Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=lanl20
Analytical Letters
ISSN: 0003-2719 (Print) 1532-236X (Online) Journal homepage: https://www.tandfonline.com/loi/lanl20
Optimization of Maceration Conditions for
Improving the Extraction of Phenolic Compounds
and Antioxidant Effects of Momordica Charantia L.
Leaves Through Response Surface Methodology
(RSM) and Artificial Neural Networks (ANNs)
Sengul Uysal, Aleksandra Cvetanović, Gokhan Zengin, Zoran Zeković,
Mohamad Fawzi Mahomoodally & Oskar Bera
To cite this article: Sengul Uysal, Aleksandra Cvetanović, Gokhan Zengin, Zoran Zeković, Mohamad Fawzi Mahomoodally & Oskar Bera (2019) Optimization of Maceration Conditions for Improving the Extraction of Phenolic Compounds and Antioxidant Effects of Momordica�Charantia L. Leaves Through Response Surface Methodology (RSM) and Artificial Neural Networks (ANNs), Analytical Letters, 52:13, 2150-2163, DOI: 10.1080/00032719.2019.1599007
To link to this article: https://doi.org/10.1080/00032719.2019.1599007
Published online: 02 Apr 2019. Submit your article to this journal
Article views: 281 View related articles
BIOANALYTICAL
Optimization of Maceration Conditions for Improving the
Extraction of Phenolic Compounds and Antioxidant Effects
of Momordica Charantia L. Leaves Through Response
Surface Methodology (RSM) and Artificial Neural
Networks (ANNs)
Sengul Uysala,b , Aleksandra Cvetanovicc , Gokhan Zengind,e , Zoran Zekovicc, Mohamad Fawzi Mahomoodallyd,e , and Oskar Berac
aErciyes University, Halil Bayraktar Health Services Vocational College, Kayseri, Turkey;bGenome and Stem Cell Research Center, Erciyes University, Kayseri, Turkey;cDepartment of Biotechnology and Pharmaceutical Engineering, Faculty of Technology, University of Novi Sad, Novi Sad, Republic of Serbia;dDepartment of Biology. Faculty of Science, Selcuk University, Konya, Turkey;eFaculty of Science, Department of Health Sciences, University of Mauritius, Reduit, Mauritius
ABSTRACT
The main goals of this research were the chemical and biological characterization of the bitter melon (Momordica charantia) isolate obtained by traditional (maceration) extraction, as well as optimiza-tion of this process using response surface methodology (RSM) and artificial neural networks (ANNs). Experiments were performed using Box–Behnken experimental design on three levels and three varia-bles: extraction temperature (20C, 40C, and 60C), solvent concen-tration (30%, 50%, and 70%) and extraction time (30, 60, and 90 min). The measurements consisted of 15 randomized runs with 3 replicates in a central point. The antioxidant activity of obtained extracts was determined by the 1,1-diphenyl-2-picrylhydrazyl (DPPH), cupric ion reducing antioxidant capacity (CUPRAC) and ferric reduc-ing antioxidant power (FRAP) assays while chemical characterization was done in terms of the total phenolic content (TPC). The method-ology shows positive influence of solvent concentration on all four observed outputs, while temperature showed a negative impact. RSM showed that the optimal extraction conditions were 20C, 70% methanol, and an extraction time of 52.2 min. Under these condi-tions, the TPCs were 20.66 milligrams of gallic acid equivalents (mg GAE/g extract), DPPH 30.22 milligrams of trolox equivalents (mg TE/g extract), CUPRAC 67.78 milligrams of trolox equivalents (mg TE/g extract), and FRAP 45.48 milligrams of trolox equivalents (mg TE/g extract). The neural network coupled with genetic algorithms (ANN-GA) was also used to optimize the conditions for each of the outputs separately. It is anticipated that results reported herein will establish baseline data and also demonstrate that that the present model can be applied in the food and pharmaceutical industries.
ARTICLE HISTORY Received 18 February 2019 Accepted 20 March 2019 KEYWORDS Momordica charantia; antioxidant properties; artificial neural network— genetic algorithm (ANN-GA); response surface methodology (RSM); total phenolic content (TPC)
CONTACT Gokhan Zengin gokhanzengin@selcuk.edu.tr Faculty of Science, Department of Biology, Selcuk University, Campus/Konya, Konya, Turkey.
Color versions of one or more of the figures in the article can be found online atwww.tandfonline.com/lanl. These authors contributed equally.
ß 2019 Taylor & Francis Group, LLC 2019, VOL. 52, NO. 13, 2150–2163
Introduction
Momordica charantia L., is a popular cultivated food crop belonging to the Cucurbitaceae family. Different traditional names of this plant have been reported, such as: bitter melon, bitter gourd, kugua, karela, balsam pear, but also in Turkey it is known as kudret narı (Kultur 2007; Habicht et al. 2011). M. charantia is spread over a wide part of the world and it is especially used in a treatment of diabetes, cancer, and micro-bial infections (Habicht et al.2014; Dandawate et al.2016).
