Tunable and Recyclable Natural Deep Eutectic Solvents for Efficient Extraction of Steviol Glycosides from Stevia rebaudiana

Abstract

Steviol glycosides (SGs), particularly rebaudioside A (RA) and stevioside (STV), are the primary natural sweeteners in Stevia rebaudiana and are increasingly utilized as zero-calorie sucrose substitutes. Although water is environmentally friendly, its poor selectivity limits efficient extraction of SGs, while conventional organic solvents raise environmental concerns. Auxiliary techniques such as ultrasound-assisted extraction often require high energy input or specialized equipment. Natural deep eutectic solvents (NADES) offer a sustainable, tunable alternative for SGs extraction. In this study, the Conductor-like Screen Model for Real Solvent (COSMO-RS) was applied to screen 403 DES combinations, revealing that NADES with strong hydrogen-bonding abilities significantly enhance SGs extraction. Experimental validation identified a betaine- propionic acid NADES (1:2 molar ratio) with 45 wt % water as optimal, achieving yields of 98.15 ± 2.62 mg/g RA, 59.83 ± 1.60 mg/g STV, and 7.16 ± 0.45 mg/g rebaudioside C (RC). To address challenges in SGs recovery from NADES, static adsorption/desorption experiments were conducted using AB-8 resin. Hydrophilic NADES were effectively recovered with deionized water washing and maintained over 85% extraction efficiency after two reuse cycles. Target SGs were efficiently desorbed with 70% ethanol, while impurities were removed with 30% ethanol. The resulting yields of RA, STV, and RC were 90.56%, 86.68%, and 65.89%, with corresponding purities of 53.23%, 29.06%, and 4.42%, respectively. These results demonstrate that the combination of the betaine-propionic acid NADES and AB-8 resin offers an effective and sustainable strategy for the extraction and purification of SGs from S. rebaudiana.

1. Introduction

S. rebaudiana (Bertoni), commonly known as honey chrysanthemum or stevia, is a perennial herbaceous plant belonging to the genus Stevia in the Asteraceae family. Native to the border region between Paraguay and Brazil, it was introduced to China in 1977 and is now widely cultivated in Shandong, Jiangsu and Xinjiang. Stevia leaves are characterized by their low caloric content (∼2.7 kcal/g) and contain over 30 diterpene glycosides, collectively referred to as steviol glycosides (SGs), which are the primary natural sweetening compounds. Among them, stevioside (STV) and rebaudioside A (RA) are the most abundant, accounting for approximately 14–18% of dry matter, and exhibit sweetness 200–300 and 350–450 times that of sucrose, respectively, with high thermal, acid–base, and fermentation stability. In 2008, the Joint FAO/WHO Expert Committee on Food Additives (JECFA) established an acceptable daily intake (ADI) of 4 mg/kg body weight. Subsequently, the European Union approved stevia as a low-calorie, low-toxicity natural sweetener suitable for use in the food and beverage industry. Driven by the rising demand for natural, noncaloric sweeteners, SGs have increasingly replaced sucrose and fructose, emerging as one of the most popular natural sweeteners and being recognized as the world’s “third-generation natural zero-calorie healthy sugar source”. Beyond SGs, S. rebaudiana leaves are rich in other phytochemicals, including phenolics, flavonoids, and carotenoids, which exhibit anti-inflammatory, antioxidant, antihypertensive, and diuretic effects. Growing evidence supports the medicinal potential of SGs. For instance, Ahmad et al. reported that stevia extract exhibits antihyperlipidemic effects in albino rats, suggesting its potential as a nutraceutical for hyperlipidemia. Maki et al. found that continuous RA consumption for 16 weeks enhanced insulin secretion, supporting its therapeutic potential in type 2 diabetes. Tadhani et al. identified phenolic compounds and alkaloids in stevia leaf extracts with strong reactive oxygen species (ROS) scavenging activity, indicating potential in cancer and cardiovascular disease prevention. The combination of natural sweetness and therapeutic benefits positions SGs as valuable resources for both pharmaceutical and food applications. Currently, most natural functional products (e.g., sweeteners, antioxidants, antimicrobials, and antidiabetic agents) are obtained through extraction.

Traditional methods for SG extraction, including Soxhlet extraction, fermentation, and boiling, commonly employ hot water or aqueous ethanol solutions, and are associated with long extraction times, low yield and purity, and operational complexity. Enhanced techniques, such as ultrasound-assisted extraction (UAE), microwave-assisted extraction, supercritical fluid extraction, and subcritical water extraction, often require energy-intensive processes or significant investment in specialized equipment. Therefore, there is an urgent need to develop novel, efficient, and environmentally friendly extraction technologies that reduce processing time, improve yield, and maintain product quality.

Deep eutectic solvents (DESs) have emerged as a sustainable alternative aligned with green extraction principles. DESs are typically formed by mixing hydrogen bond donors (HBDs), such as acids, amines, alcohols, or carbohydrates, with hydrogen bond acceptors (HBAs), such as quaternary ammonium or phosphonium salts. Extensive hydrogen bonding leads to charge delocalization and significantly lowers the melting point of the mixture compared to its individual components. Compared with ionic liquids, DESs offer advantages including lower production costs, higher biodegradability, and reduced toxicity, making them attractive for sustainable extraction protocols. Natural deep eutectic solvents (NADES), a subclass of DES, are composed of biocompatible and biodegradable natural compounds. Common HBAs in NADES include choline chloride, betaine, proline, or carnitine, while HBDs typically comprise urea, polyols, organic acids, or sugars. The synthetic flexibility of NADESs allows for the design of solvents with targeted functionality-hydrophilic, hydrophobic, acidic, basic, neutral, or biphasic-making them highly tunable. In recent years, NADES have been increasingly used for the green extraction of natural products. For example, Nam et al. discovered that a l-proline-glycerol (2:5) NADES outperformed methanol in extracting flavonoids from Sophora japonica. Bertolo et al. employed a choline chloride-lactic acid NADES to effectively extract phenolics from pomegranate peel, and Guo et al. developed a ternary DES (choline chloride:1,3-butanediol:D-sorbitol) for extracting bioactive polysaccharides from Poria cocos.

