Purification and concentration of model viruses using single‐pass tangential flow filtration

Abstract

The vaccine and viral vector industry is growing at an accelerated rate. To improve harvest and purification processes, the development of continuous membrane‐based operations, such as normal flow filtration (NFF) and single pass tangential flow filtration (SPTFF) for concentration were explored. This work was conducted using two model viruses, non‐enveloped porcine parvovirus (PPV) and enveloped Suid herpesvirus (SuHV). The viruses are in the same family as the gene therapy vectors adeno associated virus and herpes simplex virus, respectively. SPTFF design started with batch TFF for membrane selection. Hollow fiber membranes with a 100 and 300 kDa molecular weight cut off were defined for PPV and SuHV SPTFF operations, respectively. The SPTFF runs for PPV did not provide any concentration of the virus and low protein and DNA removal, unlike batch TFF. Two hollow fiber membranes run at 10 mL/min and 2 psi were the best condition for SuHV concentration, with approximately 100‐fold titer concentration and protein and DNA removal of 37% ± 5% and 32% ± 8%, respectively. This concentration was superior to the batch TFF and indicated a strong dependence on flow rate and transmembrane pressure. For NFF, filters selection and performance tests were carried out for NFF of PPV and SuHV, as well as cleaning protocols for hollow fiber membranes. The ultimate goal is to integrate this work into the continuous purification of viral vectors produced in mammalian cell cultures to reduce costs and increase throughput.

Single‐pass TFF was developed for an enveloped and non‐enveloped virus for use in continuous downstream purification processes. Enveloped virus was concentrated and purified.

1. INTRODUCTION

Vaccine manufacturing is an important segment for the biotechnological industry. The vaccine market size reported a value of 73 B USD in 2025, with a projected growth to 116 M USD between 2025 and 2030. ¹ The World Health Organization (WHO) has called on governments, industries, and international organizations to commit to the adoption of several actions to guarantee a strong market and future preparedness against potential outbreaks of vaccine preventable diseases. ² This includes the fast development of vaccine technologies and regulatory‐approved manufacturing processes. There are many forms that vaccines can take; examples include protein or carbohydrate subunit and mRNA; our focus is on viral vectors that include the influenza vaccine and some COVID vaccines. While the upstream process, where the biological molecule is produced, is often a focus for development, in vaccine manufacturing, up to 70% of the production cost is in the downstream purification ³ , ⁴ and can be a bottleneck. This bottleneck is mostly due to low recoveries, often because of inefficiencies in the chromatography for large biomolecules.

Downstream purification often relies on membrane operations. Early in the downstream process, the clarification stage is to remove large, undesirable impurities, such as whole cells, cell debris, colloids, and/or aggregates. It can also remove small impurities, such as proteins and nucleic acids. ⁵ Membrane‐based operations have the advantages of high scalability, relatively low cost, and low energy consumption. ⁵ Based on the operation mode, filtration can be classified into two types, normal flow filtration (NFF) and tangential flow filtration (TFF). NFF, also known as dead‐end filtration, occurs when the flow is directed perpendicular to the membrane. Anything removed from the liquid by the filter forms a cake layer. In contrast, TFF, also known as cross‐flow filtration, involves directing the flow across or parallel to the membrane. ⁶ , ⁷ In viral product purification, TFF has been employed mainly for concentration and diafiltration, although both modes have been used in virus clarification. ⁵ , ⁸ , ⁹

The field of virus concentration by TFF is still developing. Although several viruses have been concentrated with TFF, ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ there remains an opportunity for improvement. Most research has been performed in batch mode, where the retentate is repeatedly recirculated to a feed tank until the target concentration is reached. ⁹ , ¹⁹ A major concern in TFF is maintaining the biological activity and structural integrity of the virus, which may be affected by high shear stress or membrane fouling caused during recirculation ²⁰ and is directly influenced by operation conditions. ²¹ Enveloped viruses are especially susceptible to more damage compared to non‐enveloped viruses. ¹⁵

There is not a clear methodology to selecting membranes for virus purification operations. Even selecting the membrane pore size can be fraught. For example, small non‐enveloped viruses, like Aedes aegypti densonucleosis virus (AeDNV) (18–26 nm diameter), achieved effective concentration (10‐fold) with 30 and 50 kDa membranes. ¹¹ A 500 kDa membrane was used to recover adeno‐associated virus (AAV) serotype 1 (25 nm diameter), a similarly sized virus to AeDNV, with 98% host cell protein clarification. However, recovery was only 54.8% for AAV 1. ¹⁸ For larger enveloped viruses (100–200 nm diameter), molecular weight cut‐offs (MWCOs) up to 100 kDa have been successfully used. ¹⁰ , ¹⁵ , ¹⁷ , ¹⁹ Consequently, there is still a need to explore TFF operational design for viral clarification and concentration.

One key problem that continues to plague virus purification with TFF is viral particle inactivation. The shear stress caused by high recirculation rates used in TFF can inactivate virus. One solution is to use single‐pass tangential flow filtration (SPTFF). SPTFF uses a long flow path, low flow rates, and low transmembrane pressures (TMP) with a single pass of solution across a large area membrane. ²² , ²³ , ²⁴ Until now, SPTFF has been applied to the concentration of juice, blood cells, and DNA/RNA, with a major focus on proteins and antibodies. ²⁵ , ²⁶ , ²⁷ SPTFF use for antibody clarification has demonstrated lower shear stress and aggregation, faster processing times, higher yields, and simplicity, ²³ , ²⁸ , ²⁹ however, less exploration has been conducted for SPTFF with viruses. Conditions were identified that could be used in viral vectors SPTFF of long‐duration operation by modeling the system with 100 nm nanoparticles. ³⁰ The results were verified with almost the complete retention of AAV with a 300 kDa regenerated cellulose membrane. ²⁸ The implementation of SPTFF in viral vector manufacturing might provide increased stability during manufacturing and enable integration and intensification with other unit operations for the establishment of continuous bioprocessing. ²⁶ , ³¹ , ³² Continuous manufacturing processes have demonstrated a reduction in cost and facility size, multi‐product flexibility, while having improved quality, productivity, and control. ³³ , ³⁴ , ³⁵