Currently, M. charantia tea is very popular and used widely for medicinal purposes (Jia et al. 2017). Leaves of M. charantia are consumed as a vegetable in Taiwan, Indonesia, India, Mauritius, and Malaysia (Mahomoodally, Fakim, and Subratty 2004). At the same time, its leaves are used for treating skin diseases and burns (Lim 2012). Fruits of this plant from Anatolia have been used for stomach ulcers (Kultur et al.
2018). The composition of M. charantia has been well investigated and there is consid-erable scientific information about its bioactive components. It was documented that the bioactive compounds which are related with the activity of M. charantia include phenolics, polysaccharides, terpenoids, saponins, and proteins (Kubola and Siriamornpun 2008; Keller et al. 2011; Fang et al. 2012; Chang et al. 2015; Zhang, Lin, and Xie 2016).
Polyphenolic compounds are well-known bioactive constituents of many plants. They possess different bioactivities and have been associated with many health attributes. For these reasons, special attention has been put on these molecules and there are efforts to include them regularly as part of a normal human diet. For example, different plant extracts have been used for the preparation of functional foods and nutritional supple-ments. These foods are preferable from the health point of view because the polyphe-nols replace synthetic additives, avoiding their negative effects. The first and one of the most important steps in designing these foods is their extraction from plant matrices.
So far, different extraction techniques, including traditional but also modern techni-ques, have been used for their recovery from plant samples (Anand et al. 2005). Maceration is one of the oldest extraction techniques but still the most frequently used. Some of the common reasons for its utilization are (i) simple methodology, (ii) the great number of solvents which can be used, (iii) usually low extraction temperature, and (iv) low investment costs. Some of the parameters which influence the maceration efficiency are the choice of the solvent, extraction temperature, solvent-to-sample ratio, and time of the extraction. All of these parameters affect extraction yield, the concentra-tion of bioactive compounds in the final extracts, and thus the bioactivity. In order to obtain extracts with desirable characteristics, it is necessary to optimize the extraction conditions. For this purpose, there have been many devised approaches, including one factor at the time or different statistical approaches.
One of the commonly used statistical methods is response surface methodology (RSM), which is regarded to be a significant and effective method. RSM has been used dominantly in various fields including food, medicine, and pharmaceutical sciences (Zhang et al. 2014; Oberoi and Sogi 2017; Wong et al. 2017). Further, artificial neural networks (ANNs) became more and more popular for modeling and optimization proc-esses. This methodology permits the study of relationships between the inputs and the
outputs of the process using a limited number of experimental measurements (Desai et al. 2008).
To the best of our knowledge, there has been no report on the study of optimization of maceration of M. charantia leaves to produce bioactive crude extract. In this respect, this study was designed to assess the impact of extraction time, temperature, and solvent concentration on phenolic component and biological activities of extracts prepared from M. charantia.
Materials and methods
Sample preparation
The extractions were performed with leaves of M. charantia. The collection of plant material has been carried out from Silifke, Mersin, Turkey in 2015. After collection of plant material, the leaves were separated and dried naturally for 10 days. Dried leaves were then powdered with a laboratory mill.
These samples were macerated by methanol. Maceration was done by using three concentrations of the solvent, at three temperatures and using three extraction times. The values of methanol concentration, extraction temperature, and time are summarized inTable 1. Extraction runs were defined by different combinations of experimental con-ditions (Table 2). Obtained extracts were filtered and stored in a refrigerator for fur-ther analysis.
Total phenolic content
The total phenolic content (TPC) was determined as described previously (Slinkard and Singleton 1977) with some modifications. In a nutshell, 0.25 mL of extract (2 mg/mL) were mixed with 1 mL of Folin-Ciocalteu reagent . After 3 min, 0.75 mL of 1% sodium carbonate were added. This reaction mixture was incubated at room temperature for 2 h, and the absorbance of the reaction mixture was measured at 760 nm in order to determine the TPC. The calculation was done by interpolating the measured sample absorbance into the calibration curve defined by standard solutions of gallic acid. The results were expressed as the equivalents of gallic acid (mg GAE/g extract).
Assays for antioxidant capacity
The antioxidant and antiradical ability of the obtained extracts were explored by using three spectrophotometric methods: DPPH, CUPRAC, and FRAP.
Table 1. Natural and coded levels of independent variables used in applied RSM design.
Variable Coded level
21 0 1
Natural level
Temperature (C) 40 60 80
Solvent concentration (%) 30 50 70
For the DPPH measurements, extracts at concentrations of 2 mg/mL were used. The extracts were mixed with 0.004% DPPH in methanol in a 1:4 ratio (Sarikurkcu 2011). The obtained reaction mixtures were incubated in the dark for 30 min and immediately after the absorbance of the reaction mixture was measured at 517 nm. The ability of the extracts to neutralize DPPH radicals was expressed as milligrams of trolox equivalents (mg TE/g extract).
