Dairy products have a very large global market that is constantly growing, which creates a necessity for new membrane technologies that are required for their processing. Considerable attention has been paid to the production of lactose-free milk . Lactose is a disaccharide that comprises approximately 4–5 wt% of milk. However, a large portion of the world population is intolerant to lactose, with some regions of Asia reaching 90% of the population . Thus, dairy products with low or zero lactose content are needed. Milk is usually delactosed through lactose hydrolysis, via enzymatic conversion into galactose and glucose . However, this process increases sweetness, changes coloring, and may decrease the nutritional value [1,3]. Hence, membrane technology is utilized to avoid lactose hydrolysis and straightforwardly remove lactose. First, most lactose and proteins are separated from milk using ultrafiltration (UF), and then lactose is separated by nanofiltration (NF) membranes .
The combination of UF, NF, and electrodialysis has been explored to produce low-lactose milk powder, by H. Zang et al. . UF membranes reject proteins and fat; then electrodialysis is used to remove mineral salts; finally, NF membranes are used to separate lactose. However, the abovementioned authors have stated that membrane fouling is a major issue in their setup and it can decrease operation time. In another study, diverse commercial UF membranes were used to concentrate proteins and separate lactose. The membranes achieved 100% protein recovery with 10% lactose remaining in the concentrate . Interestingly, T. Sánchez-Moya et al. were able to reduce the fouling layer by washing with water. Typically, NF membranes are used to separate lactose; however, fouling remains a challenge for increasing operation lifetime and reduce operation costs. In a particular study, seven commercial NF membranes were used to filter dairy wastewater. It has been observed that the surface roughness and contact angle had a low correlation with permeability, whereas the pore size and thickness of the skin layer were more determinant. Nevertheless, surface roughness and contact angle had a higher influence on irreversible fouling, which suggests that adsorption of proteins is the main mechanism for fouling rather than pore blocking . On the basis of several reports in the literature we concluded that membranes that were resistant to fouling by milk proteins were essential for use in processes that include the preparation of milk-protein concentrates, concentration of whey, milk fat removal, and effluent control . Additionally, efficient antiorganic fouling properties are also required for applications in pharmaceutical, medical, and food industries, as well as in wastewater treatment [, , , ].
Nanofiltration membranes have been used widely for desalination pretreatment, wastewater treatment, and food industry, among other applications . Most of these membranes are polymer-based and prone to fouling by organic compounds that are present in the feed source. For food filtration, fouling is initiated through protein adhesion on membranes [13,14]. Additionally, fouling on membrane surfaces deteriorates their performance, reduces operation lifetime, and requires cleaning treatments that increase operation costs. Hence, the design and development of emerging materials for membrane filtration with effective antifouling properties are urgently needed.
Graphene oxide (GO) is a novel material obtained through the controlled oxidation of graphite . It is a water-dispersible colloid that can be deposited as a membrane via several techniques, including casting, filtration, shear coating, and spray coating [, , , ]. GO membranes exhibit NF level selectivity, i,e, the ability to separate molecular constituents from liquid sources . Additionally, GO membranes have shown not only excellent antifouling properties against oil, surfactants, sodium alginate, and proteins but also antimicrobial activity [, , ]. In several cases, GO has been deposited on the surface of an active layer, from a filtration membrane, without affecting its filtration performance, improving antifouling capabilities [22,23]. However, the separation capabilities and antifouling properties of membranes with active layers made entirely of GO have been scarcely studied for food filtration.
Here, we report the design and fabrication of GO membranes exhibiting higher lactose permeability and water flux recovery from milk filtration compared with commercial NF and UF membranes measured in our laboratory. Molecular dynamics (MD) simulations demonstrate that lactose exhibits a weak interaction with GO layers and thus can diffuse through the GO membrane, whereas fat and proteins cannot. The fouling layer was determined to affect the filtration performance. Hence, the interaction of GO layers with proteins, such as bovine serum albumin (BSA) and lysozyme, was studied experimentally and via MD, and showed that the highly hydrophilic GO with a negative charge had a higher affinity to lysozyme, where charge interactions were dominant in determining fouling likelihood.
