Separation, Immobilization, and Biocatalytic Utilization of Proteins by a Supramolecular Membrane

Membrane separation of biomolecules and their application in biocatalysis is becoming increasingly important for biotechnology, demanding the development of new biocompatible materials with novel properties. In the present study, an entirely noncovalent water-based material is used as a membrane for size-selective separation, immobilization, and biocatalytic utilization of proteins. The membrane shows stable performance under physiological conditions, allowing filtration of protein mixtures with a 150 kDa molecular weight cutoff (∼8 nm hydrodynamic diameter cutoff). Due to the biocompatibility of the membrane, filtered proteins stay functionally active and retained proteins can be partially recovered. Upon filtration, large enzymes become immobilized within the membrane. They exhibit stable activity when subjected to a constant flux of substrates for prolonged periods of time, which can be used to carry out heterogeneous biocatalysis. The noncovalent membrane material can be easily disassembled, purified, reassembled, and reused, showing reproducible performance after recycling. The robustness, recyclability, versatility, and biocompatibility of the supramolecular membrane may open new avenues for manipulating biological systems.


Introduction
Noncovalent materials are adaptive, unlike most conventional covalent systems, rendering them advantageous for a variety of applications. [1][2][3][4][5][6][7][8][9][10][11][12] They are particularly useful for biotechnology, where biocompatible, stimuli-responsive, and self-healing materials have been utilized as artificial tissues, drug delivery devices, and matrices for encapsulation and controlled release of cells. [13][14][15][16][17][18][19] Yet, a number of important biotechnological applications require high material stability. For instance, membrane filtration of biomolecules employs nanoporous materials that should be stable for long periods of time under the pressuredriven flux of solvent, in the presence of buffer, and at physiological pH and salt concentrations. Membrane filtration is an essential tool in the biotechnological industry and appears to be particularly useful for the purification and concentration of proteins. [20][21][22][23][24][25][26][27] Moreover, membranes can be used for immobilization and biocatalytic utilization of enzymes. As enzymes catalyze reactions under very mild conditions, exhibiting efficiency and selectivity largely unmatched by synthetic catalysts, such membrane reactors are emerging components in new, environmentally friendly industrial processes (heterogeneous biocatalysis), which may supplement or replace traditional chemical methods. [28].
Due to high stability demands, virtually all filtration membranes used today are based on polymers or ceramics. [29][30][31][32] However, while providing remarkable stability, the irreversible bonding in these systems sets limits to the scope of possible adaptive changes, including self-healing, response to external stimuli, and controlled membrane disassembly. In this respect, functionalization of polymer-based membranes with polymeric hydrogel coatings enabled stimuli-responsiveness, for instance temperature-or pHinduced shrinkage/swelling. [33] Importantly, when dealing with biological samples, adaptive changes in the membrane (e.g., for controlled release of biomolecules) should take place under mild conditions to preserve the biomolecules' structure and function.
Entirely supramolecular membranes can be very advantageous due to the adaptive character of noncovalent bonding. While supramolecular arrays have been used to template the formation of covalent membranes, [34][35][36][37] only a few supramolecular membranes have been reported that do not require structural reinforcement by covalent crosslinks in the active layer. [38][39][40][41][42][43] For instance, Lu et al. have modified polymer membranes with self-assembling amphiphiles in order to achieve unique rejection properties. [39] Dankers et al. have electro-spun self-assembling molecules to form membranes that functioned as matrices for renal epithelial cells; [43] the performance of this system in pressure-driven filtration was not an object of the study. Others have combined the well-established phase inversion membrane fabrication with self-assembly procedures to create block copolymer micelle-based membranes, [40] some of which exhibit tunable porosity [41] and self-healing properties. [42] However, these membranes employ conventional high molecular weight polymers, and those applied to biological systems require elaborate modifications of the self-assembled material prior to use. [35,44] To the best of our knowledge, robust biocompatible filtration membranes based entirely on noncovalent bonds between small amphiphilic molecules and operating under physiological conditions have not been reported.
