A multivariate study of the rGO/g-C3N4 composite for the photocatalytic degradation of bisphenol A: Characterization, degradation kinetics, products identication, and ecotoxicity

Bisphenol A (BPA), a common polymer plasticizer, is a contaminant of emerging concern with endocrine disrupting activity. Among existing abatement methods, photodegradation demands easily fabricated, inexpensive, high photoactive catalysts, leading to non-toxic byproducts after degradation. It is proposed an optimized (surface response methodology) catalyst for those goals: graphitic carbon nitride impregnated with reduced graphene oxide. The method was based on the sonication of preformed particles followed by reduction with hydrazine in reux, a methodology that allows for better reproducibility and larger specic surface areas. The catalyst removed 90% of BPA (100 mL, 100 µg L − 1 ) in 90 min under UV irradiation (365 nm, 26 W) compared to 50% with pure g-C 3 N 4 (pseudo-rst-order kinetics). Tests with radicals scavengers revealed that superoxide radical was the main oxidation agent in the system. By mass spectrometry, two major degradation products were identied, which were less ecotoxic than BPA towards a series of organisms, according to in silico estimations performed with the ECOSAR 2.0 software.


Introduction
The use of water resources is a critical topic of discussion and technology development. 61% of the world population has no access to safely managed drinking water services, and approximately 48% of the population drinks water with no kind of treatment at all. Therefore, the importance of developing water treatment technologies is undeniable (UNESCO,2019).
The traditional drinking water treatment is mainly based on the physical-chemical processes of oc-coagulation followed by decantation and disinfection. Those processes are not designed to remove pollutants that occur at concentrations ranging from ng L − 1 to µg L − 1 , which are not currently analyzed by the environmental agencies. Although those pollutants, known as Contaminant of Emerging Concern (CEC), may not cause acute effects on a variety of organisms, their chronic effects are reported in the literature (Ankley et al., 2008).
An example of CEC is bisphenol A (BPA), a plasticizer agent largely used in several industries, which presents endocrine disrupting activity. BPA-laden water consumption is associated with human health problems like infertility, obesity, cancer, and attention de cit hyperactivity disorder -ADHD (Ziv-Gal, Flaws, 2016; Wassenaar, et al., 2017;Rochester et al., 2018). Moreover, BPA is recalcitrant to chlorination and ozonation (Reddy et al., 2018).
Among the available methods for BPA abatement, heterogeneous photocatalysis is an e cient complementary technology to the traditional drinking water ones. That method is based on the light absorption capability of semiconductors, forming electron-hole pairs that perform redox reactions. The process e ciency is mainly affected by the charge carriers recombination time.
Titanium dioxide (TiO 2 ) is the most common semiconductor for that purpose due to relatively low recombination rates, but it only can be activated by ultraviolet light (UVA), rendering it less e cient for solar-based treatments (Moma, Baloyi, 2019). Several materials (ZnO, Mn x O y , CdS), with smaller bandgaps, have been reported as replacements for TiO 2 , as they may be more easily activated by solar light. However, metal-based semiconductors may ultimately contaminate the treated water with metallic ions (secondary pollution), motivating research about metal-free semiconductors (Cao et  Graphitic carbon nitride (g-C 3 N 4 ) is an adequate choice due its simple preparation from inexpensive precursors (melamine or urea) and high synthesis throughput. Although it is activated by visible light, its charge recombination rate is high (Ai et al., 2015;Chen et al., 2016). To increase its charge separation, g-C 3 N 4 can be associated with reduced graphene oxide (rGO), producing a metal-free composite, where rGO concentrates electrons, while g-C 3  rGO is produced by the reduction of graphene oxide (GO), from graphite oxide (GrO) exfoliation -an inexpensive precursor as well. It is largely associated with a variety of semiconductors, enhancing their photocatalytic activity by acting as an electron-acceptor (Mural et al., 2015;Nikokavoura, Trapalis, 2017;2018;Meng, Zhang, 2018). rGO/g-C 3 N 4 composites can be produced by thermal or sonochemical approaches (Wan et al., 2018, Aleksandrzak, et al., 2017. While the thermal path usually leads to poor distribution of rGO into de g-C 3 N 4 matrix, the sonochemical approach may generate materials with larger speci c surface areas, materials with improved reduction, besides less energy consumption in comparison to the other path. Moreover, the experimental conditions are easily controlled and reproduced . Although rGO/g-C 3 N 4 preparation conditions have been previously described, the photocatalytic e ciencies of the prepared composites were very different, even in similar works that typically report the photocatalytic degradation of synthetic dyes or phenol halides (Ai et al., 2015;Gu et al., 2018;Aleksandrzak et al., 2017). It lacks information about the e ciency of rGO/g-C 3 N 4 for degrading CECs like BPA. Moreover, all those papers used a univariate approach for improving the degradation outcome, possibly leading to pseudo-optimal conditions, and not actually the best ones (Bruns et al., 2006). Therefore, the objective of the present work was to perform a multivariate optimization of the rGO/g-C 3 N 4 composite synthesis aiming at increasing its photocatalytic activity, employing the response surface methodology and using BPA photodegradation as the response-variable. The composite was characterized and the BPA degradation kinetics and mechanism were studied. Furthermore, photodegradation products were determined and their respective ecotoxicities estimated.

