Removal of Antibiotics From Water with an All-Carbon 3D Nanofiltration Membrane

Recent industrial developments and increased energy demand have resulted in significantly increased levels of environmental pollutants, which have become a serious global problem. Herein, we propose a novel all-carbon nanofiltration (NF) membrane that consists of multi-walled carbon nanotubes (MWCNTs) interposed between graphene oxide (GO) nanosheets to form a three-dimensional (3D) structure. The as-prepared membrane has abundant two-dimensional (2D) nanochannels that can physically sieve antibiotic molecules through electrostatic interaction. As a result, the prepared membrane, with a thickness of 4.26 μm, shows both a high adsorption of 99.23% for tetracycline hydrochloride (TCH) and a high water permeation of 16.12 L m− 2 h− 1 bar− 1. In addition, the cationic dye methylene blue (MB) was also removed to an extent of 83.88%, indicating broad applications of the prepared membrane. Electronic supplementary material The online version of this article (10.1186/s11671-018-2555-9) contains supplementary material, which is available to authorized users.


Background
Recently, pharmaceutical compounds, especially antibiotics, have attracted increasing attention around the world because their occurrence in natural water poses a threat to ecosystems and public health, even at low concentrations [1,2]. To date, various technologies have been developed with the aim of eliminating antibiotics from the aquatic environment, such as oxidation processes and adsorption [3,4]. Oxidation processes, such as photocatalysis, sonolysis, and the Fenton reaction, involve complex procedures, whereas membrane-based separations are potentially simpler [5]. However, many currently available membranes for the removal of smaller antibiotic molecules are less effective because they merely operate through a size-exclusion effect [6].
In recent years, carbon-based materials have been used as adsorbents for the removal of antibiotics [7,8]. Particularly, graphene has also been widely applied to remove pollutants from water due to their one-atom-thick nature, high specific surface areas, and porous structures [9][10][11]. Graphene oxide (GO) has distinctive structural features [12], excellent hydrophilicity, strong antifouling properties [13], and high mechanical strength. These properties make it suitable for applications in water purification and desalination. In addition, GO may be produced on large scale, in contrast to pristine graphene [14]. However, because of the stacking tendency of GO nanosheets, polymeric materials or large nanoparticles need to be intercalated between them to increase the interlayer spacing [15,16]. Carbon nanotubes (CNTs), as one-dimensional (1D) materials with excellent properties and compatibilities, have proved to be ideal "nanowedges" for regulation of the interlayer spacing of GO [17]. Compared with single-walled carbon nanotubes (SWCNTs), multi-walled carbon nanotubes (MWCNTs) provide greater stability under hydrodynamic flow conditions [18]. Moreover, the increased interlayer spacing by intercalation of GO nanosheets with MWCNTs has been proved to enhance water flux. However, undesirable aggregation of CNTs in aqueous solution often hampers application of CNT/GO-based membranes. Conversely, various polyelectrolytes have been used to enhance the dispersion of CNTs through functionalization [19,20].
In this work, we propose a novel all-carbon nanofiltration (NF) membrane that consists of MWCNTs interposed between GO nanosheets. Poly diallyldimethylammonium chloride (PDDA), as a cationic polyelectrolyte, was grafted onto the MWCNTs to ensure their cationization, imparting strong antifouling properties because of the excellent dispersity. Due to the oxygen-containing functional groups appended both irregularly along the edges and on the surfaces of GO sheets, GO can be regarded as an anionic polyelectrolyte. Thus, the reaction between PDDA-MWCNTs and GO was mainly as a result of charge interaction. The prepared NF membrane was systematically characterized and used as an absorbent for the removal of tetracycline hydrochloride (TCH) and methylene blue (MB) as model organic pollutants. The concentrations of the filtered TCH and MB solutions were determined by UV/Vis spectrophotometry.

