Stable Graphene Membranes for Selective Ion Transport and Emerging Contaminants Removal in Water

Carbon‐based materials, such as graphene oxide and reduced graphene oxide membranes have been recently used to fabricate ultrathin, high‐flux, and energy‐efficient membranes for ionic and molecular sieving in aqueous solution. However, these membranes appeared rather unstable during long‐term operation in water with a tendency to swell over time. Membranes produced from pristine, stable, layered graphene materials may overcome these limitations while providing high‐level performance. In this paper, an efficient and “green” strategy is proposed to fabricate µm‐thick, graphene‐based laminates by liquid phase exfoliation in Cyrene and vacuum filtration on a PVDF support. The membranes appear structurally robust and mechanically stable, even after 90 days of operation in water. In ion transport studies, the membranes show size selection (>3.3 Å) and anion‐selectivity via the positively charged nanochannels forming the graphene laminate. In antibiotic (tetracycline) diffusion studies under dynamic conditions, the membrane achieve rejection rates higher than 95%. Sizable antibacterial properties are demonstrated in contact method tests with Staphylococcus aureus and Escherichia coli bacteria. Overall, these “green” graphene‐based membranes represent a viable option for future water management applications.


Introduction
On a global scale, the water resources available to human activities (a mere ≈ 0.03% of the total) are being vastly contaminated due to increasing and inattentive usage. According to the WHO, in 2020, only 74% of the global population (5.8 billion) safely and compositions. [4] Membranes for pressure-driven separation processes (i.e., microfiltration, ultrafiltration, nanofiltration, and reverse osmosis-RO) are typically made of synthetic organic polymers, such as polyethylene, polytetrafluorethylene, polypropylene, and cellulose acetate. Nanofiltration membranes have a pore size in the order of 0.5-2 nm, in between ultrafiltration (2-100 nm) and reverse osmosis (< 0.5 nm). Ultrafiltration membranes are commonly used for bacteria and virus removal. RO membranes mainly serve for desalination but can also efficiently remove small organic molecules and inorganic salts, which are the typical goal of nanofiltration membranes. The rejection rates and permeance values are the critical features of distinction between them. [5,6] Inorganic membranes are commonly made of ceramics, metals, zeolites, or silica. [7] They are chemically and thermally stable and widely used in industrial applications like hydrogen separation, microfiltration, and ultrafiltration.
Polymeric and ceramic-based membranes are quite widespread. Conventional polymeric membranes are rather robust but have typically low chemical resistance; while ceramicbased membranes are more fragile and difficult to produce over large scales at a reduced cost. Toward commercial applications, it is thus crucial to develop nanofiltration membranes with high durability, high liquid permeance, good stability in harsh conditions, and excellent separation efficiency, while keeping the fabrication process simple and affordable. [5] To this end, researchers focused on identifying suitable materials for high-performance membranes, combining a sensible balance between rejection rate, water permeability, and lifetime stability (which directly translate into economic viability). Additionally, marked antibacterial properties are also central for membranes with a stable filtration performance over time in practical applications. [8][9][10] Having high thermal, chemical, and mechanical stability, graphene-based materials, such as graphene oxide (GO) and reduced graphene oxide (rGO), have been widely explored for water filtration. [11,12] Permeability studies of the GO membranes for a wide selection of aqueous salt ions (with varying ionic charges and spanning a wide range of practical hydrated ionic volumes) showed two dominant mechanisms for the ion rejection: size exclusion (due to compression of the ionic hydration shell in narrow channels) and electrostatic repulsion (due to the membrane's surface charge). Such GO membranes exhibited ultrahigh charge selectivity (up to 96%), driven by the negative surface charge of the oxygen-carrying functional groups in the membrane's nanochannels. Unfortunately, operating the GO membranes led to the intercalation of 2-3 layers of water molecules among the individual GO sheets, resulting in a 50% swelling (from 9 to 13.5 Å). [13] Such swelling significantly impairs the separation efficiency since the interlayer spacing needs to be tuned and fixed to filter/separate specific ions. In this regard, a scalable method to make GO membranes with minor swelling (minimum interlayer distance of 6.4 Å, which corresponds to a reduction of 53%) relied on physical confinement, leading to ≈97% rejection of NaCl salt. [14] Further studies on the long-term stability showed another limitation of GO membranes, which disintegrated after soaking 5 days in water. By contrast, GO-based membranes with different kinds of chemical functionalization demonstrated improved stability [15] and moderate rejection rates against a short range of chemicals, [16] including antibiotics. [17] rGO-based membranes also showed higher stability (minor damage or delamination in water solutions up to 90 days) than GO-based ones. [18] Although those GO-and rGO-based membranes showed promising performance, they still have significant drawbacks regarding long-term stability and sustainability (i.e., they often require hazardous materials and complex fabrication procedures). [18,19] Pristine graphene membranes could offer a good platform for ion sieving in aqueous solutions, going beyond those limitations. Exploratory studies demonstrate the application of mechanically exfoliated graphene crystals for ion-selective transport. [20,21] Similarly, graphene produced by chemical vapor deposition (CVD) showed solid performance in selective ionic and water transport, which can be regulated by high-density, sub-nm pores in the polycrystalline structure (with an estimated water flux through an individual pore up to three molecules per pico-second). [22,23] Among the different methods, liquid phase exfoliation (LPE) allows the production of graphene and 2D materials in large quantities with high yields. [24][25][26][27] Scalable mass production is a prerequisite for the fabrication and commercialization of any technology, including filtering membranes. LPE graphene membranes generally possess a laminar structure that is highly suited for water purification (as opposed to CVD monolayer graphene, for example). To date, only a single report discusses the use of LPE graphene for water filtration: The membranes exhibited excellent Na + rejection properties (≈97%) with high water permeance (≈10 times higher than those reported for GO membranes) and stability in aqueous solutions with no observed swelling. [19] This work describes a green LPE approach for fabricating graphene-based membranes suitable for water filtration. Pristine graphene flakes are exfoliated in Cyrene by high-shear mixing, thus avoiding conventional LPE solvents (such as N-Methyl-2-pyrrolidone, dimethylformamide, and dimethyl sulfoxide), which are typically unsafe and toxic and pose health and environmental risks. [24,[28][29][30][31][32] The graphene flakes in Cyrene dispersion at different volumes (2.5, 5, and 10 mL) are vacuum-filtered to fabricate membranes with an increasing thickness (≈1.2, 2.6, and 5.7 µm), having a uniform surface made of well-packed graphene flakes with no pinholes. The membranes remain structurally intact (thickness increase < 5%) after 90 days of operation in water. Ion transport and filtration experiments show ionic conductance values for KCl solutions ranging from 0.55 to 0.45 mS. Diffusion studies confirm that the membranes operate by size and charge selection (behaving as anion selective). More than 95% rejection rates are observed in tetracycline antibiotic diffusion studies. Satisfactory anti-bacterial adhesion properties are verified with two different types of bacteria, Escherichia coli and Staphylococcus aureus. ALTANA). Hydrophilic PVDF membranes (47 mm diameter, 100 nm, and 220 nm pore size, VVLP04700) were purchased from Durapore-Merck Millipore Ltd. Graphene oxide (GO) flakes were purchased from Ragedara mines (PVT) LTD, Sri Lanka. All materials were used for the experiments without any further purification process.

