Optimization of ultrasound-assisted extraction of polysaccharides from Akebia Fruit using an artificial neural network model: Characteristics and antioxidant activity

Graphical abstract

Abstract

This study investigated the extraction, structural characterization, and antioxidant activity of polysaccharides derived from Akebia Fruit. The ultrasonic-assisted extraction (UAE) process of polysaccharides was optimized through the application of the Box-Behnken Design (BBD) in conjunction with the genetic algorithm-back propagation (GA-BP) artificial neural network model. The experimental data showed that the GA-BP model performed better than the BBD model, and more polysaccharide components could be extracted under the process parameters predicted by this model. The GA-BP model predicted the optimal extraction parameters as follows: the extraction temperature was 65 ℃, the solid-liquid ratio was 1:50 g/mL, the extraction power was 400 W. Experimental results showed that combining UAE with GA-BP artificial neural network not only enabled efficient extraction of polysaccharides but also optimized the extraction process. After purification, AFP-1 was obtained and its characterization was conducted. Structural analysis results indicated that compound AFP-1 was a homogeneous polysaccharide with a lamellar structure and a molecular weight of 13,775 Da. The polysaccharide contained a network of pyranose rings, which were interconnected to form a complex framework. The polysaccharide was composed of a mixture of monosaccharide units, specifically arranged in a specific configuration that included mannose, ribose, glucose, galactose, and fucose. Finally, the antioxidant activity of AFP-1 was preliminarily verified through in vitro experiments. Subsequent research could systematically explore the biological activities of AFP-1, by employing both in vitro and in vivo models.

1. Introduction

The Chinese medicine AKEBIAE FRUCTUS (Akebia Fruit) is the dried, nearly mature fruit of Akebia quinata (Thunb.) Decne., Akebia trifoliata (Thunb.) Koidz., or Akebia trifoliata (Thunb.) Koidz. var. australis (Diels) Rehd. Incorporated into the People's Republic of China's pharmacopoeia, it finds wide application in diverse regions, notably the Wudang region of China. It plays a crucial role in the treatment of cardiovascular diseases when used in conjunction with other traditional Chinese medicines. Despite Akebia Fruit 's long-standing use in traditional Chinese medicine for its purported health-promoting properties, including liver-qi regulation and pain-relieving effects, polysaccharides derived from Akebia Fruit (also known as Yuzhizi in China) remain largely overlooked and underexplored, even as polysaccharides have emerged as a significant area of research in natural products.

Natural polysaccharides popularity is due to the diversity of active principles that are naturally heterogeneous, such as tumor therapy, immune regulation, antioxidant activity, and anti-inflammatory properties [1]. Among carbohydrate polymers, a majority exhibit remarkable attributes including biodegradability, modifiability without immunogenicity, and low toxicity. They can undergo conjugation, cross-linking, or functional modification and subsequently be utilized as nanocarrier materials [2]. Polysaccharide-based drug delivery systems can evade phagocytosis by the reticuloendothelial system, protect biomolecules from degradation, and enhance the bioavailability of small molecules, thereby achieving effective therapeutic outcomes [3]. Plant polysaccharides are generally composed of multiple monosaccharides, including glucose, galactose, and arabinose, and are characterized by complex and diverse structures with relatively high molecular weights. Contemporary studies have confirmed that these polysaccharides possess notable pharmacological activities. However, their low content in plants leads to inefficient direct extraction [4]. Thus, the development of simple and efficient extraction methods is essential for the advancement of polysaccharide research.

The extraction process is a key component of polysaccharide research, and selecting appropriate extraction methods is essential for ensuring efficient polysaccharide recovery [5,6]. Common extraction methods for polysaccharides in traditional Chinese medicine include hot water extraction, enzymatic hydrolysis extraction, ultrasonic-assisted extraction, microwave-assisted extraction, and other methods [7]. Although hot water extraction is a widely used extraction technique, it has several limitations, including the requirement for high extraction temperatures, significant solvent usage, lengthy extraction durations, and low extraction efficiency [8]. Due to the high cost of enzymes and the harsh usage conditions, the enzymatic hydrolysis extraction method is restricted from large-scale application in the extraction of polysaccharides from traditional Chinese medicine. The local high temperature caused by microwaves leads to changes in the structure of polysaccharides, which restricts the application of microwaves in the field of polysaccharide extraction [9]. Among these methods, UAE has attracted significant attention due to its low energy consumption, short processing time, high efficiency, and advantages such as mild heating temperatures and the preservation of active ingredients during extraction [10]. The mechanism of UAE for polysaccharides relies on the mechanical and thermal effects generated by ultrasound. Mechanically, ultrasound enhances solvent penetration into the herb surface, thereby accelerating the dissolution of effective components. Thermally, process generates heat, which increases the solubility of these components. Under ultrasonic influence, particles experience intense vibrations, resulting in the disruption of cell walls and membranes, ultimately releasing polysaccharide molecules. Additionally, ultrasound triggers molecular movement and vibration in the liquid, which enhances the mass transfer rate between the solvent and the sample. This, in turn, accelerates the dissolution and diffusion process of polysaccharides [[11], [12], [13]]. The UAE technique not just improves extraction efficiency but minimizes structural damage to some extent. Therefore, in this experiment, the UAE method was adopted to extract polysaccharides, with emphasis on the selection and optimization of its parameters.