Results and Discussion
The MWCNTs/GO hybrid was used to produce a freestanding and flexible membrane based on a simple vacuum filtration method. As illustrated in Fig. 1a, PDDA, the cationic polyelectrolyte, could be easily appended on the surface of the MWCNTs, conferring a positive charge. GO, as an anionic polyelectrolyte, could then react with the positively charged MWCNTs through electrostatic interaction. Finally, an ultrathin membrane was prepared by vacuum filtration of the above dispersion.
The possible adsorption process is illustrated in Fig. 1b. The introduction of MWCNTs between the GO nanosheets opened nanoscale channels, which allowed for improved flow of water molecules [21]. TCH molecules were intercepted in the nanoscale channels due to steric hindrance and their covalent interaction with the functional groups of the prepared all-carbon membrane.
An optical image of a PDDA-MWCNTs/GO membrane is shown in Fig. 2a. The prepared all-carbon membrane was like cloth and exhibited excellent mechanical flexibility (Fig. 2b). The as-prepared membrane was shown to be hydrophilic by a water contact angle  Figure S1) [22,23]. Nevertheless, it proved to be stable when soaked in water (Fig. 2c). Moreover, the membrane could be reused more than seven times without developing any obvious cracks (Fig. 2d).
An SEM image of a cross-section of the prepared membrane is shown in Fig. 2e. The thickness of the membrane was estimated as 4.26 μm, and the MWCNTs were seen to be uniformly inserted between the GO sheets. Moreover, wrinkles on the surface of the prepared membrane were apparent from an AFM image (Additional file 1: Figure S2A), leading to a larger contact area with contaminants. TEM images of the all-carbon membrane revealed that the modified MWCNTs were well dispersed within the GO, in accordance with the SEM results.
As shown in Fig. 3a, compared with the GO membrane, the PDDA-MWCNTs/GO membrane exhibited a more porous structure. Variation of the concentration of PDDA influenced the dispersion of MWCNTs in the GO (Fig.  3b-e). Because of the strong adhesive property of PDDA [24], a high concentration thereof (20 wt%) resulted in agglomeration of the MWCNTs (Fig. 3e). However, as can be seen from Fig. 3a-d, PDDA concentrations of 0, 2, 5, or 8 wt% were insufficient to disperse 4 mg of MWCNTs, and the thickness of the membrane was affected. Possible reasons are as follows. Firstly, the membranes were thick partly because MWCNTs were easily agglomerated at low PDDA concentrations. Secondly, non-covalent interaction between a suitable concentration of PDDA and GO sheets can lead to an ultrathin membrane. Thirdly, because MWCNTs become almost enwrapped with long-chain PDDA at excessively high concentrations, large porous structures are formed. However, the mechanism of such pore formation is as yet unknown. It was concluded that a PDDA concentration of 10 wt% gave the optimal dispersion of MWCNTs in GO (Fig. 2e). The morphological characterization including SEM and TEM is presented in Additional file 1: Figure S3. It can be observed that the PDDA is successfully modified into the surface of CNTs, and the thickness of PDDA is around 5.2 nm [25]. Nitrogen adsorption-desorption isotherms were used to characterize the porosities of the MWCNTs/GO and PDDA-MWCNTs/GO membranes (Fig. 4). PDDA-MWCNTs/GO showed increased adsorption-desorption capacity compared with the MWCNTs/GO membrane. The PDDA-MWCNTs/GO NF membrane had a higher specific surface area (402.96 m 2 g −1 ) than the MWCNTs/ GO membrane (378.45 m 2 g −1 ). Moreover, a typical type IV nitrogen isotherm with hysteresis loops for the prepared NF membrane corroborated its mesoporous properties [26]. The inset image shows the corresponding pore size distribution calculated by the Barrett-Joyner-Halenda (BJH) model, which indicates that the pores of both membranes were about 3-10 nm in diameter, consistent with the N 2 isotherm.
