Removal of Antibiotics from Water by Polymer of Intrinsic Microporosity: Isotherms, Kinetics, Thermodynamics, and Adsorption Mechanism

Traces of antibiotics within domestic and industrial effluents have toxic impact on human health as well as surrounding flora and fauna. Potential increase in antibiotic resistance of microorganisms is likely to rise due to the incomplete removal of antibiotics by traditional wastewater processing, methods such as membrane filtration and biological treatment. In this study, we investigated a novel class of material termed Polymer of Intrinsic Microporosity (PIM) that is based on amorphous microporous organic materials for the application of antibiotic removal form aqueous environments. The adsorption of four commonly used antibiotics (doxycycline, ciprofloxacin, penicillin G, and amoxicillin) was evaluated and found that at least 80% of the initial concentrations was eliminated under the optimized conditions. Langmuir and Freundlich models were then employed to correlate the equilibria data; the Freundlich model fit well the data in all cases. For kinetic data, pseudo-first and second order models were examined. Pseudo-second order model fit well the kinetic data and allowed the calculation of the adsorption rate constants. Thermodynamic parameters were obtained by conducting the adsorption studies at varied reaction temperatures. Surface potential, adsorption at various solution pHs, thermogravimetric analysis (TGA), Infrared spectroscopy (IR), and surface area experiments were conducted to draw possible adsorption mechanisms. The removal of antibiotics from water by PIM-1 is likely to be governed by both surface and pore-filling adsorption and could be facilitated by electrostatic interactions between the aromatic rings and charged functional groups as well as hydrogen bond formation between the adsorbent and adsorbate. Our work shows that the application of such novel microporous material could contribute to the removal of such challenging and persistent contaminants from wastewater with further optimizations of large-scale adsorption processes.

Adsorption isotherms. Five mL solutions of the target antibiotics (neutral pH) with different concentrations (50,100,150, and 200 µM) were mixed and stirred (using magnetic bar at a speed of 400 rpm) at room temperature with the adsorbent PIM-1 (2 mg) at different adsorption times specific for each antibiotic. Equilibrium times were 24 hours for amoxicillin and penicillin-G and 5 hours for ciprofloxacin and doxycycline. Determination of equilibrium was experimentally achieved by measuring the adsorbate concentration and determining the saturation point. Antibiotic concentrations were measured as described above. Adsorption capacity of PIM-1, Q e , was calculated according to Eq. 1 51 : where Q e is the adsorption capacity at equilibrium (mg g −1 ); C 0 and C e represent the initial and equilibrium concentrations of antibiotics (mgL −1 ), respectively; V is the volume of antibiotic solutions (5 mL); and m is the mass of the adsorbent PIM-1 (mg).
Characterizations. PIM-1 molecular weight was measured by an Agilent gel-permeation chromatograph (GPC) equipped with a ZORBAX PSM 300-S column. The instrument was calibrated with standard polystyrene samples. THF was used as an eluent and was run at a flow rate of 1 mg mL −1 . The thermogravimetric analysis (TGA) was carried out using a Perkin Elmer TGA7 with a temperature range from 25 °C up to 800 °C (with an elevation rate of 10 °C min −1 ) under nitrogen constant flow. 1 H-NMR spectra were recorded at room temperature using JOEL 600 MHz and polymer solutions were prepared in CDCl 3 . Infrared spectroscopy (PERKIN ELMER16F PC FT-IR) instrument equipped with an attenuated total reflectance accessory was used to characterize PIM-1 and PIM-1-adsorbed antibiotics. Each sample was scanned 16 times at a resolution of 4 cm −1 . Water content was removed from the adsorbent through filtration and overnight drying at 40 °C. Brunauer-Emmet-Teller (BET) surface area and pore parameter measurements were conducted using Quantachrome instrument (Nova touch LX 2 model) for 50 mg of PIM-1 samples and 200 µM concentration of the target antibiotics. The samples were left under vacuum for four hours at 120 °C before the analysis. BET surface area was determined by using a multi-point method, while pore parameters were calculated by Dubinin-Astakhov (DA) method. Surface potential measurements were conducted using a Microtrac instrument (Zeta-check model) at 20 °C. Five mL of PIM-1 suspensions with various experimental conditions (different antibiotics or in different pH values) were used for the measurements.
