Ultrasound-assisted extraction for enhanced recovery and bioactivity of polyphenols from Tuber formosanum

Abstract

Tuber formosanum is a valuable edible and medicinal fungus, yet systematic research on the efficient and green extraction of its polyphenolic compounds remains limited. This study aims to establish an optimized ultrasound-assisted extraction (UAE) process for polyphenols from T. formosanum. UAE was conducted using a KS-700XDS power-adjustable ultrasonic bath at 50 kHz. Initially, UAE was validated as the most efficient method among several conventional techniques, achieving a 14.7% higher yield than heating reflux extraction while consuming 7.5-fold less energy. Subsequently, Lasso regression identified ultrasonic power, time, temperature, and ethanol concentration as the critical variables influencing total polyphenol content (TPC). Through response surface methodology (RSM), the optimal extraction conditions were determined: 95 min, 210 W, 70 ℃, and 60% ethanol. Under these parameters, the TPC reached 2.72 mg/g, representing a 16.24% improvement over pre-optimization levels. To elucidate the contribution patterns of each factor, SHAP analysis was employed, revealing that ultrasonic power was the dominant factor, while time and temperature exhibited unimodal optimal contribution trends, corroborating the RSM findings. Antioxidant assays confirmed that the extract obtained under optimized conditions demonstrated significantly enhanced DPPH and ABTS radical scavenging activity and ferric reducing power compared to unoptimized extracts. This study not only establishes an efficient UAE protocol for T. formosanum polyphenols but also provides a novel methodological framework integrating green extraction technology with interpretable machine learning for the sustainable valorization of bioactive compounds from rare fungal resources.

1. Introduction

Truffles are the edible underground fruiting bodies of certain fungi, renowned for their unique aromatic characteristics and their crucial roles in the culinary world [1], [2]. Among the multitude of truffle species, Tuber formosanum emerges as the most extensively cultivated variety, primarily due to its remarkably high market value, which places it at the forefront of truffle production [3]. Recent pharmacological research has unveiled that T. formosanum possesses a variety of advantageous biological properties. These include immunomodulatory, anti-inflammatory, and antioxidant effects, which not only enhance its appeal in culinary contexts but also suggest considerable potential for therapeutic uses [4], [5]. These findings further elevate the importance of T. formosanum, pointing to its dual role as both a gourmet ingredient and a promising candidate for health-related applications, thus broadening the scope of its significance in both gastronomy and medicine [4]. However, due to its demanding growth conditions and lengthy cultivation cycle, truffle resources are extremely scarce. Therefore, fully extracting the high-value bioactive compounds from truffles to achieve efficient resource utilization holds significant economic importance.

Polyphenolic compounds in fungal resources have garnered significant attention in recent years for their development in functional foods and pharmaceuticals due to their remarkable antioxidant, anti-inflammatory, and metabolic regulatory bioactivities [6], [7]. As a vital source of natural antioxidants, polyphenolic components in edible and medicinal fungi demonstrate immense application potential in preventing chronic diseases and enhancing immune function by scavenging free radicals, inhibiting oxidase activity, and regulating cellular signaling pathways [8], [9]. Nevertheless, despite the well-recognized culinary value of truffles, studies on their bioactive metabolites remain relatively limited. Existing research has predominantly focused on polysaccharides and volatile flavor constituents, whereas the efficient extraction, compositional characterization, and bioactivity evaluation of truffle polyphenols are still insufficiently explored. In addition, the links between extraction parameters, polyphenol yield, and antioxidant performance have not been fully established.

Eco-friendly extraction methods present numerous benefits compared to conventional approaches, such as lower solvent consumption, faster extraction durations, and reduced equipment requirements [10], [11]. A promising approach involves extracting bioactive compounds through alternative methods such as supercritical fluid extraction or pressurized liquid extraction, thereby enabling efficient extraction of active components from truffles [12], [13]. However, these methods are often constrained by long extraction durations, risks of thermal degradation, and the requirement for substantial volumes of organic solvents. To address these limitations, ultrasound-assisted extraction (UAE)—have gained increasing attention. UAE is widely regarded as a green and energy-efficient process capable of shortening extraction time, reducing solvent consumption, and minimizing waste generation [14], [15]. UAE is based on the propagation of ultrasonic waves through a liquid medium, which induces acoustic cavitation, i.e., the formation, growth, and implosive collapse of microbubbles [12]. The collapse of these bubbles generates localized high temperature, pressure, microjets, and strong shear forces, which disrupt cell structures, improve solvent penetration, and facilitate the release of intracellular target compounds [14]. Depending on the extraction system, UAE can be performed using different transducer types, such as ultrasonic baths, probe or horn systems, and plate transducers. Among these, ultrasonic baths are commonly used because of their operational simplicity and lower cost, whereas probe systems generally deliver higher energy intensity and cavitation strength. In addition, ultrasound frequency can significantly influence extraction efficiency: lower frequencies typically produce more intense cavitation and stronger mechanical effects, while higher frequencies may lead to more uniform but less violent cavitation [16], [17].

In this study, an optimized extraction process for polyphenols from T. formosanum was developed. Initially, three conventional extraction techniques were compared to determine the most effective approach. Following this, single‑factor experiments were conducted to evaluate the influence of individual parameters. Lasso regression analysis was then employed to identify the key variables influencing extraction yield, and response surface methodology (RSM) was subsequently applied to optimize the extraction conditions. To further elucidate the contribution of each variable, SHAP analysis was employed to interpret their importance and specific modes of action during the extraction process. Finally, the antioxidant activity of the extract obtained under the optimized conditions was compared with that of extracts prepared before optimization. This study not only establishes an optimized protocol for the efficient extraction of polyphenols from T. formosanum but also confirms the superior antioxidant activity of the extract obtained under these conditions. The integration of machine learning approaches with traditional optimization techniques demonstrates considerable potential for advancing the extraction of bioactive compounds.

