Limestone nanoparticles as nanopore templates in polymer membranes: narrow pore size distribution and use as self-wetting dialysis membranes

Abstract

Limestone nanoparticles can be used as nanopore templates to prepare porous polymeric films. Their application as membranes is so far strongly limited by the fact that these films are highly hydrophobic. In this study, a simple method is reported to directly produce self-wetting membranes by the template removal method. Triethyl citrate modified polyethersulfone and cellulose acetate membranes were produced using dissolvable limestone nanoparticles as pore templates. The nanoporous polymer films were used as dialysis membranes and characterized by means of buffer exchange rate, molecular weight cut-off, protein adsorption, pore size distribution and water contact angle. The herein prepared membranes were further benchmarked against commercially available dialysis membranes with comparable average pore size. They showed narrow pore size distributions, fast dialysis rates at low protein adsorption and molecular weight cut-off of around 12 kDa. Interestingly, the triethyl citrate modified polyethersulfone membranes displayed only moderate change in pore size distribution as a result of the plasticizer additive compared to pure polyethersulfone membranes. This is a matter of substantial interest considering the fact that additive modifications of membranes produced by the predominant phase inversion process typically show alterations in morphology that lead to undesired changes in membrane performance. Furthermore, dextran recovery analysis proved to meet the specific requirements for dialysis membrane characterization and benchmarking.

---

1. Introduction

Porous polymeric membranes are of continuously growing importance in the field of filtration and dialysis. Ultrafiltration and microfiltration are commonly used for separation and purification of proteins and particles where the target is smaller than the contamination.^1–3 Dialysis on the other hand is applied, when the proteins and particles, that need to be purified, are larger than the contamination of the sample.^4–7 Compared to filtration, dialysis is a pressureless, diffusion driven process. The membrane separates two liquid solutions with different protein or salt concentrations. The molecules then diffuse through the membrane following the concentration gradient as described by Fick's law:with flux J, diffusion coefficient D and dimensional concentration gradient described by . Diffusion coefficient D can be described by the following relation

with Boltzmann constant (k_B), temperature (T), dynamic viscosity (η) and hydrodynamic radius (R). Thus, we get

The difference in flux – as a result of the different hydrodynamic radius of the molecules – and the pore size of the membrane can therefore be used to separate larger from smaller molecules. Such membranes are currently being produced by the phase inversion process.^8–13 Low selectivity due to broad pore size distribution of these membranes has frequently been reported.^14–16 Research on this type of membrane has mainly focussed on chemical functionalization^17–19 or addition of particles^20–22 to increase hydrophilicity. However, phase inversion is complicated by the numerous parameters that have to be precisely controlled during production. Therefore, surface modification and functionalization often leads to undesired structure changes and degradation in membrane performance.^22–24

Recently, we presented a novel type of polymeric membrane production based on the nanoparticle template removal method.^15,25–27 This method uses a dense, interconnected nanoparticle structure as pore forming template in a polymer matrix. Later dissolution of the template leads to the formation of a nanoporous polymer. This process proved to be very versatile and was recently applied on different polymers,^28,29 showed to be functionalizable by addition of nanoparticles³⁰ and demonstrated the opportunity to obtain membranes with high ionic conductivity for use as battery separators.²⁹ Residual templating particle amount was shown to be below 0.3% wt for limestone derived membranes.¹⁵ These residual particles are well encapsulated in the polymer matrix and do not lead to any measurable contamination of the permeate.³¹ Recently, we demonstrated the first large scale production of this new membrane type for use as water filters in the production of safe drinking water.³¹ Unfortunately, water flux was rather low due to the bad wettability of the polyethersulfone (PES) membrane surface.

