Tungsten Trioxide (WO3)-assisted Photocatalytic Degradation of Amoxicillin by Simulated Solar Irradiation

This study investigates the photocatalytic degradation of amoxicillin (AMO) by simulated solar irradiation using WO3 as a catalyst. A three-factor-three-level Box-Behnken design (BBD) consisting of 30 experimental runs is employed with three independent variables: initial AMO concentration, catalyst dosage, and pH. The experimental results are analyzed in terms of AMO degradation and mineralization, the latter of which is measured using dissolved organic carbon (DOC). The results show that the photocatalytic degradation of AMO follows pseudo-first-order kinetics. AMO degradation efficiency and the pseudo-first-order rate constants decrease with increasing initial AMO concentration and pH and increase with increasing catalyst dosage. Though AMO degradation is almost fully complete under the experimental conditions, DOC removal is much lower; the highest DOC removal rate is 35.82% after 180 min. Using these experimental results, second-order polynomial response surface models for AMO and DOC removal are constructed. In the AMO removal model, the first-order terms are the most significant contributors to the prediction, followed by the quadratic and interaction terms. Initial AMO concentration and pH have a significant negative impact on the photocatalytic degradation of AMO, while catalyst dosage has a significant positive impact. In contrast, in the DOC removal model, the quadratic terms make the most significant contribution to the prediction and the first-order terms the least. The optimal conditions for the photocatalytic degradation of AMO are found to be an initial AMO concentration of 1.0 μM, a catalyst dosage of 0.104 g/L, and a pH of 4, under which almost complete removal of AMO is achieved (99.99%).

various wastewater treatment methods have been proposed for AMO degradation; a subset of these, advanced oxidation processes (AOPs), has been found to be highly efficient in removing AMO from wastewater. AOPs rely on the in-situ production of highly reactive hydroxyl radicals ( • OH) with the help of one or more primary oxidants (e.g., ozone, hydrogen peroxide, and oxygen), energy sources (e.g. ultraviolet, solar and visible light), and/or catalysts (e.g. tungsten trioxide, and titanium dioxide) [17][18][19] . Refractory pollutants in water react with • OH, leading to their decomposition or mineralization into CO 2 , H 2 O, and inorganic ions [17][18][19] . Previous studies of AOPs, some of which are summarized in Table 1, have introduced methods such as ozonation 20 , Fenton 21 and photo-Fenton 22 , UV or solar photolysis with catalysts such as TiO 2 and ZnO [23][24][25][26][27] ). In particular, heterogeneous photocatalysis has shown considerable potential as a versatile, low-cost, environmentally friendly, and sustainable AOP treatment technology for the removal of emerging contaminants, including AMO.
Heterogeneous photocatalysis involves the irradiation of a semiconductor catalyst (e.g. TiO 2 , ZnO, WO 3 , Fe 2 O 3 , CdO, CdS, GaS, GaP, SnO 2 , ZnS, and SrTiO 3 ) with a light source (i.e., ultraviolet, solar, and visible light) to generate highly reactive transitory species (i.e., • OH, •− O 2 ) for the subsequent mineralization of organic pollutants. According to [28][29][30][31] , photocatalytic degradation reactions are initiated when the semiconductor catalyst absorbs photons from visible light. Upon the absorption of light, the electrons in the valence band (VB) of the catalyst are transferred to the conduction band (CB), generating an electron-hole pair. This leads to the formation of • OH via oxidation, with H 2 O or OH − molecules reacting with VB holes + (h ) VB , while superoxide radicals •− (O ) 2 from dissolved O 2 and CB electrons − (e ) CB are formed via reduction. These active radicals are then able to break down organic contaminants in aqueous solutions. The majority of AMO photodegradation approaches use the direct excitation of molecules by UV light (e.g., UV/TiO 2 and UV/ZnO) 23,25,30 . However, the use of UV light in the treatment of large volumes of industrial effluent is not feasible or economical. Hence, researchers have focused on using simulated sunlight and developing solar/visible photocatalysts, which exhibit high activity levels under solar and visible light radiation. Because solar energy is a cheap, abundant, non-polluting, renewable, and readily available energy source in most parts of the world 32 , solar photocatalysis has become a target green technology for the treatment of water and wastewater.
