Does Fermentation Really Increase the Phenolic Content in Cereals? A Study on Millet

Millet is underutilized in Europe, despite its advantages compared to other common cereals. In Asia and Africa, millet is mainly eaten in fermented form; its consumption has beneficial properties on human health. Three millet batches were compared in terms of free and bound phenols by High Performance Liquid Chromatography-Diode Array Detector-Mass Spectrometry (HPLC-DAD-MS). The richest one in terms of bound phenols was selected for testing via a basic (0.1 M NaOH) and an acidic (1.2 M H2SO4) hydrolysis, in which 149.3 and 193.6 mg/100 g of phenols were recovered, respectively. The ability of fermentation, with yeast and Lactobacilli, to increase the content of phenolic compounds was verified. Five withdrawalswere performed to verify the influence of fermentation time on the total phenolic content. The greatest phenolic content was observed after 72 h. Fermentation increased the cinnamic acids and flavonoids contents by approximately 30%. Vitexin and vitexin 2″-O-rhamnoside contents were significantly higher in the fermented millet; these compounds partially inhibit the protein tyrosine phosphatase enzyme, which is overexpressed in type-2 diabetes. A molecular dynamic simulation showed the two flavonoids to be allosteric inhibitors. The phenolic extract from fermented millet demonstrated a higher level of antioxidant protection on human erythrocytes by ex vivo cellular antioxidant activity in red blood cells. In this context, functional foods based on fermented millet could represent a new trend in European markets.


Introduction
Millet is a generic term that includes various small grains belonging to the Poaceae family. It exhibits great advantages compared to other cereals such as wheat. Millet is able to grow almost anywhere thanks to its resistance to drought, high temperature tolerance, and low incidence of mycotoxin contamination [1]. In 2016, millet was the sixth most produced cereal worldwide. Furthermore, it contains high levels of proteins, minerals, and vitamins, that confer upon it higher nutritional value than nonmillet cereals. Its proteins are reported to possess significant levels of essential amino acids. Furthermore, compared to other cereals, it is rich in slowly digestible and resistant starch, which are important for the prevention of diseases related to type-2 diabetes, obesity, and coronary heart disease, because of their resistance to digestion, resulting in the slow release of glucose into the blood stream (c) to investigate on the mechanism of action of vitexin and vitexin 2"-O-rhamnoside through an in silico study; (d) to evaluate the antioxidant activity (CAA-RBC) of fermented and unfermented samples. The work aims to show how as a simple fermentation method can improve the nutritional and functional value of millet, a cereal not commonly used in Europe, paving the way for the design of new fermented products.
Human recombinant PTP-1B, expressed in Escherichia coli TB1 strain, and purified, and p-nitrophenyl phosphate, were purchased from Santa Cruz Biotechnology Inc. (Dallas, TX, USA).

Fermentation Process
The fermentation of the M1 and M2 samples (FM1 and FM2) was performed at the CNR of Pisa using a natural sourdough constituting a mixture of Lactobacilli, mainly L. sanfranciscensis and L. pentosus, and yeast strains in a ratio of approximately 100:1. Millet flour was obtained by grinding millet grains using a laboratory mill (M20 Universal mill IKA-Werke, Staufen im Breisga, Germany). Water (60%) was added to moisten the mixture, and then the sourdough, provided by "Lievitamente s.n.c." (Viareggio (LU), Italy), was added to initiate fermentation. Once the product was fermented, it was frozen and freeze dried for two days with a Lyovac GT 2. The fermentation temperature was maintained at around 38 • C, while the pH reached a value of 4. Different withdrawals were performed at time 0 (T0) and during the process, that is, after 24 h (T1), 48 h (T2), 72 h (T3), and 96 h (T4) of fermentation.

Extraction of Phenolic Compounds
Free phenols. All the unfermented and fermented samples were treated with aqueous ethanol as previously described in the literature [24]. Briefly, the flour obtained from three milled samples (M1, M2, M3) was defatted twice with hexane (1:10 (w/v)) and kept on a mechanical shaker for 1 h. Then, 250 mg of millet flour was then extracted with ethanol:water 80:20 v/v under magnetic stirring in an ultrasound bath for about 15 min. The extract was centrifuged at 5000 rpm for 10 min, 10 mL of the supernatant were evaporated to dryness, and the residue redissolved in 1.5 mL of acidified H 2 O (1% HCOOH).
Total phenols. The flour of the three millet samples, and the fermented sample from M1 FM1-T3 obtained after 72 h were treated with the following hydrolytic procedure. The basic hydrolysis was carried out with the aid of ultrasound, as previously described [24]; 1 g of flour was treated with 25 mL of NaOH 0.1 M in MeOH/H 2 O 7:3 v/v and sonicated (40 MHz) for 1 h at 60 • C; the pH was neutralized with acetic acid, the suspension centrifuged at 5000 rpm for 10 min, and the supernatant was recovered (Method B). The samples of M1 and M2 with the corresponding fermented products after 72 h of fermentation (FM1-T3 and FM2-T3) were also treated with an optimized acidic hydrolysis for the recovery of total phenols by applying only a single extractive step according to the method described by Balli et al. [25]. Briefly, 1 g of defatted flour was suspended in 25 mL of acidified MeOH (1.20 M by H 2 SO 4 ), the solution was sonicated for 180 min at 55 • C, and the sample was then centrifuged at 5000 rpm for 10 min (Method A).