Preparation of Few-Layer Graphene (FLG) Flakes
Graphene exfoliation was performed via high-shear exfoliation in Cyrene. The chosen method was a combination of intercalation and exfoliation steps. 20 g of natural graphite powder was added to 500 mL of Cyrene solution, to which 1 v/v % of DIS-PERBYK-2012 and DISPERBYK-190 were added. The prepared mixture was sonicated for 30 min to obtain the homogeneous dispersion. Then, the dispersion was subjected to exfoliation with a high-shear mixer (Silverson L5M, standard mixing with an axial flow head) at 6000 rpm for 6 h at room temperature (an ice bath was used to withdraw excess heat produced during the process). After exfoliation, the resultant black dispersion was centrifuged at 6000 rpm for 30 min to remove thick and un-exfoliated flakes. Finally, the supernatant containing FLG flakes was collected ( Figure S1, Supporting Information). Thermogravimetric analysis was used to estimate the concentration of the FLG flakes in the dispersion (≈ 1.2 mg mL −1 ).

Membranes Preparation
Graphene membranes were prepared by vacuum filtration on PVDF supports with 100 nm pore size, using three volumes of FLG flake dispersion (2.5, 5, and 10 mL) to make membranes with increasing thickness-indicated hereafter as Gr2.5, Gr5.0, and Gr10.0, respectively. An individual membrane was fabricated on PDVF support with 220 nm pore size using 5 mL of dispersion-indicated as Gr5.0/220. In all cases, the filtered Cyrene was collected ( Figure S2, Supporting Information) from the bottom flask of the vacuum filtration assembly. The membranes were dipped in DI water for 30 min to allow the realignment of flakes and finally dried at 60°C in an oven overnight. After a complete characterization, the membranes were kept in water for 90 days (at room temperature, in the dark) and characterized once more to probe their stability and ascertain any possible structural change (such as swelling). For comparison, GO-based membranes were fabricated by vacuum filtration (5 mL of a 25 mg mL −1 dispersion of GO flakes in ethanol, whose Raman spectrum is shown in Figure S3, Supporting Information).

