments, Norwood, MA). The antibody concentration, aggre- gation level, change variant, and N-glycan proﬁles of antibodies produced by cell line A and B were determined by various product quality analysis methods, which were previously described (Yang et al., 2014).
Results and Discussion
Glutamate-Based Feedback Control
The historical process for cell line A was an animal component free process using customized proprietary medium supplemented with hydrolysate (Table I). Based on development experience, glutamate could be used as a surrogate for total nutrient needs because its consumption proﬁle trended similarly to key amino acids, which fell to low concentrations in the historical process. Thus, maintaining the glutamate concentration at a certain level (3 mM) could prevent other amino acids from being depleted. To prove this concept,the NOVA auto-sampler was scheduled to sample and measure the glutamate level every 3 h. If the glutamate concentration fell below 3 mM, the volume of glutamate-containing nutrient feed was calculated and added using the feed pump to bring glutamate back to 3 mM, with all other nutrient components added simultaneously. As shown in Figure 1A, the glutamate-based feedback control process achieved a higher peak cell density than the historical average. The culture duration was also longer than thehistorical average (17 days compared to 14 days). The antibody titer of the controlled glutamate process was increased by 35% within the same culture duration (day 14), and 60% overall relative to the historical process (Fig. 1B).
When comparing the total quantity of nutrient feed added, it was
observed that the controlled glutamate process consumed three times more total nutrient feed than that of the historical process (data not shown). Earlier attempts had been made to increase productivity by increasing the volume of the scheduled feeds by 10%, but there was no signiﬁcant improvement in titer. Using feedback control, a better feeding strategy emerged based on ahigher total feed volume added at more frequent intervals (every 3-h feed in auto-feedback control process vs. every other day feed in historical process). Figure 2A shows that harvest product quality attributes in the glutamate-based feedback control process were comparable to the historical control.
Given the absence of automatic feedback control capability on the
manufacturing ﬂoor, the glutamate-based auto feedback control process at bench scale was treated as a development tool and translated into a pre-deﬁned daily manual bolus feed schedule
Figure 1. Cell growth and titer profile comparison between glutamate-based feedback control process and historical average. (A) Glutamate-basedfeedback control process achieved a higher peak celldensity than historical average. The cultureduration was also longer (17 days compared to 14 days) (B) The antibody titer of the controlled glutamate process increased by 35% within the same culture duration, and 60% overall relative to historical process. Error bars indicate range of analytical accuracy of measurements.
(Fig. 2B). Similar titer and productivity were achieved using the pre-deﬁned daily bolus feed strategy compared to the auto feedback control process (Fig. 2C). In addition, product quality was comparable to the historical control (Fig. 2A). The data indicates that the auto feedback control method can be used to develop feeding strategies for future processes, even if the sample/feedback methods are not utilized in the ﬁnal processes. This could be
particularly useful in manufacturing, where online feedback control may not be available.
Bio-Capacitance-Based Feedback Control
The historical process for cell line B used Biogen Idec in-house proprietary chemically deﬁned medium (Table I). Complex
Figure 2. Comparison between historical process, glutamate based feedback control process, and a manual daily bolus feed process converted from the glutamate feedback control process. (A) Product qualities. (B) Total feed volume. (C) Productivity. Error bars indicate range of analytical accuracy of measurements.
nutrients and glucose were fed separately. Using this cell cultureplatform and cell line, the glutamate consumption proﬁle was no longer an accurate surrogate for other nutrient components. Instead, the cell mass was used to determine the feed amount (Huang et al., 2010). As mentioned in Materials and Methods section, nutrient feeding was proportional to the cumulative integral of cell growth (cICG). In the historical process, daily sampling, analysis, and feed were conducted manually, which is time-consuming and requires a large amount of manual labor. The auto feedback system made it possible to automatically control the complex feed using online BC measurements. The original approach was to convert the BC reading to VCD and then feed based on the estimated cICG. In the early stage of the cell culture batch, the BC- based VCD matched the ofﬂine cell count-based VCD well. However, the linear correlation decreased as viability dropped below 90% due to the abrupt changes in the physiological state of cells, as well as the alteration in the electrochemical properties of the culture broth (Downey et al., 2014; Opel et al., 2010). The prediction discrepancy in the late stage of culture restricted the usability of BC-estimated VCD in the cICG-based feed process. To avoid the discrepancy in the late phase biomass estimation, the BC reading was directly correlated to the cICG-based feed amount. Several control runs were carried out using the ofﬂine cICG-based manual feed while BC data were collected. Two linear correlations were observed between cumulative integrated BC (cIBC) and total feed amount throughout the entire process, which made it feasible to directly use the online BC reading to control the feed (data not shown). These two slopes of
feed over cIBCmay correspond to the cell growth phase and production phase, respectively. During growth phase, cells need more nutrients to increase cell mass (higher slope). When cells enter production phase, fewer nutrients are provided to maintain nutrients at lower level and enhance protein production (lower slope). The BC-based control loop was set up in one dosing system to control the nutrient feed. Figure 3 shows that the BC-based feed process was comparable to the manual feed process in terms of cell growth and production, which was expected because the total feed amounts were also similar. In addition, product quality attributeswere also comparable.
