Simultaneous determination of three organophosphorus pesticides residues in vegetables using continuous-flow chemiluminescence with artificial neural network calibration

Talanta 72 (2007) 223–230

Simultaneous determination of three organophosphorus pesticides residues in vegetables using continuous-flow chemiluminescence with artificial neural network calibration Baoxin Li ∗ , Yuezhen He, Chunli Xu Department of Chemistry, School of Chemistry and Materials Science, Shaanxi Normal University, Xi’an 710062, PR China Received 23 June 2006; received in revised form 8 October 2006; accepted 17 October 2006 Available online 20 November 2006

Abstract In this article, a continuous-flow chemiluminescence (CL) system with artificial neural network calibration is proposed for simultaneous determination of three organophosphorus pesiticides residues. This method is based on the fact that organophosphorus pesticides can be decomposed into orthophosphate with potassium peroxodisulphate as oxidant under ultraviolet radiation and that the decomposing kinetic characteristics of the organophosphorus pesticides with different molecular structure are significantly different. The produced orthophosphate can react with molybdate and vanadate to form the vanadomolybdophosphoric heteropoly acid, which can oxidize luminol to produce intense CL emission. The CL intensity of the solution was measured and recorded every 2 s in the range of 0–250 s. The obtained data were processed chemometrically by use of a three-layered feed-forward artificial neural network trained by back-propagation learning algorithm, in which input node, hidden node and output nodes were 65, 21 and 3, respectively. The proposed multi-residue analysis method was successfully applied to the simultaneous determination of the three organophosphorus pesticides residue in some vegetables samples. © 2006 Elsevier B.V. All rights reserved. Keywords: Chemiluminescence; Organophosphorus pesticides; Artificial neural network; Simultaneous determination

1. Introduction Organophosphorus pesticides are essential in modern agriculture, and are now widely used to control pests and to increase harvest productivity. However, it is well known that they do have a high acute toxicity due to prevention of neural impulse transmission by their inhibition of cholinesterase [1]. Because of the potentially dangerous effects on human health, the control of pesticide residue in food is of great importance in order to minimize risk to consumers. So, the fast, reliable and economically viable methods are required for their detection in the environment and in agro-food products. Many methods have been developed in the last few years for the determination of organophosphorus pesticides. The most widely used methods are gas chromatography (GC) [2–4], liquid chromatography (LC) [5–7] and biosensor [8–10]. The classical GC methods have not been satisfactory in general, due to the

∗

Corresponding author. Tel.: +86 29 85300986; fax: +86 29 85307774. E-mail address: [email protected] (B. Li).

0039-9140/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.talanta.2006.10.023

thermal instability of the molecules of these compounds [11], and S- or P-detectors are required. In high performance liquid chromatography (HPLC), the usual UV or electrochemical detectors, which are utilized in the analysis of water samples of different origin as well as of samples from soils and oils, cannot provide enough sensitivity to detect the organophosphorus pesticides residue in the samples. Nowadays, mass spectrometry (MS) is often used as the GC and LC detector for identification and determination of mixtures of organophosphorus pesticides in the developed countries [12,13]. However, the expensive GC–MS and LC–MS instrumentation are rather few and cannot widely used for analysis of organophosphorus pesticides in the developing countries, yet the need for pesticide analysis for the developing countries is arguably even greater than that for the developed countries because agriculture production mainly concentrates in developing countries; furthermore, these chromatographical methods require trained staff, complicated sample pretreatments and are often not suitable for field analysis. The biosensor has now emerged as a popular alternative technique for organophosphorus pesticides, and most of the biosensors are based on the inhibition action of organophosphorus pesticides

