Producing Magnetic Nanocomposites from Paper Sludge for the Adsorptive Removal of Pharmaceuticals from Water—A Fractional Factorial Design

In view of a simple after-use separation, the potentiality of producing magnetic activated carbon (MAC) by intercalation of ferromagnetic metal oxide nanoparticles in the framework of a powder activated carbon (PAC) produced from primary paper sludge was explored in this work. The synthesis conditions to produce cost effective and efficient MACs for the adsorptive removal of pharmaceuticals (amoxicillin, carbamazepine, and diclofenac) from aqueous media were evaluated. For this purpose, a fractional factorial design (FFD) was applied to assess the effect of the most significant variables (Fe3+ to Fe2+ salts ratio, PAC to iron salts ratio, temperature, and pH), on the following responses concerning the resulting MACs: Specific surface area (SBET), saturation magnetization (Ms), and adsorption percentage of amoxicillin, carbamazepine, and diclofenac. The statistical analysis revealed that the PAC to iron salts mass ratio was the main factor affecting the considered responses. A quadratic linear regression model A = f(SBET, Ms) was adjusted to the FFD data, allowing to differentiate four of the eighteen MACs produced. These MACs were distinguished by being easily recovered from aqueous phase using a permanent magnet (Ms of 22–27 emu g−1), and their high SBET (741–795 m2 g−1) were responsible for individual adsorption percentages ranging between 61% and 84% using small MAC doses (35 mg L−1).


Introduction
Over the past decade, pharmaceuticals have been recognized as potential toxic environmental contaminants and, as a result, their occurrence in aquatic environments gained considerable attention worldwide. Resulting from their massive consumption, substantial amounts of pharmaceuticals in both their unchanged and metabolized forms are directly or indirectly discarded into aquatic systems. Therefore, concentration levels of ng L −1 to µg L −1 can be commonly found in surface water and groundwater [1][2][3] and some studies have even reported the occurrence of pharmaceuticals at mg L −1 levels [4,5]. Given the detected values and due to their persistence in natural waters, pharmaceuticals pose a longterm menace to aquatic organisms and can induce undesirable effects on both humans and ecosystems [1,2,6,7]. In order to address emerging concern regarding the contamination by pharmaceuticals, Directive 2013/39 EU set the necessity of studying their environmental risks and of protecting the aquatic environment and human health, simultaneously highlighting that the development of innovative cost effective water treatment technologies should be stimulated [8].
Several approaches have been investigated for the removal of pharmaceuticals from waters (e.g., biological treatments, reverse osmosis, nanofiltration, ozonation, and advanced oxidation processes) [9]. Among them, adsorption is a well-researched technique, and its recognition lies in the low initial investment needed, the operational simplicity and its versatility [1]. Carbonaceous adsorbents, in particular powdered activated carbon (PAC), are usually preferred for the adsorptive removal of pharmaceuticals from waters, due to their porous structure and their high specific surface area (S BET ) [1,10,11]. The production of PAC using industrial and agricultural wastes as precursors has been focus of special attention, in a way to increase the sustainability of these carbon materials [10]. Pulp and paper mill sludge is an example of a widely produced waste (50 kg of dry sludge per tonne of paper), which economic and sustainable management is a continuous challenge for the industry. Given the cellulosic nature of this residue, it presents the adequate properties to be used as precursor of carbonaceous adsorbents. In this sense, the application of paper mill sludge-derived adsorbents for water remediation has been explored by several authors regarding the removal of pharmaceuticals [12][13][14][15], and other organic [16,17] and inorganic [18] contaminants from waters. Along with the use of alternative precursors, aiming to generalize the application and increase the potential of carbonaceous adsorbents, a lot of research work has been devoted to their structural and chemical modification. In the last decade, the anchorage of ferromagnetic metal oxides (e.g., iron oxides) on its carbon matrix has arisen as a suitable solution to overcome difficulties in the after-use separation of PAC [19,20]. The inclusion of magnetic nanoparticles on PAC allows for the recovery of the adsorbent using an external magnetic field, so avoiding the filtration and/or centrifugation operations usually needed for PAC separation. In this way, the use of magnetic activated carbon (MAC) reduces the overall process cost and complexity [20,21].
Different synthesis routes can be used to produce MAC (e.g., co-precipitation, oxidative hydrolysis of ferrous salts, thermochemical and mechanical treatments), but co-precipitation is typically the selected method, owing to its low cost and simplicity [20,22]. The magnetically active components usually introduced in PAC framework by co-precipitation are magnetite (Fe 3 O 4 ) and maghemite (G-Fe 2 O 3 ). Anyhow, the magnetic, textural and morphologic properties imparted to MAC are highly dependent on the concentration of ferric (Fe 3+ ) and/or ferrous (Fe 2+ ) salts, the alkaline conditions (concentration of OH − ) [23], the synthesis temperature [23,24] used to produce iron nanoparticles and also the ratio between PAC and iron salts [25]. Therefore, the management of the above-mentioned experimental parameters has been shown to be crucial to achieve the desired features in terms of size and distribution of nanoparticles [24,26], magnetic properties (saturation magnetization, M s ) and S BET of the materials [27]. However, to the best of our knowledge, the combined effect of these synthesis conditions and their impact on the features and adsorptive performance of waste-based MACs have not been assessed yet.
The aim of this work was to determine the conditions to produce efficient magnetresponsive nanocomposite materials by in-situ iron oxide co-precipitation onto PAC prepared from paper mill sludge for application in the adsorption of pharmaceuticals from water. For that purpose, a fractional factorial design (FFD) was employed to evaluate the effect of production variables (Fe 3+ :Fe 2+ molar ratio, PAC:Fe salts mass ratio, reaction temperature, and pH conditions) on the properties (S BET and M s ) and adsorptive performance (pharmaceutical percentage of adsorption, A) of the resulting waste-based MACs. For the latter, pharmaceuticals from three different therapeutic classes, namely amoxicillin (AMX, antibiotic), carbamazepine (CBZ, antiepileptic), and sodium diclofenac (DCF, non-steroidal anti-inflammatory drug) were selected as adsorbate models. After selecting the most appropriate production conditions, the morphology, composition and surface structure of the selected MACs were assessed by scanning electron microscopy (SEM), X-ray fluorescence (XRF), and X-ray photoelectron spectroscopy (XPS) analyses, respectively.

