High-Throughput Preparation of Uncontaminated Graphene-Oxide Aqueous Dispersions with Antioxidant Properties by Semi-Automated Diffusion Dialysis

A semi-automated diffusion-dialysis purification procedure is proposed for the preparation of uncontaminated graphene oxide (GO) aqueous dispersions. The purification process is integrated with analytical-signal processing to control the purification degree online by several channels: oxidation-reduction potential, conductivity, and absorbance. This approach reduces the amounts of reagents for chemical treatment during dialysis. The total transition metal (Mn and Ti) content was reduced to a sub-ppb level (assessed by slurry nebulization in inductively coupled plasma optical atomic emission spectroscopy). Purified aqueous GO samples possess good stability for about a year with a zeta-potential of ca. −40 mV and a lateral size of ca. sub-µm. Purified GO samples showed increased antioxidant properties (up to five times compared to initial samples according to chemiluminometry by superoxide-radical (O2−) generated in situ from xanthine and xanthine oxidase with the lucigenin probe) and significantly decreased peroxidase-like activity (assessed by the H2O2–L-012 system).


Introduction
Graphene oxide (GO) is a two-dimensional material with a unique atomic structure, tunable lateral size, high specific surface area, and variety of sp 3 carbon atoms with functional oxygen-containing groups [1]. The reduction of GO to reduced graphene oxide (rGO) gives new perspectives for tunable preparation of graphene-resemblance structures [2]. The unique properties of both GO and rGO pave the way for a wide range of applications in technology, biomedicine, and environmental sciences [3,4]. Robust membranes based on GO or rGO can be prepared for desalination, separation of gaseous mixtures, and water purification [5]. In biomedical applications, GO, especially aqueous GO dispersions (aqGO) can be used as a tunable material for wearable artificial kidneys [6,7] or thin-film nanofiltration purification systems [8]. GO and rGO have a potential in diagnosis and treatment of cancer [9][10][11], drug delivery [12], in tissue regeneration and immunomodulation [13,14]. Studies of nanozyme properties of GO and rGO are being actively performed [15,16].
Many applications require pure GO-based materials, especially for biomedical and other practical uses, e.g., heat-transfer tasks or optical materials; thus, purification steps are on the cutting edge of many applications of GO and rGO [17]. E.g., graphene nanoplatelets bridge between two semiconductors (AgBr and graphitic carbon nitride) boost photoelectrochemical performance [18]. Effective photocatalysts generated the superoxide radical (O 2 − )

Enzymes, Substrates, and Chemiluminescent Probes
Catalase from bovine liver (Cat), lyophilized powder (2000-5000 units/mg protein) and peroxidase (Px) from horse radish, 10,000 U/vial with working concentrations of 7 mM and 200 nM, respectively, were prepared by dissolving in a PBS with pH 7.4 (Sigma-Aldrich, St. Louis, MO, USA). A xanthine oxidase (XO) working solution (Sigma-Aldrich, St. Louis, MO, USA) with an activity of 0.11 U/mL was obtained by diluting the PBS suspension (pH 7.4). Before chemiluminescence measurements, the stock enzyme solution was kept for 15 min at laboratory temperature. The catalytic activity of the one enzyme unit converts 1.0 µmol of xanthine to uric acid in 1 min at 25 • C and pH 7.5.
A xanthine (Xa), 3,7-dihydropurine-2,6-dione (Sigma-Aldrich, St. Louis, MO, USA) aqueous solution was prepared. An off-the-shelf ultrasound probe with a MEF93.T timer at continuous operation and 0.6 kW electrical-power mode was used for GO aqueous solution preparation (LLC MELFIZ-ul'trazvuk, Moscow, Russia). Parameters of the ultrasonic device and the operating mode are described elsewhere [46]. A Thermo Fisher Scientific FBH 612 low-temperature circulation thermostat (Thermo Fisher, Waltham, MA, USA) was used for increased ultrasound efficiency during aqGO preparation.

Certified Reference Materials (CRMs)
Elemental analysis. ICP-AM-6-A ICP Analytical Mixture with 23 components solution for instrument calibration solutions and elemental analysis on ICP-OES (High Purity Standards, North Charleston, SC, USA) was used throughout.

Dialysis Setup Equipment and Operation
We used off-the-shelf cellulose membrane tubing with a molecular weight cutoff (MWCO) at 3.5 and 14 kDa Spectra/Por™ 3 Spectrum Labs (Thermo Scientific, Waltham, MA, USA). Pre-conditioning of freshly packed tubes was performed by procedure 2 below to wash out glycerol, sulfide, and trace transition metals.
Dialysis setup was equipped with (1) a magnetic stirrer with a hot plate (ISO Lab, Isolab Laborgeräte GmbH, Eschau, Germany), (2) a 4-channel auto dosing peristaltic pump, Jebao Doser 3.4 WiFi programmable (Zhongshan Jiebao Electronic Appliance Co., Ltd., Zhongshan, China), (3) a washing solution channel compartment including three 10 L polypropylene tanks with aqueous solutions of 0.005 M EDTA, 3 wt.% H 2 O 2 , ultrapure water, and drain (see the Supplementary Materials, Figure S1). This setup requires maintenance every 500 h of peristaltic pump operation: replacement of feed tubes and the lubricate pump shaft to sustain reproducible dosing results.

Other Instrumentation
A SevenCompact™ pH/Ion S220 pH-meter (Mettler-Toledo AG, Greifensee, Switzerland) was used to prepare the phosphate buffer solution. An axial view Agilent 720 ICP-OES spectrometer (Agilent, Mulgrav, Australia) was used for elemental analysis. The ICP-AES operating conditions are given elsewhere [46]. Atomic emission lines are chosen based on ISO 11885:2007 recommendations.

Data Treatment
The measurement results are presented as mean values with confidence intervals (p < 0.05 or p = 0.95) under the requirements for the competence of testing and calibration laboratories ISO 17025:2020. Calibration parameters were calculated according to the IUPAC recommendations to present the results of the chemical analysis [51]. All statistical analysis and plot illustration were conducted by Origin 2017 SR1 b9.2.257 (OriginLab Corporation, Northampton, MA, USA).

Procedure 1. Preparation of GO aqueous dispersions
A weighted portion of GO (or rGO) samples ca. 50 ÷ 150 mg was placed into a 100 mL Erlenmeyer flask. A 75-mL portion of pure water was put into a flask, then ultrasound probe working in a continuous mode was immersed. Then, the flask was placed into a thermostat at 25 • C. The ultrasound graphene oxide exfoliation was conducted four times for 30 min with 15 min breaks between cycles. Then, ultrasound-exfoliated graphene oxide was placed into a 100.0 mL volumetric flask, and water was added to the mark.

