Novel ternary deep eutectic solvent coupled with in-situ-ultrasound synergistic extraction of flavonoids from Epimedium wushanense: machine learning, mechanistic investigation, and antioxidant activity

Abstract

This study innovatively developed a novel ternary deep eutectic solvent coupled with in-situ-ultrasound synergistic extraction (TDES-ISUSE) method for efficient extraction of total flavonoids (TF) from Epimedium wushanense. Among 20 prepared DESs, the ternary system composed of choline bitartrate, urea, and glycerol (Chb:Ure:Gly) exhibited the highest extraction efficiency. Response surface methodology-artificial neural network-genetic algorithm (RSM-ANN-GA) optimization identified the optimal extraction parameters: water content of 32 %, vortex time of 10 min, liquid/solid ratio of 34:1 mL/g, ultrasound time of 30 min, and ultrasound power of 270 W. Under these conditions, the TF yield reached 26.39 ± 0.61 mg/g, which was significantly higher than that of conventional extraction methods. Phytochemical profiling of E. wushanense via UHPLC-Q-TOF-MS (positive/negative ion modes) unambiguously annotated 15 compounds, including 9 flavonoids; concurrently, an HPLC method was established for quantitative determination of 4 key flavonoids. FT-IR spectroscopy and nuclear magnetic resonance (NMR) confirmed the successful prepared of the novel TDES, while scanning electron microscopy (SEM) and molecular dynamics (MD) simulations elucidated the underlying extraction mechanism, revealing enhanced cell wall disruption and improved solubility of flavonoids in TDES-ISUSE. Antioxidant assays further demonstrated the potential bioactivity of the TF extract. Overall, this work provides an eco-friendly, efficient, and reproducible approach for extracting bioactive flavonoids from E. wushanense, with implications for sustainable utilization of medicinal plant resources and advancement of green extraction technologies.

1. Introduction

Epimedium wushanense, a perennial herbaceous species within the Berberidaceae family, is primarily distributed in southwestern and northwestern China, including Chongqing, Sichuan, Guizhou, and Guangxi provinces [1]. E.wushanense has been extensively employed for both dietary supplementation and therapeutic interventions, demonstrating efficacy in replenishing vital essence, enhancing physical endurance, and alleviating wind-dampness syndrome. 9 flavonoids, 3 organic acids, 1 terpenoid, 1 phenylpropanoid, and 1 alkaloid,

Phytochemical investigations reveal that E.wushanense is enriched with flavonoids, phenolic acids, organic acids, phenylpropanoid, nucleosides, terpenoids, and alkaloids [2]. Among these, flavonoids represent the predominant bioactive constituents, attributable to their intrinsic biosynthetic abundance as characteristic secondary metabolites of the Epimedium genus [3]; their relatively high solubility in commonly employed extraction solvents that are efficiently permeated and agitated by ultrasonic cavitation; the potential for enhanced flavonoid yield and selectivity via UAE mechanisms, such as cell wall disruption facilitating the release of intracellular flavonoids [4,5]. The flavonoids exhibit remarkable pharmacological properties, including potent antioxidant, antimicrobial, and antitumor activities. These multifunctional properties have propelled E.wushanense to the forefront of phytochemical and pharmacological research, garnering significant scientific attention in recent decades [6].

Conventional solvents, such as ethanol and methanol have been widely employed for flavonoids extraction. However, organic reagents present notable limitations including environmental contamination, high solvent consumption, and potential toxicity. These drawbacks have prompted increasing scholarly attention toward developing novel solvent systems [7]. Deep eutectic solvents (DESs) represent a novel class of solvents composed of a hydrogen bond acceptor (HBA) and a hydrogen bond donor (HBD) at an appropriate molar ratio [8]. DESs have many encouraging advantages, including cost-effectiveness, ease of preparation, environmental friendliness, biocompatibility, sustainability, and customizable [9]. To date, DESs have emerged as a mainstream technology for the efficient recovery of target bioactive constituents from natural matrices. Their strategic applications in phytochemical extraction are increasingly recognized for achieving superior selectivity compared to conventional solvent systems. For instance, a highly relevant and comprehensive study by Xu et al. [10] provided an extensive review of UAE-DES for agro-food waste valorization, serving as a foundational reference for this field. In the extraction of bioactives from citrus waste (orange peel), the optimized UAE-DES process (ChCl: lactic acid, 30 min, 40 °C) achieved a superior recovery of total phenolic content (TPC: 36.51 mg GAE/g), total flavonoid content (TFC: 28.66 mg QE/g), and antioxidant activity (DPPH inhibition: 74.4 %), alongside valuable proteins (7.81 mg/g) and lipids (5.58 mg/g), demonstrating a holistic valorization approach far exceeding the efficiency of conventional solvents like ethanol [11]. Furthermore, the work of Chaudhary et al. [12] demonstrated the successful valorization of orange peel, using UAE with DES (ChCl-lactic acid). Their study went beyond extraction optimization by integrating the obtained bioactive-rich extract with other natural ingredients to develop a novel, clean-label functional beverage. This approach not only achieved high extraction efficiency for polyphenols and antioxidants but also addressed the critical step of utilizing these extracts in a value-added end product. Building on prior successes in utilizing deep eutectic solvents for plant bioactive extraction [13], this study engineered a novel ternary DES system based on choline bitartrate (choline bitartrate, urea and glycerol, 1:2:2 M ratio), which demonstrated superior extraction efficiency compared to the other DESs.