Preparation of Powdered Activated Carbon
Primary sludge (PS) from pulp and paper mill industry was used as precursor to prepare PAC, according to the optimized experimental conditions determined by Jaria et al. (2019) [28]. PS was collected from a pulp and paper mill that employs the kraft elemental chlorine free production process and uses eucalyptus (Eucalyptus globulus) wood. Briefly, PS was first impregnated with KOH activating agent using a 1:1 w/w ratio (batches of 15 g of PS with 15 g of KOH in 50 mL of distilled water), under ultrasonic stirring for 1 h and left to dry at room temperature in a laboratory fume hood. The dried material was then pyrolyzed in porcelain crucibles at 800 • C in a muffle for 150 min (heating rate of 10 • C min −1 ), under N 2 atmosphere. The carbonized material was washed with 1.0 M HCl (for ashes and KOH removal) and distilled water (until neutral pH of the leachate was reached) and dried overnight at 100 • C. Finally, the material was crushed in order to obtain a fine homogenous powder [28].

Preparation of Magnetic Activated Carbon
The loading of iron oxide magnetic nanoparticles onto PAC to produce MACs was performed by co-precipitation. The magnetic nanoparticles were synthetized by alkaline co-precipitation of FeCl 3 ·6H 2 O and FeSO 4 ·7H 2 O salts, and to avoid the formation of non-magnetic forms of iron oxides, the reaction was conducted in inert atmosphere. The systematic synthesis of MACs is described according to the following procedure. Firstly, the ultrapure water used in the synthesis was degassed for 30 min with N 2 to prevent from oxidation of iron salt solutions prior to MAC synthesis. A solution containing both FeCl 3 ·6H 2 O (concentration interval between 0.022-0.073 mol L −1 ) and FeSO 4 ·7H 2 O (concentration interval between 0.037−0.087 mol L −1 ) salts was prepared (for a total volume of 50 mL) and transferred to a glass reactor (during this step no color alteration was observed, nor the formation of a precipitate), followed by the addition of the PAC prepared in Section 2.2. The mixture was then heated to a pre-defined temperature, under oxygen-free conditions (N 2 flow) and stirred at 100 rpm. These conditions were kept during the whole operation. A KOH solution (~0.5 mol L −1 ) was added dropwise (for a volume varying between 39 mL and 50 mL) to achieve the desired pH and the reaction was held for 1 h, keeping the defined temperature. After magnetic decantation, the supernatant (excess of alkali solution) was discarded, and the produced MACs were thoroughly washed with Nanomaterials 2021, 11, 287 4 of 20 distilled water until neutral pH of the washing leachate was reached. The materials were then dried at 50 • C for 48 h, mechanically grinded and finally stored in a sealed container prior to their use. A total of 18 MAC materials were produced and the experimental conditions used, i.e., molar ratio between FeCl 3 ·6H 2 O and FeSO 4 ·7H 2 O salts (Fe 3+ :Fe 2+ ), mass ratio between PAC and iron salts (PAC:Fe), reaction temperature and pH conditions, are summarized in Table 1. Table 1. Experimental conditions (Fe 3+ :Fe 2+ molar ratio, PAC:Fe mass ratio, reaction temperature and pH conditions), their codified variables and factor levels of the fractional factorial design (FFD) applied to optimize the production of the 18 magnetic activated carbons (MACs).

Process Variables and Experimental Fractional Factorial Design
An experimental fractional factorial design (FFD) (relationship between input factors (variables) and output effects (responses) in a process) was used to determine the optimum set of operational variables to produce MAC materials, since it allows to perform a systematic optimization of the process with a reduced number of experiments and depletion of resources.

Responses
The magnitude and the direction of the factor effects on the response modelling were evaluated, using the following responses: S BET , M s and percentage of adsorption of AMX, CBZ and DCF from aqueous solutions (A).

(i) Specific surface area
The S BET of MACs was determined by N 2 adsorption isotherms at −196 • C using a Micromeritics Instrument, Gemini VII 2380, after outgassing the materials overnight at 120 • C. The S BET was calculated by means of Brunauer-Emmett-Teller equation [34] in the relative pressure range between 0.01 and 0.1. The N 2 adsorption isotherms obtained for PAC and selected MACs from the FFD (MAC 4, MAC 7, MAC 11, and MAC 17) are presented in Figure S1, as Supplementary Materials. Additionally, the total micropore volume (V mic ), the total pore volume (V p ) and average pore width (D) were also determined. V mic was determined by the Dubinin-Radushkevich equation [35], as for V p , this parameter was estimated from the amount of N 2 adsorbed at a relative pressure of 0.99 and D was determined by using the following equation [12]: The pore size distribution was determined by Non-Local Density Functional Theory (NLDFT) analysis assuming slit pores.

