Synthesis and Characterization of Halloysite/Carbon Nanocomposites for Enhanced NSAIDs Adsorption from Water

The adsorption of ketoprofen, naproxen, and diclofenac (non-steroidal anti-inflammatory drugs, NSAIDs) on halloysite/carbon nanocomposites and non-modified halloysite were investigated in this work. Halloysite/carbon nanocomposites were obtained through liquid phase impregnation and carbonization using halloysite as the template and saccharose as the carbon precursor. Scanning electron microscopy (SEM), X-ray diffraction (XRD), Fourier-transform infrared spectrometry (FT-IR), and low-temperature nitrogen adsorption method were employed to study the morphological and structural changes of the halloysite/carbon nanocomposites. The effects of contact time, initial concentration of adsorbates, pH of solution, and mass of adsorbent on the adsorption were studied. Adsorption mechanism was found to fit pseudo-second-order and intra-particle diffusion models. The obtained experimental adsorption data were well represented by the Langmuir multi-center adsorption model. Adsorption ability of halloysite/carbon nanocomposites was much higher for all the studied NSAIDs in comparison to non-modified halloysite. Optimized chemical structures of ketoprofen, naproxen, and diclofenac obtained by Density Functional Theory (DFT) calculation showed that charge distributions of these adsorbate molecules and their ions can be helpful to explain the details of adsorption mechanism of NSAIDs on halloysite/carbon nanocomposites.


Introduction
In recent decades, a variety of new chemical compounds have been identified in wastewater. These trace compounds, mostly organic (personal care products, endocrine disrupting compounds, pharmaceuticals, and their transformation products) are known as the "emerging contaminants". Pharmaceuticals and personal care products (PPCPs) have been detected in surface water and wastewater in the nanogram/liter to micrograms/liter concentration range worldwide [1]. Active pharmaceuticals and their biotransformation products have a tendency for bioaccumulation [2] and can cause significant consequences to ecosystems [3,4]. Hence, it is important to develop efficient and economic methods for removing these compounds from wastewater. Several techniques have been studied, including oxidation, adsorption, electro degradation, bio-degradation, and photocatalytic degradation [5]. Adsorption is one of the most promising methods for removing organic micropollutants [6]. Adsorbents, including zeolites [7,8], activated carbon [9,10], biochar [11], activated hydrochars [12], carbon nanotubes [13], and composites with activated carbon [14], were Table 1. Adsorbents used for ketoprofen, naproxen, and diclofenac removal from aqueous solutions.

Preparation of Halloysite/Carbon Composites
Modification of halloysite was carried out in the following manner: Raw halloysite (RH) was dried, powdered, and washed with deionized water. Next, the magnetic fraction containing Fe 2 O 3 was separated using a 2-T magnetic separator. The obtained non-magnetic fraction of a halloysite mineral was cleaned with distilled water, dried, and followed by reductive bleaching with sodium hydrosulfite. Chemical composition of raw (RH) and modified halloysite (H) was published in ref. [35]. Particle fractions of 0.2-0.32 mm were used for the preparation of halloysite/carbon composites. Halloysite/carbon composites were prepared through liquid phase impregnation and carbonization. In detail, 20.0 g of H was added into 100 cm 3 of saccharose aqueous solutions (5, 10, 20, 30 wt%) [5C/H, 10C/H, 20C/H, 30C/H] and placed in an ultrasonic bath for removal of gases (1 h). Next, degassing mixture was allowed to shake (130 rpm) for 24 h at room temperature to ensure that saccharose could fill the outer and inner space between halloysite particles. After removing excessive saccharose solution, the solid was dried in 100 • C (24 h) and carbonized in a tubular furnace at 800 • C for 8 h with a heating rate of 5 • C/min under N 2 atmosphere. Finally, the black product was milled and sieved. The obtained halloysite/carbon products were called 5C/H, 10C/H, 20C/H, and 30C/H.