MS Analysis
The HPLC-DAD-MS analyses of the phenolic extracts were performed according to a method described by Balli et al. [25]. The HPLC-DAD-MS system was from Waters, and comprised a 2695 HPLC, 2996 PAD, and 4 microtm MS equipped with Zspray ESI source. The ESI interface parameters were capillary 2.90 kV, cone 64 V in the first 18 min and 30 V until the end of the analysis, source temperature 120 • C, desolvation temperature 350 • C, cone gas flow 19 (L/Hr), and desolvation gas flow 350 (L/Hr). Data were acquired in negative ion mode from 110 m/z to 1000 m/z.

Inhibition Test on PTP1B Enzyme
The two flavonoids, vitexin and vitexin 2"-O-rhamnoside, were tested on the PTP1B enzyme at concentrations of 0.205 mg/mL and 0.111 mg/mL, respectively. Enzymatic assays were carried out using human recombinant PTP1B and p-nitrophenylphosphate (pNPP) as reference substrate [27]. Briefly, the pNPP (2.5 mM final concentration) was dissolved in sodium β,β-dimethyl glutarate buffer (75 mM, pH 7.0), containing 1mM ethylenediaminetetraacetic acid (EDTA) and 1 mM dithiothreitol. Reactions were started by the addition of aliquots of the enzyme and stopped by adding 2 mL of KOH 0.2 M. The released p-nitrophenolate was quantified by measuring the absorbance of the final solution at 400 nm (ε = 18,000 M −1 cm −1 ). The net hydrolysis rate was determined by subtracting the value of spontaneous hydrolysis rate of pNPP from each sample. The inhibitory power of vitexin standards was tested by adding different amounts of standard solutions, i.e., 5, 10, and 20 µL/mL for vitexin and 12, 25, and 50 µL/mL for vitexin 2"-O-rhamnoside. Furthermore, the two flavonoids mixed together (MIX-std) were tested in a concentration of 0.082 mg/mL and 0.111 mg/mL for vitexin and vitexin 2"-O-rhamnoside, respectively, adding 50 µL/mL. In addition, the aqueous extracts of unfermented (M1) and fermented flour (FM1) in a concentration of 1g/mL were both tested by adding 60 and 20 µL/mL, respectively. The percentage of inhibition was calculated by normalizing the absorbance values obtained for assays Foods 2020, 9,303 5 of 17 carried out in the presence of the inhibitor versus the control test. All the results were expressed as a mean of three independent experiments.

Molecular Dynamics Simulation for PTP1B Enzyme
A 10 ns MD simulation was performed for all the complexes using the GROMACS v5.1 program [28]. The partial atomic charge of the structures was calculated with CHIMERA [29] using the AM1-BCC method, and the topology was created with ACPYPE [30] based on the routine Antechamber [31]. The OPLS-AA/L all-atom force field [32] parameters were applied to all the structures. The simulation was performed in a vacuum. To remove bad contacts, energy minimization was performed using the steepest descent algorithm until convergence was achieved, or for a maximum of 50,000 steps. In the consecutive steps, equilibrations of the system were conducted in two phases: (1) canonical Particle Number, Volume, Temperature (constant) (NVT) ensemble, a 100 ps position-restrained of molecules at 300 K was carried out using a Temperature coupling thermostat (velocity rescaling with a stochastic term) to ensure the proper stabilization of the temperature [33]; (2) isothermalisobaric Particle Number, Pressure, Temperature (constant) (NPT) ensemble, a 100 ps position restrained of molecules at 300 K and 1 bar was carried out without using barostat pressure coupling to stabilize the system. These were followed by a 10 ns MD run at 300 K with position restraints for all protein atoms. The Lincs algorithm [34] was used for bond constraints to maintain rigid bond lengths. The initial velocity was randomly assigned according to Maxwell-Boltzman distribution at 300 K and computed with a time step of 2 fs; the coordinates were recorded every 0.1 ns.

