The concept of “design space” has been proposed in the ICH Q8 guideline and is gaining momentum in its application in the biotech industry. It has been defined as “the multidimensional combination and interaction of input variables (e.g., material attributes) and process parameters that have been demonstrated to provide assurance of quality.” This paper presents a stepwise approach for defining process design space for a biologic product. A case study, involving P. pastoris fermentation, is presented to facilitate this. First, risk analysis via Failure Modes and Effects Analysis (FMEA) is performed to identify parameters for process characterization. Second, small-scale models are created and qualified prior to their use in these experimental studies. Third, studies are designed using Design of Experiments (DOE) in order for the data to be amenable for use in defining the process design space. Fourth, the studies are executed and the results analyzed for decisions on the criticality of the parameters as well as on establishing process design space. For the application under consideration, it is shown that the fermentation unit operation is very robust with a wide design space and no critical operating parameters. The approach presented here is not specific to the illustrated case study. It can be extended to other biotech unit operations and processes that can be scaled down and characterized at small scale.
The concept of “design space” has been lately receiving a lot of attention in the biotech community (FDA Guidance, 2006). It has been defined in the ICH Q8 guideline as “the multidimensional combination and interaction of input variables (e.g., material attributes) and process parameters that have been demonstrated to provide assurance of quality.” The guideline encourages inclusion of additional information demonstrating a higher degree of understanding of the manufacturing process, which can lead to more flexible regulatory overview. This is elucidated in the following excerpt from the guideline: “Working within the design space is not considered as a change. Movement out of the design space is considered to be a change and would normally initiate a regulatory post approval change process. Design space is proposed by the applicant and is subject to regulatory assessment and approval.” It has been suggested that while applying the concept of design space would result in ease of performing process improvements, this would require extensive process characterization studies that examine parameters in a wide range to provide the necessary process understanding (Rathore et al., 2007).
Process characterization studies are performed primarily at laboratory scale with the purpose to define the “design space” within which the process can operate and still perform in an acceptable fashion with respect to product quality and process consistency. Over the past decade these studies have been widely accepted in the biotech industry as a necessary precursor to a successful process validation. This is demonstrated by recent publications on related topics. Seely and Haury published use of Failure Modes and Effects Analysis (FMEA) as a tool that provides a rational approach to evaluating a process and generating a ranked order of parameters requiring process characterization (Seely and Haury, 2003). Several publications have pointed out the need of performing scaled-down modeling and qualification as well as the guidelines for doing so (Godavarti et al., 2005; Rathore et al., 2005). Seely and Seely have published a rational stepwise approach toward characterization of biotech processes (Seely and Seely, 2003). Shukla et al. demonstrated the utility of a Design of Experiments (DOE) approach for performing process characterization of a metal-affinity chromatographic purification process for an Fc fusion protein (Shukla et al., 2001). They used the results of the DOE to design worst case studies to examine the robustness of the process. Rathore et al. have elucidated the difference between process development and process characterization studies and the use of the latter for categorization of parameters as critical, key, and non-key (Rathore et al., 2007). More recently, publications have addressed the design and approach toward process characterization of cell culture (Li et al., 2006) and ion-exchange unit operations (Kaltenbrunner et al., 2007).
This paper presents a stepwise approach for defining process design space for a biotech product. A case study, involving Pichia pastoris fermentation, is presented to facilitate this. Tools such as FMEA, scale-down modeling, and DOE have been shown as effective for performing an efficient examination of process robustness. It is also shown that well-designed, process characterization studies can serve as a foundation for a successful process validation, regulatory filing/approval, and subsequent manufacturing over the lifecycle of the product. The approach presented here is not specific to the illustrated case study. It can be extended to other biotech unit operations and processes that can be scaled-down and characterized at small scale.