The application of BBD in optimizing extraction parameters is widely recognized, as it not only evaluates the interactive effects of multiple variables but also delivers more intuitive and precise results [14,15]. While BBD is one of the most commonly used statistical approaches, it has limitations in exploring non-linear, highly interactive, or complex relationships, as it may fail to capture all influencing factors and change patterns. In practice, certain input variables may exhibit correlations, rendering optimization models assuming independence potentially inaccurate. Additionally, when the dataset for modeling is limited, there is a risk of model overfitting, where the model becomes excessively complex and yields poor predictive performance on new datasets [16]. However, AI-based optimization algorithms do not have this limitation, as they benefit from their strong nonlinear fitting capability [17]. Artificial neural networks possess the ability to learn and adapt independently, enabling them to simulate the information-processing mechanism of the human brain. The neural network model training is terminated when the relative error between the experimental value and the predicted value reaches the set minimum error [18]. It measures weights and thresholds based on information that is constantly propagated. Therefore, the results of the model are closer to the actual results. A genetic algorithm is that simulates evolutionary processes and is passed down from generation to generation. It continues to retain well-adapted individuals, but excludes poorly adapted individuals. Finally, we need to set the right values and thresholds to activate a neural network. In this way, we can avoid potential problems with local optimal solutions [19]. Numerous studies have demonstrated that the GA-BP model excels in determining optimal experimental conditions across a broad parameter space, outperforming conventional optimization methods. By leveraging independent learning, it reduces the number of required experiments and minimizes the time and effort spent on research [20]. Therefore, it is particularly well-suited for multifactorial experiments in the extraction of traditional Chinese medicine. In this experiment, the UAE process of polysaccharides was optimized by combining the BBD with a GA-BP artificial neural network model.

The structural characteristics of polysaccharides, such as molecular weight, influence their activities [21]. On the premise of maintaining a certain active spatial structure, low-molecular-weight polysaccharides are more favorable for transmembrane absorption or affinity with immune cell carbohydrate receptors, potentially resulting in higher bioactivity. This is attributed to the fact that low-molecular-weight polysaccharides possess reducible groups (such as hydroxyl and amino groups) that are readily accessible to reactive radicals and oxidants. In contrast, high-molecular-weight polysaccharides tend to have more compact structures, which restrict the exposure of their reducible groups and render them less accessible for antioxidative reactions [22]. Thus, the impact of molecular weight on the antioxidant capacity of plant polysaccharides is both substantial and intricate, necessitating further investigation [23].

In the study, for the betterment of the extraction process of Akebia Fruit, the BBD and the GA-BP neural network models were utilized. We took the composite score calculated based on the purity and transfer rate of polysaccharides as the index. Compared with the response surface model, the GA-BP model demonstrated a higher R² coefficient, alongside lower RMSE and MAE values. The GA-BP model was used to optimize the UAE process of polysaccharides from Akebia Fruit, improving the model fitting degree compared with the traditional response surface method. Experimental results demonstrated that the integration of UAE with artificial neural networks not only facilitated the efficient extraction of polysaccharides but also optimized the extraction process, providing a reliable basis for the industrial-scale production of polysaccharide extraction processes.

The neutral polysaccharide AFP-1 from Akebia Fruit was purified by ion chromatography. Subsequent systematic characterization of AFP-1 included analyses of molecular weight and monosaccharide composition. Results revealed that AFP-1 was a neutral polysaccharide with a molecular weight of 13,775 Da, composed of mannose, ribose, glucose, galactose, and fucose. Finally, the antioxidant activity of AFP-1 was evaluated in vitro, and experimental data preliminarily confirmed that AFP-1 exhibited antioxidant capacity. This study provided an experimental basis for further research on AFP-1, and aimed to lay a solid and far-reaching foundation for the large-scale development, industrial application, and further exploration of polysaccharides from Akebia Fruit.

2. Materials and methods

2.1. Material and chemicals

Decoction Pieces of Akebia Fruit were obtained from Hangzhou Kanglun Chinese Medicine Slices Co., Ltd (Hangzhou, China). NaIO_4, D₂O NaOH, Anthrone and NaCl were obtained from Macklin Biochemical Technology Co., Ltd (Shanghai, China). Glucose. DEAE-Sepharose 52 were obtained from Shanghai yuanye Bio-Technology Co., Ltd (Shanghai, China). Sulfuric acid, Hydrochloric Acid and Chloroform were obtained from the Huadong Medicine Co., Ltd (Hangzhou, China). Sephadex G100 was obtained from Cytiva (Massachusetts, United States). KBr and Sodium Hydroxide were obtained from Shanghai Aladdin Biochemical Technology Co., Ltd (Shanghai, China). Deionized water was produced with a Millipore water purification system (MA, USA). The DPPH free radical scavenging ability assay kit, T-AOC detection kit (ABTS method) and T-AOC detection kit (FRAP method) were all purchased from Nanjing Jiancheng Bioengineering Institute (Nanjing, China).