Atomic-Force Microscopy (AFM)
The thickness and morphological features of the samples were measured in tapping mode with a CSI Nano Observer. Image processing was done with Gwyddion software.

Contact Angle
Static water contact angle (WCA) measurements were performed, under ambient conditions, by sessile drop method with a Drop Shape Analysis -Contact Angle (Kruss DSA 100), using a drop (10 µl) of distilled water added via a motor-driven syringe at room temperature. Data were calculated using the average values ± S.D. (n = 5).

Scanning-Electron Microscopy (SEM)
Images were taken using a FEI Quanta 650 FEG with a cold field electron source, using an electron acceleration voltage of 5-10 kV. SEM analysis was also used to visualize the adhesion of bacteria on the graphene membranes. For that, the samples were incubated with the bacteria (10 6 colony forming units -CFU -mL −1 ) at 37°C for 24 h, fixed with a 2.5% glutaraldehyde solution overnight at 4°C, and dehydrated in a series of ethanol solutions (30,50,70,85,90,95, and 100 v/v%), sequentially for 10 min each. Before being observed in SEM, the samples were air dried overnight at room temperature and coated with gold by sputtering for 30 s at 30 mA (EM ACE200, Leica, Germany).

Transmission Electron Microscopy (STEM)
The structure of the graphene flakes was investigated using a JEOL 2100 system at 200 kV. The sample was made by drop casting the Graphene diluted dispersion on TEM copper grids (200 mesh) covered with lacey carbon and allowed to dry at 80°C for 30 min.

X-Ray Photoelectron Spectroscopy (XPS)
Data were acquired with an ESCALAB 250 Xi system with an analysis chamber maintained in ultra-high vacuum (UHV) conditions all the time.

Raman spectroscopy
Measurements were performed on an ALPHA300 R Confocal Raman Microscope (WITec) using 532 nm laser light for excitation at room temperature. The laser beam was focused on the sample by a 50x and 100x lens (Zeiss). Single acquisitions were performed using the 600 g mm −1 grating and the P Laser < 2 mW to minimize localized heating and damage to the sample.

UV-Vis Spectrophotometer
Measurements were carried out on NBI-SHIMADZU UV-2550 UV-vis SpectroPhotoMeter in the range of 200-800 nm using quartz cuvettes (1 cm path length).

Zeta Potential
The surface charge of the graphene flakes was measured with a NBI-Horiba SZ-100Z Dynamic Light Scattering (DLS) System. The graphene suspension in Cyrene was diluted to 1:10 ratio and slowly injected into a folded capillary cell having two graphite electrodes. The Zeta potential was estimated following the Henry and Smoluchowski equations. [33,34]

Ion Transport Test
To estimate the ion transport behavior through the graphenebased membranes, a custom-made apparatus (tau cell) was assembled as shown in Figure S4 (Supporting Information). The tau cell was designed by Artcam and fabricated in Teflon by CNC-High-Speed Milling System (FlexiCam Viper). Graphene membranes were pasted on an acrylic sheet (with an opening of 2 × 2 mm 2 ) and sides were sealed using stycast epoxy glue to ensure the only path given for ion transport is through the membrane named it to coin cell. The tau cell's two reservoirs (feed and permeate) were separated by a coin cell ( Figure S4, Supporting Information). It is essential to wet the surface of the graphene membranes before any transport or diffusion measurements since the graphene flakes are hydrophobic. To do this, we immersed the coin cell in 2-Propanol (IPA) for 10 min, and the tau cell was washed thoroughly with pure IPA, followed by IPA + DI water (50:50) and pure DI water. In ion transport tests, chloride solutions filled the two reservoirs so that the graphene membrane surface was fully immersed in the solution. Current-voltage (I-V) characteristics for Gr2.5, Gr5.0, and Gr10.0 was measured using a KCl solution between two liquid reservoirs (feed and permeate). Three solutions with different cation valences (KCl, CaCl 2, and AlCl 3 ) were used to estimate the ionic permeability through the membranes. The concentration ratio between feed (1 m) and permeate (10 mm) was fixed at 100. The potential was applied with the help of a Keithley 2614B source meter using Ag/AgCl electrodes, and I-V characteristics were obtained through a LabVIEW program. To estimate the diffusion potentials, we performed I-V measurements with a voltage swept in the range of 200 mV (in steps of 25 mV), and the resultant current was recorded. These experiments were repeated for both soaked and un-soaked membranes.

