OR WAIT 15 SECS
Strategies for transfer of the manufacturing process.
Since 1986, Genentech has achieved 25 successful transfers of commercial drug substance (DS) processes to both internal manufacturing sites and external manufacturing facilities such as Genentech's partners and contract manufacturing facilities. These 25 transfers include both recombinant protein and monoclonal antibody products from E. coli and Chinese Hamster ovary (CHO) host cells. More than 80% of these transfers have occurred since 2000 (1). This 26th article in the Elements of Biopharmaceutical Production series presents an overview of the various challenges that are encountered during technology transfer of biotech processes along with potential solutions that can address theses challenges. The artlcle describes the process-transfer strategy, which includes three key steps: facility fit analysis, risk-mitigation plan preparation, and risk-mitigation plan execution.
The first step, facility fit analysis, or gap analysis, is a thorough process walk-through at the receiving site based on process-requirements documents. This exercise involves a detailed step-by-step analysis of how the entire manufacturing process would be performed at the receiving site and its goal is to identify any potential process, facility, or procedure changes that may be necessary to fit the process into the receiving site facility. The output of this exercise guides the scope of the process-transfer project. Therefore, the exercise must be detailed and cover the entire manufacturing process as well as required ancillary equipment, analytical instruments, small-scale laboratory capabilities, and all relevant operational practices at the receiving site. The output of the facility fit analysis should be a comprehensive list of possible gaps and should be jointly developed by both the sending and receiving sites.
The second step is to analyze each gap, assess its potential risk to the process and product, and formulate a risk mitigation plan using the quality risk management concept. Each gap presents a potential risk to the process, product, or both. The risk is classified based on its potential effect (2). The risk classification criteria are shown in Table I. Risk assessment relies on process and product knowledge obtained throughout the lifecycle of the product. The rationale for the risk classification of each gap must be scientifically sound.
Table I: Risk category.
For risks that are considered medium or high, a risk-mitigation strategy is assembled. In this case study, four risk examples and their risk-mitigation strategies are discussed in detail. In general, the risk-mitigation effort is proportional to the risk classification (i.e. high risks require a larger effort to mitigate and control than do low risks). Ideally, the risk-mitigation effort results in the reduction of a risk. In cases where the risk effect can not be reduced, additional risk-control strategies should be put in place. The risk-assessment, risk-mitigation, and risk-control strategies are documented in the risk-mitigation plan. This document is revised as additional information becomes available.
The third step is to execute the risk-mitigation plan, which includes four sets of activities: transfer and verification of a scale-down model (SDM), evaluation of potential process changes using the SDM, execution of full-scale engineering runs, and evaluation of results from engineering runs. Establishing a SDM at the receiving site enables the receiving site to prospectively evaluate potential process changes that may be necessary for facility fit reasons, and to troubleshoot performance issues during and after the completion of a transfer. Process changes that are evaluated using the SDM should be independent of scale and of the equipment involved. Two examples of this type of change are given (see examples 2 and 4). For changes that are scale- or equipment-dependent, it is typically necessary to conduct full-scale evaluations (i.e., engineering runs) before the initiation of the qualification campaign. Examples 1 and 3 describe this type of situation. The execution of the engineering runs serves two purposes. First, it provides an opportunity to evaluate the full-scale process performance at the receiving site and to verify the findings of the small-scale studies for potential process changes. Second, it is used to assess the readiness of the receiving site for the qualification campaign to ensure a high degree of confidence that the qualification campaign will be successfully executed.
The predefined evaluation criteria for the engineering runs are developed jointly by the sending and receiving sites. Genentech uses two-tiered success criteria for this evaluation. The primary success criteria are the same as the acceptance criteria for process and product comparability. The secondary success criteria include criteria for secondary process-performance indicators (e.g., cell culture metabolic profiles), equipment-performance criteria (e.g., bioreactor temperature, agitation, pressure, dissolved oxygen, and pH control performances), and performance criteria for full-scale media and buffer solutions preparation. These secondary success criteria serve to identify any subtle performance differences between the sending and receiving sites. If the engineering runs fail any of the primary success criteria, a root-cause analysis must be performed and a path forward must be identified before the initiation of the qualification campaign.
