Enhanced Removal of Non-Steroidal Inflammatory Drugs from Water by Quaternary Chitosan-Based Magnetic Nanosorbents

Non-steroidal anti-inflammatory drugs (NSAIDs) are among the most common pharmaceuticals used worldwide. They are widely detected in natural waters due to their persistence in wastewater treatment, and their removal is desirable in wastewater management. As a contribution to tackle this challenge, this study explores magnetic quaternary chitosan-based nanosorbents for the effective magnetically assisted removal of three NSAIDs (diclofenac, naproxen, and ketoprofen) from water. Toward this goal, silane groups were grafted onto the backbone of trimethyl chitosan through the reaction with an epoxide functionalized silane. Once silanized, the modified chitosan was employed to coat Fe3O4 nanoparticles. The prepared materials were characterized using FTIR spectroscopy and solid-state 29Si and 13C NMR spectroscopy, which confirmed the encapsulation of Fe3O4 nanoparticles with a hybrid siliceous material enriched in trimethyl chitosan. The effect of the initial NSAIDs concentration, pH, and contact time in the adsorption behavior was investigated. The kinetic data were well described by the pseudo-second-order kinetic model, indicating a chemisorption mechanism. The maximum adsorption capacities estimated from the Langmuir model were 188.5 mg/g (0.5925 mmol/g), 438.1 mg/g (1.7371 mmol/g), and 221.5 mg/g (0.8710 mmol/g) for diclofenac, naproxen, and ketoprofen, respectively. These adsorption capacities are higher than those of most reported sorbents, indicating the potential of these biosorbents to remove the selected NSAIDs using low-energy magnetically assisted separation.


Introduction
Water pollution is one of the most critical environmental problems that the world urgently needs to address, since clean water is vital for every living organism [1]. There has been an increasing interest in emerging contaminants, such as pharmaceutically active compounds [2]. In the European Union (EU), about 3000 different substances are used for human medication purposes that are highly persistent in the environment and are of toxicological concern [3]. These chemicals are considered emerging pollutants because their discharge limits in the environment still remain unregulated or are in the process of regularization [4,5]. Non-steroidal anti-inflammatory drugs (NSAIDs) are among the most frequently used pharmaceuticals and are included in the watch list of compounds in the EU related to the top 10 persistent pollutants [6]. NSAIDs encompass an extensive range of pharmaceuticals used to treat pain and inflammation in different arthritic and postoperative situations [6,7]. Among the NSAIDs, diclofenac (DCF), naproxen (NAP), and ketoprofen (KET) are recurrently detected in the environment as trace contaminants [1,8,9]. NSAIDs undergo several processes during wastewater treatment, but they are not entirely removed from water using conventional procedures [6,10]. This contributes to the occurrence of high levels of NSAIDs in aquatic environments, reaching values as high as µg/L concentration at multiple locations worldwide [6,11].

Derivatization of the Quaternary Chitosan
Trimethyl chitosan (TMC) was modified by reaction with the alkoxysilane coupling agent GPTMS [19,20,33]. Hence, TMC (1 g) was previously dried at 30 • C for 24 h, and the reaction with GPTMS (5.2 mmol, 1.24 mL) was performed using dry N,N-dimethylformamide (13 mL) as solvent. The reaction was conducted at 100 • C, under a dry nitrogen atmosphere, with stirring at 500 rpm and over 24 h. The obtained material, hereafter named TMC/GPTMS, was washed thoroughly with dry methanol and dry ethanol and finally dried at room temperature.

Synthesis of Magnetic Bio-Hybrid Sorbent
The magnetic core was synthesized by oxidative hydrolysis of Fe 2 SO 4 ·7H 2 O in alkaline conditions to obtain magnetite (Fe 3 O 4 ) nanoparticles (Supporting Information) [34]. Then the magnetic core was coated using the sol-gel method by adding a mixture of the silica precursor (TEOS) with the derivatized chitosan (TMC/GPTMS). Briefly, for the coating, an ethanolic solution (38 mL) containing Fe 3 O 4 nanoparticles (40 mg) was kept immersed in an ice bath under sonication (horn Sonics, Vibracell) for 15 min. Next, the ammonia solution (2.4 mL), TEOS (0.406 mL), and TMC/GPTMS (0.3 g) were slowly added, and the reaction was maintained for 2 h under sonication in an ice bath. After the reaction, the resulting magnetic biosorbent was collected by magnetic separation using an NdFeB magnet and washed six times with ethanol. The particles were dried at room temperature, and the quaternary chitosan magnetic biosorbent (Fe 3 O 4 @SiO 2 /TMC/GPTMS) was obtained.

Synthesis of Non-Magnetic Quaternary Chitosan Particles
Non-magnetic quaternary chitosan particles (SiO 2 /TMC/GPTMS) were also prepared by employing a similar method but in the absence of Fe 3 O 4 nanoparticles. In the synthesis, 0.3 g of TMC/GPTMS and 0.406 mL of TEOS were added to a mixture of 0.9 mL of deionized water, 8.5 mL of ethanol, and 0.15 mL of ammonia [35]. The reaction was stirred (500 rpm) at RT for 24 h. Then, the particles were washed five times with deionized water and one time with dry ethanol, separated by centrifugation (10 min at 13,300 rpm), and dried. For complementary studies using solid state NMR spectroscopy (Bruker, Billerica, MA, USA), SiO 2 particles were also prepared by using the Stöber method (Supporting Information) [36].