2.2. Extraction of polysaccharide from Akebia Fruit

The dried Akebia Fruit was crushed and passed through a 30-mesh sieve. Then, it was placed in a conical flask, mixed with an appropriate quantity of water, and treated in an ultrasonic extractor (KQ-500DE, Kunshan, China) for a certain period. To the resulting solution, anhydrous ethanol was added to make the concentration reach 80 % and the mixture was precipitated at 4 ℃ for 24 h [24]. Subsequently, raw polysaccharide was isolated by low-speed centrifugation and then was dissolved in hot water. The content was ascertained using the anthrone-sulfuric acid method by UV (UV-2600i, shimadzu, Japan). Calculate the purity transfer rate and the comprehensive score using the Eq. (1), Eq. (2) and Eq. (3).

$$\mathit{polysaccharidepurity}(\%)=\frac{polysaccharidecontentration\left(g/mL\right)\times filtratevolume\left(mL\right)}{\mathit{AkebiaFruit}\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}\left(\mathrm{g}\right)}$$ (1)

$$\mathit{polysaccharidetransferrate}(\%)=\frac{polysaccharidecontentration\left(g/mL\right)\times filtratevolume\left(mL\right)}{\mathit{ThemaximummassofpolysaccharidesinAkebiaFruit}\left(\mathrm{g}\right)}$$ (2)

$$\mathit{compositescore}=polysaccharidepurity(\%)\times 50\%+polysaccharidetransferrate(\%)\times 50\%$$ (3)

2.3. Optimization of extraction process

2.3.1. BBD experiment

Utilizing the Design-Expert 13 software, the experiments with three factors at three levels were devised as indicated in Table 1. The composite score of polysaccharide was taken as an indicator [25].

Table 1.

Variables and their levels.

level	factor
	temperature /℃	Solid-liquid ratio /w/v	power /w
−1	65	1:45	350
0	70	1:50	400
1	75	1:55	450

2.3.2. GA-BP artificial neural network

The experimental data obtained from BBD were utilized for GA-BP artificial neural network model estimation. The selection criteria for artificial neural networks were determined based on regression analysis and RMSE evaluation. The quantity of neurons and activation functions in the hidden layer were selected through a trial-and-error approach [26]. The construction of the GA-BP model was carried out with MATLAB R2024a software. This model typically consisted of a three-layer architecture: the input layer, the hidden layer, and the output layer. The input layer comprised three neurons, which were associated with the extraction power, solid–liquid ratio, and extraction temperature. The output layer consisted of one neuron that was representative of the composite score. For a comprehensive assessment of the models' fitting performance, R², RMSE, and MAE were calculated [20]. Specifically, these metrics were computed by comparing the predicted values with the actual values for both the BBD model and the GA-BP model. Through a detailed analysis and comparison of these calculated values, a thorough evaluation of the two models' fitting capabilities was conducted. The R² metric is a crucial indicator used to evaluate the goodness of fit of data. Generally, the closer the value of R² is to 1, the better the goodness of fit of the data. The smaller the RMSE, the smaller the overall deviation between the predicted values and the true values, indicating stronger fitting or predictive ability of the model for the data. The smaller the MAE, the closer the predicted values are to the true values, indicating stronger fitting or predictive ability of the model for the data.

$$R^2=1-\frac{{\sum }_{i=1}^n{\left(y_i-\widehat{y_i}\right)}^2}{{\sum }_{i=1}^n{\left(y_i-\overline{y}\right)}^2}$$ (4)

$$\mathit{RMSE}=\sqrt{\frac{1}{n}{\sum }_{i=1}^n{\left(y_i-\widehat{y_i}\right)}^2}$$ (5)

$$\mathit{MAE}=\frac{1}{n}{\sum }_{i=1}^n\left|y_i-\widehat{y_i}\right|$$ (6)

Where $y_i$ represented the actual value of the dependent variable for sample i, $\widehat{y_i}$ represented the predicted value from the regression model for sample i, $\overline{y}$ represented the mean of the observed values of the dependent variable. n represented number of observations (samples).

2.3.3. Validation experiment

validation experiments were carried out separately according to the optimization conditions determined by BBD model and the GA-BP artificial neural network model.

2.4. Isolation and purification of AFP

In step 2.3, raw polysaccharide solution was taken. The Sevag reagent was mixed with it to prepare the mixture [27]. The mixture underwent shaking for 5 min before being centrifuged. The operations described above were replicated twice so as to fulfill the objective of protein removal. After removing the protein with the Sevag reagent, the polysaccharides were precipitated with ethanol to remove the Sevag reagent [28].

The DEAE-52 cellulose column chromatography enables the highly efficient isolation and purification of polysaccharides by virtue of its remarkable ion-exchange properties, especially its high adsorption capacity for ionic species [29]. The raw polysaccharide solution after deproteinization was slowly loaded onto DEAE-52 cellulose column (1.6 cm × 60 cm). Subsequently, gradient elution was carried out with a NaCl solution (0–0.5 M) as described in reference [30], and the fraction with the highest polysaccharide content was collected. All the eluates were collected at a volume of 5.6 mL per tube, and the anthrone-sulfuric acid method was employed to detect the polysaccharide content in these eluates [31]. Subsequently, the distilled-water-eluted samples were collected. The combined components were concentrated, and packed into a dialysis bag to remove NaCl [32], and then underwent freeze-drying to yield purified polysaccharides and prepared as a 10 mg/mL polysaccharide solution. At a flow rate of 0.5 mL/min, the solution was subsequently washed with deionized water, followed by further purification on a Sephadex G-100 column (2.6 cm × 100 cm) [33]. Collection of each tube took place over a 10-minute period. And the contents were measured using the same method.