Synthesis of the g-C 3 N 4
Graphitic carbon nitride (g-C 3 N 4 ) was synthesized by the methodology developed from previous works (Cadan et al., 2021). Two milligrams of melamine were put into a porcelain crucible (model A-45 with cover) and pyrolyzed in a mu e furnace (EDG 7000 coupled to an EDG heater EDGCON 3P).
Melamine was heated, from room temperature to 50ºC, at a rate of 10ºC min − 1 and held at that temperature for 30 min.
Then the powder was heated from 50 to 605ºC at a rate of 6ºC min − 1 . The system remained at that temperature for 183 min. Then, the material was naturally cooled to room temperature.
The collected material was ground in an agate mortar/pestle and transferred to Falcon® tubes (50 mL) which were kept away from light. The amount of g-C 3 N 4 needed for performing all experiments were produced and homogenized.

Synthesis of GrO
Graphite oxide, GrO, was synthesized by a modi ed Hummers methodology adapted from Chen et al. (2013) -more details are given in the Supplementary Material. Graphite powder (1.00 g) was mixed with 70 mL of concentrated sulfuric acid in a 2-L beaker. To this mixture, 9.00 g of potassium permanganate were added. The system was heated to 40 °C and magnetically stirred for 40 min. Then, 150 mL of ultrapure water was poured into it. Finally, 500 mL of ultrapure water were added to the ask. It was magnetically stirred for 15 min more and 10.0 mL of concentrated hydrogen peroxide (10 mol L −1 ) were slowly added.
This mixture was magnetically stirred for 20 min and vacuum ltered. The material was suspended in 250 mL of hydrochloric acid 1.0 mol L − 1 and vacuum ltered again. The amount of synthesized GrO required for all experiments was prepared and homogenized. The solid was previously frozen, grounded, and dialyzed with a membrane capable of removing ions from 8,000 to 14,000 Da. The external solution was renovated every 2 h during the rst day, every 4 h during the second one, and then every 12 h thereafter, until the spent water pH matched the one of ultrapure water (pH meter Marconi PA 200). The puri ed material was air-dried and lyophilized for six days. This material was ground into an agate mortar/pestle and transferred to Falcon® tubes (50 mL) which were also kept away from light.

Preparation of the rGO/g-C 3 N 4
The composite was prepared by a sonochemical route. Required amounts of g-C 3 N 4 and GrO were weighed to prepare 1.0 g L − 1 suspensions. Those suspensions were sonicated in parallel with the aid of two ultrasound (tip) devices (BRANSON model 450 and 550, for GO and g-C 3 N 4 , respectively) with power of 14 W (3 s on, 7 s off) for 1.5 h. The suspensions were then mixed and magnetically stirred and the pH was adjusted to 3.0 by the addition of H 2 SO 4 3.0 mol L − 1 . After 20 min, the mixture was sonicated again by the same routine for a speci c time. To the sonicated mixture, a certain amount of hydrazine was added. That mixture was re uxed at 98 °C for 24 h. That suspension was naturally cooled down to room temperature and vacuum ltered through 0.45 µm pore diameter cellulose acetate membranes. The composite was air-dried during 24 h, away from light, grounded using an agate mortar/pestle, and stored in 1.5-mL Eppendorf® tubes.

Optimization of the rGO/g-C 3 N 4 synthesis
The response surface methodology (RSM) was used for optimizing the composite synthesis (Bruns et al., 2006). The response-variable was BPA degradation. For more details, please check the Supplementary Material. Initially, a 2 3 full factorial design (in duplicate) was performed. Three factors were studied: weight percentage of GrO in the mixture prior to reduction (%GrO), the mixture sonication time, and the N 2 H 4 :GrO weight ratio.
Second, based on the obtained results, a polynomial was adjusted to the data and a set of experiments (according to Equation SM4) were performed along the path of steepest ascent, i.e., towards increasing BPA degradation.
Finally, over the region with the greatest BPA degradations, a central composite design (CCD) was performed. Another polynomial was adjusted to the data. The maximum of that polynomial estimates the best synthetic conditions that would maximize BPA degradation. An analysis of variance (ANOVA) was performed to check the tting of the model generated.

