|
1. |
Neural networks in multivariate calibration |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 157-178
Frédéric Despagne,
Preview
|
PDF (334KB)
|
|
摘要:
Tutorial Review Neural networks in multivariate calibration Fr�ed�eric Despagne and D. Luc Massart* ChemoAC, Pharmaceutical Institute, Vrije Universiteit Brussel, Laarbeeklaan 103, B-1090 Brussels, Belgium Received 17th July 1998, Accepted 28th August 1998 1 Introduction 2 Principle of neural networks 3 Neural networks in multivariate calibration 3.1 When to use neural networks 3.2 Alternative methods 3.2.1 Linear methods 3.2.2 Non-linear methods 3.3 Advantages and limitations of neural networks 3.3.1 Flexibility of neural networks 3.3.2 Neural networks and linear models 3.3.3 Robustness of neural networks 3.3.4 Black-box aspect of neural networks 4 Development of calibration models 4.1 Data pre-processing 4.1.1 Detection of non-linearity 4.1.2 Detection of outliers 4.1.3 Number of samples 4.1.4 Data splitting and validation 4.1.5 Data compression 4.1.6 Data scaling 4.2 Determination of network topology 4.2.1 Number of layers 4.2.2 Number of input and output nodes 4.2.3 Number of hidden nodes 4.2.4 Transfer function 4.3 Training of the network 4.3.1 Learning alogrithms 4.3.2 When to stop training 4.3.3 Model interpretation 5 Conclusion 6 Acknowledgements 7 References 1 Introduction Artificial neural networks (NNs) have now gained acceptance in numerous areas of chemistry, as illustrated by the number of applications mentioned by Zupan and Gasteiger1 in their review.In the 1996 Chemometrics fundamental review,2 NN applications were reported in sections concerning signal processing, curve resolution, calibration, parameter estimation, QSAR, pattern recognition and of course artificial intelligence.Tutorials on NNs in chemistry were proposed by Smits et al.3 and Svozil et al.4 (the latter contains an extensive list of Internet resources for NNs) and different types of applications of NNs to spectroscopy were reviewed by Cirovic.5 This tutorial is restricted to the application of NNs for multivariate calibration with chemical data, which is an important source of publications in chemometrics.The potential of NNs as modelling tools for multivariate calibration is well established, and efforts must now focus on developing proper methodologies to ensure that NNs are always used in ideal conditions; this is the goal of this tutorial. Bos et al.6 presented an excellent overview of practical aspects of NNs in quantitative analysis. Most of these aspects will be presented again here, in particular in order to establish the terminology, and we will include some recommendations according to recent results obtained in NN research.We will restrict ourselves to NNs of the multi-layer feed-forward type (also called multi-layer perceptron, MLP) with the error back-propagation learning rule that is the most popular. The tutorial is organised as follows. In Section 2, we remind readers how NNs came on to the scene and explain their basic principles. Section 3 is dedicated to the possibilities offered by NNs to analytical chemists. We present some of the most general aspects of NNs (flexibility, black-box aspect) and emphasise their main limitations.In Section 4 we consider more technical aspects and propose a methodology for the development of calibration models with NNs. A non-negligible part of this methodology is dedicated to data handling, and we try to outline the pitfalls specific to NN modelling. We consider topology optimisation and introduce techniques that can help in developing and interpreting NN models.The different aspects discussed in the tutorial are illustrated with examples of applications from the literature. Fr�ed�eric Despagne obtained an engineer degree from the Ecole Nationale Sup�erieure de Chimie et Physique de Bordeaux and a postgraduate diploma in Materials Science from Universit�e de Bordeaux in 1994. He was then sponsored by Elf Aquitaine to do research in the Chemometrics group of Professor Brown at the University of Delaware.In 1996 he joined the research group of Professor Massart at the Vrije Universiteit Brussel, where he is currently studying for a PhD His research interests are in multivariate calibration and artificial intelligence. Professor D. Luc Massart teaches analytical chemistry at the Pharmaceutical Institute of the Vrije Universiteit Brussel, where he was appointed in 1968. He is author of several books on chemometrics.Analyst, 1998, 123, 157R–178R 157R2 Principle of neural networksNNs stem from the field of artificial intelligence. An earlymotivation for developing NNs was to mimic some uniquecharacteristics of the human brain, such as the ability to learngeneral mechanisms from presentation of a reduced set ofexamples, or to retrieve correct information from missing ordistorted input data. NNs currently used in applied scienceshave little in common with their human counterparts and thescope of their possible applications is more restricted.Researchis still being carried out to establish links between neurobiologyand artificial intelligence, but a description of NNs by analogywith biological concepts, although fascinating, can lead to anerroneous perception of NNs as mysterious intelligent machines.In the framework of multivariate calibration, we willconsider NNs in a more pragmatic way and in a firstapproximation define them as non-parametric non-linear regressionestimators.7 Non-parametric methods are those methodsthat are not based on the a priori assumption of a specificmodel form.NNs allow one to estimate relationships between one orseveral input variables called independent variables or descriptorsand one or several output variables called dependentvariables or responses. Information in an NN is distributedamong multiple cells (nodes) and connections between the cells(weights).An example of MLP is displayed in Fig. 1, for amodel with four descriptors x1, x2, x3, x4 and a single responsey.The descriptors are presented to the NN at the input layer andthen weighted by the connections wAij between the input andhidden layer.Hidden layer nodes receive simultaneouslyweighted signals from input nodes and perform two tasks: asummation of the weighted inputs followed by a projection ofthis sum on a transfer function fh, to produce an activation. Inturn, hidden nodes activations are weighted by the connectionswBj between the hidden and output layer and forwarded towardsthe nodes of the output layer.Similarly to hidden nodes, outputnodes perform a summation of incoming weighted signals andproject the sum on their specific transfer function fo. In Fig. 1 asingle response y is modelled and the output layer contains onlyone node. The output of this node is the estimated response ÆÙythat can be expressed asÆÙyf w f w x o jjnhh ij iind= ¢F¢F ¢F¢F ¢F + ¢FÆÙ¢X£»£»£»= = q q +1 1(1)where nd and nh are the number of input variables and hiddennodes, respectively.Although NNs can be considered as non-parametric tools, themodels that they yield are defined by sets of adjustableparameters determined by an algorithm, not a priori by the user.Adjustable parameters are the weights wAij, wBj and biases qA, qBthat act as offset terms by shifting the transfer functionshorizontally.They are determined with an iterative procedurecalled training or learning. The adjustable parameters are firstascribed initial random values, then training starts and proceedsin two steps.First, a forward pass [Fig. 1(a)] is performedthrough the NN with a set of training samples with knownexperimental response y. At the end of the pass, the magnitudeof the error between experimental and predicted responses iscalculated and used to adjust all weights of the NN, in a backpropagationstep [Fig. 1(b)]. These two steps constitute aniteration or epoch. A new forward pass is then performed withthe training samples and the optimised parameters.The wholeprocedure is repeated until convergence is reached. This meansthat a pre-specified or acceptably low error level is reached.Training an NN is an optimisation problem, where one seeksthe minimum of an error surface in a multi-dimensional spacedefined by the adjustable parameters. Such surfaceerised by the presence of several local minima, saddlepoints or canyons. It must be accepted that the NN will probablynot find the absolute minimum of the error surface, but a localminimum relatively close to the absolute minimum andacceptable for the problem considered.The most popularalgorithm to adjust weights during training is the gradientdescent algorithm based on the estimation of the first derivativeof the error with respect to each weight.8The most important feature of NNs applied to regression isthat they are universal approximators: they can fit anycontinuous function defined on a compact domain (a domaindefined by bounded inputs) to a pre-defined arbitrary degree ofaccuracy.9 We will now see why this characteristic can beparticularly attractive in analytical chemistry.3 Neural networks in multivariate calibration3.1 When to use neural networksFor analytical chemists, a calibration model relates a series ofinstrumental measurements to the concentration or somephysico-chemical properties of one or several target analytes.10NNs can be used to build empirical multivariate calibrationmodels of the form: Y = F(X) + e.We will only consider inversecalibration models, for which X designates a matrix of analyticalmeasurements performed on a series of n samples. For a givensample, measurements are described by a set of descriptors, xi,for instance, absorbance values at a given set of wavelengths. Yis a vector or a matrix containing sample responses, for instance,the concentrations of a target analyte in a set of mixtures.Calibration sample responses are often determined experimentallywith reference methods such as the wet chemistry KjeldahlFig. 1 Feed-forward NN training: a, forward pass; b, error backpropagation.158R Analyst, 1998, 123, 157R¡V178Rmethod for the determination of protein content in wheat, or research engines for the determination of gasoline octane number. NNs should be primarily used when a data set is known or suspected to be non-linear. (From the mathematical point of view, a truly non-linear model is non-linear with respect to its parameters.We will also consider as non-linear the models where the non-linearity appears in the relationship between the response and the descriptors. In analytical chemistry, the distinction between true and apparent non-linearity cannot always be performed since most non-linearities are detected visually on calibration lines or in model residuals.) Several types of non-linearity can be observed with sensor or spectroscopic data.11,12 For instance, the Beer–Lambert law that linearly relates the absorbance of a species in a mixture to its concentration is an approximation that is only valid for dilute and non-saturated systems.Deviations from linearity can occur if a sample is highly absorbing or non-homogeneous, if the particle size is not constant in all samples (for crystalline species) or if some signals are overlapping. A non-linear detector response (especially with photoconductive detectors) or the presence of stray light (due to imperfections in the optics of a spectrometer) introduces curvature in the concentration– response function.These non-linear effects can sometimes be corrected with appropriate pre-processing such as first or second derivative, multiplicative scatter correction (MSC) or standard normal variate correction (SNV) [the last two are preprocessing techniques generally used to remove particle size or scatter effects from near-infrared (NIR) spectra13].Pre-processing has limitations, however: derivatives reduce the signal-tonoise ratio; in some situations the spectra are apparently corrected by a mathematical pre-treatment but non-linearity is introduced in the wavelength space (for instance, SNV is not a linear transformation). Other non-linear effects can be observed as a result of chemical factors such as non-symmetrical chemical equilibrium, intermolecular reactions, intermetallic reactions in electrochemistry, presence of humidity inducing hydrogen bonding, changes in temperature or solvent composition.These effects result in a shift and broadening of the absorption bands. In some situations, the X–Y relationship is known to be intrinsically non-linear even if it cannot be explicitly derived. This is the case, for instance, with the relationship between the NIR spectrum and flex modulus of an elastomer,11 or the relationship between octane number and NIR spectra or chromatograms of gasolines, most gasolines containing hundreds of different hydrocarbons with non-linear blending characteristics.14 NNs are ideal tools for such problems since they can theoretically map any measurable linear or non-linear function.NNs can also be applied when no a priori indication concerning the nature of the relationship to model is available and a model is needed rapidly. It is a situation where nonparametric statistical inference can help to delve into complex multivariate problems.However, we recommend that one always starts by modelling new data with one of the well known linear regression methods that give satisfactory results on most calibration data [multiple linear regression (MLR), principal component regression (PCR) or partial least squares (PLS)]. A decision as to whether an NN model is necessary can be based on the examination of linear model residuals: a curvature or a trend in the residuals is indicative of an un-modelled source of variance, generally a non-linearity.(With PCR and PLS, one must look for non-linearity in residuals in the first few dimensions only since the inclusion of a large number of components will generally mask the presence of non-linearity.) Finally, NNs can be recommended for monitoring on-line processes, where measured variables are likely to be blurred with noise and perturbations such as temperature effects15 that introduce non-linearity into the model.In such situations, the flexibility of NNs and their ability to maintain a decent performance even in the presence of significant amounts of noise in the input data are highly desirable. Several workers have obtained good results with NNs in the presence of noise in the analytical measurements or in the responses from the calibration data.16–18 Long et al.19 found that the addition of increasing levels of random noise to the training data did not significantly affect the model.In contrast to most techniques, the performance of NNs degrades regularly with increasing levels of random noise in the training data or with deletion or perturbation of an increasingly large number of weights.4,20 This remarkable property can be attributed first to the signalaveraging effect of small random deviations in the two summation terms in eqn. (1).19,21 In addition, the dense interconnectivity between nodes is a non-localised form of information storage that acts as a security against component damage.22,23 It also implies that adding more nodes to an NN should make it more robust with respect to random noise.This temptation must be resisted because the data sets are seldom large enough to avoid the under-determination problem encountered with an oversized NN: too few data points are available compared with the number of adjustable parameters. Even with a limited number of connections, the distribution of information in an NN allows one to reduce the influence of random noise.Another reason why NNs are more robust than several other techniques in the presence of random noise in training data is that they can build non-linear models from a limited number of descriptors whereas techniques such as PLS or PCR accommodate the non-linearities by including higher order components that are likely to be blurred with noise.24 Table 1 contains a list of references that illustrate the broadness of the application field of MLP NNs in multivariate calibration. 3.2 Alternative techniques 3.2.1 Linear methods. In order to assess the usefulness and main limitations of NNs, it is interesting to have an overview of alternative tools often applied in multivariate calibration. The most popular methods remain MLR, PCR and PLS. The attraction of MLR lies in the ease of model interpretation since the estimated parameters relate the property of interest linearly to a set of original variables.In PLS or PCR, components used for modelling are linear combinations of original variables. The projection of samples on the reduced subspace spanned by the first components allows the visualisation of outliers, atypical samples and clusters among the objects. These methods are based on the minimisation of a least squares criterion, similarly to NNs. In fact, if linear transfer functions are used in both hidden and output layers, the two successive linear combinations performed in the NN are equivalent to a single MLR regression.If a linear data set is modelled with an NN using linear transfer functions, it will converge to the MLR solution. The difference between MLR and NNs lies in the way the model parameters are estimated: by matrix inversion in MLR and by iterative optimisation in NNs. Eqn. (1) can also represent a PCR or PLS model when the transfer functions are linear.25 The weights between input and hidden layer are equivalent to the X-data loadings on the different factors and the activation produced by the hidden nodes can be compared with PCR or PLS scores.Again, the difference lies in the way the parameters are optimised. In NNs, adjustable parameters are fitted without restriction to minimise the calibration samples squared residuals. In PCR or PLS, constraints such as scores orthogonality, maximisation of Xdata variance (PCR) or X–Y covariance (PLS) are also taken into account, and therefore the parameters obtained are different.Analyst, 1998, 123, 157R–178R 159RAlthough they are linear methods, MLR, PCR or PLS can be used for the modelling of some specific types of non-linear data. If the form of the non-linear relationship between the response and the descriptors is known, a model can be linearised by taking the appropriate transform of the original variables,26 or by adding higher order and cross-terms to the regression equation. In practice, the number of situations where these approaches are successful is limited, mainly because the exact form of the non-linear relationships is not known a priori and the number of calibration samples available is not sufficient to fit a complex model with a large number of cross-terms.It is also known to PCR and PLS practitioners that in some cases these methods can accommodate non-linear relationships by using higher order components to correct for partial nonlinearities. 27 However, there is a risk of introducing a significant amount of irrelevant information in the model. 