A scientist assesses virus-filter integrity using a bubble-point test method. PHOTO COURTESY OF MILLIPORESIGMA (HTTPS://MILLIPORESIGMA.COM)
Estimating pinhole defects in large-scale membrane filters used during manufacturing is both a critical element of viral safety and a significant challenge, particularly for virus filters in large-scale multiround housings. Understanding the correlations among virus-filter structure, filter throughput, hydraulic permeability, and virus-clearance capacity is important for efficient process design and assurance of viral safety. Filter integrity tests (FITs) measure flow or pressure drops across wetted filters. Tests based on forward-flow diffusion often are used to assess virus-filter integrity for biomanufacturing applications (1–4). A pressurized gas, typically air or nitrogen, on the upstream side of the membrane provides a driving force for diffusion across a wetted membrane. Results from such testing enable detection of defects and gross leaks in individual filters and multimembrane assemblies. Filter damage can thin the liquid layer under high pressures and strong diffusive flow (5). Small holes and defects generate full convective flows that are significantly stronger than are background diffusive flows. Vendors base their filter-acceptance criteria on correlations between diffusive-flow parameters and virus retention.
During scale-up, batch-size increases necessitate larger numbers of individual virus-filter elements operating in parallel. Hence, it is critical to define an FIT strategy based on a single filter in an individual housing or on multiple filters in a multiround housing. An FIT serves to confirm membrane or membrane-assembly integrity, providing a high degree of confidence in proper filtration. Herein, we perform a technical risk assessment of whether an FIT run can provide sufficient assurance of viral clearance — expressed as virus log-reduction values (LRVs) — when multiple filters are mounted in the same housing. Using available literature, we describe the sensitivity of a forward-flow diffusion test for one filter or for multiple filters in a single multiround housing.
An FIT is a time-consuming procedure in a good manufacturing practice (GMP) facility, and increasing the number of membrane assemblies also increases the procedure’s complexity with respect to volumetric scale and installation requirements. Hence, before adopting a complex strategy for managing integrity testing of large virus-filter assemblies, it is important to understand the sensitivity of the forward-flow diffusion test for a single filter or for multiple filters in one housing.
Available literature includes theoretical estimations of the largest hole that consistently can go undetected in a single filter within a larger assembly of virus filters. In such studies, the integrity-test limited minimum LRV is based on the size of the hole as a function of the number of filters in a housing. Such studies show the minimum LRV to be largely insensitive to the number of filters within a housing. We suggest that there is minimal benefit to testing filters in individual housings or adjusting test specifications to compensate for increasingly larger numbers of filters in a housing.
A key advantage of properly engineered virus-removing membranes is that filtration depends predominantly on a size-exclusion mechanism. Validation of a virus-filtration step involves a scaled-down filter model for performing virus spiking studies (e.g., with porcine parvovirus, PPV) as well as production-scale quality assurance (QA) procedures that ensure consistency with spiking-study results. Such procedures include performance qualification (PQ) of the virus-filtration step that encompasses integrity testing.
Filter manufacturers have developed integrity tests to confirm that virus filters meet specifications or end-user virus-removal claims under processing conditions. End-user integrity tests need to detect plausible modes of filter damage — e.g., from defects, improper shipping and handling, improper installation, autoclaving/steaming-in-place, and out-of-specification (OoS) forward and/or reverse pressures.
Currently, integrity of virus-removal membranes can be measured using a particle-challenge test (e.g., colloidal gold or dextran), an air–liquid test (bubble point or air diffusion), or a liquid–liquid porosimetry test (the CorrTest procedure from MilliporeSigma). Among such methods, biologics manufacturers prefer nondestructive air-diffusion tests because they are relatively simple and provide reliable options for both detecting micrometer-level gross defects and confirming false failures.
Possible Strategies: For large-scale batches (>10 kL), efficient and robust integrity testing with multiple virus-filter elements can be challenging. Our company has used products from many filter manufacturers for processing large volumes of protein solution. Different test configurations are possible depending on specific process-control strategies. We describe four approaches below.
