Sie sind auf Seite 1von 12

ARTICLE Dening Process Design Space for Monoclonal Antibody Cell Culture

Susan Fugett Abu-Absi, LiYing Yang, Patrick Thompson, Canping Jiang, Sunitha Kandula, Bernhard Schilling, Abhinav A. Shukla Manufacturing Sciences & Technology, Bristol-Myers Squibb Co., 6000 Thompson Road, East Syracuse, New York 13057; telephone: 315-431-7926; fax: 315-432-2343; e-mail: abhinav.shukla@bms.com
Received 31 August 2009; revision received 16 February 2010; accepted 5 April 2010 Published online 16 April 2010 in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/bit.22764

ABSTRACT: The concept of design space has been taking root as a foundation of in-process control strategies for biopharmaceutical manufacturing processes. During mapping of the process design space, the multidimensional combination of operational variables is studied to quantify the impact on process performance in terms of productivity and product quality. An efcient methodology to map the design space for a monoclonal antibody cell culture process is described. A failure modes and effects analysis (FMEA) was used as the basis for the process characterization exercise. This was followed by an integrated study of the inoculum stage of the process which includes progressive shake ask and seed bioreactor steps. The operating conditions for the seed bioreactor were studied in an integrated fashion with the production bioreactor using a two stage design of experiments (DOE) methodology to enable optimization of operating conditions. A two level Resolution IV design was followed by a central composite design (CCD). These experiments enabled identication of the edge of failure and classication of the operational parameters as non-key, key or critical. In addition, the models generated from the data provide further insight into balancing productivity of the cell culture process with product quality considerations. Finally, process and product-related impurity clearance was evaluated by studies linking the upstream process with downstream purication. Production bioreactor parameters that directly inuence antibody charge variants and glycosylation in CHO systems were identied. Biotechnol. Bioeng. 2010;106: 894905. 2010 Wiley Periodicals, Inc. KEYWORDS: cell culture; design space; quality by design; monoclonal antibody

Introduction
The quality by design (QbD) paradigm for biopharmaceutical manufacturing processes has emerged from the cGMPs
Correspondence to: A. A. Shukla

for the 21st century initiative taken by the Food and Drug Administration (FDA, 2004). QbD involves three primary components (Rathore and Winkle 2009): process knowledge that includes a thorough understanding of process inputs and their impact on performance, the relationship between the process and a products critical quality attributes (CQAs) and the association between CQAs and a products clinical properties. The QbD approach is expected to build quality into the process rather than testing quality into the product. As such, the QbD philosophy is anticipated to be implemented throughout the lifecycle of a product starting with process design and denition, moving on to process characterization and validation and nally to monitoring and control of the commercial manufacturing process (Kozlowski and Swann, 2006; Rathore and Winkle, 2009). Some of the anticipated benets of QbD for the biopharmaceutical industry include a reduction in the number of post-commercial lings for process changes and a modern approach to biopharmaceutical product quality leading to greater exibility for the manufacturer. The concept of design space for the manufacturing process is inherent as one of the three primary components of the QbD paradigm (Rathore et al., 2007). The ICH Q8 guidance document denes design space as the multidimensional combination and interaction of input variables that have been demonstrated to provide an assurance of quality (FDA, 2006, 2009). The expectation of exibility within the design space is inherent in the guidance; working within the design space is not considered a change. The guidance goes on to state design space is proposed by the applicant and is subject to regulatory assessment and approval. The design space concept is better established in small molecule pharmaceutical manufacturing (Lipsanen et al., 2007; Nail and Searles, 2008). A systematic approach toward the creation of a design space understanding of a biopharmaceutical manufacturing process is essential for the creation of a robust and well-controlled process in-line with the QbD paradigm (Rathore et al., 2008, 2009). 2010 Wiley Periodicals, Inc.

894

Biotechnology and Bioengineering, Vol. 106, No. 6, August 15, 2010

Process characterization studies are performed at laboratory scale (Seely and Seely, 2003) using a qualied scaledown model (Li et al., 2006) to dene the design space. The operational parameters that will be studied during process characterization are prioritized by a risk analysis approach such as failure modes and effects analysis (FMEA) (FDA, 2009; Seely and Haury, 2005). Process characterization experiments are typically conducted using a design of experiments (DOE) approach. This has been applied toward the characterization of individual downstream processing steps (Kelley et al., 1997, 1998; Shukla et al., 2001). These techniques have also been applied for the denition of design space for microbial fermentation products (Harms et al., 2008; Rathore et al., 2008). During the process characterization exercise, each operational parameter is studied over a wide range called the characterization range. The aim is to understand process operation in a range wider than that used for routine process operation. The acceptable range for any parameter is typically between the operating range and the characterization range. The use of a design of experiments (DOE) methodology for process characterization enables a study of parameter interactions and classication of parameters based on their impact to productivity and product quality. A tiered classication of operational parameters as nonkey, key or critical and their operating and acceptable ranges form the in-process control (IPC) strategy for that product. The IPC strategy also utilizes in-process operating ranges, action limits and acceptance criteria to ensure the consistent monitoring and control of the manufacturing process. Nonkey operational parameters are easily controlled and/or have wide operating ranges. Key operational parameters are essential for process performance considerations, including productivity and scheduling. If varied outside their action limit, key operational parameters may affect process performance, but not CQAs. Critical process parameters (CPPs) are operational (input) or performance (output) parameters that could affect the CQAs of the drug substance if varied outside their acceptable range. Figure 1 shows a ow diagram for the strategy used to dene the design space for a process step. The FMEA exercise enables operational parameters with low risk priority number (RPN) scores to be eliminated from further study. Next, low resolution DOE studies using Resolution III or IV factorial designs are used for screening operational parameters and eliminating non-key parameters. A higher resolution response surface methodology (RSM) design is used to study the remaining key process parameters. For worst-case experiments, key operational parameters are combined to create a worst-case outcome with respect to a particular performance parameter for that step. This worstcase material is passed on to subsequent steps to determine if the proposed key operational parameter ranges are acceptable. Worst-case studies also enable critical process parameters to be identied. If a characteristic of the protein does not change signicantly in subsequent steps, operational parameters that affect it may be dened as CPPs in the