X-ray Powder Diffraction
The crystalline structure of the produced materials was evaluated by X-ray powder diffraction (XRD). The XRD data were collected using a PANalytical Empyrean X-ray diffractometer (Malvern, Worcestershire, UK) equipped with a Cu Kα monochromatic radiation source at 45 kV/40 mA.

Fourier Transform Infrared Spectroscopy
Fourier transform infrared (FTIR) spectra of the materials were acquired over the range of 4000-450 cm −1 using a Bruker Optics Tensor 27 spectrometer (Bruker, Billerica, MA, USA) coupled to a horizontal attenuated total reflectance (ATR) cell. The samples were placed on the ATR crystal, and 256 scans were acquired at 4 cm −1 resolution.

Transmission and Scanning Electron Microscopy
The morphology of the produced materials was analyzed by transmission electron microscopy (TEM) and scanning electron microscopy (SEM). Transmission electron microscopy was performed using a Hitachi H-9000 TEM microscope (Chiyoda, Tokyo, Japan) carried out at an accelerating voltage of 300 kV. A drop (10 µL) of a diluted suspension of the particles in ethanol was placed on a copper grid with a lacey amorphous carbon film and left to dry before TEM analysis. Scanning electron microscopy was performed using a Hitachi SU-70 instrument at 15 kV. For SEM analysis, the samples were prepared by placing an aliquot of a dilute suspension of particles in ethanol on a glass slide, which was glued to the sample holder using double-sided carbon tape, and then coated with carbon.

Elemental Analysis, Specific Surface Area, and Pore Volume
The elemental analysis of carbon, hydrogen, nitrogen, and sulfur was performed using a Leco Truspec-Micro CHNS 630-200-200 (Leco, Saint Joseph, MI, USA). The specific surface area of the particles was assessed by nitrogen adsorption/desorption measurements using a Gemini V2.0 Micromeritics instrument (Micromeritics, Norcross, GA, USA). The materials were first outgassed overnight at 80 • C/1.5 bar, and then adsorption and desorption isotherms were conducted at liquid nitrogen temperature (−196 • C). The specific surface area was determined using the Brunauer-Emmett-Teller (BET) equation for relative pressures (P/P 0 ) up to 0.3. The pore volume was evaluated from the adsorption amount using the Barret-Joyner-Halenda (BJH).

Zeta Potential
The surface charge of the colloidal nanoparticles was assessed by zeta potential measurements in a Zetasizer Nano ZS instrument (Malvern Instruments) (Malvern, Worcestershire, UK) that uses a HeNe laser operating at 633 nm and a scattering detector at 173 • . The measurements were performed in aqueous suspensions of the particles using a disposable folded capillary cell.

Nuclear Magnetic Resonance Spectroscopy
The 13 C cross-polarization (CP)/magic-angle spinning (MAS) nuclear magnetic resonance (NMR) and 29 Si MAS/CP MAS NMR spectra were recorded on a Bruker Avance III 400 MHz (9.4 T) spectrometer(Bruker, Billerica, MA, USA) at 79.49 and 100.61 MHz, respectively. 13 C CP/MAS NMR spectra were acquired with 3.65 µs 1 H 90 • pulses, 1.5 ms contact time, a recycle delay of 5 s, and at a spinning rate of 9 kHz. 29 Si MAS NMR spectra were recorded with 4.5 µs 1 H 90 • pulses, a recycle delay of 60 s, and at a spinning rate of 5 kHz. The chemical shifts are quoted in ppm relative to tetramethylsilane (TMS).

Batch Adsorption Experiments
Individual aqueous solutions of diclofenac sodium (DCF), naproxen sodium (NAP), and ketoprofen (KET), with the required concentrations, were prepared using ultra-pure water. The DCF, NAP, and KET concentrations were determined spectrophotometrically using a Jasco U 560 UV-VIS spectrophotometer, measuring the maximum absorbance at 276, 230, and 260 nm, respectively. The calibration curves were obtained from the analysis of the spectra obtained at distinct NSAID concentrations: DCF (0.12-12.0 mg/L), NAP (0.12-3.0 mg/L), and KET (0.12-12.0 mg/L) ( Figure S1, Supporting Information). The structures and characteristics of NSAIDs are shown in Table S1 (Supporting Information). The experiments were performed in glass vials, in which an exact amount of the magnetic biosorbent (0.5 mg/mL dosage) was placed in contact with solutions with known DCF, NAP, and KET concentration. The NSAID solutions containing the magnetic biosorbent were continuously stirred in an orbital shaker incubator (IKA, Staufen, Germany, KS 4000i control) at 200 rpm under controlled temperature (25.0 ± 1.0 • C) over a specific time (0 min to 300 min) in dark conditions. After adsorption for a pre-determined contact time, the biosorbent was magnetically separated using a NdFeB magnet, and the concentration of the corresponding pharmaceutical (C t ) in the supernatant was assessed by UV-VIS spectroscopy (GBC Scientific Equipment, Hampshire, IL, USA). The percentage of removal of each NSAIDs was calculated using Equation (1), where C 0 is the initial concentration (mg/L).
Additionally, control experiments (pharmaceutical solution without the sorbent) were carried out simultaneously with adsorption experiments.
2.6.1. Effect of pH on the Removal of NSAIDs DCF, NAP, and KET solutions (50 mg/L), with distinct pH values (5, 6, 7, 8, and 9), were prepared by adjusting the pH with an appropriate amount of NaOH (0.01 M) or HCl (0.01 M). The adsorption experiments were then conducted using 0.5 mg/mL of sorbent in each case. The mixtures were shaken continuously at 30 rpm for 5 h to ensure that the adsorption equilibrium was achieved. Then, the magnetic biosorbent was separated magnetically from the solution, and DCF, NAP, and KET concentrations in the supernatant were assessed by UV-VIS spectrophotometry.

