R
Internal Virus Polarization Model for Virus Retention by the UltiporV VF Grade DV20 Membrane Nigel B. Jackson Pall Europe Ltd., 5 Harbourgate Business Park, Portsmouth PO6 4BQ, U.K.
Meisam Bakhshayeshi and Andrew L. Zydney Dept. of Chemical Engineering, The Pennsylvania State University, University Park, PA 16802
Amit Mehta and Robert van Reis Genentech Inc., 1 DNA Way, South San Francisco, CA 94080
Ralf Kuriyel Pall Corporation, 20 Walkup Drive, Westborough, MA 01581 DOI 10.1002/btpr.1897 Published online in Wiley Online Library (wileyonlinelibrary.com)
Several recent studies have reported a decline in virus retention during virus challenge filtration experiments, although the mechanism(s) governing this phenomenon for different filters remains uncertain. Experiments were performed to evaluate the retention of PP7 and PR772 bacteriophage through Ultipor VF Grade DV20 virus filters during constant pressure filtration. While the larger PR772 phage was fully retained under all conditions, a 2-log decline in retention of the small PP7 phage was observed at high throughputs, even under conditions where there was no decline in filtrate flux. In addition, prefouling the membrane with an immunoglobulin G solution had no effect on phage retention. An internal polarization model was developed to describe the decline in phage retention arising from the accumulation of phage in the upper (reservoir) layer within the filter which increases the challenge to the lower (rejection) layer. Independent support for this internal polarization phenomenon was provided by confocal microscopy of fluorescently labeled phage within the membrane. The model was in good agreement with phage retention data over a wide range of phage titers, confirming that virus retention is throughput dependent and supporting current recommendations for virus retention validation studies. These results provide important insights into the factors governing virus retention by membrane filters and their dependence C 2014 American Institute of on the underlying structure of the virus filter membrane. V Chemical Engineers Biotechnol. Prog., 000:000–000, 2014 Keywords: virus filtration, antibody, bioprocessing, ultrafiltration, polarization
Introduction Virus filtration has become a well-established component of the overall viral clearance strategy applied during the production of both recombinant and plasma proteins.1 Virus filters can remove both small parvoviruses (size 20 nm) and larger retroviruses (80–120 nm), complementing both inactivation and adsorptive removal steps by providing robust size-based virus removal. Several recent studies have reported a decline in virus retention during the course of filtration through different virus filters.2–4 Lute et al.4 investigated the performance of several commercially available parvovirus filters by challenging the filters with small /X-174 and PP7 bacteriophages, both alone and in the presence of a therapeutic protein. All parvovirus filters provided high initial phage
Correspondence concerning this article should be addressed to N. B. Jackson at [email protected]. C 2014 American Institute of Chemical Engineers V
retention (6–7 orders of magnitude reduction in phage titer), but the extent of /X-174 and PP7 phage retention declined during the course of filtration, particularly under conditions of high phage loading. A decline in filtrate flux was also observed after high loading of phage. In some filters, the protein alone caused a significant decline in filtrate flux, while in others a cumulative bacteriophage load of more than 1014 pfu/m2 was required to demonstrate a significant flux decline. The authors divided the observed behavior into “protein-dominated” and “phage-dominated” regimes based on the level of the phage and protein challenge, with the nature of these regimes being specific to the different parvovirus filters. The underlying mechanisms controlling the loss of virus retention during virus filtration are still not well understood. Bolton et al.3 hypothesized that the decline in virus retention was due to preferential blockage of the smallest pores in the membrane, with the fluid flow redirected to the larger (nonretentive) pores. Experimental data obtained with Millipore 1
2
Biotechnol. Prog., 2014, Vol. 00, No. 00 R V
Viresolve NFP membranes seemed to collapse to a single curve when plotted as a function of the extent of flux decline, in good agreement with this physical model.3 This theory provided justification for the run and spike option of carrying out virus validation after prefouling membranes5 as an alternative approach to conventional virus retention validation procedures based on spiking over the full process throughput.6 However, data obtained with other virus filters are inconsistent with this physical model,4 suggesting that the mechanisms controlling this behavior may depend on the properties of the virus filter and/or the feed stream. In contrast to the decline in virus retention during protein filtration seen by Bolton et al.3 and Lute et al.,4 Khan et al.7 found that preconditioning the ViresolveV NFP membrane by filtering a 1 g/L solution of bovine serum albumin to a 60% flux decay caused an order of magnitude increase in retention of the /X-174 bacteriophage. This increase in phage retention was attributed to the formation of a cake layer on the surface of the filter, although the magnitude of this effect was also dependent on the protein concentration during the preconditioning step. Evidence for the importance of cake filtration in virus retention was previously demonstrated by Bolton et al.3 using 90 nm polystyrene beads as model aggregates to maintain a constant high log reduction value (LRV). Both fundamental understanding and practical implementation of virus filter performance continues to develop and build upon existing literature. Studies into the influence of pH8 and antibody concentration9 on virus filter performance have been combined and considered alongside further factors (ionic strength, protein purity and buffer composition) to optimize performance.10 The addition of temperature optimization and the incorporation of existing prefilter methods11 demonstrates the potential for further improvement in protein capacity12 where this can be practically implemented. One practical restriction is reproducing this optimal performance in the virus validation scale-down model. Recent studies have focused on demonstrating the robustness of some virus filters to virus spike impurities,13 improving the purification of the virus spike14 and developing novel methods to maximize prefilter performance during validation.15 Virus filter fouling testing under different membrane orientation and operating conditions16 was recently expanded to include a wider variety of membrane structures and virus retention data17 to suggest specific structures for improved virus filter design. The overall objective of this work was to develop a more fundamental understanding of the factors governing virus retention with the Ultipor VF Grade DV20 virus filter under typical virus loading. A new internal polarization model was developed to describe the observed decline in virus retention based on the accumulation of retained virus within an upper reservoir zone within the filter membrane. Independent support for this model was obtained using confocal microscopy to examine fluorescently labeled bacteriophage within the membrane. These results provide important insights into the factors governing virus retention and their dependence on the underlying structure of the virus filter membrane, and they also support current recommendations for virus filter validation. R
Materials and Methods Materials Experiments were performed with the Ultipor VF Grade DV20 virus removal filter in 47 mm disc format