Figure 1. Strategy used to establish the process design space for the unit operations of a biopharmaceutical manufacturing process. upstream step. For example, if the N-Linked oligosaccharide prole does not change through the purication steps, the operational parameters that affect it in the production bioreactor would be dened as CPPs. A typical mammalian cell culture process, as shown in Figure 2A, is comprised of a series of steps starting with vial thaw and proceeding through several stages of inoculum expansion. The inoculum is used to initiate the production bioreactor culture. The cells in the production bioreactor, typically operated in fed-batch mode, express and secrete the protein product. At the end of the production bioreactor step, the cells and cell debris are separated from the cell culture supernatant, which is taken through downstream process steps designed to remove impurities. Given its multi-stage nature, process characterization for the upstream process can be a time and resource intensive process. A streamlined approach to upstream process characterization is needed to provide a rapid understanding of the process design space and enable the creation of an appropriate IPC strategy. However, the current literature has a paucity of case studies describing the development of a design space for products produced by mammalian cell culture. Moreover, the linkage between cell culture process conditions and product quality has not been adequately explored for mammalian cell culture. This paper describes a holistic approach for dening the process design space for a monoclonal antibody derived by mammalian cell culture. A streamlined experimental design was used to link steps in the inoculum stages with the production bioreactor. A linkage between upstream process

Abu-Absi et al.: Dening Cell Culture Design Space Biotechnology and Bioengineering

895

Figure 2. A: Upstream steps of the monoclonal antibody cell culture process. B: Failure mode and effects analysis (FMEA) Ishikawa diagrams for the seed bioreactor and production bioreactor steps. Underlined parameters were studied. conditions and the downstream process was created. A comprehensive understanding of raw material variability requires signicant at-scale manufacturing experience and will be dealt with in a future report. growth rate in the seed bioreactor was calculated by linear regression to determine the slope of the natural log of viable cell density versus culture duration.

Downstream Processing

Materials and Methods


Cell Culture The monoclonal antibody studied was expressed in Chinese Hamster Ovary (CHO) cells. Cells were grown in suspension using proprietary cell culture media. Shake asks (Corning Incorporated, Corning, NY) were cultured in VWR, International (West Chester, PA) CO2 incubators using Thermo Scientic, Inc. (Waltham, MA) MaxQ shaker platforms. The shake ask studies were conducted at the same scale and in the same disposable asks used in manufacturing. The seed and production bioreactor steps were studied in 7-L stirred-tank bioreactors (Sartorius, Goettingen, Germany). The 7-L bioreactors were operated as scaledown models of the manufacturing bioreactors. Specic

Culture from the 7-L production bioreactors was harvested using centrifugation followed by a series of depth and absolute lters. CUNO Maximizer EXT lters (3M, St. Paul, MN) were loaded at 200 L/m2 followed by sterile ltration using Acrodisc lters (Pall, East Hills, NY). The harvest pool was subsequently puried via lab-scale MabselectTM (GE Healthcare, Piscataway, NJ) Protein A chromatography.

Analytical Methods A Bioprole Analyzer 400 (Nova Biomedical, Waltham, MA) was used to monitor pH, dissolved oxygen and metabolite concentrations. Cell count measurements were carried out with a Cedex automated cell counter (Innovatis

896

Biotechnology and Bioengineering, Vol. 106, No. 6, August 15, 2010

AG, Germany). Titer was measured by HPLC analysis of binding to Protein A. Size exclusion chromatography was utilized to quantify the amount of high molecular weight (HMW) species. Charge variants were measured with an analytical cation exchange chromatography (CEX) HPLC method. N-Linked glycosylation was measured by High Performance Anion-Exchange Chromatography with Pulsed Amperometric Detection (HPAEC-PAD).

