Use of MMV as a Single Worst-Case Model Virus in Viral Filter Validation Studies

Size Exclusion Mechanism of Virus Filter Retention

Virus removal filters are designed, developed, and manufactured by vendors to retain particles by size exclusion. While filter products by different vendors can vary widely in materials of construction and format, the retentive capabilities of the membrane are dictated by the complex internal porous structure of the polymeric matrix. A summary of commercially available virus-retentive filters is shown in Table III (8). These membranes are either tangential flow filters (TFFs) or normal flow filters (also known as direct flow filters, DFFs). Membranes can have very fine pore structures and narrow pore size distribution, with particle capture at the membrane surface, as typically found for TFFs. Normal flow filters contain either an open symmetric or asymmetric pore structure that allows for entrapment of the virus particles within the depth of the membrane structure. Multiple membrane layers are often incorporated to boost overall log reduction factors by providing additive clearance across the layers. Examples of virus filter pore structure via scanning electron microscope (SEM) imaging are shown in Figure 1, illustrating the complex internal pore structure that provides a balance between high flux and passage for the product of interest and high retention of viruses (17). An important design criterion for virus filters is the minimization of non-specific adsorption to the membrane surface by either using a material of construction that is inherently hydrophilic in nature, or by coating the surface with a hydrophilic polymer. This serves to minimize product loss and fouling on the membrane surface, and thus results in virus particle retention characteristics that are dominated by a size-based mechanism. The membrane chemistry and hydrophilic modifications are summarized for the different virus filters in Table III.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

SEM cross-sectional images of Ultipor® DV20 (left) and Viresolve® Pro (right) membranes. Reprinted from reference 17 with permission of the publisher.

In the design of filters with more open pore structures, as typified by the standard normal flow filters used in therapeutic protein processes, the pore size distribution and pore frequency contribute to the overall filter performance and the resultant virus retention. Therefore, it is important for filter manufacturers to control and test for the actual pore size distribution of each membrane lot. With the use of advanced analytical techniques, filter manufacturers provide a high degree of quality assurance of their virus filter product. Examples of non-destructive tests that evaluate the pore size of a filter include liquid–liquid porosimetry (LLP) (18, 19), forward flow tests (e.g. air-water diffusion) (20), and a more recently developed binary gas diffusion test (21). Destructive tests that challenge the filter with a model particle in solution, and thus provide information on the retentive pore size of the membrane, include the gold particle test (22), dextran retention test (23), and bacteriophage challenge test (24). The advantage of non-destructive tests is that the filter can be 100% pre-use tested for every membrane/device produced, whereas destructive tests can only be performed on a selective lot sampling or post-use basis. Ultimately, filter manufacturers must provide evidence that their integrity test methods and the associated acceptance criteria correlate to retention of the targeted size range of virus at the desired reduction value (12).

An example is provided for the development of a correlation between pore size testing and virus particle retention. LLP involves the intrusion of one immiscible fluid into another through the pores of the membrane at a defined transmembrane pressure. Because liquid flow through the membrane pores is proportional to the fourth power of the pore diameter (18), the overall membrane performance may be dictated by a small number of the largest pores. LLP provides an independent physical measure of flow through these largest pores, which can then be correlated to virus retention as measured by log reduction value (LRV). Figure 2 shows an example of the correlation between LLP (or LLDP, liquid–liquid displacement porosimetry) and PP7 (bacteriophage) LRV for Sartorius polyethersulfone (PES) research samples (19). For each membrane sample, the pore size distribution is measured via LLP, and the mean (r_p,mean), max (r_p,max), and min (r_p,min) pore sizes are determined. These same samples are also tested for PP7 retention. In this figure, the maximum pores r_p,max in the pore size distribution show the best correlation to PP7 LRV; the trend is as expected, with increasing LRV at smaller pore sizes. The lower coefficient of determination (r² = 0.697) seen in this study is likely due to experimental variability in the bacteriophage assay. An additional correlation is later shown between the maximum pore size r_p,max and the 90% molecular weight cut-off value for dextran molecules; this later correlation shows a higher coefficient of determination (r² = 0.866). Other references also demonstrate the ability to develop a correlation between virus particle retention and pore size testing: Phillips and DiLeo demonstrate a correlation for the Viresolve® 70 and Viresolve® 180 membranes using a LLP test (18), and Giglia and Krisnan demonstrate a correlation for the Viresolve® Pro membrane using a binary gas test (21). The ability to correlate the retention characteristics of a virus to the pore size of the membrane provides evidence that the predominant mechanism for virus removal is size-based sieving.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Correlation between LLDP (liquid–liquid displacement porosimetry) and bacteriophage PP7 LRV for Sartorius PES filter. Each membrane sample is tested for pore size distribution by LLDP and PP7 retention, and the resulting PP7 LRV is plotted against the mean (black circles), maximum (gray circles), and minimum (white circles) pore size. Reprinted from reference 19 with permission of the publisher.

