Optimization of Ultrasonic-Assisted Extraction of α-Glucosidase Inhibitors from Dryopteris crassirhizoma Using Artificial Neural Network and Response Surface Methodology

Dryopteris crassirhizoma Nakai is a plant with significant medicinal properties, such as anticancer, antioxidant, and anti-inflammatory activities, making it an attractive research target. Our study describes the isolation of major metabolites from D. crassirhizoma, and their inhibitory activities on α-glucosidase were evaluated for the first time. The results revealed that nortrisflavaspidic acid ABB (2) is the most potent α-glucosidase inhibitor, with an IC50 of 34.0 ± 0.14 μM. In addition, artificial neural network (ANN) and response surface methodology (RSM) were used in this study to optimize the extraction conditions and evaluate the independent and interactive effects of ultrasonic-assisted extraction parameters. The optimal extraction conditions are extraction time of 103.03 min, sonication power of 342.69 W, and solvent-to-material ratio of 94.00 mL/g. The agreement between the predicted models of ANN and RSM and the experimental values was notably high, with a percentage of 97.51% and 97.15%, respectively, indicating that both models have the potential to be utilized for optimizing the industrial extraction process of active metabolites from D. crassirhizoma. Our results could provide relevant information for producing high-quality extracts from D. crassirhizoma for functional foods, nutraceuticals, and pharmaceutical industries.


Evaluation of the Inhibitory Activity against α-Glucosidase
The α-glucosidase assay was performed as previously described [20]. First, 130 µL of α-glucosidase (0.16 unit/mL) in 0.1 mM phosphate buffer (pH 6.8) was added to a 96-well plate with 20 µL of test compounds dissolved in dimethylsulfoxide. Then, 50 µL of 1 mM substrate (4-nitrophenyl-D-glucopyranoside) in the buffer was added to the mixture. The intensity of p-nitrophenol converted from 4-nitrophenyl-D-glucopyranoside by αglucosidase was measured after 15 min at 405 nm using Infinite 200 PRO spectrophotometer (Tecan, Zürich, Switzerland).
where ∆C and ∆S are the intensity of control and inhibitor after 20 min, respectively. Acarbose was then selected as a positive control [21]. The inhibitory effect was elucidated via triplicate independent experiments. The results are shown as the mean ± standard error (SEM). Statistical differences were investigated using one-way ANOVA and Duncan s test with a p-value < 0.05. SigmaPlot 10.0 (Systat Software Inc., San Jose, CA, USA) was used to analyze the inhibition results.

Selection of Variables
In this context, three extraction factors, including sonication time (min), sonication power (W), and solvent-to-material ratio (mL/g), were examined in the preliminary analy-Metabolites 2023, 13, 557 4 of 13 sis. The total peak area of the bioactive components was selected as the response variable ( Figure S1).

Box-Behnken Design for Optimization
Box-Behnken design (BBD) experiment was used to examine the interaction between the factors and determine the optimum operating conditions. Based on the results of singlefactor experiments, the bounds of the variables were selected as 80-120 min for sonication time, 240-400 W for sonication power, and 40:1-120:1 for solvent-to-material ratio. A threelevel [higher (+1), middle (0), and lower (−1)] and a three-factor approach were used for the experiment design, model construction, and data interpretation. A total of 17 independent experimental runs were conducted to evaluate the effect on the bioactive compounds and predict the optimum operating parameters. The results for three independent variables were optimized by RSM using Design-Expert 12 (Stat-Ease, Minneapolis, MN, USA).

ANN
The ANN was realized in MATLAB R2020b (MathWorks, Natick, MA, USA). A combined dataset comprising 17 data points was compiled, and the operating parameters of extraction were used as independent input variables. The feedforward backpropagation network was selected along random data division, Levenberg-Marquardt (trainlm) was selected as a training function, and mean squared error (MSE) was used as a performance function with one hidden and one output layer.