The CUPRAC test was performed with 0.5 mL of extracts at a concentration of 2 mg/ mL that were mixed with 3 mL previously prepared mixture composed of 1:1:1 10 mM CuCl2:7.5 mM neocuproine:1 M ammonium acetate buffer (Apak et al.2006). The blank
was prepared by replacing the extract with water and without CuCl2 in the premixed
reaction mixture.
The absorbance values of both sample and blank were measured at 450 nm after a 30 minute incubation period. The calibration curve was prepared with trolox, and the results are expressed as milligrams of trolox equivalents (mg TE/g extract).
The ferric reducing antioxidant power (FRAP) assay described by Aktumsek et al (2013) was used in order to define reduction ability of the extracts. The FRAP reagent was prepared by mixing 0.3 M pH 3.6 acetate buffer, 10 mM 2,4,6-tris(2-pyridyl)-S-tria-zine (TPTZ) in 40 mM HCl and 20 mM ferric chloride in a ratio of 10:1:1. The FRAP reagent was mixed with 2 mg/mL of the extract solution in a ratio 20:1 (v/v).
After 30 min, the absorbance was measured at 593 nm. The FRAP activity was expressed as the milligrams of trolox equivalents (mg TE/g extract). All measurements were repeated three times.
Experimental design and statistical analysis
In order to find best conditions to obtain M. charantia extract with the maximum TPC and maximum antioxidant activity, the Box–Behnken experimental design (BBD) of RSM was applied. The temperature (20–60C), concentration of the solvent (30–70%), Table 2. Natural and coded values of independent variables for Box-Behnken design and experimen-tally observed responses (TPC, DPPH, CUPRAC, and FRAP) in the extracts.
Run
Independent variable Investigated response Temperature/C Methanol concentration (%) Extraction time/min TPC (mg/g GAE) DPPH (mg/gTE) CUPRAC (mg/gTE) FRAP (mg/gTE) 1 40 (0) 70 (1) 30 (1) 13.73 39.81 61.72 42.43 2 20 (-1) 30 (1) 60 (0) 9.86 23.88 40.23 28.53 3 20 (-1) 50 (0) 30 (1) 12.85 25.19 53.59 37.88 4 40 (0) 50 (0) 60 (0) 11.59 21.82 47.98 34.93 5 40 (0) 50 (0) 60 (0) 11.96 24.11 52.86 37.94 6 60 (1) 50 (0) 30 (1) 12.35 20.14 51.79 36.70 7 60 (1) 50 (0) 90 (1) 9.05 7.79 43.12 29.37 8 40 (0) 50 (0) 60 (0) 11.78 25.74 49.9 35.98 9 60 (1) 30 (1) 60 (0) 7.80 3.94 33.80 26.40 10 60 (1) 70 (1) 60 (0) 15.89 33.08 79.00 50.42 11 40 (0) 30 (1) 90 (1) 8.22 4.85 40.49 28.21 12 40 (0) 30 (1) 30 (1) 10.44 14.20 55.38 38.12 13 20 (-1) 70 (1) 60 (0) 15.48 31.16 77.91 49.86 14 40 (0) 70 (1) 90 (1) 15.52 30.61 85.98 52.76 15 20 (-1) 50 (0) 90 (1) 15.02 29.38 76.62 45.71
TPC: Total phenolic content; DPPH: 1,1-diphenyl-2-picrylhydrazyl radical; CUPRAC: cupric ion reducing activity; FRAP: ferric reducing antioxidant power.
and extraction time (30–90 min) were selected as independent variables and their nat-ural as well as coded values are presented in the Table 1. The response variables were fitted into the given second-order polynomial model:
Y¼ b0þ X biXiþ X biXii2þ X bijXiXj (1)
where Y represents the response variable, and Xi and Xj are the independent variables.
The regression coefficients for all three terms (intercept, linear and quadratic) are repre-sented as b0, bi, bii, and bij.
Design-Expert v.7 Trial (Stat-Ease, Minneapolis, MN, USA) was used for performing BBD while ANOVA was applied for statistical testing of obtained results at the signifi-cance level of 0.05. Three-dimensional plots were used in order to describe the relation-ship between the obtained results and the extraction parameters.
ANN optimization
All data analysis with ANNs was performed using MATLAB software (The Math Works). The ANN models with one hidden layer were designed using the MATLAB Neural Network Toolbox. Four ANNs were constructed with three inputs: temperature, methanol concentration, and time. In each ANN, one of four outputs was observed: TPC, DPPH, CUPRAC, and FRAP activity. The optimization of the extraction procedure was designed to achieve the maximal total extraction yield and antioxidant activity and conducted in MATLAB. The optimization was performed using genetic algorithm (GA).