Results and Discussion


Risk Assessment The rst step toward mapping the process design space is to identify factors (operational parameters) and to assess the
Table I. FMEA results. Parameter

risk each of them poses to process robustness. A formal risk assessment exercise that includes the development history of the process and an understanding of manufacturing controls is highly recommended. This exercise enables identication of factors that could inuence important outcomes from the cell culture process (culture health, product titer and product quality). It also enables a reduction in the number of factors that require study during process characterization. An FMEA exercise was conducted for the inoculum expansion, seed bioreactor and production bioreactor steps to identify process variables that could inuence cell culture performance. Figure 2B shows the Ishikawa diagrams that helped identify factors for further study. Table I shows the factors ranked based on the RPN scores. The threshold RPN score was determined based on an acceptable level of risk, taking into consideration the amount of process knowledge

Responses considered Effects on seed bioreactor growth rate and nal viability

S (15) 2 2 2 2 2 2 2 2 2 2 2 2 1 1 5 3 3 5 5 5 4 4 3 3 3 2 2 2 1 2 2

O (13) 2 2 2 2 1 2 2 1 1 1 1 1 1 1 2 3 2 2 1 2 2 2 2 2 1 1 1 1 1 1 1

D (13) 2 2 1 1 2 1 1 1 1 1 1 1 1 1 2 2 2 1 2 1 1 1 1 1 1 1 1 1 2 1 1

RPN (145) 8 8 4 4 4 4 4 2 2 2 2 2 1 1 20 18 12 10 10 10 8 8 6 6 3 2 2 2 2 2 2

Seed bioreactor initial viable cell densitya Seed bioreactor temperaturea Shake ask incubator carbon dioxide concentrationa Seed bioreactor pHa Shake ask initial viable cell densitya Shake ask nal viable cell densitya Shake ask incubator temperatureb Shake ask agitation speed Shake ask working volume Seed bioreactor DO Seed bioreactor aeration Seed bioreactor agitation speed Vial thaw waterbath temperature Vial thaw duration Production bioreactor temperaturea Production bioreactor initial viable cell densitya Seed bioreactor temperaturea Production bioreactor pHa Production bioreactor temp shift timinga Production bioreactor durationc Production bioreactor DOa Seed bioreactor pHa Seed bioreactor nal viable cell densitya Seed bioreactor durationc Production bioreactor feed timing and volume Seed bioreactor agitation speed Seed bioreactor aeration Seed bioreactor DO Seed bioreactor initial viable cell density Production bioreactor aeration Production bioreactor agitation speed

Effects on production bioreactor productivity and product quality

Severity of the excursion: S 1: negligible impact, 2: potential impact to manufacturing schedule, 3: potential impact to yield, 4: potential impact to product quality, 5: potential impact to yield and product quality. Occurrence of the excursion: O 1 (low)3 (high). Detection of the excursion: D 1 (high)3 (low). Risk Priority Number (RPN) S O D. Factors chosen for study are shown in bold font. a Parameters evaluated in screening study. b Shake ask incubator temperature was not studied although it received an RPN score above the threshold. For efciency in experimentation, the effect of temperature on cell growth and viability was assessed only in the seed bioreactor. c Duration in the seed bioreactor and production bioreactor were not included as factors in the studies. Duration in the seed bioreactor was indirectly evaluated by examining nal viable cell density in the seed bioreactor. Duration in the production bioreactor was indirectly evaluated by examining product quality from days 12 to 17.

Abu-Absi et al.: Dening Cell Culture Design Space Biotechnology and Bioengineering

897

Table II.

Factors studied to establish the IPC. Range (coded) evaluated during screening studies 1 1 1 1 1 1 1 1 1 1 1 1 to to to to to to to to to to to to 1 1 1 1 1 1 1 1 1 1 1 1 Range (coded) evaluated during CCD studies

Code A B C D E F G H I J K L

Factor Incubator carbon dioxide Shake ask initial viable cell density Shake ask nal viable cell density Seed bioreactor temperature Seed bioreactor pH Seed bioreactor initial viable cell density Seed bioreactor nal viable cell density Production bioreactor initial viable cell density Production bioreactor temperature Production bioreactor temperature shift timing Production bioreactor pH Production bioreactor dissolved oxygen

1 to 1.5 2 to 2 1 to 1

gained during development. As an example, the effects of the feed scheme (amount and timing of feeds) were not evaluated in the production bioreactor screening study since feeds are well controlled and were examined and optimized during process development. The process utilizes several xed volume feeds that are administered at specic culture durations. The thresholds for the inoculum screening study and production bioreactor screening study were 4 and 6, respectively. Factors chosen for study are underlined in Figure 2B and shown in bold font in Table I. Table II describes the factors studied and the code that identies them in the subsequent gures.

Scale-Down Model Qualication The 7-L scale-down production bioreactor model was qualied by comparing proles of viable cell density, viability, metabolites and titer in addition to antibody characteristics measured by SEC and N-Linked oligosaccharide proles to results from commercial scale. Results for viable cell density, viability and normalized titer are shown in Figure 3. The viable cell density and viability trends at the two scales were comparable. The titer was slightly lower at commercial scale than in the 7-L scale-down bioreactors. The protein produced by the 7-L scale-down bioreactors had

Figure 3.

The qualication of the 7-Lscale-down model included a comparison of (A) the viable cell density and viability proles and (B) normalized titer results.