2. Materials and methods

2.1. Materials and chemicals

Specimens of T. formosanum (voucher No. HMJAU80587) used in this study were procured from Jilin Agricultural University (China). Molecular identification based on the internal transcribed spacer region confirmed the taxonomic classification. The samples are preserved at the Key Laboratory of Medicinal Fungal Resources and Development Utilization, Jilin Agricultural University. Vitamin C (Tianjin Bailuns Biotechnology Co., Ltd., Tianjin, China), sodium hydroxide solution, sodium carbonate solution, ferrous sulfate, and anhydrous ethanol (Shenyang Tiangang Chemical Reagent Factory, Shenyang, China), and Folin–Ciocalteu reagent (Beyotime Biotechnology, Shanghai, China; No. ST2071) were used in this study. All reagents are analytical grade. Ultrapure water was used throughout the experiments.

2.2. Sample preparation

Fresh T. formosanum samples were cut into 2–3 mm-thick slices and dried in a drying oven at 50 ℃ for 8 h. The residual moisture content of the dried material was 4.2%. Following this drying phase, the samples were ground and subsequently sieved using a 40-mesh sieve to achieve a uniform particle size. Extractions were performed using a KS-700XDS ultrasonic bath (Kunshan Ultrasonic Instruments Co., Ltd., China) equipped with a rectangular stainless-steel chamber (500 mm × 300 mm × 200 mm; total bath volume, 30 L). The instrument operates at a fixed frequency of 50 kHz with a nominal power output of 700 W, adjustable from 10% to 100%. The heating temperature is adjustable from room temperature to 80 ℃ (accuracy ± 1 ℃). Borosilicate glass test tubes (18 mm outer diameter, 20 mL nominal volume) were used to accommodate 0.5 g of dried sample powder and 15 mL of 60% (v/v) aqueous ethanol, yielding a solid-to-liquid ratio of 1:30 (w/v). The tubes were closed with loosely fitted PTFE-lined silicone stoppers to prevent solvent evaporation while allowing pressure equilibration. No calorimetric measurement was performed to determine the effective acoustic power delivered to the solvent. The test tube rack (17  cm × 5.7  cm × 7.5  cm; 2 rows × 6 columns, 12 holes) containing 12 tubes was placed at the geometric center of the ultrasonic bath (internal dimensions: 50  cm × 30  cm). This positioning ensured a minimum distance of approximately 16.5  cm from the longer sidewalls and 12.2  cm from the shorter sidewalls, minimizing edge effects. To further compensate for any residual spatial variation in cavitation intensity across the 12 positions, at the midpoint of extraction, the ultrasonic bath was paused, and the rack was removed, manually rotated 180° horizontally, and then repositioned at the same geometric center. In addition, the positions of the tubes within the rack were randomized across independent experimental replicates. After the extraction was completed, the resulting solutions underwent a filtration step through a 0.22 μm membrane. This filtration process was essential for clarifying the extracts and removing any particulate matter, ensuring that the final extracts were pure and ready for subsequent analysis or application.

2.3. Comparison of several extraction methods

Following the initial solvent screening, ultrasonic-assisted extraction (UAE) was compared with two conventional techniques: stirring-assisted extraction (SAE) and heating reflux extraction (HRE). To ensure methodological consistency, all extractions were performed under identical parameters optimized during preliminary screening: a 90-min duration, 50% ethanol as the solvent, a solid-to-liquid ratio of 1:30 (g/mL), and a temperature of 60 ℃.

UAE was conducted using a KS-700XDS numerical control ultrasonic cleaner (Kunshan Ultrasonic Instruments Co., Ltd., China) at a fixed power of 200 W. SAE was performed on a DF-101S heating magnetic stirrer (Henan Aibot Technology Development Co., Ltd., China) with a constant agitation speed of 100 rpm. HRE was carried out in an HH-6 thermostatic water bath (Nanjing Ronghua Scientific Equipment Co., Ltd., China).

Upon completion of each extraction, the mixtures were centrifuged at 11,000 rpm for 10 min to separate the supernatant. Each extraction condition was tested in triplicate to ensure reproducibility.

2.4. Determination of total polyphenol content

The total phenolic content (TPC) in the samples was evaluated using a modified version of the Folin-Ciocalteu method, a technique that has been detailed in prior research conducted by Huo et al. [18]. In this analysis, a precise aliquot of 50 μL of the extract was combined sequentially with 500 μL of the Folin-Ciocalteu reagent, which had been diluted to one-tenth of its original concentration. Following this, 1000 μL of a 1 M sodium hydroxide solution and 500 μL of a 10% sodium carbonate solution were added to the mixture. The combined solution was subjected to thorough vortexing to ensure complete mixing of the reagents. Afterward, it was allowed to incubate in the dark for a period of 60 min to facilitate the development of color, which is critical for the subsequent absorbance measurement. Once the incubation time had elapsed, the absorbance of the solution was measured at a wavelength of 750 nm, which is optimal for detecting the phenolic compounds present. The TPC was ultimately quantified by employing a calibration curve established with the equation y = 0.0452x + 0.0137, yielding a very high correlation coefficient of R² = 0.9992, indicative of the reliability of the method used for quantification.