2. Results and discussion
2.1. Lactose removal from milk
The GO sheet size was evaluated via scanning electron microscopy (SEM) with a wide size distribution ranging from 0.005 to 0.2 μm2 (Fig. S1). To understand how much of the area was covered by each sheet size, the count number was multiplied by the area distribution. In this plot, a bimodal distribution becomes apparent at 0.06 and 0.2 μm2 (Fig. S1). The GO layer was deposited via spray coating on polysulfone (PS) and polytetrafluoroethylene (PTFE) supports under the same conditions, with support pore sizes of 20 nm and 1 μm, respectively, which resulted in thin-film composite membranes labeled as PS-GO and PTFE-GO, respectively. This spray coating methodology has previously shown good peel-off resistance at high surface water flux (1000 ml/min) because of the presence of polyvinyl alcohol, which acts as a binder and exhibits negligible change to the surface chemistry after drying at 100 °C for 1 h . The PS-GO membranes exhibited a smooth surface with few wrinkles (Fig. 1a and b), as observed via SEM and 10 nm RMS surface roughness, as determined via atomic force microscopy (AFM). In contrast, the PTFE-GO membrane has a surface roughness of 183 nm RMS (Fig. 1 c and d). The thickness of the GO layer is approximately 105 nm (Fig. 1e–g) with a surface density of 2.16 × 10−2 mg/cm2.
Surface charge and hydrophilicity are factors that can determine the separation performance of membranes [19,24]. The surface charge of all the membranes used in this study was measured using zeta potentials. Milk typically has a pH between 6.7 and 7; here, the obtained average zeta potentials were −24, −34, −49, and −31 mV for PS-GO, PTFE-GO, NF, and UF membranes, respectively (Fig. S2). Additionally, pure PS and PTFE exhibited zeta potentials of −52 and −60 mV. The contact angle was measured to understand the hydrophilic nature of the membranes, with values of 27°, 46°, 35°, and 59° for PS-GO, PTFE-GO, NF, and UF membranes, respectively. By contrast, the PS and PTFE support membranes exhibit contact angles of 69° and 120°, respectively. Noteworthily, the trend in surface charge and contact angle for the PS and PTFE support membranes was maintained even after the deposition of GO, where the PTFE-GO membrane was more negative and exhibited a higher contact angle than did the PS-GO membrane. Similarly, the effect of substrate material upon water permeation in a layer of graphene nanoplatelets was reported, which showed that the wettability of the graphene nanoplatelet layer was correlated with the substrate hydrophilicity . Hence, these results indicate that a support membrane affects the surface charge and hydrophilicity of GO membranes.
The permeate flux was monitored for all membranes (PS-GO, PTFE0GO, NF, and UF) by analyzing two independent samples of each type. The measurements were monitored after the initial permeate flux stabilization with water; then the source was alternated between water and milk (Fig. 2a–d and S3). After each milk filtration stage, all membranes tend to exhibit a decrease in water flux, possibly because of the surface adsorption of the organic matter from milk and even partial clogging of some pores, which is known as fouling. During the initial phase of fouling, organic molecules adsorb on the surface, occlude the pores on the surface of the membranes and keep building up creating a thicker fouling layer, as proposed in other dairy fouling observations [6,26]. Meanwhile, the fouling layer decreases the permeate flux as the fouling layer increases in thickness. The relative amount of nonrecovered flux is known as irreversible fouling and was measured after each milk filtration stage when the source was changed to pure water. Both PS-GO and PTFE-GO membranes exhibited a lower irreversible fouling ratio (0.05–0.63) than did NF (0.46–0.85) and UF (0.12–0.92) membranes (Fig. 2e). Remarkably, the PS-GO membrane had the highest permeate flux recovery (J/J0) of 0.89 followed by the PTFE-GO membrane (0.43), which indicates that GO membranes exhibited a higher flux recovery than did commercial NF (0.18) and UF (0.16) membranes.
The permeate solutions after milk filtration were analyzed via attenuated total reflection–Fourier transform infrared spectroscopy (ATR-FTIR) and thermogravimetric analysis (TGA), to determine the lactose and ash weight content. The TGA of the permeate solutions showed the content of water, organic matter, and remaining ashes (Fig. 3a and b). Above 80 °C the remaining solids are considered as organic material until 800 °C and above are ashes. Surprisingly, the permeate from the PTFE-GO membrane exhibits the highest content of organic matter (5.3 wt%), followed by the PS-GO (1.8 wt%), UF (1.5 wt%), and NF (1.3 wt%) membranes, whereas the milk source has 12 wt% (Fig. 3b) similarly to that reported on the milk label (12.25 wt%, Table S1). Initially, milk has a content of 1.68 wt% of remaining ashes and the permeate solutions have 0.6, 0.23, 0.55, and 0.95 wt% for PS-GO, PTFE-GO, NF, and UF membranes, respectively. The remaining ashes are attributed to the presence of salts/minerals within milk. Thus, during milk filtration, the PS-GO and PTFE-GO membranes exhibit partial salt/mineral rejection and are expected to be higher than the currently studied commercial NF and UF membranes.