898

Biotechnology and Bioengineering, Vol. 106, No. 6, August 15, 2010

comparable HMW and N-Linked oligosaccharide prole to commercial scale. Inoculum Screening Studies The purpose of the vial thaw and inoculum steps in successive shake asks is to produce the minimum number of viable cells required to inoculate the subsequent culture vessel. To achieve an efcient design, combinations of operating conditions in the shake ask stage were evaluated together with those in the seed bioreactor to study their collective impact on the entire inoculum expansion process. A DOE approach was utilized to study the impact of inoculum expansion and seed bioreactor parameters on cell culture performance at the seed bioreactor stage. A two-level Resolution IV factorial design was conducted. A total of 24 runs from vial thaw through 7-L scale-down seed bioreactor were conducted in two blocks. Factors A through F in Table II were studied and evaluated for their effect on seed bioreactor growth rate and nal viability. Multiple seed bioreactors are often utilized in cell culture manufacturing processes. The function of the rst seed bioreactor(s) is to expand culture volume. The nal seed bioreactor (the n 1 seed bioreactor) is used to inoculate the production bioreactor and has a greater potential to impact the performance of the production bioreactor. In this report, the seed bioreactor step was evaluated in two studies. During the inoculum screening study, the effects of seed bioreactor parameters on cell growth and viability in the seed bioreactor were evaluated. In the production bioreactor screening study described below, the effects of the n 1 seed bioreactor parameters on the performance of the production bioreactor were evaluated. The shake asks were cultured at different incubator carbon dioxide setpoints and were seeded at varying initial viable cell densities and passaged when the target nal viable

cell density was achieved. The average culture duration for the shake ask steps ranged from 35 to 126 h. To ensure sufcient cell doublings for the culture to show effects from sub-optimal conditions, shake asks were repeatedly passaged to at least 10 cell doublings prior to inoculation of the seed bioreactors. The seed bioreactors were inoculated at varying initial viable cell densities and operated at different temperature and pH setpoints. The viability and growth rate in the seed bioreactors were determined after 4 days. Regression models for the specic growth rate and nal viability in the seed bioreactor are shown in Table III. The equations are described in terms of coded factors. During the execution of the studies, factors were studied at low, midpoint and high levels. In the model equations, the values of the coded factors at the midpoint condition are zero and the values at the low and high conditions are 1 and 1, respectively. The R-Squared term (R2) is an indication of how well the model ts the experimental data. However, this value will inherently increase as more factors are included in the model. The adjusted R2 is also an indication of how well the model ts the experimental data, taking into account the number of factors evaluated. The predicted R2 is calculated using regression analysis and is used to indicate how well the model will predict responses for new combinations of factors within the 1 to 1 ranges evaluated to develop the model. The P-value indicates the overall signicance of the model. Since the study was a Resolution IV design, interactions of two factors (e.g., CD) were aliased with other two parameter interactions. Three parameter interactions (e.g., BCF) were also aliased with other three parameter interactions. To determine which interaction to include in the model, a comparison of the R2 values and F-values was conducted. The aliased terms were chosen so that the resulting model had the maximum F-value as well as adjusted R2 and predicted R2 values that were within reasonable agreement with each other.

Table III.

ANOVA models for seed bioreactor and production bioreactor responses from screening studies. R2 0.99 Adjusted R2 0.97 Predicted R2 0.62 Model P-value <0.0001

Response Seed bioreactor specic growth rate (day 1) Seed bioreactor nal viability (%) Titer (normalized) HMW (%) G0F (%) G1F (%) CEX Peak 1 (%) CEX Peak 6 (%)

Model 0.512 0.049E 0.044EF 0.034B 0.031C 0.031F 0.027A 0.026DE 0.022BF 0.019BC 0.019BCF 0.018BE 0.014D 0.013BEF 0.012DF 96.65 1.19F 1.18D 1.14EF 1.10BE1.03BD 1.03DE 0.63B 0.60BDE 0.56BF 0.51A 0.44BC 0.16C 0.05E 1.00 0.18D 0.13DH 0.12J 0.10L 0.09H 0.08GJ 0.08HK 0.06G 0.06K 1.51 0.24J 0.16K 0.14L 60.68 5.37K 3.40H 2.42GE 2.17EK 1.50I 1.50G 0.40E 29.14 3.87K 3.18H 2.32HJ 1.99IJ 1.43G 1.34IK 0.83J 0.47I 7.88 0.86I 0.43HK 0.34H 0.11K 3.07 0.32K 0.22L 0.20D 0.19HL 0.14I 0.14DH 0.12H 0.10E

0.99

0.97

0.73

<0.0001

0.88 0.54 0.89 0.94 0.59 0.93

0.77 0.46 0.83 0.90 0.49 0.89

0.61 0.47 0.66 0.74 0.22 0.79

0.0008 0.0035 <0.0001 <0.0001 0.0047 <0.0001

Coded factor values range from 1 to 1.

Abu-Absi et al.: Dening Cell Culture Design Space Biotechnology and Bioengineering

899

All six factors studied and several interactions had signicant effects ( P < 0.01) on specic growth rate in the seed bioreactor. Key process parameters were selected by comparing the magnitude of the effects to the results of the centerpoint conditions. The relative standard deviation of specic growth rate for the centerpoint runs was 5%. Therefore, signicant model factors with coefcients greater than 5% of the model-predicted centerpoint growth rate, (>0.05 0.512 >0.026), were considered for designation as key process parameters. These factors included E, EF, B, C, F, and A. Shake ask carbon dioxide concentration (A) was not designated as a key process parameter since it is easily controlled within the range studied and the interaction EF was not designated since it was aliased with other two-parameter interactions. Shake ask initial viable cell density (B), shake ask nal viable cell density (C), seed bioreactor pH (E) and seed bioreactor initial viable cell density (F) were selected as key process parameters due to their effects on growth rate in the seed bioreactor. The nal viability in the seed bioreactors was found to be high throughout the design, as shown in Figure 4. The model terms with signicant effects ( P < 0.01) on seed bioreactor nal viability included B, D, F, and several interactions. However, no process parameters were designated as key for impact to nal viability since the magnitude of the effects was small and the variability of the response was negligible. The relative standard deviation of nal viability for the centerpoint runs was less than 1%. The inoculum studies enabled a rapid assessment of the shake ask and seed bioreactor steps. They conrmed the robustness of the inoculum expansion steps and the appropriateness of the operating ranges.