Halloysite/Carbon Composite Characterization
Textural structure of halloysite/carbon composites was characterized using the methods of low-temperature nitrogen adsorption-desorption isotherms (−196 • C) on a volumetric adsorption analyzer ASAP 2020 by Micromeritics (Norcross, GA, USA) (Structural Research Laboratory of Jan Kochanowski University in Kielce). Before measurement, all of the samples were degassed at a temperature of 200 • C for 2 h. Specific surface area (S BET ) of investigated carbon materials was determined with the Brunauer-Emmett-Teller (BET) method at a range of relative pressure from 0.05 to 0.2, considering the surface occupied by a single molecule of nitrogen in an adsorptive monolayer (cross-sectional area equals 0.162 nm 2 ) [36,37]. Total pore volume (Vt) (the sum of micropore volume (V mi ) and mesopores (V me )) was calculated from one point of nitrogen adsorption isotherm, corresponding to the relative pressure p/p o equals 0.99 [37].
Images of halloysite/carbon composites were obtained by SEM-EDS measurements carried out on a Tescan Vega3 microscope with LaB6 as electron source. Accelerating voltage was set to 5 kV in order to avoid excessive sample charring. Samples were put on standard SEM studs covered with an adhesive copper tape. In order to carry out elemental analysis, an Oxford Instruments x-act EDS detector was used. Additional high resolution SEM images were recorded after sputtering gold on the sample and using 30 kV accelerating voltage.
Infrared spectra were recorded using a Perkin-Elmer Spectrum 400 FT-IR/FT-NIR spectrometer with a smart endurance single bounce diamond, attenuated total reflection (ATR) cell. Spectra in the 4000-650 cm −1 range were obtained by the co-addition of 500 scans with a resolution of 4 cm −1 . Before the measurements, all samples were dried and powdered in an agate mortar.
Mineralogical compositions and sample structure were determined by the XRD analysis. X-ray diffraction patterns were collected using a Panalytical Empyrean diffractometer, equipped with a copper-based anode (Cu-K α LFF HR, λ = 0.154059 nm). The instrument settings were 40 mA and 45 kV and the step of scans and counting time were 0.02 • and 2 s per step, respectively.
Total carbon (TC) was quantified using an ELTRA CHS580 elemental analyzer (Neuss, Germany) (AGH). Total carbon was analyzed by placing approximately 0.250 g of dried, ground, and homogenized sample into a clean, carbon-free combustion boat. The sample boat was loaded into the furnace at 1350 • C. The combustion of the sample was performed in an oxygen atmosphere. Carbon content was determined on the basis of the amount of CO 2 formed and the mass of the sample used for the analysis. An IR detector was used to measure CO 2 .
Functional groups on the surface 5C/H, 10C/H, 20C/H, and 30C/H composites were identified using Boehm's titration method [38]. The Boehm methodology is based on quantifying different functional groups from titrations with bases of different strength: NaOH solutions were used to titrate the sum of carboxylic acids, phenols, and lactones; Na 2 CO 3 and NaHCO 3 were used to titrate carboxylic acids and lactones. Basic groups were quantified by titration with HCl as a total sum. The procedure was given a detailed description in [39]. The pH value at point of zero charge (pH PZC ) was determined by mass titration. In particular, 0.15 g of halloysite/carbon sample was placed in a beaker with aliquots of 50 mL of 0.1 M NaCl added. The samples were stored under agitation (150 rpm) and constant temperature (25 • C) for 3 h. After that, the pH of each solution was regulated by adding 0.1 M HCl and 0.01 M NaOH measured in the range of 1 to 12 [40].