Kinetics and Equilibrium Adsorption Studies
The adsorption kinetics of single DCF, NAP, and KET solutions with different concentrations (10, 50, and 100 mg/L) were prepared, and the pH was adjusted to 5. An appropriate amount of the magnetic biosorbent was added to the mixture (0.5 mg/mL), and the adsorption process was conducted. The mixtures were shaken, and aliquots of 1 mL were collected along the time (during 5 h, 300 min) at room temperature (25.0 ± 1.0 • C). The amount of pharmaceutical adsorbed at each time t (q t , mg/g) was determined by Equation (2): where C 0 (mg/L) is the initial concentration (mg/L), C t (mg/L) is the pharmaceutical concentration in the aqueous phase at time t (min), V (L) is the volume of solution, and m (mg) is the mass of the magnetic biosorbent. Equilibrium adsorption studies were performed by dispersing 0.5 mg/mL of magnetic biosorbent, in DCF, NAP, and KET aqueous solutions, with distinct initial concentrations (ranging from 5 to 350 mg/L) at pH = 5. The experiments were conducted for 5 h (300 min) at room temperature (25.0 ± 1.0 • C). The amount of DCF, NAP, and KET adsorbed at equilibrium (q e , mg/g) was assessed by UV-VIS spectroscopy and calculated using Equation (2) for C t = C e , where C e (mg/L) is the concentration of DCF, NAP, and KET at equilibrium.

XRD Analysis
The powder X-ray diffraction (XRD) pattern of the particles used as cores (Figure 1a), along with the labelling of the respective Miller indices, indicated the presence of magnetite (Fe 3 O 4 ) with cubic inverse spinel structure (JCPDS file No. 19-0629) [37]. A similar XRD pattern was observed for the Fe 3 O 4 @SiO 2 /TMC/GPTMS particles, confirming that after surface modification, the composition of the core of the particles was maintained as Fe 3 O 4 . The XRD of the Fe 3 O 4 @SiO 2 /TMC/GPTMS particles also showed a broad peak around 23 • that is characteristic of short-range order in amorphous SiO 2 , in agreement with the presence of the siliceous network coating the magnetic cores [38]. These Fe 3 O 4 particles are ferrimagnetic and exhibit a small magnetization hysteresis loop at room temperature with a saturation magnetization of 83 emu/g [19].
The powder X-ray diffraction (XRD) pattern of the particles used as cores (Figure 1a), along with the labelling of the respective Miller indices, indicated the presence of magnetite (Fe3O4) with cubic inverse spinel structure (JCPDS file No. 19-0629) [37]. A similar XRD pattern was observed for the Fe3O4@SiO2/TMC/GPTMS particles, confirming that after surface modification, the composition of the core of the particles was maintained as Fe3O4. The XRD of the Fe3O4@SiO2/TMC/GPTMS particles also showed a broad peak around 23° that is characteristic of short-range order in amorphous SiO2, in agreement with the presence of the siliceous network coating the magnetic cores [38]. These Fe3O4 particles are ferrimagnetic and exhibit a small magnetization hysteresis loop at room temperature with a saturation magnetization of 83 emu/g [19].

FTIR and TEM Results
The FTIR spectrum of the Fe3O4 particles ( Figure 1b) showed the characteristic band at 529 cm −1 assigned to the Fe-O stretching vibration [39]. This band was also visible at

FTIR and TEM Results
The FTIR spectrum of the Fe 3 O 4 particles (Figure 1b) showed the characteristic band at 529 cm −1 assigned to the Fe-O stretching vibration [39]. This band was also visible at 559 cm −1 in the spectrum of Fe 3 O 4 @SiO 2 /TMC/GPTMS particles, although it was shifted to higher wavenumbers. FTIR analysis of the coated particles was challenging because the main vibrational bands of TMC and the siliceous network may overlap. The FTIR spectrum of TMC showed bands at approximately 3316 cm −1 attributed to -OH bonds, at 2891 cm −1 ascribed to asymmetrical stretching of C-H bonds, and peaks at 1627 and 1523 cm −1 belonging to C-O stretching (amide I) and N-H bending (amide II) vibrations, respectively [40]. Moreover, the band at 1416 cm −1 was assigned to the characteristic absorption of N-CH 3 groups in TMC [41]. Additionally, the intense bands between 1000 and 1200 cm −1 were ascribed to C-O vibrations in the polysaccharide backbone [33]. In the preparation of TMC/GPTMS, both hydroxyl and amine groups of TMC can react with the epoxide groups of the alkoxysilane GPTMS to yield ether (−CH 2 OCH 2 −) and secondary amine (−CHNHCH 2 −) covalent bonds, respectively ( Figure 2). In the FTIR spectrum of GPTMS (Figure 1b), the band at 1460 cm −1 is associated with the deformation vibrational mode of the methylene groups of the propyl unit [42,43]. The band at 1252 cm −1 was due to the epoxide ring stretching vibration, and the band at 1188 cm −1 was ascribed to the C−H deformation mode of the Si−O−CH 3 [42][43][44]. The C−O stretching vibration of the epoxide ring could be observed at 912 cm −1 , and the bands at 1075 cm −1 , 815 cm −1 , and 779 cm −1 corresponded to the stretching and bending vibrations of the Si−O−C bonds of GPTMS [42,43,45]. In the TMC/GPTMS spectrum, the absence of the band of the epoxide ring could be evidence for epoxide ring opening upon reaction with TMC [46][47][48]. The amide bands of TMC were retained and appeared shifted to higher wavenumbers (1627-1648 and 1585 cm −1 for C=O and N-H bending, respectively). This may be attributed to changes in the H-bonding network in the TMC/GPTMS. Although the FTIR results are not conclusive to clearly discern the mechanism for the reaction between TMC and GPTMS, the results suggest the formation of covalent bonding through the reaction of the epoxide ring of GPTMS. The FTIR spectrum of Fe 3 O 4 @SiO 2 /TMC/GPTMS particles also displayed the vibrational bands expected for a material comprising a siliceous network. This inorganic phase gives FTIR bands at 795 cm −1 and 446 cm −1 that can be assigned to symmetric Si-O-Si stretching and O-Si-O deformation modes of amorphous silica, respectively [20,33]. Moreover, the presence of a broad band centered at 1047 cm −1 can result from overlapped vibrational contributions of the trimethyl chitosan and amorphous silica [20,33].

