Ultrasound-assisted extraction of flavonoids from Blumea balsamifera using deep eutectic solvents: optimization, characterization, and antioxidant evaluation

Graphical abstract

Highlights

• •A green DES–UAE process was established for flavonoid extraction from Blumea balsamifera.
• •Choline chloride–glycerol DES (1:2, 40% water) was selected as the optimal solvent system.
• •ANN–GA optimization achieved a flavonoid yield of 3.52%, 73.40% higher than 50% ethanol heat-assisted extraction.
• •The DES-UAE extract exhibited strong DPPH/ABTS activity and protected L02 hepatocytes against H2O2-induced injury.
• •UPLC–Q–Orbitrap HRMS identified 15 flavonoids, including several newly reported in Blumea balsamifera.

Abstract

Blumea balsamifera, a traditional medicinal herb rich in flavonoids, is recognized for its strong antioxidant potential. In this study, a sustainable extraction strategy combining deep eutectic solvents (DES) and ultrasound-assisted extraction (UAE) was developed to enhance flavonoid recovery. Extraction parameters were optimized using response surface methodology (RSM) for DES screening, and further refined by an artificial neural network–genetic algorithm model (ANN-GA). Choline chloride-glycerol (ChCl-Gly, 1:2, 40% water) was selected as the optimal solvent. Under the ANN-GA conditions of 40:1 mL/g, 77 °C, 45 min, and 300 W, the total flavonoid yield reached 3.52 ± 0.01%, which was 73.40% higher than that obtained by 50% ethanol heat-assisted extraction. Factorial, kinetic, rated-power-based operational, and SEM analyses were consistent with ultrasound-associated matrix disruption and mass-transfer intensification, while FTIR and ¹H NMR supported hydrogen-bond network reorganization in the DES. Antioxidant evaluation through DPPH, ABTS, and reducing power assays showed superior activity of DES extracts, with DES-3 achieving a DPPH IC₅₀ of 0.1443 ± 0.0063 mg/mL. Total flavonoid content was positively correlated with DPPH (r = 0.82, p < 0.01) and ABTS (r = 0.75, p < 0.05) radical-scavenging activities. A cell-based H₂O₂-injury model in L02 hepatocytes further demonstrated cytoprotective effects of the DES-UAE extract. UPLC–Q–Orbitrap HRMS identified fifteen flavonoids, several of which were reported in this species for the first time. After three recovery and reuse cycles, the regenerated DES retained 90.57% of its initial extraction yield, with limited changes in viscosity, density, and principal ¹H NMR signals. These findings demonstrate that DES-UAE provides a green and efficient approach for extracting bioactive compounds from B. balsamifera, offering a sustainable foundation for natural antioxidant production.

1. Introduction

Antioxidants represent bioactive substances capable of counteracting free radicals and alleviating oxidative stress, which are pivotal in upholding human health [1]. Through scavenging reactive oxygen species (ROS), these compounds guard against cellular damage, DNA aberrations, and the progression of chronic maladies, encompassing cardiovascular diseases, cancer, and neurodegenerative disorders [2], [3], [4]. Such molecules are ubiquitously distributed across natural matrices, including fruits, vegetables, whole grains, and medicinal plant species [5]. Among the various types of natural antioxidants, flavonoids are polyhydroxy compounds with diverse structural configurations. They exhibit multiple antioxidant mechanisms, including direct free radical scavenging and modulation of cellular redox homeostasis, which effectively reduces oxidative damage [6], [7]. Due to their potent bioactivity and extensive natural availability, flavonoids have garnered significant interest within the antioxidant research domain.

Blumea balsamifera (L.) DC. (Asteraceae), a traditional medicinal herb, demonstrates a broad distribution across tropical and subtropical regions of Asia, including China, Thailand, and Vietnam [8]. It has been historically utilized in Chinese medicine for treating rheumatism, traumatic injuries, colds, and inflammatory skin conditions [9], [10]. Recent phytochemical investigations have identified volatile oils, flavonoids, sesquiterpenes, and phenylpropanoids as its principal bioactive constituents [9], [10], [11]. Flavonoids, in particular, represent a critical class of secondary metabolites in B. balsamifera, encompassing dihydroflavones, dihydroflavonols, flavones, flavonols, flavanols, chalcones, pterocarpans, and flavonoid dimers [12]. Pharmacological studies have demonstrated its antitumor, antioxidant, anti-inflammatory, hypoglycemic, and antimicrobial properties [10], [13], [14], [15], [16], [17], [18]. Notably, preliminary research by our group and others has highlighted its remarkable antioxidant capacity, likely attributed to its high flavonoid and polyphenol content, where phenolic hydroxyl groups efficiently scavenge free radicals [19], [20], [21], [22].

Flavonoids have attracted extensive attention in the fields of medicine, food, and cosmetics due to their diverse bioactivities. Consequently, extracting flavonoids from B. balsamifera is an important approach to harness these beneficial compounds. However, conventional extraction techniques typically rely on organic solvents such as ethanol or methanol, which raise concerns regarding safety, environmental pollution, and energy efficiency [23], [24]. These challenges have driven the exploration of sustainable alternative methodologies aligned with green chemistry tenets. Deep eutectic solvents (DES), composed of hydrogen bond donors (HBD) and hydrogen bond acceptors (HBA), have emerged as promising eco-friendly platforms for natural product extraction. DES systems offer distinct advantages, including tunable polarity, negligible volatility, low toxicity, cost efficiency, facile preparation, and biodegradability, positioning them as a focal area of contemporary research [25], [26], [27]. When integrated with ultrasound-assisted extraction (UAE), which enhances cellular disruption and solvent penetration through cavitation-induced shear forces [28], [29], DES technologies exhibit superior efficacy in isolating bioactive compounds such as flavonoids. This synergistic approach has been demonstrated to improve extraction yields by facilitating matrix breakdown and solute diffusion, as reported in recent studies [30], [31], [32]. For instance, Martinović et al. reported that DESs outperformed conventional solvents in extracting phenolic compounds from bilberry leaves, with betaine-urea DES yielding the highest flavonoid content and antioxidant activity [33]. Similarly, Gao et al. achieved a 1.6-fold increase in flavonoid yield from mung beans using glycol-choline chloride DES compared to 30% ethanol [34]. Sun et al. further optimized DES-UAE parameters for hawthorn flavonoids, achieving a yield of 57.82 ± 0.04 mg/g, significantly higher than ethanol-based extraction [35]. Despite these advances, no studies have yet explored DES-based flavonoid extraction from B. balsamifera.

Response surface methodology (RSM) is a widely used statistical tool for extraction optimization, enabling efficient evaluation of key variables and their interactions with a limited number of experiments [36]. However, DES-UAE may exhibit complex nonlinear behavior that can be inadequately described by conventional polynomial models. Artificial neural networks (ANN) provide a data-driven approach capable of capturing nonlinear relationships between process variables and extraction responses, and have increasingly been adopted to improve predictive accuracy and optimization robustness in natural-product extraction [37], [38].

This study pioneers an ultrasound-assisted DES strategy for flavonoid extraction from B. balsamifera and performs process optimization using RSM and ANN modeling, together with mechanistic interrogation via Scanning electron microscopy (SEM), Fourier-transform infrared spectroscopy (FTIR), and ¹H NMR. Through systematic screening of seven DES systems, we aim to establish a green and efficient extraction protocol. Ultra-performance liquid chromatography coupled with quadrupole-Orbitrap high-resolution mass spectrometry (UPLC-Q-Orbitrap HRMS) will be employed to characterize the chemical profiles of the extracts, while antioxidant potential will be evaluated using DPPH, ABTS, and ferric-reducing assays, together with an H₂O₂-induced injury model in L02 hepatocytes to evaluate cytoprotective effects based on cell viability. Correlation analyses will further elucidate the relationships among flavonoid yield, chemical composition, and antioxidant responses. This work provides a sustainable strategy for the valorization of B. balsamifera and contributes to the broader application of DESs in natural product extraction for nutraceutical, cosmetic, and pharmaceutical industries.

2. Materials and methods

2.1. Plant material

Dried aerial parts of B. balsamifera were procured from the Guangxi Yulin Medicinal Materials Market (China). Species identification was verified by Associate Researcher Xilong Zheng, School of Traditional Chinese Medicine Resources, Guangdong Pharmaceutical University, and a voucher specimen (No. P-2022–021) was deposited for reference. The raw material was rinsed thoroughly to eliminate surface residues and oven-dried until its moisture level was below 10%. The dried herb was subsequently ground using an FW100 mill, sieved through a 50-mesh screen, and stored as fine powder in sealed, light-protected containers until further use.

2.2. Materials and reagents

All chemicals and reagents adopted met analytical grade or higher criteria. General solvents involved methanol (furnished by Tianjin Damao Chemical Reagent Factory, Tianjin, China) and ethanol (provided by Tianjin Fuyu Fine Chemical Co., Ltd., Tianjin, China). Distilled water (manufactured by Guangzhou Watsons Food & Beverage Co., Ltd., Guangzhou, China) acted as the primary aqueous medium. Organic acids and additives such as urea, acetic acid, propylene glycol, lactic acid, oxalic acid, citric acid, 1,4-butanediol, and ethylene glycol were acquired from Guangzhou Chemical Reagent Factory (Guangzhou, China). Specialty reagents included choline chloride and rutin standard (supplied by Aladdin Co., Ltd., Shanghai, China), as well as sodium nitrite, aluminum nitrate, and sodium hydroxide (obtained from Xilong Science Co., Ltd., China). For chromatographic analyses, LC-HRMS-grade methanol, acetonitrile, and formic acid were sourced from Thermo Fisher Scientific (Waltham, MA, USA). Antioxidant assay components such as ascorbic acid, trichloroacetic acid, ferric chloride, potassium ferricyanide, DPPH, ABTS were procured from McLin Biochemical Technology Co., Ltd. (China) and Aleshan Biotechnology Co., Ltd. (Guangzhou, China). Human normal liver L02 cells were from the National Infrastructure of Cell Line Resource.