2.5. Characterization of AFP

2.5.1. Molecular weight determination

The molecular weight distribution of the polysaccharides was analyzed using HPGPC. The uniformity and molecular weight of AFP-1 were assessed through an HPGPC method utilizing an HPLC instrument (Agilent 1260, CA, USA) equipped with two serially connected 8 mm × 300 mm ultrahydrogel linear columns (KS-804, KS-802, Shodex Co., Japan) [34]. Column elution was conducted using 0.2 mol/L NaCl at a flow rate of 0.8 mL per minute. All samples were prepared at a concentration of 2.0 mg/mL. Injections of 20 µL aliquots were made for the analytical process.

2.5.2. Congo red assay

The Congo red test underwent minor modifications compared to the previously reported methods by Du et al [35]. 1 mL of the AFP-1 solution (1 mg/mL) was combined with 1 mL of Congo red solution (0.4 mmol/L). The above-mentioned solutions were respectively mixed with NaOH of different concentrations varying from 0 to 0.5 mol/L. These solutions were kept at room temperature for a while before being analyzed using the UV instrument in the 400–600 nm range. Congo red served as a control.

2.5.3. FTIR spectral and UV spectral analysis

The AFP-1 powder was mixed with KBr powder in a 1:10 wt-to-weight ratio to form a translucent sheet. The FTIR instrument (Nicolet IS50, Thermo Fisher, USA) was utilized to capture the sample's spectra in the infrared range of 400 cm⁻¹ to 4000 cm⁻¹ [36]. The absorption spectrum of AFP-1 was analyzed using the UV instrument over a wavelength range of 200 to 400 nm [37].

2.5.4. SEM analysis

The surface texture of AFP-1 was analyzed using SEM (SU8010, Hitachi, Japan) after gold coating to ensure the sample's electrical conductivity. A precise amount of AFP-1 dry powder was positioned on a conductive gel-coated slide, and any extra polysaccharide powder was carefully removed. The prepared sample was then placed into an ion sputtering vacuum chamber equipped with a gold target. Following a certain period, the sample was observed under an electron microscope, and a suitable area was selected to clearly visualize the polysaccharide structure and capture high-resolution images [38].

2.5.5. Monosaccharide composition

The monosaccharide composition test was made with slight modifications compared to the previously reported methods by Yang et al [39]. HPLC was utilized to analyze the monosaccharide composition of polysaccharides. 2 mg AFP-1 powder was hydrolyzed with 2 mol/L TFA (1 mL) at 105 ℃ for 6 h. Centrifuged and removed the supernatant. The residue after TFA removal by co-evaporation with methanol and dissolved by adding 2 mL ultra-pure water. 400 μL PMP-methanol solution (0.5 mol/L) and NaOH solution (0.3 mol/L) were added to 400 μL polysaccharide hydrolysate and kept at 70 ℃ for 1 h. Added 400 μL HCl (0.3 mol/L) solution to adjust the pH to neutral. The derived products were extracted with chloroform. Subsequently, the monosaccharide content in AFP-1 was analyzed via HPLC. A Waters Alliance e2695 liquid chromatograph, equipped with a UV detector and a Diamonsil C18 column (4.6 × 250 mm, 5 μm), was employed. The analysis was conducted with an injection volume of 10 μL at a detection wavelength of 245 nm. The mobile phase was composed of phosphate buffer solution and acetonitrile.

2.5.6. NMR analysis

The AFP-1 samples were dissolved in deuterated water (D₂O) and underwent deuterium exchange via freeze-drying. 600 MHz NMR Spectrometer (Magnet System 600′54 Ascend LH, BRUKER, Germany) was employed to conduct 1D and 2D NMR analyses, such as ¹H NMR, ¹³C NMR, H-H COSY, and HSQC analyses.

2.5.7. Periodate oxidation analysis

13.19 mg of AFP-1 was added to 25 mL of a 0.03 mol/L NaIO₄ solution, and the mixture was stored at 4 °C in the dark. Every 6 h, a 0.1 mL sample was taken from the solution, and its absorbance at 223 nm was measured by UV. When it reached a steady state, the consumption of NaIO₄ was determined using the NaIO₄ standard curve (y = 9.9133x-0.0274, R² = 0.999). Next, 2 mL of the mixed solution was taken, and the formic acid concentration was determined by titrating with a 5 mmol/L NaOH solution. The type of glycosidic bond could be ascertained through the measurement of NaIO₄ consumption and formic acid production.

2.6. Determination of antioxidant activity

In this study, the antioxidant capacity of AFP-1 at different concentrations (5.625, 11.25, 22.5 and 45 mg/mL) was evaluated via DPPH radical scavenging assays and total antioxidant capacity (T-AOC) measurements. We used the DPPH free radical scavenging ability assay kit, the T-AOC detection kit (ABTS method) and the T-AOC detection kit (FRAP method) commercial kits with the protocols provided by the manufacturer.