224

B. Li et al. / Talanta 72 (2007) 223–230

on cholinesterases, and the biosensors are rather rapid, sensitive and suitable for analysis of organophosphorus pesticides on-line and on site. However, the biosensors supply the total amount of organophosphorus pesticides in samples, and cannot distinguish the different kinds of organophosphorus pesticides; the inhibition action of every organophosphorus pesticide on the same cholinesterases is remarkably different [14], and when different organophosphorus pesticide is used as the standard solution, the amount of organophosphorus pesticides in same sample is different with such same biosensor; some heavy metal ions and proteins also inhibit the activity of cholinesterases, so the biosensor lacks enough selectivity [1]; the inhibition action of organophosphorus pesticides on cholinesterases is irreversible, so the lifetime of the biosensor is rather short and the measurement cost increases. Thus, development of reliable and fit-forpurpose methods of analysis for mixtures of organophosphorus pesticides with the use of simple and relatively inexpensive instrumentation is an appropriate objective to strive for. Chemiluminescence (CL) has been known as a powerful and important analytical technique because the analytical performance of CL detection is better than that of other common spectroscopic detection methods (such as spectrophotometry and fluorescence), higher sensitivity, lower detection limits and wider linear ranges can be achieved with simpler and cheaper instrument (no excitation light source and no spectral resolving system) [15,16]. CL methods have been applied to the determination of pesticides [14,17]. Based on the catalytic effect of organophosphorus pesticides on the luminol–H2 O2 CL reaction, the luminol–H2 O2 CL system has been used for the detection of dichlorvos [18], parathion [19], monocrotophos [20] and methyl-parathion [21], respectively. So, these CL methods for organophosphorus pesticides could be applied to the single-residue samples, and the multi-residue samples could not be detected using these CL methods. Nowadays, about 36 organophosphorus pesticides are in use in China [22], and two or three organophosphorus pesticides are often simultaneously used in agriculture production. Therefore, multi-residue analysis method of organophosphorus pesticides is far more important for food safety and environmental protection. In recent years, the use of some analytical methods combined with multivariate calibration can be considered a promising, faster, direct and relatively less expensive alternative for the multicomponent analysis of mixture [23]. In this kind of situation, where the direct determination of a analyte is difficult due to the presence of one or several other constituents, instead of eliminating the interfering species, e.g. by a separation procedure, the use of multivariate calibration makes possible the quantification of these interferences along with the analyte. Artificial neural networks (ANN) based on artificial intelligence is powerful non-parametric non-linear modeling techniques [24–27]. The ANN calibration is able to acquire information and provide models even when the information and data are complex, noise contaminated, nonlinear and incomplete. They can model nonlinear systems with no prior knowledge of the system concerned or reaction order and reaction rate coefficient of the involved analytical system. This ability makes them particularly attractive for calibration in kinetic mixture

resolution [28]. Simultaneous determination of closely related species is of continuing analytical interest and methods developed for resolving them are important. Multicomponent kinetic determinations are able to resolve these systems by using differences of behavior with respect to a common reagent [29]. The kinetic methods of analysis with some advantages such as selectivity, sensitivity, and determination of species that cannot be resolved by equilibrium-based methods have been developed. These methods can be associated with ANN to resolve multicomponent kinetic systems without requiring their prior separation [30]. This paper describes a simple continuous-flow CL system combined with ANN calibration for simultaneous determination of three organophosphorus pesticides in ternary mixtures without previous separation. This method is based on the fact that organophosphorus pesticides can be decomposed into orthophosphate with potassium peroxodisulphate as oxidant under ultraviolet radiation [31] and that the decomposing kinetic characteristics of the organophosphorus pesticides with different molecular structure are significantly different. The produced orthophosphate can react with molybdate and vanadate to form the yellow vanadomolybdophosphoric heteropoly acid (the chemical formula H4 [PMo11 –VO40 ]·xH2 O), which can oxidize luminol to produce intense CL emission [32]. The main organophosphorus pesticides are commonly classified into P–C band organophosphorus pesticides, P–O band organophosphorus pesticides, P–S band organophosphorus pesticides and P–N band organophosphorus pesticides. In the north of China, the three organophosphorus pesticides (P–C, P–O and P–S) are widely used. So, in this paper we chose the dipterex, dichlorvos and omethoate as model molecules of P–C band organophosphorus pesticides, P–O band organophosphorus pesticides and P–S band organophosphorus pesticides, respectively. The CL intensity was measured and recorded at 2-s intervals during the reaction, and the obtained data were processed by ANN calibration. The proposed multi-residue analysis method was successfully applied to the simultaneous determination of the three organophosphorus pesticides residue in some vegetables samples. 2. Theoretical background Assume that n analytes, Mi (i = 1, 2, . . ., n), react with a common reagent R to give the same product P, according to the following scheme: kMi

Mi + R−→P

(i = 1, 2, . . . , n)

(1)

If the concentration of R is much larger than that of the analytes, the reaction can be conformed to the first or pseudo-firstorder kinetics [33], and thus its rate equation can be represented as dCP = k1,t CM1 ,t + k2,t CM2 ,t + · · · + kn,t CMn ,t dt (t = 1, 2, . . . , s)

(2)

B. Li et al. / Talanta 72 (2007) 223–230

225

where, CMi ,t is the concentration of Mi at time t, ki,t the rate constant of Mi , and s is the total number of time values at which measurements are made. For this CL system, the CL intensity increases during measurement time. The CL intensity (I) measurement versus time can be expressed as: It = b1,t CM1 + b2,t CM2 + · · ·bi,t CMi + · · · + bn,t CMn + b0t (t = 1, 2, · · ·, s)