Identification of Isolated Compounds from D. crassirhizoma
The ethanolic extract of D. crassirhizoma was partitioned into fractions and isolated by combined chromatographic methods, such as silica gel, Sephadex LH-20, and RP-18 CC, to obtain compounds 1 and 2 ( Figures 1 and S2-S7). The isolated compounds (1 and 2) were confirmed to be flavaspidic acid AP and nortrisflavaspidic acid ABB, respectively, by analyzing NMR spectroscopic data and comparison with the literature [10,11].

Inhibitory Activity against α-Glucosidase
α-Glucosidase is an enzyme classified as EC 3.2.1.20 and is a member of the glycoside hydrolase family [20]. Primarily, it is found in epithelial cells of the small intestine and plays an important role in the breakdown of disaccharides and oligosaccharides into monosaccharides thereby facilitating carbohydrate digestion [22]. Inhibition of α-glucosidase is a potential therapeutic strategy for treating type 2 diabetes because it can slow the release of sugar from starch and oligosaccharides, thus leading to delayed sugar absorption and decreased postprandial blood sugar levels [23,24].
The inhibition of α-glucosidase by D. crassirhizoma was examined by investigating the inhibitory activities of different extract fractions (ethanol extract, CHCl 3 , EtOAc, and aqueous fractions) and two isolated compounds (1 and 2). The data presented in Table 1 show the inhibitory activity of four extract fractions against α-glucosidase. All four fractions exhibited inhibition at a concentration of 100 µg, with inhibition percentages ranging from 71.6 ± 1.39 to 79.9 ± 1.71. The IC 50 value, which represents the concentration of the fraction required to inhibit α-glucosidase activity by 50%, was found to be 15.2 µg/mL for the EtOAc fraction, indicating that it is the most potent activity among all tested fractions. The EtOH extract, CHCl 3 , and aqueous fractions showed weaker inhibitory effects, with IC 50 values of 34.8 ± 1.43, 25.8 ± 0.68, and 22.2 ± 1.92 µg/mL, respectively. Additional examination of the isolated compounds demonstrated that at a concentration of 100 µM, only nortrisflavaspidic acid ABB (2) showed an inhibitory effect on α-glucosidase greater than 50%. Thus, this compound was tested at different concentrations to determine its IC 50 value. Previous studies reported that trimeric phloroglucinol nortrisflavaspidic acid ABB (2) exhibited the significant inhibition of PTP1B and β-glucuronidase, with IC 50 values of 1.19 ± 0.13 and 8.0 ± 1.8 µM, respectively [5,6]. In this study, nortrisflavaspidic acid ABB (2) demonstrated a strong inhibitory effect against α-glucosidase, with an IC 50 value of 34.0 ± 0.14 µM, stronger than that of the positive control acarbose (IC 50 = 329.2 ± 0.35 µM). Our results suggest that D. crassirhizoma could be a promising natural source of active compounds for the treatment of type 2 diabetes mellitus in general and inhibition of α-glucosidase in particular.

Development of the HPLC Analysis Method
Various HPLC parameters were examined to develop a scientifically reliable and efficient method for quantifying the amounts of the two compounds in D. crassirhizoma. These parameters included the composition of the mobile phase (acetonitrile-water or methanol-water with formic acid and/or trifluoroacetic acid as buffers), the temperature of the column (25,30,35, and 40 • C), and flow rate of the mobile phase (0.6, 0.8, 1.0, and 1.2 mL/min). To obtain satisfactory resolution and separation, the UV spectra of the two components were characterized using the photodiode array detector. The optimal wavelength for detection was set to 280 nm. The optimized HPLC conditions resulted in good peak shape, separation, and resolution of the two components, indicating the suitability of this analytical method for the subsequent study using RSM (Figure 2).

Development of the HPLC Analysis Method
Various HPLC parameters were examined to develop a scientifically reliable and efficient method for quantifying the amounts of the two compounds in D. crassirhizoma. These parameters included the composition of the mobile phase (acetonitrile-water or methanol-water with formic acid and/or trifluoroacetic acid as buffers), the temperature of the column (25,30,35, and 40 °C), and flow rate of the mobile phase (0.6, 0.8, 1.0, and 1.2 mL/min). To obtain satisfactory resolution and separation, the UV spectra of the two components were characterized using the photodiode array detector. The optimal wavelength for detection was set to 280 nm. The optimized HPLC conditions resulted in good peak shape, separation, and resolution of the two components, indicating the suitability of this analytical method for the subsequent study using RSM (Figure 2).