2.7. Statistical analysis

Data were presented as mean ± standard deviation (SD). Graphs were generated using Origin 2024 software and GraphPad Prism 8 software.

3. Results and discussion

3.1. Extraction optimization

3.1.1. Results of BBD experiment

The yields of polysaccharide, along with their physicochemical properties, varied according to the diverse combinations of factors at distinct levels. In previous experimental results, it was found that extraction temperature, extraction power, and solid–liquid ratio were crucial for polysaccharide extraction. Building on single-factor experimental results, the initial ranges for extraction temperature, extraction power, and solid–liquid ratio were established. In this design, the composite score calculated based on a 50 % polysaccharide transfer rate and a 50 % polysaccharide purity was used as the response variable, aiming to further improve the extraction process and boost the overall efficiency and quality of the extraction.

The data were analyzed and modeled by Design-Expert 13 software and its results under different conditions were shown in Table 2. Additionally, the data were modeled using a quadratic polynomial regression equation, as shown in Eq. (7).

$$Y=50.77+0.3138A+0.7513B-0.4650C-1.00AB+0.6125AC+0.6775BC-7.23A^2-6.93B^2-5.26C^2$$ (7)

Table 2.

The actual and predicted data from the BBD and the predicted data from GA-BP.

Runs	A	B	C	Actual score	BBD Predicted score	GA-BP predicted score
1	70	350	1:45	37.24	38.97	38.98
2	75	450	1:50	33.33	36.68	34.73
3	70	450	1:55	41.28	39.55	41.00
4	70	350	1:55	33.26	36.69	30.82
5	65	400	1:45	37.44	39.05	37.38
6	75	350	1:50	38.99	37.17	39.17
7	75	400	1:45	38.37	38.45	38.36
8	70	400	1:50	49.47	50.77	49.23
9	70	450	1:45	42.55	39.12	46.70
10	70	400	1:50	47.18	50.77	49.23
11	65	400	1:55	36.98	36.90	38.62
12	70	400	1:50	51.74	50.77	49.23
13	65	350	1:50	37.89	34.54	36.56
14	70	400	1:50	52.04	50.77	49.23
15	70	400	1:50	53.43	50.77	49.23
16	75	400	1:55	40.36	38.75	41.75
17	65	450	1:50	36.23	38.05	35.97

Here, the composite score was denoted by Y, extraction temperature was represented by A, extraction power was designated as B, and solid–liquid ratio was assigned to C. From Eq. (7), a significant negative correlation was observed between solid–liquid ratio and composite score. In contrast, both the extraction temperature and extraction power were positively correlated with the composite score. Among them, extraction power showed the strongest positive correlation, whereas temperature exerted the most significant negative effect.

The fitted model underwent ANOVA. According to Table 3, when p < 0.05, the established model was valid and reliable. Conversely, when p > 0.05, it indicated that the model lacks an adequate fit. A low p value (0.0041, 0.0051 and 0.0188) indicated that the squared terms of extraction temperature, extraction power, and solid–liquid ratio significantly influence the composite score. The coefficient of variation (8.51) further confirmed the model's excellent reproducibility. Moreover, with an R² value of 0.8749 for the model, it could account for 87.49 % of the variation in the composite score. From the F value of the regression model, the primary and secondary order of the influencing factors on the polysaccharide extraction rate was: B > C > A.

Table 3.

ANOVE analysis of BBD model.

Source	Sum of squares	df	Mean square	F value	p value
model	614.89	9	68.32	5.44	0.0181	significant
A	0.7875	1	0.7875	0.0627	0.8095
B	4.52	1	4.52	0.3595	0.5677
C	1.73	1	1.73	0.1377	0.7215
AB	4	1	4	0.3185	0.5901
AC	1.5	1	1.5	0.1195	0.7398
BC	1.84	1	1.84	0.1462	0.7136
A2	220.01	1	220.01	17.52	0.0041
B2	202.41	1	202.41	16.12	0.0051
C2	116.32	1	116.32	9.26	0.0188
Residual	87.92	7	12.56
Lack of Fit	63.72	3	21.24	3.51	0.1283	not significant
Pure Error	24.21	4	6.05
Cor Total	702.81	16
R2	0.8749
Adj R2	0.7141
C.V. %	8.51

The quadratic regression equation was employed to build a 3D response surface plot. This plot served to elucidate both the individual impacts of each variable and the interactive effects between any two variables on the composite score. It could be seen from Fig. 1 A-C that the corresponding contour plot was circular, indicating that the interaction between AB, AC and BC had no significant effect on the comprehensive score of polysaccharides, which was consistent with the judgment of the F value.

Fig. 1.

Response 3D surface plot of the three variables to the interaction of composite score and Result of ANN.A: the interaction between extraction temperature and extraction power, B: the interaction between extraction temperature and solid- liquid ratio, C: the interaction between extraction power and solid–liquid ratio. D-F: the results of ANN, D: fitness function plot of GA-BP neural network, E: GA-BP neural network MSE for different data sets, F: Comparison Chart of BP Neural Network Predicted Values and True Values Before and After Optimization.