(3)

where bi,t is the proportional coefficient for component CMi at a time t, and b0t is the corresponding background. Let CM0 = 1, and then I0 = b0,t CM0 . Now, Eq. (3) can further be simplified as: It =

n 

bi,t CMi

(t = 1, 2, · · ·, s)

(4)

i=0

If m standard samples are prepared, Eq. (4) can be expressed in matrix form: Im×s = Cm×(n+1) B(n+1)×s

(5)

where the first row in matrix B represents the background vector. According to this equation it is possible to determine the component by a suitable chemometric method. In this paper, the experimental data were collected from experiments and then processed by the ANN calibration method. 3. Experiment

Fig. 1. Schematic diagram of continuous-flow CL system for simultaneous determination of omethoate, dichlorvos and dipterex. P1 and P2 , peristaltic pump; UV, UV photo-reactor; D, detector; PC, personal computer; (a) pesticides + K2 S2 O8 ; (b) NH4 VO3 + (NH4 )6 Mo7 O24 ; (c) luminol.

The photoreactor-lamp assembly was housed in a fan ventilated metal box. The flow cell is a flat spiral-coiled colorless glass tube (i.d. 1.0 mm; total diameter of the flow cell, 3 cm, without gaps between loops) and placed close to the window of the photomultiplier tuber (PMT). The CL signal produced in the flow cell was collected with a CR-105 PMT (Hamamatsu, Japan) of the ultra-weak Chemiluminescence Analyzer (Institute of Biophysics, Chinese Academy of Sciences, Beijing). The signal was recorded using an IBM-compatible computer, equipped with a data acquisition interface. Date acquisition and treatment were performed with BPCL software running under Windows 95. The data pretreatment was done with MATLAB for windows (Mathworks, version 6.1). The ANN program for calibration prediction and experimental design was written in MATLAB 6.1 according to the algorithm described by Xu et al. [34].

3.1. Chemicals and reagents

3.3. Procedures

All chemicals used were of analytical grade unless stated otherwise, and the water used throughout was deionized and double distilled. Pesticide standards including dipterex, dichlorvos and omethoate were obtained from Shaanxi Huawei Pesticide Co. Ltd. (Xi’an, China). KH2 PO4 , Na2 B4 O7 , K2 S2 O8 , NH4 VO3 , (NH4 )6 Mo7 O24 ·4H2 O, H2 SO4 and NaOH were obtained from Xi’an Chemical Plant (Xi’an, China). A 5 × 10−2 mol l−1 luminol stock solution was prepared by dissolving 9.32 g of luminol (Shaanxi Normal University, PR China, >95%) in 20 ml of 0.1 mol l−1 NaOH and then diluting to 1 l with water. The luminal solution was stored in the dark for 24 h prior to use ensure that the reagent properties had stabilized.

Transfer 0.25 ml of 0.4 mol l−1 Na2 B4 O7 solution and 2.50 ml of 0.1 mol l−1 K2 S2 O8 solution into a volumetric flask with the appropriate amounts of dipterex, dichlorvos and omethoate (or the treated sample solution), and dilute to 25 ml. The mixture solution was placed in the photo-reactor. As shown in Fig. 1, flow lines were inserted into mixture solution and luminol solution, respectively. The UV lamp and the pump were started at the same time, and the decomposed pesticide stream was merged with the mixture stream (NH4 VO3 + (NH4 )6 Mo7 O24 + H2 SO4 ), and then merged with luminol stream just prior to reaching a spiral flow cell, producing CL emission. The CL intensity of the solution was recorded every 2 s in the range of 0–250 s.

3.2. Apparatus and software

4. Results and discussion

The continuous-flow CL system used in this work is shown in Fig. 1. There are two HL-2 type peristaltic pumps (Shanghai Huxi Analytical Instrument Plant, China): one was used to deliver the stream (sample + K2 S2 O8 ) and the stream (NH4 VO3 + (NH4 )6 Mo7 O24 + H2 SO4 ) at a flow rate of 0.8 ml min−1 (per tube), and another was used to deliver the luminol stream at a flow rate of 0.8 ml min−1 (per tuber). PTFE tubing (0.8 mm i.d.) was used as connection material in the flow system. The UV source was a rod-shaped high pressure mercury lamp (400 W, Philips) that had a major emission line at 254 nm.