A paper by Grant and Liu (25) provides some fundamental insight into the particle removal efficiency of membranes in which size-based sieving is the dominant particle capture mechanism. The authors describe a methodology to test the particle capture efficiency of a microfilter using submicron latex particles with well-defined sizes. Their results show that particle capture is a strong function of particle diameter, as well as filter media thickness and filter loading. As expected, particle capture efficiency increases with increasing particle diameter. A sieving model was developed to predict the particle capture efficiency as a function of these variables; this model predicts the concentration of particles exiting the filter matrix relative to the concentration entering the filter. Their analysis showed that the model was able to accurately predict the same trends as the experimental results, thus demonstrating that the predominant mechanism of size-based sieving is understood.

There is direct evidence in literature showing the correlation of mammalian virus retention factor to the size of the virus. In a paper by DiLeo et al (26), mammalian viruses Polio Type-1 (32.6 nm), SV-40 (47.9 nm), Sindbis (54.1 nm), Reovirus Type-3 (77.6 nm), and MuLV (85 nm), as well as phage viruses PhiX-174 (28 nm) and Phi-6 (78 nm), were applied to a virus filter membrane, Viresolve® 70 (nominal 70 kDa pore size). While the membrane used in this study was operated in tangential flow mode, the results are still applicable towards the current industry standard normal flow filters, as both filter types operate under a predominantly size-based mechanism of removal. As shown in Figure 3, the virus retention in buffer and protein solution increases constantly with virus diameter, independent of virus class or type, from about 3.5 log₁₀ with polio to greater than 6.8 log₁₀ with MuLV. The authors describe further experiments to demonstrate the absence of adsorption effects through mass balance and filter soaking experiments. Additionally, the trend of monotonically increasing virus LRV with increasing pore size is well predicted by a hard-particle sieving model (solid lines in Figure 3). Altogether, these results support that the mechanism of retention is predominantly size-based sieving. Implicit in this mechanism is the understanding that the retention factor for a larger virus, such as MuLV, should be greater than that of a smaller virus, such as MMV; this is well supported by the experimental data.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Log reduction value (LRV) as a function of virus size for Viresolve® 70 membranes in buffer or protein solution. The retention coefficient in buffer (closed circles) is compared with that in the presence of protein (open diamonds) as a function of Stokes diameter. Rauscher MuLV was suspended in 10% FCS while the other viruses were in 0.65 mg/mL HSA. Reprinted from reference 26 with permission of the publisher.

The arguments presented thus far support the assumed dominant mechanism of size-based sieving: manufacturing of filters designed to capture virus particles within the complex pore structure, minimizing non-specific adsorption by use of hydrophilic polymers or a hydrophilic base matrix, performing pore-size integrity tests that show correlation to particle retention, and demonstrating the trend of increased virus retention as a function of particle diameter. Other mechanisms such as interception, impaction, diffusive capture, gravitational settling, electrostatic attraction or repulsion, and other non-specific adsorption, may also be present (27); however, the extent of impact of these other mechanisms is likely small given the body of evidence.

The proposal of using a single, small virus for parvovirus filtration studies shows some parallels to the current recommendation for a filter consensus rating method for small virus filters. As described in Lute et al (24), a task force was formed to develop a common nomenclature and standardized test method for the classification and identification of virus filters. The final recommendation was a rating method based on a single model particle challenge, bacteriophage PP7 (26 nm), to define the retention capabilities of the virus filter.