Procedure 2. Pre-conditioning membrane (dialysis bag) treatment for dialytic purification of graphene oxide
The necessary length of the membrane (dialysis bag) for the treatment according to the manufacturer recommendations about size vs. volume was cut. A portion of 150 mL of 5 mM EDTA put into a conical (Erlenmeyer) flask. The membrane was placed into the flask and the surface was thoroughly wetted. The flask was heated to 60 • C under stirring and left for three days. The chemicals were exchanged three times every 24 h. After the preconditioning, the membrane was flushed using 200 mL of ultrapure water.

Procedure 3. Semi-automatic dialysis purification of graphite oxide aqueous dispersions using EDTA and hydrogen peroxide
A portion of 10 mL of aqGO was placed into a pre-conditioned membrane (procedure 2), covered with a polypropylene clamp, and put into a 1 L chemical beaker of ca. 800 mL of deionized water. Automatic supplies of reagents, washing liquid, and drainage were connected to the system using three channels of the peristaltic pump (see the Supplementary Materials, Figure S1). Automated cyclic feeding reagents of 100 mL per cycle were performed in a programmable mode every 30 min (see the Supplementary Materials, Table S1). During the first four days, the dialysis was performed in a reagent-supply mode in the following channels: (1) 0.05 M EDTA, (2) 3% w/v hydrogen peroxide, and (3) drain. During dialysis, the subsequent analytical signals were monitored: (1) electric conductivity (µS/cm), ORP (mV), and absorption spectrum (a.u.) of the washing liquid in the range 190-1100 nm. The setup schematic is shown in Figure 1. After dialysis, the sample was transferred into Nanomaterials 2022, 12, 4159 6 of 27 a brown borosilicate-glass vial, and the dialysis purification efficiency was evaluated using ICP-AES, TOC, and FTIR.
Supplementary Materials, Figure S1). Automated cyclic feeding reagents of 100 mL per cycle were performed in a programmable mode every 30 min (see the Supplementary Materials, Table S1). During the first four days, the dialysis was performed in a reagent-supply mode in the following channels: (1) 0.05 M EDTA, (2) 3% w/v hydrogen peroxide, and (3) drain. During dialysis, the subsequent analytical signals were monitored: (1) electric conductivity (µS/cm), ORP (mV), and absorption spectrum (a.u.) of the washing liquid in the range 190-1100 nm. The setup schematic is shown in Figure 1. After dialysis, the sample was transferred into a brown borosilicate-glass vial, and the dialysis purification efficiency was evaluated using ICP-AES, TOC, and FTIR.

Procedure 4. Chemiluminometric assay of superoxide dismutase-like activity
Superoxide anion-radical (SAR) scavenging potential of aqGO was assessed by chemiluminometry with the xanthine and xanthine oxidase (Xa/XO) model system in the presence of the lucigenin probe as a selective enhancer for SAR, which we presented for fullerenes [52]. Briefly, aliquots of xanthine (20 µL, 20 µM), lucigenin (20 µL, 1 µM), and the analyzed GO samples were added to a cuvette with PBS (100 mM, pH 7.4). The background signals of luminescence were recorded for 30 ÷ 60 s, then an aliquot of xanthine oxidase (40 µL; activity, 4.4 mU/mL) was added. The total volume of the system was 1.00 mL. All samples were analyzed in triplicate at 37 °C, and signal registration was processed up to 15 ÷ 20 min. As the analytical signal, the ratio of areas under the curve to for the test and blank experiments (without GO) was used.

Procedure 5. Chemiluminometric assay of peroxidase-and catalase-like activities
The peroxidase-or catalase-like activity of the analyzed GO samples was studied in the luminol (or a chemiluminescence probe L-012) and H2O2 system. A solution of luminol or L-012 (40 µL, 1 mM) was prepared by dissolving a sample in PBS (pH 7.4 and 8.6). A working solution of hydrogen peroxide with a concentration of 0.1 mM was prepared by diluting the 37 w/v% H2O2 stock solution with ultrapure water. Background signals of luminescence were recorded for 30 ÷ 60 s, then an aliquot of H2O2 (10 µL, 0.1 mM) was added, then, in another 30 s, an aliquot of peroxidase (20 µL, 20 nM) or catalase (10 µL, 7 µM), finally, in another 60 ÷ 120 s, an aliquot of GO (10 µL, ca. 1 g/L). The total volume of the system was 1.00 mL. All samples were analyzed in triplicate at 37 °C, and signal

Procedure 4. Chemiluminometric assay of superoxide dismutase-like activity
Superoxide anion-radical (SAR) scavenging potential of aqGO was assessed by chemiluminometry with the xanthine and xanthine oxidase (Xa/XO) model system in the presence of the lucigenin probe as a selective enhancer for SAR, which we presented for fullerenes [52]. Briefly, aliquots of xanthine (20 µL, 20 µM), lucigenin (20 µL, 1 µM), and the analyzed GO samples were added to a cuvette with PBS (100 mM, pH 7.4). The background signals of luminescence were recorded for 30 ÷ 60 s, then an aliquot of xanthine oxidase (40 µL; activity, 4.4 mU/mL) was added. The total volume of the system was 1.00 mL. All samples were analyzed in triplicate at 37 • C, and signal registration was processed up to 15 ÷ 20 min. As the analytical signal, the ratio of areas under the curve to for the test and blank experiments (without GO) was used.

Procedure 5. Chemiluminometric assay of peroxidase-and catalase-like activities
The peroxidase-or catalase-like activity of the analyzed GO samples was studied in the luminol (or a chemiluminescence probe L-012) and H 2 O 2 system. A solution of luminol or L-012 (40 µL, 1 mM) was prepared by dissolving a sample in PBS (pH 7.4 and 8.6). A working solution of hydrogen peroxide with a concentration of 0.1 mM was prepared by diluting the 37 w/v % H 2 O 2 stock solution with ultrapure water. Background signals of luminescence were recorded for 30 ÷ 60 s, then an aliquot of H 2 O 2 (10 µL, 0.1 mM) was added, then, in another 30 s, an aliquot of peroxidase (20 µL, 20 nM) or catalase (10 µL, 7 µM), finally, in another 60 ÷ 120 s, an aliquot of GO (10 µL, ca. 1 g/L). The total volume of the system was 1.00 mL. All samples were analyzed in triplicate at 37 • C, and signal registration was up to 15 ÷ 20 min. As the analytical signal, the ratio of areas under the curve to for the test and blank experiments (without GO) was used.

