Core – shell graphene oxide – polymer hollow ﬁ bers as water ﬁ lters with enhanced performance and selectivity †

Commercial hollow ﬁ ber ﬁ lters for micro- and ultra ﬁ ltration are based on size exclusion and do not allow the removal of small molecules such as antibiotics. Here, we demonstrate that a graphene oxide (GO) layer can be ﬁ rmly immobilized either inside or outside polyethersulfone – polyvinylpyrrolidone hollow ﬁ ber (Versatile PES®, hereafter PES) modules and that the resulting core – shell ﬁ bers inherits the micro ﬁ ltration ability of the pristine PES ﬁ bers and the adsorption selectivity of GO. GO nanosheets were deposited on the ﬁ ber surface by ﬁ ltration of a GO suspension through a PES cartridge (cut-o ﬀ 0.1 – 0.2 m m), then ﬁ xed by thermal annealing at 80 (cid:1) C, rendering the GO coating stably ﬁ xed and unsoluble. The ﬁ ltration cut-o ﬀ , retention selectivity and e ﬃ ciency of the resulting inner and outer modi ﬁ ed hollow ﬁ bers (HF-GO) were tested by performing ﬁ ltration on water and bovine plasma spiked with bovine serum albumin (BSA, 66 kDa, z 15 nm size), monodisperse polystyrene nanoparticles (52 nm and 303 nm sizes), with two quinolonic antibiotics (cipro ﬂ oxacin and o ﬂ oxacin) and rhodamine B (RhB). These tests showed that the micro ﬁ ltration capability of PES was retained by HF-GO, and in addition the GO coating can capture the molecular contaminants while letting through BSA and smaller polystyrene nanoparticles. Combined XRD, molecular modelling and adsorption experiments show that the separation mechanism does not rely only on physical size exclusion, but involves intercalation of solute molecules between the GO layers. ) and ﬂ uxed volume (250 of were the same for tests. performed by


Introduction
The development of novel membrane materials for purication of uids is of great interest for the fabrication of personalized biomedical treatments (i.e. selective apheresis, dialysis), specic chemical separation (organic solutes from organic matrices), advanced water purication 1 and gas separation technologies. 2 Polymeric membranes are currently exploited at the industrial level for a variety of processes and applications, spanning blood ltration to food/drug purication, and drinking and wastewater purication. 3 The market trend for polymeric membrane ltration modules is growing, and it is expected to further increase in the next few years due to the increasing demand for advanced healthcare treatments and also drinkable water. In general, polymeric membrane ltration modules may be classied into three types, namely plate and frame, spiral wound, and hollow ber (HF) modules. Among them, hollow ber modules are the most used as separation units in industry because of their unique characteristics of self-support, high membrane packing density, and high surface/volume ratio. 4 Compared to planar membranes, the hollow-ber conguration has a much larger membrane area per unit volume of membrane module. The surface to volume ratio is about 300-500 m 2 m À3 for plate and frame modules, 600-800 m 2 m À3 for spiral wound modules, and 6000-13 000 m 2 m À3 for hollow ber modules, resulting in higher productivity. Nowadays, hollow ber congurations are widely used in basically all types of membrane separation, including gas separation, ultraltration, pervaporation, dialysis and supported liquid membrane extraction.
The ltration mechanism of these membranes mainly relies on size exclusion, and the pore size ultimately denes the cut-off range. Microltration is widely used in water treatment and plasma apheresis as a disinfection step since colloidal particles, microorganisms and other particulate material of a size larger than about 200 nm are removed. Ultraltration and nanoltration membrane modules have higher cut-offs of 1-10 nm and 100-200 nm, respectively, thus enabling decontamination of viruses/endotoxins (ultraltration) and low molecular weight molecules (nanoltration), but the throughput is much lower than what is achievable for microltration.
However, there is an urgent market and societal need to improve the removal of emerging organic contaminants (EOC) such as pesticides, pharmaceuticals, or surfactants used in large quantities in civil industrial and farming activities, which are able to contaminate water sources or food and liquids, causing severe environmental and health problems.
