Meeting Report—Workshop on Virus Removal by Filtration: Trends and New Developments

2. Virus Removal by Filtration

Damon Asher (EMD Millipore) started the session by highlighting the importance of the use of purified virus for performing filtration studies. Virus preparations produced in cell culture contain impurities such as proteins, lipids, and DNA. The impurities can change filtration performance by blocking filter membranes and changing virus retention. This was demonstrated with minute virus of mice (MVM). An “ultrapure” MVM sample was prepared by ultrafiltration, ultracentrifugation, and finally a chromatographic polishing step (1) with consistency over several preparations in virus titre (8.70 ± 0.04 log₁₀TCID₅₀/mL), protein content (14.7 ± 3.7 μg/mL), DNA content (3.54 ± 1.17 μg/mL), and a genome copy to infectivity titer (log₁₀TCID₅₀) ratio of about 1:100. Filtration experiments with Viresolve® Pro (EMD Millipore) using a monoclonal antibody (mAb, 9 g/L) spiked with “ultrapure” MVM resulted in a flux corresponding to the nonspiked baseline, whereas spiking with less purified virus showed reduced flow. Ultrapure murine leukemia virus, pseudorabies virus, and reovirus preparations that show minimal spiking impact have also been developed. The results support the recommendation to use highly purified virus for filtration experiments.

Alternative strategies for virus purification and characterization for parvoviruses was provided by Joseph V. Hughes (WuXi AppTec). The most successful purification process demonstrated superior results when tested for filtration performance on three filter brands (EMD Millipore, Asahi-Kasei, Sartorius). Crude parvovirus spikes consisted of a virus suspension from infected cells separated from cell debris by low- and mid-speed centrifugation; virus was further purified by ultracentrifugation using a glycerol shelf (Ultra 1 virus preparations), or additionally subjected to purification by membrane absorber or column chromatography (Ultra 2), or processed further after DNA digestion with a polishing step and chromatography (Ultra 3). The stepwise purification was demonstrated by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE), overall protein, and DNA concentrations, with Ultra 3 demonstrating the most consistent purification parameters. Ultra 3 lots had high recovery in the 80% range, and demonstrated good protein analyses by gels (primarily viral-related). DNA was reduced from 7 × 10⁴ ng/mL to 3.0 ng/mL, and the final step provided a concentration capability so the Ultra 3 was consistently in the 10⁹ to 10¹⁰ PFU/mL (plaque forming units per milliliter) range. An analysis by dynamic light scattering revealed the Ultra 3 virus had the highest level of expected particles (99.7%) at the correct size (ca. 22–24 nm) compared to the Ultra 2 (75–85%) or Ultra 1 (50–70%), with larger-sized particles constituting the remainder in Ultra 1 and 2 lots. Ultra 3 porcine parvovirus (PPV) was most appropriate for filtration studies when the Viresolve Pro (EMD-Millipore) virus filter was used for several proteins, which were tested with up to a 5% v/v virus spike or 10⁸ to 10⁹ PFU/mL load with no or little impact on flow decay. Planova 20 N (Asahi-Kasei) demonstrated excellent performance if Ultra 2 or Ultra 3 PPV was used—up to a 5% spike, 3 mg/mL immunoglobulin G (IgG). Only minimal differences were seen between all three preparations (Ultra 1–3) for the Virosart CPV filter (Sartorius), but Ultra 3 had the best performance at different spike levels. Overall, Ultra 1 lots demonstrated inconsistency for purity and filtration performance, there was some improvement with the Ultra 2 process, and clearly the best and most consistent purity and performance was obtained with the Ultra 3 parvoviruses.