(ii) Saturation magnetization
The magnetization measurements were conducted in a vibrating sample magnetometer (VSM EV9), with an applied magnetic field to a maximum of 22 kOe. By plotting the magnetic moment as a function of the applied magnetic field, it is possible to determine the M s of each MAC, dividing the plateau found for the magnetic moment by the MAC mass (10 mg). Prior to the analysis, the instrument was calibrated with a disk of pure nickel and applying a magnetic field of c.a.1 Oe and with dispersion on the magnetic moment inferior to 0.5%.

(iii) Adsorption percentage
Individual solutions of AMX, CBZ and DCF with an initial concentration (C i ) of 5 mg L −1 were prepared by dissolving a specific amount of each pharmaceutical in ultrapure water. Batch adsorption experiments were conducted in 50 mL polypropylene falcon tubes, into which were put into contact 40 mL of pharmaceutical solution (AMX, CBZ, or DCF, with no pH adjustment) with 1.4 mg of the corresponding MAC (dose of 35 mg L −1 ). Then, falcon tubes were shaken for 4 h (preliminary experiments with different MACs showed that such a contact time guarantees a situation of equilibrium) in an overhead shaker (Heidolph, Schwabach-Germany, Reax 2) at 80 rpm under controlled room temperature (25.0 ± 0.1 • C). After shaking during the defined time, the solutions were collected and filtered through 0.22 µm PVDF filters (Whatman) for the analytic determination of the remaining pharmaceutical concentration (C f ). The analytic quantification of AMX, CBZ, and DCF was performed by micellar electrokinetic chromatography (MEKC), according to the procedure described in Section 2.5. Blank controls containing each pharmaceutical (C i of 5 mg L −1 and without any MAC) were shaken during the same time as the adsorption experiments and used as reference for the calculation of adsorption percentages (A) for each pharmaceutical, using Equation (2): where C f (mg L −1 ) is the remaining pharmaceutical concentration in the liquid phase of adsorption experiments at the end of shaking and C 0 (mg L −1 ) is the concentration of pharmaceutical in the corresponding control experiments.

Analytic Quantification of Pharmaceuticals
The quantification of the pharmaceuticals in the aqueous solutions was carried out by MEKC, using a Beckman P/ACE MDQ (Fullerton, CA, USA) instrument, equipped with a UV-visible detection system. A silica capillary was dynamically coated according to the procedure described by Calisto et al. (2011) [36]. The electrophoretic separation was performed by direct polarity mode at 25 kV and 25 • C, during 2.5 min for AMX and DCF and during 3.0 min for CBZ. All samples and standard solutions were spiked with the internal standard ethylvanillin (final concentration of 3.34 mg L −1 ). The detection was monitored at 200 nm for DCF and AMX and 214 nm for CBZ. The separation buffer used was composed by 15 mmol L −1 of sodium tetraborate and 30 mmol L −1 of SDS for CBZ and AMX, and 15 mmol L −1 of sodium tetraborate and 50 mmol L −1 of SDS for DCF; the separation buffer was renewed every six runs. After each run, the capillary was washed with ultrapure water (60 s) and then with the separation buffer (90 s), at 20 psi. The determination of the calibration curve was carried out for each pharmaceutical using seven standard solutions with concentrations ranging between 0.500 mg L −1 and 5.00 mg L −1 . All the analyses were performed in triplicate. The detection limit of the method (3σ), within the concentration range used, was 0.307 mg L −1 , 0.263 mg L −1 and 0.289 mg L −1 for AMX, CBZ, and DCF, respectively.

Data Treatment
Analysis of variance (ANOVA) of the FFD results was applied to evaluate the significance of each experimental variable (χ 1 , χ 2 , χ 3 , and χ 4 ) tested in the selected responses (S BET , M s , A AMX , A CBZ , and A DCF ). The p-values were used as a tool to check the significance of each factor on the obtained responses, with a confidence level of 95% (p-values ≤ 0.05 indicated significant factors, while p-values > 0.05 indicated non-significant factors).
Furthermore, principal component analysis (PCA) and hierarchical cluster analysis (HCA) using centroid linkage method and Euclidean metric, were applied to the same data to identify patterns among the different MACs. Additionally, and using the loadings of the PC1 corresponding to the columns of the adsorption of AMX, CBZ, and DCF, an overall adsorption removal was defined, A pooled (%), which was calculated by means of Equation (3): Also, a linear quadratic model of A pooled versus S BET and M s was fitted and used to assess the optimal MAC regarding adsorption performance.
Matlab software R2019a (The MathWorks, Co., Natick, MA, USA) was used for all calculations and graphs.

Morphological and Chemical Characterization of Selected Materials
Materials selected from the previous FFD (based on the data analysis of Section 3.3) were further characterized, namely by the determination of the morphological and chemical composition. The morphological features of the selected MACs were evaluated by scanning Nanomaterials 2021, 11, 287 7 of 20 electron microscopy (SEM) using a S4100 Hitachi (Tokyo, Japan) equipment and an electron acceleration voltage of 20 kV. Prior to the analysis, the samples were coated with a thin film of amorphous carbon. The following magnifications were used: 500×, 3000×, and 10,000×. The chemical composition on the MACs surface was determined by X-ray diffraction (XRD), X-ray fluorescence (XRF), and X-ray photoelectron spectroscopy (XPS). XRD measurements were performed at room temperature with a PANalytical Empyrean powder diffractometer using monochromated CuKα radiation (λ = 1.541 • A) in the 80 • 2θ range at 0.02 • resolution, and 4000 acquisition points per step. The XRF analysis was performed using a Malvern Panalytical Axios spectrometer, under inert atmosphere (He) and applying a maximum voltage of 36.50 kV and 60 kV and a maximum current of 60 mA, 72 mA and 100 mA. The XPS spectra of the selected MACs were acquired in an Ultra High Vacuum (UHV) system with a base pressure of 2 × 10 -10 mbar. The system is equipped with a hemispherical electron energy analyzer (SPECS Phoibos 150, Berlin, Germany), a delay-line detector and a monochromatic AlKα (1486.74 eV) X-ray source. High-resolution spectra were recorded at a normal emission take-off angle and with a pass-energy of 20 eV, which provides an overall instrumental peak broadening of 0.5 eV.