Though a few studies have focused on the degradation of AMO via solar photocatalysis using TiO 2 26,27,33 , no research on visible/solar photocatalysis assisted by tungsten trioxide (WO 3 ) has been reported to date. WO 3 , which has a narrow band gap energy (2.4-2.8 eV), is a visible light responsive catalyst with stable physicochemical properties [34][35][36][37][38] . WO 3 is physiochemically stable and is mechanically robust in aqueous solutions, while the production of high-purity WO 3 is relatively facile and cost-effective, making it a suitable choice for the photocatalytic degradation of organic pollutants under solar irradiation [34][35][36][37][38][39][40][41] . Therefore, the objective of this study is to employ WO 3 -assisted photocatalysis in a simulated solar irradiation system to facilitate AMO degradation.
The proposed reaction mechanisms for the WO 3 -assisted photocatalytic degradation of AMO under simulated solar irradiation are presented in Fig. 1 as the set of equations Eqs (1.1) to Eq. (1.11) [40][41][42] . The VB holes of WO 3 have a high oxidation power (E VB = + 3.1-3.2 V NHE ), where E VB is the valance band edge at the normal hydrogen electrode (NHE), which facilitates the oxidation of water = + .
( ) The efficiency of this process depends on several parameters, such as the initial concentration of AMO, the catalyst dosage, pH, and the presence of competing radical scavengers. In order to understand the effects of these parameters on the degradation of AMO and to optimize the experimental conditions, we apply response surface methodology (RSM), a technique commonly used in process analysis and modeling.
RSM is a collection of mathematical and statistical techniques that are useful for developing, improving, and optimizing various processes. It can be used to evaluate the relative significance of several factors even in the presence of complex interactions 48 . The main objective of RSM is to determine the optimal operating conditions for a given system or to determine a region that satisfies operating specifications. Hence, RSM allows the optimal conditions for various reactions to be identified to reduce time, labor, and material costs. RSM also quantifies the relationships between controllable input parameters and obtained response surfaces. An adequate number of experiments is required to develop a mathematical model for predicting degradation efficiency and to determine the direct and interactive effects of the operating conditions. Central composite design (CCD) and Box-Behnken design (BBD) are the two most common experimental design methods for RSM 49,50 . The BBD is more convenient and less expensive than the CCD for the same number of factors. Therefore, we utilize RSM-BBD to optimize the operating conditions and increase the efficiency of AMO degradation.
In summary, this study aims to (i) assess the AMO degradation and mineralization efficiency of simulated solar irradiation using WO 3 as a catalyst, (ii) examine the effects of different operating conditions (initial AMO concentration, catalyst dosage, and initial pH) on the photocatalytic degradation of AMO, and (iii) optimize the operating conditions based on RSM-BBD.  Table 2. Sodium hydroxide was obtained from Deajung Chemicals (South Korea). Sulfuric acid (purity ≥ 96%) was purchased from Kanto Chemicals (Japan). All other chemicals used in this study were of analytical grade. The stock solution of AMO and other solutions were prepared using de-ionized water (DI; ≥ 18.2 Ω·cm −1 ) and diluted as required. photocatalytic experiments. The photocatalytic solar irradiation set-up consisted of a light source and a photocatalytic reactor (Fig. 2). The light source was a 300-W Xenon lamp installed in a solar simulator (Model: SLB300A, Sciencetech, Canada). The photocatalytic reactor was a 250-mL glass beaker with a double-layer jacket. The solution in the photocatalytic reactor was placed on a magnetic stirrer and uniformly mixed. During the experiments, the temperature was maintained at 25 °C using a refrigerated bath circulator (Daihan Scientific, South Korea). The photocatalytic degradation of AMO was assessed using a 200-mL working solution containing an initial AMO concentration of 1.0, 1.5, or 2.0 μM, a WO 3 concentration of 0.1, 0.3, or 0.5 g/L, and an initial pH of 4, 6, or 8, which was adjusted using 1 M H 2 SO 4 and 1 M NaOH. The reaction time was 180 min, and 1-mL aliquots were taken at regular time points (0, 10, 30, 60, 90, 120, 150, and 180 min) for liquid chromatography coupled with mass spectrometry (LC-MS/MS) analysis. All samples were filtered through 0.20-μm PTFE syringe filters to remove WO 3 and were stored in amber glass vials at 4 °C until analysis. The working solutions were continuously stirred during the experiment.