In a traditional manufacturing process, the feed amount is ﬁxed without considering the variation in cell growth and nutrient requirements, which could lead to over- or under-feeding and which can consequently impact cell culture performance, titer, and product quality. Kshirsagar et al. (2012) and Gilbert et al. (2013) reported previously that overfeeding led to higher trisulﬁde in the molecules, as well as higher lactate production. A BC-based auto feed process can mitigate this risk by automatically adjusting the feed amount in real time to accommodate nutrient demand variation due to different cell growth. To illustrate this, bioreactors were intentionally seeded at low cell densities using cell line B. One bioreactor was fed following the ﬁxed feed strategy used in the current manufacturing process; the other wasfed based on the BC measurements. Figure 4Ashows that the ﬁxed feed process caused overfeeding,inhibited cell growth,and caused the viability to drop faster resulting in a twoday truncation of the batch.The BC-based
Figure 3. The performance of cIBC-based auto feed is comparable to manual feed control process. (A) Growth profile comparison. (B) Titer profile comparison. (C) Product quality. Error bars indicate range of analytical accuracy of measurements.
Figure 4. The comparison between cIBC-based auto feed and fixed feed at low seed density. The cIBC auto-feedback control low seed process showed similar performance as the control regular seed process. However, the fixed feed low seed process crashed earlier due to over-feeding, which caused high ammonia and high lactate. (A) Growth profile comparison. (B) Ammonium profile comparison. (C) Lactate profile comparison. (D) Titer profile comparison. (E) Product quality. Error bars indicate range of analytical accuracy of measurements.
feed process reduced feed accordingly and cells performed similarly to the control process. Signiﬁcantly higher ammonium and lactate levels were observed in the low seed ﬁxed feed process, causing the viability drop (Fig. 4B and C). As expected, the higher biomass in the BC-based feed process resulted in signiﬁcantly higher titer (Fig. 4D). Product quality data also show the ﬁxed feed process resulted in higher impurities, glycation, and trisulﬁdes due to overfeeding, whereas the auto feed process showed comparable product quality to the control process as the BC-based feeding strategy prevented overfeeding (Fig. 4E). This study demonstrated the capability and beneﬁt of the real-time feed adjustment using BC-based feedback control, which can improve process robustness, provide consistency in productivity and product quality, and rescue manufacturing batches. To further ensure manufacturing robust- ness, it would also be important to determine the acceptable range of the amount of feed added per cell, as the feed volume is determined by the cell mass.
In addition to nutrient feed control, an additional dosing code was applied to control the glucose feed. The glucose level was controlled at a predetermined value using a NOVA Flex autosampler with a second dosing system to add glucose as needed. Feeding
glucose as necessary based on a set point as opposed to a daily boluslowered the overall glucose concentrations in the media, which lowered the risk of glycation which is positively correlated with culture glucose concentration (Yuk et al., 2011). Thus, the process was converted to a fully automated process. Due to the ﬂexibility and modularity of the monitoring and controlling instruments, the fully automated process could also be easily converted to continuous and semi-continuous feed strategies to reduce the negative impact of nutrient and osmolality spikes from the current daily bolus feed process, should that be desired.
Glucose- and Lactate-Based Feedback Control
Batch-to-batch variability was often observed in the historical process using cell line C. In lower-performing batches, a higher level of lactate was observed and was believed to be the main cause of low yield. In order to improve process robustness, feedback control strategies have been applied to maintain low lactate production by controlling glucose at a low level (Omasa et al., 1992; Zhou et al., 1995; Zhou and Hu, 1994). A Nova Bioproﬁle Automated Samplersystem was used by Lu et al. (2013) to control the glucose level at
Figure 5. The glucose and lactate auto-control profiles using BioPAT and the comparison of BioPAT online and BioHT offline glucose and lactate measurements.