2.5. Single-factor experiment

In this study, the effects of six independent parameters on the TPC extracted from T. formosanum were investigated. The selected factors included ultrasonic treatment duration (60, 90, 120, and 150 min), solid-to-liquid ratio (1:20, 1:30, 1:40, and 1:50 w/v), ethanol concentration (40%, 50%, 60%, and 70%), ultrasonic power (150, 200, 250, and 300 W), extraction temperature (50, 60, 70, and 80 ℃), and number of extraction cycles (1, 2, and 3).

2.6. Lasso regression analysis

To identify the key variables significantly influencing polyphenol yield, LASSO regression analysis was performed on the data obtained from single-factor experiments using the HandyBioPlot platform (https://handybioplot.cn). LASSO is a penalized linear regression method that introduces an L1 regularization term to shrink regression coefficients and select important variables. The objective function is expressed as follows:

$$\begin{array}{c}\mathit{min}\\ {\beta }_0,\beta \end{array}\left\{\frac{1}{2n}\sum _{i=1}^n{\left(y_i-{\beta }_0\sum _{j=1}^p{x}_{\mathit{ij}}{\beta }_j\right)}^2+\left(\lambda \sum _{j=1}^p\left|{\beta }_j\right|\right\}\right)$$

where $y_i$ denotes the response variable (polyphenol yield), $x_{\mathit{ij}}$ represents the j-th predictor of the i-th sample, ${\beta }_0$ is the intercept, ${\beta }_j$ are the regression coefficients, n is the number of observations, p is the number of predictors, and $\lambda$ is the regularization parameter controlling the trade-off between model fit and sparsity. By adjusting $\lambda$, variables with negligible contributions are shrunk to zero and excluded from the model, whereas variables with larger effects retain non-zero coefficients. The absolute magnitude of the coefficients was used to assess the relative importance of each factor. The optimal $\lambda$ was selected through cross-validation as implemented in the HandyBioPlot platform. Since LASSO regression is a statistical variable-selection method rather than a physical or spatial field model, no boundary conditions or contour conditions were applied.

2.7. Design of RSM

The experimental data were analyzed using a quadratic model, which is mathematically expressed by equation (1).

$$\begin{array}{c}Y={\beta }_0+{\sum }_{j=1}^k{\beta }_j{X}_j+{\sum }_{j=1}^k{\beta }_{\mathit{jj}}{X}_j^2+{\sum }_i{\sum }_{<j=2}^k{\beta }_{\mathit{ij}}{X}_i{X}_j+e_i\end{array}$$ (1)

The optimization of the results was executed using Design-Expert 13.0 software, a powerful tool developed by Stat-Ease, a company located in Minneapolis, Minnesota, USA. This software is specifically designed to facilitate the analysis and optimization of experimental data. The experiments themselves were meticulously planned and organized following a Box-Behnken Design (BBD), a statistical approach grounded in response surface methodology (RSM). This methodology is particularly effective for exploring the relationships between multiple factors and their effects on response variables. A comprehensive overview of this four-factor experimental matrix is presented in Table 1, which outlines the specific parameters and configurations utilized during the study.

Table 1.

Comparison of extraction methods.

Extraction Method	Polyphenol content (mg/g)	Energy consumption (J)
UAE	2.34 ± 0.21	1.08 × 106
SAE	2.04 ± 0.02**	5.40 × 106
HRE	2.21 ± 0.02*	8.10 × 106

Each experiment was performed in triplicate (n = 3), and results are expressed as mean ± standard deviation (SD). *p < 0.05 and **p < 0.01 versus UAE group.

2.8. SHAP analysis

To elucidate the complex nonlinear relationships between process parameters and polyphenol yield, while enabling model interpretability, an Adaptive Boosting (AdaBoost) algorithm was employed to construct a predictive model. As an ensemble learning method, AdaBoost iteratively adjusts sample weights to effectively capture intricate patterns within mixed datasets, exhibiting robust performance even with limited sample sizes. Model interpretation was conducted using SHAP (SHapley Additive exPlanations) analysis via the 'kernelshap' package in R (version 4.4.3). Visualization tools, including SHAP dependence plots and force plots, were utilized to intuitively illustrate the contribution and influence patterns of each feature on TPC extraction.

2.9. Anti-oxidant activity

The antioxidant capacity of extracts obtained from T. formosanum was thoroughly evaluated using a variety of assays, including DPPH, ABTS, FRAP, and hydroxyl radical scavenging tests. These evaluations adhered to a modified methodology as specified by Huo et al. [19], ensuring a rigorous and standardized approach to the assessment of antioxidant properties. In the case of the DPPH assay, a specific volume of 100 μL of the plant extract was mixed with 100 μL of a DPPH solution, which was prepared at a concentration of 50 μmol/L. This combination was allowed to incubate in darkness for a period of 30 min, enabling sufficient reaction time for the antioxidant components within the extract to interact with the DPPH radicals. Following this incubation, the absorbance of the solution was measured at a wavelength of 517 nm, providing crucial data to determine the antioxidant capacity of the extracts under investigation. The scavenging activity of ABTS was assessed utilizing a commercial kit specifically designed for this purpose. In the procedure, a volume of 10 μL of the sample was mixed with 170 μL of the ABTS working solution, along with 20 μL of the application solution. This mixture was allowed to react for a duration of 6 min at room temperature. Subsequently, the absorbance of the solution was measured at a wavelength of 405 nm to evaluate the scavenging activity. The FRAP assay was performed following the specific guidelines outlined in the kit instructions. In this procedure, 5 µl of the extract were combined with 180 µl of the FRAP working solution. The mixture was then incubated at a temperature of 37 degrees Celsius for a duration of 5 min. After the incubation period, the absorbance was measured at a wavelength of 595 nm, with the results being evaluated against a standard curve created using ferrous sulfate (FeSO₄). The hydroxyl radical scavenging activity was assessed through a systematic addition of several components to the reaction mixture. Initially, 5 μL of the extract was introduced, followed by the incremental addition of 0.05 mL of a 6 mmol/L solution of FeSO₄. This was succeeded by the inclusion of 0.05 mL of 6 mmol/L salicylic acid, which was previously dissolved in ethanol, and finally, 0.05 mL of a 6 mmol/L solution of hydrogen peroxide (H₂O₂) was added. The complete mixture was then incubated at a temperature of 37 ℃ for a duration of 30 min. After this incubation period, the absorbance of the solution was measured at a wavelength of 510 nm to evaluate the scavenging activity of the hydroxyl radicals.