Production Bioreactor Screening Studies Factors D, E, and G through L were studied for their effect on the performance of the production bioreactor. Operating conditions in the nal seed bioreactor and in the production bioreactor were studied in combination to enable an efcient experimental design. In addition, the study enabled an assessment of the impact of seed bioreactor operational parameters on the production bioreactor. A two-level Resolution IV factorial design was conducted in 24 runs in two blocks. The operating ranges for A, B, C, and F were in alignment with the control strategy dened by the inoculum study. The performance of the production bioreactors was assessed by examining nal product titer, percentage of high molecular weight (HMW) aggregate species, N-Linked oligosaccharide prole and analytical CEX charge variant prole. In addition to the desired protein, process-related impurities and product variants are produced in cell culture. Controlling these species is crucial for achieving the desired product quality in drug substance. This is particularly important for the demonstration of product comparability with a previous version of the manufacturing process. HMW aggregate levels represent a key area of concern for biopharmaceuticals due to their potential immunogenicity. This product-related impurity poses a challenge for clearance in the downstream purication process. Therefore, it is important to quantify the impact of production bioreactor operating conditions on HMW levels. While glycosylation has not been directly linked to efcacy for this product, N-Linked glycosylation patterns

Figure 4.

Summary of nal viability results from the seed bioreactor study. All combinations of conditions resulted in cell viabilities greater than 90%, demonstrating the robustness of the inoculum expansion process.

900

Biotechnology and Bioengineering, Vol. 106, No. 6, August 15, 2010

have been shown to impact effector functions for monoclonal antibodies (Jefferis, 2005). Glycosylation is known to be directly inuenced by the cell culture process conditions and is typically not signicantly altered through downstream purication for monoclonal antibodies. Hence, control of the upstream process is important to ensure the correct ratio of glycoforms. The N-Linked oligosaccharide prole was quantied in terms of percent distribution between three isoforms, G0F, G1F, and G2F. The isoforms are designated as GXF, where X represents the number of terminal galactose units on the biantennary structure (Jefferis, 2005). Charge variants are monitored to assess the heterogeneity of the product. Methods such as isoelectric focusing (IEF) are typically included in the product release analytical testing for biopharmaceuticals. An analytical CEX method was developed to characterize the charge variants for this product. The prole results in several peaks. Peak 1 has been identied as deamidated forms of the antibody. Peak 6 has been shown to correlate with increased basic banding in IEF. Use of the CEX method provided quantitative results by which to characterize the production bioreactor step. In addition, the CEX prole remains unchanged through the downstream process (not shown), so control at the production bioreactor step is necessary. The centerpoint 7-L production bioreactor runs attained a day 14 normalized titer range of 1.41.8 and a HMW range of 1.72.3%. Titer results were normalized to the modelpredicted centerpoint titer. The factorial runs attained a day 14 normalized titer range of 0.51.8 and a day 14 HMW range of 1.02.6%. The amount of HMW remained low (<3.1%) for the duration of each of the runs (up to 17 days). The overall N-Linked oligosaccharide prole showed little variation with culture duration as shown in Figure 5. In general, an overall minor increase was observed for the G0F peak, while an overall minor decrease was observed for the G1F peak and G2F peak (not shown) between days 12 and 17. The G0F results for the factorial conditions trended lower than the centerpoint conditions and G1F factorial results trended higher than centerpoint conditions. Therefore, when interpreting the results of the studies, low G0F and high G1F were considered unfavorable. The models generated from this dataset for normalized titer, HMW, N-Linked oligosaccharide prole and CEX prole are shown in Table III. Most of the factors studied for the production bioreactor had an inuence on the nal product titer. In contrast, only a few parameters had an inuence on the high molecular weight aggregate level and CEX Peak 1 at harvest. Several parameters inuenced the N-Linked oligosaccharide prole and CEX Peak 6. The model terms with signicant ( P < 0.05) effects on titer were D, J, L, H, DH, and GJ. The relative standard deviation of day 14 titer from the centerpoint runs was 8.6%. Therefore, signicant model factors with coefcients greater than 8.6% of the model-predicted centerpoint titer, (>0.086 1.0 >0.086), were selected as key process parameters. The interaction of DH was not identied as key due to the fact that it was aliased with other two-

Figure 5. Trends in N-Linked oligosaccharide prole with culture duration for factorial and centerpoint conditions. A: Factorial points for G0F trended lower than the centerpoint results and (B) factorial points for G1F trended higher than the centerpoint results.