Filters can be installed in parallel in individual housings. This strategy allows for individual integrity testing of each filter element. However, the process of installing individual components and performing integrity testing of each element is time-consuming, resource-intensive, and operationally nonrobust.
Multiple filters can be placed in a single housing, and the test specification for the assembly can be corrected statistically to include more filters in a given housing, effectively reducing the test specification for each element. This approach increases the likelihood of detecting a single failing filter in a large assembly, but potential for false failures also increases with assembly size.
Filters can be placed in a single housing, and information about the release test for each filter can be used to build a unique specification for each assembly.
The simplest approach to managing multiple filter elements is to enclose them within a single housing and to set the integrity-test specification for the assembly by multiplying the specification for a single filter by the number of filters applied. The primary drawback to this approach is that increasing numbers of filters diminish the test’s ability to detect a single filter with failing gas flow in the entire assembly. Significant damage to one element in a large assembly would generate substantial, detectable flows. However, a single, marginally damaged filter might exhibit test flows that are above an individual filter specification but are not high enough to push the total test flow over the specification for an entire assembly.
Because each strategy has pros and cons, our company has performed risk assessment for logarithmic virus reduction based on scientific evidence with respect to FIT values. Below, we discuss the industry-preferred approach for large-scale manufacturing-volume processes: a single FIT based on multiple membranes in one holder and multiplying the specification for a single filter by the number of filters applied. That method has been used in a number of published experiments and hypothetical worst-case scenarios provided by virus-filter suppliers around the globe, including MilliporeSigma, Pall (now part of Cytiva), and Asahi Kasei.
FIT Based on Gas–Liquid Diffusion — Theoretical Justification: Gas–liquid tests often are used to evaluate sterilizing-grade and virus filters. Excess flow through a wetted membrane indicates a leak. Here, convective flux must be differentiated from background noise caused by diffusive flux. A change in diffusive flux shifts the threshold flow rate between integral and nonintegral membranes. Figure 1 illustrates principles of diffusive flux according to Fick’s law and convective flux under laminar and choked flow (6).
Figure 1: Gas–liquid diffusion test theory (figure ADAPTED FROM 6)
Pore size is the controlling factor in virus retention. According to Fick’s law, the rate of diffusive flow (Q) depends not on pore size, but rather on membrane porosity. Thus, a test based on gas–liquid diffusion can detect filter leaks but is insensitive to pore-size distribution.
Influence of Defect Size on Virus Retention: We investigated correlations between pore defect size (e.g., pinhole damage) and rate of virus removal. Integrity testing returned LRVs of ≈4.5 for clean membranes and ≈4.3 for membranes fouled at 75% (Figure 2). Accordingly, LRVs are ensured up to 75% in a fouled membrane compared with values for defect-free, clean membranes.
Figure 2: (top) Effect of defect size on virus retention, assessed in terms of logreduction value (LRV); (bottom) relationship between virus removal rate and pinhole diameter, as shown by Sekine et al. (7); an arrowhead indicates that a value exceeded the LRV upper detection limit. The solid line represents the curve fit (approximating a power function) for empirical data for 0.1-m2 filters. The dotted lines indicate extrapolation to the 1.0- and 4.0-m2 filters using that approximation.
Sekine et al. studied virus clearance with respect to pinhole size and membrane surface area (7). Their results demonstrate that viruses are likely to pass through artificially generated, micrometer-sized pinhole defects, but permeate flows through such defects are too low compared with overall permeate flow to measure LRV impacts appreciably.
For 0.1-m2 virus filters, Sekine’s team observed a decline in LRV for PPV beginning at defect sizes of about 3 μm, and the reduction in LRVs increased in magnitude with larger pinhole diameters. However, for filters with larger effective surface areas, pinhole defects had less of an impact on PPV LRV than they did for the 0.1-m2 filters. In Figure 2 (bottom), that result is seen most clearly for pinholes of 8–9 μm. For such defects, LRVs for PPV were 3.3 in 0.1-m2 filters and 5.0 for filters with effective surface areas of 1.0 and 4.0 m2. Thus, single pinhole defects have relatively less of an impact on the LRV of filters with large effective surface areas.