In this study, PIM-1 was produced following the high temperature approach, due to the fact that it can be obtained in a shorter time, and the produced PIM-1 was fully characterized. The appearance of as prepared PIM-1 shows significant fluorescent yellow color. The produced PIM-1 has a molecular weight of 1.9 × 10 5 g mol −1 and a polydispersity index of 1.90 (as detailed in the experimental section). IR and 1 H-NMR (detailed in the experimental section) measurements indicated that the prepared PIM-1 was pure and similar to those previously reported 45,50,52 . The characteristic nitrile (CN) stretches at 2240 cm −1 with the aromatic and aliphatic C-H stretches around 3000 cm −1 were observed. Additionally, the presence of six different proton environments in 1 H NMR spectrum was displayed (more details are mentioned in the experimental section). Thermal stability of PIM-1 was confirmed by conducting TGA experiment where the polymer backbone degradation was observed at 450 °C, indicating a thermally stable adsorbent material. Furthermore, the nitrogen gas adsorption and desorption isotherms with their pore size distribution of PIM-1 were studied. The observed isotherm can be characterized as a type II according to the IUPAC classification 53 . BET specific surface area was found to be 630 m 2 g −1 and the total pore volume was 0.514 cm 3 g −1 . Pore size distribution, based on density functional theory, shows characteristics of microporosity (<2 nm). These data are comparable to the previously reported figures of PIM-1 45,50,52 . The lack of efficient backing of the solid state macromolecules forming PIM-1 arises from its contorted shape and extreme rigidity. Degradation of the porous structure as a result of bond rotation is prohibited, due to the presence of spirocenters and the rigid fused-ring Adsorption isotherms. An adsorption isotherm characterizes the distribution of the adsorbate molecules between the liquid phases at various equilibrium concentrations 54 . Important insights into the adsorption process, such as how the interaction between adsorbent and adsorbate occurs, are obtained by finding the suitable model that fits well the adsorption data. The most popular adsorption models are Langmuir 55 and Freundlich 56 . These two models were employed in their linear forms 57,58 , shown in Eqs 2 and 3: where C e (mg L −1 ) is the equilibrium concentration of antibiotics, Q e (mg g −1 ) is the amount of antibiotics adsorbed per gram of adsorbent PIM-1 under equilibrium, Q m (mg g −1 ) is the theoretical maximum adsorption capacity of PIM-1 for antibiotics, and K L (L mg −1 ) is a constant describing the affinity in the process of Langmuir adsorption; K F is the Freundlich empirical constant reflecting the relative adsorption capacity of the PIM-1, and 1/n is a constant indicative of the intensity of the Freundlich adsorption 57,58 . As shown in Fig. 2, the adsorption data of the studied antibiotics onto PIM-1 were fitted with both the Langmuir and Freundlich isotherms. They are known as two-parameter models, which can provide information on the adsorption capacity and constants related to the adsorption affinity. The Langmuir model assumes that the adsorption takes place in a homogeneous surface (sites of equal accessibility and energy) resulting in the formation of a monolayer of adsorbate on the surface of the material that saturates the pores and prevents the transmigration. The Freundlich model assumes that adsorption is not ideal and reversible, occurring through formation of multilayers of adsorbate on a non-uniform heterogeneous surface 57,59 .