Water Permeability Test
Water permeability through the membranes (before and after soaking) was measured using a dead-end vacuum filtration setup with an effective area of 25 mm 2 , maintaining a pressure difference of 1.0 bar with a feed solution (DI water) of 250 mL at room temperature. The permeance J (L m −2 h −1 bar −1 ) was calculated using Equation 1: where V (L) was the volume of permeated water, A (m 2 ) was the effective membrane area, T (h) was the total permeate time, and P (bar) was the pressure difference.

Molecular Separation Test
The aforementioned tau cell was used for antibiotic dynamic diffusion studies. The procedure used in Experimental Section 2.5 was followed: the graphene membranes were placed facing the reservoir (feed) containing the antibiotic (AB) solution (25 mg L −1 tetracycline in DI water), while the second reservoir (permeate) was filled with DI water. Peristaltic pumps were used to recirculate (with a flow rate of 30 mL min −1 ) the solutions in both reservoirs of the tau cell, as shown in Figure S5 (Supporting Information). Concentration variation between the reservoirs would indicate the diffusion of antibiotic molecules through the membranes. The tetracycline concentration in water was estimated by UV-vis, tracking the main absorption peak at 367 nm (via a calibration curve, shown in Figure S6, Supporting Information). Experiments were performed by installing new membranes at the reported time intervals and filling new solutions on both the feed and permeate sides of the tau cell.

Bacterial Adhesion Studies
The bacterial adhesion properties on graphene membranes of Gr2.5, Gr5.0, and Gr5.0/220 (made on PVDF with 220 nm pore size) were analyzed. Moreover, the same procedure was used for GO membranes and the reverse osmosis membrane (50 GPD, Geekpure) was used as a control. These studies were assessed using gram-positive Staphylococcus aureus (S. aureus) and gramnegative Escherichia coli (E. coli), typically waterborne bacteria selected to probe the antibacterial properties of materials. Before all experiments, the membranes were cut with a puncher of 6 mm diameter and sterilized in 75% aqueous ethanol solution for 15 min or UV sterilization. E. coli (CECT 434) and S. aureus (CECT 435) were purchased from the Spanish Type Culture Collection (Valencia, Spain). The two bacterial colonies grew overnight in 4 mL trypcase soy broth (TSB) at 37°C. The bacteria suspension concentration was estimated by measuring the absorbance at 600 nm (Nanodrop 2000c) and then adjusted to an initial concentration of 10 6 CFU mL −1 by dilution. The ISO 22 196:2007 norm was adapted to test the antibacterial adhesion on the surfaces (contact method). [35] Briefly, the sample disks were fully incubated in separate wells of a sterile 48-well plate with 150 µL of the bacterial suspension (10 6 CFU mL −1 ) for 24 h at 37°C. After washing the samples 3 times with phosphate buffer (PBS, 10 mm), the attached bacteria were dissociated from the surfaces by sonicating twice with PBS for 2 min (water-bath ultrasonicator Elmasonic P, 37 kHz frequency, 100% power) and agitating in a vortex for 2 min. A serial dilution was performed for each sonicated sample and 20 µL of each dilution was added to Trypticase soy agar (TSA) plates and incubated at 37°C overnight. The CFU was counted to obtain the number of viable bacteria per volume unit and expressed in CFU mL −1 . Experiments were performed in triplicate. The contact method results were analyzed in terms of reduction of bacterial adhesion (BA) according to the following Equation 2: where C is the average number of viable bacteria recovered from control (GO-based and RO membranes) after 24 h, and S is the number of viable bacteria recovered from the graphene membranes after 24 h.