Relating Air-Diffusion Flow Rate to Virus Retention: The choked-flow formula helps to model the relation of air diffusion to LRV, an indicator of virus-filter performance. Figure 3 depicts the effect of defect leak rate to virus retention. We followed this approach presented by Giglia et al. (6):
• For nonintegral (defect-bearing) membranes, excess-flow rate is used to calculate a defect size that is consistent with that flow rate. This calculation assumes that a defect penetrates all membrane layers and that, according to choked-flow principles, particle speed limits the flow rate for defects of >10 µm.
• The resulting defect size is used to calculate the liquid flowthrough.
• Assuming nonretentive flow through a defect, its impact on virus retention can be calculated based on known rates of flux and retention through integral pores.
Figure 3: Effect of defect leak rate on virus retention, assessed as log-reduction values (LRVs) (FIGURE ADAPTED FROM 6)
Cd: discharge coefficient, d: pore diameter, LRV: log-reduction value, Pf: feed-side pressure, Pp: permeate-side pressure, Q: diffusive-flow rate, VI: volume of feed
passing through the integral portion of a membrane, Vd: volume of feed passing through a defect.
Applying those conditions, excess flow rate can be used to estimate virus-retention loss (Figure 3). Those calculations translate to >4 LRV assurance for filters with integral pores (LRVI) for an excess-flow rate of 30 sccm in make-A virus-filter devices (Figure 4).
Figure 4: Relationship between excess-flow rate and virus retention; membrane area = 1.5 m2, membrane permeability = 200 LMH/psi, liquid transmembrane pressure = 30 psi, air differential transmembrane pressure = 50 psi.
If LRVI is extremely high — e.g., if the volume passing through an integral membrane is significantly higher than that moving through a defect (VI >>> Vd) — then the LRV-assurance equation in Figure 3 simplifies to LRV = log(VI ÷ Vd). The volume passing through an integral membrane is proportional to filter area, pore thickness, and change in membrane pressure (VI = A × Lp × ∆P). For purposes of risk assessment, Vd represents the maximum liquid-leak flow rate. The liquid flow rate is proportional to the gas flow rate, which is proportional to the filter area. Thus, both VI and Vd are proportional to filter area, and LRV assurance is independent of filter area or module.
Theoretical LRV Comparison of Single- and Multimembrane FIT Results: According to the theory described above, LRV is irrelevant to filter area and to the number of filters mounted onto a holder. For multifilter systems, vendors recommend that diffusion specifications be determined by multiplying the diffusion specification of a single device by the total number of devices in the system.
A key question during integrity testing concerns whether an assembly can mask a defect in a single filter. Just as one large device can be considered as an assembly of multiple smaller devices, multiple devices manifolded together can be evaluated as one large device. Because both the integral liquid-flow rate and detectable leak rate (excess flow rate) are essentially proportional to filter area, assurance of virus retention in terms of LRV (e.g., ~log(VI Vd)) is practically independent of both filter area and the number of devices within an assembly. It is not that defect masking does not occur, but rather that any masking of a small defect within one membrane is offset completely by the added integral flow rate of other devices within a manifold.
A couple of hypothetical worst-case scenarios can illuminate the effect on virus-retention assurance of a single device within an assembly that exhibits marginally passable diffusive flow (8). The left side of Figure 5 shows a manifold of three make-A devices, each of which exhibits a marginally passing diffusion value of 60 sccm. Devices can exhibit passing values as low as 30 sccm, so in this worst-case example, each device has an excess-diffusion rate of <30 sccm. Excess of that magnitude can result in a virus LRV as low as 4.1 for each device within the assembly — and therefore as low as 4.1 for the entire assembly.
Figure 5: Hypothetical worst-case scenarios for assemblies of three make-A filtration devices; (left) all three devices score marginally
passing air-diffusion values. (right) Two devices show low-end diffusive values, while one device presents a high excess flow rate (pfu = plaque-forming unit) (figure ADAPTED FROM 8).
