Rapid extraction of copper ions in water, tea, milk and apple juice by solvent-terminated dispersive liquid–liquid microextraction using p-sulfonatocalix (4) arene: optimization by artificial neural networks coupled bat inspired algorithm and response surface methodology

Abstract

A bat inspired algorithm with the aid of artificial neural networks (ANN-BA) has been used for the first time in chemistry and food sciences to optimize solvent-terminated dispersive liquid–liquid microextraction (ST-DLLME) as a green, fast and low cost technique for determination of Cu²⁺ ions in water and food samples using p-sulfonatocalix (4) arene as a complexing reagent. For this purpose, the influence of four important factors four factors which was influenced on the extraction efficiency such as salt addition, solution pH and disperser and extraction solvent volumes were investigated. Central composite design (CCD) as a comparative technique was employed for optimization of ST-DLLME efficiency. The ANN-BA optimization technique was regarded as a superior model due to its higher value of extraction efficiency (about 7.21%) compared to CCD method. Under ANN-BA optimal conditions, the limit of quantitation (S/N = 10), limit of detection (S/N = 3) and linear range were 0.35, 0.12 and 0.35–1000 µg L⁻¹, respectively. In these circumstances, the percentage recoveries for drinking tea, apple juice, milk, bottled drinking water, river and well water spiked with 0.05, 0.1 and 0.2 mg L⁻¹ of Cu²⁺ ions were in the acceptable range (91.4–107.1%). In comparison to other methods, the developed ST-DLLME method showed the lowest solvent and sample consumption, shortest value of extraction time, most suitable determination and detection limits and linear range with simple and low cost apparatus. Additionally, the use of bat inspired algorithm as a powerful metaheuristic algorithm with the aid of artificial networks is another advantage of the present work.

Electronic supplementary material

The online version of this article (10.1007/s13197-019-03892-6) contains supplementary material, which is available to authorized users.

Introduction

Copper as one of the most important heavy metals has found a large scale use in various industries and its pollution can caused many serious problems in high concentrations for human health and the environment (Chen 2012; Duruibe et al. 2007). In food and water samples, the preconcentration of trace heavy metal ions is needed before final assay. For this purpose, some pretreatment methods including liquid–liquid extraction (Karadaş and Kara 2014) and solid phase extraction (Ghaedi et al. 2013) were developed. The use of these methods has been restricted by the need for expensive laboratory equipment and high consumption of organic solvents. To overcome these problems, dispersive liquid–liquid microextraction (DLLME) as an efficient green and miniaturized microextraction technique was presented by Rezaee and coworkers (Rezaee et al. 2006). The extraction solvent in this technique is limited to the solvents with high density and also poisonous property which can be collected in aqueous solution via centrifugal separation. Chen and coworkers developed a solvent terminated dispersive liquid–liquid microextraction (ST-DLLME) using a low density extraction solvent (Chen et al. 2010). In the ST-DLLME method, the injection of demulsifier solvent like acetonitrile into the final turbid solution of aqueous sample and the extraction solvent was performed as an alternative approach to centrifugation. One of the main problems among scientists and researchers is to choose suitable optimization technique for obtaining the most appropriate response. Recently, metaheuristic algorithms as evolutionary optimization techniques were developed based on inspiration of nature, society and other very disciplined behaviors. Various metaheuristic algorithm such as genetic algorithm (GA) (Zheng et al. 2017), particle swarm optimization (PSO) (Ghaedi et al. 2015b) and bees algorithm (Ghaedi et al. 2015a) have been satisfactory employed for solving several areas of optimization problems. Bat inspired algorithm as a successful metaheuristic optimization method based on the mimicking of the microbat echolocation behavior was introduced by Yang and coworkers in 2010 (Yang 2010). This technique has been applied in restricted areas, and accordingly few reports have been published on this subject (Taha et al. 2013; Topal and Altun 2016).

In our previous projects, DLLME technique has been used for extraction of several analytes in various samples (Farajvand et al. 2018; Kiarostami et al. 2014; Maham et al. 2013a, b, c, 2014; Moradi et al. 2017). As far as we know, no work has been performed on the microextraction of Cu²⁺ ions by ST-DLLME with p-sulfonatocalix (4) arene as a complexing agent. Furthermore, a bat inspired algorithm with the aid of artificial neural networks (ANN-BA) approach has not been used in the field of chemistry and food sciences. So, this study describes the use of ANN–BA to optimize the ST-DLLME of Cu²⁺ ions with p-sulfonatocalix (4) arene as a ligand in water and food samples (river, well water, bottled drinking water, drinking tea and apple juice). In addition, a comparative optimization technique was also performed using central composite design.

