6266
Langmuir 2006, 22, 6266-6272
In Situ Three-Dimensional Characterization of Membrane Fouling by Protein Suspensions Using Multiphoton Microscopy David J. Hughes, Zhanfeng Cui,* Robert W. Field, and Uday K. Tirlapur Department of Engineering Science, UniVersity of Oxford, Parks Road, Oxford, U.K., OX1 3PJ ReceiVed December 14, 2005. In Final Form: February 21, 2006 Fouling of microfiltration membranes leads to severe flux declines and the need to clean or replace the membrane. In situ 3D characterization of protein fouling both on the surface and within the pores of the membrane was achieved using multiphoton microscopy. Time-lapse images of the fouled membrane were obtained for single suspensions and mixtures of fluorescently labeled bovine serum albumin and ovalbumin. Deposited protein aggregates were visible on the membrane and evidently play an important role in fouling. A combination of 3D images and resistance versus time data was used to identify the dominant fouling mechanism. Fouling is initially internally dominated, but after 1 and 15 min for ovalbumin and bovine serum albumin, respectively, the fouling becomes externally dominated. This is in good agreement with two-stage protein fouling models.
1. Introduction Membrane filtration of suspensions containing proteins is an important evolving technology. Membrane fouling has long been an issue in membrane processes1,2 and still remains a current problem.3 In constant transmembrane pressure (TMP) operation, fouling leads to a reduction in flux and protein transmission with time, reducing the efficiency of the separation and eventually resulting in the need to clean or replace the membrane.3 Fouling results from the accumulation of rejected material on the top surface of the membrane (external fouling) and/or the deposition and accumulation of smaller species at the openings of pores or within the pore structure (internal fouling). Microfiltration (MF) is being used increasingly in biotechnology applications such as the separation of plasma proteins from blood or the removal of cells from broths that contain proteins, where the goal is to achieve high protein recovery (through a combination of high fluxes and protein transmission) while retaining other components such as cells.4 The pores of MF membranes (typically 0.1-1 µm diameter) are much larger than native proteins (typical dimensions of a few nanometers),5 hence little rejection of proteins would be expected at the surface of the membrane and protein transmission should be high. However, numerous studies have shown that the fouling of MF membranes by proteins during filtration can be severe, resulting in low fluxes and protein transmission.6-11 Numerous techniques have been used to characterize the protein fouling of membranes. They were recently thoroughly reviewed by Chan and Chen.3 The pertinent issues when considering characterization techniques are the sample preparation required, the ability of the technique to discriminate between foulant species, * Corresponding author. E-mail:
[email protected]. (1) Belfort, G.; Davis, R. H.; Zydney, A. L. J. Membr. Sci. 1994, 96, 1-58. (2) Marshall, A.; Munro, P.; Traegaardh, G. Desalination 1993, 91, 65-108. (3) Chan, R.; Chen, V. J. Membr. Sci. 2004, 242, 169-188. (4) Chan, R.; Chen, V. J. Membr. Sci. 2001, 185, 177-192. (5) Guell, C.; Czekaj, P.; Davis, R. J. Membr. Sci. 1999, 155, 113-122. (6) Ferrando, M.; Rozek, A.; Zator, M.; Lopez, F.; Guell, C. J. Membr. Sci. 2005, 250, 283-293. (7) Kluge, T.; Rezende, C.; Wood, D.; Belfort, G. Biotech. Bioeng. 1999, 65, 649-658. (8) Guell, C.; Davis, R. J. J. Membr. Sci. 1996, 119, 269-284. (9) Herrero, C.; Pra´danos, P.; Calvo, J. I.; Tejerina, F.; Herna´ndez, A. J. Colloid Interface Sci. 1997, 187, 344-351. (10) Ho, C.-C.; Zydney, A. L. J. Membr. Sci. 1999, 155, 261-275. (11) Kelly, S. T.; Zydney, A. L. J. Membr. Sci. 1995, 107, 115-127.