For the extraction of SGs, Milanir et al. employed UAE using a NADES composed of tetraethylammonium chloride and ethylene glycol (1:2, molar ratio), achieving efficient recovery of STV and RA from S. rebaudiana, with extraction yields of 14.21 ± 0.21 mg/g and 21.4 ± 1.3 mg/g, respectively. In another study, Suresh et al. developed quaternary NADES systems combined with UAE. Among the tested systems, NADES 50 (lactic acid/glycerol/malic acid/glucose, 1:1:1:1) exhibited the highest extraction efficiency for RA and STV, yielding 38.24 ± 2.22 mg/g and 114.58 ± 5.89 mg/g, respectively. Although the overall extraction yields of RA and STV were relatively high, the recovery of RA remained unsatisfactory considering its higher natural abundance in stevia leaves and its superior sweetness profile. While NADES have demonstrated considerable potential for SG extraction, existing studies generally rely on auxiliary extraction techniques or complex multicomponent DES systems. These approaches often involve energy-intensive processes, substantial equipment investment, challenges in maintaining solvent stability and compositional consistency during extraction, and limited improvement in RA extraction yields. Moreover, DES systems commonly suffer from high viscosity, difficulties in the separation and purification of target compounds, and complicated solvent recovery processes, all of which hinder their practical application. Therefore, it is necessary to further systematically investigate the applicability, extraction performance, and process characteristics of NADES systems for the extraction of SGs.

Traditional methods for separating and purifying SGs include recrystallization, microporous resin adsorption (MRA), ion-exchange chromatography, high-performance liquid chromatography (HPLC), thin-layer chromatography, high-speed counter-current chromatography, and membrane separation technology. Some of these methods, however, suffer from high solvent consumption, long processing times, complex procedures, limited sample throughput, and industrial-scale limitations. Therefore, developing efficient and selective separation techniques is crucial. MRA has gained attention due to its physical adsorption selectivity, large surface areas, and interactions like van der Waals forces. Macroporous resins are reusable, cost-effective, and suitable for large-scale production. For instance, Yang et al. used a choline chloride-malonic acid NADES to extract steroidal saponins from Dioscoreae Nipponicae Rhizoma, followed by purification with D101 resin, achieving recovery rates of 67.27–79.90%. Wu et al. employed a lactic acid-glucose NADES combined with DM130 resin to isolate rhubarb anthraquinones, achieving purity and recovery exceeding 90% and 78%, respectively. Similarly, Wang et al. enriched flavonoids from Carthamus tinctorius using NKA-2 resin in the ethylene glycol-choline chloride biphasic system, demonstrating high adsorption and desorption efficiency.

This study aims to develop and optimize tunable, recyclable NADES for extracting SGs (specifically STV and RA) from S. rebaudiana leaves. The Conductor-like Screening Model for Real Solvent (COSMO-RS) model was employed to predict the optimal binary NADES composition to maximize SGs solubility and extraction efficiency. Subsequently, MRA was employed to purify the crude NADES extract, achieving high recovery and purity of the target glycosides. The reusability of NADES was also evaluated to reduce production costs and minimize waste generation. This research is expected to advance the deep processing of S. rebaudiana and other natural plants in response to the growing demand for natural sweeteners, while promoting environmentally friendly extraction technologies.

2. Materials and Methods

2.1. Materials and Chemicals

Stevia leaves were purchased from Shaanxi Gaochang Biomedicine Co., Ltd. The standard substances of RA, STV, and RC­(≥98% purity)­were all purchased from Beijing Huabo Station Bioanalysis Technology Co., Ltd. Carnitine (98% purity), choline chloride (98% purity), citric acid (99.5% purity), formic acid (HPLC grade), and acetonitrile (HPLC grade) were purchased from Shanghai Aladdin Co., Ltd. Propionic acid, methanol, and ethanol were supplied by Qiguang Kemao Industrial Co., Ltd. (Shaanxi), all with purity >99%. Proline (99%, Shanghai Macklin Biochemical Technology Co., Ltd.), betaine (98%, Tianjin Baima Technology Co., Ltd.), tetraethylammonium chloride (98%, Shanghai Yuanye Bio-Technology Co., Ltd.), ethylene glycol (≥99%, Tianjin Kemiou Chemical Reagent Co., Ltd.), urea (≥99%, Sinopharm Chemical Reagent Co., Ltd.), levulinic acid (99%, Shanghai Meirui Biochemical Technology Co., Ltd.), oxalic acid (99.99%, Shanghai Shenze Chemical Technology Co., Ltd.), tartaric acid (≥98%, Shanghai Haohong Biomedical Technology Co., Ltd.), and resorcinol (99%, Kaimate (Tianjin) Chemical Technology Co., Ltd.).

