- Email: [email protected]
Quantitative analysis of internal flow distribution and pore interconnectivity within asymmetric virus filtration membranes
Journal of Membrane Science xxx (xxxx) xxx
Contents lists available at ScienceDirect
Journal of Membrane Science journal homepage: http://www.elsevier.com/locate/memsci
Quantitative analysis of internal flow distribution and pore interconnectivity within asymmetric virus filtration membranes Fatemeh Fallahianbijan a, Sal Giglia b, Christina Carbrello b, Andrew L. Zydney a, * a b
Department of Chemical Engineering, The Pennsylvania State University, University Park, PA, 16802, USA MilliporeSigma, Bedford, MA, 01730, USA
A R T I C L E I N F O
A B S T R A C T
Keywords: Membrane morphology Pore interconnectivity Lateral flow Asymmetric membrane Electron microscopy
Several studies have demonstrated that the filtration performance and fouling characteristics of porous mem branes can be significantly influenced by the pore interconnectivity. However, there are no available techniques that can quantify the pore connectivity of highly asymmetric membranes with small pore size used in ultrafil tration and virus removal filtration. In this study, a novel approach was developed to measure the pore inter connectivity from SEM images of gold nanoparticles captured within a membrane in which flow through the exit (skin side) was partially blocked by a stainless steel support. The pore interconnectivity parameter was then evaluated by comparison of the observed capture profile with numerical simulations of the flow and particle capture. Results for the Viresolve® Pro membranes showed much greater pore interconnectivity than for the Viresolve® NFP membranes. SEM images of the Ultipor® DV20 membrane showed nanoparticle capture only at regions of the inlet located directly over the open portions of the membrane exit, indicating that there is minimal lateral flow in this membrane. These results provide the first quantitative measurements of the extent of pore interconnectivity within virus filtration membranes having highly asymmetric pore structures.
1. Introduction The effects of membrane pore size, density, and length (membrane thickness) on the performance characteristics of ultrafiltration and microfiltration membranes are very well-established [1]. However, the pore interconnectivity can also play a major role in membrane systems. This is clearly true in “lateral flow” membranes used for immunoassays in which solutes have to diffuse laterally through the membrane, i.e., in a direction perpendicular to the traditional direction of filtration [2]. In addition, the rate of membrane fouling is a strong function of the membrane pore interconnectivity. For example, Ho and Zydney [3–5] showed that fouling of membranes with straight-through, non- interconnected, pores was much more rapid than membranes with highly interconnected pores, since the pore interconnectivity allows fluid to flow around and under any pore blockage on the membrane surface. Ho and Zydney [5] developed a model for the rate of flux decline accounting for this pore interconnectivity based on the solution of Laplace’s equation for the local pressure. Li et al. [6] subsequently extended the model to describe the fouling behavior of asymmetric membranes with arbitrary values chosen for the pore interconnectivity parameter in the skin and macroporous support.
There is also anecdotal evidence that the membrane pore inter connectivity plays a significant role in determining the virus retention characteristics of virus filtration membranes [7–9]. For example, Woods and Zydney [10] hypothesized that the increase in virus transmission after a process disruption during virus filtration through the Ultipor® DV20 membrane was due to lateral diffusion of previously captured virus within the porous structure of the membrane. Yamamato et al. [11] hypothesized that this type of lateral diffusion caused the observed reduction in virus clearance for the Planova virus filters when operated at low transmembrane pressure. However, neither of these studies pro vided any direct measure of the magnitude of the lateral diffusivity or the extent of pore interconnectivity. Ho and Zydney [12] developed a simple approach for evaluating the pore interconnectivity of homogeneous microporous membranes by measuring the water flow rate, or diffusive solute flux, through a membrane in which the upper and lower surfaces were partially blocked by “overlapping” tape. Thus, a membrane with straight-through, non- interconnected, pores would have zero flow/diffusion, while mem branes with highly interconnected pores would show significant transport. The rate of flow (or diffusion) was then used to evaluate the ratio of the hydraulic (or diffusive) permeabilities in the directions
* Corresponding author. E-mail address: [email protected] (A.L. Zydney). https://doi.org/10.1016/j.memsci.2019.117578 Received 14 August 2019; Received in revised form 16 October 2019; Accepted 16 October 2019 Available online 18 October 2019 0376-7388/© 2019 Elsevier B.V. All rights reserved.
Please cite this article as: Fatemeh Fallahianbijan, Journal of Membrane Science, https://doi.org/10.1016/j.memsci.2019.117578
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
parallel to and normal to the membrane surface based on a mathemat ical solution for the velocity/pressure profiles. However, this technique cannot be applied to composite or asymmetric membranes in which the permeability varies with position through the depth of the membrane. Tanis-Kanbur et al. [13] developed a modified evapoporometry method that can be used to characterize the pore size distribution of both continuous (pores that go entirely through the membrane) and dead-end pores, with the ratio of the continuous to total pore volumes assumed to provide information on the pore connectivity. Data obtained with polyethersulfone and polyvinylidene fluoride membranes indicate that the continuous pores have a smaller average pore size than the dead-end pores, with both membranes showing a high degree of pore connectivity. However, this technique only provides values for the pores in the membrane skin layer, with no information on the variation of the pore connectivity through the depth of the membrane. Several groups have used synchrotron radiation [14], X-ray or electron [15] tomography for the characterization of the 3-dimensional membrane pore structure. However, these methods have not yet been successfully applied to the evaluation of the pore interconnectivity in the relatively small (<50 nm) pores of virus filtration membranes. The objective of this study was to develop a new approach for measuring the pore interconnectivity based on direct observation of gold nanoparticles captured within a porous membrane in which the mem brane exit (skin) surface was partially blocked. Experiments were per formed using Viresolve® Pro, Viresolve® NFP, and Ultipor® DV20 membranes using a mixture of different size gold nanoparticles, with the captured nanoparticles imaged by scanning electron microscopy (SEM) following the procedures described by Nazem-Bokaee et al. [8]. The extent of pore interconnectivity was then evaluated by comparing the observed nanoparticle capture with the calculated flow streamlines and predicted capture profiles determined by numerical analysis of the flow and pressure profiles based on Darcy’s law expressions in the directions normal to and parallel to the membrane surface.
stainless steel screen with a circular pattern of voids was used as a support; this screen blocked approximately 60% of the outlet surface of the membrane. Experiments were conducted in normal flow filtration mode at a constant pressure of 210 kPa (30 psig), which was maintained by nitrogen pressurization of the feed reservoir. The membranes were initially flushed with at least 40 L/m2 of deionized distilled water ob tained from a Direct-Q® 3 UV Water Purification System (Milli poreSigma, Burlington, MA) to ensure complete wetting of the membrane and to remove any trapped air bubbles in the system. Gold nanoparticles (20, 40, 100, and 200 nm) were obtained from MilliporeSigma (Burlington, MA) as stabilized suspensions in citrate buffer. Nanoparticle suspensions were diluted with DI water and ultra sonicated for a short time to insure complete dispersion before being used to challenge the filter. At the end of the filtration run, the filter was removed from the stainless steel holder, rinsed gently with DI water, and then prepared for electron microscopy as discussed below. 2.3. Scanning electron microscopy (SEM)
2. Methods and materials
The location of the captured nanoparticles was examined by SEM using previously developed protocols [9]. Briefly, the membranes were cut into small strips (10 � 3 mm), dehydrated using progressively more concentrated ethanol solutions, infiltrated with LR White medium grade embedding resin (Electron Microscopy Sciences, Hartfield, PA), and transferred to a clean snap-fit gelatin capsule (Ted Pella. Inc., Rodding, CA). Approximately 1 μm slices were obtained using a Leica EM UC6 Ultramicrotome (Leica Biosystems Inc., Buffalo Grove, IL) and coated with a thin layer of iridium using an Emitech K575X sputter coater (Quorum Technologies Ltd, UK). The membrane cross-section was examined at high-magnification using a Zeiss Sigma VP-FESEM (Carl Zeiss Microscopy, Thornwood, NY) equipped with a retractable five-diode backscattering detector. The locations of the individual nanoparticles were determined using ImageJ 1.50I processing software (http://imagej.nih.gov/ij).