One method to improve the longevity of continuous SPTFF is to use a prefiltration step in NFF mode. NFF can use either depth or regular filters. ⁸ Depth filters often retain larger particles, such as cells or cellular debris. ⁸ , ³⁶ However, there is a risk that the virus product will also adhere to the filter. Polymeric membranes are often preferred for viral vector purification because there is less virus adsorption. Filters with a 0.45 μm pore size are often chosen such that a variety of viral vectors can pass through, such as poliovirus (30 nm diameter) and rabia (150 nm diameter) virus. ³⁷ More work is needed to understand the effect of membrane MWCO and composition on filtration operations that desire the virus to flow through the membrane.

Here, we present the development of a SPTFF operation for the concentration of viruses produced in mammalian culture, using a model non‐enveloped virus (porcine parvovirus, PPV) and a model enveloped virus (Suid herpesvirus, SuHV). The viruses are models for the gene therapy vectors adeno associated virus (AAV) and herpes simplex virus (HSV), respectively. PPV is 18–26 nm in diameter and SuHV is 110–180 nm, giving them unique membrane selection needs. Pore‐size selection was carried out in the traditional TFF design, followed by characterization of virus recovery with different feed flows and transmembrane pressure (TMP) in SPTFF mode using different membrane surface areas. In addition, clarification prior to TFF was explored for a NFF and a preferred filter identified. This process development provides a framework for continued development of continuous viral vector downstream operations.

2. MATERIALS AND METHODS

2.1. Materials

All the components for culture, Eagle's minimum essential medium (EMEM), sodium bicarbonate, fetal bovine serum (FBS), 0.25% trypsin–EDTA, and penicillin–streptomycin (pen/strep) (10,000 U/mL) were purchased from Life Technologies (Grand River, NY). The reagents for the cell viability assay were 2‐(3,5‐diphenyltetrazol‐2‐ium‐2‐yl)‐4,5‐dimethyl‐1,3thiazole bromide (MTT, 98%) from Fisher Scientific (Waltham, MA), sodium dodecyl sulfate (Bioreagent, ≥98.5%) from Sigma Aldrich (St. Louis, MO) and hydrochloric acid, certified ACS Plus, 36.5 to 38.0% from Fisher Chemical™ (Ottawa, Ontario, Canada). In addition, contaminant concentrations were determined with the Coomassie Plus™ Protein Assay Reagent from Thermo Fisher Scientific (Waltham, MA) and the Quant‐iT™ Picogreen™ dsDNA Assay Kit from Invitrogen (Waltham, MA). Phosphate saline buffer (PBS) pH 7.2 was purchased from Gibco™ (Grand Island, NY). Bovine serum albumin standard (BSA) was acquired from Thermo Fisher Scientific.

The cleaning solution for membranes sanitization used sodium hydroxide (NaOH, 99%) purchased from Sigma Aldrich and sodium hypochlorite commercial solution purchased from Thermo Scientific (Waltham, MA).

All solutions were prepared in deionized water (>18 MΩ) from a Milli Q system from Thermo Scientific (Waltham, MA).

TFF experiments were performed in a KrosFlo® KR2i TFF System (Repligen Corporation, Waltham, MA, USA), equipped with a peristaltic pump, size 16 Tygon tubing, an Automatic Backpressure Valve, vessels and scales for retentate and permeate. The data were recorded at 5 s intervals using the KF Comm 2 V2.1 Workbook. The membranes used were modified polyethersulfone (m‐PES) hollow fiber membranes in MicroKros® modules (20 cm length and 0.5 mm fiber inner diameter) with different MWCOs (50, 100, 300, 500 kDa), obtained from Repligen Corporation.

In NFF tests, an IPS‐12 series syringe pump (Cambridge, MA) and a variety of syringe filters were used. Supor® 0.1 μm, Supor® 0.2 μm, Supor® 0.45 μm, Supor® EKV 0.2 μm, Supor® 0.8 μm, Supor® 0.8/0.2 μm (Pall Corporation®) and Puradisc® 1.0 μm (Whatman®) were acquired from Cytiva (Marlborough, MA) while the Nalgene® 0.8 μm CA was obtained from Thermo Scientific (Waltham, MA).

2.2. Methods

2.2.1. Cell culture, viruses, and virus titration

Porcine kidney (PK‐13) cells (ATCC, CRL‐6489) and Vero cells (ATCC, CCL‐81) were acquired from ATCC (American Type Culture Collection, Manassas, VA). Porcine parvovirus (PPV) strain NADL‐2 was a generous gift from Dr. Ruben Carbonell (North Carolina State University, Raleigh, NC). Suid herpesvirus 1 strain Auzjeszky (SuHV‐1, ATCC VR‐135) was purchased from ATCC.