Results and discussion
The efficiency of the extraction of phenolic compounds, as well as the components which influence the antioxidant activity of the M. charantia, was studied by RSM. This
Table 3. Estimated coefficient of second-order polynomial models for investigated responses.
Regression coeficient TPC DPPH CUPRAC FRAP
b0 11.78 þ23.89 þ50.25 þ36.28 Linear b1 0 5.58 5.08 2.39 b2 þ3.04 þ10.97 þ16.84 þ9.28 b3 0.19 3.34 þ2.97 þ0.11 Cross product b12 þ0.62 þ5.47 þ1.88 þ0.67 b13 1.37 4.13 7.92 3.79 b23 þ1.00 þ0.038 þ9.79 þ5.06 Quadratic b11 þ0.41 1.31 þ1.44 0.22 b22 þ0.070 þ0.43 þ6.05 þ2.74 b33 þ0.13 1.96 þ4.60 þ1.35 R2a 0.9576 0.9627 0.9319 0.9295 CVb 7.62 15.28 12.42 9.79
TPC: Total phenolic content; DPPH: 1,1-diphenyl-2-picrylhydrazyl radical; CUPRAC: cupric ion reducing activity; FRAP: fer-ric reducing antioxidant power.
aCoefficient of multiple determination. b
Coefficient of variance (%). Highly significant p-value (< 0.01). Significant p-value (0.01< p < 0.05).
statistical method estimates the individual and combined effects of process parameters and proposed the optimal and most cost-effective extraction conditions which provide the highest possible content of polyphenolic compounds and antioxidant activity. The design of extractions and obtained values of TPC and antioxidant activities of different M. charantia extracts are presented in Table 2.
For each dependent variable, the multiple coefficients of regression were generated and are presented in Table 3. In the same table, the values of coefficients of determin-ation (R2) as well as coefficient of variation values are found. The lack of the fit of the proposed model together with ANOVA results are summarized in Table 4.
Mathematical models applied for all responses were statistically important, which was concluded according to the significant regression for the model (p< 0.05). In the case of antioxidant activity (DPPH, CUPRAC, and FRAP assays), the lack of fit was insignifi-cant indicating good correlation between experimental data and applied mathematical model. The coefficients of determination in all cases were very high, in the range from 0.92 to 0.96, indicating good representation of experimental values. The values for coef-ficients of variation were between 7.62 and 15.28, demonstrating the suitability of the model and its reproducibility.
Influence of independent variables
The TPC results extracted under different conditions were in the range from 7.80 to 15.89 mg/g GAE (Table 2). The lowest TPC was obtained using an extraction tempera-ture of 60C in 30% methanol for 60 min. On the other hand, the following
Table 4. Analysis of variance (ANOVA) of the fitted second-order polynomial model for TP content, DPPH, CUPRAC, and FRAP test.
Source Sum of squares Degrees of freedom Mean of square F-value p-value TPC Model 96.04 9 10.67 12.55 0.0062 Residual 4.25 5 0.85 Lack of fit 4.18 3 1.39 40.72 0.0241 Pure error 0.068 2 0.034 Total 100.30 14 DPPH Model 1510.52 9 167.84 14.35 0.0045 Residual 58.48 5 11.70 Lack of fit 50.72 3 16.91 4.36 0.1922 Pure error 7.76 2 3.88 Total 1568.99 14 CUPRAC Model 3393.96 9 377.11 7.61 0.0189 Residual 247.91 5 49.58 Lack of fit 235.83 3 78.61 13.01 0.0722 Pure error 12.09 6.04 Total 3641.88 14 FRAP Model 929.34 9 103.26 7.33 0.0205 Residual 70.46 5 14.09 Lack of fit 65.79 3 21.93 9.40 0.0977 Pure error 4.67 2 2.33 Total 999.80 14
TPC: Total phenolic content; DPPH: 1,1-diphenyl-2-picrylhydrazyl radical; CUPRAC: cupric ion reducing activity; FRAP: fer-ric reducing antioxidant power.
combination of parameters: temperature of 60C, 70% methanol concentration, and extraction time of 60 min, led to extracts with the highest TPC.
The TPC was highly influenced by linear term of solvent concentration (p< 0.01) and significantly influenced by the linear temperature term as well as by cross product of temperature and time (p< 0.05). Good fit of the experimental and theoretical values was confirmed by the high coefficient of multiple determination value equal to 0.9576. The coefficient of variance (7.62) was rather low, supporting the fitness of the model. The predicted second-order polynomial model for TPC was given by
TPC¼ 11:78 þ 3:04X2 0:19X3þ 0:62X1X2 1:37X1X3
þ 1X2X3þ 0:41X21þ 0:07X22þ 0:13X32
(2)
Figure 1 shows the influence of all three extraction parameters on the TPC in the obtained extracts. An increase in the TPC was recorded with increasing solvent concentration.