3.2.2 Non-linear methods. Non-linear variants of PCR or PLS also exist (polynomial PCR,28 quadratic PLS29). Their main limitation is that they are based on the assumption that a simple (e.g., quadratic) relationship exists between the response modelled and the components. This assumption is sometimes violated since components are already linear combinations of original variables.30 Locally weighted regression (LWR) is based on the decomposition of a global non-linear model in a series of local linear PLS or PCR models.It was found to perform well in multivariate calibration, especially on clustered data sets.31,32 However in LWR a data set cannot be described with a unique set of components and loadings, since each sample is fitted with a local model built with its nearest neighbours only. One must also take the risk that local model parameters are less stable than global model parameters since they are estimated with a reduced set of objects. Other techniques exist for non-linear regression but they are not yet as popular as NNs and the above-mentioned techniques.A review of non-parametric non-linear regression methods [alternating conditional expectations (ACE), smooth multiple additive regression technique (SMART), non-linear partial least squares (NLPLS), classification and regression trees (CART), multivariate adaptive regression splines (MARS) and spline partial least squares (SPL-PLS)] can be found in the report of Sekulic et al.33 and in Frank’s tutorial.34 These methods can perform well on non-linear data but are computationally more complex than linear methods and share with NNs the limitation of being prone to overfitting. Their performance also depends heavily on the amount and quality of data available.34 3.3 Advantages and limitations of neural networks 3.3.1 Flexibility of neural networks.We have seen that NNs are not the only tools to handle non-linear multivariate data.However, their flexibility is often a decisive asset compared with parametric techniques that require the assumption of a specific hard model form. Hard models cannot be developed with NIR data owing to the significant overlap of combination and overtone bands in the spectra. Other types of analytical data Table 1 Examples of application of NNs to multivariate calibration Property modelled Descriptors Ref.Alditols in binary mixtures (%) 1H NMR spectra 18 Apparent metabolic energy of barley Measured physical and chemical characteristics of barley 60 Components in simulated binary and ternary mixtures UV/VIS spectra with simulated instrumental perturbations 12 Active ingredients in drugs UV/VIS spectra Components in simulated binary and ternary mixtures Absorption spectra with simulated non-linear effects 30, 67 Components in rhodamine mixtures UV/VIS spectra Components in simulated binary mixtures UV/VIS spectra 19 Active ingredients in drugs UV/VIS spectra Protein in wheat Near-infrared spectra [H2O] in meat Near-infrared spectra 25 Flex modulus of polymers Near-infrared spectra Ethanol in mixtures containing latex Near-infrared spectra 24 Fat in port meat Near-infrared spectra Mineral charge in polymer Near-infrared spectra 70 Gasoline octane number Near-infrared spectra [KOH] in polyether polyols Near-infrared spectra Constituents in paper coatings Near-infrared spectra 64, 65 [OH], [NH], grind size in cereals Simulated near-infrared spectra 35 Methanol in water mixtures Near-infrared spectra 41 Composition of organic extract Near-infrared spectra Hydroxyl in cellulose esters (%) Near-infrared spectra Property of polymer pellets Near-infrared spectra Solvents in aqueous process stream Near-infrared spectra Aromaticity of brown coals Fourier transform infrared spectra 72 Color change in emulsion paints Measured concentration of oxide ingredients 61 RNA, DNA or lysozyme in binary mixtures containing glycogen Pyrolysis mass spectra 16 Bacteria in tertiary mixtures Pyrolysis mass spectra Adulteration of cow’s milk with goat’s or ewe’s milk Pyrolysis mass spectra 17 Penicillin and buffer ion concentrations in solutions with different buffer ion concentrations Measurement signals of enzyme field effect transistor flow injection analysis 96 Urea and glucose in solutions at different pH Measurement signals of enzyme field effect transistor flow injection analysis 23 [SO2] and relative humidity of water vapour in sample gas Frequency response of a piezoelectric crystal gas sensor 97 Cu/Zn in simulated two-component system with formation of intermetallic compounds Anodic stripping voltammograms 98 Cu/Pb/Cd/Zn in experimental four-component system Anodic stripping voltammograms Ionic concentrations in mixtures Measurements from ion-selective electrode arrays 99 Characteristics of the physical structure of polymer yarns Parameters describing mechanical properties of the yarns 38 Metals in Fe/Ni/Cr systems X-ray fluorescence spectra 86 Fe/Ni in thin films X-ray fluorescence spectra Gasoline octane number Gas chromatograms 14 160R Analyst, 1998, 123, 157R–178R(e.g., UV/VIS spectra) are more easily interpretable from the spectroscopic point of view, but the a priori specification of a hard model rarely incorporates the non-linear effects that may occur in practice.Non-linearity in a data set can be detected with graphical methods but identification of its source is more challenging and sometimes impossible. Thanks to their ability to learn and derive X–Y relationships from the presentation of a set of training samples, NNs avoid the time-consuming and possibly expensive task of hard model identification. In addition, the fundamental principle of distributing information among several weights and nodes renders the NN model robust with respect to random noise in the input data (as already explained) and allows one to have several NNs with different topologies converging to qualitatively equivalent results.If one is not careful, however, a drawback of the flexibility of NNs is their tendency to overfit calibration data and the resulting lack of generalisation ability, that is, the capability of a model to produce a valid estimate of the correct output when a new input is presented to the NN.Also, the flexibility of NNs can lead to unreliable results in situations of extrapolation. Although NNs proved to perform better than PLS on extrapolated non-linear data in some applications,24 they were found to be equivalent to or less reliable than methods such as MLR, PCR, PLS or LWR in comparative studies of calibration methods where extrapolations occurred.35,36 The dangers of strong extrapolation with NNs are illustrated in Fig. 2(a)–(c), which show results obtained for the modelling of a cosine function with different numbers of hidden nodes and test points (+) outside the calibration domain.The calibration domain contains the X-values in the range [22, +2]. The NN builds an empirical model to fit objects in the calibration space only and test points are badly predicted. With analytical data, such strong extrapolations rarely occur and one generally has the situation represented in Fig. 2(d), where the prediction error is less dramatic.It is possible to use NNs to perform small or mild extrapolations on such non-linear data but NNs should not be considered as generally suitable for extrapolation, as with any other chemometrics technique. 3.3.2 Neural networks and linear models. One may wonder what happens if an NN is used to model a linear data set. For instance, a model may be wrongly considered as non-linear owing to an incorrect estimation of linear PLS or PCR model complexity. It is also tempting to take advantage of the flexibility of NNs and let them do the work with any kind of data, even when they should be linear. From the point of view of prediction, if the data are linear an NN with non-linear transfer functions should nevertheless converge to a solution that approximates a linear model solution, since the linear portion of the transfer functions can be activated in that case (see Fig. 3). This was confirmed by the results of a recent comparative study carried out to evaluate the performance of several linear and non-linear modelling methods on real industrial data.32 Each of the four industrial data sets consisted of a series of NIR spectra (X-data) and a specific property to be predicted (Y-data).Some results of this comparative study are listed in Table 2, which contains the root mean square error of prediction (RMSEP) values obtained with stepwise MLR, PCR, PLS and NN. NNs outperform linear methods for the strongly non-linear data set, which is not surprising, but their performance on slightly non-linear and linear data is comparable to the performance of linear methods such as PLS or PCR.This is in agreement with the observations of Gemperline et al.,12 who stated that ‘Artificial neural networks having the appropriate Fig. 2 NN predictions within and outside calibration space: a–c, cosine function; d, quadratic function. Model with, a, three; b, six; c, nine; and d, two hidden cones. o, Actual training, +, actual test, *, predicted.Analyst, 1998, 123, 157R–178R 161Rarchitecture can be used to develop linear calibration models that perform as well as linear calibration models developed by PCR or PLS’. It was said that when NNs are used to model linear relationships, they require a long training time since a nonlinear technique is applied to linear data.33 In theory this is true in the sense that the apparently linear portion of the non-linear transfer functions is not perfectly linear, and therefore the learning algorithm must perform continuous adjustments to correct for this slight deviation.For a perfectly linear and noisefree data set, the NN performance tends asymptotically towards the linear model performance and it generally converges to the intrinsic precision of the computer. However, in this case the curve of NN error as a function of the number of iterations is almost perfectly flat and an acceptable solution can be reached relatively early during the training.Moreover, perfectly linear and noise-free data sets are seldom available so that in practice NNs can reach a performance qualitatively similar to that of linear methods in a reasonably short training time. In spite of these reassuring results, it does not make sense intuitively to apply a complex and possibly time-consuming method when simpler tools are likely to perform as well. MLR with stepwise variable selection can give excellent prediction results on linear data sets (see Table 2) and its interpretation properties for the analyst are optimal compared with all other methods.In practice, using a highly flexible tool to model linear phenomena can lead to rapid overfitting of the measurement noise. Artefacts can also occur if the topology of the NN is not carefully designed. As an illustration, Fig. 4 shows distortions appearing when a perfectly linear and noise-free model is fitted with an NN containing too many hidden nodes and a non-linear node instead of a linear node in the output layer. 3.3.3 Robustness of the models. NNs are sometimes recommended for their ‘robustness’,37 but this term is rarely defined with precision. Unlike analytical procedures for which official definitions of the term exist, there is not a unique definition of the robustness of a multivariate calibration model, as illustrated by some controversial statements.38,39 It seems reasonable to follow Frank and Todeschini’s40 definition of robustness in the framework of regression analysis: robust methods are those methods that are insensitive to small deviations from the distributional assumptions.This definition applies in particular to methods designed to cope with outliers present in the calibration set. Methods to detect or handle outliers are presented in Section 4.1.2. Robustness of an NN is also challenged when predictions are performed on new samples outside the calibration domain in the X-space or in the Y-space.We underlined in Section 3.3.1 that NNs often perform relatively poorly in situations of extrapolation. In all these situations, deviations from a priori assumptions (data set free of outliers and of systematic noise) affect the training samples. Some authors consider robustness from a different perspective, in situations where a model has been developed with training data that fulfil initial assumptions but perturbations affect new objects to be predicted.38,41 Different types of perturbations must be considered.The appearance of higher levels of random noise in the test samples is usually not catastrophic.42 Derks et al.38 related quantitatively the variance of predicted responses to the variance of random noise added to the input variables. The influence of instrumental perturbations that have a more systematic effect than random noise (e.g., baseline or wavelength shift) can be more catastrophic and is difficult to anticipate.Indeed, it depends on a number of parameters: the curvature of the relationship between each descriptor and the response and the position of the perturbed samples on the descriptors axes. When a strongly non-linear relationship is being modelled, the NN can have either an attenuating effect with respect to perturbations (compared with linear models) because of the squashing effect of the nonlinearity, or a catastrophic effect on high leverage points, as illustrated in Fig. 5. Since the exact shape of the model and the position of future samples in input space cannot always be known, a solution consists of identifying possible sources of degradation and including them either in the training set42 or in the monitoring set.41 This allows one to avoid large prediction errors after the appearance of small perturbations that can be expected in practice. 3.3.4 Black-box aspect of neural networks. NNs can perform at least as well as any other technique in terms of prediction, but a major criticism remains their black-box aspect.To be fair, it should be pointed out that this limitation is not peculiar to NNs only. For instance, it is often impossible to visualise clusters and outliers by projecting scores on component axes in LWR since the samples belong to local models Fig. 3 Usual non-linear transfer functions: hyperbolic tangent; sigmoid. Table 2 RMSEP of different multivariate calibration methods applied to industrial data Property y Nature of data MLR PCR PLS NN Moisture in wheat Linear 0.1860 0.2147 0.2150 0.1981 Hydroxyl number of polyether polyol Linear 0.90 1.15 1.31 0.88 Octane number of gasoline Slightly non-linear 0.1355 0.1426 0.1461 0.1459 Mineral charge in a polymer Strongly non-linear 0.0797 0.0477 0.0445 0.0096 162R Analyst, 1998, 123, 157R–178Rdefined with different objects. However, model interpretation with an NN is still considered much more complex than with, e.g., PLS or PCR.This is due to the operations (summation and projection on transfer function) performed successively in the hidden and output layer, that prevent one from deriving simple analytical expressions between input and output variables [see eqn. (1)]. In addition, unlike QSAR applications, where input variables are heterogeneous original variables, the input variables used in multivariate calibration are often scores compressing spectral information, which complicates even further model interpretation.Methods to ease model interpretation will be presented in Section 4.3.3, but it is clear that model interpretability remains an active research area for the NN community and the danger of incorrect inference (common to all non-parametric techniques) must not be overlooked. 4 Development of calibration models We will now examine in more detail the way in which an NN model should be developed, according to our experience. The different steps in method development are summarised in the flow chart in Fig. 6. It will come as no surprise that data pre-processing (Fig. 6, left) governs closely the quality of results that can be expected. We propose some tools to help in optimising parameters such as the number of input variables or the number of hidden nodes. NN construction (Fig. 6, right) is based on alternating removal of input and hidden nodes, starting from a large NN. The procedure described in this flow chart is very general and of course other strategies are applicable.Short cuts can be made through the flow chart by including a priori knowledge, or as the user acquires more experience with topology optimisation. 4.1 Data pre-processing 4.1.1 Detection of non-linearity. As a general rule, one should not try to build an NN model unless the situation is one of those mentioned in Section 3.1. Therefore, some diagnostic tools are necessary to detect the presence of non-linearity in a data set.The simplest approach—which in many cases is sufficient to detect the presence of non-linearity—is to plot the property of interest versus the different measurement variables, or combinations of these variables such as PC scores. If these plots are inconclusive then one should build a linear model with MLR, PCR or PLS. Visual inspection of the residuals (y 2 �y) of the linear model versus each descriptor xi retained in the model, versus the experimental response y and versus the estimated response �y should then be performed to detect non-linearities. Recently, Centner et al.43 reviewed a number of more sophisticated graphical and numerical methods to detect nonlinearities.They cited the Mallows augmented partial residuals plot (MAPRP) combined with a runs test as the most promising approach for non-linearity detection. The MAPRP is the plot of the term (e + bixi + biixi 2), called augmented residuals, versus xi. The e are residuals of the linear regression y = f(x1,.. .,xi,. . .,xn, xi 2). The regression should be performed on all variables xi in the model (original variables or principal component scores). Curvature in the MAPRP plot indicates that higher variables xj (j > i) correct for the non-linear (quadratic) nature of the relationship between y and the variable xi. In that case the variable xj is undesirable because it makes the model less robust. The runs test is used to detect series of residuals with the same sign, called runs.Long runs indicate the presence of a trend in residuals that may be a systematic bias or non-linearity. From the total number of positive and negative residuals, one calculates a z-value that is compared with a tabulated value. A significant value of |z| indicates a trend in the residuals. As an illustration, we performed the detection of non-linearity between the spectra of a series of 104 diesel oil NIR spectra and their viscosity.We built a 10-component PCR model, and for each principal component (PC) we looked at the MPARP plot combined with the runs test. For PC2, PC3 and PC4, the |z| values indicate a non-linearity between the augmented residuals and the variable (Fig. 7). A limitation of the MPARP plot is that it allows only the detection of non-linearities that can be described or approximated by a quadratic term. Centner et al.43 emphasised the need for careful outlier detection before drawing conclusions about the presence of non-linearity in a data set.Outliers with high leverage can pull the regression line and lead to an incorrect estimation of the number of runs. Conversely, some outlier detection methods can wrongly flag as outliers samples that are high leverage points responsible for non-linearity in the data.33 This will be illustrated in the next section. 4.1.2 Detection of outliers. Actually, the term ‘outlier detection’ encompasses two steps: first, atypical object detection, followed by an outlier identification.Although numerical methods allow flagging of samples that are outliers on statistical grounds, the positive identification of an atypical object as a true outlier requires knowledge of the process or data acquisition procedure, or interaction with the person in charge of this acquisition. It is recommended to keep all flagged samples unless they are positively identified as outliers on experimental grounds. It is beyond the scope of this paper to review all methods for outlier detection proposed in the literature, but we will suggest a few guidelines.One must make a distinction between different types of outliers. Outliers in X can be due to accidental process upsets, experimental errors during acquisition of spectra or transcription errors during the labelling of samples or file Fig. 4 Predictions for linear model with incorrect NN topology. Fig. 5 Attenuation or amplification of Y-prediction error in a non-linear model compared with a linear model, depending on the sign of the error in X.Analyst, 1998, 123, 157R–178R 163Rmanipulation. Outliers in Y are due to incorrect measurements of reference values or transcription errors also. Atypical objects, i.e., possible outliers in X or in Y, can be flagged before performing any modelling. By contrast, outliers in the X–Y relationship can only be detected after building a complete model. The simplest tool to flag atypical objects before modelling is the visual observation of the X and Y data available.One should look at the original set of sample spectra, the vector of responses and score plots on the first PCs. To detect outliers in the X space, it is recommended to examine the leverage of each sample to detect possible outliers. The leverage of a sample is a measure of its spatial distance to the main body of the samples in X.44 For a given data matrix X, the leverage of sample i is given by the diagonal term pii of the prediction matrix P, also called Hat matrix: P = X (XTX)21 XT (2) Fig. 6 Strategy for construction of NN model: left, data handling; right, network construction. 164R Analyst, 1998, 123, 157R–178RWhen there are more variables than objects in X, the prediction matrixt be calculated with the matrix TA of sample scores on the A first significant PCs: P = TA (TA TTA)21 TA T (3) High leverage points have large values of pii (diagonal elements of the P matrix) and special attention should be paid to these points.They have a strong influence on parameter estimation and can alter the model dramatically if they happen to be true outliers. The limitation of this approach is that it is not Fig. 6–Continued Analyst, 1998, 123, 157R–178R 165Rstraightforward to determine A. Several methods (see Section 4.2.2) can be applied to perform this determination.45–49 An alternative approach for the a priori detection of atypical objects in X is to apply Grubb’s test on Rao’s statistic.50 Rao’s statistic D2( k) (yi) is a value calculated for each sample i and for each PC k.It accumulates all variations described by PCs k + 1 to p. For each k, the Rao’s statistic is used as input to flag possible outliers in X with the univariate Grubb’s test. After flagging possible outliers in X or in Y, one must check if these samples are outliers in the X–Y relationship. Centner et al.50 proposed a procedure based on the development of PLS leave-one-out cross-validation models after flagging possible outliers with a Grubb’s test performed on the Rao’s statistic.The goal of the cross-validation is to discriminate situations where a true outlier alters the models resulting in a large cumulative cross-validation error, from situations where the large value of the cross-validation error is simply due to the incorrect prediction of a high leverage point that is not an outlier.A limitation of this approach is that the identification is based on linear cross-validation models (it will be explained in Section 4.1.4 why cross-validation should not be performed with NNs). A sample that is an outlier to a linear model might not be an outlier to a non-linear model.33 The final decision should be made on the basis of a comparison of prediction results for NN models with and without the flagged samples in the training set. To illustrate the difficulty of outlier detection in non-linear models, we report in Fig. 8 a PC scores plot for the NIR data set used to model viscosity of diesel oil. Applying Grubb’s test on Rao’s statistic, the sample marked with an asterisk was identified as an atypical object. Using leave-one-out crossvalidation on PLS models, the flagged sample (which has the highest Y-value in the data set) is positively identified as an outlier to the PLS model. If we compare the PLS and NN test results depending on the inclusion or not of this flagged sample in the training set, we obtain the RMSEP values reported in Table 3.When the flagged sample is included in the training set, the NN performance in prediction improves whereas the PLS performance degrades. This illustrates how non-linear information can be extracted by the NN from a high leverage sample that is not an outlier. Since outlier detection is not always successful, it is possible to design NNs that can handle outliers present in the training set. For instance, Walczak51 proposed to use error thresholding functions adjusted iteratively during training with respect to the median of residuals.Wang et al.52 also applied a thresholding function adjusted with respect to the assumed proportion of outliers among the ranked residuals. In both approaches, the idea is to prevent outlier residuals from influencing weight estimations during training. 