Fractional Factorial Design
The responses assessed in the FFD analysis, namely S BET , M s and A for each of the considered pharmaceuticals, are shown in Figure 1 for the eighteen produced MACs.
Regarding the structural properties of the eighteen produced MACs, the S BET ranged between 475 m 2 g −1 and 899 m 2 g −1 , being these values lower than that obtained for the bare PAC (1438 m 2 g −1 ). Still, S BET of the here produced MACs are similar or even higher than most of values determined for waste-based magnetic carbons in the literature [22,37]. The lower S BET of MACs when compared with PAC can be explained by the decrease and/or blockage of pores on PAC framework by the iron oxide nanoparticles produced by co-precipitation [27,38,39], and also by the very small S BET values typically found in the literature for Fe 3 O 4 (19 m 2 g −1 [39]) and Fe 2 O 3 (64 m 2 g −1 [27]).
Furthermore, and according with the data presented in Table S2 of Supplementary Materials, the incorporation of magnetic iron nanoparticles onto the carbonaceous matrix was followed by a decrease on the total pore volume (V p ) and micropore volume (V mic ). The trend observed for the V p and V mic values was the same as that of S BET ; the lowest V p and V mic were obtained for the MACs that had the lowest S BET , conversely the materials presenting the highest S BET are characterized by having a porous structure. Also, it is possible to infer from these results the formation of well-developed pore structures in the eighteen produced MACs, with the micropore volume accounting for ca. 37% to 48% of the total pore volume. Regarding the pore size distribution, despite some differences that could be found (particularly for MACs with distinct S BET values, as shown in Figure S2), typically all MACs exhibited identical profiles, with pores sizes ranging between 2 and 40 nm (mesopores), and with a higher incidence of pores between 2 nm and 15 nm. Additionally, macropores of ca. 63 nm were also detected in all MAC's.
As for the magnetic properties of the produced MACs, the M s values found for the overall MACs ranged between 2.0 emu g −1 and 44.2 emu g −1 , the last one being closer to the values of 60 emu g −1 reported in the literature for Fe 3 O 4 [25] and of 58 emu g −1 for Fe 2 O 3 [27], both produced by the co-precipitation method. Besides that, as it may be seen in Figure 1, most of the produced MACs have M s values over 16 emu g −1 (see the dashed line in Figure 1B) which according to Wang et al. (2014) [40], ensures a proper magnetic separation [40]. Furthermore, Figure 1A,B showed that, in some cases, decreasing the mass ratio of PAC to iron salts (factor χ 2 ) and maintaining constant the Fe 3+ :Fe 2+ molar ratio (factor χ 1 ), S BET decreases and consequently, M s increases (e.g., MAC 1 to 3, MAC 13 to 15, and MAC 16 to 18).
This trend was also evidenced by other authors and was explained by the increase on the amount of magnetic iron oxide nanoparticles onto the AC framework [27,41,42].
However, this tendency was not shown by the remaining set of MAC materials (MAC 5 to 7, MAC 7 to 9 and MAC 10 to 12), probably because they are also affected by the temperature and pH of the reaction medium, also considered as variables in the synthesis of MACs. that assures an effective magnetic separation, according to [40].
As for the magnetic properties of the produced MACs, the Ms values found for the overall MACs ranged between 2.0 emu g −1 and 44.2 emu g −1 , the last one being closer to the values of 60 emu g −1 reported in the literature for Fe3O4 [25] and of 58 emu g −1 for Fe2O3 [27], both produced by the co-precipitation method. Besides that, as it may be seen in Figure 1 4 5 6 7 8 9 10 11 12 13 14 15 16 17  that assures an effective magnetic separation, according to [40].
MAC 16 (χ 1 of 2:1 (w/w), χ 2 of 1:3 molar ratio, χ 3 of 60 • C and χ 4 of 13.5), which was the material that presented the highest S BET (899 m 2 g −1 ), also displayed the highest adsorption efficiencies for all the pharmaceuticals ( Figure 1C), with percentages of adsorption Nanomaterials 2021, 11, 287 9 of 20 of (76.6 ± 4.5)% for AMX, (76.7 ± 5.8)% for CBZ and (85.5 ± 3.9)% for DCF. Despite these results, the low M s of MAC 16 (6.2 emu g −1 ) resulted in a weak magnetic separation and recovery of the material from solution. On the other hand, MAC 15 presented a M s of 44 emu g −1 , which assures a fast and effective magnetic separation from aqueous media, but a relatively low S BET (539 m 2 g −1 ) that negatively affected the adsorptive performance of this material, with adsorption efficiencies of (45.5 ± 4.1)% for AMX, (39.5 ± 5.3)% for CBZ and (37.1 ± 4.7)% for DCF. Preliminary experiments showed that the bare magnetic iron oxide nanoparticles have almost no adsorption affinity towards these pharmaceuticals, and as for PAC, the adsorption of these three contaminants was almost complete (>95%).