The antioxidant activity of the extracts was measured using the DPPH, CUPRAC, and FRAP assays. The ability of the extracts to neutralize DPPH radicals was measured to be from 3.94 to 39.81 mgTE/g. Extracts obtained at 60C in 30% methanol for 60 min provided the lowest antiradical activity toward DPPH radicals. On the other hand, lower temperature (40C) and time (30 min) in a combination with higher solv-ent concsolv-entration (70%) resulted in extracts with the highest potsolv-ential to neutralize DPPH free radicals. The antiradical activity toward DPPH radicals was highly influ-enced by linear term of temperature and solvent concentration. It was significantly influenced by the linear term of the extraction time as well as by cross product of tem-perature/time and temperature/solvent concentration.
The experimental measurements tend to correlate with the theoretical values that were confirmed by the value of coefficient of multiple determination of 0.9627 pre-sented in Table 3. In addition, the coefficient of variation was rather low, supporting the fitness of the model. The equation used in order to predict behavior of the extrac-tion system is given by
DPPH¼ 23:89 5:58X1þ 10:97X2 3:34X3þ 5:47X1X2 4:13X1X3
þ 0:038X2X3 1:31X12þ 0:43X22 1:96X32
(3)
Figure 2 shows the influence of the experimental conditions on the DPPH activity. An increase in temperature produces decreases in antiradical activity, while increases in the methanol concentration has an opposite effect.
Figure 1. Response surface contour plots showing the combined effects of the extraction parameters on the TPC.
The results obtained using the CUPRAC assay showed that the extracts possess high levels of antioxidant activity. Extraction of M. charantia with methanol at a concentra-tion of 30% at 60C for 60 min resulted in extracts with the lowest antioxidant ability based on the CUPRAC assay (33.80 mgTE/g). On the other hand, the utilization of higher methanol concentrations (70%) and prolonged extraction times (90 min) at 40C produced the extract with the highest activity according to CUPRAC (85.98 mg TE/g). As in the two previous assays, the calculated coefficient of multiple determination pro-duced a high value of 0.9627 which is a suitable indication for good fit of experimental and theoretical values (Table 3). To predict the behavior of the system, the following equation was generated:
CUPRAC¼ 50:25 5:08X1þ 16:84X2þ 2:97X3þ 1:88X1X2 7:92X1X3
þ 9:79X2X3þ 1:44X12þ 6:05X22þ 4:60X23
(4)
The analysis of p-value of linear, cross product, and quadratic terms suggested that the extraction of antioxidants responsible for the high antioxidant activity according to CUPRAC assay is highly influenced by the linear terms of solvent concentration and significantly influenced by the cross product term of solvent concentration and time. The influence of all investigated parameters is depicted by constructing the three-dimensional plot (Figure 3). The figure clearly shows that when the methanol concen-tration increases, the CUPRAC activity also increased. The same result was observed for the extraction time.
The third assay (FRAP) was conducted in order to determine the antioxidant poten-tial of the M. charantia extract. The results obtained from this experiment suggested that combination of extraction parameters (60C/30%/60 min) was found to be insig-nificant and thus, this condition was reflected the lowest reducing ability by the FRAP assay (26.40 mg TE/g). However, performing extraction at 40C with 70% methanol for 90 min produced extracts with highest reduction capacity according to the FRAP assay (52.75 mg TE/g). The reduction capacity of the obtained extracts was highly influenced by the linear term of the methanol concentration. The cross product term of the metha-nol concentration and extraction time also showed a significant influence on reduc-tion capacity.
According to Joglekar and May (1987), the R2 value should be at least 0.80 for an adequate model. In the case of FRAP assay, this value was 0.9295, which shows a high coefficient of multiple determination value, demonstrating good fit between the experi-mental and theoretical values. Moreover, the low value of the coefficient of variation
Figure 2. Response surface contour plots showing the combined effects of extraction parameters on the DPPH radical scavenger activity.
equal to 9.79 also suggests the fitness of the model. The influence of the investigated inputs on the reduction capacity (FRAP test) of the explored extracts may be described by
FRAP¼ 36:28 2:39X1þ 9:28X2þ 0:11X3þ 0:67X1X2 3:79X1X3
þ 5:06X2X3 0:22X21þ 2:74X22þ 1:35X32
(5)
The influence of the extraction conditions on FRAP results in Figure 4 clearly shows an increase in the activity by increasing the solvent concentration. The relationships between the FRAP results and experimental conditions are also summarized in
Figure 4.
ANN analysis
ANN has been widely used as a cutting-edge tool for the simulation and optimization of extraction of bioactive compounds from different matrices (Aliakbarian et al. 2012; Xi et al. 2013). In this study, ANN was used in combination with experimental design in order to produce the extract with the maximum yield of total phenols with highest antioxidant activity and to perform simultaneous optimization using GA methodology.