4.1.3 Number of samples. The number of samples available is often a limiting factor when using NNs.Like other regression methods, there are constraints concerning the number of samples required to develop an NN model. The number of adjustable parameters is usually such that the training set is rapidly overfitted if too few samples are available. We consider that when this number is less than 30, an alternative modelling technique should be considered. Unfortunately, this is not always obvious for inexperienced users, who can be deceived by the extreme flexibility of NNs since they can fit the training data with arbitrary precision.It is possible to obtain excellent training results for the modelling of data sets with less than 15 samples. However, if these models are validated on new independent samples, a significant degradation of the results is observed due to a lack of generalisation ability. To estimate the minimum number of training samples allowing theoretical generalisation, one can use a parameter called the Vapnik–Cervonenkis dimension (VCDim).For an MLP with one hidden layer, the lower bound of the VCDim is approximated as twice the total number of weights in the NN.53 It is possible to reach good generalisation if the number of Fig. 7 Mallows augmented partial residual plots for PCR models of diesel oil viscosity: PC1; PC2; PC3; PC4. 166R Analyst, 1998, 123, 157R–178Rtraining samples is at least equal to this lower bound. When the number of samples available does not fulfil this requirement, NN can still be used to find an acceptable local minimum close enough to the absolute minimum of the error function.However, the ratio of the number of samples to the number of adjustable parameters should be kept as high as possible, in order to avoid under-determination of the problem. The number of samples is generally imposed or limited by practical constraints, but one can partly solve the under-determination problem by reducing the number of weights in the NN as much as possible, as will be explained in Section 4.1.5. 4.1.4 Data splitting and validation. An important step in the development of any calibration model is the splitting of the available data into two subsets: a training set (used to estimate model parameters) and a validation set or test set (used to check the generalisation ability of the model of new samples). For NNs the problem is more complex because they fit to arbitrary precision the training data, provided that the number of hidden nodes is sufficient and the training time long enough. Therefore, an additional monitoring set is necessary to stop the training before the NN learns idiosyncrasies present in the training data.4,54,55 The monitoring set must be representative of the population under study in order to avoid NN overtraining that leads to overfitting (see Fig. 9). Ideally, for a number nt of training samples, the monitoring set and the test set (if it is available) should contain between nt/2 and nt samples each.The repartition of samples between these sets and the terminology used in several papers are the source of many confusions. When prediction errors are reported in the literature concerning NNs, are the authors referring to training error, monitoring error or validation error? The performance of an NN should not be judged by its performance on training data that can always be fitted perfectly. Often, the problem is to know whether the reported results have been obtained on a monitoring set or a validation set.Data sets are seldom large enough to be split into three subsets, so that authors often report results on a monitoring set that they call the ‘validation set’ or ‘test set’. There is no reason why results obtained on a monitoring set could not be reported, as long as it is made clear that these results were obtained on the data set used to evaluate the training end-point. One must be aware of the limitations of this approach: a true validation error is a better estimator of the NN generalisation ability than a monitoring error.4 If one decides to favour the modelling power of the NN by using only two subsets (training and monitoring) instead of three subsets of smaller size (training, monitoring and validation), very good results may be obtained on the monitoring set but the model has not been truly validated in the sense that the monitoring data were used to optimise one of the model parameters (number of iterations for training).However, the monitoring results can be considered as indicative of the modelling power to expect from the NN model, and they can be compared with, e.g., PLS results with cross-validation. We summarise the comparison between different situations for PLS and NNs in Fig. 10. Some authors mention leave-k-out (often k = 1) crossvalidation as a way of estimating the generalisation ability of the NN, for instance, when only few calibration samples are available.3,6,56 We believe that this approach is not adapted to NNs37,41,54 and we do not recommend it.The procedure can be suitable for parametric linear models characterised by a quadratic bowl-shaped smooth error surface. With such models, the perturbation caused by the removal of one or a few samples from the training set has little influence on the model parameters, and therefore the cumulative cross-validation error obtained is a reliable validation error estimate for the model constructed with all samples.The situation changes for NNs applied to non-linear problems characterised by complex error surfaces.53 Unlike PLS or PCR, which are constrained to produce orthogonal components, no constraint is imposed on NN adjustable parameters and it tends to perform a point-bypoint fit of all training samples. Solutions obtained when two different samples are removed from the training set can differ significantly from each other.4 In this case one cannot consider that the global model is validated, and it is even possible that none of the models developed during cross-validation describe the same region of the error surface as the global model.Therefore, if too few calibration samples are available to create a monitoring set, it is better to consider an alternative method to NNs. Ideally, the monitoring and validation set should be independent of each other and of the training set.This can only be achieved if the samples in each of these subsets are selected randomly. However, it is important to include as many sources of variance as possible in the training set. If not, extrapolation may occur in the prediction phase and this should be avoided with any modelling method. Specific algorithms can be used to select training samples that are representative of the total population and contain high leverage points that carry information about the main sources of variance.A limitation of this approach is that the subsets selected are no longer independent since mathematical criteria are applied to discriminate training samples from the other samples. It is important to keep Fig. 8 Score plot of diesel oil samples. Table 3 Influence of the presence of a single training sample on RMSEP obtained with PLS and NN models Method RMSEP (flagged sample not in training set) RMSEP (flagged sample in training set) PLS 0.31 0.39 NN 0.28 0.23 Fig. 9 Typical evolution of training and monitoring errors as a function of number of iterations. Analyst, 1998, 123, 157R–178R 167Rthis restriction in mind when results are reported. We will now present some algorithms to perform automatic subset selection. The D-optimality criterion selects the n calibration samples that provide regression coefficients with the lowest variance of all the subsets Xn of n samples. Selection is performed by maximising the determinant of the information matrix (Xn TXn).When the number of samples available is large, Ferr�e and Rius57 proposed the use of Fedorov’s exchange algorithm to select the D-optimal subset. Samples selected with this criterion are located at the border of the calibration domain. If a small number of samples are retained, the interior of the calibration domain is not appropriately sampled and the set obtained is not representative of the whole population. The Kennard–Stone algorithm58 is an alternative method that allows the selection of a subset of representative samples. Samples are selected iteratively by maximising the Euclidean distance between the last selected point and its nearest previously selected neighbour.The first samples selected with this method are generally the same as with the D-optimality criterion and they describe the border of the calibration domain. As the number of selected samples increases, their repartition becomes more homogeneous and the subset selected is more representative of the global population. These two algorithms ensure that monitoring and/or validation samples are within the domain covered by the training samples, so that the model does not extrapolate. This type of sample selection does not match the not-so-ideal situation sometimes encountered in practice, where it is not guaranteed that all new samples fall within the calibration domain.The duplex algorithm59 allows a more realistic repartition of samples than the two previous methods.Samples are selected in the same way as with the Kennard–Stone method, but they are alternatively assigned to the training set and the validation (or monitoring) set. Thus, not all samples located at the border of the calibration domain are placed in the training set; some are found in the validation set. However, if some samples at the border of the domain are very close to each other, duplex splitting can be misleading because each training sample will have its nearest neighbour in the validation set.This can lead to overfitting and over-optimistic estimation of the validation error. For the same reason, with any splitting method all replicates of a sample should be assigned to the same subset. Sample selection is often performed in the PC space on the scores matrix T instead of on the original matrix X, which allows one to reduce the computational burden. To illustrate the principle of the three selection methods (Kennard–Stone, Doptimal and duplex), we represented the sets of 30 training samples from a non-linear data set (prediction of viscosity of diesel oil samples from their NIR spectra) selected with each method.We first performed a PCA decomposition of the original X matrix (104 3 795), then the 30 training samples were selected in the subspace spanned by the first ten PCs. Fig. 11 represents the position of the training samples (asterisks) selected in the PC1–PC2 plane.If one wants to compare the efficiency of several modelling methods, samples can be selected with D-optimal or Kennard– Stone designs. If a model has to be developed for an application for which there is no guarantee that only interpolation will be performed, then duplex design will lead to more pessimistic but reliable results. It is also possible to perform the splitting after projecting the samples on a two-dimensional map with a Kohonen NN.60.61 The advantage of such a projection is that an estimation of the relevant number of dimensions is not required and the essential topological features of the data set are preserved in two dimensions, which allows rapid visualisation of the data structure.Fig. 10 Repartition of samples for internal and external validation with PLS and NN. 168R Analyst, 1998, 123, 157R–178RWith strongly clustered data, subset selection should be performed on each cluster separately in order to ensure good representativity between the training and test data.After data splitting, one can apply the methods presented by Jouan- Rimbaud et al.62,63 for estimating numerically the representativity of two data sets. These methods provided indices varying between 0 and 1 to compare direction, covariance and centroids of two data sets. 4.1.5 Data compression. As pointed out earlier, the ratio of the number of samples to the number of adjustable parameters in the NN should be kept as large as possible.One way of overdetermining the problem is to compress input data, especially when they consist of absorbances recorded at several hundred wavelengths. In addition to reducing the size of input data, compression allows one to eliminate irrelevant information such as noise or redundancies present in a data matrix. Successful data compression can result in increased training speed, a reduction of memory storage, better generalisation ability of the model, enhanced robustness with respect to noise in the measurements and simpler model representation.The latent variables calculated with the PLS algorithm are designed to project data points on a lower dimensional subspace descring all relevant sources of variance. While PCs are designed to maximise the explained variance in the X-space only, PLS latent variables are built so as to maximise the covariance between X and Y. Some authors have used PLS to calculate input socres for NN training.64 However, the latent variables are designed to conserve information linearly correlated with the response and some relevant non-linear information might be rejected in higher order latent variables that are not retained in the model.24,65 For this reason, we do not recommend pre-processing data with PLS before NN modelling.The most popular method for data compression in chemometrics is principal component analysis (PCA). In addition to summarising almost all variance in the X-matrix on a few axes only (the PCs), it has the property that these axes are mutually orthogonal, which allows inversion of the variance–covariance matrix in linear regression models (PCR).Orthogonality of input variables is not so critical for NNs that can handle collinear input data. However, most NN applications in quantitative analysis with spectral data use PC scores as input variables.24,30,41,66–70 For the determination of the optimum number A of input PCs to retain, one can use the same PC selection procedures as for PCR, although the choice is not so critical since NN models are built iteratively by successive optimisations of the NN topology.One possible approach consists in performing initial calculations with a deliberately large number of PCs and progressively reducing this number. This point will be detailed in Section 4.2. When compressing data with PCA, one must be aware of some theoretical limitations. PCA is a linear projection method that fails to preserve the structure of a non-linear data set.If there is some non-linearity in X (or between X and Y), this nonlinearity can appear as a small perturbation on a linear solution and will not be described by the first PCs as in a linear case. A non-linear transformation of the X-matrix or PC scores matrix can be performed to restore the least-squares approximation property, but the resulting non-linear PCs are strongly dependent upon the pre-selected non-linear form and may not ensure the best representation of distances between points in the original space.71 In practice, PC scores are often successfully used as inputs without transformation because all relevant information about X is usually contained in the first 15 PCs.Alternatively, it is possible to use Fourier analysis,35,41 Hadamard transform72 or wavelet analysis73 to pre-process spectral data before NN modelling. An attractive feature of wavelets is their ability to describe optimally local information from the spectrum, whereas Fourier decomposition is global.If this localised information is related to the non-linearity present in the data, an improvement can be expected if the input matrix is described with wavelet coefficients instead of PC scores or Fourier coefficients. A difficulty lies in the selection of one of the numerous wavelet bases for spectral decomposition. A scheme based on the optimisation of the minimum description length (MDL) criterion in multivariate calibration was explained by Walczak and Massart.74 Whatever the compression method retained, the new subspace (PCs, Fourier coefficients, wavelet coefficients) for sample description must be determined on the training set only.Then the monitoring and test samples can be projected in this subspace to calculate their scores or coefficients. 4.1.6 Data scaling. Once the input variables have been selected or calculated, one must ensure that they can be used for Fig. 11 Data splitting: selection of calibration samples (*) in PC space: a, D-optimum design; b, Kennard–Stone design; c, duplex design. Analyst, 1998, 123, 157R–178R 169Refficiently estimating NN parameters. It is not necessary to mean-center input variables before training since the biases act as offsets in the model. NN training is not based on variance– covariance maximisation, and therefore it is not necessary to scale the different variables to unit variance, even when they are heterogeneous.This is an advantage over methods such as PCR or PLS that require auto-scaling when variables are of different nature. For instance, in process control applications where some variables are continuous and others are binary, the binary variables can be artificially given more weight than the continuous variables because of auto-scaling, and the model interpretation is incorrect. The only constraint for NNs is to scale each input variable so that training starts within the active range of the non-linear transfer functions.Usually, samples are range-scaled with a linear mapping called min–max scaling. Scaling parameters must be determined on the training samples. All samples must be scaled with respect to these parameters. Let Xmin train and Xmax train be the extreme values of variable X in the training set, and let rmin and rmax define the limits of the range where we want to scale variable X.Any sample Xi (from the training, monitoring or test set) must be scaled to a new value Ai as follows: A X X X X r r r i i = - ( ) - - ( )+ min train max train min train max min min (4) For NNs with sigmoid or hyperbolic tangent transfer functions, rmin and rmax are set to 21 and 1, respectively. One must also ensure that the initial weights wi 0 are reasonably small to avoid saturating the transfer functions in the first iterations. We suggest setting them so that 0 < |wi 0| < 0.1.If non-linear transfer functions are used in the output layer, the Y-values must also be range-scaled so that outputs produced by the NN are not in the flat regions of the transfer function (see Fig. 2). In these regions, the derivatives used for weight adjustment are almost zero and learning stops. For a sigmoid transfer function, range-scaling Y to [0.2, 0.8] is recommended, whereas for hyperbolic tangents range-scaling must be performed in the range [20.8, 0.8].In theory, when linear transfer functions are used no range-scaling is needed since they are not bounded. In practice, we found that in the early steps of learning there is a risk that unscaled responses lead to divergent wild steps for weight adjustments that can only be slowly recovered, especially with noisy data. Therefore, we suggest also rangescaling responses to an arbitrary small range. To calculate training, monitoring or test error one must perform an inverse range-scaling to return the predicted responses to their original scale and compare them with experimental responses. 