In the case study presented here, the drug substance (DS) manufacturing process for a recombinant glycoprotein product was transferred to the receiving site. The product had been manufactured commercially at the sending site for over a decade. The DS process consists of cell culture and recovery operations. Cell culture includes three types of operations:
Recovery includes four types of operations:
The formulated bulk is filtered through a sterilizing-grade filter and stored frozen in storage vessels until further processing for fill and finish.
To illustrate how the quality risk-management concept can be applied to process transfers, four gaps identified as part of this case study and their associated risk assessment and mitigation strategies are discussed below. Two of the gaps were from the cell culture portion of the process, while the other two were from the recovery portion. Their risk classifications ranged from medium to high, depending on their potential effect on the process and product as outlined in Table I.
Gap example # 1: production bioreactor gap
During the process walkthrough, certain aspects of the bioreactor design configurations were found to differ between the sending and receiving sites, even though the production bioreactor designs at the sending and receiving sites were highly similar, having identical aspect ratios and baffle configurations. Specifically, one of the two impellers in the production bioreactor was different, and the maximum sparged air-flow rate for the production bioreactor at the receiving site was lower than that at the sending site.
The typical bioreactor operating parameters can be categorized into volume-independent and volume-dependent parameters (3). Volume-independent parameters include pH, dissolved oxygen (DO), and temperature-control set points, which remain constant during a process transfer or scale change (i.e., scale up or scale down). Volume-dependent parameters include parameters that can be scaled linearly, such as nutrient feed volumes, and parameters that can not be scaled linearly, such as agitation speed and sparged gas-flow rate in bioreactors. The criterion applied here to select the production bioreactor agitation speed at the receiving site was to maintain approximately the same energy dissipation rate as at the sending site, which was estimated by using the empirical correlation between energy dissipation rate and agitation speed (4).
In a bioreactor, oxygen and carbon dioxide are typically sparged to control the culture DO level and pH, respectively. For large-scale bioreactors (i.e., > 1,000 L), air is also sparged to reduce the CO2 concentration in the sparged gas and therefore increase the driving force for CO2 transfer from the culture liquid to the sparged gas (i.e., CO2 stripping). For large-scale bioreactors, mass transfer between the culture fluid and gas phase is mainly driven by sparging and the contribution to mass transfer from the overlay gas flow is negligible (5). The sparging mass-transfer coefficient can be empirically correlated with the energy-dissipation rate and the sparged gas-flow rate (6). Even though the energy-dissipation rate was maintained constant between the sending and receiving sites, the difference in the maximum sparged air-flow rate between the sending and receiving sites would likely result in different CO2 stripping effects and therefore different pCO2 levels. Differences in pCO2 levels could in turn cause differences in osmolality because the culture pH is controlled by based addition. Elevated pCO2 and osmolality levels have been shown by others to negatively affect cell growth and productivity in CHO cells (7, 8). Depending on the cell line, product-quality attributes such as glycoform distributions may also be affected (9). Additionally, elevated pCO2 and osmolality levels may contribute to poor cell culture process performance when coupled with other factors such as raw material issues (10). Therefore, this gap in the production-bioreactor configuration was classified as a medium risk.
Figure 1a: Production culture pCO2 profiles of the engineering runs. The black lines represent the expected range based on historical data from the sending site.
The risk-mitigation approach taken here was to first compare the existing standard sparged gas-flow control strategies between the sending and receiving sites. Based on the available process data from both sites (e.g. pH setpoint, bicarbonate concentration of the cell culture media, peak oxygen uptake rate, etc.), a sparged air-flow rate was selected for the full-scale engineering runs at the receiving site. The goal was to maintain the production culture pCO2 profile within the expected range that had been established based on the historical manufacturing data from the sending site. Additionally, a response plan was put in place before the initiation of the engineering runs so that the sparged air-flow rate could be adjusted if the real-time pCO2 profile of the first engineering run had deviated from the expected range. The pCO2 and osmolality profiles of the engineering runs are shown in Figures 1a and 1b, respectively. Both profiles were maintained well within the expected ranges, thereby demonstrating that the selected air-flow rate at the receiving site was adequate to achieve an equivalent pCO2 stripping effect to that of the sending site process. Therefore, the risk associated with the gap in the production bioreactor configuration was successfully mitigated.
Figure 1b: Production culture osmolality profiles of the engineering runs. The black lines represent the expected range based on historical data from the sending site.