Specific Surface Area and Elemental Analysis
The BET specific surface area decreased from 27.4 m 2 /g in Fe3O4 to 13.6 m 2 /g in Fe3O4@SiO2/TMC/GPTMS particles (Table 1), which is consistent with an increase of the average particle size due to the formation of the outer hybrid shells. Table 1 shows the elemental microanalysis results of the prepared materials. As expected, the Fe3O4 particles revealed negligible carbon (<0.2%) and nitrogen (<0.04%) contents. In contrast, the Fe3O4@SiO2/TMC/GPTMS particles had a carbon content of 30 wt% and a relevant nitrogen amount (5.9 wt%), confirming the incorporation of the organic (TMC) component in the final particles.  The resulting particles were composed of a magnetite spheroidal core (56 ± 11 nm) coated by amorphous shells with irregular surfaces (Figure 1c,d). The thickness of the shells was around 37 ± 6 nm. The XRD and FTIR results and the TEM images confirmed the encapsulation of the Fe 3 O 4 particles. Note that the TEM image provided was representative of the sample, always showing denser nanoparticles (magnetite) surrounded by a material with lower image contrast (biopolymer/silica material), resulting in small aggregates with several magnetic cores.

Specific Surface Area and Elemental Analysis
The BET specific surface area decreased from 27.4 m 2 /g in Fe 3 O 4 to 13.6 m 2 /g in Fe 3 O 4 @SiO 2 /TMC/GPTMS particles (Table 1), which is consistent with an increase of the average particle size due to the formation of the outer hybrid shells. Table 1 shows the elemental microanalysis results of the prepared materials. As expected, the Fe 3 O 4 particles revealed negligible carbon (<0.2%) and nitrogen (<0.04%) contents. In contrast, the Fe 3 O 4 @SiO 2 /TMC/GPTMS particles had a carbon content of 30 wt% and a relevant nitrogen amount (5.9 wt%), confirming the incorporation of the organic (TMC) component in the final particles.

NMR Analysis
Solid-state NMR spectroscopy analysis of the non-magnetic quaternary chitosan particles (SiO 2 /TMC/GPTMS) was performed to inquire about the shells' chemical nature of the Fe 3 O 4 @SiO 2 /TMC/GPTMS particles. A brief characterization of SiO 2 /TMC/GPTMS particles by SEM and FTIR spectroscopy is included in Figure S2 (Supporting Information). Figure 3a shows the 13 C CP/MAS NMR spectra for TMC and SiO 2 /TMC/GPTMS particles, and the corresponding chemical-shift assignments are listed in Table S2 (Supporting Information). The 13 C NMR spectrum of TMC showed the following evidences: δ = 24.3 ppm attributed to the carbon atom of the methyl moieties of the acetyl groups; δ = 54.8 ppm, which correspond to carbon atoms of N,N-dimethylated groups; δ = 55.2 ppm attributed to carbon atoms of N,N,N-trimethylated groups; δ = 57.8 ppm, where two overlapped signals were observed and ascribed to carbon C 6 and C 2 ; δ = 71.2 ppm, representing the carbon atoms in the O-methylated groups; δ = 74.7 ppm assigned to carbons C 5 and C 3 ; δ = 81.1 ppm ascribed to the carbon C 4 ; δ = 99.7 ppm assigned to carbon C 1 ; and finally δ = 173.8 ppm that corresponds to the carbon of carbonyl of acetyl group [49,50]. The 13 C NMR spectrum of SiO 2 /TMC/GPTMS showed the resonances associated with the carbons of GPTMS that were overlapped with the NMR peaks assigned to the carbon atoms of TMC and that were observed in the range δ = 22.8-173.8 ppm. The methoxy groups (Si-O-CH 3 ) associated with GPTMS are usually identified by a 13 C resonance at δ = 50 ppm. However, this resonance was absent, suggesting that the hydrolysis of GPTMS to form silanol groups proceeded to completion [23]. The signals of the carbon atoms C a and C b of the Si-bonded propyl chain linked to GPTMS at δ = 22.8 ppm were overlapped with the signal of the carbon of the methyl moieties of the acetyl groups, which were attributed to the reaction of epoxide ring with the primary amine of TMC (-NH 2 ) to form a secondary amine [24]. Due to many different C-O species, the spectrum was saturated with resonances in the range of 50-120 ppm, making difficult the assignment of the resonances of C c , C d , C e , and C f in this region. Nevertheless, the literature concerning the reaction of different biopolymers with GPTMS suggests that the resonances of the carbons C c , C d , C e , and C f might be assigned in this region [51][52][53]. 29 Si NMR spectroscopy was used to provide further insight into the structure of hybrid materials. The 29 Si MAS NMR and 29 Si CP/MAS NMR spectra of the non-magnetic SiO2/TMC/GPTMS particles are shown in Figure 3b,c, respectively. For comparison, the NMR spectra of amorphous SiO 2 particles prepared using the same approach but in the absence of TMC (SEM and FTIR results were included in Figure S2, Supporting Information) were also included in Figure 3b,c. A specific region could be clearly observed from δ = −90 to −120 ppm, corresponding to the silica Q species. The Q species describes the connectivity of the silica network, and the Q n denotes a silicon atom with n bridging oxygens (Figure 3d) [54]. Therefore, a higher value of n in Q species indicates a more connected silica network. From the 29 Si MAS NMR spectrum of SiO 2 /TMC/GPTMS, it was clear that the Q 2 (δ = −91.9 ppm), Q 3 (δ = −101.2 ppm), and Q 4 (δ = −111.3 ppm) structures were the dominant Q species, in accordance with the literature (Figure 3b) [54]. The degree of condensation of the silica network could be obtained using the fraction of silanol groups ((Q 2 + Q 3 )/Q 4 ) calculated from the 29 Si MAS NMR spectra, and is given in Table S2, Supporting Information. The degree of condensation was 0.52 in SiO 2 particles and decreased to 0.48 in the SiO 2 /TMC/GPTMS particles. Since the degree of condensation decreased, it means evidence for the covalent bonding of the precursor TMC/GPTMS with the surface of the particles [35]. From the 29 Si CP/MAS NMR spectrum of SiO 2 /TMC/GPTMS (Figure 3c), two distinct regions could be clearly observed from δ = −50 to −70 ppm and from δ = −90 to −120 ppm, corresponding to the silica T and Q species, respectively. The amount of T species indicates the degree of cross-linking between GPTMS and the silica network. T n denotes a silicon atom bonded to carbon with n bridging oxygens (-Si-O-Si-) [24]. Therefore, a higher value of n in T and Q species indicates a more connected silica network. From the 29 Si CP/MAS NMR spectra (Figure 3c) it was possible to observe that the Q 3 and Q 4 structures were the dominant Q species in the SiO 2 /TMC/GPTMS particles with no Q 1 species detected, implying a highly condensed silica network [24,54]. Moreover, the 29 Si CP/MAS NMR spectrum of SiO 2 /TMC/GPTMS particles showed two resonances ascribed to the T 2 (δ = −57.6 ppm) and T 3 (δ = −64.2 ppm) sites [55]. Additionally, the absence of T 1 species indicated that the alkoxy portion of the GPTMS molecules underwent a high degree of condensation, confirming the covalent bonding of TMC/GPTMS to the siliceous network.