Analytical methods. The residual AMO concentration was analyzed using Shimadzu LC-MS 8045. Sample separation was performed using an HSS C18 column (particle size 1.8 μm, 2.1 × 100 mm) with a gradient elution program using a mobile phase consisting of a mixture of water containing 0.1% formic acid and acetonitrile. Table 3 provides details of the analytical conditions. The mineralization of AMO was evaluated by measuring the dissolved organic carbon (DOC) using a TOC-V CPH analyzer (Shimadzu, Japan). The standard solution for DOC calibration was prepared using potassium hydrogen phthalate in the range of 1-20 mgC/L.
The photocatalytic degradation kinetics of AMO can be described with a pseudo-first-order kinetic model, as expressed in Eq. (2.1). The degradation and mineralization efficiency in terms of AMO and DOC removal were determined using Eq. (2.2). In addition, the change in degradation efficiency (∆ degradation ) and change in the pseudo-first-order rate constant (∆ k ) of AMO under different experimental conditions were calculated using Eqs (2.3) and (2.4), respectively.
www.nature.com/scientificreports www.nature.com/scientificreports/   www.nature.com/scientificreports www.nature.com/scientificreports/ experimental design. In this study, RSM based on BBD was employed to optimize the photocatalytic degradation of AMO. The RSM process can be divided into five stages ( Fig. S2(a)): (1) selecting independent variables and possible responses, (2) selecting an experimental design strategy, (3) conducting the experiments and recording the results, (4) fitting a model equation to the experimental data and producing response surface graphs, and (5) determining the optimal operating conditions. A three-factor-three-level BBD consisting of 30 experimental runs was implemented in the present study. Three main factors associated with the operating conditions were chosen as independent variables: (A) initial AMO concentration (µМ), (B) catalyst dosage (g/L), and (C) pH. The ranges and levels of the independent variables determined by BBD are presented in Table 4 ( Fig. S2(b)).
The experimental data were analyzed using Minitab statistical software (Version 18, Minitab Inc., State College, PA) and fitted to a second-order (quadratic) polynomial model as follows: where Y is the predicted response, X i and X j are the independent variables, β 0 is the constant coefficient, β i is the linear coefficient, β ij is the interaction coefficient, and β ii is the quadratic coefficient. For statistical calculations, the real values of the variables were transformed into coded values using the following equation 48 : where X 0 is the real value of the independent variable at the center point, X i is the real value of the independent variable, and δX is the step change between the low (−1) and high (+1) levels (as shown in Table 4).
In this study, we have three independent variables (denoted as A, B, and C for initial AMO concentration, catalyst dosage, and pH, respectively), and the mathematical relationship model is written exclusively as To evaluate the full quadratic approximation of the BBD response surface model, an analysis of variance (ANOVA) for the experimental response was conducted. The F-values and associated p-values were used to determine the order of the model (linear, square, or full quadratic) 51 . A multiple regression analysis was performed to fit the response function to the experimental data. The significance of each coefficient was determined from the t-values and associated p-values. Coefficients with t-values greater than 95% or p-values smaller than 0.05 were considered statistically significant.