4–5 g/L with a 6-h sampling interval. However, this system is not able to control glucose at lower levels at bench scale because the sampling frequency is constrained by sampling volume. In our study, the BioPAT1 Trace instrument was integrated into the control loop, which could read online glucose and lactate as frequently as every minute without impact on the culture volume
(see Materials and Methods section). The glucose target wasset at
0.5 g/L before day 8 and 0.2 g/L after day 8 with a 15-min sampling interval. Once the glucose level fell below the set point, a glucose solution was automatically added to bring the concentration back to the pre-determined value. Figure 5 shows that the glucose was well controlled at a low level. As expected, the lactate concentration plateaued after day 6 when the glucose concentration fell below1 g/L. However, the lactate level increased after day 8 even though the glucose concentration was maintained at 0.5 g/L. In response to the observed lactate accumulation, the glucose control level was reduced to 0.2 g/L on day 9, and the lactate accumulation slowed down brieﬂy before gradually increasing again. This interesting response to glucose concentration indicates that the lactate metabolism did not shift from production to consumption despite the low glucose environment, which may be because glucose was preferentially consumed to generate lactate even at low glucose level. Interestingly, Siegwart et al. (1999) showed that lactate can be adequately controlled in their HEK293 process by maintaining glucose at the level of 1 mM, which is in contrast to our observation. This maybe due to differences in the cellclone, media, or process. Meanwhile, a glutamine decrease was observed from day 9 to day 10, whereas the lactate level was increased, which also suggests that lactate was probably produced from glutamine (Zagari et al., 2013). However, the relatively low lactate concentration of the BioPAT-controlled condition still beneﬁted the process (Fig. 6A).
Figure 6. The comparison between glucose auto-controlled process and historical process. (A) Glucose and lactate comparison: Glucose and lactate were controlled at the lower levels compared with historical process. However, the lactate level was still high even though the glucose level was approximately 0.2 g/L; (B) Growth profile comparison: Compared with historical process, the low glucose feedback control process showed a higher peak cell density. The culture duration also was able to be extended to 13 days due to higher viability compared to 11 days of regular process (C) Normalized titer profile comparison: Higher titer was observed in low glucose feedback control process. (D) Normalized specific productivity comparison. The slope reflected the specific productivity. Error bars indicate range of analytical accuracy of measurements.
Figure 7. Glucose and lactate profiles in glucose/lactate dual-controlled cell line C process. Lactate was consumed after day 6 when glucose was depleted.
Figure 8. The growth profile comparison between glucose-only controlled process, glucose/lactate dual-controlled process, and historicalprocess. The glucose/ lactate dual-controlled process showed higher cell growth than the historical process, but slightly lower than that of the glucose-only controlled process. Error bars indicate range of analytical accuracy of measurements.
Compared with the historical process, the low glucose feedback control process achieved a higher peak VCD. The culture duration could also be extended to 13 days due to higher viability comparedto 11 days for the historical process (Fig. 6B). The titer of the low glucose control process increased rv11% within the same culture duration, and rv32% overall relative to historical process (Fig. 6C). The speciﬁc productivity was comparable, so the titerincrease was mainly due to higher biomass in the low glucose feedback control process (Fig. 6D).
To further improve the process, a dual control strategy was set up in the feedback control system using a customized logic/comparison module. The glucose stock solution was only added if both the glucose level was lower than 0.5 g/L and the lactate level was lower than 1.5 g/L. This control strategy forced cells to consume lactate when glucose was depleted. Figure 7 shows that glucose was completely depleted after day 7 and cells started to consume lactate. Compared with the historical process, better cell growth was observed in the glucose/lactate dual-controlled process (Fig. 8). However, the extent of improvement in the glucose/lactate dual-controlled process was less than that of the glucose single-controlled process. This is likely due to the low lactate concentration. Cells were forced to consume lactate in the dual-control process, which caused the pH to remain at the upper dead band (Fig. 9 A). Consequently, alarge amount of CO2 was added to control the pH, and more ammonium accumulated as a result (Fig. 9B and C). In addition, the higher ammonium could also be attributed to amino acid consumption when the glucose was exhausted in media. Both high pCO2 and high ammonium can suppress cell growth (Dezengotita et al., 1998; Yang and Butler 2000; Zhu et al., 2005). Another possible explanation is that cell line C may not be able to efﬁciently utilize lactate as a carbon source, causing a shortage of energy (Altamirano et al., 2006). In this case, lowering lactate further may not translate to a better process. For some cell lines, sufﬁcient lactate levels may be required to stimulate cell growth (Choi et al., 2007; Omasa et al., 1992). Choi et al. (2007) reported sodium lactate addition enabled longer cultures and increased EPO production by more than a 2.7-fold in recombinant Chinese hamster ovary cells. Sodium lactate added at 40 mM enabled a lactate dehydrogenase
enzyme to catalyze the oxidation of lactate to pyruvate at a relatively high rate, which led to greater accumulation of nicotinamide adenine dinucleotide and increased energy yield.