and the sensitivity or resolution available. In light of these issues, the major techniques for characterizing protein fouling will be briefly assessed. Techniques can be conveniently categorized as in situ or ex situ. Ex situ techniques involve the removal of the membrane from the module before analysis. The most common method of ex situ visualization of protein fouling is scanning electron microscopy (SEM).3,4,8,10 This technique offers high 2D spatial resolution but requires considerable sample preparation, and differentiation between foulant species is difficult if not impossible. An alternative ex situ approach with the potential to produce 3D images of fouled membranes is confocal scanning laser microscopy (CLSM).6,12 This technique requires that the foulant species and/or the membrane must be labeled with fluorophores. CLSM uses either UV or visible lasers and single-photon excitation from three separate laser sources to evoke red, green, and blue fluorescence. Serial optical sections in the xy plane are taken at a number of depths and then reconstructed to generate a 3D image. In the xy plane, spatial resolutions close to those of SEM are possible. The penetration depth in the z direction is limited to around 50 µm (and is to some degree sample-dependent) because CSLM uses a single high-energy photon to evoke fluorescence. An appropriate choice of fluorophores allows the identification of fouling proteins. Ex situ techniques that allow the characterization of fouling without the visualization of the membrane include atomic force microscopy (AFM), matrix-assisted laser desorption ionization mass spectrometry (MALDI-MS), and attenuated total reflection Fourier transform infrared spectroscopy (ATR-FITR). In AFM, a sharp tip attached to a cantilever is moved over the sample surface. A laser beam is reflected from the cantilever to an optical sensor, allowing deflections of the cantilever to be detected. AFM can be used to produce topographic maps of the fouled surface with a resolution under ideal conditions down to fractions of a nanometer.13,14 Samples require no preparation before analysis, and quantitative information on electrochemical interactions can be gained. However, this technique cannot be (12) Reichert, U.; Linden, T.; Belfort, G.; Kula, M.; Tho¨mmes. J. J. Membr. Sci. 2002, 199, 161-166. (13) Ge´san, G.; Daufin, G.; Merin, U.; Labbe´ J. P.; Que´merais, A. J. Membr. Sci. 1993, 80, 131-145. (14) Bowen, W. R.; Doneva, T. A.; Stoton, J. A. G. Colloids Surf., B 2003, 27, 103-113.
10.1021/la053388q CCC: $33.50 © 2006 American Chemical Society Published on Web 06/13/2006
In Situ 3D Characterization of Membrane Fouling
used to differentiate between fouling species. MALDI-MS involves dispersing the sample in a matrix of laser-absorbing material and firing short pulses of laser energy at the matrix. The sample is ionized, desorbs, and passes through a time-of-flight mass spectrometer. Quantitative information on the type and quantity of fouling species can be obtained, but the sample requires significant preparation before analysis.15,16 ATR-FITR requires the specimen to be pressed against an internal reflection element (IRE) under vacuum conditions. Infrared radiation is then fired at one end of the IRE and undergoes total internal reflection before leaving the IRE at the opposite end. On each total internal reflection, the infrared radiation interacts with the sample pressed against the IRE, forming an evanescent wave at the interface. The frequency at which infrared radiation is absorbed is a characteristic of the chemical bonds or functional groups present in the sample. ATR-FITR allows depth profiling of the sample up to 5 µm.3,17,18 The technique cannot be used to differentiate between protein foulants because many proteins contain similar functional groups.19 In situ techniques have the advantage of analyzing the fouling without the removal of the membrane from the module and may give real-time information about the development of the fouling. To use in situ techniques, a special module must often be fabricated. Small-angle neutron scattering (SANS) is an in situ technique where neutrons are fired at a membrane while filtration is occurring and their scattering is detected. From this, the location and quantity of the foulant can be determined.20-22 Only a limited range of membranes may be used because the technique requires that the membranes be transparent to neutrons. Measurements of streaming potential can be used to calculate an “apparent” zeta potential of a membrane using the Helmholtz-Smoluchowski equation.3 Filtration modules have been combined with equipment to measure the streaming potential to monitor, in real time, the changes in the apparent zeta potential as the membrane is fouled by proteins.23,24 In this study, multiphoton microscopy (MPM) is used to image in situ the fouling of track-etched polycarbonate membranes by fluorophore-labeled bovine serum albumin (BSA) and ovalbumin. Two or more near infrared photons are focused to a femtoliter volume where fluorescence is excited. Through sectioning in the xy plane at different depths, this technique produces high-quality 3D images. The analysis of the 3D images allows protein fouling species and the location of the fouling to be determined. The images have been used in conjunction with a dead-end filtration model to identify the dominant fouling mechanism.