Providing an extreme situation in a different direction, the right side of Figure 5 depicts an assembly with three make-A devices. Two of those devices exhibit flow rates at the low end of passable diffusive values (30 sccm). The third device shows excess diffusive flow at the highest possible level that still allows the entire assembly to pass an FIT (diffusive flow of 120 sccm, excess flow of 90 sccm). Note that the LRVs of both assemblies in Figure 5 are identical: 4.1.
Dowd analyzes in detail the impact on integrity-test sensitivity of multiple filters installed into an assembly (9). He states that a gross failure of any individual device will cause the entire assembly to fail a test. If one device shows a marginally high diffusion value, there is a possibility that the entire assembly will show a passing diffusion value — but in such a case, that assembly will have an acceptable LRV. Data from his study confirm that even if one filter’s air-diffusion rate is high, the assembly’s virus-removal rate is preserved by the combination of other filters therein. He also concludes that no significant benefits come from test-specification compensation for multiple filters.
Dowd defines an FIT limited detectable failing flow rate as the flow rate that will cause 99.9% of all possible filter assemblies to exceed the specification by the smallest margin (9). As Figure 6 shows, only 0.1% of all possible filter assemblies would pass an FIT despite the presence of a single hole discharging gas at the detectable failing flow rate.
Figure 6: Definition of detectable failing flow for any sized assembly of filters in a single housing (9).
The likelihood that a damaged filter will be applied increases with higher numbers of filters in a single housing (Figure 7). But the aggregate failure rate still should be extremely low given failure-rate estimates for a single cartridge. Available failure-rate estimates suggest that even one undetectable filter breach is unlikely to occur over the commercial lifetime of a biotherapeutic.
Bolton et al. provide a methodology for relating gas flow through a hole in a virus filter to LRV loss, and they support their calculations with experimental data (10). The authors present a statistical approach for determining the minimum detectable gas flow through a hole crossing all virus-filter layers. Measuring defect gas flow enables estimation of defect size, the proportion of liquid flowing through the hole, and the resulting worst-case minimum LRV. The authors show that value to be largely insensitive to the number of filters incorporated into a single housing.
Figure 7: Integrity-test limited minimum log-reduction values (LRVs) dependent on the number of 30-inch filter units in a single housing; trends are shown for cases in which liquid flow through a membrane hold is treated as flowing through a cylindrical channel and through an orifice (9).
Drug manufacturers must perform their own risk assessments for each type of virus filter using vendor-validation support documentation. Evaluation should account for potential defects, including improper pore-size distribution and gross (pinhole) damage that may affect a filter’s virus-clearance capability. Such risks should be scored based on their severity, rate of occurrence, and difficulty of detection as per standard quality risk assessment rubrics. Filter manufacturers usually perform quality testing before product shipment. Each lot is inspected for proper pore-size distribution with a pressure-hold test, and the absence of defects is confirmed using a leakage test based on air/water diffusion or binary gas diffusion. That said, end users also must evaluate the necessity of checking for filter pore-size distribution and/or evaluate filters for gross defects after receipt and before use.
Once a filter has been shipped, its pore-size distribution should not shift under vendor-recommended use conditions, assuming that the filter has sufficient mechanical strength. Filter capability for targeted virus removal can vary across medicinal products and operating procedures. Such factors should be part of each risk-assessment exercise. Acceptance criteria should be validated for each production process at an end user’s site.
Based on gas–liquid diffusion test theory and modeling approaches for relating air-diffusion flow rate to virus retention, we have established that an FIT limited minimum LRV is largely insensitive to the number of filters that are placed in a single housing. Even if an assembly contains a nonintegral membrane, an FIT for a multifilter assembly should return an LRV of at least 4. That conclusion derives from the choked-flow formula.
Regarding defect masking within an assembly, just as one large device can be considered as an assembly of several smaller devices, an assembly of manifolded devices can be evaluated as one large device. Because integral liquid-flow rate and detectable-leak rate (excess flow rate) are proportional to filter area, virus retention (LRV) is practically independent of both filter area and the number of devices within an assembly. An assembly can mask a defect in a single filter, but any effect of such masking is offset by the added integral flow rate of other devices within the manifold.