4.2 Determination of network topology The topology of an NN is determined by the number of layers in the NN, the number of nodes in each layer and the nature of the transfer functions. Optimisation of NN topology is probably the most tedious step in the development of a model. To understand the difficulty of topology optimisation, let us first consider the well known bias/variance decomposition of the mean squared error (MSE) for regression problems. It can be demonstrated that E(�y 2 y)2 = E[�y 2 E(�y)]2 + E[E(�y) 2 y]2 (5) where E( ) denotes expectation with respect to the distribution function of a pair (x, y).The first term on the right-hand side of eqn. (5) is related to the variance in the model, whereas the second term describes the bias introduced to counter-balance model flexibility and avoid overfitting.7 The composite contribution of bias and variance to the MSE in a regression model can be represented as a function of model complexity, as in Fig. 12. NNs can perform unbiased estimation of the training set to arbitrary precision and achieve asymptotic consistency. Universal approximation has a cost, however: a truly unbiased NN model (for instance, an NN with an infinite number of hidden nodes) would exhibit a very large variance, would be extremely sensitive to the idiosyncrasies in the training set and could only perform well on noise-free data.7 To attenuate the influence of noise that affects real analytical measurements, one has to constrain NN topology and allow some bias in the is can be done by the following means: reducing the number of layers, nodes and connections in the NN, constraining the form of the transfer functions or using a monitoring set to stop training. We have said earlier that in a first approximation, NNs could be defined as non-parametric models.7,38 This definition is ambiguous and some authors consider NNs as parametric models.25 We can refine the definition now that we have presented the concept of bias in NNs.A model with one nonlinear hidden node is strongly biased and reduces to a parametric sigmoidal regression model. A model with two nonlinear hidden nodes is also strongly biased and will only fit the class of functions that can be modelled by combining the two non-linear transfer functions. As more hidden nodes are added to an NN, the bias is reduced and the number of functions that can be fitted increases exponentially.However, the term semiparametric seems more adapted to NNs used in multivariate calibration, where one tries to build models as parsimonious as possible. 4.2.1 Number of layers. The terminology used to describe NN topology can vary according to the authors, some of them considering the input layer as a simple buffer. We designate the NN represented in Fig. 1 as a three-layer NN, with a 4–3–1 architecture (four input nodes, three hidden nodes, one output node). The theoretical property of universal approximation has been proved for NNs with only one hidden layer.9 In practice, we never obtained better results on calibration problems by using two hidden layers instead of one, even if learning is sometimes faster. A similar observation was made by other authors,3,4 and it is therefore recommended that one uses only one hidden layer in multivariate calibration, unless the relationship to the model seems to be discontinuous.37 In this case an additional hidden layer is necessary.It is possible to add direct connections between the input and output layer of an NN24 as illustrated in Fig. 13. When the input variables have a mixed contribution to the response (some linear and some non-linear), direct connections can handle the linear part and the classical NN builds the nonlinear part of the model. This approach can be interesting with NIR spectroscopic data where the non-linear effects observed generally correspond to small deviations from a linear solution. 24 Direct connections may speed up the learning process and ease model interpretation in situations where descriptors are heterogeneous. Blank and Brown30 compared the performances Fig. 12 Evolution of mean squared error of as a function of the complexity of a model. 170R Analyst, 1998, 123, 157R–178Rof NNs with and without direct connections for the development of multivariate calibration models with non-linear simulated data.They found that directly connected NNs learned more quickly in the initial and intermediate training phases, but NNs without direct connections converged to lower calibration and prediction errors. Dolmotova et al.65 recently compared NNs with and without direct connections for the simultaneous determination of the concentration of three main components in paper coating. The results obtained with both methods were approximately similar.In theory, an NN without direct connections can achieve the same prediction performance as an NN with direct connections, and we therefore prefer NNs without direct connections to reduce the number of adjustable parameters. 4.2.2 Number of input and output nodes. Although NNs have the property to model multiple responses simultaneously, it is recommended that one models only one response at a time and therefore have a single output node.The only exception to this rule is for situations where one wants to predict several correlated responses, such as the concentrations of different constituents of a mixture in a closed system. In that case, all responses can be modelled simultaneously with an NN having one output node per response. To set the initial number of input nodes, two approaches are possible: the stepwise addition approach consists of starting with a deliberately small number of input variables and adding new variables one at a time until the monitoring and/or prediction performance of the NN does not improve any more; the stepwise elimination approach consists of starting with a deliberately large number of input scores and gradually removing (pruning) some of them until the monitoring and/or prediction performance of the NN stops improving.Both approaches are used in practice and no definite recommendation can be given as to which one is better, since they both have advantages and limitations.If PCs are selected according to eigenvalues and the scores used as inputs, the stepwise addition method often leads to quick and satisfactory results, because all necessary information is usually contained in the first few PCs. However, it can happen that most information is contained in, e.g., PC1 to PC5, but some important additional information is also contained in PC10. During stepwise addition, the NN performance will stagnate or degrade between PC6 and PC9 and there are few chances that PC10 is included in the final model.When stepwise elimination is performed, one must include a deliberately large number of input variables in the initial set. Irrelevant variables can be eliminated later, but relevant variables that have not been included in the initial model will not be tested subsequently. Here again, working with PC scores as inputs is advantageous. Using classical techniques (e.g., Malinowski’s factor indication function and reduced eigenvalue test45 or cross-validation75), one can estimate the pseudo-rank of the input data matrix.Then, one selects a few additional PCs (five or six) that may account for possible non-linearity, and the NN training can be started with this initial training set. For calibration problems, the size of the initial set should typically vary between 10 and 15 PCs. The drawback of the stepwise elimination approach is that it can be extremely time consuming, if input variables are tentatively removed by trial and error, because of the large number of possible combinations.60 In neural computation, the relevance of a variable to a model is called its sensitivity.The optimisation of the set of input variables can be accelerated if a method to estimate the sensitivity of each variable is implemented. Several methods have been proposed. The most common is often referred to as Hinton diagrams. It consists of ascribing to each input variable a sensitivity proportional to the average magnitude of its associated connections in the NN, represented on a twodimensional map by square boxes of varying size.Candidate variables to be deleted are those with the lowest sensitivity. In spite of its popularity, this method exhibits severe theoretical and practical limitations.70,76 It is based on an analogy with the classical MLR approach, where the magnitude of a regression coefficient reflects the importance of the relationship between the associated descriptor and the response. In an NN model, input variables that have a linear contribution to the response will be modelled in the linear portion of the sigmoidal transfer function associated with small or medium magnitude weights, whereas the non-linear variables will be modelled in the concave portion of the transfer function associated with large magnitude weights.Therefore, the Hinton diagram ranking method is not based on the intrinsic relevance of a variable to a model, but simply on the nature of its contribution to the response.Linear input variables are systematically flagged as unimportant even when they explicitly contribute to the model. This approach can only give reliable results when the data set is entirely linear, in which case there is no point in using an NN. For the same reason, we are not in favour of training methods based on the principle of weight decay4 that consists of adding to the cost function a term penalising large weights.The approach based on estimation of saliencies is theoretically more stringent.76 The saliency of a weight is the measure of the increase in the NN cost function caused by the deletion of this weight. It is estimated at the end of the training. Deletion of an individual weight wi in an NN can generally be considered as a small perturbation. First, the change in cost function caused by this small perturbation to the weight matrix is approximated by a second-order Taylor series expansion.Ideally, the training is stopped when the NN has converged to a minimum, and therefore the change in cost function can be described using only Hessian terms (second partial derivatives of the error function with respect to weights) in the approximation of the change in error. Hassibi and Stork77 proposed calculating the saliency of a weight k as s w k k kk = [ ] - 1 2 2 1 H (6) where H21 is the inverse of the Hessian matrix. Once the saliency of each weight in the NN is obtained, we use the sum of the saliencies of weights connected to input variable i to determine the sensitivity Si of this variable:76 S s i k k = (7) The saliency estimation method has already been used to optimise NN topology in multivariate calibration.68 It can lead to unstable results in situations where the assumptions made for saliency estimation (small magnitude of weights, training stopped when training error is at a minimum) are not fulfilled.70 Two variance-based approaches for input variable sensitivity determination were proposed recently.70 They are designed for situations where input variables are orthogonal, which is the Fig. 13 Example of three-layer 4–3–1 NN with direct connections. Analyst, 1998, 123, 157R–178R 171Rcase with PC scores. The methods are based on the estimation of the individual contribution of each input variable to the variance of the predicted response.In the first approach, this contribution is determined by partial modelling. First, the NN is trained to estimate the parameters of the model: �y = f(x1, x2,. . .,xn) (8) After training, the sensitivity of each input variable xi is calculated as the variance of the response �y(xi) predicted with the trained NN when all input variables except xi are set to zero: �y(xi) = f(xi) (9) Si = s 2� y(xi) (10) In the second approach, the separate contribution of each input variable to the variance of the estimated response is derived from a variance propagation equation for non-linear combinations of variables. In the case of a two-variable model (x1, x2), this equation is s s s y x x x x y x y x y x y x COV 2 1 2 2 2 2 2 1 2 1 2 1 2 2 = Ê Ë Á � � � + Ê Ë Á � � � + ( ) ¶ ¶ ¶ ¶ ¶ ¶ ¶ ¶ � � � � (11) Since PC scores are orthogonal, the covariance term can be neglected and the sensitivity of input variable xi is calculated as S y x i i xi = Ê Ë Á � � � ¶ ¶ 2 2 s (12) Applying the chain rule several times, one obtains an analytical expression that allows one to determine Si at the end of training.The most interesting characteristic of these two variance-based methods (partial modelling and variance propagation) is that they give extremely stable results. When NNs with the same topology are trained with different sets of initial random weights, they can converge to different local minima on the error surface that are qualitatively equally good and close to each other.In that case the two variance-based methods give similar results, which is not always the case with Hinton diagrams or with the saliency estimation method. Once the sensitivity of each input variable has been estimated, we recommend that one should first try to remove the variable with the lowest sensitivity, and retrain the NN. If the monitoring error decreases after removing the flagged variable, it can be considered as irrelevant for the model and permanently removed, otherwise it must be replaced and another flagged variable must be tentatively removed.Since parsimonious models should be preferred in multivariate calibration, we propose the following methodology for the stepwise elimination of input variables. Let ME(k) be the monitoring error at the kth trial and ME(k + 1) the monitoring error at the next trial after removal of a flagged input variable. Then, If ME(k + 1) @ t 3 ME(k), then remove the flagged variable Else, replace the flagged variable and try to remove the next variable with lowest sensitivity Here t is a tolerance factor that can be adjusted to different values; we suggest t = 1.1.Increasing this factor will result in removing more input variables from the model, at the risk of losing some relevant sources of variance; t should not be lower than 1, otherwise the NN could have a poor generalisation ability. For a given set of input variables, the NN performance will also vary with the number of hidden nodes.Therefore, optimisation of the number of input variables and of the number of hidden nodes should be performed in conjunction: at each step, one should optimise the number of input variables, then the number of hidden nodes, then optimise again the number of input variables and proceed so until the monitoring error stops decreasing. 4.2.3 Number of hidden nodes. A study performed by Tetko et al.55 suggested a fairly wide tolerance of NNs to the number of hidden nodes, provided that overtraining be avoided with an external validation set.However, an upper bound on the number of hidden nodes is of the order of the number of training samples used.53 It was further proved that an NN with n sigmoidal hidden nodes could approximate the response of 2n 2 1 samples.78 These results support the idea that it is not necessary to use large numbers of hidden nodes to fit complex multivariate relationships.On the contrary, large numbers of hidden nodes often accentuate the risk of overfitting.79 To circumvent the problems of overfitting and local minima trapping characteristic of complex networks, Jiang et al.66 proposed a recursive algorithm to add a reasonable number of hidden nodes to an already trained NN. The idea is that an augmented NN is capable of the same approximation as a smaller one, and convergence can be improved with additional hidden nodes. The augmented NN is trained with a modified genetic algorithm (MGA) instead of the usual back-propagation algorithm to avoid local minima.However, the initial topology to be augmented remains to be determined. Conversely, Kanjilal and Banerjee80 presented a strategy for reducing the number of hidden nodes in an NN. The method is based on orthogonalisation of the hidden layer output matrix with singular value decomposition (SVD), after a crude convergence has been reached. Zhang et al.69 presented an algorithm based on a similar concept, that allows one to use all calibration samples for NN training without need for a monitoring set.The initial postulate is that NNs with large numbers of hidden nodes are relatively insensitive to initial conditions, but their generalisation ability is worse than NNs with a hidden layer of reduced size. The proposed scheme consists of starting NN training with a deliberately large hidden layer until an arbitrarily low error is reached, then perform SVD on the hidden layer output matrix H: Hk3h = Uk3k·Sk3h·VTh 3h (13) where h is the number of hidden nodes and k the number of training samples. The number r of dominant singular values in the diagonal S matrix (determined by a variance ratio criterion) is considered as the number of hidden nodes necessary for the NN.A new NN is built, with only r < k hidden nodes, and the new initial weight matrices are determined by least squares fit so that the hidden layer output matrix is HA = [U1U2.. .Ur] (14) Training is then resumed on this pruned NN with improved generalisation ability. We have studied the influence of the number of hidden nodes on the NN error on four non-linear NIR data sets, for which the optimum set of input variables (PC scores) had previously been identified. The first two data sets consist of diesel oil spectra with their corresponding values of viscosity and pour point (eight and four input variables, respectively).The third data set contains spectron of a mineral charge in this polymer as dependent variable (three input variables). The fourth data set contains spectra of gasoline samples and their corresponding octane numbers (thirteen input variables). The first three sets can be considered as strongly non-linear, whereas the last one is only slightly non-linear.32 For each set, models with different numbers of hidden nodes have been designed.Each model was repeated five times to avoid chance correlations due to the random initialisation of the weights. Fig. 14 shows the evolution of average calibration error (CE), monitoring error (ME) and test error (TE) as a function of the number of hidden nodes in the NN, for each of the four data sets. For the three highly non-linear data sets [Fig. 14(a)–(c)], there is first a sharp decrease in error as the second and/or third hidden node are added to the model, whereas for the modelling of octane number [Fig. 14(d), slightly non-linear], the error 172R Analyst, 1998, 123, 157R–178Rcurves remain relatively flat between 1 and 20 hidden nodes. The high initial error values observed in Fig. 14(a)–(c) for one hidden node indicate a situation where the NN is not flexible enough to model highly non-linear relationships. The situation is equivalent to fitting a second- or third-order polynomial with a first-order model. One could think of simply selecting an arbitrary large number of hidden nodes and keep it constant, since the error curves in Fig. 14 remain stable for high numbers of hidden nodes. However, the test samples in these examples are all within the calibration domain. The situation changes significantly when NN are used in extrapolation. For instance, in Fig. 15(a) the CE, ME and TE values are reported for the modelling of diesel oil viscosity, when the test set contains samples with extreme X values. The monitoring and test errors increase as more hidden nodes are added, in contrast to what was observed in Fig. 14(a). The main reason is that several samples that describe the nonlinearity are now in the test set, and the calibration samples Fig. 14 Evolution of NN calibration, monitoring and test error as a function of the number of hidden nodes: a, viscosity data; b, pour point data; c, polymer data; d, gasoline data. Fig. 15 Evolution of calibration, monitoring and test errors as a function of the number of hidden nodes for viscosity data, when some test samples are outside calibration space: a, error; b, standard deviation of error.Analyst, 1998, 123, 157R–178R 173Rmainly describe the linear portion of the viscosity range. One hidden node is sufficient to fit the mild non-linearity present in the calibration set. The fit is slightly better if a second hidden node is added (lower CE), but we already start to overfit the training data, which leads to higher ME and TE values.The situation is now equivalent to fitting a first-order polynomial with a second- or third-order model. If we consider only the TE values, models with one or six hidden nodes give equivalent results, but the one hidden node model has the advantage of producing very stable results: Fig. 15(b) represents the standard deviation of errors on five trials with different initial sets of random weights. A model obtained with one hidden node is quasi-independent from the set of initial weights (standard deviation almost zero).As more hidden nodes are added, different sets of initial random weights can lead to different combinations of transfer functions to build empirical models.81 These models are generally equivalent within the calibration domain, but can lead to different results in extrapolation, as was seen in Fig. 4: when the number of hidden nodes was increased to six or nine, the calibration fit improved slightly but the performance in prediction degraded.We therefore recommend systematically reducing the number of hidden nodes as much as possible, in order to achieve simpler and more robust models. It is always a good idea to compare the performance of a one hidden node model with the performance of a more complex model since many data sets in multivariate calibration are only slightly non-linear. The advantage of models with one hidden node is that the results they produce are stable and independent of the set of initial random weights.81 Moreover, a model with one hidden node reduces to a sigmoidal regression that can be easily interpreted.In an extrapolation calibration study,36 the prediction error of the NN on one data set was reduced by 50% by using one hidden node only. 4.2.4 Transfer function. Kolmogorov’s theorem states that an NN with linear combinations of n 3 (2n + 1) monotonically increasing non-linear functions of only one variable is able to fit any continuous function of n variables.82 The most currently used nonlinear transfer functions in the hidden layer are the sigmoid or hyperbolic tangent functions that are bounded, easily differentiable and exhibit a linear-like portion in their centre, so that data sets that are only slightly non-linear can also be modelled (see Fig. 2). These two functions are popular because they allow one to fit a large number of non-linearities, but other functions can be tried. For instance, Gemperline et al.12 performed multivariate calibration with NNs on UV/VIS data using in their hidden layer combinations of linear, sigmoid, hyperbolic tangent and square functions, to accommodate different types of non-linear response in different spectral regions.The transfer function(s) in the output layer can be linear or non-linear. In many situations, if the number of hidden nodes is sufficient, all modelling is done in the hidden layer. It was observed that in some situations where data were mainly linear, non-linear output transfer functions could introduce distortion in the predicted responses,16 as illustrated in Fig. 3(a). If a linear output transfer function is used, any linear node in the hidden layer can be replaced with a direct connection between input and hidden layer (because two successive linear transformations can be reduced to a single one), which reduces the number of adjustable parameters in the NN. The safest procedure is try both types of output transfer functions (linear and non-linear) during topology optimisation and to base the decision on the shape of residuals for models constructed with the same input variables. 