Analysis of Variance
To evaluate the effects of the process factors on the studied responses, allowing for a more systematic analysis of the results, an ANOVA statistical analysis was performed, and the obtained results are depicted in Table S3 in Supplementary Materials.
The p-values above 0.05 obtained for Fe 3+ :Fe 2+ molar ratio (χ 1 ), reaction temperature (χ 3 ) and reaction pH (χ 4 ) suggest that none of these factors has a significant effect on the responses. For the factor χ 2 , the p-values below 0.05 (identified in bold in Table S3) attained for S BET and M s , clearly indicate the effect of the mass ratio between PAC and iron salts used in the synthesis of MACs on both responses. Such an effect may be visualized in the relationship graphical representations displayed in Figure S3, as Supplementary Materials. The important role of the PAC:Fe ratio is in agreement with some previous studies, which refer the mass ratio between the carbonaceous precursor and the iron salts as one of the experimental conditions affecting the magnetic and structural properties of MACs produced by co-precipitation [27,41]. Moreover, the adsorption percentage of CBZ and DCF is also influenced by factor χ 2 (p-values ≤ 0.05), and since the adsorptive removal of pharmaceuticals from aqueous solutions generally depends on and is positively correlated with S BET , this behavior was somehow expected [25,27,42]. This was not the case for AMX adsorption, since the p-values > 0.05 obtained from the ANOVA analysis point that the variables tested in the synthesis of MAC did not markedly affect this response.
To identify patterns in the obtained responses, highlighting similarities and differences and allowing for the grouping of MACs, PCA and HCA analyses were performed, with the obtained results being shown in Figure 2. The score graph of the first two principal components (PC1 and PC2) explained a cumulative proportion of the data variance of 84.7%, corresponding to 73.0% and 11.7% respectively for PC1 and PC2 (Figure 2A). It is evident that M s and S BET values of the produced MAC follow opposite trends, which is related to the decrease in S BET values resultant from the incorporation of the magnetic iron oxides nanoparticles in the carbonaceous framework (further information in Section 3.3). The adsorption percentages of CBZ and DCF and S BET are grouped in the same quadrant, which reflects a strong and usually positive correlation between these responses. As for the adsorption of AMX, no correlation with the other responses was observed. However, other factors, namely the surface chemistry of MACs and the characteristics inherent to each pharmaceutical, have an important role in the removal efficiency. This might be a possible explanation for the fact that an increment in S BET not always imply an increase in the adsorption percentage and the absence of correlation between A AMX and the other responses.
From the PCA plot ( Figure 2A) and according to the dendrogram of the HCA results using centroid linkage method and Euclidean metric ( Figure 2B), five distinct clusters can be defined. Two groups presenting distinct and opposite M s and S BET values can be identified, the first one including MAC 15, MAC 12, and MAC 9 (group 1) and the second one, MAC 16 (group 2). Group 1 is characterized by MACs with the highest M s , with values varying between 37.2 emu g −1 and 44.2 emu g −1 , the high amount of iron oxide magnetic nanoparticles dispersed on the PAC structure resulting in relatively low S BET , in particular for MAC 15 and MAC 12. Due to these low S BET values, MAC 15, MAC 12, and MAC 9 exhibited the lowest adsorption percentages for AMX (between 37% and 44%), CBZ (between 40% and 48%) and DCF (between 46% and 59%). Group 2 is comprised solely by MAC 16, which has the highest S BET value, and consequently the highest adsorption percentages towards AMX (65% to 76%), CBZ (67% to 77%) and DCF (between 84% and 86%). Three additional groups can be defined in the PCA biplot: which reflects a strong and usually positive correlation between these responses. As for the adsorption of AMX, no correlation with the other responses was observed. However, other factors, namely the surface chemistry of MACs and the characteristics inherent to each pharmaceutical, have an important role in the removal efficiency. This might be a possible explanation for the fact that an increment in SBET not always imply an increase in the adsorption percentage and the absence of correlation between AAMX and the other responses.

Quadratic Regression
In order to find the optimal MACs regarding adsorption performance, a quadratic linear regression model relating a weighted average of the adsorption of the three pharmaceuticals as a function of M s and S BET was fitted. The weights of each pharmaceutical are the components of the first eigenvector, i.e., the first principal component PC1 (73.0%). Equation (4) describes the relation between S BET , M s and the overall adsorption removal (A pooled , calculated using Equation (3)), with a satisfactory correlation coefficient R 2 of 0.891: From the 3D graphical representation of the three responses (S BET , M s and A pooled ), which is depicted in Figure 3, it is clear a positive correlation between A pooled and S BET , and also that higher S BET values typically result in MACs with lower M s , and vice-versa. This information corroborates the results from the PCA analysis, in which MAC 12 and MAC 15, and MAC 16 and MAC 13, present opposite trends regarding the M s magnetic properties and both S BET and A pooled . ticals, all of them having reasonably good responses (SBET > 640 m g , MS > 20 emu g and A > 50 % for either AMX, CBZ or DFC, as shown in Figure 1).

Quadratic Regression
In order to find the optimal MACs regarding adsorption performance, a quadratic linear regression model relating a weighted average of the adsorption of the three pharmaceuticals as a function of Ms and SBET was fitted. The weights of each pharmaceutical are the components of the first eigenvector, i.e., the first principal component PC1 (73.0 %). Equation (4) describes the relation between SBET, Ms and the overall adsorption removal (Apooled, calculated using Equation (3) From the 3D graphical representation of the three responses (SBET, Ms and Apooled), which is depicted in Figure 3, it is clear a positive correlation between Apooled and SBET, and also that higher SBET values typically result in MACs with lower Ms, and vice-versa. This information corroborates the results from the PCA analysis, in which MAC 12 and MAC 15, and MAC 16 and MAC 13, present opposite trends regarding the Ms magnetic properties and both SBET and Apooled.  Furthermore, based on this 3D graphical representation, it was possible to observe that MAC 4 and MAC 7 (χ 2 of 1:3 and χ 4 of 9.5) and MAC 11 and MAC 17 (χ 2 of 1:4 and χ 4 of 13.5), are the materials that gather the best compromise between magnetic properties and binding capacity towards the pharmaceuticals from different classes here considered. For these MAC, the M s values ranged between 22 emu g −1 and 27 emu g −1 , allowing their efficient separation from aqueous media, and the values of S BET ranging between 741 m 2 g −1 and 795 m 2 g −1 were quite satisfactory, allowing the adsorption of 61% to 70%, 69% to 77%, and 80% to 84% for AMX, CBZ, and DCF, respectively.