2.10. Statistical analysis

The extraction process was conducted in triplicate to ensure accuracy and reliability of the results. The outcomes of these extractions were reported as the mean value accompanied by the standard deviation, which provides a clearer understanding of the variability within the data. Furthermore, for the purpose of statistical analysis and graphical representation of the findings, the Origin 2021 software developed by OriginLab Corporation, located in Northampton, MA, USA, was utilized. This software facilitated the analysis and visual interpretation of the data, enhancing the overall presentation of the results.

3. Results and discussion

3.1. Selecting the optimal extraction methods

As presented in Table 2, UAE yielded a significantly higher polyphenol content (2.34 ± 0.21 mg/g) compared to both HRE (2.21 ± 0.02 mg/g) and SAE (2.04 ± 0.02 mg/g). This represents a substantial increase of approximately 5.9% and 14.7% over HRE and SAE, respectively. This superior performance is unequivocally attributable to the unique cavitation phenomena generated by ultrasonication. The implosive collapse of cavitation bubbles produces localized extreme pressures and shear forces, which are highly effective in disrupting resilient cell walls and facilitating the efficient solubilization of intracellular polyphenols [20]. Furthermore, UAE exhibited a markedly lower energy footprint, consuming only 1.08 × 10⁶ J. This value is significantly less than the energy required for HRE (8.10 × 10⁶ J) and SAE (5.40 × 10⁶ J), corresponding to a 7.5-fold and 5-fold reduction, respectively. In contrast, the efficiency of HRE is constrained by its primary reliance on thermal energy and passive solvent diffusion. This approach is not only less effective at breaching rigid cellular architectures but also subjects thermolabile polyphenolic compounds to potential degradation, thereby diminishing the final yield [21]. The intrinsically high energy demand of sustained heating further underscores its inefficiency. Similarly, SAE, while employing mechanical agitation, lacks the critical cell-disruption mechanism of cavitation. The absence of ultrasonic shock waves limits the extent of cell wall fragmentation, resulting in an incomplete release of polyphenols from the lignocellulosic matrix [22]. The extended agitation times necessary to partially compensate for this inefficiency lead to the substantially higher energy consumption observed. Consequently, both HRE and SAE are demonstrably less efficient than UAE, which achieves a superior extraction yield at a fraction of the energy cost. These findings underscore the dual advantage of UAE as a highly effective and energy-sustainable technology for the recovery of bioactive polyphenols from biomass.

Table 2.

Response surface design test factor level and coding.

Factors	Levels
	−1	0	1
X1: Ultrasonic time (min)	60	90	120
X2: Ultrasonic temperature (℃)	60	70	80
X3: Ethanol concentration (%)	50	60	70
X4: Ultrasonic power (W)	150	200	250

3.2. Single-factor experiment

3.2.1. Effect of ultrasonic time

As illustrated in Fig. 1A, prolonged ultrasonic extraction enhanced cavitation effects, leading to more extensive disruption of fungal cell walls. This increased cellular fragmentation significantly improved extraction efficiency [23]. Within an optimal range, extending the treatment duration promoted the release and recovery of TPC, thereby increasing yield. However, beyond 90 min, extraction efficiency declined. Excessive ultrasonic duration elevates the solution temperature, which may induce oxidative polymerization of polyphenolic compounds, compromising their structural stability and reducing the detectable active components [24]. Accordingly, an extraction time of 90 min was selected as the central point for subsequent optimization experiments.

Fig. 1.

Effects of various extraction parameters on polyphenols yield. The influence of (A) ultrasonic time, (B) ratio of solid to liquid, (C) ethanol concentration, (D) ultrasonic power, (E) ultrasonic temperature, (F) and extraction times on polyphenol yield. n = 3. All data are expressed as mean ± standard deviation (SD).

3.2.2. Effect of solid to liquid ratio

The solid-to-liquid ratio is a critical determinant of extraction efficiency, as it governs the mass transfer driving forces [25]. As depicted in Fig. 1B, polyphenol yield increased markedly as the ratio was raised from 1:20 to 1:30 g/mL, but declined upon further increase. This trend reflects the opposing effects of solvent volume on mass transfer kinetics. A larger solvent volume enhances mass transfer by increasing the contact surface area, elevating osmotic pressure, reducing viscosity, and intensifying ultrasonic cavitation. However, excessive solvent can attenuate ultrasonic wave propagation, thereby diminishing extraction efficiency [26]. Considering these competing mechanisms, a solid-to-liquid ratio of 1:30 g/mL was identified as optimal for maximizing polyphenol yield in this study.