Recently, membrane doping with nanomaterials has been reported as a promising strategy to tune the selectivity and enhance the efficiency of polymeric membranes. 5 Among nanomaterials, graphene oxide (GO) is particularly suitable for promoting selective recognition processes due to its intrinsic 2-D conguration, high surface area and abundance of surface chemical groups. For instance, the addition of a small amount of GO in polysulfone-based membranes obtained by phase inversion increased their hydrophilicity and antimicrobial activity, reduced the biofouling, 6 promoted arsenate rejection, 7 and allowed for oil-water separation. 8 In general, graphene containing membranes are receiving increasing attention because they exhibit enhanced separation performance with enormous potential outcomes for ion sieving, desalination and water purication applications. [9][10][11][12] Graphene oxide also has excellent adsorption properties toward EOC (including pharmaceuticals and personal care products) [13][14][15][16] and metal ions, even at very low concentration. 17 This feature has led to the development of 3D structures with removal efficiencies superior to those of other nanomaterialbased adsorbents including Granular Activated Carbon (GAC), the industrial standard, for some metal ions and organic compounds. Moreover, covalent chemical modication of the oxygen-based functionalities of GO allows tuning of the adsorption selectivity of GO-based structures. 18 In this direction, enhanced adsorption of heavy metal ions and organic dyes in water and wastewater have been reported for EDTA, 19 sulfonated 20 and amino-rich 21 graphenes.
Aiming to exploit both the adsorption properties and membrane enhancing effects of GO to develop new multifunctional lters, we recently demonstrated the superior efficiency of GO-doped polysulfone porous structures toward hydrophilic organic contaminants including dyes and drugs. 22 We also described a simple method to x GO on scraps from the production of polysulfone (PSU) ultraltration membranes. 23 The process involves the partial removal under vacuum of water from a GO and PSU suspension, followed by thermal xation. 24 This material showed enhanced removal capability (up to seven times) toward polar organic contaminants (e.g. ooxacin and rhodamine B) thanks to the high hydrophilicity of the GO layer exposed to the surface in contact with water. The coating process allows GO to be xed on the PS surface by means of supramolecular interactions, by exploiting the spontaneous aggregation of GO sheets on PSU and enabling up to 50% of the PSU total surface area to be covered. The lter could effectively capture EOC, but poor retention was obtained for larger chemical moieties with respect to hollow ber-based ultraltration cartridges.
Typically, polymer composites containing graphene or GO are prepared by mixing or co-extrusion, then shaped in the nal form. GO or graphene could in some cases be applied on the surface of simple shapes, such as powders 24 or at sheets; 25 until now, it has however never been possible to apply GO coatings on the surface of nite commercial devices such as lters.
We thus developed a completely different approach, exploiting the ltration capability of commercial hollow ber lters to achieve a uniform coating on a geometrically complex substrate. We could thus obtain GO coatings on polyethersulfone-polyvinylpyrrolidone hollow bers (HF) made of a commercial polymer, Versatile PES® ( Fig. 1a-d), already assembled in a working lter cartridge (commercial name 'Plasmart'). Then, we used these lters for purication of water solutions and showed the possibility to selectively remove small molecules (including two antibiotics of current environmental concern). Stable coating of the outer or inner walls of PES hollow bers with a GO membrane and controlled membrane thickness was achieved by ltration of a GO suspension in dead-end conguration (Fig. 2), followed by thermal xation by annealing in an oven.
The retained microltration capability was assessed for HF-GO by ltering a mixture containing: (1) Nanoscopic objects of different sizes: protein (Bovine Serum Albumin, BSA, M w ¼ 66 kDa) and polystyrene beads (52 nm and 303 nm sizes); (2) molecular EOC contaminants. As realistic test molecules we chose ooxacin and ciprooxacin (two quinolonic antibiotics under monitoring by the EU) and rhodamine B (a textile dye, Fig. S1, ESI †).