Notably, this study pioneers a novel dual-mechanism strategy integrating in situ extraction with ultrasonication. Ultrasonication serves as a pivotal driver in advancing sustainable green chemistry and extraction protocols, leveraging cavitation-induced mechanical shear forces to enhance mass transfer efficiency [14]. This method achieved rapid extraction, while demonstrating high reproducibility, reduced solvent consumption, streamlined processing steps, and enhanced efficiency. The crux of this process lied in the cavitation effect, a phenomenon driven by ultrasonic-induced pressure oscillations that generate alternating compression [15]. These pressure fluctuations induced the formation and subsequent collapse of microbubbles within the liquid medium, generating localized high-energy jets and shear forces that disrupt cellular ultrastructure, thereby enhancing the solubilization and release of target bioactive compounds [16]. Consequently, this synergistic extraction mechanism facilitates rapid and efficient liberation of phytochemical constituents from botanical matrices.

Considering the intricate higher-order synergies and non-linear interdependencies among extraction parameters, conventional Response Surface Methodology (RSM) may exhibit suboptimal predictive accuracy. In order to transcend this limitation, a hybridized computational framework integrating RSM, artificial neural networks (ANN), and genetic algorithms (GA) were engineered [17].

This study innovatively employed a ternary deep eutectic solvent coupled with in-situ-ultrasound synergistic extraction (TDES-ISUSE) strategy to extract total flavonoids from E.wushanense. The systematic investigation comprised seven key phases: first, 20 DESs were rigorously evaluated through comparative extraction efficiency assays. A hybrid RSM-ANN-GA model was constructed to precisely identify critical parameter thresholds and the novel approach was compared with the traditional approaches. Furthermore, UHPLC-Q-TOF-MS was first employed to characterize principal phytoconstituents. An HPLC method was established to determine four flavonoids in E.wushanense. Moreover, the extraction mechanisms were elucidated via FT-IR, NMR, SEM, and molecular dynamics simulations. Finally, the antioxidant activity of E.wushanense extract was investigated. In summary, this work first established the ISUSE extraction strategy in conjunction with the novel TDES, offering a straightforward yet robust alternative for natural product extraction. Moreover, it addressed the escalating demand for sustainable and green chemical solutions within the realm of natural product chemistry.

2. Materials and methods

2.1. Materials and reagents

The E.wushanense was acquired from Guanyang Town, Wushan, Chongqing in July 2024, which was identified by associate researcher YiQuan Zhou. Dried E.wushanense was powdered and sieved with a 40 mesh sieve. 1,1-Dipheny-2-picrylhydrazyl (DPPH) and 2,2′-Azino-bis (3-ethylbenzothiazoline-6-sulfonic acid) (ABTS), Choline bitartrate, Choline dihydrogencitrate, butane-1,4-diol, xylitol, malic acid, ethylene glcol, lactic acid, urea, epimedin A(EA), epimedin B(EB), epimedin C(EC), and icariin (IA) were procured from Shanghai Yuanye Biotechnology Co.

2.2. Preparing and screening the DESs

Based on prior research [18], 24 DESs were prepared in this study. Each DES was prepared by mixing a HBA and a HBD, followed by continuous stirring at 70 °C for 1–2 h until a homogeneous mixture formed. If necessary, a controlled volume of water was added [19]. Detailed compositions and molar ratios of the DESs are summarized in Table 1.

Table 1.

The formula of various DESs.

No.	Abbreviation	HBA:HBD	Molar ratio
DES-1	Cdc:Mal	Choline dihydrogen citrate: Malic acid	1:2
DES-2	Cdc:Lac	Choline dihydrogen citrate: Lactic acid	1:2
DES-3	Cdc:Gly	Choline dihydrogen citrate: Glycerol	1:2
DES-4	Cdc:XyL	Choline dihydrogen citrate: Xylitol	1:2
DES-5	Cdc:Ace	Choline dihydrogen citrate: Acetic acid	1:2
DES-6	Chb:Ure	Choline bitartrate: Urea	1:2
DES-7	Chb:Lac	Choline bitartrate: Lactic acid	1:2
DES-8	Chb:Mal	Choline bitartrate: Malic acid	1:2
DES-9	Chb:Ace	Choline bitartrate: Acetic acid	1:2
DES-10	Chb:Xyl	Choline bitartrate: Xylitol	1:2
DES-11	Cdc:But:Gly	Choline dihydrogen citrate:1,4-Butanediol: Glycerol	1:2:2
DES-12	Cdc:Pro:Gly	Choline dihydrogen citrate: Propylene glycol: Glycerol	1:2:2
DES-13	Cdc:Eth:Gly	Choline dihydrogen citrate: Ethylene glycol: Glycerol	1:2:2
DES-14	Cdc:Mal:Ure	Choline dihydrogen citrate: Malic acid: Urea	1:2:2
DES-15	Cdc:Lac:Gly	Choline dihydrogen citrate: Lactic acid: Glycerol	1:2:2
DES-16	Cdc:Ace:Gly	Choline dihydrogen citrate: Acetic acid: Glycerol	1:2:2
DES-17	Chb:Ure:Gly	Choline bitartrate: Urea: Glycerol	1:2:2
DES-19	Chb:But:Gly	Choline bitartrate:1,4-Butanediol: Glycerol	1:2:2
DES-20	Chb:But:Lac	Choline bitartrate:1,4-Butanediol: Lactic acid	1:2:2

3.0 g of E.wushanense powder was mixed with 60 mL DES and TDES containing 30 % (v/v) water. Subsequently, this sample was vortex-mixed 10 min, followed by ultrasonication for 30 min (ultrasound power: 270 W, frequency: 40 kHz, and temperature: 33 °C) [8]. Finally, the mixture was filtered and centrifuged (4000 rpm, 10 min) using a centrifuge (80–2, Jiangsu Jinchengguosheng Co., Ltd., China) to obtain the supernatant.

2.3. Analytical methods

2.3.1. Determination of total flavonoids content

The total flavonoids (TF) content of E.wushanense was quantitatively determined by ultraviolet spectrophotometry. Rutin was employed as the standard, and a calibration curve (Y = 12.124X − 0.0316, R² = 0.9994) was established over the concentration range of 40 to 200 μg/mL, which demonstrated excellent linearity. The TF content was denoted as milligrams of rutin equivalents per gram of dried E. wushanense (mg/g).