4.1. Photo-oxidation kinetics of organophosphorus pesticides Preliminary investigations showed that the UV irradiation of solutions of dipterex, dichlorvos and omethoate in the presence of peroxydisulphate or TiO2 led to the conversion of the organophosphorus compounds into orthophosphate, and that the photochemical reaction using peroxydisulphate as oxidant was faster possibly because the rate of homogenous reaction is higher than its of heterogeneous reaction (using TiO2 as photo-oxidant).

226

B. Li et al. / Talanta 72 (2007) 223–230

Fig. 2. Kinetic data for omethoate, dichlorvos, dipterex and the reagent blank. Omethoate, 5 × 10−7 g ml−1 ; dichlorvos, 5 × 10−7 g ml−1 ; dipterex, 5 × 10−7 g ml−1 ; K2 S2 O8 , 0.05 mol l−1 ; NH4 VO3 , 4 × 10−4 g ml−1 ; 5 × 10−4 mol l−1 ; H2 SO4 , 0.01 mol l−1 ; luminol, (NH4 )6 Mo7 O24 , 5 × 10−4 mol l−1 ; NaOH, 0.06 mol l−1 .

So, we chose peroxydisulphate as the oxidant. The strong acid or alkaline peroxydisulphate solution was suitable for decomposing organophosphorus compound, but problem arose with the use of peroxydisulphate in acid medium due to the noise caused by gas bubbling generated within sample solution. Alkaline K2 S2 O8 has the advantage that the CO2 generated by the photo-oxidation of the organophosphorus pesticide is predominantly in the carbonate form and hence problem of bubble is resolved. Sodium tetrahydroborate was selected as photooxidation media. In this system, organophosphorus pesticide is photodegradated quantitatively to PO4 3− , and the produced PO4 3− can react with molybdate and vanadate to form the yellow vanadomolybdophosphoric heteropoly acid, which can oxidize luminol to produce intense CL emission. So, the CL signal of the luminol system can be used to character the process of the photodegradation reaction. Fig. 2 shows the kinetic of the reaction of each of the three pesticides in the presence of excess amounts of other reagents (such as K2 S2 O8 and luminol). The experimental results showed that the reaction rates of the three pesticides were different. In addition, it can also be seen that the intensity response of the ternary mixture of the pesticides is lower than the sum of the intensities recorded for each individual pesticide, probably because of the synergy action of

the three pesticides in the photo-reaction. This observed difference in the kinetic behavior of the three pesticides result form the difference of their molecule structures (Table 1). The order of photo-degradation rate of the three pesticides was dipterex > dichlorvos > omethoate. For omethoate, almost no CL signal was detected at beginning of degradation (0–60 s), possibly because omethoate could not directly transform to orthophosphate, and omethoate would firstly transform to one intermediate, which could not react with molybdate and vanadate to form the vanadomolybdophosphoric heteropoly acid to oxidize luminol, producing CL emission. The different reaction rates in the photo-degradation reaction with K2 S2 O8 provide the possibility for resolving their mixtures and enabling their quantitative analysis. In this system, we chose the ANN calibration for simultaneous determination of the three organophosphorus pesticides. 4.2. Optimization of the photo-reaction conditions In this system, K2 S2 O8 is oxidant in the photo-degradation of the organophosphorus pesticides, and the concentration of K2 S2 O8 affects the redox reaction. At first, the concentration of K2 S2 O8 must be much more than that of pesticides in order that the reaction can be conformed to the first or pseudofirst-order kinetics. The results showed that the CL intensity increased with increasing K2 S2 O8 concentration, and at the K2 S2 O8 concentration of 0.01 mol 1−1 , there is a relatively high CL intensity of three pesticides. Thus, a K2 S2 O8 concentration of 0.01 mol l−1 was selected for simultaneous determination of dipterex, dichlorvos and omethoate. In this photo-reaction, Na2 B4 O7 is the media [31], so the effect of Na2 B4 O7 concentration on the CL intensity was also investigated. It was found that the optimal concentration of was 0.004 mol l−1 . Therefore, 0.004 mol l−1 Na2 B4 O7 was chosen as the reaction media (the pH was about 10.3) in this photo-reaction. 4.3. Optimization of forming condition of vanadomolybdophosphoric heteropoly acid The concentrations of ammonium molybdate, ammonium vanadate and H2 SO4 affect the form ratio of vanadomolyb-

Table 1 Chemical structure of the three organophosphorus pesticides Pesticides

Molecular formula

Omethoate (OME)

C5 H12 O4 NSP

Dichlorvos (DIC)

C4 H7 O4 PCl2

Dipterex (DIP)

C4 H8 O4 PCl3

Structure

B. Li et al. / Talanta 72 (2007) 223–230

227

Table 2 Concentration data for different mixtures used for calibration of omethoate, dichlorvos and dipterex Mixture