The permeate solutions were dried to conduct ATR-FTIR spectroscopy. The ATR-FTIR spectrum of milk has three distinguishable and nonoverlapping broad peaks that are assigned to fat (ester groups of triglycerides; 1745 cm−1), proteins (1640 cm−1), and C–O stretching of lactose (1040 cm−1) (Fig. 3c) . From the ATR-FTIR spectra, the intensity of the peak assigned to lactose exhibits higher intensity than those of fat and proteins in all permeate solutions. To perform a semiquantitative analysis of the lactose/amino acid concentration in the permeate, we prepared a calibration curve using lactose and BSA mixtures measured via ATR-FTIR, with different ratios between the main peaks located at 1040 and 1640 cm−1, respectively (Fig. S4). ATR-FTIR measurements indicate that lactose is the main organic component in all permeate solutions, all with values of the ATR-FTIR intensity ratio of lactose/amino acid larger than those of milk (Fig. 3d). The intensity ratio of lactose/amino acid signals was correlated with the lactose concentration in the organic matter via our calibration curve. Milk was calculated to exhibit 59.08 wt% of lactose from the mixture of lactose and amino acids, which is in accordance with that reported on the milk label (59.28 wt%, Table S1). The permeate from the PTFE-GO and UF membranes exhibits the highest lactose content (93 wt%) in the organic matter, followed by the NF (89 wt%) and PS-GO (81 wt%) membranes; meanwhile, the UF membrane permeate solution shows a small presence of fat (Fig. 3c).
The lactose concentration in permeate solutions was calculated from the FTIR and TGA data, (Fig. 3e); then, the membrane lactose permeate flux was calculated gravimetrically (Fig. 3f). In milk, the lactose content (considering water, organic matter, and salts) was estimated to be 7.7 wt%; the permeate solutions of the PS-GO and PTFE-GO membranes yield 1.4 and 4.9 wt% of lactose, respectively, whereas the permeate solutions of the NF and UF membranes exhibited 1 and 2.3 wt% of lactose, respectively. Using these values, the obtained lactose permeate flux was calculated for the PS-GO and PTFE-GO membranes to be 0.2 and 2.9 kg m−2 day−1, respectively; whereas that for the NF and UF membranes was 0.64 and 1.60 kg m−2 day−1, respectively. Therefore, the PTFE-GO membrane has the highest lactose concentration in the permeate solution and consequently produced the highest lactose concentration permeate flux; this may be due to the large pore size of 1 μm in the PTFE substrate to allow higher permeation after diffusing through the GO layer and the hydrophilicity of GO.
In GO membranes, the permeation mechanism is controlled by the spacing between GO sheets, because the pores on GO sheets are below 2.5 nm and cover less than 1% of the GO sheet area . The interlayer spacing of the dry and milk hydrated state GO membranes was studied by X-ray diffraction (Fig. S5). Initially, the PS-GO membrane exhibits an interlayer spacing of 9.5 Å in the dry state, whereas it reaches 12.8 Å after milk hydration. Upon hydration in milk, the 001 plane in XRD exhibits an FWHM of 1.45° 2θ, which corresponds to the variation of ±2.75 Å in the interlayer distance. The interlayer spacing after hydration in milk is lower than that of similar GO membranes hydrated in water (14.0 Å) . The distance between GO layers and strong surface charge of GO can yield partial salt rejection , and allow the diffusion of lactose.
Additionally, the GO microstructure should exhibit nanochannels because of the mismatch alignment during GO spray coating. Spray coating is conducted using random deposition microdroplets, where each microdroplet creates an island of GO. After repeated spray coating the density of GO islands increases and a continuous film is formed and increases in thickness. This random overlap of GO islands creates nanochannels and wrinkles. During filtration, it is expected that the GO layer undergoes pressurized compaction [28,29] and the size of the channels can decrease. Nevertheless, these nanochannels exhibit proper size to allow the diffusion of lactose. Additionally, these nanochannels have been reported to allow the diffusion of molecular dyes [29,30].
2.2. Membrane surface analysis after milk filtration
Milk fouling is a complex mixture of fat, proteins, saccharides, and minerals , and its interaction with a membrane surface varies because of hydrophilicity, surface roughness, charge, and chemical composition. Additionally, the growth of a fouling layer alters the filtration performance; thus, the surface characterization after milk filtration is required. The fouling layer was characterized by SEM, which showed that all membranes exhibited a fouling layer with varying thicknesses of 23, 36, 65, and 19 μm for the PS-GO, PTFE-GO, NF, and UF membranes, respectively (Fig. 4). Furthermore, the morphology of fouling layers varies. The GO membranes exhibited a fouling layer with larger pores than did the NF and UF membranes. Hence, it is expected that GO membranes can exhibit higher water flux recovery because of the unique porous fouling layer.