Materials and methods

Chemical and reagents

All reagents were analytical grade and purchased from Merck. p-Sulfonatocalix (4) arene, also named 4-sulfonatocalix (4) arene (C₂₈H₂₄O₁₆S₄, P ≥ 97.0%) was obtained from Sigma-Aldrich. Copper ions stock solutions with concentration level of 1000 mg L⁻¹ were made by dissolution of CuCl₂ (Titrasol, Merck) in deionized water. Dilutions of heavy metal stock solutions in deionized water were carried out to make calibration solutions. pH adjustment was controlled by adding sodium hydroxide (0.1 M) and hydrochloric acid (0.1 M). The well and river water were sampled from agricultural areas (Islamshahr, Iran). In addition, all the other samples such as bottled milk (Damdaran, Iran), apple juice (Sunich, Iran), bottled drinking water (Vata, Iran) and black tea (Ahmad, England) were obtained from local market and employed to study the proposed ST-DLLME efficiency. A sample of drinking tea was traditionally made by putting 5 g of the tea leaves in 250 mL of boiling water. The real samples were stored individually in a food container and kept at 4 °C. Before the ST-DLLME procedure, they were centrifuged for 25 min at 4000 rpm and afterwards the supernatants were passed through a 0.45 µm membrane filter (Millipore Co, MA, USA). After filtration, the cleaned milk and apple juice were mixed with deionized water for dilution in the ratio of 1:4. Finally, the prepared aqueous samples were employed for ST-DLLME process.

Instruments

Varian spectra 240 fs (USA) atomic absorption spectrometer with an air-acetylene burner and deuterium background correction was used for the determination of copper ions. pH meter Jenway 3510 (United States of America) was used to determine the solution pH.

ST-DLLME procedure

To perform ST-DLLME procedure, a 10 mL of deionized water spiked with 0.5 mg L⁻¹ Cu²⁺ ions with adjusted pH of 6 was transferred into a 10 mL glass volumetric flask. To this solution, a triple mixture of 100 µL p-sulfonatocalix (4) arene (1 × 10⁻⁴ mol L⁻¹) as a complexing reagent, 225 µL toluene as an extraction solvent and 250 µL methanol as a disperser solvent was injected by a Hamilton syringe (1 mL) and a turbid solution was produced. Then, with injection of a demulsifier (0.5 mL, Acetonitrile) into the turbid solution, the phase separation occurred. A Hamilton syringe was used to collect the upper layer phase which was evaporated using a N₂ gas gentle stream. After the evaporation, 1 mL of 0.1 M nitric acid was added and the final solution nebulized into the flame atomic absorption spectrometer for Cu²⁺ ions analysis.

Optimization methods

For optimization of three independent factors such as disperser, extraction solvents types and extraction time, one variable at a time (OVAT) approach was applied. Multivariate optimization of other four variables such as salt addition, pH and the volumes of extraction and dispersive solvents were carried out by bat inspired algorithm with the aid of artificial neural networks (ANN-BA) and response surface methodology (RSM) with regard to the ST-DLLME efficiency.

Central composite design (CCD)

In this project, a CCD approach with four variables and three levels as a response surface methodology with α = 2 including 30 treatment combinations with eight axial points, six replicates at center point and 16 factorial points was performed using Design Expert (7) software. The low and high levels of four independent variables in CCD such as the volumes of extraction (V₁), disperser (V₂) solvents, pH (V₃) and salt addition (V₄) were 75 and 375 µL, 50 and 450 µL, 2 and 8 and 0 and 7.5%, respectively. The optimal conditions were obtained by solving the regression equation (Eq. 1) whilst the response was reached its maximum (Bajpai et al. 2012).

$$Y={\mu }_0+{\Sigma }_i{\mu }_i{v}_i+{\Sigma }_i{\mu }_{\mathit{ii}}{v}_i^2+{\Sigma }_i{\Sigma }_j{\mu }_{\mathit{ij}}{v}_i{v}_j+\eta$$ 1

where v_i and v_j are the coded independent variables, Y is the predicted response, µ_i, µ_ii, µ_ij, µ_o, n, and η are the linear coefficient, quadratic coefficient, interaction coefficient, model coefficient, number of factors, and model error, respectively.