In another parallel, sterile filters are used in biopharmaceutical processing to produce a sterile product solution through removal of bacteria and other microorganisms. The design of these microfilters is similar to viral filters in that the predominant mechanism for particle removal is through size-based sieving. A sterilizing-grade filter is defined as one that produces a sterile filtrate when challenged by 10⁷ colony-forming units per square centimeter of filter area of Brevundimonas diminuta (previously classified as Pseudomonas diminuta) (28). A standard test procedure for performing sterilization validation is outlined under ASTM F838-83 (now F838-05) (29). That procedure recommends the use of a single test microorganism because of its small size, and to produce a consistent standard across the industry.

Selection of Model Viruses for Viral Clearance Studies

The choice of viruses used for viral clearance studies is based on the cell substrate used for production and the potential risk of introduction of adventitious viruses. Relevant viruses that are known contaminants, and model viruses with a diverse set of physico-chemical characteristics are used to evaluate the viral clearance capacity of purification process unit operations. General considerations include the nature of endogenous viruses produced by the cell substrate as well as the susceptibility of the cell substrate to adventitious agents that may be introduced through raw materials (1). Chinese hamster ovary (CHO) cells are widely used cell substrates for production of recombinant protein therapeutics (30). Compared to other less commonly used production cells, such as baby hamster kidney (BHK) cells, CHO cells are less susceptible to infection by viruses based on literature surveys (31). Although viral contaminations in cell culture processes are rare, a few instances have been reported (10, 32⇓⇓–35). In addition to MMV, contamination events in CHO cell culture processes with Cache Valley virus (CVV), Vesivirus 2117, and Reo-3 virus have been reported. Elimination of primary animal-derived raw materials, such as fetal bovine serum and porcine-derived trypsin, and use of chemically defined media for cell culture processes, has significantly reduced the risk of inadvertent introduction of adventitious viral contaminants. However, even in the absence of any primary animal-derived raw material used in cell culture processes, certain viruses such as murine parvoviruses still pose a risk due to their resistance to inactivation and ability to survive in the environment (36). Most of the reported contamination events are hypothesized to have been associated with use of contaminated animal-derived raw materials. However, Moody et al. attributed an MMV contamination of CHO cell culture process to a recombinantly derived raw material (36). Therefore, for mAbs produced in CHO cells without the use of primary animal-derived raw materials, MMV is a worst-case relevant virus because it is a small, non-enveloped virus (18–26 nm), is highly resistant to inactivation, can be a contaminant of non-animal-derived raw materials, and readily replicates in CHO cells. Contamination events with smaller (∼17 nm) porcine circovirus (PCV) have been reported (37). However, these events were associated with the use of contaminated porcine-derived trypsin in a Vero cell culture system that was capable of supporting PCV replication. Porcine-derived trypsin is not used in the production process for CHO-derived protein therapeutics, and CHO cells have been shown to be not permissive for PCV infection (Amgen unpublished data).

Virus morphology is another important consideration in the selection of a worst-case model virus. Most virus particles are spherical in nature. However, there are some viral families that include representative members that are filamentous and are rod- or kidney-shaped. These include filoviridae, rhabdoviridae, and toroviruses in the coronaviridae family. Although filoviridae are rod-shaped and variable in length, they have diameter that is approximately 80–100 nm. The filoviridae family of viruses contains a single genus Filovirus. The members of this family cause severe hemorrhagic fevers in humans and non-human primates. Rhabdoviridae is a large family of RNA viruses that are also rod-shaped with variable length (100–430 nm). The members of this family are more uniform in diameter and range between 45 and 100 nm. Torovirus is a genus of Coronaviriae and is pleomorphic. They may be rod- or kidney-shaped and the diameter of the viral particles ranges from 120–140 nm (38). Although the morphology of these viruses are different in that they are not spherical, they are generally larger in diameter than parvoviruses and represent viruses with negligible risk with respect to being a contaminant in chemically defined media used in cell culture processes.