Materials and Methods
Yeast Strain and Fermentation Process. A methylotrophic yeast Mut+, Komagataella (Pichia) pastoris (P. pastoris), expressing a recombinant protein was used in this study. The production of the protein is methanol-inducible and secreted into the growth medium by the host cells. Bench top fermentation runs (referred to as scaled-down runs throughout this article) as well as two 300 L runs completed in the pilot facility were performed following the process flow diagram shown in Figure 1. The process utilized here was similar to those typically used in other published studies on Pichia fermentation (Cereghino et al., 2002; Sinha et al., 2003; Trinh et al., 2003). The process starts with thawing a working cell bank vial into a 2 L Erlenmeyer flask containing the inoculum expansion medium. During the inoculum development phase, the flask is incubated and agitated until a desired optical density (OD) is reached. The production fermentor is inoculated with the inoculum flask and then grown in batch phase until OD reaches a preset target. Then, Feed 1 is added at predefined rate to promote oxidative growth by limiting the broth glucose concentration. Throughout the batch and fed-batch growth phase, the culture pH and temperature are controlled at the same set points. Again, once a predetermined OD target is reached, the induction phase starts. The induction phase of the process consists of two distinctive phases, adaptation phase and production phase. During the adaptation phase, in addition to Feed 1, Feed 2 containing the inducer is also added. Once the duration of the adaptation phase reaches a predetermined target, Feed 1 is aborted and the flow rate of Feed 2 is adjusted to promote product formation. At the initiation of induction phase, the culture temperature is reduced to decrease the likelihood of proteolysis.
Harvest and Purification. Once the induction phase duration has reached a preset target, the culture is chilled before centrifugation. To harvest the culture from the scaled-down runs, cell broth was dispensed into 1 L centrifuge bottles and centrifuged for 1 h at 4000 RPM using a Beckman J6-HC centrifuge. After centrifugation was complete, the supernatant was collected, clarified using a Millipore A1HC depth filter, and purified using immuno-affinity chromatography (IMAC).
Off-Line Measurements. Both the YSI 2700 select and NOVA Biomedical BP300 machines were used for off-line measurements of glucose, acetate, lactate, Na+, K+, PO43-, and NH4+.
Percent Solids Determination. Percent solids in the fermentation broth was calculated based on weight using an analytical balance. A sample was weighed in a 2 mL microcentrifuge tube and centrifuged for 10 min at 4500 RPM in a Beckman model TJ-6 centrifuge. The supernatant was discarded and the weight of the pellet was measured. The % solids in the sample was then calculated.
Optical Density. To measure the optical density (OD), a sample was diluted with NaCl solution and measurement was performed with a Shimadzu 1201 spectrophotometer (600 nm wavelength).
Protein Concentration and Purity. Protein concentration at different time points was determined using an anion exchange high performance liquid chromatography (AE-HPLC1). This method utilized an Agilent 1100 equipped with a standard 10 mm flow cell; a DEAE-5PW column (Tosoh Bioscience 07164); gradient of solvent A [20 mM Tris, pH 8.5] and solvent B [20 mM Tris, 0.25 M sodium chloride, pH 8.5] at 0.5 mL/min in 35 min with detection at 230 nm.
Absorbance at wavelength 280 nm (A280) was also used to measure protein concentration of the IMAC column load and pool material.
The protein purity of the IMAC column pool was determined by reverse phase high performance liquid chromatography (RP-HPLC) and a second anion-exchange method (AE-HPLC2). The RP-HPLC method utilized an Agilent 1100 equipped with a standard 10 mm flow cell; a C4 column (Vydac 214TP54); gradient of solvent A [0.1% trifluoroacetic acid (TFA), 10% acetonitrile (AcN) in water] and mobile phase B [0.09% TFA, 2% isopropyl alcohol (IPA), 88% AcN in water] at 0.6 mL/min in 40 min with detection at 214 nm. The second AE HPLC method was very similar to the AE HPLC method described above except with a longer run time and a shallower gradient to allow for better resolution between the different species.