Influence of pH and Adsorption Mechanism
The pH of the solution is a key factor in the adsorption process and interactions at the sorbent-sorbate interface. It influences the surface chemistry of the sorbent and the speciation of the solutions. The pKa range of the three targeted NSAIDs was between 4.00 and 4.45 (pKa DCF = 4.00; pKa NAP = 4.19; pKa KET = 4.45) [7], as shown in Figure 4a-c. The selected NSAIDs are classified as weak acid compounds, and thus the pH will affect their chemical speciation in solution. At pH < pKa, the NSAIDs are mainly protonated (neutral), while at pH > pKa the NSAIDs are deprotonated (negatively charged). Measurement of zeta potential values of the Fe 3 O 4 @SiO 2 /TMC/GPTMS particles as a function of pH was also performed to provide information on the surface charge of the sorbent (Figure 4d). The point of zero charge (pH pzc ) is the pH in which the net charge on the sorbent surface is zero, and it was found to be between pH = 8 and 9, which means that at pH < pH pzc the surface was positively charged (Figure 4d) and could interact electrostatically with deprotonated NSAIDs species. At pH values not higher than 8, the particles presented positive zeta potentials due to the contribution of cationic trimethylammonium groups (−N + (CH 3 ) 3 ) from TMC present at the particles' surfaces. Adsorption of DCF, NAP, and KET showed identical pH-dependent patterns (Figure 4d). This indicates that these NSAIDs may undergo the same type of interactions with the surface of Fe 3 O 4 @SiO 2 /TMC/GPTMS particles. As shown in Figure 4d, high adsorption capacity was observed at pH = 5 for DCF, NAP, and KET, and then the subsequent adsorption experiments were performed at this pH. At this pH, the particles are positively charged, and the NSAIDs are mainly in the deprotonated form, which promotes the electrostatic interactions and increase the adsorption capacity [56]. When the pH increased from 5 to 9, the adsorption capacity (q t ) decreased for all the NSAIDs tested, most likely because the surface charge of the nanoparticles also decreased. At pH = 9 the NSAIDs and the surface of the particles were negatively charged, and repulsive forces could occur. However, the particles still adsorbed NSAIDs, indicating that electrostatic interactions are not the only mechanisms that govern the adsorption at this pH.
There are several possible mechanisms for the adsorption of NSAIDs onto the Fe 3 O 4 @SiO 2 / TMC/GPTMS particles: (i) electrostatic interactions between the cationic trimethylammonium groups from TMC and the carboxylate groups of the NSAIDs molecules [57,58]; (ii) H-bonding interactions among the NSAIDs (carboxyl, amine, and ester groups) with the hydroxyl groups of trimethyl chitosan [58,59]; and (iii) hydrophobic interactions in which the hydrophobic dimethylated groups (−N(CH 3 ) 2 ) of trimethyl chitosan can capture the NSAID molecules [59][60][61]. Previous studies have explained the adsorption of NSAIDs on other sorbents based on hydrogen bonding [59,60]. Taking into account the results in Figure 4d, at pH = 9, the adsorption capacity of each NSAID towards Fe 3 O 4 @SiO 2 /TMC/GPTMS particles decreased in the order DCF > NAP > KET. The extent of the interaction through hydrogen bonding has been well explained based on two factors: differences in the chemical structures of each NSAIDs, and the interactions between each functional group present in the NSAIDs [60,62]. As can be seen by the molecular structures of DCF, NAP, and KET (Table S1, Supporting Information), these molecules showed three bonding sites each. Hydrogen bonding increased as the accessibility of the available sites for interaction increases in each NSAID molecule. Furthermore, among the NSAIDs tested, DCF presented a higher affinity towards sorbent particles, most likely because it contains both H-donor (amine) and H-acceptor (oxygencontaining) functional groups that can establish H-bonds with Fe 3 O 4 @SiO 2 /TMC/GPTMS particles [60]. In the case of NAP, the oxygen of the methoxy group may lead to more hydrogen bonding than the keto-group in KET [60].
Coatings 2021, 11, x FOR PEER REVIEW 9 of 20 which correspond to carbon atoms of N,N-dimethylated groups; δ = 55.2 ppm attributed to carbon atoms of N,N,N-trimethylated groups; δ = 57.8 ppm, where two overlapped signals were observed and ascribed to carbon C6 and C2; δ = 71.2 ppm, representing the carbon atoms in the O-methylated groups; δ = 74.7 ppm assigned to carbons C5 and C3; δ = 81.1 ppm ascribed to the carbon C4; δ = 99.7 ppm assigned to carbon C1; and finally δ = 173.8 ppm that corresponds to the carbon of carbonyl of acetyl group [49,50]. The 13 C NMR spectrum of SiO2/TMC/GPTMS showed the resonances associated with the carbons of GPTMS that were overlapped with the NMR peaks assigned to the carbon atoms of TMC and that were observed in the range δ = 22.8-173.8 ppm. The methoxy groups (Si-O-CH3) associated with GPTMS are usually identified by a 13 C resonance at δ = 50 ppm. However, this resonance was absent, suggesting that the hydrolysis of GPTMS to form silanol groups proceeded to completion [23]. The signals of the carbon atoms Ca and Cb of the Sibonded propyl chain linked to GPTMS at δ = 22.8 ppm were overlapped with the signal of the carbon of the methyl moieties of the acetyl groups, which were attributed to the reaction of epoxide ring with the primary amine of TMC (-NH2) to form a secondary amine [24]. Due to many different C-O species, the spectrum was saturated with resonances in the range of 50-120 ppm, making difficult the assignment of the resonances of Cc, Cd, Ce, and Cf in this region. Nevertheless, the literature concerning the reaction of different biopolymers with GPTMS suggests that the resonances of the carbons Cc, Cd, Ce, and Cf might be assigned in this region [51][52][53].  29 Si MAS NMR spectra and (c) 29 Si CP/MAS NMR spectra of SiO2 and SiO2/TMC/GPTMS particles; and (d) schematic representation of SiO2/TMC/GPTMS particles with the labelling of Si sites according to NMR spectroscopy notation (acetyl groups are not represented in the TMC molecule). tions and increase the adsorption capacity [56]. When the pH increased from 5 to 9, the adsorption capacity (qt) decreased for all the NSAIDs tested, most likely because the surface charge of the nanoparticles also decreased. At pH = 9 the NSAIDs and the surface of the particles were negatively charged, and repulsive forces could occur. However, the particles still adsorbed NSAIDs, indicating that electrostatic interactions are not the only mechanisms that govern the adsorption at this pH. There are several possible mechanisms for the adsorption of NSAIDs onto the Fe3O4@SiO2/TMC/GPTMS particles: (i) electrostatic interactions between the cationic trimethylammonium groups from TMC and the carboxylate groups of the NSAIDs molecules [57,58]; (ii) H-bonding interactions among the NSAIDs (carboxyl, amine, and ester groups) with the hydroxyl groups of trimethyl chitosan [58,59]; and (iii) hydrophobic interactions in which the hydrophobic dimethylated groups (−N(CH3)2) of trimethyl chitosan can capture the NSAID molecules [59][60][61]. Previous studies have explained the adsorption of NSAIDs on other sorbents based on hydrogen bonding [59,60]. Taking into account the results in Figure 4d, at pH = 9, the adsorption capacity of each NSAID towards Fe3O4@SiO2/TMC/GPTMS particles decreased in the order DCF > NAP > KET. The extent of the interaction through hydrogen bonding has been well explained based on two factors: differences in the chemical structures of each NSAIDs, and the interactions between