An evaluation of different filter brands was presented by Scott Lute (CDER/FDA). Small virus–retentive filters produced by Pall (Ultipor DV 20), EMD Millipore (Viresolve NFP), Sartorius (Virosart CPV), and Asahi-Kasei (Planova 20 N) were tested for fluid flow and retention of bacteriophages. The model viruses phage φX174 (Microviridae, produced in Escherichia coli), and PP7 (Leviviridae, produced in Pseudomonas aeroginosa) purified by cesium chloride gradient ultracentrifugation had high titres of 10¹² to 10¹³ PFU/mL and were monodisperse. Spiking with φX174 (26–30 nm; up to 10¹⁰ PFU/mL) and combined spiking with φX174 and PP7 (30–33 nm), both up to 10¹⁰ PFU/mL, generally demonstrated a log reduction value (LRV) that declined with overloading, but was independent of phage load for one filter type. Flux decline was observed early and correlated with phage load for one filter type (Viresolve NFP), but it occurred only at very high phage load for others (DV20, Planova 20 N) or was independent of a phage load up to 10⁹ φX174 PFU/mL for Virosart CPV filters. It should be noted that the above are challenge conditions unlikely to be encountered during routine manufacture. All filters were subjected to the PPV retention PDA Test Method outlined in PDA Technical Report 41 (2). Five filter types, three lots each, were tested in total to show that this method was valid; all filter types (Planova 20 N, 15N; Pegasus SV4; Virosart CPV and Viresolve NFP) met the criteria related to intravenous IgG passage and PP7 phage retention. However, the above studies also showed that there were differences between the filter brands related to the pattern of LRV vs flux decay or volumetric throughput.

Focusing on LRV decline with different filter brands, Amit Mehta (Genentech) investigated the underlying mechanism of action in collaboration with Pall Corp. and Penn State University. The small pore plugging model, that is, the hypothesis that preferential plugging of small pores results in more flow through the large pores causing higher virus passage, was investigated. Studies with and without IgG prefouling of the filter (challenge with ∼700 g IgG/m²) demonstrated that an LRV decline from 6 to 4 observed with the Pall DV20 filter was independent of prefouling and observed even in the absence of flux decline during phage challenge. The “small pore plugging” hypothesis could therefore not explain the phenomenon. Other hypotheses such as adsorption and pore entrapment, which suggest that saturation of virus binding sites could result in virus passage, that is, LRV decline, were considered. As the observed LRV decline with volumetric throughput was independent of phage PP7 concentration in the load (3 × 10⁶ to 2 × 10⁷ PFU/mL), these hypotheses were dismissed also. Furthermore, the external and internal polarization hypotheses were tested: external (surface) polarization refers to the retention of virus particles by the membrane resulting in higher concentration of virus near the membrane surface affecting virus passage and LRV. If this is the underlying mechanism, then residual retentate in the virus filter holding (virus retained on the filter surface) should have a higher virus concentration than the feed. Phages of different size, PR772 and PP7, and 20 nm colloidal gold particles were used in these experiments. Virus retention at the surface was higher with the large phage (PR772, ∼50 to 75%) but much less (∼20%) with the small phage (PP7). Consistent with the PP7 phage data, external polarization was not observed with the colloidal gold particles. Experiments with fluorescently labeled phage demonstrated that PP7 phage accumulates in a “reservoir zone” within the filter (internal polarization), which is followed by a retentive layer. Higher virus load increases the virus concentration in the reservoir zone resulting in higher virus passage with increasing throughput. Model calculations performed with the internal polarization hypothesis accurately predicted phage retention with DV20 filter as a function of throughput. Additionally, data in the literature (3) also confirmed model predictions.

