10.8
CiteScore
 
5.3
Impact Factor
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
Filter by Categories
Corrigendum
Current Issue
Editorial
Erratum
Full Length Article
Full lenth article
Letter to Editor
Original Article
Research article
Retraction notice
Review
Review Article
SPECIAL ISSUE: ENVIRONMENTAL CHEMISTRY
10.8
CiteScore
5.3
Impact Factor
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
Filter by Categories
Corrigendum
Current Issue
Editorial
Erratum
Full Length Article
Full lenth article
Letter to Editor
Original Article
Research article
Retraction notice
Review
Review Article
SPECIAL ISSUE: ENVIRONMENTAL CHEMISTRY
View/Download PDF

Translate this page into:

Original article
02 2022
:16;
104425
doi:
10.1016/j.arabjc.2022.104425

Optimization of Portulaca oleracea L. extract using response surface methodology and artificial neural network and characterization of bioactive compound by high-resolution mass spectroscopy

Department of Food Science and Biotechnology, Graduate School, Kyungpook National University, Daegu 41566, Republic of Korea
Food and Bio-Industry Research Institute, Inner Beauty/Antiaging Center, Kyungpook National University, Daegu 41566, Republic of Korea
Department of Chemistry, Kyungpook National University, Daegu 41566, Republic of Korea
Mass Spectrometry Converging Research Center and Green-Nano Materials Research Center, Daegu 41566, Republic of Korea

⁎Corresponding author at: Department of Food Science and Biotechnology, Kyungpook National University, Daegu 41566, Republic of Korea. sang@knu.ac.kr (Sang-Han Lee)

Disclaimer:
This article was originally published by Elsevier and was migrated to Scientific Scholar after the change of Publisher.

Peer review under responsibility of King Saud University.

Abstract

The well-known medicinal plant Portulaca oleracea L. (PO) is used as a traditional medicine and culinary herb to treat various diseases. Fatty acids, essential oils, and flavonoids were extracted from PO seeds and leaves using ultrasonic, microwave, and supercritical fluid extraction with RSM techniques. However, investigations on the secondary metabolites and antioxidant capabilities of the aerial part of PO (APO) are scarce. In order to extract polyphenols and antioxidants from APO as effectively as possible, this study used heat reflux extraction (HRE), response surface methodology (RSM), and artificial neural network (ANN) modeling. It also used high-resolution mass spectrometry to identify the APO secondary metabolite. A central-composite design (CCD) was used to establish the ideal ethanol content, extraction time, and extraction temperature to extract the highest polyphenolic compounds and antioxidant activity from APO. According to RSM, the highest amount of TPC (8.23 ± 1.06 mgGAE/g), TFC (43.12 ± 1.15 mgCAE/g), DPPH-scavenging activity (43.01 ± 1.25 % of inhibition) and FRAP (35.98 ± 0.19 µM ascorbic acid equivalent) were obtained at 60.0 % ethanol, 90.2 % time, and 50 °C. Statistical metrics such as the coefficient of determination (R2), root-mean-square error (RMSE), absolute average deviation (AAD), and standard error of prediction (SEP) revealed the ANN's superiority. Ninety-one (91) secondary metabolites, including phenolic, flavonoids, alkaloids, fatty acids, and terpenoids, were discovered using high-resolution mass spectrometry. In addition, 21 new phytoconstituents were identified for the first time in this plant. The results revealed a significant concentration of phytoconstituents, making it an excellent contender for the pharmaceutical and food industries.

Keywords

Antioxidant
Artificial neural network
Portulaca oleracea
Response surface methodology
Secondary metabolites
1

1 Introduction

Portulaca oleracea L. (PO) is a well-known medicinal plant used both as a traditional medicine and as an edible herb to treat various ailments. This herb is widely used in European folk medicine. Additionally, PO is mentioned in some pharmacopeias, such as the Ayurvedic Pharmacopoeia of India and the Pharmacopoeia of PR China (Iranshahy et al., 2017). Pharmacological investigations have demonstrated that PO has a wide range of biological effects, including anti-inflammatory, a bronchodilator, anti-microbial, antioxidant, and neuroprotective characteristics (Malek et al., 2004, Wang et al., 2007, Hozayen et al., 2011, Karimi et al., 2011, Du et al., 2017). Animal studies have demonstrated its hepatoprotective, antiulcerogenic, and antifertility benefits (Kumar et al., 2010, Nayaka et al., 2014, Eidi et al., 2015). Additionally, investigations on phytochemistry have revealed that this plant includes minerals, vitamins, fatty acids, flavonoids, alkaloids, and terpenoids (Sakai et al., 1996, Xiang et al., 2005, Yan et al., 2012, Petropoulos et al., 2016).

Numerous studies have demonstrated that the solvent concentration, incubation time, and temperature affect the effectiveness of polyphenol extraction, while some thermolabile bioactive substances may degrade during extraction (Saha et al., 2011). There are several extraction strategies for bioactive molecules, including reflux, soxhlet, microwave-aided, ultrasonicator-assisted, and supercritical fluid extraction (Pandey and Banik 2012). Heat reflux extraction has several benefits over more traditional extraction methods, (1) the solvent is replenished in the extraction; (2) the mass transfer driving force is more substantial, (3) the extraction takes less time, (4) the solvent is used less because it has already been used, and (5) the extraction yield is increased. This technology is a promising substitute for extracting bioactive natural compounds due to its benefits over traditional extraction methods (Gong et al., 2014, Ma et al., 2022).

Extraction is the initial and most crucial step in collecting and purifying bioactive chemicals from plant sources; yet, lengthy extraction times and low extraction efficiency limit these approaches (Samuel and Emovon 2018, Sedraoui et al., 2020). Analytical techniques were optimized using multivariate statistical methodologies to address this problem. Response surface methodology (RSM) combines mathematical and statistical methods that have proven effective in developing, improving, and optimizing complex processes (David Samuel et al., 2021). RSM provides a wealth of information and is more cost-effective because it reduces the required experiments. In addition, RSM assesses the simultaneous influence of several factors and anticipates the system's response to each new condition to find the optimal circumstances for the predicted response (Kusuma et al., 2021, Kusuma et al., 2022). Nevertheless, nonlinearity and inaccurate data are not handled precisely by RSM approaches. It has also been demonstrated that artificial neural networks (ANNs) are effective data-driven computational tools with the flexibility to capture complex and non-linear data (Okwu et al., 2020, Okwu et al., 2021). The operation of ANN as a prediction tool is similar to that of the human brain. The brain's neurons, basic processing units connected by networks and used to transmit messages between the neurons, served as the model's primary source of inspiration. The sigmoid function controls the network (Samuel and Okwu 2019, Okwu et al., 2021, Zadhossein et al., 2021). However, the black box learning technique associated with the ANN cannot be utilized to correlate input factors and output variables (Gupta and Sharma 2014). This problem is circumvented by incorporating an additional method, such as RSM, to analyze the interaction between the input and response variables. Hence, Combining RSM with ANN resulted in a more precise forecast (Samuel and Okwu 2019).

Mounting studies over the past few decades have shown how to extract fatty acids, essential oils, and flavonoids from PO seeds and leaves using ultrasonic, microwave, and supercritical fluid extraction with RSM methods (Stroescu et al., 2013, Wang et al., 2014, Sodeifian et al., 2018). Most of the research disclosed just process optimization. However, the authors did not compare the efficacy of predictive modeling with better methodologies, such as ANN, and there was a dearth of secondary metabolite profiling of the improved extracts. To the best of our knowledge, however, heat reflux extraction (HRE) using RSM and ANN was used for the first time in this study to increase the polyphenol content and antioxidant activity of the aerial portions of the Portulaca oleracea (APO). This study aimed to examine and improve extraction parameters, including extraction temperature and duration, as well as ethanol concentration, using the RSM central composite design (CCD) tool to obtain the highest polyphenolic content and antioxidant potential from APO. Additionally, for the first time, we have profiled the secondary metabolites of APO using high-resolution mass spectrometry analysis.

2

2 Materials and methods

2.1

2.1 Sample collection and preparation

Wild Portulaca oleracea L. was collected in September 2021 in Daegu, Korea. The Department of Food Science and Technology, Kyungpook National University, Daegu, Korea (voucher specimen # FT1005), identified the sample. Heat reflux extraction (HRE) was done in an oven with a condenser (Soxhlet water bath C-WBS-D6, Changshin Science, Seoul, Korea). Dry powder samples (10.0 g) were extracted using 100 ml of solvent following the instructions in supplemental Table 1. The extracted materials were filtered on Whatman No. 1 filter paper (Schleicher & Schuell, Keene, New Hampshire) and then dried in a freeze drier (Il-shin Biobase, Goyang, Korea). The APO extract was kept at −20 °C for the ensuing investigations.