The surface chemistry of membranes before and after milk filtration was characterized by Raman, ATR-FTIR, X-ray photoelectron spectroscopy (XPS), and contact angle measurements to study changes within the GO layer and the fouling layer. The Raman measurements of all membranes exhibited an increase in peak intensity near 2900 cm−1 (Fig. S6), corresponding to CH vibrations from aliphatic and aromatic molecules . Regarding the GO membranes, the ID/IG ratio showed a slight increase after milk filtration for the PS-GO membrane from 0.60 to 0.63 and for the PTFE-GO membrane from 0.98 to 1.00; the FWHM of the G-band in the PS-GO membrane varied from 77 to 70, and in the PTFE-GO membrane from 98 to 75. GO was deposited similarly on PS and PTFE; thus, changes in Raman spectra can arise from interactions between GO and the support membranes. Surprisingly, large peak intensities corresponding to lactose (1110–1220 cm−1) and aromatic amino acids (935–1035 and 1515 cm−1) were observed (Fig. S6c) . GO has been previously observed to exhibit enhanced Raman spectroscopy, which is attributed to the electronegativity of oxygen functional groups, and favorable enhancement by increasing the number of GO layers . On the other hand, aromatic amino acids have a resonance effect at an excitation wavelength of 532 nm . Thus, the porous structure of the fouling layer and the support PTFE membrane allow low occlusion of the GO membrane with adsorbed lactose and amino acids. For commercial membranes, the peak intensities corresponding to the membranes were diminished, whereas the peaks for the fouling layers were dominant, which suggests that the fouling layer can be thicker or denser on commercial membranes.
Regarding the ATR-FTIR spectra (Fig. S7), the PS-GO membrane exhibited a lower peak intensity for lactose (1040 cm−1) than those for fat and proteins, and the PTFE-GO membrane exhibited a higher lactose peak intensity than those of other components of milk (fat and proteins) (Fig. 5a and S7). The spectra of the PTFE-GO membrane after milk filtration suggest a higher content of lactose within the membrane and the fouling layer compared with the other membranes, which is in accordance with the TGA and ATR-FTIR analyses of permeate solutions (Fig. 3). Interestingly, the fouling on commercial NF and UF membranes exhibits a similar composition as that observed for dry milk.
The surface analysis via XPS may detect up to 6 nm in penetration depth; thus, it mainly provides information for the fouling layers after milk filtration. Before milk filtration, the PS-GO and PTFE-GO membranes exhibit similar chemical bonds in C1s and O1s spectra (Fig. S8). This indicates that the variations observed in Raman spectroscopy are due to the interaction between GO and support membranes. Overall, all membranes after milk filtration exhibit similar chemical bonds for C1s and O1s (Figs. S8–S10), as observed by ATR-FTIR (Fig. S7), because of the presence of the fouling layer on the surface of membranes. Interestingly, there is a larger intensity for CO bonds in the PTFE-GO membrane after milk filtration than in other membranes, because of the higher presence of lactose, which agrees with ATR-FTIR data (Fig. 5a and S8).
The hydrophilicity of membranes after milk filtration may provide further information regarding the nature of the foulant, specifically whether it is composed mainly of lactose, protein, or fat. Thus, the contact angle was measured for all membranes before and after milk filtration; in all membranes, the contact angle increased after milk filtration (Fig. 5b), which indicated the presence of nonpolar compounds within the fouling layer. Change in contact angle was higher in the PS-GO and PTFE-GO membranes possibly because of the porous structure of the foulant layer since the molecular structure of all foulant layers is similar, and the presence of a hierarchical structure may increase the contact angle . The surface chemistry analysis of the membranes before and after milk filtration indicates that GO does not exhibit significant changes in its molecular structure and the foulant layer on all membranes exhibits a similar composition as milk, i.e., it is composed of lactose, proteins, and fat, as shown via XPS and ATR-FTIR. Interestingly, salts were not detected through XPS. Particularly, the foulant layer on the PTFE-GO membrane has the highest lactose content, which is in accordance with the chemical analysis of permeate solutions.
2.3. Protein antifouling
During the initial stages of milk filtration, proteins are generally adsorbed on membranes and promote further growth of the fouling layer [13,14]. Protein fouling was evaluated on the PS-GO membrane (Fig. 6a) to reduce the influence of surface roughness from the PTFE support membrane. The PS-GO membrane was kept in a static dispersion of BSA and lysozyme at 200 ppm for 3 days at pH 8. Typically, protein fouling is observed as a lump deposit with an irregular shape . The membrane surface remained smooth with some wrinkles similar to those of the pristine membrane (Fig. 6b and c), and we did not observe fouling for either protein, possibly because of the low surface roughness of PS-GO membranes.