Morphologic and Chemical Features of the Optimal MAC
The morphologic features of MAC 4, MAC 7, MAC 11, and MAC 17 were evaluated by SEM, with the obtained images being represented in Figure 4. The micrographs of these four materials show small iron oxide nanoparticles of different shapes and sizes, distributed onto the surface of PAC. Besides that, it is clear from these images that the pores of the carbonaceous matrix of PAC are partially obstructed/blocked by the magnetic nanoparticles (see Figure S4, as SM) and this explains the decrease observed in both total pore volume (between 24% to 37%) and micropore volume (between 45% to 50%), and consequently in the S BET values of these materials. Still, the SEM images of these MACs revealed the formation of well-developed porous structures, with the micropore volume accounting for ca. 39% to 46% of the total pore volume (according with the S BET data provided in Table S2). the carbonaceous matrix of PAC are partially obstructed/blocked by the magnetic nanoparticles (see Figure S4, as SM) and this explains the decrease observed in both total pore volume (between 24 % to 37 %) and micropore volume (between 45 % to 50 %), and consequently in the SBET values of these materials. Still, the SEM images of these MACs revealed the formation of well-developed porous structures, with the micropore volume accounting for ca. 39 % to 46 % of the total pore volume (according with the SBET data provided in Table S2). The presence of magnetic iron oxides in the produced materials was evaluated by XRD, obtained results being depicted in Figure 5. The absence of peaks in the PAC highlighted the lack of a measurable crystallographic order, as this material is composed by non-graphitic and non-graphitizable carbon (Pereira et al., 2020) [15].  The presence of magnetic iron oxides in the produced materials was evaluated by XRD, obtained results being depicted in Figure 5. The absence of peaks in the PAC highlighted the lack of a measurable crystallographic order, as this material is composed by non-graphitic and non-graphitizable carbon (Pereira et al., 2020) [15]. As for MAC 4, MAC 7, MAC 11, and MAC 17 all exhibited a characteristic XRD pattern with peaks at around 30.2 • , 35.7 • , 43.3 • , 54.3 • , 57.3 • , and 62.9 • , associated with the cubic spinel structure of magnetite and maghemite [15,32,43]. Although XRD results prove the existence of iron magnetic nanoparticles (magnetite and/or maghemite), they do not allow for distinguishing between these two crystalline components.
The elemental composition of PAC and the four MACs was evaluated by XRF ( Table 2). The obtained data indicate that MACs are mainly composed by Fe (>95%), a value considerably higher than that of PAC (16%). These results support the successful incorporation of iron magnetic nanoparticles on the carbonaceous framework of PAC.
Evaluating their oxide composition, the results indicate the presence of Fe 2 O 3 in MACs. Since these materials exhibited magnetic properties (M s ranging between 22 emu g −1 and 27 emu g −1 ) it is likely that the iron oxide nanoparticles are in the form of maghemite (G-Fe 2 O 3 ), and not the non-magnetic form hematite (α-Fe 2 O 3 ). Besides that, the occurrence of maghemite instead of magnetite might result from the incomplete removal of oxygen from the reaction medium (despite of the N 2 flux used), contributing to the oxidation of magnetite (usually reported as being unstable in an oxidizing environment [43,44]). Other elements were also found in MACs, namely Si, Ca, S, K, Ti, Cl, and P and their presence could derive from the precursor itself and from the chemical activation (PAC production) and co-precipitation procedures, to which the precursor was subjected. Furthermore, the very low percentages found for Cl (<0.12%) and S (<0.75%), suggests the nearly total absence of unreacted iron sulphate/chloride salts in the carbon matrix (probably eliminated during the last stage of H 2 O washing) and that the Fe content was due to the presence of iron oxide nanoparticles in MAC's composition. In the case of PAC, the Fe, Si, Ca, S, P, K, and Ti were detected with mass percentages varying between 4.0 and 22%, accounting for 96% of the elements detected. Just like in MACs, the presence of these elements in PAC's composition can derive from both raw material and procedure applied to its production.
the Fe, Si, Ca, S, P, K, and Ti were detected with mass percentages varying between 4.0 and 22 %, accounting for 96 % of the elements detected. Just like in MACs, the presence of these elements in PAC's composition can derive from both raw material and procedure applied to its production.
The surface chemical composition of MAC 4, MAC 7, MAC 11, and MAC 17 was further examined by XPS and compared with that of PAC. The XPS spectra obtained for PAC, MAC 4, MAC 7, MAC 11, and MAC 17 are presented in Figure S2, as Supplementary Material. The XPS spectrum of PAC shows two major peaks, which are the C1s and O1s, while in the case of MAC 4, besides these two, Fe2p and Fe3s peaks were also identified. Other small peaks were also visible, namely Si2p, Si2s, and N1s for all the analyzed (Figure S5).   The surface chemical composition of MAC 4, MAC 7, MAC 11, and MAC 17 was further examined by XPS and compared with that of PAC. The XPS spectra obtained for PAC, MAC 4, MAC 7, MAC 11, and MAC 17 are presented in Figure S2, as Supplementary Materials. The XPS spectrum of PAC shows two major peaks, which are the C1s and O1s, while in the case of MAC 4, besides these two, Fe2p and Fe3s peaks were also identified. Other small peaks were also visible, namely Si2p, Si2s, and N1s for all the analyzed ( Figure S5).
The C1s, O1s and Fe2p peaks of MAC were deconvoluted and compared with those of PAC. Table 3 presents the component groups, binding energies and atomic concentrations of PAC and of the four MACs, while Figure 6 presents the deconvoluted C1s and O1s peaks of PAC and C1s, O1s, Fe2p peaks for MAC 4 (as for MAC 7, MAC 11 e MAC 17 the information is depicted in Figure S6, as Supplementary Materials).   Intensity (a.u.) Binding Energy (eV) O1s Figure 6. Deconvolution of C1s, O1s, and Fe2p XPS peaks of PAC and MAC 4: Experimental peak (-), adjusted peak (-) and the component groups (-). Table 3, the deconvolution of the C1s region of PAC and the four selected MACs showed the presence of 5 peaks with binding energies ranging between 284 eV and 289 eV. These binding energies were assigned to the graphitic carbon sp 2 , the presence of C-C sp 3 and C-H and other surface functional groups including C-O (ether and alcohol), C=O (carbonyl, quinones, and ketones) and O-C=O (carboxylic acids and carboxylic anhydride) [45,46]. In both PAC and MACs, the peaks with the lowest binding energies (graphitic C and C-C sp 3 /C-H) represent the prevailing peaks of C1s, except for MAC 7. By deconvoluting the O1s peak of PAC, a main peak at 533 eV appears and is attributed to O-H in hydroxyl groups and carbonyl (C=O) oxygen atoms in lactone and anhydrides. Besides that, two other peaks appear at 534 eV and 531 eV due to non-carbonyl oxygen atoms O-C=O in lactone and anhydrides and the carbonyl group and C=O in quinones; a small peak at 536 eV appears due to the contribution from the physiosorbed water [45,46]. As for MAC materials, the deconvolution of O1s results in an additional peak at 530 eV, resulting from Fe-O [42,47]. The other peaks at 531 eV, 533 eV, and 534 eV are attributed to the oxygen in the same functional groups as in PAC and no peaks were detected at ca. 536 eV. Regarding the surface chemistry, and from the functionalities identified in O1s and C1s deconvoluted peaks (Table 3), a decrease was observed in the percentage of carbonyl, carboxyl and hydroxyl groups in the four MACs when compared with PAC. Since these surface functional groups can act as binding sites, their content may have a significant impact on the adsorption of pharmaceuticals.