parameter interactions. Seed bioreactor temperature (D), production bioreactor temperature shift timing (J), production bioreactor pH (L) and production bioreactor initial viable cell density (H) were selected as key process parameters due to their individual effects on titer. The seed bioreactor temperature and timing of the temperature shift in the production bioreactor had a signicant inuence on titer due to their effect on peak viable cell density. Lower seed bioreactor temperatures and later temperature shifts correlated with higher peak viable cell density (not shown). Likewise, higher titer correlates with higher peak viable cell density. HMW levels did not vary signicantly for the factorial conditions compared to the centerpoint conditions, and hence the correlation to operational parameters was weak. This is an important result since it shows that this productrelated impurity level was not signicantly affected by the seed bioreactor or production bioreactor parameters. The model terms with signicant ( P < 0.05) effects on HMW were J and K. However, no process parameters were designated as key for impact to HMW because the magnitude of the effects was small and the variability of the response was negligible. The relative standard deviation of HMW for the centerpoint runs was 10%. The model terms with signicant ( P < 0.05) effects on the G0F peak were K, H, GE, and EK. The relative standard deviation of the G0F results for the centerpoint runs was 3%.

Abu-Absi et al.: Dening Cell Culture Design Space Biotechnology and Bioengineering

901

Therefore, signicant model factors with coefcients greater than 3% of the model-predicted centerpoint G0F peak, (>0.03 60.7 >1.8), were selected as key process parameters. The interactions GE and EK were not identied as key due to the fact that they were aliased with other twoparameter interactions. Production bioreactor pH (K) and initial viable cell density (H) were selected as key process parameters due to their individual effects on G0F. For the G1F peak, the model terms with signicant effects ( P < 0.05) were K, H, G, HJ, IJ and IK. The relative standard deviation of the day 14 G1F peak from the centerpoint runs was 5%. Therefore, signicant model factors with coefcients greater than 5% of the model!predicted centerpoint G1F peak, (>0.05 29.1 >1.5), were selected as key process parameters. The interactions HJ and IJ were not identied as key due to the fact that they were aliased with other twoparameter interactions. Production bioreactor pH (K) and initial viable cell density (H) were selected as key process parameters due to their individual effects on G1F. The relative standard deviation of the day 14 CEX results for the centerpoint runs was 8%. Therefore, factors with effects greater than 8% of the model-predicted centerpoint value, (>0.08 7.9 >0.63; >0.08 3.1 >0.25), were considered to be key. Production bioreactor temperature (I) was the only model term with signicant ( P < 0.05) effect on CEX Peak 1. It was designated as a key parameter, since the magnitude of its effect was 11% of the model-predicted centerpoint value. Factors with signicant ( P < 0.05) effect on CEX Peak 6 were K, L, D, HL, I, DH, H and E. Production bioreactor pH (K) was designated as a key parameter, since the magnitude of its effect was 10% of the model-predicted centerpoint value.

Figure 6 shows overlay plots summarizing the results of the production bioreactor screening study. Regions of concern in terms of high levels of CEX Peaks 1 and 6 and unfavorable trends in N-Linked oligosaccharide prole are shaded dark gray. The normal operating ranges are indicated in the plots and are a safe distance from the regions of concern. Of the key parameters identied (B, C, D, E, F, H, I, J, K, and L) three parameters (H, I, and K) were chosen for further study due to their potential to impact the critical quality attributes of the drug substance. An RSM study was designed to verify the designation of these parameters as CPPs and to conrm their acceptable ranges.

Impurity Clearance Through the Downstream Process In addition to HMW levels described earlier, it is important to determine if the levels of other process-related impurities, such as DNA and host cell protein (CHOP) from the Chinese hamster ovary cells, can be correlated to upstream operational parameters. This can help assess the burden on the downstream process to clear impurities. For this process, only a weak correlation was observed between in the overall level of CHOP in the cell culture harvest and the nal nonviable cell density in the production bioreactor (not shown). However, a better correlation was observed between the CHOP levels after the rst chromatography step in the downstream purication process (Protein A chromatography) and the cell culture performance in the production bioreactor (not shown). A small percentage of cells in the bioreactor lyse prior to and during harvest, releasing host cell protein impurities into the cell culture uid. Increased

Figure 6. Overlay plots showing results of production bioreactor study. A: Production bioreactor pH versus temperature and (B) initial viable cell density versus temperature. Shaded areas represent regions of concern in terms of N-Linked oligosaccharide and CEX proles. The normal operating ranges are also shown.

902

Biotechnology and Bioengineering, Vol. 106, No. 6, August 15, 2010

nal non-viable cell density implies a larger release of CHOP from dying cells. It has been shown that monoclonal antibodies can bind CHOP from cell culture harvest uid and carry these proteins through Protein A purication (Shukla and Hinckley, 2008). Higher levels of CHOP in the harvest uid are often not independently observed due to the high overall CHOP level and the relative insensitivity of CHOP assays when high levels are present. However, this can be manifested through higher levels of CHOP in the Protein A eluate pool due to co-purication with the product. The downstream process was shown to be capable of clearing the extreme levels of CHOP and host cell DNA produced during the production bioreactor characterization experiments. As a result, no critical operational parameters were identied in the upstream process based upon effects to these impurities. However, since a correlation was observed between nal viability and CHOP levels, and to ensure that longer culture duration does not impact CHOP levels, harvest viability was designated as a CPP. The acceptable