Software. JMP software version 6.0.2 (SAS Institute) was used to perform statistical analysis using the standard least-square model fit method and a p-value ≤ 0.1 was considered statistically significant.
Approach toward Process Characterization
The overall approach toward process characterization involved four key steps. First, risk analysis via FMEA was performed to identify parameters for process characterization. Second, small-scale models were created and qualified prior to their use in these experimental studies. Third, studies were designed using DOE in order for the data to be amenable for use in defining the design space. Fourth, the studies were executed and the results analyzed for decisions on the criticality of the parameters as well as on establishing design space. These steps will be discussed in more detail in the follow sections.
FMEA. FMEA is a useful tool to assess the degree of risk for every operation step in a systematic manner and to prioritize the experiments to evaluate their impact on the overall process performance. An approach to perform this analysis for biotech processes has been published earlier (Seely and Haury, 2005). Assessment is performed by a multifunctional team for severity, occurrence and detection. This team includes process experts and manufacturing equipment engineers who are familiar with equipment design and limitations. For each operating parameter in the unit operation, the severity score measures the impact on the process performance of an excursion that is three times the operating range. Occurrence and detection scores are based on an excursion just outside the operating range. While the occurrence score measures how frequently the excursion might occur, the detection score indicates the probability of timely detection and correction of the excursion. All three scores are multiplied to provide a Risk Priority Number (RPN) and the RPN scores are then ranked to identify the parameters with a high enough risk to merit process characterization.
Scale-Down Model Development and Qualification. To meet the accelerated drug development timeline and cost effectiveness, process characterization studies are primarily performed at laboratory scale. Therefore, developing a representative scale-down model is crucial to the success of process characterization. A general strategy for scale-down model development is to linearly scale down the scale-dependent operating parameters while keeping the scale-independent parameters at the same control set point as the large-scale process, assuming similar vessel geometries (Godavarti et al., 2005; Rathore et al., 2005; Schmidt, 2005; Li et al., 2006). For fermentation, scale-independent parameters include process temperature, pH, inoculation percentages (v/v) for each step, and timing of feed media additions. If oxygen transfer rates between the scales are equivalent, then the dissolved oxygen control setpoint and vessel backpressure should also be held constant.
Design of Process Characterization Studies. The objective of the process characterization studies is to evaluate process robustness and thus to define the “design space” within which the process can operate and still perform in an acceptable fashion with respect to product quality and process consistency. Data generated in these studies serves as a basis for categorization of parameters into critical (those that impact product quality), key (those that impact process consistency), and non-key (those that are neither critical nor key) and for establishing acceptance criteria for process validation. As mentioned in the section above, these studies are performed using qualified scale-down models and parameters identified via FMEA. A variety of DOE can be used depending on the number of parameters that need to be examined and the resolution required for the study (Kaltenbrunner et al., 2007; Li et al., 2006; Seely and Seely, 2003; Shukla et al., 2001; Rathore et al., 2003). These studies may be followed up by worst case studies, which are designed to examine the combination of worst case conditions within the operating space of the process.
Data Analysis and Definition of Design Space. At the conclusion of the process characterization studies, data are analyzed using statistical software to assess the variability in critical quality attributes caused by changes in the operating parameters. This assessment leads to definition of the process design space inside which the capability of the process to perform consistently and generate product of acceptable quality has been demonstrated. Figure 2 shows an illustration of a desirable outcome of a process characterization effort. Characterized space is examined and based on process sensitivities, design space is established. It is desirable to have the operating space nested within the design space. Once it has been established, the design space can be utilized for process validation and regulatory filing.
Results and Discussion
FMEA. The fermentation process was divided into three phases for risk assessment: pre-induction, induction, and induction +10 h. RPN scores were calculated for each of the three phases. Figure 3 shows the Pareto chart for the pre-induction phase as an example. Operating parameters with an RPN score of higher than 50 were characterized using a qualified scaled-down model. These included, pH, temperature, DO, OD inoculum, Feed 1 start OD, induction start OD, Feed 1 rates during fed-batch and adaptation, and Feed 2 rates during adaptation and production.