As seen in
The XPS peak of Fe2p shows two peaks, one corresponding to Fe2p 3/2 peak with a binding energy between 711.4 eV and 711.8 eV and the other to Fe2p 1/2 appearing between 724.8 and 725.2 eV. The separation between these two peaks for the four MACs varies between 13.2 and 13.8 eV and the ratio between the area of Fe2p 3/2 and Fe2p 1/2 peaks is ca. 2. Besides that, a satellite peak of Fe2p 3/2 appears at around 719.6 eV and is located just about 8 eV higher than the main Fe2p 3/2 peak. According to Yamashita and Hayes (2008), the presence of this Fe2p 3/2 satellite peak on the XPS spectra of MAC materials confirms the formation of Fe 2 O 3 (maghemite) nanoparticles, since this peak does not appear in the XPS spectrum of Fe 3 O 4 (magnetite) [48]. The formation of Fe 2 O 3 instead of Fe 3 O 4 corroborates the results obtained from XRF analysis. Besides, another peak appeared at 729.5 eV corresponding to the satellite peak for Fe2p 1/2 . The Fe2p of MAC 4 was deconvoluted, as showed in Figure 6. The Fe2p 3/2 and Fe2p 1/2 peaks and their corresponding satellites can be deconvoluted into Fe 2+ and Fe 3+ peaks and the lowest binding energies were always attributed to Fe 2+ (see Table 3).
The total atomic percentage of each peak, along with the elemental composition, for PAC, MAC 4, MAC 7, MAC 11, and MAC 17 is listed in Table 3. The XPS data indicate contents of 74.9% of carbon and 17.8% of oxygen for the bare PAC. As for the composite MACs, the content of carbon decreased ca. 16% to 24% after loading the carbonaceous framework with iron oxide nanoparticles, and this variation was followed by an increase in both oxygen and iron contents, corresponding to the formation of iron oxides on their surface. Only trace amounts of nitrogen (<0.8%) were detected on the surface of all materials. Besides that, the XPS spectra identified the presence of a peak at around 102 eV, corresponding to the presence of SiO 2 in all the five materials. The presence of silicon in these materials, with atomic concentrations varying between 4.7% and 6.3%, as stated before, might be due to the own composition of the carbonaceous precursor, namely PS, and the activation procedure applied to it.