Influence of Contact Time and Initial NSAIDs Concentration
The influence of contact time on the adsorption of DCF, NAP, and KET at variable initial concentration is shown in Figure 5. The adsorption process was rapid in the first half-hour due to the presence of a high number of available surface sorption sites. The number of existing available sites decreased, and the adsorption sites became saturated for longer contact times. The equilibrium was achieved in 30 min for the lowest concentration tested. To ensure that the equilibrium was reached regardless of the NSAID concentration, the adsorption time was extended up to 5 h. The adsorption capacity increased with the increase of the NSAIDs' initial concentrations.
The affinity of the NSAID compounds towards the magnetic particles was different and followed the trend DCF > NAP > KET. More specifically, the maximum experimental adsorption capacities reached were 100.3 mg/g (0.3153 mmol/g), 78.4 mg/g (0.3108 mmol/g), and 63.7 mg/g (0.2505 mmol/g) for DCF, NAP, and KET, respectively, for an initial concentration of 100 mg/L. The maximum NSAID removal ( Figure S3, Supporting Information) was ca. 80%, 60%, and 42% of DCF, NAP, and KET, respectively, for an initial concentration of 10 mg/L. These adsorption experiments were performed at pH = 5, and the surfaces of the particles were positively charged at this pH. As shown in Figure 4a-c, at pH = 5, the NSAID that had a higher fraction of negative species was DCF (90.8% DCF − , 87.4% NAP − , and 78.9% KET − ), which promoted the electrostatic interactions and increased the adsorption capacity. Furthermore, as mentioned above, more prominent H-bonding between DCF and particles surface may contribute to increasing adsorption. Blank experiments without magnetic particles were carried out in parallel, and no relevant losses of DCF, NAP, or KET (<2%) were detected along the time ( Figure S4, Supporting Information). This is evidence that the decrease of the NSAIDs concentrations in the presence of the trimethyl chitosan magnetic sorbent was ascribed to adsorption phenomena. Moreover, the removal of the NSAIDs with bare Fe 3 O 4 particles ( Figure S5, Supporting Information) was minimal (<3%, q max = 2.7 mg/g), which confirms the relevance of the TMC as a surface modifier for the removal of NSAIDs from water. Overall, the pseudo-second-order equation provided good fits for all the NSAIDs (R 2 between 0.98 and 0.99 and low χ 2 value), supporting the chemisorption mechanism. Moreover, the values of qt obtained from the pseudo-second-order model were closer to the