Hydrophobic interactions [45][46][47] between large nonpolar groups of amphiphilic molecules in aqueous solution can be remarkably strong, driving self-assembly towards very stable supramolecular systems. [3,[48][49][50] Based on this idea, we have previously designed bolaamphiphile PP2b [51] (Figure 1). Its extended hydrophobic core contains two perylene-3,4,9,10-tetracarboxylic acid diimide (PDI) [52] units. PP2b self-assembles in aqueous media into a robust three dimensional (3D) fibrous network, resulting in a stable and multiple-stimuli-responsive material. Recently, we have fabricated PP2b-based supramolecular ultrafiltration membranes capable of size-selective separation of metal and semiconductor nanoparticles. [38] The membranes are readily prepared within 15 minutes via one-step deposition of an aggregated PP2b solution on a microfiltration support. This simple, fast and robust method represents a significant advantage over most fabrication techniques for covalent membranes, which are time-consuming and often require meticulously controlled conditions. Owing to its noncovalent nature, the material is easily disassembled by organic solvent (e.g. ethanol), the retained particles are released, and the membrane material itself can be recycled and reused multiple times.
Are hydrophobically stabilized noncovalent materials such as PP2b membranes suitable for biotechnological applications that require robustness? In general, hydrophobic groups can interfere with biomolecules by affecting their stability or by forming hydrophobic surfaces that will stick to proteins. Also, presence of high salt concentrations may significantly alter the structure of supramolecular assemblies, [53,54] and weak hydrophobic assemblies in particular. Nevertheless, strong noncovalent bonds are sometimes adequate to produce very robust supramolecular arrays. [55][56][57][58] In this respect, the PP2b assembly is based on very strong hydrophobic interactions, preventing exposure of the hydrophobic moieties to bulk water. [1,59] It is also enclosed by a shell of polyethylene glycol (PEG) groups (Figure 1), [51] which are known to preserve the native structure of proteins and resist undesired biomolecule adsorption (the latter would result in rapid membrane fouling). [60][61][62] Thus, in water, the PP2b-based network is potentially both robust and biocompatible, while retaining recyclability due to its noncovalent nature.
The present study demonstrates how the PP2b-based membranes can be employed for the manipulation of proteins. First, protein filtration experiments were carried out to characterize the membranes' performance in bioseparation of protein mixtures, and to determine its molecular weight cutoff. Next, its biocompatible properties were utilized towards immobilization of enzymes that were used in heterogeneous biocatalysis. Such membrane/immobilized enzyme arrays bear great potential to carry out biocatalytic reactions under continuous flow of reactants (i.e. the enzyme's substrate) that are converted ''en passant'' into a desired product, combining biocatalytic reaction and catalyst removal in a single step. [63][64][65][66] As described herein, the PP2b Figure 1. Illustration of membrane fabrication and recycling. Molecular structure of PP2b, schematic depiction of supramolecular fibers, 3D network, and membrane. Hydrophobic groups of PP2b are located in the fibers' core (red), whereas their hydrophilic PEG shell (blue) provides a biocompatible interface. Recycling of the membrane is achieved by disaggregation or physical removal of the supramolecular layer from the support, followed by purification, and subsequent reassembly in aqueous solution. doi:10.1371/journal.pone.0063188.g001 membranes are versatile for the manipulation of proteins, allowing recycling of the membrane material and retrieval of proteins. To the best of our knowledge, this is the first report of a noncovalent membrane that performs separation, immobilization and biocatalytic utilization of proteins, thus demonstrating the great potential of hydrophobically stabilized noncovalent materials in biotechnology.

Fabrication and Characterization of PP2b Supramolecular Membranes
The membranes used in this study were fabricated by depositing 1 ml of an aggregated solution of PP2b (0.5 mM) in water on a cellulose acetate (CA) support (0.45 mm pore size; effective filtration area: 5.7 cm 2 ), analogously to the reported procedure (see Materials and Methods). [38] Hence, each supramolecular membrane was composed of 1.5 mg PP2b. Importantly, the membranes remain robust under physiological conditions, e.g. aqueous NaCl (150 mM), and 3-(N-morpholino) propanesulfonic acid (MOPS) buffer solution, as revealed by monitoring the solution flux through the membrane over time ( Figure 2). The flux was virtually identical for different aqueous solutions and remained constant during the experiment (30 min), indicating that neither membrane disassembly nor significant compression of the material took place. Fulfilling this first requirement ensured that the membranes are sufficiently stable for typical laboratoryscale filtrations of biological systems. During the course of our experiments, supramolecular membranes showed reproducible performance for filtration and sustained continuous solvent flux for up to 6 hours (see below). The flux of 20 Lh 21 m 22 at 0.8 bar transmembrane pressure was somewhat lower than usual commercial ultrafiltration membranes with similar rejection properties, which exhibit typical fluxes of 50-100 Lh 21 m 22 . [67] The membrane thickness was approximately 6 mm, as revealed by cryogenic scanning electron microscopy (cryo-SEM) ( Figure 3). Higher magnification of its cross-section indicates that the material retains its nanoporous structure under physiological conditions ( Figure 3B, inset), which is critical for size-selective bioseparations.