From the data fits shown in Fig. 2 and the correlation coefficient values (R 2 ) presented in Table 1, the Freundlich model seems to be the best fit for the adsorption process of penicillin G and amoxicillin. Doxycycline and ciprofloxacin are almost equally fitted with the Langmuir and Freundlich models. Characterizing the adsorption to be chemisorption (monolayer) or physisorption (multilayer) is not only dependent on the adsorbent www.nature.com/scientificreports www.nature.com/scientificreports/ material, but also on the adsorbate. Fitting the isothermal data alone would not result in firm conclusions regarding the nature of the adsorption 60 . However, with the data presented later in this study (Section 3.5), it can be concluded that penicillin G and amoxicillin are adsorbed only by physisorption mechanism (as Table 1 shows very poor fitting for chemisorption mechanism). On the other hand, doxycycline and ciprofloxacin are well-described by both adsorption mechanisms. Previous studies have shown that both adsorption mechanisms could coexist in the same adsorption system 58,61 . Thus, the variations encountered in modeling the adsorption of the studied antibiotics is attributed to their chemical nature and their affinity for adsorption onto PIM-1. According to the Freundlich model, the values of the constant 1/n were less than 1 (Table 1) and indicated favorable adsorption 60 .
The highest adsorption capacity (Q m ) for the studied antibiotics is reported for penicillin G followed by amoxicillin, which were found to follow the physisorption mechanism with multilayer adsorption. These two compounds belong to the same antimicrobial family (β-Lactam) and differ only from one another by two functional groups (amine and hydroxyl), their chemical structures are presented in Fig. 1. The modeling of their uptakes resulted in similar isothermal characteristics in both Langmuir and Freundlich models, shown in Table 1. Ciprofloxacin resulted in the lowest Q m value (33.10 mg g −1 ) which could be attributed to the limited number of functional groups compared to the other studied compounds. Generally, the reported Q m values are better or comparable to other adsorbent materials previously reported for the removal of the same classes of the studied compounds. For example, the value of Q m was found to be 47 mg g −1 for amoxicillin, 315 mg g -1 for penicillin G, and 44.4 mg g −1 for tetracycline when bentonite, activated carbon, and mesoporous silica were used, respectively [62][63][64] .
Furthermore, the favorable nature of adsorption can be expressed in terms of the dimensionless separation factor R L 65 for an adsorbate that obeys the Langmuir model 60 . It can be calculated using the Langmuir constant K L presented in Table 1, according to R L = 1/(1 + K L C 0 ) 60 . R L values indicate the following: irreversible isotherm (R L = 0), favorable (0 < R L < 1), linear (R L = 1) or unfavorable (R L > 1). The dimensionless separation factors calculated for doxycycline and ciprofloxacin are 0.002 and 0.019, respectively. R L values were less than 1 and greater than zero indicating favorable adsorption.
Adsorption kinetics. Designing an effective and sustainable adsorption system requires deeper understanding of the dynamic of the reaction in terms of orders of the rate constants 66 . Therefore, adsorption kinetics data for the studied antibiotic systems were analyzed (as shown in Fig. 3) by two different kinetic models using pseudo-first order 67 and pseudo-second order 68 using their linear forms 60 : where Q e and Q t denote the adsorption capacity of antibiotics at the equilibrium state and at time of t, respectively; K 1 (min −1 ) and K 2 (g mg −1 min −1 ) are the modulus of pseudo-first-order and pseudo-second-order adsorption, respectively 60 . Table 2 shows the parameters obtained from the fits of the adsorption kinetic models. The correlation coefficients (R 2 ) of the pseudo-second order model for all the studied antibiotics were greater than 0.99, with the exception of amoxicillin (R 2 = 0.88) indicating the applicability of this fitting model. The implication of this finding is that the adsorption of antibiotics by PIM-1 is likely to be kinetically controlled as a second-order reaction and the adsorption is dependent on the concentration of both adsorbent and adsorbate. According to the pseudo-second-order rate constant, the fastest adsorbing compound is ciprofloxacin followed by amoxicillin, penicillin-G, and doxycycline; respectively. The reported rate constants are broadly consistent with those reported for the same antibiotics 59,61,69,70 . The pseudo-second order kinetic model suggests that chemical adsorption is the dominant mechanism, involving electrostatic attraction, which is consistent with the variable pH experiments presented in the coming section.