2.3.2. HPLC analysis

The four flavonoids, including epimedin A(EA), epimedin B(EB), epimedin C(EC), and icariin (IA)were separated using an Agilent 1220 HPLC system equipped with a Supersil ODS-B C18 column (5 μm, 250 × 4.6 mm). The gradient elution program was detailed in Table 2. Chromatographic conditions were 25 °C, 1 mL/min and 10 μL. Fig. 1 displayed the chromatographic peaks of EA, EB, EC, and IA, respectively.

Table 2.

HPLC gradient elution.

Time (min)	Ultrapure water (%)	Acetonitrile (%)
0	76	24
40	74	26

Fig. 1.

The four flavonoids from mix standards (a) and E.wushanense extract (b). The peaks marked with 1, 2, 3, 4were EA, EB, EC and IA, respectively.

2.3.3. Identification of E. wushanense extract

The chemical profiling of E. wushanense extract was performed using ultra-high-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry (UHPLC-Q-TOF-MS). The analysis was conducted on an Agilent 1290 UHPLC system interfaced with an Agilent 6546 Q-TOF high-resolution mass spectrometer, equipped with a Waters ACQUITY UPLC HSS T3 column (2.1 × 100 mm, 1.8 μm). The mobile phase consisted of acetonitrile (A) and 0.1 % formic acid in water (B), with a gradient elution program detailed in Table S1. Chromatographic separation was achieved at a column temperature of 35 °C, with a detection wavelength of 254 nm and an injection volume of 1 μL. The electrospray ionization (ESI) source was operated in both positive and negative ion modes.

2.4. Extraction optimization

2.4.1. Single factor experiments

To ensure efficiency, only primary factors influencing bioactive compounds were prioritized. As detailed in Table S2, five critical elements were systematically investigated to identify the ideal condition combinations for maximizing the yields of TF yields. In single-factor experiments, 3.0 g of E. wushanense powder was added to the optimal DES to evaluate effects of various factors on the contents of TF. Meanwhile, only one variable was altered in a same time while keeping the other conditions.

2.4.2. Response surface methodology (RSM) experiment

A Box-Behnken design (BBD) was employed to explore the interactions of critical variables. The three-factor-two-level BBD experiment was established to optimize extraction conditions for maximizing the yields of TF. The experiment took water content (20–30–40 %), vortex time (8–10–12 min), and liquid/solid ratio (20–30–40 mL/g) as the main influencing variables. The levels of 17 variables were listed in Table 3. Additionally, the equation was as follows:

$$Y={\beta }_0+\sum _{i=1}^i\left({\beta }_i{x}_i\right)+\sum _{\mathit{ij}=1}^{\mathit{ij}}\left({\beta }_{\mathit{ij}}{X}_i{X}_j\right)+\sum _{1=1}^i\left({\beta }_{\mathit{ii}}{X}^2\right)$$ (1)

where Y: yield, ${\beta }_0$: constant, ${\beta }_{\mathit{ij}}$: coefficients and X: factor.

Table 3.

ANOVA for quadratic model.

Source	Sum of Squares	df	Mean Square	F-value	p-value	significant
Model	26.81	7	3.83	44.27	< 0.0001	significant
A	2.71	1	2.71	31.38	0.0003
B	0.32	1	0.32	3.7	0.0866
C	8.08	1	8.08	93.4	< 0.0001
BC	1.28	1	1.28	14.76	0.004
A2	0.3266	1	0.3266	3.77	0.0839
B2	1.34	1	1.34	15.45	0.0035
C2	11.93	1	11.93	137.93	< 0.0001
Residual	0.7786	9	0.0865
Lack of Fit	0.4943	5	0.0989	1.39	0.3856	not significant
Pure Error	0.2843	4	0.0711
Cor Total	27.59	16

2.5. Response surface methodology-artificial neural network-genetic algorithm (RSM-ANN-GA) modeling

While RSM remained a cornerstone of extraction optimization, its quadratic regression-based modeling inherently constrains capturing complex nonlinear interactions pivotal to ultrasonic-assisted systems [20]. To address this, we developed an RSM-ANN-GA hybrid framework: Box-Behnken design under RSM generated systematic training data; the multi-layer perceptron architecture of artificial neural networks (ANN) modeled TF yield with superior nonlinear fidelity; GA navigated ANN-predicted response surfaces for global optimization. This integration preserves the experimental rigor of RSM while markedly enhancing predictive accuracy and optimization robustness [21].

In this study, a sophisticated backpropagation artificial neural network (BP-ANN) model was constructed through systematic optimization using a GA integrated within the Toolbox™ of MATLAB R2024a. The model was rigorously trained on a dataset generated via RSM, which encapsulated the multivariate relationships between extraction parameters and target compound yields. The training protocol was designed to incorporate three critical input variables: water content (w/w, %), vortex time (min), and liquid/solid ratio (mL/g). These input parameters were specifically tailored to predict the key output variable, namely the extraction yield of TF.

A meticulously designed 3-9-1 ANN architecture—comprising 3 input neurons, 9 hidden neurons, and 1 output neuron, was developed under the guidance of minimizing the mean squared error (MSE) function, as delineated in Fig. 2. This configuration was determined through systematic sensitivity analysis to balance model complexity and generalization capability. The Levenberg-Marquardt algorithm was employed for network training due to its superior computational efficiency in error surface navigation. For robust model development, the total RSM-generated samples were strategically partitioned: 70 % were allocated to the training phase to enable the network to learn intrinsic patterns and non-linear correlations within the data; 15 % were reserved for validation, and the remaining 15 % were set aside for testing. The optimal ANN model was identified based on the lowest MSE and highest correlation coefficient (R). Subsequent GA optimization was implemented to refine model parameters. This GA-driven refinement further enhanced the model’s predictive accuracy and stability [17].