Tomoko Hongo-Hirasaki (Asahi Kasei Medical Co., Ltd.) addressed the influence of protein concentration, ionic strength, pH, and level of IgG aggregates on the filtration performance of Planova 20 N (20 N) and BioEX filters using model IgG. Filtration was performed under standard pressure (78.4–98 kPa for Planova 20 N; 294 kPa for BioEX); in some experiments low pressure (19.6 kPa) was studied. Using a protein concentration of 10 mg/mL IgG, increase of flux (L/m²/h) was observed when NaCl was added in a range of 50 to 500 mM. At higher protein concentration (30 mg/mL IgG), flux decay was observed at higher salt concentration, and flux was enhanced only in the range of 10–100 mM NaCl. Effect of pH on flux and throughput were tested at 10 mg/mL IgG for 20 N and 30 mg/mL IgG for BioEX. 20 N kept high flux at pH 4–8 with a maximum at pH 4. Weekly acidic conditions of pH 4–6, below the isoelectric point (pI) of the antibody, were more desirable to maintain higher throughput. For BioEX, a pH range of 4–6 gave optimal conditions, whereas a further increase in pH resulted in flux decay. Increase in multimer level (multimer 0–0.8%; dimer ca. 4%, monomer 95–96%, 10 mg/mL IgG) correlated with a decrease in throughput, with BioEX being more sensitive to a higher multimer level than 20 N. Experiments with PPV (produced under serum-free conditions, 0.5% v/v spike, pH 4.5, 100 mM NaCl, standard pressure 78.4 kPa and 294 kPa for 20 N and BioEX, respectively) demonstrated that the antibody concentration (in the range 1–50 mg/mL) did not influence PPV retention for either filter. Changing ionic strength (up to 500 mM NaCl) influenced flux but not PPV removal (LRV ≥ 5). Variation in the pH (4.5 to 8) influenced filterability of the spiked material but did not influence PPV removal by both filters. Low pressure, however, resulted in virus breakthrough at specific solution conditions (20 N, pH 4–6, 10–100 mM NaCl, 19.6 kPa). Such flux phenomena were explained by electrostatic and hydrophobic surface interactions. The basis for virus removal is mainly size exclusion. The velocity of the particles through the filter is controlled by the pressure, and the hydrodynamic force controls the Brownian motion of virus particles. At low velocity/pressure the Brownian motion can be stronger and allow particles to move possibly to larger pores. The data support the conclusion that antibody concentration, ionic strength, and pH may have little influence on virus removal under standard pressure but may affect the flow; optimal ranges of these parameters are different for both filters (20 N and BioEX).

Arick Brown (Genentech) described experiments to improve the capacity of a small size virus-retentive filter (Viresolve Pro, EMD Millipore). The sequence of purification steps in the downstream processing of mAbs has a direct effect on filter performance, but it can be different with different mAb products. Complete removal of the parvovirus MVM was found in all experiments with Viresolve Pro using Viresolve Shield as a prefilter. Five different mAbs were used under variable conditions (pH 5.0 to 5.5, conductivity 6.7 to 11.8 mS/cm, throughput of 2.2 to 7.0 kg/m², flux 150 or 200 L/m²h); virus reduction of ≥4 log₁₀ was demonstrated in all experiments and a residual signal (< limit of quantitation [LOQ] by quantitative polymerase chain reaction [qPCR]) was detected in the filtrate in only one case. Small-packed cation exchange columns were used before the virus filter with large beads (100–300 μm bead size, 300–400 Å pore size) or traditional small beads (50 μm bead size, 500–1000 Å pore size); both resulted in a significant improvement of Viresolve Pro throughput. The combination of Protein A gradient elution before anion exchange chromatography translates downstream to greater product volume throughput; further enhancement can be realized by combining gradient elution and charged prefiltration (or pre-chromatographic column). Surfactants (polysorbate 20 and Triton X100) have an effect as well, but combining surfactant with prefiltration may provide synergistic effects and result in exceptionally high throughput. The approaches, while improving capacities, do not appear to affect MVM LRV.

Marcel Asper (Charles River Laboratories) tested the influence of pressure release during virus filtration on virus break-through. After spiking—PPV, 1.0% v/v in 1% bovine serum albumin (BSA) or 0.1% PPV in mAb solution (5 mg/mL)—the process fluid was subjected to prefiltration through a 0.1 μm filter and then processed through small virus-retentive filters. Five filter types were used, and the filtrate was collected in 10 to 11 fractions. The course of filtration was interrupted twice and the virus concentration measured. When spiked BSA solution was filtered, differences in virus retention between filter types were clearly shown; pressure release resulted in higher virus content in the filtrate with filters 1, 2, and 3; the effect was less with filter 4 and not detectable with filter 5. Both interruptions resulted in higher virus content in the following fractions. The same effect was demonstrated if mAb solution was used on filter 2 and an additional 6^th filter, or if the filter was washed with buffer at the end of the filtration process to obtain a better yield (filter 2). Pressure release during filtration reduced the reduction factor by 1.1 log₁₀ (4.6 log₁₀ without pressure release vs 3.5 log₁₀ with twofold pressure release, filter 3). The effect of virus break-through after pressure release was also demonstrated for encephalomyocarditis virus (EMCV) (22–30 nm in size), although the effect was not noticeable with a larger virus, MuLV (80–100 nm in size). The influence of pressure release was demonstrated also with one process filter. The data are of practical relevance; virus break-through must be taken into consideration if the filtration process is interrupted for technical reasons. Virus reduction values determined in small-scale studies without pressure release may then not be applicable.