Gap example # 2: bioreactor pH
Another gap that was identified during the facility fit analysis was that both the bioreactor inline pH probes and the bench-top pH meters were different between the sending and receiving sites. The inline pH probes measure the culture pH in the bioreactors. The probe signal is used to control the culture pH at a target set point via additions of both base and acid (or CO2 gas for bicarbonate-buffered media). During the set-up of a bioreactor, the glass pH probes are calibrated with two pH standards. They are then inserted into the bioreactor and steam sterilized in place along with the bioreactor, before the cell culture medium is transferred through sterilizing-grade filters into the bioreactor. As the steam-sterilization process affects the glass membranes of the inline pH probes and causes the probe-sensing signal to drift, the pH probes are typically calibrated again after the medium is batched into the bioreactor (11). This calibration is performed by taking a medium sample from the bioreactor, measuring the sample's pH using a bench-top pH meter, and then adjusting the pH probe reading to match that of the bench-top pH meter. Therefore, the differences in both the inline pH probes and the bench-top pH meters between the sending and receiving sites could result in an unintended shift in the culture pH. Culture pH is known to affect cell culture growth performance, cell specific productivity and glycosylation patterns for recombinant proteins produced in CHO cells (3, 12). For this reason, this gap in pH instruments presented a high risk to the success of the process transfer.
To mitigate this risk, a two-pronged approach was taken. First, because the bench-top pH meters were used to calibrate the bioreactor inline pH probes, it was important to understand if there was any offset in the pH readings between the different bench-top pH meters. Second, a study was performed to expand the understanding of the pH effect in the current state of the process. As mentioned, this process was approved more than a decade ago and several process improvements have been made since then. It was therefore important to understand the current state of the process when evaluating potential process changes.
A side-by-side comparison of the two different pH meters used at the sending and receiving sites was performed. Both cell free media samples and cell culture fluid (i.e., cell containing) samples were used in the comparison and an offset of about 0.1 pH unit was found (see Figure 2). A larger variability was observed in the pH readings of the cell culture fluid samples compared with the cell free media samples. This was not unexpected because cells continued to metabolize during the period between when the cell culture fluid samples were taken from bioreactors and when the pH readings were obtained from the pH meters, and cell metabolites such as lactic acid can change the sample pH.
Figure 2: pH offset between the receiving site and sending site bench-top pH meters (pH offset=receiving site pH meterâsending site pH meter). Open diamonds represent cell free media samples and closed circles represent cell culture fluid samples. The dashed line represents the average pH offset of all samples.
In the study to expand the understanding of the effect of pH, four different pH set points were compared (e.g., low-low, low, control, and high). The range of pH set points studied encompassed the control set point of the current process at the receiving site and the offset observed in the pH meter comparison study. A scale-down model was used for the study, the performance of which was previously demonstrated to be comparable to the full-scale process performance (data not shown). Cell culture performance indicators such as total cell mass, final cell viability and final product titer were evaluated for all four conditions. The culture fluid was purified so that product-quality analysis could be also performed. There was no significant effect of culture pH on total cell mass or final cell viability (data not shown). As shown in Figure 3, there was a significant effect of culture pH on cell specific productivity, where a higher pH set point resulted in lower specific productivity (p <0.05). Several product-quality attributes were measured for all tested conditions. The analysis of one glycosylation attribute, an indicator of glycoform distribution of the product protein, is shown in Figure 4 as an example. This glycosylation attribute was not significantly affected by pH, similar to the other product-quality attributes that were evaluated.
Figure 3: Bivariate fit of cell specific productivity (Qp) by pH set point (p=0.04).
Based on the results of the pH meter comparison study and the small-scale pH set-point study, it was determined that the bioreactor pH control set-point at the receiving site should be adapted to account for the offset from the pH meters. This correction would ensure that the actual culture pH at the receiving site matched that at the sending site. Furthermore, the small-scale study on different pH set-points may be leveraged to widen the proven acceptable pH set-point range for the production culture operation. In conclusion, the potential risk from the pH instrument gap was mitigated by adapting the pH control set-point.
Figure 4: Oneway analysis of glycosylation attribute by pH set point. Red lines represent the bulk product release specification.