2. Fouling Mechanisms Many authors have suggested that the initial stages of fouling under constant TMP cross-flow filtration can be well approximated by simple dead-end filtration theory.25 At the start (15) Chan, R.; Chen, V.; Bucknall, M. P. Desalination 2002, 146, 83-90. (16) Chan, R.; Chen, V.; Bucknall, M. P. Biotech. Bioeng. 2004, 85, 190-201. (17) Kim, K. J.; Fane, A. G.; Nystro¨m, M.; Pihlajamaki, A. J. Membr. Sci. 1997, 134, 199-208. (18) Pihlajama¨ki, A.; Va¨isa¨nen, P.; Nystro¨m, M. Colloids Surf., A 1998, 138, 323-333. (19) Nystro¨m, M.; Aimar, P.; Luque, S.; Kulovaara, M.; Metsa¨muuronen, Colloids Surf., A 1998, 138, 185-205. (20) Su, T. J.; Lu, J. R.; Cui, Z. F.; Thomas, R. K.; Heenan, R. K. Langmuir 1998, 14, 5517-5520. (21) Su, T. J.; Lu, J. R.; Cui, Z. F.; Bellhouse, B. J.; Thomas, R. K.; Heenan, R. K. J. Membr. Sci. 1999, 163, 265-275. (22) Su, T. J.; Lu, J. R.; Cui, Z. F.; Thomas, R. K. J. Membr. Sci. 2000, 173, 167-178. (23) Nystro¨m, M.; Pihlajama¨ki, A.; Ehsani, N. J. Membr. Sci. 1994, 87, 245256. (24) Causserand, C.; Nystro¨m, M.; Aimar, P. J. Membr. Sci. 1994, 88, 211222.
Langmuir, Vol. 22, No. 14, 2006 6267
of constant TMP microfiltration, fluxes are high, and the convective flux of material toward the membrane greatly exceeds the material removed by cross-flow action. Thus, conditions approximate dead-end filtration, and dead-end filtration models can be applied to determine the fouling mechanism. Hermia26 introduced a characteristic form of the blocking filtration laws for dead-end constant pressure filtration
( ) () d2t dt )K 2 dV dV
n
(1)
where V is the volume of filtrate collected at time t and K and n are constants depending upon the fouling mechanism. The four fouling mechanisms incorporated into the model are complete pore blocking (n ) 2), standard blocking (n ) 1.5), intermediate blocking (n ) 1), and cake filtration (n ) 0). These mechanisms maybe classified as internal, external, or intermediate depending on whether the fouling occurs within the pores of the membrane or on the membrane surface. Darcy’s law provides a relationship between the transmembrane pressure drop and the permeate flux
J)
∆P µRT
(2)
where J is the permeate flux, ∆P is the TMP, µ is the permeate viscosity, and RT is the total resistance to permeate flow, which maybe subdivided as
RT ) Rm + Rif + Rc
(3)
where Rm is the intrinsic membrane resistance, Rif is the resistance due to internal fouling, and Rc is the cake resistance due to an external cake. A combination of Hermia’s blocking laws (eq 1) and Darcy’s law (eqs 2 and 3) leads to equations describing the variation of total resistance with time for each mechanism. The analysis is similar to that used by Tracey and Davis.27 Internal fouling occurs within the internal pore structure of the membrane, and the mechanisms incorporated into the model to describe the phenomena are standard blocking and complete pore blocking. With these mechanisms, there is no external cake formation; therefore, Rc ) 0. Standard blocking assumes even deposition of foulant on the pore walls, resulting in a uniform decrease in pore diameter. For standard blocking, integration and recasting of eq 1 yields
RT ) Rm(1 + KSBMQ00.5t)2
(4)
where t is time, Q0 is the permeate flow rate at t ) 0, and KSBM is a constant. The second internal fouling mechanism is the complete pore blocking mechanism where the number of blinded or plugged pores increases proportionally to the filtrate volume. The diameter of pores that have not been plugged is assumed to be constant. This again increases the resistance of the membrane. Integration and recasting of eq 1 yields
RT ) Rme(KPBMt)
(5)
where KPBM is a constant. For both internal fouling mechanisms, RT and dRT/dt increase with time. External fouling occurs when a species accumulates on the upper surface of the membrane, forming a cake. This is commonly (25) Keskinler, B.; Yildiz E.; Erhan, E.; Dogru, M.; Bayhan, Y. K.; Akay, G. J. Membr. Sci. 2004, 233, 59-69. (26) Hermia J. Trans. 1nd. Chem. Eng. 1982, 60, 183-187. (27) Tracey, E.; Davis, R. H. J. Colloid Interface Sci. 1994, 167, 104-116.