3.2.3. Effect of ethanol concentration

Ethanol, as a polar solvent, facilitates the dissolution and extraction of polyphenols from matrices [27]. The ethanol concentration plays a pivotal role in determining extraction efficiency. As illustrated in Fig. 1C, polyphenol yield increased markedly as the ethanol volume fraction was raised from 40% to 60%, beyond which a plateau or decline was observed within the 60–70% range. Excessive ethanol concentrations may reduce solvent polarity, thereby diminishing the solubility of polar polyphenolic compounds [28]. Based on these findings, 60% ethanol was selected as the optimal concentration for subsequent extraction experiments.

3.2.4. Effect of ultrasonic power

As ultrasonic power increased, TPC yield initially rose, reaching a plateau beyond 200 W (Fig. 1D). Enhanced cavitation at higher power levels intensifies cellular disruption, thereby improving extraction efficiency [29]. Within an optimal range, increased ultrasonic output effectively promotes cell wall rupture and facilitates polyphenol release. However, excessive power can impede shock wave propagation due to the formation of excessive cavitation bubbles [30]. These bubbles may coalesce into larger clusters, diminishing the intensity of micro-jet formation and collapse [31]. Additionally, overly intense cavitation can induce chemical bond cleavage, molecular degradation, and free radical generation in polyphenolic compounds. As shown in Fig. 1D, no statistically significant difference in TPC yield was observed between 200 W and 300 W (p > 0.05). Therefore, 200 W was selected as the optimal ultrasonic power to balance extraction efficiency with energy consumption.

3.2.5. Effect of ultrasonic temperature

Ultrasonic temperature critically influences extraction efficiency by modulating mass transfer driving forces [32]. In this study, the extraction temperature was actively maintained using a thermostatic water bath (accuracy ± 1 ℃) throughout the ultrasonic process, thereby excluding uncontrolled temperature increases caused by ultrasonic energy dissipation. As illustrated in Fig. 1E, polyphenol yield increased markedly as the temperature rose from 50 to 70 ℃, beyond which a significant decline was observed. Temperature plays a dual role in the extraction of bioactive compounds from mushrooms: it facilitates extraction by softening cellular structures [33], yet excessive heat may compromise the stability of phenolic compounds. Within an optimal range, elevated temperature enhances solvent penetration by reducing viscosity and surface tension [34], thereby improving the diffusion of phenolic compounds. Furthermore, ultrasonic cavitation generates high shear forces that disrupt cell walls, facilitating the release of intracellular contents into the solvent [35]. The synergistic effects of thermal softening and cavitation-induced disruption contribute to the enhanced extraction efficiency observed at 70 ℃. Based on these findings, 70 ℃ was selected as the optimal temperature for subsequent experiments.

3.2.6. Effect of extraction times

The influence of the number of extraction times (1, 2, and 3 times) on the TPC was investigated to determine the most efficient process. As illustrated in Fig. 1F, increasing the number of extraction times from one to three resulted in a marginal increase in the cumulative TPC yield. This incremental rise can be attributed to the further dissolution of polyphenols from the matrix into the fresh solvent during subsequent times. However, a detailed statistical analysis of the data revealed that there was no statistically significant difference in the incremental TPC obtained from the first, second, and third extraction times (p > 0.05). The lack of significant difference between successive times suggests that the primary extraction event is highly efficient, depleting the bulk of the soluble phenolic compounds in the initial stage [36]. The additional polyphenols released in the second and third times were minimal and did not contribute to a statistically meaningful enhancement of the total yield. Consequently, while employing multiple extraction cycles can theoretically lead to a slightly higher cumulative yield, the marginal benefit does not justify the substantial increase in time, solvent consumption, and energy required. A single extraction cycle is therefore determined to be the most practical and efficient approach. Following a comprehensive evaluation of yield against resource expenditure, one extraction cycle is identified as the ideal condition for optimizing overall process efficiency.

3.3. Lasso regression analysis

The Lasso regression analysis conducted in this study yielded significant insights into the various factors influencing polyphenol extraction efficiency. The predictive performance of the model was notably strong, achieving an area under the curve (AUC) of 0.875 during cross-validation, as illustrated in Fig. 2A. This high AUC signifies that the model is effective in predicting outcomes based on the input variables. To establish the optimal value of the regularization parameter λ, cross-validation was employed, as shown in Fig. 2B. The x-axis of this graph indicates the logarithmic scale of λ. An increase in λ enhances the regularization effect, which results in greater discrepancies between the actual values and the predicted values. Consequently, this leads to lengthier error bars and a decline in predictive accuracy. The application of Lasso regression serves to control model complexity through the mechanism of regularization, which is instrumental in preventing overfitting of the data. Accordingly, the optimal λ identified through this process was utilized to set the factor levels for the subsequent response surface methodology experiment. The careful tuning of parameters was effective in stabilizing the model’s error rate, thereby reinforcing the reliability of the findings [37]. Further analysis revealed that among all the factors investigated, the coefficients associated with ultrasonic power, ultrasonic time, ultrasonic temperature, and ethanol concentration demonstrated a slower rate of shrinkage towards zero, as depicted in Fig. 2C. This observation highlights that these four parameters are crucial for efficient extraction of polyphenols. Conversely, the coefficients corresponding to the solid to liquid ratio and extraction times diminished entirely to zero, indicating their minimal contribution to the extraction results. This finding substantiates their exclusion from further experimental considerations. To further clarify the situation, the significance of the retained factors was ranked according to their resistance to shrinkage along the regularization path (Fig. 2D). The ordering of importance for these factors is thus established, with ultrasonic power being the most significant, followed by ultrasonic time, ultrasonic temperature, and finally, ethanol concentration.