GO immobilization and xation
Graphene oxide powder (<35 mesh, purchased from Abalonyx, sheet lateral size about 1 mm, many primary single sheets declared) was suspended in Milli-Q water (2 mg mL À1 ) and sonicated for 4 hours. Then, the GO solution was ltered through commercial HF lters (Plasmart 25, Medica). Each lter was composed of ca. 275 PES bers, with each ber having a length of ca. 4.5 cm, an inner diameter of z280-300 mm and an outer diameter of z360-400 mm. Thanks to the approach used, we could choose to coat the inner surface of the HF (Fig. 2a) or the outer surface ( Fig. 2c), using two different dead-end ltration modalities. Aer ltration of 5 mL of solution containing about 10 mg of GO, the cartridges were kept in oven at 80 C overnight to give HF-GO1i samples, i.e. hollow bers containing about 1% w/w of GO with respect to the PES membrane weight in the inner surface. Hereaer, we will name samples as HF-GO followed by the % of GO loading and the letter e/i, indicating whether the coating is placed on the outer or inner surface of the hollow ber. Following this nomenclature, we repeated the ltration-xation cycle to obtain samples HF-GO1e/i, HF-GO5e/i and HF-GO10e/i, varying the coating from about 1% to 5% and 10%, either on the inner or outer surface.
The HF-GO lters are shown in Fig. 3. For bers coated on the outside wall, the dark color of the coating is clearly visible by increasing the amount of GO loading from 1% to 10%. Fibers with inside GO coating showed no apparent change of colour (Fig. 3b), but a black coating could be observed in the inner wall by cutting the bers (Fig. 3d).
The stability of the GO membrane coating was tested by owing deionized water (1 L) through the cartridges before and aer thermal xation and by performing UV-vis absorption spectroscopy on the ltered water, comparing  the results to what was obtained with calibrated solutions of GO at known concentrations (Fig. S2, ESI †). No evidence of GO nanosheets was found in the ltered water (detection limit 2-5 ppm, Fig. S2, ESI †), conrming that the xation process we already used on the powders is effective on the hollow bers as well. 24 We also performed standard chemical potability tests (certied analysis of salts, metal ions, taste, total organic carbon) on tap water ltered through HF-GO5e/i cartridges, conrming the potability of the ltered water and the absence of any dangerous contaminants, in accordance with current legal limits (D. Lgs. 31/01 Agg. D.M. 14/06/2017, Table S1, ESI †).

Membrane characterization
A combination of optical microscopy, SEM and micro-Raman analyses was carried out to investigate the homogeneity of the coating, while XRD measurements were performed to estimate the periodic stacking in the GO coatings in HF-GO bers and the number of GO layers. Optical microscopy on the HF-GO bers ( Fig. 4a and S3, ESI †) showed a black coating on the whole ber surface. GO coating was not uniform at the lowest GO load (1%), while uniform coating was found for all the other samples. Accordingly, SEM analysis on HF-GOe bers (Fig. 4b and c) showed the presence of a GO layer covering the ber surface. Fig. 4b shows the case of HF-GO5e. Notably, some open (uncoated) pores (about 1 mm size) were also observed ( Fig. 4c and S4, ESI †). This is highly benecial, since it ensures that the membrane is not clogged due to GO coating. Micro-Raman analysis performed on HF-GO1e and HF-GO10e (Fig. S5, ESI †) showed a limited inltration depth of GO within the ber pore channels independent of the amount of GO used for coating. Indeed, in both outer modied HF, at 5 mm from the GO layer on the outer surface wall it was possible to detect Raman peaks of GO that completely disappeared in the bulk of the ber section (at about 25 mm from the external wall, Fig. S5, ESI †). Fig. 5a shows the XRD patterns of HF-GO bers. The bell-shaped proles centred at 18.1 (2-theta) were due to the amorphous PES component. A signal due to the stacked GO nanosheets was observed at about 11.7 (d ¼ 0.75 nm), visible as a shoulder or peak depending on the GO loading, and better evidenced aer data treatment (Fig. 5b). This distance is slightly smaller than that calculated for the pristine GO powder (10.5 , d ¼ 0.84 nm), and was ascribed to partial dehydration during the annealing treatment. 24 The thickness of the stacked crystalline domains was estimated from peak width using the Scherrer equation. 25 The stacked domains of GO had an average thickness of 6-8 layers on all observed bers, indicating that even thicker coatings do not form a continuous, perfectly stacked layer (Fig. 5c and Table S2, ESI †). Rather, the coating is formed by a number of these crystalline nanometre sized regions assembled together in a compact structure of different thicknesses.