The PS-GO membrane was also monitored during the cross-flow operation with proteins (BSA and lysozyme) for 3 days at 0.5 MPa and washed with distilled water within the cross-flow system. Subsequently, the membranes revealed a larger number of wrinkles on the surface because of the shear stress and hydration during the water cross-flow. Neither membrane exposed to BSA and lysozyme exhibited evident protein deposition by SEM (Fig. 6d and e). AFM images of GO membranes after cross-flow filtration indicate that the PS-GO membrane exposed to BSA has a morphology that is similar to that of the PS-GO membrane exposed to pure water (Fig. 6f and g). However, few irregular deposits were observed when lysozyme was present in the source (Fig. 6h). The GO membrane has a negative charge at the pH level used (7–8) , whereas BSA and lysozyme are known to have negative and positive charges, respectively . Microscopy observations indicate that lysozyme was attracted to the GO surface, because of the presence of carboxyl functional groups yielding a negative charge on the PS-GO membrane and the positive charge of lysozyme . By contrast, BSA yielded no observable deposit and showed a charge-dependent protein fouling, which has been also previously observed on silica and GO surfaces [36,39]. Additionally, the hydrophilic nature of GO can promote stronger interaction with water via H-bonds compared with organic molecules because of the presence of carboxylic bonds, thus yielding low fouling . Another factor that promotes fouling is surface roughness; surface roughness should be significantly larger than the protein size to exhibit an effective higher surface area and lower shear force to promote fouling . Therefore, the low surface roughness of PS-GO membrane (10 nm Rms) and the small size of the studied proteins (<10 nm)  yield low fouling. Because of the abovementioned reasons, milk fouling can be initiated via negative protein adsorption on the PS-GO and PTFE-GO membranes. Furthermore, the fouling layer affects the lactose separation performance.
To understand the lactose permeation mechanism, the van der Waals and Coulomb interaction energies of GO with lactose were calculated. Lactose-GO interacts favorably via van der Waals forces because it exhibits higher interaction energy (−1.567 eV) than Coulomb interactions (0.735 eV). The interaction of lactose with water was also studied (shown in Fig. S11a), this interaction occurs mainly through hydrogen bonds between the atoms of lactose and O atoms of water, as well as O atoms of lactose and H atoms of water. On average the lactose molecules had 4 hydrogen bonds with water. Both radial distribution functions (RDF) for the atoms involved in hydrogen bonds (Figs. S11b and S11c) exhibit similar trends, the first peak appears at 1.92 Å because of the hydrogen bond. The RDF of the carbon atoms shows a double shell of water with weak interaction. Thus, the van der Waals dimensions for lactose can be considered for steric diffusion, its dimensions are approximately 5 Å × 9 Å × 12 Å. This result suggests that nanochannels in the GO membrane microstructure should be larger than 5 Å to allow the diffusion of lactose. We carried out a simulation with GO sheets, water, and lactose molecules under 0.5 MPa (Fig. S12a). The GO sheets exhibit an interlayer distance of 10.5 Å, close to that observed by XRD (Fig. S5). The hydrogen bond analysis indicates a stronger interaction between GO and water, and lactose and water than lactose and GO. The diffusion of lactose through the GO membranes is because of (1) the weak interaction between oxygen functionalities in GO and lactose (Fig. S12c), (2) the dimensions of the nanochannels in the GO microstructure, and (3) the externally applied pressure allows lactose diffusion through the GO membrane. The simulations and experimental results indicate the viability of lactose diffusion between GO sheets and through nanochannels (Fig. S12b), as confirmed by XRD (Fig. S5).
During milk filtration, the membranes exhibited fouling layers initiated via protein adsorption, and then influenced the filtration performance [13,14]. Among the main components present in milk, positively charged proteins (e.g., lysozyme) are more likely to initiate fouling because of Coulomb interactions, whereas fat globules and many proteins in milk are negatively charged. To improve the understanding of protein fouling, MD simulations were performed with GO in the XY plane kept within a solvated box with BSA and lysozyme; the system was studied under low (4.34 × 10-6 eV/Å per atom) and high (4.34 × 10−5 eV/Å per atom) water force, to resemble the cross-flow filtration process where there can be a variance in water flux across the surface because of turbulence (Fig. 7 a,b). The total simulation times were 1 and 10 ns for high and low applied water forces, respectively. The protein velocity for BSA and lysozyme was monitored for both applied water forces (Fig. 7 c,d), which showed that the BSA protein had a higher velocity despite its larger molecular weight compared with that of lysozyme (Table S2). This indicates that lysozyme has a stronger interaction with GO than GO-BSA. Protein velocity gradually increased with time before reaching a stable point near the end of the simulation. Along the z-axis, perpendicular to the GO sheet, the water velocity is lower near the GO surface because of the hydrogen bonding established between the oxidation sites of GO and water (Fig. S13) . Notably, the water velocity increases as the distance of water from GO increases; later, it decreases because of the periodicity of the simulation box in which another GO sheet is on the top side. Compared with polymeric membranes, to which the nonpolar parts of BSA have a strong affinity, the strong hydrophilicity of GO has been considered as a key advantage for reducing protein fouling . Additionally, the charge-repulsion effect may contribute to fouling [36,39].