Gap example # 3: primary recovery equipment
In the primary-recovery step of the process, the cell culture fluid from the production bioreactor is fed via a rotary pump to a multi-membrane tangential flow filtration (TFF) system. The filtrate (i.e., permeate) is collected in the harvest tank while the retentate is recycled back to the production bioreactor, as shown in Figure 5. The facility fit analysis indicated that the TFF feed pump was different between the sending and receiving sites. Because the receiving site was a multi-product facility and the feed pump was used for several products, it was highly desirable to adapt the process to the existing equipment at the receiving site if possible. The difference in the TFF feed pump could potentially change the turbulent-eddy size distribution to which the cells were exposed in the TFF flow path and thereby cause different levels of cell lysis during the primary recovery operation (13, 14). The performance of the primary recovery step, such as step yield and processing time, could be affected as a result. Additionally, cellular enzymes, such as glycosidases, proteases, or reductases, may be released as a result of varied cell lysis and could potentially affect product quality. An extremely high energy-dissipation rate due to the feed pump difference could also affect product quality. However, this was considered unlikely given the type of pump at the receiving site and available information from the literature on the effect of high shear on proteins (15, 16). Therefore, this gap in the TFF equipment was classified as a medium risk.
Figure 5: Process flow diagram of the tangential flow filtration step.
Laboratory-scale TFF systems typically do not represent the performance of manufacturing-scale systems well. Additionally, manufacturing-scale pump performance cannot be reproduced in the laboratory. Therefore, the risk-mitigation plan was to perform full-scale engineering runs to assess the potential effect on process performance and modify any TFF process parameters if necessary.
The TFF operation is run by controlling the filtrate flow rate at a predetermined target until the trans-membrane pressure (TMP) reaches a maximum limit, after which the filtrate-flow rate is reduced to maintain the TMP at the maximum limit. The switch between the concentration phase and the diafiltration phase is based on a target-concentration factor. TFF operation ends once a target diafiltration volume has been reached. The TMP and filtrate turbidity profiles of the engineering runs were compared with two typical manufacturing runs from the sending site in Figures 6a and 6b, respectively. It was evident from the first two engineering runs that the equipment difference likely resulted in a higher level of cell lysis and thus caused higher filtrate turbidity and TMP. In fact, the first engineering run was terminated before reaching the target diafiltration volume because of the low filtrate flowrate. For the third engineering run, the initial filtrate flow-rate target and the maximum TMP limit were reduced to improve the performance of the TFF step.
Figure 6a: Tangential flow microfiltration trans-membrane pressure (TMP) profiles of the engineering runs. Gray and back lines are typical TMP profiles from the sending site.
As shown in Figures 6a and 6b, although the total processing time was increased as a result of reducing the filtrate flow rate, these modifications improved the filtrate turbidity profile of the third engineering run. The overall step yield and product-quality results of the third engineering run were comparable to those of the typical runs from the sending site (data not shown). Therefore, the performance of the TFF system at the receiving site using the modified process parameters was deemed acceptable and the risk from the TFF equipment gap was successfully mitigated.
Figure 6b: Filtrate turbidity profiles of the engineering runs. Gray and back lines are typical profiles from the sending site.
Gap example # 4: viral clearance strategy
One risk that was identified early on during the transfer was not caused by facility fit constraints, but by the gap between the licensed process and the current industry standards and regulatory expectations around viral clearance. This risk was classified as a high risk. To mitigate this risk, two process improvements were made: improving the robustness of the acid-treatment step for viral inactivation, and changing the virus-removal filter to improve the process capability for removing adventitious viruses.
The effectiveness of viral inactivation via acid treatment depends on temperature, pH, and duration of the treatment. Although pH 3.8 inactivates model retroviruses, subsequent studies have found that the performance of viral inactivation is more robust at pH 3.6 and below (17, 18). An exploratory study was performed at small scale to evaluate the effects of lower pH and longer durations for the acid-treatment step. As expected, an inverse correlation was found between pH and the glycoprotein aggregate level in the pH neutralized pool. It was observed that lower pH resulted in higher aggregate levels, as shown in Figure 7. However, the subsequent chromatography steps were able to reduce the percent aggregate level in the purified pool well below the limit for the bulk-product release specification. In a follow-up study, several product-quality attributes including the aggregate level were analyzed to validate the acceptable pH range and duration for the acid-treatment step. Product-quality analysis for the engineering runs also demonstrated that the full-scale performance of this step was acceptable.
Figure 7: Effect of the target pH for the acid-treatment step on aggregate levels in the pH-neutralized pool and purified pool. The red line represents the limit in the bulk-product release specification.