Filtration selectivity and efficiency
Water permeability tests were rstly performed on HF-GO cartridges in the same dead-end conguration described in the experimental part. Each cartridge was lled with osmotic water, the pressure value was measured at the lter inlet, the amount of water microltered in 1 minute was weighed and the ltration coef-cient (K f ) was calculated. As expected, the permeability of PES decreased as long as the amount of GO loading increased. We observed the lowest K f of 0.42 AE 0.24 mL min À1 mmHg À1 m À2 for the HF coated with 10% GO on the outside. The ideal K f for ltering tests was obtained for inside coatings of 1% and 5% (Fig. 6).
Besides measuring the permeability of water, we also measured (dry) air permeability, in order to distinguish the contribution of water transport across swollen GO or polymer with respect to the transport in macroscopic pores. 26 The  obtained air permeability measured conrmed the porous structure of the HF, revealing an incomplete coating of the PES HF at the lowest GO concentration, thus featuring several holes; for thicker coatings, a complete impermeable coating could also not be detected, even though a signicantly more compact structure was obtained, and the diffusing molecules have to proceed along a very tortuous path to cross coating, wiggling around the GO layers (see ESI, Fig. S6 for more details †). The cut-off of the PES hollow ber pores used is in the range 0.1-0.2 mm, optimal for microltration of biological samples, blocking colloids and microorganisms of size >1.000 kDa (Fig. S7 and S8 ESI †). To establish the cut-off of the HF-GO bers, ltration tests were performed on water spiked with BSA and polystyrene standard nanoparticles with sizes below and above the cut-off of PES cartridges (i.e. PS NPs, 52 nm and 303 nm sizes). BSA (about 15 nm, 66 kDa) and 52 nm PS NPs are expected to cross a microltration membrane, while 303 nm sized PS NPs are expected to be retained. Fig. 7 shows that all lters blocked larger particles and let through smaller ones, as expected, and the retention of 52 nm PS NPs was basically equal to that of the bare HF modules (about 20%). This indicates that no clogging effect of GO occurred and that there were pores in the range 52-303 nm available for ltration.
A partial retention of BSA was observed in the HF-GOi membranes (up to 15-20%), while no signicant effect was detected from the cartridges with GO coatings on the external surface. This can hardly be attributed to a size exclusion mechanism, and effective nanoltration operation was excluded, in view of the smaller size of BSA with respect to 52 nm PS NPs. Additional BSA and ltration experiments are reported in the ESI (Fig. S7 and S8, † respectively).
Once we veried that the GO coating does not affect the size-dependent ltering performance of the lters, we measured their ability to retain, instead, small contaminant molecules (RhB, OFLOX and ciprooxacin, 5 mg L À1 in water), which cannot be blocked by the standard lters due to their nanometric size, which is much smaller than BSA protein.
We measured the removal efficiency of the lters for such molecular contaminants by uxing through the lter 250 mL of solutions contaminated with each molecule (5 mg L À1 ) at 15 mL min À1 , then analysing the ltered solution by HPLC/UV analysis (details in ESI †).
Uncoated PES lters (GO load 0%) showed insignicant ltering effects for the standard contaminants inspected, as shown in Fig. 8, with removal efficiencies <10% in all cases. Conversely, HF-GO bers showed signicant removal ability, up to about 80% in the best case (Ciproox ltered by HF-GO10i). Removal performance increased with GO loading, showing a monotonous increase for the outercoated bers (Fig. 8a), and a saturation plateau for the inner-coated ones (Fig. 8b). The best performances were reached with a lower amount of GO in inner-coated bers, which showed a signicant removal performance even at low GO loading, i.e. removing 50% of Ciproox vs. about 20% for outer-coated bers at 1% GO loading. At the highest loadings (GO 10% w/w), the performances of the two lters (inner or outer coatings) are equivalent.