The hydrogen bonds established between the GO layers and proteins were examined to better understand the fouling mechanism. At a high-water force (Fig. 7e), BSA quickly decreases the number of H-bonds to zero, allowing it to move freely without interacting with GO. In contrast, lysozyme exhibits H-bonds for a longer time and eventually exhibits no interaction with GO due to the high-water force. The number of hydrogen bonds for GO-BSA was lower than that for GO-lysozyme because of the negative charge of BSA, although it has a higher number of interaction points (Fig. 7e and f). The larger number of hydrogen bonds for GO-lysozyme was sufficiently strong to produce lower lysozyme velocity than that of BSA. MD analysis confirms the experimental observations that GO-BSA exhibits a much lower fouling than does GO-lysozyme. It also confirms that positively charged molecules within milk can initiate fouling during filtration by blocking the surface of GO and modifying its net surface charge.
From the experimental and theoretical analysis of milk filtration with GO membranes, the following observations can be determined. GO membranes can filter lactose from milk at a higher flux than commercial NF and UF membranes used in this study. Water that is present in milk diffuses through the membrane and acts as a driving force for lactose diffusion through the GO membranes because of weak van der Waals interactions, the interlayer distance between GO sheets, and nanochannels that are present in the GO microstructure. Meanwhile, fat and proteins remain above the membrane because of their larger molecular weight, and salts can be partially rejected because of Coulomb interactions with GO (Fig. 8a). It has been previously reported that proteins initiated fouling on membranes and later affected the filtration performance of food [13,14]. Protein fouling on GO membranes is affected by Coulomb interactions where negatively charged proteins exhibit negligible fouling, whereas positively charged proteins demonstrate a larger interaction and promote the growth of the fouling layer, which ultimately affects the lactose permeate flux. Surface roughness and chemistry can also determine the properties of the fouling layer. The PS-GO membrane exhibits a smooth hydrophilic (contact angle 27°) surface. Conversely, the PTFE-GO membrane surface is rougher, less hydrophilic (contact angle 46°), and more electronegative; however, both membranes exhibited a porous fouling layer possibly because of the carboxylic acid functional groups of GO, whereas commercial membranes exhibited fewer carboxylic acid functional groups (Figs. S8–S10), and a more dense fouling layer. The unique surface chemistry of GO can determine the formation of a porous fouling layer (Fig. 8b) and allow higher permeation of lactose compared with the commercial NF and UF membranes despite exhibiting higher contact angles on the foulant layers than on pristine GO membranes. By contrast, commercial NF and UF membranes exhibit denser fouling layers, which hinders the lactose permeate flux.
In this study, our novel GO membranes (PS-GO and PTFE GO) were compared with commercial NF and UF membranes. UF polymer-based membranes can permeate up to 100% of lactose from dairy sources (Table S3) ; however, they are not ideal for lactose removal from milk, because fat can diffuse through the membrane as observed via FTIR spectroscopy. By contrast, NF polymer-based membranes are typically used for lactose removal from milk, which allows the passage of a low amount of lactose (<10%) (Table S3) [1,6,, , ]. The obtained results indicate that the developed novel PTFE-GO membranes yield higher lactose concentration in the permeate (approximately 4.86%) than do NF membranes (approximately 1%), and higher lactose permeate flux (2.87 kg m−2 day−1) than do UF membranes (1.61 kg m−2 day−1) without the passage of fat globules. Lactose permeation through the GO membrane is attributed to the nanochannels in the GO microstructure, which is caused by the mismatch in GO sheet alignment during spray coating, and the weak van der Waals interaction between GO and lactose.