In the past decade, more virus removal filters have become commercially available (19–21). The process transfer presented an opportunity to change the virus removal filter. A small virus-retentive filter was selected to replace the current virus-removal filter. This selection was based on considerations for the existing process capability for viral clearance, characteristics of the product pools (e.g., protein concentration), and current industry standards. Sizing of the new small virus-retentive filter was performed at small-scale. The performance of the viral-filtration step was also evaluated during the engineering runs to confirm appropriate sizing of this filter and acceptable process performance at full-scale.
Overall, the potential gap in the viral-clearance strategy of the process was closed by improving process robustness and leveraging current process technologies.
As demonstrated in this case study, taking a structured, documented, and disciplined approach to process transfers is crucial to the success of a transfer. While some process changes are inevitable due to facility fit reasons or differences in operational practices, it is important to avoid introducing unnecessary process changes as part of the transfer. Applying the quality risk-management concept in the process-transfer strategy ensures that the potential risk from each potential change is analyzed rigorously and the risk-mitigation and control measures are commensurate with the risks. With the proper use of both scale-down models and full-scale engineering runs, the chance of surprises during the qualification campaign can be greatly reduced.
The authors gratefully acknowledge Eric Ordonez, Tom Lecocq, Mark Iversen, Shirin Fuller, Sean Forestell, Raymond Arnold, Robert Kiss, and Harry Lam for their valuable technical contributions and critical review. We also thank Boehringer Ingelheim Pharma GmbH & Co. KG for their collaboration on the work presented in this article.
Jean Harms* is a technical manager in quality operations, and Purav Dave is an engineer II in global biologics manufacturing science and technology, both at Genentech, a member of the Roche Group. *To whom correspondance should be addressed, firstname.lastname@example.org.
1. K. Carswell, presentation at the IBC 24th International Conference, (San Diego, CA, 2009).
2. A. Mire-Sluis et al., BioProcess Int. 8 (3), 20–32 (2010).
3. F. Li et al., Biotechnol. Prog. 22 (3), 696– 703 (2006).
4. M. Winkler, "Problems in Fermenter Design and Operation," in Chemical Engineering Problems in Biotechnology, M. Winkler Ed, (Elsevier Science 1990), pp. 222–350.
5. S. Mostafa and X. Gu, Biotechnol. Prog. 19 (1), 45–51 (2003).
6. K. van't Riet, Trends in Biotechnol. 1, 113–119 (1983).
7. V. DeZengotita, A. Schmelzer, and W. Miller, Biotechnol. Bioeng. 77 (4), 369–380 (2002).
8. M. Zhu et al., Biotechnol. Prog. 21 (1), 70–77 (2005).
9. R. Kimura and W. Miller, Biotechnol. Prog., 13 (3), 311–317 (1997).
10. A. Kirdar, K. Green, and A. Rathore, Biotechnol. Prog. 24 (3), 720–726 (2008).
11. V. Saucedo et al., Biotechnol. Prog. 27 (3), 885–890 (2011).
12. M. Gawlitzek et al., Biotechnol. Bioeng. 103 (3), 1164–1175 (2009).
13. R. Godoy-Silva et al., Biotechnol. Bioeng., 103 (6), 1103–1117 (2009).
14. B. Vickroy, K. Lorenz, and W. Kelly, Biotechnol. Prog., 23 (1), 194–199 (2007).
15. Y. Maa and C. Hsu, Biotechnol. Bioeng., 51, 458–465 (1996).
16. C. Thomas, "Problems of Shear in Biotechnology," in Chemical Engineering Problems in Biotechnology, M. Winkler Ed, (Elsevier Science, 1990), pp. 25–93.
17. K. Brorson et al., Biotechnol. Bioeng. 82 (3), 321–329 (2003).
18. G. Miesegaes, S. Lute and K. Brorson, Biotechnol. Bioeng. 106 (2), 238–246 (2010).
19. G. Bolton, J. Basha. and D. LaCasse, Biotechnol. Prog. 26 (6), 1671–1677 (2010).
20. T. Hongo-Hirasaki, M. Komuro, and S. Ide, Biotechnol. Prog. 26 (4), 1080–1087 (2010).
21. G. Kern and Krishnan, BioPharm Int. 19 (10), 32–41 (2006).