In ANN generated results are directly dependent on the initial parameters, but the number of hidden neurons also influences the ANN model outcomes. In current work, the number of neurons in the hidden layer was varied from 1 to 10 and the training process of each network was repeated five times with random initial values of weights and biases. Thus, the total number of created networks was 50 for all four measure-ments (TPC, DPPH, CUPRAC, and FRAP). From the total number of ANNs, for fur-ther analysis only ANNs with the coefficient of determination (R2) higher than 0.5 were used. Furthermore, based on the minimal SSer and maximal R2 values, the number of hidden layers was optimized.
The optimal number of hidden neurons for TPC was shown to be 10 (R2 of 0.9985; SSer of 0.1787). On the other hand, in case of antiradical activity against DPPH radicals, the optimal number of hidden neurons was 6 (R2 of 0.9929; SSer of 12.59). Furthermore, in case of the CUPRAC and FRAP antioxidant assays, the best fit was achieved using a neural network with 7 hidden neurons (R2of 0.9748, SSer of 125.2, R2 of 0.9556; SSer of 52.95, respectively). Although the higher number of neurons in hid-den layer increases the R2 values, the number of hidden neurons used for the optimiza-tion were much lower preventing possible overfitting.
Figure 3. Response surface contour plots showing the combined effects of extraction parameters on the CUPRAC antioxidant activity.
The comparison between experimental and calculated results is shown in Figure 5. The results show that predictive accuracy of the ANN models is very high.
The relative influence (RI) of temperature, methanol concentration, and time on the TPC, DPPH, CURAC, and FRAP outputs evaluated by
RIij½ ¼% Pn k¼0ðWikWkjÞ Pm i¼0ABS Pn k¼0ðWikWkjÞ 100% (6)
where RIij represents the relative importance of the ith input on the jth output, Wikis
the weight between the ith input and the kth hidden neuron, and Wkj is the weight
between the kth hidden neuron and the jth output. The mean values of calculated RI and standard deviations are presented in Figure 6.
Figure 6 shows for a relative influence exceeding 60%, the methanol concentration has the greatest influence on the TPC extraction. This result was in accordance with the RSM measurements that show the TPC extraction increased with solvent concentration. On the other hand, temperature and extraction time have a negative influence on the
Figure 4. Response surface contour plots showing the combined effects of extraction parameters on the FRAP antioxidant activity.
TPC extraction. Temperature (relative influence greater than 20%) has a greater effect on the TPC than the extraction time (relative influence of approximately 3%) which has a negligible effect.
Furthermore, for the DPPH assay, the ability of the extracts to neutralize free DPPH radicals increased with the methanol concentration. On the other hand, the temperature and extraction time show similar negative influence on the DPPH, so the antiradical ability of observed extracts against DPPH radicals decreased with increased temperature and extraction time. In the CUPRAC and FRAP reduction ability assays, the methanol concentration and time positively influenced the reduction ability, while temperature had negative influence. The reason for different influences of time for CUPRAC and FRAP in comparison to DPPH may be because CUPRAC and FRAP are antioxidant analyses while DPPH is anti-radical assay. The temperature had a similar influence on the CUPRAC and FRAP assays.
RSM and ANN optimization
The final step in this work was to find the optimal combination of examined extraction parameters in order to ensure the best extraction process. The optimization was per-formed in MATLAB using the ANN model and in Design Expert using the RSM model and was constrained within the experimental range. The RSM optimization was devel-oped in order to maximize the yields of TPC, DPPH, CUPRAC, and FRAP results with the highest level of importance for the all of the responses. For the final optimization result, the value for the three inputs, temperature, methanol concentration, and time, to provide the highest yield of TPC antioxidant activity were 20C, 70% methanol, and an extraction time of 52.2 min. Under these conditions, the measured value of TPC was 20.66, indicating highest content of these bioactive compounds in the extract obtained
under optimized conditions. In terms of antioxidant ability, the following results were obtained: DPPH 30.22 mg TE/g, CUPRAC 67.78 mg TE/g, and FRAP 45.48 mg TE/g. The DPPH, CUPRAC, and FRAP results confirmed the high antiradical and antioxidant activities of the extracts.
In addition, ANN was performed in order to calculate the maximal possible value for each output separately. Table 5 gives obtained optimal combination of experimental parameters for all four cases together with their predicted values. In both RSM and ANN optimization, the optimal value of methanol concentration was 70%. In case of temperature, the RSM suggested 20C which was optimal in the ANN optimization for TPC, CUPRAC and FRAP.