Photodegradation experiments
Photodegradation experiments were performed in a 250-mL, temperature-controlled open-jacketed reactor maintained at 20ºC. Typically, 100 mL of BPA 100 µg L − 1 at pH 6.0 (adjusted with aqueous ammonia) and 5.0 mg of catalyst (composite) were put into the reactor and kept under magnetic stirring and air bubbling (approximately 270 mL min − 1 ) in the dark for 30 min. Then, a black light blue (BLB) lamp (Empalux®, 25 W) placed 15 cm over the suspension surface was turned on for 60 min. Afterwards, the irradiated suspension was ltered through 0.45 µm pore diameter cellulose acetate membranes, stored in amber asks, and kept in the fridge (5ºC) until the extraction step. Adsorption and photolysis experiments were performed in the same manner, except for the absence of light and catalyst, respectively.

Extraction step and HPLC analyses
Extraction was performed by the vortex-assisted liquid-liquid microextraction (VALLME) technique, based on Yantzi et al.  (Table 1). Table 1 -Scavengers and respective inhibited species

Pre-optimization experiments (Zeta potential determination)
Zeta potential (ZP) was measured at different pH values in order to determine the pH in which the precursors had opposite charges, since the critical steps of the composite formation is the electrostatic agglomeration of pre-formed particles. At pH 3.0, the surface charge of g-C 3 N 4 and the exfoliated GrO (called graphene oxide, GO) are approximately + 7.96 and − 32.7 mV, respectively, as shown in Figure SM1. That pH value was used in all syntheses.
3.2 Optimization of the formation of rGO/g-C 3 N 4 Initially, a 2 3 (3 factors, 2 levels) full factorial design in duplicate was performed, generating 16 experiments, as shown in Table SM1. The estimated experimental error was 1.9% (Equation SM2). That error was low as it encompasses several steps (catalyst synthesis, photocalytic experiments, extraction, and HPLC analyses).
Main and interaction effects, along with their statistical signi cance (95% con dence interval), are presented in Figure  SM2. This might mean that, in both levels, hydrazine was in excess. Thus, the amount of hydrazine was set at its lower level in the following experiments.
The experimental conditions with which the best BPA removal was achieved, along the path of steepest ascent (Table  SM2), were used as the central point of a CCD (Table SM3). The model proposed for de CCD is shown in Eq. (2).