Ex Vivo Cellular Antioxidant Activity (CAA-RBC) Assay in Red Blood Cells
Human blood samples from healthy volunteers were collected in EDTA-treated tubes and centrifuged for 10 min at 2300× g at 4 • C. Plasma and buffy coat were discarded and erythrocytes were washed twice with PBS pH 7.4. Fermented and unfermented millet samples were extracted with 10% of dimethyl sulfoxide in distilled water according to a method described by Gabriele et al. [35]. Briefly, the samples were centrifuged for 10min at 2300 g at 4 • C in a Jouan CR3i centrifuge (Newport Pagnell, UK), and the supernatant was collected, filtered on 0.2 µm cellulose filter (VWR International PBI, Milan, IT), and kept at 4 • C in the dark. The antioxidant activity of millet extract was tested at three different concentrations (0.01, 0.1 and 1 mg/mL) on human erythrocytes as described by Frassinetti et al. [36]. Trolox was used as a reference standard. The fluorescence was read at 485 nm excitation and 535 nm emission using a Victor TM X3 Multilabel Plate Reader (Waltham, MA, USA), and each value was expressed using the following formula: CAA unit = 100 − ( SA/ CA) × 100, where SA is the integrated area of the sample curve and CA is the integrated area of the control curve. Data are displayed as the average value of independent experiments performed on three different subjects. Blood samples were collected from healthy donors according to the regulations of "Fondazione G. Monasterio CNR-Regione Toscana" for the use of residual blood for research purposes. Informed consent was obtained from all participants and/or their legal guardians.

Statistical Analysis
Each experiment was performed in triplicate, and the results were expressed as the mean values ± SD; the EXCEL software in-house routines were applied. Differences between fermented and unfermented millet effects on human erythrocytes were analyzed by one-way analysis of variance (ANOVA) with Dunnett's multiple comparison test and by unpaired t-test. A p-value lower than 0.05 was considered as statistically significant.

Proximate Analysis
Dietary fiber (soluble and insoluble) content was determined according to the AOAC Method 991.43 (determination of soluble, insoluble, and total dietary fiber in foods and food products; final approval: 1991).

Phenolic Characterization of Nonfermented Flours
The HPLC profiles obtained from the extraction of free phenols from the three analyzed batches were very similar from a qualitative point of view (Figure 1), with compounds 3, 4, 6, and 7 being recognized as flavonoids (f), and compounds 1, 5, and 8-12 as cinnamic (c) derivatives. The HPLC profiles obtained from the extraction of free phenols from the three analyzed batches were very similar from a qualitative point of view ( Figure 1), with compounds 3, 4, 6, and 7 being recognized as flavonoids (f), and compounds 1, 5, and 8-12 as cinnamic (c) derivatives.  Table  1; c= cinnamic derivatives; f= flavonoids.
All the molecules listed in Table 1 and recognized by HPLC-DAD and MS analysis have already been identified in millet [25]. Chromatographic profiles showed differences in relative abundances among the three batches, while with regard to cinnamic derivatives, compound c5, identified as ferulic acid, was only present in M1 and M3 (Figure 1b). The extraction of total phenols was carried out on the whole defatted flour with a basic hydrolysis by NaOH 0.1 M. The applied hydrolytic conditions were milder than those proposed in the literature [19,37].   Table 1; c = cinnamic derivatives; f = flavonoids. All the molecules listed in Table 1 and recognized by HPLC-DAD and MS analysis have already been identified in millet [25]. Chromatographic profiles showed differences in relative abundances among the three batches, while with regard to cinnamic derivatives, compound c5, identified as ferulic acid, was only present in M1 and M3 (Figure 1b). The extraction of total phenols was carried out on the whole defatted flour with a basic hydrolysis by NaOH 0.1 M. The applied hydrolytic conditions were milder than those proposed in the literature [19,37].
From a quantitative point of view, M1 showed total free phenols content of 127.4 mg/100 g, 87% of which were flavonoids; M2 showed a total of 48.2 mg/100 g with 68% flavonoids, and M3 showed a total of 64.0 mg/100 g with 83% flavonoids ( Table 2).  The data are expressed in mg/100 g of dry weight as a mean of three independent extractions. TCC: total cinnamic content; TFC: total flavonoids content; TPC: total phenolic content; RSD < 5%. Different letters (a, b and c) indicate significant differences at p < 0.05. Table 2 shows M1 to be the richest in terms of total phenols (149.3 mg/100 g), while for bound phenols, calculated as a difference between total and free phenols, M3 was the richest one (59.7 mg/100 g). To better evaluate the amount of bound phenols, M1 was also treated with an acidic hydrolysis which was recently optimized on a millet sample by the Response Surface Methodology according to Balli et al. [25]. Acidic hydrolysis guaranteed a recovery of the same phenols detected after the basic procedure but in a higher amount, i.e., 193.6 mg/100 g instead of 149.3 mg/100 g (Table 3). Our results confirm that flavonoids in millet are almost completely present in free forms, in agreement with previous findings [25]; this makes them more stable in acidic conditions, with 111.6 mg/100 g recovered flavonoids compared to 71.9 mg/100 g obtained in a basic medium. Analogously, the acidic hydrolysis guaranteed a higher release of cinnamic derivatives bound to cell wall structures, reaching 82 mg/100 g.