limit for harvest viability was set based upon the lowest harvest viabilities generated during the study. Production Bioreactor RSM Study A face-centered central composite design (CCD) was conducted to study production bioreactor temperature (I), pH (K) and initial viable cell density (H), due to the demonstration of their impact to glycosylation and charge variants, described earlier. The goal of the CCD study was to conrm the previous results and ensure that the CPP ranges were appropriately set. The CCD design enabled the inclusion of curvature in the regression models. The inclusion of several center point runs also conrmed the robustness of the operating conditions selected for the production process. To explore the edge of failure, the ranges for temperature and initial viable cell density were expanded from what was evaluated in the screening study. The temperature range studied in the CCD was twice as wide as the range in the

Figure 7. Contour plots showing results of CCD study of the production bioreactor for (A) Normalized Titer, (B) CEX Peak 1, (C) CEX Peak 6, and (D) Final viability. The acceptable ranges for temperature and initial viable cell density were constrained to ensure acceptable levels of CEX Peak 1 (deamidation), CEX Peak 6 (correlates to basic banding in IEF gels) and nal viability (correlates to CHOP levels).

Abu-Absi et al.: Dening Cell Culture Design Space Biotechnology and Bioengineering

903

Figure 8.

Prediction proler results for the effect of temperature on N-Linked oligosaccharide prole. Constraint of temperature range ensures acceptable levels of G0F and G1F.

screening study, and the upper range of the initial viable cell density was increased. Figure 7 shows contour plots of the results of the RSM study with coded values for the three factors. During development, the process had been optimized for titer, as demonstrated by the bulls-eye near the centerpoint conditions for temperature and pH on the plot of normalized titer. The results of the CCD study shown in Figure 7A also corroborated the nding in the screening study that production bioreactor temperatures between 1 and 1 do not signicantly affect titer. The CEX results indicated high Peak 1 percentages at high temperature and high Peak 6 levels at high initial viable cell density combined with high pH. The effect of high initial viable cell density on Peak 6 was not identied in the screening study because of the more narrow range evaluated. In addition, nal viability was lower at high temperature, especially when combined with high pH.
Table IV.

Temperature, pH and initial viable cell density were conrmed to impact N-Linked oligosaccharide proles. Temperature was not identied as a key parameter for its effect on N-Linked glycosylation in the screening study. However, the evaluation of a wider temperature range enabled a demonstration of the edge of failure with respect to the N-Linked oligosaccharide prole and resulted in the designation of this parameter as a CPP. Figure 8 shows the prediction proler for the G0F and G1F species versus production bioreactor temperature. G0F decreases and G1F increases with increasing temperature. At a lower temperature set point, the effects of pH and initial viable cell density on glycosylation are minimized (not shown), thus improving overall process robustness. Of course, a lower production bioreactor temperature also decreases product titer. The prediction models revealed an optimization of product titer in conjunction with maintenance of comparable glycosylation patterns. The factors found to impact glycosylation for the monoclonal antibody are in general agreement with process parameters indicated in the literature to be important for glycosylation. Production bioreactor temperature, duration of the cell culture run and timing of inducer addition were found to be important for N-Linked glycosylation in CHO cell culture (Andersen et al., 2000). Studies on other glycoproteins indicate that culture pH (Gawlitzek et al., 2009) and culture temperature (Ahn et al., 2008) are the most signicant factors that impact glycosylation. Due to the results of the CCD study, the upper limits for temperature and initial viable cell density were constrained. A second CCD study was then conducted to ensure that the rened edges of the design space were acceptable (results not shown). Production bioreactor temperature (I), pH (K) and initial viable cell density (H) were designated as CPPs. Table IV summarizes the factors that were studied and their

Factors of statistical signicance and designations of key and critical operational parameters. Critical process parameters identied in CCD studies

Result No key or critical process parameters identied Key parameters identied Critical process parameters identied

Response Seed bioreactor nal viability HMW Seed bioreactor growth rate Titer CEX prole

Factors of statistical signicance in screening studies B, D, F, BC, BD, BE, BF, DE, EF, BDE J, K A, B, C, D, E, F, BC, BE, BF, DE, EF, BCF D, H, J, L, GJ, DH I E, D, H, I, K, L, DH, HL H, K, GE, EK G, H, K, HJ, IJ, IK

Key process parameters identied in screening studies

B, C, E, F D, H, J, L I K H, K H, K

Peak 1 Peak 6 G0F G1F

I Ha, K H, Ib, K H, Ib, K

N-Linked oligosaccharide prole


a

The upper limit of H, production bioreactor initial viable cell density, evaluated in the CCD study was higher than the upper limit evaluated in the screening study. It was not identied as a key parameter for effects on Peak 6 in the screening study. However, the results of the CCD study revealed the edge of failure for the upper limit of H, so it was deemed as a CPP for its effect on Peak 6. b The range of I, production bioreactor temperature, evaluated in the CCD study was two times wider than the range evaluated in the screening study. It was not identied as a key parameter for effects on N-Linked oligosaccharide prole in the screening study. However, the results of the CCD study revealed the edge of failure for the upper limit of I, so it was deemed as a CPP for its effect on N-Linked oligosaccharide prole.