Detailed analysis of FIT sensitivity to the number of filters in a multiround assembly has shown that a gross failure of any individual device will cause the entire assembly to fail an integrity test (9). Thus, there is no significant benefit to adjusting FIT specifications for an individual filter housing or multiple filters. Available literature also shows that the FIT limited minimum LRV is largely insensitive to the number of filters incorporated into one housing (10).
Under normal process/experimental conditions, filters that pass a leakage test achieve suitable virus removal. The incidence of detrimental shifting in pore-size distribution is negligible under normal filtration conditions for materials from most virus-filter manufacturers. Thus, end-user integrity testing needs to cover only damage that may have occurred during shipping and handling, such as tears and broken fibers. Our analysis of virus-filter experimental data also suggests minimal benefit to placing filters in individual housings or adjusting the test specification to compensate for larger numbers of filters in a given housing. We conclude that multidevice FIT presents low risk for evaluation of virus-clearance effectiveness.
However, risk assessment must be performed in individual cases to establish test specifications, number of filters in a housing, and necessary level of virus reduction for patient safety. By considering the low risk of evaluating the integrity of single or multiple filters in a given housing, drug manufacturers can focus limited resources on areas that are best positioned to improve process efficiency and patient safety outcomes.
We thank the anonymous referees for their suggestions; Chris Dowd for permission to use content from his study; and Sal Giglia for his contributions to our risk assessment and permission to use data for our manuscript.
1 Volume 4: Good Manufacturing Practice (GMP) Guidelines, Annex 1 — Manufacture of Sterile Products. European Commission: Brussels, Belgium, 22 August 2022; https://health.ec.europa.eu/document/download/e05af55b-38e9-42bf-8495-194bbf0b9262_en?filename=20220825_gmp-an1_en_0.pdf.
2 PDA/BioPhorum Sterilizing Filtration Quality Risk Management Consortium. Points To Consider for Implementation of Pre-Use Post-Sterilization Integrity Testing. Parenteral Drug Association: Bethesda, MD, 2020.
3 Technical Report 41 (Revised): Virus Retentive Filtration. Parenteral Drugs Association: Bethesda, MD, 2022.
4 Chaloupka K, Pathak N, Kundu S. To FIT or Not To FIT, That Is the Question. BioPharm Int. 22(11) 2009: 40–45; https://www.biopharminternational.com/view/fit-or-not-fit-question.
5 Jornitz MW. Integrity Testing. Adv. Biochem. Eng. Biotechnol. 98, 2006: 143–180; https://doi.org/10.1007/b104248.
6 Giglia A, et al. Air–Water Binary Gas Integrity Test for Sterilizing and Virus Filters. PDA J. Pharm. Sci. Technol. 70(6) 2016: 577–590; https://journal.pda.org/content/70/6/577.
7 Sekine, et al. Integrity Testing of Planova™ BioEX Virus Removal Filters and in the Manufacture of Biological Products. Biologicals 43(3) 2015: 186–194; https://doi.org/10.1016/j.biologicals.2015.02.003.
8 Giglia S. Air–Water Diffusion Integrity Test Sensitivity for Viresolve Pro: Effect of Multi-Device Assemblies (webinar). MilliporeSigma, 29 March 2021.
9 Dowd CJ. Multi-Round Virus Filter Integrity Test Sensitivity. Biotechnol. Bioeng. 103(3) 2009: 574–581; https://doi.org/10.1002/bit.22275.
10 Bolton G, et al. Integrity Testing of Normal Flow Parvovirus Filters Using Air–Liquid Based Tests. Bioprocess J. 5(1) 2006: 52–57; https://doi.org/10.12665/J51.Bolton.
Hyejee Jang is lead downstream scientist, Seonmi Kim is senior scientist of downstream processing in manufacturing science and technology (MSAT), Jihyun Lee is group lead of MSAT 2, Yonghwan Jin is vice president of MSAT, and corresponding author Lalit Saxena ([email protected]) is team lead for MSAT, all at Samsung Biologics, 300 Songdo bio-daero, Yeonsu-gu, Incheon 21987, Republic of Korea 82-32-455-3114; https;//samsungbiologics.com.
You May Also Like