The optimal response surface fitting conditions were set at 428 W, 1:53 g/mL, 72.6 ℃, and the predicted composite score was 44.614. Based on the simplicity of operation, the predicted optimal conditions were appropriately adjusted, as indicated in Table 4.

Table 4.

Results of verification experiments and comparison between the two models. ($\overline{\mathrm{x}}$ ± SD, n = 3).

Model	Optimal conditions			Composite score		R2	RMSE	MAE
	A	B	C	predict	actual
BBD	73	450	1:53	44.614	42.18 ± 0.36	0.87	2.28	1.99
GA-BP	65	400	1:50	50.21	51.22 ± 0.46	0.90	2.04	1.57

The BBD model helped establish a quantitative relationship between dependent and independent variables. This relationship not only revealed the priority ranking of the independent variables and their interactions but also allowed for data prediction, providing a solid foundation for assessing the model's applicability and predictive power [40].

3.1.2. Results of GA-BP neural network

MATLAB R2024a was used to analyze the experimental data. Thirteen data sets from the BBD experiments were used for training, and the other five data sets were reserved for testing. The relationship between each generation and fitness values was calculated, as shown in Fig. 1 D. The optimal fitness value was 5.09855, and the average fitness value was 1138.5. Fig. 1 E displayed the performance of the training set and test set under different numbers of epochs. The number of neurons in the hidden layer which was regarded as optimal was found to be four. Fig. 1 F showed that the predicted values of the GA-BP model were closer to the true values compared to the BP model. In contrast, the GA-BP model could adaptively adjust weights and parameters, granting it strong generalization abilities. These capabilities led to better predictive accuracy, more solid forecasting of superior process outcomes, and a more thorough evaluation of model performance [19].

This combination was identified through the advanced predictive capabilities of the GA-BP model within the framework of the BBD experimental design, aiming to achieve the most efficient and effective extraction outcome. The GA-BP model generated a predicted composite score of 50.21 and identified the optimal extraction parameters (1:50 g/mL, 400 W, and 65 ℃) through experimentation.

3.1.3. Validation experiment

Subsequently, validation experiments were carried out separately according to the optimization conditions determined by BBD model and the GA-BP neural network model. A larger coefficient of R², accompanied by smaller values of RMSE and MAE, indicated a more superior performance of the model. Compared with the BBD models, the GA-BP model demonstrated a higher R² coefficient, alongside lower RMSE and MAE values. As demonstrated in Table 4, the experimental validation results demonstrated that the GA-BP model outperformed the BBD model. The enhanced accuracy of the ANN approach might be attributed to its inherent universal capacity to approximate nonlinear systems. By contrast, the BBD model was founded solely on second-order polynomial regression. Validation experiments’ results closely aligned with the values predicted by the GA-BP model, confirming that the optimized process parameters of the model were both reliable and stable. Experimental results showed that combining UAE with artificial neural networks not only enabled efficient extraction of polysaccharides but also optimized the extraction process, making it a viable guidance for industrial production of polysaccharide extraction processes, providing data support for the extraction of other polysaccharides.

What’s more, the intelligent algorithm module could be migrated to the extraction optimization of other traditional Chinese medicine components such as flavonoids and alkaloids, shortening the process development cycle.

3.2. AFP-1 characterization

3.2.1. Preparation of AFP-1

As depicted in Fig. 2 A-B, the polysaccharides present in Akebia Fruit underwent separation and purification via the DEAE-52 Cellulose column chromatography method and SephadexG-100 column chromatography method. Subsequently, four distinct fractions, namely AFP-1, AFP-2, AFP-3, and AFP-4, were acquired following elution with deionized water and sodium chloride solutions at concentrations of 0.1–0.3 mol/L. AFP-1 possessed the highest content among all tested samples, thus making it suitable for follow-up research. AFP-1 was further purified using a Sephadex G-100 column, as depicted in Fig. 2 B. Then the AFP-1 was finally obtained and preliminarily identified as a neutral polysaccharide. This accomplishment not only validated the efficacy of the employed extraction and purification methodologies but also established a robust experimental basis for further in-depth investigations.

Fig. 2.

Preparation of AFP-1 and the characterization of AFP-1. A: DEAE-52 chromatography of AFP, B: Sephadex G-100 chromatography of AFP-1, C: Measurement of molecular weight, D: Congo red experiment, E: FTIR spectra of AFP-1, F: UV spectra of AFP-1.

3.2.2. Measurement of molecular weight

Polysaccharides with lower molecular weights had weaker intramolecular hydrogen bonding, leading to freer amino and hydroxyl groups, which were more conducive to exerting their active effects [23]. As depicted in Fig. 2 C, it was demonstrated that AFP-1 was a homogeneous polysaccharide. The standard curve for dextran was y = -0.00540373x^3 + 0.288705x^2–5.42501x + 39.6686 (R² = 0.996). AFP-1 had a relative molecular weight of 13,775 Da, indicating it was a polysaccharide with a relatively small molecular weight.