Dialysis Dispersion Purification and Monitoring
GO aqueous dispersions were prepared by Procedure 1 using ultrasound-assisted exfoliation [53] and GO powder. The total time of all the stages of ultrasound preparation of Nanomaterials 2022, 12, 4159 7 of 27 dispersions is 3.5 h. We used a thermostat to sustain no attenuation of ultrasound power in solution [54], increase the reproducibility of the preparation yield, and reduce evaporation. Due to light irradiation effects on the decomposition of graphene oxide [55], the storage was in the dark.
After preparation, GO dispersion was purified. The purification strategy can be summarized as: (i) the removal of any contaminants by selecting a ready-to-use target reagent for a two-step treatment based on preliminary sample characterization and impurity affinities; and (ii) analytical control and monitoring: handling and controlling the reagent supply; monitoring specific parameters to help spotting a purification problem at any stage; and the precise quality control of the final product. Dialysis purification includes several steps ( Figure 1): (1) dialysis bag preconditioning (procedure 2); (2) dialysis purification with multiparametric monitoring and end-point dialysis estimation (procedure 3); and (3) pure aqGO characterization.
During the dialysis stage, the feeding of reagents for purification was carried out in an automatic mode. Reagent containers were pre-soaked with target reagents for 30 days (conditions are approximately the same as [56]) to remove plasticizers and unrelated impurities. The results of plasticizer quantities (phthalates) after 60-day storage are presented, which do not exceed 0.05 ppm in some cases (Supplementary Materials, Section S1.4, Figures S7-S10, Table S4). The following analytical signals from probes were also monitored during the process: conductivity ( Figure 2), ORP (Figure 3), and absorbance spectra ( Figure 4). The quality control procedure depends on the type of contaminants, transition metals used as oxidation products, and extra exfoliation procedure.
in solution [54], increase the reproducibility of the preparation yield, and reduce evaporation. Due to light irradiation effects on the decomposition of graphene oxide [55], the storage was in the dark.
After preparation, GO dispersion was purified. The purification strategy can be summarized as: (i) the removal of any contaminants by selecting a ready-to-use target reagent for a two-step treatment based on preliminary sample characterization and impurity affinities; and (ii) analytical control and monitoring: handling and controlling the reagent supply; monitoring specific parameters to help spotting a purification problem at any stage; and the precise quality control of the final product. Dialysis purification includes several steps ( Figure 1): (1) dialysis bag preconditioning (procedure 2); (2) dialysis purification with multiparametric monitoring and end-point dialysis estimation (procedure 3); and (3) pure aqGO characterization.
During the dialysis stage, the feeding of reagents for purification was carried out in an automatic mode. Reagent containers were pre-soaked with target reagents for 30 days (conditions are approximately the same as [56]) to remove plasticizers and unrelated impurities. The results of plasticizer quantities (phthalates) after 60-day storage are presented, which do not exceed 0.05 ppm in some cases (Supplementary Materials, Section S1.4, Figures S7-S10, Table S4). The following analytical signals from probes were also monitored during the process: conductivity ( Figure 2), ORP (Figure 3), and absorbance spectra ( Figure 4). The quality control procedure depends on the type of contaminants, transition metals used as oxidation products, and extra exfoliation procedure.
The optimum conditions to prepare highly pure GO dispersions are summed up in Table 1. Using a 0.05 M EDTA and a 3% w/v hydrogen peroxide as reagents, followed by washing with distilled water for 3 days, allows purifying GO samples from manganese and aluminum thoroughly and reduces the titanium (by 90%) and iron (by 80%) to the ppb level. Using membranes with a 14 kDa molecular weight cutoff (MWCO) is optimal.  The optimum conditions to prepare highly pure GO dispersions are summed up in Table 1. Using a 0.05 M EDTA and a 3% w/v hydrogen peroxide as reagents, followed by washing with distilled water for 3 days, allows purifying GO samples from manganese and aluminum thoroughly and reduces the titanium (by 90%) and iron (by 80%) to the ppb level. Using membranes with a 14 kDa molecular weight cutoff (MWCO) is optimal. value of conductivity when filling the purification reagents (non-electrolyte H2O2 decrease of κ and electrolyte EDTA increase of κ) between 2.5 and 3 days.
(a) (b)     More specific comments are presented in Table 1. Excess purification reagents (EDTA and H 2 O 2 ) were removed by dialysis by an ultrapure water supply after 3 days following the reagent-supply stage. Below we describe each step in more detail.

Monitoring Metal Contents in Membrane during Preconditioning
Due to the high content of metals [57] and sulfides [58], there is a need of preliminary dialysis bag preconditioning (procedure 2). Preconditioning dialysis membranes with a 0.05 M EDTA solution reduced the degree of contamination of GO dispersions with transition metals. An untreated membrane increases copper and zinc contents up to 400%.

Conductivity Monitoring
Electrical conductivity (EC) was used as the primary channel, and the dialysis was stopped when this parameter reached the threshold of 10 µS/cm (Figure 2a). EC measurement in dialysate solution shows a behavior similar to ORP, see below. The beginning of signal registration started at values of less than 5 µS/cm. The overall profile is shown in Figure 2a, there was a gradual increase in EC to the steady state. Adding uncharged hydrogen peroxide lowers EC (dilution), charged EDTA increases the EC value (Figure 2b).

Oxidation-Reduction Potential Monitoring
ORP has a sharp increase at the beginning of reagent feeding, then, for one day, it dropped to the steady-state value (Figure 3a). The increase is associated with launching catalytic decomposition of hydrogen peroxide by manganese dioxide (which is a component of crude graphene oxide). Long return to steady-state values is most likely due to adsorption of oxidation products on the platinum electrode.
Between one and four days, it tends to have a gradual increase. A detailed examination of the signal shows up-and-down behavior due to the supply of purification reagents (Figure 3b). Adding uncharged hydrogen peroxide lowers ORP by dilution; adding a charged EDTA, on the contrary, increases the potential as EDTA increased the solution ionic strength. There was a rapid decrease in the final washing-out phase. As well, a slight increase in ORP after adding each portion of water is observed (Figure 3a). On the fifth day, the potential remained stable, indirectly indicating the completion of the main purification stage (Figure 3a). After five days, the relative standard deviation of ORP measurements increases (the signal-to-noise ratio decreases) due to the insufficient ionic strength of the solution [59]. A similar behavior was observed for a 14-day experiment.