Separation Performance
A mixture of six purified proteins was used to carry out filtration experiments: 1. N-terminal domain of EIIBCA-Bgl residues 2-84 (EIIBCA), 2. In silico designed Kemp eliminase (KE70), [68,69] 3. L-carnitine dehydratase (LCD), 4. L-Fuculose-1-Phosphate Aldolase (Aldolase), 5. Citrate Synthase (CS), 6. Bovine Serum Albumin (BSA). This protein mixture represents a broad range of masses, making it suitable for characterization of the filtration cutoff. The proteins were dissolved in MOPS buffer solution (MOPS, 20 mM; KCl, 70 mM; MgCl 2 , 10 mM; pH = 7.0) with an overall protein concentration of 1.8 mg/ml (0.3 mg/ml for each protein). In order to evaluate the actual size of the individual proteins in solution, they were each analyzed by gel filtration chromatography (GFC) ( Figure S3 in Text S1), and their hydrodynamic diameters (D h ) were determined using dynamic light scattering (DLS) ( Figure  S4 in Text S1). We note that not all proteins in the mixture were monomeric; CS exhibited a molecular weight of approximately 288 kDa (from GFC), corresponding to its well-known hexameric form (301 kDa), [70,71] Aldolase was a tetramer, [72] and the commercially obtained BSA was oligomeric ($400 kDa, determined by GFC). These findings are consistent with DLS measurements, showing a size increase in the order EIIB-CA,KE70, LCD,Aldolase,CS,BSA (Table 1).
Filtration experiments were performed employing a flow of the protein mixture (1.5 ml) in aqueous MOPS buffer solution through a freshly prepared supramolecular membrane at 0.8 bar transmembrane pressure. As the feed solution was passing the membrane, the filtrate was collected (1.5 ml, F1). Subsequently, additional clean buffer solution (6 ml) was filtered to rinse out remaining proteins from the membrane, and another four fractions were collected successively (461.5 ml, F2-F5). The dead volume of the filter chamber was 0.8 ml. UV/Vis spectra of all fractions were recorded ( Figure 4B, Figure S7 in Text S1). Relative protein concentrations for the collected fractions were quantified via absorbance at 280 nm. The fractions did not exhibit any absorption feature in the visible spectrum, showing that no detectable amount of PP2b (l max,1 = 393 nm, l max,2 = 538 nm) was leaching out of the membrane during filtration. The first two filtrate fractions (F1 and F2) contained considerable amounts of protein; F3 contained only minor amounts, whereas F4 and F5 were virtually protein-free ( Figure 4C). Altogether, 3864% of the proteins passed the membrane. In a control experiment, the protein mixture was also passed through the pristine CA membrane without the PP2b layer. As confirmed by UV/Vis spectra of the filtrate, the CA membrane itself did not retain any proteins ( Figure 4A), showing that the retention is due to the supramolecular membrane only and not caused by adsorption on the support. Importantly, proteins that were retained on the supramolecular membrane could be partially recycled by dispersing the used membrane in buffer solution (using a vortex mixer), followed by removing the PP2b supramolecular material via centrifugation. The resulting supernatant contained retained proteins while the pellet contained PP2b together with some proteins that could not be retrieved. The mechanical dispersion of the supramolecular membrane is enabled by its reversible noncovalent bonding, representing a mild alternative to common release mechanisms in covalent polymers, such as pH and temperature-induced shrinkage/swelling. [32,33] Notably, the enzymatic activity of proteins recycled from the PP2b membrane is fully retained (see below).