As shown in Fig. 5a, a characteristic diffraction peak for the GO sample was observed at 11.02°(001), indicating a distance between the nanosheets of 0.80 nm [27], whereas for the MWCNTs peaks were observed at 26.96°(002) and 44.89° [28], in accordance with previous literature reports. After the incorporation of MWCNTs into GO, the characteristic diffraction peaks of both components significantly decreased due to inhibition of the restacking of GO nanosheets and of aggregation of MWCNTs, reflecting the low propensity for crystallization of the hierarchical NF membrane.
Moreover, the diffraction peak of the GO nanosheets slightly shifted from 11.02°to 10.63°, corresponding to an increase in interlayer spacing from 0.81 to 0.87 nm. Bands in the wide-scan XPS spectra of the respective membranes confirmed the increased N 1 s content in the PDDA-MWCNTs/GO membrane. Figure 6a shows the UV/Vis absorption spectra of the initial TCH solution (20 mL, 500 μm) and of the filtrates obtained after passage through the MWCNTs/ GO and PDDA-MWCNTs/GO membranes. After filtration through the all-carbon membranes, the solution showed lower absorption intensity in the region up to 420 nm. Combined with the inset image, the remaining concentrations of TCH after filtration through the MWCNTs/GO and PDDA-MWCNTs/GO membranes were 18.78 and 6.74 μM, respectively. The adsorption capacities could be converted into adsorption percentages, which were evaluated as 95.04% for MWCNTs/GO and 99.23% for PDDA-MWCNTs/GO, following a single filtration through each membrane. Thus, compared with the MWCNTs/GO membrane, the PDDA-MWCNTs/GO membrane caused a more pronounced decrease in absorption intensity. From these results, we can conclude that TCH filtration involves both the interfacial functional groups and a synergistic effect. In addition, the water permeation flux for the PDDA-MWCNTs/GO membrane was evaluated as 16.12 L m − 2 h − 1 bar − 1 after functionalization, around twice that for the MWCNTs/ GO membrane. The PDDA-MWCNTs/GO membrane clearly demonstrated the best results in terms of both high adsorption capability and water permeation flux. Figure 6b shows the static adsorption of TCH. In this experiment, the mass of the PDDA-MWCNTs/GO membrane was 1 mg. After static adsorption, the amount adsorbed on the PDDA-MWCNTs/GO membrane was 436.13 mg g −1 , confirming its high capacity for TCH removal from water.
Stability is important for practical application of NF membranes. Here, we conducted adsorption experiments in harsh environments, namely, basic, acidic [29], and ionic conditions (Fig. 7). We anticipated that pH would affect the electrostatic interaction by regulating the charges on both TCH and the membrane. It is found that the zeta potential of the membrane is around − 45 mV, while TCH has a positive and negative charge in the acid and alkaline conditions, respectively [30]. When the pH raised from 2 to 4 or from 8 to 10, the adsorption of TCH was slightly decreased (Fig. 7a). This may due to membrane swelling or electrostatic repulsion [31]. Neutral pH proved to be optimal, and all further  As we could see, for the PDDA-MWCNTs/GO membrane, the adsorption behavior was only slightly influenced. From this, it could be surmised that the main adsorption mechanism is capture of molecules in the nanochannels. As shown in Fig. 7b, the adsorption of TCH decreased with increasing salt concentration. Saltinduced membrane swelling and salting-out effects may have synergistically influenced the adsorption performance [32]. Nevertheless, the prepared membrane showed a moderate tolerance for saline ions.
We used an MB dye as a positively charged molecule to further study the adsorption mechanism. We concluded that a synergistic effect was operative, involving molecular sieving in the nanochannels, the hydrophobicity of the non-oxide nanochannels in the membrane, and charge interaction between the solute and the membrane [33]. The ζ-potential for the PDDA-MWCNTs/GO NF membrane is at around pH 7, indicating good stabilization, and it showed greater adsorption of the zwitterionic TCH (99. 23%) than that of the positively charged MB dye (88.23%). We attribute this to size exclusion being the major factor in the controlled nanoscale channels [34,35], consistent with the results of pH experiments.