2.2. COSMO-RS Simulation Screening and Computational Details

COSMO-RS is a powerful thermodynamic tool used to predict solubility and activity coefficients by calculating the chemical potential of molecules in solution. Based solely on molecular structure, COSMO-RS can forecast solubility and related thermodynamic properties, making it valuable for designing efficient extraction processes. Using COSMO-RS for solvent screening can enhance experimental efficiency and enable the prediction of SG solubility in designed NADES systems. Based on literature reports, 13 HBAs commonly used for the extraction of natural compounds were selected: glycerol, xylose, maltose, glucose, carnitine, arginine, proline, glycine, alanine, ammonium acetate, betaine, choline chloride, and tetraethylammonium chloride; and 31 HBDs: ethylene glycol, glycerol, urea, lactic acid, oxalic acid, malonic acid, malic acid, propionic acid, maleic acid, levulinic acid, ascorbic acid, glutamic acid, tartaric acid, aconitic acid, citric acid, caffeic acid, erythritol, xylitol, D-sorbitol, L-rhamnose, resorcinol, D-mannose, fructose, galactose, xylose, glucose, sucrose, maltose, L-sorbose, trehalose, and raffinose. The molecular structures of all HBAs, HBDs, and target compounds (STV and RA) were constructed using GaussView 5.0. Subsequently, the optimal geometries of each component were calculated and optimized using Gaussian 09 with the DFT/B3LYP/6–31G­(d,p) basis set. The resulting output files (.out files) were converted into.mol files in GaussView and imported into TURBOMOLE 6.4 software. Further geometry optimization was performed using the BP-TZVP basis set, generating *.cosmo files for subsequent analysis in COSMOthermX. For any extraction process, efficiency is directly related to solute solubility in the solvent. COSMOthermX was used to predict σ-surfaces, σ-profiles, σ-potentials of HBAs, HBDs, and target compounds, as well as the relative solubility of solid SGs in various NADES at infinite dilution (noniterative computation). In this study, COSMO-RS was used to predict the relative solubility of two SGs (STV and RA) in NADES at 70 °C and infinite dilution conditions. The relative solubility was calculated using eq ()

$$\mathit{lg}{X}_i=\mathit{lg}\{\exp [({{\mu }_i}^{\mathrm{pure}}-{{\mu }_i}^{\mathrm{solvent}}-\Delta G_{G\mathrm{fusion}})/\mathit{RT}]\}$$ 1

where X _i represents the relative solubility of the solute, μ _i ^pure and μ _i ^solvent are the chemical potentials of the pure solute i (i.e., steviol glycoside) and the solute i at infinite dilution in the solvent; ΔG _fusion is the Gibbs free energy change of fusion; R is the universal gas constant; and T is the temperature in Kelvin.

2.3. Preparation of NADES

Through COSMO-RS screening, 14 promising binary DES combinations were selected (Table S1). The DESs were prepared following the method described by Milani et al. Briefly, the required amounts of HBA and HBD (Table S1) were accurately weighed and mixed at the specified molar ratio in a glass vessel. The mixture was then heated at 82 °C under continuous magnetic stirring until a clear and homogeneous liquid was obtained. After cooling to room temperature, the resulting DES remained clear and transparent, indicating successful formation. The prepared DESs were subsequently stored at room temperature for 20 days to evaluate their physical stability. Visual inspection was conducted daily, and no signs of phase separation, turbidity, or precipitation were observed. Furthermore, microscopic examination (×400) of samples collected on days 1, 10, and 20 revealed no particles, crystal formation, or phase separation, confirming that the DESs remained homogeneous and transparent throughout the storage period (Figure S1).

2.4. Extraction Protocol and HPLC Quantification

The extraction of SGs from S. rebaudiana leaves was performed as follows: First, the dried leaves were ground using an electric grinder (Yongkang Jiashante Trading Co., Ltd.) and sieved through an 80-mesh sieve. Subsequently, the resulting powder was dried at 80 °C in an oven until constant weight was achieved. Next, 1.00 g of the powder was mixed with 10 mL of a 50% NADES aqueous solution. Extraction was carried out with stirring at 70 °C for 60 min. Following extraction, the supernatant was collected, transferred to a 2 mL centrifuge tube, and centrifuged at 9000 g for 15 min. The resulting supernatant was diluted with methanol and filtered through a 0.45 μm PTFE filter prior to analysis. Quantitative analysis of the extracts was performed using high-performance liquid chromatography with a diode array detector (HPLC-DAD). A Waters Symmetry C₁₈ column (250 × 4.6 mm², 5 μm) was used as the separation column, with a column temperature of 40 °C. The mobile phase consisted of (A) 0.1% formic acid in water and (B) acetonitrile, with a gradient elution program detailed in Table S2. This method achieved complete separation of the three target compounds within 20 min. The flow rate was 1 mL/min, the injection volume was 5 μL, and the detection wavelength was set at 200 nm.

2.5. SEM Analysis

The morphology of the Stevia powder after DES extraction, water extraction, and untreated powder, were examined using a scanning electron microscope (SEM, MAIA3 LMH, TESCAN, Czech Republic). Samples from each group were mounted on a specimen stub, sputter-coated with gold, and imaged under high vacuum conditions.