3.2.3. Congo red assay

Under alkaline conditions, polysaccharides with a triple helix structure can form complexes with Congo red, causing a red shift in the maximum absorption wavelength [41], and may exhibit enhanced anticancer, antioxidant, and other biological activities [42]. As depicted in Fig. 2 D, AFP-1 exhibited a red shift in the maximum absorption wavelength compared to the single Congo red solution, indicating the presence of a triple helix structure.

3.2.4. FTIR spectral and UV spectral analysis

Understanding the structure–property relationship is crucial, as the complex glycosyl and glycosidic linkages determine their diverse biological functions.

The initial characterization of the structural features of AFP-1 was conducted through FTIR spectroscopy. As depicted in Fig. 2 E, a prominent absorption peak at 3396.2 cm⁻¹ was attributed to the stretching vibration of hydroxyl groups involved in intermolecular or intramolecular hydrogen bonding. The absorption peak at 2926.63 cm⁻¹ corresponded to the characteristic stretching vibration of C-H bonds within the polysaccharides. Notably, the absence of a characteristic peak at 1700 cm⁻¹ suggested the neutrality of the polysaccharide in question.

The absorption peak recorded at 1638.63 cm⁻¹ was ascribed to the stretching vibration of bound water (H-O-H), whereas the peak situated at 1371.68 cm⁻¹ corresponded to the bending vibration of C-H bonds in polysaccharides. Two absorption peaks, specifically at 1152.87 cm⁻¹ and 1035.51 cm⁻¹, were likely associated with the functional groups C-OH and C-O-C, respectively, in polysaccharides. It was noteworthy that the absence of significant absorption peaks in the regions of 890 cm⁻¹ and 925 cm⁻¹ implies that AFP-1 did not contain furanose units. Conversely, the absorption peak at 607.89 cm⁻¹ signified the presence of pyranose rings in the structure of AFP-1.

Fig. 2 F displayed the UV absorption spectrum of AFP-1 spanning from 200 to 400 nm. No absorption peaks were observed between 260–280 nm, demonstrating that AFP-1 did not contain nucleic acids or proteins.

3.2.5. Scanning electron microscopy

Polysaccharides, being complex biomolecules, carried more detailed structural information than proteins, nucleic acids, and lipids. The extraction and drying methods had a considerable impact on both the surface and internal structure of polysaccharides. The microstructures of AFP-1 were observed by SEM at different magnifications of 1,000×, 2,000×, and 8,000 × .

As illustrated in Fig. 3 A-C, the AFP-1 exhibited a layered structure composed of sheets, rods, and spherical forms, which were stacked. Despite their smooth surfaces, these structures exhibited surface perforations.

Fig. 3.

Characterization of AFP-1. A-C: SEM analysis of AFP-1 with different magnifications, including 1 000×, 2 000×, and 8 000×; D-E: HPLC result of monosaccharide composition for AFP-1, D: blank reagent, F: mixed standard, E: AFP-1 simple. (1: D-Mannose; 2: D-ribose; 3: rhamnose; 4: D-glucuronic acid; 5: D-galacturonic acid; 6: D-(+)-Glucose; 7: D-galactose; 8: Arabinose; 9: L-Fucose).

3.2.6. Monosaccharide composition

As illustrated in Fig. 3 D-F, it was known that AFP-1 contained five monosaccharides: mannose, ribose, glucose, galactose and fucose, with a molar ratio of 1.81:1.00:6.31:4.46:1.97. The proportion of glucose was 40.49 %, the proportion of mannose was 11.61 %, the proportion of ribose was 5.85 %, the proportion of galactose was 28.61 %, and the proportion of fucose was 11.53 %. Therefore, AFP-1 was a heteropolysaccharide, and glucose was the main monomer.

3.2.7. NMR analysis

The chemical shift of the hydrogen proton (H-1) on the anomeric carbon (C-1) of polysaccharides occurs in a lower magnetic field, generally within 4.5–5.5 ppm. This characteristic facilitates structural analysis [43]. Specifically, the chemical shift of H-1 in α-pyranose is greater than 4.95 ppm, in β-pyranose it is less than 4.95 ppm, and in furanose it is approximately 5.4 ppm [44]. These values can be utilized to determine the configuration of the sugar ring.

As illustrated in Fig. 4 A-B, the anomeric region appeared in 4.66–5.23 ppm and 95.56–105.30 ppm, which suggested that both the α-and β-conformations were present in AFP-1. In ¹H NMR, 4.79 ppm was the chemical shift of solvent D₂O.The range of 3.2 ppm to 4.19 ppm was H-2 to H-6 on the sugar ring C-2 to C-6 of AFP-1 In ¹³C NMR, the carbon signals of polysaccharides are generally distributed in the range of 60–110 ppm. The chemical shifts in the chemical shift positions of C-2 to C-6 were approximately in the range of 60–90 ppm [45]. For ¹³C NMR, the overlapping of complex carbon signals in the range of 60.58–77.77 ppm corresponded to the overlapping of a large number of hydrogen signals in the range of 3.0–5.5 ppm in 1H NMR. This was because the chemical shifts of the same atoms had severe signal overlap in one-dimensional nuclear magnetic resonance. The resonance of the unsubstituted C-6 of the glucosyl residue gave rise to the signal near 60 ppm. The signal peak at 105.30 ppm indicated the presence of a β-configuration in the polysaccharide polymer. It was likely the signal peak generated by C-1 in β-galactose. In combination with the data from H-H COSY (Fig. 4 C) and HSQC (Fig. 4 D), the chemical shifts of the rest of the protons and their associated carbon atoms were attributed. Moreover, the H-H COSY experiment revealed correlations between the protons of A(H1-H2), B(H1-H2), and C(H1-H2). Meanwhile, the HSQC experiment demonstrated correlations between AH1-CC1, BH1-BC1, and CH1-AC1.