Scale-Down Model Development and Qualification. All scale-dependent operational control set points, except for agitation, were adjusted linearly by a scale factor equivalent to the ratio of overall process volumes. The volume-dependent parameters included pre- and post-sterilization volumes of growth media, feed media delivery rates, total airflow, oxygen flow rate and agitation. Equivalent vessel geometries and sparger design were assumed, and agitation was set to provide an equivalent power input per unit volume (P/V) (eq 1), which should result in similar mixing profiles across scales, and thus similar oxygen transfer and dispersion:
where Np = impeller power number, ρ = fluid density, N = impeller speed, D = impeller diameter, and VT = the total working volume of the reactor.
Three runs were performed using the scale-down model. Figure 4 shows a comparison between the scale-down model and the pilot-scale runs with respect to product accumulation (Figure 4a), dissolved oxygen profiles (Figure 4b), growth profiles (Figure 4c), and percent solids formation (Figure 4d). The optical density profiles of the three scaled-down runs were very similar to one of the pilot-scale runs (Figure 4c). For one of the pilot-scale runs, considerable variability in optical density was observed and this was due to variation in dilution and inaccuracy of the spectrophotometer at high cell densities. The consistency in % solids production and dissolved oxygen profiles between the pilot-scale and scale-down runs further demonstrated the comparability of oxygen demand and cell growth in the two scales as shown in Figure 4b and d. The product accumulation and final titer of the scale-down runs were within the ranges observed at pilot scale (Figure 4a).
Design and Results of Process Characterization Studies. As mentioned above, parameters were identified for process characterization based on FMEA. Due to the large number of parameters, three different studies were designed: optical density and feed rate screening, culture parameters, and protein stability. The optical density and feed rate study focused on characterizing the fermentation inoculum density, optical density based process triggers and feed rates during the batch and fed-batch stages. The culture growth parameter study was designed to investigate the effects of pH, temperature and dissolved oxygen levels on the growth kinetics during the batch and fed-batch growth phases of the fermentation. The protein stability study tested the effects of pH, temperature and dissolved oxygen levels on growth and productivity during the induction phase and the effects of pH, temperature and hold time during the post-induction hold step.
Optical Density and Feed Rate Screening. In methylotrophic yeasts such as Pichia, the alcohol oxidase promoter is activated by methanol and at the same time repressed by carbon source such as glucose or glycerol. The yeast is also sensitive to methanol. It has been reported that concentrations above 0.4% inhibit growth and production (Zhang et al., 2000; Trinh et al., 2003). Controlling specific growth rate and specific production rate through feed rates has been widely studied (Cereghino et al., 2002; Cos et al., 2006; Sinha et al., 2003; Trinh et al., 2003). In this study, the effects of fermentation inoculum density, optical density process triggers, and feed rates on culture growth and productivity during the batch and fed-batch stages were characterized. A resolution IV fractional factorial screening design was used to evaluate the effects of each parameter and their interactions with each other. A total of eight factors were tested at two levels in fours blocks with one center point per experimental block. In this study design, main effects are not confounded with either other main effects or two-factor interactions. However, two-factor interactions can be confounded with other two-factor interactions.
JMP analysis of the study results yielded two statistically significant effects. First, the effect of Feed 1 adaptation rate, Feed 2 induction rate, and the interaction of OD inoculation and OD adaptation on final OD (p = 0.04, 0.03 and 0.05, respectively) and second, the effect of Feed 2 induction rate on % solids (p = 0.02). The leverage plots for these parameters and their interactions shown in Figure 5 demonstrated that while these effects are statistically significant, they are of small magnitude. Further, there was no statistically significant impact on product titer from any of the parameters screened. Based on the results of this study, all parameters that were examined were classified as non-key operating parameters.