2.3. Instrumentation

A suite of instruments from domestic and international suppliers was employed for experimental implementation. In sample preparation, a vortex shaker (QL-861, Haimen Qilin Bell Instrument Manufacturing Co., Ltd., Haimen, China), a desktop CNC ultrasonic cleaner (KQ-600 DE, Dongguan Keqiao Ultrasonic Equipment Co., Ltd., Dongguan, China), and a medical centrifuge (L550, Hunan Xiangyi Laboratory Instrument Development Co., Ltd., Hunan, China) were utilized. Particle size reduction was achieved through a pulverizer (FW100, Beijing Zhongxing Weiye Instrument Co., Ltd., Beijing, China). For precision measurement tasks, an ultra-micro balance (ATY124, Shimadzu Enterprise Management (China) Co., Ltd., Beijing, China) and a magnetic stirrer (DF-101 S, Gongyi Yuhua Instrument Co., Ltd., Gongyi, China) were adopted. Analytical characterization involved: a multifunctional microplate reader (SYNERGY H1, Agilent BioTek Company, Winooski, VT, USA) for spectroscopic assays; a scanning electron microscope (Apreo 2S, Thermo Fisher Scientific, Waltham, MA, USA) for morphological analysis. FTIR spectra were recorded using an IRAffinity-1S FTIR spectrometer (Shimadzu Corporation, Japan), and ¹H NMR spectra were acquired using a Bruker AVANCE 400 MHz NMR spectrometer (Bruker BioSpin GmbH, Rheinstetten, Germany). The viscosity and density of fresh and regenerated DES were measured using an NDJ-8S rotational viscometer and an LC-LDB-3002 electronic density meter, respectively (Shanghai Lichen Bangxi Instrument Technology Co., Ltd., Shanghai, China). The LC-HRMS system specifically consisted of Vanquish Flex UHPLC coupled with Orbitrap Exploris 120 HRMS (Thermo Fisher Scientific, Waltham, MA, USA), equipped with a Hypersil GOLD C₁₈ column (100 mm × 2.1 mm, 1.9 µm) to complete chromatographic separation.

2.4. Preparation of DESs

DESs were formulated following established protocols [39], [40]. Choline chloride (ChCl) was chosen as the hydrogen-bond acceptor because it is among the most widely used HBAs for DES preparation and has been extensively applied in DES-based extraction [41]. The seven ChCl-based DESs were rationally selected to represent three common hydrogen-bond donor (HBD) families, including polyols (ethylene glycol, 1,4-butanediol, glycerol), organic acids (lactic acid, oxalic acid, citric acid), and an amide (urea), thereby covering distinct polarity/viscosity and hydrogen-bonding characteristics. Each mixture (1:2 M ratio of HBA:HBD) was stirred magnetically at 80 °C for 0.5–2 h until a clear and homogeneous solution was obtained. The prepared DES formulations are listed in Table 1.

Table 1.

Types of prepared DES.

No.	HBA	HBD
DES-1	Choline chloride	Citric acid
DES-2		Oxalic acid
DES-3		Glycerol
DES-4		Lactic acid
DES-5		1,4-Butanediol
DES-6		Urea
DES-7		Ethylene glycol

2.5. Extraction procedure and flavonoid quantification

Approximately 0.1 g of B. balsamifera powder was mixed with 6 mL of the designated extraction solvent in a sealed polypropylene centrifuge tube with an internal diameter of 17 mm. Ultrasound-assisted extraction was conducted using a KQ-600 DE ultrasonic bath operated at a fixed frequency of 40 kHz with continuous output. The internal dimensions and nominal capacity of the bath were 50 × 30 × 15 cm and 22.5 L, respectively. During operation, the bath contained 15 L of water at a liquid height of approximately 10 cm. Eighteen tubes were fixed in the central region of the bath using a sample rack. Bath temperature was controlled using circulating water within ± 1 °C, and extraction time was counted after the designated temperature had been reached. The mixture was centrifuged twice at 2200 × g for 10 min, and the resulting supernatant was collected for flavonoid quantification using the colorimetric method described in section 2.6. Extraction yield (W, %) was calculated according to:

$$\text{W}=\left(\frac{\text{C}\times \text{V}\times \text{D}}{\text{1000M}}\right)\times 100\%$$ (1)

In this formula, W indicates the extraction yield (%), C is the concentration (mg/mL) obtained from the standard curve, V refers to the total volume of the reaction mixture (mL), D represents the dilution factor, and M stands for the mass of the sample (g).

2.6. Determination of total flavonoid content

Total flavonoid concentration was measured using the sodium nitrite–aluminum nitrate colorimetric procedure [42]. A rutin calibration curve was prepared by dissolving 0.0105 g of rutin in 60% ethanol to obtain a 0.21 mg/mL stock solution. Aliquots of 0.5–6.0 mL of the standard were reacted sequentially with 1 mL 5% sodium nitrite, 1 mL 10% aluminum nitrate, and 10 mL 4% sodium hydroxide, followed by dilution to 25 mL with 60% ethanol. After incubation for 15 min at room temperature, absorbance was recorded at 500 nm to establish the standard curve.

2.7. Optimization of total flavonoid extraction from b. Balsamifera

2.7.1. Single-factor experiments

Seven DES formulations were compared with ethanol, 50% ethanol, and water as controls. Extraction parameters were initially fixed at a 1:2 M ratio (HBA:HBD), 30% water content, 60 °C, 30 min, ultrasonic power of 300 W, and ultrasonic frequency (40 kHz). Subsequently, the influence of individual parameters—molar ratio (1:1–1:5), water content (10–50%), ultrasonic power (100–600 W), time (10–60 min), ultrasonic temperature (40–90 °C), and liquid–solid ratio (10:1–60:1 mL/g)—was systematically evaluated.

2.7.2. Kinetic and rated-power-based operational analysis

The extraction-time data obtained at 10, 20, 30, 40, 50, and 60 min were used to evaluate the extraction kinetics. Because the total flavonoid yield increased continuously from 10 to 50 min but decreased at 60 min, the monotonic stage from 10 to 50 min was fitted using the Peleg model:

$${\text{Y}}_{\text{t}}=\frac{\text{t}}{{\text{K}}_1+{\text{K}}_{2}t}$$ (2)

where Y_t is the total flavonoid yield at time t, and K₁ and K₂ are the Peleg rate and capacity constants, respectively.

The predicted equilibrium yield and the time required to reach 90% of the predicted equilibrium yield were calculated as Y_eq = 1/K₂ and t₉₀ = 9 K₁/K₂, respectively. Model performance was evaluated using R², RMSE, and average absolute deviation. The 60-min value was excluded from parameter estimation and used as an external diagnostic point. For the ultrasonic-power experiment, the rated-power–time index was calculated as:

$${\text{E}}_{\text{RP}}=\frac{{\text{P}}_{\text{rat}\text{e}\text{d}}\text{×t}}{\text{1000}}$$ (3)

The bath-volume-normalized rated-power–time index was calculated as:

$${\text{E}}_{\text{RP,V}}=\frac{{\text{E}}_{\text{RP}}}{{\text{V}}_{\text{bath}}}$$ (4)

where E_RP is the rated-power–time index (kJ), P_rated is the rated device power (W), t is the treatment time (s), E_RP,V is the bath-volume-normalized rated-power–time index (kJ/L), and V_bath is the actual bath-water volume of 15 L. These calculated quantities were used solely as comparative indices of the selected operating conditions. Because electrical consumption and the acoustic power transmitted to the bath and extraction mixture were not measured, the indices do not represent actual electrical energy consumption, calorimetric acoustic energy, local acoustic-energy density, or cavitation dose.

2.7.3. RSM based on Box-Behnken design

Based on single-factor results, a Box–Behnken design was implemented to optimize the process. The three variables considered were liquid–solid ratio (A), ultrasonic temperature (B), and extraction time (C). The coded and actual levels are presented in Table 2.

Table 2.

Factor levels design for RSM in total flavonoid extraction from B. balsamifera.

Factor	Level
	–1	0	1
A: Liquid–solid ratio (mL/g)	40:1	50:1	60:1
B: Ultrasonic temperature (°C)	70	80	90
C: Ultrasonic time (min)	40	50	60

2.7.4. ANN and GA modeling

The ANN model was developed in MATLAB (R2025b) using the Box–Behnken design dataset, with liquid–solid ratio (A), ultrasonic temperature (B), and ultrasonic time (C) as inputs and total flavonoid yield (%) as the output. Prior to modeling, all data were normalized to the range of –1 to 1 to improve training stability. The dataset was randomly divided into training, validation, and testing subsets at a ratio of 70%, 15%, and 15%, respectively. A feed-forward backpropagation network trained using the Levenberg-Marquardt algorithm was employed, and the number of hidden-layer neurons was optimized by trial and error. The final ANN architecture was determined as 3–9-1. To further search for the optimal extraction conditions, a genetic algorithm (GA) was coupled with the trained ANN model as the fitness function, with total flavonoid yield as the optimization objective. Model performance was evaluated by R², mean square error (MSE), root mean square error (RMSE), and average absolute deviation (AAD). Considering the limited size of the BBD dataset, the ANN-GA model was used as an auxiliary nonlinear optimization tool rather than a standalone predictive model.