In this work, we report on the modification of pure PES membranes by use of a plasticizer and on the application of cellulose acetate (CA) as membrane polymer targeting the formation of hydrophilic, self-wetting dialysis membranes. Both, PES and CA are well known polymers for membrane production, especially for the phase inversion based processes.^32–34 PES was altered by addition of 30 wt% of triethyl citrate (TEC). Environmentally friendly TEC³⁵ successfully decreased the water contact angle of PES by more than 3° and turned the PES more water-wettable without any drastic change in pore structure and performance. CA was for the first time used as membrane polymer using nanoparticles as pore templates and showed faster dialysis rates than comparable commercial membranes. All tested membranes were fully characterized regarding dialysis rate, protein adsorption properties, pore size distribution, water contact angle and molecular weight cut-off (MWCO). MWCO's of ultrafiltration membranes are generally determined by the so called dextran rejection profile test.³⁶ This membrane characterization method is widely accepted among membranologists.^37–39 However, even though ultrafiltration and dialysis membranes are basically the same from a morphological point of view, the separation processes of filtration and dialysis are completely different. Eqn (3) describes the dependence of the diffusion process on the size of the molecules. In a typical dialysis step, a biomolecule (10–100 nm) is separated from a highly concentrated salt solution (monovalent cations, appr. 0.6–0.8 nm). This difference in size leads to a flux ratio of 10 to 150 in favour of the salt molecules. Therefore, pore size may be the determining factor in filtration. However, separation processes in dialysis are much more flux dependent than pore size dependent. Hence, dialysis membranes should not be characterized by a filtration test. This is why we here propose a simple dextran recovery analysis. A dextran standard (12 kDa) was used to challenge the membranes as direct dialysis solution, instead as permeate in a filtration experiment. The solutions were then analyzed using gel permeation chromatography (GPC).

2. Experimental

2.1. Nanoparticle–polymer dispersions

The herein applied nanoparticle dispersions were prepared as described in detail earlier.^15,31 The PES (Veradel A-201, Dolder AG, Switzerland) and the CA (CA-398-6, Eastman, USA) dispersions were prepared as follows: 10 g of polymer were dissolved in 90 g of N,N-dimethylacetamide (DMAc, ABCR Chemicals, Germany) by stirring overnight. Then, 15 g of calcium carbonate nanoparticles were added (CaCO₃, American Elements, USA) and mixed to reach a particle–polymer ratio of 1.5. The average particle diameter and particle size distribution was determined by SEM. More than 400 particles were counted optically using scanning electron microscopy (SEM, STEM mode, 30 kV, FEI NovaNanoSEM450). This mixture was then shaken vigorously and sonicated for 1 minute at 400 W using an ultrasonic finger (Hielscher UP400 s, Germany) to generate a stable dispersion. The PES TEC dispersion was prepared similarly, except that 3 g of TEC was added before the sonication step. Dispersions were applied to form membranes within 48 h after preparation to prevent the formation of particle agglomerates.

2.2. Membrane preparation

The membranes with approximate dimensions of 10 × 20 cm were produced using a doctor knife. Therefore, roughly 5 mL of the above described dispersion was cast onto a glass substrate. The solvent was subsequently evaporated in a circulating air oven at 80 °C for 10 minutes. Then the polymer–particle composite was given into a bath of 1 M HCl for 1 minute to remove the particle template. The membranes were then washed thoroughly under deionized water and dried on air. Commercial dialysis membranes with similar or slightly larger pore size for comparison were purchased from Millipore and Spectrum laboratories (Millipore V50 and Spectrum Spectra/Por 1000 kDa). These commercial membranes were chosen for comparison since they show the largest average pore size offered by these providers and therefore guarantee the fastest dialysis rate.

2.3. Pore structure and contact angle measurement

The morphology of the membranes was analysed using SEM to determine the pore size distribution. At least 400 pores on the top side of the membrane were optically evaluated (using commercially available imaging software). Cross section images reveal the pore structure inside the membranes. Thus, pieces of each membrane were immersed in liquid nitrogen and subsequently broken to create a clean cross section for later SEM analysis. Hydrophilicity of the membranes was tested by determining the water contact angle of flat, non-porous polymer surfaces prepared from CA, PES and PES TEC. Non-porous surfaces were chosen in order to explain the effect on the material itself, rather than displaying the influence of surface roughness on the contact angle. 150 μL of deionized water were carefully pipetted onto the polymer surfaces that were previously cast on glass slides. Photographs were taken and contact angle was measured optically. The average of 4 replicates (angles measured on both sides of the drop) and the corresponding standard error are displayed in Fig. 2.

2.4. Preparation of dialysis devices

Membrane pieces of each membrane type were glued onto one side of an acrylonitrile butadiene styrene (ABS) ring produced by rapid prototyping (Stratasys UPrint SE Plus). The ring had an outer diameter of 30 mm, an inner diameter of 15 mm and a height of 9 mm giving a volume of 1.6 mL and a dialysis surface area of 1.76 cm². This device was capable of floating on the dialysate (e.g. deionized water) surface while the ring acted as a reservoir for the dialysis solution. The integrated membrane separated the solution from the dialysate.