The quality attributes of the virus stock preparation, such as titer and purity, are also important considerations in the selection of virus used in the spiking study (39). Regulatory guideline document ICH Q5A recommends the use of high-titer stocks to demonstrate adequate viral clearance (1). High-purity stocks are also desirable in virus spiking studies so that the scale-down model system can more accurately represent the actual manufacturing process. This is especially relevant for the viral filtration unit operation for two reasons. First, in the potential event of an actual viral contamination, any virus particles that would reach the viral filter would be of relatively high purity, given the location of the unit operation further downstream in the purification process. Second, there are a number of published studies that provide evidence for the negative impact of virus spike impurities on filter performance (11, 39⇓⇓–42). A review of current literature shows MMV stock preparations to be high in purity and titer (9, 16, 39, 40). It is generally recognized that MMV can be grown to a higher titer and purified more readily as compared to xMuLV (16); further evidence for this assertion is provided in a survey of virus stock preparations from various contract laboratories showing that MMV preparations exhibit relatively low levels of protein impurity and host-cell DNA impurity at high infectious titer (9). Recent work on optimization of MMV stock preparations suggest that virus stock titers and purity can be even further enhanced with modifications to the virus production process (43, 44). Thus, the selection of MMV as a single worst-case model virus is well suited to not only provide high titers to demonstrate the maximum retentive capability of the virus filter, but also high purity so as to minimize non-representative fouling of the filter during challenge studies. Porcine parvovirus (PPV) is another model parvovirus of similar size to MMV and is also commonly used in viral filter challenge studies. In the same survey of virus stock preparations, PPV was similarly observed to exhibit low levels of host-cell impurities at high infectious titer (9). An FDA meta-analysis of viral filtration records from regulatory submissions compared LRVs between MMV and PPV; the mean clearance values were found to be quite similar (MMV 4.6, n = 99; PPV 4.9, n = 27; p = 0.2) (11). While the use of MMV as a single model virus has its merits based on the points described previously, the data reviewed here suggests that PPV could be an acceptable alternative.

Datasets Comparing Clearance of Different Model Viruses

The evidence presented thus far has focused on the broader aspects of the size-based removal mechanism for virus filters as a class, as well as discussion of the rationale for choosing MMV as a worst-case virus to assess parvovirus filter clearance. This section summarizes datasets for currently available commercial parvovirus filters, and compares the removal efficiency of smaller viruses, such as MMV, to that of larger viruses.

Amgen Planova™ 20N Dataset

A review of Amgen's historical Planova™ 20N (Asahi Kasei Medical) validation data was performed for viral clearance studies between 2003 and 2011, and includes results from 23 mAb processes (Figure 4). The data include immunoglobulin G1 (IgG1) and IgG2 antibodies, and represent both early-stage and late-stage clinical process development with a variety of process and solution conditions. Buffer salts include sodium acetate, sodium phosphate, sodium citrate, sodium sulfate, sodium chloride, and Tris, with the pH of solution ranging from pH 5.0 to 7.5. Filter loadings were generally around 200 L/m², but ranged as high as 400 L/m². Virus spiking was performed with MMV and xMuLV, with either in-house or vendor-sourced virus preparations. In a majority of the cases (39 of 41 runs), MMV was not detected in the filtrate for the chosen process conditions, and for all of the runs, xMuLV was not detected in any of the filtrate samples. No difference in virus retention performance was observed between the sources of virus stock, suggesting that the quality attributes of the virus preparation did not have a large role in determining achievable LRV on the virus filter. Consistent with the data shown by Genetech and Pfizer, these results show that LRV achieved with xMuLV are comparable to MMV, with differences in LRV attributed to virus stock titer and dilution required to eliminate cytotoxicity of the test article in the virus assay. Given the size difference of the viruses, it is generally expected that a greater clearance value should be attainable for xMuLV; however, the achievable LRV is often more limited for xMuLV due to its lower virus stock titers and lower percentage spike used to avoid filter plugging from the virus spike impurities. An example of this scenario is illustrated in Table IV, in which the LRV for the larger xMuLV is demonstrated to be lower than the smaller MMV due to the items listed above and including a higher assay dilution factor due to indicator cell cytotoxicity. In both cases, no virus is detected in the filtrate sample (output titer), but the experimental conditions dictate the value reported for the assay sensitivity limit. Taken together, these results present a compelling case for applying the LRV obtained with MMV, the smallest of the panel of model viruses used for filter validation, to claim LRV across this unit operation for larger viruses such as xMuLV.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Summary of Amgen Planova™ 20N virus clearance datasets with MMV and xMuLV. LRV data is shown for 23 mAb processes; some datasets were performed with duplicate runs. Plus signs represent xMuLV LRV with no virus detected, closed and open circles represent MMV LRV with and without virus detected in the permeate, respectively.

