By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Because biotherapeutic products are complex and difficult to fully characterize, the process by which they are created is an important factor in defining the final drug product. As such, the characterization and validation of the processes are critical parts of ensuring that products are consistently safe and effective for patients. Quality-by-design (QbD) is a regulatory initiative that encourages a thorough understanding of the process variables, which affect the quality of the product and how those parameters interact with one another.1 Once approved, the QbD approach may then allow for more regulatory flexibility in manufacturing process changes. The process knowledge required for approval of a QbD filing involves the formation of a design space, which is defined as “the multidimensional combination and interaction of input variables and process parameters that have been demonstrated to provide assurance of quality.”1 As long as changes are made within the design space, the change can be dealt with by internal quality controls rather than requiring regulatory approval. A general guidance for the development of design spaces has been provided,1 but few examples of their development in biotechnology settings have been described.2–4 Additional case studies of design space development will provide the industry with the concepts needed to build systematic methods and take full advantage of the QbD regulatory initiative.
The first step in the QbD process involves determining critical quality attributes (CQAs), which are those attributes of a product that characterize its quality, safety, and efficacy. Although many products may have distinct CQAs, viral safety should be considered for all products, which are produced in mammalian cell culture. These processes are known to produce endogenous retroviruses or retrovirus-like particles (RVLPs),5–7 and they have the potential to be infected by adventitious viruses during mAb production.8 Any viral impurities that are or could be present in the product pools must be removed or inactivated by the product purification process.9–11 One unit operation that plays an important role in providing acceptable levels of the viral clearance CQA is the anion exchange chromatography (AEX) step. AEX performed in product flow-through mode has been shown to be capable of providing a high level of removal of both nonenveloped and enveloped viruses, including the endogenous RVLPs that are produced by CHO-cells.12–14 This process has also been shown to be very robust. It has been tested over wide ranges of parameters and shown to consistently provide log reduction values greater than 4.15, 16 In addition, our group has validated this process for many different mAb products and processes (Table 1). In any case, no specific product effect has been observed, and despite many differences between those processes, the step has consistently provided excellent LRV values. The robustness observed for viral clearance by AEX makes it an excellent candidate for the application of QbD concepts.
Table 1. LRV Values Obtained from Small-Scale Virus Reduction Experiments
In this work, we discuss strategies for developing design spaces for viral clearance and present examples of how we have applied these concepts to AEX processes. We first develop a formalized risk assessment to cover multiple AEX processes by using previous knowledge of the effects of parameters on viral clearance as well as the mechanism of action of the process. We then present experimental methods, which utilize that risk assessment as well as DOE techniques, to develop an understanding of process variability on viral clearance. Using a highly fractionated experimental design, we show that the AEX can provide complete removal of the endogenous RVLPs present in CHO-cell derived feedstocks over wide ranges of many process parameters. Finally, we present a full factorial design space in which very high LRV values are obtained for three different model viruses for all combinations of parameters within ranges typical of AEX manufacturing processes.
Materials and Methods
RVLP design space
Stock monoclonal antibody intermediate protein A pools were obtained from the mAb15 commercial-scale manufacturing operations. For each chromatography run, pool conditions were adjusted to varying levels as indicated in Table 2. The pool was first adjusted to the appropriate pH using tris-base or HEPES acid. The conductivity of each pool was then measured using a temperature-adjusted conductivity meter (SevenMulti, Mettler Toledo, Columbus, OH), and either WFI or sodium acetate was added to achieve the desired conductivity. Selected loads were also spiked to 0.1% V/V with harvested cell culture fluid (HCCF) to introduce increased levels of host cell impurities. The feedstock was then filtered through a 0.45 μm PVDF filter (Millipore, Billerica, MA), and the protein concentration was determined by optical density at 280 nm (OD280).
Table 2. RVLP Design Space Parameters
Both load density values were obtained for all runs.
A small-scale chromatography column (0.66-cm diameter, Bio-Chem Valve/OmniFit, Boonton, NJ) was packed with naive QSFF resin to 20-cm bed height. Chromatography runs were performed with varying parameters as described in Table 2. Column equilibration was first performed using tris-acetate equilibration buffers with varied compositions and for varied durations. The feedstock pool was then loaded onto the column, and collection of the flow-through fraction was started when the OD280 reached 0.1 unit above baseline. Protein was loaded to a total of 200 g mAb/mL resin, changing the collection vessel after the initial 50 g mAb/mL resin was loaded and allowing for analysis of both load densities from each chromatography run. The column was then washed with three column volumes (CV) of equilibration buffer, followed by regeneration with four CV of 0.5 N NaOH and three CV of 0.1 N NaOH. Aliquots of the product pools and remaining feedstock pools were collected and stored at −80°C. Extraction of viral genomic material was performed either using the Virus Mini Kit v2.0 on a BioRobot EZ1 (Qiagen, Valencia, CA) or the MagAttract Virus Mini Kit v1.3 on a BioRobot M48 (Qiagen, Valencia, CA). In both cases, extractions were followed by treatment with RNase-free DNaseI (Strategene, Santa Clara, CA) to remove any contaminating CHO-cell genomic DNA, and QPCR assays were performed as described previously.17
Model virus design space
Stock monoclonal antibody intermediate cation chromatography pools were obtained from mAb1 commercial-scale manufacturing operations. For each chromatography run, load pools were conditioned to varying levels as indicated in Table 4. The pool was first adjusted to the appropriate pH using 1.5 M tris-base. The conductivity of each pool was then measured using a temperature-compensated conductivity meter (SevenMulti, Mettler Toledo, Columbus, OH), and either WFI or 5 M NaCl was added to adjust the conductivity to the desired level. The feedstock was then filtered through a 0.22 μm PES filter, and the protein concentration was determined by optical density at 280 nm (OD280).