Fig. 2.

Screening of variables based on Lasso regression. (A) ROC curve diagram, (B) Selection of the optimal value of the parameter λ, (C) The variation characteristics of the coefficient of variables, (D) Screening of factor importance.

3.4. Response surface experimental results

3.4.1. Model building

The ANOVA results for TPC are summarized in Table 4. A low p-value coupled with a high F-value indicates a statistically significant effect of the corresponding variable. The TPC model was highly significant (p < 0.0001), confirming that the selected factors substantially influence extraction yield. The lack-of-fit was non-significant (p = 0.5702 > 0.05), indicating that the model adequately fits the experimental data. Among the model terms, the linear coefficients (X₁, X₄), interaction terms (X₁X₂, X₁X₃, X₁X₄), and quadratic coefficients (X₁², X₂², X₃², X₄²) exhibited extreme significance (p < 0.001). The interaction term (X₃X₄) was significant at p < 0.05, meriting further consideration. Other terms showed no significant influence (p > 0.05). The model demonstrated excellent fit, with an R² of 0.9959 and a coefficient of variation (C.V.) of 0.006552, indicating high precision and low variability in TPC predictions. Accordingly, the regression model for TPC was established in terms of coded factor levels.

$$Y=2.70+0.0642X_1+0.0025X_2-0.0008X_3+0.0842X_4+0.0400X_1{X}_2+0.0425X_1{X}_3+0.0350X_1{X}_4+0.0125X_2{X}_3+0.0050X_2{X}_4+0.0175X_3{X}_4-0.1883X_1^2-0.2283X_2^2-0.1958X_3^2-0.1783X_4^2$$ (2)

Table 4.

ANVOA analysis for RSM model.

Source	Sum of squares	df	Square	F value	P-Value	Significance
Model	0.8132	14	0.0581	240.35	< 0.0001	***
X1	0.0494	1	0.0494	204.45	< 0.0001	***
X2	0.0001	1	0.0001	0.3103	0.5863	ns
X3	8.333E-06	1	8.333E-06	0.0345	0.8553	ns
X4	0.0850	1	0.0850	351.76	< 0.0001	***
X1X2	0.0064	1	0.0064	26.48	0.0001	***
X1X3	0.0072	1	0.0072	29.90	< 0.0001	***
X1X4	0.0049	1	0.0049	20.28	0.0005	***
X2X3	0.0006	1	0.0006	2.59	0.1301	ns
X2X4	0.0001	1	0.0001	0.4138	0.5304	ns
X3X4	0.0012	1	0.0012	5.07	0.0409	*
X12	0.2301	1	0.2301	952.02	< 0.0001	***
X22	0.3382	1	0.3382	1399.37	< 0.0001	***
X32	0.2488	1	0.2488	1029.36	< 0.0001	***
X42	0.2063	1	0.2063	853.61	< 0.0001	***
Residual	0.0034	14	0.0002
Lack of Fit	0.0024	10	0.0002	0.9533	0.5702	ns
Pure error	0.0010	4	0.0002
Sum	0.8166	28

The adjusted R-squared value R²_adj was in close agreement with the coefficient of determination (R²) for the TPC regression model, indicating that the model adequately captures the variability in the experimental data. The non-significant lack-of-fit further confirmed the model's reliability in predicting TPC across various combinations of input parameters. These statistical metrics collectively underscore the robustness and predictive accuracy of the model, supporting its applicability for practical optimization purposes.

3.4.2. Effect of extraction parameters on TPC

The superior extraction efficiency of UAE compared to conventional methods arises from the synergistic effects of acoustic cavitation, mechanical cell disruption, and enhanced mass transfer. Based on the regression model, three-dimensional response surface plots were generated to visualize the interactions between variables (Fig. 3). A significant interaction was observed between ultrasonic time (X₁) and ultrasonic temperature (X₂), (p = 0.0001, Table 3), with TPC initially increasing then declining as both factors rose (Fig. 3A). This trend aligns with findings from Jiang et al. on anthocyanin extraction from blueberries [38]. Similarly, the interaction between ultrasonic time (X₁) and ethanol concentration (X₃) was highly significant (p < 0.0001), where TPC peaked before decreasing with further increases in either parameter (Fig. 3B). Comparable results were reported for flavonoid extraction from Dendrobium chrysotoxum using ultrasonic-assisted methods [39]. The interaction between ultrasonic time (X₁) and ultrasonic power (X₄) followed a similar pattern, with TPC initially increasing then declining beyond optimal levels (Fig. 3C), consistent with alkaloid extraction studies on safflower seeds [40], where cavitation intensity is governed by the product of time and applied electrical power. A moderate power–time product maximizes mechanical damage to the matrix, sharply elevating the extractable phenolic content [41], [42]. In contrast, excessive power combined with prolonged exposure intensifies both the frequency and violence of cavitational collapse, leading to overgeneration of free radicals and localized hotspots [43], [44]. This accelerates phenolic oxidation and polymerization, outweighing any further gain in mass transfer, consistent with alkaloid extraction studies on safflower seeds [45]. For ethanol concentration (X₃) and ultrasonic power (X₄), TPC exhibited analogous behavior, reaching a maximum before declining with further increases (Fig. 3F), corroborating findings on triterpenoid extraction from Ganoderma lucidum [46].

Fig. 3.