2.1. Membranes
3. Theoretical modeling
Experiments were performed using commercially available flat sheet virus filtration membranes: highly asymmetric Viresolve® Pro and Viresolve® NFP membranes were provided by MilliporeSigma (Bedford, MA) and the relatively homogeneous Ultipor® DV20 membrane was from Pall Corp. (Port Washington, NY). The Viresolve® Pro membrane is a surface-modified polyethersulfone (PES) while the Viresolve® NFP and DV20 membranes are both surface-modified polyvinylidene difluoride (PVDF) (Table 1). These flat sheet membranes were cut into 47 mm disks from large sheet-stock and used in single layer format; the actual commercial modules employ multiple layers of membrane to provide the high levels of viral clearance needed in bioprocessing.
The flow distribution within the membranes was evaluated by nu merical solution of the Darcy flow equations: Vr ¼
1 ∂ ∂Vz ðrVr Þ þ ¼0 r ∂r ∂z
kr ∂ ∂P ∂2 P ðr Þ þ kz ¼0 r ∂r ∂r ∂z
Thickness (μm)
Manufacturer
Viresolve® Pro Viresolve® NFP Ultipor® DV20
Polyether-sulfone (PES) Polyvinylidene fluoride (PVDF) Polyvinylidene fluoride (PVDF)
Asymmetric
140
MilliporeSigma
Asymmetric
140
MilliporeSigma
Symmetric
40
Pall
(2)
(3)
Equation (3) was solved numerically using the finite element analysis in COMSOL Multiphysics software (version 5.3) for the system geometry shown in Fig. 1, which is identical to that examined by Li et al. [6]. The dark region is one of the open spaces in the stainless steel support, with the surrounding space representing the blocked region; the grid is defined for the entire membrane above both the blocked and open re gions. No flow boundary conditions were applied over the region covered by the stainless steel support, with symmetry conditions applied at the pore axis and the outer boundary of the cylindrical region. The pressures at the upper and lower surfaces of the membrane were assumed to be uniform. A detailed description of the boundary equations
Table 1 Properties of virus filtration membranes. Asymmetry
(1)
Substitution of Eq. (1) into Eq. (2) yields the following partial dif ferential equation for the local pressure:
Membranes were placed in a 47 mm stainless steel filter holder (MilliporeSigma, Burlington, MA), with the retentive layer (shiny side) facing downstream as per the manufacturer’s recommendation. A
Material
∂P ∂P ; V ¼ kz ∂r z ∂z
where kr and kz are the Darcy permeabilities in the directions parallel to and perpendicular to the membrane surface, respectively. The velocities Vr and Vz must satisfy the continuity equation for an incompressible fluid which is given by:
2.2. Nanoparticle filtration
Membrane
kr
2
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
local pore size first became equal to the particle size; the pore size profile for the Viresolve® Pro and Viresolve® NFP membranes were evaluated by SEM image analysis as presented in a previous publication [16]. A total of 100,000 nanoparticles were placed at the entrance of the filter (z ¼ 140 μm) and distributed based on the magnitude of the local flow velocity as determined by solution of Equation (3) assuming that the nanoparticles have no affect on the flow. 4. Results 4.1. Viresolve® pro membrane Fig. 2 shows the lower surface of the Viresolve® Pro membrane after filtration of 15 L/m2 of a suspension of 20 nm gold nanoparticles at a constant pressure of 210 kPa (30 psig). The membrane was used with the shiny-side (retentive layer) placed on a stainless steel screen (left panels) or on a microporous support (right panels). In both cases, the filtrate flux remained essentially constant (within 5%) throughout the filtration, consistent with the very dilute suspensions of gold nanoparticles (1010 nanoparticles/mL) used in these experiments. The dark red color shows the regions filled with captured gold nanoparticles, with the regular array of red dots in the left hand panels representing the regions that were unblocked by the stainless steel support, consistent with the model geometry shown in Fig. 1. The membrane that was placed on the microporous support shows an essentially uniform red color due to the high porosity and small pore spacing in the support. Fig. 3 shows an SEM image of a cross-section through the Viresolve® Pro membrane, focusing on the region immediately above the boundary between the open and blocked regions of the stainless steel support. The membrane was challenged at a constant pressure of 210 kPa (30 psig) with a suspension containing a mixture of the 20, 40, 100, and 200 nm gold nanoparticles (with mean size of 25 � 8 nm, 42 � 12 nm, 100 � 30 nm, and 200 � 45, respectively, as determined by dynamic light scattering). The white dots are the individual gold nanoparticles;
Fig. 1. Schematic of lower surface of partially blocked membrane. Dark circle represents the open region at the filter exit.
is available in the literature [12]. The governing equations were solved numerically for different values of the pore interconnectivity, defined as K ¼ kkzr , where K ¼ 0
corresponds to a membrane with non-interconnected pores (no lateral flow) while K ¼ 1 corresponds to a membrane with an isotropic struc ture with equal resistance to flow in both directions. A value of K ¼ 0.1 indicates that the velocity in the lateral direction would be one-tenth that in the normal direction for equivalent pressure gradients. The predicted distribution of captured nanoparticles within the membrane was evaluated as follows. The membrane was “challenged” with a feed containing a distribution of nanoparticles (particle radius described by a simple Gaussian function). Each nanoparticle was released at a given location and assumed to follow the flow streamline until it reached a point within the depth of the membrane where the
Fig. 2. Optical images of the lower surface of the Viresolve® Pro membrane after challenge with 15 L/m2 of a suspension of 20 nm gold nanoparticles. Left panels are low and high magnification images of a membrane on the stainless steel support. Right panels are corresponding images of a membrane on a microporous support. 3
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
Fig. 3. High contrast SEM image of the cross-section of a Viresolve® Pro membrane (filter exit at the top surface) after challenging with 60 L/m2 of a sus pension containing 20, 40, 100 and 200 nm gold nanoparticles at 210 kPa with the skin side placed on top of the stainless steel screen spacer (top panel). The distribution of captured nanoparticles (shown by red circles) within the Viresolve® Pro membrane as determined by ImageJ software is shown in the bottom panel. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)
energy dispersive X-ray spectroscopy was used to confirm that these dots were elemental gold as discussed elsewhere [8]. There was no evidence of any nanoparticle aggregation, either in the feed suspension (as determined by dynamic light scattering) or within the membrane (as seen in the SEM images). The SEM image shows distinct “bands” for the different size nanoparticles, with the 20 nm particles captured right near the filter exit (top surface) while the 200 nm nanoparticles are captured around 25 μm further into the filter. More strikingly, the bands for the larger particles are spread out into the space above the blocked region of the membrane, with the “span” for the band increasing as the size of the nanoparticle increases, i.e., as one moves further away from the mem brane exit. This behavior is shown more clearly in the bottom panel of Fig. 3 in which the location of each nanoparticle, as determined by the ImageJ software, is shown by a red circle (irrespective of the particle size). The net result is that the left-hand edge of the particle bands generates a boundary (shown by the red curve, which is simply drawn by eye) that defines the capture profile within the Viresolve® Pro mem brane; the shape of this curve is determined by the extent of lateral flow within this region of the membrane. The calculated values of the flow streamlines within the Viresolve® Pro membrane were evaluated numerically using COMSOL Multiphysics software for different values of the pore connectivity: K ¼ 1, 0.1, and 0.01, with results shown in Fig. 4. At the smallest value of K, the streamlines are directed almost perpendicular to the feed/permeate
surfaces of the membrane, similar to what would be expected for flow through a membrane with straight-through non-interconnected pores, with essentially no bulk flow through the regions upstream of the blocked surface. As K increases, the streamlines begin to move out into the portion of the membrane above the blocked region due to the in crease in the lateral flow through the porous membrane (similar to the profile for the gold nanoparticles seen in Fig. 3). For K ¼ 1, there is a measurable flow over almost the entire entrance to the membrane even though more than 60% of the exit region is blocked by the stainless steel screen. A more quantitative analysis of the pore interconnectivity in the Viresolve® Pro membrane was performed by simulating the nano particle capture in MATLAB using the velocities calculated in COMSOL for specific values of K. The membranes were “challenged” with a dis tribution of particles, in this case defined by a mixture of particles with mean size of 20, 40, 100, and 200 nm having standard deviations of 4, 6, 10, and 15 nm, respectively. Each particle was captured at a location (zvalue) at which the local membrane pore size was equal to the size of the nanoparticle. The resulting simulation results are overlaid on the flow streamlines in Fig. 4. As K increases, the nanoparticles are captured in a more diffuse band spreading out over the blocked region of the mem brane as expected. The best fit value of K for the Viresolve® Pro membrane was esti mated by comparison of the experimental results for nanoparticle 4
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
Fig. 4. Internal flow streamlines within the Viresolve® Pro membrane (oriented with the filter exit at the top surface) for (A) K ¼ 1, (B) K ¼ 0.1, and (C) K ¼ 0.01. Shaded regions show calculated locations of captured nanoparticles corresponding to each value of the pore interconnectivity.