(FTKDV20047; Pall, Port Washington, NY). The DV20 virus filter consists of two layers of a hydrophilic modified polyvinylidene fluoride membrane with a fairly homogenous (isotropic) pore structure. Virus filtration experiments were performed using direct flow (also known as normal flow or dead-end) filtration, with the DV20 membrane placed with the shiny-side down (away from the feed) in a stainless steel housing specifically validated for virus retention studies (FTK200; Pall). Gold colloid experiments were carried out in a stirred cell (Amicon 8010; Millipore, MA) using 20 nm gold colloidal particles (AGP-HA20; Asahi Kasei Medical, Japan) at a concentration of 6.2 3 1014 particles/mL. The concentration of dilute gold colloid suspensions was determined spectrophotometrically using a NanoDropTM 1000 UV–Vis spectrophotometer (Thermo Scientific, Wilmington, DE) with the absorbance measured at 520 nm based on the red color of the particle suspension. Additional details on the use of gold colloids are provided by Bakhshayeshirad.18 All bacteriophage and host organisms were obtained from the American Type Culture Collection. PP7 (model parvovirus) and PR772 (model retrovirus) stock solutions were generated by infecting an agar plate of the host organism (Pseudomonas aeruginosa and Escherichia coli, respectively) to achieve confluent lysis, corresponding to greater than 1,000 plaque forming units (pfu) per 90 mm plate. The top layer of the plates was removed and resuspended in tryptone soya broth (TSB). The phage were purified by centrifugation at 3,000 rpm for 10 min, decanting to recover the supernatant, then recentrifugation and filtration through a 0.2 mm rated sterilizing grade filter to remove contaminants. For the confocal microscopy experiments, the PP7 stock preparation was further purified by dialysis into pH 7.4 phosphate buffered saline using a 10 kDa nominal molecular weight cut-off membrane. The resulting phage suspensions were refrigerated until use. Fouling experiments were performed using human immunoglobulin (IgG) solutions prepared by dissolving human gamma globulin powder (HS-470; Seracare Life Sciences, CA) in a 200 mM acetate buffer at pH 5.5. The resulting solution was prefiltered through a 0.2 mm rated sterilizing grade filter to remove insoluble aggregates.
Virus filtration The volume of stock solution required to achieve the target phage load for each experiment was calculated based on the concentration of phage in the stock solution. The actual phage concentration in the feed was determined by a plaque forming assay at both the start and end of the experiment to ensure that there was no host contamination or significant loss of infectivity over the course of the experiment. A 1 mL sample was mixed with 4.5 mL of molten tryptone soya agar (TSA) and 0.5 mL of host cell suspension, P. aeruginosa for PP7 and E. coli for PR772. The resulting mixture was poured onto a TSA 90 mm plate and incubated for 12 h before counting the number of plaques. When necessary, the original phage samples were diluted 10-fold (serially) in TSB to obtain a countable number of plaques. Virus filtration was performed at a constant pressure of 210 kPa (30 psi). Filtrate flow rate was evaluated by timed collection. Except where otherwise stated, 10 mL aliquots of filtrate were collected with the phage concentration determined by the plaque forming assay. The LRV was calculated
Biotechnol. Prog., 2014, Vol. 00, No. 00
3 R
Figure 1.
Filtrate flux and LRV as a function of volumetric throughput during filtration of a 8.0 3 106 pfu/mL PP7 bacteriophage feed through an Ultipor VF Grade DV20 membrane.
labeled phage were then purified using a Bio-Rad Bio-GelV P-30 fine size exclusion resin followed by diafiltration as described by Bakhshayeshi et al.19 The DV20 membrane was challenged with a suspension of the fluorescently labeled PP7 phage for a desired volumetric throughput. The membrane was then removed from the holder, cut into small pieces of about 5 mm 3 5 mm, and mounted on a glass bottom dish (MatTek Corp., Ashland, MA). Confocal images were obtained using an Olympus FluoViewTM 1000 confocal scanning laser microscope (Olympus American). Excitation was performed with a 488nm blue laser directed through a pinhole aperture with the emitted fluorescent light collected through a separate pinhole aperture using a 1003 oil objective lens. The sample was scanned by automatically moving the objective lens. A three-dimensional fluorescent image of the filter was reconstructed by stacking the optical sections using the Olympus FluoViewTM viewer software (Olympus American). Additional details are provided by Bakhshayeshi et al.19
Results and Analysis LRV behavior
Figure 2.
Log reduction value (LRV) as a function of volumetric throughput for clean (0%) and 50% prefouled Ultipor VF Grade DV20 membranes during phage challenge by a protein-free solution containing 1.5 3 106 pfu/mL PP7 bacteriophage.
from the ratio of the phage concentration in the filtrate and feed as Cfiltrate LRV 52log10 Cfeed
(1)
where the feed concentration was evaluated as the average of the values determined at the start and end of the experiment. Confocal microscopy Virus retention within the Ultipor VF Grade DV20 membranes was also examined using confocal microscopy. The PP7 phage were labeled using an Alexa FluorV 488 protein labeling kit (Invitrogen-A10235, Eugene, OR) designed for conjugation to the lysine amine groups of IgG. 0.5 mL of a 2.6 3 1010 pfu/mL suspension of PP7 phage in 0.1 M bicarbonate buffer at pH 8.3 were mixed with one vial of the Alexa FluorV 488 dye and stirred overnight at 4 C. The R
R
Figure 1 shows the filtrate flux and LRV for the PP7 bacteriophage as a function of volumetric throughput through a DV20 membrane (total collected filtrate volume divided by the membrane cross-sectional area). Data were obtained for a feed solution containing 8.0 3 106 pfu/mL PP7 and 4.2 3 104 pfu/mL PR772 phage. The error bars represent the mean6 standard deviation for three repeat experiments under identical conditions; error bars were omitted when they were smaller than the size of the symbol. There were no detectable PR772 phage in any of the filtrate samples, demonstrating that there was no by-pass flow or defects in the DV20 membrane. The initial aliquot LRV for the PP7 phage was nearly 6, after which the LRV declined to approximately 4.3 after filtration of 18 L/m2 before approaching at very high throughput what appears to be an asymptote slightly greater than LRV 5 3. There was no measurable decline in filtrate flux over the course of these experiments, with the flux remaining between 20.8 and 23.5 L/(m2h). The 2-log decline in LRV, in the absence of any measurable flux decline, is inconsistent with the pore plugging model proposed by Bolton et al.3 The magnitude of LRV decline in Figure 1 is similar to that observed by Lute et al.4 for DV20 under typical phage loading with a therapeutic protein and minimal observed flux decay. Figure 2 shows the LRV behavior for a series of membranes challenged with 1.5 3 106 pfu/mL PP7 phage. The data from the control experiment (denoted as 0% fouling) are similar to the results in Figure 1. The other two membranes were first used to filter a 16 g/L IgG solution that contained no virus or phage. This IgG filtration was performed at a constant transmembrane pressure of 210 kPa (30 psi) until the flux had declined to approximately 50% of its initial value, corresponding to a mass throughput of 700 6 10 g of IgG/m2 of membrane area. The membrane was gently rinsed with acetate buffer and then installed into a clean stainless steel holder before being challenged with a protein-free solution of the PP7 bacteriophage. The results for these two experiments are plotted as the mean 6 standard deviation in the LRV. The LRV profiles during the phage challenge for these prefouled membranes were essentially
4
Figure 3. Confocal scanning image of the cross-section of a DV20 membrane after filtration of 11 L/m2 (top panel) and 44 L/m2 (bottom panel) of a suspension containing approximately 108 pfu/mL fluorescently labeled PP7 bacteriophage.
