Emerging manufacturing modalities
Newer manufacturing technologies (e.g. continuous manufacturing, gene therapy vectors) require special consideration when designing processes and scale-down viral clearance studies for viral filtration.
For very small gene therapy vectors, like AAV, it is conceivable to use large virus retentive filters in downstream manufacturing as a risk mitigation barrier for larger contaminating viruses. AAV is predicted to pass through these filters, while larger viruses like RVLPs are entrapped. While there are some studies assessing the use of viral filters in this field (Adams, Bak et al. 2020, Barone, Wiebe et al. 2020), this concept is being adopted on a case-by-case basis in this therapy class.
Inherent to continuous manufacturing is the seamless transition from one unit operation to the next, which presents a challenge for the viral filtration unit operation. Besides the unique spiking strategies, which are discussed below, integrated unit operations also have to consider the impacts of longer processing times, continuous but potentially low flow rates, and the higher level of impurities and product titer that will be experienced in an end-to-end processing unit operation linkage. One potential strategy to avoid the potential issues of clogging and filter overload is to implement a parallel switch-in and switch-out filtration scheme before the filters reach a validated total volumetric throughput. Monitoring of cumulative flow volume, in addition to traditional filtration pressure monitoring, would allow for control of switching from the first filter to the second once the established filter capacity is reached. Several groups have conducted proof-of-concept studies for such studies and showed equitable clearance values to traditional batch viral filtration (Patt, Dong et al. 2015, Swalec, Feng et al. 2015, Johnson, Brown et al. 2017).
The concept of “filter trains” for continuous viral filtration have also been posited as a potential strategy to allow for increased impurity clearance and improve overall viral filter lifetimes in the extended processing times (Arnold, Lee et al. 2019). Several filters are placed in series with no line or pressure breaks for clarification by depth filtration, TFF for product concentration, a charged membrane filter as a low pressure option for further impurity removal, and finally viral filtration for primary viral clearance. The train allows for reduced facility footprint as well as potentially faster processing times, but discussions on how to validate orthogonal viral clearance steps within these trains is required.
As discussed above, the development of the scale down models for integrated continuous viral filtration may pose a few challenges compared to batch viral filtration. These challenges include 1) performing virus filtration under constant flow, 2) extended volumetric throughputs and extended processing times, and 3) the potential for a dynamic product fluid stream potentially due to significant fluctuations in protein and buffer concentrations (Lute, Kozaili et al. 2020). While the use of constant flow may not be a difficult challenge, as this mode of operation can also occur in batch mode, proper modeling of the pump behavior may be required to avoid pulsations at small scale. A bigger challenge is how to perform virus spiking for the extended processing volumes and processing times. The traditional approach to virus spiking (bolus spike) may be prohibitive because the increased throughputs would require a higher virus spike concentration or a larger volume of virus. This could also lead to overloading of the filter with virus, which was previously discussed as a known concern for some virus filters (Lute, Riordan et al. 2008). Conversely, a low titer virus spike may be implemented to avoid overloading; however, this may lower the clearance values that can be achieved due to assay sensitivity and lowered linear range. Another challenge for virus spiking is maintaining the infectivity of the virus over the extended processing time. A bolus spike may experience a significant loss of infectivity over the course of the filtration study (David, Maiser et al. 2019, Lute, Kozaili et al. 2020). Alternative spiking strategies, have been proposed (Johnson and Roush 2018, Lute, Kozaili et al. 2020). Proposed bracketed spiking strategies involve spiking a high virus load in a small volume at the beginning and end of the filtration study, with either no virus or low titer virus spike for the majority of the process volume. This approach was able to achieve similar total virus loads and clearance values, for up to 2300 L/m2, while avoiding potential virus overloading and loss of infectivity by only spiking virus at the beginning and end of the filtration study (Johnson and Roush 2018).
Another approach to avoiding a loss of infectivity is to have a fresh daily spike of virus throughout the filtration study. In this approach, a fresh spike can be applied to the filter every 24 hours for the duration of the experiment (Lute, Kozaili et al. 2020). To avoid overloading of the filter with virus, the total virus spike should be determined prior to the study and back calculated for a reasonable load titer per day. Care should be taken to seamlessly integrate the fresh load and avoid the introduction of air bubbles or pressure fluctuations that may negatively impact filtration study. This approach was able to demonstrate an LRV of >6 log10 for up to 4 days and 2900 L/m2(Lute, Kozaili et al. 2020).
One final spiking approach mixed the virus stock with the feed solution to desired viral titer, then prefiltered the final feed solution through a sterile 0.1µm filter to mimic the practice of prefilter usage for aggregates. The feed was then pumped at a continuous rate through the viral filter(s) with a constant flux of 0.3L/m2/hour for up to 72 hours. Load samples were taken twice a day (log10 Day) and overnight without fresh spike being added to the feed solution. To account for potential loss of viral titer, the study calculated LRVs as equal to the log10of the daily load titer minus the log10 daily pool sample titer (David, Maiser et al. 2019).
These studies have provided frameworks for spiking strategies and provide strong supportive evidence that valid models for continuous viral filtration exist. Though the extended processing times in these scale-down models do not appear to affect viral reduction, other aspects like filter leachables need consideration.