Stocks of virus samples (PPV and SuHV) were generated by propagation of viruses in corresponding cell lines (PK‐13 cells for PPV and Vero cells for SuHV), as stated previously ³⁸ , ³⁹ with EMEM‐supplemented media containing 10% v/v FBS and 1% v/v pen/strep. Cells were infected with virus stocks at 3 log₁₀ (MTT₅₀/mL) for PPV and at 4 log₁₀ (MTT₅₀/mL) for SuHV and incubated at 37°C for six (PPV) and five (SuHV) days. The flasks were subjected to two freeze–thaw cycles at −20°C (room temperature), scrapped with a cell scrapper, and then the supernatant was centrifuged at 5500 × g using a Sorvall ST16R centrifuge with a TX‐400 swing‐bucket rotor (Thermo Scientific, Waltham, MA), for 15 min at 4°C. Glycerol was added to the SuHV stock at a final concentration of 10% v/v. ³⁸ , ³⁹ Virus stocks were stored at −80°C until use.

The virus titer was determined by the colorimetric MTT cell viability assay as stated previously. ³⁸ , ³⁹ Briefly, PK‐13 and Vero cells were seeded at a density of 8 × 10⁴ and 1 × 10⁵ cells/mL, respectively, in 96‐well plates with 100 μL/well for 24 h. The plates were infected with virus samples in quadruplicates with serial dilutions at a 1:5 ratio across the plate. The plates were incubated for 6 days and 5 days respectively, at 37°C, 5% CO₂ and 100% humidity. Afterwards, 10 μL/well of MTT reagent was added, and 4 h later, 100 μL/well of 0.01 M HCl with 10% SDS was added and incubated overnight. The absorbance was measured at 550 nm on a Synergy Mx microplate reader from BioTek (Winooski, VT). The value of 50% cell death was calculated and expressed in the units of log₁₀ (MTT₅₀/mL). ³⁸

Density and viscosity were determined in triplicate for freshly thawed, clarified PPV and SuHV samples. Density was determined by weighing 500 μL of sample with a calibrated 1 mL micropipette on an Mettler Toledo MS204S analytical balance (Columbus, OH). The viscosity measurements were performed using a stress‐controlled, oscillatory rheometer (HAAKE MARS 60, Thermo Fisher Scientific, Waltham, MA) equipped with a stainless‐steel parallel plate geometry (25 mm diameter). 0.5 mL of sample was pipetted on to the sample holder and the measurement gap was set to 1 mm. A pre‐shear at 100 s⁻¹ was applied for 30 s, followed by a 5‐min rest period. Subsequently, viscosity was recorded using a shear rate (υ) sweep experiment with υ ranging from 10 to 1000 s⁻¹.

Spent cellular media used for preliminary TFF runs was obtained by taking media from cell cultures of PK‐13 and Vero cells that were not exposed to the virus. The spent media was centrifuged at 300 × g for 5 min using the Sorvall ST16R centrifuge.

2.2.2. Protein and DNA quantification

The Bradford assay measured protein concentration and was used according to the manufacturer instructions. ⁴⁰ Each sample was assayed in triplicate and samples were diluted with PBS buffer, as needed. The results were obtained by measuring the absorbance at 595 nm on a Synergy Mx microplate reader.

DNA quantification was done by the fluorescent Quant‐iT™ Picogreen™ dsDNA Assay Kit following manufacturer's instructions. ⁴¹ The excitation and emission wavelengths were 480 and 520 nm, respectively, and the fluorescence was measured with a Synergy Mx microplate reader.

2.2.3. SPTFF development

The general procedure for the development of the SPTFF is depicted in Figure 1. First, the TFF was flushed with water, and a normalized water permeability (NWP) test ⁴² was conducted at 10 mL/min, measuring the flux as a function of TMP for each new hollow fiber membrane. Some of the membranes used in this project were re‐used after sanitization. The sanitization procedure was integrated into three steps: a deionized water rinse (200 mL), cleaning with 200 mL of washing solution (0.5 N NaOH + 150 ppm NaOCl), and a final deionized water rinse (1200 mL). The solutions were pumped at 12 mL/min (0.2 cm³/s), a transmembrane pressure (TMP) of 2 psi, and in the case of the sanitization solution, it was in contact with the membrane for 1 h using the configuration of Figure 2b. For the cleaning step, three different protocols were evaluated with a 100 kDa hollow fiber membrane, as shown in Table S1. The volume of water in the final water rinse was defined by monitoring the pH of the permeated effluent until it reached neutrality. Based on the results of the evaluated protocols for a single hollow fiber membrane, a modified protocol was implemented for the sanitization of the two membranes in series, using 300 mL of water, 300 mL of washing solution, and 1600 mL of water. NWP tests following cleaning were used to ensure a membrane recovered at least 90% of the initial flux to be considered for reuse.

FIGURE 1.

Process development for SPTFF operation. Pore size and operating condition selection were carried out in batch mode. SPTFF was operated in continuous mode.

FIGURE 2.

Process diagrams for (a) traditional, recirculating TFF and (b) SPTFF.

In preliminary tests for membrane selection, three MWCOs were tested for PPV (50, 100, and 300 kDa) and three for SuHV (100, 300, and 500 kDa) in a traditional TFF configuration with recirculation to the feeding vessel (Figure 2a) using the KrosFlo® KR2i TFF. The feeding vessel contained 120 mL of virus sample. The operation began with 10 mL/min feed flow and a retentate pressure (TMP) of 3.5 psi. The filtration was stopped when a volumetric concentration factor (VCF) of 3 was reached.

Preliminary runs for each model virus in TFF (Figure 2a) were performed with spent cellular culture media at four different feed flows (8, 10, 12, and 15 mL/min) and six P_Rs (0, 1, 2, 3, 5, 7, 9, 12 psi). For each flow rate, the lowest TMP was at the beginning of the run, and it was increased stepwise after 3 min, allowing relative equilibrium to be achieved at each pressure before the change in TMP. Once conditions were determined, in the third test, longer runs (with different sample volumes) were performed in SPTFF (Figure 2b) at a constant flow rate and a starting P_R. Runs were completed with either one membrane or two membranes in series. Immediately after each run, samples were kept at −20°C until analysis.