identical to that for the un-fouled membrane; the 50% decline in filtrate flux due to IgG fouling had no effect on either the initial LRV or the rate of LRV decline during the subsequent phage challenge. These data also demonstrate the need to perform virus spiking studies over the full process throughput, since the initial LRV after prefouling is unlikely to be representative of the true process LRV. The results in Figure 2 demonstrate that LRV for the DV20 membrane is strongly dependent on throughput of the virus challenge, in contrast to the strong dependence on flux decline seen by Bolton et al.3 The different behavior observed in these studies is likely due to differences in the morphology of the virus filters; the DV20 membrane is relatively homogeneous while the Viresolve NFP membranes examined by Bolton et al.3 are highly asymmetric with a very open structure (with micron-size pores) at the inlet region of the filter. It is also possible that the results are affected by differences in the proteins; Bolton et al.3 prefiltered their IgG through the Viresolve Prefilter media while the IgG used in this work was prefiltered only through a 0.2mm sterilizing grade filter. SEC analysis of the IgG used in this work showed no measurable aggregates >40 nm in size, suggesting that the protein fouling was not due to cake formation. Note that Khan et al.7 observed an increase in LRV with increasing throughput due to protein cake formation, which is exactly opposite to the behavior seen in this study.
Confocal microscopy Figure 3 shows a fluorescent image of a cross-section through a single layer of two DV20 membranes after filtra-
Biotechnol. Prog., 2014, Vol. 00, No. 00
tion of 11 L/m2 (top panel) and 44 L/m2 (bottom panel) of a suspension containing approximately 108 pfu/mL of the fluorescently labeled PP7 bacteriophage. These cross-sectional views were constructed from 54 in-focus images of the x–y planes inside the membrane. These images were stacked along the z-axis and then sliced through an arbitrary x–z plane, approximately 1 mm thick, using the Olympus FluoViewTM viewer software. The fluorescently labeled PP7 bacteriophage are easily visible throughout the upper region of the DV20 membrane, extending approximately one-quarter of the way through the approximately 40 mm thick membrane. The depth of penetration was very similar at the two loadings, with the image at 44 L/m2 showing a much greater fluorescent intensity. Additional details on the confocal microscopy are provided by Bakhshayeshi et al.19 Independent experiments were performed to verify that the majority of filtered bacteriophage actually enter the DV20 membrane. 100 mL of a solution containing a mixture of 2.8 3 107 pfu/mL PP7 and 2.5 3 104 pfu/mL PR772 bacteriophages were filtered through a DV20 membrane. The filtrate was collected in four samples, each containing approximately 24 mL, with approximately 4 mL of retentate left in the housing above the filter. No PR772 phage were observed in the filtrate samples and less than 0.01% of the PP7 phage passed through the membrane. The residual retentate was collected by draining the housing, with any remaining phage collected by circulating 200 mL of acetate buffer through the filter headspace upstream of the membrane at a flow rate of approximately 40 mL/min. Approximately 20 6 15% of the PP7 phage were recovered in the residual retentate plus buffer rinse, with the remainder being retained within the filter (consistent with the confocal image in Figure 3). In contrast, 75 6 25% of the large PR772 bacteriophage were recovered in the residual retentate plus the buffer rinse. The large error bars on the calculated values of the percent recovery reflect the inherent uncertainties in the plaque assay. Internal polarization model The confocal images in Figure 3 suggest that the retained bacteriophage accumulate within the upper region of the membrane, which we refer to as the “reservoir zone” within the filter. This accumulation of retained virus would lead to a reduction in LRV as the “rejection zone” of the membrane is challenged by a continually increasing concentration of virus from the reservoir zone. As with classical membrane processes, the polarization occurs when the virus is selectively retained by the membrane so that the concentration increases near the surface of the rejection layer. In contrast to standard concentration polarization theory, rejection in internal polarization occurs within the membrane after convective transfer of viruses through the reservoir zone. The thickness of this internal concentration polarization layer is thus determined by the physical dimensions of the reservoir zone as opposed to the balance between convection and diffusion in a concentration polarization boundary layer. The slightly asymmetric structure of the Ultipor VF Grade DV20 membrane, with a more open pore size in the reservoir zone and a tighter pore size near the filter exit, has been observed previously. In contrast to scanning electron microscopy cross-sections showing relative homogeneity, dextran retention tests performed with the membrane oriented with the shiny-side up yielded contrasting results to tests with the shiny-side down.20 Further demonstration of this asymmetry
Biotechnol. Prog., 2014, Vol. 00, No. 00
5
was provided by filtration of the 20 nm gold colloid nanoparticles.18 When the DV20 membrane was challenged in the normal orientation (shiny-side down), all of the gold particles were retained within the membrane, with no measureable increase in the concentration of gold particles in the bulk solution in the stirred cell. In contrast, there was a several-fold increase in the bulk particle concentration when the orientation was reversed (shiny-side up) since most of the particles were unable to enter the DV20 membrane in this orientation. A simple mathematical model for this internal polarization phenomenon was developed as follows. The virus rejection zone of the membrane was assumed to have a constant virus sieving coefficient, S, which is equal to the fractional transmission of the virus through the rejection zone. The reservoir zone of the membrane is assumed to have a constant volume (VR) and a uniform virus concentration (C), where C refers to the concentration of “free” (mobile) virus in the reservoir zone. The transient mass balance for the concentration of free virus in the reservoir zone is thus VR
dC 5xqCF 2qSC dt
(2)