Results and Discussion
A dispersion of FLG flakes in Cyrene was prepared by highshear exfoliation, as described in Experimental Section 2.2. The dispersion was diluted 50 times to check the Tyndall effect using a 635 nm laser light (inset in Figure 1a): The beam of light was visible in the graphene dispersion, indicating the excellent homogeneity of the graphene flakes in Cyrene. The Zeta potential of the graphene dispersion was measured as +30 mV. The Zeta potential value is essential to understand the ion transportation mechanism through the channels arising in the graphene laminates. The chemical composition of the exfoliated FLG flakes was ascertained by X-ray photoelectron spectroscopy ( Figure 1a). The survey spectrum shows a C1s peak at 284.4 eV, a minor O1s peak at 534 eV, and no other trace elements: This reveals the high purity of the FLG flakes ( Figure S7, Supporting Information). [36] The high-resolution C1s spectrum shows the asymmetric profile typical of graphene materials. The C1s spectrum was deconvoluted into three components: an intense one at 284.4 eV for sp 2 C, and two others at 285.1 and 286.5 eV representative of oxidized carbon, as typically expected in exfoliated flakes. [37] Figure 1b shows the Raman spectra of the FLG flakes compared to the natural graphite used for the exfoliation. [38][39][40] The three characteristic peaks are observed: D, G, and 2D bands. The G band corresponds to the highfrequency E 2g phonon at the Brillouin zone center (due to inplane vibrational modes of sp 2 carbon domains). The D band is due to the breathing modes of sp 2 rings and requires a defect for its activation by double resonance; as such, it is generally related to disorder and defects. In exfoliated graphene flakes, it can identify defects within the basal plane and along the flake edges. [41] The 2D band is the second order of the D band; it can provide information about the layer number and the presence of doping. Raman spectra of natural graphite (NG) show G band centered at ≈1579 cm −1 , and 2D band at ≈2717 cm −1 . The 2D band redshifts and become broader and symmetrical after the exfoliation, indicating the graphite thinning into few-layer flakes. [42] The increase in the D peak intensity is due to the lateral size decrease, which in turn entail a higher ratio between edge and basal sites. The defect density was quantified by the intensity ratio of the D and G bands (I D /I G ), which is calculated as 0.35 in the exfoliated flakes, much higher than the natural graphite's value (0.015). The Raman parameters are tabulated in Table 1. Figure 1c,e shows SEM and TEM micrographs of the exfoliated FLG flakes. The flakes exhibit a smooth, even, and thinlayered structure with a lateral size of a few hundred nm. We statistically analyzed the flake size distribution by measuring the lateral size of 60 individual flakes. The graph in Figure 1d shows the lateral size histogram, fitted with a Gaussian distribution centered at 0.8 µm. Additionally, we carried out HR-TEM analysis: which further proves the existence of a few layers (≈2-4 layers, Figure 1f) in the exfoliated FLG flakes. Graphene membranes were prepared by vacuum filtration on PVDF support as described in Experimental Section 2.3. Figure 2a,e,i shows photos of the membranes Gr2.5, Gr5.0, and Gr10.0, respectively. At a visual inspection, the membranes appear highly flexible and robust. SEM inspection was employed to study each membrane's surface morphology (Figure 2b,f,j). The membrane surfaces seem similar: The exfoliated graphene flakes are randomly orientated in-plane, forming an interconnected surface with no pinholes within or among the flakes. AFM shows slight changes in the surface roughness (Figure 2c,g,l): Gr2.5 has higher roughness (≈177 nm) than Gr5.0 (≈140 nm) and Gr10.0 (≈85 nm). The cross-sectional SEM fractography in Figure 2d,h,l provides the section morphology and the membrane thickness. The PVDF support can be easily identified in each image by its bright, porous structure. On top of the PVDF support, the laminates appear regularly formed by stacking the FLG flakes (assembled via vacuum force, as described in Experimental Section 2.3). The membrane thickness, highlighted in red, is uniform in each sample: ≈1.2, 2.6, and 5.7 µm for Gr2.5, Gr5.0, and Gr10.0, respectively.
The stability and robustness of the membranes were evaluated by operating them in water for 90 days (see Experimental Section 2.3 and Figure S8, Supporting Information). Highresolution cross-sectional images of prepared and soaked membranes were acquired and compared ( Figure S9, Supporting Information). Based on these images, thickness values were estimated and plotted ( Figure S10, Supporting Information). AFM topography, amplitude, and phase diagrams (before and after soaking) were also acquired for all samples (Figures S11 and S12, Supporting Information). The samples' roughness values were also measured and displayed in Figure S13 (Supporting Information). In all three cases, the graphene laminates composing the membrane appear unchanged (neither breaking nor swelling occurs), retaining the initial stacking configuration. The degree of swelling (D s ) was estimated using the following equation.
where T a represents the average thickness of the membranes after 90 days of soaking and T b is the average thickness of asprepared membranes. The degree of swelling is calculated as 4.8%, 2.6%, and 1.1% for Gr2.5, Gr5.0, and Gr10.0 membranes, respectively. As opposed to GO and rGO membrane, this structural stability could be granted by the minimal oxygen content in the pristine graphene flakes (as evidenced by XPS in Figure 1a), which hinders the water uptake within the laminate. Interestingly, the thickest membranes retained the maximum structural stability. The wetting properties of the membrane surface were characterized by the water contact angle (WCA). The bare PVDF support has a WCA of 71.1 ± 1.6°. The three graphene membranes (Gr2.5, 5.0, 10.0) have higher WCA: 94.0 ± 1.2°, 87.5 ± 1.4°, 98.2 ± 2° respectively. Gr10.0 is the most hydrophobic of the lot. [43] Contact angles of ≈90° have also been reported in previous studies using hydrophobic graphene sheets [44] or density functional theory, [45] in agreement with our results. The water permeance through the membranes (before and after soaking) was evaluated, as described in Experimental   thick membrane [15] and 8-27.6 L m −2 h −1 bar −1 for 14 nm thick membrane). [16] rGO membranes (60 nm thick) were reported, with water permeance over 10 000 L m −2 h −1 bar −1 (up to 1000 times higher than those of the reported GO-based membranes and commercial membranes). [18] However, this kind of membrane involves a complex fabrication process (via reduction and cross-linking agents). Also, upon increasing the thickness of GO/rGO-based membranes, the water permeance is known to decrease several times. [15,18] As shown, our membranes do not follow this trend, not even after operating in water for 90 days ( Figure S14, Supporting Information).