The relation between coded and actual values is given by Eq. (2).

$$v_i=\frac{V_i-V_{{}_{\mathit{avg}}}}{V_{\text{max}i}-V_{\text{min}i}}$$ 2

where v_i is the ith coded independent factor, V_i is the ith uncoded independent factor, V_mini is the ith uncoded independent factor at the minimum level and V_maxi is the ith uncoded value at the maximum level.

Artificial neural networks-bat inspired algorithm (ANN-BA)

The nonlinear relation between the independent variables (input data) and the dependent variables (output data) was obtained by multilayer preceptrons (MLP) artificial neural network using MATLAB R2010a software. Training of the network was performed by using the levenberg–marquardt back propagation method. Tansig and pureline were utilized as activation functions in the hidden and output layers, respectively. The nonlinear equation between output and input variables is indicated in Eq. 3 (Pappu and Gummadi 2017).

$$\text{Re}sponse=Pureline(k_2\times tansig(k_1\times Inputvector)+{\beta }_1)+{\beta }_2$$ 3

where β₁, k₁ and β₂, k₂ are the biases and weights of middle and output layers, respectively. The percentage removal as a predicted response is shown by Response.

Bat inspired algorithm (BA) as a metaheuristic optimization algorithms was developed by Yang and coworkers in 2010 (Yang 2010). The algorithm is mimicked from the microbats echolocation behavior while catching their prey. Bat algorithm integrates the benefits of present algorithms, particularly harmony search and particle swarm optimization. When flying and hunting, bats produce sound pulses and listen for the returning echoes. The shapes, distance and the location of the prey were determined by using of the information obtained by the returning echoes (Chakri et al. 2017). The echolocation signals will enable bats to detect obstacles and insects. Therefore, bats can obtain a clear image from the surroundings (Topal and Altun 2016). The important aim of the bat echolocation is to identify the distance. Bats fly by using the delay between the sound production and the returning echoes from the environment. This behavior can be modelled by using mathematical relations as an optimization algorithm with the fitness function to be optimized (Yang and Papa 2016).

Yang suggested three rules for modelling of bat echolocation behaviors:

1. All microbats utilize echo sounding to notice distance, and they also sense the dissimilarities among background obstacles, prey and food in an anonymous approach.

2. Bats randomly move with a fixed frequency f_min and velocity v_i at position x_i, changeable wavelength λ and loudness A₀ to find for food. They can regulate the pulse production r from [0,1], relying on the closeness to their object.

3. The loudness of microbats can be changed in many different ways and consequently the loudness alters from A₀ (Large value) to A_min (minimum value).

In the bat algorithm, an artificial bat has its position x^t_i, velocity v^t_i, frequency f_i 0 [f_min + f_max], loudness A^t_i, and production pulse rate r^t_i, which are modified during the process. Initialization of the bat population is performed randomly. Beginning with a primary population of microbats spread over the space, the algorithm continues in iterations. The new solutions $x_i^t$ are carried out by transferring virtual bats as stated in the Eqs. 4–6.

$$f_i=f_{\text{min}}+\left(f_{\text{max}}-f_{\text{min}}\right)\beta ,$$ 4

$$v_i^t=v_i^{t-1}+\left(x_i^{t-1}-x_{\ast }\right)f_i,$$ 5

$$x_i^t=x_i^{t-1}+{\upsilon }_i^t,$$ 6

where β 0 [0,1] is a random vector obtained from a uniform distribution and x_* is the current global best solution (Yang and Papa 2016). While a microbat approaches closer to the victim, the pulses rate increases and the loudness decreases (Topal and Altun 2016). The pulse rate and loudness are shown in Eqs. 7 and 8, respectively.

$$A_i^{t+1}=\alpha A_i^t,$$ 7

$$r_i^{t+1}=r_i^0\left[1-exp-\gamma t\right],$$ 8

where γ and α are constants which can be set to 0.9. For easiness, the values of A_min and A₀ can be set at 0 and 1, respectively, where A_min = 0 means that a bat has just obtained the food (Yang and Papa 2016) and provisionally stopped the production of any sound (Yang and He 2013).