where q is the volumetric flow rate, CF is the virus concentration in the challenge (feed) solution, and x is the fraction of viruses that enter the reservoir zone that remain in free solution. Thus, x 5 1 would imply that all of the viruses concentrate in the reservoir zone while x 5 0 would correspond to a situation in which all of the viruses are retained by size exclusion in retentive pores within the matrix of the reservoir zone and are no longer “free”. It is assumed that x remains constant over the phage loadings in this study as the total capacity for retained virus remains in excess and essentially constant throughout the filtration. Equation 2 can be directly integrated assuming that the reservoir initially contains no virus, yielding C x SV 5 12exp 2 CF S VR
(3)
where V is the cumulative filtrate volume. Equation 3 predicts that the virus concentration in the filtrate solution of a single layer membrane (Cfiltrate 5 S C) increases from zero at the start of filtration to xCF as the volumetric throughput approaches infinity. The LRV can be evaluated directly from Eq. 3 giving
SC V 52log ðxSÞ2log LRV 52log CF VR
(4)
where the second expression is developed using the first term in the Taylor series expansion for the exponential and is thus valid when SV VR 1. This is equivalent to neglecting virus transport out of the reservoir zone, i.e., assuming that the second term on the right-hand side of Eq. 2 is negligible. Equation 4 suggests that the LRV is determined by the properties of the membrane and the total volumetric throughput (i.e., V/VR), independent of the virus titer or the flux decay. The term 2log(xS) can be thought of as the intrinsic virus retention characteristic of the membrane; it is directly related to the fraction of the virus that remain in free solution within the reservoir zone of the filter and to the sieving coefficient of the rejection zone. Equation 4 is only valid for a single layer membrane. The LRV for the two-layer DV20 membrane is developed in an
analogous fashion, with the transient mass balance for the reservoir zone in the second layer given as VR
dC2 5xqSC1 2qSC2 dt
(5)
where C1 and C2 are the concentrations of free virus in the reservoir zones in Layers 1 and 2, respectively. Equation 5 can be solved analytically to give 2 C 2 x2 SV x V SV 2 5 exp 2 12exp 2 VR VR VR CF S
(6)
The LRV for the two-layer membrane is given as SC2 V ½22 logðxSÞ1logð2Þ22 log LRV 52log VR CF
(7)
where the final expression is again developed using a Taylor series expansion for the exponentials. The LRV of an n-layer membrane (n 2) was developed by extending this formulation to a membrane with n separate layers giving " # i21 # " X n SV 1 SV SV n n LRV 52log x 12exp ð2 Þ 2 x exp ð2 Þ VR ði21Þ! VR VR i52
(8)
Equation 8 can again be simplified for small values of SV/ VR giving LRV 5½2n logðxSÞ1logðn!Þ2n log
V VR
(9)
The LRV for an individual aliquot was calculated by integration of the expression for the virus concentration in the filtrate solution, SC2 with C2 given by Eq. 6, over the volume of the aliquot (from V1 to V) 0
1 2 V 2 2VDV1 DV 3 A LRV DV 5½22 logðxSÞ1logð2Þ2log@ VR2
(10)
where DV 5 V 2 V1. Equation 10 is valid for a two-layer membrane at small values of SV/VR; it reduces to Eq. 7 when DV V as expected. The experimental data in Figure 1 along with data for experiments using both lower (2.7 3 106 pfu/mL) and higher (23 3 106 pfu/mL) phage concentrations are shown in Figure 4 as an explicit function of the volumetric throughput. These concentrations were selected to cover an order of magnitude difference in phage concentration at typical values as defined by the methods for nomenclature ratings for virus filtration.6 The data for the different phage challenges collapse to a single curve when plotted in this manner, consistent with the form given by Eq. 7 or 9. There was again no measureable flux decline in any of these experiments. The solid curve in Figure 4 represents the internal polarization model given by Eq. 10; the results using the analytical and approximate solutions were essentially identical over the volumetric throughput range examined in Figure 4. The volume of the reservoir zone was evaluated as VR/A 5 5 mm 5 0.005 L/m2 based on the 10 mm thickness of the phage accumulation layer determined from the confocal image in Figure 3 and an assumed membrane porosity of 50%. The product of the fraction of the phage remaining in free
6
Biotechnol. Prog., 2014, Vol. 00, No. 00
Filter flushing
Figure 4. LRV for PP7 bacteriophage as a function of volumetric throughput during filtration of phage solutions having different concentrations through Ultipor VF Grade DV20 membranes. Solid curve shows calculated values of the LRV given by the internal polarization model (Eq. 10).
Two additional sets of experiments were performed to provide additional validation of the internal virus polarization model. First, an Ultipor VF Grade DV20 membrane was challenged with 43 mL of a feed containing 1.6 3 107 pfu/mL of the PP7 bacteriophage. This membrane was then extensively flushed with phosphate buffered saline, with the flow in the reverse direction, to recover any “mobile” phage from within the filter. This reverse flushing yielded a total of 2.8 3 108 pfu, which is 41% of the initial phage challenge (although it should be recognized that the error bars on this phage recovery are quite large due to the inherent variability in the plaque assay). These data cannot be used to obtain an accurate estimate for x due to potential phage release and also subsequent retention that can occur within the filtration matrix when the flow is reversed. However, the results do show that a substantial proportion of phage is not permanently retained by size exclusion or adsorbed to the pore surface and therefore could be concentrated within the reservoir zone. A second set of experiments was performed in which the DV20 was challenged with 48 L/m2 of a feed containing approximately 7 3 107 pfu/mL PP7. At this point the feed was replaced with a pure acetate buffer (containing no phage), and the filtration was continued at a constant transmembrane pressure of 30 psi. Permeate samples were taken throughout the buffer flush, with the results for two repeat experiments shown in Figure 5. The phage concentration in the permeate solution initially increases compared to the final aliquot during the phage challenge, and then either slowly declines or remains relatively constant throughout the buffer flush. Variation from the final phage aliquot is
Internal Virus Polarization Model for Virus Retention by the UltiporV VF Grade DV20 Membrane Nigel B. Jackson Pall Europe Ltd., 5 Harbourgate Business Park, Portsmouth PO6 4BQ, U.K.