Phenolic Characterization of Fermented Flours
Millet is usually decorticated before consumption to improve its edible properties, but this practice leads to a reduction in minerals, fibers, and antioxidants as phenolic compounds located in the peripheral parts of the grains [38]. Fermentation can be applied in order to treat the whole nondecorticated flour, thereby avoiding losing part of its functional components. Contradictory data about the effectiveness of fermentation in increasing phenolic content are present in the literature. Some studies reported that fermentation is able to increase the bio-conversion of phenolic compounds from their linked or conjugated forms to the corresponding free forms [17,39,40], while other works reported a decrease in the total phenolic content attributable to the rearrangement of some phenolic structures after a self-polymerization processes favored by the acidic conditions induced by fermentation [8,41].
In this context, the use of HPLC-DAD and pure standards for the quantification of phenolic compounds before and after fermentation represents a more accurate method to estimate the real content. Herein, the extractions of both free and bound phenols were performed in order to verify whether fermentation was able to guarantee an almost complete release of the bound forms linked to the cellular structures, as cellulose or lignin. The M1 was selected to undergo the fermentation process. The HPLC-DAD profiles of free phenols from the fermented matrix showed a lower number of peaks compared to those obtained from the nonfermented sample; in particular, only flavonoids were detected ( Figure 2). The absence of phenolic acids in the fermented flours could be attributable to the ability of Lactobacilli to metabolize some of the phenolic acids present in millet [42]. Some studies reported that fermentation is able to increase the bio-conversion of phenolic compounds from their linked or conjugated forms to the corresponding free forms [17,39,40], while other works reported a decrease in the total phenolic content attributable to the rearrangement of some phenolic structures after a self-polymerization processes favored by the acidic conditions induced by fermentation [8,41].
In this context, the use of HPLC-DAD and pure standards for the quantification of phenolic compounds before and after fermentation represents a more accurate method to estimate the real content. Herein, the extractions of both free and bound phenols were performed in order to verify whether fermentation was able to guarantee an almost complete release of the bound forms linked to the cellular structures, as cellulose or lignin. The M1 was selected to undergo the fermentation process. The HPLC-DAD profiles of free phenols from the fermented matrix showed a lower number of peaks compared to those obtained from the nonfermented sample; in particular, only flavonoids were detected ( Figure 2). The absence of phenolic acids in the fermented flours could be attributable to the ability of Lactobacilli to metabolize some of the phenolic acids present in millet [42]. The total content of free phenols changed during fermentation, passing from 126.9 mg/100 g at T0 to a maximum of 145.3 mg/100 g at T3; the amount after 96 h (T4) remained almost unchanged ( Figure 3). According to this result, T3 was selected as the maximum time for fermentation. The total content of free phenols changed during fermentation, passing from 126.9 mg/100 g at T0 to a maximum of 145.3 mg/100 g at T3; the amount after 96 h (T4) remained almost unchanged ( Figure 3). According to this result, T3 was selected as the maximum time for fermentation. The Folin Ciocalteu assay for FM1-T3 identified a significantly higher level of total phenols than in the M1 sample (270.1 ± 34.2 vs. 150.3 ± 8.1 mg GAE/100 g DW, p < 0.05). These results demonstrated an increase of phytochemical content after the fermentation process.
After the basic and acidic hydrolyses applied to the FM1-T3 sample, the presence was confirmed of the same phenolic compounds detected before fermentation. The basic hydrolysis allowed us to extract a higher amount of phenols (206.2 mg/100 g) compared to the basic hydrolysis of the unfermented sample (149.3 mg/100 g). Analogously, the acidic hydrolysis previously optimized for unfermented millet [25] was also performed on FM1-T3, again confirming the ability of this method to more efficiently extract the total phenols from cereals, obtaining 278.3 mg/100 g (Table 3). Even though a real comparison with the literature data should consider different environmental aspects, varieties and analytical methods, our results highlight that the total recovered phenols obtained by the acidic hydrolysis are in the higher range of those proposed in the literature, i.e., from 36.7 mg/100 g to 377.2 mg/100 g (Table S1). The higher total phenolic values in the fermented sample could be attributed to the action of microorganisms that, using the soluble and fermentable fiber for their growth, released "mechanically trapped" phenolic compounds in the polymeric structures of the fiber. The evaluation of soluble fiber allowed us to verify that the M1 sample contained 2g/100 g, which, after fermentation, strongly decreased to <0.10 g/100 g, demonstrating that fermentation was a useful treatment to release phenolic compounds which are mechanically or chemically trapped in the soluble fiber. For the M1 sample, the fermentation, induced an increase of up to 35% in the amount of these molecules with respect to their total content in unfermented flour. To confirm this result, another batch of millet (M2) was fermented using the same mixture of yeast and Lactobacilli, and free phenols and total phenols were evaluated after 72 h of fermentation by acidic hydrolysis. The results The Folin Ciocalteu assay for FM1-T3 identified a significantly higher level of total phenols than in the M1 sample (270.1 ± 34.2 vs. 150.3 ± 8.1 mg GAE/100 g DW, p < 0.05). These results demonstrated an increase of phytochemical content after the fermentation process.
After the basic and acidic hydrolyses applied to the FM1-T3 sample, the presence was confirmed of the same phenolic compounds detected before fermentation. The basic hydrolysis allowed us to extract a higher amount of phenols (206.2 mg/100 g) compared to the basic hydrolysis of the unfermented sample (149.3 mg/100 g). Analogously, the acidic hydrolysis previously optimized for unfermented millet [25] was also performed on FM1-T3, again confirming the ability of this method to more efficiently extract the total phenols from cereals, obtaining 278.3 mg/100 g (Table 3). Even though a real comparison with the literature data should consider different environmental aspects, varieties and analytical methods, our results highlight that the total recovered phenols obtained by the acidic hydrolysis are in the higher range of those proposed in the literature, i.e., from 36.7 mg/100 g to 377.2 mg/100 g (Table S1).
The higher total phenolic values in the fermented sample could be attributed to the action of microorganisms that, using the soluble and fermentable fiber for their growth, released "mechanically trapped" phenolic compounds in the polymeric structures of the fiber. The evaluation of soluble fiber allowed us to verify that the M1 sample contained 2 g/100 g, which, after fermentation, strongly decreased to <0.10 g/100 g, demonstrating that fermentation was a useful treatment to release phenolic compounds which are mechanically or chemically trapped in the soluble fiber. For the M1 sample, the fermentation, induced an increase of up to 35% in the amount of these molecules with respect to their total content in unfermented flour. To confirm this result, another batch of millet (M2) was fermented using the same mixture of yeast and Lactobacilli, and free phenols and total phenols were evaluated after 72 h of fermentation by acidic hydrolysis. The results again highlighted an increase of 36% in the total phenolic content, passing from 66.5 mg/100 g to 90.7 mg/100 g for the unfermented and fermented matrix respectively (Figure 4). again highlighted an increase of 36% in the total phenolic content, passing from 66.5 mg/100 g to 90.7 mg/100 g for the unfermented and fermented matrix respectively (Figure 4).

