Development of small‐scale models to understand the impact of continuous downstream bioprocessing on integrated virus filtration

Abstract We designed small‐scale virus filtration models to investigate the impact of the extended process times and dynamic product streams present in continuous manufacturing. Our data show that the Planova 20N and BioEX virus filters are capable of effectively removing bacteriophage PP7 (>4 log) when run continuously for up to 4 days. Additionally, both Planova 20N and BioEX filters were able to successfully process a mock elution peak of increased protein, salt, and bacteriophage concentrations with only an increase in filtration pressure observed during the higher protein concentration peak. These experiments demonstrated that small‐scale viral clearance studies can be designed to model a continuous virus filtration step with specific process parameters.


| INTRODUCTION
Biotechnology manufacturing has gradually evolved over the years from traditional batch mode operation to more continuous modes of operation with the implementation of continuous perfusion cell culture. 1,2 Recently, with technical advancements in continuous chromatography and an increased focus on single-use systems, focus has shifted to an integrated upstream and downstream continuous process.
However, due to technical constraints and lack of small-scale models, most theoretical continuous manufacturing designs focus on a hybrid continuous system with one or more dedicated virus removal/inactivation steps remaining in batch mode via traditional hold tanks (e.g., low pH inactivation) or as a dedicated offline step (e.g., virus filtration). To facilitate the design and implementation of a fully continuous downstream process, one must understand the differences and challenges between traditional batch mode purification and integrated continuous purification. In particular, it is necessary to understand how these differences can impact the performance of each unit operation and how to design small-scale studies of each continuous unit operation. Recent studies have focused mostly on implementation of continuous capture chromatography and continuous viral inactivation 3,4 with little research on the integration of virus filtration into continuous processes, aside from theoretical process design strategies. 5 Virus filtration, also called nanofiltration, is an effective process step for removal of small parvovirus-sized or larger virus particles, predominantly through a size-based mechanism. 6 Virus filters are single use and are typically run under constant pressure to a target throughput. This strategy has proven to be robust and effective despite the discovery of a few technical vulnerabilities and failure modes. 7,8 However, adapting a batch mode filtration strategy to continuous processes may prove challenging. Understanding how the unique technological parameters of continuous processing impact the performance of virus filters may lead to the development of both integration strategies and small-scale process models.
To develop a small-scale model for continuous virus filtration, it is necessary to consider two key differences between the batch unit operation and the integrated continuous unit operation. The first key difference is the concept of a discrete input and output versus dynamic input and output. In traditional batch mode purification, the downstream process flow path typically has hold tanks between each process step which allows for (a) load homogeneity, (b) discrete input volumes/concentrations for the subsequent unit operation, (c) control of optimized flow rate and pressure for each unit operation and for the virus filtrations step in particular, and (d) accommodation for filter replacement based on total throughput limits per filter (e.g., <1,000 L/m 2 ) or processing time (e.g., <8 hr). For an integrated continuous purification process, there are no traditional hold tanks and the fluid flow is constant from one unit operation to another, creating a dynamic product stream with fluctuations in protein concentration, pH, and conductivity due to the inherent periodic elution peaks from one or more bind and elute steps. While these fluctuations may be dampened with the use of surge tanks, in instances where no surge tank is implemented, the fluctuating fluid streams have the potential to negatively impact virus filter performance.
Previous studies have shown that some virus filters are susceptible to virus particle passage or reduced throughput with high protein concentrations or high ionic strength buffers. [9][10][11] Additionally, in the unlikely event of a contamination, an elution peak may theoretically contain an increased concentration of virus, which may lead to a reduction in virus removal due to filter membrane overloading. 7,12 The second key difference is the system process parameters such as flow rate and pressure. In batch mode, each unit operation is a discrete process step that can be operated under optimal flow rates, process times, or pressure. In continuous mode, the process steps and flows are linked with the whole system flow typically governed by the flow rate of the continuous capture step.
This difference can result in a virus filter operated for an extended processing time under low flow and/or low-pressure parameters, as well as pressure fluctuations due to periodic elution peaks, which may be a concern for viral safety. 8

| Viruses and assays
Crude bacteriophage PP7 was prepared and plaque assays were performed according to PDA TR-41: Virus Filtration. 6 Load material was freshly spiked with PP7 at a target 10 6 PFU/ml at the start of each day, prefiltered using 0.2 μm Nalgene Rapid Flow PES bottles, and applied to the virus filters daily for 4 days. Aliquots of spiked feed and 12 ml samples of the filtrate collected daily were stored at −80 C until plaque assays were conducted. Proportions of the filtrates from each day were pooled together to create a simulated pool at the end of each filtration. Large volume testing was performed on simulated pool samples. Log 10 reduction values (LRV) were calculated by subtracting the log 10 of the total virus in the filtrate from the log 10 of the virus load applied to the filter.