Omethoate (×10−7 g ml−1 )

Dichlorvos (×10−7 g ml−1 )

Dipterex (×10−7 g ml−1 )

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.83 0.83 0.83 0.83 1.76 1.76 1.76 1.76 4.25 4.25 4.25 4.25 11.80 11.80 11.80 11.80

0.77 1.82 5.67 9.36 0.77 1.82 5.67 9.36 0.77 1.82 5.67 9.36 0.77 1.82 5.67 9.36

0.60 1.54 3.95 8.82 1.54 0.60 8.82 3.95 3.95 8.82 0.60 1.54 8.82 3.95 1.54 0.60

dophosphoric heteropoly acid, on which the oxidation activity of vanadomolybdophosphoric heteropoly acid depends. So the effect of ammonium molybdate concentration was studied over 5 × 10−5 to 1 × 10−3 mol l−1 . The results showed that the CL response increased up to 7 × 10−4 mol l−1 ammonium molybdate, above which the response decreased due to a low acid/molybdate ration [35]. At the same time, the effect of NH4 VO3 concentration and H2 SO4 concentration on the CL intensity, and the experimental results showed that the optimal concentrations of NH4 VO3 and H2 SO4 were 0.043 and 0.01 mol l−1 , respectively. 4.4. Optimization of CL reaction condition As the luminescence reagent in this system, the luminol concentration affects the response. The effect of luminol concentration on the determination of the pesticides was studied over the range 1 × 10−5 to 4 × 10−3 mol l−1 . The results showed that the CL intensity increased from 1 × 10−5 to 5 × 10−4 mol l−1 , but no appreciable increase in CL intensity was observed above this concentration due to saturation. A luminol concentration of 5 × 10−4 mol l−1 was therefore used for all subsequent experiments. The CL response varied with the age of the luminol solution [36] and therefore it was always prepared 24 h before use. Luminol chemiluminescence is particularly dependent on the reaction pH [37]. In view of the nature of luminol CL reaction, which is more favoured under basic conditions, an alkaline medium was introduced to improve the sensitivity of the system. Several medium solutions such as NaOH, NaHCO3 –Na2 CO3 , Na2 B4 O7 –NaOH and NH3 –NH4 Cl were studied. The result showed that the strong and stable CL signal was obtained in the NaOH. In the experiments, the alkalinity of the reaction medium was adjusted by preparing luminol with a suitable concentration of sodium hydroxide. The effect of NaOH concentration was also studied in the range 0.02–0.12 mol l−1 . It was found that 0.07 mol l−1 NaOH was optimum reaction media and chosen for further work. In this system, vanadomolybdophosphoric

heteropoly acid formed in 0.01 mol−l H2 SO4 media, so some NaOH was first needed to neutralize the acid, resulting in higher concentration NaOH in this system. 4.5. Multivariate calibration Multivariate calibration methods such as ANN require a suitable experimental design of the standard belong to the calibration set in order to provide good prediction. Two sets of standard solutions were prepared as calibration set and prediction set. The former was used to train the network and the latter was used to validate the learned network. Base on our primary experiment, the CL intensity versus the pesticides concentration was linear in the range 1 × 10−8 to 5 × 10−6 g ml−1 . The calibration set consisting of 16 standards with different concentrations of omethoate, dichlorvos and dipterex was used. Their concentrations are shown in Table 2. For prediction set, 12 test mixtures in Table 4 were used, and the concentration of the three pesticides solution covered the whole concentration range of the three pesticides in calibration set. The CL intensity of each ternary mixture solution was detected every 2 s from 0 to 250 s. An experimental calibration data matrix with 16 by 125 was prepared. The kinetic data were processed by ANN, which was trained with the back-propagation of errors learning algorithm. Its basic theory and application to chemical problems can be found in the literature [38]. The neural network performs a non-linear iterative fit of data. The structure of the network is comprised of three node layers: an input layer, a hidden layer and output layer. The node in the input layer transferred the input data to all nodes in the hidden layer. These nodes calculate a weighted sum of the inputs that is subsequently subjected to a non-linear transformation:   l  Oj = f (6) si wij i=1

where si is the input to the node i in the input layer, l the number of nodes in the input layer, wij (weights) are the connection

228

B. Li et al. / Talanta 72 (2007) 223–230

Table 3 Statistical data for the optimization matrix using the ANN model Item

Value

Input nodes Hidden nodes Output nodes Number of iterations Hidden layer transfer function Output layer transfer function