UV/vis Absorbance Spectra with Fiber Optical-Probe Monitoring
A contour map of absorption spectra during the purification is shown in Figure 4 (ca. 1000 spectrums). First, one can see an increase in absorption due to EDTA feeding. The absorption in the range of 190-300 nm remains unchanged owing to a constant concentration of this substrate. It is also worth noting that there is a minor contribution from graphite oxide, which also passes through the dialysis membrane. After adding water for washing, the absorption decreases. In the end of the process, the significant absorption is observed only in the range of 190-220 nm. As well, 254 nm has been selected for carbonaceous content release during dialysis (see the Supplementary Materials, Figure S2A).

Total Yield
The total organic carbon (TOC) content estimation after preparation of aqGO (Procedure 1) made it possible to check that the share of the non-dispersed and unstable GO fraction was less than 5%. Due to the high lateral size of GO, filtering through 0.45 µm removed almost all its fraction ( Table 2). Total losses of the carbon fraction were up to 70%. A low TOC loss is shown for the 3.5 kDa membrane to be approx. 25%. Reusing a 3.5 kDa membrane decreased TOC up to 2.5 times. For 14 kDa membranes, the maximum TOC loss was approx. 50%. Reusing a 14 kDa membrane decreased TOC up to 1.05 times. On average, the losses for a 14 kDa membrane were 1.6 times higher than for a 3.5 kDa membrane. In terms of dialysis bag stability, we evaluated the change and shift in IR absorption spectra by ATR-FTIR, which coincided with all presented bands before and after the dialysis in the range 4000-100 cm −1 ( Figure S3A,B in the Supplementary Materials). No additional bands are observed in FTIR spectra, not even graphene oxide bands, which were visually retained in the membrane.

Sample Stability Lateral Size and Zeta-Potential
After purification, we estimated the sedimentation stability by UV/vis spectra and zeta-potential measurements ( Table 2). The wavelength of 254 nm was selected as characteristic for the GO samples in analogy with dissolved organic carbon [60]. In all cases, absorbance decreased to 10% for 8 months, showing excellent stability. In all cases, the absorbance of purified samples at 254 nm decreased no more than 10% for 8 months, showing excellent stability after purification.

pH and Conductivity
The acidity of GO samples increases after purification, reflecting a decrease in sulfate from sulfuric acid used for the graphite exfoliation process ( Table 2). Measurements of anions conducted by ion chromatography after 5 min centrifugation at 4 krpm showed that, with both 3 and 14 kDa membranes, sulfate and nitrate contents decreased up to 97%. These data correlate with conductivity measurements (decreased on average to 9 times).

UV/Vis Absorption Spectra
All absorption bands registered in the UV/vis region (Supplementary Materials, Figure S2B) coincide with the previously known for the UV/vis region [61]. The plasmon peak at 210 nm belongs to π→π* of C=C bonds and 310 nm, to n→π* transitions due to organic peroxide (R-O-O-R) or epoxide (C-O-C) moieties. The weak bands at 600 and 730 nm correspond to GO in solution [61]. After dialysis, there is no band shifting; however, increasing absorbance in normalized absorption spectra is revealed. Figure 5 shows attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR) of purified and non-purified GO dispersions. The ATR-FTIR spectrum (Figure 5a) after dialysis demonstrates a series of bands. Band assignments for GO samples are shown in Figure 5b. We distinguish four characteristic ranges: 4000 ÷ 2000, 2000 ÷ 1500, the fingerprint region 1500 ÷ 900, and below 900 cm −1 .

Sample Stability Lateral Size and Zeta-Potential
After purification, we estimated the sedimentation stability by UV/vis spectra and zeta-potential measurements ( Table 2). The wavelength of 254 nm was selected as characteristic for the GO samples in analogy with dissolved organic carbon [60]. In all cases, absorbance decreased to 10% for 8 months, showing excellent stability. In all cases, the absorbance of purified samples at 254 nm decreased no more than 10% for 8 months, showing excellent stability after purification.

pH and Conductivity
The acidity of GO samples increases after purification, reflecting a decrease in sulfate from sulfuric acid used for the graphite exfoliation process ( Table 2). Measurements of anions conducted by ion chromatography after 5 min centrifugation at 4 krpm showed that, with both 3 and 14 kDa membranes, sulfate and nitrate contents decreased up to 97%. These data correlate with conductivity measurements (decreased on average to 9 times).

UV/Vis Absorption Spectra
All absorption bands registered in the UV/vis region (Supplementary Materials, Figure S2B) coincide with the previously known for the UV/vis region [61]. The plasmon peak at 210 nm belongs to π→π* of C=C bonds and 310 nm, to n→π* transitions due to organic peroxide (R-O-O-R) or epoxide (C-O-C) moieties. The weak bands at 600 and 730 nm correspond to GO in solution [61]. After dialysis, there is no band shifting; however, increasing absorbance in normalized absorption spectra is revealed.  The band of ν C-O-H appears as a fine structure after dialysis at 1384 cm −1 . Shifting to higher frequencies (blueshift) for bands of ν C=C 1621→1611, ∆ν = 10 cm −1 , and lower frequencies (redshift) ν C=O 1730→1728, ∆ν = 2 cm −1 , is registered.

Chemiluminescence Assays
For chemiluminescence measurements, we chose the most purified GO dispersions, which are produced by the dialysis with a 14 kDa MWCO membrane.

SOD-like Activity Using Lucigenin/Xanthine/Xanthine Oxidase System
To assess superoxide dismutase-like (SOD-like) activity, the lucigenin/xanthine/xanthine oxidase system was used (Procedure 4). Xanthine oxidase catalyzes the oxidation of xanthine with the formation of superoxide anion. Lucigenin is a highly sensitive and specific probe for superoxide anion. Therefore, measurements were conducted as lucigenindependent CL for superoxide anion radical quantification ( Figure 6).

Chemiluminescence Assays
For chemiluminescence measurements, we chose the most purified GO dispersions, which are produced by the dialysis with a 14 kDa MWCO membrane.