Meisam Bakhshayeshi and Andrew L. Zydney Dept. of Chemical Engineering, The Pennsylvania State University, University Park, PA 16802
Amit Mehta and Robert van Reis Genentech Inc., 1 DNA Way, South San Francisco, CA 94080
Ralf Kuriyel Pall Corporation, 20 Walkup Drive, Westborough, MA 01581 DOI 10.1002/btpr.1897 Published online in Wiley Online Library (wileyonlinelibrary.com)
Several recent studies have reported a decline in virus retention during virus challenge filtration experiments, although the mechanism(s) governing this phenomenon for different filters remains uncertain. Experiments were performed to evaluate the retention of PP7 and PR772 bacteriophage through Ultipor VF Grade DV20 virus filters during constant pressure filtration. While the larger PR772 phage was fully retained under all conditions, a 2-log decline in retention of the small PP7 phage was observed at high throughputs, even under conditions where there was no decline in filtrate flux. In addition, prefouling the membrane with an immunoglobulin G solution had no effect on phage retention. An internal polarization model was developed to describe the decline in phage retention arising from the accumulation of phage in the upper (reservoir) layer within the filter which increases the challenge to the lower (rejection) layer. Independent support for this internal polarization phenomenon was provided by confocal microscopy of fluorescently labeled phage within the membrane. The model was in good agreement with phage retention data over a wide range of phage titers, confirming that virus retention is throughput dependent and supporting current recommendations for virus retention validation studies. These results provide important insights into the factors governing virus retention by membrane filters and their dependence C 2014 American Institute of on the underlying structure of the virus filter membrane. V Chemical Engineers Biotechnol. Prog., 000:000–000, 2014 Keywords: virus filtration, antibody, bioprocessing, ultrafiltration, polarization
Introduction Virus filtration has become a well-established component of the overall viral clearance strategy applied during the production of both recombinant and plasma proteins.1 Virus filters can remove both small parvoviruses (size 20 nm) and larger retroviruses (80–120 nm), complementing both inactivation and adsorptive removal steps by providing robust size-based virus removal. Several recent studies have reported a decline in virus retention during the course of filtration through different virus filters.2–4 Lute et al.4 investigated the performance of several commercially available parvovirus filters by challenging the filters with small /X-174 and PP7 bacteriophages, both alone and in the presence of a therapeutic protein. All parvovirus filters provided high initial phage
Correspondence concerning this article should be addressed to N. B. Jackson at [email protected]. C 2014 American Institute of Chemical Engineers V
retention (6–7 orders of magnitude reduction in phage titer), but the extent of /X-174 and PP7 phage retention declined during the course of filtration, particularly under conditions of high phage loading. A decline in filtrate flux was also observed after high loading of phage. In some filters, the protein alone caused a significant decline in filtrate flux, while in others a cumulative bacteriophage load of more than 1014 pfu/m2 was required to demonstrate a significant flux decline. The authors divided the observed behavior into “protein-dominated” and “phage-dominated” regimes based on the level of the phage and protein challenge, with the nature of these regimes being specific to the different parvovirus filters. The underlying mechanisms controlling the loss of virus retention during virus filtration are still not well understood. Bolton et al.3 hypothesized that the decline in virus retention was due to preferential blockage of the smallest pores in the membrane, with the fluid flow redirected to the larger (nonretentive) pores. Experimental data obtained with Millipore 1
2
Biotechnol. Prog., 2014, Vol. 00, No. 00 R V
Viresolve NFP membranes seemed to collapse to a single curve when plotted as a function of the extent of flux decline, in good agreement with this physical model.3 This theory provided justification for the run and spike option of carrying out virus validation after prefouling membranes5 as an alternative approach to conventional virus retention validation procedures based on spiking over the full process throughput.6 However, data obtained with other virus filters are inconsistent with this physical model,4 suggesting that the mechanisms controlling this behavior may depend on the properties of the virus filter and/or the feed stream. In contrast to the decline in virus retention during protein filtration seen by Bolton et al.3 and Lute et al.,4 Khan et al.7 found that preconditioning the ViresolveV NFP membrane by filtering a 1 g/L solution of bovine serum albumin to a 60% flux decay caused an order of magnitude increase in retention of the /X-174 bacteriophage. This increase in phage retention was attributed to the formation of a cake layer on the surface of the filter, although the magnitude of this effect was also dependent on the protein concentration during the preconditioning step. Evidence for the importance of cake filtration in virus retention was previously demonstrated by Bolton et al.3 using 90 nm polystyrene beads as model aggregates to maintain a constant high log reduction value (LRV). Both fundamental understanding and practical implementation of virus filter performance continues to develop and build upon existing literature. Studies into the influence of pH8 and antibody concentration9 on virus filter performance have been combined and considered alongside further factors (ionic strength, protein purity and buffer composition) to optimize performance.10 The addition of temperature optimization and the incorporation of existing prefilter methods11 demonstrates the potential for further improvement in protein capacity12 where this can be practically implemented. One practical restriction is reproducing this optimal performance in the virus validation scale-down model. Recent studies have focused on demonstrating the robustness of some virus filters to virus spike impurities,13 improving the purification of the virus spike14 and developing novel methods to maximize prefilter performance during validation.15 Virus filter fouling testing under different membrane orientation and operating conditions16 was recently expanded to include a wider variety of membrane structures and virus retention data17 to suggest specific structures for improved virus filter design. The overall objective of this work was to develop a more fundamental understanding of the factors governing virus retention with the Ultipor VF Grade DV20 virus filter under typical virus loading. A new internal polarization model was developed to describe the observed decline in virus retention based on the accumulation of retained virus within an upper reservoir zone within the filter membrane. Independent support for this model was obtained using confocal microscopy to examine fluorescently labeled bacteriophage within the membrane. These results provide important insights into the factors governing virus retention and their dependence on the underlying structure of the virus filter membrane, and they also support current recommendations for virus filter validation. R
Materials and Methods Materials Experiments were performed with the Ultipor VF Grade DV20 virus removal filter in 47 mm disc format