Additionally, the trend of adsorption capacity at equilibrium predicted by the pseudo-second-order model (listed in Table 2) is consistent with the trend of maximum adsorption capacity found by Langumir model, listed in Table 1. Both pseudo-second-order model and Langumir model found that the highest adsorption capacity is www.nature.com/scientificreports www.nature.com/scientificreports/ four β-Lactam antibiotics followed by doxycycline, and the least adsorption capacity is for ciprofloxacin. It should be noted that the limits of solubility of the studied antibiotics in aqueous solvents are higher than the highest target concentration (200 µM) used herein [71][72][73] . Thus, aggregation and precipitation do not contribute to any of the observed results.
BET surface area, IR and TGA characterizations of antibiotic adsorption. The large surface area of PIM-1 promotes it as a promising candidate for adsorption of antibiotics from aqueous environments. BET surface area measurements showed that PIM-1 powder has an apparent surface area of about 630 m 2 g −1 . It shows a type II isotherm at low pressure indicating significant microporosity. A hysteresis behavior was observed upon N 2 desorption that could be due to polymer swelling, as shown in Fig. 4. It is well known that the surface area of PIM-1 is strongly influenced by the processing history 74 . Treatment of PIM-1 with methanol for 48 hours resulted in higher surfer area (775 m 2 g −1 ) 75 .
Detailed N 2 adsorption/desorption isotherms and pore parameters data are provided in Fig. 4. The large surface area determined by N 2 gas adsorption is due to the adsorption of the gas molecules not only to the surface of PIM-1, but also inside the pores. BET surface area measurements were repeated for PIM-1 after adsorption of  www.nature.com/scientificreports www.nature.com/scientificreports/ the studied antibiotics when the adsorption process was terminated at equilibrium (specific for each antibiotic, as detailed in Sections 2.3 and 2.4).
As shown in Fig. 4A, different degrees of surface area reduction were recorded upon adsorbing the investigated antibiotics and modeling the N 2 gas adsorption/desorption. The values of surface area are inserted in Fig. 4A. Active sites responsible for N 2 gas adsorption were saturated by the presence of adsorbed antibiotics. It could be assumed that the pore filling adsorption takes place due to the moderately surface area reduction due to antibiotics adsorption. For example, there is approximately 60 m 2 g −1 surface area reduction when doxycycline was adsorbed. It is difficult to assume that the adsorption process occurs only on the surface while PIM-1 pores are freely accessible by the small-sized antibiotic molecules, smaller than the PIM-1 pore size 59,69 . Evidently, Fig. 4B shows that there were different degrees of pore volume reduction of PIM-1 after the adsorption of antibiotics. Furthermore, the isothermal data fitted by Freundlich model supports the heterogeneous nature of adsorption with uneven distribution of adsorbed molecules over the active sites, which suggests pore filling adsorption. Similar conclusions were reported with the adsorption of phenol molecules by PIM-1 49 .
Further investigation of antibiotic adsorption was conducted by IR spectroscopy. As shown in Fig. 5, a typical PIM-1 IR spectrum is shown with the characteristic nitrile (CN) stretches at 2240 cm −1 and the aromatic and aliphatic C-H stretches around 3000 cm −1 48 . The IR spectrum for PIM-1 with adsorbed antibiotics resulted in the appearance of a new peak at ~1740 cm −1 , which is attributed to a carbonyl group (C = O) present in the molecule structures of adsorbed antibiotics, shown in Fig. 1. No other peaks identifying the presence of the antibiotics could be recorded due to 1) lack of sensitivity of IR spectroscopy to proportionally small adsorbed antibiotic quantities, and 2) shielding effect of PIM-1 polymeric porous structure to antibiotic signals.
The data obtained by BET surface area and IR spectroscopy are further supported by TGA measurements. Figure 6 shows a TGA curve of powder PIM-1 sample. Consistent with previous reports 50 , PIM-1 is stable up to 460 °C, after which polymer backbone degradation occurs. TGA experiments were repeated for PIM-1 powder after the adsorption of antibiotics, shown in Fig. 6A-D. Around 2-8% weight loss is observed for PIM-1 loaded with antibiotics below 400 °C, which is due to the decomposition of the adsorbed antibiotics, as their decomposition was found to sharply occur between 100-300 °C (shown in Fig. 6A-D). The gradual thermal decomposition behavior of the adsorbed antibiotics compared to the sharper decomposition of the free forms suggests encapsulation within PIM-1 pores that provide protection against thermal decomposition. TGA results are in line with the abovementioned BET surface area and IR data.