Inhibition Test on PTP1B Enzyme
The two standards, i.e., vitexin and vitexin 2"-O-rhamnoside, were in vitro evaluated for their inhibitory effect on PTP1B, known as a negative regulator of insulin receptors. The choice of testing these flavonoids was made due to the inhibitory activity already shown by this class of molecules on the PTP1B enzyme [23]. In contrast, the cinnamic derivatives that were more abundant in millet were unable to inhibit the enzyme [43]. Previous literature has already reported the ability of vitexin to inhibit the PTPB1B enzyme at a concentration of 7.62 µM [44]. Considering the multiple structures of the previously used PTP1B enzyme, we decided to test the two standards of vitexin on our enzyme model. The results indicated that vitexin and vitexin 2"-O-rhamnoside were not able to inhibit the enzyme at concentrations of 2.3 µM and 4.7 µM, respectively, but an inhibition of 25% and 32% was identified for both of these flavonoids at 9.5 µM (Figure 5a). In light of this finding, the other three samples were tested: a mixture (MIX-std) of the two flavonoids at a concentration of 9.5 µM and the total unfermented extract (M1) and fermented extract (FM1) containing vitexin and vitexin 2"-O-rhamnoside, both at a concentration of approximately of 9.5 µM; see Figure 5b. The results highlighted inhibitory activity against the PTP1B enzyme of approximately 40% in the mixture extract (MIX-std), suggesting a sort of sum effect of the two flavonoids. Concerning the unfermented and fermented extracts (M1 and FM1), the results indicated an inhibition of 13% and 25%, respectively. These data suggest the presence, mainly in the unfermented flour, of other compounds that act as antagonists of the vitexin flavonoids, thereby reducing their activity on the PTP1B enzyme. a) Figure 4. Free and total phenolic amount, by acidic hydrolysis, before and after fermentation in M2 sample. The fermented sample was withdrawn after 72 h of fermentation. Data are expressed in mg/100 g as a mean of three independent extraction.