I-V characteristics of Gr2.5, Gr5.0, and Gr10.0 were measured using a KCl solution (0.1 m) between two liquid reservoirs, as described in Experimental Section 2.5 and depicted in Figure 3a. The data in Figure 3b and used to determine the ionic conductance (Figure 3c) by measuring the slope of the respective curves. The ionic conductance values are in the range of 0.43-0.55 mS, which is one order of magnitude lower than the ≈ 8 mS reported for analogous membranes (i.e., made of pristine graphene flakes and 3 µm thick). [19] The much lower ionic conductance values imply a more efficient ion filtering than previously reported, for the effect of the nanochannels formed in the graphene laminates. The ionic conductance decreases as the laminate thickness increases: Thicker samples are more effective in filtering because of a longer ion path length. [46] The ionic conductance through the membrane was measured using KCl solutions with different salt concentrations (from 10 −5 to 1 m, Figure 3d) to elucidate the role of the membrane's surface charge on ion transport. At decreasing concentrations (from ≈1 to 10 −4 m), the conductance falls linearly, following the bulk conductance value. At lower concentrations (<10 −4 m), the conductance deviates from the bulk value and saturates, implying that the nanochannel surfaces possess a charge (thus showing an electrokinetic phenomenon). This is in good agreement with the Zeta potential value of the graphene dispersion (+30 mV) used to fabricate the laminate, which is reflected in the inner surface charge of the laminate itself-as observed in membranes based on nanochannel slits, [47,48] interlayer spaces in graphite, [49] nanopores in 2D materials, [50,51] GO and LPE graphene membranes. [13,14,19] The membranes were tested for selective ion transport, using aqueous solutions of three chloride salts (KCl, CaCl 2, and AlCl 3 ), as shown in Figure 4. The corresponding cation hydration radii are in the order K + < Ca 2+ <Al 3+ (K + :3.3 A°, Ca 2+ : 4.3 A°, and Al 3+ : 9 A°), [52] while the hydrated Cl − radius remains the same in all the solutions. [47] I-V characteristics are shown in Figure 4a-c. Due to the concentration gradient, ions start moving from higher to lower concentrations through the filtering membrane. A diffusion current appears when the diffusivities of cations and anions are different. The open circuit potential V m (i.e., the potential at zero current, also known as "membrane potential") decreases with increasing cation radius (in hydrated form). This indicates the rejection of cations and no transportation through the channels. [47] In all cases, the V m is always shifted to positive values. This would indicate that the surface of the graphene channels is positively charged, thus operating as anion-selective. To further confirm this assumption, the mobility ratios were measured, as described below.  redox potential is present at the electrode/electrolyte interface, which needs to be subtracted from the measured membrane potential to get the diffusion potential. Therefore, the diffusion potential (Figure 4d) was calculated as V diff = V m −V R , where V R is the redox potential. The ion selectivity of the membrane (Figure 4f) can be calculated using the Nernst Equation 4: where S = t − −t + is the selectivity and t − and t + are the transport number of anions and cations, R, T, and F are the universal gas constant, temperature, and Faraday constant, respectively, and ΔC is the concentration ratio. It was observed that the diffusion potential increases as the cation size increases. Further studies were made to elucidate this point. The mobility ratio was calculated ( Figure 4e) using the Henderson Equation 5: where , _ µ µ + are the mobilities and Z + , Z − are the valencies of cations and anions, respectively. We observe that the mobility ratio decreases with an increase in the hydrated cation radius, in agreement with previous reports. [49] We can attribute this effect to a slower movement of cations than anions through the membrane, as the anion Cl − remains the same in all the solutions. The reason lies in the formation of complex nanochannels with a high positive surface charge that suppresses the transportation of cations with high charge selectivity (Figure 4f). All these observations align with the highly positive zeta potential value of +30 mV. The rejection of cations by positively charged nanochannels is schematically shown in Figure S15 (Supporting Information). Similar studies for soaked membranes are shown in Figure S16 (Supporting Information).
To further test the Gr5.0 membranes in terms of molecular separation, we used tetracycline antibiotic (a common and hazardous water pollutant) as a probe molecule (as described in Experimental Section 2.7). Generally, antibiotic removal studies are performed on cross flow cells in static mode, while we performed a dynamic study in a tau cell configuration, which has not been previously reported. The tetracycline molecular size (≈1.1 nm) is bigger than the hydrated cations tested in the previous sections (K + : 3.3, Ca 2+ : 4.3, and Al 3+ : 9 Å). The membranes are thus expected to reject tetracycline via size selection. Figure 5 shows the UV-vis absorption spectrum of the water solution (permeate side as described in Experimental Section 2.7) with different time intervals to probing the AB molecules. With increasing time, no absorption peaks appeared in either the un-soaked or soaked membranes since, in both cases, the absorption was always below the detection limit given by the calibration curve of the antibiotic. The rejection rates were estimated to be > 95% in all cases. Considering that tetracycline has no charge in DI water, [53] we can assume that the rejection mechanism is driven by adsorption, i.e., the antibiotic molecules are trapped on the membrane surface, while no electrostatic repulsion can occur. [17] To further test this assumption, we allowed the solution in continuous contact with the membrane surface for 100 h (as described in Experimental Section 2.7). In this experiment, the concentration of the tetracycline feed side reduced over time ( Figure S17, Supporting Information), proving that the molecules got progressively adsorbed to the membrane.