2.2.4. Filtration using NFF

Pre‐filtration prior to SPTFF was explored using dead‐end filtration. NFF was carried out by pumping 3 mL of virus stock at 1 mL/min using the IPS‐12 series syringe pump, and different syringe filters were tested: Supor® 0.1 μm, Supor® 0.2 μm, Supor® 0.45 μm, Supor® EKV 0.2 μm, Supor® 0.8 μm, Supor® 0.8/0.2 μm, Nalgene® 0.8 μm CA, and Puradisc® 1.0 μm. All syringe filters had a 25 mm diameter, and these were previously flushed with 3 mL of sterile water. The filtrate solution was collected in sterile cryovials and stored at −20°C until further analysis.

Selected syringe filters (Supor® 0.45 μm for PPV and Puradisc® 1.0 μm for SuHV) had a longer filtration run using spent culture media at the same flow velocity of 1 mL/min. The filtrate volume was collected and measured over time in duplicate.

2.2.5. Statistical analyses

All statistical analyses were performed using Minitab software version 22.2.1. One‐way ANOVA followed by Tukey's multiple comparisons test was conducted for virus titer and protein removal data in pre‐filtration experiments for each virus. A p‐value of <0.05 was used to indicate statistical significance in results.

3. RESULTS AND DISCUSSION

There were two goals in this project, the development of a continuous concentration step using SPTFF and a pre‐filtration step using NFF for viral vectors. These steps were designed using two different model viruses. Porcine parvovirus (PPV) is a non‐enveloped, single‐stranded DNA virus, with an approximate diameter of 20 nm. ⁴³ This virus is often used for viral clearance studies as a small, non‐enveloped virus and is also in the Parvoviridae family with adeno‐associated virus (AAV). The second virus was the enveloped, double‐stranded DNA Suid herpesvirus (SuHV), which belongs to the Herpesviridae family, ⁴⁴ and has a reported diameter ranging from 110 to 180 nm. ⁴⁵ SuHV is also used for virus clearance studies and is a safer version of the enveloped human herpes virus.

3.1. Membrane selection in traditional TFF operation

Membrane MWCO selection was performed with batch experiments in standard recirculating TFF mode shown in Figure 2a. The recovery of the virus infectious titer and the removal of host cell protein and host cell DNA were used as metrics to determine the desired MWCO. The system was run at a constant 10 mL/min feed flow rate and stopped once a VCF of 3 was reached. Results are shown in Figure 3. Complete numerical data for virus titers and contaminants (protein and DNA) can be found in Tables S2 and S3. All of the MWCOs tested were able to recover all of the PPV in the retentate, as shown in Figure 3a. The SuHV input was recovered for all MWCOs tested (Figure 3b) within the error of the MTT assay of 0.5 logs, ⁴⁶ which was used to determine the virus titer. No measurable SuHV was detected in the permeate for the 100 and 500 kDa membranes.

FIGURE 3.

Membrane MWCO selection for PPV (left side in blue) and SuHV (right side in green) samples in traditional TFF for retentate (R) and permeate (P) streams. (a) Virus titer of PPV. (b) Virus titer of SuHV. Dotted line represents the starting titer of the virus. (c) Protein recovery from PPV sample. (d) Protein recovery from SuHV sample. (e) DNA recovery from PPV sample. (f) DNA recovery from SuHV sample. Data based on one filtration run and error bars are the standard error of mean (SEM) of triplicate analysis for titer and protein recovery and duplicate analysis for DNA recovery. * at the limit of detection (1.6 logs).

Since virus recovery was not variable between the MWCOs studied, the removal of protein from the retentate was explored as a factor to determine the MWCO to use for further development. The protein recovery from the PPV in the permeate fractions increased as the membrane MWCO increased (Figure 3c). However, the recovery of the protein in the retentate did not close the mass balance for the 50 and 100 kDa membranes. Thus, there was likely fouling of the membrane, although that was not tested at this time since we were only screening for the best MWCO to continue into further development. For SuHV, the recovery of protein in the retentate did not change with MWCO (Figure 3d), which remained at 29%–37% of the input protein. This demonstrated that for SuHV, TFF could also be considered a purification step rather than just a concentration step.

The removal of DNA (most likely host cell DNA) was also analyzed in the screening process. There was no clear trend in DNA recovered for PPV as a function of pore size, as shown in Figure 3e. Fouling is the most likely reason for low DNA recovery in both the retentate and permeate for the 100 kDa membrane. For SuHV, there were no great differences in DNA recoveries between the MWCOs (Figure 3f), however, there was removal of >50% of the DNA, providing purification as seen by the removal of host cell protein. There were no pre‐treatment steps, like enzyme digestion, prior to this step.

Due to the high removal of host cell protein (59% ± 2%) and DNA (72.% ± 0.1%) in the retentate of the 100 kDa hollow fiber membrane, this MWCO was chosen for subsequent TFF experiments with PPV. It was also a concern that the 300 kDa membrane may be too close to the size of PPV and may show more passage of the virus through the membrane when larger volumes were tested. The selection of MWCO to concentrate non‐enveloped viruses of similar size has been variable, as discussed in the introduction. We also confirmed that all of the feeds had similar density (1.01 ± 0.01 g/mL) and viscosity and that the solutions were Newtonian fluids (see Figure S1), allowing comparisons between virus solutions.

Interestingly, it was found in all the tests that the amount of contaminants in the retentate and permeate did not sum up to the initial contaminant amount. It is likely that fouling of the membrane occurred, but the experiment was not long enough to see an increase in TMP due to fouling. Even though the hydrophilic modified PES membrane is less susceptible to high protein binding, ⁴⁷ retention of protein and DNA from the virus solutions was observed.