Fig. 2.

Structure diagram of BP neural network model.

2.6. Comparison of traditional methods

2.6.1. Ethanol assisted in-situ-ultrasound synergistic extraction (ISUSE)

3.0 g of raw material was combined with 60 mL ethanol. Subsequently, this sample was vortex-mixed for 10 min, followed by ultrasonication for 30 min (ultrasound power: 270 W, frequency: 40 kHz, and temperature: 33 °C). Finally, the mixture was filtered and centrifuged (4000 rpm, 10 min) using a centrifuge (80–2, Jiangsu Jinchengguosheng Co., Ltd., China) to obtain the supernatant. The extract was subsequently analyzed via UV to quantify the content of TF.

2.6.2. Ultrasound-assisted ethanol extraction (UAE)

3.0 g of raw material was mixed with 60 mL of ethanol. This sample was subjected to ultrasonic-assisted extraction at 270 w for 30 min (frequency: 40 kHz, and temperature: 33 °C), and then centrifugated (4000 rpm,15 min) to obtain the supernatant. The extract was subsequently analyzed via UV to quantify the content of TF.

2.6.3. Maceration extraction (ME)

3.0 g of raw material was weighed into a 100 mL flask and mixed with 60 mL of ethanol. This sample was subjected to extraction in soaking at 25 ℃ for 8 h and then centrifugated (4000 rpm,15 min) to obtain the supernatant. The extract was subsequently analyzed via UV to quantify the content of TF [22].

2.7. Extraction mechanism

2.7.1. FT-IR spectroscopy

To investigate the hydrogen bond formation characteristics of the novel TDES, the optimal DES formulation was analyzed using a Fourier transform infrared (FTIR) spectrometer within the wavenumber range of 4000–400 cm⁻¹ [13].

2.7.2. SEM

The morphological characteristics of the processed E.wushanense powder were observed adopting SEM. Comparative analysis was performed across four experimental conditions: (1) powder obtained via TDES-ISUSE, (2) powder obtained via ISUSE, (3) powder obtained via UAE (4) powder obtained via ME [13].

2.7.3. Molecular dynamics (MD) analysis

MD simulations, a cornerstone of computational chemistry, enabled the exploration of temporal evolution and intermolecular interactions within atomic-scale systems [23]. These simulations were grounded in the framework of classical mechanics, utilizing Newtonian equations of motion to predict particle trajectories under predefined energy potentials. Comprehensive simulation parameters were archived in the Supplementary Material [24,25].

2.8. Antioxidant activity

2.8.1. DPPH

Refer to the research of Lei et al [26]. Briefly, sample solutions were prepared to achieve five concentrations (0.1–0.5 mg/mL). After that, 2 mL sample was added to 2 mL freshly prepared DPPH solution (0.15 mmol/L in methanol). The reaction mixture was incubated in darkness for 30 min and detected at 517 nm. The activity was calculated as follows:

$$\text{S}\text{cavenging rate}\left(\text{\%}\right)=\left[\frac{\text{Ac-As}}{\text{Ac}}\right]\text{*100\%}$$ (2)

where Ac represented the blank control, As represented the sample.

2.8.2. ABTS

Refer to the research of Liu et al [27]. Briefly, 250 mL ABTS solution (7 mmol/L) was added to 250 mL potassium persulfate solution (2.45 mmol/L). The mixture was incubated in darkness 16 h. The solution was freshly prepared by diluting the stock with ethanol to an absorbance of 0.70 ± 0.02 at 734 nm using a UV–Vis spectrophotometer. For the assay, 0.5 mL of sample solution (0.1–0.5 mg/mL) was reacted with 3 mL of ABTS·⁺ working solution. The reaction mixture was maintained in darkness for 6 min, and detected at 734 nm. The activity was calculated as follows:

$$\text{S}\text{cavenging rate}\left(\text{\%}\right)=\left[\frac{\text{A}\text{0}\text{-A}\text{1}}{\text{A}\text{0}}\right]\text{*100\%}$$ (3)

where A0 represented the blank control, A1 represented the sample.

2.8.3. Ferric ion reducing assay

Refer to the research of Jiang et al [28]. Briefly, the sample solutions were prepared at five concentrations (0.1–0.5 mg/mL). Subsequently, 3 mL of each sample was mixed with 2 mL of sodium phosphate buffer (0.2 mol/L, pH 6.6) and 2 mL of potassium ferricyanide solution (1 % w/v). The mixture was incubated in a water bath(50 °C, 30 min). After reaction termination, 2 mL of trichloroacetic acid (10 % w/v) was added to the solution, followed by centrifugation at 3000 rpm for 10 min. A 2 mL aliquot of the supernatant was combined with 0.1 mL of ferric chloride solution (0.1 % w/v) and 2 mL water. This sample was allowed to react for 10 min and detected at 700 nm.

2.9. Statistical analysis

The RSM results were statistically analyzed using one-way analysis of variance (ANOVA). Data are expressed as mean ± standard deviation (SD), with significance levels denoted as ***p < 0.001, **p < 0.01, and *p < 0.05.

3. Results and discussion

3.1. Selecting the optimal DES

20 dess were rationally designed and prepared adopting choline dihydrogen citrate and choline hydrogen tartrate as key hydrogen bond acceptors (HBAs) to extract total flavonoids from E.wushanense, encompassing 10 binary DESs and 10 ternary DESs. As depicted in Fig. 3, distinct extraction efficiencies for the total flavonoids were revealed by systematic screening of all 20 DESs (30 % water content). The DES composed of Choline bitartrate, urea and Glycerol (Chb:Ure:Gly) demonstrated superior performance. Notably, ethanol exhibited the extraction capacity of ethanol could reach 22.24 ± 1.21 mg/g. This observation aligns with the “like dissolves like” principle [29]. On the whole, Chb:Ure:Gly was the optimal solvent to extract total flavonoids from E.wushanense, so it was employed for subsequent process optimization and mechanistic studies

Fig. 3.