6268 Langmuir, Vol. 22, No. 14, 2006
Hughes et al.
known as cake fouling. Here, there is no internal fouling of the membrane (Rif ) 0), and the cake causes an additional resistance in series with the membrane. Integration and recasting of eq 1 yields
RT ) Rm(1 + KCFMQ02t)0.5
(6)
where KCFM is a constant. For external fouling, RT increases with time, and dRT/dt decreases with time. Intermediate fouling may be considered to be an amalgamation of complete pore blocking and external fouling. Foulants are assumed to deposit on the surface in a random fashion with some completely blocking pores whereas others deposit over an already blocked pore. Deposition above already blocked pores will lead to an external fouling layer. Integration and recasting of eq 1 yields
RT ) Rm(1 + Q0kIMt)
(7)
where KIM is a constant. RT increases linearly with time, and dRT/dt is constant. In the present work, the shape of the resistance versus time curves will be correlated with 3D MPM images. MPM images allow both internal and external fouling to be clearly identified. 3. Materials and Methods 3.1. Cross-Flow Filtration Module. A small cross-flow filtration module was designed and integrated onto the TE300 upright microscope stage (Nikon, U.K. Ltd., Surrey, U.K.) of the MPM system in the imaging plane of the microscope objective. The module consists of two Perspex plates that hold thin rubber gaskets above and below the membrane. The feed and permeate channels are formed from two thin rubber gaskets. The feed channel has a width of 10 mm, a length of 45 mm, and a height of 0.5 mm. The equivalent hydraulic diameter of the feed channel is 0.95 mm. The permeate channel has a width of 10 mm, a length of 37 mm, and a height of 1 mm. The permeate channel is shorter than the feed channel to minimize inlet effects over the active filtration area. The active area of the membrane within the module is 3.7 cm2. The filtration module above the area to be imaged must be free from defects and totally transparent. To accommodate this requirement, the upper plate of the module has a recessed window into which a 22 mm × 40 mm microscope coverslip is placed and sealed with rubber cement. A schematic of the system can be seen in Figure 1. Feed is drawn from a reservoir by an Ismatec MP peristaltic pump. The volumetric flow rate through the system is set by the rotational speed of the pump. The cross-flow velocity in all experiments was 0.20 m s-1, which corresponds to a Reynolds number of 190. The TMP was taken as the average of the pressures at the inlet and outlet of the module, which were measured using two Cole Parmer high-accuracy pressure transducers. The TMP was set to 0.10 bar in all experiments by adjusting a needle valve at the module exit. The permeate was collected, and the weight was measured using a Denver Instruments AL 400 balance connected to a PC. The permeate was returned to the feed reservoir at regular intervals. 3.2. Membranes. The membranes used in the study were 0.22µm-pore-size polycarbonate track-etched membranes (cat. no. GTTP 047 00) supplied by Millipore. All membranes used in the study were supplied as 47-mm-diameter disks. A new membrane was used for each experiment. 3.3. Proteins. Nonfluorescent BSA and ovalbumin were purchased from Sigma-Aldrich (cat. no. A-3803) and Fisher (cat. no. 400450500), respectively. BSA-fluorescein conjugate and ovalbuminTexas red conjugate were purchased from Sigma-Aldrich (cat. no. A-9771) and Invitrogen (cat. no. O-23021), respectively. BSAfluorescin conjugate fluoresces green whereas ovalbumin-Texas red conjugate fluoresces red. Nonfluorescent proteins are included in the study to examine the effect of conjugating the fluorophore to the protein on filtration performance. Protein suspensions were
Figure 1. Schematic of the MPM setup. PT1 and PT2 denote pressure transducers. prepared in 10 mM phosphate buffer at a pH of 7.5. Experiments were performed with suspensions of single proteins or 50:50 mixtures by mass. In all experiments, the total protein concentration was 50 mg L-1. 3.4. MPM Experimental Procedure. Noninvasive 3D two-photon images of all membrane protein fouling were generated using a modified BioRad/Zeiss Radiance 2100 MP multiphoton laser scanning system (Zeiss, GmbH, Jena, Germany), as partially detailed elsewhere.28 The near-infrared (NIR) laser beams (λ ) 800 nm) were obtained from a tuneable 76 MHz femtosecond Ti:sapphire laser (Mira 900-F, Coherent, Ely, U.K.) pumped by a 7 W multiline argon-ion laser (Verdi, Coherent, Ely, U.K.). Appropriate mean laser powers were electronically regulated via the pockol cell of the BioRad beam-conditioning unit (BCU) and were recorded29,30 using the Fieldmaster FM power meter (Coherent, Ely, U.K.) interfaced between the BCU and the MPM scan head. The NIR femtosecond pulsed laser beams were focused to a diffraction-limited spot using a high-numerical-aperture (N.A. 1.3) 60× Nikon water-immersion objective (with a working distance of ca. 1 mm). The excited fluorescence from the sample was returned either to the fluorescence detectors for 3D imaging or to the SpectraCube (Applied spectral imaging, Israel) for spectral analysis. For 3D imaging, the x, y, z resolutions used were either 0.16 µm × 0.l6 µm × 0.70 µm or 0.1 µm × 0.1 µm × 0.52 µm. The former was used for most images, and the latter was used for the high-definition examination of deposited aggregates. The SpectraCube allows the precise determination of the wavelength of the emitted fluorescence from the sample. This spectral detector has a resolution of Analyze Particles tool then determines the fractional area of the image, which is black. This is the fractional area covered by protein deposits.
