Retention modeling and resolution optimization for a group of N-phenylpyrazole derivatives in micellar electrokinetic chromatography using empirical and physicochemical models

Carmen García-Ruiz Olga Jiménez Maria L. Marina Departamento de Química Analítica, Facultad de Química, Universidad de Alcalá, Alcalá de Henares (Madrid), Spain Retention modeling and resolution optimization for a group of N-phenylpyrazole derivatives in micellar electrokinetic chromatography using empirical and physicochemical models The optimization of the separation resolution for a group of N-phenylpyrazole derivatives in micellar electrokinetic chromatography (MEKC) as a function of the separation buffer composition (surfactant and organic modifier concentration) has been performed. In order to achieve our purpose, the first step has been the prediction of the migration times of the electroosmotic flow (t0) and micelles (tm), and the retention factors of solutes (k), as a function of surfactant (sodium dodecyl sulfate) and alcohol (n-propanol or n-butanol) concentrations, by means of empirical equations. Also, some physicochemical models have been applied to relate the retention factors to the surfactant and the organic modifier concentrations in order to optimize the separation resolution and to increase our knowledge of the separation process. Finally, a comparison of the resolution optimization through the use of the physicochemical and empirical models selected has been made in order to obtain the optimum separation buffer composition for the separation of a group of 17 N-phenylpyrazole derivatives as test solutes. Keywords: Empirical equations / Micellar electrokinetic chromatography / Physicochemical model / Pyrazole derivatives / Resolution optimization / Retention prediction EL 5262 1 Introduction The optimization of the separations performed by micellar electrokinetic chromatography (MEKC) is a complex task due to the great number of variables affecting the process: pH, type and concentration of buffer, surfactant, and organic modifiers. In recent years, in order to predict the optimal separation conditions with the minimum number of experiments, several strategies in MEKC have been reported [1, 2]. As an example, it can be cited the overlapping resolution mapping (ORM) [3–9], iterative regression strategies [10], physicochemical approaches [11– 14], empirical equations [15–18], and artificial neural networks (ANNs) [19]. In order to carry out these studies, several designs can be used, e.g., the Plackett-Burman design [20, 21] and the orthogonal array design (OAD) [22] which are factorial designs suitable for screening the influence of many parameters and to monitor possible interactions among a large number of factors, or the central composite design [23, 24] that can provide a response surface for the prediction of areas of optimum performance. From the different strategies cited, the ORM has been one of the most used [3–9] because it allows the deduction of the optimal separation conditions from an overlay of all the graphs obtained plotting the resolution versus different separation conditions. However, migration of solutes is not followed. Also, a great number of experiments is required to carry out the optimization process. Physicochemical models describing the migration behavior of individual solutes have been reported. Thus, Khaledi et al. [11–14] introduced physicochemical models describing the migration behavior of both acidic and basic solutes as a function of the separation buffer composition and physicochemical constants. First, a description of migration in terms of pKa, micelle-water binding constant, and mobility of the anionic solutes in the absence of micelles has been performed [11]. A description of the solute mobility in terms of physical and chemical constants of each solute, the pH of the buffer, and the micelle concentration in the buffer has also been reported [12]. Finally, a model describing the migration of ionizable (acidic and basic) solutes as a function of the simultaneous variation of the pH and micelles concentration has Correspondence: Dr. Maria Luisa Marina, Departamento de Química Analítica, Facultad de Química, Universidad de Alcalá, E-28871 Alcalá de Henares (Madrid), Spain E-mail: [email protected] Fax: 34-91-8854971 Abbreviations: MPRE, mean prediction relative errors; PC, principal component Electrophoresis 2003, 24, 325–335 325 2003 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 0173-0835/03/0302–325 $17.501.50/0 CE and CEC 326 C. García-Ruiz et al. Electrophoresis 2003, 24, 325–335 been proposed [13]. Recently, our research team has developed a physicochemical model to predict the retention of neutral compounds as a function of micellized surfactant and the organic modifier concentration [14]. Although these models usually require the use of nonlinear fitting software during the process of optimization, physicochemical migration characteristics of compounds are obtained. Insight is gained in the mechanism of migration in MEKC. On the other hand, empirical equations have been considered very valuable tools to predict the retention behavior of solutes as a function of different variables in MEKC. Pyell and Bütehorn [15] have proposed linear first-degree models for the migration time of the electroosmotic flow marker, the logarithm of the migration times of the micelles and the logarithm of the retention factors of the solutes. They used these models to predict the resolution between peaks, which were calculated assuming a constant plate number. In practice, the predicted optimum diverges from the real optimum, indicating that the underlying linear function is not able to describe the retention behavior accurately. Later, Bütehorn and Pyell [16] introduced the interaction between the variables considered (surfactant and modifier concentrations) to model the migration time of the electroosmotic flow marker, the logarithm of the migration times of the micelles and the retention factors of the solutes, however, the number of parameters in their empirical equations was equal to the number of experiments achieved, so their models appeared to have better descriptive than predictive quality. Recently, Zomeren et al. [18] studied which response should be modeled preferently to enable resolution optimization in MEKC (resolution, separation factor, apparent and effective mobility). Although their results are very promising too much data were required (14 data to obtain the model parameters). ANNs can be considered as soft models because they do not need mathematical equations [19, 25] and have been usually applied to classification, modeling, association, and mapping [19, 26]. Havel et al. [19] have examined the modeling capabilities of the ANN approach in MEKC, with comparison to hard models and the use of ANNs in combination with suitable designs to facilitate the optimization and/or prediction of electrophoretic mobilities in MEKC. Jalali-Heravi and Garkani-Nejad [27] have compared the prediction power of ANNs and multiple linear regression in capillary electrophoresis, pointing out that their results were superior with the ANN. Farková et al. [28] have shown that ANNs can be used to estimate peak parameters and which experimental design can be applied for efficient prediction of optimal separation conditions. Srecnik et al. [29] have developed an ANN model, which can be generalized and used in a variety of applications for retention modeling in ion chromatography. Agatonovic-Krustin et al. [30] have compared the usefulness of ANNs for response surface modeling in HPLC optimization with multiple regression methods. Gao et al. [25] have carried out the optimization of gas chromatographic experimental parameters and compared their results with that obtained by the orthogonal method. Loukas [31] has examined the behavior of a series of training algorithms in the behavior of ANNs and the results were compared from the partial least square (PLS) method. Zhao et al. [32] have applied an ANN to model the retention behavior of several solutes in ternary systems in HPLC and to predict two groups of different liver and bile diseases. Madden et al. [33] have used ANNs to predict the retention times of anions when eluted with linear hydroxide gradients of varying slope. Jiménez et al. [34] have examined the usefulness of ANNs to model the retention behavior of organic solutes in micellar liquid chromatography. From the results comparison with other empirical and theoretical models it seems that ANNs are the best choice but some drawbacks must be clear. The use of ANNs is not a simple task. Previously, the ANN architecture must be optimized and it means to optimize the number of hidden layers, the number of neurons in the hidden layer(s), the normalization or transformation of the experimental data, the data used to train the network (not only the representativity but also the number of them), the momentum and the learning rate, the training algorithm, the transfer functions for the hidden layer(s) and the output layer, etc. There are no rules of thumb and “trial and error” must be adopted. Moreover, although the network structure could be carefully optimized some other problems must be taken in mind, that is, the effect of overfitting and, what it is more important, there is no guarantee to find the global minimum. So, recently, Siouffi and PhanTan-Lun [2] in a very interesting review about optimization methods in chromatography and capillary electrophoresis have pointed out that the trend is towards the use of ANNs but they are still in infancy when applied in separations. Moreover, if a hard model (empirical or theoretical model) is available and it can predict accurately the magnitude we are interested in, the use of ANNs is not justified. So, our aim in this work has been to perform a global resolution optimization by using empirical and physicochemical models to predict the retention factors of 17 N-phenylpyrazole derivatives as a function of the surfactant and modifier concentrations in the buffer composition. Also, it has been considered of great importance to predict the migration time of the electroosmotic flow and micelles due to the scarce data found in the literature. Electrophoresis 2003, 24, 325–335 Resolution optimization in MEKC 327 2 Materials and methods 2.1 Chromatographic data Retention data for the group of 17 N-phenylpyrazole derivatives considered as model solutes (Table 1 shows their structures and names) when separation buffer contains 0.08 Mof 2-(N-cyclohexylamino)ethanesulfonic acid (CHES) in alkaline medium (pH 10) with SDS as surfactant and n-propanol or n-butanol as the organic modifier have been used in this work [35]. Although the experimental conditions employed to perform the separation of these compounds are widely explained in [35], they are summarized below: an Applied Biosystems capillary electrophoresis instrument 279A-HT model (Norwalk, CT, USA), with UV detection at 238 mn and temperature controlled at 307C was used to obtain the electropherograms of the compounds under study. The dimensions of the fused-silica capillary (Polymicro Technologies, Phoenix, AZ, USA) used were 75 cm of total length and 50 cm of effective length with 25 mm of inner diameter and 375 mm of outer diameter. The applied voltage was 15 kV. Table 1. Identification numbers, names and structures of N-phenylpyrazole derivatives studied No. Name R3R4R5R2’ 1DNPP HHHNO 2 2 3-Methyl DNPP Me H H NO2 3 4-Methyl DNPP H Me H NO2 4 4,5-Dimethyl DNPP H Me Me NO2 5 3-Ethyl DNPP Et H H NO2 6 4,5-Dimethyl pNPP H Me Me H 7 3,4,5-Trimethyl DNPP Me Me Me NO2 8 4-Methyl pNPP H Me H H 9 3-Methyl-4-nitro-5cloro DNPP Me NO2Cl NO2 10 3,5-Dimethyl pNPP Me H Me H 11 4-Bromo pNPP H Br H H 12 3-Bromo-4-methyl DNPP Br Me H NO2 13 3,5-Dimethyl-4-bromo DNPP Me Br Me NO2 14 5-Methyl-4-bromo pNPP H Br Me H 15 3-tert-Butyl pNPP t-BuHHH 16 3,4-Dibromo DNPP Br Br H NO2 17 3-Ethyl-4-bromo DNPP Et Br H NO2 DNPP, dinitrophenylpyrazole; pNPP, paranitrophenylpyrazole 2.2 Data treatment Data have been divided into two sets, the model and the prediction data sets following the factorial design plotted in Fig. 1. Model data set (crosses) permits the calculation of the parameters of the different equations and the test data set (circles) have been used to evaluate the prediction capability of the empirical and physicochemical equations used in this work. It should be mentioned that when the buffer consisted of 0.03 MSDS and 0.03 as the volume fraction of n-butanol, results obtained have proved to be outliers and were not used to build any of Figure 1. Composition of the MEKC systems using a 0.08 MCHES (pH 10) buffer. Circle data means the prediction data set and crosses the model data set. 328 C. García-Ruiz et al. Electrophoresis 2003, 24, 325–335 the models. The evaluation of empirical equations has been made using box plots performed by the Sigma Plot software [36]. Data treatment was made using Microsoft Excel [37]. Multiple regression analysis and tests performed in this work to compare the error averages obtained in different experimental conditions were carried out using the Statgraphics Plus software [38]. Nonlinear regressions were performed by using the Sigma Plot software [36]. The performance of equations considered in this work has been evaluated comparing the relative errors defined as follows: PRE %kcal kexp  kexp 100 (1) MPRE % X iN i1 PREi N(2) where kcal is the calculated retention factor by the equation that is being testing, kexp is the experimental retention factor, Nis the number of different separation buffers (with different concentrations of surfactant and alcohol). 3 Results and discussion As mentioned, the optimization of separations performed by MEKC is complex and difficult due to the high number of parameters affecting the separation process. Thus, resolution of two closely adjacent peaks (Rs), in MEKC, is dependent on the selectivity factor, the mean retention factor and the ratio of the migration time of the electroosmotic flow to the migration time of the micelles according to the following equation proposed by Terabe and Cheng [39]: RS N p 4 a1 a k k1 1t0 tm 1t0 tm k (3) where Nis the theoretical plate number; kis the mean retention factor of the two peaks considered; ais the selectivity factor (kj/kibeing kj.ki), and t0and tmare the migration times of the electroosmotic flow and micelles, respectively. According to Eq. (3), resolution optimization requires the knowledge of t0,tm,k, and Nvalues. In the following calculations, the efficiency of the chromatographic system (N) has been considered independent of the electrolyte composition [15, 16]. However, the values of t0,tm,k, and aare dependent on the electrolyte composition. In this study, the prediction of these parameters has been performed by means of empirical and physicochemical equations in order to optimize the resolution. 3.1 Use of empirical equations for the prediction of electroosmotic flow and micelles migration times Pyell and Bütehorn [15] first stated the importance of predicting the migration times of the electroosmotic flow (t0) and micelles (tm) as a function of the buffer composition. However, although they have proposed the use of different equations [15, 16] no validity studies have been performed. So, in this work a systematic study on equations relating migration times of the electroosmotic flow and micelles with the surfactant concentration and the volume fraction of the organic modifier (mand f, respectively) has been performed (Table 2). It must be mentioned that two of them have been used previously by Pyell and Bütehorn [15], Eq. (22), and by Bütehorn and Pyell [16], Eq. (4). In order to evaluate the empirical equations (Eqs. 4–33) that best predict the migration times of the electroosmotic flow and micelles, multiple regression analysis has been performed by using the model data set shown in Fig. 1. Once the parameters have been calculated, the prediction has been performed in the conditions shown for the test data set (Fig. 1). Table 2. Empirical equations used for the prediction of electroosmotic flow (t0) and micelles (tm) migration times Basic equation zvalue Equation No. zvalue Equation No. z=A1Bm1Cf 1Dmf t0 1/t0 lg t0 4 6 8 tm 1/tm lg tm 5 7 9 z=A1Bmf t010 tm11 1/t012 1/tm13 lg t014 lg tm15 z=A1Bf1Cmf t016 tm17 1/t018 1/tm19 lg t020 lg tm21 z=A1Bm1Cft022 tm23 1/t024 1/tm25 lg t026 lg tm27 z=A1Bm1Cmf t028 tm29 1/t030 1/tm31 lg t032 lg tm33 In order to establish the validity of the tested equations, different criteria have been taken into account. That is, the most simple equation for which the lowest mean prediction relative errors are obtained is our objective. Nevertheless, we must take in mind that all the terms must be statistically significant. In this respect, for the prediction of t0when n-propanol is used, the lowest mean prediction Electrophoresis 2003, 24, 325–335 Resolution optimization in MEKC 329 relative errors are obtained for Eqs. 4, 6, 8, 16, 22, 24, and 26 (no statistically significant differences, P.0.05) as can be observed in Fig. 2a. From these equations, we have chosen Eq. (4) because all the terms are statistically significant (indicating that it is not possible to simplify it to Figure 2. MPRE for electroosmotic flow (t0) and micelle (tm) migration times by using empirical equation in a MEKC system with (a) n-propanol or (b) n-butanol as the organic modifier. These plots are defined in terms of percentiles and take a quick look at the median and spread of the data, as well as the mean, minimun, and the maximum values for the variable studied. Outliers of percentiles are also represented by circles. The lengths of the upper and lower lines associated to each box show how stretches the tails of the distribution are. obtain Eqs. 16, 22, 24, or 26) and due to it is the simplest one compared with Eqs. (6) and (8). Following the same reasoning exposed previously, we have chosen Eq. (4) to predict t0(Fig. 2b) when n-butanol is considered as the organic modifier and Eqs. (23) and (7) (Figs. 2a and b) to predict tmwhen n-propanol and n-butanol are considered as the organic modifier, respectively. It can be noted that the equation that best predicts the migration time of the separation buffer is the same when the organic modifier used is n-propanol and n-butanol (Eq. 4), but that the equations differ when the prediction of the migration time of the micelle is achieved. This fact is in agreement with Van Hove et al. [40] that considers these two alcohols belonging to different categories. 3.2 Use of empirical equations for retention prediction In order to predict the retention of the compounds under study, four empirical equations (Eqs. 34–37) proposed by Jiménez et al. [17] and another one empirical equation (Eq. 38) proposed by Pyell and Bütehorn [15] relating retention factors with the total surfactant concentration and the volume fraction of the alcohol (mand f, respectively) have been employed: 1 kAB mCf(34) 1 kAB mCf m(35) 1 kAB mCf2 m(36) 1 kAB mCf mDf2(37) lnkABln mCf(38) Multiple regression analysis has been performed to obtain the equation parameters (Eqs. 34–38) by using the model data set shown in Fig. 1. Then, retention factors for the prediction data set (Fig. 1) were calculated. From the comparison of the mean prediction relative errors it can be observed (Figs. 3a and b) that when n-propanol was used, the lowest errors were obtained by means of Eqs. (35–38) (no statistically significant differences observed, P.0.05). When n-butanol is considered, the lowest errors were obtained by means of Eq. (37) and (38) (no statistically significant differences observed, P.0.05). However, the multiple regression analysis shows that Eq. (37) contains a nonsignificant term, reason for which Eq. (38) was selected as the most appro- 330 C. García-Ruiz et al. Electrophoresis 2003, 24, 325–335 Figure 3. MPRE for the retention factors by using empirical equations in a MEKC system with (a) n-propanol or (b) n-butanol as the organic modifier. priate to predict the retention of the solutes considered (prediction errors 3.9–11.5% for n-propanol and 5.4– 15.1% for n-butanol). Nevertheless, Jiménez et al. [17] found that Eq. (35) was the most appropriate to predict the retention behavior of a group of dihydropyridines in similar MEKC systems using the same five empirical equations employed in this work. These results could be explained by the different characteristics of each group of compounds. As an example, the literature shows that the group of the N-phenylpyrazole derivatives studied in this work are less hydrophobic (with values of the logarithm of the octanol-water distribution coefficient, log Pow, ranging from 2.08 to 2.60 [41]) than the group of dihydropyridines (with values of the log Pow ranging from 2.43 to 4.31 [42]). 3.3 Use of physicochemical models for retention prediction The retention prediction through the use of a physicochemical model relating the retention factor with the micellized surfactant (SDS) and the organic modifier (npropanol and n-butanol) concentrations has been performed. The physicochemical model [14] considers different interactions among the species present in the system according to several equilibria. Relating these equilibria with the solute retention obtained by MEKC the following equation can be obtained: kfK11K3Aaq  Mm  1K2Aaq  K2K4Aaq  2(39) where, kis the retention factor, fis the phase ratio, [Aaq] and [Mm] are the alcohol and the micellized surfactant concentrations (molar concentrations), respectively, and K1,K2,K3, and K4are equilibrium constants. These constants take into account the association of the solute with the micelle to form a complex in the micellar pseudophase (K1), the enhancement of solubility of solute in the separation buffer modified by alcohols (K2), the formation of complexes among the solute, the alcohol and the micelle in the micellar pseudophase (K3), and the complex formation between the solute-alcohol complex and other molecules of alcohol (K4), as it is shown in the following equilibria: Saq Mm ! K1SMm Saq Aaq ! K2SAaq SMmAaq ! K3SAMm SAaq Aaq ! K4SA2;aq Nevertheless, some simplifications to this equation can be done, according to the following approximations [14]: (i) If 1 .. K3[Aaq] the simplified expression is: kfK1Mm  1K2Aaq  K2K4Aaq  2(40) (ii) If 1 ,, K3[Aaq] the simplified expression is: kfK1K3Aaq  Mm  1K2Aaq  K2K4Aaq  2(41) Electrophoresis 2003, 24, 325–335 Resolution optimization in MEKC 331 (iii) If 1 .. K4[Aaq] the simplified expression is: kfK11K3Aaq  Mm  1K2Aaq  (42) In order to check the validity of this model, the retention data for a group of test solutes (17 N-phenylpyrazole derivatives) have been used. First of all, nonlinear regressions were achieved to obtain the equilibrium constant values from Eqs. (39)–(42) in order to predict the retention factors. These values were used to calculate the mean prediction relative errors according to the four physicochemical equations described. Mean relative errors for the four equations considered in MEKC systems with n-propanol (5.7, 20.2, 31.4, 5.8 for Eqs. 39, 40, 41, and 42, respectively) and n-butanol (4.8, 5.4, 30.1, and 6.9 for Eqs. 39, 40, 41, and 42, respectively) have been compared with a multiple comparison procedure [38] to establish the equation that best model the experimental retention behavior. The results from the comparison indicate that Eqs. (39) and (42) are not statistically different when n-propanol is used as the organic modifier and Eqs. (39), (40) and (42) are not statistically different when n-butanol is employed as the organic modifier. Although for the two systems considered (n-propanol and n-butanol), Eqs. (39) and (42) do not differ significantly, the mean error values are lower when Eq. (39) is used, so it can be considered that Eq. (39) is the best equation, although the use of nonlinear regression is necessary. From the results exposed in the two preceding sections we can conclude that Eqs. (38) and (39) are, respectively, the best equations to predict retention factors by means of empirical and physicochemical models. Moreover, from the comparison of mean prediction errors obtained, the statistical test indicates that the physicochemical model is the best to explain the retention behavior of the solutes under study (confidence level of 95%). 3.4 Resolution optimization through the use of physicochemical and empirical models The evaluation of all these equations has permitted to choose the best equations for the prediction of k,t0, and tmwith the aim of optimizing the resolution (Rs) between two consecutive peaks using Eq. (3). In this equation it has been considered an Nvalue constant and equal to 150 000 [43]. The mean retention factor of the two peaks considered has been calculated according to the empirical Eq. (38) or the physicochemical Eq. (39); and t0and tm have been obtained, respectively, from empirical Eqs. (4) and (23) for n-propanol, and from Eqs. (4) and (7) for n-butanol. The resolution has been optimized according to the criterion of global resolution given in [44]. This optimizing criterion is based on the normalized product, r, of different properties, Xi, i11, associated to pairs of consecutive peaks. rY n1 i1 Xi;i1 P n1 i1 Xi;i1n1  n1(43) In this work, Xis the resolution predicted with the abovementioned equations and nis the number of solutes. Figure 4 shows the response surfaces obtained for systems containing SDS as the surfactant and n-propanol or n-butanol as the organic modifier. The equations used for the prediction of retention factors are Eqs. (38) (Figs. 4a and b) and (39) (Figs. 4c and d), respectively. In both cases, the prediction of t0and tmwas achieved by empirical Eqs. (4) and (23) when n-propanol was used and Eqs. (4) and (7) when n-butanol was employed. It can be observed that the maximum resolution to separate the group of N-phenylpyrazole derivatives studied using empirical equations corresponds to a 0.08 MCHES buffer (pH 10) with 0.02 Min SDS and 0.03 volume fraction of n-propanol (Fig. 4a). On the other hand, the maximum resolution obtained for the model solutes studied using the empirical equations selected correspond to 0.08 M CHES buffer (pH 10) with 0.05 MSDS and 0.03 volume fraction of n-butanol (1), or with 0.01 MSDS without n-butanol (2) (Fig. 4b). When the physicochemical model (Eq. 39) has been used, the response surface for n-propanol shows two maxima which correspond to 0.08 MCHES buffer (pH 10) with 0.05 MSDS and a 0.05 volume fraction of n-propanol (1) or with 0.02 MSDS and a 0.01 volume fraction of n-propanol (2) (Fig. 4c). The best resolution is obtained for maximum 2. On the other hand, when n-butanol is used as organic modifier only a maximum is obtained, corresponding to 0.08 MCHES buffer (pH 10) with 0.02 MSDS and a 0.03 volume fraction of n-butanol (Fig. 4d). Figure 5 shows the experimental electropherograms for a selected mixture of solutes and for the best conditions chosen from the response surfaces shown in Fig. 4. From the experimental electropherograms shown in Fig. 5 some conclusions can be drawn: (i) As expected, Eq. (39) leads to the best optimal conditions for the separation of the solutes considered in this study if we compare with the results obtained by means of Eq. (38). For example, if n-propanol is considered (Figs. 5a and c) solutes 5 and 6 are resolved (although not completely) but when Eq. (38) is used in the retention prediction these peaks coelute. Moreover, when butanol is used as the organic modifier (Figs. 5b and d), and Eq. (39) is used, solutes 15 and 16 are resolved (although not completely) but when Eq. (38) is used these two solutes coelute. (ii) It is interesting to 332 C. García-Ruiz et al. Electrophoresis 2003, 24, 325–335 Figure 4. Response surfaces obtained for the group of N-phenylpyrazole derivatives studied in the MEKC system with SDS as surfactant and n-propanol or n-butanol as the organic modifier using the predicted retention factors with (a, b) the empirical Eq. (38) or (c, d) the physicochemical Eq. (39). Conditions of maximum resolution: (a) 0.08 MCHES buffer (pH 10) with 0.02 MSDS and 3% n-propanol; (b) 0.08 M CHES buffer (pH 10) with 0.05 M SDS and 3% n-butanol (1) or with 0.01 MSDS without n-butanol (2); (c) 0.08 MCHES buffer (pH 10) with 0.05 MSDS and 5% n-propanol (1) or with 0.02 MSDS and 1% n-propanol (2); (d) 0.08 MCHES buffer (pH 10) with 0.02 MSDS and 3% n-butanol. Figure 5. Electropherograms corresponding to the injection of a mixture of 15 N-phenylpyrazole derivatives in the optimal resolution conditions obtained in Fig. 4. DMF: dimethylformamide; BaP, benzo[a]pyrene. note that n-propanol (Fig. 5c) better separates the firstmigrating compounds in the electropherograms while n-butanol better separates the last-migrating compounds as has been reported previously in [35]. So, these two alcohols work complementarily and the most adequate one would depend on the solutes we are interested in. 3.5 Mechanistical approaches through the physicochemical model The values of the physicochemical constants obtained from Eq. (39) are shown in Table 3. These results clearly show that the constant values depend not only on the solute nature but also on the alcohol nature. From the fK1 values obtained, we can conclude that although the hydrophobicity is the main driving force affecting the chromatograhic behavior, it is not the only one. Thus, a plot of the fK1values versus log Pow shows a moderately strong relationship between the variables (correlation coefficient, r, 0.8297). A better correlation is obtained when the logarithm values of fK1are plotted versus the log Pow (r= 0.9564), but solutes with the same hydrophobicity differ in fK1values, probably due to electrostatic interactions (i.e., solutes 2 and 4, both with log Pow 2.16, solutes 11 and 16, both with log Pow 2.43, and solutes 8 and 12, both with log Pow 2.41). Electrophoresis 2003, 24, 325–335 Resolution optimization in MEKC 333 Table 3. Equilibrium constants obtained from physicochemical model considered (Eq. 39) Solute No. n-Propanol n-Butanol fK1K2K3K4fK1K2K3K4 1 38 2.52 0.89 38 3.06 0.27 2 67 2.67 0.87 67 3.43 0.19 3 96 5.15 2.11 67 3.06 0.03 4 109 3.46 0.94 109 4.18 5 162 2.16 0.36 161 3.43 0.44 6 188 6.11 1.98 188 4.04 2.83 2.49 7 196 4.75 1.71 196 4.29 0.05 8 233 4.35 1.70 232 3.81 0.09 9 236 17.50 8.28 236 3.53 10 268 4.85 1.46 267 5.63 0.06 11 363 2.13 0.48 357 2.75 0.70 12 470 1.61 0.45 403 1.78 1.22 13 493 4.19 1.48 757 4.41 14 757 6.15 2.69 753 3.86 0.41 15 795 1.12 0.38 784 7.95 7.71 3.91 16 740 1.59 0.99 1.18 736 2.28 24.70 17 1550 5.01 5.50 1550 0.44 18.70 The values of K2and K3for the solute number nine are significantly higher than those for the other solutes when n-propanol is considered, indicating that favorable interactions between the alcohol and the nitrogen atom in the pyrazole ring of this compound are possible, both in the micellar phase and in the electrophoretic buffer. This fact leads to contrary effects to the retention, first the formation of complexes in the micellar pseudophase increases the retention in this phase and, second, the formation of complexes in the electrophoretic buffer drifts the retention towards the hydro-organic phase. When n-butanol is considered, the complex formation between the alcohol, the solute and the micelle is negligible. This alcohol is more hydrophobic than n-propanol, so the aqueous phase polarity and the micelle surface charge density of the micelle diminish [45]. As a consequence, the electrostatic interactions are diminished, fact that can be viewed by the lower value of K2and the negligible value of K3constants. Generally, K3values are lower for n-butanol than for n-propanol with two exceptions (solutes 6 and 15). These solutes have a very voluminous substituent in the R3position (see Table 1), so perhaps steric impediments can be expected. n-Butanol could enhance the retention of this compound (K3value higher in n-butanol than in n-propanol) due to the expansion of the micelle or to the better solubilization power of this solvent inside the core ofthe micelle. The highest values of K4(solute 16 for systems containing n-propanol and solutes 16 and 17 for systems containing n-butanol) can be attributed to interactions between the alcohol head group and the positive charge density of the two nitrogen atoms (in the pyrazole ring). For the other compounds, the K4values are low or negligible, so in these cases the retention could be explained by means of a more simplistic model (Eq. 42). In order to clarify the relationship among the retention behavior and the structural properties of the solutes, principal component (PC) and cluster analysis have been achieved. Massart et al. [46] highly recommended that a clustering method is combined with a PC output. So, in this work the four constant values have been reduced to a lower number of variables by means of the PC analysis (two and three components have been extracted when n-propanol and n-butanol are considered, respectively). The PCs extracted were: (i) for n-propanol: PC1 = 0.999998fK120.000781284K210.00159507K3 10.000170362K4; PC2=20.0000369037fK110.888103K210.459474K3 20.0125016K4. (ii) For n-butanol: PC1 = 0.999924fK120.000893064K210.000838577K3 10.0122405K4; PC2=20.0121667fK120.127968K220.0691081K3 10.989293K4; PC3=20.00177059fK110.563511K210.815836K3 10.129861K4. Then, the cluster analysis has been applied to the components extracted in every case. In this way, we can classify the compounds with the same retention behavior. In Fig. 6, the cluster scatterplots for systems containing n-propanol and n-butanol are shown (PC 2 vs. PC 1).