SOD-like Activity Using Lucigenin/Xanthine/Xanthine Oxidase System
To assess superoxide dismutase-like (SOD-like) activity, the lucigenin/xanthine/xanthine oxidase system was used (Procedure 4). Xanthine oxidase catalyzes the oxidation of xanthine with the formation of superoxide anion. Lucigenin is a highly sensitive and specific probe for superoxide anion. Therefore, measurements were conducted as lucigenin-dependent CL for superoxide anion radical quantification ( Figure  6). Before dialysis, GO samples with low concentrations (5 and 12.5 mg/L) caused luminescence suppression; a higher concentration (25 mg/L) increased chemiluminescence (Figure 6a). After purification, GO in all concentrations had approximately equally reliable suppressed CL signals (Figure 6b). The shape of kinetic Before dialysis, GO samples with low concentrations (5 and 12.5 mg/L) caused luminescence suppression; a higher concentration (25 mg/L) increased chemiluminescence (Figure 6a). After purification, GO in all concentrations had approximately equally reliable suppressed CL signals (Figure 6b). The shape of kinetic curves is similar to chemiluminograms for superoxide dismutase (Figure 6c). The area under the curve reflects the number of free radicals; thus, it is a parameter for constructing a calibration plot (Figure 6d). A dose-dependent decrease in luminescence intensity was observed (Figure 6d). From the calibration, 1 mg/L high-purity GO exhibits a superoxide-intercepting capacity equivalent to 2.00 ± 0.01 nM SOD activity units. Unpurified graphene at 25-50 mg/L did not show antioxidant properties, presumably due to impurities. The absence of a concentration dependence is most likely due to a low sensitivity coefficient or the activity saturation. Thus, GO exhibits antioxidant properties with respect to the superoxide radical. The mechanism of this action should be investigated further.

Catalase/Peroxidase Activity
In relation to hydrogen peroxide, nanozymes can exhibit either prooxidant (peroxidase) or antioxidant (catalase) activity. The study of peroxidase-like (Px-like) or catalase-like (Cat-like) activity (Procedure 5) was conducted with the L-012/H 2 O 2 model (Figure 7). curves is similar to chemiluminograms for superoxide dismutase (Figure 6c). The area under the curve reflects the number of free radicals; thus, it is a parameter for constructing a calibration plot (Figure 6d). A dose-dependent decrease in luminescence intensity was observed (Figure 6d). From the calibration, 1 mg/L high-purity GO exhibits a superoxideintercepting capacity equivalent to 2.00 ± 0.01 nM SOD activity units. Unpurified graphene at 25-50 mg/L did not show antioxidant properties, presumably due to impurities. The absence of a concentration dependence is most likely due to a low sensitivity coefficient or the activity saturation. Thus, GO exhibits antioxidant properties with respect to the superoxide radical. The mechanism of this action should be investigated further.

Catalase/Peroxidase Activity
In relation to hydrogen peroxide, nanozymes can exhibit either prooxidant (peroxidase) or antioxidant (catalase) activity. The study of peroxidase-like (Px-like) or catalase-like (Cat-like) activity (Procedure 5) was conducted with the L-012/H2O2 model (Figure 7).  The sensitivity of the L-012 probe is 45-fold higher than that of the luminol model, which is frequently used in CL measurements ( Figure S4A, Luminol, and Figure S4B, L-012 probes data, Supplementary Materials). Thus, L-012 was selected as a probe for low concentrations.
Concentration dependence of the CL signal and kinetic behavior of GO in the reaction with H 2 O 2 is shown in Figure 7a,b. For two concentrations (1 and 2.5 mg/L), similar results were obtained. Before dialysis, GO caused an increase in CL by 5-7 times, thus exhibiting peroxidase-like properties. On the contrary, after dialysis purification, GO suppressed chemiluminescence, thus exhibiting catalase-like properties. It is highly probable that "peroxidase-like" properties of unpurified GO are due to the presence of impurities, particularly metals.
This conclusion can be confirmed by experiments in the presence of native horseradish peroxidase and catalase. Purified graphene oxide sharply (by a factor of five in amplitude) reduces peroxidase chemiluminescence (Figure 7c). Similarly, in the presence of high-purity graphene oxide, the chemiluminescence intensity in the H 2 O 2 + catalase + L-012 system is almost twofold lower (Figure 7d). Presumably, high-purity graphene oxide quickly neutralizes hydrogen peroxide. A direct effect of GO on enzymes is not excluded, but it is weak, as follows from the analysis of the effect of unpurified graphene.
Thus, the oxidative properties of purified and unpurified GO are essentially different. Unlike commercial graphene oxide, dialysis-purified GO exhibits antioxidant properties to the superoxide anion radical and hydrogen peroxide.

Discussion
To develop the purification procedure, we used as-prepared GO samples, most common in practice (Procedure 3, Table 1). We aimed to create a setup and developed the procedure to purify aqGO from various contaminants that may be present in various quantities depending on the GO preparation technique [21,63]. Primary contaminants (their quantity depends on preparation techniques) are reagents in Hummers' method utilizing KMnO 4 , NaNO 3 , and H 2 SO 4 [64]. Secondary contaminants (their quantity depends on subsequent exfoliating techniques) are substances appearing in ultrasound-assisted exfoliating, which brings forth TiO 2 as we previously showed for aqueous fullerene dispersions of C 60 , C 70 , and Gd@C 82 by ultrasound probe operating in a continuous mode [52]. The third type are carbonaceous debris continuously formed during GO synthesis as a byproduct of the main reaction [29]. This requires the development of tools and methodologies for (i) online signals processing to control all the purification stages; (ii) comprehending the endpoint of the dialysis process; (iii) the assessment of purified samples; and (iv) an estimation of biological relevance and applicability of purified GO samples.