2.8. Comparative extraction experiments

Five extraction modes were compared to benchmark the optimized DES-UAE: heat-assisted extraction with stirring using 50% (v/v) aqueous ethanol (EtOH-HS), reflux extraction using 50% aqueous ethanol (EtOH-REF), UAE using 50% aqueous ethanol (EtOH-UAE), heat-assisted extraction with stirring using the selected deep eutectic solvent, ChCl-Gly (1:2; 40% water content, w/w) (DES-HS), and DES-UAE. The liquid–solid ratio, extraction time, and temperature used for EtOH-HS, EtOH-UAE, DES-HS, and DES-UAE were matched to the optimized DES-UAE conditions. EtOH-REF was heated to the boiling point of 50% aqueous ethanol (approximately 85 °C) and maintained under gentle reflux for 60 min. A condenser was used throughout the extraction to minimize solvent loss and maintain the solvent composition. After extraction, the mixtures were filtered and the supernatants were used for total flavonoid content (TFC) determination. All experiments were carried out in triplicate and results are reported as mean ± SD. To separate the contributions of solvent type and ultrasound, EtOH-HS, EtOH-UAE, DES-HS, and DES-UAE were additionally analysed as a 2 × 2 factorial design, with solvent type and ultrasound treatment as the two factors. EtOH-REF was excluded from this factorial analysis because reflux extraction did not represent the same non-ultrasonic treatment level as heat-assisted stirring. The interaction contribution was calculated as:

$${\mathrm{\Delta }}_{\text{interaction}}=\left({\text{Y}}_{\text{DES - UAE}}-{\text{Y}}_{\text{DES - HS}}\right)-\left({\text{Y}}_{\text{EtOH - UAE}}-{\text{Y}}_{\text{ETOH - HS}}\right)$$ (5)

2.9. DES recovery and reusability

The reusability of the selected choline chloride-glycerol (ChCl-Gly) DES (1:2, 40% water content, w/w) was evaluated over three consecutive recovery and reuse cycles. After each DES-UAE extraction performed under the optimized conditions described above, the extract was centrifuged, and the resulting supernatant was collected for DES recovery. A glass chromatographic column (2 cm i.d. × 50 cm) packed with 30.00 g of pretreated AB-8 macroporous resin was used. The packed bed volume was 100 mL. The supernatant was loaded onto the column at a flow rate of 1 BV/h. Because the flavonoids were retained by the resin, the initial column effluent was collected, after which the column was washed with 5 BV of deionized water. The initial effluent and water washings were combined as the DES-rich fraction. The recovered fraction was concentrated by rotary evaporation under reduced pressure at 50 °C to remove excess water. Fresh deionized water was then added gravimetrically to the dewatered recovered DES to reconstitute the DES to a nominal water content of 40% (w/w). The recovered solvent was weighed, and the amount required to maintain the same liquid–solid ratio was used in the subsequent extraction cycle. The regenerated DES was reused under the same optimized conditions, and the TFC values of the extracts obtained using fresh DES and regenerated DES from cycles 1–3 were determined in triplicate (n = 3). Aliquots of the fresh and regenerated DES were collected after each cycle, and their viscosity and density were measured at 50 °C to evaluate changes in the bulk physical properties of the solvent during recycling.

2.10. FTIR and ¹H NMR characterization of ChCl-Gly DES

FTIR spectra of choline chloride, glycerol, and the prepared ChCl-Gly DES were recorded using an IRAffinity-1S FTIR spectrometer (Shimadzu, Japan) over the range of 4000–400 cm⁻¹ at a resolution of 4 cm⁻¹. All samples were analyzed using the KBr pellet method. ¹H NMR spectra of choline chloride, glycerol, freshly prepared ChCl-Gly DES, and regenerated DES samples collected after reuse cycles 1–3 were recorded on a Bruker AVANCE 400 MHz NMR spectrometer (Bruker BioSpin GmbH, Rheinstetten, Germany) using DMSO‑d₆ as the solvent. Chemical shifts were reported in ppm and referenced to the residual DMSO‑d₆ signal at δ 2.50 ppm.

2.11. SEM analysis

To visualize cellular alterations induced by different extraction modes, untreated powder (unextracted) and the solid residues obtained after DES-UAE, DES-HS, EtOH-UAE, EtOH-HS, and EtOH-REF were collected for SEM observation. After extraction, the residues were filtered and quickly rinsed to remove residual solvent/DES components (ethanol-based residues were rinsed with 50% aqueous ethanol; DES-based residues were rinsed with deionized water followed by 50% aqueous ethanol), freeze-dried for 24 h, sputter-coated with gold, and examined using SEM at a magnification of 500 × .

2.12. Antioxidant activity assays

2.12.1. DPPH radical scavenging assay.

Antioxidant activity was determined using the DPPH radical assay [43]. Different concentrations of extract (0.003–3.333 mg/mL) were reacted with equal volumes of 2 mmol/L DPPH in ethanol. After 30 min of dark incubation at room temperature, absorbance was measured at 517 nm. Ascorbic acid served as the positive control, and scavenging activity was computed as:

$$\text{DPPH scavenging (\%)}=\left(1-\frac{{\text{A}}_1-{\text{A}}_2}{{\text{A}}_0}\right)\times 100\%$$ (6)

Here, A₁ denotes the absorbance of the mixture with 1 mL sample solution and 1 mL DPPH solution; A₂, the absorbance of the mixture with 1 mL sample solution and 1 mL ethanol; A₀, the absorbance of the blank control (1 mL blank extraction solvent + 1 mL DPPH solution).

2.12.1. ABTS radical scavenging assay

ABTS radicals were generated by reacting 7 mmol/L ABTS with 2.45 mmol/L potassium persulfate in darkness for 12–16 h [44]. The working solution was diluted to an absorbance of 0.70 ± 0.02 at 734 nm. Then, 1 mL of sample (0.003–0.5 mg/mL) was mixed with 1 mL of ABTS solution and reacted for 30 min in the dark. Absorbance was recorded at 734 nm, and radical scavenging rate was calculated analogously to the DPPH assay.

$$\text{ABTS scavenging (\%)}=\left(1-\frac{{\text{A}}_1-{\text{A}}_2}{{\text{A}}_0}\right)\times 100\%$$ (7)

In this formula, A₁ represents the absorbance of the mixture with 1 mL sample solution and 1 mL ABTS solution; A₂, the absorbance of the mixture with 1 mL sample solution and 1 mL ethanol; A₀, the absorbance of the blank control (1 mL blank extraction solvent + 1 mL ABTS solution).

2.12.2. Reducing power assay

The reducing capacity of extracts was evaluated following the method of Dai et al. [43]. Sample aliquots (0.333–33.333 mg/mL) were combined with PBS (pH 6.0) and 1% potassium ferricyanide, incubated at 50 °C for 20 min, and terminated with 10% trichloroacetic acid. After centrifugation, the supernatant was mixed with distilled water and 0.1% ferric chloride. Absorbance was read at 700 nm, where higher readings indicated stronger reducing power. Ascorbic acid was used as a reference antioxidant.

2.12.3. Cell-based protective effect against H₂O₂-induced injury

The H₂O₂-induced injury model and cytoprotective assay in L02 hepatocytes were performed as previously described, with minor modifications [45]. Briefly, cells were exposed to graded H₂O₂ concentrations to select the modeling dose yielding ∼ 50% viability (600 μmol·L⁻¹). The DES-UAE stock extract used for cell assays was the crude DES extract prepared at a liquid–solid ratio of 40:1 (mL/g), which was then serially diluted (10 × –100000 × ) for cytotoxicity screening; a matched DES blank prepared at the same dilution as the 100 × extract was included as the vehicle control. For the protection assay, H₂O₂-injured cells were treated with the DES-UAE extract at 100 ×, 200 ×, 400 ×, and 800 × dilutions. The 100-fold diluted DES blank was included as the highest-exposure vehicle control because all other extract treatments contained proportionally lower DES concentrations. Cell viability was determined. All experiments were performed in triplicate and results are expressed as mean ± SD.

2.13. Uplc-q-orbitrap HRMS analysis

2.13.1. Sample preparation

Following the optimized DES extraction protocol, 200 μL of the B. balsamifera extract was transferred into a 2 mL centrifuge tube. Subsequently, 800 μL of 50% methanol was added. The mixture was fully vortexed and filtered through a 0.22 μm syringe filter. The obtained filtrate was stored at 4 °C until analysis.

2.13.1. Chromatographic conditions

Chromatographic separation utilized a Vanquish Flex UHPLC system equipped with a Hypersil GOLD C₁₈ column (100 mm × 2.1 mm, 1.9 µm). The column was maintained at 35 °C. The mobile phase consisted of solvent A (acetonitrile with 0.1% formic acid) and solvent B (water with 0.1% formic acid). Gradient elution was applied as follows: 0–5 min, 95%–75% B; 15–25 min, 75%–5% B; 25–27 min, 5% B; 27.001–30 min, 95% B. The flow rate was set at 0.3 mL/min, the injection volume was 2 μL, and the autosampler was kept at 8 °C.

2.13.2. Mass spectrometric conditions

Mass spectrometric detection was conducted using an Orbitrap Exploris 120 high-resolution mass spectrometer, operating in both positive and negative electrospray ionization (ESI) modes. The instrument was set to switch between modes, with a positive ion spray voltage of 3.5 kV and a negative ion spray voltage of 2.8 kV. The capillary temperature was maintained at 325 °C, and the auxiliary gas (nitrogen) was heated to 350 °C. Mass scans were recorded in the m/z range of 100–1200, with a full-scan resolution of 60,000 and MS/MS (dd-MS²) resolution of 15,000. The flow rates of sheath, auxiliary, and sweep gas were set at 50, 8, and 1 arbitrary units, respectively, and collision energies were applied at normalized values of 20%, 40%, and 60%.

2.14. Data processing

All experiments were performed in triplicate, and results are expressed as mean ± SD. Differences among ultrasonic-power levels, extraction times, temperatures, and extraction modes were evaluated using one-way ANOVA followed by Tukey’s HSD test. The effects of solvent type, ultrasound treatment, and their interaction were evaluated using two-way ANOVA. Nonlinear least-squares regression was used for kinetic-model fitting. Differences were considered significant at p < 0.05. Statistical analysis and RSM modeling were carried out using Design Expert 11 (Stat-Ease, USA), SPSS 26.0 (IBM, USA), and GraphPad Prism 8 (GraphPad Software, USA). Figures were prepared in Origin 8.0. Pearson correlation analysis was used to evaluate the relationships between total flavonoid yield and DPPH, ABTS, and reducing-power responses. Mass spectral data were processed via Compound Discoverer 3.3 (Thermo Fisher Scientific, USA) using peak extraction, alignment, and library matching against mzCloud and mzVault databases. Metabolite identification required a mass deviation < 5 ppm and a matching score ≥ 80.

3. Results and discussion

3.1. Total flavonoid standard curve

Accurate quantification of TFC relies on a robust calibration curve. A rutin standard solution (0.21 mg/mL) was serially diluted to prepare concentrations within the linear range. Absorbance readings at 500 nm exhibited excellent linearity with rutin concentration, described by the regression equation:

$$\text{A}=8.4719\text{C}-0.0023$$ (8)

where A represents the value of absorbance, while C represents the concentration of the rutin standard product (mg/mL). With a correlation coefficient of R² = 0.9992, linearity was confirmed across a concentration range of 0.0042–0.0504 mg/mL. This validation ensures accurate quantification of flavonoids in subsequent experimental evaluations.