65 21 3 5500 Sigmoid Purelin

between each node i in the input layer and each node j in the hidden layer, Oj the output of node j in the hidden layer, and f is a non-linear function. The output of the network is a weighted sum of the outputs of the hidden layer and it is the calculated concentration. During training (calibration) of the network, the weights are iteratively calculated in order to minimize the sum of the squared difference between the known concentration and the calculated concentration. Overfitting is avoided by using two sample sets; thus weights are calculated from a calibration set while the concentration of another sample set (the test set) is being simultaneously predicted. In addition, the number of the data values used for training must exceed that of weights determined in the network; this entails using a large number of samples for calibration if the number of input variables is also large. In this work, the CL intensity data versus the time were centred and normalized with pressed function in MATLAB as the input for ANN, and the output of the network were the calculated concentrations related to the CL intensity data. For the optimized model, three parameters were selected to test the prediction ability of the chemometric model for each component. The root mean square difference (RMSD), the square of the correlation coefficient (R2 ) and the relative error of prediction (REP), which can be calculated for each component as:  n 0.5 1 2 RMSD = (ˆci − ci ) (7) n i=1

n (ˆci − c¯ i )2 R = i=1 n ¯ i )2 i=1 (ci − c 2

 REP (%) =

100 c¯ i

(8)

0.5   n 1 (ˆci − ci )2 n

(9)

i=1

where ci is the true concentration of the analyte in the sample i, cˆ i represented the estimated concentration of the analyte in the sample i, c¯ i the mean of the true concentration in the prediction set, and n is the total number of sample used in the prediction set. To enhance the prediction ability of ANN, neural network models were optimized. Function sigmoid hidden layer functions that can be used to model a variety of relationships were found to be optimum for calculation. The different number of input nodes was changed to sieve the data. The result showed that the relative errors of both calibration set and prediction set gradually decreased when the number of input nodes was increased. When the number of input nodes was 65, the network had the highest degree of approximation. When the number of input nodes exceeded 65, the relative standard error of the calibration set decreased and that of the prediction set increased. Thus the degree of approximation evidently decreased. This result indicates an overfitting phenomenon is affecting the network. Because there are three kinds of pesticides in sample, the output layer contained three neurons. The number of hidden nodes also has great effect on the predictive result. The proper number of nodes in the hidden layer was determined by training ANN with different numbers of nodes in the hidden layer. A minimum in RMSE occurred when 21 nodes were used in the hidden layer. Continued training beyond 5500 iterations frequently resulted in a slight increase in error of prediction as the learning iteration increased whereas the error of calibration leveled off or continued to increase slightly. The construction of these ANN models is summarized in Table 3. From the CL intensity data of the calibration sets of the above mixtures, whose concentrations were selected randomly,

Table 4 Prediction results for synthetic mixtures of omethoate, dichlorvos and dipterex Synthetic mixture

1 2 3 4 5 6 7 8 9 10 11 12 RMSD R2 REP (%)

OME (×10−7 g ml−1 )

DIC (×10−7 g ml−1 )

DIP (×10−7 g ml−1 )

Added

Added

Added

0.91 0.91 0.91 1.62 1.62 1.62 3.75 3.75 3.75 9.50 9.50 9.50

Found 0.91 0.89 0.82 1.59 1.53 1.54 3.66 3.60 3.89 9.23 9.63 9.37 0.1227 0.9995 3.110

0.73 2.00 8.70 0.73 4.90 8.70 2.00 4.90 8.70 0.73 2.00 4.90

Found 0.75 1.89 8.51 0.71 5.04 8.87 2.08 5.12 8.53 0.76 1.92 4.71 0.1368 0.9989 3.3509

0.68 1.86 7.90 1.86 7.90 4.15 4.15 0.68 1.86 7.90 4.15 0.68

Found 0.70 1.95 7.76 1.78 8.12 3.98 4.26 0.63 1.76 7.63 4.32 0.68 0.1405 0.9987 3.8519

B. Li et al. / Talanta 72 (2007) 223–230

229

Table 5 Determination results of organophosphorous pesticides residues in vegetable samples Found (×10−7 g ml−1 ) original sample

Added (×10−7 g ml−1 ) addition

Found (×10−7 g ml−1 ) addition

Recovery (%)

OME

DIC

DIP

OME

DIC

DIP

OME

DIC

DIP

OME

DIC

DIP

9.4 1.1 0.4

7.7 4.1 1.7

10.5 4.8 2.0

5.0 3.0 1.5

5.0 3.0 1.5

5.0 3.0 1.5

4.6 2.8 1.6

5.2 2.9 1.6

5.0 3.2 1.5

92 93 107

104 97 107

100 107 100

Rape 1# 2# 3#

15.3 2.5 0.9

7.1 4.2 2.6

5.5 2.3 0.6

5.0 3.0 1.5

5.0 3.0 1.5

5.0 3.0 1.5

5.3 3.3 1.4

5.1 3.1 1.7

4.9 3.3 1.5

106 110 93

102 103 113

98 110 100

Spinach 1# 2# 3#

14.6 2.7 0.7

33.6 4.7 1.9

13.9 1.7 0.4

5.0 3.0 1.5

5.0 3.0 1.5

5.0 3.0 1.5

4.8 2.8 1.6

4.9 2.8 1.4

5.2 2.6 1.5

96 93 107

98 93 93

104 87 100

Leek 1# 2# 3#

18.6 3.3 1.2

24.7 2.1 0.4

41.9 7.3 1.6

5.0 3.0 1.5

5.0 3.0 1.5

5.0 3.0 1.5

5.4 3.1 1.7

5.1 3.2 1.6

5.2 3.3 1.4

108 103 113

102 107 107

104 110 93

Samplea

Lettuce 1# 2# 3#

a

1# , the first eluted solution; 2# , the second eluted solution; 3# , the third eluted solution.

the ANN model was optimized and the data of the prediction sets were used to evaluate the performance of the resulting ANN model. The prediction, RMSD, REP, and R2 are summarized in Table 4. The obtained values of the statistical parameters show the ability of the chosen method for simultaneous determination of analytes. 4.6. Interference studies The interference of foreign substances was tested by analyzing a standard mixture solution of 1 × 10−7 g ml−1 omethoate, 1 × 10−7 g ml−1 dichlorvos and 1 × 10−7 g ml−1 dipterex. The tolerable concentration ratios for interference at the 5% level were over 1000 for Na+ , Ca2+ , K+ , Mn2+ , Cl− , SO4 2− , CO3 2− and NO3 − ; 100 for Ba2+ , Pb2+ , Mg2+ , Al3+ and Zn2+ ; 5 for Cu2+ , Fe3+ , Cr3+ , Co2+ and Ni2+ ; and 1 for PO4 3− . In this system, some substances with strong reducing property (such as ascorbic acid and SO3 2− ) did not cause interference because in the photo-oxidation these substances would be oxidized. Although some ions (such Cu2+ , Fe3+ , Cr3+ , Co2+ , Ni2+ , PO4 3− ) caused interference severely, it could be easily discriminated by the cation-exchange and anion-exchange resins. 4.7. Application of the model to determination of the three pesticides in vegetable sample Some fresh commercial vegetable samples obtained from a supermarket in Xi’an were free from pesticides. About 2 ml mixture standard solution of 1 × 10−3 g ml−1 omethoate, 1 × 10−3 g ml−1 dichlorvos and 1 × 10−3 g ml−1 dipterex was sprayed on the surface of about 50 g each samples, and the treated vegetables were placed at ventilation cabinet for 2 h at 25 ◦ C. Then the each of vegetables was immersed in 50 ml water to wash the pesticides for 5 min, and the first eluted solution were collected and measured. At last, each of vegetables

samples was again eluted two times using 50 ml water, and the second eluted solution and the third eluted solution were collected and measured, respectively. In order to evaluate the validity of the proposed method for the simultaneous determination of omethoate, dichlorvos and dipterex, recovery studies were carried out on the eluted solutions to which known amounts of omethoate, dichlorvos and dipterex were added. The results are given in Table 5. 5. Conclusions In this paper, the continuous-flow CL system combined with artificial neural network calibration was successfully applied to simultaneous determination of omethoate, dichlorvos and dipterex in the vegetable samples without any prior separation. The detection limit of the proposed method was less than 1 × 10−8 g ml−1 for the pesticides. Compared with the reported multi-residue analysis methods of organophosphorus pesticides (such as CG and LC), this method offers the potential advantages of high sensitivity, simplicity and rapidity for multi-residue analysis of organophosphorus pesticides. Furthermore, this paper shows the possibilities of the combination of ANN calibration and CL method, and it shows a guide that the CL method is used to synchronously determinate multi-analytes in one sample. Acknowledgements This study was supported by the National Natural Science Foundation of China (Grant No. 20405009) and by the Program for New Century Excellent Talents in University. References [1] L. Wang, L. Zhang, H. Chen, Prog. Chem. 18 (2006) 440. [2] F. Ahmadi, Y. Assadi, S.M.R. Milani Hosseini, M. Rezaee, J. Chromatogr. A 1101 (2006) 307.