Therefore, the mechanism of capture of these substances does not rely on size exclusion, but it rather is given by adsorption onto the GO layer surface, which is able to interact with such molecules. The larger the GO amount on the HF module, the larger the EOC removal. The key aspect of the ltration step is the accessibility of the adsorbing sites in the GO layer, thus allowing these molecules to be intercalated. The more open structure of inner coated lters, as indicated by both water and air permeability tests ( Fig. 6 and S6, ESI † respectively), and their larger porosity promote contaminant transport in the coating with more effective contact with the adsorbing sites. Previous experiments conrmed that small quantities of GO can capture hundreds of mg of OFLOX and RhB (isotherms), ca. 590 mg g À1 for RhB and 360 mg g À1 for OFLOX, but such experiments were always performed in static conguration, with a prolonged (24 h) contact time between GO and the contaminant solution. Kinetics studies of EOC removal showed how equilibrium conditions can be achieved aer different times, 5-10 minutes for RhB 27 and 60-80 minutes for OFLOX; 28 such a timescale is incompatible with continuous ow ltering, for practical applications. In our setup, with a 15 mL min À1 ow and 2.5 mL cartridge volume, the contact time is z10 seconds, thus indicating that the operative conditions of the lters are very far from thermodynamic equilibrium. However, we could achieve signicant removal even if our contact time was one order of magnitude smaller than what is reported in the above cited works. We attribute this remarkable performance to the synergic action of the ber pores and the GO coating, with the solution being forced to pass through the GO coating (see scheme in Fig. 2), thus in an ideal condition for the intercalation and trapping of the EOC molecules in between overlapped GO layers in the GO layers, as described in previous work. 24 Aer demonstrating the improved performance of the HF-GO lters compared to standard PES, we also calculated the lter longevity, i.e. the ability to lter signicant amounts of solution before having to be replaced. The concentration of EOC pollutants is usually in the sub ppb range, thus we estimated our lter consumption using a contaminant concentration of 0.2 mg L À1 for Ciproox. Fig. 9 shows the removal efficiency of the lters as a function of cumulative mass uxed. This plot allows us to estimate the amount of Ciproox removed by a single cartridge and normalise the removed EOC mass on the mass of the active material (GO). We see that HF-GO5i can have a reasonable removal of about 90% with 15 mg g GO À1 of Ciproox, 14 mg g GO À1 of OFLOX and 7 mg g GO À1 of RhB.
This is not the adsorbed amount at equilibrium at high concentration, but the effective amount of EOC adsorbed by the operative lter. It is remarkable that such values are obtained with quite a short contact time (seconds), while the references in water treatment plants are usually 10-20 minutes, with ca. 20 mg g À1 of Ciproox removed by using traditional powdered activated carbon. 29 In turn, the lter HF-GO5i still has a removal efficiency ca. 90% aer the ow of a total mass of 0.5 mg of contaminant, corresponding to ca. 2500 L of realistic contaminated solution with 0.2 ppb of Ciproox; this proves the suitability of HF-GO lters for realistic commercial applications (see also Table S7, ESI †). Fig. 9 compares the removal efficiency of inner coated HF-GOi lters toward the three EOC molecules. As expected, the lters are rapidly saturated with only 1% GO loading, while 5% and 10% loading give the best performances, in

Simultaneous ltration and adsorption test and working mechanism
Eventually, we performed ltration of a complex, realistic matrix of BSA and OFLOX solution, to conrm the capability of HF-GO lters to work simultaneously as a physical lter (cut-off depending on the pore size) and adsorbent (mediated by chemical interactions). The ltration tests were performed with both water (Fig. 10) and bovine plasma matrices ( Fig. S8 and S12, ESI †). As a representative case study, in Fig. 10 we show a HF-GOi lter that features almost quantitative removal of OFLOX (z90%) and negligible (<1%) removal of BSA. Similar results were found in bovine plasma matrix containing BSA and other proteins with a total concentration of about 6-8 g dL À1 , with no signicant reduction of the BSA and TP amounts occurring aer ltration (Fig. S13, ESI †).