Inhibition Test on PTP1B Enzyme
The two standards, i.e., vitexin and vitexin 2"-O-rhamnoside, were in vitro evaluated for their inhibitory effect on PTP1B, known as a negative regulator of insulin receptors. The choice of testing these flavonoids was made due to the inhibitory activity already shown by this class of molecules on the PTP1B enzyme [23]. In contrast, the cinnamic derivatives that were more abundant in millet were unable to inhibit the enzyme [43]. Previous literature has already reported the ability of vitexin to inhibit the PTPB1B enzyme at a concentration of 7.62 µM [44]. Considering the multiple structures of the previously used PTP1B enzyme, we decided to test the two standards of vitexin on our enzyme model. The results indicated that vitexin and vitexin 2"-O-rhamnoside were not able to inhibit the enzyme at concentrations of 2.3 µM and 4.7 µM, respectively, but an inhibition of 25% and 32% was identified for both of these flavonoids at 9.5 µM (Figure 5a). In light of this finding, the other three samples were tested: a mixture (MIX-std) of the two flavonoids at a concentration of 9.5 µM and the total unfermented extract (M1) and fermented extract (FM1) containing vitexin and vitexin 2"-O-rhamnoside, both at a concentration of approximately of 9.5 µM; see Figure 5b. The results highlighted inhibitory activity against the PTP1B enzyme of approximately 40% in the mixture extract (MIX-std), suggesting a sort of sum effect of the two flavonoids. Concerning the unfermented and fermented extracts (M1 and FM1), the results indicated an inhibition of 13% and 25%, respectively. These data suggest the presence, mainly in the unfermented flour, of other compounds that act as antagonists of the vitexin flavonoids, thereby reducing their activity on the PTP1B enzyme.
It is noteworthy that the concentrations of the two flavonoids in the fermented extract were 2-3 times higher than those needed to exert an enzymatic inhibition by 30%, i.e., 18 µM for vitexin and 22 µM for vitexin 2"-O-rhamnoside. Furthermore, the concentrations of the two flavonoids were higher in the fermented extracts (FM1-T3) with respect to the unfermented millet (MF1). After the recovery of total phenols in acidic condition, vitexin was 55% higher and vitexin 2"-O-rhamnoside was 60% higher. In light of these findings, we affirm that the daily consumption of fermented millet containing higher amounts of flavonoids with respect to unfermented flour could help to reduce the incidence of type-2 diabetes by the inhibition of the PTP1B enzyme, which is overexpressed in these subjects. Further analyses will be conducted to better understand whether the entire fermented extract is able to inhibit the PTP1B through a synergic action between the single components of the fermented flour.
2"-O-rhamnoside, both at a concentration of approximately of 9.5 µM; see Figure 5b. The results highlighted inhibitory activity against the PTP1B enzyme of approximately 40% in the mixture extract (MIX-std), suggesting a sort of sum effect of the two flavonoids. Concerning the unfermented and fermented extracts (M1 and FM1), the results indicated an inhibition of 13% and 25%, respectively. These data suggest the presence, mainly in the unfermented flour, of other compounds that act as antagonists of the vitexin flavonoids, thereby reducing their activity on the PTP1B enzyme.

a)
Foods 2020, 9, x FOR PEER REVIEW 11 of 17 It is noteworthy that the concentrations of the two flavonoids in the fermented extract were 2-3 times higher than those needed to exert an enzymatic inhibition by 30%, i.e., 18 µM for vitexin and 22 µM for vitexin 2"-O-rhamnoside. Furthermore, the concentrations of the two flavonoids were higher in the fermented extracts (FM1-T3) with respect to the unfermented millet (MF1). After the recovery of total phenols in acidic condition, vitexin was 55% higher and vitexin 2"-O-rhamnoside was 60% higher. In light of these findings, we affirm that the daily consumption of fermented millet containing higher amounts of flavonoids with respect to unfermented flour could help to reduce the incidence of type-2 diabetes by the inhibition of the PTP1B enzyme, which is overexpressed in these subjects. Further analyses will be conducted to better understand whether the entire fermented extract is able to inhibit the PTP1B through a synergic action between the single components of the fermented flour.

Molecular Dynamic Simulation for PTP1B Enzyme
Thanks to the crystallographic structure of a fragment of PTP1B ( Figure 6) complexed with 3-(3,5-dibromo-4-hydroxy-benzoyl)-2-ethyl-benzofuran-6-sulfonic acid (4-sulfamoyl-phenyl)-amide allosteric inhibitor (BFS) [45], it was possible to clarify the role of WPD-Loop [46]. The structure of the complex between the benzofuran derivative (BFS) and the PTB1B fragment [45] is deposited in the Protein Data Bank (PDB ID: 1t49). It is noteworthy that the concentrations of the two flavonoids in the fermented extract were 2-3 times higher than those needed to exert an enzymatic inhibition by 30%, i.e., 18 µM for vitexin and 22 µM for vitexin 2"-O-rhamnoside. Furthermore, the concentrations of the two flavonoids were higher in the fermented extracts (FM1-T3) with respect to the unfermented millet (MF1). After the recovery of total phenols in acidic condition, vitexin was 55% higher and vitexin 2"-O-rhamnoside was 60% higher. In light of these findings, we affirm that the daily consumption of fermented millet containing higher amounts of flavonoids with respect to unfermented flour could help to reduce the incidence of type-2 diabetes by the inhibition of the PTP1B enzyme, which is overexpressed in these subjects. Further analyses will be conducted to better understand whether the entire fermented extract is able to inhibit the PTP1B through a synergic action between the single components of the fermented flour.