2.6. Screening of Extraction Conditions

To enhance the extraction efficiency of steviol glycosides from S. rebaudiana, the extraction conditions for the selected NADES system were systematically screened and validated. Five key parameters were investigated for their impact on extraction performance by Suresh et al., including HBA/HBD molar ratio (1:1, 1:2, 1:3, 2:1, and 3:1), water content (5, 10, 25, and 50%, v/v), solid-to-liquid ratio (0.25:10, 0.5:10, 1:10, and 2:10, w/v), extraction temperature (25, 50, 60, 70, and 90 °C), and extraction time (30, 60, 90, 120, and 150 min). Each parameter was evaluated independently using the extraction yields of the target SGs. STV, RA, and RC were quantified via HPLC and all experiments were performed in triplicate.

2.7. Selection of Macroporous Resins

Four macroporous resins (X-5, AB-8, D101, and NKA-9, which were purchased from Tianjin Nankai Hecheng Technology Co., Ltd.) were evaluated for their adsorption and desorption capabilities. For each test, 1 g of resin was added to 10 mL of extract and shaken (135 rpm, 30 °C, 24 h). After adsorption, the resin was filtered, rinsed with water, and desorbed with 20 mL of 70% ethanol (120 rpm, 24 h, room temperature). Concentrations of RA and STV were measured by HPLC. All experiments were performed in triplicate, adsorption and desorption capacities were calculated as follows

$$q_e=\frac{(C_0-C_e)V_i}{W}$$ 2

$$A(\%)=\frac{C_0-C_e}{C_0}\times 100\%$$ 3

$$q_d=\frac{C_d{V}_d}{W}$$ 4

$$D(\%)=\frac{C_d{V}_d}{(C_0-C_e)V_i}\times 100\%$$ 5

where q _e and q _d (mg/g) refer to the adsorption and desorption capacities, respectively; C ₀ and C _e (mg/mL) refer to the initial and equilibrium concentrations, respectively; V _i and V _d (mL) refer to the sample and eluent volumes, respectively; A (%) and D (%) refer to the adsorption ratio and desorption ratio, respectively; and W (g) refers to the dry weight of the resin.

2.8. Effect of pH on Adsorption

The effect of pH on adsorption efficiency was investigated across a range of values (pH = 5, 6, 7, 8, 9, 10). For each test, 1 g AB-8 resin was added to 20 mL of pH-adjusted crude SGs extract. The mixtures were then shaken at 135 rpm and 40 °C for 13 h. The adsorption performance was evaluated based on the adsorption quantity and the adsorption ratio.

2.9. Selection of Desorption Solvent

The effect of the desorption solvent was investigated using ethanol aqueous solutions (20–100% ethanol). After adsorption, the resin was filtered, rinsed with water, and then desorbed with 20 mL of the respective solvent under room temperature agitation (120 rpm, 24 h). Desorption performance was evaluated based on the desorption quantity and the desorption ratio.

2.10. Selection of Wash Solvent

An appropriate wash solution can effectively remove impurities adsorbed on the resin without causing significant loss of the target components, thereby enhancing the purity of STV, RA, and RC during separation and purification. AB-8 resin was packed into a column (2.4 cm i.d. × 8 cm height). The column was rinsed with pure water, then loaded with 20 mL of pH-adjusted crude SGs extract. NADES was efficiently removed by washing with deionized water. Impurities were subsequently eluted with 20 mL wash solvent, and finally, the SGs were desorbed with 40 mL of 70% ethanol solution. The flow rate was consistently maintained at 0.5 mL/min throughout the entire process. The effect of different wash solventsethanol solutions ranging from 10% to 40%was evaluated based on the recovery rate and purity of the three target SGs.

2.11. Statistical Analysis

All statistical analyses were carried out using SPSS software (Version 25.0, IBM Inc., New York). Data were presented as mean ± standard deviation (n = 3). Differences between groups were assessed by one-way ANOVA and Tukey test, with significance set at P < 0.05.

3. Results and Discussion

3.1. Screening of DES via COSMO-RS Model

3.1.1. The Polarity Distribution of the Molecular Surfaces

COSMO-RS predicts the thermodynamic properties of compounds by analyzing the polarity distribution across their molecular surfaces, characterized through σ-surface, σ-profile, and σ-potential descriptors. The σ-surface provides a three-dimensional visualization of the shielded charge distribution on the molecular surface. This distribution is further converted into the σ-profile, a histogram quantifying the proportion of surface area exhibiting a specific polarity value σ. The σ-potential, μS­(σ), represents the chemical potential of a surface segment with polarity σ and indicates the affinity of a solvent system “S” for that polarity. The σ-profile and σ-potential distributions are typically divided into three regions: the hydrogen bond (H-bond) donor region, the nonpolar region, and the hydrogen bond acceptor region. Figure (a) shows the σ-profiles of two steviol glycosides RA and STV. The shielded charge distribution spans [-0.022, + 0.019] e/Ų for RA and a slightly narrower [-0.02, + 0.019] e/Ų for STV. Both compounds exhibit distributions in the H-bond donor and acceptor regions, with a slightly broader distribution observed in the donor region. The σ-potential curves for RA and STV are highly similar (Figure (b)). In the H-bond acceptor region, the minimum μ­(σ) values range from −0.488 to −0.476 kcal/mol·Å², confirming their acceptor capability. In the H-bond donor region, the minimum μ­(σ) values range from −0.62 to −0.617 kcal/mol·Å², indicating stronger donor abilities for both RA and STV, consistent with their σ-profiles. Based on the above analysis, RA and STV are strongly polar compounds that act as both hydrogen bond donors and acceptors, although with a slightly stronger donor capacity.