TF yields of E.wushanense extracted with DESs.

3.2. Single factor experiments

3.2.1. Effect of water content

The rheological properties of DESs, particularly viscosity, critically govern their extraction performance by modulating mass transfer dynamics. Generally, the viscosity of DESs decreased with increasing water content. Optimal water content enhanced the content of TF. Accordingly, the effects of water contents in the optimal DES were investigated to extract the TF.

After systematic tuning of the water content, a non-linear correlation between aqueous content and the TF from E.wushanense was revealed. Specifically, as shown in Fig. 4A, the TF increased significantly as the water content in the DES rose from 20 % to 30 %. This enhancement was likely attributed to the reduced solution viscosity caused by higher water content, thereby improving the extraction efficiency of flavonoids [30,31]. Nonetheless, the exceed water content (40–50 %) led to a marked decline in the TF, presumably due to excessive water in the DES system, which may weaken the intermolecular interactions between flavonoids and DES components, thereby limiting their solubilization [32].

Fig. 4.

Single factor experiments.

3.2.2. Effect of vortex time and other factors

Vortex facilitated rapid homogenization of herbal powder and solvent, while accelerating the dissolution of target components, thereby enabling the achievement of a relatively high extraction yield within a short timeframe [20]. As depicted in Fig. 4B, the TF yield increased with the extension of extraction time, peaking at 10.0 min—an observation indicative of the rapid initial diffusion of target compounds. With further prolongation of vortex duration, substantial impurities were likely eluted, which might have interfered with the dissolution of TF and consequently led to a decline in its content.

The liquid/solid ratio modulated the diffusion behavior of solutes in the solvent system. An increased liquid/solid ratio could enhance extraction efficiency by maximizing the driving force. However, it concurrently elevated solvent consumption and promoted the co-extraction of impurities. As illustrated in Fig. 4C, the TF yield increased with the increment of the liquid/solid ratio, reaching a maximum value at 30 mL/g. Nevertheless, a further increase in the liquid/solid ratio resulted in a decline in TF yield, which might be ascribed to the excessive contact between the solvent and medicinal herb matrix, thereby triggering the massive dissolution of interfering impurities. In addition, excessive solvent volumes attenuated the propagation of ultrasound energy, which in turn reduced the disruption of plant cell walls and ultimately decreased the extraction efficiency [33].

Ultrasound time also exerted a critical influence on the TF yield, as ultrasound required sufficient time to achieve thorough disruption of plant cell walls. As depicted in Fig. 4D, the TF yield increased progressively with the extension of ultrasound time and attained a relatively high plateau at 30 min. However, further prolongation of the ultrasound duration resulted in negligible changes in the TF yield. This could be attributed to the fact that the plant cell wall structure had been adequately disrupted and the solvent had reached a mass transfer equilibrium with the target compounds [8]. Consequently, additional extension of the ultrasound time failed to elicit a significant enhancement in extraction efficiency.

The intensity of ultrasound cavitation was dependent on the applied power level, with an optimal power regime facilitating efficient and targeted disruption of the intricate plant cell wall matrix. As depicted in Fig. 4E, the TF yield increased concomitantly with the elevation of ultrasound power, attaining a maximum value at 270 W. Upon further increments in power, the yield exhibited negligible variation, a phenomenon that could be ascribed to the ultrasonic power having reached a saturation threshold, where the cavitation intensity had already maximized the disruption of cell wall structures and subsequent mass transfer of flavonoids [8].

During this experiment, all other experimental parameters were maintained at a constant state [24,34]. Based on the results shown in Fig. 4A–E, the TF yields were significantly influenced by these variables. Consequently, water content (20–40 %), vortex time (8–12 min) and liquid/solid ratios of (20–40 mL/g) were selected for following experiments.

3.3. RSM analysis

A BBD was adopted to optimize the three critical variables: water content, vortex time, and liquid/solid ratio. The DesignExpert software was employed to design the experiments.

The variables and responses were shown in Table S3. ANOVA analysis for model prediction was presented in Table 3, which indicated that the model was statistically significant (P < 0.001). Furthermore, the coefficient of determination (R²) for the response variable was 0.9718, with an adjusted R² (Adj. R²) of 0.9498, demonstrating high predictive accuracy of the model for future outcomes. The lack-of-fit test yielded a non-significant result (P > 0.05), indicating the predictive model adequately represented the experimental data [35]. To enhance the regression of the optimized model, the negligible terms with p-values greater than 0.1 in the model were removed [36].

Furthermore, the following equation represents the quadratic model for total flavonoid yield, with less significant terms excluded based on coded values [37].

Y = 25.73 + 0.5825A + 0.2B + 1.01C + 0.565BC-0.2785A²-0.5635B²-1.68C²

As illustrated in Fig. 5, the 3D response surface plots visually elucidated the impacts of variables on the TF yield. It could be inferred that water content of 36.47 %; vortex time of 11.13 min, liquid/solid ratio of 35.35:1 mL/g were selected. Simultaneously, the optimal content was 26.23 mg/g. Fore more scientific operation, the conditions were adjusted (water content of 36 %; vortex time of 11 min, liquid/solid ratio of 35:1 mL/g).

Fig. 5.

Response surface plots.

As illustrated in Fig. 5, the total flavonoid (TF) yield exhibited an initial upward trend with increasing vortex duration and liquid-to-solid ratio, peaking at an optimal point before declining thereafter. Moderate extensions of vortex time and elevations in liquid-to-solid ratio enhanced ultrasonic cavitation effects, thereby facilitating cell wall disruption, reducing the viscosity of the deep eutectic solvent (DES), and accelerating mass transfer dynamics, which collectively promoted TF dissolution. However, excessive vortex duration and overly high liquid/solid ratio led to the co-extraction of substantial impurities, which competitively inhibited TF elution. Notably, the influence of liquid-to-solid ratio on yield was more pronounced than that of vortex time, highlighting its more critical role in governing the extraction efficiency.