Table 3. LRV Values from Fractional Factorial Design Space of RVLP Removal
For each run, a small-scale chromatography column (0.66-cm diameter, Bio-Chem Valve/OmniFit) was packed with naive QSFF resin to 20 cm bed height and equilibrated with eight CV equilibration buffer (25 mM tris, NaCl, pH and conductivity adjusted to match load pool). The feedstock pool was then loaded onto the column, and collection of the flow-through fraction was started when the OD280 reached 0.1 unit above baseline. The feedstocks were loaded at varying flow rates and to varying load densities as indicated in Table 4, and the columns were then washed with three CV equilibration buffer. On completion of the chromatography run, the protein concentration of the product pool was determined by OD280, and aliquots of this pool as well as the remaining feedstock pool were collected and stored at −80°C. The samples were subject to treatment with DNaseI (Applied Biosystems, Foster City, CA), followed by extraction of viral genomic material using the EZ1 Virus Mini Kit v2.0 on a BioRobot EZ1 (Qiagen, Valencia, CA). QPCR assays used to quantify X-MuLV, SV40, and MMV viral particles were performed as previously described.13, 18, 19
Results and Discussion
As a multidimensional study of input variables, design space experiments have the potential to be quite extensive. When performed in product flow-through mode during mAb production, AEX is relatively straightforward compared with other bind-and-elute chromatography steps.20 However, we have identified 18 sources of variability for this unit operation. Using two-level screening designs to test variables at the high and low ends of a range, full factorial experiments to investigate all combinations require at least 2n runs, where n is the number of parameters. For AEX, testing all combinations of the 18 variables identified would require over 260,000 runs, an unfeasible number, especially for viral clearance experiments, which, due to the high costs of virus stocks, are generally expensive to perform. Therefore, we utilized two tools to reduce the number of runs required to perform design space studies: development of a formalized risk assessment and utilization of design-of-experiments (DOE) techniques.
The goal of a risk assessment is to provide an unbiased ranking of variables so that experiments can be devised that focus on the factors most likely to have an impact on process performance or have significant interactions with other factors. Because of the robust nature of viral clearance by AEX with regard to both products and process parameters, we focused on constructing a risk assessment based on the ranges of variables from the AEX processes for many different mAb products. These processes are often optimized on a product-specific basis for yield and impurity removal, and many parameters can vary significantly between products. This strategy led to increased parameter ranges and consideration of many categorical parameters, such as buffer and salt compositions, which are generally locked for a specific process but are varied between different processes. We used a variation of an FMEA risk assessment tool, taking into account the severity that variability in a parameter may have on viral clearance and the probability that the parameter would be variable. The effects of each factor were first analyzed using previous data, which indicated their effects on viral clearance. For instance, our laboratory has previously determined that feedstock conductivity, pH, and salt composition can have effects on LRV values, and they interact with one another,15, 16 whereas variables such as the buffer compositions and pooling criteria were not observed to impact viral clearance. In addition, while considering the effects of certain parameters, we also took into account the mechanism of viral clearance by this process. The binding interactions that allow removal of viruses from mAb products are primarily due to electrostatic forces.21 This knowledge emphasizes the impact that variability in factors, which are related to electrostatic forces, such as conductivity and pH, may have on viral clearance. In addition, the mechanism suggests that impurities that compete for the electrostatic binding sites, such as anion host cell proteins, may also impact viral clearance. This observation increases the impact of parameters such as impurity level, as well as load density since it relates directly to total impurity load. For each parameter, we also gauged the probability that it would vary between different processes by comparing the many AEX processes previously developed in-house. The results of our risk assessment indicate that feedstock conductivity and load density are the parameters, which are most likely to impact viral clearance by AEX, while feedstock pH, impurity levels, and buffer conductivity also may have some impact (Figure 1). This risk assessment provides a useful tool for comparing the viral clearance ability of different processes or predicting the effects of process changes. Furthermore, the strategy presented for the formation of a risk assessment provides the opportunity to apply previous knowledge and experience with a process to guide the development of design space experiments.