The filtration experiment results were analyzed by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) ( Figure 5A). Individual protein concentrations in filtrate and retentate were estimated densitometrically from the gel (Figures   S5, S6 in Text S1). Filtration experiments were repeated five times using independently deposited membranes, which showed consistent separation performance ( Figure 5, Figures S6, S7 in Text S1). Figure 5C shows the retention (the percentage of proteins filtered from the solution) as a function of the molecular weight. It is known for the PP2b membrane that retained particles are captured in the interior of the membrane rather than on the surface (depth filtration). [38] In depth filtration, the average observed pore sizes of the membrane are usually an order of magnitude larger than the cutoff value, since particles are captured at constrictions within the membrane or by adsorption on pore surfaces. [73] The retention of proteins in the PP2b membrane is clearly size-dependent and follows a typical sigmoid curve, indicating that size-selective capture (e.g. through mechanical sieving) plays a major role in the filtration process, rather than specific (e.g. electrostatic) protein adsorption. However, adsorption might play a minor role, for instance in the retention of small amounts (,10%) of EIIBCA. The molecular weight cutoff, defined as the theoretical weight of a molecule with 90% retention, was determined from the retention curve ( Figure 5C), as 150 kDa. In terms of hydrodynamic diameter, the membrane cutoff is 8 nm (Figure 5D), which is somewhat larger than the previously reported 5 nm cutoff for spherical gold nanoparticles. [38] Yet, the latter value refers to the nanoparticles' rigid core, which does not include organic capping agent, counter ions, and hydration shell. The available protein crystal structures [69,70,72] reveal that the actual protein dimensions are significantly smaller than the corresponding hydrodynamic diameters ( Figure 5C, Table 1). For instance the smallest dimension of KE70 is only 3.8 nm, enabling it to pass the membrane, whereas the smallest dimension of Aldolase is 5.5 nm, resulting in .90% retention. Thus, good agreement with previous nanoparticle filtration experiments is observed. [38] We note that size-selective retention in a membrane can depend on shape and deformability of filtered particles. [73]. Figure 5B details the concentration of the proteins in each of the filtrate fractions and that of the retained proteins recycled from the membrane. Notably, permeation of the three smaller proteins EIIBCA, KE70, and LCD, showed a size-dependent trend: The smallest (EIIBCA) passed the membrane with a short time delay and was completely collected in F1 and F2, whereas the larger (KE70) exhibited a longer delay, and small amounts were detected in F3 as well. Finally, the largest of these three (LCD) was found in considerable amounts in F3, and its traces were detected in F4. For the retained proteins, the efficiency of recycling was not equal. For example, about 66% of CS was regained from the filter, whereas only 25% of retained Aldolase could be recycled, possibly due to aggregation or irreversible entrapment within the supramolecular fibers of the membrane material. The membrane itself can be disassembled in water/ethanol (2:3, v/v) and cleaned from contaminations via chromatography over a short silica column (see Figure S2 in Text S1, showing the 1 H-NMR of PP2b before and after recycling). It is easily re-assembled and re-used as a membrane, having similar thickness and identical separation performance as compared to the membranes prepared from non-recycled PP2b ( Figure S8 in Text S1).
Facile recycling and reproducible separation performance after recycling can be regarded as a direct result of the noncovalent nature of the membrane material: its nanoscopic structure (i.e. the three dimensional network of supramolecular fibers) [51] is encoded at the molecular level, allowing simple self-assembly and deposition procedures.
The cutoff value of the supramolecular membrane is in the upper range of commonly used ultrafiltration membranes in biotechnology, allowing the retention of large proteins, nucleic acids, lipids and other large lysate components. [20,21,23] We emphasize that the described filtration characteristics can be particularly useful for the rapid separation of protein monomers from protein aggregates. In order to specifically demonstrate monomer/aggregate separation, oligomeric BSA was mixed with specially prepared monomeric BSA and the mixture was filtered over the PP2b supramolecular membrane. Due to their small size, the protein oligomers (D h , 12.8 nm) cannot be removed by a standard desktop centrifuge (20800 g). However, as revealed by GFC, the supramolecular membrane efficiently removed oligomeric BSA from the mixture, resulting in a filtrate of pure BSA monomer ( Figure 6).

Biocatalytic Utilization of the Supramolecular Membrane
In order to examine whether the filtration process affected protein function, we tested the enzymatic activity of KE70 at 25uC before and after filtration. Using a standard assay, [68] the activity of the enzyme solution before filtration, in the filtrate, and in neat buffer solution (background reaction) was determined by measuring the kinetics of the enzyme-catalyzed isomerisation of 5-Nitrobezisoxazole via absorbance of the product at 380 nm (OD 380 ) (Figure 7, Figure S10 in Text S1). Due to dilution with residual buffer solution in the filter chamber, the concentration of KE70 in the filtrate was 90% with respect to the feed solution. In comparison, its relative activity was 88%, revealing a nearquantitative (,98%) conservation of activity (see section S10, Figure S9, and Table S1 in Text S1).