3.2. Single-factor experiments

To identify the most effective extraction conditions, several single-factor experiments were conducted to examine the effects of solvent type/DES formulation and key operating parameters. For DES screening, all DESs were prepared at a fixed HBA:HBD molar ratio of 1:2 and adjusted to 30% water content (w/w) to reduce viscosity and ensure comparable handling and mass transfer.

3.2.1. Solvent screening and DES selection

Seven ChCl-based deep eutectic solvent systems were systematically constructed through molecular design principles, utilizing glycerol, lactic acid, and 1,4-butanediol as HBDs. These designer solvents were benchmarked against conventional extraction media including aqueous, hydroalcoholic (50% ethanol), and pure alcoholic systems. As illustrated in Fig. 1, the ChCl-Gly combination (1:2 M ratio, DES-3) demonstrated superior extraction capability, achieving a total flavonoid yield of 3.09% ± 0.12%. This performance significantly surpassed conventional 50% ethanol (2.85% ± 0.09%). The statistically significant enhancement (p < 0.05) in phytochemical recovery confirmed DES-3′s optimized solvation properties, leading to its selection as the optimal extraction medium for subsequent experimental phases.

Fig. 1.

Effect of different solvents on the extraction yield of total flavonoids from B. balsamifera. Different lowercase letters indicate significant differences (P < 0.05).

3.2.2. Effect of DES molar ratio

DESs display differing physicochemical properties at various molar ratios, and these property changes affect the solvent's viscosity, density, surface tension, and intermolecular forces [46]. As shown in Fig. 2a, when the ChCl-Gly molar ratio alters within the 1:1 to 1:5 range, the total flavonoid extraction yield first increases and then decreases. Notably, at a 1:2 ChCl-Gly molar ratio, the total flavonoid extraction yield reaches a maximum of 3.14%. Expanding the glycerol amount from a 1:1 to 1:2 M ratio boosts the extraction yield, but further increases beyond 1:2 lead to a reduction. This reveals that an inadequate or excessive hydrogen-bond donor quantity impacts the flavonoid extraction yield, closely tied to changes in the DES's viscosity, density, surface tension, and intermolecular forces. This hydrogen-bonding environment in the ChCl-Gly DES is supported by the FTIR and ¹H NMR results (Section 3.7). Therefore, selecting an optimal molar ratio enhances the dissolution of target components. Thus, a 1:2 ChCl-Gly molar ratio is selected for subsequent experimental studies.

Fig. 2.

Influence of various factors on the extraction yield of total flavonoids from B. balsamifera. (a) Effect of DES molar ratio; (b) Effect of water content; (c) Effect of ultrasonic power; (d) Effect of ultrasonic time; (e) Effect of ultrasonic temperature; (f) Effect of liquid–solid ratio.

3.2.3. Influence of water content

Water content in DES is a key factor impacting the extraction yield of target compounds, primarily influencing solvent polarity and viscosity. Suitable water content can reduce DES viscosity while increasing its surface tension. However, an excess of water disrupts the hydrogen bonds between DES components, leading to a decrease in the extraction yield [47]. As shown in Fig. 2b, when the water content ranges from 10% to 50%, the extraction yield of total flavonoids from B. balsamifera gradually increases with the rise in water content. The extraction yield reaches the highest value of 3.01% at a water content of 40%. When the water content is further increased to 50%, the extraction yield decreases significantly. Therefore, ChCl-Gly with a 40% water content is selected for subsequent experiments.

3.2.4. Effect of ultrasonic power

As shown in Fig. 2c, the total flavonoid yield increased from 2.41 ± 0.01% at 100 W to 3.13 ± 0.01% at 300 W, corresponding to an increase of 29.88%. The yields obtained at 400, 500, and 600 W were 3.17 ± 0.03%, 3.16 ± 0.0246%, and 3.19 ± 0.02%, respectively. The overall effect of rated device power was significant (p < 0.001). However, 300 W did not differ significantly from 400 or 500 W, whereas the small difference between 300 and 600 W was statistically significant.

At the fixed treatment time of 30 min, the rated-power–time indices (E_RP) calculated for 100, 300, and 600 W were 180.00, 540.00, and 1080.00 kJ, respectively, and the corresponding bath-volume-normalized indices were 12.00, 36.00, and 72.00 kJ/L (Table S1). These derived values were used only to compare the selected operating conditions and do not represent measured electrical consumption, calorimetrically determined acoustic power, acoustic energy absorbed by the extraction mixture, or cavitation dose. Although increasing the rated power from 300 to 600 W doubled the calculated indices by definition, the flavonoid yield increased by only 1.92%. This diminishing process-level return supports 300 W as a practical operating setting but does not establish an acoustic-energy optimum.

The marked increase in flavonoid yield from 100 to 300 W indicates a positive extraction response to increasing rated device power and is consistent with ultrasound-assisted process intensification. Previous calorimetric studies have experimentally shown that increasing ultrasonic output generally increases the power transferred to the treatment medium under fixed operating conditions, although this relationship may be nonlinear and system-dependent [48]. Chemat et al. reported that cavitation-associated microstreaming, microjets, and shear forces can disrupt plant tissues, reduce external boundary-layer resistance, and improve solvent penetration and solute transfer at low-to-moderate power, whereas further increases in operating power frequently produce diminishing extraction returns [49]. Similarly, Milani et al. demonstrated that coupling ultrasound with DES can enhance extraction through the combined effects of cavitation-associated phenomena, solvent–matrix interactions, and improved mass transfer, with the extraction response approaching a plateau after readily accessible solutes have been released and the major transport limitations have been reduced [50]. The limited yield improvement above 300 W may also be associated with the complex cavitation behavior of viscous DES media. As reported by Jacobson et al., cavitation development in DES systems is influenced by viscosity, water content, bubble-cloud interactions, and acoustic shielding and therefore may not increase proportionally with the applied operating power [51]. However, because the transmitted acoustic power and cavitation activity were not directly measured in the present system, these mechanisms should be regarded as literature-supported interpretations rather than experimentally verified explanations for the observed power–yield relationship.

3.2.5. Effect of ultrasonic time

As shown in Fig. 2d, the total flavonoid yield increased from 2.63 ± 0.01% at 10 min to a maximum of 3.28 ± 0.03% at 50 min, corresponding to an increase of 24.71%. However, the incremental yield gradually decreased during the later extraction stage. The increases during 20–30, 30–40, and 40–50 min were 0.25, 0.12, and 0.09 percentage point, respectively, indicating a progressive reduction in the net extraction yield.

To further characterize this time-dependent behavior, the monotonic extraction stage from 10 to 50 min was fitted using the Peleg model. The fitted parameters were K₁ = 0.99 min/%, K₂ = 0.29%⁻¹, and Y_eq = 3.44%. The model showed acceptable fitting performance, with R₂ = 0.92, RMSE = 0.07, and an average absolute deviation of 1.94%. The predicted time required to reach 90% of the model equilibrium yield was 30.52 min. The kinetic results support a transition from rapid release of readily accessible flavonoids to a slower stage increasingly controlled by internal diffusion and a reduced concentration gradient between the plant matrix and the bulk solvent. During the initial stage, ultrasound accelerates matrix disruption and continuously renews the solvent layer surrounding the particles, thereby reducing external mass-transfer resistance. As extraction proceeds, depletion of the readily extractable fraction and increasing diffusion resistance within the remaining plant matrix progressively reduce the extraction yield. Similar kinetic behavior has been reported for ultrasound-assisted extraction of phenolic compounds from plant materials [52], [53].

When the sonication time was extended to 60 min, the yield decreased to 3.11 ± 0.01%, which was 5.18% lower than that obtained at 50 min and 0.14 percentage point below the Peleg-predicted value. Thus, the late-stage response could not be explained solely by a monotonic approach to equilibrium. Prolonged sonication may increase energy consumption and thermal accumulation and may also promote compound transformation or re-adsorption within the plant matrix [54]. Because these processes were not directly measured, they are considered possible explanations rather than confirmed degradation mechanisms. Accordingly, 50 min was selected as the optimal extraction time in the single-factor experiment.

3.2.6. Effect of ultrasonic temperature

Temperature elevation reduces viscosity and surface tension, thereby accelerating molecular motion, enhancing solvent penetration, and promoting target component dissolution. Both excessively low and high temperatures impact the extraction yield [55]. Previous studies have shown that flavonoid compound extraction in most plants generally requires relatively high temperatures [56], [57]. As depicted in Fig. 2e, the extraction yield increased progressively as the ultrasonic temperature rose from 40 to 80 °C, indicating that elevated temperature favors flavonoid solubilization and diffusion in the DES system. The highest yield was obtained at 80 °C, whereas a further increase to 90 °C led to a slight decline, suggesting that overly high temperature may reduce the net extraction efficiency. The slight decrease in yield at higher temperature may be associated with reduced cavitation collapse intensity, because increasing liquid temperature raises the vapor content in cavitation bubbles and enhances the vapor-cushioning effect during collapse [58], and it may also reflect the potential thermal sensitivity of some flavonoid constituents, together resulting in diminishing extraction gains [59]. Therefore, 80 °C is identified as the optimal extraction temperature.

Taken together, the power-, time-, and temperature-dependent responses indicate that ultrasound-assisted extraction was governed by a balance between cavitation-enhanced mass transfer and diminishing benefits under excessive operating conditions. Duan et al. reported that dual-frequency ultrasound can generate complementary acoustic fields and improve extraction efficiency compared with single-frequency treatment [60]. In contrast, the present study used a fixed 40 kHz ultrasonic bath. Although the yield trends are consistent with literature-reported cavitation-associated microstreaming, shear, and matrix disruption, these mechanisms were not directly verified. Because the actual acoustic energy density and cavitation-collapse intensity were not measured, the calculated rated-power–time indices should be interpreted only as comparative operational indices for the device settings used.

3.2.7. Effect of liquid–solid ratio

The ratio of DES volume to plant material mass influences bioactive compound mass transfer. As shown in Fig. 2f, the total flavonoid extraction yield from B. balsamifera increases with the rise in DES quantity. Within the 20:1–40:1 mL/g liquid–solid ratio range, the extraction yield grows rapidly, and the upward trend slows down from 40:1 to 50:1 mL/g. When the solvent amount is further increased to a 60:1 mL/g liquid–solid ratio, the flavonoid extraction yield slightly decreases. This may be attributed to the fact that an appropriate solvent increase enhances material-solvent contact, promotes molecular movement, strengthens solvent penetration into the material, and facilitates target compound dissolution. Conversely, excessive solvent volume can dilute the ultrasonic energy delivered to the system, lowering the effective energy density and thereby weakening cavitation intensity and micro-mixing, which ultimately limits mass transfer and may reduce the extraction yield [61], [62]. Thus, a 50:1 mL/g liquid–solid ratio is selected as the optimal extraction ratio.