Fractionated design space for RVLP clearance
Another way to reduce the overall number of runs necessary to explore a design space with many parameters is through the use of DOE experiments. DOE techniques ensure that the efficiency of an experiment matches the desired outcome, maximizing reliable information gained while minimizing redundancy between runs. For initial design space experiments, two-level screening designs are well-suited to give information on which parameters have main effects on the desired outcome, and often give information as to the interactions of those variables. For a given number of runs, DOE approaches range from highly fractionated designs to nonfractionated full factorial designs. Highly fractionated designs can test many different parameters, providing information on main effects of most parameters but only providing interactions between a few. Because many parameters can be tested with these designs, they rely less on the risk assessment to choose which parameters should be tested. However, because many parameters are adjusted for each run, the experiments can be logistically difficult, and since the results contain more information than the familiar one-factor-at-a-time experiments, some researchers and regulatory personal may require the assistance of a statistician to help in analysis and interpretation of the data, particularly when using designs with complex aliasing structures. Full factorial designs, on the other hand, provide both main effects and high-level interactions for all parameters. For a given number of runs, fewer parameters can be tested with these designs, so although they rely heavily on the risk assessment in choosing the most important factors, they are logistically easier to carry out. In addition, since all combinations of all the included parameters are tested, it is easy to grasp the full meaning of the resulting data. Our experimental strategy included an iterative approach, including a highly fractionated design to test many of the AEX parameters, re-evaluation of the risk assessment based on the outcome of that data, and then a full factorial design to test the most significant parameters.
We first investigated the effects of many AEX variables using a highly fractionated two-level screening design. To minimize the cost associated with this preliminary experiment, we chose to test removal of in-process RVLPs using a protein A pool as a feedstock. This methodology allows for the determination of LRV values by the AEX process without the need for expensive virus spiking, although the maximum LRV values which can be obtained are limited by the starting RVLP titers. We used a fractional factorial design to test nine different parameters over large ranges. The design consisted of 16 runs plus four center point runs, and the runs were grouped to form a split plot design with similar feedstocks and column temperature runs performed in pairs, allowing two runs per day. In addition, we collected multiple flow-through pools for each run allowing for LRV values to be calculated for different load densities. This design strategy allowed for 10 of the 18 AEX variables to be varied (Table 2), while two others, feedstock and buffer salt compositions, were held at worse case conditions for viral clearance.16 We observed that virus titers in all of the product pools were below the limit of quantification (LOQ) for the QPCR assay, indicating that all 20 runs achieved complete or near-complete removal of RVLPs for all load densities tested. LRV values for the runs, which were calculated from the input titers and QPCR LOQ values (Table 3), varied from ≥2.9 to ≥3.7, with variations primarily stemming from the QPCR assays of the feedstocks. Since complete removal was observed for all of the runs, the data cannot be used to determine the effects of the varied chromatography parameters. Re-evaluation of the risk assessment based on these data did not result in changes to the previous order of parameters. However, the data does indicate that variation of even the most important AEX variables does not have a measurable impact on its ability to remove RVLPs. Furthermore, the robust removal of RVLPs over these broad ranges of AEX parameters establishes a minimal design space where consistent RVLP removal can be achieved by this process.
Full factorial design space for model virus clearance
The design space from the first experiment provides strong evidence that viruses can be removed over wide ranges of many variables and that none of the variables appear to completely disrupt viral clearance within those ranges. However, the limited LRV values obtained by that methodology are not optimal to determine the full potential of the process to remove viruses or the effects that variables may have on higher levels of clearance, and the parameter ranges tested are wider than those found in actual manufacturing processes. We therefore performed a second experiment using a full factorial design to determine removal of spiked model viruses over parameter ranges more representative of those used in actual manufacturing settings (Table 4). The model viruses used for this experiment, X-MuLV, SV40, and MMV, represent a broad range of virus classes including both enveloped and nonenveloped, and RNA and DNA containing viruses. To limit the number of runs while maximizing the number of variables tested, some parameters were varied together. For instance, the conductivity of the feedstock and equilibration buffer were matched for each run, as were the pH of the feedstock and equilibration buffer, a methodology previously used to test these parameters.15 In addition, bed heights and flow rates were combined into contact times, which were then included experimentally as variation in flow rate. In this way, seven different AEX parameters were combined into four factors in the full factorial design, which, with the addition of two center point runs, were studied in an 18-run experiment. For all of those runs, we observed complete or near-complete removal of all three model viruses to levels below the LOQ's for their respective assays (Table 5). Calculation of LRV's using the input titers and assay LOQ's resulted in LRV's of ≥5.6 to 6.1 for X-MuLV, 4.9 to ≥5.9 for SV40, and ≥5.3 to ≥6.0 for MMV, with variations again due to the QPCR assay. These data indicate that this AEX process is highly robust over the design space and that there are no practically significant effects or interactions of any of the parameters tested over relevant ranges.
Table 5. LRV Values from Full Factorial Design Space of Model Virus Removal
We have developed strategies to create design spaces for the removal of viruses during downstream purification of mammalian-cell derived mAb products. Through the systematic development of a risk assessment and two carefully designed experiments, we have shown that the AEX process can remove the endogenous RVLP impurities over ranges of parameters much wider than those typically found in production processes, and we have provided ranges of process variables where AEX can also be assured of providing complete or near-complete removal of potential adventitious virus contaminants.