This study also demonstrated that the customized dosing
codehas the ﬂexibility tobe easily integrated with different online monitoring instruments. The dosing code was combined with the BioPAT instrument to control glucose and lactate at a high samplingfrequency without impact on culture volume and resulted in a more productive and robust cell line C process. The BioPAT provides proof-of-concept that cell culture processes can beneﬁt from real- time feedback control of glucose and lactate levels. However, the instrument is difﬁcult to implement in large-scale GMP manufacturing due to its intricate setup. Currently, Raman spectroscopy-based glucose and lactate models are being developed at Biogen Idec (Berry et al.,2015), with the potential for providing real-time feedback control for glucose and lactate in a GMP manufacturing-friendly format.
Three different auto-feedback control strategies were implemented in this report in which cell metabolism and physiology were actively monitored and used to dynamically control the level of nutrients delivered to the culture (Table I). VCD and culture duration were increased using glutamate or glucose/lactate-based feedback control to decrease metabolite waste accumulation, resulting in signiﬁcant increases in antibody production. In addition, the feed strategy generated by the glutamate-based feedback loop could be translated into a manufacturing-friendly manual feed strategy without impact on product quality. BC-based feedback control demonstrated remarkable enhancement in process robustness over traditional ﬁxed fed-batch processes. The feed amount was automatically adjusted in response to process variations in real time to avoid over-/under-feeding due to seed density variation. This report also describes the successful use of a glucose/lactate online analyzer for cell culture process control with higher sampling frequency. Together, these strategies form a toolbox for developing the next generation of cell culture manufacturing processes that can
Figure 9. The metabolite profile comparison between glucose-only controlled process, glucose/lactate dual-controlled process, and historical process. Higher pH, pCO2, and ammonium were observed in glucose/lactate dual-controlled process dueto forced lactate consumption,which might negatively impact cell growth. (A) pH profilecomparison. (B) pCO2 profile comparison. (C) Ammonium profile comparison. Error bars indicate range of analytical accuracy of measurements.
take advantage of automated, real-time controlof nutrient and metabolite levels. In the future, by combining the BC probe to control nutrient feeds and the Raman probe to control glucose and lactate levels, we can design a fully automated process to dynamically and adaptively control a the feeding schedules of a variety of cell lines in GMP manufacturing.
The authors thank Angela X. Wang, Amy Choi, and Jimmy Huynh for their contributions to this work, and William Yang, Weimin Lin, Weiwei Hu, and John Paul Smelko for excellent technical discussions.
Abu-Absi NR, Kenty BM, Cuellar ME, Borys MC, Sakhamuri S, Strachan DJ, Hausladen MC, Li ZJ. 2010. Real time monitoring of multiple parameters in mammalian cell culture bioreactors using an in-line Raman spectroscopy probe. Biotechnol Bioeng 108:1215–1221.
Altamirano C, Illanes A, Becerra S, Cair'o JJ, G、odia F. 2006. Considerations on the
lactate consumption by CHO cells in the presence of galactose. J Biotechnol 125(4):547–556.
Berry B, Moretto J, Matthews T, Smelko J, Wiltberger K. 2015. Cross-scale predictive modeling of CHO cell culture growth and metabolites using Raman spectroscopy and multivariate analysis. Biotechnol Prog 31(2):566–577.
Bittner C, Wehnert G, Scheper T. 1998. In situ microscopy for on-line determination of biomass. Biotechnol Bioeng 60:24–35.
Braasch K, Nikolic-Jaric M, Cabel T, Salimi E, Bridges GE, Thomson DJ, Butler M.
2013. The changing dielectric properties of CHO cells can be used to determine early apoptotic events in a bioprocess. Biotechnol Bioeng 110(11):2902–2914.
Chee Furng Wong D, Tin Kam Wong K, Tang Goh L, Kiat Heng C, Gek Sim Yap M.
2005. Impact of dynamic online fed-batch strategies on metabolism, productivity and N-glycosylation quality in CHO cell cultures. Biotechnol Bioeng 89(2):164–177.