Conclusion
In recent years, optimization studies on plant extracts have great potential to provide high recovery of bioactive components. From this point, the present study was aimed to investigate the influence of temperature, time, and solvent concentration on the macer-ation extraction of bitter melon leaves. A comprehensive optimizmacer-ation study of bitter melon leaves was performed using RSM and ANN. In recent decades, RSM has been widely used to optimize extraction conditions for phytochemicals. In addition, ANN may be a better technique in terms of prediction and estimation capabilities.
High values of the coefficient of determination (R2) for all four investigated responses confirmed the accordance between experimentally obtained values and predicted values. In case of TPC, DPPH, FRAP, and CUPRAC, the values of coefficient of determination (R2) were 0.9576, 0.9627, 0.9295, and 0.9319, respectively, confirming the accordance between experimentally and predicted values of investigated responses. Insignificant val-ues for the lack of fit indicated good correlation between experimental data and applied mathematical model.
Depending on the extraction conditions, the content of extracted phenolic com-pounds were from 7.80 to 15.89 mg/g GAE and highly influenced by the linear term of solvent concentration and significantly influenced by the linear term of temperature as well as by the cross product of temperature and time. The scavenger capacity of the extracts was highly influenced by the linear temperature and solvent concentration terms, and significantly influenced by the linear extraction time term as well as by the cross product of the temperature/time and temperature/solvent concentration. On the other hand, the reduction capacity was highly influenced by the linear term of methanol concentration, while the cross product term of methanol concentration and extraction time showed significant influence on reduction capacity.
The optimization of extraction parameters was performed in MATLAB using the ANN model, and in Design Expert using the RSM model. The best conditions for RSM
Table 5. GA optimization results.
Parameter Temperature (C) Methanol concentration (%) Extraction time (min) Maximal value TPC 20 70 90 17.324 mg/g GAE DPPH 45.754 70 30.673 40.456 mg/g TE CUPRAC 20 70 90 91.117 mg/g TE FRAP 20 70 90 55.085 mg/g TE
were a temperature of 20C, methanol concentration of 70%, and extraction time of 52.2 min. The maximal values were: TPC 20.66 mg GAE/g, DPPH 30.22 mg TE/g, CUPRAC 67.78 mg TE/g, and FRAP 45.48 mg TE/g. In case of ANN, GA optimization results showed that 70% methanol was the optimal concentration of the solvent in case of all outputs, while 20C and a time of 90 min were optimal for TPC, CUPRAC, and FRAP.
According to the literature, this research can be considered as the first report to study the optimization of maceration extract from M. charantia leaves. It is anticipated the results reported herein will establish baseline data and also demonstrate that the sug-gested conditions can be applied in food and pharmaceutical industries.
ORCID
Sengul Uysal http://orcid.org/0000-0003-4562-1719
Aleksandra Cvetanovic http://orcid.org/0000-0001-5621-1788
Gokhan Zengin http://orcid.org/0000-0002-5165-6013
Mohamad Fawzi Mahomoodally http://orcid.org/0000-0001-6548-7823
References
Aktumsek, A.,. G. Zengin, G. O. Guler, Y. S. Cakmak, and A. Duran. 2013. Antioxidant poten-tials and anticholinesterase activities of methanolic and aqueous extracts of three endemic Centaurea L. species. Food and Chemical Toxicology 55:290–296. doi:10.1016/j.fct.2013.01.018. Aliakbarian, B., A. Fathi, P. Perego, and F. Dehghani. 2012. Extraction of antioxidants from
win-ery wastes using subcritical water. The Journal of Supercritical Fluids 65:18–24. doi:10.1016/ j.supflu.2012.02.022.
Anand, R., N. Verma, D. K. Gupta, S. C. Puri, G. Handa, V. K. Sharma, and G. N. Qazi. 2005. Comparison of extraction techniques for extraction of bioactive molecules from Hypericum perforatum L. plant. Journal of Chromatographic Science 43 (10):530–531. doi:10.1093/ chromsci/43.10.530.
Apak, R., K. G€uc¸l€u, M. €Ozy€urek, S. Esin Karademir, and E. Erc¸ag.2006. The cupric ion reducing antioxidant capacity and polyphenolic content of some herbal teas. International Journal of Food Sciences and Nutrition 57 (5/6):292–304. doi:10.1080/09637480600798132.
Chang, C. I., Chou, C. H. M. H. Liao, T. M. Chen, C. H. Cheng, R. Anggriani, R. C. P. Tsai, H. I. Tseng, and H. L. Cheng. 2015. Bitter melon triterpenes work as insulin sensitizers and insulin substitutes in insulin-resistant cells. Journal of Functional Foods 13:214–24. doi:10.1016/ j.jff.2014.12.050.
Dandawate, P. R., D. Subramaniam, S. B. Padhye, and S. Anant. 2016. Bitter melon: a panacea for inflammation and cancer. Chinese Journal of Natural Medicines 14 (2):81–100. doi:10.1016/ S1875-5364(16)60002-X.
Desai, K. M., S. A. Survase, P. S. Saudagar, S. S. Lele, and R. S. Singhal. 2008. Comparison of artificial neural network (ANN) and response surface methodology (RSM) in fermentation media optimization: case study of fermentative production of scleroglucan. Biochemical Engineering Journal 41 (3):266–73. doi:10.1016/j.bej.2008.05.009.
Fang, E. F., C. Z. Zhang, J. H. Wong, J. Y. Shen, C. H. Li, and T. B. Ng.2012. The MAP30 pro-tein from bitter gourd (Momordica charantia) seeds promotes apoptosis in liver cancer cells in vitro and in vivo. Cancer Letters 324 (1):66–74. doi:10.1016/j.canlet.2012.05.005.
Habicht, S. D., V. Kind, S. Rudloff, C. Borsch, A. S. Mueller, J. Pallauf, R. Yang, and M. B. Krawinkel. 2011. Quantification of antidiabetic extracts and compounds in bitter gourd vari-eties. Food Chemistry 126 (1):172–176. doi:10.1016/j.foodchem.2010.10.094.
Habicht, S. D., C. Ludwig, R. Y. Yang, and M. B. Krawinkel. 2014. Momordica charantia and type 2 diabetes: from in vitro to human studies. Current Diabetes Reviews 10 (1):48–60. doi:
10.2174/1573399809666131126152044.
Jia, S., M. Shen, F. Zhang, and J. Xie.2017. Recent advances in Momordica charantia: Functional components and biological activities. International Journal of Molecular Sciences 18 (12):2555. doi:10.3390/ijms18122555.
Keller, A. C., J. Ma, A. Kavalier, K. He, A. M. Brillantes, and E. J. Kennelly.2011. Saponins from the traditional medicinal plant Momordica charantia stimulate insulin secretion in vitro. Phytomedicine 19 (1):32–37. doi:10.1016/j.phymed.2011.06.019.
Kubola, J., and S. Siriamornpun.2008. Phenolic contents and antioxidant activities of bitter gourd (Momordica charantia L.) leaf, stem and fruit fraction extracts in vitro. Food Chemistry 110 (4):881–890. doi:10.1016/j.foodchem.2008.02.076.
Kultur, S. 2007. Medicinal plants used in Kırklareli province (Turkey.). Journal of Ethnopharmacology 111 (2):341–364.
Kultur, S., O. Altınbasak, S. Anıl, and G. Melikoglu. 2018. T€urkiye’de mide €ulserinde kullanılan tıbbi bitkiler. Marmara Pharmaceutical Journal 22 (1):1–14.
Lim, T. K.2012. Edible medicinal and non-medicinal plants. Vol. 1. New York: Springer
Joglekar, A. M., and A. T. May.1987. Product excellence through design of experiments. Cereal Foods World 32 (12):857–868.
Mahomoodally, M. F., A. G. Fakim, and A. H. Subratty. 2004. Momordica charantia extracts inhibit uptake of monosaccharide and amino acid across rat everted gut sacs in-vitro. Biological and Biological & Pharmaceutical Bulletin 27 (2):216–218. doi:10.1248/bpb.27.216. Oberoi, D. P. S., and D. S. Sogi.2017. Utilization of watermelon pulp for lycopene extraction by
response surface methodology. Food Chemistry 232 :316–321. doi:10.1016/ j.foodchem.2017.04.038.
Sarikurkcu, C. 2011. Antioxidant activities of solvent extracts from endemic Cyclamen mirabile Hildebr. tubers and leaves. African Journal of Biotechnology 10 (5):831–839.
Slinkard, K., and V. L. Singleton. 1977. Total phenol analysis: Automation and comparison with manual methods. American Journal of Enology and Viticulture 28 (1):49–55.
Wong, K. H., G. O. Li, K. M. Li, V. Razmovski-Naumovski, and K. Chan.2017. Optimisation of Pueraria isoflavonoids by response surface methodology using ultrasonic-assisted extraction. Food Chemistry 231:231–237. doi:10.1016/j.foodchem.2017.03.068.
Xi, J., Y. Xue, Y. Xu, and Y. Shen. 2013. Artificial neural network modeling and optimization of ultrahigh pressure extraction of green tea polyphenols. Food Chemistry 141 (1):320–326. Zhang, X., J. Chen, M. Mao, H. Guo, and Y. Dai. 2014. Extraction optimization of the
polysac-charide from Adenophorae Radix by central composite design. International Journal of Biological Macromolecules 67 :318–322. doi:10.1016/j.ijbiomac.2014.03.039.
Zhang, F., L. Lin, and J. Xie. 2016. A mini-review of chemical and biological properties of poly-saccharides from Momordica charantia. International Journal of Biological Macromolecules 92: 246–253. doi:10.1016/j.ijbiomac.2016.06.101.