Molecular Dynamic Simulation for PTP1B Enzyme
Thanks to the crystallographic structure of a fragment of PTP1B ( Figure 6) complexed with 3-(3,5-dibromo-4-hydroxy-benzoyl)-2-ethyl-benzofuran-6-sulfonic acid (4-sulfamoyl-phenyl)-amide allosteric inhibitor (BFS) [45], it was possible to clarify the role of WPD-Loop [46]. The structure of the complex between the benzofuran derivative (BFS) and the PTB1B fragment [45] is deposited in the Protein Data Bank (PDB ID: 1t49). In this work, the complex BFS-PTP1B was submitted to a MD simulation and the trajectory was compared with those obtained by complexes of PTP1B (obtained from 1t49) with four potential allosteric inhibitors with a flavonoidic structure, i.e., apigenin (API), schaftoside (SCH), vitexin (VIT) and vitexin-2"-O-rhamnoside (VRA) (Figure 7). Foods 2020, 9, x FOR PEER REVIEW 12 of 17 In this work, the complex BFS-PTP1B was submitted to a MD simulation and the trajectory was compared with those obtained by complexes of PTP1B (obtained from 1t49) with four potential allosteric inhibitors with a flavonoidic structure, i.e., apigenin (API), schaftoside (SCH), vitexin (VIT) and vitexin-2′'-O-rhamnoside (VRA) (Figure 7). The docking position (pose) of each flavonoid with respect to the PTP-1B enzyme was determined by AUTODOCK [47]. The calculation found for all the flavonoids selected in this work, along with other possible structures of docking, the same pose of the BFS allosteric inhibitor. The MD simulation for all complexes was carried out in a vacuum at 300 K per 10ns by GROMACS package [28]. The mean mobility of WPD-Loop and of S-loop in the complex with BFS, calculated as a value of the B factor, was compared with those obtained for the complexes with the selected flavonoids, yielding the results summarized in the histogram of Figure 8. The B-factor values were calculated by the GROMACS routine [28], starting from the root mean square fluctuation (RMSF) value of alpha carbon positions in the trajectory.  The docking position (pose) of each flavonoid with respect to the PTP-1B enzyme was determined by AUTODOCK [47]. The calculation found for all the flavonoids selected in this work, along with other possible structures of docking, the same pose of the BFS allosteric inhibitor. The MD simulation for all complexes was carried out in a vacuum at 300 K per 10ns by GROMACS package [28]. The mean mobility of WPD-Loop and of S-loop in the complex with BFS, calculated as a value of the B factor, was compared with those obtained for the complexes with the selected flavonoids, yielding the results summarized in the histogram of Figure 8. The B-factor values were calculated by the GROMACS routine [28], starting from the root mean square fluctuation (RMSF) value of alpha carbon positions in the trajectory.
Foods 2020, 9, x FOR PEER REVIEW 12 of 17 In this work, the complex BFS-PTP1B was submitted to a MD simulation and the trajectory was compared with those obtained by complexes of PTP1B (obtained from 1t49) with four potential allosteric inhibitors with a flavonoidic structure, i.e., apigenin (API), schaftoside (SCH), vitexin (VIT) and vitexin-2′'-O-rhamnoside (VRA) (Figure 7). The docking position (pose) of each flavonoid with respect to the PTP-1B enzyme was determined by AUTODOCK [47]. The calculation found for all the flavonoids selected in this work, along with other possible structures of docking, the same pose of the BFS allosteric inhibitor. The MD simulation for all complexes was carried out in a vacuum at 300 K per 10ns by GROMACS package [28]. The mean mobility of WPD-Loop and of S-loop in the complex with BFS, calculated as a value of the B factor, was compared with those obtained for the complexes with the selected flavonoids, yielding the results summarized in the histogram of Figure 8. The B-factor values were calculated by the GROMACS routine [28], starting from the root mean square fluctuation (RMSF) value of alpha carbon positions in the trajectory.  As shown in Figure 8, the mean mobility of WDP-Loop in the complexes with the flavonoids SCH, VIT, and VRA was very similar to that obtained for the complex with the allosteric inhibitor BFS, while in the complex with API, the fluctuation in the WPD-Loop was significantly higher. The higher value obtained for the latter flavonoid is in agreement with the lower inhibitory activity, i.e., approximatively three times higher with respect to that of VIT [47]. The fluctuation of the WDP-Loop for VIT and its glucoside (VRA) were almost identical, suggesting that the inhibitory activity was of the same entity. Similarly, the behavior of the S-Loop in the case of API showed a significantly higher fluctuation compared to the values obtained for the other flavonoids and for the allosteric inhibitor BFS. These in silico results were in perfect agreement with the experimental results obtained for the selected flavonoids, which showed VIT and VRA with a very similar inhibitory activity of PTP1B, while API was less effective.