capture with the calculated profiles shown in Fig. 4. An example is shown in Fig. 5. The red circles are the positions of the individual gold nanoparticles (irrespective of particle size) determined using ImageJ software, while the gray curves are the flow streamlines for K ¼ 1. The shape of the calculated streamlines, and the simulated nanoparticle capture bands, are well-aligned with the particle locations, suggesting that K ¼ 1 provides a reasonable estimate of the pore interconnectivity for the Viresolve® Pro membrane. Note that the calculated streamlines do not define the regions with and without flow; the fluid velocity is not actually zero except at the exit of the blocked region (and there are streamlines in the regions to the left of the streamlines shown in Fig. 5). However, the flow streamlines for simulations using either smaller, or
larger, values of K show clear discrepancies with the shape of the experimentally-determined nanoparticle capture profiles. 4.2. Viresolve® NFP membrane Corresponding results for the Viresolve® NFP membrane are shown in Figs. 6 and 7. The top panel of Fig. 6 shows an SEM image after filtration of 60 L/m2 of a mixture of the 20, 40, 100, and 200 nm gold particles at a constant pressure of 210 kPa (30 psi). The 20 nm gold particles are captured very close to the membrane exit (skin layer) over the open region of the stainless steel support. The 40 and 100 nm par ticles were captured in bands that are mostly above the open region but
Fig. 5. Comparison of captured nanoparticle locations within the Viresolve® Pro membrane (determined by ImageJ software) and the calculated flow streamlines from solution of Equation (3) using K ¼ 1. Membrane exit is at upper surface. Particle capture is only shown over 50 μm region near filter exit – total membrane thickness is approximately 140 μm. 5
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
Fig. 6. High contrast SEM image of the cross-section of a Viresolve® NFP membrane with filter exit at top surface after challenging with 60 L/m2 of a suspen sion containing 20, 40, 100 and 200 nm gold nano particles at 210 kPa with the skin side placed on top of the stainless steel support (top panel). The distri bution of captured nanoparticles (in green circles) within the Viresolve® NFP membrane near the retentive layer as determined by ImageJ analysis (bottom panel). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)
do extend slightly into the space above the blocked region. In contrast, the 200 nm particles are scattered throughout the depth of the mem brane over both the open and blocked flow regions, reflecting the nearly uniform pore size in the microporous support on which the Viresolve® NFP membrane is cast. The bottom panel of Fig. 6 shows the distribution of captured nanoparticles obtained by ImageJ analysis of the SEM image in the top panel; the software is able to identify the individual nanoparticles much more effectively by properly adjusting the threshold limits. Again, the nanoparticles are “aligned” with the open region on the membrane based on the location of the “last” 20 nm particle visible in the SEM image. The furthest most nanoparticle seen in the ImageJ output is located 12 μm to the left of the open region, reflecting the small amount of lateral flow within the Viresolve® NFP membrane. Fig. 7 shows the calculated flow streamlines in the Viresolve® NFP membrane for K ¼ 1, 0.1, 0.01, and 0.001. As expected, the flow bends out further into the space above the blocked region as K increases, with the streamlines for K ¼ 0.001 remaining almost entirely over the open region of the membrane. The flow streamlines for K ¼ 0.01, and the corresponding nanoparticle capture simulations, are in fairly good agreement with the experimental results in Fig. 6. This indicates that the
pore interconnectivity for the Viresolve® NFP membrane is approxi mately K ¼ 0.01, which is two orders of magnitude smaller than the value determined for the Viresolve® Pro membrane. 4.3. Ultipor® DV20 membrane In contrast to the Viresolve® Pro and Viresolve® NFP membranes, the Ultipor® DV20 membrane has a fairly uniform pore size throughout the depth of the 40 μm thick membrane; dextran sieving data [18] and fluorescent nanoparticle capture profiles [7] show only a small decrease in pore size as one moves through the depth of the Ultipor® DV20 membrane. Fig. 8 shows an SEM image of a cross-section through a Ultipor® DV20 membrane after challenging with the nanoparticle mixture following the same procedures as used for the Viresolve® Pro and Viresolve® NFP membranes. In contrast to the results with the Viresolve® membranes, all of the nanoparticles were captured near the filter inlet, directly opposite the blockage of the exit surface by the stainless steel support, with the 20 and 40 nm particles penetrating only a few microns into the depth of the membrane. The capture pattern on the upper surface of the Ultipor® DV20 membrane mirrors the circular pattern of open spaces defined by the stainless steel support located 6
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
Fig. 7. Internal flow streamlines within the Viresolve® NFP membranes for (A) K ¼ 1, (B) K ¼ 0.1, (C) K ¼ 0.01, (D) K ¼ 0.001. Filter exit located at top surface.
to fully wet the pores, with the permeability evaluated as 1.1 � 10 10 m based on the flow rate at 210 kPa (30 psi). The top and bottom surfaces of the membrane were then partially covered with tape, with the two taped sections overlapping by approximately 1 mm. The membrane was then placed in the base of a stirred cell and re-pressurized to 210 kPa (30 psi). No measurable filtrate flux was observed after 180 min of filtration, corresponding to a permeability of less than 2 � 10 16 m based on the approximate volume of one drop of water. This low permeability cor responds to K < 10 3 based on the solution of Equation (3), in good agreement with the results obtained from the gold nanoparticles, providing further confirmation of this methodology. 5. Discussion The work described in this study presents a new method for evalu ating the internal flow distribution and pore interconnectivity in asymmetric membranes based on the capture profiles for different size gold nanoparticles as visualized by scanning electron microscopy. The membranes were used with the skin-side placed directly on top of a stainless steel spacer that blocked a significant fraction of the pores in the membrane skin (filter exit), forcing the fluid to flow through the open (unblocked) regions. The captured nanoparticles in the Viresolve® Pro membrane were clearly visible in the region above the open region of the membrane, with the nanoparticle “bands” spreading out further into this region as one moves away from the filter exit, providing a picture of the flow streamlines within the interior of the membrane. ImageJ analysis of the gold nanoparticles in the Viresolve® Pro mem brane were in good agreement with the calculated values of the flow streamlines and capture profile evaluated for a membrane with a permeability ratio of K ¼ kkzr ¼ 1, consistent with a high degree of lateral
Fig. 8. High contrast SEM image of the cross-section of an Ultipor® DV20 membrane challenged with 60 L/m2 of a suspension containing 20, 40, 100 and 200 nm gold nanoparticles at 210 kPa showing nanoparticle capture near the filter entrance (lower surface).