3.4. RSM-ANN-GA analysis

A BP-ANN was trained using a dataset derived from RSM analysis, incorporating three critical factors: water content, vortex time, and liquid/solid ratio—as input parameters, with the extraction yield of TF designated as the output variable. The hyperbolic tangent function (tansig) was employed as the activation function bridging the input and hidden layers, while a linear function served as the transfer function linking the hidden layer to the output layer. The Levenberg–Marquardt backpropagation algorithm (trainlm) was utilized as the training function, and the network was meticulously calibrated through a protocol comprising 1000 epochs, with 25 iterations per epoch, targeting a training error of 0.0001. As illustrated in Fig. S1, the optimal validation performance was achieved at epoch 4, with a MSE of 0.0033553. As shown in Fig. 6, the correlation coefficients (R) associated with the training, validation, testing, and aggregated datasets all exceeded the 0.90, demonstrating the BP-ANN model’s notable predictive potential across training, validation, and testing samples.

Fig. 6.

Training, validation, testing and fit of all data to the BP-ANN.

To achieve output maximization through optimization, network-generated data derived from the RSM-ANN-GA integrated framework was employed as the objective function. The three independent variables were conceptualized as matrix variables, subject to boundary constraints, which are specified as follows:

[20; 8; 20] ≤ [X₁; X₂; X₃] ≤ [40; 12; 40]

The optimal extraction parameters determined through the intricate integration of RSM-ANN-GA are delineated as follows: a water content of 31.75 %, a vortex time of 10.33 min, and a liquid/solid ratio of 34.12 mL/g, yielding an optimal TF yield of 26.36 mg/g. In order to enhance the feasibility of experimental implementation, these conditions were judiciously adjusted during the validation phase to a water content of 32 %, a vortex time of 10 min, and a liquid/solid ratio of 34 mL/g.

3.5. Validation experiment

Table 4 illustrated that both the RSM and RSM-ANN-GA models exhibited strong predictive performance for the TF yield, with only minimal discrepancies from experimentally measured values. The experimental results optimized via traditional RSM yielded 26.17 ± 1.01 mg/g, with a relative error of 0.19 % and a relative RSD of 0.69 %. In contrast, the hybrid RSM-ANN-GA model achieved an experimental yield of 26.39 ± 0.61 mg/g, with a lower relative error of 0.12 % and a significantly reduced RSD of 0.24 %. These findings underscore the hybrid model’s superior accuracy and stability compared to the standalone RSM approach, attributed to its enhanced capacity to capture complex.

Table 4.

Validation experiments.

	Water content (%)	Vortex time (min)	Liquid/solid ratio (mL/g)	Experimental value	RSD(%)
RSM	36	11	35:1	26.17 ± 1.01	0.69
RSM-ANN-GA	32	10	34:1	26.39 ± 0.61	0.24

3.6. Comparation with traditional methods

A significant comparative assessment was conducted to validate the superior efficacy of TDES-ISUSE extraction in enhancing the TF from E. wushanense. The TF obtained via this novel method were systematically benchmarked against those achieved through conventional techniques, including ISUSE, UAE and ME [27]. As evidenced in Fig. 7, the TDES-ISUSE demonstrated dual advantages over ME: a significant reduction in extraction time and an apparent enhancement in the TF yields. Moreover, the TF in TDES-ISUSE was higher than the value in ISUSE (22.24 ± 1.21 mg/g) and UAE (16.21 ± 1.37 mg/g). The results demonstrated that the TDES-ISUSE combined rapid processing, high efficiency, and environmental sustainability, exhibiting significant potential for optimizing the TF from E. wushanense.

Fig. 7.

Comparation with four methods.

3.7. UHPLC-Q-TOF MS analysis

To date, UHPLC-Q-TOF-MS has been well-established as a reliable and high-throughput technique for the identification of bioactive components in natural products, leveraging its superior sensitivity, high-resolution separation capability, and accurate mass measurement performance [38]. For the first time, the UHPLC-Q-TOF-MS strategy was implemented in this study to conduct chemical profiling of E. wushanense extracts, with the objective of elucidating its bioactive components. As illustrated in Fig. 8, total ion current (TIC) chromatograms were acquired under both positive and negative ion modes. By matching MS fragments against the PCDL MS database and integrating with references, a total of 15 compounds were identified, encompassing 9 flavonoids, 3 organic acids, 1 terpenoid, 1 phenylpropanoid, and 1 alkaloid, as detailed in Table 5. Among these, compounds 8 to 11 were designated as EA, EB, EC, and IA, respectively, with EC being the predominant component. This endeavor was anticipated to lay a solid scientific basis for the product development and bioactivity-oriented investigations of E. wushanense.

Fig. 8.

UHPLC–Q-TOF-MS extracted ion chromatograms of the compounds from S E.wushanense. positive (A) and negative(B) ion modes.

Table 5.

compounds were characterized using UHPLC-Q-TOF MS.