Classification of Metal Impurities and Their Removal Mechanism
Evaluation should be done for biomedical applications for in vitro and in vivo purity levels. The most appropriate recommendations with specific minimum requirements for GO dispersions purity are proposed in ISO 23500-3:2019 for water to be used in hemodialysis and related therapies. Results of the inorganic impurity estimation by ICP-AES and IC (Table 1) indicate that we achieved the content of elements below the recommendations in the standard. It is worth noting that Zn (<0.1 ppm) and Al (<0.01 ppm) as well as nitrate (<2 ppm) and sulfate (<100 ppm) contents are significantly reduced during the purification process. In addition, it should be noted that such elements as Fe, Mn, and Ti are not standardized in their contents, although we achieved their reduction to a sub-ppb level.
Ion chromatography shows that sulfate and nitrate have been released in the first two days of the process, which agrees well with the existing data [34], but in case of purifying of carbon dots it should be approx. 15 days [65] due to predominantly random and chaotic carbonization processing [66]. Sulfate in the GO structure can generate an epoxide bridge and a free SO 3 2− radical [26], which retains sulfate in GO samples. As for nitrate, there is no specific role of this anion in the process of GO formation that allowed it to be released completely during the purification.
Due to a high sorption capacity of GO [67,68] and specific metal binding [69], it is unfeasible to conduct the purification by water only. It is necessary to use special reagents; in this work, EDTA was selected as the most used chelation agent and H 2 O 2 as an oxidation agent. EDTA can bind metals at six points (at two N and four O atoms) and form an elaborate complex with five-membered rings [70].
The regulation of Fe, Mn, and Ti removing deals with chemical transformation to appropriate ions, which are readily removed from GO sheets. We suggest that interlayer ions chelated with GO surface moieties, mainly carboxyl and hydroxyl functional groups, would provide competitive binding with excess EDTA and then released due to diffusion [71]. It is well-known that GO coordinate Mn 2+ preventing its hydrolysis [72]. It is impossible to unequivocally answer which form of the element is associated with EDTA; the speciation analysis [73] was beyond the scope of this work. A presumable reaction pathway for ions release bonded with GO surface of free ions is presented below.
As for a GO binding site, the presence of carboxyl groups response to metal sorption activity [68,79] and vic-diol transformed to enols can bind divalent metal ions [72]. Presumably, the substantial factor for the high efficiency of EDTA for GO purification is its excess and much more sorption and binding capacities in comparison with GO. Compared with manganese, the EDTA quantity was at a 400-fold excess. As well, for highly oxidized samples, 1 g GO contains 5 to 8 mmol of acidic groups [72], which correlates to the found sub-mmol/g quantity (1.4 ± 0.2 mmol/g), see the Supplementary Materials, Section S1.1 ( Figure S5); it is much lower than the EDTA concentration (up to 500 times). Therefore, the thermodynamic equilibrium shifts towards a more stable EDTA complex.
In terms of pore size, 3.5 and 14 kDa membranes correspond to diameters of 2 and 3 nm for proteins, respectively [80]. Regarding graphene oxide flat structure and metal removal efficiency, a 14 kDa membrane provided much better results than a 3.5 kDa membrane (bag). The oxidative debris and metal throughput for a 14 kDa membrane are higher, with the main 1 µm fraction remaining in the solution. Although some approaches have been developed to determine the pore size distribution of nanofiltration membranes [81] for graphene oxide, such techniques have not yet been validated.

Purification Setup Requirements
We propose the following requirements for the setup to carry off the semi-automated purification process. A setup should include the four basic units (Figure 1).

Membrane (Bag)
Currently, dialysis bags [25] or centrifugation-filtration [17] are used to purify GO. A pre-treated membrane (Procedure 2) was used for reducing contaminant contents to prevent their introduction to dialyzed GO samples. According to the manufacturer data, membranes may contain up to 15 ppm of Pb, Zn, and Fe, less than 0.1 ppm of Cd, Cr, Cu, Ni, glycerol to prevent brittleness, and sulfides. EDTA is most appropriate due to the best stability constants for transition metals [82], and H 2 O 2 is the best reagent for slight oxidation of sulfides, which can be easily removed as free sulfate. Experiments without a pre-treated protocol showed increasing zinc content up to 400% in GO after the dialysis.

Peristaltic Pump for Reproducible Reagent Dosing
Most dialytic purification treatments and approaches use changing water (or reagent solutions) manually and it spreads out the process up to 30 days [25,84]. If the process is unchecked, the endpoint is undetermined. Therefore, we used an automatic reagent supplier using a multichannel peristaltic pump programmable via a Wi-Fi protocol. The pump is needed to deliver the reagents evenly and to regulate the amounts within certain limits. Reducing both quantities and fluxes of reagents are in demand for long-term sustainability of the dialysis process [85]. On average, it takes 25 ÷ 30 L of liquid (pure water and reagents) to purify (approx. 100 mL, up to 0.2 g scale) a GO dispersion in the semi-automation mode for 7 days. The limiting factor in purification and reducing the reagent amounts and their efficiency is the surface area of the dialysis bag [86], which cannot be infinite. The evaluation of the fluid fluxes shows that the process automation ( f auto regents , L per h) reduces the amounts of reagents for purification by up to 25% compared to manual dialysate replacement ( f manual regents , L per h), which was not reported previously for GO. Using Float-A-Lyzer™ [87] or Slide-A-Lyzer™ Dialysis Cassettes, e.g., the total area can be increased, but it has not been applied for GO purification yet. Although it is worth noting recent works on the closest topic of carbon dots, where similar membranes are successfully used [65]. As well, a crossflow membrane akin Vivaflow ® 50 Laboratory Cross Flow Cassette can increase the purification efficiency but still has not been widely used for GO purification due to the complexity and high operating expense of this process.
Another advantage of gradient-supplying reagents is prevented activation of GO coagulation-flocculation-sedimentation cascades due to the increased ionic strength. Coagulation processes were observed during reagent purification of as-prepared GO samples when reagents were added at once (like HCl, EDTA, etc.) due to disruption of electrostatic interactions and steric repulsion [88].

Real-Time Online Signal Recording
Rapid separation of low molecular weight impurities from GO dispersions requires continuous signal processing of the dialysate. Continuous signal processing would simplify the purification and make it more cost-effective by saving human resources, but not environmentally friendly due to electricity and other resource consumption. Automation of GO purification is essential protocol in high demand on an industrial scale. Attempts have been reported for isolation, purification, and characterization of GO [34] and humic substances [89]; however, not enough methods and devices were used to control the whole process. Attention should be paid to comprehensive automation processes and data collection [90] to reduce: (i) labor-intensive operations, (ii) the duration of the entire process, even while maintaining the number of reagents consumed, (iii) accuracy, and (iv) reproducibility. We suggest a continuous signal registration approach. Previously, merely conductometry monitoring [91] or simultaneous conductivity and pH [27] were used for total online control of dialysis purification in scaling processes [34]. EC is the best option for signal registration because the primary contaminants of GO aqueous dispersions are charged particles. The cheapest option is immersion probes, but their significant disadvantage is the impossibility of continuous operation. The registration of electric conductivity during purification is the most affordable and most accurate option providing the degree of leaching of contaminants (cationic oxidants) and reagents used in dialysis, such as EDTA. However, we propose to use extra ORP and absorbance registration to assess dialysis purification of GO, which makes the process much more efficient. The proposed signal registration provides insight into not only the charged particles (inorganic ions or surface moieties), but also the ab-sorbing particles released during purification, either graphite debris or any other. ORP online registration is a relatively cheap alternative to conductivity measurements. The use of UV/vis probes expands the interpretation of the dialysis process, and this approach has not been used previously for GO purification.