Results and discussions

Optimization methods

The extraction solvent type

For investigating the influence of extraction solvent type, several solvents with low density including cyclohexane, toluene, n-hexane, cyclohexanone and xylene were studied. Triplicate experiments of 10 mL sample were carried out under conditions of 300 µL ethanol, 100 µL extraction solvent, no salt addition and pH 5.5. Figure 1a reveals that the maximum efficiency was obtained significantly with toluene (p < 0.05, Single factor ANOVA). So, toluene was selected as the extraction solvent in the next experiments.

Fig. 1.

Effect of a type of the extraction solvent, b type of the dispersive solvent, c the extraction time on the ST-DLLME efficiency (Error bar was obtained based on standard deviation, N = 3)

The disperser solvent type

The disperser solvent type influences the extraction efficiency, viscosity of the solvent and generation of the droplet. Ethanol, methanol and acetone were studied as disperser solvents. As illustrated in Fig. 1b, methanol gave the maximum efficiency with significant difference (p < 0.05, Single factor ANOVA). Moreover, methanol had the minimum toxicity with the lowest price compared to others and consequently, was chosen as the disperser solvent in the further studies.

The time of extraction

The influence of time of extraction was also investigated at the time interval between 2 and 10 min time interval. Figure 1c, illustrates the percentage relative recovery of Cu²⁺ ions against the time of extraction. As presented in Fig. 1c, the optimum extraction was achieved at time of 6 min with the maximum extraction efficiency (p < 0.05, one-way ANOVA).

Central composite design (CCD) optimization

The CCD was utilized to optimize the disperser and extraction solvents volumes, solution pH and salt addition for ST-DLLME of Cu²⁺ ions from aqueous solutions. The quadratic response surface models were confirmed for parameters of regression by the analysis of variance (ANOVA) as shown in Table S1 (Supplementary Table 1). Table S1 indicates the high F-value (4.37) with a low value of probability (p < 0.001) for the model and consequently, the model was significant. According to the values of p, the V₂, V₃, V²₁ and V²₂ for Cu²⁺ ions were significant. The lack of fit and the rest of the variables and interactions were insignificant (p > 0.05). The coded quadratic equations based on the CCD analysis for Cu²⁺ ions are given by Eq. 9.

$$\begin{array}{cc}\hfill R& =88.78+1.14V_1+2.20V_2-1.47V_3-0.52V_4+0.41V_1{V}_2+1.78V_1{V}_3+0.34V_1{V}_4+0.89V_2{V}_3\hfill \\ \hfill & +0.93V_2{V}_4+0.79V_3{V}_4-2.18V_1^2-3.41V_2^2-0.97V_3^2-0.31V_4^2\hfill \\ \hfill \end{array}$$ 9

Based on the optimal conditions acquired from the CCD for multivariate optimization (250 µL methanol, 225 µL toluene, 3.75% salt addition and pH 6), the predicted response, experimental response and absolute error were 94%, 92.5% and 1.5 for Cu²⁺ ions, respectively.

Bat algorithm with the aid of artificial neural networks (ANN-BA)

ANN-BA was also used for optimizing the ST-DLLME of Cu²⁺ ions according to the nonlinear equations which, were obtained from artificial neural networks. Figure 2 shows the structure of multilayer artificial neural networks with a population of four variables such as volumes of disperser and extraction solvents, salt addition and pH that are outlined by the CCD. The values of bias and weight for each layer of ANN model were computed and corresponding equations for Cu²⁺ ions are presented as Eq. 10.