4.3 Training of the network 4.3.1 Learning algorithms. Two general modes of learning can be distinguished: incremental learning and batch learning. Incremental learning consists of successively updating the weights in the NN after estimating the error associated with the response predicted for each sample presented in a random order.In the batch learning mode the errors of all training samples over each iteration are first summed and the parameters are adjusted with respect to this sum. The former approach has the advantage that it superimposes a stochastic component on the weight update. This can help the NN escape from local minima on the error surface in the hyperspace of the weights. A drawback is that the method is prone to the phenomenon of thrashing: the NN can take successive steps in opposite directions that may slow learning. Batch learning provides a more accurate estimate of the gradient vector4 and faster convergence, but it also requires more memory storage capacity. The relative efficiency of both approaches is usually data set dependent.The incremental approach seems particularly suited for very homogeneous training sets21 or for on-line process control applications4 where the composition of the training set is constantly modified.Training an NN is an optimisation problem, and several methods are available for this task. It is not possible to review in detail all algorithms available, but the main types of algorithms will be summarised and their particularities outlined. The gradient descent algorithm performs a steepest-descent minimisation on the error surface in the adjustable parameters Fig. 16 Detection of representativity problems between training and monitoring set on r.m.s. error curves: a, lack of representativity; b, chance correlation with initial set of weights. 174R Analyst, 1998, 123, 157R–178Rhyperspace. This algorithm was described and popularised by Rumelhart and McClelland83 in 1986. The excessively slow convergence of the basic algorithm and its tendency to become trapped in the numerous local minima of the error surface triggered the need for improvements such as the addition of a momentum term in the weight update, that allows one to smooth the error surface and to attenuate oscillations in the bottom of steep valleys.The speed of the algorithm can be significantly enhanced by using adaptive parameters (learning rate and momentum rate) for each weight in the NN. This is the basis of the delta-bar-delta84 and extended delta-bar-delta85 algorithms, that have been successfully applied in multivariate calibration. 30 Faster convergence can be reached with second-order optimisation methods, based on the determination or approximation of the Hessian matrix of partial second derivatives of the cost function: these methods typically have a convergence time one order of magnitude smaller than the gradient method or its derivatives.In the Newton–Raphson method, the Hessian matrix is used to adjust the descent direction at each step, and convergence is reached in a single step if the error surface is quadratic, with ellipsoidal contours. Currently, one of the most popular and efficient second-order methods for NN training is the Levenberg–Marquardt algorithm,8 which is a compromise between gradient descent and Newton–Raphson optimisation.At each step, an adaptive parameter allows the algorithm to transit smoothly between the gradient direction and the Newton–Raphson direction. The inverse Hessian matrix is only estimated and iteratively updated to avoid tedious calculations. Applications of this algorithm for NN training in multivariate calibration have recently been reported.32,68,70,79 Conjugate gradient optimisation is an alternative second-order technique that also uses the Hessian matrix, but the algorithm is formulated in such a way that the estimation and storage of the Hessian matrix are completely avoided.8 With conjugate gradient optimisation, each new search direction is chosen so as to spoil as little as possible the minimisation achieved by the previous one, in contrast to the winding trajectory observed with the gradient method.This method is guaranteed to locate the minimum of any quadratic function of n variables in at most n steps. Genetic algorithms (GA) have been used for NN training.66,86 This global search method allows one to overcome the problem of becoming trapped in local minima, but at the expense of a long computing time because each individual in the population represents a different NN model. In addition, a number of parameters must be set to define the population size and evolution mode, and therefore this approach cannot be easily implemented.Random optimisation consists of taking successive random steps in the weight space and discarding all steps that do not reduce the cost function. In contrast to the classical backpropagation algorithm, random search is guaranteed to find a global minimum,87 but the computation time is so high that the method is never used in practice. Instead, GA or random optimisation can be used as preliminary techniques to optimise the initial set of weights in the NN, then the training is continued with a back-propagation-based method. 4.3.2 When to stop training. As mentioned previously, a monitoring set has to be used in order to reduce the tendency of NN to overtrain and therefore overfit the training data. The evolution of the monitoring error must be followed during training. The frequency of monitoring error estimation has to be determined by the user; ideally it should be performed after each iteration.Consecutive monitoring error values are stored in a vector, and several criteria can be applied to retain the optimum set of weights: train the NN for a pre-defined large number of iterations and retain the set of weights corresponding to the minimum of the monitoring error curve; stop training and retain the last set of weights as soon as the monitoring error is below a pre-specified threshold; or stop training and retain the last set of weights as soon as the decrement between two successive monitoring errors is below a pre-specified threshold.One must also check that the training error is reasonably low at the number of iterations retained, and that the representativity between the training and the monitoring set is ensured. A useful way to detect lack of representativity between training and monitoring set is when the r.m.s. error curves for both sets are separated by a large gap in the region where they flatten, as shown in Fig. 16(a).3,88 Alternatively, it is possible that the optimum monitoring error is reached while the training error is still relatively high [Fig. Fig. 17 Visualisation of sample repartitions on hidden nodes (hn) output maps for ICP data: a, hn1–hn2; b, hn1–hn3; c, hn2–hn3. Analyst, 1998, 123, 157R–178R 175R16(b)]. This can be due to chance correlation, for instance when the initial set of random weights brings the model near a local minimum on the monitoring error surface. Chauvin89 demonstrated that in NNs with complex architectures, late validation minima could sometimes be deeper than the first local minimum.In both cases (large gap between monitoring and training error curves, or early minimum for monitoring), a different splitting of data between the two subsets should be considered. The sensitivity of the NN solution to initial conditions is a well known issue that was discussed by Kolen and Pollack.81 To overcome effects due to chance correlation, several trials must be performed with different sets of initial random weights.55 At least five trials are recommended.The topology corresponding to the lowest average monitoring error should be retained, provided that the variability of predictions is not significantly higher than with other topologies. Once the topology has been established, any set of weights leading to an acceptable monitoring error can be retained for the final model. It is recommended, however, to test it against a validation set, if available, before performing predictions on unknown samples.Some approaches have been presented that avoid the need for a monitoring set, such as the method based on hidden node pruning presented in Section 4.2.3.69.80 Since no overfitting is observed in the later stage of training with this approach, it is claimed that no monitoring set is necessary. This seems particularly attractive for situations where the number of calibration samples is low.In practice, we found that the method was giving very good results when no particular overfitting problem was observed with a classical NN, but in situations where we had difficulty with a classical NN a monitoring set was also necessary with the hidden node pruning approach. 4.3.3 Model interpretation. NNs have more to offer than a simple empirical model. The sensitivity plots that we have presented earlier describe the relative influence of the different input variables in the final model.In addition, examination of the projection of the samples on the hidden nodes of the NN is often informative.37 We performed a calibration model for the quantitative analysis of traces of lead in water, using inductively coupled plasma atomic emission spectrometry (ICP-AES) data as input (14 descriptors). At the end of training, if we display the activation of hidden nodes versus each other, we obtain plots comparable to score plots (Fig. 17). The five measurement replicates marked with asterisks are easily identified as probable outliers.Such plots are instructive and also allow visualisation of clusters present in the data, but they are rarely used. When data must first be compressed, visualisation is performed on the scores before modelling instead. We displayed in Fig. 18(a)–(c) the activation of the three hidden nodes at the end of training for the ICP-AES data NN model. Fig. 18(d) and (e) show the activation of the two hidden nodes in the non-linear model for polymer charge concentration.To estimate the relative importance of each hidden node in the final model, we have reported the value of the magnitude of the weight between this hidden node and the output node in parentheses. This is possible because all hidden nodes are connected to one output node only. Therefore, the magnitude of the connecting weights can directly be compared, which is not the case for weights connected to input nodes.The activation of hidden nodes for ICP-AES data indicates that this data set is mainly linear, whereas the transfer functions Fig. 18 Visualisation of hidden nodes activation: a, ICP data, hn1, w = 20.36; b, ICP data, hn2, w = 20.54; c, ICP data, hn3, w = 0.60; d, polymer data, hn1, w = 20.12; e, polymer data, hn2, w = 0.33. 176R Analyst, 1998, 123, 157R–178Rfor the modelling polymer data are activated in their strongly non-linear portion. Thus we obtain information on the degree of non-linearity of a given data set, even when the exact form of the model is unknown.Recently, several groups have investigated the assessment of statistical confidence intervals for predictions with NNs. Dathe and Otto72 derived confidence intervals using the bootstrap method. After finding the optimum topology of the NN, they erase a portion of the calibration matrix and randomly fill it with replicate samples from the remaining portion. An arbitrary number nsets of calibration matrices is created, and nsets models are built with the pre-defined topology.An external test set is used to predict the responses with each of the bootstrapped NN models, and standard deviations of predicted responses can be calculated. Derks and Buydens90 also worked on the calculation of confidence intervals and compared three forms of bootstrapping. The advantage of the bootstrap approach is that the derived confidence intervals contain all sources of variability (experimental noise, model errors, effect of different sets of random weights), thus yielding a worst case estimation.The drawback is that the confidence intervals derived correspond to an NN topology, not to a single model with a fixed set of weights. 5 Conclusions As is often the case in chemometrics, data pre-treatment and presentation (number of samples, detection of outliers, data compression and splitting) are critical issues that should not be overlooked.Experience has proved that several failures of NNs for modelling were indeed due to inappropriate problem formulation. Such issues can be circumvented by focusing on prior model identification, in particular the detection of nonlinearity. Proper a priori non-linearity detection is one of the major difficulties and methods existing so far often fail in the presence of outliers. NNs should become part of the standard toolkit of analytical chemists concerned with multivariate calibration, but it is important to have a clear understanding of their capabilities and limitations.One should not consider NNs as black boxes, but as regression models whose flexibility will depend on the topology defined by the user. In recent years, numerous research efforts have been focused on improving the speed of algorithms used for NN training. With the availability of faster personal computers, the emphasis is no longer on the speed of algorithms but rather on the development of tools to ease topology optimisation, visualisation and model interpretation.The design of an optimum topology is certainly critical and time consuming, but this is true also for the optimisation of parameters for other methods (form of the model in polynomial PCR or PLS, complexity of soft models, number of nearest neighbours in LWR, variables to retain/eliminate in methods based on feature selection/elimination), although it is less emphasised. Moreover, the comment that NNs do not allow inference is somewhat unfair.Some simple plots can provide information on the nature and form of the problem tackled and on the presence of possible clusters or outliers. Several recent research efforts aimed at combining the flexibility and auto-adaptive ability of NNs with the superior interpretability and inference capability of PLS models.91–94 So far, it seems that these methods also combine the pitfalls of both approaches and their application generally requires an optimisation of a large number of parameters.Radial basis function (RBF) networks offer interesting alternatives to MLP in the sense that they allow local training and the final models can be interpreted in terms of logical rules.38,53,95 Another approach to gain insight into a complex problem is to combine the use of classical MLP (for prediction) with counter-propagation NNs to obtain contour plots of the input and output variables.60,61 6 Acknowledgements The authors are grateful to Vita Centner and Fr�ed�eric Estienne for fruitful discussions.This work received financial support from the European Commission (SMT Programme contract SMT4-CT95-2031) and the Fonds voor Wetenschappelijk Onderzoek (FWO, Fund for Scientific Research). 7 References 1 J. Zupan and J. Gasteiger, Anal. Chim. Acta, 1991, 248, 1. 2 S. D. Brown, S. T. Sum, F. Despagne and B. K. Lavine, Anal. Chem., 1996, 68, 21R. 3 J. R. M. Smits, W. J. Melssen, L.M. C. Buydens and G. Kateman, Chemom. Intell. Lab. Syst., 1992, 22, 165. 4 D. Svozil, V. Kvasnicka and J. Posp�ýchal, Chemom. Intell. Lab. Syst., 1997, 39, 43. 5 D. A. Cirovic, Trends Anal. Chem., 1997, 16, 148. 6 M. Bos, A. Bos and W. E. van der Linden, Analyst, 1993, 118, 323. 7 S. Geman, E. Bienenstock and R. Doursat, Neural Comput., 1992, 4, 1. 8 R. Fletcher, Practical Methods of Optimisation, Vol. 1: Unconstrained Optimisation, Wiley, New York, 1980. 9 K. Hornik, M.Stinchcombe and H. White, Neural Networks, 1989, 2, 359. 10 E. Thomas, Anal. Chem., 1994, 66, 795A. 11 C. E. Miller, NIR News, 1993, 4, 3. 12 P. J. Gemperline, J. R. Long and V. G. Gregoriou, Anal. Chem., 1991, 63, 2313. 13 M. S. Danhoa, S. J. Lister, R. Sanderson and R. J. Barnes, Near Infrared Spectrosc., 1994, 2, 43. 14 J. A. van Leeuwen, R. J. Jonker and R. Gill, Chemom. Intell. Lab. Syst., 1994, 25, 325. 15 F. W�ulfert, W. T. Kok and A. K. Smilde, Anal. Chem., 1998, 70, 1761. 16 R. Goodacre, M. J. Neal and D. B. Kell, Anal. Chem., 1994, 66, 1070. 17 R. Goodacre, Appl. Spectrosc., 1997, 51, 1144. 18 S. R. Amendolia, A. Doppiu, M. L. Ganadu and G. Lubinu, Anal. Chem., 1998, 70, 1249. 19 J. R. Long, V. G. Gregoriou and P. J. Gemperline, Anal. Chem., 1990, 62, 1791. 20 T. J. Sejnowski and C. R. Rosenberg, Complex Syst., 1987, 1, 145. 21 J. Hertz, A. Krogh and R. Palmer, Introduction to the Theory of Neural Computation, Addison Wesley, Redwood City, CA, 1991. 22 S. Biswas and S. Venkatesh, in Advances in Neural Information Processing Systems, ed. R. P. Lippmann, J. E. Moody and D. S. Touretzky, Morgan Kaufmann, San Mateo, CA, 1991, Vol. III. 23 B. Hitzmann, A. Ritzka, R. Ulber, T. Scheper and K. Sch�ugerl, Anal. Chim. Acta, 1997, 348, 135. 24 C. Borggaard and H. H. Thodberg, Anal. Chem., 1992, 64, 545. 25 T. Naes, K. Kvaal, T. Isaksson and C. Miller, J. Near Infrared Spectrosc., 1993, 1, 1. 26 J. Verd�u-Andr�es, D. L. Massart, C.Menardo and C. Sterna, Anal. Chim. Acta, 1997, 349, 271. 27 H. Martens and T. Naes, Multivariate Calibration, Wiley, Chichester, 1989. 28 N. B. Vogt, Chemom. Intell. Lab. Syst., 1989, 7, 119. 29 S. Wold, N. Kettaneh-Wold and B. Skagerberg, Chemom. Intell. Lab. Syst., 1993, 65, 3081. 30 T. B. Blank and S. D. Brown, Anal. Chem., 1993, 65, 3081. 31 T. Naes and T. Isaksson, NIR News, 1994, 5, 7. 32 V. Centner, J. Verdu-Andres, B. Walczak, D. Jouan-Rimbaud, F. Despagne, L. Pasti, R.Poppi, D. L. Massart and O. E. de Noord, submitted for publication. 33 S. Sekulic, M. B. Seasholtz, Z. Wang, B. R. Kowalski, S. E. Lee and B. R. Holt, Anal. Chem., 1993, 65, 835A. 34 I. E. Frank, Chemom. Intell. Lab. Syst., 1995, 27, 1. 35 P. H. Hindle and C. R. R. Smith, J. Near Infrared Spectrosc., 1996, 4, 119. 36 L. Pasti, B. Walczak, F. Despagne, D. Jouan-Rimbaud, D. L. Massart and O. E. de Noord, submitted for publication. Analyst, 1998, 123, 157R–178R 177R37 T.Masters, Practicork Recipes in C++, Academic Press, Boston, 1993. 38 E. P. P. A. Derks, M. S. Sanchez Pastor and L. M. C. Buydens, Chemom. Intell. Lab. Syst., 1995, 28, 49. 39 K. Faber and B. R. Kowalski, Chemom. Intell. Lab. Syst., 1996, 34, 293. 40 I. E. Frank and R. Todeschini, The Data Analysis Handbook, Elsevier, Amsterdam, 1994. 41 P. J. Gemperline, Chemom. Intell. Lab. Syst., 1997, 39, 29. 42 M. Hartnett, D. Diamond and P. G. Barker, Analyst, 1993, 118, 347. 