Bacterial Adhesion Studies
Biofilm formation typically reduces the membrane lifetime and increases the operational costs (e.g., by increasing energy consumption due to higher required pressures). [54] The antibacterial properties of the membranes were studied using two bacterial models: gram-positive (S. aureus) and gram-negative (E. coli). Three graphene membranes were selected for this study: the Gr2.5 and Gr5.0 analyzed in the previous section, and one additional membrane made with 5 mL of FLG flake dispersion (the volume that resulted in the optimal thickness for filtering purposes). The Gr5.0/220 membrane (having analogous ion transport properties to Gr5.0, Figure S18, Supporting Information) was tested to explore further the influence of wetting properties and roughness on bacterial adhesion. For a robust benchmarking against commercially available products, the bacterial adhesion studies of our membranes was also performed and compared to GO-based and RO membranes. Figure 6 shows the SEM images and the concentration of viable bacteria obtained by contact methods for the different membranes.
Gr2.5, Gr 5.0, and Gr5.0/220 show similar surfaces with very few S. aureus bacteria adhered. The GO surface appears smoother and a higher number of S. aureus is visible, while only very few appear on the RO membrane. In the case of E. coli bacteria, all surfaces appear significantly covered by them, with the exception of the Gr5.0/220 membrane, where a smaller number appears. The RO membrane seems particularly prone to the adhesion of E. coli bacteria. The concentrations of viable bacteria on Gr2.5 and Gr5.0 are 2.1 × 10 5 and 3.1 × 10 5 CFU mL −1 for S. aureus, and 2.2 × 10 6 and 2.9 × 10 6 CFU mL −1 for E. coli, respectively. No significant differences between the two membranes are found. The two concentrations on Gr5.0/220 decrease to 5.9 × 10 3 and 1.6 × 10 6 CFU mL −1 , respectively. This translates into an impressive reduction of S. aureus bacterial adhesion (−98% with respect to Gr5.0). In the case of E. coli, the concentration is also lower but the adhesion decreases less significantly (−44% with respect to Gr5.0). The antimicrobial behavior of Gr5.0/220 is superior to that of well-established GO membranes (−99% and −63% for S. aureus and E. coli, respectively) and commercially available RO membranes (−67% and −95%, respectively). It is worth noting that even Gr2.5 and Gr5.0 membranes showed a decrease in bacterial adhesion between 92% and 94% for E. coli in comparison to RO membranes. The contact method and SEM analysis revealed a lower adhesion of S. aureus than E. coli in all the graphene membranes. The differential adhesion of the two species may be related to the surface charge of the membranes. E. coli is known to have a more negatively charged surface than S. aureus. [55] A higher adhesion of more negatively charged bacteria can be thus expected on the positively charged surfaces of the graphene membranes due to electrostatic attraction. [56,57] Moreover, the bacterial species possess different membrane structures that can influence their capacity to tolerate stresses, such as reactive oxygen speciesrelated lipid peroxidation and membrane disruption. S. aureus has a cytoplasmic membrane with a thick peptidoglycan layer, whereas E. coli possesses a thinner peptidoglycan layer but has a stiffer outer membrane made of lipopolysaccharides. [58] For this reason, E. coli were found to be more resistant to membrane damage caused by the sharp edges of graphene-based materials when compared to S. aureus, which lacks an outer membrane. [59] Different factors may affect the interaction between bacteria and a graphene-based surface with different physicochemical characteristics, such as lateral size and thickness, morphology, roughness, surface area and chemistry, wetting properties, etc. [44] Here, the same production/deposition methods and experimental conditions were used; therefore, the surface properties (e.g., surface charge, hydrophobicity, and roughness) of the membranes should be the determining factor for bacterial adhesion. The contact angle of Gr2.5 (94 ± 1.2°) was slightly higher than Gr5.0 (87.5±1.4°), as well its roughness (177 nm for Gr2.5 and 140 nm for Gr5.0). These results suggest that the minor discrepancies in roughness and wettability between Gr2.5 and Gr5.0 did not significantly influence bacterial adhesion. By contrast, Gr5.0/220 showed a rougher surface (R rms = 212 nm) with some irregularities (AFM image in Figure S20, Supporting Information), which is also more hydrophobic than the others (WCA = 113°). [60] With a PVDF support with a pore diameter (220 nm) commensurate to the FLG flake size (using the same method as described in Experimental Section 2.3), this membrane was expected to present a high density of sharp edges and overall higher roughness/ hydrophobicity than membranes fabricated on PVDF-100 nm. S. aureus is expected to adhere less on rough surfaces and to be mechanically damaged by sharp edges (possibly piercing the bacterial membrane). [61][62][63] Additional damage may also arise from the formation of pores in the membrane, causing osmotic pressure and bacterial death. [64] Regarding E.coli, our results indicate that hydrophobicity can be the main factor attributed to their higher adhesion to the membranes. The gram-negative bacterial cell wall is more hydrophobic than the gram-positive cell wall, and it generally has a higher affinity for hydrophobic surfaces. [44] This would justify a higher adhesion of E. coli on the Gr5.0/220 membrane than S. aureus. Nevertheless, the sharpness and roughness of the Gr5.0/220 also could be less efficient in disrupting the thicker cell wall of E. coli. Previous works have claimed antibacterial effects in a number of GO surfaces, [58,65,66] but no consensus exists yet on the GO biocompatibility itself and the mechanism behind its antimicrobial activity. [44,67,68] The purity and the chemical state (oxygen functionalization level) of GO can also affect its antibacterial activity (e.g., no antibacterial activity was observed in highly purified GO). [69] The deposition, attachment, and proliferation of microorganisms as well as biofilm growth were previously reported on polyamide RO membranes, posing a significant challenge for RO filtration in water treatment. [70,71] Our results highlight the potential use of our graphene-based membranes (particularly the Gr5.0/220 that presents a high antibacterial effect), in long-term water treatment applications.