The mechanism of ltration was investigated by XRD analysis (Fig. 11a). We extracted the bers from the lters aer ltration of the contaminated solutions, and estimated GO stacking using XRD, comparing them with pristine, unused bers. We observed a shi of the GO peak towards a larger stacking distance aer ltration of EOC (0.75 nm / 0.89 nm), and a decrease in intensity. This suggests that the removal mechanism is due to the strong affinity between the aromatic core of the EOC with the sp 2 structure of the GO layers that induces partial swelling and exfoliation of the layers, as discussed in detail in previous work. 24 Atomistic molecular mechanics simulations may reveal insights into the adsorption and packing of molecules interacting with carbon nanomaterials. 30 Thus, molecular modelling simulations of GO-ooxacin interactions were performed for a deeper understanding of the removal mechanism. First, simulations show that the spacing of 0.75 nm observed for GO (Fig. 11, black curves) can be explained by considering the uptake of a water monolayer in the interlayer space of GO. 31 Indeed, GO preserves the layered structure of graphite, but in contrast to graphite, it is hydrophilic, thus water molecules are hardly completely removed from GO layers. Consequently, spacings of carbon layers in the range from 0.6 to 1.2 nm are observed in GO, depending on the water content of the samples. 32 This water layer is crucial to reach equilibrium spacing between the GO layers because an attractive force between the two negative GO sheets is needed to overcome the electrostatic repulsive force between the GO layers. Intercalation of ooxacin between the GO layers (Fig. 11b) generates an increase in the spacing between the two GO layers from 0.75 to 0.89 nm. These values are in perfect agreement with observed XRD data (Fig. 11a, red proles).
In fact, considering that (i) the size of a water molecule is $0.25 nm and the size of ooxacin is $0.4 nm, (ii) during the intercalation process, the GO is locally dehydrated, an increase of $0.15 nm is expected upon intercalation, as observed experimentally by XRD measurements and by molecular modelling simulations.
Collectively, the multitarget ltration experiments show that the synergic action of the PES membrane and GO coating porosity (micrometric and nanometric pores, respectively) allows the removal from solution of large biological objects and small molecules at the same time. In contrast to simple mixtures or bi-layered GO-polymer composites previously described, 22 the approach  described here allows the nano-adsorbent component (i.e. the GO) to be positioned exactly on the surface more exposed to the solution, in particular in proximity to the nanopores. The results obtained show that this geometry allows a signicant removal of EOC to be obtained even with a short contact time and for low GO loadings, in particular when the GO is placed in the inner surface of the bers, where contact time with the contaminated solution is maximal (see scheme in Fig. 12).

Conclusions
In conclusion, GO coating was achieved by a simple and mild procedure on already commercially available microltration PES hollow ber modules. While unmodied lters could stop large objects and let BSA and small molecule EOCs pass through, only HF-GO hollow ber lters were able to selectively capture three target EOCs of environmental relevance. Air permeation tests revealed that diffusing molecules are forced to travel around the GO sheets along tortuous paths, depending on the in-plane distance between two near GO sheets and the intrinsic aspect ratio of the 2-D materials. This is of course not useful for the ltration of large molecules like BSA, with a size of 20 nm, but could be useful for the selective ltration of smaller molecules such as EOCs. Accordingly, combined XRD analysis of virgin and used membranes and molecular modelling simulations revealed intercalation of organic molecules through the GO layers as the mechanism of adsorption. This work demonstrates that HF-GO modules can be useful for removing antibiotics from water and plasma matrices while letting proteins and nanoobjects pass though the HF pores, thus paving the way toward selective separation processes for biomedical and water treatment applications.

Conflicts of interest
There are no conicts to declare.