Effect of experimental conditions: insights into the adsorption mechanism. Effect of solution
pH. Solution pH could play a major role in the adsorption process of a molecule, as pH variation could promote changes in the charges of adsorbent and adsorbate 50,60 . In the present study, an important aspect of antibiotic www.nature.com/scientificreports www.nature.com/scientificreports/ adsorption is the charge distribution on PIM-1 surface under different pH media. Surface potential measurements can provide information about the charge and charge distribution on the surface of PIM-1. Figure 7A shows different surface potential values for PIM-1 polymeric powder suspended in solutions with different pHs values. The isoelectric point (IP) was determined to be 3.3, where PIM-1 carries zero net charge 76 . At pH lower than the IP, PIM-1 is positively charge while solutions with pHs higher than the IP result in a negatively charged PIM-1.
Furthermore, Fig. 7B shows that the adsorption of the studied antibiotics at neutral pH resulted in alternation of the original surface potential of PIM-1. The highest surface potential change was observed with the adsorption of ciprofloxacin and the smallest change associated with the adsorption of amoxicillin. Such variation could be  www.nature.com/scientificreports www.nature.com/scientificreports/ attributed to the fact that the studied antibiotics could present different charges at neutral pH due to the differences in the values of acid dissociation constants (pKa), shown in Table 3. It is worth noting that the major contributor to the resolved surface potential values are the surface charges. This is due to the dependence of phase analysis light scattering technique (used to measure surface potential) on electrophoretic movability influenced by solution-polymer interphase 77 . Thus, surface potential data indicates that the adsorption of antibiotics occurs on the exterior phase of PIM-1 surface; while pore filling adsorption cannot be ruled out and was previously confirmed by the surface area measurements.
The effect of solution pH on the adsorption of antibiotics onto PIM-1 was further investigated by repeating the kinetic experiments at acidic or basic conditions. As shown in Fig. 8, a common feature across all the investigated antibiotics is that the adsorption was greatly diminished at a solution pH of 12 compared to acidic and neutral conditions. At pH 12, the surface of PIM-1 is highly negatively charged with a surface potential of -86.7 mV. Additionally, all the reported pK a values of the investigated compounds fall below pH 12 which indicate that the functional groups including hydroxyl, amine, and carboxylic acid; that are abundantly available in the molecular structure of the adsorbates; will be deprotonated bearing negative charges 58,69 . Thus, electrostatic repulsion between the negatively charged PIM-1 and the negatively charged adsorbate could be the dominant effect explaining the diminished adsorption activity at pH 12.
Additionally, hydrolysis of nitrile group (CN) and the formation of amide in the polymer backbone was previously reported in basic sodium hydroxide solution; carboxylic acid, ammonium carboxylate, and sodium carboxylate structures were also confirmed 45 . The formation of these functional groups was observed within few hours, which is likely to occur in the batch adsorption studies conducted herein and could additionally explain the decreased adsorption functionality of PIM-1 towards the studied antibiotic at pH 12.   www.nature.com/scientificreports www.nature.com/scientificreports/ Due to complex chemical nature of the studied antibiotics, it is expected that distinct adsorption behavior for each molecule could be observed at different experimental conditions. As can be seen in Fig. 8, the adsorption of doxycycline slightly prefers the neutral pH compared to acidic pH, although both conditions reached nearly the same removal percentage (75%) if enough adsorption time is provided. On the contrary, ciprofloxacin, Penicillin G, and amoxicillin were adsorbed more effectively at pH 2 compared to other pH solutions. At pH 2, PIM-1 is positively charged with a surface potential of 16.4 mV and this characteristic could increase the hydrophobicity of PIM-1 in acidic solution, which leads to a further enhancement in the adsorption of the moderately hydrophobic antibiotics (seen in Fig. 8). The pK a values of the aliphatic carboxylic acid group of amoxicillin and penicillin G are close to pH 2, as shown in Table 3. Thus, this group is not fully protonated bearing partial negative charges and could facilitate electrostatic attraction between the adsorbent and adsorbate. A common feature of these three compounds is their moderate hydrophobicity with positive log P values, shown in Table 3. Previous studies of PIM-1 surface modification have shown that replacing the nitrile group with more hydrophilic groups (amide, carboxylic acid, ammonium carboxylate and sodium carboxylate) at high pH values increased its solubility in more polar organic solvents 45 . Although this study conducts the experiments in aqueous media and PIM-1 remains insoluble in all experimental conditions, it could be speculated that acidic pH could increase the hydrophobicity of the polymer 58 , which might contribute to the enhanced adsorption of the moderately hydrophobic antibiotics (completely dissolved in water under the current experimental conditions). Similar trends were observed during the adsorption of phenol and its derivatives onto activated carbon 58 .