Greg Buczynski (Grifols Inc., formerly Talecris Biotherapeutics, Inc.) investigated the robustness of PPV removal by Planova 15 N (Asahi Kasei) filters with and without prefiltration. Virus removal under set-point manufacturing conditions was determined for enveloped viruses—human immunodeficiency virus type 1 (HIV-1), bovine viral diarrhea virus (BVDV), pseudorabies virus (PRV), vesicular stomatitis virus (VSV)—and nonenveloped viruses: reovirus type 3 (reo3), hepatitis A virus (HAV), and PPV. Complete virus removal was observed with HIV-1, BVDV, reo3, and HAV with filter loads up to 65 L Product A/m². PPV removal was robust, but the LRV declined slightly with higher filter loads (LRV 5.0 at 25L/m², 4.2 at 65 L/m²). Robustness studies with PPV revealed no effect of low or high protein load (≤5 mg/mL vs >10 mg/mL) on flux or virus removal. Another brand of virus-retentive filter was used for Product B in combination with a prefilter that was needed to maintain flux. Robust PPV clearance was observed; between 1.6 and 3.3 log₁₀ PPV was removed by the prefilter and 3.5 to 4.8 log₁₀ PPV was removed by the virus-retentive filter alone, which resulted in a cumulative reduction factor between 6.2 and 7.2 log₁₀ under variation of pH, pressure, or protein content. These results initiated discussion on the acceptance of this cumulative reduction factor, as the mechanism of action of the prefilter is unclear and no integrity test is available to assure consistent performance by each prefilter module.

A strategy to select the right filter combination for a fibrin sealant consisting of thrombin (36 kD) and fibrinogen (340 kD) was discussed by Francisco J. Belda (Intituto Grifols, SA) with a goal of achieving robust clearance of ≥4 log₁₀ for small viruses (HAV and parvovirus). Virus-retentive filters of two manufacturers, A and B, were used with single and double filtration being applied. Thrombin could be filtered through two A15 filters or two B “small” filters in series, revealing LRVs of 6.56 (HAV) and 6.14 (PPV) (A15 filters) and ≥4.19 (HAV) and ≥5.63 (PPV) (B “small” filters). Implementation of virus-retentive filters in the production of fibrinogen is difficult, and the large molecule requires the combination of filters with large and small pore size. A combination of B “big” and B “small” filters resulted in only limited removal of PPV (0.96 log₁₀); the combination of A35 (large) and A15 (small) provided satisfactory results of 5.22 log₁₀ (HAV) and 4.37 log₁₀ (PPV) reduction, in contrast to the results obtained with “B” filters in the thrombin solution, which showed inconsistent removal of virus. Ultimately, filters of Manufacturer A were used for production of thrombin (A15 & A15) and fibrinogen (A35 & A15).

Shirley Stagg (Bio Products Laboratory Ltd.) provided a case study of the approach taken to select a virus for filtration studies of a plasma protein under development consisting of dimers or multimers with a molecular weights range of ∼98 to ∼600 kDa. Filters from EMD Millipore (Viresolve Pro), Asahi-Kasei (Planova 15 N and 20 N; BioEX) and Sartorius (Virosart CPV) were tested. Canine parvovirus (CPV), clarified from infected cell supernate and passed through a Planova 35 N filter without virus retention, was used in these experiments. The effect of virus spike on flux, protein load, ionic strength, and pressure was investigated. It was shown that virus retention properties of the filters could not be predicted from other proteins and that product-specific validation is needed to assure the effectiveness of the selected filter. Planova 20 N gave a good combination of virus removal and protein throughput, with a combination of two filters in series providing ≥4 log₁₀ removal of CPV. The transmembrane pressure had to be carefully controlled during serial filtration because parvovirus removal was affected by lower pressure.

3. Panel Discussion

A panel discussion followed the presentations with representatives of regulatory authorities from the EU (Glenda Silvester, EMA), Germany (Johannes Blümel, PEI), Spain (Sol Ruiz, AEMPS) and the USA (Kurt Brorson, FDA) joining the speakers. Questions from the audience or questions provided beforehand and listed in the program were discussed. Discussion focused on (i) the mechanism of virus retention and patterns of breakthrough, (ii) issues related to virus spike preparations, and (iii) the function and regulatory implication of prefilters.