Choi YS,Lee DY, Kim IY, Park HW, Choe TB, Kim IH. 2007. Enhancement of erythropoietin production in recombinant Chinese hamster ovary cells by sodium lactate addition. Biotechnol Bioprocess Eng 12(1):60–72.
Crowley J, Arnold SA, Wood N, Harvey LM, McNeil B. 2005. Monitoring a high cell density recombinant Pichia pastoris fed-batch bioprocess using transmission and reﬂectance near infrared spectroscopy. Enzyme Microb Technol 36: 621–628.
Cruz HJ, Moreira JL, Carrondo MJT. 1999. Metabolic shifts by nutrient manipulation in continuous cultures of BHK cells. Biotechnol Bioeng 66:104–108.
Derfus GE, Abramzon D, Tung M, Chang D, Kiss R, Amanullah A. 2010. Cell culture monitoring via an auto-sampler and an integrated multi-functional off-line analyzer. Biotechnol Prog 26(1):284–292.
Dezengotita VM, Kimura R, Miller WM. 1998. Effects of CO2 and osmolality on hybridoma cells: Growth, metabolism and monoclonal antibody production. Cytotechnology 28(1–3):213–227.
Dorresteijn RC, Numan KH, de Gooijer CD, Tramper J, Beuvery EC. 1996. On-line estimation of the biomass activity during animal-cell cultivations. Biotechnol Bioeng 51:206–214.
Downey BJ, Graham LJ, Breit JF, Glutting NK. 2014. A novel approach for using dielectric spectroscopy to predict viable cell volume (VCV) in early process development. Biotechnol Prog 30(2):479–487.
Finn B, Harvey LM, McNeil B. 2006. Near-infrared spectroscopic monitoring of biomass, glucose, ethanol and protein content in a high cell density baker’s yeastfed-batch bioprocess. Yeast 23:507–517.
Food and Drug Administration. 2004. Guidance for industry PAT — A framework for innovative pharmaceutical development, manufacturing, and quality assurance.
Gagnon M, Hiller G, Luan YT, Kittredge A, DeFelice J, Drapeau D. 2011. High-end pH-controlled delivery of glucose effectively suppresses lactate accumulation in CHO fed-batch cultures. Biotechnol Bioeng 108:1328–1337.
Gawlitzek M, Papac DI, Sliwkowski MB, Ryll T. 1999. Incorporation of 15N from ammonium into the N-linked oligosaccharides of an immunoadhesin glycoprotein expressed in Chinese hamster ovary cells. Glycobiology 9(2):125–131.
Gawlitzek M, Ryll T, Lofgren J, Sliwkowski MB. 2000. Ammonium alters N-glycan structures of recombinant TNFR-IgG: Degradative versus biosynthetic mechanisms. Biotechnol Bioeng 68(6):637–646.
Gilbert A, McElearney K, Kshirsagar R, Sinacore MS, Ryll T. 2013. Investigation of metabolic variability observed in extended fed batch cell culture. Biotechnol Prog 29(6):1519–1527.
Glacken MW, Fleischaker RJ, Sinskey AJ. 1986. Reduction of waste product excretion via nutrient control: Possible strategies for maximizing production and cell yields on serum in cultures of mammalian cells. Biotechnol Bioeng 28:1376–1389.
Gong X, Li D, Li X, Fang Q, Han X, Wu Y, Yang S, Shen BQ. 2006. Fedbatch culture optimization of a growth-associated hybridoma cell line in chemically deﬁned protein-free media. Cytotechnology 52(1):25–38.
Guez JS, Cassar JP, Wartelle F, Dhulster P, Suhr H. 2004. Real time in situ microscopy for animal cell-concentration monitoring during high density culture in bioreactor. J Biotech 111:335–343.
Harris CM, Tood RW, Bungard SJ. 1987. Dielectric permittivityof microbial suspensions at radio frequencies: A novel method for the real-time estimation of microbial biomass. Enzyme Microb Technol 9:181–186.
Huang YM, Hu W, Rustandi E, Chang K, Yusuf-Makagiansar H, Ryll T. 2010.
Maximizing productivity of CHO cell-based fed-batch culture using chemically deﬁned media conditions and typical manufacturing equipment. Biotechnol Prog 26(5):1400–1410.
Kshirsagar R, McElearney K, Gilbert A, Sinacore M, Ryll T. 2012. Controlling trisulﬁde modiﬁcation in recombinant monoclonal antibody produced in fed- batch cell culture. Biotechnol Bioeng 109(10):2523–2532.