In addition to KE70, we also investigated the enzymatic activity of hexameric CS after filtration. In contrast to KE70, CS is completely retained and can be recycled from the membrane. In case of retained enzymes, their structure and function might be affected by the extensive contact with the supramolecular material, and by the subsequent recycling procedure. CS activity before filtration and after recycling from the membrane was quantified at 25uC according to an activity assay based on a literature procedure: [74] measuring the kinetics of Citrate formation from Oxaloacetate and Acetyl Coenzyme A (Acetyl-CoA) by detecting the indicator of the reaction, 2-nitro-5-thiobenzoate (TNB) at 412 nm (OD 412 ) ( Figure S11 in Text S1). Taking into account changes in concentration (determined via Bradford test), [75] the kinetic measurements revealed quantitative retention of enzymatic activity of CS, underscoring the membrane's biocompatibility and the mild conditions of enzyme release from the membrane (see section S11, Tables S2, S3 in Text S1).
Our experiments show that filtration of large enzymes results in their immobilization within the supramolecular network, keeping them exposed to the flux of solvent and solutes. Such an array (membrane/immobilized enzyme) may represent a versatile system to carry out biocatalytic reactions in a heterogeneous manner. Addressing this concept, we investigated the enzymatic activity of membrane-immobilized b-Galactosidase (b-Gal, 465 kDa). b-Gal is significantly larger than the membrane cutoff, resulting in near-quantitative retention. It is ubiquitous in nature and widely utilized in molecular biology, catalyzing the hydrolysis of the glycosidic bond of b-Galactopyranosides. We note that b-Gal and many other hydrolases do not require presence of coenzymes, making them the most relevant enzyme class for biotechnological applications in industry, [28] e.g., b-Gal has wide application in the dairy industry for the production of low-lactose milk. [28].
For immobilization, 1.5 ml of b-Gal (0.2 mg/ml) was filtered over a freshly prepared supramolecular membrane, and rinsed with 7.5 ml of clean buffer solution. Once the enzyme was retained on the membrane, the solution flux slightly dropped from 20 L h 21 m 22 to 17 L h 21 m 22 . Subsequently, a typical activity assay solution [76] containing the substrate o-Nitrophenyl b-Dgalactopyranoside (ONPG, 0.05 mg/ml) was passed through the membrane. As the colorless ONPG feed solution passed through the membrane it turned yellow, indicating the b-Gal-catalyzed conversion of ONPG into o-Nitrophenol (ONP, l max = 420 nm) ( Figure 8A). In order to quantify the conversion and to study performance under constant substrate flux for prolonged periods of time, the filter was connected to a UV/Vis flow cuvette, and the absorbance of the filtrate at 420 nm was recorded (Figures S14, S15 in Text S1). Importantly, the reaction showed stable conversion over several hours of uninterrupted substrate flow with an average yield of ,90% ONP ( Figure 8B).
Small amounts of enzyme were observed to leach out of the membrane over time, as expected in immobilization that does not involve covalent attachment of the enzyme to the support. [66] However, enzyme leaching was very low (decreasing from 0.24% to 0.046% activity with respect to the original enzyme solution; see section S15, Figure S16, Table S5 in Text S1) and did not influence the overall reaction yield. Emphasizing its remarkable robustness and biocompatibility, the supramolecular membrane/ embedded enzyme array sustained stable operation under a constant flux of solution for overall 6 hours (incl. preparation, rinsing steps, and test of enzyme leaching), out of which 3 hours were dedicated to biocatalytic substrate conversion.
In an additional experiment, hexameric CS was immobilized and tested for its biocatalytic activity. While b-Gal facilitates the degradation of a complex molecule (ONPG) into smaller parts (Galactose, ONP), CS represents a distinctly different case: a simple precursor (Oxaloacetate) is converted into a more complex product (Citrate), requiring the presence of a coenzyme (Acetyl-CoA). CS immobilization was performed by simply filtering 2 ml enzyme solution (0.3 mg/ml) over a freshly prepared supramolecular membrane, followed by rinsing with 10.5 ml clean buffer solution. We employed a common assay solution [74] to test activity of immobilized CS. When running the mixed solution of Oxaloacetate, Acetyl-CoA, and 5,59-Dithiobis(2-nitrobenzoate) (DTNB) through the membrane, biocatalytic formation of Citrate was indicated by the yellow color of the filtrate, which resulted from the subsequent reaction between HS-CoA and DTNB, forming TNB (l max = 412 nm) ( Figure 9). Thus, conversion of Oxaloacetate into Citrate under constant flow of substrate and coenzyme was readily achieved. Similar to b-Gal, leaching of CS was low (,0.23% activity with respect to the original enzyme solution; see section S13, Figures S12, S13, Table S4 in Text S1).