Table 1 Central composite design (CCD) for independent variables and corresponding response values (experimental).
Run Independent variables Responses
(X1) (X2) (X3) TPC (Y1) TFC (Y2) DPPH (Y3) FRAP (Y4)
1 50 140 50 5.09 ± 0.68 21.20 ± 0.73 20.87 ± 0.72 19.03 ± 0.05
2 50 40 50 3.89 ± 0.83 19.42 ± 0.52 15.36 ± 0.39 16.36 ± 0.02
3 50 90 50 8.12 ± 0.34 41.89 ± 0.25 41.75 ± 0.37 35.19 ± 0.16
4 50 90 30 3.15 ± 0.26 29.07 ± 0.65 9.18 ± 0.24 12.04 ± 0.10
5 0 90 50 1.25 ± 0.41 19.76 ± 0.32 0.66 ± 1.00 3.04 ± 0.07
6 75 60 60 6.58 ± 0.82 29.09 ± 0.10 27.21 ± 0.21 26.02 ± 0.08
7 50 90 50 7.61 ± 1.02 41.75 ± 0.56 41.93 ± 0.25 35.19 ± 0.06
8 75 120 40 5.06 ± 0.49 31.02 ± 0.26 25.07 ± 0.45 23.12 ± 0.02
9 75 60 40 5.03 ± 0.42 32.02 ± 0.95 22.70 ± 0.54 22.26 ± 0.04
10 25 120 60 5.06 ± 0.23 25.75 ± 0.35 20.08 ± 0.98 17.01 ± 0.08
11 50 90 50 7.59 ± 0.62 43.20 ± 0.26 40.56 ± 0.10 35.19 ± 0.04
12 100 90 50 5.75 ± 0.06 31.50 ± 0.33 30.25 ± 1.02 25.93 ± 0.02
13 50 90 50 7.49 ± 0.04 43.02 ± 0.53 40.05 ± 0.56 35.17 ± 0.06
14 75 120 60 6.84 ± 0.08 29.02 ± 0.35 29.88 ± 0.46 27.96 ± 0.16
15 25 60 60 3.17 ± 0.06 24.52 ± 0.15 14.02 ± 0.29 15.33 ± 0.13
16 50 90 70 5.53 ± 0.24 31.01 ± 0.60 22.53 ± 0.37 21.33 ± 0.19
17 25 60 40 2.81 ± 0.68 23.13 ± 0.72 7.01 ± 0.19 10.46 ± 0.15
18 50 90 50 7.98 ± 0.68 43.11 ± 0.72 41.89 ± 0.73 35.09 ± 0.07
19 50 90 50 8.12 ± 0.83 43.23 ± 0.39 41.05 ± 0.52 36.09 ± 0.05
20 25 120 40 2.82 ± 0.34 21.03 ± 0.37 8.51 ± 0.25 9.19 ± 0.16

X1. Ethanol concentration (%); X2. time (min); X3. temperature (°C); TPC. total phenolic content (mg gallic acid equivalent/g dry weight extract); TFC. total flavonoid content (mg catechin equivalent/g dry weight extract); DPPH. DPPH radical scavenging activity (% inhibition); FRAP. ferric reducing antioxidant power (µM ascorbic acid equivalent).

2.2

2.2 Total phenolic content (TPC), total flavonoid content (TFC) and antioxidant activities

The TPC and TFC of APO extracts were assessed using the Folin-Ciocalteu assay and the aluminum chloride colorimetric method, respectively (Alam et al., 2017). The corresponding regression equations for the calibration curves were used to determine the TPC (y = 0.0582x + 0.0038; r2 = 0.9955) and TFC (y = 0.059x + 0.0081; r2 = 0.9879). The gallic acid equivalent (mg)/dry weight sample (g) and catechin equivalent (mg)/dry weight sample (g) were used as the units of measurement for the TPC and TFC, respectively. DPPH-radical scavenging test and ferric reducing antioxidant power (FRAP) assay were used to assess the antioxidant properties of APO extracts (Alam et al., 2021).

2.3

2.3 RSM design and extraction process

The RSM model was designed to extract phenolic chemicals from APO using ethanol concentration (X1), extraction duration (X2), and temperature (X3) as independent process factors. Respondent factors included TPC, TFC, DPPH-scavenging activity and FRAP (Y1-Y4, respectively). A three-component, five-layer CCD was employed for the extractions (supplementary data Table S1). The CCD is widely utilized due to its adaptability. The early experimental results from a factorial design can be used in a CCD that only makes the axial points, eliminating resource waste. Nonetheless, the most distinctive parameters are the axial points (−α, α). These locations are outside the minimum and maximum limits of the factors, guaranteeing the response surface's curvature and allowing the construction of ideal conditions. In contrast to the BBD, a CCD can distinguish between axial point estimates based on orthogonal and rotational characteristics. An orthogonal design allows for an independent examination of the primary, interaction, and quadratic effects, simplifying the interpretation of the results. The second-order polynomial model equation (Eq.1) describes the link between independent factors and replies.

(1)
Y = β 0 + i = 1 n β i X i + i = 1 n β ii X ii 2 + i n - 1 j n β ij X ij where Y, Xi and Xj as well as β0 represents the response variable, independent coded variables and the constant coefficient, respectively, while βi, βii, and βij represent the coefficients of linear, quadratic, and interaction effects, respectively.

The model's adequacy was assessed using the determination coefficient (R2), the adjusted determination coefficient (Adj.R2), and the lack of fit test. The F-value (p < 0.05) was significant. Three-dimensional (3D) surface plots showed each factor's effect on response value. The RSM analysis and multiple linear regression were carried out using Design Expert 11 (Stat-Ease, Minneapolis, Minnesota, USA).

2.4

2.4 Artificial neural networks (ANN) modeling

The ANN modelling was systematically conducted by using the dataset presented in Table 1. MATLAB R2020a software (The Neural Network Toolbox, Inc., USA) was employed to create the ANN model. The ANN design consists of data collection; model development using different functions and algorithms; configuration of the model; weights and biases initialization; training, testing, and validation of the model. The MLP (multilayer perceptron network) topology consists of input, hidden, and output layers. Independent variables (X1, X2, and X3) were used as input vectors, and four responses (Y1, Y2, Y3, and Y4) were applied as target vectors (Fig. 1A). The data were divided into three subsets, where 70, 15, and 15 % of the whole data points were utilized for training, validation, and testing, respectively. In the training step, the feed-forward network and cascade feed-forward network with the Broyden-Fletcher-Goldfarb-Shanno algorithm (BFGS) and Levenberg-Marquardt back-propagation algorithm were used to lessen the mean square error (MSE). The MSE was calculated using Eq. (2)

(2)
MSE = 1 N N i = 1 Y ANN - Y Exp 2 where YExp is the experimental outcomes, N represents to sample number and YANN is the predicted value. A hyperbolic tangent sigmoid transfer function was used for pattern recognition and network modeling using Eq. (3)
(3)
f x = t a n s i g n = 2 1 + e - 2 x - 1
ANN model topology. The best ANN model in terms of architecture (A), network training curves for trained subsets with epoch numbers for TPC (B), TFC (C), DPPH (D) and FRAP (E) using the MATLAB software.
Fig. 1
ANN model topology. The best ANN model in terms of architecture (A), network training curves for trained subsets with epoch numbers for TPC (B), TFC (C), DPPH (D) and FRAP (E) using the MATLAB software.

2.5

2.5 Comparison of the RSM and ANN models' predictive abilities

To evaluate the estimation skills of RSM and ANN, several statistical metrics such as R2, RMSE, AAD, and SEP were calculated using the following equations.

(4)
R 2 = 1 - i = 1 n ( Y p - Y e ) 2 i = 1 n ( Y m - Y e ) 2
(5)
RMSE = i = 1 n ( Y p - Y e ) 2 n
(6)
AAD = i = 1 n Y p - Y e / Y e n x 100
(7)
SEP = RMSE Y m x 100
where Yp is the predicted response; Ye is the observed response; Ym is the average response variable; n is the number of experiments.

2.6

2.6 Model validation

The optimal extraction parameters were determined using response surface and desirability function analysis. A series of three experiments was conducted under ideal conditions to ensure the model's accuracy, with the average experimental results compared to predictions. In addition, the electrospray ionization mass spectrometry (ESI-MS)/MS profiles of bioactive compounds were identified under optimum conditions.

2.7

2.7 Identification of bioactive compounds by ESI-MS/MS analysis

The negative (-) mode ESI-MS was performed on a Q-Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific Inc., San Jose, CA, USA). Immersing the sample in the ESI source required a 500 L graded syringe (Hamilton Company Inc., Reno, NV, USA) and a 15 L/min syringe pump (Model 11, Harvard, Holliston, MA, USA). The normal negative mode ESI-MS conditions were as follows: mass resolution of 140,000 (full width at half maximum, FWHM), sheath gas flow rate of 5, seep gas flow rate of 0, auxiliary gas flow rate of 0, spray voltage of 4.20 kV, capillary temperature of 320 °C, S-lens Rf level, and automatic gain control of 5 E 6. The MS/MS investigations used the same apparatus with three stepwise normalized collision energies (10, 20, and 30) (Alam et al., 2021). Mass spectrum data was processed using the Xcalibur 3.1 with Foundation 3.1 (Thermo Fisher Scientific Inc., Rockford, IL, USA). The compounds were probably found by comparing the calculated (exact) masses of deprotonated (M−H) adducts with the m/z values and ESI-MS/MS fragmentation patterns from the in-house MS/MS database and online databases like FooDB (Naveja et al., 2018), METLIN (Guijas et al., 2018), CFM-ID 4.0 (Wang et al., 2021). The ChemDraw Professional 15.0 (PerkinElmer, Waltham, MA, USA) was used to draw the chemical structure.

2.8

2.8 Statistical analysis

All data were reported as the mean ± standard deviation of at least three independent experiments (n = 3), each with three sample replicates. Differences were considered significant at p < 0.001, p < 0.01, and p < 0.05.

3

3 Results and discussion

3.1

3.1 Fitting of the RSM and ANN models

For each extraction circumstance, Table 1 describes the experimental settings and conclusions. All response variables were transformed into second-order quadratic polynomial equations to account for changes in answers as a function of extraction factors. ANOVA was used to determine whether the fitted second-order quadratic model equations were statistically significant. The regression coefficient (β), adjusted correlation factor (R2), coefficient of variation (CV), and adequate precision were used to describe how well the model fit (Table 2). The nonsignificant terms (p > 0.05) were taken out of the models to improve the fit and predictions. We used the p-values to figure out how vital each coefficient was. When the p-values were<0.05, 0.01, and 0.001, the model terms were significant, very significant, and strikingly significant, respectively.

Table 2 ANOVA for quadratic model.
ANOVA for quadratic model for TPC
Source RC SS DF MS F-value p-value
Model 81.54 9 9.06 102.01 < 0.0001 Significant
Intercept 7.82
Linear terms
X1 1.17 21.74 1 21.74 244.77 < 0.0001 Significant
X2 0.3091 1.30 1 1.30 14.58 0.0034 Significant
X3 0.6681 7.14 1 7.14 80.42 < 0.0001 Significant
Interaction terms
X1X2 −0.2012 0.3240 1 0.3240 3.65 0.0852
X1X3 0.0912 0.0666 1 0.0666 0.7500 0.4068
X2X3 0.2637 0.5565 1 0.5565 6.27 0.0313 Significant
Quadratic terms
X12 −1.08 29.78 1 29.78 335.27 < 0.0001 Significant
X22 −1.20 19.41 1 19.41 218.56 < 0.0001 Significant
X32 −0.8696 19.32 1 19.32 217.52 < 0.0001 Significant
Lack of Fit 0.4767 5 0.0953 1.16 0.4379 not significant
Pure error 0.4115 5 0.0823
R2 0.9892
Adjusted R2 0.9795
Adeq Precision 31.5543
C.V. % 5.47
ANOVA for quadratic model for TFC
Model 1414.37 9 157.15 249.82 < 0.0001 Significant
Intercept 42.67
Linear terms
X1 3.14 157.50 1 157.50 250.38 < 0.0001 Significant
X2 0.0757 0.0778 1 0.0778 0.1236 0.7324
X3 0.2375 0.9025 1 0.9025 1.43 0.2586
Interaction terms
X1X2 −0.0250 0.0050 1 0.0050 0.0079 0.9307
X1X3 −1.38 15.24 1 15.24 24.22 0.0006 Significant
X2X3 0.5325 2.27 1 2.27 3.61 0.0868
Quadratic terms
X12 −4.30 472.67 1 472.67 751.40 < 0.0001 Significant
X22 −8.14 894.88 1 894.88 1422.58 < 0.0001 Significant
X32 −3.12 248.70 1 248.70 395.35 < 0.0001 Significant
Lack of Fit 3.93 5 0.7861 1.67 0.2946 Non-Significant
Pure error 2.36 5 0.4720
R2 0.9956
Adjusted R2 0.9916
Adeq Precision 41.8660
C.V. % 2.54
ANOVA for quadratic model for DPPH
Model 3330.37 9 370.04 412.69 < 0.0001 Significant
Intercept 41.13
Linear terms
X1 7.15 818.25 1 818.25 912.55 < 0.0001 Significant
X2 1.61 35.01 1 35.01 39.04 < 0.0001 Significant
X3 3.41 186.32 1 186.32 207.80 < 0.0001 Significant
Interaction terms
X1X2 −0.3150 0.7938 1 0.7938 0.8853 0.3689
X1X3 −1.16 10.72 1 10.72 11.95 0.0061 Significant
X2X3 0.6075 2.95 1 2.95 3.29 0.0997
Quadratic terms
X12 −6.52 1085.41 1 1085.41 1210.51 < 0.0001 Significant
X22 −8.49 974.81 1 974.81 1087.16 < 0.0001 Significant
X32 −6.42 1052.36 1 1052.36 1173.65 < 0.0001 Significant
Lack of Fit 5.90 5 1.18 1.92 0.2449 Non-significant
Pure error 3.07 5 0.6132
R2 0.9973
Adjusted R2 0.9949
Adeq Precision 60.2976
C.V. % 3.78
ANOVA for quadratic model for FRAP
Model 1975.64 9 219.52 900.96 < 0.0001 Significant
Intercept 35.30
Linear terms
X1 5.82 542.31 1 542.31 2225.80 < 0.0001 Significant
X2 0.5651 4.33 1 4.33 17.77 0.0018 Significant
X3 2.49 99.35 1 99.35 407.77 < 0.0001 Significant
Interaction terms
X1X2 0.2988 0.7140 1 0.7140 2.93 0.1177
X1X3 −0.5112 2.09 1 2.09 8.58 0.0151 Significant
X2X3 0.5038 2.03 1 2.03 8.33 0.0162 Significant
Quadratic terms
X12 −5.23 698.20 1 698.20 2865.65 < 0.0001 Significant
X22 −6.39 551.56 1 551.56 2263.76 < 0.0001 Significant
X32 −4.68 559.01 1 559.01 2294.37 < 0.0001 Significant
Lack of Fit 1.72 5 0.3435 2.39 0.1806 Non-significant
Pure error 0.7190 5 0.1438
R2 0.9988
Adjusted R2 0.9977
Adeq Precision 93.2690
C.V. % 2.14

RC. Regression coefficient; SS. sum of squares; MS. mean square.

Table 2 shows that model terms are significant (p < 0.0001). The R2 values (0.9892–0.9988) of the built regression models indicate a high level of statistical significance. The appropriate precision indicates a signal-to-noise ratio, and > 4 is ideal (Alam et al., 2022). The ratio ranged between 31.55 and 93.26, showing a significant signal and suitability for this method. The coefficient of variation (CV) measures the repeatability of a model, and the range of 2.14 to 5.47 indicates that the model is reproducible. Multiple linear regression equations were used to create 3D surfaces and contour plots to show independent variable interactions (Fig. 2A-D).

The three-dimensional (3D) response surface plots of APO extraction for TPC (A), TFC (B), DPPH-radical scavenging activity (C), and FRAP (D) for ethanol concentration, time, and temperature as a function of key interaction factors for RSM.
Fig. 2
The three-dimensional (3D) response surface plots of APO extraction for TPC (A), TFC (B), DPPH-radical scavenging activity (C), and FRAP (D) for ethanol concentration, time, and temperature as a function of key interaction factors for RSM.

Mounting evidence revealed that ANN modeling is superior and more sophisticated than RSM, and ANNs are a feasible alternative to RSM for complicated nonlinear multivariate modeling. ANNs are more exact than RSM at fitting experimental responses, predicting, and modeling biological processes (Huang et al., 2017). ANN modeling was used to verify the experimental values. The trained ANN model's predicted values are in Table S2. The ANN predicts nonlinear relationships between extraction parameters (X1, X2, and X3) and response variables (Y1, Y2, Y3, and Y4). The ANN model predicted values that were pretty close to the actual values, proving its accuracy. By comparing network training and testing errors, the hit-and-try strategy modified the number of hidden layer neurons. The experiment investigated the lowest practicable error between training and testing and the minimal number of epochs to prevent model overfitting; the results were consistent with earlier efforts (Choi et al., 2022). The Levenberg-Marquardt approach produced the best validation result for all dependent variables Y1, Y2, Y3, and Y4 (Fig. 1B-E.

3.2

3.2 Comparison of the RSM and ANN models' predictive abilities

Both the RSM and ANN models' prediction and estimation skills were examined. Comparative similarity plots were utilized to examine the ANN model's four target response predictions (Y1, Y2, Y3, and Y4). In terms of fitting experimental data to all target responses, the ANN model was more accurate, precise, and assessable than the RSM model (supplementary data Table S2). The RSM model had a larger discrepancy between projected and actual data, whereas the ANN model's residuals remained steady.

To compare RSM with ANN, R2, RMSE, AAD, and SEP were calculated (Table 3). A better model has lower RMSE, AAD, and SEP while higher R2. R2 values of the trained ANN model were greater than those of the RSM model, suggesting the ANN model's superiority in predicting all four dependent variables. The AAD gauges the deviation between projected and actual data, while RMSE shows model fit. The ANN outperformed RSM by having lower AAD and RMSE values. The ANN model also showed low SEP values, which ranged from 0.0813 to 0.3126. The ANN model is more predictive than the RSM model because it can approximate nonlinear systems, while the RSM model requires second-order polynomial regression. The ANN model is also unaffected by experimental design and calculates several replies in a single run, while the RSM model takes multiple runs for multi-response optimization (Dadgar et al., 2015).

Table 3 Comparison of the prediction abilities of the RSM and ANN models.
Parameters TPC TFC DPPH FRAP
RSM ANN RSM ANN RSM ANN RSM ANN
R2 0.9892 0.9976 0.9956 0.9997 0.9973 0.9985 0.9988 0.9995
RMSE 0.1299 0.0753 2.3413 1.7888 1.6655 1.0867 1.3608 1.1492
AAD (%) 2.2909 1.1408 6.6496 3.0372 28.8295 14.4468 4.9237 3.1221
SEP (%) 0.1401 0.0813 0.4091 0.3126 0.4675 0.3050 0.3567 0.3012

R2. correlation coefficient; RMSE. root-mean-square error; AAD. absolute average deviation; SEP. standard error of prediction.

3.3

3.3 Influence of HRE parameters on TPC and TFC

In APO extracts, TPC and TFC contents ranged from 1.25 ± 0.41 to 8.12 ± 0.34 mgGAE/g and 19.76 ± 0.32 to 43.23 ± 0.91 mgCAE/g, respectively (Table 1). Both the TPC and TFC exhibited a substantial linear influence of X1 and the quadratic component of (X12), (X22), and (X32) (supplementary data Figure S1). The second-order polynomial equations in eqs. (8) and (9) illustrate the relationships between TPC, TFC, and their variables.

(8)
T P C Y 1 = 7.82 + 1.17 X 1 + 0.3091 X 2 + 0.6681 X 3 - 1.08 X 1 2 - 1.20 X 2 2 - 0.8696 X 3 2 - 0.2012 X 1 X 2 + 0.0912 X 1 X 3 + 0.2637 X 2 X 3
(9)
T F C Y 2 = 42.67 + 3.14 X 1 + 0.0757 X 2 + 0.2375 X 3 - 4.30 X 1 2 - 8.14 X 2 2 - 3.12 X 3 2 - 0.0250 X 1 X 2 - 1.38 X 1 X 3 + 0.5325 X 2 X 3

The TPC and TFC showed nonsignificant lack of fit values (F = 1.16 and 1.67, respectively) showing the model accurately predicted R2 = 0. 9892 (TPC) and 0.9956 (TFC) and Adj.R2 = 0. 9795 (TPC) and 0.9949 (TFC) (Table 2). The RSM model accurately predicted the parameter impacts on TPC and TFC of the APO extract. As depicted in Fig. 2(A, B), at 50 °C, 50 % ethanol produced the most TPC and TFC in 90 min. Previous studies revealed that medium-concentration ethanol may make the solvent more polar, dissolving more polar and moderately polar phenolic compounds (Sedraoui et al., 2020). Moderate ethanol in water can affect the architecture and structure of membrane phospholipids. This affects plant cell penetrability, allowing higher polyphenol extraction and diffusion (Gurtovenko and Anwar 2009). Experiments in a prior comparison investigation found that extraction of phenolic compounds from green tea leaves under high hydrostatic pressure increased with ethanol in the solvent; peaked at 50 % ethanol and fell after that (Xi and Wang 2013).

3.4

3.4 Effect of HRE parameters on the in vitro antioxidant capacity (AC)

A linear significant influence of ethanol content (X1), a quadratic effect of concentration (X1), time (X2) and temperature (X3) as well as interaction between concentration and temperature (X1X3) (supplementary data Figure S1) on antioxidant activity were found using DPPH radical scavenging activity and FRAP analyses. Eqs. (10) and (11) display the fitted second-order polynomial equations for DPPH (% inhibition) and FRAP (ascorbic acid equivalent μM):

(10)
D P P H Y 3 = 41.13 + 7.15 X 1 + 1.61 X 2 + 3.41 X 3 - 6.52 X 1 2 - 8.49 X 2 2 - 6.42 X 3 2 - 0.3150 X 1 X 2 - 1.16 X 1 X 3 + 0.6075 X 2 X 3
(11)
F R A P Y 4 = 35.30 + 5.82 X 1 + 0.5651 X 2 + 2.49 X 3 - 5.23 X 1 2 - 6.39 X 2 2 - 4.68 X 3 2 + 0.2988 X 1 X 2 - 0.5112 X 1 X 3 + 0.5038 X 2 X 3

The AC values ranged from 0.66 ± 1.00 % to 41.89 ± 0.22 % inhibition of DPPH and from 3.04 ± 0.07 to 36.09 ± 0.34 μM ascorbic acid equivalent (Table 1). The ANOVA results show that the data fitted the model results for DPPH (R2 = 0.9973 and Adj. R2 = 0.9949) and FRAP response (R2 = 0.9988 and Adj. R2 = 0.9977), and the lack of fit was nonsignificant (F = 1.92 for DPPH and 2.39 for FRAP) (Table 2). As depicted in Fig. 2(C, D), at 50 °C, 50 % ethanol produced the highest DPPH inhibition and FRAP value in 90 min. This indicates that the capacity for electron and proton donation improves with increasing amounts of the organic solvent. This outcome is in line with the earlier discovery for TFC that maximum extraction calls for 75 percent ethanol (Do et al., 2014). The extraction of considerable polyphenolics from APO, both in terms of quality and quantity, is made possible by raising the ethanol concentration. There is growing evidence that ethanol concentrations affect antioxidant activity and polyphenolic compound quality and amount (Zhu et al., 2011, Do et al., 2014).

3.5

3.5 Model validation

The desirability function optimizes TPC, TFC, DPPH, and FRAP simultaneously. Derringer's desirability function was used to anticipate the parameters, allowing a multivariate analysis to find the best level for all replies in a single extraction. In this study, the following conditions (X1, 60 %), (X2: 90.5 min), and (X3, 50 °C), was used to achieve the maximal overall desirability D = 0.999 (on a scale of 0 to 1). Under these optimal conditions, the predicted values for TPC, TFC, percentage inhibition of DPPH, and µM ascorbic acid equivalent FRAP are 8.12 mgGAE/g, 43.23 mgCAE/g, and 42.98 %, and 36.81, respectively. To verify the sufficiency of the model equations, a duplicate experiment was conducted in the optimal conditions predicted by Derringer's desire model. The following results were obtained: TPC = 8.23 ± 1.06 mgGAE/g, TFC = 43.12 ± 1.15 mgCAE/g, % inhibition of DPPH = 43.01 ± 1.25 %, and µM ascorbic acid equivalent FRAP = 35.98 ± 0.19. The model efficiently optimized the common extraction parameters for all responses, as evidenced by the good agreement between experimental and expected values (supplementary data Table S3).

Additionally, comparison research between this study and earlier studies was carried out to confirm the high extractability of hydro-alcoholic solvents for polyphenols and the antioxidant properties of APO. The hydro-alcoholic solvent had more TPC and DPPH scavenging action than other solvents, as indicated in Table 4. When compared to alternative solvents, which were typically utilized in earlier studies, these comparisons showed that the HRE technique using hydro-alcoholic solvent was a high-efficiency technique.

3.6

3.6 Identification of secondary metabolites in APO by high-resolution mass spectroscopy

The ESI-MS/MS in negative ionization techniques detected secondary metabolites in APO extracts. Table 4 shows that 93 compounds were identified in negative mode utilizing MSn data from the precursor ion mass, fragments, recognized fragmentation patterns for the provided classes of compounds, and neutral mass loss, as well as literature and online database searches. The confidence level determined the significance of these results. Level 2 shows the likely structure of the detected substance, whereas Level 3 denotes a speculative candidate. (Schymanski et al., 2014).

Table 4 Comparative study of the polyphenolic content and antioxidant activity of APO with prior study report.
Methods Solvent TPC (mgGAE/g) DPPH IC50 (mg/ml) Ref.
HRE HM 8.23 1.01 Present study
HRE M 4.78 1.78
HRE M 6.98 2.52
HRE E 3.60 3.56
HRE W 4.41 2.35

HM. 60% methanol; M. 100% methanol; E. 100% ethanol and W. 100% aqueous solvent.

3.6.1

3.6.1 Phenolic acids

A phenolic acid can lose methyl (15 Da), hydroxyl (18 Da), or carboxyl (44 Da) to form a fragment ion. Fragmentation of a phenolic acid glycoside begins with cleavage of the glycosidic bond to give phenolic acid m/z and sugar loss (neutral mass loss of 162 Da) (Choi et al., 2022). In addition, hydroxycinnamic acid conjugates yield quinate (m/z 191) by loss of the acyl group and dehydrated quinate (m/z 173), coumarate (m/z 163), caffeate (m/z 179), ferulate (m/z 193) and sinapate (m/z 223) through β-elimination of a carboxylic acid (Jaiswal et al., 2010, Parveen et al., 2011). Thus, compounds 1–3, 5, 7, 8, 10–16, and 20 were tentatively identified as hydroxy benzoic acid, coumaric acid, ferulic acid, caffeic acid phenethyl ester, ellagic acid, protocatechuic acid glucoside, coumaroylquinic acid, caffeic acid hexoside, ferulylshikimic acid, ferulic acid hexoside, syringoylquinic acid, caffeic acid derivatives, feruloyl galactaric acid and hexosyl caffeoyl hexose (Islam et al., 2020, Alam et al., 2021, Fernández-Poyatos et al., 2021, Choi et al., 2022). In addition, by comparing the fragmentation patterns to those previously published in the literature, compound 4, 6, 17, 18 and 21 was identified as maclurin (C13H10O6), uralenneoside (C12H14O8), picraquassioside A (C18H22O10), methylpicraquassioside A (C19H24O10) and arillatose B (C22H29O14), respectively which yielded a precursor ion [M−H]¯ at m/z 261.0401, 285.0612, 397.1142, 411.1302, and 517.1563, respectively (Berardini et al., 2004, Abdelrahman et al., 2017, Llorach et al., 2019, Tang et al., 2020). All of these substances were intriguingly discovered for the first time in APO. Furthermore, compound 9 generated a monoisotopic ion [M−H]¯ at m/z 333.0618 and produced fragment ions at m/z 289.07 by lose of carboxyl (-44 Da) group and at m/z 271.06 through successive loss of H2O. It also yielded a characteristic ion at m/z 167.03 by cleavage of ether bond between C7-C9 and tentatively confirmed as berceloneic acid B, which has been identified as first time in APO (Fig. 3A). In addition, compound 19 was tentatively identified as paederol B with molecular formula (C20H28O12), generated a deprotonated ion at m/z 459.1511 and yielded the following fragment ions: m/z at 399.12 ([M−H−61 Da]), 341.08 ([M−H−C4H10O3−CH3]), 281.08 ([M−H−178 Da]) and 193.05 (ferulate ion) through β-elimination of a carboxylic acid (Fig. 3B). This compound has also been first time identified in APO.

Possible mass fragmentation pattern of (A) barceloneic acid B, (B) paederol B and (C) Oleracein E.
Fig. 3
Possible mass fragmentation pattern of (A) barceloneic acid B, (B) paederol B and (C) Oleracein E.

3.6.2

3.6.2 Flavonoids

According to a prior study, each subgroup of flavonoids exhibits a distinct fragmentation pattern during mass analysis. The most common fragmentation of flavonoids is the cleavage of the C-ring bonds (retro-Diels-Alder, i.e., RDA mechanism), which forms ions with the A- or B-ring and a portion of the C-ring. There may also be significant losses of tiny neutral molecules, such as CO (28 Da), C2H2O (42 Da), COO (44 Da), 2CO (56 Da), CO + COO (72 Da), and 3CO (84 Da). A unique ion [M−H−CH3], distinguished by the loss of 15 Da, is also present in methylated flavonoids (Alam et al., 2021, Choi et al., 2022). Flavonoids typically undergo glycosylation. O-glycosides, C-glycosides, and OC—glycosides are formed when the glycoside residues are connected to the O and C atoms of the flavonoids. Hexoses (162 Da), deoxyhexoses (146 Da), pentoses (132 Da), and an aglycone ion are the neutral species that result from the usual fragmentation of O-glycosides. As opposed to this, C-glucosides result in a series of fragments due to the cleavage of the C—C bonds with the sugar moiety. Some examples of these fragments include [M−H−60], [M−H−90], and [M−H−120], which are used as the distinctive diagnostic ions of glycone (Vukics and Guttman 2010, Kachlicki et al., 2016). Compounds 22–29 were identified as eriodictyol, catechin, dactylorhin C, taxifolin-7-sulfate, diadzin, cajanone, phenethylrutinoside, and kaempferol glucoside respectively, based on the similarities noticed in their fragmentation behaviors and the behaviors mentioned in the literature (Alam et al., 2021, Islam et al., 2021, Choi et al., 2022).

3.6.3

3.6.3 Alkaloids

APO contains oleraceins, a type of indoline amide glycosides. Many of these compounds are glucosylated and have 5,6-dihydroxyindoline-2-carboxylic acid N-acylated with cinnamic acid derivatives such as hydroxybenzoyl, coumaroyl, caffeoyl, feruloyl, and sinapoyl. The following fragment ions indicate the types of hydroxy cinnamic acid N-linked to the indoline core, at m/z 340.08, 356.07, and 370.09 for coumaroyl, caffeoyl and feruloyl, respectively. Furthermore, oleraceins also yielded characteristics ions by neutral loss of CO (28 Da), COO (44 Da), hydroxybenzoyl (120 Da), coumaroyl (146 Da), caffeoyl (162 Da), feruloyl (176 Da), sinapoyl (206 Da), glucosyl (162 Da), double glucosyl (324 Da) and triple glucosyl (486 Da). The first oleracein compound found in this study was oleracein E (compound 30), which has the chemical formula C12H13NO3 and produces a deprotonated ion at m/z 218.0833. The loss of an HCHO molecule and an HCOCH = CH2 molecule, respectively, resulted in the production of the fragment ions with m/z 188.07 and 162.05 respectively. It also undergoes i-cleavage of the middle ring's phenyl and C—C connections, producing fragment ions with m/z values of 135.04 and 121.02, respectively (Fig. 3C). Moreover, compounds 36–44 were identified as oleracein U, A, B, C, I, P, N/S, L/J, and O, respectively, based on commonalities seen between their fragmentation behaviors and those reported in the literature (Voynikov et al., 2021).

Furthermore, hydroxycinnamic acid amide yielded the base ion at m/z 147.04 (coumaroyl), m/z 163.04 (caffeoyl), m/z 177.05 (feruloyl) and m/z 207.06 (sinapoyl) by elimination of tyramine (137 Da) moiety. Further fragmentation was generated by the loss of a molecule of CO from the base peak. In addition, the tyramine moiety was further loss of NH3 to yield ion at m/z 121 (Liu et al., 2021). On the basis of the above fragmentation behavior, compound 32–36 was identified as feruloylglycine, coumaroyltyramine, caffeoyltyramine, feruloyltyramine and feruloyloctopamine, respectively (Zhou et al., 2015).

3.6.4

3.6.4 Carboxylic acids, fatty acids and amino acids

From comparisons of the mass and the fragmentation behaviors of the precursor ion based on mass spectroscopic analysis reported in literature and various online databases, compounds 45–49, and 52–64 were identified as carboxylic acids, and fatty acids, respectively (Table 5). In addition, compound 67–74 were characterized as amino acids (Guijas et al., 2018, Naveja et al., 2018, Nematallah et al., 2018, Ruan et al., 2019, Islam et al., 2020, Alam et al., 2021, Najm et al., 2021, Wang et al., 2021). Furthermore, molecule 50, 51, 65, and 66 were recognized as jasmonic acid and its derivatives (tuberonic acid, tuberonic acid glucoside, and methyl tuberonic acid glucoside) based on the mass fragmentation behavior described by Quirantes-Pine et al (Quirantes-Piné et al., 2010) (Fig. 4A). The jasmonic acid and its all derivaties are discovered for the first time in APO.

Table 5 List of possible identified compounds of the optimized extract of Potulaca oleracea by ESI-MS/MS.
No. Compound name EF OM (m/z)- CM (m/z)- MS/MS (negative mode) CL
Phenolic acid 1 4-Hydroxy benzoic acid C7H6O3 137.0253 137.0244 119.03, 93.01 2
2 Coumaric acid C9H8O3 163.0402 163.0401 119.04, 2
3 Ferulic acid C10H10O4 193.0521 193.0506 179.03, 149.06, 135.04 2
4 Maclurin# C13H10O6 261.0401 261.0405 151.00, 107.01 3
5 Caffeic acid phenethyl ester C17H16O4 283.0967 283.097 265.08, 239.07, 179.03, 163.04, 135.04 3
6 Uralenneoside# C12H14O8 285.0612 285.0616 153.01, 109.02 3
7 Ellagic acid C14H6O8 300.9893 300.9984 283.99, 245.00, 229.01, 200.01, 185.02 2
8 Protocatechuic acid glucoside C13H16O9 315.072 315.0716 162.02, 153.01 2
9 Barceloneic acid B# C16H14O8 333.0618 333.0618 289.07, 271.06, 167.03 3
10 Coumaroylquinic acid C16H18O8 337.0924 337.0626 191.05, 163.03 2
11 Caffeic acid hexoside C15H18O9 341.1084 341.0872 215.03, 179.06, 161.04, 135.04 2
12 Ferulylshikimic acid C17H18O8 349.0927 349.0923 193.05, 177.01, 173.04, 155.03, 129.02 2
13 Ferulic acid hexoside C15H16O10 355.0666 355.0665 193.05, 179.02, 149.05, 134.02 2
14 Syringoylquinic acid C16H20O10 371.0981 371.0978 353.08, 191.05, 173.04, 135.04 2
15 caffeic acid derivatives C18H18O9 377.0885 377.0878 341.11, 215.03, 179.05, 161.03 2
16 Feruloyl-galactaric acid C16H18O11 385.0828 385.0776 341.08, 209.03, 191.03, 147.02 2
17 Picraquassioside A# C18H22O10 397.1142 397.1135 235.06, 217.05, 191.07, 187.04, 177.05 3
18 Methylpicraquassioside A# C19H24O10 411.1302 411.1297 397.11, 249.07, 231.06, 219.06, 201.05 3
19 Paederol B# C20H28O12 459.1511 459.1508 399.12, 341.12, 281.10, 193.05 3
20 Hexosyl caffeoyl hexose C21H28O14 503.1393 503.1401 341.0.08, 179.03, 161.02 2
21 Arillatose B# C22H29O14 517.1563 517.1557 313.05, 193.05 3
Flavonoids 22 Eriodictyol C15H12O6 287.0565 287.0555 179.01, 151.00, 135.04, 125.01, 107.03 2
23 Catechin C15H14O7 289.0721 289.0712 245.08, 205.05, 179.03, 135.04 2
24 Dactylorhin C C14H24O10 351.1293 351.1291 189.07, 179.05, 171.06, 163.06, 127.07 2
25 Taxifolin-7-sulfate C15H12O10S 383.0105 383.0079 303.05, 285.04, 275.05, 151.00, 125.03 2
26 Daidzin C21H20O9 415.1029 415.1035 253.05, 235.04, 225.02, 135.00, 119.05 2
27 Cajanone C25H26O6 421.1662 421.1651 383.12, 217.05, 197.09, 165.07, 151.00 2
28 Phenethylrutinoside C20H30O10 429.1767 429.1761 249.09, 205.01, 161.04, 145.05, 119.05, 2
29 Kaempferol-3-O-glucoside C21H20O11 447.0932 447.0928 285.04, 271.06, 256.02, 240.04, 151.00 2
Alkaloids 30 Oleracein E C12H13NO3 218.0833 218.0823 200.07, 190.08, 160.04, 121.02 2
31 Feruloylglycine C12H13NO5 250.0724 250.0721 206.08, 191.07, 177.05, 149.06 2
32 Coumaroyltyramine C17H17NO3 282.1128 282.1123 279.01, 162.03, 145.03, 134.02, 119.02 2
33 Caffeoyltyramine C17H17NO4 298.1085 298.1079 280.09, 178.05, 160.04, 136.07, 121.06 2
34 Feruloyltyramine C18H19NO4 312.1235 312.1241 177.05, 149.06, 136.07, 121.06, 119.05 2
35 Feruloyloctopamine C18H19NO5 328.1181 328.1184 310.02, 161.05, 133.02 2
36 Oleracein U C18H15NO6 340.0831 340.0827 322.07, 296.09, 194.04, 145.02, 132.04 2
37 Oleracein A C24H25NO11 502.1351 502.1349 340.08, 296.09, 194.05, 145.02 2
38 Oleracein B C25H27NO12 532.1463 532.1455 370.09, 326.10, 194.05, 175.04, 161.02 2
39 Oleracein C C30H35NO16 664.1883 664.1877 502.13, 340.08, 296.09, 194.04, 145.02 2
40 Oleracein I C31H37NO17 694.1989 694.1983 518.15, 370.09, 326.10, 194.04, 175.04 2
41 Oleracein P C36H45NO21 826.2395 826.2405 664.18, 502.13, 340.08, 194.04, 145.02 2
42 Oleracein N/S C40H43NO19 840.2336 840.2351 694.19, 664.18, 340.08, 194.04, 145.02 2
43 Oleracein L/J C40H43NO20 856.2276 856.2300 694.19, 518.15, 326.10, 194.04, 161.02 2
44 Oleracein O C41H45NO20 870.2443 870.2456 694.19, 518.15, 194.04, 175.04, 161.02 2
Fatty acids 45 Citramalic acid C5H8O5 147.0319 147.0299 129.01, 115.00, 103.04 2
46 2-Isopropylmalic acid C7H12O5 175.0625 175.0612 157.05, 115.04, 113.06 2
47 Citric acid C6H8O7 191.0217 191.0197 145.01, 129.01, 111.00 2
48 Oxaloglutaric acid C7H8O7 203.0189 203.0197 141.01, 97.02, 69.03 2
49 Homocitric acid C7H10O7 205.0351 205.0354 161.04, 143.04, 117.05 2
50 Jasmonic acid# C12H18O3 209.1176 209.1183 165.09, 133.01, 109.03 3
51 Tuberonic acid# C12H18O4 225.1125 225.1127 207.10, 181.12, 163.11, 135.08 3
52 Palmitic acid C16H32O2 255.2314 255.2324 237.22, 211.24, 195.21, 59.01 2
53 2-Hydroxypalmitic acid C16H32O3 271.2274 271.2273 253.21, 227.12, 2
54 Linolenic acid C18H30O2 277.2165 277.2169 259.20, 233.22, 205.21, 179.25, 165.23 2
55 alpha-Linoleic acid C18H32O2 279.2331 279.233 261.22 2
56 Oleic acid C18H34O2 281.2487 281.2486 263.25, 181.21, 127.25 2
57 Stearic acid C18H36O2 283.2643 283.2637 265.24, 239.25, 209.22, 183.19, 171.12 2
58 Hydroxy octadecatrienoic acid C18H30O3 293.2112 293.2116 275.20, 223.03, 195.13, 183.13, 171.10 2
59 Hydroxy octadecadienoic acid C18H32O3 295.2312 295.2276 277.20, 253.02, 223.03, 167.05 2
60 Hydroxy octadecenoic acid C18H34O3 297.2433 297.2429 279.23, 255.12, 225.05, 127.05 2
61 Arachidonic acid C20H32O2 303.2326 303.2324 285.22, 269.19, 259.24, 205.12 2
62 Dihydroxy octadecatrienoic acid C18H30O4 309.2075 309.2069 291.19, 199.85, 179.14, 110.03 2
63 Trihydroxy-octadecadienoic acid C18H32O5 327.217 327.2171 299.12, 285.21, 229.14, 211.13, 171.10 2
64 Pinellic acid# C18H34O5 329.2329 329.2328 229.14, 211.13, 171.10 3
65 Tuberonic acid glucoside# C18H27O9 387.1656 387.1655 207.10, 163.11, 101.02 3
66 Methyl tuberonic acid glucoside# C19H30O9 401.1823 401.1817 239.12, 221.11, 207.10, 163.06 3
Amino acids 67 Phenylalanine C9H11NO2 164.0732 164.0717 147.04, 120.08 2
68 Tyrosine C9H11NO3 180.0674 180.0666 163.04, 134.06 2
69 3,4-Dihydroxyphenylalanine C9H11NO4 196.0571 196.0615 181.05, 152.07 2
70 N-acetyl phenylalanine C11H13NO3 206.0816 206.0823 164.07, 147.04 2
71 N-Acetyl tyrosine C11H13NO4 222.0766 222.0772 180.06, 178.08, 163.04 2
72 N-benzoylaspartic acid C11H11NO5 236.0558 236.0564 218.05, 192.06, 174.05, 120.04, 115.00 2
73 N-glucosyl ethanolamine C8H17NO7 238.0927 238.0932 220.08, 202.07, 139.00 2
74 N-Acetyl tryptophan C13H14N2O3 245.0925 245.0932 203.08, 185.07, 170.06, 116.05, 2
Terpenoids 75 Triptophenolide A1# C20H24O3 311.1682 311.1653 295.13, 283.16, 267.17, 251.14, 237.12, 3
76 Menthane-1,2,8,9-tetrol 2-glucoside# C16H30O9 365.1807 365.1812 204.13, 186.12, 168.11 3
77 α,g-Onoceradienedione C30H46O2 437.3426 437.342 219.17, 205.15 2
78 4,5-dioxo 10-epi-4,5-seco-γ-eudesmol 2′-O-acetyl-fucopyranoside# C23H38O8 441.2508 441.2488 399.23, 253.18, 221.15, 191.14 3
79 Oleanolic acid C30H48O3 455.353 455.3525 407.33, 391.30, 377.28, 363.26 2
Others 80 Glucose C6H12O6 179.0572 179.0561 163.06, 147.06, 115.04 2
81 Psoralen C11H6O3 185.025 185.0244 157.02, 141.03, 129.03, 115.01 2
82 Gluconic acid C6H12O7 195.0522 195.051 177.01, 151.06, 129.02, 121.04 2
83 Ethyl glucoside C8H16O6 207.0854 207.0847 179.05, 163.06, 147.06, 115.04 2
84 Bargapten C12H8O4 215.0348 215.0344 185.02, 157.02, 141.03, 129.03, 115.01 2
85 Glucosylglycolate C8H14O8 237.0619 237.0616 220.05, 207.05, 193.07,163.06, 147.02 2
86 Oxyresveratrol C14H12O4 243.0656 243.0663 225.05, 199.05, 161.06, 135.04 2
87 2-deoxy-2,3-dehydro-N-acetylneuraminic acid# C11H17NO8 290.0876 290.0876 230.06, 200.05, 169.01, 128.07 3
88 Diphyllin C21H16O7 379.0823 379.0817 363.05, 347.01, 333.04, 319.06, 305.04 2
89 Piceatannol glucoside C20H22O9 405.1172 405.1178 243.06, 201.05, 159.04 2
90 Benzyl alcohol dihexoside C19H28O11 431.1564 431.1553 341.10, 269.10, 251.09, 179.05, 163.06 2
91 Daphylloside C19H26O12 445.135 445.1346 409.11, 387.12, 267.08, 179.05 2

EF. elemental formula; OM. observed mass; CM. calculated mass; CL. confidence level; (-). Negative mode. # First time identification in Portulaca oleracea.

Possible mass fragmentation pattern of (A) tuberonic acid glucoside, (B) menthane-1,2,8,9-tetrol glucoside and (C) 4–5-dioxo 10-epi-4,5-seco-γ-eudesmol 2′-O-acetyl-fucopyranoside.
Fig. 4
Possible mass fragmentation pattern of (A) tuberonic acid glucoside, (B) menthane-1,2,8,9-tetrol glucoside and (C) 4–5-dioxo 10-epi-4,5-seco-γ-eudesmol 2′-O-acetyl-fucopyranoside.

3.6.5

3.6.5 Terpenoids

For the first time in APO, compounds 75–79 was tentatively identified as terpenoids (Table 5). Compounds 75 was identified as triptophenolide A1 (m/z 311.1682) with molecular formula C20H24O3, based on the mass fragmentation behaviour described by Li et al., (Li et al., 2008). Compounds 76 produced the deprotonated ion [M−H]- at m/z 365.1812, yielded a fragment ion at m/z 204.13 by losing the glucosyl (162 Da) moiety, which was followed by the loss of one and two molecules of H2O to form the fragment ions at m/z 186.12 and 168.11, respectively. Accordingly, the compound was tentatively identified as menthane-1,2,8,9-tetrol glucoside (Fig. 4B) (Matsumura et al., 2002). Moreover, compound 78 generated a monoisotopic mass [M−H]- at m/z 441.2504, yielded fragment ions at m/z 399.23 ([M−H−acetyl]-), 253.18 ([M−H−acetyl−rhamnosyl]-), 221.15 ([M−H−253.18-CH3OH]-) and 191.14 ([M−H−221.15-CO]-) and was tentatively identified as 4,5-dioxo 10-epi-4,5-seco-γ-eudesmol 2′-O-acetyl-fucopyranoside with molecular formula C23H38O8 (Fig. 4C).

3.6.6

3.6.6 Others

compounds 80–91 were identified as glucose (m/z 179.0572), psoralen (m/z 185.0250), gluconic acid (m/z 195.0522), ethyl glucoside (m/z 207.0854), bergapten (m/z 215.0384), glucosylglycolate (m/z 237.0619), oxyresveratrol (m/z 243.0656), 2-deoxy-2,3-dehydro-N-acetylneuraminic acid (m/z 290.0876), diphyllin (m/z 379.0823), piceatannol glucoside (m/z 405.1178), benzyl alcohol glucoside (m/z 431.1564) and daphylloside (m/z 445.1346), based on the similarities noticed in their fragmentation behaviors and the behaviors mentioned in the literature.

4

4 Conclusions

This work, which was the first investigation into optimizing the HRE conditions on APO using two modeling approaches (RSM and ANN), revealed the presence of phenolic acids, flavonoids, alkaloids, fatty acid and terpenoids, through high-resolution mass spectroscopy examination. The ANN model is more accurate and sophisticated than the RSM model, as evidenced by the fact that it had a higher R2 and lower RMSE, AAD, and SEP values than the latter. The ideal parameters were identified as 60 % ethanol, extraction time of 90.5 min of extraction time, and 50 °C of extraction temperature. The highest values of TPC, TFC, DPPH radical scavenging effect, and ascorbic acid equivalent FRAP were found as 8.23 ± 1.06 mgGAE/g, 43.12 ± 1.15 mgCAE/g, 43.01 ± 1.25 %, and 35.98 ± 0.19, respectively, under these circumstances. These results lead us to the conclusion that APO, a viable candidate for an antioxidant functional food, can be widely used commercially in the nutraceutical food and pharmaceutical industries.

CRediT authorship contribution statement

Fanar Alshammari: Methodology, Formal analysis, Investigation, Writing – original draft. Md Badrul Alam: Conceptualization, Investigation, Formal analysis, Project administration, Writing – review & editing. Marufa Naznin: Methodology, Investigation, Formal analysis. Sunghwan Kim: Conceptualization, Supervision, Writing – review & editing. Sang-Han Lee: Conceptualization, Methodology, Supervision, Funding acquisition, Project administration, Writing – review & editing.

Acknowledgments

This study was supported by the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (2020R1A2C2011495 and 2021R1IA1A01058062). Fanar Alshammaria received the financial support for his Ph.D. studying project from Education Ministry of Kingdom of Saudi Arabia (EMSA).

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.

References

  1. , , , . Isolation and characterization of Cepa2, a natural alliospiroside A, from shallot (Allium cepa L. Aggregatum group) with anticancer activity. Plant Physiol. Biochem. : PPB.. 2017;116:167-173.
    [CrossRef] [Google Scholar]
  2. , , , . DNA protecting activities of Nymphaea nouchali (Burm. f) flower extract attenuate t-BHP-induced oxidative stress cell death through Nrf2-mediated induction of Heme Oxygenase-1 expression by activating MAP-kinases. Int. J. Mol. Sci.. 2017;18
    [CrossRef] [Google Scholar]
  3. , , , . High resolution mass spectroscopy-based secondary metabolite profiling of Nymphaea nouchali (Burm. f) stem attenuates oxidative stress via regulation of MAPK/Nrf2/HO-1/ROS pathway. Antioxidants. 2021;10:719.
    [Google Scholar]
  4. , , , . High resolution mass spectroscopy-based secondary metabolite profiling of Nymphaea nouchali (Burm. f) stem attenuates oxidative stress via regulation of MAPK/Nrf2/HO-1/ROS pathway. Antioxidants (Basel Switzerland). 2021;10
    [CrossRef] [Google Scholar]
  5. , , , . Response surface methodology (RSM)-based optimization of ultrasound-assisted extraction of Sennoside A, Sennoside B, Aloe-Emodin, Emodin, and Chrysophanol from Senna alexandrina (Aerial Parts): HPLC-UV and antioxidant analysis. Molecules (Basel, Switzerland).. 2022;27
    [CrossRef] [Google Scholar]
  6. , , , . Characterization of gallotannins and benzophenone derivatives from mango (Mangifera indica L. cv. 'Tommy Atkins') peels, pulp and kernels by high-performance liquid chromatography/electrospray ionization mass spectrometry. Rapid Commun. Mass Spectromet. : RCM.. 2004;18:2208-2216.
    [CrossRef] [Google Scholar]
  7. , , , . Optimization of the extraction conditions of Nypa fruticans Wurmb. using response surface methodology and artificial neural network. Food Chem.. 2022;381:132086
    [CrossRef] [Google Scholar]
  8. , , , . Comparison between artificial neural network and response surface methodology in the prediction of the parameters of heat set polypropylene yarns. J. Text. Inst.. 2015;106:417-430.
    [CrossRef] [Google Scholar]
  9. , , , . Prandtl number of optimum biodiesel from food industrial waste oil and diesel fuel blend for diesel engine. Fuel. 2021;285:119049
    [CrossRef] [Google Scholar]
  10. , , , . Effect of extraction solvent on total phenol content, total flavonoid content, and antioxidant activity of Limnophila aromatica. J. Food Drug Anal.. 2014;22:296-302.
    [CrossRef] [Google Scholar]
  11. , , , . Flavonoids extract from Portulaca oleracea L. induce Staphylococcus aureus death by apoptosis-like pathway. Int. J. Food Prop.. 2017;20:S534-S542.
    [Google Scholar]
  12. , , , . Hepatoprotective effects of Portulaca oleracea extract against CCl4-induced damage in rats. Pharm. Biol.. 2015;53:1042-1051.
    [CrossRef] [Google Scholar]
  13. , , , . Phytochemical composition and antioxidant activity of Portulaca oleracea: influence of the steaming cooking process. Foods (Basel, Switzerland). 2021;10
    [CrossRef] [Google Scholar]
  14. , , , . Optimization of the ethanol recycling reflux extraction process for saponins using a design space approach. PLoS One. 2014;9:e114300.
    [Google Scholar]
  15. , , , . METLIN: a technology platform for identifying knowns and unknowns. Anal. Chem.. 2018;90:3156-3164.
    [CrossRef] [Google Scholar]
  16. , , . A survey on stock market prediction using various algorithms. Int. J. Comput. Technol. Appl.. 2014;5:530-533.
    [Google Scholar]
  17. , , . Interaction of ethanol with biological membranes: the formation of non-bilayer structures within the membrane interior and their significance. J. Phys. Chem. B. 2009;113:1983-1992.
    [CrossRef] [Google Scholar]
  18. , , , . Effects of aqueous Purslane (Portulaca oleracea) extract and fish oil on gentamicin nephrotoxicity in albino rats. J. Nat. Sci.. 2011;9:47-62.
    [Google Scholar]
  19. , , , . RSM and ANN modeling-based optimization approach for the development of ultrasound-assisted liposome encapsulation of piceid. Ultrason. Sonochem.. 2017;36:112-122.
    [Google Scholar]
  20. , , , . A review of traditional uses, phytochemistry and pharmacology of Portulaca oleracea L. J. Ethnopharmacol.. 2017;205:158-172.
    [CrossRef] [Google Scholar]
  21. , , , . Identification of secondary metabolites in Averrhoa carambola L. bark by high-resolution mass spectrometry and evaluation for α-glucosidase, tyrosinase, elastase, and antioxidant potential. Food Chem.. 2020;332:127377
    [CrossRef] [Google Scholar]
  22. , , , . Metabolite profiling of Manilkara zapota L. leaves by high-resolution mass spectrometry coupled with ESI and APCI and in vitro antioxidant activity, α-Glucosidase, and Elastase inhibition assays. Int. J. Mol. Sci.. 2021;22:132.
    [Google Scholar]
  23. , , , . Profile and characterization of the chlorogenic acids in green Robusta coffee beans by LC-MS n: identification of seven new classes of compounds. J. Agric. Food Chem.. 2010;58:8722-8737.
    [Google Scholar]
  24. , , , . Structural characterization of flavonoid Glycoconjugates and their derivatives with mass spectrometric techniques. Molecules. 2016;21
    [CrossRef] [Google Scholar]
  25. , , , . Protective effects of aqueous and ethanolic extracts of Nigella sativa L. and Portulaca oleracea L. on free radical induced hemolysis of RBCs. Daru. 2011;19:295-300.
    [Google Scholar]
  26. , , , . Antiulcerogenic effect of ethanolic extract of Portulaca oleracea experimental study. Pharmacologyonline. 2010;1:417-432.
    [Google Scholar]
  27. , , , . Evaluation of extract of Ipomoea batatas leaves as a green coagulant–flocculant for turbid water treatment: Parametric modelling and optimization using response surface methodology and artificial neural networks. Environ. Technol. Innov.. 2021;24:102005
    [CrossRef] [Google Scholar]
  28. , , , . A comparative evaluation of statistical empirical and neural intelligence modeling of Manihot esculenta-derived leaves extract for optimized bio-coagulation-flocculation of turbid water. Ind. Crop. Prod.. 2022;186:115194
    [CrossRef] [Google Scholar]
  29. , , , . Analysis of triptophenolide and its related compounds from Tripterygium wilfordii Hook.f by electrospray ionization tandem mass spectrometry. Int. J. Mass Spectrom.. 2008;278:38-49.
    [CrossRef] [Google Scholar]
  30. , , , . Identification of α-glucosidase inhibitors from Cortex Lycii based on a bioactivity-labeling high-resolution mass spectrometry–metabolomics investigation. J. Chromatogr. A. 2021;1642:462041
    [Google Scholar]
  31. , , , . Comparative metabolite fingerprinting of legumes using LC-MS-based untargeted metabolomics. Food Res. Int. (Ottawa, Ont.). 2019;126:108666
    [CrossRef] [Google Scholar]
  32. , , , . Reflux extraction optimization and antioxidant activity of phenolic compounds from Pleioblastus amarus (Keng) Shell. Molecules (Basel, Switzerland). 2022;27
    [CrossRef] [Google Scholar]
  33. , , , . Bronchodilatory effect of Portulaca oleracea in airways of asthmatic patients. J. Ethnopharmacol.. 2004;93:57-62.
    [Google Scholar]
  34. , , , . Water-soluble constituents of caraway: carvone derivatives and their glucosides. Chem. Pharm. Bull.. 2002;50:66-72.
    [CrossRef] [Google Scholar]
  35. , , , . Identification of phytochemicals of Phoenix dactylifera L. Cv Ajwa with UHPLC-ESI-QTOF-MS/MS. Int. J. Fruit Sci.. 2021;21:848-867.
    [CrossRef] [Google Scholar]
  36. Naveja, J. J., M. P. Rico-Hidalgo and J. L. Medina-Franco, 2018. Analysis of a large food chemical database: chemical space, diversity, and complexity. F1000Research. 7, Chem Inf Sci-993. https://doi.org/10.12688/f1000research.15440.2.
  37. , , , . Evaluation of potential antifertility activity of total flavonoids, isolated from Portulaca oleracea L on female albino rats. Int. J. PharmTech. Res.. 2014;6:783-793.
    [Google Scholar]
  38. , , , . Polyphenols LC-MS2 profile of Ajwa date fruit (Phoenix dactylifera L.) and their microemulsion: potential impact on hepatic fibrosis. J. Funct. Foods. 2018;49:401-411.
    [CrossRef] [Google Scholar]
  39. , , , . Development of ternary models for prediction of biogas yield in a novel modular biodigester: a case of fuzzy Mamdani model (FMM), artificial neural network (ANN), and response surface methodology (RSM) Biomass Convers. Biorefin. 2020
    [CrossRef] [Google Scholar]
  40. , , , . Estimation of biogas yields produced from combination of waste by implementing response surface methodology (RSM) and adaptive neuro-fuzzy inference system (ANFIS) Int. J. Energy Environ. Eng.. 2021;12:353-363.
    [CrossRef] [Google Scholar]
  41. , , , . Predictive ability of Response Surface Methodology (RSM) and artificial Neural Network (aNN) to approximate Biogas Yield in a Modular Biodigester. Springer International Publishing; .
  42. , , . Optimization of extraction conditions for colchicine from Gloriosa superba tubers using response surface methodology. J. Agric. Technol.. 2012;8:1301-1315.
    [Google Scholar]
  43. , , , . Isolation, identification and quantitation of hydroxycinnamic acid conjugates, potential platform chemicals, in the leaves and stems of Miscanthus× giganteus using LC–ESI-MSn. Phytochemistry. 2011;72:2376-2384.
    [Google Scholar]
  44. , , , . Phytochemical composition and bioactive compounds of common purslane (Portulaca oleracea L.) as affected by crop management practices. Trends Food Sci. Technol.. 2016;55:1-10.
    [Google Scholar]
  45. , , , . Characterization of phenolic and other polar compounds in a lemon verbena extract by capillary electrophoresis-electrospray ionization-mass spectrometry. J. Sep. Sci.. 2010;33:2818-2827.
    [CrossRef] [Google Scholar]
  46. , , , . Comprehensive chemical profiling in the ethanol extract of Pluchea indica aerial parts by liquid chromatography/mass spectrometry analysis of its silica gel column chromatography fractions. Molecules (Basel, Switzerland). 2019;24
    [CrossRef] [Google Scholar]
  47. , , , . Response surface optimisation of antioxidant extraction from kinema, a Bacillus-fermented soybean food. Food Chem.. 2011;129:507-513.
    [CrossRef] [Google Scholar]
  48. , , , . Portuloside A, a monoterpene glucoside, from Portulaca oleracea. Phytochemistry. 1996;42:1625-1628.
    [Google Scholar]
  49. , , . The development of expeller for kernel based seed and its oil characterization. Int. J. Integr. Eng.. 2018;10
    [Google Scholar]
  50. , , . Comparison of Response Surface Methodology (RSM) and Artificial Neural Network (ANN) in modelling of waste coconut oil ethyl esters production. Energy Sources Part A. 2019;41:1049-1061.
    [CrossRef] [Google Scholar]
  51. , , , . Identifying small molecules via high resolution mass spectrometry: communicating confidence. Environ. Sci. Tech.. 2014;48:2097-2098.
    [CrossRef] [Google Scholar]
  52. , , , . Optimization of the ultrahigh-pressure–assisted extraction of phenolic compounds and antioxidant activity from palm dates (Phoenix dactylifera L.) Food Anal. Methods. 2020;13:1556-1569.
    [CrossRef] [Google Scholar]
  53. , , , . Properties of Portulaca oleracea seed oil via supercritical fluid extraction: experimental and optimization. J. Supercrit. Fluids. 2018;135:34-44.
    [Google Scholar]
  54. , , , . Optimization of fatty acids extraction from Portulaca oleracea seed using response surface methodology. Ind. Crop. Prod.. 2013;43:405-411.
    [Google Scholar]
  55. , , , . Analysis of chemical constituents in Ficus Hirta Vahl. by LCMS-IT-TOF and GC-MS. IOP Conf. Ser.: Mater. Sci. Eng.. 2020;730:012027
    [CrossRef] [Google Scholar]
  56. , , , . UHPLC-Orbitrap-MS tentative identification of 51 Oleraceins (Cyclo-Dopa Amides) in Portulaca oleracea L. cluster analysis and MS(2) filtering by mass difference. Plants (Basel Switzerland). 2021;10
    [CrossRef] [Google Scholar]
  57. , , . Structural characterization of flavonoid glycosides by multi-stage mass spectrometry. Mass Spectrom. Rev.. 2010;29:1-16.
    [CrossRef] [Google Scholar]
  58. , , , . Protective effect of Portulaca oleracea extracts on hypoxic nerve tissue and its mechanism. Asia Pac. J. Clin. Nutr.. 2007;16:227.
    [Google Scholar]
  59. , , , . Optimization of ultrasonic-assisted extraction of flavonoid from Portulaca oleracea L. by response surface methodology and chemical composition analysis. J. Korean Soc. Appl. Biol. Chem.. 2014;57:647-653.
    [Google Scholar]
  60. , , , . CFM-ID 4.0: more accurate ESI-MS/MS spectral prediction and compound identification. Anal. Chem.. 2021;93:11692-11700.
    [CrossRef] [Google Scholar]
  61. , , . Optimization of ultrahigh-pressure extraction of polyphenolic antioxidants from green tea by response surface methodology. Food Bioproc. Tech.. 2013;6:2538-2546.
    [CrossRef] [Google Scholar]
  62. , , , . Alkaloids from Portulaca oleracea L. Phytochemistry. 2005;66:2595-2601.
    [Google Scholar]
  63. , , , . Homoisoflavonoids from the medicinal plant Portulaca oleracea. Phytochemistry. 2012;80:37-41.
    [Google Scholar]
  64. , , , . Exergy and energy analyses of microwave Dryer for Cantaloupe slice and prediction of thermodynamic parameters using ANN and ANFIS algorithms. Energies. 2021;14:4838.
    [Google Scholar]
  65. Zhou, Y. X., H. L. Xin, K. Rahman, et al., 2015. Portulaca oleracea L.: a review of phytochemistry and pharmacological effects. BioMed research international. 2015, 925631. https://doi.org/10.1155/2015/925631.
  66. , , , . Antioxidant activities and total phenolic contents of various extracts from defatted wheat germ. Food Chem.. 2011;126:1122-1126.
    [CrossRef] [Google Scholar]

Appendix A

Supplementary material

Supplementary data to this article can be found online at https://doi.org/10.1016/j.arabjc.2022.104425.

Appendix A

Supplementary material

The following are the Supplementary data to this article:

Supplementary data 1

Supplementary data 1

Show Sections