View this table:

• View inline
• View popup

Table IV

Example of Experimental Conditions Resulting in Lower LRV for xMuLV as Compared with MMV

Amgen Viresolve® NFP Dataset

An example of the viral clearance capability of the Viresolve® NFP filter is presented with an in-house viral clearance study for mAb B. Runs were performed using the model viruses MMV, xMuLV, PRV, and Reo-3. Virus retention on the NFP viral filter is known to be affected by flux decay, which is defined by the final flow rate normalized to the initial flow rate. As described by Bolton et al (42), flux decay on the filter indicates pore plugging, and the progressive plugging of the smaller pores of the viral filter leads to a proportional shift in flow through the larger pores, which results in higher virus breakthrough in the permeate. The effect of flux decay on the filter was studied for both the MMV and Reo-3 viruses for this process.

Virus spiking studies were performed with MMV in duplicate and Reo-3 in triplicate in which grab samples of the permeate were tested at specific flux decay points (Figure 5). For MMV, virus was observed in most of the permeate grab samples, with increasing virus breakthrough as flux decay increased. This is consistent with the findings reported in Bolton et al (42). The resulting pool also showed MMV breakthrough with LRV of 4.07 and 4.28 out to 80% flux decay for the duplicate runs. In contrast, no Reo-3 was detected in any of the grab or pool samples even after extensive flux decay (up to 80%), with a final pool LRV of ≥4.62. The Reo-3 and MMV datasets further provide supporting evidence for the size-based sieving mechanism because the smaller MMV (18–26 nm) results in higher virus breakthrough as a result of the selective plugging of the smaller membrane pores, whereas the larger Reo-3 virus (60–80 nm) does not break through.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

Grab or pool sample LRV for MMV (in duplicate) and Reo-3 (in triplicate with identical results) as a function of flux decay on the Viresolve® NFP filter. Open circles indicate no virus detected, closed circles indicate virus detected. Three runs performed with Reo-3 (triangles) showed identical results with no virus detected.

Viral clearance for xMuLV (80–110 nm) and PRV (120–200 nm) was also assessed, and the data is summarized in Table V. As expected, these larger viruses are fully retained by the NFP filter, although as discussed previously, the lower values for the LRV are a function of the spike and assay conditions. These two datasets shown for the panel of four viruses demonstrate that a virus challenge with MMV is capable of achieving at least 4 log₁₀ LRV for effective, robust clearance, and that this clearance value is similar to that which can be practically achieved for the larger viruses.

Conclusions and Recommendations

This paper has described the concept of virus validation on a parvovirus-grade filter with a single worst-case model virus and presented multiple supporting arguments in favor of this approach, using both scientific literature and Amgen's own data. There is a general consensus in the literature that virus filters provide virus removal predominantly through a size-based sieving mechanism. In the design and production of these filters, non-specific adsorption to the membrane surface is minimized, and the pore size distribution of the membranes is controlled and tested to show correlation to virus retention. Examples were provided from the literature to show a direct correlation between membrane pore size, virus size, and virus retention. Furthermore, the practice of defining the retention capability of a filter with a single model test organism has been well established in the area of sterile filter validation. Because larger viruses are expected to be retained more effectively than smaller viruses, the reduction value obtained for the smallest model virus is a conservative estimate for that of larger viruses.

The choice of a parvovirus, MMV, as the worst-case model virus is well supported through literature assessment of virus contamination events applicable to CHO-based cell culture processes with chemically defined media and no animal-derived raw materials. MMV is a worst-case relevant virus because it is small (18–26 nm), non-enveloped, highly resistant to physico-chemical inactivation, a known contaminant of non-animal-derived raw materials, and readily replicates in CHO cells. In addition, MMV stock preparations offer the additional benefits of high titer and high purity, resulting in virus spiking studies that better simulate representative manufacturing conditions and demonstrate high log reduction factors.