Effect of temperature and thermodynamic parameters. Effect of adsorption process temperature and determination of thermodynamic parameters were conducted at three different temperatures, i.e. 5, 25 and 40 °C for 3 hours. As shown in Fig. 9, three main observations could be drawn from the behavior of antibiotic adsorption by PIM-1 at increasing adsorption temperature: A) adsorption is favored at room temperature (the case of doxycycline and penicillin G), B) increasing adsorption temperature does not improve antibiotic uptake (the case of amoxicillin), and C) increasing adsorption temperature improved the antibiotic uptake (the case of ciprofloxacin). Such complex behavior highlights the challenging nature of the adsorption of this class of contaminants and the need to accurately optimize the adsorption process if a given compound is being treated. It was hypothesized that specific temperature dependence is associated with given adsorbates as their solubility could be decreased or increased with temperature variation, and therefore influencing their solid phase adsorption 78 . It is worth mentioning that all these three scenarios were previously reported for the adsorption of various organic pollutants by various adsorbents 79 . www.nature.com/scientificreports www.nature.com/scientificreports/ The thermodynamic parameters of antibiotic adsorption onto PIM-1, such as Gibbs free energy (ΔG 0 ), enthalpy (ΔH 0 ) and entropy (ΔS 0 ) were determined using the variation of solute distribution coefficient between the solid and liquid phases (K d ), through the following equations 60,78 : where R is the universal gas constant (8.314 J mol −1 K −1 ) and T is the absolute temperature of the system (K), and the factor 1000 is multiplied by K d to make it dimensionless 78,80,81 . K d must be a dimensionless parameter since the unit of ΔG 0 , gas constant, and temperature are J mol −1 , J mol −1 K, and K, respectively. Since the adsorption of antibiotics was conducted using aqueous solutions with very low concentration of the target compounds, the dimensionality of the K d (L g −1 ) can be easily converted into dimensionless values by the multiplication of the distribution coefficient by 1000 (as 1 L = 1000 g, and the solution density is 1 g mL −1 ) 78,80,81 . Furthermore, System's ΔH 0 and ΔS 0 values were determined from the slope and intercept of van't Hoff plot (ln (1000 x K d ) versus 1/T), respectively. The calculation of ΔG 0 resulted in negative values in all cases indicating spontaneous adsorption 69,78 , as shown in Table 4. The spontaneity (increase in the negativity of ΔG 0 values) was increased with the more favorable adsorption temperature, as seen in the case of ciprofloxacin adsorption. The current study applies the dimensionless distribution coefficient (K d ) to predict the system's ΔH 0 and ΔS 0 , as shown in Eq. 8. Their values were determined from the slope and intercept of van't Hoff plots (ln (1000 x K d ) versus 1/T). It was found that the use of the dimensionless K d in our system to calculate ΔH 0 and ΔS 0 is not appropriate, resulting in plots with very poor correlation coefficients (R 2 value of doxycycline is 0.18, ciprofloxacin is   Table 4. Thermodynamic parameters (as average values with uncertainty not exceeding 5%) for antibiotics adsorption onto PIM-1.