beneath the membrane (at the membrane exit), consistent with the absence of any significant pore interconnectivity in the DV20 mem brane. Model simulations for the flow streamlines and nanoparticle capture suggest that K < 10 3 for the Ultipor® DV20 membrane; larger values of K show significantly more diffuse particle capture than that shown in the SEM image in Fig. 8. In order to confirm the very low degree of lateral flow within the DV20 membrane, an independent measure of the pore interconnectivity was obtained following the general approach described by Ho and Zydney [12]. An Ultipor® DV20 membrane was first flushed with water 7
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
flow within the membrane. In contrast, the nanoparticle capture profiles in the Viresolve® NFP membrane were best fit using K ¼ 0.01, indicating that the permeability in the direction normal to the membrane surface is much larger than that in the transverse (lateral) direction. Experiments performed with the Ultipor® DV20 membrane showed particles captured on the upper (inlet) surface of the membrane, but only over the regions of the membrane that were open on the downstream (exit) surface. This clearly indicates that there is minimal lateral flow within the Ultipor® DV20 membrane. This very low degree of pore interconnectivity was confirmed using the “overlapping tape” method described previously by Ho and Zydney [12] for symmetric (homoge neous) microfiltration membranes, with no measurable permeability (flow) when the tape on the upper and lower surfaces of the membrane overlap. This provides additional validation of the gold nanoparticle methodology developed in this study. However, the “overlapping tape” method cannot be used with highly asymmetric virus filtration mem branes since the high degree of pore interconnectivity in the membrane support structure will obscure the effects of the interconnectivity in/near the membrane skin, which is the region of interest in under standing the performance characteristics of these asymmetric mem branes. Note that the approach developed in this work could easily be extended to evaluate the pore interconnectivity of asymmetric ultrafil tration or microfiltration membranes by adjusting the size of the gold nanoparticles to properly map out the flow through the depth of these membranes. Interestingly, the devices containing Viresolve® NFP and Ultipor® DV20 membranes, both of which have very low pore interconnectivities as determined from the gold nanoparticle capture data, show a signifi cant loss in virus retention after a process disruption [10,17]. In contrast, the Viresolve® Pro devices show highly robust virus retention, both following a process disruption and over a broad range of filtration conditions [17]. Additional studies will be required to identify the
relationship between the pore interconnectivity and the virus retention characteristics of these virus removal membranes. 6. Conclusions This paper demonstrates a novel methodology for evaluating the pore interconnectivity of highly asymmetric membranes using SEM to visualize gold nanoparticles captured within a membrane in which flow through the exit (skin side) was partially blocked, in this case by a stainless steel support. The method was validated using data for the Ultipor® DV20 with very low pore interconnectivity and then extended to provide the first measurements of the pore interconnectivity for the Viresolve® Pro and Viresolve® NFP virus filtration membranes. The data clearly demonstrate that there are significant differences in the pore interconnectivity for these membranes, all of which have very similar retention characteristics. The approach can be readily extended to other asymmetric membranes by proper choice of the gold nanoparticle size, providing a general methodology for evaluating the pore inter connectivity in both ultrafiltration and microfiltration membranes. Declaration of competing interest Two of the authors (SG and CC) are employed by MilliporeSigma, which provided some of the membranes examined in this study. Acknowledgements The authors would like to acknowledge MilliporeSigma for their financial support, including donation of the Viresolve® Pro and Vir esolve® NFP membranes. The authors would also like to thank Farzad Mohajerani for the thoughtful discussion and assistance in performing the numerical simulations.
Nomenclature Notation δm D K k L P Pfeed Ppermeate V
Membrane thickness (m) Open region diameter (m) Permeability ratio (-) Darcy permeability in membrane (m3 s/kg) Distance between open regions (m) Local hydrostatic pressure (N/m2) Feed pressure (N/m2) Filtrate pressure (N/m2) Fluid velocity (m/s)
Subscripts r direction parallel to membrane surface z direction normal to membrane surface
References
[5] C.C. Ho, A.L. Zydney, Theoretical analysis of the effect of membrane morphology on fouling during microfiltration, Separ. Sci. Technol. 34 (1999) 2461–2483, https://doi.org/10.1081/SS-100100785. [6] W. Li, C. Duclos-Orsello, C.-C. Ho, Theoretical analysis of the effects of asymmetric membrane structure on fouling during microfiltration, AIChE J. 55 (2009) 857–866, https://doi.org/10.1002/aic.11763. [7] F. Fallahianbijan, S. Giglia, C. Carbrello, A.L. Zydney, Use of fluorescently-labeled nanoparticles to study pore morphology and virus capture in virus filtration membranes, J. Membr. Sci. 536 (2017) 52–58, https://doi.org/10.1016/j. memsci.2017.04.066. [8] H. Nazem-Bokaee, F. Fallahianbijan, D. Chen, S.M. O’Donnell, C. Carbrello, S. Giglia, D. Bell, A.L. Zydney, Probing pore structure of virus filters using scanning electron microscopy with gold nanoparticles, J. Membr. Sci. 552 (2018) 144–152, https://doi.org/10.1016/j.memsci.2018.01.069. [9] M. Bakhshayeshi, N. Jackson, R. Kuriyel, A. Mehta, R. van Reis, A.L. Zydney, Use of confocal scanning laser microscopy to study virus retention during virus filtration,
[1] W.M. Deen, Hindered transport of large molecules in liquid-filled pores, AIChE J. 33 (1987) 1409–1425, https://doi.org/10.1002/aic.690330902. [2] A.L. Ahmad, S.C. Low, S.R.A. Shukor, W.J.N. Fernando, A. Ismail, Hindered diffusion in lateral flow nitrocellulose membrane: experimental and modeling studies, J. Membr. Sci. 357 (2010) 178–184, https://doi.org/10.1016/j. memsci.2010.04.018. [3] C.C. Ho, A.L. Zydney, Protein fouling of asymmetric and composite microfiltration membranes, Ind. Eng. Chem. Res. 40 (2001) 1412–1421, https://doi.org/10.1021/ ie000810j. [4] C.C. Ho, A.L. Zydney, Effect of membrane morphology on the initial rate of protein fouling during microfiltration, J. Membr. Sci. 155 (1999) 261–275, https://doi. org/10.1016/S0376-7388(98)00324-X.
8
F. Fallahianbijan et al.
[10] [11] [12] [13]
[14]
Journal of Membrane Science xxx (xxxx) xxx
J. Membr. Sci. 379 (2011) 260–267, https://doi.org/10.1016/j. memsci.2011.05.069. M.A. Woods, A.L. Zydney, Effects of a pressure release on virus retention with the Ultipor DV20 membrane, Biotechnol. Bioeng. 111 (2014) 545–551, https://doi. org/10.1002/bit.25112. A. Yamamoto, T. Hongo-Hirasaki, Y. Uchi, H. Hayashida, F. Nagoya, Effect of hydrodynamic forces on virus removal capability of PlanovaTM filters, AIChE J. 60 (2014) 2286–2297, https://doi.org/10.1002/aic. C.C. Ho, A.L. Zydney, Measurement of membrane pore interconnectivity, J. Membr. Sci. 170 (2000) 101–112, https://doi.org/10.1016/S0376-7388(99) 00360-9. M.B. Tanis-Kanbur, F. Zamani, W.B. Krantz, X. Hu, J.W. Chew, Adaptation of evapoporometry (EP) to characterize the continuous pores and interpore connectivity in polymeric membranes, J. Membr. Sci. 575 (2019) 17–27, https:// doi.org/10.1016/j.memsci.2018.12.050. J.C. Remigy, M. Meireles, X. Thibault, Morphological characterization of a polymeric microfiltration membrane by synchrotron radiation computed
[15] [16]
[17] [18]
9
microtomography, J. Membr. Sci. 305 (2007) 27–35, https://doi.org/10.1016/j. memsci.2007.06.059. K.-L. Tung, K.-S. Chang, T.-T. Wu, N.-J. Lin, K.-R. Lee, J.-Y. Lai, Recent advances in the characterization of membrane morphology, Curr. Opin. Chem. Eng. 4 (2014) 121–127, https://doi.org/10.1016/j.coche.2014.03.002. F. Fallahianbijan, S. Giglia, C. Carbrello, D. Bell, A.L. Zydney, Impact of protein fouling on nanoparticle capture within the Viresolve® Pro and Viresolve® NFP virus removal membranes, Biotechnol. Bioeng. (2019), https://doi.org/10.1002/ bit.27017. S.K. Dishari, A. Venkiteshwaran, A.L. Zydney, Probing effects of pressure release on virus capture during virus filtration using confocal microscopy, Biotechnol. Bioeng. 112 (2015) 2115–2122, https://doi.org/10.1002/bit.25614. M. Bakhshayeshi, D.M. Kanani, A. Mehta, R. van Reis, R. Kuriyel, N. Jackson, A. L. Zydney, Dextran sieving test for characterization of virus filtration membranes, J. Membr. Sci. 379 (2011) 239–248, https://doi.org/10.1016/j. memsci.2011.05.067.