$$\text{Re}sponse=Pureline(K_2\times tansig(K_1\times [V(1);V(2);V(3);V(4)]+{\beta }_1)+{\beta }_2)$$ 10

where, β₁ and K₁ are the bias and weight of middle layer with 10 neurons, respectively, and β₂ and K₂ are the bias and weight of output layer with 1 neuron. The biases and weights for Cu²⁺ ions are presented in Table 1. These data were used to obtain an objective function for ANN–BA optimization technique with 1000 maximum number of iterations, 0.01 Lamda, 0.99 alpha, 0.4924 a₀, 0.5 r₀, q_max 0.5, q_min 0, q 0.0143 and 20 npop. The variations of the best fitness values against iterations are shown in Fig. 3. As illustrated in Fig. 3, after 712 iterations, the fitness value becomes 2.719 × 10⁻⁸ as the best fitness value and then remained constant. The percentage recovery (R %) for ST-DLLME of Cu²⁺ ions under optimal conditions obtained by ANN-BA (Extraction solvent volume 335 µL, Disperser solvent volume 250 µL, pH 5.5 and salt addition 3%) was found to be 99.81%. Therefore, the ANN-BA is considered as a better optimization method in comparison to CCD which gave a higher percentage removal of 7.21%.

Fig. 2.

The structure of three layer artificial neural networks

Table 1.

The weight and bias of each artificial neural networks layer

k1				kT2	β1	β2
0.7396	− 1.0614	− 0.9768	1.0042	0.4151	− 3.3291	− 0.1342
− 1.3894	1.0477	0.6961	− 1.7000	0.9458	1.8200
− 1.5306	− 0.7229	0.8584	− 1.7946	− 0.6000	1.9896
2.0568	1.0274	− 1.5983	− 1.2702	− 0.7708	− 0.0597
− 0.9253	0.8348	2.0857	1.3907	− 0.3095	0.5685
1.2793	1.4241	− 1.4394	− 0.2904	0.1832	0.3583
1.2859	0.1188	− 2.3915	1.4060	− 0.3984	0.3487
2.6033	− 0.5059	− 1.0399	− 0.0634	0.8027	0.9681
0.7846	2.4551	− 0.9461	0.3683	0.9511	1.3183
− 1.6453	− 0.0112	1.1070	1.5945	0.4696	− 2.3878

k₁, the weight matrix of hidden layer (10 × 4); k₂, the weight vector of output layer (1 × 10); k₂T, the matrix transpose of w₂, β₁, the bias vector of hidden layer (10 × 1); β₂, the scalar bias of output layer

Fig. 3.

The variations of the best fitness values versus iterations

Figures of merit

In accordance with ANN-BA optimal conditions, some important figures of merit for ST-DLLME method such as quantitation (S/N = 10) and detection (S/N = 3) limits, regression line, linear range and coefficient of determination (R²) in spiked water samples were acquired and the results are presented in Table S2. The results indicate the suitable values for figures of merit.

Study of real samples

Several real samples such as drinking tea, apple juice, milk, bottled drinking water, river water and well water were investigated to obtain microextraction efficiency and accuracy. Real samples were tested with spiking levels of 50, 100 and 200 µg L⁻¹ of Cu²⁺ ions. As presented in Table 2, the mean percentage recoveries with three replicate experiments for each sample were found to be in the range of 91.4 to 107.1% which were in the acceptable range with standard deviations of 0.11 to 0.41. Therefore, the suggested ST-DLLME method can be employed successfully for the determination of Cu²⁺ ions in real samples.

Table 2.

Percentage relative recoveries of Cu²⁺ ions in the spiked water, tea, milk and apple juice

Sample	Initial concentration (µg L−1)	Spiking (µg L−1)	Found (µg L−1), means ± s (N = 3)	Relative recovery (%)
Bottled drinking water	N.D.	50	46.9 ± 0.34	93.80
Bottled drinking water	N.D.	100	98.9 ± 0.14	98.90
Bottled drinking water	N.D.	200	212.1 ± 0.34	106.10
River water	N.D.	50	45.7 ± 0.25	91.40
River water	N.D.	100	98.5 ± 0.17	98.50
River water	N.D.	200	197.5 ± 0.12	98.75
Well water	5	50	50.9 ± 0.12	91.80
Well water	5	100	103.1 ± 0.16	98.10
Well water	5	200	214.9 ± 0.24	104.95
Milk	N.D.	50	45.7 ± 0.18	91.40
Milk	N.D.	100	97.7 ± 0.15	97.70
Milk	N.D.	200	214.2 ± 0.26	107.10
Tea	N.D.	50	46.6 ± 0.26	93.20
Tea	N.D.	100	93.1 ± 0.28	93.10
Tea	N.D.	200	197.8 ± 0.11	98.90
Apple juice	N.D.	50	46.6 ± 0.18	93.20
Apple juice	N.D.	100	97.1 ± 0.41	97.10
Apple juice	N.D.	200	196.4 ± 0.39	98.20