3.2. SPTFF

The next step was defining an operational space for each virus sample between two different SPTFF configurations: one single hollow fiber membrane and two hollow fiber membranes in series. The curves of the normal water permeability (NWP) tests performed for new membranes are shown in Figure S2. For one single membrane of 100 and 300 kDa, the increase in permeate flux was linear with respect to the TMP up to 4.6 and 3.9 psi (Figure S2A,B). The NWP values used as reference for comparing membrane performance between runs were 105 LMH/psi and 118 LMH/psi, respectively. For two serial membranes of 100 and 300 kDa, the permeate fluxes were decreased in comparison with one single membrane, and the linearity between permeate fluxes and TMPs was 3.3 and 2.5 psi (Figure S2C,D), using as reference NWP values of 109 LMH/psi and 122 LMH/psi, respectively. While there are many different configurations for SPTFF, we chose a simple linear configuration to increase the membrane area to determine the impact on virus concentration.

For each filter configuration and virus, several feed flow rates were tested to determine the permeate flux at multiple TMPs and related TMPs, as shown in Figure 4. Each curve presented two different flow regimes. At lower TMP, the permeate flux is dependent on the TMP. At higher TMP, the flux became independent of TMP and was likely in a gel layer control region, caused by the accumulation of rejected solutes on the membrane surface. ⁴⁸ , ⁴⁹ Since clarified lysate was used in the experiments, there was a significant amount of host cell protein and DNA to create a gel layer. The gel formation was observed for PPV at TMP greater than 4 psi, with an average permeate flux in this region between 47 and 55 liters per meter squared per hour (LMH), shown in Figure 4a. For SuHV, there was no permeate flux below 1 psi for the single 300 kDa membrane (Figure 4b). However, similar to PPV, the gel formation likely began at 4 psi for all the feed flows tested.

FIGURE 4.

Permeate flux versus TMP at different feed flows in SPTFF for one single hollow fiber membrane (a), (b) and two hollow fiber membranes in series (c), (d). (a) PPV with one single 100 kDa membrane. (b) SuHV with one single 300 kDa membrane. (c) PPV with two serial 100 kDa membranes. (d) SuHV with two 300 kDa membranes in series. (▲) Feed flow at 8 mL/min, (■) Feed flow at 10 mL/min, (●) Feed flow at 12 mL/min, (♦) Feed flow at 15 mL/min. Error is the standard deviation given over a 2 min period.

Because the behavior in the permeate flux versus TMP curves with one membrane was very similar regardless of flow rate, only the feed flow rates of 8 and 10 mL/min were tested for the two serial membranes. A marked decrease in the permeate flux was observed for PPV in the two serial membranes (Figure 4c) with respect to a single membrane. A linear relation of permeate flux with TMP was found from 1 to 4 psi, as observed with a single membrane. At a TMP greater than 4 psi, a stable permeate flux between 39 and 46 LMH was found. As observed, in the linear region, the permeate flux tended to decrease when two serial membranes were used with respect to one membrane. This is expected since the filtration area increased from 20 to 40 cm², and the area and permeate flux are inversely proportional in the linear regime. This follows the basic relationship for a filtration process, which states that the area (A) can be determined by the quotient of the permeate volume (V) by the product of the permeate flux (J) and the process time (t). ⁴⁹ The permeate flux reached up to 53 LMH for SuHV at feed flows of 8 and 10 mL/min (Figure 4d), and with no significant decrease with respect to one membrane in the linear region in comparison with one membrane. Wolf and collaborators reported permeate flux curves for HIV VLPs over TMPs in the range of 3.6 to 18.1 psi at three distinct volumetric flows (9, 13.5 and 18 mL/min) and with four membranes (100, 300, 500 and 750 kDa). They found that for HIV VLPs, the membrane MWCO and TMP did not influence the flux, but flux was only a function of flow rate. ¹⁷ In our system, we were not able to achieve high enough feed flow rates and TMPs to observe this relationship. Also, the feed material was different in each case. This type of optimization has been carried out with proteins, ²⁰ and in particular with antibodies. ⁵⁰ However, this is one of the first explorations with virus solutions.

Two conditions were selected to test each virus using higher volumes with either one or two membranes. For PPV, both the flow rates and TMPs tested for the single membrane showed similar infectious virus titer before and after the experiment in the retentate, as shown in Table 1. For the two membranes in series, the lower flow rate and TMP led to no detectable virus in the permeate, but at a flow rate of 10 mL/min and 2.5 psi TMP, some virus was detected in the permeate. In all cases, the virus was not concentrated, as expected. The inclusion of a higher feed volume (420 mL) with the two membranes aimed to observe if a longer operation time at steady‐state would show a time‐dependent concentration. However, this was not observed. It is possible that either larger volumes need to be tested to observe concentration since virus was likely lost during the build up of the gel layer, or the MWCO was too large for the virus.

TABLE 1.

PPV recovery and purification in SPTFF with a 100 kDa MWCO membrane. Each run was completed once using the volume indicated.