43 V. Centner, D. L. Massart and O. E. de Noord, Anal. Chim. Acta, in press. 44 S. Chatterjee and A. S. Hadi, Stat. Sci., 1986, 1, 379. 45 E. R. Malinowski, Factor Analysis in Chemistry, Wiley, New York, 2nd edn., 1991. 46 J. E. Jackson, A User’s Guide to Principal Components, Wiley, New York, 1991. 47 A. H�oskuldsson, Prediction Methods in Science and Technology; Vol. 1: Basic Theory, Thor Publishing, 1996. 48 G. B. Dijksterhuis and W. J. Heiser, Food Qual. Pref., 1995, 6, 263. 49 A. G. Gonz�alez and D. Gonz�alez Arjona, Anal. Chim. Acta, 1995, 314, 251. 50 V. Centner, D. L. Massart and O. E. de Noord, Anal. Chim. Acta, 1996, 330, 1. 51 B. Walczak, Anal. Chim. Acta, 1996, 322, 21. 52 J. H. Wang, J. H. Jiang and R. Q. Yu, Chemom. Intell. Lab. Syst., 1996, 34, 109. 53 D. R. Hush and B. G. Horne, IEEE Signal Process. Mag., 1993, 1, 8. 54 Neural Networks in QSAR and Drug Design, ed. J. Devillers, Academic Press, London, 1996. 55 I. V. Tetko, D. J. Livingstone and A. I. Luik, J. Chem. Inf. Comput. Sci., 1995, 35, 826. 56 J. H. Wikel and E. R. Dow, Bioorg. Med. Chem. Lett., 1993, 3, 645. 57 J. Ferr�e and X. Rius, Anal. Chem., 1996, 68, 1565. 58 R. W. Kennard and L. A. Stone, Technometrics, 1969, 11, 137. 59 R. D. Snee, Technometrics, 1977, 19, 415. 60 J. Lozano, M. Novic, F. X. Rius and J. Zupan, Chemom. Intell. Lab. Syst., 1995, 28, 61. 61 N. Majcen, K. Rajer-Kanduc, M. Novic and J. Zupan, Anal. Chem., 1995, 67, 2154. 62 D. Jouan-Rimbaud, D. L. Massart, C. A. Saby and C. Puel, Anal. Chim. Acta, 1997, 350, 149. 63 D. Jouan-Rimbaud, D. L. Massart, C. A. Saby and C. Puel, Chemom. Intell. Lab. Syst., 1998, 40, 129. 64 N. Dupuy, C. Ruckebush, L. Duponchel, P. Beurdeley-Saudou, B. Amram, J. P. Huvenne and P. Legrand, Anal. Chim. Acta, 1996, 335, 79. 65 L. Dolmotova, C. Ruckebusch, N. Dupuy, J. P. Huvenne and P. Legrand, Chemom. Intell. Lab. Syst., 1997, 36, 125. 66 J. H. Jiang, J. H. Wang, X. H. Song and R. Q. Yu, J. Chemom., 1996, 10, 253. 67 T. B. Blank and S. D. Brown, Anal. Chim. Acta, 1993, 277, 273. 68 R. J. Poppi and D. L. Massart, Anal. Chim. Acta, in the press. 69 L. Zhang, J. H. Jiang, P. Liu, Y. Z. Liang and R. Q. Yu, Anal. Chim. Acta, 1997, 344, 29. 70 F. Despagne and D. L. Massart, Chemom. Intell. Lab. Syst., 1998, 40, 145. 71 J. H. Jiang, J. H. Wang, X. Chu and R. Q. Yu, Anal. Chim. Acta, 1996, 336, 209. 72 M. Dathe and M. Otto, Fresenius’ J. Anal. Chem., 1996, 356, 17. 73 E. R. Collantes, R. Duta, W. J. Welsh, W. L. Zielinski and J. Brower, Anal. Chem., 1997, 69, 1392. 74 B. Walczak and D. L. Massart, Chemom. Intell. Lab. Syst., 1997, 36, 81. 75 S. Wold, Technometrics, 1978, 20, 397. 76 I. V. Tetko, A. E. P. Villa and J. Livingstone, J. Chem. Inf. Comput. Sci., 1996, 36, 794. 77 B. Hassibi and D. G. Stork, in Advances in Neural Information Processing Systems, ed. J. D. Cowan and C. L. Giles, Morgan Kaufmann, San Mateo, CA, 1993, vol. V. 78 E. D. Sontag, in Advances in Neural Information Processing Systems, ed. R. P. Lippmann, J. E. Moody and D. S. Touretzky, Morgan Kaufmann, San Mateo, CA, 1991, vol. III. 79 E. P. P. A. Derks and L. M. C. Buydens, Chemom. Intell. Lab. Syst., 1998, 41, 171. 80 P. P. Kanjilal and D. N. Banerjee, IEEE Signal Process. Mag., 1993, 1, 8. 81 J. F. Kolen and J. B. Pollack, in Advances in Neural Information Processing Systems, ed. R. P. Lippmann, J. E. Moody and D. S. Touretzky, Morgan Kaufmann, San Mateo, CA, 1991, vol. III. 82 R. P. Lippmann, IEEE Trans. Neural Networks, 1995, 5, 1061. 83 D. E. Rumelhart and J. L. McClelland, Parallel Distributed Processing, MIT Press, Cambridge, MA, 1986, vol. 1. 84 R. A. Jacobs, Neural Networks, 1988, 1, 226. 85 A. A. Minai and R. D. Williams, in International Joint Conference on Neural Networks, 1990, vol. 3, p. 676. 86 M. Bos and H. T. Weber, Anal. Chim. Acta, 191, 247, 97. 87 C. G. Looney, Pattern Recognition Using Neural Networks: Theory and Algorithms for Engineers and Scientists, Oxford University Press, New York, 1997. 88 G. Kateman and J. R. M. Smits, Anal. Chim. Acta, 1993, 277, 179. 89 Y. Chauvin, in Advances in Neural Information Processing Systems, ed. R. P. Lippmann, J. E. Moody and D. S. Touretzky, Morgan Kaufmann, San Mateo, CA, 1991, vol. III. 90 E. P. P. A. Derks and L. M. C. Buydens, Chemom. Intell. Lab. Syst., 1990, 41, 185. 91 R. Bro, J. Chemom., 1995, 9, 423. 92 G. Andersson, P. Kaufmann and L. Renberg, J. Chemom., 1996, 10, 605. 93 S. J. Qin and T. J. McAvoy, Comput. Chem. Engng., 1992, 16, 379. 94 T. R. Holcomb and M. Morari, Comput. Chem. Engng., 1992, 16, 393. 95 B. Walczak and D. L. Massart, Anal. Chim. Acta, 1996, 331, 177. 96 B. Hitzmann and T. Kullick, Anal. Chim. Acta, 1994, 294, 243. 97 H. Wei, L. Wang, W. Xing, B. Zhang, C. Liu and J. Feng, Anal. Chem., 1997, 69, 699. 98 H. Chan, A. Butler, D. M. Falck and M. S. Freund, Anal. Chem., 1997, 69, 2373. 99 M. Bos, A. Bos and W. E. van der Linden, Anal. Chim. Acta, 1990, 233, 31. Paper 8/05562I 178R Analyst, 1998, 123, 157R&nd
ISSN:0003-2654
DOI:10.1039/a805562i
出版商:RSC
年代:1998
数据来源: RSC
|
2. |
Determination of fluorene in sea-water by room temperature phosphorescence in organised media† |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 2217-2221
Manuel Algarra González,
Preview
|
PDF (76KB)
|
|
摘要:
Determination of fluorene in sea-water by room temperature phosphorescence in organised media† Manuel Algarra Gonz�alez and Miguel Hern�andez L�opez* Department of Analytical Chemistry, Faculty of Sciences, University of M�alaga, Campus de Teatinos s/n, 29071 M�alaga, Spain. E-mail: sulphur@uma.es Received 4th June 1998, Accepted 18th August 1998 Fluorene, insoluble in water, forms an inclusion compound in aqueous media with b-cyclodextrin, with an equilibrium constant of 2290 ± 150 l mol21 at 15 °C.The inclusion phenomenon was studied by fluorimetric and phosphorimetric techniques. A schematic host–guest model to explain the inclusion complex structure is reported. The phosphorescence spectra showed maximum excitation and emission wavelengths at 304 and 460 nm, respectively. The phosphorescence lifetimes were calculated employing different organic and inorganic perturber atoms, and was 185 ms for 3-bromopropan-1-ol. Optimum conditions of the method were [b-cyclodextrin] = 8 3 1023 m, pH = 6.65, 3-bromopropan-1-ol as heavy atom, sodium sulfite–sulfurous acid as oxygen scavenger, b-cyclodextrin, heavy atom and buffer as addition order, temperature 15 °C and td and tg 0.1 and 13 ms, respectively.The main figures of merit were linear dynamic range 15–2000 ng ml21, detection limit 4.5 ng ml21 and RSD 2.5%. The method has a moderate selectivity against other PAHs and aromatic molecules and a considerable increase in selectivity in comparison with fluorimetric measurements is observed.An application of this technique to fluorene determination in environmental sea-water samples with successful results is described. Introduction Polycyclic aromatic hydrocarbons (PAHs) are a family of compounds which appear as complex mixtures in the environment. Owing to their ubiquity, the characterization, quantification and study of their properties are of great importance.1–4 The main sources of these compounds are anthropogenic pyrolytic and combustion processes related to industrial plants, domestic heating and automobile traffic; from these sources they enter aquatic biota, where they are rapidly taken up and accumulated by both fish and algae.They are particularly interesting because at least some of them have been found to be toxic and a number of epidemiological studies have shown that they can induce the development of certain forms of tumours.5 Fluorene, a three-ring diaromatic hydrocarbon, a precursor of other PAHs is a common component of crude and processed fossil fuels and it has been identified as a carcinogenic agent and an important component of hydrocarbon pollution of aquatic ecosystems, and it appears to be highly toxic to fish and algae.5–7 Its determination in sea-water has been achieved by gas chromatography (GC) with flame ionization detection (GCFID) 5 and mass spectrometry (GC-MS)8 and liquid and micellar chromatography with fluorescence detection (LC-F).9,11 Lebo and Smith12 determined fluorene in fish tissues, plants and sediments.Luminescence spectroscopy is a particularly useful and powerful technique for this purpose because of its high sensitivity;13 other studies have been based on work at low temperature using Shpol’skii spectroscopy with very low limits of detection.14,15 Because of the ability of cyclodextrins and their derivatives to form stable inclusion compounds with numerous species in aqueous media,16,17 they have been widely used as stabilizing and solubilizing systems and moreover to enhance the luminescence phenomena of molecules introduced into the hydrophobic internal cavity.Deoxygenation is a key step for observing phosphorescence and chemical deoxygenation employing sodium sulfite has been successfully used.18 Room temperature phosphorescence of fluorene has been observed by means of bcyclodextrin (b-CD) and micelles but only qualitative data have been reported.19–22 In this work, luminescent characteristics of the inclusion compound of fluorene in b-CD aqueous solution were studied with the object of characterizing the inclusion process involved and proposing a phosphorimetric method for its determination in sea-water.The procedure proposed compared favourably with the fluorimetric technique in terms of selectivity, good limit of detection, low RSD and a wide linear dynamic range. Experimental Reagents b-Cyclodextrin (b-CD) and a-cyclodextrin (Sigma, St. Louis, MO, USA), g-cyclodextrin (Fluka, Buchs, Switzerland), methyl-b-cyclodextrin, hydroxypropyl-a-cyclodextrin, hydroxypropyl- b-cyclodextrin and hydroxypropyl-g-cyclodextrin (Aldrich, Milwaukee, WI, USA) were used in aqueous solution (1022 m) without further purification.Stock standard solutions (2000 mg ml21) of fluorene, phenanthrene, anthracene, pyrene, chrysene, 1,2-benzanthracene, 1,2 : 5,6-benzanthracene, benzo[a]pyrene, benzo[e]pyrene, triphenylene, fluoranthene, acridine and carbazole (Aldrich), naphthalene, acenaphthene (Sigma), biphenyl, 1-naphthol, 2-naphthol (Merck, Darmstadt, Germany), 2-bromofluorene, dibenzothiophene, dibenzofuran and 9-bromofluorene (Acros, Geel, Belgium) were prepared in propan-1-ol (Merck).Organic heavy atoms used were 1,3-dichloropropan- 2-ol, 2,3-dichloropropan-1-ol, 2,2,2-trichloroethanol, 3-chloropropan- 1-ol, 1,4-dibromobutane, dibromomethane, 3-bromopropan- 1-ol, 1,3-dibromopropane, 1,2-dibromethane, 1-bromopropane, 2-bromoethanol, 2,3-dibromopropan-1-ol, 2-bromoethylammonium bromide, iodomethane, 2-iodopropane, diiodomethane, 1,2-diiodoethane, 1-iodopropane, all of ‘for synthesis’ grade from Merck; 1,3-dibromopropan-2-ol of † Presented at the VIIIth International Symposium on Luminescence Spectrometry in Biomedical and Environmental Analysis, Las Palmas de G.C., Spain, May 26–29, 1998.Analyst, 1998, 123, 2217–2221 2217‘for synthesis’ grade (Aldrich), bromoform, of analyticalreagent grade (Janssen, Geel, Belgium) and chloroform, of analytical-reagent grade (Merck).Heavy inorganic salts were potassium bromide, potassium iodide, mercury(ii) bromide, mercury(ii) nitrate monohydrate, mercury(ii) iodide, silver nitrate, thallium(i) nitrate, lead(ii) nitrate, all of analyticalreagent grade from Merck. In all cases, the final concentration of heavy atom was 0.15 m. Sodium sulfite (2 m) and sulfurous acid (5–6% SO2), purchased from Merck, were used for appropriate chemical deoxygenation.Methanol, ethanol, propan-1-ol, propan-2-ol, tert-butyl alcohol, isobutanol, butan- 1-ol, butan-2-ol, cyclohexanol, cyclopentanol, glycerine (Merck) and ethylene glycol monomethyl ether and pentan-1-ol (Sigma) were of analytical-reagent grade and were used without further purification. Other chemicals used were of analyticalreagent grade and all solutions were prepared with ultrapure water (obtained from a Milli-Q/Milli-Q2 system, Millipore, Bedford, MA, USA).Apparatus All excitation and emission temperature phosphorescence (RTP) spectra were obtained with a Perkin-Elmer (Norwalk, CT, USA) LS-5 luminescence spectrophotometer, equipped with a xenon pulsed discharge lamp (9.9 W) pulsed at line frequency (10 ms half-width, 50 Hz) and an F/3 Monk– Guilleson-type monochromator. The spectrometer was connected to a Perkin-Elmer Model 3600 data station microcomputer, provided with PECLS II applications software. The photomultiplier signal was measured with gated electronics.The delay time (td) and the gate time (tg) chosen were 0.1 and 13.0 ms, respectively, and the excitation and the emission slits were set to 5 and 10 nm, respectively. The phosphorescence lifetimes were obtained by employing the Obey–Decay application program. A quartz cuvette (10 3 10 mm) with PTFE stopper to avoid contact with air was used. In order to homogenize the turbid samples in the cuvette, a Model 333 magnetic stirrer (Hellma, Mulheim, Germany) was employed.The temperature was maintained at 15 ± 0.5 °C by an ultrathermostatic water-bath circulator (Frigiterm S-382, Selecta, Barcelona, Spain). A Bransom Model 5200 ultrasonic water-bath (Branson Ultrasonics Co., Danbury, CT, USA) was used to homogenize the samples. General proceduof inclusion phenomena. To aliquots of fluorene (0.05 mg ml21) in acetone solution, gently heated in order to eliminate the solvent, were added increasing volumes of 0.01 m b-CD solution and de-ionised water to give a final volume of 10 ml.These samples were sonicated for 10 min and their fluorescence spectra, at different temperatures, were recorded with excitation at 261 nm. Phosphorimetric determination of fluorene. Into a 10 ml calibrated flask were transferred aliquots of a sample containing 0.05–70 mg of fluorene in propan-1-ol, propan-1-ol to a final volume of 50 ml, 9 ml of 0.01 m b-CD solution, 150 ml of 3-bromo-1-propanol, 0.4 ml of 2 m sodium sulfite and 0.4 ml of sulfurous acid.The samples were sonicated in an ice-bath for 5 min, left until they reached 15 °C and then the phosphorescence intensity was measured at 461 nm with excitation at 304 nm against a reagent blank. Determination of fluorene in sea-water. Aliquots of seawater samples were placed in a 10 ml calibrated flask and 50 ml of propan-1-ol, a mass of solid b-CD (final concentration 8.85 3 1023 m), 150 ml of 3-bromo-1-propanol, 0.4 ml of sodium sulfite and 0.4 ml of sulfurous acid were added.The phosphorescence was measured as described above. Results and discussion Study of inclusion process Fluorene forms an inclusion complex with b-CD of stoichiometry 1 : 1 obtained by the continuous-variations method (Job plot) from absorbance and fluorescence measurements. The equilibrium constants for the complexation between fluorene and b-CD at pH 7.0 and at different temperatures were calculated by the Benesi–Hildebrand method.23 The linear relationship between DF21 and [b-CD]21 gave K, and the thermodynamic parameters DH, DS and DG° for the formation inclusion complex were determined from the temperature dependence of the association constants.24 Inclusion constants were 2290, 757, 470 and 166 l mol21 at 288, 293, 298 and 303 K, respectively, and the thermodynamic parameters were DH = 212.10 kJ mol21, DS = 0.35 kJ mol21 K21 and DG° = 2116.40 kJ mol21, so the inclusion process is governed by a positive entropy change; the process essentially involves a hydrophobic interaction and the fluorene molecule tends to be introduced into the internal cavity of the cyclodextrin to attain the maximum order.Moreover, molecular modelling system software (Hyperchem, from Hypercube, Waterloo, Canada) was used to minimize the energy of the inclusion process and so to optimize the geometry of the complex; to remove intentionally the guest from the internal cavity of the cyclodextrin and to minimize the system, this evolves as a structure of minimum energy (Fig. 1). Other cyclodextrins assayed, all at a concentration of 0.01 m in aqueous solution, such as a-cyclodextrin, hydroxypropyl-a-cyclodextrin, methyl-b-cyclodextrin, hydroxypropyl- b-cyclodextrin, g-cyclodextrin and hydroxypropyl-gcyclodextrin, showed slight changes in the intensity and maximum wavelength of the spectra recorded in their absence, smaller values of the inclusion constants being obtained.Phosphorimetric study Fluorene shows intense phosphorescence in b-CD solution with maximum excitation at 270 and 304 nm and emission maxima at 434 and 461 nm (Fig. 2). Negligible phosphorescence is given by a blank under identical conditions. In other cyclodextrins and derivatives assayed cited earlier phosphorescence emission was not observed, probably owing to a poor host–guest fit (of Fig. 1 Structure of b-cyclodextrin–fluorene complex. 2218 Analyst, 1998, 123, 2217–2221different size) or to steric hindrance of substituents. Other PAHs and aromatic molecules were studied under the same conditions as for fluorene and some spectral characteristics are given in Table 1. Sample treatment. The previous immersion of the sample in an ice-bath and sonication for 5 min produced an immediate and intense emission in relation to the sample prepared at room temperature. In Fig. 3 it can be observed that the kinetics for the sample prepared at room temperature are slow, reaching equilibrium (curve 1) under these conditions at a time longer than 15 min; similar results were obtained when samples were sonicated at room temperature for 5 or 10 min (curves 2 and 3).The sample introduced for 5 min in an ice-bath without sonication (curve 4) attained equilibrium after 5 min, whereas for the sample sonicated for 5 or 10 min (curves 5 and 6) it was instantaneous.Subsequent work was carried out with sonication for 5 min in an ice-bath. Influence of solvent, pH and oxygen scavenger. To avoid elimination of the solvent by heating the sample, the effects on the kinetics and phosphorescence signal in different solvents were studied. An aliquot of 50 ml of each solvent was added to a final volume of 10 ml and it was observed that short and nonlinear alcohols (methanol, ethanol, propan-2-ol, tert-butyl alcohol, isobutanol, butan-2-ol and glycerine) gave a smaller signal than longer and cyclic alcohols (propan-1-ol, butan-1-ol, pentan-1-ol, cyclopentanol and cyclohexanol).The best results were obtained with propan-1-ol; the other solvents gave similar signals with slow kinetics. Different concentrations of propan- 1-ol were studied to evaluate their influence on the kinetics and equilibrium of the inclusion process. Up to 6.6 3 1022 m the emission was instantaneous and for higher concentrations an induction period (ti) was observed (for 0.132 m, ti = 250 s; for 0.396 m, ti = 600 s) (Fig. 4). The molecule of fluorene can compete between the apolar cavity of b-CD and its affinity or solubility in propan-1-ol. An optimum concentration of 0.066 m was chosen. In order to study the influence of pH on phosphorescence, several mixtures with different ratios of 2 m sodium sulfite solution and sulfurous acid were employed. In the pH range 6.3–6.9, the complex developed maximum and constant phosphorescence (Fig. 5). A pH of 6.65 corresponding to a molar ratio of 1 : 2.5 (sulfurous acid : sodium sulfite) was chosen for subsequent work. Other buffers were studied (acetate–acetic acid, citrate–citric acid and phosphate), also adding sodium sulfite to obtain the same concentration as above. In these cases it was necessary to employ high concentrations of buffer to maintain the pH after the addition of sulfite; moreover, the kinetics were the slowest, requiring 20 min for equilibrium.This can be explained by the greater competition between buffer and cyclodextrin in relation to the analyte. On the other hand, this mixture acts as an oxygen scavenger; in the absence of sulfurous acid (pH 9.2) a weak phosphorescence signal was observed and with increasing emission, as for sulfurous acid, with a sulfurous acid : sodium sulfite molar ratio range of 0.35–0.65 equilibrium was obtained immediately and the signal was maximum and constant (pH 6.3–6.9); at higher ratios the signal decreased drastically. Fig. 2 Excitation and emission spectra of fluorene in b-cyclodextrin solution. [Fluorene] = 1 mg ml21; [propan-1-ol] = 0.066 m; [b-CD] = 8.5 3 1023 m; [3-bromopropan-1-ol] = 0.17 m; [sodium sulfite] = 0.08 m; [sulfurous acid] = 0.032 m; pH = 6.65. Table 1 Phosphorescent spectral characteristics of some PAHs and aromatic molecules in b-cyclodextrin aqueous solution Species lex/lem/nm Relative phosphorescence intensity Acenaphthene 293/482, 516 60 Anthracene — — Biphenyl — — 9-Bromofluorene — — Naphthalene 290/474, 508 50 Benzo[a]pyrenea — — Benzo[e]pyrenea — — Pyrenea 323, 339/589 15 Fluoranthenea — — Fluorene 270, 304/434, 461 195 1-Naphthol 300/491, 521 20 Benzo[k]fluoranthene 309/565 17 1,2 : 5,6-Benzanthracene — — 1,2-Benzanthracene — — Acridine — — Dibenzofuran 292/417, 443 580 Dibenzothiophene 289, 323/420, 445 132 Carbazole 295, 320/416, 441 232 Phenazine — — Triphenylene 300/433, 461 52 2-Naphthol 324/489, 519 12 2-Bromofluorene — — Phenanthrene 294/463, 497 152 Chrysene 306/511 17 a Measured with g-cyclodextrin.Fig. 3 Effect of temperature and sonication time on the phosphorescence emission: (1), (2) and (3) at room temperature with sonication for 0, 5 and 10 min, respectively; (4) in an ice-bath for 5 min without sonication; (5) and (6) in an ice-bath with sonication for 5 and 10 min, respectively. [Fluorene] = 1 mg ml21; pH = 6.65; lem = 461 nm. Fig. 4 Effect of the concentration of 1-propanol: (1), 0; (2) 0.066; (3) 0.099; (4) 0.132; (5) 0.264; and (6) 0.396 m.[Fluorene] = 1 mg ml21; pH = 6.65; lem = 461 nm; [3-bromopropan-1-ol] = 0.17 m. Analyst, 1998, 123, 2217–2221 2219Influence of heavy atom. Different organic and inorganic molecules were studied as heavy atom perturbers. Mineral salts [TlNO3, Pb(NO3)2, Hg(NO3)2, etc.] do not produce emission phosphorescence. This was also observed with alkyl halides and halogenated aliphatic alcohols containing chlorine and iodine; only bromine derivatives of these compounds produced emission and intense phosphorescence was observed with 1,3-dibromopropane, 1,3-dibromopropan-2-ol and 3-bromopropan- 1-ol.The results obtained are presented in Table 2, where the relative phosphorescence intensities are given. For the rest of the work 3-bromopropan-1-ol was chosen as its kinetics are faster than those for 1,3-dibromopropan-2-ol with similar intensity emission and lifetime, probably owing to steric hindrance of bromine atoms making the coupling less effective.The effect of the concentration of 3-bromopropan-1-ol on the kinetics of the formation of the b-CD–fluorene–heavy atom trimolecular complex and on the intensity of the phosphorescence signal at equilibrium were studied; a 0.17 m heavy atom concentration was optimum. It is important to note that 2- and 9-bromofluorene do not show phosphorescence emission, that is, the presence of the bromine atom in these molecules does not produce the expected internal heavy atom effect, and even on adding 3-bromopropan-1-ol, these compounds probably do not form inclusion complexes.The lifetimes were calculated by employing the Obey–Decay application program with delay times between 0.1 and 11 ms with 10 measurements and with a correlation of at least 0.99 considering a monoexponential decay (t for 3-bromopropan-1-ol is 185 ms). A delay time of 0.1 ms and a gate time of 13 ms were chosen as giving the best signal-to-noise ratio.Influence of cyclodextrin concentration, temperature and order of addition. The phosphorescence intensity increases and the time to attain equilibrium becomes shorter as the b-CD concentration increases. In all subsequent experiments the b- CD concentration was maintained at 8.85 3 1023 m. A slight decrease in the phosphorescence intensity was observed in the interval 9–20 °C and a greater decrease above 25 °C. The work reported here was carried out at 15 ± 0.5 °C.The order of addition is important and the sequence sample, b-CD, heavy atom, sodium sulfite and sulfurous acid is recommended. The inclusion compound is stable for at least 1 d. Figures of merit of the proposed method By employing the optimum values of the variables, a calibration graph obtained by the least-squares treatment was Ip = 1.70 + 1.35[fluorene] with a correlation coefficient of 1.00, and where Ip is the phosphorescence intensity and the concentration of fluorene is expressed in ng ml21.The linear dynamic range was established between 15 and 7000 ng ml21. The detection limit (4.5 ng ml21) was calculated as three times the standard deviation of seven blanks divided by the slope of the calibration graph. The precision was determined by analysing 10 samples containing 50 ng ml21 of fluorene and it was 2.5% (as relative standard deviation). Interference study In order to evaluate the selectivity of the proposed method, the effect of various PAHs and molecules with similar structures on the phosphorimetric determination of fluorene at the 50 ng ml21 level was examined over a wide range of concentrations.The tolerance criterion established was a deviation from the expected concentration of ±3s, where s is the standard deviation of the analytical signal for the procedure. The results are given in Table 3. The method tolerates a concentration ratio of 200 : 1 of biphenyl and moderate concentrations of naphthalene, acenaphthene, benzo[a]pyrene, benzo[e]pyrene and anthracene.It is important to note the different tolerance ratios for 9- and 2-bromofluorene, probably due to steric hindrance for the bromine atom position in relation to the b-CD cavity. Rubio Barroso et al.25 determined, by spectrofluorimetry, several analytical parameters of fluorene included in b-CD, such as linear dynamic range (0.45–33.2 ng ml21), RSD (2.6%) and detection limit (0.1 ng ml21).No interference study was reported in relation to the effect of other species on the fluorescence of fluorene. The fluorimetric measurements were made under the same conditions as for the phosphorimetric interference study but no heavy atom was added (Table 3). The great increase in selectivity in the phosphorimetric determination of fluorene in relation to the fluorimetric technique {in some cases the increase in tolerance ratio is hundreds of times Fig. 5 Influence of pH on the phosphorescence intensity.[Fluorene] = 1 mg ml21; [b-CD] = 8.85 3 1023 m; pH = 6.65; lem = 461 nm. Table 2 Influence of heavy atoms on the phosphorescence of fluorene Heavy atoma Ip t/ms Heavy atoma Ip t/ms 1,3-Dichloropropan-2-ol — — 2,3-Dibromopropan-1-ol 2 — 2,3-Dichloropropan-1-ol — — Iodomethane — — 3-Chloropropan-1-ol 37 — 2-Iodopropane — — 2,2,2-Trichloroethanol — — Diiodomethane — — 1,4-Dibromobutane 12 120 1,2-Diiodoethane — — Dibromomethane — — 1-Iodopropane — — 1,3-Dibromopropane 122 158 1,3-Dibromopropan-2-ol 183 180 1,2-Dibromethane 40 164 Bromoform — — 1-Bromopropane 60 132 Chloroform 5 — 2-Bromoethanol 33 124 3-Bromopropan-1-ol 180 185 a Other heavy atoms assayed were TlNO3, KBr, HgI2, KI, Pb(NO3)2, HgBr2, AgNO3, Hg(NO3)2 and 2-bromoethylammonium bromide. 2220 Analyst, 1998, 123, 2217–2221(naphthalene, 9-bromofluorene, acenaphthene, benzo[a]pyrene)} is notable; this is due to the intrinsic nature and more restrictive conditions of the phosphorescence emission.Moreover, a synthetic coal tar (fluorene, benzo[a]pyrene, benzo[e]- pyrene, fluoranthene, pyrene and benzo[k]fluoranthene, all of them at a concentration of 50 ng ml21) was analysed for fluorene and the relative error was 5%. Application to sea-water samples The procedure was applied to determination of fluorene in natural and synthetic sea-water. The synthetic sea-water contained per litre the following salts: 0.12 g of CaCO3, 1.75 g of CaSO4·H2O, 29.7 g of NaCl, 2.48 g of MgSO4, 3.32 g of MgCl2, 0.55 g of NaBr and 0.53 g of KCl.Sea-water samples were collected at different locations in the bay of Malaga (southern Spain). The samples, without filtration (the turbidity does not affect the phosphorescence emission), were spiked with fluorene and their phosphorescence intensities measured as described. Table 4 summarises the results; the good results obtained for the phosphorimetric determination of fluorene in sea-water demonstrate its applicability for routine analysis.Acknowledgements We thank the Consejeria de Educaci�on y Ciencia de la Junta de Andalucia (Group FQM 0148) for supporting this study and Professors F. J. Lopez Herrera and S. Pino Gonzalez for designing the molecular model for the host–guest system studied. References 1 J. M. Neff, Polycyclic Aromatic Hydrocarbons in the Aquatic Environment: Sources, Fate and Biological Effects, Applied Science, London, 1979. 2 M. L. Lee, M. V. Novotny and K. D. Bartle, Analytical Chemistry of Polycyclic Aromatic Compounds, Academic Press, London, 1981. 3 A. Bjorseth, Handbook of Polycyclic Aromatic Hydrocarbons, Marcel Dekker, New York, 1983. 4 C. Y. Raoux and P. Garrigues, in Organic Geochemistry: Advances and Applications in the Natural Environment; 15th Meeting of the European Association of Organic Geochemists, ed. D. A. C. Manning, Manchester University Press, Manchester, 1991, pp. 2. 5 P.A. Cerutti, Science, 1985, 227, 375. 6 D. L. Vassilaros, P. W. Stoker, G. M. Booth and M. L. Lee, Anal. Chem., 1982, 54, 106. 7 P. Thomas, H. W. Wofford and J. M. Neff, Aquat. Toxicol., 1981, 1, 329. 8 Y. Hsieh, M. B. Thompson and C. H. Ward, Dev. Ind. Microbiol., 1980, 2, 401. 9 G. Green, J. H. Skerrat, R. Leeming and P. D. Nichols, Mar. Pollut. Bull., 1992, 25, 293. 10 S. A. Wise, M. M. Schantz, B. A. Benner, M. J. Hays and M. J. Schiller, Anal. Chem., 1995, 67, 1171. 11 S. Lopez, S.Rubio and L. M. Polo, J. Liq. Chromatog., 1995, 18, 2397. 12 J. A. Lebo and L. M. Smith, J. Assoc. Off. Anal. Chem., 1986, 69, 944. 13 M. Blanco, V. Cerd�a, J. Coello, J. Gene, H. Iturriaga, S. Maspoch and M. T. Oms, Anal. Lett., 1996, 29, 1603. 14 C. Gooijer, I. Kozin and N. H. Velthorst, Mikrochim. Acta, 1997, 127, 149. 15 P. Garrigues and M. Ewald, Chemosphere, 1987, 16, 485. 16 D. Duchene, Cyclodextrins and Their Industrial Uses, Editions de Sant�e, Paris 1987. 17 K. A. Connors, Chem. Rev., 1997, 97, 1325. 18 M. E. Diaz Garcia and A. Sanz Medel, Anal. Chem., 1986, 58, 1436. 19 S. Scypinski and L. J. Cline-Love, Anal. Chem., 1984, 56, 322. 20 Y. Wei, W. J. Jin, R. H. Zhu, G. W. Xing, C. S. Liu, S. S. Zhang and B. L. Zhou, Spectrochim. Acta, Part A, 1996, 52, 683. 21 V. B. Nazarov, V. I. Gerko and M. V. Alfimov, Russ. Chem. Bull., 1996, 45, 2109. 22 W. Jin and C. S. Liu, Anal. Chem., 1993, 65, 863. 23 H. Benesi and J. Hildebrand, J. Am. Chem. Soc., 1949, 71, 2703. 24 F. Garc�ýa S�anchez, M. Hern�andez L�opez and J. C. M�arquez, J. Inclus. Phenom., 1990, 8, 389. 25 S. Rubio Barroso, L�opez L�opez, C. Val Ontillera and L. Polo D�ýez, Qu�ým. Anal., 1991, 10, 127. Paper 8/04222E Table 3 Tolerable limits of foreign species (fluorene taken, 50 ng ml21) Tolerance ratio (species to fluorene, m/m) Species added Phosphorescence Fluorescence Biphenyl 200 5 9-Bromofluorene 100 0.1 Naphthalene, acenaphthene 40 0.05, 0.1 Anthracene, benzo[a]pyrene, benzo[e]pyrene 20 0.5, 0.2, 0.5 Pyrene 4 1 Fluoranthene, 1-naphthol 2 0.1, 0.1 Benzo[k]fluoranthene 2 1 1,2 : 5,6-Benzanthracene, 1,2-benzanthracene 2 0.5, 1 Acridine, dibenzofuran, 1 0.02, 0.02, dibenzothiophene 0.02 Carbazole, phenazine, triphenylene 1 0.5, 0.2, 0.2 2-Naphthol, 2-bromofluorene 1 0.1, 0.1 Phenanthrene, chrysene 0.5 0.1, 0.1 Table 4 Determination of fluorene in sea-water Sea-water sample Fluorene added/ ng ml21 Fluorene found ± ng ml21 Synthetica 50 49 ± 1.5 Sacaba beachb 50 51 ± 2.1 Harbourb 50 52 ± 3.0 Malagueta beachb 50 50 ± 1.5 El Candado beachb 50 51 ± 1.9 Rincon de la Victoria beachb 50 52 ± 2.2 a Ten determinations. b Three determinations. Analyst, 1998, 123
ISSN:0003-2654
DOI:10.1039/a804222e
出版商:RSC
年代:1998
数据来源: RSC
|
3. |
Physico-chemical characterization of liposomes with covalently attached hepatitis A VP3(101–121) synthetic peptide† |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 2223-2228
M. Muñoz,
Preview
|
PDF (114KB)
|
|
摘要:
Physico-chemical characterization of liposomes with covalently attached hepatitis A VP3(101–121) synthetic peptide† M. Muñoz,a M. García,b F. Reig,b M. A. Alsinaa and I. Haro*b a Unitat de Físicoquímica, Facultat de Farmàcia, Pius XII s/n, 08028 Barcelona, Spain b Departament de Química de Pèptids i Proteïnes, CID, CSIC, Jordi Girona 18–26, 08034 Barcelona, Spain Received 16th June 1998, Accepted 2nd September 1998 The covalent conjugation of a 20-mer peptide belonging to the VP3 capsid protein of hepatitis A virus to the surface of preformed liposomes was investigated.Three different bonds (disulfide, thioether and amide) were established between the peptide sequence and liposomes bearing at their surface appropriate reactive groups. The effect of the relative concentration of the N-[4-(p-maleimidophenyl)butyryl]dipalmitoylphosphatidylethanolamine anchor in liposomes on stability during coupling of the peptide sequence was studied. The interaction of the three liposomal preparations with phospholipids in a biomembrane model system, monolayers at the air–water interface, is also reported.The results showed that although the peptides associate with liposomes in similar yields for the three strategies studied, differences can be observed when their interaction with phospholipid monolayers composed of dipalmitoylphosphatidylcholine is analysed. Introduction Liposomes, or phospholipid bilayer vesicles, were first described in 1965.1 They are composed of substances with low intrinsic toxicity, and they can be formulated in a large range of sizes and chemical compositions.Potential uses of liposomes for drug entrapment and their controlled release became apparent soon after their development. Although in many applications liposomes have not been as powerful tools as expected, they have been shown clearly to potentiate the humoral and the cell-mediated immune responses to different protein antigens.2,3 It is generally accepted that a physical association between liposomes and antigens is a prerequisite for adjuvanticity to occur.However, the method of choice, encapsulation within the aqueous interior of the liposomes or covalent linkage of the antigen to the liposome surface, is not clear. On the one hand, the administration of the encapsulated antigen protects the peptide molecule from enzymatic hydrolysis and can render it more effective. On the other hand, the attachment of a peptide to the liposome surface can better stimulate the immunological process, in spite of the possibility of altering the active conformation of the molecule caused by the covalent linkage.All these aspects must be taken into account when preparing liposome–peptide associates in order to optimize the biological response.4,5 In contrast to the efforts dealing with liposomal surface attached proteins,6 not many studies have been carried out with synthetic peptides. In this sense, peptides can be coupled to liposomes by chemically well defined procedures.7 In this work, we studied in detail the conjugation of the highly immunogenic peptide sequence (101–121) belonging to the VP3 capsid protein of hepatitis A virus8 to the surface of preformed vesicles by means of disulfide, thioether or amide bonds between the liposomes and the synthetic peptide. Moreover, the physico-chemical characterization of the conjugates was performed by measuring the surface activity and the insertion of the three liposomal preparations into monolayers of dipalmitoylphosphatidylcholine.Experimental Chemicals N-a-Fluorenylmethoxycarbonylamino acids, HMPB linker [4-(4-hydroxymethyl-3-methoxyphenoxy)butyric acid] and methylbenzhydrylamine resin (MBHA resin) were obtained from Novabiochem (Nottingham, UK). Dimethylformamide (DMF), dichloromethane (DCM) and piperidine–20% DMF were purchased from Milligen (Eschborn, Germany). Washing solvents such as propan-2-ol, acetic acid and diethyl ether were obtained from Merck (Poole, Dorset, UK).Trifluoroacetic acid (TFA) was supplied by Fluka (Buchs, Switzerland). 2-(1HBenzotriazol- 1-yl)-1,1,3,3-tetramethyluronium hexafluorophosphate (HBTU), N-hydroxybenzotriazole (HOBt), benzotriazol- 1-yloxytris(dimethylamino)phosphonium hexafluorophosphate (BOP), and N,NA-diisopropylcarbodiimide (DIPCDI) coupling reagents were obtained from Novabiochem. Dipalmitoylphosphatidylcholine (DPPC), dipalmitoylphosphatidylethanolamime (DPPE), N-succinimidyl 4-(p-maleimidophenyl)butyrate (SMPB), N-succinimidyl 3-(2-pyridyldithio)propionate (SPDP), and glutaric acid and cholesterol (Chol) were supplied by Sigma (St.Louis, MO, USA). Carboxyfluorescein (CF) was obtained from Eastman-Kodak (Rochester, NY, USA) and was purified as described previously.9 Chloroform and methanol were purchased from Merck. Water was doubly distilled. NHydroxysulfosuccinimide (NHSS) and N-(3-dimethylaminopropyl)- NA-ethylcarbodiimide (EDC) were supplied by Fluka.Peptide synthesis Solid phase synthesis of the VP3(101–121) peptide sequence was carried out following a 9-fluoromethylmethoxycarbonyl– tert-butyl (Fmoc/tBut) strategy as described previously.10 As reported, a cysteine protected with trityl group [Fmoc–Cys(Trt)] was incorporated at the N-terminus in order to obtain a thiol † Presented at the VIIIth International Symposium on Luminescence Spectrometry in Biomedical and Environmental Analysis, Las Palmas de G.C., Spain, May 26–29, 1998. Analyst, 1998, 123, 2223–2228 2223terminal group that allowed the formation of a disulfide or thioether bond with preformed liposomes containing N- [succinimidyl-3-(2-pyridylthio)propionyl]dipalmitylphosphatidyl (PDP-DPPE) or N-[4-(p-maleimidophenyl)butyryl]dipalmitoylphosphatidylethanolamine (MPB-DPPE). Additionally, the two cysteinyl residues present in the original sequence at positions 105 and 109 were substituted by aminobutyric acid (Abu) in order to avoid secondary reactions during the conjugation processes.Purified synthetic peptide was successfully characterized by analytical HPLC, amino acid analysis and electrospray mass spectrometry (Table 1). Liposome preparation Small unilamellar vesicles (SUV liposomes) were prepared with DPPC containing the appropriate derivatized phospholipids. The molar ratio for lipids was DPPC : Chol : derivatised DPPE = 9 : 10 : 1.MPB-DPPE,11 PDP-DPPE12 and NG-PE13 were prepared and purified as referenced. Either the different thiol-reactive dipalmitoylphosphatidylethanolamine derivatives or N-glutarylphosphatidylethanolamine (NG-PE) were incorporated in the composition of liposomes and then phospholipid vesicles were obtained following procedures described previously. 14 Covalent coupling of VP3(101–121) synthetic peptide to liposomes Coupling of VP3(101–121) peptide to preformed vesicles was established by means of disulfide, thioether or amide bonds as shown in Fig. 1. A disulfide or a thiother bond was obtained when the cysteinyl-synthetic peptide in 0.01 m borate buffer (pH 8 or pH 6, respectively) was conjugated to the liposomes, by mixing freshly prepared PDP–DPPE or MPB–DPPE containing liposomes with an equimolar amount of peptide dissolved in 3% DMSO–borate buffer. The coupling reaction was stirred overnight at room temperature and under a nitrogen atmosphere.In order to remove non-coupled peptide, liposomes were dialyzed against 0.01 m borate buffer (pH 8) for SUV–S–S– peptide or against 0.01 m borate buffer (pH) 6 in the case of thioether bonds. To obtain an amide bond between the preformed SUV containing NGPE derivative and VP3(101–121) peptide, liposomes were activated with N-hydroxysulfosuccinimide (NHSS) and N-(3-dimethylaminopropyl)-NA-ethylcarbodiimide (EDC) at room temperature for 1 h. Then an equimolar amount of peptide dissolved in 3% DMSO–0.01 m borate buffer (pH 6) was added.The final pH of the incubation medium was adjusted to 7.5 with 0.2 m NaOH and the conjugation reaction was carried out overnight at 4 °C. Free peptide was separated from liposomes by dialysis against borate buffer (pH 7.5). Characterization of liposome–peptide conjugates Vesicles size was determined before and after the conjugation to the peptide by measurement of the sample diffusion coefficient by photon correlation spectroscopy.The phospholipid content was determined as described previously.15 Determination of liposome-associated peptide was achieved by quantitative amino acid analyses. The analyses were carried out in a Pico-Tag system (Waters, Stockport, UK). Samples were hydrolyzed in 6 m HCl at 110 °C for 24 h. Traces of non-covalent coupled peptide (adsorbed on the vesicle surface) were determined by detection of free peptide by analytical HPLC after disruption of liposomes with methanol.These analyses were performed on a Spherisorb ODS (10 mm) column, eluted with acetonitrile (ACN)–H2O (0.05% TFA) mixtures. Conditions used for liposome samples were 5 min isocratic with 60% ACN–H2O (80 + 20) followed by a gradient from 60 to 100% in 30 min. Fig. 1 Strategies for preparation of liposome-linked VP3(101–121) peptide. 2224 Analyst, 1998, 123, 2223–2228Fluorescence study of liposomal integrity Small unilamellar liposomes containing 1, 2.5 and 5% of MPBDPPE were prepared as described above, the lipid being resuspended in 100 mm 5(6)-carboxyfluorescein (CF).Nonencapsulated CF was removed by dialysis. Liposomes were used immediately for thiol coupling to Cys-VP3(101–121) peptide. Control liposomes were obtained by incubation with non-Cys-containing peptide, and were therefore unable to bind to MPB-DPPE. For fluorescence measurements, aliquots of 20 ml of liposomes were diluted into 1.5 ml of 0.25 m sodium acetate (pH 7.4) and then 20 ml of this solution was diluted into 2.5 ml of the same solution.CF was excited at 490 nm and emission was read at 520 nm. The total liposome encapsulated CF was determined by lysing the liposomes with 2% v/v of 10% Triton. Latency was determined by applying the equation Latency (%) = t F F F t - ¥ 0 100 (1) where Ft is the total fluorescence after liposome lysis and F0 is the initial fluorescence. Surface activity of peptides The surface activity of peptides was studied following widely described procedures16–19 by using a cylindrical 70 ml PTFE minicuvette.The aqueous subphase was HEPES (10 mm, pH 7.4). To measure the equilibrium spreading pressures, increasing volumes of concentrated solutions of peptides were injected beneath the aqueous surface and pressure increases were recorded after 120 min. Insertion of liposomal peptides into monolayers Stock standard solutions of lipid (DPPC) were prepared in chloroform, and different volumes of these solutions (at a concentration approximately of 1 mg ml21) were spread at the air–water interface using a Hamilton syringe, to attain a 5, 10 or 20 nN m21 surface pressure.After a stabilization period of 10 min, the liposomal preparation, at a slightly lower liposomal peptide concentration than that corresponding to the spreading pressure, was injected into the subphase. These concentrations, obtained by means of surface activity curves, were 0.980, 1.050 and 2.192 mm for disulfide, amide and thioether bond, respectively.The equilibrium value for surface pressure increases was determined 60 min after injection. Results and discussion Solid phase synthesis of the HAV-VP3 peptide sequence was successfully carried out using a Fmoc/tBut strategy as described. 