Ex Vivo Cellular Antioxidant Activity on Human Erythrocytes
Erythrocytes play a key role in the body in terms of anti-inflammatory and antioxidant protection, and represent a powerful tool to assess the radicals scavenging activity of many natural compounds [48]. In this study, the biological effects of millet, before and after the fermentation process, was evaluated in an ex vivo model of human erythrocytes under oxidative conditions using the CAA-RBC assay ( Figure 9). Foods 2020, 9, x FOR PEER REVIEW 13 of 17 As shown in Figure 8, the mean mobility of WDP-Loop in the complexes with the flavonoids SCH, VIT, and VRA was very similar to that obtained for the complex with the allosteric inhibitor BFS, while in the complex with API, the fluctuation in the WPD-Loop was significantly higher. The higher value obtained for the latter flavonoid is in agreement with the lower inhibitory activity, i.e., approximatively three times higher with respect to that of VIT [47]. The fluctuation of the WDP-Loop for VIT and its glucoside (VRA) were almost identical, suggesting that the inhibitory activity was of the same entity. Similarly, the behavior of the S-Loop in the case of API showed a significantly higher fluctuation compared to the values obtained for the other flavonoids and for the allosteric inhibitor BFS. These in silico results were in perfect agreement with the experimental results obtained for the selected flavonoids, which showed VIT and VRA with a very similar inhibitory activity of PTP1B, while API was less effective.

Ex Vivo Cellular Antioxidant Activity on Human Erythrocytes
Erythrocytes play a key role in the body in terms of anti-inflammatory and antioxidant protection, and represent a powerful tool to assess the radicals scavenging activity of many natural compounds [48]. In this study, the biological effects of millet, before and after the fermentation process, was evaluated in an ex vivo model of human erythrocytes under oxidative conditions using the CAA-RBC assay ( Figure 9). In this assay, following 1 h pretreatment with millet extracts (0.01, 0.1 and 1 mg/mL), human erythrocytes were exposed to an oxidative insult induced by the thermal decomposition of AAPH in peroxyl radicals. Our results demonstrated good, dose-dependent antioxidant protection of human erythrocytes from all tested samples with the exclusion of millet flour at the highest tested concentration (1 mg/mL), that induced erythrocyte hemolysis (Figure 9). The fermented millet obtained after 72 h of fermentation (FM1-T3) and unfermented millet (M1) extracts significantly In this assay, following 1 h pretreatment with millet extracts (0.01, 0.1 and 1 mg/mL), human erythrocytes were exposed to an oxidative insult induced by the thermal decomposition of AAPH in peroxyl radicals. Our results demonstrated good, dose-dependent antioxidant protection of human erythrocytes from all tested samples with the exclusion of millet flour at the highest tested concentration (1 mg/mL), that induced erythrocyte hemolysis (Figure 9). The fermented millet obtained after 72 h of fermentation (FM1-T3) and unfermented millet (M1) extracts significantly raised the cellular antioxidant activity of human erythrocytes compared to the control, that refers to only AAPH-exposed cells (CAA = 0; ** p < 0.01), with CAA values being comparable to or higher than Trolox 10 and 50 µM, used as a reference standard. Additionally, the millet fermented extract exhibited, at all tested doses, significantly higher antioxidant protection than the respective unfermented ones (# p < 0.05).

Conclusions
This work aimed to increase knowledge about the effects of fermentation on millet. According to our results, fermentation carried out for three days using a mixture of natural microorganisms consisting mainly of Lactobacilli increased the total amount of phenolic compounds in the millet by more than 30%, presumably favoring the release of phenols trapped mechanically or chemically in the rate of soluble and fermentable fiber. In fact, soluble fiber has been shown to be used by Lactobacilli and yeasts for their growth. Furthermore, the acid hydrolysis with respect to basic hydrolysis guaranteed a higher recovery of total cinnamic derivatives and flavonoids from fermented millet, as already demonstrated for the unfermented samples. It is reasonable to say that, so far, the total phenolic content in cereals could be underestimated, due to the application of unsuitable extractive methods and due to the mechanical entrapment of these molecules by the fiber components.
Regarding the in vitro analyses, vitexin and vitexin 2"-O-rhamnoside partially inhibited the PTP1B enzyme, which is overexpressed in type-2 diabetes. In silico evaluations confirmed that both the main flavonoids in millet interact with WDP-Loop as allosteric inhibitors of PTP1B enzyme, suggesting that a daily intake of fermented millet may help reduce the risk of type-2 diabetes.
Our results also demonstrated a dose-dependent antioxidant protection of human erythrocytes exerted by both unfermented and fermented millet samples. According to the phenolic enrichment confirmed by chemical analyses, sourdough fermentation significantly raised the ex vivo antioxidant activities in fermented millet flour with respect to the unfermented one.
In light of our results, fermented millet, thanks to its high content of bioactive compounds, particularly flavonoids, could be proposed as a new product which is suitable for all types of consumers, and which is useful in the reformulation of recipes for celiac products for which the demand for new ingredients is constantly increasing.

Funding:
The study was partially funded by the Tuscany Region NUTRATOSCAFRICA project, DD650/2014.

Conflicts of Interest:
The authors declare no conflict of interest.