Sample	Volume (mL)	Feed volume reduction (%)	Virus titer (log MTT50/mL)				Protein recovery (%)			DNA recovery (%)
			Feed stock	Retentate	Permeate 1	Permeate 2	Retentate	Permeate 1	Permeate 2	Retentate	Permeate 1	Permeate 2
One single membrane (F = 8 mL/min, PR 3.5 psi)	120	90	7.2 ± 0.2	7.0 ± 0.3	1.6 a	‐	87 ± 2	10.0 ± 0.1	‐	73 ± 11	1.9 ± 0.1	‐
One single membrane (F = 10 mL/min, PR 5.0 psi)	120	85	8.0 ± 0.1	8.2 ± 0.5	1.6 a	‐	85.4 ± 0.6	12 ± 1	‐	55 ± 6	1.3 ± 0.1	‐
Two membranes in series (F = 8 mL/min, PR 2.0 psi)	200	79	7.9 ± 0.1	7.8 ± 0.2	1.6 a	1.6 a	99 ± 4	3.9 ± 0.3	6.0 ± 0.4	97.5 ± 0.1	1.6 ± 0.1	2.3 ± 0.8
Two membranes in series (F = 10 mL/min, PR 2.5 psi)	420	66	7.7 ± 0.1	7.4 ± 0.2	2.7 ± 0.1	3.0 ± 0.1	69 ± 1	6.8 ± 0.1	10.2 ± 0.4	67 ± 1	2.0 ± 0.1	2.3 ± 0.1

Note: Values are shown as the mean ± standard error of a triplicate measurement.

^a Limit of detection (1.6 logs).

The protein recovery in retentate was high for the majority of selected conditions. Only in two in series membranes at 10 mL/min and 2.5 psi, a protein recovery of 69% ± 1% was found. These are the same conditions that also detected PPV in the permeate. Host cell DNA removal was moderate, with a recovery between 55% and 73%. The condition at two in series membranes at 8 mL/min and 2 psi was an outlier with a recovery of 98%. The registered TMP for the four experiments was relatively constant with the target TMP, as shown in Figure S3. The profiles for the permeate flux are depicted in Figure S4. A 100 kDa PES membrane was able to concentrate canine parvovirus 2 (CPV‐2), a non‐enveloped virus with 20–26 nm in diameter, from 7.39 ± 0.53 to 8.7 ± 0.28 log₁₀ TCID₅₀/mL. ¹⁶ The parvovirus was concentrated from 500 mL of harvested DMEM media in traditional, recirculating TFF (as shown in Figure 2a). Based on the results obtained, PPV maintained its titer better in the traditional TFF format (Figure 3a) as compared to SPTFF. This could be due to the hydrophobicity of the virus, ⁵¹ causing the virus adsorption to the membrane. Once the membrane is saturated with the virus, the virus loss is reduced. We did not likely reach this level of saturation of the membrane area for SPTFF, but did for batch TFF.

For SuHV, four conditions were evaluated with one and two membranes. For one membrane at 8 mL/min and a TMP of 3.5 psi, there was no concentration of the SuHV between the retentate and feed stock, as shown in Table 2. However, the titer increased almost one log for one membrane at 10 mL/min and 5 psi, and two logs for both conditions with two serial membranes. Thus, concentration was achieved for SuHV. While the mass balance does not close due to the high error in the infectivity measurement, it still demonstrates that SuHV can be concentrated under these conditions and still remain infectious. No virus was found in any of the permeates. The protein recovery was above 80% for all conditions except for the two membranes in series at 10 mL/min and 2 psi. In this condition, the protein recovery was 63% ± 5%. The DNA recovery was also lower for this condition (68 ± 8%). Therefore, two membranes in series run at 10 mL/min and 2.0 P_R would be the most recommended for the concentration of SuHV. When P_R was increased from 2 to 5 psi, the remaining host cell protein and DNA greatly increased. The increased R_P likely increased the gel‐layer formation and retained more impurities in retentate. The related TMPs and permeate fluxes for SuHV are in Figures S5 and S6, respectively, showing the constant values during the runs.

TABLE 2.

SuHV recovery and purification in SP‐TFF with a 300 kDa MWCO membrane. Each run was completed once using the volume indicated.

Sample	Original volume (mL)	Feed Volume reduction (%)	Virus titer (log MTT50/mL)				Protein recovery (%)			DNA recovery (%)
			Feed stock	Retentate	Permeate 1	Permeate 2	Retentate	Permeate 1	Permeate 2	Retentate	Permeate 1	Permeate 2
One single membrane (F = 8 mL/min, PR 3.5 psi)	120	95	4.9 ± 0.1	5.2 ± 0.2	1.6 a	‐	94 ± 2	7.4 ± 0.3	‐	73 ± 11	1.9 ± 0.1	‐
One single membrane (F = 10 mL/min, PR 5.0 psi)	120	78	3.3 ± 0.1	4.2 ± 0.1	1.6 a	‐	86 ± 1	13.7 ± 0.4	‐	55 ± 6	1.3 ± 0.1	‐
Two membranes in series (F = 10 mL/min, PR 2.0 psi)	400	66	3.4 ± 0.2	5.6 ± 0.1	1.6 a	1.6 a	63 ± 5	10.0 ± 0.1	10.5 ± 0.4	68 ± 8	3.1 ± 0.1	3.0 ± 0.1
Two membranes in series (F = 10 mL/min, PR 5.0 psi)	400	79	2.8 ± 0.1	4.9 ± 0.1	1.6 a	1.6 a	84 ± 1	4.3 ± 0.1	5.6 ± 0.2	81.9 ± 0.9	1.2 ± 0.1	1.6 ± 0.1

Note: Values are shown as the mean ± standard error of a triplicate measurement.

^a Limit of detection (1.6 logs).