1.

σ-profiles (a) and σ-potentials (b) of RA and STV; σ-potentials of HBAs (c); σ-potentials of HBDs (d–f).

As shown in Figure (c), the σ-potential distributions for 13 HBAs have minimum μ­(σ) values range from −1.571 to −0.293 kcal/mol·Å² in the H-bond acceptor region. Betaine (−1.508 kcal/mol·Å²), carnitine (−1.571 kcal/mol·Å²), and tetraethylammonium chloride (−1.27 kcal/mol·Å²) show the most negative values, indicating vary strong H-bond acceptor capabilities. This computational finding correlates directly with their subsequent superior performance in solubilizing the target substances (Figure ). The donor capabilities of the HBAs, assessed at σ = 0.03 e/Ų, were more varied. Figure (d–f) shows the σ-potential profiles for 31 HBDs. The 13 organic acids consistently showed higher (less negative) μ­(σ) values in the acceptor region than in the donor region (Figure (d)). Their minimum μ­(σ) values in the donor region ranged from −1.235 to −0.846 kcal/mol·Å², with tartaric acid, citric acid and oxalic acid exhibiting the strongest donor capabilities. Among the alcohols and urea, resorcinol exhibited the lowest μ­(σ) value of −1.079 kcal/mol·Å² at 0.03 e/Ų, demonstrating the strongest donor capability (Figure (e)). The 11 sugars (Figure (f)) displayed moderate acceptor (min μ­(σ): −0.611 to −0.364 kcal/mol·Å²) and donor (min μ­(σ): −0.809 to −0.56 kcal/mol·Å²) capabilities, indicating a bifunctional nature with a generally stronger propensity to act as donors. Overall, organic acids were identified as the strongest H-bond donors.

2.

Relative solubility of SGs in 403 NADES (HBA/HBD = 1:2) at 70 °C predicted by COSMO-RS. (a) RA; (b) STV.

3.1.2. Relative Solubility Predicted via COSMO-RS

The relative solubility of RA and STV was predicted for 403 NADES systems constructed from the 13 HBAs and 31 HBDs at a 1:2 molar ratio, using eq (). The results are presented as a heatmap in Figure , where red and blue represent high and low predicted solubility, respectively. The solubility trends for RA and STV across the 403 DESs were remarkably similar. This is expected due to their closely related tetrahydro-diterpenoid structures and their nearly identical σ-profile and σ-potential curves, which form the primary basis for the COSMO-RS calculation. The heatmap identified specific HBA-HBD combinations that yielded high solubility for both RA and STV. High performance was consistently associated with HBAs possessing strong H-bond acceptor abilities, such as betaine, carnitine and tetraethylammonium chloride, paired with HBDs exhibiting strong H-bond donor abilities, such as oxalic acid, tartaric acid, propionic acid, ethylene glycol and urea. This result suggests that hydrogen bonding interactions between the solvent components and the solutes play a crucial role in the dissolution process of SGs in these NADES.

3.1.3. Experimental Verification of the Screening Results

Based on the COSMO-RS prediction results, 14 DESs (Table S1) were selected for experimental validation. Extraction was performed at 70 °C with a liquid–solid ratio of 10 mL/g for 120 min. The steviol glycosides content in the extracts, quantified by HPLC, is shown in Figure . The results indicated a betaine-propionic acid (HBA/HBD) aqueous solution yielded the highest steviol glycosides content of 169.8 mg/g, significantly surpassing the water control. Therefore, the betaine-propionic acid DES was selected as the extraction solvent for subsequent experiments.

3.

Extraction of steviol glycosides using 14 selected DES and water (n = 3). (solid–liquid ratio: 1:10 g/mL, time: 120 min, and temperature: 70 °C).

3.2. Morphological Properties

SEM analysis was employed to examine morphological changes in stevia leaf powder before and after extraction. As shown in Figure A, the untreated powder exhibited a relatively dense, uneven surface. After water extraction (Figure B), the leaf surface remained largely intact, with slight shrinkage of the cell wall and the appearance of some pores, indicating partial disruption of fibrous tissue and limited extraction of SGs. In contrast, DES extraction (Figure C) caused pronounced fragmentation and collapse of the leaf surface, with severe disruption of the cell wall and exposure of internal tissues. Numerous voids, fissures, and dense micropores were observed at higher magnification (Figure S2), demonstrating that DES effectively created pathways for solvent penetration. These observations confirm that DES induces more extensive and efficient disintegration of leaf structure compared to water, correlating with its superior extraction performance.

4.

SEM photographs of stevia powder untreated (A), after water extraction (B) and after DES (C) extraction.

3.3. Screening of Extraction Conditions

To maximize the yield of SGs, key extraction parameters, including HBA/HBD ratio, water content, solid–liquid ratio, extraction temperature, and extraction time were systematically screened via single-factor experiments. The results are summarized in Figure .

5.

Effects of five factors on SGs extraction yields (n = 3). (a) HBA-HBD ratio (water content: 50% v/v; solid–liquid ratio: 1:20 g/mL; temperature: 30 °C; time: 30 min); (b) water content (HBA/HBD = 1:2; solid–liquid ratio: 1:20 g/mL; temperature: 30 °C; time: 30 min); (c) solid–liquid ratio (HBA/HBD = 1:2; water content: 45% v/v; temperature: 30 °C; time: 30 min); (d) extraction temperature (HBA/HBD = 1:2; water content: 45% v/v; solid–liquid ratio: 1:10 g/mL; time: 30 min); and (e) extraction time (HBA/HBD = 1:2; water content: 45% v/v; solid–liquid ratio: 1:10 g/mL; temperature: 70 °C).