3.3. RSM optimization

3.3.1. Response surface optimization design and results

After analyzing the outcomes of single-factor experiments for total flavonoid extraction from B. balsamifera, the extraction solvent was fixed as DES with a 40% water content (ChCl-Gly molar ratio of 1:2). For other influencing factors—liquid–solid ratio (A), ultrasonic temperature (B), and ultrasonic time (C)—a response surface optimization experiment was carried out to explore the optimal extraction scheme. The experimental design and RSM results are presented in Table 3.

Table 3.

Response surface design and experimental/predicted results for total flavonoid extraction from B. balsamifera.

Test Number	A: Liquid–solid ratio (mL/g)	B: Ultrasonic temperature (°C)	C: Ultrasonic time (min)	Y: Total flavonoid yield (%)	RSM Predicted	RSM residual	ANN Predicted	ANN residual
1	40:1	70	50	3.03 ± 0.04	3.08	−0.05	3.07	−0.04
2	60:1	70	50	2.85 ± 0.02	2.84	0.01	2.88	−0.03
3	40:1	90	50	2.91 ± 0.05	2.92	−0.01	2.91	0
4	60:1	90	50	2.80 ± 0.03	2.75	0.05	2.81	−0.01
5	40:1	80	40	3.21 ± 0.05	3.18	0.03	3.20	0.01
6	60:1	80	40	2.81 ± 0.02	2.84	−0.03	2.83	−0.02
7	40:1	80	60	2.93 ± 0.06	2.90	0.03	2.92	0.01
8	60:1	80	60	2.82 ± 0.03	2.84	−0.02	2.87	−0.05
9	50:1	70	40	3.14 ± 0.05	3.12	0.02	3.14	0
10	50:1	90	40	2.85 ± 0.03	2.87	−0.02	2.85	0
11	50:1	70	60	2.87 ± 0.02	2.85	0.02	2.86	0.01
12	50:1	90	60	2.83 ± 0.06	2.85	−0.02	2.84	−0.01
13	50:1	80	50	3.30 ± 0.04	3.33	−0.03	3.35	−0.05
14	50:1	80	50	3.33 ± 0.07	3.33	0	3.35	−0.02
15	50:1	80	50	3.32 ± 0.03	3.33	−0.01	3.35	−0.03
16	50:1	80	50	3.36 ± 0.06	3.33	0.03	3.35	0.01
17	50:1	80	50	3.35 ± 0.04	3.33	0.02	3.35	0

3.3.2. Analysis of variance of the regression model

Based on the experimental data in Table 4, a multiple quadratic regression model is established to depict the correlation among three influencing factors—A (liquid–solid ratio), B (ultrasonic temperature), C (ultrasonic time)—and the total flavonoid extraction yield (Y). The developed equation is shown as follows:

$$\text{Y}=3.33-0.1\text{A}-0.0625\text{B}-0.07\text{C}+\text{0.0175AB}+0.0725\text{AC}+0.0625\text{BC}-0.{\text{2073A}}^2-0.2273{\text{B}}^2-0.1822{\text{C}}^2$$ (9)

Table 4.

Analysis of variance (ANOVA) for response surface model.

Source	Sum of squares	Degrees of Freedom	Mean square	F-value	p-value	Significance
Model	0.7891	9	0.0877	53	< 0.0001	**
A	0.08	1	0.08	48.36	0.0002	**
B	0.0312	1	0.0312	18.89	0.0034	**
C	0.0392	1	0.0392	23.7	0.0018	**
AB	0.0012	1	0.0012	0.7405	0.418	NS
AC	0.021	1	0.021	12.71	0.0092	**
BC	0.0156	1	0.0156	9.45	0.018	*
A2	0.1809	1	0.1809	109.32	< 0.0001	**
B2	0.2174	1	0.2174	131.44	< 0.0001	**
C2	0.1399	1	0.1399	84.54	< 0.0001	**
Residual	0.0116	7	0.0017	−	−	−
Lack-of-fit	0.0093	3	0.0031	5.44	0.0677	NS
Pure Error	0.0023	4	0.0006	−	−	−
Total	0.8006	16	−	−	−	−
R2	0.9855	−	−	−	−	−
Radj2	0.9669	−	−	−	−	−
Precision	18.5769	−	−	−	−	−

** Highly significant (P < 0.01); * Significant (P < 0.05); NS: Not significant.

Table 4 illustrates that the model exhibited high significance (P < 0.0001) alongside an insignificant lack-of-fit (P = 0.0677 > 0.05), thereby validating its reliability for process optimization. The high determination coefficient (R² = 0.9855) and adjusted R² (0.9669) indicated excellent model fitness, explaining 98.55% of the response variation. The low coefficient of variation (1.34%) suggested minimal experimental interference, while adequate precision (18.5769 > 4) confirmed model reliability for prediction. ANOVA outcomes indicated that linear terms (A, B, and C) and quadratic terms (A², B², C²) exhibited strong statistical significance (P < 0.01), while both the BC and AC interactions showed moderate significance (P < 0.05). Factor influence order was determined as: A (liquid–solid ratio) > C (time) > B (temperature) based on F-values.

3.3.3. Response surface interaction analysis

Three-dimensional response surfaces and contour plots, as presented in Fig. 3, illustrate interactive effects among variables. The B/C interaction (ultrasonic temperature/ultrasonic time, Fig. 3b) and the A/C interaction (liquid–solid ratio/ultrasonic time, Fig. 3c) demonstrated significant influence through elliptical contours and steep response surfaces, consistent with ANOVA results (P < 0.05). In contrast, the A/B interaction (liquid–solid ratio/ultrasonic temperature, Fig. 3a) showed circular contours with gentle slopes, indicating (P > 0.05).

Fig. 3.

Interaction effects of factors on the extraction yield of total flavonoids from B. balsamifera. (a) Interaction between Factors A and B; (b) Interaction between Factors B and C; (c) Interaction between Factors A and C.

3.3.4. Model validation and verification

The model determined the optimal conditions to be a liquid–solid ratio of 54.657:1 mL/g, an ultrasonic temperature of 78.642 °C, and an extraction time of 48.770 min, projecting a total flavonoid extraction yield of 3.246%. Practical adjustments were made to 55:1 mL/g and 49 min for operational feasibility. Triplicate verification experiments under modified conditions (40% water content, ChCl-Gly 1:2) yielded 3.30% ± 0.12% total flavonoids, demonstrating excellent agreement with predicted values and confirming model reliability for DES-based extraction optimization.

3.4. Predictive performance and ANN-GA optimization

To further describe the nonlinear relationship between extraction variables and total flavonoid yield, an ANN model was established based on the BBD dataset and compared with the RSM model. The number of hidden-layer neurons was optimized by trial and error, and the best performance was obtained when nine neurons were used in the hidden layer, resulting in a 3–9–1 network topology. As shown in Fig. 4(a,b), the best validation performance was achieved at epoch 3 after four training iterations. The regression coefficients for the training, validation, testing, and overall datasets were 0.99825, 0.99587, 0.99779, and 0.99368, respectively, indicating good agreement between the ANN-predicted and experimental values. The optimized network structure, including the corresponding weight and bias matrices, is shown in Fig. 4(c).

Fig. 4.

ANN model performance and network parameters. (a) Training performance curve; (b) regression plots of ANN-predicted and experimental total flavonoid yields; (c) weight and bias matrices of the optimized 3–9–1 ANN topology.

The trained ANN model was further coupled with GA to perform global optimization of the DES-UAE process. As shown in Fig. S1, the GA fitness curve gradually converged after 300 generations, indicating that an optimal or near-optimal solution had been reached. The ANN-GA model predicted the optimal extraction conditions as a liquid–solid ratio of 40.006:1 mL/g, ultrasonic temperature of 76.488 °C, and ultrasonic time of 45.023 min, with a predicted total flavonoid yield of 3.534%. For operational convenience, the predicted conditions were adjusted to 40:1 mL/g, 77 °C, and 45 min for experimental verification. Under these practical conditions, the measured total flavonoid yield was 3.52 ± 0.01%, which closely agreed with the predicted value and confirmed the feasibility and reliability of the ANN-GA optimum.

The predictive performance of RSM and ANN-GA was compared using the statistical parameters summarized in Table 5. The RSM model predicted optimal conditions of 54.657:1 mL/g liquid–solid ratio, 78.642 °C ultrasonic temperature, and 48.770 min ultrasonic time, with a predicted yield of 3.246% and a measured yield of 3.30 ± 0.12%. In comparison, ANN-GA produced a higher predicted yield and a closely matched experimental value. The R² value of ANN-GA (0.9914) was slightly higher than that of RSM (0.9855), while the MSE, RMSE, and AAD values of ANN-GA (0.0006, 0.0245, and 0.5815%, respectively) were lower than those of RSM (0.0007, 0.0265, and 0.7828%, respectively). These results suggest that ANN-GA provided slightly improved nonlinear fitting and optimization performance compared with RSM.

Table 5.

Comparison of optimized extraction conditions and predictive performance of RSM and ANN-GA models.

Model	Process Parameter			Total flavonoid yield (%)		Model performance metrics
	A	B	C	Predicted	Measured	R2	MSE	RMSE	AAD (%)
RSM	54.657:1	78.642	48.770	3.246	3.30 ± 0.12	0.9855	0.0007	0.0265	0.7828
ANN	40.006:1	76.488	45.023	3.534	3.52 ± 0.01	0.9914	0.0006	0.0245	0.5815

The optimized conditions were generally consistent with the single-factor results. The single-factor experiments identified favorable levels of approximately 50:1 mL/g, 80 °C, and 50 min, whereas both models converged on relatively narrow temperature and time ranges of approximately 76–79 °C and 45–49 min, respectively. This agreement indicates that moderate-to-high temperature reduced the viscosity of the aqueous ChCl-Gly system and enhanced solvent penetration and diffusion, while avoiding the unfavorable cavitation and thermal effects observed at 90 °C. Similarly, the optimized treatment times were sufficient for ultrasound-induced matrix disruption and rapid flavonoid release, whereas prolonged sonication produced diminishing extraction gains. The main difference between the models was the liquid–solid ratio. The ANN-GA optimum of 40.01:1 mL/g was substantially lower than the RSM value of 54.66:1 mL/g. This suggests that ultrasound-enhanced mixing, matrix disruption, and shortened diffusion pathways allowed efficient extraction without excessive solvent addition. Therefore, compared with RSM, ANN-GA provided a more resource-efficient condition by reducing DES consumption and shortening the extraction time while maintaining a higher flavonoid yield. Because the ANN-GA liquid–solid ratio was close to the lower boundary of the design range, its reliability was primarily supported by the experimental verification result.