RSM Optimization
Based on the results of our single-factor experiment, RSM with a BBD was conducted using Design-Expert 12 software to optimize three independent experimental parameters, including sonication time (A), sonication power (B), and the ratio of solvent (EtOH) to material (C). A total of 17 experiments were performed, and the BBD matrix for the three individual variables and response results (the sum peak area of compounds 1 and 2) is presented in Table 2. The 17 designed experiments were conducted, the obtained data

RSM Optimization
Based on the results of our single-factor experiment, RSM with a BBD was conducted using Design-Expert 12 software to optimize three independent experimental parameters, including sonication time (A), sonication power (B), and the ratio of solvent (EtOH) to material (C). A total of 17 experiments were performed, and the BBD matrix for the three individual variables and response results (the sum peak area of compounds 1 and 2) is presented in Table 2. The 17 designed experiments were conducted, the obtained data were statistically analyzed, and the results are presented in Table 3. In a statistical model, a higher F-value and a lower p-value indicate a more significant effect [25]. The F-value of 60.05 indicates that this model is significant. The model term is considered significant when its p-value is below 0.05. In this study, model terms A, B, AB, AC, BC, A 2 , B 2 , and C 2 are significant. The f-value for the lack of fit is 5.67, indicating that the lack of fit is non-significant in comparison to the pure error. The p-value of 0.0634 indicates that there is only a 6.34% possibility that a lack of fit F-value could occur because of noise.
The adequacy of the developed model was assessed to evaluate the data analysis of the experiment. Figure 3A presents the predicted peak area of the active compounds, which agrees with the actual experimental results. The coefficient of determination is remarkably high, indicating a strong correlation between the predicted and actual values. As shown in Figure 3B, the normal percentage probability plot for externally studentized residuals of variables A, B, and C exhibits a uniform distribution with no variance, indicating that these individual variables behave well. The adjusted and simple coefficients of determination (Adjusted R 2 and R 2 , respectively) are 0.9969 and 0.9872, respectively. These values indicate that the RSM model accounts for 99.69% of the variability of the corresponding variable, indicating that this model is a good fit for the experimental results. Therefore, the results suggest that the RSM model successfully describes the data and that the employed methodology is acceptable.
Equation ( where Y is the sum of the peak areas of two active compounds (1 and 2). The adequacy of the developed model was assessed to evaluate the data analysis of the experiment. Figure 3A presents the predicted peak area of the active compounds, which agrees with the actual experimental results. The coefficient of determination is remarkably high, indicating a strong correlation between the predicted and actual values. As shown in Figure 3B, the normal percentage probability plot for externally studentized residuals of variables A, B, and C exhibits a uniform distribution with no variance, indicating that these individual variables behave well. The adjusted and simple coefficients of determination (Adjusted R 2 and R 2 , respectively) are 0.9969 and 0.9872, respectively. These values indicate that the RSM model accounts for 99.69% of the variability of the corresponding variable, indicating that this model is a good fit for the experimental results. Therefore, the results suggest that the RSM model successfully describes the data and that the employed methodology is acceptable.
Equation (2) shows the relationship between the response and three individual variables: where Y is the sum of the peak areas of two active compounds (1 and 2).

Combined Effect of Solvent Concentration, Power, and Extraction Time
The data points on each ramp shown in Figure 4 show the anticipated response of the best choice for extracting the maximum amount of the two compounds. The optimal settings are attained at the extraction time of 103.03 min, sonication power of 342.69 W, and solvent-to-material ratio of 94.00 mL/g.