Pros and Contras of Dialysis-Monitoring Methods
UV/vis probes and ORP have not previously been shown to be capable of monitoring dialysis purification of aqueous dispersions. To sum up, these two probes can be considered universal with high accuracy and rapid responses. The practical implementation of these probes provides the maximum economic effect and suitability. To prove that the purification process is over, the analysis of the dialysate for 7-day and 14-day processes shows that all significant contaminants are washed out within the first 48 h of the purification process. The steady-state values of EC, ORP, or absorbance should be achieved within 6 to 12 h; usually, they will be reached on day 3 or day 4 of the process. After that, it is necessary to characterize the prepared sample with physicochemical methods.

ORP Probe
Advantages of ORP monitoring are (i) a wide working temperature range for conducting dialysis between −10 and 100 • C due to the working temperature of Ag|AgCl reference electrode; (ii) the stability of the signal, RSD is lower than 1%; and (iii) visual control and adjustment of all the purification reagent feed fluxes. Adding uncharged hydrogen peroxide lowers ORP by dilution, and adding a charged EDTA, on the contrary, increases the potential due to fluctuation of ionic strength (µ) of the solution, that changes the potential values [92,93].
The main issue in the recording of ORP is the instability (potential difference) for several reasons. The first factor is the noise at low ionic strengths, which correspond to relative conductivity <10 µS/cm [94]. The next is the parasitic leakage current of the power grid [95], which affects the accuracy of measurements [96]. Furthermore, the instability factors are magnetic stirrers and dissolved oxygen and its effect on the platinum electrode [97]. A possible solution to improve the signal quality is a Faraday cage to shield the setup from external electromagnetic fields [98] and the application of uninterruptible power supplies with double conversion (not less than 1500 VA of full power) to improve potential reproducibility [99] and reduce the probability of loss of the registered data [100].

UV/Vis Probe
Continuous measurements require the use of flow cells or spectrophotometric fiberoptic probes. In this work, the UV/vis probe showed high reproducibility of the longterm signal at 254 nm for 10 h, which was~0.1%. Additionally, the probe accuracy check by a Ferroin solution spectrum and calculation of the apparent molar absorptivity ε 510 = 11,100 ± 200 M −1 cm −1 coincided with existing data [101] (ε 510 = 11,080 M −1 cm −1 ). Thus, UV/vis probes could be used for washing control from organic reagents, which absorb in 190-400 nm. The dialysate solution was limpid and colorless; the endpoint criteria of absorbance of 0.05 corresponds to <1 ppm of dissolved substances [102].
The disadvantages of UV/vis probe are the instability of operation when air bubbles are formed with poorly degassed solvents, which leads to deviations from Beer's law and an increased proportion of the scattering component in the absorption spectra.

Characterization of Purified GO Dispersions
Considering the detailed cycle of high-purity GO dispersions preparation from graphite to graphene oxide material and to high-purity graphene oxide aqueous dispersions, each stage requires proper characterization of the resulting sample. These stages include (1) chemical graphite oxidation exfoliation, (2) GO isolation and characterization, (3) extra exfoliation, if any, and further purification and complete characterizations. For the second stage, the methods to verify GO structure using standard protocols [103] are XRD, XPS, SEM, and Raman spectroscopy. Vis-à-vis the third stage, this is much more challenging and requires individual quality standards and criteria. The development trend of such methods and techniques lies in the study of the qualitative and quantitative GO composition of both major and minor, both organic and inorganic impurity components and the analysis of their long-term stability.
For further GO biomedical testing and hemocompatibility, we recommend the chemical information for the three-level characterization of aqGO: (1) elemental composition by ICP-AES, anionic composition by IC and pH-metry; (2) absorption spectra and FTIR for the presence of organic substances or colorants; (3) colloidal stability by DLS. For GO dispersions, we set the water requirements for hemodialysis. Insufficient purity in terms of metal content can set off a clotting cascade [104,105].
The FTIR spectrum of GO dispersions displays the presence of typical GO bands [106]. Several peculiarities deal with absorption bands at 1729 and 1619 cm −1 , which are distinctive features of any GO sample. There is some ambiguity in the interpretation of the band at 1619 cm −1 , which could be assigned to C=C [107][108][109]. However, this band corresponds to the absorption of water molecules embedded in graphite layers, proven by isotopic exchange with D 2 O [110][111][112]. As well, dialysis samples showed some FTIR band redshifts up by 30 cm −1 . More bands in purified samples compared to the unpurified are due to less interaction with contaminants (Figure 5c), which is consistent with the purified proteins [113] and chemical states of edges (in our case purified edges) play a crucial role in electrical transport within graphene-based materials [114].
Higher band intensities at 1619 cm −1 are concerned with an increasing quantity of embedded H 2 O dipoles, increased intra-and interlayer H-bonding network between hydroxy-and epoxy-moieties [115] and H-bonds with other water molecules (Figure 5c). These sites play a crucial role as binding sites of removed transition metals. In addition, the GO surface is highly reactive, including epoxide opening and alkoxide shuttling, and their interactions with water as well as proton abstraction events from interfacial waters are possible [116]. The increase in the water band intensities presumably reflects better water sorption capacity owing to shorter diffusion pathways [117] in the purified state of GO sheets. However, more detailed investigations of this sorption capacity are required, that were beyond the scope of this study. As for carbonyl stretching modes at 1729 cm −1 , its changes are related to the dissociation of the lactone-ether pairs [118].
Raman scattering stands as a standard technique of the quality of GO materials. Raman spectroscopy confirmed the GO structure preservation during the purification process (see the Supplementary Materials, Figure S11). Samples had both G and D vibration bonds of sp 2 carbon. Raman spectra of our materials presented D and G bands centered at about 1330 and 1590 cm −1 , respectively. Band intensity ratio (I D /I G ) had the same order of magnitude, but after the dialysis process, increased by 3.5%, which indicated the better alignment of carbon atom structure to a perfect hexagonal honeycomb. In addition, the total area under the curve decreased after the purification process proves changes in samples. The crystallinity structure band D shifted from 1330 to 1339 cm −1 (∆ = 9 cm −1 ).
Zeta-potential value is deteriorated after the dialysis process by 20%. For dialysis using both 3 and 14 kDa membranes, zeta-potential shows the same magnitude. The experimental values between −40 to −31 mV correspond to moderate stability (no agglomerates) [119], which correlates to [120]. The lateral size of GO samples stayed the same, slightly increasing after purification. The PDI index has decreased, which tends to be characteristic of monodisperse systems, and the size and shape of the distribution are unchanged.
As a whole, we have achieved significant improvements in the purification process that would provide a broad range of GO applications in the extended phases of purified GO by genotoxicity, cytotoxicity, and biocompatibility tests. The proposed approach to GO characterization by three levels extends the guidelines described elsewhere [121].