Contents lists available at ScienceDirect
Journal of Membrane Science journal homepage: http://www.elsevier.com/locate/memsci
Quantitative analysis of internal flow distribution and pore interconnectivity within asymmetric virus filtration membranes Fatemeh Fallahianbijan a, Sal Giglia b, Christina Carbrello b, Andrew L. Zydney a, * a b
Department of Chemical Engineering, The Pennsylvania State University, University Park, PA, 16802, USA MilliporeSigma, Bedford, MA, 01730, USA
A R T I C L E I N F O
A B S T R A C T
Keywords: Membrane morphology Pore interconnectivity Lateral flow Asymmetric membrane Electron microscopy
Several studies have demonstrated that the filtration performance and fouling characteristics of porous mem branes can be significantly influenced by the pore interconnectivity. However, there are no available techniques that can quantify the pore connectivity of highly asymmetric membranes with small pore size used in ultrafil tration and virus removal filtration. In this study, a novel approach was developed to measure the pore inter connectivity from SEM images of gold nanoparticles captured within a membrane in which flow through the exit (skin side) was partially blocked by a stainless steel support. The pore interconnectivity parameter was then evaluated by comparison of the observed capture profile with numerical simulations of the flow and particle capture. Results for the Viresolve® Pro membranes showed much greater pore interconnectivity than for the Viresolve® NFP membranes. SEM images of the Ultipor® DV20 membrane showed nanoparticle capture only at regions of the inlet located directly over the open portions of the membrane exit, indicating that there is minimal lateral flow in this membrane. These results provide the first quantitative measurements of the extent of pore interconnectivity within virus filtration membranes having highly asymmetric pore structures.
1. Introduction The effects of membrane pore size, density, and length (membrane thickness) on the performance characteristics of ultrafiltration and microfiltration membranes are very well-established [1]. However, the pore interconnectivity can also play a major role in membrane systems. This is clearly true in “lateral flow” membranes used for immunoassays in which solutes have to diffuse laterally through the membrane, i.e., in a direction perpendicular to the traditional direction of filtration [2]. In addition, the rate of membrane fouling is a strong function of the membrane pore interconnectivity. For example, Ho and Zydney [3–5] showed that fouling of membranes with straight-through, non- interconnected, pores was much more rapid than membranes with highly interconnected pores, since the pore interconnectivity allows fluid to flow around and under any pore blockage on the membrane surface. Ho and Zydney [5] developed a model for the rate of flux decline accounting for this pore interconnectivity based on the solution of Laplace’s equation for the local pressure. Li et al. [6] subsequently extended the model to describe the fouling behavior of asymmetric membranes with arbitrary values chosen for the pore interconnectivity parameter in the skin and macroporous support.
There is also anecdotal evidence that the membrane pore inter connectivity plays a significant role in determining the virus retention characteristics of virus filtration membranes [7–9]. For example, Woods and Zydney [10] hypothesized that the increase in virus transmission after a process disruption during virus filtration through the Ultipor® DV20 membrane was due to lateral diffusion of previously captured virus within the porous structure of the membrane. Yamamato et al. [11] hypothesized that this type of lateral diffusion caused the observed reduction in virus clearance for the Planova virus filters when operated at low transmembrane pressure. However, neither of these studies pro vided any direct measure of the magnitude of the lateral diffusivity or the extent of pore interconnectivity. Ho and Zydney [12] developed a simple approach for evaluating the pore interconnectivity of homogeneous microporous membranes by measuring the water flow rate, or diffusive solute flux, through a membrane in which the upper and lower surfaces were partially blocked by “overlapping” tape. Thus, a membrane with straight-through, non- interconnected, pores would have zero flow/diffusion, while mem branes with highly interconnected pores would show significant transport. The rate of flow (or diffusion) was then used to evaluate the ratio of the hydraulic (or diffusive) permeabilities in the directions
* Corresponding author. E-mail address: [email protected] (A.L. Zydney). https://doi.org/10.1016/j.memsci.2019.117578 Received 14 August 2019; Received in revised form 16 October 2019; Accepted 16 October 2019 Available online 18 October 2019 0376-7388/© 2019 Elsevier B.V. All rights reserved.
Please cite this article as: Fatemeh Fallahianbijan, Journal of Membrane Science, https://doi.org/10.1016/j.memsci.2019.117578
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
parallel to and normal to the membrane surface based on a mathemat ical solution for the velocity/pressure profiles. However, this technique cannot be applied to composite or asymmetric membranes in which the permeability varies with position through the depth of the membrane. Tanis-Kanbur et al. [13] developed a modified evapoporometry method that can be used to characterize the pore size distribution of both continuous (pores that go entirely through the membrane) and dead-end pores, with the ratio of the continuous to total pore volumes assumed to provide information on the pore connectivity. Data obtained with polyethersulfone and polyvinylidene fluoride membranes indicate that the continuous pores have a smaller average pore size than the dead-end pores, with both membranes showing a high degree of pore connectivity. However, this technique only provides values for the pores in the membrane skin layer, with no information on the variation of the pore connectivity through the depth of the membrane. Several groups have used synchrotron radiation [14], X-ray or electron [15] tomography for the characterization of the 3-dimensional membrane pore structure. However, these methods have not yet been successfully applied to the evaluation of the pore interconnectivity in the relatively small (<50 nm) pores of virus filtration membranes. The objective of this study was to develop a new approach for measuring the pore interconnectivity based on direct observation of gold nanoparticles captured within a porous membrane in which the mem brane exit (skin) surface was partially blocked. Experiments were per formed using Viresolve® Pro, Viresolve® NFP, and Ultipor® DV20 membranes using a mixture of different size gold nanoparticles, with the captured nanoparticles imaged by scanning electron microscopy (SEM) following the procedures described by Nazem-Bokaee et al. [8]. The extent of pore interconnectivity was then evaluated by comparing the observed nanoparticle capture with the calculated flow streamlines and predicted capture profiles determined by numerical analysis of the flow and pressure profiles based on Darcy’s law expressions in the directions normal to and parallel to the membrane surface.
stainless steel screen with a circular pattern of voids was used as a support; this screen blocked approximately 60% of the outlet surface of the membrane. Experiments were conducted in normal flow filtration mode at a constant pressure of 210 kPa (30 psig), which was maintained by nitrogen pressurization of the feed reservoir. The membranes were initially flushed with at least 40 L/m2 of deionized distilled water ob tained from a Direct-Q® 3 UV Water Purification System (Milli poreSigma, Burlington, MA) to ensure complete wetting of the membrane and to remove any trapped air bubbles in the system. Gold nanoparticles (20, 40, 100, and 200 nm) were obtained from MilliporeSigma (Burlington, MA) as stabilized suspensions in citrate buffer. Nanoparticle suspensions were diluted with DI water and ultra sonicated for a short time to insure complete dispersion before being used to challenge the filter. At the end of the filtration run, the filter was removed from the stainless steel holder, rinsed gently with DI water, and then prepared for electron microscopy as discussed below. 2.3. Scanning electron microscopy (SEM)
2. Methods and materials
The location of the captured nanoparticles was examined by SEM using previously developed protocols [9]. Briefly, the membranes were cut into small strips (10 � 3 mm), dehydrated using progressively more concentrated ethanol solutions, infiltrated with LR White medium grade embedding resin (Electron Microscopy Sciences, Hartfield, PA), and transferred to a clean snap-fit gelatin capsule (Ted Pella. Inc., Rodding, CA). Approximately 1 μm slices were obtained using a Leica EM UC6 Ultramicrotome (Leica Biosystems Inc., Buffalo Grove, IL) and coated with a thin layer of iridium using an Emitech K575X sputter coater (Quorum Technologies Ltd, UK). The membrane cross-section was examined at high-magnification using a Zeiss Sigma VP-FESEM (Carl Zeiss Microscopy, Thornwood, NY) equipped with a retractable five-diode backscattering detector. The locations of the individual nanoparticles were determined using ImageJ 1.50I processing software (http://imagej.nih.gov/ij).