Culture Growth Parameters. Dissolved oxygen (DO) or partial oxygen pressure, pH, and temperature are among the most relevant physiological parameters for a fermentation process (Cereghino et al., 2002; Graumann and Premstaller, 2006). This study characterizes their effect on growth kinetics during the batch and fed-batch growth phases of the fermentation process. At the low end of the air flow rate range, DO could not be controlled adequately and fluctuated from 0 to 200%, resulting in stopped culture growth. The high flow rate condition did not have a significant impact on cell growth or productivity (data not shown). Therefore, only temperature and pH were characterized using a two level full factorial design.
Figure 6a and 6b illustrates OD growth and % solids profiles. A 4-h lag during the batch phase was observed at the lower temperature conditions compared to center point and the higher temperature conditions. The overall trends for both OD and % solids were comparable among all conditions tested. The final titer results are shown in Figure 6c. All titer values were normalized against the control. JMP analysis indicated that temperature had a statistically significant effect on titer with a p-value of 0.10. The magnitude of this effect was significant as shown in the leverage plot of temperature (Figure 6d). Overall, none of the characterized parameters had significant effect on culture growth. Since temperature had a significant effect on titer, it is categorized as a key operating parameter.
Protein Stability. Proteolytic degradation of secreted recombinant proteins in high cell density fermentation has been widely reported (Kobayashi et al., 2000; Sinha et al., 2003; Sinha et al., 2004). In methylotrophic yeasts such as Pichia, the proteolytic activity and degradation of recombinant proteins is attributed to methanol metabolism along with cell lysis toward the end of fermentation (Sinha et al., 2004). It has also been reported that proteolytic degradation can be minimized through careful manipulation of pH and temperature (Jahic et al., 2003). In this study, the effects of pH, temperature and dissolved oxygen levels on growth and productivity during the induction phase and on post-induction protein stability were examined. A two-level full factorial experimental design was used to estimate all main and interaction effects for the growth and productivity in the induction phase. The final fermentation optical density and % solids were examined using JMP to determine statistical significance. JMP analysis of the results showed the interaction between temperature and DO had a statistically significant effect on % solids and titer (p = 0.004 and 0.04, respectively, Figure 7). Temperature was also shown to have a statistically significant effect on titer (p = 0.07, Figure 7c). The scaled estimates for pH and temperature on titer (Figure 7c) showed that the effects of both pH and temperature were of a large magnitude. It is seen from examining the normalized titer results presented in Figure 8 that two runs (− − + and − + +) showed the lowest titer numbers. These batches both experienced transient pH spikes during the induction phase of the fermentation and based on prior experience, this was the primary cause for the atypical behavior. On the basis of the results presented in Figures 7 and 8, temperature, pH, and DO were categorized as key parameters for growth.
Post-induction protein stability was examined using experiments run through the induction production phase at center point. The temperature and pH ranges were characterized in a full factorial experiment. Samples were taken during a 24-h hold post-induction for titer analysis to characterize the hold time between the end of fermentation and the beginning of harvest. As shown in Figure 9, the 24-h post-induction titer was found to be similar for all temperature and pH conditions tested. Thus neither pH, temperature nor the interaction between pH and temperature was considered to have a significant effect on post-induction product protein concentration for this process. This study further demonstrated that the product is stable and there was no substantial proteolytic degradation post-induction.
Worst Case. Worst case conditions for the process were developed based on the results of the process characterization studies. This was achieved by picking the edge of the operating range such that it would have the greatest negative impact on cell growth and titer. As seen in the process characterization studies, the parameters that negatively affected cell growth also caused a decrease in titer. Therefore, a single worst case study could encompass the worst case conditions for both growth and titer. The process was run with the cumulative worst case conditions for each stage of the process (shake flask, seed fermentor, and production fermentor) to determine the effect of all worst case values on process performance. The worst case condition settings for the production fermentor are shown in Table 1. Two reactors were run with worst case conditions, and at the center point conditions. After fermentation was complete, the cell broth was harvested and subsequently purified in order to assess the impact of the worst case conditions on product quality.