Results and Discussion
photocatalytic degradation of amoxicillin. Photocatalytic degradation kinetics. The photodegradation kinetics of AMO followed a pseudo-first-order kinetic model by plotting ln(C/C 0 ) against time (t) at different concentrations, as expressed by Eq. (2.1). The pseudo-first-order rate constant (k) was determined from the slope of the linear regression line of ln (C 0 /C) versus time. The pseudo-first-order rate constants and the linear regression coefficient (R 2 ) for the photodegradation kinetics of AMO are given in Fig. 3 and Table 5.
As seen in Fig. 3 and Table 5, degradation plots for the different operating conditions exhibited an almost linear relationship (R 2 ≥ 0.989). This indicates that the degradation of AMO in both simulated solar photolysis and WO 3 /simulated solar photocatalysis follows pseudo-first-order kinetics with high linear regression coefficients. It was also found that WO 3 photocatalysis was much more efficient than direct photolysis in terms of AMO degradation as the pseudo-first-order rate constants for WO 3 photocatalysis increased at least 22-fold compared to that for direct photolysis (Table 5). Furthermore, the photocatalytic degradation of AMO increased with reaction time, with degradation occurring particularly rapidly in the first 90 minutes before slowing. For example, for the experimental conditions of 1.0 µМ of C 0 , 0.3 g/L of WO 3 , and a pH of 4, the AMO degradation rate was rapid at first, reaching 94.14% after the first 90 min of irradiation, before finally reaching 99.39% after 180 min.
Effect of initial amoxicillin concentration. To investigate the effects of initial AMO concentration (C 0 ) on photocatalytic degradation, experiments with initial concentrations of 1.0 and 2.0 µМ were carried out. As shown in Figs 4 and S3, the AMO degradation efficiency decreased with an increase in C 0 when the other two parameters (catalyst dosage and pH) were unchanged. AMO degradation was found to follow a pseudo-first-order kinetic model in all cases. Pseudo-first-order rate constants (k) were calculated from the slope of degradation plots (R 2 > 0.986), and it was found that they decreased as C 0 increased. More specifically, as the C 0 increased  Table 4. Experimental ranges and levels of the independent operating variables.  www.nature.com/scientificreports www.nature.com/scientificreports/ It is clear that the higher the dose of the catalyst, the greater number of active sites on the catalyst surface available, thus facilitating the formation of • OH, and subsequently increasing the removal efficiency and rate constants 31,54 .
Effect of pH. The pH of the aqueous solution is an important parameter in the photocatalytic degradation of organic compounds because it determines the charge of the catalyst molecules, the size of the aggregates, the charge of the organic pollutants, the adsorption of the organic pollutants onto the catalyst surface, and the concentration of • OH radicals 54,55 . To investigate the effect of pH on the photocatalytic degradation of AMO, experiments with an initial pH of 4 and 8 were conducted (Figs 6 and S5). It was observed that the photocatalytic reaction followed a pseudo-first-order reaction. Figure 6 shows that the removal efficiency and the rate constants decreased with an increase in pH, as calculated using Eqs This can be explained by changes to the surface charges of WO 3 and AMO. The pH at the point of zero charge (pH ZPC ) for WO 3 is 1.9 54 . Therefore, the WO 3 surface is negatively charged above a pH of 1.9. On the other hand, ionic AMO species are positively charged at acidic pH and negatively charged at alkaline pH 23,56,57 . At an acidic pH, AMO and WO 3 have opposing charges, thus the electrostatic attraction between the AMO molecules and the catalyst surface increases the absorption rate of AMO onto the WO 3 surface, leading to an increase in degradation efficiency and the rate constants. At an alkaline pH, both AMO and WO 3 are negatively charged, leading to repulsive forces between the AMO molecules and the WO 3 surface. Thus, adsorption onto the surface of WO 3 is limited, meaning lower degradation efficiency and rate constants. Therefore, when the pH increased from 4 to 8, the repulsive forces hindered the absorption of AMO onto the WO 3 surface. As a result, the AMO degradation efficiency and pseudo-first-order rate constants decreased.  