To illustrate the retention capability of different parvovirus-grade filters for a range of model viruses in comparison with MMV, datasets were summarized from the 2009 Viral Clearance Symposium white paper, as well as Amgen data with the Planova™ 20N, Viresolve® Pro, and Viresolve® NFP filters. In these studies, the parvovirus filters consistently showed full retention of the larger viruses, with no virus detected in the permeate pool. In general, MMV was also fully retained, except in the case of the NFP filter, which exhibited virus breakthrough as a function of flux decay. This result is expected based on previous literature, and in fact it provides further evidence for the size-based sieving mechanism, as this small model virus exhibits virus breakthrough as a result of the selective plugging of the smaller membrane pores. In contrast, no breakthrough was observed for larger viruses such as Reo-3, xMuLV, and PRV. This differential virus retention behavior was directly demonstrated by researchers at Eli Lilly in virus co-spiking experiments, which showed that retrovirus is fully retained on the virus filter even when parvovirus is exhibiting breakthrough. While these examples bolster the argument for a predominant size-based sieving mechanism, the occurrence of parvovirus breakthrough, and thus lower achieved clearance values on the filter, underscores the need to carefully consider the case for when a single worst-case parvovirus challenge approach can be used. Given that the parvovirus clearance value will be used as a conservative estimate for clearance of larger viruses, it may result in an underestimation of actual clearance capacity for the larger viruses. Higher clearance factors may be necessary for certain large viruses, such as retroviruses, in order to meet an overall clearance target. It is recommended to proceed with this streamlined approach only when the expected parvovirus clearance will provide the desired level to claim for larger viruses.

This approach of using a single worst-case model virus has been discussed in a number of recent scientific forums, and published reports from the 2009 Viral Clearance Symposium and the 2011 PDA Virus & TSE Safety Forum Workshop on Virus Removal by Filtration indicate a gaining acceptance (11, 16). Certain regulatory authorities are accepting clinical trial applications with this single worst-case model virus approach (46). Amgen has successfully filed clinical trial applications with this approach. Amgen has also used this approach in support of a parvovirus filter addition to the purification processes of two globally marketed products as a post-approval change. Specifically, the viral clearance data obtained for MMV from the virus filter step was approved as an acceptable surrogate value for the determination of the virus clearance capabilities of the entire purification process for the larger model viruses. Justification was given that for the size-based mechanism of clearance, the larger virus is expected to equal or exceed the level demonstrated for MMV, and therefore use of the MMV value is considered a conservative estimate. These examples of broad acceptance of this scientific justification by regulatory reviewers suggests that use of a single worst-case model parvovirus with accompanying justification based on the mechanism of action could be used to support marketing applications as appropriate. The guideline document ICH Q5A (1), which provides guidance for marketing applications and was published in 1998, suggests that the viral clearance robustness of a step is demonstrated through use of viruses exhibiting a range of biochemical and biophysical properties. The document also describes a need to justify the choice of viruses used in the studies, and that this choice is influenced by the production process. If the purpose of testing multiple viruses is to characterize the robustness of the process, given that the mechanism of removal for these parvovirus filters is known and understood to be size-based, then testing of viruses larger than the filter size rating would not provide additional robustness information about the process step. This is in line with the guideline's allowance that process step validation testing that does not provide value-added robustness information about the process need not be performed. Given the body of scientific evidence that now exists in support of the size-based mechanism of removal for parvovirus-grade filters, updating the guidelines based on current knowledge and practice should be considered. A recent FDA presentation suggests that “integration of data from multiple sources within the scientific community supporting mechanism of action and reliability of clearance obtained could lead to changes in approach” (47). It is the authors' desire that the results presented herein are one step in that direction.

In summary, this paper provides a strong scientific rationale for using a single worst-case model virus, MMV, for parvovirus filter virus clearance studies. The data were presented in a comprehensive manner to comment on filter design and the size-based sieving mechanism, and on supporting evidence from in-house and external datasets of the retentive capabilities of these filters. This new level of understanding for the underlying viral clearance mechanism should support the adoption of this new approach, which will ultimately allow for more efficient and streamlined viral clearance studies.

Declarations

The contents of this publication do not represent an Amgen endorsement of any products described herein and are not meant to imply that Amgen uses any of these products for clinical or commercial manufacturing.

The authors declare that they have no competing interests related to this manuscript, financial or otherwise.