3.3.1. Effect of HBA/HBD Ratio

Figure (a) illustrates the effect of different HBA:HBD ratios on extraction efficiency. The results demonstrated that the extraction yields for all three steviol glycosides (RA, STV, and RC) initially increased and then decreased with varying HBA:HBD ratios. The highest yields were achieved at a 1:2 ratio. DSC analysis was performed on DES formulations with HBA:HBD molar ratios of 1:1, 2:1, 3:1, and 1:2 to evaluate their eutectic behavior (Figure S3). The 1:2 ratio exhibited a single, sharp endothermic peak at 45.9 °C, confirming the formation of a true eutectic phase between HBA and HBD (pure propionic acid melts at −22 °C, pure betaine at ∼300 °C). The 1:1 ratio showed a small peak at −2.1 °C, indicating supercooling and minimal crystallizationa useful solvent property, though not a classical eutectic. For HBA-rich compositions, the 2:1 ratio had a single peak at 47 °C, while the 3:1 ratio showed two peaks at 4.8 and 36 °C, suggesting phase separation and noneutectic behavior. These results confirm the 1:2 ratio as the genuine eutectic composition, whose enhanced extraction performance can be attributed to favorable physicochemical properties, such as hydrogen-bonding networks and optimal viscosity, rather than chance. Consequently, an HBA/HBD ratio of 1:2 was selected for subsequent experiments.

3.3.2. Effect of Water Content

Viscosity is a crucial property of DESs, which significantly influences mass transfer and extraction efficiency. Water addition effectively reduces viscosity. Florindo et al. found that incorporating water can reduce the viscosity of Choline Chloride-based DESs (using carboxylic acids as HBDs) by factors ranging from 10 to 200, depending on the specific HBD. As shown in Figure (b), extraction yields of steviol glycosides increased with water content up to 45%, reaching maxima of 96.36 mg/g for RA, 59.10 mg/g for STV, and 7.18 mg/g for RC. Beyond this point, further dilution reduced yields, likely due to altered solvent polarity or disruption of the DES hydrogen-bonding network. Therefore, 45 wt % water was identified as optimal.

3.3.3. Effect of Solid–Liquid Ratio

The solid–liquid ratio significantly impacts extraction efficiency. Figure (c) illustrates the effects of solid–liquid ratio (1:40, 1:20, 1:10, and 2:10 g/mL) on extraction yields of RA, STV, and RC. As the solid–liquid ratio decreased from 1:40 to 1:10, the extraction yields showed an upward trend. However, increasing the ratio to 2:10 g/mL resulted in a significant decrease in yield. Therefore, a solid–liquid ratio of 1:10 g/mL was determined to be optimal, achieving extraction yields of 98.84 mg/g for RA, 59.21 mg/g for STV, and 7.31 mg/g for RC.

3.3.4. Effect of Extraction Temperature

Temperature affects solvent properties such as viscosity, solubility and diffusion coefficient. As shown in Figure (d), extraction yields increased with temperature up to 70 °C, achieving 98.34 mg/g for RA, 50.19 mg/g for STV, and 6.92 mg/g for RC. This enhancement is likely due to increased temperature accelerating diffusion, reducing DES viscosity, and facilitating plant cell wall disruption. Further increasing temperature beyond 70 °C (70–90 °C) led to a decline in extraction efficiency, potentially due to heat-induced degradation of some components or solvent evaporation. Therefore, 70 °C was selected as the optimal extraction temperature for steviol glycosides. This is close to the 75 °C reported by Lòpez-Carbon et al. as the optimal temperature for stevioside extraction.

3.3.5. Effect of Extraction Time

Extraction time is crucial for achieving complete solute recovery. Figure (e) shows the effect of extraction time (30–150 min) on extraction efficiency. The results indicated that yields of steviol glycosides increased up to 120 min and plateaued thereafter, indicating complete extraction. Therefore, 120 min was selected as optimal duration. In summary, the optimized extraction conditions were determined as follows: betaine–propionic acid (1:2) with 45 wt % water, solid–liquid ratio of 1:10 g/mL, temperature of 70 °C, and extraction time of 120 min. Under these conditions, the extraction yields of RA, STV, and RC from steviol glycosides reached 98.15 ± 2.62, 59.83 ± 1.60, and 7.16 ± 0.45 mg/g, respectively.

3.4. Purification via Microporous Resin Adsorption

3.4.1. Resin Screening

The adsorption and desorption performance of macroporous resins is influenced by their physicochemical properties, such as surface polarity, particle size, pore radius, and surface area. Both adsorption and desorption capacity are generally considered as the main criteria for selecting an appropriate resin for purifying SGs. Four resins with distinct characteristics were evaluated for their adsorption and desorption of RA, STV, and RC. As summarized in Table , NKA-9 exhibited the highest adsorption capacity for the three targets, likely due to its polar phenolic hydroxyl groups enhancing affinity via the principle of “like dissolves like”. D101, conversely, exhibited the best desorption capacity, but the poorest adsorption. Although NKA-9 and X-5 displayed nearly comparable adsorption capacities, potentially attributed to their high surface areas (Table S3), they performed poorly in desorption. AB-8 offered good adsorption (comparable to NKA-9 and X-5) and the second-best desorption capacity, likely facilitated by its large pore size. Considering both adsorption and desorption performance comprehensively, AB-8 was selected for subsequent studies.