The improved performance of ANN-GA is consistent with previous studies showing that ANN-based models can more effectively capture nonlinear relationships in complex extraction systems, whereas RSM remains valuable for interpreting factor effects and interactions [37], [38]. Recent studies have further demonstrated the applicability of model-guided DES-UAE to the extraction of bioactive constituents from medicinal plants [63], [64]. Cui et al. combined tailored DES systems with ultrasound, RSM, and ANN-GA to optimize the intensified extraction of flavonoids from Scutellaria Radix, demonstrating the value of integrating solvent design, ultrasonic cavitation, and intelligent process optimization [65]. Similarly, Tian et al. applied DES-UAE together with single-factor experiments, RSM, and ANN to optimize the extraction of polyphyllins from Paris polyphylla var. chinensis, further supporting the applicability of DES-UAE and nonlinear modelling to complex medicinal plant matrices [66]. Nevertheless, the ANN model was developed using a relatively limited BBD dataset. ANN-GA should therefore be regarded as a complementary nonlinear optimization approach rather than a standalone predictive framework. The final conditions were supported by experimental verification and interpreted together with the RSM results, extraction kinetics, and mechanistic evidence for DES-assisted ultrasonic mass transfer.

3.5. Comparison of extraction modes for total flavonoid yield

Under the specified comparative conditions, significant differences were observed among the five extraction modes (Fig. 5). DES-UAE produced the highest total flavonoid yield of 3.52 ± 0.01%, followed by EtOH-UAE (2.85 ± 0.01%), EtOH-REF (2.57 ± 0.02%), DES-HS (2.42 ± 0.02%), and EtOH-HS (2.03 ± 0.01%) (p < 0.05). When the DES-UAE yield was normalized to 100%, the relative yields of EtOH-UAE, EtOH-REF, DES-HS, and EtOH-HS were 80.97%, 73.01%, 68.75%, and 57.67%, respectively. The direct comparisons within the same solvent system clearly demonstrated the contribution of ultrasound to extraction efficiency. In the ChCl-Gly system, ultrasound increased the yield from 2.42% for DES-HS to 3.52% for DES-UAE, corresponding to an absolute increase of 1.10 percentage points and a relative improvement of 45.45%. Similarly, in the 50% ethanol system, ultrasound increased the yield from 2.03% for EtOH-HS to 2.85% for EtOH-UAE, representing an increase of 0.82 percentage point or 40.39%. EtOH-UAE also produced a 10.89% higher yield than conventional ethanol reflux extraction, further indicating that ultrasound enhanced recovery beyond that achieved by prolonged thermal treatment alone.

Fig. 5.

TFC obtained under five extraction modes. Different lowercase letters indicate significant differences (p < 0.05).

When averaged across the two solvent systems, ultrasound treatment increased the total flavonoid yield by 0.96 percentage point, whereas replacing 50% ethanol with DES produced an average increase of 0.53 percentage point. Two-way ANOVA confirmed significant effects of solvent type and ultrasound treatment (p < 0.001), indicating that ultrasound was a major factor governing extraction efficiency.

The superiority of DES-UAE over EtOH-UAE further indicates that the aqueous ChCl-Gly system provided a more favorable environment for flavonoid desorption and solubilization under ultrasound. DES-UAE increased the yield by 23.51% relative to EtOH-UAE, while DES-HS exceeded EtOH-HS by 19.21%. A similar advantage of DES-UAE over ethanol-UAE has been reported for flavonoid extraction from hawthorn, supporting the broader applicability of DES–ultrasound coupling in plant extraction [35]. Previous studies have also shown that the polarity, hydrogen-bonding capacity, and wetting properties of DES facilitate solvent penetration and bioactive-compound recovery [57], [67].

The four matched EtOH-HS, EtOH-UAE, DES-HS, and DES-UAE treatments were further analysed using a 2 × 2 factorial design. The difference-in-differences was 0.28 percentage point, and the interaction between solvent type and ultrasound was significant (p < 0.001). In the absence of interaction, the expected additive DES-UAE yield was 3.24%, whereas the observed yield was 3.52%, which was 8.64% above the additive expectation. These results indicate that ultrasound not only exerted a strong independent effect but also enhanced the extraction capability of the DES system. The combined enhancement can therefore be explained by complementary physical and solvent effects. Ultrasound promoted matrix disruption, solvent circulation, and mass transfer, while the ChCl-Gly-water system improved flavonoid desorption and solubilization through its polarity and hydrogen-bonding environment. Similar process-level synergy between DES and ultrasound has been reported for steviol glycoside extraction [50].

3.6. Reusability of DES over consecutive recovery–regeneration cycles

The reusability of the ChCl-Gly DES was evaluated by monitoring the total flavonoid yield and the physical properties of the recovered solvent over three consecutive recovery and reuse cycles. Using the extraction yield from fresh DES as the benchmark (3.50% ± 0.06%, set to 100%), the regenerated DES afforded extraction yields of 3.41% ± 0.05% in cycle 1 (97.43% retention), 3.28% ± 0.04% in cycle 2 (93.71% retention), and 3.17% ± 0.05% in cycle 3 (90.57% retention). To assess the physical stability of the regenerated DES, viscosity and density were measured at 50 °C. The viscosity values of the fresh DES and DES recovered after cycles 1–3 were 22.30, 20.70, 20.30, and 21.10 mPa·s, respectively, while the corresponding density values were 1.106, 1.091, 1.093, and 1.108 g/cm³. The maximum deviations from the fresh DES were 8.97% for viscosity and 1.36% for density, and neither parameter showed progressive deterioration during recycling.

Comparative ¹H NMR analysis showed that the regenerated DES samples retained the principal proton signals of the fresh ChCl-Gly DES without pronounced changes across the three reuse cycles (Fig. S2), indicating that no evident large-scale decomposition of the major DES components occurred under the applied recovery conditions. Nevertheless, the recovered solvent gradually changed from colorless to pale yellow. These observations suggest the accumulation of small amounts of co-extracted plant constituents or other impurities during repeated use. The qualitative ¹H NMR spectra do not, however, quantitatively confirm an unchanged ChCl/Gly molar ratio or cycle-specific water content.

The gradual decrease in flavonoid yield therefore cannot be definitively attributed to a single factor. It may reflect the combined effects of mechanical solvent losses during recovery, accumulation of co-extracted constituents, and small compositional variations introduced during solvent recovery and gravimetric reconstitution [68]. Although the regenerated DES retained 90.57–97.43% of the initial extraction yield and exhibited relatively limited changes in viscosity, density, and its principal ¹H NMR signals, these results should be regarded as preliminary evidence of short-term laboratory-scale reusability rather than proof of stable DES composition or long-term recyclability. Because the water content was controlled through dewatering followed by gravimetric reconstitution, rather than independently determined by Karl Fischer titration, further studies involving direct water-content measurement, quantitative compositional analysis, impurity characterization, and additional reuse cycles are required.

3.7. FTIR and ¹H NMR analysis of ChCl-Gly DES

To probe intermolecular interactions in the ChCl-Gly DES, FTIR and ¹H NMR analyses were performed (Fig. 6). In the FTIR spectra (Fig. 6a), the DES exhibits a broadened O–H stretching band relative to the pure components, indicating a strengthened and diversified hydrogen-bonding environment upon DES formation. In addition, the band assigned to C–O/C–N stretching in the fingerprint region shows a clear shift from 1092 cm⁻¹ (pure ChCl) to 1083 cm⁻¹ in the DES, suggesting altered local bonding/electronic environments consistent with hydrogen-bond network reorganization. The ¹H NMR spectra recorded in DMSO‑d₆ (Fig. 6b) further support this interpretation: upon DES formation, the hydroxyl signal of ChCl shifts from 5.77 to 5.36 ppm, while the hydroxyl signals of glycerol shift from 4.50/4.57 to 4.57/4.63 ppm. These opposite shift directions indicate that hydroxyl protons experience distinct changes in shielding/deshielding, consistent with redistribution of hydrogen-bond interactions in the mixture. Collectively, the FTIR and ¹H NMR results support the formation of a new hydrogen-bonding network in the ChCl-Gly DES, likely involving chloride ions and hydroxyl groups, in agreement with previous reports [69], [70].

Fig. 6.

FTIR and ¹H NMR analyses of the ChCl-Gly DES (ChCl-Gly = 1:2). (a) FTIR spectra of ChCl-Gly DES, Gly, and ChCl. (b) ¹H NMR spectra (DMSO‑d₆) of ChCl-Gly DES, Gly, and ChCl.

3.8. SEM analysis

To investigate microstructural changes induced by different extraction modes, the untreated powder and the solid residues obtained after DES-UAE, DES-HS, EtOH-UAE, EtOH-HS, and EtOH-REF were examined by SEM. As shown in Fig. 7a (500 × ), the untreated powder displayed a relatively intact cellular framework with a smooth surface. The EtOH-HS residue (Fig. 7e) remained largely recognizable, indicating limited disruption under conventional heat-assisted extraction. The EtOH-REF residue exhibited a comparatively loosened and irregular structure, suggesting that prolonged heating and solvent contact caused greater structural alteration than heat-assisted stirring alone. In contrast, ultrasound-assisted treatments produced markedly stronger structural damage: both EtOH-UAE and DES-UAE residues exhibited pronounced surface erosion and fragmentation (Fig. 7d and Fig. 7b). These morphological changes are consistent with the mechanical effects associated with acoustic cavitation, including microstreaming, microjets, and shear forces, which can disrupt plant tissues, shorten internal diffusion pathways, and facilitate solvent penetration and solute release. Moreover, DES-HS (Fig. 7c) showed a roughened morphology compared with EtOH-HS, indicating that the ChCl-Gly medium may improve matrix wetting and penetration even in the absence of ultrasound. The most extensive structural disintegration was observed for DES-UAE (Fig. 7b), which is consistent with the combined effects of the DES solvent environment and ultrasound-assisted matrix disruption.