Overall, immobilization on the supramolecular membranes is versatile and applicable to various enzymes provided that they are large enough to be retained. In order to extend the method to smaller enzymes, one can use standard procedures to enlarge the enzyme without affecting its activity, either by fusing the enzyme to a large inactive protein, or by formation of cross-linked enzyme aggregates (CLEAs) prior to filtration. [64,66,77] Since the design of effective immobilization techniques has been described as one of the main obstacles for industrial-scale biocatalysis, [78] we note that enzyme entrapment in the supramolecular membrane is accomplished in a simple filtration step within 20 minutes. The reactant conversion can be controlled by adjusting the amount of enzyme deposited on the supramolecular membrane. Having a substantial thickness of ,6 mm the noncovalent matrix functions as a depth filter, allowing very high enzyme loading (e.g. 0.4 g  enzyme/1 g PP2b membrane) without membrane clogging. In contrast, most commercial ultrafiltration membranes retain particles on their surface (screen filtration), which enables high solution flux but limits the membranes' capacity for loaded enzymes. [73] Since biocatalysis in membrane reactors takes place within the short time of the substrate's passing through the membrane layer, high enzyme loading is important in order to achieve satisfying reaction yields. Heterogeneous biocatalysis also offers an elegant method to facilitate more complex cascade reactions, if two or more enzymes are immobilized in a sequential manner. [66,79] As membrane fabrication and enzyme immobilization are carried out by simple deposition steps, preparation of complex layered structures with alternating membrane/enzyme arrays [79] can be feasible.
Employing a fully noncovalent membrane reactor for enzyme biocatalysis has several advantages: no synthetic modification or covalent attachment of the enzyme to a stationary phase is necessary. Moreover, enzymes can be easily retrieved from the membrane by disassembly of the noncovalent membrane material, which is important considering the high cost of enzyme synthesis. These factors represent a significant advantage over the commonly employed covalent attachment of enzymes to a support, which is a slow and complex (multistep) process, can affect enzyme activity, and prohibits simple enzyme retrieval. [28].

Conclusions
In summary, we have demonstrated the application of a noncovalent membrane for separation, immobilization and biocatalytic utilization of proteins. To the best of our knowledge, this is the first example in which these functions were achieved using an entirely noncovalent water-based material. The membrane exhibits stable performance under physiological conditions, allowing size-selective separation with a 150 kDa molecular weight cutoff (,8 nm hydrodynamic diameter cutoff). Importantly, separated proteins -both passing and retained -remain functionally active, underlining the biocompatibility of the noncovalent material and separation process. Although the membrane operates under rather low pressure (0.8 bar) and flow rate (20 Lh 21 m 22 ), it has several significant advantages over conventional polymer membranes: 1) Membrane fabrication via deposition of self-assembled fibers is significantly easier and faster than common polymer processing methods. [29] 2) The noncovalent material can be disassembled under benign conditions, e.g. in case of physical damage, clogging, or for intentional protein release. Subsequently, the membrane material can be easily purified, reassembled, and reused multiple times, [38] thus being fully recyclable and cost-efficient. 3) Retained proteins can be retrieved using simple procedures.
Facile membrane fabrication followed by deposition of large proteins represents a simple and fast alternative to covalent enzyme immobilization. Such membrane-immobilized enzymes can be utilized in heterogeneous biocatalysis, which may allow ''green'' industrial processes and development of enzyme-based sensors. The noncovalent nature of the supramolecular membranes is expected to enable further modifications [14] and realization of more complex tandem reactions. Overall, we demonstrate that self-assembled noncovalent materials based on strong hydrophobic interactions are advantageous for biotechnological applications, where a combination of facile fabrication, robustness, and adaptivity provide new opportunities for manipulating biomolecules.

Materials and Methods
Solvents and reagents were purchased from commercial sources and used as received, unless otherwise specified. Synthesis of PP2b was reported previously. [51] For all aqueous mixtures doubledistilled water was used (Barnstead NANOpure Diamond TM water system). Cellulose acetate (CA) syringe filters were purchased from Whatman (Puradisc FP 30/0.45 CA-S). Protein crystal structures were obtained from the PDB (http://www.rcsb.org) and visualized with RasMol (v. 2.7.4.2.). [80,81]. 1 H-NMR spectra were recorded at room temperature on a 300 MHz spectrometer (Bruker).