Adsorption Measurements
Batch adsorption experiments were carried out in a 100 cm 3 Erlenmayer's flask containing a proper amount of adsorbent and 30 mL ultrapure water from the Milli Pore system. Next, 30 ml of the investigated adsorbate solution was added and the flask was put into the incubator for a defined period of time: 10, 15, 30, 45, 60, 120, 240, 300, 360, and 1440 minutes. The measurements were performed at 25 • C and a mixing rate of 150 rpm. After that, adsorbent was separated from the solution with cup-type centrifuge. Concentrations of ketoprofen, diclofenac, and naproxen solutions before and after the adsorption were determined with the spectrophotometric method, using a UV Shimadzu UV-1800 spectrophotometer. The wavelengths used to determine the studied adsorbate concentrations were as follows: 260.5 nm (ketoprofen), 275.5 nm (diclofenac), and 230.5 nm (naproxen).
The removal efficiency (R, %) and the amount of NSAIDs adsorbed at equilibrium (adsorption capacity, q e , mg/g) were calculated using Equations (1) and (2): where C 0 and C e (mg/L) are the initial and equilibrium concentrations of adsorbate solutions, V (L) is the volume of the adsorbate solution, and m (g) is the mass of adsorbent. Adsorbate concentration as a function of the adsorbed amount at equilibrium is an important relation in determination of attraction nature for a given adsorption system. In addition, the analysis of equilibrium data is useful for designing the adsorption unit. We checked the fit of the experimental data to the following adsorption models: Langmuir (one-center and multi-center) [41], Freundlich [42], Temkin [43], and Dubinin-Radushkevich [10]; together with three-parameter isotherms like Sips [44]. The names and equation forms of common adsorption models are collected in Table 3. Table 3. The names and equation forms of common adsorption models.

Adsorption isotherms
Freundlich q e = K F C 1/n e K F -Freundlich constant for a heterogeneous adsorbent (mg −1/n (dm 3 ) 1/n g −1 ); 1/n-the heterogeneity factor C e -equilibrium concentration of a solute in an aqueous solution (mg dm −3 ); q e -the amount of a solute adsorbed per gram of the adsorbent at equilibrium (mg g −1 ), q m -maximum monolayer coverage capacity (mg g −1 ); Langmuir (multi-center) q e = qm(KLFCe) ; q e -the amount of a solute adsorbed per gram of the adsorbent at equilibrium (mg g −1 ), q m -maximum monolayer coverage capacity (mg g −1 ); K LF -Langmuir-Freundlich isotherm constant (dm 3 g −1 ), n-adsorption model index Table 3. Cont.

Kinetics models
Pseudo-first-order ln(q e −q t ) = lnq e −k 1 t k 1 -pseudo-first-order rate constants (min −1 ); t-time of contact between the adsorbent and adsorbate (min); q e -amount of adsorbate at equilibrium (mg g −1 ); q t -amount of adsorbate at time t (mg g −1 ) Pseudo-second-order Intraparticle diffusion q t = k d t 1/2 + c k id -intra-particle diffusion rate constant (mg g −1 min −1/2 ); c-intercept, which represents the thickness of the boundary layer (mg g −1 )

Computational Methodology
The molecules and their corresponding ions were modeled by DFT (Density Functional Theory) calculations using a commercial Scigress program (version FJ 2.7). The structures of the studied compounds were subject to geometry optimization using the B88-LYP GGA functional and the DZVP basis [45,46].