Peak	RT(min)	Model	Compound	Molecular formula	Adduct (m/z)	Major fragment ion	References
1	8.870	+	chlorogenic acid	C16H18O9	355.1029	163.0392,145.0282,135.0439,117.0333	[39]
2	9.696	−	Neochlorogenic acid	C16H18O9	353.0884	173.0457,135.0452,111.0451	[40]
3	12.601	−	3-O-p-coumaroylquinic acid	C16H18O8	337.0935	191.0563,163.0402,119.0502	[41]
4	13.086	+	Magnoflorine	C20H24NO4	342.1707	297.1119,265.0855,191.0851	[42,43]
5	19.551	+	quercetin 3-O-glucoside	C21H20O12	465.1032	302.0501,285.0389,145.0495	[44]
6	23.721	+	Eugenol rutinoside	C22H32O11	490.229	351.1022,349.1262,331.1007	[45]
7	28.694	+	diphylloside B	C38H48O19	809.2878	663.2291,517.1710,335.1175,129.0547	[46,47]
8	33.853	+	Epimedin A	C39H50O20	839.2984	677.2442,531.1863,369.1329,129.0544	[48]
9	34.198	+	Epimedin B	C38H48O19	809.2888	677.2451,531.1871,369.1335,313.0706,255.0759	[48]
10	34.626	+	Epmedin C	C39H50O19	823.3036	677.2451,531.1872,369.1334,313.0704,129.0546	[48]
11	34.907	+	Icariin	C33H40O15	677.2468	531.1883,369.1342,313.0710,129.0549	[48]
12	42.613	−	Baohuoside VII	C33H40O15	675.2294	367.1187,351.0884,323.0926	[49,50]
13	43.743	+	Epicornuin F	C25H26O7	439.1759	421.1642,403.1541,367.1177	[51,52]
14	43.973	−	Baohuoside I	C27H30O10	513.1773	366.1111,323.0935,217.0512	[47,53]
15	48.473	+	epimedokoreanin B	C25H26O6	423.1804	405.1691,349.1060,261.2208,127.0388	[52,54]

3.8. Method validation

Based on the results of HPLC, four flavonoids (EA, EB, EC and IA) were discovered in E. wushanense. Quantitative analysis of the aforementioned compounds was conducted under optimized conditions to precisely evaluate the quality of E. wushanense extracts. Calibration curves were constructed using standard solutions to assess linearity, with the linear range and correlation coefficient (R²) determined to validate the analytical reliability of the method [55].

The standard curves of EA, EB, EC and IA were Y_EA = 32455x + 36.861.

(R² = 0.9994), Y_EB = 21976x + 21.8214 (R² = 0.9996), Y_EC = 23452x + 26.3687 (R² = 0.9993), Y_IA = 30547x + 15.4277 (R² = 0.9998), respectively. As presented in Table 6, the four flavonoids exhibited excellent linearity over the range of 3.12–100.00 μg/mL, with the R² exceeding 0.999. Furthermore, the LODs and LOQs for these compounds were determined to be 0.33–0.46 μg/mL and 1.01–1.37 μg/mL.

Table 6.

Method validation.

Analyte	LOD(μg/ mL)	LOQ(μg/mL)	Intraday precision	Interday precision	Stability	Repeatability	Recovery
EA	0.38	1.14	0.16	1.72	1.65	1.03	99.06
EB	0.46	1.37	0.63	1.15	1.30	1.25	99.41
EC	0.41	1.21	0.39	1.09	1.47	1.72	98.20
IA	0.33	1.01	0.64	1.43	1.68	1.46	99.84

Intraday precision was evaluated through six consecutive analyses of the same sample within a 24-hour period, while inter-day precision was determined by performing the assay over six consecutive days. As presented in Table 6, the RSD values for both intraday and interday precision were below 2 %, unequivocally demonstrating the exceptional precision of the novel methodology. The stability of samples was systematically assessed under 25 °C at 0, 3, 6, 9, and 12 h. The results demonstrated exceptional stability over 12 h with RSD values consistently below 2 %.

Six identical samples were analyzed to assess method repeatability, yielding RSD values ranging from 1.03 % to 1.72 %, which conclusively validates the high reproducibility of the proposed methodology. To evaluate recovery rates, a known quantity of the four flavonoids was spiked into 3 g of the raw material, and the recovery efficiency was calculated in Table 6.

3.9. Extraction mechanism

3.9.1. Characterization of DES

TDESs are formulated as eutectic mixtures comprising one HBA and two HBDs. In this study, fourier transform infrared spectroscopy (FT-IR) was employed to characterize the molecular interactions and structural features of the individual components (choline bitartrate, urea, glycerol) and the synthesized TDESs. The FT-IR spectra (Fig. 9A) revealed characteristic peaks corresponding to various functional groups in the system: –OH (3,500–3,200 cm⁻¹), C-H (3,000–2,500 cm⁻¹), and C-O (1,112–1,000 cm⁻¹). In the synthesized TDESs, the C-N stretching vibration peak of choline bitartrate (originally at 1085 cm⁻¹) persisted but shifted slightly to 1044 cm⁻¹. Furthermore, compared with choline bitartrate, urea, and glycerol, the –OH vibrational absorption peak in TDESs broadened and underwent a blue shift, confirming the formation of extensive hydrogen bonds within the system. Collectively, these FT-IR spectral features, including the shifted C-N peak, broadened and blue-shifted –OH band, and preservation of key functional group signatures-provided direct evidence for the successful formation of TDESs.

Fig. 9.

FT-IR spectra (A)and ¹H NMR spectrum (B)of choline bitartrate, urea and glycerol and TDES.

As presented in Fig. 9B, ¹H NMR spectroscopy offered compelling evidence for the formation of the TDES. All characteristic proton signals corresponding to the individual components were retained in the TDES spectrum, with no new peaks emerging at distinct chemical shifts that would indicate the formation of covalent bonds. This spectral feature confirmed that the TDES was constructed through non-covalent intermolecular interactions, primarily hydrogen bonding between the hydrogen bond donor and acceptor, rather than undergoing chemical reactions that would alter the intrinsic molecular structures of the constituents [56]. 

3.9.2. Scanning electron microscopy (SEM)

Vortex induced turbulent mixing at the solid–liquid interface, thereby augmenting mass transfer kinetics and expediting the detachment of target analytes from the plant matrix [57]. Furthermore, the efficacy of ultrasonic extraction is primarily attributed to the cavitation effect: a phenomenon driven by ultrasound-induced pressure oscillations, characterized by alternating cycles of compression and rarefaction in the liquid medium [58,59]. Hence, the present study integrated vortexing with ultrasonic extraction to achieve synergistic extraction. This combined approach leverages the turbulent mixing of vortexing (enhancing solvent penetration) and the mechanical shear of ultrasonic cavitation (intensifying cellular rupture), resulting in more thorough cellular disruption as visually corroborated by microscopic observations in Fig. 10A.