(FTKDV20047; Pall, Port Washington, NY). The DV20 virus filter consists of two layers of a hydrophilic modified polyvinylidene fluoride membrane with a fairly homogenous (isotropic) pore structure. Virus filtration experiments were performed using direct flow (also known as normal flow or dead-end) filtration, with the DV20 membrane placed with the shiny-side down (away from the feed) in a stainless steel housing specifically validated for virus retention studies (FTK200; Pall). Gold colloid experiments were carried out in a stirred cell (Amicon 8010; Millipore, MA) using 20 nm gold colloidal particles (AGP-HA20; Asahi Kasei Medical, Japan) at a concentration of 6.2 3 1014 particles/mL. The concentration of dilute gold colloid suspensions was determined spectrophotometrically using a NanoDropTM 1000 UV–Vis spectrophotometer (Thermo Scientific, Wilmington, DE) with the absorbance measured at 520 nm based on the red color of the particle suspension. Additional details on the use of gold colloids are provided by Bakhshayeshirad.18 All bacteriophage and host organisms were obtained from the American Type Culture Collection. PP7 (model parvovirus) and PR772 (model retrovirus) stock solutions were generated by infecting an agar plate of the host organism (Pseudomonas aeruginosa and Escherichia coli, respectively) to achieve confluent lysis, corresponding to greater than 1,000 plaque forming units (pfu) per 90 mm plate. The top layer of the plates was removed and resuspended in tryptone soya broth (TSB). The phage were purified by centrifugation at 3,000 rpm for 10 min, decanting to recover the supernatant, then recentrifugation and filtration through a 0.2 mm rated sterilizing grade filter to remove contaminants. For the confocal microscopy experiments, the PP7 stock preparation was further purified by dialysis into pH 7.4 phosphate buffered saline using a 10 kDa nominal molecular weight cut-off membrane. The resulting phage suspensions were refrigerated until use. Fouling experiments were performed using human immunoglobulin (IgG) solutions prepared by dissolving human gamma globulin powder (HS-470; Seracare Life Sciences, CA) in a 200 mM acetate buffer at pH 5.5. The resulting solution was prefiltered through a 0.2 mm rated sterilizing grade filter to remove insoluble aggregates.
Virus filtration The volume of stock solution required to achieve the target phage load for each experiment was calculated based on the concentration of phage in the stock solution. The actual phage concentration in the feed was determined by a plaque forming assay at both the start and end of the experiment to ensure that there was no host contamination or significant loss of infectivity over the course of the experiment. A 1 mL sample was mixed with 4.5 mL of molten tryptone soya agar (TSA) and 0.5 mL of host cell suspension, P. aeruginosa for PP7 and E. coli for PR772. The resulting mixture was poured onto a TSA 90 mm plate and incubated for 12 h before counting the number of plaques. When necessary, the original phage samples were diluted 10-fold (serially) in TSB to obtain a countable number of plaques. Virus filtration was performed at a constant pressure of 210 kPa (30 psi). Filtrate flow rate was evaluated by timed collection. Except where otherwise stated, 10 mL aliquots of filtrate were collected with the phage concentration determined by the plaque forming assay. The LRV was calculated
Biotechnol. Prog., 2014, Vol. 00, No. 00
3 R
Figure 1.
Filtrate flux and LRV as a function of volumetric throughput during filtration of a 8.0 3 106 pfu/mL PP7 bacteriophage feed through an Ultipor VF Grade DV20 membrane.
labeled phage were then purified using a Bio-Rad Bio-GelV P-30 fine size exclusion resin followed by diafiltration as described by Bakhshayeshi et al.19 The DV20 membrane was challenged with a suspension of the fluorescently labeled PP7 phage for a desired volumetric throughput. The membrane was then removed from the holder, cut into small pieces of about 5 mm 3 5 mm, and mounted on a glass bottom dish (MatTek Corp., Ashland, MA). Confocal images were obtained using an Olympus FluoViewTM 1000 confocal scanning laser microscope (Olympus American). Excitation was performed with a 488nm blue laser directed through a pinhole aperture with the emitted fluorescent light collected through a separate pinhole aperture using a 1003 oil objective lens. The sample was scanned by automatically moving the objective lens. A three-dimensional fluorescent image of the filter was reconstructed by stacking the optical sections using the Olympus FluoViewTM viewer software (Olympus American). Additional details are provided by Bakhshayeshi et al.19
Results and Analysis LRV behavior
Figure 2.
Log reduction value (LRV) as a function of volumetric throughput for clean (0%) and 50% prefouled Ultipor VF Grade DV20 membranes during phage challenge by a protein-free solution containing 1.5 3 106 pfu/mL PP7 bacteriophage.
from the ratio of the phage concentration in the filtrate and feed as Cfiltrate LRV 52log10 Cfeed
(1)
where the feed concentration was evaluated as the average of the values determined at the start and end of the experiment. Confocal microscopy Virus retention within the Ultipor VF Grade DV20 membranes was also examined using confocal microscopy. The PP7 phage were labeled using an Alexa FluorV 488 protein labeling kit (Invitrogen-A10235, Eugene, OR) designed for conjugation to the lysine amine groups of IgG. 0.5 mL of a 2.6 3 1010 pfu/mL suspension of PP7 phage in 0.1 M bicarbonate buffer at pH 8.3 were mixed with one vial of the Alexa FluorV 488 dye and stirred overnight at 4 C. The R
R
Figure 1 shows the filtrate flux and LRV for the PP7 bacteriophage as a function of volumetric throughput through a DV20 membrane (total collected filtrate volume divided by the membrane cross-sectional area). Data were obtained for a feed solution containing 8.0 3 106 pfu/mL PP7 and 4.2 3 104 pfu/mL PR772 phage. The error bars represent the mean6 standard deviation for three repeat experiments under identical conditions; error bars were omitted when they were smaller than the size of the symbol. There were no detectable PR772 phage in any of the filtrate samples, demonstrating that there was no by-pass flow or defects in the DV20 membrane. The initial aliquot LRV for the PP7 phage was nearly 6, after which the LRV declined to approximately 4.3 after filtration of 18 L/m2 before approaching at very high throughput what appears to be an asymptote slightly greater than LRV 5 3. There was no measurable decline in filtrate flux over the course of these experiments, with the flux remaining between 20.8 and 23.5 L/(m2h). The 2-log decline in LRV, in the absence of any measurable flux decline, is inconsistent with the pore plugging model proposed by Bolton et al.3 The magnitude of LRV decline in Figure 1 is similar to that observed by Lute et al.4 for DV20 under typical phage loading with a therapeutic protein and minimal observed flux decay. Figure 2 shows the LRV behavior for a series of membranes challenged with 1.5 3 106 pfu/mL PP7 phage. The data from the control experiment (denoted as 0% fouling) are similar to the results in Figure 1. The other two membranes were first used to filter a 16 g/L IgG solution that contained no virus or phage. This IgG filtration was performed at a constant transmembrane pressure of 210 kPa (30 psi) until the flux had declined to approximately 50% of its initial value, corresponding to a mass throughput of 700 6 10 g of IgG/m2 of membrane area. The membrane was gently rinsed with acetate buffer and then installed into a clean stainless steel holder before being challenged with a protein-free solution of the PP7 bacteriophage. The results for these two experiments are plotted as the mean 6 standard deviation in the LRV. The LRV profiles during the phage challenge for these prefouled membranes were essentially
4
Figure 3. Confocal scanning image of the cross-section of a DV20 membrane after filtration of 11 L/m2 (top panel) and 44 L/m2 (bottom panel) of a suspension containing approximately 108 pfu/mL fluorescently labeled PP7 bacteriophage.
identical to that for the un-fouled membrane; the 50% decline in filtrate flux due to IgG fouling had no effect on either the initial LRV or the rate of LRV decline during the subsequent phage challenge. These data also demonstrate the need to perform virus spiking studies over the full process throughput, since the initial LRV after prefouling is unlikely to be representative of the true process LRV. The results in Figure 2 demonstrate that LRV for the DV20 membrane is strongly dependent on throughput of the virus challenge, in contrast to the strong dependence on flux decline seen by Bolton et al.3 The different behavior observed in these studies is likely due to differences in the morphology of the virus filters; the DV20 membrane is relatively homogeneous while the Viresolve NFP membranes examined by Bolton et al.3 are highly asymmetric with a very open structure (with micron-size pores) at the inlet region of the filter. It is also possible that the results are affected by differences in the proteins; Bolton et al.3 prefiltered their IgG through the Viresolve Prefilter media while the IgG used in this work was prefiltered only through a 0.2mm sterilizing grade filter. SEC analysis of the IgG used in this work showed no measurable aggregates >40 nm in size, suggesting that the protein fouling was not due to cake formation. Note that Khan et al.7 observed an increase in LRV with increasing throughput due to protein cake formation, which is exactly opposite to the behavior seen in this study.
Confocal microscopy Figure 3 shows a fluorescent image of a cross-section through a single layer of two DV20 membranes after filtra-
Biotechnol. Prog., 2014, Vol. 00, No. 00
tion of 11 L/m2 (top panel) and 44 L/m2 (bottom panel) of a suspension containing approximately 108 pfu/mL of the fluorescently labeled PP7 bacteriophage. These cross-sectional views were constructed from 54 in-focus images of the x–y planes inside the membrane. These images were stacked along the z-axis and then sliced through an arbitrary x–z plane, approximately 1 mm thick, using the Olympus FluoViewTM viewer software. The fluorescently labeled PP7 bacteriophage are easily visible throughout the upper region of the DV20 membrane, extending approximately one-quarter of the way through the approximately 40 mm thick membrane. The depth of penetration was very similar at the two loadings, with the image at 44 L/m2 showing a much greater fluorescent intensity. Additional details on the confocal microscopy are provided by Bakhshayeshi et al.19 Independent experiments were performed to verify that the majority of filtered bacteriophage actually enter the DV20 membrane. 100 mL of a solution containing a mixture of 2.8 3 107 pfu/mL PP7 and 2.5 3 104 pfu/mL PR772 bacteriophages were filtered through a DV20 membrane. The filtrate was collected in four samples, each containing approximately 24 mL, with approximately 4 mL of retentate left in the housing above the filter. No PR772 phage were observed in the filtrate samples and less than 0.01% of the PP7 phage passed through the membrane. The residual retentate was collected by draining the housing, with any remaining phage collected by circulating 200 mL of acetate buffer through the filter headspace upstream of the membrane at a flow rate of approximately 40 mL/min. Approximately 20 6 15% of the PP7 phage were recovered in the residual retentate plus buffer rinse, with the remainder being retained within the filter (consistent with the confocal image in Figure 3). In contrast, 75 6 25% of the large PR772 bacteriophage were recovered in the residual retentate plus the buffer rinse. The large error bars on the calculated values of the percent recovery reflect the inherent uncertainties in the plaque assay. Internal polarization model The confocal images in Figure 3 suggest that the retained bacteriophage accumulate within the upper region of the membrane, which we refer to as the “reservoir zone” within the filter. This accumulation of retained virus would lead to a reduction in LRV as the “rejection zone” of the membrane is challenged by a continually increasing concentration of virus from the reservoir zone. As with classical membrane processes, the polarization occurs when the virus is selectively retained by the membrane so that the concentration increases near the surface of the rejection layer. In contrast to standard concentration polarization theory, rejection in internal polarization occurs within the membrane after convective transfer of viruses through the reservoir zone. The thickness of this internal concentration polarization layer is thus determined by the physical dimensions of the reservoir zone as opposed to the balance between convection and diffusion in a concentration polarization boundary layer. The slightly asymmetric structure of the Ultipor VF Grade DV20 membrane, with a more open pore size in the reservoir zone and a tighter pore size near the filter exit, has been observed previously. In contrast to scanning electron microscopy cross-sections showing relative homogeneity, dextran retention tests performed with the membrane oriented with the shiny-side up yielded contrasting results to tests with the shiny-side down.20 Further demonstration of this asymmetry
Biotechnol. Prog., 2014, Vol. 00, No. 00
5
was provided by filtration of the 20 nm gold colloid nanoparticles.18 When the DV20 membrane was challenged in the normal orientation (shiny-side down), all of the gold particles were retained within the membrane, with no measureable increase in the concentration of gold particles in the bulk solution in the stirred cell. In contrast, there was a several-fold increase in the bulk particle concentration when the orientation was reversed (shiny-side up) since most of the particles were unable to enter the DV20 membrane in this orientation. A simple mathematical model for this internal polarization phenomenon was developed as follows. The virus rejection zone of the membrane was assumed to have a constant virus sieving coefficient, S, which is equal to the fractional transmission of the virus through the rejection zone. The reservoir zone of the membrane is assumed to have a constant volume (VR) and a uniform virus concentration (C), where C refers to the concentration of “free” (mobile) virus in the reservoir zone. The transient mass balance for the concentration of free virus in the reservoir zone is thus VR
dC 5xqCF 2qSC dt
(2)
where q is the volumetric flow rate, CF is the virus concentration in the challenge (feed) solution, and x is the fraction of viruses that enter the reservoir zone that remain in free solution. Thus, x 5 1 would imply that all of the viruses concentrate in the reservoir zone while x 5 0 would correspond to a situation in which all of the viruses are retained by size exclusion in retentive pores within the matrix of the reservoir zone and are no longer “free”. It is assumed that x remains constant over the phage loadings in this study as the total capacity for retained virus remains in excess and essentially constant throughout the filtration. Equation 2 can be directly integrated assuming that the reservoir initially contains no virus, yielding C x SV 5 12exp 2 CF S VR
(3)
where V is the cumulative filtrate volume. Equation 3 predicts that the virus concentration in the filtrate solution of a single layer membrane (Cfiltrate 5 S C) increases from zero at the start of filtration to xCF as the volumetric throughput approaches infinity. The LRV can be evaluated directly from Eq. 3 giving
SC V 52log ðxSÞ2log LRV 52log CF VR
(4)
where the second expression is developed using the first term in the Taylor series expansion for the exponential and is thus valid when SV VR 1. This is equivalent to neglecting virus transport out of the reservoir zone, i.e., assuming that the second term on the right-hand side of Eq. 2 is negligible. Equation 4 suggests that the LRV is determined by the properties of the membrane and the total volumetric throughput (i.e., V/VR), independent of the virus titer or the flux decay. The term 2log(xS) can be thought of as the intrinsic virus retention characteristic of the membrane; it is directly related to the fraction of the virus that remain in free solution within the reservoir zone of the filter and to the sieving coefficient of the rejection zone. Equation 4 is only valid for a single layer membrane. The LRV for the two-layer DV20 membrane is developed in an
analogous fashion, with the transient mass balance for the reservoir zone in the second layer given as VR
dC2 5xqSC1 2qSC2 dt
(5)
where C1 and C2 are the concentrations of free virus in the reservoir zones in Layers 1 and 2, respectively. Equation 5 can be solved analytically to give 2 C 2 x2 SV x V SV 2 5 exp 2 12exp 2 VR VR VR CF S
(6)
The LRV for the two-layer membrane is given as SC2 V ½22 logðxSÞ1logð2Þ22 log LRV 52log VR CF
(7)
where the final expression is again developed using a Taylor series expansion for the exponentials. The LRV of an n-layer membrane (n 2) was developed by extending this formulation to a membrane with n separate layers giving " # i21 # " X n SV 1 SV SV n n LRV 52log x 12exp ð2 Þ 2 x exp ð2 Þ VR ði21Þ! VR VR i52
(8)
Equation 8 can again be simplified for small values of SV/ VR giving LRV 5½2n logðxSÞ1logðn!Þ2n log
V VR
(9)
The LRV for an individual aliquot was calculated by integration of the expression for the virus concentration in the filtrate solution, SC2 with C2 given by Eq. 6, over the volume of the aliquot (from V1 to V) 0
1 2 V 2 2VDV1 DV 3 A LRV DV 5½22 logðxSÞ1logð2Þ2log@ VR2
(10)
where DV 5 V 2 V1. Equation 10 is valid for a two-layer membrane at small values of SV/VR; it reduces to Eq. 7 when DV V as expected. The experimental data in Figure 1 along with data for experiments using both lower (2.7 3 106 pfu/mL) and higher (23 3 106 pfu/mL) phage concentrations are shown in Figure 4 as an explicit function of the volumetric throughput. These concentrations were selected to cover an order of magnitude difference in phage concentration at typical values as defined by the methods for nomenclature ratings for virus filtration.6 The data for the different phage challenges collapse to a single curve when plotted in this manner, consistent with the form given by Eq. 7 or 9. There was again no measureable flux decline in any of these experiments. The solid curve in Figure 4 represents the internal polarization model given by Eq. 10; the results using the analytical and approximate solutions were essentially identical over the volumetric throughput range examined in Figure 4. The volume of the reservoir zone was evaluated as VR/A 5 5 mm 5 0.005 L/m2 based on the 10 mm thickness of the phage accumulation layer determined from the confocal image in Figure 3 and an assumed membrane porosity of 50%. The product of the fraction of the phage remaining in free
6
Biotechnol. Prog., 2014, Vol. 00, No. 00
Filter flushing
Figure 4. LRV for PP7 bacteriophage as a function of volumetric throughput during filtration of phage solutions having different concentrations through Ultipor VF Grade DV20 membranes. Solid curve shows calculated values of the LRV given by the internal polarization model (Eq. 10).
Two additional sets of experiments were performed to provide additional validation of the internal virus polarization model. First, an Ultipor VF Grade DV20 membrane was challenged with 43 mL of a feed containing 1.6 3 107 pfu/mL of the PP7 bacteriophage. This membrane was then extensively flushed with phosphate buffered saline, with the flow in the reverse direction, to recover any “mobile” phage from within the filter. This reverse flushing yielded a total of 2.8 3 108 pfu, which is 41% of the initial phage challenge (although it should be recognized that the error bars on this phage recovery are quite large due to the inherent variability in the plaque assay). These data cannot be used to obtain an accurate estimate for x due to potential phage release and also subsequent retention that can occur within the filtration matrix when the flow is reversed. However, the results do show that a substantial proportion of phage is not permanently retained by size exclusion or adsorbed to the pore surface and therefore could be concentrated within the reservoir zone. A second set of experiments was performed in which the DV20 was challenged with 48 L/m2 of a feed containing approximately 7 3 107 pfu/mL PP7. At this point the feed was replaced with a pure acetate buffer (containing no phage), and the filtration was continued at a constant transmembrane pressure of 30 psi. Permeate samples were taken throughout the buffer flush, with the results for two repeat experiments shown in Figure 5. The phage concentration in the permeate solution initially increases compared to the final aliquot during the phage challenge, and then either slowly declines or remains relatively constant throughout the buffer flush. Variation from the final phage aliquot is