www.nature.com/scientificreports www.nature.com/scientificreports/ Mineralization of amoxicillin. To assess the photocatalytic degradation of AMO in terms of mineralization, DOC was measured. Of the 30 experimental runs using three-factor-three-level BBD, the highest DOC removal rate was 35.82% at a C 0 of 1.5 µМ, a WO 3 dosage of 0.1 g/L, and a pH of 4. Although AMO removal was high under the same conditions (95.61%), DOC removal was quite low, with a ~40% lower removal rate. This was due to the formation of intermediates that could not be completely mineralized under these experimental conditions. Other studies have reported similar discrepancies between AMO and DOC removal using different AOPs (Table 1). Elmolla et al. 23 used UV/H 2 O 2 /TiO 2 photocatalysis under UVA irradiation and reported the complete degradation of AMO in 20 min and DOC removal of 13.9% after 300 min. Similarly, Elmolla et al. 25 adopted a UV/ZnO photocatalytic process and achieved complete AMO degradation, while DOC removal was only 9.7% after 180 min. Andreozzi et al. 20 utilized ozonation and obtained a removal rate of more than 90% for AMO, but a total organic carbon (TOC) removal rate of only 18.2% after 20 min. Using Fenton's reagent treatment, Ay et al. 21 reported complete AMO degradation and a TOC removal rate of 37% after 15 min. Therefore, the degradation of AMO does not necessarily lead to the complete mineralization of the products and transformed intermediates of AMO. It should be kept in mind that intermediate products can be toxic in aquatic environments 58,59 , thus the optimization of the process parameters should also target complete mineralization as an end goal.

Response surface methodology. Box-Behnken design.
A three-factor-three-level Box-Behnken design (BBD) consisting of 30 experimental runs was adopted to optimize the experimental data. Three operational parameters -initial AMO concentration (µМ; A), catalyst dosage (g/L; B), and pH (C) -were chosen as independent variables for BBD (Table 4). Because photocatalytic performance was assessed in terms of the degradation and mineralization of AMO, the removal of AMO and DOC were employed as the responses for the experimental runs in the present study. Therefore, four predicted responses -Y 1 (AMO removal after 30 min), Y 2 (AMO removal after 90 min), Y 3 (AMO removal after 180 min), and Y 4 (DOC removal after 180 min) -were expressed as second-order polynomial equations. The complete Box-Behnken design matrix, with the three independent variables (A, B, and C) and the predicted and experimental values of the four responses (Y 1 , Y 2 , Y 3 , and Y 4 ), is presented in Table 6. The second-order polynomial equations for the four predicted responses are provided in Table 7 (Eqs 3.1-3.4).
The quality of the model fit was evaluated using analysis of variance (ANOVA); the results are presented in Table 8. The ANOVA revealed that the four second-order quadratic regression models were highly significant, because the with F-values (17.52, 28.40, 48.02, and 21.14 for the Y 1 , Y 2 , Y 3 , and Y 4 models, respectively) greater than the tabular F-value for α = 0.05 (F 0.05(9,9) tabular = 3.18 60 ). The p-values of the four models were very low (0.0000 for all models), which confirms that the models were statistically significant at a 5% level. The F-value is a statistically valid measure of how well the factors describe the variation in the data in that the mean and estimated factor effects are real. The p-value is relatively low (p < 0.05), demonstrating the significance of the model. Thus, with their large F-values and small p-values, the four models explain the measured data well, with the corresponding coefficients demonstrating high significance.
To evaluate whether the models could successfully be used for prediction, the lack of fit (LOF) was assessed. The LOF for the models Y 1 and Y 4 was statistically significant (p < 0.05), meaning there was a greater than 5% chance of prediction failure compared to experimental data for AMO removal after 30 min and DOC removal after 180 min. In contrast, the LOF for the models Y 2 and Y 3 was statistically insignificant (p = 0.200 and p = 0.288, respectively), meaning Y 2 and Y 3 can be successfully used for prediction and optimization. Furthermore, to determine whether the models fit the experimental data well and whether the independent variables had a significant effect on the responses, coefficients of determination (R 2 ) were used. R 2 represents the proportion of the variation in the response that is explained by a model and is a statistical measure used for the goodness-of-fit for a model 61 . R 2 always falls between 0 and 1. The closer R 2 is to 1, the more closely the model fits the experimental data and the better it can predict the response 61,62 . In the present study, R 2 was found to be 0.8874, 0.9274, 0.9558, and 0.9049 for the Y 1 , Y 2 , Y 3 , and Y 4 models, respectively, indicating that the quadratic polynomial models were a good fit for the experimental results.
Adjusted and predicted R 2 also explain the goodness-of-fit for a model. Adjusted R 2 is used to compare the goodness-of-fit for a regression model that contains a differing number of intendant variables. A high value for adjusted R 2 indicates high significance, and the smaller the gap between R 2 and adjusted R 2 , the stronger the goodness-of-fit for a model 61 . Predicted R 2 is used to determine how well a regression model is able to make a response prediction. In the present study, adjusted R 2 was 0.8368, 0.8948, 0.9359, and 0.8621 for the Y 1 , Y 2 , Y 3 , and Y 4 models, representing high significance. Likewise, the small gap between R 2 and predicted R 2 shown in Table 8 indicates that the models almost perfectly explained the variation in the experimental data and can thus be used for prediction.
In addition, the diagnostic plots presented in Fig. 7, which were employed to determine the residual analysis of the response surface design, confirmed the close association between the statistical assumptions and the analyzed data. The experimental data and the responses predicted by the models are compared in Fig. 7(a). Plots of normal probability versus internally studentized residuals, an important diagnostic tool to determine whether the assumptions underlying the statistical analysis are met, were constructed for the four models ( Fig. 7(b)). As seen in Fig. 7, the predicted and experimental results were in good agreement. Therefore, the ANOVA, LOF, and R 2 results together confirmed that all of four models were statistically significant and could be used to predict the degradation of AMO using photocatalysis.
The quantitative effects of the independent variables in each of the four models were evaluated using the ANOVA results and percentage contributions (PCs) for each individual factor ( Table 9). The PCs for each individual variable were calculated using the sum of squares (SS), and the total percentage contributions (TPCs) for the first-order, interaction, and quadratic terms were determined using Eqs (3.5-3.7) 61   www.nature.com/scientificreports www.nature.com/scientificreports/ where TPC i , TPC ij , and TPC ii are the total percentage contribution of the first-order, interaction, and quadratic terms, respectively, and SS i , SS ij , and SS ii are the computed sum of squares for the first-order, interaction, and quadratic terms, respectively. Figure 8 presents the quantitative effects of the independent variables in the four models on AMO removal (after 30, 90, and 180 min) and DOC removal after 180 min. As seen in Fig. 8(a) and Table 9, as the reaction time increased from 30 min to 180 min, the contributions of initial AMO concentration (A) and catalyst dosage (B) to AMO removal increased by 8.50% and 12.88%, while the contribution of pH (C) decreased by 30.21%. The first-order terms accounted for a higher contribution than the other terms, with a total percentage contribution (TPC i ) of 90.18% for Y 1 , 87.98% for Y 2 , and 81.35% for Y 3 . The second highest contribution was exhibited by the interaction terms, with a total percentage contribution (TPC ij ) of 5.52% for Y 1 , 6.43% for Y 2 , and 13.31% for Y 3 . As the reaction time increased from 30 min to 180 min, the contributions of the first-order terms decreased by 8.83%, while those of the quadratic and interaction terms increased by 1.04% and 7.79%, respectively, indicating that the combined effects of the variables on AMO removal increased with increased reaction time, though they were still much less influential than the effects of the variables individually.
In the Y 1 model (AMO removal after 30 min), the three first-order terms (A, B, and C) and one interaction term (AC) were found to be statistically significant (p < 0.05). In particular, pH (C) exhibited the highest level of significance, with a contribution of 54.05%. The TPC i for the first-order terms was 90.18%, with the lowest contribution shown by the quadratics terms.
In the Y 2 model (AMO removal after 90 min), three first-order terms (A, B, and C) and two interaction terms (AB and BC) were statistically significant (p < 0.05). The first-order terms had the highest level of significance (with a contribution of 87.98%), followed by the interaction terms and the quadratics terms (Fig. 8a).
In the Y 3 model (AMO removal after 180 min), all of the terms (A, B, C, A 2 , B 2 , C 2 , AB, AC, and BC) were statistically significant for AMO removal (p < 0.05). Based on the results of the three time-specific models (Y 1 , Y 2 , and Y 3 ), it was found that the first-order terms made the greatest contribution to AMO degradation over the entire reaction time. For the three first-order terms, in the early stages of the reaction process (i.e., 30 min), AMO removal was significantly affected by initial pH, with a contribution of 54.05%, while the roles of other two factors became more important as the reaction time increased.
In the 180-min DOC removal model (Y 4 ), all of the terms except C (p = 0.400) and AB (p = 0.087) were statistically significant. In contrast to Y 1 -Y 3 , the first-order terms accounted for only 18 Table 8. ANOVA results from the response surface quadratic models for AMO removal (%) and DOC removal (%).
www.nature.com/scientificreports www.nature.com/scientificreports/ TPC ij , respectively; Fig. 8b). This difference from the other models might be because of the unknown concentration of intermediates produced during the reaction. As mentioned above, intermediates containing organic carbon that are present due to the incomplete mineralization of AMO should be taken into consideration by DOC removal models to allow for more accurate prediction. Given the presence of these intermediates, the interaction and quadratic terms became more important to the prediction of DOC removal. More specifically, the quadratic term A 2 demonstrated the highest contribution (33.43%), followed by B 2 and C 2 . In addition, BC accounted for 22.00% of the TPC ij of 26.85%. www.nature.com/scientificreports www.nature.com/scientificreports/ Response surface analysis. Over the 30 experimental runs, the highest AMO degradation was 99.64% with C 0 = 1.0 µМ, WO 3 = 0.3 g/L, and pH 4 for a reaction time of 180 min. As a result, the Y 3 model (with a reaction time of 180 min) was selected and analyzed to determine the optimal conditions based on RSM and BBD for AMO degradation efficiency in the proposed photocatalysis process.
Three-dimensional (3D) response surface plots and two-dimensional (2D) contour plots for AMO degradation efficiency (%) are presented in Fig. 9 for the following combinations: C 0 versus catalyst dosage ( Fig. 9(a)), C 0 versus pH ( Fig. 9(b)), and catalyst dosage versus pH (Fig. 9(c)). Figure 9(a) presents the combined effects of initial AMO concentration (A) and catalyst dosage (B) on AMO removal (%). As seen, at a constant catalyst dosage, AMO removal decreased as the initial AMO concentration increased, and this trend became more obvious at lower catalyst dosages. For example, with a catalytic load of 0.1 g/L, AMO removal decreased by 18.42% when AMO C 0 increased from 1.0 μM to 2.0 μM, while AMO removal only dropped by 6.38% at a dosage of 0.5 g/L. The negative coefficient (−6.21) for A in the response function (Eq. 3.3) also confirmed the antagonistic effect of initial AMO concentration on AMO removal. As mentioned in the previous section, one reason for this inverse relationship between initial AMO concentration and AMO degradation efficiency might be the reduction in the number of available active sites on the photocatalyst surface and the deactivation of the catalyst due to the accumulation of products and reactants on its surface. Conversely,  www.nature.com/scientificreports www.nature.com/scientificreports/ when the catalyst dosage increased, AMO removal increased when the initial AMO concentration was constant. Specifically, at an initial AMO concentration of 1.0 μM, AMO removal increased by 5.84% when catalyst dosage increased from 0.1 g/L to 0.5 g/L. Similarly, at an initial AMO concentration of 2.0 μM, the AMO removal increased by 17.89% for the same increase in catalytic dosage. The positive coefficient (5.93) for B in Eq. (3.3) also confirmed that catalytic dosage had no antagonistic effect on the response. A possible explanation for this may be that there was an increase in the number of available active sites on the photocatalyst, resulting in an increased in • OH generation, and thus greater AMO degradation. Figure 9(b) summarizes the influence of initial AMO concentration (A) and pH (C) on AMO removal (%) at a constant catalyst dosage (WO 3 = 0.3 g/L). It can be observed that both initial AMO concentration and pH had a negative effect on AMO removal. At pH 4, AMO removal decreased by 7.29%, compared to a decrease of 17.51% at pH 8, when the initial AMO concentration increased from 1.0 μM to 2.0 μM. Similarly, when pH increased from 4 to 8, AMO removal decreased by 5.94% at an initial AMO concentration of 1.0 μM and by 16.15% at an initial AMO concentration of 2.0 μM. The inverse relationship between these two factors and AMO degradation efficiency was also illustrated by the negative coefficients for A (−6.21) and C (−5.53) in Eq. (3.3).
The interaction effects of catalyst dosage and pH on AMO degradation efficiency are presented in Fig. 9(c). The results show that, at a constant AMO concentration of 1.5 μM, catalyst dosage had a positive effect on AMO removal, while pH had a negative effect. When the pH was increased from 4 to 8, AMO removal decreased by 19.65% and 2.44% at catalyst dosage of 0.1 g/L and 0.5 g/L, respectively. The inverse relationship between pH (C) and AMO degradation was also illustrated by the negative coefficient for C (−5.53) in the response function (Eq. 3.3). In contrast, when the catalyst dosage increased from 0.1 g/L to 0.5 g/L, AMO removal increased by 3.26% at a pH of 4 and by 20.47% at a pH of 8. AMO degradation was directly proportional to catalyst dosage (B), as indicated by the positive coefficient (5.93) for B in Eq. (3.3).
Optimization of AMO degradation efficiency. One of the main objectives of this study was to find the optimal conditions for the target parameters in order to maximize AMO degradation. Thus, the desired goal was defined as a maximum AMO degradation of 100%. Using Minitab software, the optimal conditions for maximum AMO degradation were found to be an initial AMO concentration of 1.0 μM, a catalyst dosage of 0.104 g/L, and a pH of 4. Under these conditions, AMO and DOC removal after 180 min were predicted to be 99.99% and 24.75%, respectively. www.nature.com/scientificreports www.nature.com/scientificreports/

Conclusions
In this study, the WO 3 -assisted photocatalytic degradation of AMO with three influencing factors (initial AMO concentration, WO 3 dosage, and pH) was investigated using simulated solar irradiation. Based on AMO and DOC removals, the following conclusions were drawn: • The photocatalytic degradation of AMO followed pseudo-first-order kinetics. The pseudo-first-order rate constants and degradation efficiencies decreased with increasing initial AMO concentration and pHs, while showed increasing with more WO 3 dosage. Compared to almost compete AMO degradations, much lower www.nature.com/scientificreports www.nature.com/scientificreports/ mineralization (i.e., low DOC removals) was found, presumably due to the intermediate products formed via AMO oxidation.
• The second-order polynomial regression models revealed good fit to the experimental data. The total percentage contributions showed that the highest contributions to AMO removals were by the first-order terms, however, in the regression models of DOC removals, contributions by the quadratic terms were significantly increased. • Initial AMO concentration of 1.0 μM, WO 3 dosage of 0.104 g/L, and pH of 4 were found to be optimum conditions for complete removal of AMO.