Reactivity with Respect to the Superoxide Anion Radical
The main result of experiments in the superoxide anion radical generation model is that highly purified GO exhibits antioxidant properties in all studied concentrations. The suppression of the CL signal (Figure 6b) cannot be explained by enzyme inactivation. GO did not inhibit xanthine oxidase activity that confirms both electrochemically measured uric acid and O 2 content during the reaction [122]. The formation of uric acid, i.e., xanthine oxidase activity, was not inhibited by any GO samples (the data are not shown).
Thus, after purification, GO possessed significant radical-scavenging activity, which previously was observed for rGO only [123]. Though few-layer-graphene (FLG) or graphene oxide was shown to induce significant oxidative damage at the mitochondrial level consequent to NADH dehydrogenase-and xanthine oxidase-dependent ROS with concentrations from 0.4 to 100 ppm [124], it was commercially available samples without any purification treatment that may be comprised of any contaminants. Thus, when studying the oxidative properties of graphene oxide, it is particularly important to work with highly purified samples to obtain the intrinsic effect of graphene, and not highly free radical-reactive impurities. The presumable action of GO can be related to H-atom donation from -OH, electron transfer [125] from graphene [126], and basal adduct formation [78] as anti-oxidant mechanisms. The graphene-like domain shows highly efficient catalytic conversion of O 2 •− into O 2 and H 2 O 2 [127], which becomes more available after the purification. The activity decreases in the row FLG > rGO > GO, mainly caused by the fact that the main active sites are associated with the pristine graphene network [78] rather than with the oxygen-containing functional groups. As we have previously shown for non-functionalized fullerenes [52], the electron affinity in the row C 60 < C 70 < Gd@C 82 plays a crucial role in attaching radicals. However, the quenching effect of chemiluminescence during GO examination should not be underestimated [128].
As the reaction mechanism is complex and multistage, we have limited ourselves to qualitatively comparing the reactivity of purified and unpurified GOs towards superoxide anion radical. Mechanistic studies and quantitative examination of nanozyme SOD-like activity may be the aim of further studies using other methods like EPR, etc.

Reactivity with Respect to Hydrogen Peroxide
Many studies concern nanomaterials with enzyme-like characteristics making of artificial enzymes. Origins of the peroxidase mimicking activities of graphene oxide from first principles were elucidated [129]. GO can activate H 2 O 2 -generating hydroxyl radicals with higher oxidizabilities. Here, unpurified GO samples exhibit much more peroxidase mimicking activities (Figure 7a,b), while purified GO exhibits catalase-like properties. On the one hand, this can be explained by the presence of variable-valence metals (manganese) in solution. On the other hand, this can be justified by a change in the graphene-oxide structure itself. Thus, the authors [130] claimed that carboxyl-modified graphene oxide (GO-COOH) possesses intrinsic peroxidase-like activity that can catalyze the reaction of peroxidase substrate 3,3,5,5-tetramethylbenzidine in the presence of H 2 O 2 , and the observed peroxidase-like activity is not related to the trace metal catalyst in the sample but is caused by its own intrinsic property instead. Efficient analytical systems based on peroxidases and copper ions with GO have been proposed [131,132]. The paper [133] thoroughly considers the possible mechanism of the catalytic action of graphene oxide with respect to hydrogen peroxide. It has been argued that carbonyl C=O is the active peroxidase catalytic center [134], while −COOH is much less effective [129]. The ether is very stable and not a catalytic center; the epoxy, hydroxyl, and endoperoxide are also not catalytic centers. We showed for purified samples that the amount of these acidic centers is much lower than in non-purified samples by titration (see the Supplementary Materials, Section S1.1, Table S2). According to DLS and zeta-potential data, GO lateral size became slightly higher, up to 15% (Table 2) due to slight aggregation activity, which should cause a decrease in activity. On the other hand, compared to large GO nanosheets, GO quantum dots efficiently reduce reactive oxygen species and H 2 O 2 in vitro and in vivo [135]. Enlarged repulsive interaction by decreasing ionic strength or increasing pH diminish the coagulation of GO platelets [136]. GO samples which may not have been sufficiently purified, which could lead to the formation of small flocculates or aggregates of GO sheets. [137].
Presumably, deep crowding and hydration effects and changes in enzyme conformations [138] are more expressed in purified samples due to much better binding. However, molecular docking and dynamic approaches should be developed further. Some studies show the dual catalytic activity, peroxidase-or catalase-, at different pHs [139]. Here, we studied only physiologically relevant conditions at pH 7.4.

Conclusions
This study is the first step to developing procedures for estimating and regulating graphene oxide activity. We believe that it is especially essential to extend the study to other types of rGOs or GOs, such as Brodie's methods, where fewer carboxyl groups and point defects are present [140], for a more rigorous understanding of the catalytic cycle of the process. The developed approach to the purification of GO samples can be adapted to other synthetic carbon nanomaterials (rGO, nanotubes, carbon dots etc.) under bench-scale batches that have significant potential for real-life applications. Highly purified materials would be highly demanded for materials science and in optoelectronics. Challenges ahead also deal with developing a standard purification protocol for carbon-based nanomaterial for biomedical diagnostic purposes. It requires further research on regulating antioxidant or catalytic activity.
Additionally, we believe that an important finding of this study is dialysis-purified graphene oxide itself. Unlike commercial GO, it exhibits antioxidant properties to the superoxide anion radical and hydrogen peroxide, which is of importance for subsequent biocompatibility and toxicity studies. In our opinion, modeling the free-radical kinetics involving purified GO may help draw conclusions about the mechanism of GO action in free-radical models.
It also seems promising to use graphene oxide to create platforms for a remarkable chiral separation (enantioseparation) capacity [141] with an accessible interlayer spacing of 8 Å [142]. Primarily, the dialysis purification would provide specific fractions to create more capacitive sorbents and improve enantiomeric excess.