TFF processes have been performed on enveloped viruses with equivalent diameters. Conventional TFF of the influenza A virus yielded a 4‐fold increase in titer using a 300 kDa flat sheet membrane at a flow rate of 150 mL/min and an average TMP of 5.8 psi. ¹⁹ This high flow rate was possible because a higher media volume (500–600 mL) was used and the system was in standard recirculation mode. In addition, the initial feed was diluted 1:1 with PBS buffer and the production media was serum‐free. So, the protein content in the influenza feed was lower than in our SuHV feed by approximately 6 times, leading to lower membrane fouling and higher influenza concentration. Lentivirus was concentrated with a benchtop TFF using a 300 kDa flat sheet membrane at a TMP of 14.5 psi, in which 66‐fold concentration was measured with a 90%–100% recovery by measuring the transducing titer units. ¹⁰ In this traditional lentivirus TFF operation, a serum‐free medium and a large volume (2 L) were used. Moreover, there are no details about the feed flow nor any protein or DNA quantification. In summary, a 100‐fold concentration was achieved for SuHV with the recovery measured in infectious virus units, demonstrating that SPTFF can concentrate viral particles and maintain their activity.

Based on our findings, it is evident that a feed flow of 10 mL/min and low P_R provided the highest SuHV retention. Although this is not the typical operational conditions for virus, filtration theory might explain these observations. ⁵² Higher velocities tend to enhance the virus lift force by convection of the membrane surface. ⁵² , ⁵³ The low R_P found best for SuHV might be the critical R_P that avoids the fouling of proteins and virus particles by balancing the deposition of the virus and proteins on the membrane due to perpendicular movement of the fluid across the membrane and the sweeping force of the tangential flow. If tangential and lift forces are magnified with the particle sizes, ⁵² it is not strange that the smaller virus (PPV) had reduced forces and the total force balance was unfavorable for its concentration. Other potential factors, such as the interaction of the virus with the contaminants, may be crucial; however, more research is needed to validate this hypothesis.

In this project, membranes were reused with the purpose of saving on operating costs. ⁵⁴ Since it has been estimated that half of the total time in a TFF process is invested in preparation and cleaning, ⁵⁵ an optimized and standardized protocol for sanitizing was developed. Three different cleaning protocols were assayed for the 100 kDa membrane. The results showed that the membrane cleaned following protocol 3 had the highest recoverability (94.3% ± 0.1%) (Figure S7), while protocols 1 and 2 were less effective. Unlike the efficient cleaning observed in PES cassettes used in an SPTFF antibody process when no multiple solution passages were done, ⁵⁶ protocol 2, with more than 1 passage and change of the flow direction, led to slightly better membrane recoverability. The increment of washing solution temperature (50°C or 323.15 K) in protocol 3 allowed the recovery of the membrane performance, likely due to the efficient cleaning of protein contaminants in the membrane through increased solubility of the foulants, reduced viscosity, and consequently improved mass transport. ⁴⁹ , ⁵⁷ , ⁵⁸ Testing of the same selected protocol 3 for cleaning 300 kDa membranes and two membranes in SPTFF also yielded superior recoverability to 90% (data not shown) of the starting NWPs, so this same protocol was applied as the standard cleaning procedure, only adapting the volumes in SPTFF operation.

3.3. NFF of virus samples and filter performance

It was desired to have a pre‐filtration step to determine if protein and DNA could be removed prior to SPTFF to lower fouling during the SPTFF. Different syringe filters were used to explore this NFF step, and the results are shown in Figure 5. The complete data for titers and protein removal in NFF are shown in Table S4. No significant change was observed in the PPV titer with respect to the initial titer (Figure 5a). Also, there was no difference between the protein removal of the filtrates, ranging from 31.6% to 41.8% (Figure 5b). PPV was expected to fully pass through the filter, which it did. Similarly, minute virus of mice (MVM), a parvovirus with a diameter up to 34 nm, had no losses when a 0.1 μm filter was used, even when the filter was made of polyvinylidene difluoride (PVDF). ⁵⁹ Therefore, the Supor® 0.45 μm was chosen due to its largest pore size and the likelihood to avoid early fouling. This pore‐size choice agrees with the recovery of virus from mammalian cell cultures, such as poliovirus (30 nm diameter) and rabia virus (150 nm diameter). ³⁷

FIGURE 5.

Filtration for virus using syringe filters. (a) Virus titer for PPV. (b) Protein removal for PPV. (c) Virus titer for SuHV. (d) Protein removal for SuHV. * limit of detection (1.6 logs). Data performed in duplicate. Error is the standard error. **p‐value <0.05.

Different results for NFF pre‐filtration were found for the larger virus. SuHV was not detected in the filtrate of the Supor® 0.1 μm and 0.8/0.2 μm filter (Figure 5c), which was not surprising since SuHV has a reported diameter size ranging from 110 to 180 nm. ⁴⁵ For the other filters tested, differences in the final titer were found. The 0.45 μm Supor® filter had a 1 log removal of SuHV. The remaining filters had less than 1 log of removal. Statistical analysis did not reveal a difference in the protein removal for all the filters, and it ranged from 12% to 21% (Figure 5d). Combining SuHV titer and protein removal data, it was decided to use the Puradisc® 1.0 μm for the pre‐filtration of SuHV since the larger the pore size, the longer the filter should run prior to membrane fouling.

An additional experiment for studying the filtration flux of selected filters in a continuous manner was conducted (Figure 6). PPV flux reached a constant filtrate flux of 200–210 LMH, with the inconvenience that one of the replicates failed after 20 min of continuous PPV filtration at a feed flow rate of 1 mL/min (Figure 6a). This situation may be a consequence of the fouling of the membrane and overpressure in the filter. The other run was stopped at 30 min due to sample limitations.

FIGURE 6.

Filtrate flux profiles for the syringe filters (a) Supor® 0.45 μm tested with PPV and (b) Puradisc® 1 μm tested with SuHV.