2.1. Membranes
3. Theoretical modeling
Experiments were performed using commercially available flat sheet virus filtration membranes: highly asymmetric Viresolve® Pro and Viresolve® NFP membranes were provided by MilliporeSigma (Bedford, MA) and the relatively homogeneous Ultipor® DV20 membrane was from Pall Corp. (Port Washington, NY). The Viresolve® Pro membrane is a surface-modified polyethersulfone (PES) while the Viresolve® NFP and DV20 membranes are both surface-modified polyvinylidene difluoride (PVDF) (Table 1). These flat sheet membranes were cut into 47 mm disks from large sheet-stock and used in single layer format; the actual commercial modules employ multiple layers of membrane to provide the high levels of viral clearance needed in bioprocessing.
The flow distribution within the membranes was evaluated by nu merical solution of the Darcy flow equations: Vr ¼
1 ∂ ∂Vz ðrVr Þ þ ¼0 r ∂r ∂z
kr ∂ ∂P ∂2 P ðr Þ þ kz ¼0 r ∂r ∂r ∂z
Thickness (μm)
Manufacturer
Viresolve® Pro Viresolve® NFP Ultipor® DV20
Polyether-sulfone (PES) Polyvinylidene fluoride (PVDF) Polyvinylidene fluoride (PVDF)
Asymmetric
140
MilliporeSigma
Asymmetric
140
MilliporeSigma
Symmetric
40
Pall
(2)
(3)
Equation (3) was solved numerically using the finite element analysis in COMSOL Multiphysics software (version 5.3) for the system geometry shown in Fig. 1, which is identical to that examined by Li et al. [6]. The dark region is one of the open spaces in the stainless steel support, with the surrounding space representing the blocked region; the grid is defined for the entire membrane above both the blocked and open re gions. No flow boundary conditions were applied over the region covered by the stainless steel support, with symmetry conditions applied at the pore axis and the outer boundary of the cylindrical region. The pressures at the upper and lower surfaces of the membrane were assumed to be uniform. A detailed description of the boundary equations
Table 1 Properties of virus filtration membranes. Asymmetry
(1)
Substitution of Eq. (1) into Eq. (2) yields the following partial dif ferential equation for the local pressure:
Membranes were placed in a 47 mm stainless steel filter holder (MilliporeSigma, Burlington, MA), with the retentive layer (shiny side) facing downstream as per the manufacturer’s recommendation. A
Material
∂P ∂P ; V ¼ kz ∂r z ∂z
where kr and kz are the Darcy permeabilities in the directions parallel to and perpendicular to the membrane surface, respectively. The velocities Vr and Vz must satisfy the continuity equation for an incompressible fluid which is given by:
2.2. Nanoparticle filtration
Membrane
kr
2
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
local pore size first became equal to the particle size; the pore size profile for the Viresolve® Pro and Viresolve® NFP membranes were evaluated by SEM image analysis as presented in a previous publication [16]. A total of 100,000 nanoparticles were placed at the entrance of the filter (z ¼ 140 μm) and distributed based on the magnitude of the local flow velocity as determined by solution of Equation (3) assuming that the nanoparticles have no affect on the flow. 4. Results 4.1. Viresolve® pro membrane Fig. 2 shows the lower surface of the Viresolve® Pro membrane after filtration of 15 L/m2 of a suspension of 20 nm gold nanoparticles at a constant pressure of 210 kPa (30 psig). The membrane was used with the shiny-side (retentive layer) placed on a stainless steel screen (left panels) or on a microporous support (right panels). In both cases, the filtrate flux remained essentially constant (within 5%) throughout the filtration, consistent with the very dilute suspensions of gold nanoparticles (1010 nanoparticles/mL) used in these experiments. The dark red color shows the regions filled with captured gold nanoparticles, with the regular array of red dots in the left hand panels representing the regions that were unblocked by the stainless steel support, consistent with the model geometry shown in Fig. 1. The membrane that was placed on the microporous support shows an essentially uniform red color due to the high porosity and small pore spacing in the support. Fig. 3 shows an SEM image of a cross-section through the Viresolve® Pro membrane, focusing on the region immediately above the boundary between the open and blocked regions of the stainless steel support. The membrane was challenged at a constant pressure of 210 kPa (30 psig) with a suspension containing a mixture of the 20, 40, 100, and 200 nm gold nanoparticles (with mean size of 25 � 8 nm, 42 � 12 nm, 100 � 30 nm, and 200 � 45, respectively, as determined by dynamic light scattering). The white dots are the individual gold nanoparticles;
Fig. 1. Schematic of lower surface of partially blocked membrane. Dark circle represents the open region at the filter exit.
is available in the literature [12]. The governing equations were solved numerically for different values of the pore interconnectivity, defined as K ¼ kkzr , where K ¼ 0
corresponds to a membrane with non-interconnected pores (no lateral flow) while K ¼ 1 corresponds to a membrane with an isotropic struc ture with equal resistance to flow in both directions. A value of K ¼ 0.1 indicates that the velocity in the lateral direction would be one-tenth that in the normal direction for equivalent pressure gradients. The predicted distribution of captured nanoparticles within the membrane was evaluated as follows. The membrane was “challenged” with a feed containing a distribution of nanoparticles (particle radius described by a simple Gaussian function). Each nanoparticle was released at a given location and assumed to follow the flow streamline until it reached a point within the depth of the membrane where the
Fig. 2. Optical images of the lower surface of the Viresolve® Pro membrane after challenge with 15 L/m2 of a suspension of 20 nm gold nanoparticles. Left panels are low and high magnification images of a membrane on the stainless steel support. Right panels are corresponding images of a membrane on a microporous support. 3
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
Fig. 3. High contrast SEM image of the cross-section of a Viresolve® Pro membrane (filter exit at the top surface) after challenging with 60 L/m2 of a sus pension containing 20, 40, 100 and 200 nm gold nanoparticles at 210 kPa with the skin side placed on top of the stainless steel screen spacer (top panel). The distribution of captured nanoparticles (shown by red circles) within the Viresolve® Pro membrane as determined by ImageJ software is shown in the bottom panel. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)
energy dispersive X-ray spectroscopy was used to confirm that these dots were elemental gold as discussed elsewhere [8]. There was no evidence of any nanoparticle aggregation, either in the feed suspension (as determined by dynamic light scattering) or within the membrane (as seen in the SEM images). The SEM image shows distinct “bands” for the different size nanoparticles, with the 20 nm particles captured right near the filter exit (top surface) while the 200 nm nanoparticles are captured around 25 μm further into the filter. More strikingly, the bands for the larger particles are spread out into the space above the blocked region of the membrane, with the “span” for the band increasing as the size of the nanoparticle increases, i.e., as one moves further away from the mem brane exit. This behavior is shown more clearly in the bottom panel of Fig. 3 in which the location of each nanoparticle, as determined by the ImageJ software, is shown by a red circle (irrespective of the particle size). The net result is that the left-hand edge of the particle bands generates a boundary (shown by the red curve, which is simply drawn by eye) that defines the capture profile within the Viresolve® Pro mem brane; the shape of this curve is determined by the extent of lateral flow within this region of the membrane. The calculated values of the flow streamlines within the Viresolve® Pro membrane were evaluated numerically using COMSOL Multiphysics software for different values of the pore connectivity: K ¼ 1, 0.1, and 0.01, with results shown in Fig. 4. At the smallest value of K, the streamlines are directed almost perpendicular to the feed/permeate
surfaces of the membrane, similar to what would be expected for flow through a membrane with straight-through non-interconnected pores, with essentially no bulk flow through the regions upstream of the blocked surface. As K increases, the streamlines begin to move out into the portion of the membrane above the blocked region due to the in crease in the lateral flow through the porous membrane (similar to the profile for the gold nanoparticles seen in Fig. 3). For K ¼ 1, there is a measurable flow over almost the entire entrance to the membrane even though more than 60% of the exit region is blocked by the stainless steel screen. A more quantitative analysis of the pore interconnectivity in the Viresolve® Pro membrane was performed by simulating the nano particle capture in MATLAB using the velocities calculated in COMSOL for specific values of K. The membranes were “challenged” with a dis tribution of particles, in this case defined by a mixture of particles with mean size of 20, 40, 100, and 200 nm having standard deviations of 4, 6, 10, and 15 nm, respectively. Each particle was captured at a location (zvalue) at which the local membrane pore size was equal to the size of the nanoparticle. The resulting simulation results are overlaid on the flow streamlines in Fig. 4. As K increases, the nanoparticles are captured in a more diffuse band spreading out over the blocked region of the mem brane as expected. The best fit value of K for the Viresolve® Pro membrane was esti mated by comparison of the experimental results for nanoparticle 4
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
Fig. 4. Internal flow streamlines within the Viresolve® Pro membrane (oriented with the filter exit at the top surface) for (A) K ¼ 1, (B) K ¼ 0.1, and (C) K ¼ 0.01. Shaded regions show calculated locations of captured nanoparticles corresponding to each value of the pore interconnectivity.