GO membranes exhibit high water flux recovery because of the porous fouling layer. The protein fouling results demonstrate that fouling was lower for BSA, a negatively charged protein, compared with that of lysozyme, a positively charged protein. Additionally, MD studies indicated that protein and GO layer interactions were dominated by hydrogen bonding. The obtained results confirm that the charge of the solute can affect antifouling in GO membranes. Interestingly, the support membrane for GO greatly affects the permeate flux. The PTFE-GO exhibits support membrane exhibits a larger pore size (1 μm), which yields a more porous foulant layer and increases the lactose permeate flux (2.87 kg m−2 day−1) by more than one order of magnitude compared with the PS-GO support membrane containing 20 nm pores (0.20 kg m−2 day−1). The developed membrane technology allows the use of GO membranes in the dairy industry for lactose removal and concentration of fat and proteins, with better performance than that of typical polymeric NF membranes. This unique selectivity can help preserve the original milk flavor, and the developed technology is be expected to remove sugars from other beverages to improve the nutrition content of diverse commercial products. In the future, GO membranes can be extended to wastewater treatment, bioreactors, filtration of natural water sources, and medical devices.
4. Experimental section
4.1. Graphene oxide synthesis
GO membranes were prepared in the same manner as previously reported . Briefly, GO was synthesized according to the method reported by Marcano et al.  First, 5 g of graphite was dispersed in 200 ml of H2SO4 and 40 mL of H3PO4 under magnetic stirring, and 25 g of potassium permanganate (KMnO4) was slowly added. The reaction was kept at 40 °C; after 1 h, graphite was exfoliated, and the reaction media turned into a paste. Subsequently, the reactor was mixed with a Teflon rod every 5–10 min. Then, the mixture was cooled to room temperature after 3.5 h of oxidation and slowly dispersed in a solution of 600 ml of cold water and 40 ml of 35 wt% H2O2. This step was accompanied by heating and bubbling. Finally, the dispersion was kept static overnight to allow sedimentation of GO.
To purify the GO dispersion, the clear supernatant was discarded, and the remaining sediment was diluted in 1 L of 1.0 wt% H2SO4. Then, the dispersion was mixed with a roller to avoid damaging the GO sheets. Sulfuric acid was removed by several cycles of centrifugation and redispersion in distilled water. When the acid content is decreased, GO expands and separates into two types of sediment during centrifugation, with the bottom of the tube containing a solid pellet of large nonexfoliated particles, and the top of the tube consisting of exfoliated GO sheets remaining as a thick gel. The thick gel was collected, redispersed in water, and then centrifuged to remove any remaining large particles. Next, the GO dispersion was centrifuged to remove water and obtain a slurry with a pH of 3.0–3.5 and 0.9 wt% of solids. Finally, a solution of 1 mg/ml GO in water was ultrasonicated for 20 h.
4.2. Membrane preparation
The support membrane of PS or PTFE was kept in a solution of poly (vinyl alcohol) at 1 wt% for 1 h, and then left to dry at room temperature. Then, the GO dispersion was spray-coated onto the support membrane at 0.2 MPa with an airbrush (ANEST IWATA) held at 15 cm from the surface. Subsequently, the GO membrane was dried at room temperature and kept at 100 °C for 1 h, followed by soaking in a mixture of ethanol:water (1:3 V/V) for 1 h and finally left to dry at room temperature.
4.3. Membrane structure characterization and filtration performance
The GO dispersion was drop cast onto a Si substrate for SEM and histogram analysis with HR-SEM, using a JSM-7000F/1V microscope (JEOL). Protein fouling was performed using BSA or lysozyme in water at 200 ppm and pH 8 as model proteins. Commercial milk from Meiji (4.95 wt % lactose, 3.9 wt% fat, 3.4 wt% protein, 0.16 wt% minerals, and 87.59 wt% water) acquired from a local supermarket was used for milk filtration tests (Table S1). Initially, distilled water was used as a source for the cross-flow filtration equipment (Fig. S14) operated at 0.5 MPa and 300 ml/min water flow above the membrane surface (2.5 cm diameter). After the permeate flux was stabilized the protein solution or milk was used as the feed for several hours, and then changed back to water with an intermediate water washing step with a 1000 ml/min water flow. This cycle was repeated two or three times to evaluate fouling, milk filtration, and membrane washing. Irreversible fouling was calculated as:
J is the permeate flux at the end of the water filtration stage and J0 is the permeate flux at the end of the initial membrane stabilization in water.
For static fouling, GO membranes were kept in a protein solution for three days and then gently rinsed with water. The membranes before and after static fouling and cross-flow filtration were observed by SEM and AFM (Agilent Technologies AFM 5500, Santa Clara, CA, USA) under the tapping mode. Lactose, casein, lysozyme, BSA, milk and permeate were separately dispersed in water and drop cast onto Si substrates for ATR-FTIR (Thermo Scientific Nicolet 6700 FT-IR, Waltham, MA, USA). X-ray diffraction of the GO membrane in the dry state and hydrated in milk after one day was performed with Rigaku SmartLab (Cu radiation, kα = 1.54 Å). Milk and permeate solutions were analyzed by thermogravimetry using a Hitachi STA7300 instrument; the solution was evaporated at 80 °C for 1 h and then heated up to 1000 °C at 10 °C/min. Zeta potentials were measured by an electrokinetic analyzer (SurPASS3, Anton Paar, Graz, Austria). Two pieces of each sample, with the dimensions of 1.0 × 2.0 cm, were fixed with a waterproof adhesive tape and then set facing each other with a separation of 110 μm. For titration, a mixture of KOH (0.05 mol/L) and HCl (0.05 mol/L) was used, and KCl (1.0 mmol/L) was prepared as an electrolyte solution. X-ray photoelectron was measured with a Kratos Axis-Ultra (Manchester, UK) equipped with an Al Kα line and using pass energy for the detector at 160 and 20 eV for high-resolution scans.