Characterization
According to the XRD analysis ( Fig. 1) both precursor materials were formed and the composite crystalline structure did not differ from the one of g-C indicating that rGO alters the speci c surface area of the composite, probably by unpacking g-C 3 N 4 layers due an electrostatic effect. As a matter of fact, the expected SA for the simple mixture of g-C 3 N 4 and rGO in the same content (16% rGO) is 47 m 2 .g − 1 , con rming the SA increasing effect.
The pH of the composite formation (set in 3.0) also can be associated to the SA since pH affects how well the exfoliation and agglomeration of the material occur (Silva et al., 2017). The ZP curve ( Figure SM4) of the composite shows that the optimized material follows the same pro le of pristine g-C 3 N 4 , previously shown ( Figure SM1).
By the EDS results ( Figure SM5), one can observe that pure g-C 3 N 4 was formed, considering that the employed solidphase synthesis required only one reagent. The small content of impurities (1.92%) is derived from the sample holder. Apparently, the contact with atmospheric air during pyrolysis allowed a partial oxidation during the annealing. That would explain the presence of oxygen in the EDS analyses.
GrO ( Figure SM6) was still impregnated with a small amount of sulfur, chlorine, and manganese, even after the exhaustive puri cation. Nevertheless, as that residual contamination is small (2.21%), the material was considered suited for the next synthetic steps. Again, 1.19% of impurities came from the sample holder. Finally, looking at the optimized material (Fig. 3), it did not show impurities, except for 3.25% from the sample holder.
Aiming at estimating the rGO content in the synthesized photocatalyst, CHNSO analyses were performed. First, the g-C 3 N 4 precursor was analyzed and the nitrogen content was determined. Second, the composite was also analyzed and, using that previous nitrogen content, the amount of g-C 3 N 4 was calculated. By subtracting the g-C 3 N 4 mass from the composite one, the rGO mass was calculated. Therefore, the rGO content in the composite was estimated as approximately 4.0%.
Precursors and the optimized material were also characterized by their FTIR spectra (Fig. 4, Table 2). Bands were identi ed in accordance with the work of Aleksandrzak et al. (2017). Reduced graphene oxide was almost free from oxygen and it was mainly composed of carbon and hydrogen (small electronegativity difference). Therefore, the spectrum of the optimized material is similar to the one of g-C 3 N 4 , as the magnet dipoles of the chemical bonds are not high enough for coupling with the infrared waves emitted by the source. Reduced graphene oxide changes from bright yellow g-C 3 N 4 to a dark gray material, which explains the lower re ectance of the composite (Fig. 5a). In the DRS analysis, one can observe that the band gap of g-C 3 N 4 did not change signi cantly by the presence of rGO in the structure (Fig. 5b).
rGO is a conductor species (no signi cant charge separation) with a transition considered to be direct and allowed (r = 0.5). Therefore, oxidation reactions are promoted by g-C 3 N 4 , suggesting that the improved photocatalytic activity of the material may be due to increased speci c surface areas and low recombination rates -charge transfer to rGO (Malagutti et al., 2009).
Scavenging tests ( On the other hand, it is straightforward why • OH did not play any signi cant role in the degradation process. The oxidation potential of H 2 O into • OH is + 2.80 eV (Bauer and Fallmann, 1997), approximately 2 times higher than the VB potential (+ 1.55 eV). However, the material is able to oxidize H 2 O into O 2 , as the VB potential is greater than + 1.23 eV (Zhao et al., 2014). In summary, the BPA degradation mechanism was based on the superoxide radical chemistry, formed when the composite CB reduces dissolved molecular oxygen.
Two concurrent degradation mechanisms were proposed in Fig. 8, as the major oxidizing agent acting in the treatment was the superoxide radical. In path (1), superoxide radical abstracts a proton from BPA (−OH moiety) forming a hydroperoxyl radical. The deprotonated oxygen atom can form a double bond, allowing the hydroperoxyl radical to attack the carbon vicinal to the formed ketone. The added hydroperoxyl radical removes a proton from that carbon, triggering the elimination of water and forming a double bond with the remaining oxygen atom.
In path (2), superoxide radical adds a radicalar oxygen to BPA (vicinal carbon to the −OH moiety) and a hydroxyl ion is eliminated. The radicalar oxygen can abstract a proton from water forming hydroxyl radical in the process. Therefore, DP1 is formed when BPA reacts by paths (1)   the concentrations in which a single exposition to a chemical species, during a speci c period of time, promotes death or deleterious effects to 50% of a tested population. On the other hand, ChV is the chronic value, i.e., the concentration in which a continuous exposition promotes chronic effects (Gupta, 2018). According to the Globally Harmonized System of Classi cation and Labeling of Chemicals (GHS), the ratio (R) between the acute effect (LC 50 or EC 50 ) and As expected, BPA was very toxic or toxic to all of the test-organisms, both for acute and chronic exposures. One can see that DP2 was less toxic for all the organisms for both levels of exposure, being even not harmful for shes within acute level. DP1 was also less harmful than BPA for all the organisms in both levels. Therefore, the proposed treatment is effective for BPA removal because it not just can degrade the BPA molecule but also convert it to less toxic species.

Conclusions
The multivariate approach was helpful for determining the conditions in which materials with optimized photocatalytic properties could be synthesized.
The use of the sonochemical approach was useful for keeping the experimental conditions reproducible, what is re ected by the low experimental error of the experimental design. It was also helpful for obtaining a material with large speci c surface area.
The use of bisphenol A as model-pollutant at environmental concentrations helped in getting closer-to-real-life treatment conditions, although that makes analytical procedures more challenging.
Characterizing this kind of composite, in which precursors have similar structures and compositions, is a true challenge because it is hard to notice any differences in a variety of techniques (FTIR, XRD, SEM, etc.).
Although the analytical signals of rGO were hard to perceive, its presence in the composite could be inferred by the increased speci c surface area and the photocatalytic activity of the composite.
The photocatalytic BPA degradation exhibited a pseudo-rst-order kinetics, both for g-C 3 N 4 and the composite. The degradation rate, using the synthesized composite as the photocatalyst, was twice as fast as the one with g-C 3 N 4 .
The main oxidizing species generated by the composite was O 2 •− , according to the calculated band positions and scavenging tests.
The synthesized material was able to successfully degrade and partially detoxify BPA in a relatively short period of time.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. supplementarymaterialESRP.docx