Fig. 10.

Microstructure of E.wushanense powders treated by different methods (A: TDES-ISUSE, B: ISUSE, C: UAE and D: ME).

Cellulose, the primary structural component of plant cell walls, is efficiently disrupted and solubilized by DESs through selective hydrogen-bond interactions, outperforming conventional solvents in both dissolution capacity and biomass pretreatment efficacy. DESs were composed of HBAs and HBDs, which could interact synergistically with cellulose chains to markedly enhance the dissolution process. This cooperative action disrupts the hydrogen-bond network within the cellulose structure, loosening its dense crystalline framework and ultimately enabling complete solubilization. Comparative analysis of Fig. 10A-D revealed that the TDES-ISUSE treatment induced significantly greater disruption of plant cell walls compared to conventional methods, thereby enhancing the extraction efficiency of marker bioactive compounds [60].

3.9.3. Molecular dynamics simulation analysis

To elucidate the mechanism of flavonoid extraction from E.wushanense via TDES-ISUSE, molecular dynamics simulations were conducted using the most effective TDES (Chb:Ure:Gly) and a conventional solvent (ethanol), focusing on their molecular interactions with EC, the predominant flavonoid constituent in E.wushanense. The molecular interactions between solvent molecules and target compounds critically govern solubility parameters. Accordingly, the extraction rate was enhanced [61].

Fig. 11A and B depicted the spatial distribution of EC within the TDES and EtOH systems at 50 ns, respectively. It could be concluded that EC exhibited homogeneous dispersion in TDES, whereas localized aggregation was observed in EtOH, indicating superior solubility of EC in TDES due to enhanced. Fig. 11C illustrated the temporal evolution of hydrogen-bond interactions between EC and the TDES, as well as between EC and EtOH, over the 0–50 ns. The results demonstrate a significantly higher number of hydrogen-bond interactions between EC and the TDES compared to those between EC and EtOH, with the TDES system exhibiting markedly reduced fluctuation and enhanced stability [62]. Furthermore, as depicted in Fig. 11D, the hydrogen-bond lifetimes between EC and the TDES are significantly prolonged compared to those between EC and EtOH, demonstrating enhanced stability of hydrogen-bond interactions in the DES system [27]. In summary, molecular dynamics (MD) simulations conclusively demonstrated the enhanced solubility of EC in TDES.

Fig. 11.

Snapshots of EC dissolved in TDES (A) and EtOH solution (B); the average lifetime of hydrogen bonds (C) and average lifetime of hydrogen bonds (D) the average lifetime of hydrogen bonds.

3.10. Antioxidant activity

To explore natural antioxidants, three oxidative assays were employed to comprehensively evaluate the in vitro antioxidant activity of E.wushanense extracts. As demonstrated in Fig. 12A and 12B, the extract demonstrated significant DPPH⁺ and ABTS⁺ radical scavenging capacities within the tested range (0.1–0.5 mg/mL). At 0.5 mg/mL, the E.wushanense extract exhibited maximum scavenging percentages of 86.66 %±1.84 % and 98.61 %±1.31 % in the DPPH⁺ and ABTS⁺ assays, respectively. Furthermore, both positive control (ascorbic acid) demonstrated scavenging percentages exceeding 90 % [63]. The reducing power assay was typically employed to measure the ability of ferric complexes to be reduced to ferrous ions. As demonstrated in Fig. 12C, it could be concluded that the absorbance of the extract reached 0.558 ± 0.011when the concentration reaches the maximum level of 0.5 mg/mL.

Fig. 12.

Antioxidant activities (a) DPPH; (b) ABTS; (c) reducing ability.

In a word, E.wushanense extract exhibited potential antioxidant activity, which could be further developed as an antioxidant.

4. Conclusions

In this study, a novel natural TDES was synthesized, and a new extraction method (ISUSE) was developed for the extraction of TF from E.wushanense. After systematic screening, Chb:Ure:Gly was identified as the optimal TDES. Following optimization via the RSM-ANN-GA framework, the optimal extraction conditions were determined as follows: water content of 32 %, vortex time of 10 min, liquid/solid ratio of 34:1 mL/g, ultrasound time of 30 min, and ultrasound power of 270 W, yielding a TF content of 26.39 ± 0.61 mg/g. Compared with conventional extraction techniques, TDES-ISUSE exhibited significantly superior extraction efficiency. UHPLC-Q-TOF-MS was first employed for chemical constituent identification of E.wushanense extract, leading to the unambiguous identification of 15 compounds in total, among which 9 were characterized as flavonoids. Concomitantly, a validated HPLC method was successfully established for the simultaneous determination of 4 main flavonoids in E.wushanense. Furthermore, FT-IR spectroscopy confirmed the successful synthesis of the novel TDES. SEM analysis of post-extraction plant residues provided structural evidence supporting the enhanced extraction efficiency of the TDES-ISUSE method. MD simulations further demonstrated enhanced solubility of EC in the TDES. Three in vitro antioxidant assays demonstrated that the extract exhibited potent antioxidant activity, highlighting its potential for development as functional food ingredients or pharmaceutical agents. Collectively, TDES-ISUSE has been successfully established as a novel, green extraction approach for the efficient extraction of TF from E.wushanense.

CRediT authorship contribution statement

Cheng Liu: Writing – original draft, Investigation, Funding acquisition. Jie Lei: Writing – review & editing, Methodology. Xinyu Liu: Formal analysis. Zhaolin Huang: Investigation. Ying Zhao: Project administration.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: