ANN modeling and RSM optimization of ultrasound-assisted extraction of protodioscin-rich extracts from Tribulus terrestris L.

Abstract

Tribulus terrestris L. is a herb renowned for its abundance of saponins, flavonoids, and alkaloids, which are utilized in treating various health conditions. Protodioscin functions by enhancing the conversion of testosterone into potent dihydrotestosterone, stimulating an increase in libido, red blood cell production from the bone marrow, and muscle development. Contemporary ultrasound-assisted-extraction (UAE) process employing green extraction solvents was selected to design the required separation system. The experimental plan was developed based on the independent operating variables – extraction time, operating temperature and solvent system composition in order to determine the influence of defined parameters and their interactions on the extraction yield and the presence of protodioscin. The 3D-RSM approach was introduced to determine the optimal values of studied independent variables in the area of maximal extraction yield. UAE process performed at optimal operating conditions generated maximal extraction yield (31 %, w/w) and protodioscin content of 5.9 mg/g dry plant matrix. Experimental data was used to develop an ANN for the defined extraction system using the operating variables values as the input matrix and observed yield as target matrix. Successfully designed and trained ANN generated high correlation (r = 0.9992) between observed data and predictive model’ outputs, and MSE value of 0.29107.

Nomenclature

a.

regression coefficient.

b.

regression coefficient.

r.

correlation coefficient.

r^2.

coefficient of determination.

T.

temperature [℃]

t.

time [min]

$y_m$

yield of total extract [%, w/w]

$y_u$

yield of total extract [%, w/w]

$y_e$

yield.

Greek letters

${\beta }_0-{\beta }_9$

regression coefficients.

Abbreviations

2D.

two dimensional.

3D.

three dimensional.

ANN.

Artificial Neural Networks.

CSE.

conventional solvent extraction.

DHT.

dihydrotestosterone.

EtOH.

ethanol solvent.

HPLC-DAD.

high-performance liquid chromatography with diode-array detection.

LH.

luteinizing hormone.

MAE.

microwave-assisted extraction.

RSM.

response surface methodology.

SFE.

supercritical fluid extraction.

TT.

Tribulus terrestris L.

UAE.

ultrasound-assisted extraction.

1. Introduction

Zygophyllaceae is a family of flowering plants containing 285 species in 22 genera [1]. Plants of the family Zygophyllaceae are trees, shrubs, or herbs, mostly found in dry habitats. They are typically short annual plants, ranging from 10 to 60 cm in height, with hirsute or silky hairy herbs with opposite, unequal, paripinnate leaves and elliptic or oblong-lanceolate leaf shapes [2], [3]. The fruits of this plant are ball-shaped with sharp, spines-like tips that are 3 to 6 mm long, and radially distributed with a diameter of 7 to 12 mm. This plant is widespread, it thrives in all conditions, even deserts, and on poor soils, it is found in Africa, West Asia, China, Japan, Korea, and Europe [4], [5]. Some of them are cultivated ornamental plants, such as species of the genera Guaiacum, Zygophyllum, Tribulus, and Larrea. In the subfamily of Tribuloideae [6], one of the most famous representatives is Tribulus terrestris L. (TT). In south Asia and north Africa, the fruits are used to treat infertility, and the roots are considered cardiotonic, and in some cultures this herb is also used to protect against glomerular inflammation and other acidic conditions, as well as a diuretic [7]. Traditional pharmacological studies prove that Tribulus terrestris L. [7] has a wide range of applications: for the treatment of inflammation, drying of the skin and itching, diseases related to the heart and cleaning vessels, cancers, infections caused by microbes, oxidative damage, hormonal problems and acts on the characteristics. These characteristics are due to all the compositions that include saponins [8], flavonoids and alkaloids [9].

The composition of the bioactive components present in the plant material depends on several aspects, such as geographical origin, type of soil, water quality. The composition of the bioactive total extracts also varies, depending on the selected extraction techniques, utilized solvent or defined optimal operating conditions for the designed separation system. The main groups of chemical constituent that Tribulus terrestris L. contains are saponins, flavonoids and alkaloids.

Steroidal saponins find wide application in the pharmaceutical industry due to their reported association with sex hormones, cortisone, diuretic steroids, vitamin D and cardio glycosides. Some of them are used as starting materials for organic synthesis in the pharmaceutical industry processes [10].

One of the components of interest when analysing the bioactive components from Tribulus terrestris L. is protodioscin, a steroidal saponin, an active component that can be used as an aphrodisiac. Protodioscin affects androgenic hormones such as testosterone and dihydrotestosterone by increasing their concentration. Tribulus terrestris L. contains bioactive components that are considered to be testosterone booster [11]. Saponins from TT bind to the receptors of the hypothalamus that detect sex hormones [12]. It partially blocks the misinterpretation of receptors that lead to hypothalamus sex hormone levels in the body are lower than they really are. The hypothalamus signals to start producing luteinizing hormone (LH) [13]. When the level of LH is increased, the natural production of testosterone also increases. LH is a hormone that is also responsible for sex drive [14]. Protodioscin is a steroidal saponin bioactive component that increases testosterone and dihydrotestosterone (DHT) [15].

In addition to protodioscin, other important bioactive components of interest that are reported in total extracts from this herbal material are prototribestin, dioscin and ruthein [16].

Various discontinuous or continuous, conventional or non-conventional extraction techniques have been employed for the separation of biologically active compounds from TT, such as maceration, percolation, ultrasound-assisted extraction (UAE), microwave-assisted extraction (MAE), and supercritical fluid extraction (SFE) [17], [18], [19]. Conventional extraction methods are inefficient, energy-intensive, and ecologically harmful processes that often require extended durations to complete. The opposite of that, modern extraction techniques, such as UAE, are more environmentally friendly, and selective, and provide a higher yield of bioactive compounds of interest.

In the UAE, acoustic cavitation is a significant phenomenon. When bubbles collapse near the surface of plant cells due to intense ultrasound waves, they create powerful forces that physically disrupt the cell wall and membrane. This disruption is instrumental in improving the penetration of solvent into the plant cell and enhancing contact between the solvent and bioactive compounds contained within. Extraction time, temperature, ultrasound frequency, pressure, particle size, and solvent-material ratio are often critical working parameters that influence the effectiveness of acoustic cavitation in the UAE [20]. Determining the effects of working parameters and optimizing their values are crucial steps in maximizing the yield and quality of extracted bioactive compounds, while also minimizing energy consumption and processing time.

A study focused on Tribulus terrestris L. extracts obtained by MAE reports of a maximum antioxidant activity of generated extracts achieved at 3 min extraction time. The extract displayed values of 54.86 ± 0.51 mg GA/g for TPC, 544.33 ± 4.50 mg QE/g for TFC, 87.67 ± 0.57 % for DPPH, 1051.67 ± 3.21 mM Fe²⁺/g for FRAP, and 86.34 ± 2.08 % for ABTS. These findings suggest that the antioxidant potential of the T. terrestris L. extract is confirmed by additional contemporary extraction technique. According to Mubarik et al. [18] the geographical and climate region have effect on the chemical composition, antioxidant potential and values of total phenolic compound in Tribulus terrestris L. extract. Ali et al. reported the value for total phenolic compounds (TPC) as 14.48 ± 0.16 mg GA/g of Tribulus terrestris extract.

In this study, the influence of the initial concentration and polarity of the solvent, extraction time, and temperature on the total yield of extracts rich in protodioscin from Tribulus terrestris L. were researched. The main goals were the design and optimization of the process of solid–liquid UAE of bioactive components from the plant matrix Tribulus terrestris L.

The novelty of the submitted manuscript in view of the existing literature can be observed through several aspects of modern chemical and control engineering. First aspect is the investigated herb material – Tribulus terrestris L.; although there are some articles investigating some properties and potential health benefits of this plant, authors considered its’ potential as a rich resource of bioactive components as a solid investigative point. Especially considering the fact that among other reported medical benefits, its’ use reflecting the aphrodisiac properties of the extract is not sufficiently investigated. Furthermore, this plant is abundantly present in the geographical region of southern and south-eastern Europe thus emphasizing the sustainability aspect of this investigation. Second, the experimental analysis was performed according to the basic principles of contemporary chemical process engineering, using ultrasound-assisted-extraction with environmentally benign solvent systems. This process design fits in the strict frames of green process engineering related to production of high quality products with added value. The final aspect of this work is a significant part of the entire work that includes the introduction of adequately designed artificial neural network for predictive modeling of the described green extraction process where the final product is an extract rich in bioactive compounds with highly sought-after properties and medical benefits.

2. Experimental procedures

2.1. Plant material

Aerial parts, leaves, stems, and fruits of the plant Tribulus terrestris L. were used as raw material for the separation of bioactive compounds of interest. The material was collected during the spring season of 2022 in the region of Staro Nagorichane in North Macedonia. The procured raw material samples were mixed and properly homogenized by a commercial mill. The moisture of the prepared matrix was determined following Japanese Pharmacopoeia 17 [21], which was less than 11 %. The ash content was determined in a laboratory furnace at 800℃ for 2 h, according to the Korean Pharmacopoeia 12 [22]. Furthermore, the experimental matrices were adequately stored in a dark, isolated place at room temperature (≈ 22℃), protected from external influences.

2.2. Chemicals

The chemicals used in the experimental work were purchased from a supplier according to the technical properties required for the defined analyses. The deionized water (Supelco, LC-MS grade) and ethanol 96 % (Alkaloid AD, Ph. Eur.) were employed to prepare the solvents. Solvent solutions used in experimental research processes were predefined according to the presence of deionized water (DIW) in the extraction system: 100 % DIW, 45 % DIW, 30 % DIW, and 4 % DIW. The mobile phase for HPLC-DAD analysis required acetonitrile (Supelco, hypergrade for LC-MS) and water (Supelco, LC-MS grade). Polyphenols analysis was performed using deionized water, Folin-Ciocalteu reagent, and ${\mathrm{N}\mathrm{a}}_2{\mathrm{C}\mathrm{O}}_3$ 20 % (w/w).

2.3. Conventional solvent extraction process (CSE)

The conventional solvent extraction (CSE) process for the isolation of bioactive components from the experimental matrix of T. terrestris L. was conducted in laboratory glass extractor with continuous mixing at 500 rpm. The stated CSE-maceration process was based on previously defined solvent systems, performed at working temperatures of 25, 35, and 45℃ and extraction times of 5, 10, 15, 30, 60, and 90 min. The observed experimental results for $y_M$ were used for process modelling. The experimental results regarding the yield of the total extract are further employed for defining the kinetics of the designed extraction process. The 3D-RSM method was introduced to mathematically describe the influence of experimental variables and their interactions on the response variable.

2.4. Ultrasound-assisted extraction (UAE)

UAE of protodioscin from pre-prepared T. terrestris L. matrixes (TT matrixes) was applied as a non-conventional extraction technique to achieve the separation of protodioscin-rich extracts from the analysed raw material. Coupled with the use of green extraction solvents, since DIW and ethanol both fit in the frames of green solvents, the defined UAE model was aligned with the essential concepts of green process engineering.

The UAE was performed in a laboratory-scale ultrasonic bath (Ei Niš OOUR-RC) at the constant operating frequency of 40 kHz and the constant material-to-solvent ratio of 1:20 (w/v). The experimental plan was designed to determine the influence of the following working parameters on the monitored response-extraction yield ($y_U$):

• -extraction time (5, 10, 15, 30, 60 and 90 min);
• -temperature (25, 35, and 45 °C);
• -solvent composition − ${\mathrm{H}}_2\mathrm{O}:{\mathrm{C}}_2{\mathrm{H}}_5\mathrm{O}\mathrm{H}$ ($x_{\mathit{DIW}}$ [v/v] of 0.04, 0.3, 0.45, and 1).

According to the experimental plan related to the UAE of protodioscin-rich extracts from T. terrestris L., the raw material and solvent system (5 g and 100 mL) were placed into a glass extraction vessel. Each unit was equidistantly positioned in the ultrasonic bath, which had been prefilled with distilled water at 2/3 of the working volume. After each of the extraction time sequences, samples were filtered through filter paper with a pore size of 25 μm, to separate the liquid phase containing the isolated extract from the solid matrix residue. The extraction solvent was evaporated from the resulting UAE extracts in a rotary vacuum evaporator (Büchi R-200) to the constant weight of the extraction vessel. The developed experimental plan was created to facilitate the study of the dynamics of the separation process. An adequate 3D-RSM approach was used to depict the functional dependency of the process output (extraction yield) from the input data presented through the values of experimental parameters. Obtained extracts were subjected to HPLC-DAD analysis to determine the presence and content of protodioscin.

2.5. HPLC-DAD analysis of protodioscin from Tribulus terrestris L. extract

Qualitative and quantitative determination of the protodioscin presence in selected T. terrestris L. extracts, was achieved through HPLC-DAD chromatographic analysis. The chromatographic analysis [23] was performed using a High-Performance Liquid Chromatography (HPLC) system (Hitachi Chromaster HPLC), equipped with a Diode Array Detector (DAD) (Hitachi Chromaster 5430), an automatic injector (Hitachi Chromaster 5260), and a column temperature controller (Hitachi Chromaster 5310). The mobile phase consisted of acetonitrile and water in a gradient mode (initially a linear gradient with acetonitrile from 10 % to 60 % (v/v) that was used in the initial 15 min, followed by isocratic mode with 60 % (v/v) acetonitrile for additional 5 min.

According to the method for the determination of protodioscin, an analysis was made at optimal extraction points with maximal extraction yield observed for the CSE and UAE processes.

2.6. Analysis of polyphenolic compounds from Tribulus terrestris L. Extract

The most common technique used to determine the total content of polyphenols is UV/Vis spectrophotometry using the Folin-Ciocalteu reagent.

A 0.1 mL of diluted extracts was transferred into test tubes and their volumes were made up to 0.5 mL with deionized water. After the addition of 0.25 mL Folin-Ciocalteu reagent (0.2 N) and 1.25 mL of 20 % aqueous sodium carbonate solution, tubes were mixed for 1 min. The absorbance of the blue-colored solution was measured at 765 nm after 40 min of incubation at room temperature in the dark against a blank containing 0.1 mL of the extracting solvent instead of the tested samples using a spectrophotometer (Spectroquant Prove 600). All samples and readings were prepared and measured in triplicate. Gallic acid was used as standard. 0.5 mg/mL stock standard solution of gallic acid was prepared by dissolving 250 mg of dry gallic acid in 1 mL of extracting solvent and then diluted to 500 mL of distilled water. The stock solution was stored at 4℃. Working standards with concentrations ranging from 0 and 0.10 mg/mL were prepared by diluting the stock solution with deionized water. The standard calibration curve of gallic acid (0.00–––0.10 mg/mL) was constructed by plotting the absorbance against the concentration, following the equation:

$$x=(y-0.0163)/0.0108$$ (1)

where: x is concentration of polyphenolics (mg/L) in the extract and y is absorbance at 765 nm.

2.7. Modelling of extraction kinetics

Experimental data referring to the CSE and UAE processes kinetics was mathematically described using the function given in Eq. (2). The resulting extraction yield function was best fitted to independent variables’ values through the stated line of best fit.

$$y_k=at/(1+bt)$$ (2)

where: $y_k$ is a predicted value of the yield of total extract (%), t is extraction time (min), a and b are regression coefficients. The determination of the regression coefficients was done using the MATLAB-Curve Fitting software package.

2.8. Design application of modelled values

For the analysis of the influence of extraction time (t), temperature (T), and solvent system composition ($x_{\mathit{DIW}}$) on the reported yield of total extract, the response surface method (RSM) was employed. This method facilitated the development of a predictive mathematical model and the creation of response surfaces to optimize both CSE and UAE processes. The software package Stratigraphic Centurion XVI was used to apply the design of the predicted values.

Described analyses were based on a multilevel-factorial design with three input factors representing the working parameters. The value of the observed yields of the total extract was used as a dependent variable and model output. The generated 3D-RSM model is defined by the following expression representing a polynomial equation that correlates the experimentally obtained output as a function of working parameters and their interactions:

$$y={\beta }_0+{\beta }_{1}t+{\beta }_{2}T+{\beta }_3{x}_{\mathit{DIW}}+{\beta }_4{t}^2+{\beta }_{5}tT+{\beta }_{6}t{x}_{\mathit{DIW}}+{\beta }_7{T}^2+{\beta }_{8}T{x}_{\mathit{DIW}}+{\beta }_9{x_{\mathit{DIW}}}^2$$ (3)

where: y is the response or output value of the modelled yield of total extract (%), t is the extraction time (min), T is the working temperature (℃),$x_{\mathit{DIW}}$ is the presence of water in the solvent, ${\beta }_0$, ${\beta }_1$, ${\beta }_2$, ${\beta }_3$, ${\beta }_4$, ${\beta }_5$, ${\beta }_6$, ${\beta }_7$, ${\beta }_8$, ${\beta }_9$ are the corresponding regression coefficients. To demonstrate the influence of the working parameters t, T, and $x_{\mathit{DIW}}$ on the yield of total extract, a 3D response surface was created, while the optimal value of the modelled process is shown in detail through the contours of the response surface.

3. Results and discussion

The particle size of the ground material fraction used in the experimental part was 0.177 – 0.4 mm (40 – 80 Mesh). The moisture content in the raw material was 10.69 % and the ash content was 9.8 %.

3.1. Design of CSE process

The obtained laboratory results regarding the yield of total extract $y_M$, presented in Table 1, were used to produce an adequate mathematical model for the analysed system. The obtained model was successfully employed to study the process kinetics and define the functional dependence of the yield of the extract from the experimental parameters.

Table 1.

Yield of total extract from T. terrestris L. using the CSE process.

Working parameters		Yield of total extract $y_M$(%)
t (min)	T (℃)	$x_{\mathit{DIW}}=1$	$x_{\mathit{DIW}}=0.45$	$x_{\mathit{DIW}}=0.3$	$x_{\mathit{DIW}}=0.04$
5	25	10.3	4.6	1.6	0.2
10	25	13.2	5.6	2.2	0.5
15	25	15.5	7.6	3.6	0.6
30	25	19.2	9.2	4.7	0.7
60	25	21.4	10.3	6.3	0.9
90	25	22.6	11.2	7.1	0.9
5	35	11.5	5.0	2.9	0.8
10	35	14.3	8.2	4.9	0.9
15	35	16.2	8.9	5.5	1.2
30	35	18.5	11.3	8.4	1.6
60	35	20.1	13.5	10.4	1.9
90	35	20.7	13.4	11.9	2.0
5	45	15.5	8.1	4.9	1.9
10	45	17.8	10.2	6.4	2.4
15	45	19.2	11.6	7.6	2.7
30	45	20.8	13.4	10.4	3.4
60	45	21.5	14.8	12.0	3.6
90	45	22.0	15.6	12.8	3.8

The results presented in Table 1 showed that the yield of total extract obtained from the experimental matrix through the maceration process is directly proportional and positively affected with the increase of extraction time and temperature and inversely proportional (negatively affected) with the presence of ethanol. The kinetics data of the CSE process at 25℃, 35℃, and 45℃ based on the experimentally obtained results is fitted using Eq. (2). The characteristics of the empirical function for the different temperatures in the process of CSE with different solvents are given in Table 2.

Table 2.

Model characteristics of the kinetic function for the CSE process.

Solvent composition	T (℃)	a	b	r2	RMSE
	25	3.1249	0.1290	0.9900	0.5401
$x_{\mathit{DIW}}=1$	35	4.5508	0.2111	0.9925	0.03445
	45	9.5801	0.4288	0.9940	0.2143
	25	1.2610	0.1036	0.9803	0.4113
$x_{\mathit{DIW}}=0.45$	35	1.519	0.1037	0.9868	0.4275
	45	2.9021	0.1789	0.9892	0.3337
	25	0.3484	0.0386	0.9910	0.2298
$x_{\mathit{DIW}}=0.3$	35	0.6755	0.0470	0.9926	0.3313
	45	1.2036	0.0836	0.9892	0.4056
	25	0.0785	0.0749	0.9879	0.0312
$x_{\mathit{DIW}}=0.04$	35	0.1835	0.0814	0.9869	0.0646
	45	0.6751	0.1709	0.9860	0.0947

Fig. 1A showed that during the first fifteen minutes, the process is in the kinetic phase, primarily due to the complex interaction of process parameters – extraction time coupled with high polarity of the solvent. This phase represents kinetically controlled phase of the extraction process depicting rapid extraction dynamics, where most of the extract from the analysed material is released to the extraction medium (solvent). This phenomenon results from the complex influence of the extraction system on the process dynamics, where most of the extracted components derive from the external zone of the tissues thus influencing the mass transfer between the operating plant matrix and the solvent system. The second phase (15–90 min), the process dynamics reduces and enters diffusion controlled region since most of the bioactive components are already extracted from the experimental matrix and mass transfer phenomena occurs in the internal zone of the analysed material. In the region of 30–90 min, it enters equilibrium, where only residual components from the cell cytoplasm are extracted. The results presented in Fig. 1B indicated that after the initial rapid phase, dynamic equilibrium phase slowly begins to establish. It can be observed from Fig. 1C that this phenomenon requires more time for the CSE process with 30 % DIW. Fig. 1D showed that the process starts to slow down after thirty minutes and slowly enters an equilibrium.

Fig. 1.

Kinetics of separation of total predicted extract from T. terrestris L. matrixes using the CSE process. A − $x_{\mathit{DIW}}=1$, B −$x_{\mathit{DIW}}=0.45$, C –$x_{\mathit{DIW}}=0.3$ and D −.$x_{\mathit{DIW}}=0.04.$

The resulting empirical Eq. (3) of the second order, with values of the regression coefficients shown in Table 3, definite the function of the two operating variables t and T and the dependent $y_M$ for the discussed process.

Table 3.

Regression coefficients of the polynomial model for the designed CSE process.

Coefficient	Estimate value
${\beta }_0$	−5.4033
${\beta }_1$	0.1653
${\beta }_2$	−0.0122
${\beta }_3$	22.0173
${\beta }_4$	−0.0013
${\beta }_5$	−0.0003
${\beta }_6$	0.0715
${\beta }_7$	0.0030
${\beta }_8$	−0.0317
${\beta }_9$	−6.2680

For the graphical presentation of the importance/meaning of the estimated impact of the independent variables, arranged in descending order, a standardized Pareto chart was created. The Pareto chart represented the absolute values of the standardized effects using the F-statistic, which tests the null hypothesis that the effect is 0. While the reference line α is used to indicate the effect with statistical significance. For a perfect orthogonal design, all factors will be equal to 1. The created standardized Pareto chart for the considered process with a reference line of significance α = 0.05 is shown in Fig. 2. According to the created model and based on the analysis, the solvent composition, the extraction time, the temperature, and the interaction between extraction time and solvent composition have a positive statistically significant influence on the yield of total extract from TT. However, the extraction time-squared and solvent composition-squared have a negative statistically significant influence.

Fig. 2.

Standardized influence of working parameters on the yield of total extract for CSE process. A- extraction time, B- temperature, C- solvent composition.

To demonstrate the influence of operational parameters t, T, and $x_{\mathit{DIW}}$ on the yield of total extract as well as optimization of the studied process, a 2D and 3D response surface was created.

2D and 3D response surfaces presented in Fig. 3 showed that the optimal yield of total extract of 24.1 % was achieved at the extraction time of 87.2 min, the temperature of 45℃, and DIW presence, $x_{\mathit{DIW}}=1$ as a solvent system.

Fig. 3.

3D response surface plots (A, B, C) and 2D contour plots (A’, B’, C’) of the CSE process extraction yields as a function of the different interactions of the extraction parameters.

3.2. Design of UAE process

UAE as a non-conventional extraction technique was performed to determine the influence of t, T, and $x_{\mathit{DIW}}$ on the yield of total extract. The experimentally obtained results are given in Table 4.

Table 4.

Yield of total extract from T. terrestris L. matrixes using the UAE process.

Working parameters		Yield of total extract $y_U$ (%)
t (min)	T (℃)	DIW (100 %)	DIW (45 %)	DIW (30 %)	DIW (4 %)
5	25	17.7	10.3	5.5	1.4
10	25	19.5	11.9	6.7	1.6
15	25	20.4	13.3	8.2	1.8
30	25	21.6	14.3	9.3	2.0
60	25	22.5	15.1	9.9	2.1
90	25	22.7	15.7	10.6	2.2
5	35	19.2	15.8	2.9	1.6
10	35	21.9	17.6	4.9	1.9
15	35	23.5	18.6	5.5	2.2
30	35	25.7	19.8	8.3	2.5
60	35	26.4	20.5	10.3	2.7
90	35	27.3	20.9	11.9	2.8
5	45	20.5	14.3	8.7	1.9
10	45	23.6	16.7	10.5	2.4
15	45	26.3	18.6	12.2	2.7
30	45	28.5	20.0	13.0	3.2
60	45	30.3	20.9	14.1	3.6
90	45	31.0	21.7	14.6	3.8

The kinetics of the UAE at T of 25℃, 35℃, and 45℃ based on the results obtained is modelled by using Eq. (2). In Table 5 are the empirical function characteristics for the different working temperatures in the UAE process with different solvent systems.

Table 5.

Model characteristics of the kinetic function for the UAE process.

Solvent	T [℃]	a	b	r2	RMSE
	25	14.4611	0.6315	0.9830	0.2817
$x_{\mathit{DIW}}=1$	35	11.6927	0.4231	0.9867	0.3982
	45	10.6880	0.3381	0.9847	0.5602
	25	5.4158	0.3406	0.9839	0.2882
$x_{\mathit{DIW}}=0.45$	35	11.8082	0.5605	0.9832	0.2811
	45	0.3475	0.3475	0.9875	0.3491
	25	2.0071	0.1822	0.9853	0.2642
$x_{\mathit{DIW}}=0.3$	35	0.6755	0.0469	0.9914	0.3535
	45	3.9312	0.2619	0.9837	0.3191
	25	0.6678	0.2959	0.9860	0.0415
$x_{\mathit{DIW}}=0.04$	35	0.6277	0.2134	0.9935	0.0436
	45	0.6666	0.1691	0.9862	0.0937

The results presented in Fig. 4A and 4B complex interaction of process parameters – cavitation effect coupled with high polarity of the solvent. In the initial phase (0–10 min), the process is in a kinetically controlled extraction phase (rapid phase) and in the second phase (10–90 min) it enters diffusion controlled mass transfer dynamics. This phase represents kinetically controlled phase of the extraction process depicting rapid extraction dynamics, where most of the extract from the analysed material is released to the extraction medium (solvent). This phenomenon results from the complex influence of the extraction system on the process dynamics, where the ultrasonic cavitation effect causes disruption of the material cell wall thus promoting the mass transfer between the operating plant matrix and the solvent system. After the fifteenth minute, the process dynamics reduces and enters diffusion controlled region since most of the bioactive components are already extracted from the experimental matrix. In the last sixty minutes, it enters equilibrium, where only residual components from the cell cytoplasm are extracted. The results presented in Fig. 4B indicated that after the initial rapid phase, dynamic equilibrium phase slowly begins to establish. Analysis of the kinetics curve presented on Fig. 4C and 4D suggest that this phase requires more time for the UAE process with 30 % DIW and is followed by the second phase (diffusion controlled extraction kinetics).

Fig. 4.

Kinetics of separation of total predicted extract from T. terrestris L. matrixes using the UAE process. A − $x_{\mathit{DIW}}=1$, B −$x_{\mathit{DIW}}=0.45$, C –$x_{\mathit{DIW}}=0.3$ and D −..$x_{\mathit{DIW}}=0.04$

The resulting empirical equation of the second order, with values of the regression coefficients shown in Table 6, gives the function of the two operating variables t and T and the dependent $y_U$ for the analysed process.

Table 6.

Regression coefficients of the polynomial model for the designed UAE process.

Coefficient	Estimate value
${\beta }_0$	−4.1454
${\beta }_1$	0.0948
${\beta }_2$	0.0616
${\beta }_3$	32.0741
${\beta }_4$	−0.0011
${\beta }_5$	0.0011
${\beta }_6$	0.0556
${\beta }_7$	−0.0001
${\beta }_8$	0.2335
${\beta }_9$	−18.4373

Although the kinetics data for the investigated extraction process can be fitted using other mathematical models of higher order, selected model was chosen within the frames of this work. Selected kinetic model was chosen to mathematically describe experimentally obtained values for the separation process kinetics due to 2 main reasons: First, this model provides adequate fit for experimental data with excellent values for the correlation coefficient (r > 0.99) for each experimental curve. Further, this model is adequate for the designed separation system (extraction process involving plant matrixes) and is desirable for physical interpretation of the model parameters. Model coefficients are indicators for the mass transfer phenomena occurring during the extraction process. The first coefficient, a [% w/w · time-1] depicts the rate of the mass transfer and the second coefficient, b [time-1] represents the separation factor for the analyzed extraction process.

The results in Table 6 showed each of the estimated effects, that is, the estimated impact, as well as their interactions. Meanwhile, a standardized Pareto chart was chosen for a graphical representation of the importance of the estimated impact of the independent variables. The created standardized Pareto chart for the considered process with a reference line of significance α = 0.05 is shown in Fig. 5. The Pareto chart showed that the solvent composition, temperature, extraction time, interaction between temperature and solvent composition, the interaction between extraction time and solvent composition, and the interaction between extraction time and temperature solvent composition influence positively, while the solvent composition-squared, extraction time-squared and temperature-squared influence negatively on the yield of total extract.Fig. 6..

Fig. 5.

Standardized influence of working parameters on the yield of total extract for UAE process. A- extraction time, B- temperature, C- solvent composition.

Fig. 6.

3D response surface plots (A, B, C) and 2D contour plots (A’, B’, C’) of the UAE process extraction yields as a function of the different interactions of the extraction parameters.

3D and 2D response surfaces were created to show the effects of studied working parameters on the yield of total extract and for optimization of the UAE process.

The created response surfaces showed that UAE with water provides the higher yield of total protodioscin-rich extracts, as the level of the ethanol module in the solution increases, the total extract yield decreases. It is evident from the response surfaces that in the fast phase the increase in T in the interval of 25-30℃ does not have a great impact on the yield of extract. However, in the diffusion phase, the influence is much more pronounced, that is, with an increase in T, the yield also increases. The results suggested that the optimum yield is 31.7 % achieved at the extraction time of 89.4 min, operating temperature of 45℃, and DIW content of $x_{\mathit{DIW}}=1$ used as a solvent.

3.3. ANN predictive modelling for the designed UAE process

The integrated approach to the analysis of the analysed separation system incorporates a successfully designed ANN predictive model. For that purpose, the MATLAB-Neural Network Toolbox was used to create the feed-forward backpropagation model trained using the Levenberg-Marquardt training algorithm. The ANN structure includes 3 neural layers – the input layer, the hidden (operative) layer, and the output layer. The input layer of neurons consists of 3 neurons – the actual independent experimental variables (extraction time, temperature, and solvent composition) whereas the output layer represents the following process response – extraction yield. The optimal number of neurons in the operative neural layer (10) was determined through the minimization of the MSE (Fig. 7). The utilized transfer function that was assigned to the operating neurons was the sigmoid transfer function. The optimal ANN structure is given in Fig. 8.

Fig. 7.

MSE vs number of hidden neurons in the ANN structure.

Fig. 8.

The optimal ANN structure.

The designed ANN model includes the neurons' weights and biases as presented in Table 7. Weights and biases for the designed ANN predictive model.

Table 7.

Weights and biases for the designed ANN predictive model.

Weight	Value	Weight	Value	Weight	Value	Bias	Value	Bias	Value
w11	0.25036	w16	−0.15244	w1	0.85399	b11	2.3798	b21	−1.482
w21	2.0233	w26	2.0447	w2	1.0881	b12	3.7545
w31	−3.6434	w36	1.6678	w3	−0.34406	b13	−0.97462
w12	0.064038	w17	0.80387	w4	0.02748	b14	−1.4988
w22	−4.0044	w27	−2.9455	w5	0.082147	b15	2.9257
w32	−3.4773	w37	−1.9977	w6	0.68272	b16	−1.2048
w13	−0.53754	w18	−3.1414	w7	−0.16786	b17	−4.9121
w23	0.52955	w28	3.24	w8	−0.016091	b18	−2.4417
w33	−2.3336	w38	−1.7426	w9	−0.56264	b19	−3.7132
w14	3.1948	w19	−3.0994	w10	1.1212	b110	−0.0065918
w24	2.07	w29	0.15259
w34	2.715	w39	0.29526
w15	−0.97652	w110	−0.10215
w25	−4.6295	w210	0.16472
w35	2.5509	w310	3.6707

Described ANN was created and trained using data matrices obtained through the adequate experimental plan and the input data matrix included the predefined values for the operating parameters (extraction time, operating temperature, and solvent composition).

The experimentally obtained data used as input data matrix for the ANN design is based on the independent operating parameters: extraction time data (0, 5, 10, 15, 30, 60, and 90 min), temperature (25, 35, and 45℃) and DIW content in the solvent system (0.04, 0.3, 0.45 and 1).

The designed ANN training procedure generated a high correlation between experimentally obtained values for extraction yield and model output values for the observed response.

The training state diagram for the created ANN is presented in Fig. 9. The ANN performance plot depicting the MSE for each of the assigned data matrices (training, validation, and test data set) is presented in Fig. 10.

Fig. 9.

Training state diagram for the designed ANN model.

Fig. 10.

ANN performance plot.

To achieve a successful ANN predictive model that enables high accuracy of predicted response–extraction yield and appropriate reproducibility of its performance, the input data matrix was divided into training, validation, and test data, representing 70 %, 15 %, and 15 %, respectively. The error histogram for the resulting ANN model is given in Fig. 11. The regression plot containing the correlation between experimental values and ANN model-generated response is presented in Fig. 12.

Fig. 11.

Error histogram for the training, validation, and test sets of input data.

Fig. 12.

Regression plot for the designed ANN model.

The interpretation of graphically presented results suggests that the created ANN was designed to produce a very high correlation between experimental results and output data (Fig. 12) with an overall r = 0.9992. The ANN predictive model has the required architecture and structural properties to generate highly accurate output data (extraction yield) for the analysed separation system.