1. It was agreed that the major mechanism of virus retention by filtration is size exclusion. Virus-retentive filters of different manufacturers (EMD Millipore, Sartorius, Asahi-Kasei, and Pall) are different in membrane structure and device assembly, which probably explains the different patterns of flux decay and virus breakthrough under stress conditions. However, consistency in the data was demonstrated if a filter is used for a specific product under controlled conditions, and virus-retentive filters are reliable and powerful tools to remove possible contaminants from the product. The mechanism underlying filtration is complex, and protein and virus can interact with the membrane in multiple ways, any of which may contribute to virus retention (e.g., adsorption). Pressure interruption during filtration resulted in enhanced virus passage; this has practical implications and needs to be considered when it happens during production. The effect is mechanistically not well understood and needs to be studied further.

2. Filters are believed to remove viruses largely by a size exclusion mechanism, and so it was discussed whether a small-size virus, like parvovirus PPV or MVM, can be used as surrogates for large viruses, like MuLV, as it has never been reported that MuLV passes through a membrane if PPV or MVM are removed. An advantage of the use of PPV or MVM is that highly purified preparations with titers up to 10¹⁰ TCID₅₀/mL can be produced, and so much higher reduction factors could be claimed for this step if they are accepted to be used as surrogates for the retrovirus, MuLV, which is more difficult to prepare cleanly. From a regulatory perspective, it makes sense that a small virus can be used to provide a claim for a large virus, such as MuLV, considering the experience and the mechanism of size exclusion for MuLV. This relies on the assumption that there is no aggregation or interaction of model parvoviruses with proteins that lead to aggregates larger than retroviruses.

   In the ideal situation, the properties of the virus should reflect the real contaminant. The different surface of the virus may change its interaction with proteins in the process fluid after extensive purification of the virus spike. The surrogate virus may also possess a higher tendency for aggregation. There is, therefore, no blanket recommendation from regulatory agencies regarding which specific virus preparation should be used for the validation of a specific step; this must be considered case by case. Also, the conditions of filtration may in theory influence virus retention. It was shown, for instance, that viruses shrink at high salt concentration, and this may happen with the protein also. Of great help would be the comparison of virus removal using spike preparations differing in purity and titer, while the use of “preconditioned” spike preparations would also provide useful data. The panel agreed that preparation of the virus stocks used for spiking experiments is important and should be described better in study reports than is current practice. Characterization of the virus spike in relationship to the major quality attributes, such as protein, DNA, and aggregation stage, is important also. Such data would help in the interpretation of data provided for virus retention by filters.

3. Several prefilters are used at present or are under investigation to improve flow rate and throughput of filtration in large-scale manufacture. Small product aggregates can cause filter clogging and reduction in the overall throughput. Companies have proposed several concepts such as using a large-size virus-retentive filter as a “prefilter”; others use depth filters, adsorbers (i.e., membranes containing charged linker), or chromatographic columns to clear these fouling components. It was discussed whether the prefilter and the virus-retentive filter can be viewed as one unit operation so that virus retention by both can be claimed as the viral clearance capacity of this manufacturing step. This question engendered some controversy. Whereas on one hand it was held that as the major mechanism of virus removal is the effect of the virus-retentive filter and both are used in production, then the combination can be studied in small-scale experiments with the subsequent reduction factor of the combined step reflecting the overall capacity of the combined step. On the other hand, some participants held that as the underlying mechanism of virus removal by the prefilter and the virus-retentive filter may be different, they should be measured separately. If the relevant mechanism (i.e., the size exclusion by the virus-retentive filter) needs to be determined, then this must be achieved by testing the material immediately before it is loaded onto the virus filter. Robustness and reliability of virus removal by the prefilter should be demonstrated as well, but separately. These and other relevant issues must be considered before a reduction factor calculated from a combined step, in this case prefilter and virus filter, can be accepted.

All together, the workshop was a very well-received forum for the discussion of the different aspects of virus removal by virus-retentive filters. The discussion should be continued to examine the growing mass of virus clearance data generated by the industry and to better understand the complex mechanism of virus removal by virus-retentive filters. It would be appropriate to combine a virus filtration workshop with future PDA meetings on Virus & TSE Safety.

Conflict of Interest Declaration

The authors declare that they have no competing interests.