904

Biotechnology and Bioengineering, Vol. 106, No. 6, August 15, 2010

designation as key or critical operational parameters. In addition to the factors in the study, harvest viability was designated as a CPP based on the ability of the downstream process to clear CHOP.

Conclusions
A tiered approach to process characterization provided a streamlined mechanism for studying the design space for a cell culture process. The FMEA exercise was followed by characterization of the inoculum and production bioreactor stages of the process. An integrative study of shake ask and seed bioreactor parameters and of seed bioreactor and production bioreactor parameters minimized the total number of experiments while yielding valuable information for parameter classication. The impact of worst-case cell culture process operation on product quality after downstream purication was considered. Production bioreactor parameters that impacted product glycosylation included pH, temperature and initial VCD. Production bioreactor temperature also impacted deamidation. Charge variants indicative of basic banding in IEF gels were inuenced by production bioreactor initial viable cell density and pH. Upstream operating conditions were shown to have no other impact on product quality, but harvest viability correlated to CHOP levels in the Protein A eluate. The development of predictive models for the cell culture process demonstrated optimization of process operating conditions to balance high product titer with control of product glycosylation.

References
Ahn WS, Jeon JJ, Jeong YR, Lee SJ, Yoon SK. 2008. Effect of culture temperature on erythropoietin production and glycosylation in a perfusion culture of recombinant CHO cells. Biotechnol Bioeng 101(6):12341244. Andersen DC, Bridges T, Gawlitzek M, Hoy C. 2000. Multiple cell culture factors can affect the glycosylation of Asn-184 in CHO-produced tissue-type plasminogen activator. Biotechnol Bioeng 70(1):2531. US Food and Drug Administration. 2004. Pharmaceutical cGMPs for the 21st century; a risk based approach. US Food and Drug Administration. 2006. Guidance for industry; Q8 pharmaceutical development. US Food and Drug Administration. 2009. Guidance for industry; Q8 pharmaceutical development (Revision 1).

Gawlitzek M, Estacio M, Furch T, Kiss R. 2009. Identication of cell culture conditions to control N-glycosylation site-occupancy of recombinant glycoproteins expressed in CHO cells. Biotechnol Bioeng 103(6):1164 1175. Harms J, Wang X, Kim T, Yang X, Rathore AS. 2008. Dening process design space for biotech products: Case study of Pichia pastoris fermentation. Biotechnol Prog 24(3):655662. Jefferis R. 2005. Glycosylation of recombinant antibody therapeutics. Biotechnol Prog 21(1):1116. Kelley B, Jennings P, Wright R, Briasco C. 1997. Demonstrating process robustness for chromatographic purication of a recombinant protein. Biopharm Int 10(10):3647. Kelley B, Shi L, Bonam D, Hubbard B. 1998. Robustness testing of a chromatographyic purication step used in recombinant Factor IX manufacture. In: Kelly B, Ramlemeier A, editors. Validation of biopharmaceutical manufacturing processes. Washington, DC: American Chemical Society. p 93113. Kozlowski S, Swann P. 2006. Current and future issues in the manufacturing and development of monoclonal antibodies. Adv Drug Deliv Rev 58(56):707722. Li F, Hashimura Y, Pendleton R, Harms J, Collins E, Lee B. 2006. A systematic approach for scale-down model development and characterization of commercial cell culture processes. Biotechnol Prog 22(3):696703. Lipsanen T, Antikainen O, Raikkonen H, Airaksinen S, Yliruusi J. 2007. Novel description of a design space for uidised bed granulation. Int J Pharm 345(12):101107. Nail S, Searles J. 2008. Elements of quality by design in development and scale-up of freeze parenterals. Biopharm Int 21(1):4452. Rathore AS, Winkle H. 2009. Quality by design for biopharmaceuticals. Nat Biotechnol 27(1):2634. Rathore A, Branning R, Cecchini D. 2007. Design space for biotech products. Biopharm Int 20(5):3640. Rathore A, Saleki-Gerhardt A, Montgomery S, Tyler S. 2008. Quality by design: Industrial case studies on dening and implementing design space for pharmaceutical processesPart 1. Biopharm Int 21(12): 3747. Rathore A, Saleki-Gerhardt A, Montgomery S, Tyler S. 2009. Quality by design: Industrial case studies on dening and implementing design space for pharmaceutical processesPart 2. Biopharm Int 22(1):40 44. Seely R, Haury J. 2005. Application of FMEA to biotechnology manufacturing processes. In: Rathore A, Sofer G, editors. Porcess validation in manufacturing of biopharmaceuticals. Boca Raton, FL: Taylor and Francis. p 3168. Seely J, Seely R. 2003. A rational step-wise approach to process characterization. Biopharm Int 16(8):2435. Shukla AA, Hinckley P. 2008. Host cell protein clearance during protein A chromatography: Development of an improved column wash step. Biotechnol Prog 24(5):11151121. Shukla AA, Sorge L, Boldman J, Waugh S. 2001. Process characterization for metal-afnity chromatography of an Fc fusion protein: A design-ofexperiments approach. Biotechnol Appl Biochem 34(Pt 2):7180.

Abu-Absi et al.: Dening Cell Culture Design Space Biotechnology and Bioengineering

905