4.4. Computational methodology
The GO molecular structure was modeled by randomly combining different GO building blocks bearing functional groups similar to the method reported in the literature . The GO membrane model was constructed of three GO layers in which the lateral dimensions were 12.3 nm × 12.3 nm. The structures of BSA, lysozyme, and lactose were extracted from PDB files downloaded from the Protein Data Bank (4F5S for BSA, 1DPX for lysozyme, and LBT for lactose).
All calculations were performed via classical MD using the LAMMPS simulation package . We used the DREIDING force field for GO , CHARMM force field  for BSA and lysozyme, and for the lactose-water solvation study. For GO, atom charges were set with estimated values using the NWCHEM code with the 6-31G∗∗ level basis and the Hartree–Fock method . For the GO-lysozyme and GO-BSA simulations, counterions were added to adjust the system’s net charge. For water molecules, the SPC/E model was adopted . The interactions between molecules were calculated using the Lennard-Jones potential and Coulomb interactions with a particle-particle mesh solver . All calculations were performed under periodic boundary conditions with time steps of 1 fs.
For fouling simulations, we used an NPT ensemble at 1 atm and 300 K with the Nosé–Hoover method . To simulate the behavior of proteins against the water flow, we placed proteins on GO and added unidirectional forces to water molecules in parallel to the GO surface. The system net charge was adjusted by adding counterions to confer a negative charge to GO and BSA, and a positive charge to lysozyme. The NVT ensemble was used for energy calculation of lactose and its complex system with GO. It was constructed by setting lactose on GO with at a distance of 15 Å in vacuum and equilibrated by MD with a time step of 0.1 fs. Then the interaction energies were calculated according to the following equation:
Here, EGO-X corresponds to the energy of the complex system, and EGO, EX to the energy of the individual components.
Simulations of the GO/water/lactose structure were carried out packing two GO sheets (5 × 5 nm) with 400 molecules of water and 20 lactose molecules. We used the ReaxFF force field developed for biomolecules in water solution  as implemented in LAMMPS- (29-oct-20) . The molecules were loosely arranged in a cell of 6 × 6 × 4 nm using packmol , and the cell was later equilibrated for 100 ps then slowly compressed over 240 ps to reach the targeted GO density (1.6), then equilibrated for 100 ps at 6 atm under PVT conditions. The last 20 ps were used to analyze the hydrogen bonds and order in the structure, ant the images were prepared with VMD .
Credit author contribution statement
Aaron Morelos-Gomez: designed the experiments, performed FTIR, SEM, filtration, and fouling experiments, participated in discussions and the manuscript writing. Souya Terashima: performed FTIR, SEM, filtration, and fouling experiments, performed molecular dynamics simulations. Ayaka Yamanaka: performed molecular dynamics simulations. Rodolfo Cruz-Silva: carried out GO synthesis, participated in discussions and the manuscript writing. Josue Ortiz-Medina: performed AFM. Roque Sánchez-Salas: carried out GO synthesis, performed X-ray diffraction. Juan L. Fajardo-Díaz: performed XPS and Raman characterizations. Emilio Muñoz-Sandoval: participated in discussions and the manuscript writing. Florentino López-Urías: participated in discussions and the manuscript writing. Kenji Takeuchi: participated in discussions and the manuscript writing. Mauricio Terrones: participated in discussions and the manuscript writing. Morinobu Endo: participated in discussions and the manuscript writing.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
A.M.G., R.C.S., K.T., S. Tejima and M.E acknowledge that this work was supported by the Center of Innovation Program, Global Aqua Innovation Center for Improving Living Standards and Water Sustainability, from the Japan Science and Technology Agency (JST). We are grateful to Asbury Carbons for providing the graphite used for this research, and would like to thank Takumi Araki for his valuable discussions and aid in MD simulations. The numerical calculations were carried out at Shinshu University and the Research Institute for Information Technology in Kyushu University.
Appendix A. Supplementary data
The following is the Supplementary data to this article: