Comparison of BLLS-RBL, PLS-RBL and PARAFAC for pesticides and metabolites determinations in wine by SPE and HPLC-DAD.

ALEJANDRO CESAR OLIVIERI; B. BRAGA, JEZ W.; G. BOTTOLI, CARLA B.; C. GOICOECHEA, HÉCTOR; J. POPPI, RONEI

Campinas

Congreso; 10ª Conferencia Internacional de Quimiometría y Química Analítica; 2006

Sociedad Internacional de Quimiometria

      HPLC is a standard method in a large number of routine analyses. However, in complex samples the occurrence of overlapping of the peaks if interest with interferences is frequent, particularly when environmental and food samples are analyzed. When a HPLC-DAD is used, a second-order data is generated and this problem can be overcome using methods that present the so-called second-order advantage. The aim of this work is to compare the results obtained by bilinear least squares (BLLS) and unfolded partial least squares (PLS), both using the residual bilinearization process (RBL) and PARAFAC in a proposed chromatographic method (PM) for determination of 5 pesticides and 2 metabolites in wine samples by HPLC-DAD. For comparison purposes the results of a chromatographic reference method (RM) is also presented.       The pesticides determined are carbaryl (CBL), methyl thiophanate (TIO), simazin (SIM), carbendazin (CBZ), dimethoate (DMT) and the metabolites are 3,5-dichloroaniline (DCA) and phthalimide (PTA). In all analyses a C18 column (150x4,6 mm, 4mm) was used. PM used isocratic elution, 50:50(v/v) acetonitrile:water as mobile phase and 15 minutes of analysis time. RM used gradient elution with acetonitrile:water and 60 minutes for analysis. A solid phase extraction (SPE) method was used for cleanup and concentration of 2.5 times. For both methods time shifts was observed that were corrected by the method proposed by Prazer et al. when second-order calibration is employed. Six standards was used for model building, 9 mixtures for validation and 10 wine samples were determined. For a better evaluation of the prediction errors in 6 of the 10 wine samples the compounds of interest were spiked into an extract obtained after the SPE, therefore no loss in the extraction method was considered in these 6 extracted samples (EX). In the last 4 wine samples the compounds were spiked before SPE, therefore the losses are considered (WI). All samples were analyzed in duplicate.         For the PM it was observed that the compounds PTA and DMT are severally overlapped with interferences present in wine, TIO and SIM presented results partially overlapped with presence of interferences in less quantity. CBL resolved the standards but showed a small interference on wine samples and CBZ and DCA all samples without interference. Therefore, second order calibration was employed in 5 of the 7 compounds of interest. After background and time shift correction the results gave 9, 8, 9, 1 and 3 % of relative prediction errors on the EX samples for DMT, PTA, SIM, TIO and CBL, respectively, using the PLS-RBL model, while for CBZ and DCA errors of 7 and 5 % were observed using univariate regression. The WI samples presented the best results for CBL and SIM presenting errors below 10 % and standard deviations of 3 and 6 %, respectively, on the 4 samples. The other compounds were not completely eluted during SPE, showing that the SPE method should be improved for application in routine analyzes.          On the RM all compounds of interest are resolved in the standards. However, when a wine sample is analyzed interferences occur in SIM, CBZ, PTA and DMT. For the first two compounds second-order calibration was successfully employed. BLLS-RBL presented lightly better results with errors of 2 and 8 %, respectively. CBL, TIO and DCA were determined by univariate regression presenting errors of 11, 5,  4 %. For PTA and DMT no feasible results could be obtained with wine samples.       The second-order calibration models were successful developed in 5 distinct cases all presenting the second-order advantage. No significant differences were observed with three calibration models and the two chromatographic methods (PM and RM) can be considered equivalent, except for SIM.