Combined Effect of Solvent Concentration, Power, and Extraction Time
The data points on each ramp shown in Figure 4 show the anticipated response of the best choice for extracting the maximum amount of the two compounds. The optimal settings are attained at the extraction time of 103.03 min, sonication power of 342.69 W, and solvent-to-material ratio of 94.00 mL/g. RSM was employed to examine the 3D response surface graphs, which were generated by examining the impact of three UAE properties on the sum of peak areas of two compounds, 1 and 2 ( Figure 5). Figure 5A shows the effects of extraction time (A) and sonication power (B) on the total peak area of the two compounds at a constant solventto-material ratio. Sonication power was observed to have a weak effect on the yield of the two compounds. This observation is confirmed by the response surface plots shown in Figure 5C, which exhibit the impact of sonication power (B) and solvent-to-material ratio (C) on the yield of the two compounds. The plateau and curvature in Figure 5B,E show that the solvent-to-material ratio and the extraction time considerably affect extraction efficiency. RSM was employed to examine the 3D response surface graphs, which were generated by examining the impact of three UAE properties on the sum of peak areas of two compounds, 1 and 2 ( Figure 5). Figure 5A shows the effects of extraction time (A) and sonication power (B) on the total peak area of the two compounds at a constant solvent-to-material ratio. Sonication power was observed to have a weak effect on the yield of the two compounds. This observation is confirmed by the response surface plots shown in Figure 5C, which exhibit the impact of sonication power (B) and solvent-to-material ratio (C) on the yield of the two compounds. The plateau and curvature in Figure 5B,E show that the solvent-to-material ratio and the extraction time considerably affect extraction efficiency.

ANN
An ANN was used to confirm the RSM results. The ANN training algorithm used herein was trainlm, which is considered one of the fastest backpropagation algorithms in the toolbox for training the ANN [26]. The results of the comparative analysis of training, validation, and testing datasets, as well as a combined set of experimental and predicted data, are shown in Figure S8. The results demonstrate a high-level agreement between the predicted and experimental data, with all R values exceeding 0.985. The network was effectively trained, realizing a coefficient of determination of 0.99631. The regression coefficient (0.98637), which measures the relation between the output and objective, is close to 1, indicating excellent performance. Furthermore, the data predicted by the ANN agree with the experimental data, with minimal error. Therefore, this ANN model can be used to accurately predict the total amount of composites extracted from D. crossirhizoma.

Validation of the Optimal Conditions
To confirm the accuracy of the response equation and determine the optimal experimental parameters, the experiments were performed in triplicate, and the resulting data are presented in Tables 4 and 5. Both the RSM and ANN models demonstrate high predictability of the UAE parameters, with the ANN model delivering more accurate predictions than the RSM model. The degree of similarity between the predicted and experimental values was analyzed, and the ANN model demonstrated a matching value of 97.51%, compared with that of 97.15% in the RSM model. The ANN technique can capture any type of nonlinearity, overcoming the limitations of RSM, and can be developed without a predefined experimental design. The RSM technique provides regression equations for prediction and identifies insignificant parameters or interaction factors, it reduces the complexity of the problem [27]. However, overall, the difference in matching values is slight; it indicates that both techniques can predict the yield of UAE with a high degree of accuracy (matching higher than 97%).

Conclusions
Flavaspidic acid AP (1) and nortrisflavaspidic acid ABB (2), the most representative compounds from the ethanolic extract of D. crassirhizoma, have been successfully isolated and purified. The inhibitory activity of two secondary metabolites isolated from D. crassirhizoma against α-glucosidase was evaluated via in vitro assays, and nortrisflavaspidic acid ABB demonstrated higher inhibitory activity, indicating that this compound may potentially be used as an α-glucosidase inhibitor. When combined with the results of the previous studies, the obtained data suggest that D. crassirhizoma is a promising source for the production of functional foods for treating type 2 diabetes. A sensitive and rapid HPLC-UV method has been successfully established to quantify the amounts of these compounds in D. crassirhizoma. The parameters of the ultrasonic extraction of these compounds were optimized using RSM and ANN. The optimal conditions determined using the RSM are the extraction time of 103.03 min, sonication power of 342.69 W, and solvent-to-material ratio of 94.00 mL/g. The results predicted by the ANN demonstrated a high level of agreement with the experimental data, with all R values higher than 0.985. The RSM and ANN models were also validated. The agreement between the predicted and experimental values was higher than 97%, indicating that both models may be employed for optimizing the industrial extraction of active compounds from D. crassirhizoma.

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