Fig. 7.

SEM of B. balsamifera powder (untreated) and residues after different extraction modes (×500). (a) Untreated powder before extraction; (b) residue after DES-UAE extraction; (c) residue after DES-HS extraction; (d) residue after EtOH-UAE extraction; (e) residue after EtOH-HS extraction; (f) residue after EtOH-REF extraction.

These observations agree with the extraction-mode results, in which UAE increased the total flavonoid yield in both the DES and ethanol systems, while DES-UAE produced the highest yield. Together, the higher UAE yields and the SEM-observed morphological disruption are consistent with a physical contribution of ultrasound to matrix disruption and mass-transfer enhancement. However, SEM does not directly quantify acoustic power or cavitation intensity. [71], [72].

3.9. In vitro antioxidant activity

Antioxidant capacity serves as a critical parameter for assessing the biological activity of plant-derived extracts. In this investigation, the antioxidant potential of diverse solvent extracts was evaluated through DPPH radical scavenging, ABTS radical cation decolorization, and reducing power assay measures. These analyses aimed to highlight the benefits of DES-assisted extraction in facilitating the green isolation of flavonoids from B. balsamifera by comparing its performance against conventional solvent systems.

3.9.1. DPPH radical scavenging activity

The DPPH assay represents a commonly employed technique for evaluating antioxidative potential, where elevated scavenging efficiency corresponds to enhanced antioxidative potential (Table 6 and Fig. 8a). Collectively, all tested extracts demonstrated varying degrees of DPPH radical scavenging capability, though none matched the performance of vitamin C (VC). Within the concentration spectrum of 0–1.667 mg/mL, scavenging efficiency rose steeply with increasing concentration until reaching a plateau. Beyond 1.667 mg/mL, most extracts approached saturation levels. Notably, VC attained a peak scavenging efficiency of 96.5% at a concentration of 3.333 mg/mL.

Table 6.

IC₅₀ Values (mg/mL, Mean ± SD, n = 3) of different solvent extracts of B. balsamifera and VC for DPPH and ABTS radical scavenging.

Samples※	DPPH	ABTS
DES-1	0.2463 ± 0.0217c	0.0986 ± 0.0052 e
DES-2	0.2799 ± 0.0248c	0.0446 ± 0.0023c
DES-3	0.1443 ± 0.0063b	0.0417 ± 0.0041c
DES-4	0.2377 ± 0.0114c	0.0750 ± 0.0043 d
DES-5	0.2525 ± 0.0192c	0.0229 ± 0.0039b
DES-6	0.2725 ± 0.0155c	0.0207 ± 0.0037b
DES-7	0.2692 ± 0.0131c	0.0276 ± 0.0060b
Water	0.5283 ± 0.0256 e	0.0770 ± 0.0057 d
50% Ethanol	0.3892 ± 0.0236 d	0.0691 ± 0.0058 d
Ethanol	UC 1	UC 1
VCΔ	0.0104 ± 0.0009 a	0.0035 ± 0.0001 a

^※denotes the mass of dry matter powder per 1 mL of solvent; ^Δ represents the mass of the compound per 1 mL of solvent; ^a-e Different lowercase letters within the same column indicate significant differences at the P < 0.05 level; ¹ Unable to calculate.

Fig. 8.

Evaluation of antioxidant activity of B. balsamifera extracts from different solvents and VC. (a) DPPH radical scavenging ability of extracts from different solvents; (b) ABTS radical scavenging ability of extracts from different solvents; (c) Reducing power of extracts from different solvents; (d) Reducing power of VC.

Among the DES extracts, all but DES-2 (which reached a maximum of approximately 85%) achieved high scavenging rates (exceeding 90%), with DES-1, DES-3, and DES-4 showing rates greater than 95%. In contrast, only the 50% ethanol extract among traditional solvents approached a similar level (near 90%), while the ethanol extract displayed markedly lower activity (below 50%), precluding the calculation of its IC₅₀. The concentration-dependent antioxidant efficacy was quantified through IC₅₀ measurements, with lower values indicating stronger activity. VC displayed the most potent DPPH radical scavenging capacity (IC₅₀ = 0.0104 ± 0.0009 mg/mL). Notably, DES-3 outperformed other solvent extracts in bioactive compound enrichment, achieving an IC₅₀ of 0.1443 ± 0.0063 mg/mL. These results suggest that DES extraction, with its strong solubilization ability, significantly enhances the recovery of flavonoids with robust DPPH scavenging activity.

3.9.2. ABTS radical scavenging activity

The ABTS assay offered supplementary validation of the extracts' antioxidative capacity (Table 6 and Fig. 8b). Across the concentration spectrum from 0 to 0.333 mg/mL, scavenging efficiency demonstrated a rapid upward trend with increasing concentration. At 0.5 mg/mL, VC approached complete radical elimination. The majority of extracts, including those derived from water, 50% ethanol, and multiple DES systems, exhibited robust ABTS radical scavenging activity, with rates surpassing 95%. Among these, specific DES extracts such as DES-3, DES-5, DES-6, and DES-7 displayed particularly outstanding performance, achieving scavenging efficiencies exceeding 99%. In contrast, the ethanol extract displayed significantly lower activity, with a scavenging rate below 50%. The corresponding IC₅₀ values further confirmed that VC exhibited the highest antioxidant activity, while certain DES systems, particularly DES-6, also showed strong performance.

3.9.3. Reducing power assays

The reducing power assay measures a sample's ability to donate electrons or hydrogen atoms, which is closely linked to its antioxidant potential (Fig. 8c). VC displayed a rapid increase in reducing power within the 0–0.33 mg/mL range, reaching a saturation plateau beyond 1 mg/mL with a maximum absorbance of 2.197 at 1.67 mg/mL. While all solvent extracts demonstrated concentration-dependent enhancement of reducing power, the 50% ethanol extract exhibited superior reducing capacity comparable to the deep eutectic solvent extract DES-5. In contrast, ethanol extract displayed markedly inferior performance in this assay. Furthermore, certain DES extracts, such as DES-1 and DES-4, also showed relatively low reducing power. This discrepancy may be attributed to the specific properties of the solvents. For instance, citric acid, a component in DES-1, potentially forms stable coordination complexes with Fe³⁺ ions, which may interfere with the assay by preventing effective electron transfer or altering colorimetric detection [73].

3.9.4. Antioxidant activity analysis and discussion

B. balsamifera contains diverse phytochemicals, particularly flavonoids, that have been associated with its antioxidant activity. Nessa et al. demonstrated that the antioxidant capacity of B. balsamifera extracts varies with solvent polarity, showing the order methanol > chloroform > petroleum ether, with quercetin identified as the key flavonoid contributor [13]. In the present study, the antioxidant responses of extracts obtained using different DESs and conventional solvents were compared under identical assay conditions. Similarly, Liu et al. reported that DES coupled with ultrasound efficiently enhanced the recovery of polyphenols from pomegranate peel and enabled the identification of numerous phenolic constituents by UPLC-Q-TOF-MS/MS, supporting the effectiveness of combining green extraction with comprehensive profiling of antioxidant-related compounds [74].

In the DPPH radical scavenging assay, most DES extracts exhibited stronger activity than the conventional-solvent extracts. Notably, DES-3 exhibited an IC₅₀ of 0.1443 ± 0.0063 mg/mL, significantly lower than ethanol extracts (IC₅₀ uncalculable) and 50% ethanol (0.3892 ± 0.0236 mg/mL). While previous studies using methanol extraction reported a lower IC₅₀ (105.1 ± 0.503 μg/mL) [43], direct comparison requires caution due to methodological differences (e.g., plant material preparation and assay conditions). Crucially, DES-3′s IC₅₀ approached that of VC (0.0104 ± 0.0009 mg/mL) in our study, with a 13.9-fold activity gap versus a 102.1-fold difference between methanol extracts and VC in prior work [43]. Jirakitticharoen et al. reported that 50% ethanol extracts of B. balsamifera exhibited relatively high antioxidant activity when compared to extracts obtained using 95% ethanol, hexane, and water [20]. Antioxidant activity has also been reported in other Blumea species. For example, a B. laciniata extract showed DPPH and ABTS EC₅₀ values of 0.41 and 0.13 mg/mL, respectively [75]. More recently, Kholiya et al. evaluated the shoot and root essential oils of B. mollis using DPPH radical-scavenging, H₂O₂-scavenging, and Fe(II)-complexing assays, with the root essential oil exhibiting higher overall antioxidant activity than the shoot essential oil [76]. These findings demonstrate the broad variation in antioxidant activity among Blumea extracts, fractions, and essential oils. However, direct potency ranking across studies is not appropriate because of differences in species, plant parts, extract types, extraction conditions, assay protocols, and result expression. Under the standardized conditions used in the present study, multiple DES extracts also exhibited stronger ABTS radical scavenging responses than the tested conventional-solvent extracts, consistent with the DPPH results. These findings indicate that DES extraction can improve the recovery of constituents associated with in vitro radical-scavenging activity from B. balsamifera. However, the observed activity cannot be attributed exclusively to flavonoids, and the results do not establish the general superiority of DES beyond the solvents and experimental conditions evaluated in this study.

3.9.5. Correlation analysis

To delve deeper into the relationship between the chemical composition and antioxidant capacity of B. balsamifera extracts, Pearson correlation analysis was performed to examine associations among TFC and antioxidant activities evaluated via DPPH, ABTS, and reducing power assays (Fig. 9). Results indicated a substantial positive correlation between total flavonoid yield and DPPH radical scavenging activity (r = 0.819, P < 0.01), alongside a significant positive relationship with ABTS radical scavenging activity (r = 0.749, P < 0.05). These findings align with previous research identifying flavonoids as major radical scavengers in B. balsamifera [13], [22]. These findings are consistent with the established radical-scavenging activity of flavonoids, including quercetin derivatives previously reported in this species [77]. These associations support total flavonoid yield as a relevant compositional marker of the DPPH and ABTS responses across the tested extracts. However, correlation does not establish causality, confirm a specific radical scavenging mechanism, or quantify the proportional contribution of flavonoids to the overall antioxidant activity.