| Equipment and setup
In order to conduct continuous filtration experiments for up to 4 days, the filtration setup was configured to have two load reservoirs con-

| Filtration conditions
Virus filtration runs were conducted at constant flow starting at around 7 and 28.4 psi for Planova 20N and BioEX filters, respectively.
Filtrations were ended if the pressure reached the manufacturer recommended pressure limits of 14.2 or 49.7 psi, respectively. Prior to filtration, the filter was primed with 10-15 ml of deionized water. For each filter type, commercially available h-IgG was diluted at a relatively low concentration due to low purity of the material in 50 mM acetate and 20 mM sodium chloride buffer at pH 6.0. The diluted h-IgG solution was prepared for each filtration and stored at 4 C for the duration of the run. Aliquots were obtained daily, spiked with PP7, pre-filtered, then loaded into the virus filter at a specific flow rate ( Table 1). The load was switched to a freshly spiked container once daily to ensure consistent PP7 loading throughout the run. The filtrate was collected once daily as well. All runs were performed in duplicate.

| Proteins, buffers, and reagents
The peak mimicking studies had three components of the process fluid (protein concentration, salt concentration, and PP7 concentration) adjusted to create a buffer peak across the filters using two different load materials, A and B. The diluted h-IgG concentrations and buffer concentrations can be seen in Table 2.

| Viruses and assays
Load materials A and B were freshly spiked with PP7 to the target shown in Table 2 at the start of each day, prefiltered using 0.2 μm Nalgene Rapid-Flow PES filter bottles and applied to the virus filters.
The target spike for the high protein and high salt concentration runs was 10 7 PFU/ml for both load materials A and B. For high phage and the triple spike runs, load material A had a PP7 spike target of 10 6 PFU/ml, while load material B had a PP7 spike target of 10 8 PFU/ml to provide a 100-fold increase in virus particles without overloading the filter membrane. 7 Spiked load materials were sampled and held at room temperature while filtrate samples were collected. Both load and filtrate samples were assayed for titer using PP7 plaque assay immediately after each run. Remaining sample volumes were subsequently stored at 4 C for repeated plaque assays, if required.

| Equipment and setup
In order to mimic the properties of an elution peak, the virus filters were connected to the column selection valve of an AKTA Avant 25 (GE Healthcare Life Sciences), a baseline load was applied using pump A, and a step gradient of the simulated elution peak was applied using buffer pump B (Figure 2). Samples were manually collected, as shown in Figure 3, for each filtration experiment. The AKTA Avant was programmed with the Unicorn 6.1 software to run the following sequence for each filter experiment: The AKTA fraction collector was not used to minimize potential contamination and to minimize backpressure from AKTA tubing. All experiments were performed in duplicate, with select conditions having a third run with an in-line pressure sensor to verify the pressure readings from the AKTA and to determine the impact of the in-line pressure sensor on filtration. Additionally, each experiment had one representative run without sampling whereby the virus filter was integrated into the AKTA Avant system with the permeate flow connected to the AKTA UV and conductivity sensors to allow collection of data and assure that a mimicked elution peak was achieved.

| Extended processing studies with Planova 20N filters
The goal of these studies was to use Planova 20N filters in a continuous virus filtration setup. Planova 20N filtrations with PP7-spiked h-IgG solutions were carried out for 4 days. A starting pressure of 7 psi was chosen due to previously determined low-pressure limit. 14

| Extended processing studies with Planova BioEX filters
Planova BioEX filtrations were performed using the conditions previously described and shown in Table 1   were determined to potentially spike in concentration in an elution peak and impact filter performance were protein concentration, conductivity and virus titer. As shown in Table 2, these factors were tested as single independent variables and also combined as one total In addition to monitoring pressure for filter performance evaluation, filtrate samples were manually collected throughout each experiment as shown in Figure 3 and assayed for PP7 titer by plaque assay.

| Peak mimicking filtrations for Planova 20N and BioEX filters
As virus carryover is a concern in AKTA systems, the buffer flush sample served as a negative control and an installation check. Any filtration run with PP7 detected in the buffer flush sample was discarded, the system was re-sanitized, and the run was repeated with a new filter and new buffers. The pre-spike sample of 20 ml provided a baseline load of protein and PP7 to the filter prior to the spike challenge. The 10 ml of spike load was fractioned into 5 × 2 ml samples to track the potential PP7 passage as the mock peak passed through the filter. The final 20 ml of baseline load were collected in 2 × 10 ml fractions, Post-Spike 1 and Post-Spike 2 (Figures 6 and 7). It should be noted that Post Spike 1 samples were designed to collect residual spike load and contain a ratio of the Load A and Load B fluid stream due to inherent hold-up volume of the AKTA system and the filter.  studies would be to find a balance between the spike level as to not overload the virus filter and the viral clearance to be achieved. What these studies demonstrate is that viral clearance is robust and achievable despite the effects of subtle differences in feedstock on virus filter capacity. These effects will need to be further understood especially in light of the potential for perfusion processes going over multiple weeks.