2.5. Buffer exchange

The herein prepared membranes were tested in a buffer exchange experiment and benchmarked against the two commercially available dialysis membranes. The membranes were used dry, except the PES membrane in one experiment was wetted in ethanol prior to use (PES EtOH). Ethanol is readily miscible in water. It therefore enables fast wetting of an originally hydrophobic membrane. The commercial membranes were pretreated according to the manufacturer's requirements prior to use. The spectrum membrane was soaked in de-ionized water for 30 minutes to remove glycerol and then rinsed thoroughly in de-ionized water, the Millipore membrane was used as received. 1 mL of 1 M potassium permanganate (KMnO₄) was pipetted into the above described dialysis rings that were previously put onto the surface of 1 L of 1 M phosphate buffered saline (PBS). 50 μL of potassium permanganate solution were pipetted out of the dialysis ring every ten minutes over a period of 110 minutes. These samples were then diluted in 1 mL of deionized water and further analysed spectroscopically (Tecan infinite F200, Switzerland) to determine the rate of buffer exchange for each membrane type. Furthermore, pictures of the dialysis progress were taken from every membrane after 10 s and 1 min of dialysis to optically determine wettability and rate of dialysis.

2.6. Dextran recovery test for dialysis membranes

The sample recovery of the membranes was investigated by applying a dextran test formerly known from the characterization of ultrafiltration membranes.^36–38 Commercially available dextran standard molecules (Fluka Analytical, Switzerland) with a weight of 12 kDa were applied. The dextran concentration was 1.0 g L⁻¹ in 1 M aqueous NaNO₃. 1 mL of dextran solution was pipetted onto the different types of membranes (dialysis rings) and dialyzed against 1 L of deionized water over a period of 110 minutes. After this period, the solution was pipetted out of the dialysis ring and compared to the original dextran standard using gel permeation chromatography (GPC). Two PSS (Suprema 100 and Suprema 10 000, 2 × 8 × 300 mm) GPC columns were therefore connected in series and calibrated against the 12 kDa dextran standard. For analysis, a Hitachi L-7000 HPLC system was applied, equipped with an isocratic pump and RI detector, operated at 25 °C with 0.1 M aqueous NaNO₃ as mobile phase at 0.5 mL min⁻¹. Dextran recovery curves were obtained from the GPC data following well-known procedures.^36–38 The GPC curve of the dialyzed solution was compared to the GPC curve of the starting dextran standard used to challenge the membranes. The characteristic rejection of the tested membranes was calculated as the ratio of dextran molecule concentration in the dialyzed solution to dextran molecule concentration in the starting solution.

2.7. Protein adsorption

Protein adsorption to the dialysis membrane surface possibly leads to unwanted loss of the purified target protein. Therefore, the amount of protein adsorption to the membrane during a typical dialysis period is a crucial element of the overall membrane performance. The herein applied membranes were tested for protein adsorption. A solution containing 1 g L⁻¹ bovine serum albumin (BSA, Sigma Aldrich, USA) was used as model protein. 1 mL of the BSA solution was pipetted onto the membrane surface of the above described dialysis ring. After a time period of 110 minutes the solution was pipetted out and compared to the original BSA solution. BSA concentration was determined using high performance liquid chromatography (HPLC, Agilent 1100 series). The applied column was a Zorbax Eclipse 4.6 × 150 mm run at 40 °C, with a flux of 0.75 mL min⁻¹ and injection volume of 5 μL. All experiments were carried out in duplicate. Average values are given and normalized to μg of BSA cm⁻².

3. Results and discussion

3.1. Pore structure and water contact angle

The size distribution of the applied CaCO₃ nanoparticles and the pore size distribution of the herein prepared and commercially available membranes were evaluated in Fig. 1. The PES membrane shows the smallest average pore size (30 nm) and the most narrow pore size distribution of all tested membranes. The addition of TEC to the PES membrane resulted in a moderate increase of the average pore size (39 nm). However, the pore size distribution remained narrow. The CA membrane shows an average pore size of 48 nm, whereas the Millipore V50 membrane exhibits an average pore size of 73 nm and the Spectrum Labs membrane 57 nm, respectively. Fig. 1 clearly depicts that the pore size distribution is generally narrower for membranes prepared by nanoparticle pore templating than for the commercial membranes produced by ordinary phase inversion. The correlation between pore templating particles and the resulting pore structure has been discussed earlier.¹⁵ Narrow pore size distribution of the herein prepared membranes can be explained by a lack of agglomerates in the applied dispersion. The size distribution of the pores was only slightly larger than the size distribution of the templating nanoparticles. This underlines the significance of the availability of templating particles that combine exceptional dispersibility with narrow size distribution. As can be seen from Fig. 1a–d and graphically explained in Scheme 1, the pore morphology seems to be depending on the affinity of the polymer to the template as well. CA tends to form more spherically shaped pores, whereas the PES derived membrane pores seem more angular similar to the shape of the templating particles. Good templating properties of the CaCO₃ – PES system has also been reported earlier.¹⁵ Addition of the TEC plasticizer to the PES membrane resulted in a slight shift of the pore size distribution towards larger pores. However, the overall pore size distribution did not seem to broaden significantly and the angular character of the pores remained. A graphical representation of the pore formation process in dependence of the polymer–particle affinity is given in scheme 1.