3.4. Results of HPLC-DAD analysis of protodioscin from Tribulus terrestris L. extract

The retention time of protodioscin is about 12.3 min. A chromatographic column Supelco C18 (250x4.6 mm, 5 µm), the flow of 1 mL/min, injected sample intake of 30 µL, column temperature of 40℃, and detection wavelength of 200 nm was used for analysis. The identification of protodioscin in the extracts was made by comparing the chromatograms with those of the standard solutions and using the characteristic spectra. Three standard solutions of protodioscin were prepared using the protodioscin standard (analytical standard ≥ 98 %). Fig. 13a shows a chromatogram of a protodioscin standard solution with a concentration of 0.1 mg/mL. While, for an extract obtained under optimal maceration conditions, the chromatogram is shown in Fig. 13b, where the concentration of protodioscin is concentrated at 0.55 %. If the extract was obtained under optimal conditions of ultrasonic extraction, the chromatogram is shown in Fig. 13c, where the concentration of protodioscin is 0.59 %.

Fig. 13.

HPLC chromatogram of a protodioscin. (a)- standard solution with a concentration of 0.1 mg/mL, (b)- extract obtained using CSE, (c)- extract obtained using UAE.

3.5. Results of analysis of polyphenolic compounds from Tribulus terrestris L. extract

Bioactive components of Tribulus terrestris L. that have medical properties are saponins, flavonoids and phenolic compounds show numerous studies. The confirmation of antioxidant activity in Tribulus terrestris L. extracts through the action of flavonoids and polyphenol carboxylic acids presents compelling evidence for their therapeutic potential. Understanding these mechanisms could pave the way for new dietary recommendations or supplements aimed at oxidative stress reduction. Polyphenols are capable of scavenging hydroxyl (${\mathit{HO}}^{\bullet }$), peroxyl (${\mathit{RO}}_2^{\bullet }$), and superoxide ($O_2^{\bullet -}$) radicals [24].

Polyphenols were analysed at optimum points for each response procedure, CSE, and UAE. The results are shown in Table 8. Analysis of the UV–Vis results regarding the observed presence of polyphenolics confirms the presence of these compounds in the total extract.

Table 8.

Determination of total polyphenols.

Extraction technique	UV–VIS comments	Reporting value	Polyphenols [mg/L]
CSE	1.431	1.418	1297.87
UAE	1.130	1.117	1019.167

4. Conclusion(s)

The analysed experimental material, aerial parts (leaves, stems, flowers, and fruits) of Tribulus terrestis L. represents a reliable source of bioactive components and can be used to obtain natural product extract for the needs of the pharmaceutical, food or chemical industry. Investigated raw material proved to be a valuable source of protodioscin, a bioactive component that enhances the conversion of testosterone into potent dihydrotestosterone, thereby stimulating an increase in libido, red blood cell production from the bone marrow, and muscle development. Confirmed presence of stated compound ensures significant added value to the presented research and serves as an engineering basis for the consequential experimental planning to design and optimize an UAE process as an efficient extraction system.

The interpretation of the experimental data related to the CSE and UAE process kinetics analysed at predefined operating conditions − extraction time (0, 5, 10, 15, 30, 60, and 90 min), temperature (25, 35, and 45℃) and DIW content in the solvent system (0.04, 0.3, 0.45 and 1) suggests two phases. The initial phase of 0–15 min for the UAE based system and 0–30 min for the CSE, represents a rapid extraction phase whereas the second phase is interpreted as diffusion controlled phase (15–90 min for the UAE and 30–90 min for the CSE process).

The dependence on the intended yield of total $y_e$ extract from extraction time t and temperature T is mathematically described through an empirical polynomic second-row model and demonstrated through appropriately created 3D response surfaces of dependence $y_e=f(t,T)$. While observing kinetic curves and 3D responses to ultrasound extraction, it can be noted that the presence of ultrasonication leads to a significant increase in the monitored process response − extraction yield. The general conclusion established in accordance with experimental data regarding the UAE process produces the engineering justification for the selected separation system due to high extraction efficiency observed at optimal conditions. The UAE process resulted in high protodioscin content that was observed at reduced extraction time (15 min compared to 30 min at CSE) and mild operating temperatures. From the obtained results, we can conclude that the extracts obtained by both procedures have a large quantity of polyphenols, confirmed by the UV–Vis analytical method.

Obtained experimental results regarding the extraction yield as a function of the studied input variables (extraction time, operating temperature and water content in the solvent system) were successfully used to define appropriate input data matrix and create an ANN predictive model for the monitored output. Designed ANN was trained, validated and tested using randomly selected data points. Resulting high correlation of the ANN model outputs with experimentally produced responses, with an overall data correlation of $r=0.99919$, suggests that the network was efficiently structured with an optimal neural architecture of the input, hidden and output layers (3–10-1) and efficiently trained using the LM training algorithm.

CRediT authorship contribution statement

Stefan Kuvendziev: Writing – original draft, Visualization, Validation, Supervision, Software, Resources, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Isidora Dimitrievska: Writing – original draft, Resources, Methodology, Investigation, Data curation. Martin Stojchevski: Software, Investigation, Formal analysis, Data curation. Mirko Marinkovski: Writing – original draft, Software, Methodology.

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.