1. Adsorption Capacity/Rate and Desorption Capacity/Rate of Resins.

target analytes	resins	adsorption capacity qe (mg/g resin)	adsorption rate	desorption capacity qd (mg/g resin)	desorption rate
RA	X-5	22.51	96.04	16.29	72.38
	AB-8	22.06	94.16	19.35	87.71
	D101	21.17	89.46	19.52	92.20
	NKA-9	22.55	96.23	17.84	79.11
STV	X-5	11.32	96.02	8.49	74.99
	AB-8	11.10	94.08	10.04	90.47
	D101	10.55	89.41	9.97	94.58
	NKA-9	11.37	96.39	9.46	83.23
RC	X-5	1.75	100.00	1.19	67.87
	AB-8	1.75	100.00	1.36	77.61
	D101	1.58	90.02	1.42	90.12
	NKA-9	1.75	100.00	1.31	74.71

3.4.2. Effect of Sample Solution pH on Adsorption

pH significantly influences adsorption by altering the interaction between SGs and resin binding sites. As shown in Figure , the adsorption capacity remained stable within a pH range of 7 to 8. A significant decrease was observed at pH exceeding 10, indicating that strongly alkaline conditions are unfavorable for the adsorption of SGs. Therefore, a weakly alkaline environment (pH = 8) was identified as the optimal condition for SGs. Given the inherent acidity of the NADES (composed of betaine and propionic acid), the pH of the extractant solution was adjusted to 8 before sampling onto the AB-8 resin column for purification.

6.

Effect of pH on SGs adsorption capacity (n = 3).

3.4.3. Selection of Desorption Solvent

SGs, as tetracyclic diterpenoid compounds, are readily soluble in water and slightly soluble in ethanol, with solubility varying significantly in different ethanol–water mixtures. As shown in Figure , when the ethanol concentration increases to 70%, the desorption quantity and desorption rate of all three glycosides increased markedly: RA from 1.27 mg/g to 15.54 mg/g (desorption rate from 7.59% to 92.88%), STV from 2.10 mg/g to 8.23 mg/g (desorption rate from 25.18% to 98.53%), and RC from 0.31 mg/g to 1.14 mg/g (desorption rate from 24.76% to 89.99%). Beyond 70% ethanol, the changes in desorption quantity and desorption rate become negligible. Therefore, 70% ethanol was selected as the optimal desorption solvent.

7.

Selection of desorption solvent (n = 3). (a) RA; (b) STV; (c) RC.

3.4.4. Selection of Wash Solvent

From the yield and purity graphs of RA, STV, and RC using different ethanol concentrations as wash solvents, it can be observed that, based on the objective of eluting more impurities without desorbing SGs, analysis reveals that when 10%–30% ethanol is used as the wash solvent, the yield of SGs is the highest and the difference is minimal (Figure (a)). As the ethanol concentration increases, the yield of SGs begins to decline, which may be due to the high concentration of ethanol eluting a large amount of the target substances, consistent with the conclusion in section . When 30% ethanol is used as the wash solvent, the purity of SGs is the highest (Figure (b)). Comprehensive analysis suggests selecting 30% ethanol as the wash solvent.

8.

Yields (a) and purities (b) of RA, STV, and RC with different wash solvents (n = 3).

3.5. Reusability of NADES

Considering economic factors, the reusability of the NADES solvent was evaluated. After extraction, the pH of the extractant was adjusted to 8 and it was loaded onto an AB-8 resin column. The hydrophilic NADES was then recovered by washing the column with deionized water, followed by rotary evaporation to remove water. The regenerated NADES solvent was reused under optimized extraction conditions. As shown in Figure , the betaine/propionic acid NADES retained over 85% of its initial extraction efficiency after two cycles. Even after the third and fourth cycles, extraction efficiencies remained high at 76.7% and 74.66%, respectively, outperforming the control group (72.85%). These results indicate the excellent reusability of NADES in the extraction of SGs.

9.

Effect of NADES reuse on extraction efficiency of SGs (n = 3).

3.6. Integrated Extraction and Purification Process

The overall extraction and purification scheme developed in this study is illustrated in Figure . Briefly, dried stevia leaves were ground and sieved to obtain stevia powder, which was then extracted using the optimized NADES system (betaine/propionic acid = 1:2 + 45 wt % water). The resulting crude extract adjusted to pH 8 was then subjected to purification using AB-8 macroporous resin. Hydrophilic NADES was effectively washed from the resin column using deionized water. The collected NADES-water effluent can be concentrated via rotary evaporation to recover and recycle the NADES for subsequent extraction cycles. Impurities such as inorganic salts, proteins, and organic acids were eluted using 30% ethanol. Finally, the target SGs were desorbed efficiently using 70% ethanol. The yields of RA, STV, and RC were 90.56%, 86.68%, and 65.89%, respectively; the purities were 53.23%, 29.06%, and 4.42%, respectively (Figure ).

10.

Scheme for the extraction and purification of SGs from stevia leaves.