Calculations
Electrostatic potential energy maps for naproxen, ketoprofen, diclofenac, and their anions are presented in Figure 1. In dependence on pH, the studied molecules exist in solutions in an undissociated form or as anions as indicated by the calculated degree of dissociation values collected in Table 4. Mainly undissociated molecules occur in highly acidic environments. As pH increases, more dissociated molecules appear. In an alkaline environment, 43% and 34% of the naproxen and ketoprofen molecules are dissociated, respectively. Therefore, it is advisable to include the electronic description of the molecules to explain its interaction with the adsorbent surface. The used B88-LYP GGA functional does not fully include long-range-type electrostatic effects [47][48][49][50][51], but assuming that van der Walls forces will work on all molecules, the absolute values of partial molecular charges allow comparison of their electron distributions and capture general trends. Considering the interaction of different molecules and anions with the same adsorbent surface, electrostatic potential energy maps for adsorbates were prepared ( Figure 1). These visualizations allow one to see different electronic density distribution of adsorbates and its anions. To gain deeper insight into divergent adsorption of the studied compounds, energy values of HOMO and LUMO orbitals were used to calculate the HOMO-LUMO energy gap, ∆E HL , and global indexes like chemical potential, µ, chemical hardness (η), and electrophilicity index, ω [52,53]: where E H and E L are the HOMO and LUMO orbital energies, respectively.   Theoretical values of electronic descriptors of all adsorbates are given in Table 5. For pH equal to 1, one should compare electronic properties of undissociated molecules, because they occur mainly in strongly acidic solutions ( Table 4). The lower value of energy gap had ketoprofen (2.96 eV), which indicates its greatest reactivity. Ketoprofen was also the most electrophilic with µ equal to 4.06 eV and 4.10 eV and ω equal to 5.57 eV and 5.68 eV for isomer S and R, respectively. However, these high values of electrophilicity could cause a decrease in its reactivity, because the adsorbent surface at pH = 1 is positively charged. Higher values of the chemical hardness for undissociated molecules than for its anions show that electron densities of undissociated adsorbates were harder to modify. Nevertheless, the values of η for undissociated molecule are not very high, and that is why these adsorbates can react by electrostatic interactions. As the pH increases, the number of dissociated molecules increases as well, which is shown in Table 4. At pH equals 4 and 6, a large amount of anions means that there are more and more electron donors and, at the same time, there is still a positive charge on the surface of the adsorbent. Anions of all the studied molecules have low hardness values, which means that they are "soft" molecules and their electron densities can be easily changed. This causes that the predicted adsorption mechanism is based on electron sharing between acid surface and adsorbate anions.
In an alkaline environment, there is a negative charge on the adsorbent surface. That is why strongly electrophilic molecules interact with this surface. The highest values are for (R)-and (S)-ketoprofen, next (S)-naproxen. Diclofenac appears to be the least adsorbed in alkaline pH. Electrostatic potential energy maps (Figure 1) show that there is much more positive charge on adsorbate surface, which allows the molecules to interact with the surface in various configurations.

Characterization of Adsorbents
The morphologies of H and 30C/H samples were observed by means of SEM and EDX, as shown in Figure 2. It can be seen that sample H consists of tubular, blocky, and platy particles (Figure 2a). Tube lengths were up to a few hundred nanometers and were arranged in various directions. In the case of the 30C/H sample, one can see numerous carbon particles with irregular shapes deposited on the halloysite surface (Figure 2b). Using EDS analysis (Figure 2c,d), the presence of carbon in the synthesized composite 30C/H was confirmed. Carbon content in the 30C/H sample was equal to 6 wt%, and no carbon was found in the H sample.

Characterization of Adsorbents
The morphologies of H and 30C/H samples were observed by means of SEM and EDX, as shown in Figure 2. It can be seen that sample H consists of tubular, blocky, and platy particles (Figure 2a). Tube lengths were up to a few hundred nanometers and were arranged in various directions. In the case of the 30C/H sample, one can see numerous carbon particles with irregular shapes deposited on the halloysite surface (Figure 2b). Using EDS analysis (Figure 2c,d), the presence of carbon in the synthesized composite 30C/H was confirmed. Carbon content in the 30C/H sample was equal to 6 wt%, and no carbon was found in the H sample.   [35], which is the rest of the iron not totally removed during the preparation. Diffraction patterns of H, 5C/H, and 30C/H samples are compared in Figure 3 (the diffractograms of 10C/H and 20C/H are very similar to diffraction patterns of 5C/H and 30C/H samples). The main reflections of the halloysite, kaolinite, and hematite are identified for the H sample. XRD patterns of 5C/H and 30C/H samples confirm that amorphous carbon covered the halloysite surface. These diffractograms are very similar to diffraction patterns of amorphous carbons produced using glucose, lactose, and saccharose, as mentioned by Myronyuk et al. [21] with the main reflection at 2θ = 22 • and very weak at 43 • . The reflection at 25 • and 48 • can be assigned to the crystalline structure of formed activated carbon. This connected to alignment of disordered graphitic carbon layers, which form the crystalline turbostatic structure [54]. The XRD line at 2θ = 36 • for 5C/H corresponds to hematite [35], which is the rest of the iron not totally removed during the preparation.   [35], which is the rest of the iron not totally removed during the preparation.   Experimental adsorption-desorption N2 isotherms of the H sample and halloysite/carbon composites ( Figure 5) are type IV according to the IUPAC classification [57], which indicates a mesoporous character of these materials. Structural parameters calculated from adsorption isotherms are presented in Table 6. BET specific surface area (SBET) was higher for the H sample (73 m 2 /g) in comparison to halloysite/carbon composites. SBET for all composites was comparable and equals 53 m 2 /g for the 5C/H sample up to 55 m 2 /g for the 30C/H sample. Carbon deposition on the halloysite surface did not significantly affect the total pore volume. The volume of micropores increased for 20C/H and 30C/H samples versus the H sample. Experimental adsorption-desorption N 2 isotherms of the H sample and halloysite/carbon composites ( Figure 5) are type IV according to the IUPAC classification [57], which indicates a mesoporous character of these materials. Structural parameters calculated from adsorption isotherms are presented in Table 6. BET specific surface area (S BET ) was higher for the H sample (73 m 2 /g) in comparison to halloysite/carbon composites. S BET for all composites was comparable and equals 53 m 2 /g for the 5C/H sample up to 55 m 2 /g for the 30C/H sample. Carbon deposition on the halloysite surface did not significantly affect the total pore volume. The volume of micropores increased for 20C/H and 30C/H samples versus the H sample.   SBET-specific surface area; V-single-point total pore volume calculated at p/po = 0.99; Vmemesopore volume calculated by subtracting Vmi from V; Vmi-volume of micropores obtained by αsmethod; Mesoporosity-the percentage of the mesopore volume in relation to the total pore volume.
In order to investigate the influence of functional groups present on the carbon surface on the adsorption of NSAIDs, Boehm analysis was carried out for the 30C/H sample ( Table 7). The amount of acid groups was considerably higher than the basic groups on the surface of the 30C/H sample. Carboxyl, carbonyl, phenolic, and lactone groups were identified on the adsorbent surface. The point of zero change (pHpzc) of 30C/H equaled 6.56. In order to investigate the influence of functional groups present on the carbon surface on the adsorption of NSAIDs, Boehm analysis was carried out for the 30C/H sample ( Table 7). The amount of acid groups was considerably higher than the basic groups on the surface of the 30C/H sample. Carboxyl, carbonyl, phenolic, and lactone groups were identified on the adsorbent surface. The point of zero change (pH pzc ) of 30C/H equaled 6.56.

Adsorption Study
Removal efficiencies of diclofenac, ketoprofen, and naproxen from aqueous solution on adsorbents H, 5C/H, 10C/H, 20C/H, and 30C/H are presented in Figure 6. The results show that halloysite/carbon composites adsorb diclofenac, ketoprofen, and naproxen better than halloysite (H). Removal efficiency increased with the increase of saccharose concentration for all studied NSAIDs. The best results were obtained for adsorption on 30C/H adsorbent for all studied adsorbates. The values of removal efficiency for 30C/H adsorbent increased in the following order: Diclofenac (72%) < naproxen (85%) < ketoprofen (90%).
halloysite/carbon composites adsorb diclofenac, ketoprofen, and naproxen better than halloysite (H). Removal efficiency increased with the increase of saccharose concentration for all studied NSAIDs. The best results were obtained for adsorption on 30C/H adsorbent for all studied adsorbates. The values of removal efficiency for 30C/H adsorbent increased in the following order: Diclofenac (72%) < naproxen (85%) < ketoprofen (90%).