Fig. 1 Particle size distribution (a) and pore structure top and cross section views for (b) PES membrane, (c) PES TEC membrane, (d) CA membrane, (e) Millipore V50 and (f) Spectrum Labs 1000 kDa membrane. The black curve indicates a log normal distribution.

Scheme 1 Particle template removal method: nanoparticles are dispersed in a dissolved polymer (a), casting and subsequent drying leads to the formation of a particle–polymer composite (b and c), dissolution step reveals interconnected porous structure (d and e) defined by particle size and particle–polymer affinity.

Cross section views taken by SEM show that the pore structure inside the membranes is determined by the pore templating particles and their affinity to the applied polymer as well. Passage through pores seems to be formed by bottlenecks which represent particle contact areas before template removal was conducted. Good interconnectivity of templating particles is essential for the membrane formation and allows for uniform and homogeneous pore structure from the top to the bottom of the membrane. Lower polymer particle affinity of the cellulose acetate – CaCO₃ templating particle system can again be observed in the looser bulk structure of the CA membrane compared to the PES and PES TEC membranes.

Fig. 2 displays the water contact angles of the non-porous CA, PES and PES TEC flat polymer surfaces. The addition of TEC to the PES membrane led to a decrease of the contact angle by 3° compared to untreated PES (six times larger than standard error). CA polymer surface showed the lowest water contact angle of all three surfaces.

Fig. 2 Membrane contact angle measurements for (a) PES surface, (b) PES TEC surface, (c) CA surface. Standard error is given for each measurement.

3.2. Buffer exchange rate and wettability

The buffer exchange rates were analysed over a time period of 110 min. As becomes apparent in Fig. 3, the untreated PES membrane shows a very slow dialysis rate. After 110 min, only 50% buffer exchange was achieved. This can be directly ascribed to the low wettability of PES. Fig. 1 shows smooth surfaces of all prepared membranes, thus wettability can be attributed to the chemistry of the membrane and the pore size. If pretreated in ethanol (PES EtOH) or modified by TEC (PES TEC), the rate of buffer exchange increased drastically. The slightly faster dialysis rate of PES-TEC over PES EtOH on the other hand has to be attributed to the generally slightly larger pores of the former.

Fig. 3 Buffer exchange rate (1 M KMnO₄ vs. 1 M PBS) for different types of membranes. The grey line depicts 90% of buffer exchanged.

Only the CA, PES TEC and Millipore V50 membrane achieve over 90% buffer exchange within the given time period. This proves that TEC modified PES and CA are well suited membrane materials. However, the ethanol wetted PES membrane and the spectrum membrane achieved only slightly lower dialysis rates. ESI, Fig. S1a–c† display the generally weak wettability of modified and unmodified PES membranes whereas CA, Millipore V50 and Spectrum membranes were readily wetted and buffer exchange started within only 10 s (see ESI, Fig. S1d–f†).

3.3. Membrane sample recovery and protein adsorption

Besides a fast dialysis rate, sample recovery and protein adsorption are key factors concerning the performance of dialysis membranes. Protein or antibody sample recovery should be as high as possible to prevent loss of precious target molecules. Hence, undesired protein adsorption to the membrane surface due to affinity or hydrophobic interactions should be as low as possible. Evaluation of sample recovery of the applied membranes was performed by implementing a variation of the well-known dextran rejection profile test.³⁶ Here, the dextran standard molecule was not filtered through the membranes but used as model compound in the dialysis solution. Since dialysis is basically a pressureless process, this variation simulates membrane sample recovery more accurately. The results of the sample recovery for each of the tested membrane are summarized in Table 1.