Fig. 9.

Pearson correlation analysis between total flavonoid yield and antioxidant activity of B. balsamifera extracts from different solvents.

It should be noted that the present correlation analysis focuses on TFC as the key compositional descriptor; therefore, the potential contribution of non-flavonoid components (e.g., volatile constituents, sesquiterpenes, phenylpropanoids, and other phenolics) to antioxidant activity was not quantitatively dissected in this study. The modest correlation observed between TFC and reducing power (r = 0.277) underscores the significant limitations of relying on single-metric antioxidant assessments. This difference stems from two key factors. The weak association may reflect both compositional and matrix effects. Non-flavonoid constituents, may contribute to Fe³⁺ reduction [43], [78]. In addition, citric acid in DES-1 may alter iron complexation and the Fe³⁺ reaction environment, potentially affecting the measured reducing-power response. Because this matrix effect was not independently verified, it is considered a possible explanation rather than a confirmed cause. The present study was designed around the extraction and optimization of total flavonoids, and total phenolic content was therefore not determined as an additional compositional endpoint. Consequently, the contribution of non-flavonoid constituents cannot be quantitatively resolved, and the antioxidant activity should not be attributed exclusively to flavonoids.

3.9.6. Cytoprotective effect of the DES-UAE extract against H₂O₂-induced oxidative injury in L02 hepatocytes

L02 cell viability decreased with increasing H₂O₂ concentration, and 600 μmol·L⁻¹ was selected to establish the injury model, reducing viability to 53.97% ± 3.62% versus the control (102.35% ± 4.25%). The DES-UAE extract showed acceptable cell compatibility at the working dilution (100 × ), and the matched DES blank (100 × ) showed no marked cytotoxicity (92.11% ± 3.24%) (Fig. 10a). Although this value was slightly lower than those observed at higher extract dilutions, it remained above 90%, indicating limited cytotoxicity of the diluted DES matrix. Under H₂O₂ injury, the DES-UAE extract improved cell viability in a dilution-dependent manner, increasing viability from 53.97% ± 3.62% (model) to 88.54% ± 3.94% at 100×, while the effect weakened with further dilution and approached the model level at 800× (54.91% ± 3.25%) (Fig. 10b). Importantly, the matched DES blank under H₂O₂ treatment was not significantly different from the model group, indicating that the DES itself did not provide measurable cytoprotection. Therefore, the improved viability observed with the DES-UAE extract was mainly attributable to the extracted bioactive constituents rather than to the DES vehicle. This concentration-dependent cytoprotection is consistent with the strong antioxidant capacity observed in the DPPH and ABTS assays, supporting the antioxidant-related protective potential of the DES-UAE extract under oxidative stress conditions.

Fig. 10.

Cytotoxicity and cytoprotective effects of the DES-UAE extract in L02 hepatocytes. (a) Cytotoxicity at dilution factors of 10 × –100000 × with the matched DES blank (100 × ). (b) Protective effect against H₂O₂-induced injury at 100×, 200×, 400×, and 800 × with the matched DES blank (100 × ). Different lowercase letters indicate significant differences (P < 0.05).

3.10. Chemical characterization by UPLC-Q-Orbitrap HRMS

A comprehensive chemical profiling of B. balsamifera extract was conducted using UPLC-Q-Orbitrap HRMS, with the total ion chromatogram presented in Fig. S3. This analysis successfully identified 15 flavonoid compounds, including 7 flavonols, 4 flavones, and 4 dihydroflavones. Detailed information regarding the identified flavonoid compounds is provided in Table 7. Notably, six compounds, such as Quercetin-3,7-O-diglucopyranoside and 6,8-di-C-glucopyranosylnaringenin, were newly identified within the Blumea genus. Additionally, one compound, Rhamnocitrin, was identified in B. balsamifera for the first time. These findings substantially broaden the known phytochemical diversity of this medicinal plant.

Table 7.

Flavonoid compounds identified in B. balsamifera using UPLC-Q-Orbitrap HRMS.

No.	RT/min	Compound	Molecular Formula	Error/ppm	m/z	Ion Mode	Fragments (m/z)	Types
1	3.11	Quercetin-3,7-O-diglucopyranoside※	C27H30O17	−0.37	625.14065	[M − H]−	463.08887; 301.03537; 299.01981	flavonols
2	3.34	6,8-di-C-glucopyranosylnaringenin※	C27H32O15	−0.28	595.16665	[M − H]−	475.12476; 385.09308; 355.08252; 201.07132; 119.05032	flavones
3	4.00	Rutin	C27H30O16	−0.27	609.14571	[M − H]−	300.02765; 271.02502; 255.03011	flavonols
4	4.17	Hyperoside	C21H20O12	−0.54	463.08780	[M − H]−	303.04993; 85.02834; 61.02850	flavonols
5	4.18	Quercetin	C15H10O7	0.30	303.05004	[M + H]+	257.04489; 229.04951; 153.01828; 137.02327	flavonols
6	4.30	Kaempferol-3-O-rutinoside	C27H30O15	0.14	593.15106	[M − H]−	285.04050; 284.03278; 255.03303; 227.03516; 201.07144	flavonols
7	4.34	Narcissin※	C28H32O16	0.70	623.16217	[M − H]−	315.05145; 299.01956; 285.04190; 271.02505	flavones
8	4.52	Scutellarein※	C15H10O6	0.56	287.05522	[M + H]+	228.90855; 202.05618; 110.93182	flavonols
9	4.52	Astragalin	C21H20O11	−0.20	447.09308	[M − H]−	287.05505; 85.02839; 61.02853	flavones
10	5.46	RhamnocitrinΔ	C16H12O6	0.35	301.07077	[M + H]+	273.07516; 245.08018; 149.02299	flavonols
11	5.47	Cedeodarin※	C16H14O7	−1.45	317.06606	[M − H]−	320.08459; 302.89930; 302.56461; 301.07034	dihydroflavones
12	6.07	Eriodictyol	C15H12O6	−0.37	287.05593	[M − H]−	267.29251; 228.91705; 145.04977; 89.05982	dihydroflavones
13	8.23	Jaceosidin※	C17H14O7	0.09	329.06662	[M − H]−	316.05746; 301.03412; 273.04004; 245.04466; 228.91710	flavones
14	8.92	Blumeatin	C16H14O6	−0.51	301.07150	[M − H]−	167.03369; 163.03882; 135.04396	dihydroflavones
15	10.56	Sakuranetin	C16H14O5	0.21	287.09148	[M + H]+	167.03377; 147.04393; 119.04927	dihydroflavones

^※First discovered in plants of the genus Blumea; ^Δ First discovered in B. balsamifera.

Structural characterization revealed that the identified flavonoids, particularly rutin, astragalin, and cedeodarin, possess distinct polyhydroxyl substitution patterns (Fig. S4). These molecular architectures may enhance antioxidant capacity through hydrogen atom transfer mechanisms, thereby disrupting free radical propagation cascades. Rutin, a well-known antioxidant, serves as a representative example, exhibiting potent radical scavenging activity due to its abundant phenolic hydroxyl groups. The structural diversity of these flavonoids not only explains the strong antioxidant effects observed in B. balsamifera extracts but also identifies previously unreported phytochemical markers, providing a basis for quality standardization and further pharmacological investigations.

4. Conclusions

This study developed a sustainable and efficient method for extracting flavonoids from B. balsamifera using a DES-UAE system. ChCl-Gly at a molar ratio of 1:2 and a water content of 40% was selected as the optimal solvent. ANN-GA showed slightly better nonlinear fitting and optimization performance than RSM. Under the experimentally verified conditions of 40:1 mL/g, 77 °C, 45 min, and 300 W, the total flavonoid yield reached 3.52 ± 0.01%, representing a 73.40% increase over 50% ethanol heat-assisted extraction. Matched comparisons across both DES and ethanol systems consistently showed higher flavonoid yields under ultrasound-assisted conditions, confirming the important contribution of ultrasound to extraction efficiency. The power, kinetic, temperature, factorial, and SEM results were consistent with literature-reported physical effects of ultrasound, including matrix disruption and mass-transfer intensification; however, the acoustic field and cavitation activity were not directly measured. FTIR and ¹H NMR analyses further supported the formation and reorganization of hydrogen-bond interactions in the ChCl-Gly DES, which favored flavonoid desorption and solubilization. The DES-UAE extract exhibited strong radical-scavenging activity, and total flavonoid yield was positively correlated with DPPH (r = 0.82, p < 0.01) and ABTS (r = 0.75, p < 0.05) activities. At 100-fold dilution, the extract increased the viability of H₂O₂-injured L02 hepatocytes to 88.54 ± 3.94%, whereas the matched DES blank showed no measurable cytoprotective effect under the same injury conditions. UPLC-Q-Orbitrap HRMS identified 15 flavonoids, including compounds newly reported in the Blumea genus and in B. balsamifera. After three recovery and reuse cycles, the regenerated DES retained 90.57% of its initial extraction yield. The maximum deviations in viscosity and density were 8.97% and 1.36%, respectively, while its principal ¹H NMR signals remained largely unchanged, suggesting limited physicochemical changes during short-term recycling. These results provide preliminary evidence of the extraction efficiency, antioxidant-related bioactivity, and short-term reusability of the DES-UAE process under laboratory conditions. The present findings demonstrate laboratory-scale feasibility rather than industrial scalability. Direct acoustic-power measurement, longer-term DES recycling with composition monitoring, and pilot-scale validation are required before industrial application.

CRediT authorship contribution statement

Wei Dai: Writing – review & editing, Writing – original draft, Software, Funding acquisition, Conceptualization. Liping Dai: Writing – original draft, Methodology, Data curation. Lei Jin: Validation, Investigation, Data curation. Rui Pang: Validation, Investigation. Sina Wang: Software, Resources. Hongfeng Chen: Writing – review & editing, Project administration, Methodology, Funding acquisition, Conceptualization. Xilong Zheng: Writing – review & editing, Project administration, Funding acquisition, Conceptualization.

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:

Supplementary Data 1

The supplementary material contains the nominal energy input data, ANN-GA model optimization and prediction results, comparative 1H NMR spectra of fresh and regenerated DES, UPLC–Q-Orbitrap HRMS total ion chromatograms, and the chemical structures of the identified flavonoids.