Effect of Adsorbent Dose and pH on NSAID Adsorption
The effect of adsorbent dose (H and 30C/H composite) on the adsorption of NSAIDs studied was investigated. Solutions of ketoprofen, naproxen, and diclofenac at a concentration of 50 mg/dm 3 and an adsorbent dose from 0.1 to 1.0 g were used. Results are presented in Figure 7. Removal efficiency of all used adsorbates was higher for the adsorbent 30C/H compared to the H adsorbent, and increased together with the rise of the adsorbent dose, reaching a maximum value for 1.0 g of the 30C/H adsorbent. Removal efficiency of diclofenac, ketoprofen, and naproxen for halloysite (H) and halloysite/carbon adsorbents (concentration of adsorbate solutions-50 mg/dm 3 ; mass of adsorbent-0.5 g, temp. 25 • C).

Effect of Adsorbent Dose and pH on NSAID Adsorption
The effect of adsorbent dose (H and 30C/H composite) on the adsorption of NSAIDs studied was investigated. Solutions of ketoprofen, naproxen, and diclofenac at a concentration of 50 mg/dm 3 and an adsorbent dose from 0.1 to 1.0 g were used. Results are presented in Figure 7. Removal efficiency of all used adsorbates was higher for the adsorbent 30C/H compared to the H adsorbent, and increased together with the rise of the adsorbent dose, reaching a maximum value for 1.0 g of the 30C/H adsorbent. The adsorption of NSAIDs from an aqueous solution on carbonaceous adsorbents was mainly due to the interactions between the functional groups in the NSAID molecule and the functional groups on the adsorbent surface [58]. Ketoprofen, diclofenac, and naproxen, weak organic acids, coexist in both ionized and non-ionized forms in aqueous solution. The distribution of these forms is dependent on the pH of solution, and their interaction with adsorbent surface may or may not favor The adsorption of NSAIDs from an aqueous solution on carbonaceous adsorbents was mainly due to the interactions between the functional groups in the NSAID molecule and the functional groups on the adsorbent surface [58]. Ketoprofen, diclofenac, and naproxen, weak organic acids, coexist in both ionized and non-ionized forms in aqueous solution. The distribution of these forms is dependent on the pH of solution, and their interaction with adsorbent surface may or may not favor the adsorption process. The concentration of these conjugates depends on the solution pH and pK a . Based on pK a values of adsorbates (see Table 4), and using the Henderson-Hasselbalch equation, the amount of acidic and basic forms of adsorbate molecules was calculated. In distilled water (pH = 6), the ionization degree was 34.22% for ketoprofen and 42.91% for naproxen. Diclofenac exists only as anion, because we used its sodium salt in adsorption experiments. The value of pH PZC represents the pH of the solution, at which the net surface charge was neutral. At a solution pH lower than the pH PZC , total surface charge will be, on average, positive, whereas at a higher solution pH, they will be negative because of the deprotonation of functional groups [58]. pH of adsorbate solution had a significant influence on NSAID adsorption on halloysite/carbon composites, because it was one of the key factors that controlled the adsorption process of organic weak electrolytes on carbon materials, especially in the case of the electrostatic interactions between the adsorbent and the adsorbate. The effect of solution pH on removal efficiency (R%) of ketoprofen, diclofenac, and naproxen on 30C/H adsorbent is shown in Figure 8. Removal efficiency of ketoprofen practically did not change in the range of pH 1-9, only for a pH of 13 was it slightly decreased. For naproxen, the value of R was the highest in the ranges pf pH 1-4 and 9-14, and clearly lower for pH 6. The effect of pH on diclofenac adsorption was definitely larger. The highest values of R were obtained for the range of pH 4-6, whereas the lowest for pH equal to 1, 9, and 13. In the range of pH 4-13, naproxen and ketoprofen were dissociated from 20% to about 43%, and the neutral, as well as dissociation form molecules, existed in solution. Both forms may interact with a positive-or negative-charged adsorption surface, so the effect of pH on their removal efficiency was lower than for diclofenac. At basic pH, the uptake for diclofenac was lower, because of electrostatic repulsions between the negative surface charge and the diclofenac anions and between diclofenac anions in solution. At a solution pH between 4 and 6 (lower than the pH PZC for 30C/H adsorbent), the total surface charge was, on average, positive, and the electrostatic interaction was stronger between diclofenac anions and adsorbent surface, resulting in higher uptake. Adsorption mechanism concerning ketoprofen, naproxen, and diclofenac from dilute aqueous solutions on halloysite/carbon material includes electrostatic and non-electrostatic interactions. Electrostatic interactions appeared when adsorbate molecule electrolytes were ionized in the experimental conditions used. Non-electrostatic interactions were due to dispersion and hydrophobic interactions between non-dissociated adsorbate molecules and the adsorbent surface containing mainly oxygen acidic groups.
Electrostatic interactions do not completely explain the mechanism of adsorption of NSAIDs on the 30C/H adsorbent. One of the features of the adsorption mechanism is that aromatic compounds on carbon materials may include the π-π dispersion interaction, the possibility of H-bond formation with surface oxygen groups, such as carboxyl or carbonyl groups or donor-acceptor electron complex formation [59]. These interactions between NSAID molecules or their anions and adsorbent surfaces are also possible, and can further complicate adsorption mechanism.
The following conditions were selected as optimal conditions for NSAID adsorption on studied adsorbents: Reaction temperature 25 °C, pH 6, and catalyst dosage 0.5 g. The pseudo-first-order kinetic model [59], pseudo-second-order kinetic model [60], and intra- Adsorption mechanism concerning ketoprofen, naproxen, and diclofenac from dilute aqueous solutions on halloysite/carbon material includes electrostatic and non-electrostatic interactions. Electrostatic interactions appeared when adsorbate molecule electrolytes were ionized in the experimental conditions used. Non-electrostatic interactions were due to dispersion and hydrophobic interactions between non-dissociated adsorbate molecules and the adsorbent surface containing mainly oxygen acidic groups.
Electrostatic interactions do not completely explain the mechanism of adsorption of NSAIDs on the 30C/H adsorbent. One of the features of the adsorption mechanism is that aromatic compounds on carbon materials may include the π-π dispersion interaction, the possibility of H-bond formation with surface oxygen groups, such as carboxyl or carbonyl groups or donor-acceptor electron complex formation [59]. These interactions between NSAID molecules or their anions and adsorbent surfaces are also possible, and can further complicate adsorption mechanism.
The following conditions were selected as optimal conditions for NSAID adsorption on studied adsorbents: Reaction temperature 25 • C, pH 6, and catalyst dosage 0.5 g.

Kinetic Models.
The pseudo-first-order kinetic model [59], pseudo-second-order kinetic model [60], and intra-particle diffusion model [61] (Table 3) were investigated for the adsorption of ketoprofen, naproxen, and diclofenac on H and 30C/H adsorbents. Adsorption kinetics for ketoprofen, naproxen, and diclofenac on H and 30C/H adsorbents are presented in Figure 9a. Adsorption equilibrium was obtained after 80 min for all of the studied NSAIDs. The pseudo-first-order and pseudo-second-order rate constants, k 1 and k 2 , and correlation coefficients (R 2 ) are collected in Table 8. The values of correlation coefficients obtained for the pseudo-first-order kinetic and pseudo-second-order kinetic models clearly indicate that adsorption of the studied compounds on H and 30C/H adsorbents obey the pseudo-second-order kinetic model, suggesting that the electrochemical interactions play an important role in the adsorption process. The values of pseudo-second-order rate constant, k 2 , were greater for all adsorbates for 30C/H adsorbent in comparison to the H adsorbent and decreased in the following order: Ketoprofen > naproxen > diclofenac for both adsorbents.   The Weber-Morris diffusion model was used in order to investigate adsorption mechanism of ketoprofen, naproxen, and diclofenac on the H and 30C/H adsorbents. The diffusion model is presented by the equation given in Table 3. The Weber-Morris diffusion model was used in order to investigate adsorption mechanism of ketoprofen, naproxen, and diclofenac on the H and 30C/H adsorbents. The diffusion model is presented by the equation given in Table 3.
The values of k d1 , k d2 and c 1 , c 2 determined from the slopes and intercepts of the first and second linear part of graph (Figure 9b-d) are given in Table 9. Constant values k d1 decreased in the following order: Naproxen > ketoprofen > diclofenac for H and ketoprofen > naproxen > diclofenac for 30C/H adsorbents. For both adsorbents, the rate of diffusion was the smallest for diclofenac, i.e., for the adsorbate of the highest molecular weight. The dependency q t vs. t 1/2 multi-linear plot (broken line on the graph) indicates that in the adsorption process, several steps are involved. The first section on the graph corresponds to the faster step, attributed to the diffusion of adsorbate molecules to adsorbent outer surface while the second part of the graph corresponds to slower adsorption, where intra-particle diffusion is a controlling step of the whole adsorption process [62].

Adsorption Isotherms
The results show a complete lack of correlation of experimental data for following models: Temkin, Dubinin-Radushkevich, and Sips. In the interpretation of the experimental data, three different models were used: the Freundlich, the Langmuir one active center without dissociation (one-center), and the Langmuir adsorption model on multiple active centers without dissociation (multi-center) [62,63]. Fitting experimental data to the isotherm models given in Table 3 was performed using non-linear regression (Levenberg-Marquardt least squares method with the Origin Microcal software). The results of the fitting as regards experimental data to the Freundlich and Langmuir (one-center and multi-center) isotherm models are shown in Figure 10. The Freundlich and Langmuir (one-center and multi-center) equation parameters, as well as correlation coefficients R 2 for the adsorption of ketoprofen, diclofenac, and naproxen on H and 30C/H adsorbents, are collected in Tables 10-12. The highest values of correlation coefficients for ketoprofen, diclofenac, and naproxen adsorption on H and 30C/H adsorbents were obtained for the Langmuir adsorption model on multiple active centers without dissociation (multi-center) model applied to fit experimental data. Adsorption constant values were lower for adsorbent H than for adsorbent 30C/H, and decreased for both adsorbents in the following order: Ketoprofen > diclofenac > naproxen. An increase in temperature caused a decrease in the these constant values, indicating the exothermic nature of the adsorption process. The n values were fractional for all adsorbates, pointing out the adsorption mechanisms of the NSAID molecules with a different number of adsorptive centers on H and 30C/H surfaces. The shape of adsorption isotherms is similar in the case of ketoprofen and naproxen, while it is different in the case of diclofenac, suggesting differences in the mechanism of adsorption of these compounds on 30C/H adsorbent.

Conclusions
The adsorption of ketoprofen, naproxen, and diclofenac from an aqueous solution on halloysite and new halloysite/carbon nanocomposite was studied. Removal efficiency of all the studied NSAIDs for halloysite/carbon nanocomposites was significantly higher than for non-modified halloysite. Adsorption kinetics for ketoprofen, naproxen, and diclofenac on the halloysite and halloysite/carbon nanocomposites can be described with the pseudo-second-order kinetic model. The adsorption

Conclusions
The adsorption of ketoprofen, naproxen, and diclofenac from an aqueous solution on halloysite and new halloysite/carbon nanocomposite was studied. Removal efficiency of all the studied NSAIDs for halloysite/carbon nanocomposites was significantly higher than for non-modified halloysite. Adsorption kinetics for ketoprofen, naproxen, and diclofenac on the halloysite and halloysite/carbon nanocomposites can be described with the pseudo-second-order kinetic model. The adsorption process of ketoprofen, naproxen, and diclofenac on halloysite and halloysite/carbon adsorbents proceeded in compliance with the Langmuir (multi-center) adsorption model (Langmuir adsorption model on multiple active centers without dissociation).
Theoretical values of electronic descriptors showed the greatest reactivity of ketoprofen, lower for naproxen, and the lowest for diclofenac, which is in agreement with the order of changes in rate constants and adsorption constants for all studied adsorbates.
Halloysite as a cost-saving, environmentally-friendly nano-carrier combined with carbon material may be a probable suitable adsorbent of all the studied pharmaceuticals for large scale application.