230 [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21]

B. Li et al. / Talanta 72 (2007) 223–230 Q. Xiao, B. Hu, C. Yu, L. Xia, Z. Jiang, Talanta 69 (2006) 848. D.H. Kim, G.S. Heo, D.W. Lee, J. Chromatogr. A 824 (1998) 63. X. Zhu, J. Yang, Q. Su, J. Cai, Y. Gao, J. Chromatogr. A 1092 (2005) 161. T. P´erez-Ruiz, C. Mart´ınez-Lozano, V. Tom´as, J. Mart´ın, Anal. Chim. Acta 540 (2005) 383. C.P. Sanz, R. Halko, Z.S. Ferrera, J.J.S. Rodr´ıguez, Anal. Chim. Acta 524 (2004) 265. V.B. Kandimalla, H. Ju, Chem. Eur. J. 12 (2006) 1074. J.M. Abad, F. Pariente, L. Hern´andez, H.D. Abru˜na, E. Lorenzo, Anal. Chem. 70 (1998) 807. M. Shi, J. Xu, S. Zhang, B. Liu, J. Kong, Talanta 68 (2006) 1089. E.D. Magallona, Res. Rev. 56 (1975) 1. P.-S. Chen, S.-D. Huang, Talanta 69 (2006) 669. S. Dulaurent, F. Saint-Marcoux, P. Marquet, G. Lachˆatre, J Chromatogr. B 831 (2006) 223. P. Moris, I. Alexandre, M. Roger, J. Remacle, Anal. Chim. Acta 302 (1995) 53. B. Li, D. Wang, J. Lv, Z. Zhang, Talanta 69 (2006) 160. B. Li, Z. Zhang, Y. Jin, Anal. Chem. 73 (2001) 1203. L. G´amiz-Gracia, A.M. Garc´ıa-Campa˜na, J.J. Soto-Chinchilla, J.F. Huertas-P´erez, A. Gonz´alez-Casado, Trends Anal. Chem. 24 (2005) 927. J. Wang, C. Zhang, H. Wang, F. Yang, X. Zhang, Talanta 54 (2001) 1185. X. Liu, J. Du, J. Lv, Luminescence 18 (2003) 245. J. Du, X. Liu, J. Lv, Anal. Lett. 36 (2003) 1209. Z. Rao, J. Wang, L. Li, X. Zhang, Chin. J. Anal. Chem. 29 (2001) 373.

[22] Encyclopedia of Chinese Chemical Products, Chemical Industry Press, Beijing, 2005. [23] G.M. Escandar, P.C. Damiani, H.C. Goicoechea, A.C. Olivieri, Mirochem. J. 82 (2006) 29. [24] A.A. Ensafi, T. Khayanmian, A. Benvidi, E. Mirmomtaz, Anal. Chim. Acta 561 (2006) 225. [25] B. Li, Y. He, J. Lv, Z. Zhang, Anal. Bioanal. Chem. 383 (2005) 817. [26] K. Petritis, L.J. Kangas, P.L. Ferguson, G.A. Anderson, L. Psa-Yoli, M.S. Lipton, K.J. Auberry, E.F. Strittmatter, Y. Shao, R. Zhao, R.D. Smith, Anal. Chem. 75 (2003) 1039. [27] Y. Ni, G. Zhang, S. Kokot, Food Chem. 89 (2005) 465. [28] S. Ventura, M. Silva, D. P´erez-Bendito, Anal. Chem. 67 (1995) 4458. [29] B.M. Quencer, S.R. Crouch, Crit. Rev. Anal. Chem. 24 (1993) 243. [30] Safavi, O. Moradlou, S. Maesum, Talanta 62 (2004) 51. [31] T. P´erez-Ruiz, C. Mar´ınez-Lozano, V. Tom´aa, J. Mart´ın, Anal. Chim. Acta 442 (2001) 147. [32] O.V. Zui, J.W. Birks, Anal. Chem. 72 (2000) 1699. [33] R.J. Garmon, C.N. Reilley, Anal. Chem. 34 (1962) 600. [34] D. Xu, Z. Wu, System Analysis and Design based on Matlab 6X-Neural Network, Xidian University Press, Xi’an, 2003. [35] J.Z. Zhang, C.J. Fischer, P.B. Ortner, Talanta 49 (1999) 293. [36] W.D. King, H.A. Lounsbury, F.J. Millero, Environ. Sci. Technol. 29 (1995) 818. [37] K. Robards, P.J. Worsfold, Anal. Chim. Acta 266 (1992) 147. [38] J. Zupan, J. Gasteiger, Anal. Chim. Acta 248 (1991) 1.