3.7. Comparison of the Proposed Method with Other Extraction Approaches

Current research has increasingly focused on developing environmentally friendly, safe, and efficient methods for extracting SGs from stevia leaves. In this study, a natural, low-toxicity, and sustainable betaine-propionic acid NADES system was developed for SG extraction, aligning with green chemistry principles. The optimal extraction conditions were determined as follows: HBA/HBD molar ratio of 1:2, water content of 45%, solid–liquid ratio of 1:10 g/mL, extraction temperature of 70 °C, and extraction time of 120 min. Under these conditions, the extraction yields of RA and STV reached 98.15 ± 2.62 mg/g and 59.83 ± 1.60 mg/g, respectively. Compared with previously reported methods, the proposed NADES system exhibited superior extraction performance. Rouhani et al. employed glycerol combined with UAE and achieved a maximum SG extraction yield of 79% (8.81 mg/g) under optimized conditions, which was substantially lower than the yields obtained in the present study. Similarly, Milani et al. used a DES composed of tetraethylammonium chloride and ethylene glycol (1:2) coupled with ultrasound treatment, obtaining STV and RA contents of 38 ± 0.1 mg/g and 45.1 ± 1.7 mg/g, respectively, after 40 min of extraction. In the study by Suresh et al., the extraction yields of RA and STV using a quaternary NADES were 38.24 ± 2.22 mg/g and 114.58 ± 5.89 mg/g, respectively, indicating higher extraction efficiency for STV. This result differs from most other studies, which may be attributed to the specific composition of only HBDs used in their system. Although the total extracted amount of RA and STV in our study is only slightly higher than that reported by Suresh et al., our method achieved a substantially higher extraction yield for RAthe component with both higher natural abundance and superior sweetness quality. This selective enhancement facilitates the production of higher-quality SGs.

MRA are frequently employed for the purification of SGs from extraction solutions, as well as for the purification of RA from SG products. Liu et al. reported that diatomite effectively removed proteins and flavonoids but caused severe SG loss because of nonselective adsorption. Calcium hydroxide treatment combined with centrifugal filtration significantly improved impurity removal while limiting SG loss to below 15%, with protein removal efficiency reaching 98%. In addition, mixed-bed adsorption or chemically modified adsorbent have shown enhanced selectivity toward RA. − Ye et al. developed a functionalized resin (SD-2) containing amino and phenylboronic acid groups for the separation of RA from STV in aqueous solution. This resin exhibited maximum momolayer adsorption capacity (q _m ) of 101.0 mg/g for RA and 175.4 mg/g for STV at 308 K, representing substantial improvements over the unmodified resin. However, studies on the separation and purification of SGs from NADES extraction solutions remain relatively limited. Suresh et al. separated SGs from the NADES 50 extract using HP-20 microporous resin in column chromatography, with ammonium acetate buffer as the elution solvent and ethanol as the desorption solvent. In contrast, the present study employed AB-8 macroporous resin, using 30% and 70% ethanol as the optimal elution and desorption solvents, respectively. Under these conditions, the yield and purity of RA reached 90.56% and 53.23% respectively, while those of STV were 86.68% and 29.06%. These results demonstrate that the effective separation of SGs from NADES systems achieved by resin adsorption facilitates the industrial application of NADES-based extraction methods. Furthermore, rational chemical modification of adsorption resins could significantly enhance the selective separation of RA and STV, representing a promising avenue for future research.

4. Conclusion

This study established an integrated process for the extraction and purification of SGs from S. rebaudiana leaves by combining NADES with macroporous resin adsorption. Guided by the polarity characteristics of key SGs (RA and STV) and solvent components, COSMO-RS theory was employed to predict the relative solubility of SGs in DES and to elucidate the underlying dissolution mechanisms, enabling a rational solvent screening strategy. The optimal NADES systembetaine:propionic acid (1:2)was identified through experimental validation, consistent with the COSMO-RS predictions, and highlighted the crucial role of hydrogen-bonding interactions between solvent components and polar solute molecules. Subsequent single-factor experiments determined the optimal extraction conditions: 45% water content, solid-to-liquid ratio of 1:10 g/mL, temperature of 70 °C, and extraction time of 120 min. Under these optimized conditions, superior extraction performance was achieved, yielding 98.15 ± 2.62 mg/g RA, 59.83 ± 1.60 mg/g STV, and 7.16 ± 0.45 mg/g RC. For purification, AB-8 resin was most suitable for purifying SGs. Furthermore, hydrophilic NADES was effectively recovered from the SGs extractant through deionized water washing from the resin column, and maintained an extraction efficiency of over 85% even after two cycles of reuse, demonstrating its economic viability and effectiveness. Impurities were eluted with 30% ethanol, and the target SGs were efficiently desorbed using 70% ethanol. This process yielded RA, STV, and RC at recovery rates of 90.56%, 86.68%, and 65.89%, with corresponding purities of 53.23%, 29.06%, and 4.42%, respectively. This work provides a sustainable and efficient strategy for the extraction and purification of SGs, combining theoretical prediction with experimental validation. The approach offers an alternative to conventional methods and supports the advancement of deep-processing technologies for Stevia and other natural plant resources.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.6c01053.

• Composition of NADES systems (Table S1); HPLC elution program (Table S2); physical characteristics of resins (Table S3); optical microscope images of DES samples (Figure S1); SEM photographs of stevia powder extracted by NADES (Figure S2); DSC thermograms of DES formulations with different HBA:HBD molar ratios (Figure S3) (PDF)

⊥.

J.B., and Y.H. contributed equally to this work. Conceptualization, L.K., X.D.; writing–original draft preparation, J.B., Y.H.; Writing–review and editing, M.F.; M.C.; Visualization, Z.L.; Z.O. All authors have read and agreed to the submitted version of the manuscript.

The authors declare no competing financial interest.