UV/Vis absorption measurements were carried out on a Cary-5000 spectrometer (Varian).
MALDI-TOF mass spectrometry was carried out using a REFLEX TM reflector time-of-flight instrument with SCOUT TM GFC was performed on an AKTA purifier using an analytical Superdex 200 10/300 GL column (GH Healthcare). The column was equilibrated with buffer containing 20 mM Tris pH 7.5 and 100 mM NaCl. The proteins were injected individually and their migration was compared to commercial molecular weight markers.
DLS was carried out on a Viscotek 802 DLS (Malvern Instruments). Samples were measured in a 12 ml quartz cuvette. Data analysis was performed using OmniSIZE (v. 3,0,0,292). 544 correlator channels were used. The data acquisition time was set to 5-10 seconds. A mass-weighted size distribution was determined from a single correlation function, which was obtained from the average of at least 8 experiments.
SDS-PAGE was performed using 15% gels and stained with Coomassie blue. 5 ml sample buffer was added to 15 ml of each protein sample (0.3 mg/ml) from which 15 ml was applied to each lane. The individual bands were quantified using the QuantityOne volume analysis package (BioRad version 4.4.0). The bands were analyzed densitometrically via the absorbance of UV light integrated over selected band areas. The background absorbance subtraction was set to ''local'', which reduced the error arising from uneven background brightness.
Cryo-SEM sample preparation involved the high pressure freezing (HPF) technique, which is designed to minimize ice crystal formation. [82][83][84][85] For this purpose, a ,161 mm small rectangle was cut out from the membrane and placed inside the inner cavity of an aluminium planchette (size = 3.0 mm (diameter)60.5 mm (thickness); inner cavity = 2.0 mm (diameter)60.15 mm (depth)). The vacant space in the cavity was filled with 1-hexadecene and it was capped with the flat side of another aluminium planchette. HPF was carried out using a Bal-Tec HPM 010 high pressure freezing machine. Subsequently, the sandwich was transferred into a Bal-Tech BAF 060 freeze fracture system where it was fractured with a pre-cooled razorblade and solvent was allowed to sublime (2105uC, 60 min). Subsequently, it was coated with Pt/C, employing double axis rotary shadowing (DARS). Images were taken in a Zeiss Ultra 55 cryogenic scanning electron microscope operated at 1-5 kV with an aperture size set to 10 mm, utilizing an in-lens secondary electron detector.

Protein Mixture for Filtration Experiments
The protein mixture was prepared by dissolving 0.51 mg of each of the following proteins (from stock solutions) in MOPS buffer solution (1.7 ml): EIIBCA, KE70, LCD, Aldolase, CS, BSA (protein concentrations were determined relative to BSA using Bradford reagent [75]). The solution was vortexed and centrifuged for 5 min at 20800 g, in order to assure complete protein dispersion. 1.5 ml of the clear protein mixture was used for filtration experiments within one hour.

Preparation of PP2b with Optimized PEG Length
PP2b was synthesized following our previously reported procedure. [51] Silica column chromatography (eluent: CHCl 3 / MeOH, 0-6% MeOH gradient) was used to fractionate the material into two batches with different average PEG lengths (PP2b(s) and PP2b(l)). As determined by MALDI-TOF mass spectrometry, PP2b(s) (m/z[M+Na + ] = 2973.46, found: 2973.51) is furnished with PEG chains of 18.5 units length on average ( Figure S1A in Text S1) and PP2b(l) (m/z[M+Na + ] = 3150.57, found: 3150.85) carries PEG chains with 20.75 units on average ( Figure S1B in Text S1). While aggregates of pure PP2b(s) in water are not stable over time and begin to precipitate from the solution shortly after preparation, PP2b(l) forms a stable homogeneous aqueous solution, but its supramolecular structures are not suitable as ultrafiltration membrane, since they are not efficiently retained by the CA support membrane. The optimum average PEG-length in PP2b was adjusted by mixing PP2b(s) and PP2b(l). The mixture of 0.5 mg (0.17 mmol) PP2b(s) and 1.0 mg (0.32 mmol) PP2b(l) was found to produce a stable and sufficiently robust supramolecular solution, which was used for fabrication of supramolecular ultrafiltration membranes.