Table 1 Detailed parameters of each membrane

Membrane	Buffer exchange after 110 min	12 kDA dextran recovery	BSA adsorption [μg cm^−2]	Average pore diameter	Water contact angle
PES	51%	n.a.	0	30 nm	74°
PES EtOH	86%	89%	n.a.	30 nm	n.a.
PES TEC	92%	91%	0	39 nm	71°
CA	99%	88%	0	48 nm	59°
Millipore V50	97%	85%	85	73 nm	n.a
Spectrum 1000	88%	98%	0	57 nm	n.a.

---

The sample recovery of the 12 kDa dextran standard molecule was about 90% for all the tested membranes. This is in good agreement with the generally similar average pore sizes and diffusion rates through these membranes. CA (88%), PES (89%) and PES TEC (91%) were very close to one another. The Spectrum membrane showed the highest sample recovery of 98%. This might be attributed to its slightly slower dialysis rate (Fig. 3) and generally low protein adsorption properties as is presented later. The V50 membrane from Millipore seemed to show opposite effects. The low sample recovery of 85% may have been caused by the interaction of a fast dialysis rate (Fig. 3) with adsorptive properties of the nitrocellulose membrane material. The ethanol pretreated PES membrane had a lower dextran recovery than the PES TEC membrane. This may be ascribed to hydrophobic interactions that may have arisen after the wetting effect of ethanol. A model protein adsorption (by BSA) showed little to no protein adsorption (within error range) for all of the tested membranes. Only the Millipore V50 showed a low protein adsorption value of 80 μg cm⁻². A summary of the overall performance of the herein tested membranes is given in Table 1.

4. Conclusions

Following the nanoparticle template removal method, CA was implemented as a membrane polymer and showed superior buffer exchange rates. For comparison a PES membrane was altered by the addition of a citrate plasticizer. The addition of 30 wt% of TEC reduced the water contact angle by more than 3°. This improved wettability and turned the PES TEC membrane into a readily self-wetting dialysis membrane. Besides, this chemical modification only led to slight changes in the pore structure. Therefore, formerly reported undesired structure changes by additive modifications of membranes produced by the phase inversion process could be overcome by nanoparticle pore templating. This might be of fundamental interest for the filtration community. Rendering filtration membranes hydrophilic at unchanged performance by the addition of environmentally friendly TEC could be used to address membrane fouling issues more efficiently in the future. To our knowledge, the self-wetting and hydrophilizing effect of TEC is presented here for the first time. Additionally, the here presented membranes show narrower pore size distributions than commercial membranes produced by phase inversion. In principle, the uniformity of the pores is mostly dependent on the particle size distribution and could be significantly increased if appropriate particles would be used as templates. Furthermore, a variation of the generally accepted dextran rejection test was applied on the herein prepared membranes and on two commercially available membranes. This dextran recovery analysis proved to meet the circumstances of the specific dialysis separation process and might therefore become a valid future tool for dialysis membrane characterization.

Conflict of interest

Three authors declare personal financial interest. C.R.K. and W.J.S. declare financial interest in the form of a patent application licensed to Novamem Llc, of which C.R.K, C.M.S. and W.J.S are shareholders. The other involved authors have no competing financial interests.

Acknowledgements

We would like to thank the group of Prof. Dr M. Morbidelli for their help with the GPC experiments.

References

1. R. van Reis and A. Zydney, Curr. Opin. Biotechnol., 2001, 12, 208–211.
2. S.-A. Lee, K.-H. Choo, C.-H. Lee, H.-I. Lee, T. Hyeon, W. Choi and H.-H. Kwon, J. Ind. Eng. Chem., 2001, 40, 1712–1719.
3. A. G. Fane, C. J. D. Fell and A. G. Waters, J. Membr. Sci., 1983, 16, 211–224.
4. T. Reinard and H.-J. Jacobsen, Anal. Biochem., 1989, 176, 157–160.
5. R. Marusyk and A. Sergeant, Anal. Biochem., 1980, 105, 403–404.
6. M. Torres-Lugo, M. García, R. Record and N. A. Peppas, J. Controlled Release, 2002, 80, 197–205.
7. M. Thomas and A. M. Klibanov, Proc. Natl. Acad. Sci. U. S. A., 2003, 100, 9138–9143.
8. K. Kimmerle and H. Strathmann, Desalination, 1990, 79, 283–302.
9. L. Broens, F. W. Altena, C. A. Smolders and D. M. Koenhen, Desalination, 1980, 32, 33–45.
10. C. S. Tsay and A. J. McHugh, J. Polym. Sci., Part B: Polym. Phys., 1990, 28, 1327–1365.
11. Materials Science of Synthetic Membranes, ed. D. R. Lloyd, American Chemical Society, 1985.
12. H. Chae Park, Y. Po Kim, H. Yong Kim and Y. Soo Kang, J. Membr. Sci., 1999, 156, 169–178.
13. D. R. Lloyd, K. E. Kinzer and H. S. Tseng, J. Membr. Sci., 1990, 52, 239–261.
14. M. Ulbricht, Polymer, 2006, 47, 2217–2262.
15. C. R. Kellenberger, N. A. Luechinger, A. Lamprou, M. Rossier, R. N. Grass and W. J. Stark, J. Membr. Sci., 2012, 387–388, 76–82.
16. S. Y. Yang, J. Park, J. Yoon, M. Ree, S. K. Jang and J. K. Kim, Adv. Funct. Mater., 2008, 18, 1371–1377.
17. H. Wang, T. Yu, C. Zhao and Q. Du, Fibers Polym., 2009, 10, 1–5.
18. J.-J. Qin, F.-S. Wong, Y. Li and Y.-T. Liu, J. Membr. Sci., 2003, 211, 139–147.
19. T. Gullinkala and I. Escobar, J. Membr. Sci., 2010, 360, 155–164.
20. M.-L. Luo, J.-Q. Zhao, W. Tang and C.-S. Pu, Appl. Surf. Sci., 2005, 249, 76–84.
21. M. Sun, Y. Su, C. Mu and Z. Jiang, J. Ind. Eng. Chem., 2009, 49, 790–796.
22. G. Wu, S. Gan, L. Cui and Y. Xu, Appl. Surf. Sci., 2008, 254, 7080–7086.
23. A. Gugliuzza and E. Drioli, Eur. Polym. J., 2004, 40, 2381–2389.
24. R. A. Damodar, S.-J. You and H.-H. Chou, J. Hazard. Mater., 2009, 172, 1321–1328.
25. G. Wulff, Angew. Chem., Int. Ed., 1995, 34, 1812–1832.
26. A. Wilke and J. Weber, J. Mater. Chem., 2011, 21, 5226–5229.
27. A. Thomas, F. Goettmann and M. Antonietti, Chem. Mater., 2008, 20, 738–755.
28. K. Jiao, C. L. Graham, J. Wolff, R. G. Iyer and P. Kohli, J. Membr. Sci., 2012, 401–402, 25–32.
29. H. Y. Liu, L. L. Liu, C. L. Yang, Z. H. Li, Q. Z. Xiao, G. T. Lei and Y. H. Ding, Electrochim. Acta, 2014, 121, 328–336.
30. A. C. C. Rotzetter, C. R. Kellenberger, C. M. Schumacher, C. Mora, R. N. Grass, M. Loepfe, N. A. Luechinger and W. J. Stark, Adv. Mater., 2013, 25, 6057–6063.
31. C. R. Kellenberger, S. C. Hess, C. M. Schumacher, M. Loepfe, J. E. Nussbaumer, R. N. Grass and W. J. Stark, Roll-to-Roll Preparation of Mesoporous Membranes by Nanoparticle Template Removal, J. Ind. Eng. Chem., 2013, 53, 9214–9220.
32. K.-V. Peinemann and S. Pereira Nunes, in Membranen, Wiley-VCH Verlag GmbH & Co. KGaA, 2006,pp. 1–21,  DOI:10.1002/3527609008.ch1.
33. H. Liu and C. Tang, Polym. J., 2006, 39, 65–72.
34. R. Riley, H. Lonsdale, C. Lyons and U. Merten, J. Appl. Polym. Sci., 1967, 11, 2143–2158.
35. H.-M. Park, M. Misra, L. T. Drzal and A. K. Mohanty, Biomacromolecules, 2004, 5, 2281–2288.
36. G. Tkacik and S. Michaels, Nat. Biotechnol., 1991, 9, 941–946.
37. A. L. Zydney and A. Xenopoulos, J. Membr. Sci., 2007, 291, 180–190.
38. S. R. Wickramasinghe, S. E. Bower, Z. Chen, A. Mukherjee and S. M. Husson, J. Membr. Sci., 2009, 340, 1–8.
39. P. Mulherkar and R. van Reis, J. Membr. Sci., 2004, 236, 171–182.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c4ra12613k

---