Extraction conditions: extraction solvent and its volume, toluene, 335 µL, dispersive solvent and its volume, methanol, 250 µL, pH, 5.5, extraction time, 6 min and 3% salt addition s- Standard deviation (n = 3), N.D. Not detected

Comparison with other methods

As presented in Table 3, several methods for the determination of Cu²⁺ ions that included in the literatures were compared to the proposed ST-DLLME based on the figures of merit and type of optimization techniques. Amongst the data that are revealed in Table 3, the proposed ST-DLLME method in comparison to other methods, used the lowest solvent and sample consumption, most suitable determination and detection limits and linear range. Unlike the conventional DLLME method, in ST-DLLME, there is no need for any centrifugation step to phase separation and consequently, the extraction time is much shorter. As illustrated in Table 3, in comparison to other methods, the proposed method can be recognized as a low cost technique due to the use of simple, inexpensive equipment and the lowest consumption of liquid solvents. In addition, one variable at a time (OVAT) optimization technique (OVAT) was used in the most presented methods in Table 3. In the presence of interactions between the variables, OVAT method is not an effective way to find optimal conditions. Chemometric optimizations methods such as metaheuristic algorithm and experimental design can be used successfully for solving this difficulty. Consequently, the suggested method showed the suitable potential of ANN–BA and CCD as chemometric for evaluation of the optimal conditions.

Table 3.

Comparison of characteristic data for ST-DLLME of Cu²⁺ ions and other methods

Method	Sample type	Optimization method	LOD (µg L−1)	LOQ (µg L−1)	RSD (%)	Sample volume (mL) organic solvent volume (µL)	Extraction time (min)	LDR (µg L−1)	References
aSPE using bTw and FAAS	Tobacco and water sample	kjOVAT	0.4	2	1.9	(50 mL–5000 µL)	4	2–900	Mohammadi et al. (2016)
cSE using dOVAC and FAAS	Food and water sample	OVAT	0.26	2	–	10 mL–6700 µL	–	2–8	Karadaş and Kara (2013)
eCPE using fPAR and gICP-OES	Water sample	OVAT	1.2	10	1.3–1.9	15 mL–500 µL	50	10–500	Silva et al. (2009)
hNano-TiO2–MBT and FAAS	Water and ore samples	OVAT	0.12	0.2	4.63–3.54	250 mL and 3000 µL	10	0.2–20	Pourreza et al. (2014)
iCAF-SPE using ZrO2/B2O3 and FAAS	Tea and Tab water	OVAT	3.3	11	1.9	50 mL–10,000 µL	5	11–5000	Yalçınkaya et al. (2011)
ST-DLLME usins jPSC and FAAS	Water and food samples	lANN–BA and mCCD	0.12	0.35	0.5–2	10 mL–335 µL	6	0.35–1000	This work

^aSolid phase extraction. ^bTween 80. ^cSolvent extraction. ^d8-Hydroxy-2-quinoline carboxaldehyde. ^eCloud point extraction. ^f4-(2-pyridylazo)-resorcinol. ^gInductively coupled plasma optic emission spectrometry. ^hNano-TiO₂ modified with 2-mercaptobenzothiazole. ⁱChelating agent free-solid phase extraction. ^jp-Sulfunatocalix (4) arene, ^kOne variable at a time. ^lBat algorithm with the aid of artificial neural networks. ^mCentral composite design

Conclusion

The present ST-DLLME method for extraction of Cu²⁺ ions can be regarded as a safe and green method due to a very low consumption of hazardous solvents (335 µL) and low volume of sample (10 mL) with suitable figures of merit and acceptable percentage relative recoveries. In addition, the suggested method was performed with simple equipment and low consumption of solvent volume in a short period of time along with the application of a powerful bat metaheuristic optimization algorithm for the first time in food science and chemistry. Thus, the developed ST-DLLME method followed by flame atomic absorption spectrometer as a most commonly used instrument in all analytical laboratories can be considered as an inexpensive, highly accurate and fast method for analysis of Cu²⁺ ions in aqueous samples.

Electronic supplementary material

Below is the link to the electronic supplementary material.