Fig. 4.

NMR spectrum of AFP-1. A: ¹H NMR spectrum. B: ¹³C NMR spectrum, C: H-H COSY, D: HSQC.

3.2.8. Periodate oxidation analysis

AFP-1 produced 84 ± 3 μmol of formic acid. The formation of formic acid supported the presence of a 1 → 6 linkage. AFP-1 consumed 3.727 ± 0.081 mmol of NaIO₄. The amount of NaIO₄ consumed was more than double the quantity of formic acid produced. This suggested the existence of 1 → 2 and 1 → 4 glycosidic bonds, and it was likely that 1 → 3 glycosidic bonds were also present.

The biological activity of polysaccharides in traditional Chinese medicine is inherently intertwined with their complex structural architecture [9]. This experiment involved systematic characterization of the primary structure of AFP-1, encompassing analyses of molecular weight, monosaccharide composition, and molar ratio. For follow-up investigations, the exploration of the polysaccharide's higher-order structures will be prioritized to deepen mechanistic understanding.

3.3. Antioxidant activity analysis

Extensive research had demonstrated that oxidative stress plays a critical role in the onset and progression of many chronic human diseases, such as cardiovascular diseases [46]. Growing evidence indicate that polysaccharides derived from natural sources can exert antioxidant effects, reduce oxidative stress in the body, and thus hold therapeutic potential in diseases associated with oxidative damage [47]. Thus, the assessment of polysaccharides' antioxidant capacity is pivotal for evaluating their biological efficacy.

The total antioxidant capacity of AFP-1 at different concentrations was shown in Fig. 5 A-B. The scavenging ability of AFP-1 at different concentrations on DPPH free radicals was shown in Fig. 5 C. Within a specific concentration range, the scavenging ability of polysaccharides on ABTS free radicals and their total antioxidant capacity gradually increased, demonstrating a concentration-dependent behavior. What’s more, as the polysaccharide concentration went up, the rate of increase in the scavenging effect on DPPH free radicals slowed down, as shown in Fig. 5 C. The results could have preliminarily demonstrated that AFP-1 had antioxidant effects.

Fig. 5.

A: the result of Total antioxidant capacity (ABTS method), B: the result of Total antioxidant capacity (FRAP method), C: the capacity of scavenging DPPH radicals. ($\overline{\mathrm{x}}$ ± SD, n = 3).

Subsequent research could systematically explore the biological activities of AFP-1, by employing both in vitro and in vivo models.

4. Conclusion

With a thorough contrast, the BBD method combined with the GA-BP model was selected to optimize the extraction process of Polysaccharides from Akebia Fruit. Experimental results demonstrated that the extraction process predicted by the GA-BP model yielded a higher content of polysaccharides. By integrating the BBD with the GA-BP model, the extraction process could be more accurately predicted to achieve the optimal solution, providing a viable approach for the extraction of polysaccharides from other traditional Chinese medicines. A new homogeneous neutral polysaccharide AFP-1 was isolated and purified from Akebia Fruit by ion chromatography. AFP-1 was observed to have a smooth surface with perforations, and its molecular weight was determined to be 13,775 Da. It presented a triple-helical structure and did not contain nucleic acids and proteins. As important bioinformatic molecules, plant polysaccharides exhibited biological activities, and their mechanisms of action were closely linked to their complex structures. Therefore, structural characterization of AFP-1 was performed, revealing that it had a triple-helical structure and contained no nucleic acids or proteins. Monosaccharide composition analysis and infrared spectroscopy analysis showed that AFP-1 was composed of five monosaccharides, and β-pyranose rings were present in its structure. NMR was used to determine the types and linkage patterns of glycosidic bonds in AFP-1. Preliminary detection confirmed that AFP-1 had antioxidant efficacy, thus hold therapeutic potential in diseases associated with oxidative damage These findings provided a solid basis for exploring polysaccharide bioactivities, structure–activity relationships, and potential pharma/nutraceutical applications. This basis was essential for guiding future studies to uncover AFP-1′s functional mechanisms and optimize its use in research and industry, thereby advancing modern polysaccharide-based natural product development.

CRediT authorship contribution statement

Yusang Chen: Writing – original draft, Validation, Methodology, Investigation, Formal analysis, Data curation. Meiling Wu: Writing – original draft. Xiao Xu: Investigation. Shunyao Zhu: Methodology. Mengdan Shen: Data curation. Anting Ma: Formal analysis. Zhennan She: Validation. Senlin Shi: Data curation. Xi Han: Writing – review & editing, Supervision, Project administration. Ting Zhang: Writing – review & editing, Supervision, Project administration.

Declaration of competing interest

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