Kinetic Adsorption Studies
The experimental adsorption data were analyzed using the most common kinetic models to understand the nature of the adsorption process: the pseudo-first-order [63], the pseudo-second-order [64], and the Elovich models [65]. The non-linear forms of the kinetic equations (Equations (S1)-(S3), Supporting Information) were fit to the data. The goodness of fit was determined based on the coefficient of determination (R 2 ) and the Chi-square test value (χ 2 ) (Equations (S4) and (S5), Supporting Information). The kinetic parameters and the evaluation of the goodness of fit, obtained by non-linear regression analysis, are reported in Table S2 (Supporting Information), and the kinetic fittings are shown in Figure 5.
Overall, the pseudo-second-order equation provided good fits for all the NSAIDs (R 2 between 0.98 and 0.99 and low χ 2 value), supporting the chemisorption mechanism. Moreover, the values of q t obtained from the pseudo-second-order model were closer to the experimental results than the qt values predicted from the pseudo-first-order and Elovich models. The pseudo-second-order model assumes that the adsorption rate is reactioncontrolled [66]. Thus, these results suggest that the adsorption rate can be governed by the electrostatic interactions between the cationic groups of trimethyl chitosan with the anionic DCF, NAP, and KET molecules [64]. Similar observations have been reported regarding the kinetic profile of NSAID adsorption for other materials [67][68][69].

Equilibrium Adsorption Studies
The equilibrium adsorption capacity of DCF, NAP, and KET (q e , mg/g), as a function of the liquid-phase equilibrium concentration of DCF, NAP, and KET (C e , mg/L), is depicted in Figure 6. In accordance with Giles et al. (1960), isotherms can be divided into four main groups in terms of shape: L, S, H, and C [70]. The isotherm data obtained for all NSAIDs were the L-type (Langmuir), which is characterized by an initial concave region relative to the concentration axis ( Figure 6). The equilibrium data were analyzed using the isotherm models of Langmuir [71] and Freundlich [72] (Equations (S6) and (S7), Supporting Information, respectively), which are two-parameter isotherms, and the Sips model [73] (Equation (S8), Supporting Information), which is a three-parameter isotherm. The fitted model parameters are depicted in Table S3 (Supporting Information). Based on the values of R 2 (0.9820-0.9940) and χ 2 , it can be concluded that overall, the Sips isotherm provided a good fit of the experimental equilibrium data for DCF, NAP, and KET.
The Langmuir isotherm is one of the most used adsorption isotherm models, and it supposes monolayer coverage of NSAIDs over a homogeneous sorbent surface. The Langmuir monolayer maximum adsorption capacity calculated was 188.5 mg/g (0.5925 mmol/g), 438.1 mg/g (1.7371 mmol/g), and 221.5 mg/g (0.8710 mmol/g) for DCF, NAP, and KET, respectively. The Freundlich isotherm is an empirical model, which assumes the existence of heterogeneous adsorption sites on the surface of the sorbent. Moreover, the Freundlich constant K F is related to the affinity of the NSAIDs to the Fe 3 O 4 @SiO 2 /TMC/GPTMS particles. Thus, a high K F value for DCF (K F = 12.5) suggests higher affinity when compared to NAP (K F = 5.3) and KET (K F = 4.8). The Sips (or Langmuir-Freundlich) isotherm is a combination of Langmuir and Freundlich isotherms. At low sorbate concentration, the Sips equation reduces to a Freundlich isotherm, while at high sorbate concentrations, it predicts the sorption capacity of a monolayer, characteristic of the Langmuir isotherm [67]. The Sips isotherm equation is characterized by the dimensionless heterogeneity factor (β S ), which varies from 1, in a homogeneous surface, to β S < 1, in a heterogeneous surface [74]. For Fe 3 O 4 @SiO 2 /TMC/GPTMS particles, the results showed that β S was smaller than 1 for DCF, NAP, and KET, which indicates heterogeneous surfaces in these sorbents.
For further understanding of the adsorption process, Langmuir isotherm parameters can be used to predict the affinity between the NSAIDs and sorbent particles using a dimensionless constant called separation factor or equilibrium parameter (R L ), which is expressed by Equation (3) [75] as follows: where K L (L/mg) is the Langmuir constant, and C 0 (mg/L) is the initial sorbate concentration. The value of R L shows if the adsorption process is favorable or not as follows: if R L = 0, the adsorption process is irreversible, and this occurs if K L is very large, indicating very strong adsorption; if 0 < R L < 1, the adsorption is favorable, and this is the standard case; R L = 1 indicates that the adsorption isotherm is a straight line, and so it is called linear adsorption; for R L > 1, the adsorption is unfavorable, meaning that desorption happens [67,76]. Figure 6d shows the calculated R L values versus the initial concentrations of DCF, NAP, and KET. All the R L values were between 0 and 1, indicating that the adsorption of DCF, NAP, and KET over the Fe 3 O 4 @SiO 2 /TMC/GPTMS particles was favorable at the conditions studied. Additionally, lower R L values at higher initial DCF, NAP, and KET concentrations indicated that adsorption was more favorable at higher sorbate concentrations.
Coatings 2021, 11, x FOR PEER REVIEW 15 of 20 ditions studied. Additionally, lower RL values at higher initial DCF, NAP, and KET concentrations indicated that adsorption was more favorable at higher sorbate concentrations.