4. Results and Discussion 4.1. Filtration Performance of Fluorescent and Nonfluorescent Proteins. The flux decline curves of 50 mg L-1 suspensions of native and fluorophore-conjugated proteins are shown in Figure 2. The BSA-fluorescein conjugate behaves differently from native BSA, with a noticeable flux decline over the first 30 min, whereas the native BSA has an almost constant flux for 30 min and then begins to decline. The conjugation of fluorescein may change the surface properties of BSA, increasing its tendency to aggregate or altering the protein membrane interactions and thus increasing the rate of fouling. Kelly and Zydney11 noted that between lots of native BSA from the same (31) Kang, S. K.; Subramani, A.; Hoek, E.; Deshusses, M.; Matsumoto, M. J. Membr. Sci. 2004, 244, 151-165.
Figure 3. Time-lapse images of fouling by BSA, ovalbumin. and 50:50 BSA/ovalbumin, with an xy resolution of 0.16 µm × 0.16 µm. Scale bar, 20 µm; CFV, 0.20 m s-1; total concentration, 50 mg L-1; TMP, 0.10 bar.
supplier there maybe some difference in the filtration performance. The filtration performance of ovalbumin and the 50:50 mixture of BSA/ovalbumin was almost unaffected by conjugating fluorophores to the proteins. From Figure 2, it is clear that ovalbumin fouls the membrane much more severely than BSA and that the fouling of ovalbumin dominates the fouling behavior of the 50:50 mixture. Thus, the similarity of the fouling curves of the native and fluorophore-conjugated 50:50 mixture is not unexpected because the ovalbumin, whose behavior is unaffected by conjugated fluorophore, is dominant. 4.2. In Situ Imaging of Protein Fouling. The MPM system allows time-lapse images of the membrane to be taken. The variation of foulant species and location can be determined from the images. The top-down images in Figure 3 are of polycarbonate track-etched membranes with 0.22 µm pores challenged by singleprotein solutions of BSA and ovalbumin and a 50:50 mixture. Figure 4 shows the associated flux decline curves and the percentage of the membrane covered by protein or protein aggregates. The fluxes after 60 min of BSA, ovalbumin, and the 50:50 mixture are 107.5, 48.5, and 66.8 L m-2 h-1, respectively.
6270 Langmuir, Vol. 22, No. 14, 2006
Figure 4. Flux decline curves and percentage coverage of the membrane by proteins and/or protein aggregates. These curves correspond to the images in Figure 4. Flux: ([) BSA, (9) ovalbumin, and (2) 50:50 BSA/ovalbumin. Percentage coverage of the membrane: (]) BSA, (0) ovalbumin, and (4) 50:50 BSA/ovalbumin. CFV, 0.20 m s-1; total concentration, 50 mg L-1; TMP, 0.10 bar.
Even with the relatively low concentration of 50 mg L-1, the proteins still cause large reductions in the flux under constant TMP operation. The flux decline curves correlate well with the MPM images. In the first minute of BSA filtration, the membrane is initially fouled in small patches that cover less than 10% of the membrane surface. The track-etched polycarbonate membranes have low porosity (