Conclusions
In summary, we have proposed a novel 3D all-carbon NF membrane with tremendous properties, namely, ultrathin nanosheets with high adsorption, stabilization with extraordinary anti-fouling properties, and rapid water permeation. The synthesis is rapid and environmentally friendly, making it a promising method for the fabrication of NF membranes. The functionalized PDDA-MWCNTs/GO NF membrane exhibited superior properties compared to the MWCNTs/GO NF membrane due to the high dispersion of MWCNTs and charge interaction between the components. The high adsorption performance can be attributed to a synergistic effect between molecular sieving, the hydrophobicity of the non-oxide nanochannels in the membrane, and the charge interaction between the solute and the membrane. The simple preparation process combined with the many extraordinary properties makes this functionalized MWCNTs/GO NF membrane a promising candidate for chemical separation applications.

Apparatus
Scanning electron microscope (SEM) images of the prepared all-carbon membrane were acquired with a fieldemission scanning electron microscope (FESEM, Ultra 55, Carl Zeiss, Germany). Suspensions of graphene oxide and MWCNTs were dropped onto carbon-coated copper grids, and the volatiles were evaporated under ambient conditions. Transmission electron microscopy (TEM) was performed using a Hitachi H-800 electron microscope (Japan) operated at an accelerating voltage of 200 kV. UV/Vis spectra were recorded on a Lambda-25 spectrometer (Perkin-Elmer Inc. USA). Brunauer-Emmett-Teller (BET) measurements were performed at 77 K on an Autosorb-iQ-C analyzer (Quantachrome Instruments, USA). X-ray diffraction (XRD) patterns were obtained with a Shimadzu XD-3A diffractometer (Japan), employing CuKα radiation, λ = 0.15418 nm. Various elements in the samples were determined by Xray photoelectron spectroscopy (XPS, PHI 5000 Versa Probe, Japan). Static water contact angle measurements were performed at 25°C using a contact angle meter (Rame-Hart-100, USA) employing drops of pure deionized water. Zeta potentials of the membranes were tested by means of a SurPASS Electrokinetic Analyzer (Austria) with a clamping cell at 300 mbar. A Bruker Multimode 8 atomic force microscope (AFM, Germany) was employed to characterize the prepared nanomaterials, which had been coated on a mica substrate.

Synthesis of PDDA-Functionalized MWCNTs
PDDA-functionalized MWCNTs were prepared as described previously [36]. MWCNTs (4.0 mg) were first dispersed in deionized water (1 mL) with the aid of ultrasonication, and PDDA (10 wt%) in water was added dropwise. The centrifuged product was then washed several times with deionized water and dried in a vacuum oven at 70°C for 24 h.

Assembly of the MWCNT-Interposed GO (MWCNTs/GO) Membrane
MWCNTs (4.0 mg) were added to an aqueous suspension of GO (24 mL, 0.5 mg mL −1 ) under stirring and sonication. The homogeneous dispersion was then vacuum-filtered onto a porous polyvinylidene fluoride membrane with a pore size of 0.22 μm. Finally, the membrane was dried in a vacuum oven at 60°C for 3 min and could be easily peeled from the polyvinylidene fluoride membrane after soaking with anhydrous ethanol.

Antibiotic Adsorption Experiments Using the All-Carbon Membrane
In order to evaluate the adsorption performance towards antibiotics, TCH solution (20 mL, 500 μM) was vacuumfiltered at 0.9 bar through the prepared membrane. The concentration of the filtrate was determined by UV/Vis spectrophotometry. According to the determined concentration, then the rejection rates of the TCH molecules could be calculated through the following equation: where C o represents the concentration of TCH in the original solution and C p is the concentration of TCH in the permeate solution. All of the data were calculated based on the results from at least three experiments.

Additional file
Additional file 1: Figure S1. The water contact angle of the PDDA-MWCNTs/GO membrane. Figure S2 The AFM images of (A) PDDA-MWCNTs/GO membrane and (B) GO membrane. Figure S3