SuHV was tested on the Puradisc® 1.0 μm filter for continuous operation. SuHV had similar properties, viscosity (Figure S2) and density, to PPV, and the flux was stable for the duration of the 80 min tested. Both duplicates showed an average filtrate flux of 145 LMH at a feed flow rate of 1 mL/min (Figure 6b). It was concluded that pore size is one of the most important properties associated with fouling behavior. ⁶⁰ However, another important difference between the filters is the filtration area: 2.8 cm² for Supor® 0.45 μm and 4.2 cm² for the Puradisc® 1.0 μm; it may suggest that a reduced filtration area might have led to faster membrane rupture in the Supor® 0.45 μm due to the high flux. Notably, there was no decrease in the flux over the tested period for both filter types. Based on the presented results, Supor® 0.45 μm may not be adequate for filtering higher sample volumes to 20 mL due to the observed failure in one of the replicates. In this case, a higher filter area is needed.

For practicality, these studies were done at a flow of 1 mL/min, but different flow velocities may be considered in the future to better characterize the performance of the filters, since flow velocity is a factor affecting the performance and membrane fouling. ⁶⁰

4. CONCLUSIONS

The present research was carried out to establish a future continuous pre‐filtration and SPTFF concentration for viral vaccines and vectors. Model viruses (PPV and SuHV) were used to represent non‐enveloped and enveloped viruses, respectively. With little data on the development of SPTFF for continuous viral vector concentration, we explored some of the key parameters in the development of continuous filtration operations. NFF was explored as a pre‐filtration step to better understand the parameters affecting the filtration of the model viruses.

Preliminary experiments concluded that hollow fiber membranes with a 100 kDa MWCO were best for the concentration of PPV, and 300 kDa for SuHV. These filters conserved the virus titer while providing the highest recovery of contaminants (protein and DNA) in the retentate. No significant concentration of PPV was achieved in the SPTFF conditions tested. This could be due to virus being lost because of adhesion and cake formation on the membrane and could possibly be remedied by using a larger volume. Another reason could be that the pore size is close to the virus size and virus is getting caught in the membrane. Unlike PPV, SuHV titer was improved in three of the SPTFF tests, and the best condition increased the SuHV titer 100‐fold, showing better results than those for similar viruses in the batch TFF configuration.

We also established a cleaning protocol for the re‐usability of hollow fiber membranes, considering as re‐use qualification a recoverability of ≥90% of the initial normal water pressure. This protocol was integrated by three steps: water rising, cleaning with washing solution (0.5 N NaOH and 150 ppm NaOCl) at 50°C in forward‐backward‐forward directions, and a second water rising.

In NFF as a pre‐filtration step, the PPV sample was assayed with different Supor® syringe filters, without a difference in permeate titer and permeate protein removal. For SuHV, larger pore size filters were needed, and high virus recovery in the permeate occurred in most of the filters tested. In continuous flow at 1 mL/min, Supor® 0.45 μm for PPV operated during 20 min before fouling. This pore size was likely too small for continuous operation. For SuHV, Puradisc® 1.0 μm did not see a change in flux during the 80 min test.

The SPTFF concentration experiences presented here might be useful as an orientation for the development and implementation of SPTFF in continuous mode with other similar viruses of biotechnological or clinical importance, for example, picornaviruses or hepatitis A (considering PPV model) and HIV, influenza, or lentiviruses (considering SuHV model). Full modeling of SPTFF operation may help predict a desired recovery, fold concentration, and contaminant removal level robustly, using the studied process conditions. The membrane pore size for PPV may need further optimization, or it is possible that a longer run will show virus concentration after the filter cake is established. Only a hollow fiber membrane was used in this work; flat sheets may provide a different perspective on SPTFF operations for viral vectors. In the case of NFF, these results demonstrate the importance of pore size on long‐running continuous operation and the need to have larger pore size. Or it is possible that a depth filter may improve long‐running operation. Nevertheless, more experiments at different flow velocities and parameter measures might contribute to filtration optimization and its performance. Integration of membrane‐based operations (NFF and TFF) with model viruses presented here, with upstream and subsequent purification stages, will allow the adoption of future efficient vaccine and viral vector continuous bioprocesses.

AUTHOR CONTRIBUTIONS

Lynn Manchester: Investigation. Luis Alberto Mejía‐Manzano: Conceptualization; methodology; data curation; investigation; formal analysis; writing – original draft; writing – review and editing. Kan Wang: Investigation. Caryn L. Heldt: Conceptualization; data curation; supervision; funding acquisition; writing – review and editing. Grace Wallis: Investigation. Bruce P. Lee: Supervision; writing – review and editing. Michael Betenbaugh: Supervision; writing – review and editing; funding acquisition.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Supporting information

Table S1. Proposed cleaning protocols in membranes sanitization.

Table S2. Virus titer data for membrane pore size selection for PPV and SuHV in traditional TFF.

Table S3. Contaminants (protein and DNA) from membrane pore size selection for PPV and SuHV in traditional TFF.

Figure S1. Rheogram for determining dynamic viscosity using a parallel plate oscillatory.

Figure S2. Pre‐run WPTs for SPTFF with one single hollow fiber membrane (A‐B) and two hollow.

Figure S3. TMP profiles for selected conditions with PPV in SPTFF using one single 100 kDa.

Figure S4. Permeate flux profiles for selected conditions with PPV in SP‐TFF using one single.

Figure S5. TMP profiles for selected conditions with SuHV in SP‐TFF using one single 300 kDa.

Figure S6. Permeate flux profiles for selected conditions with SuHV in SP‐TFF using one single.

Figure S7. Recoverability of initial NWP after implementing three modifications to the cleaning.

Table S4. Result data of the titers and protein removal in viruses filtration tests. Data was performed in duplicate and the error is the standard error.

DATA AVAILABILITY STATEMENT

The data that supports the findings of this study are available in the supplementary material of this article.