Conclusion
We have used LPE graphene flakes to fabricate laminates as active material in membranes for water filtration/purification. Graphene laminates with increasing thickness (≈1.2, 2.6, and 5.7 µm, corresponding to 2.5, 5, and 10 mL of graphene flake dispersion) were formed on PVDF membrane by vacuum filtration. The membrane surfaces showed a well-packed mat of graphene flakes with no pinholes, having decreasing roughness at increasing volumes (≈177, ≈140, and ≈85 nm, respectively). The analysis after 90 days of operation in water evidenced that the membranes are structurally stable and do not swell much over time (the degree of swelling was calculated as 4.8%, 2.6%, and 1.1%, respectively). This is in stark contrast to the widely reported GO and rGO membranes. The water permeance of Gr5.0 (≈751.9 L m −2 h −1 bar −1 ) is higher than that of the other two samples (≈667 L m −2 h −1 bar −1 for Gr2.5, and ≈280 L m −2 h −1 bar −1 for Gr10.0). This is consistent with the water contact angles of the membranes, indicating Gr5.0 as the most hydrophilic of the lot. Overall, these water permeance values are higher than those of conventional GO membranes. In ion transport tests, the membranes are performed by a combination of size exclusion and charge selectivity. In particular, they showed anion-selective properties, possibly due to the positive surface charge of the nanochannels forming the laminates. Additionally, antibiotic (tetracycline) diffusion studies under dynamic conditions showed rejection rates higher than 95% for the Gr5.0 membrane, driven by size selection. The membranes (Gr5.0/220 in particular) displayed a consistent antibacterial activity in contact method tests with S. aureus and E. coli bacteria, with a reduction of 99% and 95% for Staphylococcus aureus and Escherichia coli, with respect to well-established graphene oxide and reverse osmosis membranes.