Comparison with Waste-Based MAC in the Literature
The features of waste-based MAC 4, MAC 7, MAC 11 and MAC 17, regarding M s and S BET , were compared with values reported in the literature ( Table 4). The magnetic characteristics evidenced by the referred materials were typically superior than those obtained by other carbon materials containing magnetic iron oxide nanoparticles and applied in the adsorption of pharmaceuticals, with M s values 1.4 to 4.8 times higher [30,41,42,47,49,50]. Besides that, their S BET were also higher than the values reported in the literature for other MACs [30,41,42,47,[49][50][51]. Differently, the MAC produced by Baghdadi et al. (2016) [25] presented a S BET value 1.6 to 1.7 times higher than that of MAC 4, MAC 7, MAC 11, and MAC 17, yet, its M s was ca. 5 times lower than for the MACs of the present work and, according to Wang et al. (2014) [40], insufficient to ensure an efficient magnetic separation (M s < 16 emu g −1 ). In the case of the material obtained by Rai and Singh (2018) [19], despite of the 1.4-1.6 times higher M s , the S BET was at least 2.6 lower than the values presented by the here selected MACs.
Very few information exists in the literature regarding the application of MAC in the removal of the AMX, CBZ, and DCF from aqueous media, which highlights the need for further research concerning this issue. Published data on the adsorption of these pharmaceuticals by different MACs are shown in Table S4 (depicted as Supplementary Materials) together with results obtained in this work. As it may be observed, all the A (%) in Table S4 are in the same order of magnitude, with the A AMX , A CBZ , and A DCF of the MACs produced in this work being in the range of values published in the literature [18,36,39]. However, it must be pointed out that strict comparisons between the performance of the several materials cannot be done since different experimental conditions were used in each work. This is especially evident in the case of the pharmaceutical's initial concentration and the adsorbent dose, with the lowest values being used in this study.
All the above considerations are indicative of the potential displayed by the wastebased MACs produced in this work, since they are capable of effectively adsorb AMX, CBZ and DCF from aqueous media, even when a small dose of material (35 mg L −1 ) is used, and are, simultaneously, easily and quickly recovered from the treated water by applying an external magnetic field. Due to the advantages associated with these materials, in particular their effective after-use separation, they may represent a suitable and sustainable alternative to PAC (avoiding the time consuming, costly, and inefficient separation stages) in water treatment. Considering such advantages, future work is to be carried out to further evaluate the performance of the selected MACs produced in this work, namely by determining the kinetic and equilibrium profiles towards the removal of different pharmaceuticals from WWTP effluents and exploring their regeneration/reutilization capacity by environmentally-friendly processes. Table 4. Values of M s and S BET obtained for several MAC aimed at the adsorption of pharmaceuticals from water and produced by loading iron oxide nanoparticles onto different carbon precursors by different synthesis methods.

Conclusions
In the present work, nanocomposite materials exhibiting well-developed porous structures and containing nano-sized magnetic iron oxides in their framework were produced by a simple and cost-effective co-precipitation method and using primary paper mill sludge as PAC precursor. A FFD with mixed levels was applied and the combined effect of the synthesis conditions (Fe 2+ :Fe 3+ salts molar ratio, PAC:Fe salts mass ratio, temperature, and pH) and their impact on both MAC's features (S BET and M s ) and adsorptive removal of three pharmaceuticals (the antibiotic AMX, the antiepileptic CBZ and the non-steroidal anti-inflammatory DFC) was assessed. The data obtained from the multivariate analysis revealed that the mass ratio between PAC and the iron salts used in MAC's production was the only factor significantly affecting all responses, except for the percentage of AMX adsorbed, which was not affected by any of the considered conditions. Furthermore, the obtained PCA score graph showed the definition of five distinct clusters and highlighted an opposite correlation between S BET and M s and a strong correlation between A CBZ , A DCF , and S BET responses. As for the adsorption of AMX, no correlation with the other responses was observed. Finally, a 3D graphical analysis of S BET , M s and A pooled responses allowed for the selection of the production conditions used for MAC 4, MAC 7, MAC 11, and MAC 17 as the most favourable, with these materials exhibiting high S BET , large adsorption percentages (at relatively low MAC doses) ranging between 61-70%, 69-77%, and 80-84% for AMX, CBZ, and DCF, respectively, and still guaranteeing an efficient after-use recovery by magnetic separation (M s between 22 emu g −1 and 27 emu g −1 ).

Supplementary Materials:
The following are available online at https://www.mdpi.com/2079-4 991/11/2/287/s1, Table S1: Physical and chemical properties of amoxicillin tri-hydrate (AMX), carbamazepine (CBZ) and sodium diclofenac (DCF), Table S2: Values obtained for the specific surface area (S BET ) and textural properties (total pore volume V p , micropore volume V mic , and average pore diameter D) of PAC and the eighteen MACs, Table S3: Results of ANOVA analysis: Sum of square (SS) and mean of square (MS) values, degree of freedom (df ), F-test and the p-value (confidence level of 95%), Table S4: Values of A (%) obtained for AMX, CBZ, and DCF using MACs prepared using different carbon precursors, along with the experimental conditions used in the adsorption experiments (Ci of each pharmaceutical, dose of MAC, pH, temperature and contact time), Figure  S1: N2 adsorption isotherms expressed as (A.) adsorption (cm 3 STP g −1 ) vs relative pressure (p/p 0 ) and (B.) adsorption (cm 3 STP g −1 ) vs -log (p/p 0 ) of the following materials: PAC, MAC 4, MAC 7, MAC 11, and MAC 17, Figure S2: Pore size distribution of PAC, MAC 3, MAC6, MAC 13, and MAC 16, Figure S3. Values obtained for SBET (A.) and Ms (B.) as a function of PAC:Fe salts ratio (w/w), Figure S4: SEM images of the MAC 7 using a magnification of 20,000 (A.) and 40,000× (B.), Figure S5: Overall XPS spectra of PAC, MAC 4, MAC 7, MAC 11, and MAC 17, Figure S6: Deconvolution of C1s, O1s, and Fe2p XPS peaks of MAC 7, MAC 11, and MAC 17: Experimental peak, adjusted peak and the component groups.