Three-dimensional response surface diagrams of the influence of different factors interaction on TPC. (A) Interaction between ultrasonic temperature and ultrasonic time, (B) Interaction between ethanol concentration and ultrasonic time, (C) Interaction between ultrasonic power and ultrasonic time, (D) Interaction between ethanol concentration and ultrasonic temperature, (E) Interaction between ultrasonic power and ultrasonic temperature, (F) Interaction between ultrasonic power and ethanol concentration.

Table 3.

Design and results of RSM.

Run Order	Factors				Polyphenols content (mg/g)
	X1 (min)	X2 (℃)	X3 (%)	X4 (W)
1	0	0	1	1	2.42
2	−1	0	1	0	2.21
3	0	0	−1	−1	2.27
4	−1	0	−1	0	2.28
5	0	0	−1	1	2.41
6	0	1	−1	0	2.27
7	0	1	0	−1	2.19
8	0	0	0	0	2.69
9	0	−1	−1	0	2.29
10	1	0	1	0	2.42
11	1	0	0	−1	2.29
12	0	−1	1	0	2.27
13	1	1	0	0	2.40
14	0	1	1	0	2.30
15	1	0	−1	0	2.32
16	−1	1	0	0	2.19
17	1	−1	0	0	2.30
18	0	0	0	0	2.70
19	0	0	0	0	2.71
20	−1	0	0	−1	2.23
21	−1	−1	0	0	2.25
22	0	−1	0	−1	2.21
23	−1	0	0	1	2.32
24	0	0	1	−1	2.21
25	0	−1	0	1	2.37
26	0	0	0	0	2.68
27	0	0	0	0	2.72
28	1	0	0	1	2.52
29	0	1	0	1	2.37

The RSM model predicted an optimal TPC yield of 2.7173 mg/g under the following conditions: ultrasonic time 96.03 min, ethanol concentration 60.31%, ultrasonic power 212.88 W, and temperature 70.27 ℃. To validate the model, these parameters were adjusted to practical values: 95 min, 60% ethanol, 210 W, and 70 ℃. Triplicate experiments under these conditions yielded 2.72 ± 0.14 mg/g, closely matching the predicted value and confirming the model's reliability. Compared to pre-optimization conditions, TPC yield increased from 2.34 mg/g to 2.72 mg/g, representing a 16.24% enhancement in extraction efficiency.

3.5. SHAP analysis

The developed model exhibited satisfactory predictive performance for polyphenol yield. For the training set, the coefficient of determination (R²) was 0.434 with RMSE of 0.102 and MAPE of 3.58% (Table 5). Notably, the test set yielded improved metrics with R² = 0.595, RMSE = 0.096, and MAPE = 3.17%, indicating good generalization capability without overfitting. The RPD value of 1.58 for the test set suggests that the model is suitable for semi-quantitative prediction of polyphenol content under the studied conditions. The relatively modest R² values can be attributed to the inherently narrow range of polyphenol yields and unavoidable experimental variations, as evidenced by replicate measurements. Nevertheless, the low prediction errors (MAE < 0.08) confirm the practical utility of the model for optimizing extraction parameters.

Table 5.

Performance evaluation of the predictive mode.

Indicator	Training Set	Test set
R2	0.434	0.595
RMSE	0.102	0.096
RMSE	0.084	0.076
MAPE	3.58%	3.17%
RPD	1.33	1.58

SHAP analysis was employed to elucidate the contribution patterns and interaction mechanisms of the four extraction parameters (ultrasonic power, ultrasonic time, ultrasonic temperature, and ethanol concentration) identified by Lasso regression, providing mechanistic insights beyond conventional response surface methodology. The mean absolute SHAP values ranked ultrasonic power as the most influential factor (0.21), followed by ultrasonic time (0.18), ultrasonic temperature (0.11), and ethanol concentration (0.09), underscoring the dominant role of cavitation intensity in polyphenol release (Fig. 4A). The SHAP swarm plot (Fig. 4B) displays the distribution of SHAP values across all samples, with the color gradient from blue to red representing feature values from low to high. For ultrasonic power, high values (>0.6, reddish hues) predominantly correspond to positive SHAP values, indicating that increasing power within the experimental range (150–300 W) significantly promotes polyphenol release, attributable to enhanced cavitation-induced cell wall disruption. Ultrasonic time exhibits a more complex pattern: moderate values (60–90 min) contribute positively, while excessively high values (>120 min) yield negative SHAP values, suggesting that prolonged extraction may lead to polyphenol degradation or oxidation. Ultrasonic temperature displays a unimodal trend, with positive contributions most concentrated in the 60–70 ℃ range, confirming the existence of an optimal extraction temperature for heat-sensitive polyphenols. Ethanol concentration, despite its relatively lower overall importance, shows the most pronounced positive contributions in the 50–60% range, reflecting the critical role of solvent polarity in polyphenol solubility. The dependence plots (Fig. 4C) revealed distinct contribution trends: ultrasonic power exhibited a monotonically positive effect across the tested range (150–300 W), confirming that enhanced cavitation promotes cell disruption without reaching a detrimental threshold; ultrasonic time displayed a unimodal pattern with peak contributions at 60–90 min, beyond which prolonged extraction led to declining SHAP values due to possible polyphenol degradation; ultrasonic temperature similarly showed an optimum at 60–70 ℃, reflecting the trade-off between thermal solubilization and heat-induced degradation; ethanol concentration also followed a single‑peak trend centered at 50–60%, indicating the necessity of matching solvent polarity to polyphenol solubility. Two‑dimensional dependence plots further revealed synergistic interactions: at moderate‑to‑high power levels (>180 W), extending extraction time substantially increased SHAP values, whereas at low power this synergy was negligible; likewise, elevated temperatures required moderate power to avoid thermal degradation, while lower temperatures demanded higher power to compensate for reduced molecular motion. Notably, the SHAP analysis independently corroborated the RSM‑optimized conditions. For a representative sample under these optimal parameters, the cumulative SHAP contributions from ultrasonic power (+0.32), time (+0.28), temperature (+0.15), and ethanol concentration (+0.08) summed to + 0.83 above the baseline E[f(x)] = 2.38 mg/g, producing a predicted value of 2.71 mg/g that closely matched the experimental yield (2.72 mg g). This excellent agreement validates the reliability of both analytical approaches and demonstrates that the identified contribution patterns accurately reflect the underlying extraction mechanisms. Collectively, the SHAP analysis reveals that polyphenol extraction from T. formosanum is primarily governed by cavitation intensity and its temporal coordination, with secondary but critical roles played by thermal energy and solvent polarity, and confirms that the RSM‑optimized conditions represent a synergistic combination where each factor operates within its most favorable range.