Comparison with Other Adsorbents
The efficiency of the proposed particles for the removal of DCF, NAP, and KET was compared with other sorbents reported in the literature. The maximum adsorption capacity of DCF, NAP, and KET was 188.5 mg/g (0.5925 mmol/g), 438.1 mg/g (1.7371 mmol/g), and 221.5 mg/g (0.8710 mmol/g), respectively. As can be seen in Table 2, the particles here developed are very effective in the removal of the selected NSAIDs and present maximum adsorption capacity higher than most of the reported sorbents. In addition, these particles offer the advantage of fast separation and recovery from water using low-energy magnetic separation. Hence, these results demonstrate the potential of the proposed particles for removing NSAIDs from water.

Comparison with Other Adsorbents
The efficiency of the proposed particles for the removal of DCF, NAP, and KET was compared with other sorbents reported in the literature. The maximum adsorption capacity of DCF, NAP, and KET was 188.5 mg/g (0.5925 mmol/g), 438.1 mg/g (1.7371 mmol/g), and 221.5 mg/g (0.8710 mmol/g), respectively. As can be seen in Table 2, the particles here developed are very effective in the removal of the selected NSAIDs and present maximum adsorption capacity higher than most of the reported sorbents. In addition, these particles offer the advantage of fast separation and recovery from water using low-energy magnetic separation. Hence, these results demonstrate the potential of the proposed particles for removing NSAIDs from water.

Conclusions
We have reported here studies on batch mode for the adsorption of three selected NSAIDs frequently detected in aquatic media, using new magnetic nanosorbents based on trimethyl chitosan. The resulting nanosorbents removed diclofenac, naproxen, and ketoprofen efficiently from aqueous solutions, denoting high adsorption capacity. This improved efficiency is related to the high affinity of functional groups from trimethyl chitosan, used here as a surface modifier, to NSAIDs molecules, along with reduced particle dimensions and high surface-to-volume ratios. It was observed that the affinity of the NSAID compounds towards the magnetic particles followed the trend DCF > NAP > KET. Therefore, this magnetic biosorbent based on quaternary chitosan offers new possibilities for water purification contaminated with diclofenac, naproxen, and ketoprofen by applying magnetic-assisted cleaning technologies. Future perspectives for integrating these new nanosorbents in adsorption technologies are promising due to their effectiveness as compared to other materials described in the literature.

Supplementary Materials:
The following are available online at https://www.mdpi.com/article/10 .3390/coatings11080964/s1, Figure S1: Calibration curves for diclofenac (0.12-12.0 mg/L), naproxen (0.12-3 mg/L), and ketoprofen (0.12-12.0 mg/L) in ultra-pure water, providing a linear relation between the absorbance and the concentration, Figure S2: SEM images of (a) SiO 2 /TMC/GPTMS and (c) SiO 2 particles and ATR-FTIR spectra of (b) SiO 2 /TMC/GPTMS and (d) SiO 2 particles, Figure S3: Time profile of removal percentage of NSAIDs at variable (a) DCF, (b) NAP, and (c) KET initial concentration (10, 50, and 100 mg/L), using the particles Fe 3 O 4 @SiO 2 /TMC/GPTMS, for 5 h (300 min), Figure S4: Variation of diclofenac, naproxen, and ketoprofen concentration on control experiments performed in absence of adsorbent particles to assess the loss of DCF, NAP, and KET caused by other phenomena than adsorption on sorbents, Figure S5: Time profile of removal percentage and (b) adsorption capacity of diclofenac, naproxen, and ketoprofen at 50 mg/L, using Fe 3 O 4 particles, for 5 h (300 min), Figure S6: Kinetic parameters estimated from pseudo-first and -second order and Elovich models, and evaluation of their fittings for diclofenac, naproxen, and ketoprofen, using Fe 3 O 4 @SiO 2 /TMC/GPTMS particles, Table S1: Selected characteristics and structures of NSAIDs, Table S2: 13 C CP/MAS, 29 Si MAS, and 29 Si CP/MAS NMR chemical shifts for TMC, SiO 2 /TMC/GPTMS, and SiO 2 particles and quantification of the 29 Si Q n resonances, Table S3: Equilibrium model parameters obtained from model fitting to experimental sorption data of Fe 3 O 4 @SiO 2 /TMC/GPTMS for the removal of DCF, NAP, and KET, together with the goodness of fit.