Kuwae S, Ohda T, Tamashima H, Miki H, Kobayashi K. 2005. Development of a fed- batch culture process for enhanced production of recombinant human antithrombin by Chinese hamster ovary cells. J Biosci Bioeng 100(5):502–510. Larson TM, Gawlitzek M, Evans H, Albers U, Cacia J. 2002. Chemometric evaluation of on-line high-pressure liquid chromatography in mammalian cell cultures:
Analysis of amino acids and glucose. Biotechnol Bioeng 77(5):553–563.
Li L, Li Mi L, Feng Q, Liu R, Tang H, Xie L, Yu XL, Chen ZN. 2005. Increasing the culture efﬁciency of hybridoma cells by the use of integrated metabolic control of glucose and glutamine at low levels. Biotechnol Appl Biochem 42:73–80.
Ljunggren J, Haggstrom L. 1994. Catabolic control of hybridoma cells by glucose and glutamine limited fed batch cultures.Biotechnol Bioeng 44:808–818.
Lu F, Toh PC, Burnett I, Li F, Hudson T, Amanullah A, Li J. 2013. Automated dynamic fed-batch process and media optimization for high productivity cell culture process development. Biotechnol Bioeng 110(1):191–205.
Omasa T, Higashiyama KI, Shioya S, Suga KI. 1992. Effects of lactate concentration on hybridoma culture in lactate-controlled fed-batch operation. Biotech Bioeng 39:556–564.
Opel CF, Li JC, Amanullah A. 2010. Quantitative modeling of viable cell density, cell size, intracellular conductivity, and membrane capacitance in batch and fed- batch CHO processes using dielectric spectroscopy. Biotechnol Prog 26(4):1187–1199.
Ryll T, Valley U, Wagner R. 1994. Biochemistry of growth inhibition by ammonium ions in mammalian cells. Biotechnol Bioeng 44(2):184–193.
Sauer PW, Burky JE, Wesson MC, Sternard HD, Qu L. 2000. A highyielding, generic fed-batch cell culture process for production of recombinant antibodies. Biotechnol Bioeng 67(5):585–597.
Siegwart P, C^ot'e J, Male K, Luong JH, Perrier M, Kamen A. 1999. Adaptive control at
low glucose concentration of HEK-293 cell serum-free cultures. Biotechnol Prog. 15(4):608–616.
Wu P, Ozturk SS, Blackie JD, Thrift JC, Figueroa C, Naveh D. 1995. Evaluation and applications of optical cell density probes in mammalian cell bioreactors. Biotechnol Bioeng 45(6):495–502.
Yang M, Butler M. 2000. Effects of ammonia on CHO cell growth, erythropoietin production, and glycosylation. Biotechnol Bioeng 68(4):370–380.
Yang WC, Lu J, Nguyen NB, Zhang A, Healy NV, Kshirsagar R, Ryll T, Huang YM.
2014. Addition of valproic acid to CHO cell fed-batch cultures improves monoclonal antibody titers. Mol Biotechnol 56(5):421–428.
Yu M, Hu Z, Pacis E, Vijayasankaran N, Shen A, Li F. 2011. Understanding the intracellular effect of enhanced nutrient feeding toward high titer antibody production process. Biotechnol Bioeng 108(5):1078–1088.
Yuk IH, Zhang B, Yang Y, Dutina G, Leach KD, Vijayasankaran N, Shen AY, Andersen DC, Snedecor BR, Joly JC. 2011. Controlling glycation of recombinant antibody in fed-batch cell cultures. Biotechnol Bioeng 108(11):2600–2610.
Zagari F, Jordan M, Stettler M, Broly H, Wurm FM. 2013. Lactate metabolism shift in CHO cell culture: The role of mitochondrial oxidative activity. N Biotechnol 30(2):238–245.
Zhou W, Hu WS. 1994. On-line characterization of a hybridoma cell culture process.
Biotechnol Bioeng 44:170–177.
Zhou W, Rehm J, Hu WS. 1995. High viable cell concentration fed-ba
tch cultures of hybridoma cells through on-line nutrient feeding. Biotechnol Bioeng 46(6): 579–587.
Zhu MM, Goyal A, Rank DL, Gupta SK, Vanden Boom T, Lee SS. 2005. Effects of elevated pCO2 and osmolality on growth of CHO cells and production of antibody-fusion protein B1: A case study. Biotechnol Prog 21(1):70–77.