Fig. 4.

SHAP Analysis Plots of Four Important Single Factors. (A) SHAP bar plot, (B) SHAP dependence plot, (C) Feature Dependency Graph for Each Variable.

3.6. Assays of antioxidants

DPPH is a stable paramagnetic radical that is reduced to its diamagnetic form (N–H) upon electron donation by antioxidants. In ethanolic solution, this reduction is accompanied by a visible color change from deep purple to pale yellow, which can be quantified spectrophotometrically to evaluate electron transfer efficiency [47]. The DPPH scavenging activity of T. formosanum extracts exhibited a clear dose-dependent relationship. Over a concentration range of 0.2–1.4 mg/mL, a linear correlation was observed between extract concentration and radical scavenging activity, indicating stoichiometric interactions between the extract components and free radicals (Fig. 5A). The ABTS assay, analogous to the DPPH method, is widely employed to assess antioxidant capacity [48]. Within the same concentration range (0.2–1.4 mg/mL), a positive correlation was observed between extract concentration and ABTS radical scavenging activity, with the highest tested concentration (1.2 mg/mL) demonstrating substantial antioxidative potential (Fig. 5B). This observation is consistent with previous reports on Flammulina velutipes extracts, which have also shown dose-dependent radical scavenging activity and have been linked to the presence of phenolic and other reductive constituents [49]. The FRAP assay measures reducing capacity based on the reduction of ferric to ferrous ions by electron-donating antioxidants, resulting in the formation of a blue chromophore [50]. The reducing power of T. formosanum extracts increased progressively with concentration, exhibiting the strongest association within the 0.2–1.4 mg/mL range. At 1.2 mg/mL, the extracts demonstrated comprehensive reducing capability, highlighting their potential as effective antioxidants (Fig. 5C). Such a response is in agreement with earlier studies on Malpighia glabra fruits, where higher extract concentrations generally corresponded to stronger ferric reducing activity [51]. Hydroxyl radicals, generated as byproducts of metabolic processes, are among the most reactive and deleterious free radicals [52]. The T. formosanum extracts exhibited a significant positive correlation between concentration and hydroxyl radical scavenging activity across the 0.2–1.4 mg/mL range. At the maximum tested concentration (1.2 mg/mL), the extracts demonstrated potent scavenging capacity against stable hydroxyl radicals (Fig. 5D). Collectively, these findings underscore the efficacy of T. formosanum extracts as natural antioxidants capable of mitigating oxidative stress.

Fig. 5.

Antioxidant capacity analysis of ultrasound-assisted extraction (UAE), and optimized ultrasound-assisted extraction (OUAE). (A) DPPH radical scavenging ability, (B) ABTS⁺ radical scavenging ability, (C) FRAP assay, (D) Hydroxyl radical scavenging ability. n = 3. All data are expressed as mean ± S.D.

4. Conclusion

In this study, the ultrasound-assisted extraction of polyphenols from T. formosanum was systematically optimized by integrating conventional single-factor experiments with Lasso regression analysis, RSM, and SHAP analysis. A comparative evaluation of three conventional extraction techniques identified the most efficient method for subsequent optimization. Through Lasso regression analysis, the key variables significantly influencing polyphenol yield were identified, and RSM was employed to establish the optimal extraction conditions. The optimal parameters were determined as follows: ultrasonic time of 95 min, concentrations of ethanol 60%, power of ultrasonic waves 210 W, and ultrasonic temperature 70 ℃, yielding a polyphenol content of 2.72 mg/g under these conditions. Furthermore, SHAP analysis was utilized to interpret the contribution and impact patterns of individual factors during the extraction process, providing mechanistic insights beyond traditional optimization approaches. Comparative assessment of antioxidant activity demonstrated that the extract obtained under the optimized conditions exhibited significantly enhanced activity relative to extracts prepared before optimization, validating the effectiveness of the established protocol. This integrated approach holds promise for advancing the sustainable utilization of food-derived bioactive ingredients in nutraceutical and functional food applications. Nevertheless, this study has some limitations. First, the chemical characterization of individual polyphenolic compounds was not fully performed. Second, the scalability and industrial applicability of the optimized UAE process were not evaluated in this study.

CRediT authorship contribution statement

Huimin Huo: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Conceptualization. Haiying Bao: Writing – review & editing, Writing – original draft, Supervision, Project administration, Methodology, Funding acquisition, Conceptualization. Tianrui Liu: Software, Resources, Funding acquisition. Sitegele Wu: Visualization, Validation, Software. Baihui Yang: Software, Investigation. Ruicong Gao: Visualization, Software. Mingjie Song: Software. Hong Kan: Software.

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.