capture with the calculated profiles shown in Fig. 4. An example is shown in Fig. 5. The red circles are the positions of the individual gold nanoparticles (irrespective of particle size) determined using ImageJ software, while the gray curves are the flow streamlines for K ¼ 1. The shape of the calculated streamlines, and the simulated nanoparticle capture bands, are well-aligned with the particle locations, suggesting that K ¼ 1 provides a reasonable estimate of the pore interconnectivity for the Viresolve® Pro membrane. Note that the calculated streamlines do not define the regions with and without flow; the fluid velocity is not actually zero except at the exit of the blocked region (and there are streamlines in the regions to the left of the streamlines shown in Fig. 5). However, the flow streamlines for simulations using either smaller, or
larger, values of K show clear discrepancies with the shape of the experimentally-determined nanoparticle capture profiles. 4.2. Viresolve® NFP membrane Corresponding results for the Viresolve® NFP membrane are shown in Figs. 6 and 7. The top panel of Fig. 6 shows an SEM image after filtration of 60 L/m2 of a mixture of the 20, 40, 100, and 200 nm gold particles at a constant pressure of 210 kPa (30 psi). The 20 nm gold particles are captured very close to the membrane exit (skin layer) over the open region of the stainless steel support. The 40 and 100 nm par ticles were captured in bands that are mostly above the open region but
Fig. 5. Comparison of captured nanoparticle locations within the Viresolve® Pro membrane (determined by ImageJ software) and the calculated flow streamlines from solution of Equation (3) using K ¼ 1. Membrane exit is at upper surface. Particle capture is only shown over 50 μm region near filter exit – total membrane thickness is approximately 140 μm. 5
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
Fig. 6. High contrast SEM image of the cross-section of a Viresolve® NFP membrane with filter exit at top surface after challenging with 60 L/m2 of a suspen sion containing 20, 40, 100 and 200 nm gold nano particles at 210 kPa with the skin side placed on top of the stainless steel support (top panel). The distri bution of captured nanoparticles (in green circles) within the Viresolve® NFP membrane near the retentive layer as determined by ImageJ analysis (bottom panel). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)
do extend slightly into the space above the blocked region. In contrast, the 200 nm particles are scattered throughout the depth of the mem brane over both the open and blocked flow regions, reflecting the nearly uniform pore size in the microporous support on which the Viresolve® NFP membrane is cast. The bottom panel of Fig. 6 shows the distribution of captured nanoparticles obtained by ImageJ analysis of the SEM image in the top panel; the software is able to identify the individual nanoparticles much more effectively by properly adjusting the threshold limits. Again, the nanoparticles are “aligned” with the open region on the membrane based on the location of the “last” 20 nm particle visible in the SEM image. The furthest most nanoparticle seen in the ImageJ output is located 12 μm to the left of the open region, reflecting the small amount of lateral flow within the Viresolve® NFP membrane. Fig. 7 shows the calculated flow streamlines in the Viresolve® NFP membrane for K ¼ 1, 0.1, 0.01, and 0.001. As expected, the flow bends out further into the space above the blocked region as K increases, with the streamlines for K ¼ 0.001 remaining almost entirely over the open region of the membrane. The flow streamlines for K ¼ 0.01, and the corresponding nanoparticle capture simulations, are in fairly good agreement with the experimental results in Fig. 6. This indicates that the
pore interconnectivity for the Viresolve® NFP membrane is approxi mately K ¼ 0.01, which is two orders of magnitude smaller than the value determined for the Viresolve® Pro membrane. 4.3. Ultipor® DV20 membrane In contrast to the Viresolve® Pro and Viresolve® NFP membranes, the Ultipor® DV20 membrane has a fairly uniform pore size throughout the depth of the 40 μm thick membrane; dextran sieving data [18] and fluorescent nanoparticle capture profiles [7] show only a small decrease in pore size as one moves through the depth of the Ultipor® DV20 membrane. Fig. 8 shows an SEM image of a cross-section through a Ultipor® DV20 membrane after challenging with the nanoparticle mixture following the same procedures as used for the Viresolve® Pro and Viresolve® NFP membranes. In contrast to the results with the Viresolve® membranes, all of the nanoparticles were captured near the filter inlet, directly opposite the blockage of the exit surface by the stainless steel support, with the 20 and 40 nm particles penetrating only a few microns into the depth of the membrane. The capture pattern on the upper surface of the Ultipor® DV20 membrane mirrors the circular pattern of open spaces defined by the stainless steel support located 6
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
Fig. 7. Internal flow streamlines within the Viresolve® NFP membranes for (A) K ¼ 1, (B) K ¼ 0.1, (C) K ¼ 0.01, (D) K ¼ 0.001. Filter exit located at top surface.
to fully wet the pores, with the permeability evaluated as 1.1 � 10 10 m based on the flow rate at 210 kPa (30 psi). The top and bottom surfaces of the membrane were then partially covered with tape, with the two taped sections overlapping by approximately 1 mm. The membrane was then placed in the base of a stirred cell and re-pressurized to 210 kPa (30 psi). No measurable filtrate flux was observed after 180 min of filtration, corresponding to a permeability of less than 2 � 10 16 m based on the approximate volume of one drop of water. This low permeability cor responds to K < 10 3 based on the solution of Equation (3), in good agreement with the results obtained from the gold nanoparticles, providing further confirmation of this methodology. 5. Discussion The work described in this study presents a new method for evalu ating the internal flow distribution and pore interconnectivity in asymmetric membranes based on the capture profiles for different size gold nanoparticles as visualized by scanning electron microscopy. The membranes were used with the skin-side placed directly on top of a stainless steel spacer that blocked a significant fraction of the pores in the membrane skin (filter exit), forcing the fluid to flow through the open (unblocked) regions. The captured nanoparticles in the Viresolve® Pro membrane were clearly visible in the region above the open region of the membrane, with the nanoparticle “bands” spreading out further into this region as one moves away from the filter exit, providing a picture of the flow streamlines within the interior of the membrane. ImageJ analysis of the gold nanoparticles in the Viresolve® Pro mem brane were in good agreement with the calculated values of the flow streamlines and capture profile evaluated for a membrane with a permeability ratio of K ¼ kkzr ¼ 1, consistent with a high degree of lateral
Fig. 8. High contrast SEM image of the cross-section of an Ultipor® DV20 membrane challenged with 60 L/m2 of a suspension containing 20, 40, 100 and 200 nm gold nanoparticles at 210 kPa showing nanoparticle capture near the filter entrance (lower surface).