Table Table 1.. Design of Worst Case Studies
worst case condition
culture growth temperature
OD and feed rates
The OD growth profiles for center point and worst case fermentations were similar with the exception of a 4-h lag present during the batch phase of the worst case fermentations (Figure 10a). This same trend was observed during the culture growth parameter study (Figure 6a) and was the result of the lower temperature set point. The worst case fermentation no. 2 OD toward the end of the fermentation was low compared to other batches. This was likely due to large DO fluctuations during this fermentation lot. The percent solids values for worst case fermentations were consistently around 3% lower than center point fermentations at comparable time points (Figure 10b, 11); however, the overall percent solids trends were similar for both conditions. The final OD and percent solids for the worst case conditions were about 8% and 3% lower, respectively, than the center point conditions. These minor differences in final OD and percent solids along with the similar profiles for OD and percent solids indicated the worst case conditions did not have a significant effect on cell growth. Supernatant from the center point and worst cases was recovered via centrifugation as described in the methods and materials section. The supernatant was processed using a scale down purification process to analyze purification productivity and product quality of the worst case conditions compared to center point conditions. The product protein yields through the IMAC step as determined by A280 and AE HPLC1 were similar for worst case conditions and the center point (data not shown). The product quality and impurity profiles as analyzed by RP-HPLC and AE HPLC2 were similar for both conditions (Table 2). The normalized product titer for 40-, 50-, and 60-h post-induction and final purified bulk product are shown in Figure 10c. All center point and worst case titer values are comparable and represented acceptable process performance. Overall, the worst case conditions did not have a substantial effect on product titer or quality, demonstrating the robustness of the process.
Table Table 2.. Normalized Product Titer and Quantity Results from Worst Case Study.
a DCCB: diafiltered concentrated clarified broth.
titer by A280 (mg/mL)
titer by AE HPLC1 (mg/mL)
% impurity 1
% impurity 2
% impurity 3
% impurity 4
worst case 1
main IMAC pool
worst case 2
main IMAC pool
main IMAC pool
Data Analysis and Definition of Design Space. On the basis of the results of the process characterization studies, it was found that none of the parameters have significant impact on productquality (no critical parameters). Further, temperature, pH and DO were found to impact cell growth and titer and thus were classified as key process parameters. As per the ICH Q9 guideline, process design space was established using the acceptable ranges for temperature, pH, and DO and this is illustrated in Figure 12. It can also be seen that the operating space is well nested inside the design space illustrating the robustness of the fermentation process.
Risk analysis was successfully utilized to identify operating parameters for process characterization. On the basis of the FMEA, pH, temperature, DO, OD inoculum, Feed 1 start OD, induction start OD, Feed 1 rates during fed-batch and adaptation and Feed 2 rates during adaptation and production were examined in process characterization studies. A 15 L fermentor model was developed and qualified to mimic the 300 L pilot-scale production process with respect to the growth kinetics, oxygen transfer, and protein of interest production profiles. Using the scaled-down model and design of experiments (DOE), characterization studies were performed. Of all operating parameters characterized, temperature, pH, and DO were identified as key operating parameters. All other parameters were non-key; no critical parameters were identified. A worst case study was performed using parameters identified from process characterization studies as having statistically significant impact on cell growth and titer. The overall cell growth, productivity, and product quality for worst case conditions were comparable to the center point runs. The results of the worst case study further demonstrated that the production fermentation process is robust. Finally, based on the results of the small-scale studies, acceptable ranges were set for the characterized operating parameters to define the design space for the product.
The approach presented here is not specific to the illustrated case study. It can be extended to other biotech unit operations and processes that can be scaled-down and characterized at small scale. For unit operations that cannot be accurately scaled down, alternate approaches, such as characterization at manufacturing scale, need to be employed to define design space.
The authors gratefully acknowledge Stephanie Tozer, Rajesh Krishnan, and Stephen Decker for their contribution to this work.