Self-assembly of PP2b
An aqueous PP2b supramolecular solution (0.5 mM) was prepared similarly to the previously reported procedure: [38] In a typical experiment, 26 mg (8.8 mmol) of PP2b(s) and 53 mg PP2b(l) (16.8 mmol) were dissolved in THF (2.56 ml). Then water (48.6 ml) was added, followed by vigorous shaking of the solution and sonication for 5 min. After aging for 2 h, THF was removed by evaporating ,15-20% of the solution at r.t. in high vacuum. The solution was refilled with water to its original volume (51 ml); it was shaken, sonicated for 5 min, and aged for at least 12 h. The mixture was used as a stock solution for preparation of many supramolecular membranes. It was stored at 4uC and could be used for ,2 weeks, producing membranes with consistent separation performance. During storage for prolonged periods of time (.2 weeks), significant amounts of PP2b began to precipitate out of the solution, making it necessary to dry and re-assemble the material as described above.

Fabrication of Supramolecular Membranes
Membrane fabrication was similar to previously reported. [38] PP2b supramolecular solution (1 ml) was diluted in water (2 ml) and the mixture was filtered over a syringe filter (Whatman Puradisc FP 30, CA; effective filtration area: 5.7 cm 2 ; pore size: 0.45 mm). During filtration, PP2b was deposited on the CA support, as seen by the color change from dark purple feed solution to colorless filtrate, and formation of a dark purple layer on the CA support. The transmembrane pressure was created by opening the gauge to a pressurized nitrogen gas cylinder. It was slowly increased to 0.8 bar during deposition of PP2b. Care was taken to avoid air bubbles in the filter chamber, which would hamper the formation of a continuous supramolecular layer. Subsequently, water (2 ml) was filtered through the membrane, which ensures uniform packing of the supramolecular material. [38] In order to adjust pH and ionic strength in the filter chamber to the pH and ionic strength of the protein solution that was to be filtered, neat buffer solution (3 ml) was run through the membrane. The filtrate of that buffer (F0, 1.5 ml) was collected as a reference for UV/Vis spectroscopy. The filter housing was kept filled with buffer solution and the supramolecular membrane was used directly for filtration experiments.

Filtration Experiments
The protein mixture (1.5 ml) was filtered over a freshly prepared supramolecular membrane. Then, 6 ml of protein-free MOPS buffer solution was filtered in order to rinse out residual proteins from membrane and filter housing. Fractions were collected (561.5 ml, F1-F5). To test reproducibility of the filtration outcome, the experiments were carried out five times, each time using another freshly prepared membrane. For consistency, the same transmembrane pressure of 0.8 bar was used in all experiments. The collected fractions were analyzed by UV/Vis ( Figure S7 in Text S1) and SDS-PAGE ( Figure S6 in Text S1). The reported values for protein concentrations and protein retentions (Figure 4, Figure 5) are mean values with error bars representing their standard deviations.

Recycling of Retained Proteins and PP2b
After the filtration experiment, the filter chamber was opened and the PP2b supramolecular layer was scratched off the CA support and suspended in MOPS buffer solution (1.5 ml). It was vortexed for 1 h in order to disentangle the PP2b supramolecular fibers, thus releasing retained proteins (R). In order to separate PP2b from the retained proteins, the solution was centrifuged for 15 minutes at 20800 g (using a desktop centrifuge). The supernatant contained retained proteins while the pellet (P) contained PP2b together with some retained proteins that could not be extracted. Both supernatant and pellet were analyzed separately by SDS-PAGE ( Figure S6 in Text S1), whereby the pellet was highly concentrated due to its small volume. In order to purify PP2b in the pellet from residual proteins and other organic and inorganic contaminations, ten such pellets were dissolved in 20 ml of a water/ethanol mixture (2:3, v/v). Subsequently, PP2b was extracted with dichloromethane (40 ml). The colorless aqueous phase was discarded. The organic phase was dried in high vacuum. The solid was washed with hexane (3650 ml), and with water (1650 ml). Subsequently, it was purified by silica column chromatography, eluting successively with: 1) 200 ml CHCl 3 , 2) 200 ml CHCl 3 /MeOH (98:2, v/v), 3) 200 ml CHCl 3 / MeOH (92:8, v/v). The product contained pure PP2b, as revealed by 1 H-NMR ( Figure S2 in Text S1). Nearly quantitative recycling was achieved.

Supporting Information
Text S1 Further experimental procedures and specifications, Supporting Figures S1-S16, and Supporting Tables S1-S5.