beneath the membrane (at the membrane exit), consistent with the absence of any significant pore interconnectivity in the DV20 mem brane. Model simulations for the flow streamlines and nanoparticle capture suggest that K < 10 3 for the Ultipor® DV20 membrane; larger values of K show significantly more diffuse particle capture than that shown in the SEM image in Fig. 8. In order to confirm the very low degree of lateral flow within the DV20 membrane, an independent measure of the pore interconnectivity was obtained following the general approach described by Ho and Zydney [12]. An Ultipor® DV20 membrane was first flushed with water 7
F. Fallahianbijan et al.
Journal of Membrane Science xxx (xxxx) xxx
flow within the membrane. In contrast, the nanoparticle capture profiles in the Viresolve® NFP membrane were best fit using K ¼ 0.01, indicating that the permeability in the direction normal to the membrane surface is much larger than that in the transverse (lateral) direction. Experiments performed with the Ultipor® DV20 membrane showed particles captured on the upper (inlet) surface of the membrane, but only over the regions of the membrane that were open on the downstream (exit) surface. This clearly indicates that there is minimal lateral flow within the Ultipor® DV20 membrane. This very low degree of pore interconnectivity was confirmed using the “overlapping tape” method described previously by Ho and Zydney [12] for symmetric (homoge neous) microfiltration membranes, with no measurable permeability (flow) when the tape on the upper and lower surfaces of the membrane overlap. This provides additional validation of the gold nanoparticle methodology developed in this study. However, the “overlapping tape” method cannot be used with highly asymmetric virus filtration mem branes since the high degree of pore interconnectivity in the membrane support structure will obscure the effects of the interconnectivity in/near the membrane skin, which is the region of interest in under standing the performance characteristics of these asymmetric mem branes. Note that the approach developed in this work could easily be extended to evaluate the pore interconnectivity of asymmetric ultrafil tration or microfiltration membranes by adjusting the size of the gold nanoparticles to properly map out the flow through the depth of these membranes. Interestingly, the devices containing Viresolve® NFP and Ultipor® DV20 membranes, both of which have very low pore interconnectivities as determined from the gold nanoparticle capture data, show a signifi cant loss in virus retention after a process disruption [10,17]. In contrast, the Viresolve® Pro devices show highly robust virus retention, both following a process disruption and over a broad range of filtration conditions [17]. Additional studies will be required to identify the
relationship between the pore interconnectivity and the virus retention characteristics of these virus removal membranes. 6. Conclusions This paper demonstrates a novel methodology for evaluating the pore interconnectivity of highly asymmetric membranes using SEM to visualize gold nanoparticles captured within a membrane in which flow through the exit (skin side) was partially blocked, in this case by a stainless steel support. The method was validated using data for the Ultipor® DV20 with very low pore interconnectivity and then extended to provide the first measurements of the pore interconnectivity for the Viresolve® Pro and Viresolve® NFP virus filtration membranes. The data clearly demonstrate that there are significant differences in the pore interconnectivity for these membranes, all of which have very similar retention characteristics. The approach can be readily extended to other asymmetric membranes by proper choice of the gold nanoparticle size, providing a general methodology for evaluating the pore inter connectivity in both ultrafiltration and microfiltration membranes. Declaration of competing interest Two of the authors (SG and CC) are employed by MilliporeSigma, which provided some of the membranes examined in this study. Acknowledgements The authors would like to acknowledge MilliporeSigma for their financial support, including donation of the Viresolve® Pro and Vir esolve® NFP membranes. The authors would also like to thank Farzad Mohajerani for the thoughtful discussion and assistance in performing the numerical simulations.
Nomenclature Notation δm D K k L P Pfeed Ppermeate V
Membrane thickness (m) Open region diameter (m) Permeability ratio (-) Darcy permeability in membrane (m3 s/kg) Distance between open regions (m) Local hydrostatic pressure (N/m2) Feed pressure (N/m2) Filtrate pressure (N/m2) Fluid velocity (m/s)
Subscripts r direction parallel to membrane surface z direction normal to membrane surface
References
[5] C.C. Ho, A.L. Zydney, Theoretical analysis of the effect of membrane morphology on fouling during microfiltration, Separ. Sci. Technol. 34 (1999) 2461–2483, https://doi.org/10.1081/SS-100100785. [6] W. Li, C. Duclos-Orsello, C.-C. Ho, Theoretical analysis of the effects of asymmetric membrane structure on fouling during microfiltration, AIChE J. 55 (2009) 857–866, https://doi.org/10.1002/aic.11763. [7] F. Fallahianbijan, S. Giglia, C. Carbrello, A.L. Zydney, Use of fluorescently-labeled nanoparticles to study pore morphology and virus capture in virus filtration membranes, J. Membr. Sci. 536 (2017) 52–58, https://doi.org/10.1016/j. memsci.2017.04.066. [8] H. Nazem-Bokaee, F. Fallahianbijan, D. Chen, S.M. O’Donnell, C. Carbrello, S. Giglia, D. Bell, A.L. Zydney, Probing pore structure of virus filters using scanning electron microscopy with gold nanoparticles, J. Membr. Sci. 552 (2018) 144–152, https://doi.org/10.1016/j.memsci.2018.01.069. [9] M. Bakhshayeshi, N. Jackson, R. Kuriyel, A. Mehta, R. van Reis, A.L. Zydney, Use of confocal scanning laser microscopy to study virus retention during virus filtration,
[1] W.M. Deen, Hindered transport of large molecules in liquid-filled pores, AIChE J. 33 (1987) 1409–1425, https://doi.org/10.1002/aic.690330902. [2] A.L. Ahmad, S.C. Low, S.R.A. Shukor, W.J.N. Fernando, A. Ismail, Hindered diffusion in lateral flow nitrocellulose membrane: experimental and modeling studies, J. Membr. Sci. 357 (2010) 178–184, https://doi.org/10.1016/j. memsci.2010.04.018. [3] C.C. Ho, A.L. Zydney, Protein fouling of asymmetric and composite microfiltration membranes, Ind. Eng. Chem. Res. 40 (2001) 1412–1421, https://doi.org/10.1021/ ie000810j. [4] C.C. Ho, A.L. Zydney, Effect of membrane morphology on the initial rate of protein fouling during microfiltration, J. Membr. Sci. 155 (1999) 261–275, https://doi. org/10.1016/S0376-7388(98)00324-X.
8
F. Fallahianbijan et al.
[10] [11] [12] [13]
[14]
Journal of Membrane Science xxx (xxxx) xxx
J. Membr. Sci. 379 (2011) 260–267, https://doi.org/10.1016/j. memsci.2011.05.069. M.A. Woods, A.L. Zydney, Effects of a pressure release on virus retention with the Ultipor DV20 membrane, Biotechnol. Bioeng. 111 (2014) 545–551, https://doi. org/10.1002/bit.25112. A. Yamamoto, T. Hongo-Hirasaki, Y. Uchi, H. Hayashida, F. Nagoya, Effect of hydrodynamic forces on virus removal capability of PlanovaTM filters, AIChE J. 60 (2014) 2286–2297, https://doi.org/10.1002/aic. C.C. Ho, A.L. Zydney, Measurement of membrane pore interconnectivity, J. Membr. Sci. 170 (2000) 101–112, https://doi.org/10.1016/S0376-7388(99) 00360-9. M.B. Tanis-Kanbur, F. Zamani, W.B. Krantz, X. Hu, J.W. Chew, Adaptation of evapoporometry (EP) to characterize the continuous pores and interpore connectivity in polymeric membranes, J. Membr. Sci. 575 (2019) 17–27, https:// doi.org/10.1016/j.memsci.2018.12.050. J.C. Remigy, M. Meireles, X. Thibault, Morphological characterization of a polymeric microfiltration membrane by synchrotron radiation computed
[15] [16]
[17] [18]
9
microtomography, J. Membr. Sci. 305 (2007) 27–35, https://doi.org/10.1016/j. memsci.2007.06.059. K.-L. Tung, K.-S. Chang, T.-T. Wu, N.-J. Lin, K.-R. Lee, J.-Y. Lai, Recent advances in the characterization of membrane morphology, Curr. Opin. Chem. Eng. 4 (2014) 121–127, https://doi.org/10.1016/j.coche.2014.03.002. F. Fallahianbijan, S. Giglia, C. Carbrello, D. Bell, A.L. Zydney, Impact of protein fouling on nanoparticle capture within the Viresolve® Pro and Viresolve® NFP virus removal membranes, Biotechnol. Bioeng. (2019), https://doi.org/10.1002/ bit.27017. S.K. Dishari, A. Venkiteshwaran, A.L. Zydney, Probing effects of pressure release on virus capture during virus filtration using confocal microscopy, Biotechnol. Bioeng. 112 (2015) 2115–2122, https://doi.org/10.1002/bit.25614. M. Bakhshayeshi, D.M. Kanani, A. Mehta, R. van Reis, R. Kuriyel, N. Jackson, A. L. Zydney, Dextran sieving test for characterization of virus filtration membranes, J. Membr. Sci. 379 (2011) 239–248, https://doi.org/10.1016/j. memsci.2011.05.067.