10 Coupling of the synthetic peptide to preformed vesicles was established as described previously above by means of disulfide, thioether or amide bonds. As shown in Tables 2 and 3, the vesicle size, the phospholipid content and the amount of peptide associated with each liposome preparation were determined.Our results indicate a similar content of peptide (0.35–0.45 mg peptide per mg phospholipid) for the three preparations, the adsorbed peptide in all cases being invaluable. The liposomes were uniformly distributed in size. However, the large increase in vesicle size of the MPB-DPPE containing liposomes, measured by photon correlation spectroscopy after conjugation of the VP3 peptide sequence, should be noted.In agreement with previously reported data,20 the coupling of protein fragments to liposomes containing 5 mol% MPB-DPPE caused a concentration-dependent increase in size of the liposomes. Effect of relative concentration of the MPB-DPPE anchor on liposome stability Two different parameters were measured to determine the stability of liposomes containing MPB-DPPE during and after peptide coupling: vesicle size and relative release of the encapsulated material measured by fluorescence (latency of entrapped CF).As shown in Fig. 2, the conjugation of a peptide Table 1 Analytical data for synthetic peptides. Eluents: A, H2O (0.1% TFA); B, H2O–ACN (1 + 4) (0.1% TFA). Gradient: 5 min isocratic 70% B, 70–99% B in 20 min Amino acid analysis (theoretical values in parentheses) HPLC (retention time, tR) ESI-MS D = 2.10 (2) 18.4 min M+ = 2593.2 S = 0.60 (1) Q = 2.04 (2) G = 0.98 (1) A = 0.99 (1) V = 2.18 (2) M = 0.74 (1) I = 0.97 (1) L = 1.91 (2) F = 3.87 (4) R = 0.96 (1) W and C not determined Table 2 Vesicle size and phospholipid content of liposome–peptide conjugates Bond Disulfide Amide Thioether Liposome Size/nm 105.40 ± 40.20 105.30 ± 41.04 141.67 ± 91.64 Polydispersity 0.15 ± 0.06 0.28 ± 0.09 0.33 ± 0.19 Liposome–peptide Size/nm 241.23 ± 70.78 180.67 ± 17.12 2980.8 ± 210 Polydispersity 0.31 ± 0.05 0.47 ± 0.08 0.39 ± 0.21 Phospholipid content/mg ml21 2.33 ± 0.41 2.18 ± 0.56 2.97 ± 0.24 Table 3 Determination of the amount of peptide associated with liposomes Bond Parameter Disulfide Amide Thioether Total peptide/ mg ml21 0.88 ± 0.20 0.86 ± 0.28 1.36 ± 0.53 Adsorbed peptide/ mg ml21 0.08 ± 0.03 0.06 ± 0.03 0 Covalent yield (%) 50.75 ± 13.22 60.84 ± 10.77 66.45 ± 1.44 mg peptide/ mg phospholipid 0.35 ± 0.12 0.37 ± 0.05 0.45 ± 0.15 Analyst, 1998, 123, 2223–2228 2225by a thioether bond apparently causes a leakage of about 30–50% of the entrapped CF, the higher percentage corresponding to 5% MPB-DPPE liposomes.Fluorescence measurements were repeated every 24 h for 10 d after the coupling. A similar stability for the three liposomes–peptide preparations was obtained, a 10% release of the encapsulated material being measured. On the other hand, the coupling of peptide to the 5% MPBDPPE liposomes caused a five-fold increase in the initial mean diameter. This result suggests an aggregation of the vesicles that, as shown in Fig. 2, was not reversed after the conjugation process. In contrast, liposome size was stable during the experiment for 1 and 2.5% MPB-DPPE compositions. In Fig. 3, the same experiment was carried out but without the possibility of establishing a peptide–liposome thioether bond, owing to the absence of the cysteinyl residue located at the amino terminus of the peptide. As suggested by the results, the non-binding peptide apparently maintains the stability of the 1% MPB-DPPE containing liposomes.In contrast, in the case of the 5% derivative containing liposomes, very similar latency and vesicle size values were obtained compared with Fig. 2. This proves that the peptide coupling is not responsible for the release of CF and the increase in vesicle size detected. Probably an intrinsic desestabilization is caused by the high content of MPB-DPPE. Similar results were observed for the control preparation of liposomes that were not incubated with peptide and served as a blank during the experiment.To sum up, it seems clear that there is a need to employ a concentration of MPB-DPPE lower than 5% in the liposome formulation in order to avoid aggregation and disruption during the peptide coupling. Physico-chemical characterization of peptide–liposome conjugates Surface activity. The surface activity of the peptide coupled to the liposomes was determined by injecting different volumes of liposome–peptide preparations into the HEPES-buffered surface and recording the surface pressures, p, achieved.The experimental curves were used to determine the peptide concentration to be employed in the kinetics of penetration experiments. The chosen concentration was slightly lower than the saturation concentration. The peptide showed a gradual adsorption at low concentrations. The higher the peptide concentration in the subphase, the faster was the incorporation of the peptide in the interface. Nevertheless, experiments were carried out for 120 min to ensure that the system had reached equilibrium. Results are given in Fig. 4. These experiments were carried out for thioether, amide and disulfide bonds. Liposomes without peptide were used as a control, the pressure increase being less than 10% of the maximum measured. In the case of the thioether bond, the surface pressure achieved was twice that reached with amide and disulfide bonds. A latency time in the range 6–20 min, which decreased as the peptide concentration increased, was observed on mathematical treatment of the experimental data.This time was longer than that recorded for the free peptide but shorter than that obtained when the peptide was encapsulated into multilamellar liposomes. 14 Therefore, a shorter delay in the incorporation of the peptide at the air–water interface can be assumed when peptide is covalently linked at the liposome surface in comparison with the encapsulated peptide.Kinetics of penetration. Monolayer binding properties. The ability of the VP3(101–121) peptide coupled to SUV liposomes to associate with phospholipid monolayers was studied by injecting a slightly lower concentration than the saturation concentration (obtained by surface activity measurements) beneath DPPC monolayers spread at 5, 10 and 20 mN m21 initial surface pressure and recording the changes in the surface pressure when the peptide interacted with the phospholipid.Fig. 2 Effect of relative MPB-DPPE concentration on the integrity of liposomes during peptide coupling measured as latency of entrapped CF and increase in vesicle size during and after thioether reaction. Fig. 3 Effect of relative MPB-DPPE concentration on the integrity of liposomes incubated with a non-reactive peptide, [105,109Abu]VP3(101– 121), measured as latency of entrapped CF and increase of vesicle size during and after the peptide incubation. 2226 Analyst, 1998, 123, 2223–2228In order to obtain more information about peptide–phospholipid interactions, we applied a mathematical treatment to the experimental data: D D p p = + mt K t (2) where Dpm is the maximum pressure achieved at saturation and K is the time to achieve Dpm/2. As shown in Tables 4–6, pm/2 was achieved in less than 15 min for all the liposome preparations.These results are in agreement with the previously described14 high affinity of the free peptide for phospholipid monolayers.Fig. 5 shows that the maximum surface pressure values were achieved for the liposome preparations with the peptide coupled by a thioether bond, the smallest Dpm corresponding to the disulfide peptide–liposome bond formulation. As the interacting molecule is the same in all samples, these differences could only be explained by a variability in the peptide presentation when coupled to liposomes by means of the different bonds studied. These results could be in agreement with the differences observed when the interaction of the above described peptide–liposome preparations with IgG antibodies generated against the peptide sequence was studied, measured by surface plasmon resonance based techniques (data not shown).Furthermore, previous results obtained in our laboratory by circular dichroism for a 12-mer peptide belonging to the (110–121) sequence of VP3 capsid protein of HAV21 proved that the thiother peptide–liposome preparation retains the main conformation obtained for the free peptide in a membrane environment22 to a greater extent than with disulfide or amide bonds.In the present case, (for the VP3(101–121) peptide sequence, further conformational studies are needed to confirm this point. In this respect, circular dichroism experiments are in progress with the liposome–peptide preparations reported here, in order to evaluate whether any alteration in the peptide preferent conformation can be caused by the covalent linkage to preformed liposomes.Acknowledgements This work was supported by grants BIO95-0061-CO3-02 and BIO95-0061-CO3-03 from CICYT, Spain. A predoctoral Fig. 4 Monolayer surface pressure increases for several initial peptide concentrations in the subphase induced by (a) SUV–thioether–peptide, (b) SUV–amide–peptide and (c) SUV–disulfide–peptide. Table 4 Parameters obtained by the mathematical treatment of experimental data from penetration kinetic curves for peptide thioether coupled to SUV liposomes DPPC p0/mN m21 Dpm/mN m21 K/min r2 5 10.30 ± 0.13 7.17 ± 0.42 0.996 10 7.11 ± 0.28 12.12 ± 1.66 0.981 20 4.77 ± 0.10 9.38 ± 0.85 0.991 Table 5 Parameters obtained by the mathematical treatment of experimental data from penetration kinetic curves for peptide disulfide coupled to SUV liposomes DPPC p0/mN m21 Dpm/mN m21 K/min r2 5 5.90 ± 0.21 16.37 ± 1.78 0.989 10 4.53 ± 0.19 17.88 ± 2.14 0.987 20 1.68 ± 0.09 16.44 ± 2.77 0.973 Table 6 Parameters obtained by the mathematical treatment of experimental data from penetration kinetic curves for peptide amide coupled to SUV liposomes DPPC p0/mN m21 Dpm/mN m21 K/min r2 5 8.39 ± 0.16 11.73 ± 0.82 0.995 10 6.55 ± 0.19 14.82 ± 1.38 0.991 20 5.38 ± 0.23 15.12 ± 2.04 0.981 Fig. 5 Pressure increases recorded after the injection of liposome–peptide preparations under DPPC molayers spread at 5, 10 and 20 mN m21 initial surface pressure. Analyst, 1998, 123, 2223–2228 2227CIRIT grant (FI/95-8109) awarded to M.García is also acknowledged. References 1 A. D. Bangham, M. M. Standish and J. C. Watkins, J. Mol. Biol., 1965, 13, 238. 2 G. Gregoriadis, Clin. Immunol. Newsl., 1981, 2, 33. 3 C. R. Alving, J. Immunol. Methods, 1991, 140, 1. 4 G. Gregoriadis, Immunol. Today, 1990, 11(3), 89. 5 J. Allen, Drugs, 1997, 54(4), 8. 6 T. M. Allen, A. K. Agrawal, I. Ahmad, C. B. Hansen and S. J. Zalipsky, Liposome Res., 1994, 4, 1. 7 B. Boeckler, S. Frisch, S. Muller and F. Schuber, J. Immunol. Methods, 1996, 191, 1. 8 R. M. Pintó, J. F. González-Dankaart, G. Sánchez, S. Guix, M. J. Gómara, M. García, I. Haro and A. Bosch, Astrovir. Res., submitted for publication. 9 A. F. Sikorski, K. Michalak and M. Bobrowska, Biochim. Biophys. Acta, 1987, 904, 55. 10 M. Garcia, I. B. Nagy, M. A. Alsina, G. Mezö, F. Reig, F. Hudecz and I. Haro, Langmuir, 1998, 14(7), 1861. 11 F. J. Martin and D. Papahadjoupoulos, J. Biol. Chem., 1982, 257(1), 286. 12 F. J. Martin, T. D. Heath and R. R. C. New, in Liposomes: a Practical Approach, ed. R. R. C. New, Oxford University Press, New York, USA, 1990, pp. 163–182. 13 V. T. Kung and C. T. Redeman, Biochim. Biophys. Acta, 1986, 862, 435. 14 M. Garcia, M. Pujol, F. Reig, M. A. Alsina and I. Haro, Analyst, 1996, 121, 1583. 15 C. W. F. McClare, Anal. Biochem., 1971, 39, 527. 16 I. Martin, I. Haro, F. Reig and M. A. Alsina, Langmuir, 1994, 10(3), 784. 17 K. Bogdam, M. A. Alsina, I. Haro, I. Martin and F. Reig, Langmuir, 1994, 10, 4618. 18 I. Haro, J. A. Perez, F. Reig, C. Mestres, M. A. Egea, M. A. Alsina, Langmuir, 1996, 12(21), 5120. 19 F. M. Mota, M. A. Busquets, F. Reig, M. A. Alsina and I. Haro, J. Colloid Interface Sci., 1997, 188, 81. 20 R. Bredehorst, F. S. Ligler, A. W. Kusterbeck, E. L. Chang, B. P. Gaber and C. W. Vogel, Biochemistry, 1986, 25, 5693. 21 A. Vea, M. J. Gomara, F. Reig and I. Haro, in Solid Phase Synthesis, in the press. 22 J. A. Perez, J. Canto, F. Reig, J. J. Perez and I. Haro, Biopolymers, 1998, 479. Paper 8/04560G 2228 Analyst, 1998, 123, 2223–2228
ISSN:0003-2654
DOI:10.1039/a804560g
出版商:RSC
年代:1998
数据来源: RSC
|
4. |
Spectroscopic studies of the interfacial binding ofHumicola lanuginosalipase† |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 2229-2233
Yolanda Cajal,
Preview
|
PDF (87KB)
|
|
摘要:
Spectroscopic studies of the interfacial binding of Humicola lanuginosa lipase† Yolanda Cajal,*a Josefina Prat,a Allan Svendsen,b Jordi De Bolósa and M. Asunción Alsina*a a Department of Physical Chemistry, School of Pharmacy, University of Barcelona, Av. Joan XXIII s/n, 08028 Barcelona, Spain b Enzyme Research, Novo-Nordisk A/S, 2880 Bagsvaerd, Denmark Received 16th June 1998, Accepted 13th August 1998 The interaction of Humicola lanuginosa lipase (HLL) with small unilamellar vesicles of 1-palmitoyl-2-oleoylglycero-sn-3-phosphoglycerol (POPG) and in the presence of tributyrin (TB) as a substrate was studied by the use of steady-state fluorescence techniques.An inactive mutant with the serine from the catalytic triad changed by alanine (S146A) was used in experiments with TB to avoid interferences from product formation. HLL binds to POPG vesicles in an active or open form for the catalytic turnover, and therefore POPG provides a suitable system for studying the conformational changes involving the movement of the loop of amino acids that covers the active site of the enzyme in solution.Tryptophan (Trp) fluorescence shows that HLL binding to POPG occurs with a change in the environment of Trp residue(s) and that there is only one type of bound form, even in the presence of TB. Accessibility to aqueous quenchers indicates shielding of Trp in the membrane. Fluorescence anisotropy of the enzyme increases on binding to the vesicles, indicating restricted rotational freedom for the Trp due to penetration in the bilayer.Resonance energy transfer experiments using an interfacial membrane probe, 1-[4-(trimethylammonio)phenyl]-6-phenylhexa-1,3,5-triene p-toluenesulfonate (TMA-DPH), and an internal membrane probe, 1,6-diphenylhexa-1,3,5-triene (DPH), indicate that HLL does not penetrate very deeply in the hydrophobic core of the membrane, but preferentially stays close to the lipid interface. Addition of substrate (TB) does not result in any additional changes in the spectroscopic properties of HLL. It is suggested that the observed changes are due to the ‘opening of the lid’ on binding to POPG vesicles, leaving the active site accessible for the substrate to bind.Introduction Lipase from Humicola lanuginosa is a hydrolytic enzyme which breaks down triacylglycerols into free fatty acids and glycerols. Lipase activity is depressed in monomeric substrates, but it is greatly increased in the lipid–water interface, a phenomenon known as interfacial activation.1 The majority of the known structures of lipase have the catalytic center buried beneath a surface loop or “lid” that renders it inaccessible to solvent and substrate molecules.The lid changes its conformation on binding to a lipid–water interface, exposing the active site and creating a non-polar surface which can stabilize the contact between the enzyme and the lipid interface.2 A better understanding of the movement of the surface loop is important because this is believed to be one of the main events in the interfacial activation process. A high-resolution (1.8 Å) three-dimensional structure of Humicola lanuginosa lipase has been determined by X-ray crystallography.3 The structure shows the a/b hydrolase fold characteristic of all lipases, and the consensus catalytic triad Ser–His–Asp forming the active site, which is covered by a lid of nine amino acids with an a-helical structure.This form of the enzyme is catalytically inactive and is referred to as the closed conformation of the enzyme. This lid must move to allow access of the substrate to the active site, forming the open enzyme conformation. In some lipases, such a molecular change was described in molecular detail from the X-ray structure of lipase– inhibitor complexes, for example for Rhizomucor miehei4,5 and more recently for Humicola lanuginosa.6 However, although these crystallographic data are a first step towards elucidating the molecular mechanism underlying lipase activation, the inhibitors used for crystallography are usually very different structurally from the natural substrates.For this and other reasons, the structure of lipase in these complexes cannot be considered equivalent to the active form generated in the lipid– water interface under physiological conditions. The actual process of interfacial activation involving the displacement of the lid when bound to the lipid surface is very complex and it has not been fully elucidated.In this paper, the conformational changes that take place in Humicola lanuginosa lipase (HLL) upon binding to the lipid vesicles of 1-palmitoyl-2-oleoylglycero-sn-3-phosphoglycerol, and also in the presence of tributyrin as a substrate, were studied by fluorescence techniques. The catalytic activity of the enzyme under the same conditions as used for fluorescence was measured spectroscopically by UV spectrophotometry using pnitrophenylbutyrate as a substrate.An inactive mutant, with active site Ser in position 146 of the chain mutated by Ala (S146A HLL), was used in the fluorescence experiments to avoid interference from product formation. Changes in the fluorescence emission of the protein on binding to the vesicles and accessibility to aqueous quenchers iodide and acrylamide were used to study the interaction with the bilayer. Fluorescence anisotropy and resonance energy tranfer measurements were used to demonstrate the incorporation of the lipase in the vesicle membrane.† Presented at the VIIIth International Symposium on Luminescence Spectrometry in Biomedical and Environmental Analysis, Las Palmas de G.C., Spain, May 26–29, 1998. Analyst, 1998, 123, 2229–2233 2229Experimental Chemicals Humicola lanuginosa wild type (HLL), and its mutant in which the catalytic serine is changed to alanine (S146A HLL), were obtained from Novo Nordisk (Denmark).For the calculations of enzyme concentration, a molar absorptivity of 43 000 l mol21 cm21 with an Mr of 32 kDa were used.7 p-Nitrophenylbutyrate (PNPB) and tributyrin (TB) were obtained from Sigma (St. Louis, MO, USA). N-(7-Nitro- 2,1,3-benzoxadiazol-4-yl)dioleoylphosphatidylethanolamine (NBD-PE) and 1-palmitoyl-2-oleoylglycero-sn-3-phosphoglycerol (POPG) were purchased from Avanti Polar Lipids (Alabaster, AL, USA) and 1,6-diphenylhexa-1,3,5-triene (DPH) and 1-[4-(trimethylammonio)phenyl]-6-phenylhexa- 1,3,5-triene p-toluenesulfonate (TMA-DPH) from Molecular Probes (Eugene, OR, USA).Vesicle preparation Small unilamellar vesicles (SUVs) of POPG alone or with the fluorescent probes NBD-PE, TMA-DPH or DPH were prepared by evaporation of a mixture of the lipids in CHCl3–CH3OH (2 + 1 v/v). The dried film was hydrated for a lipid concentration of 20 mm, and then sonicated in a G112SPIT bath type sonicator (Laboratory Supplies, Hicksville, NY, USA) above the gel– fluid transition temperature until a clear dispersion was obtained (typically 2–4 min).Vesicles were annealed for 1 h above their transition temperature before use. Kinetic protocols The kinetics of hydrolysis of PNPB were monitored as the change in the absorbance at 400 nm, corresponding to the absorption maximum of the p-nitrophenolate anion with a molar absorptivity of 14 000 l mol21 cm21, as described elsewhere.8 Data acquisition and manipulation were carried out in a Hewlett-Packard (Avondale, PA, USA) diode array spectrophotometer with a 0.5 s acquisition time.Measurements were made in 0.7 ml of 10 mm TRIS buffer (pH 8.0) at 25 °C in a quartz cuvette. Vesicles of POPG were added to the cuvette followed by PNPB from a stock solution in tetrahydrofuran, and the reaction was initiated by addition of the lipase and gentle mixing of the cuvette. The rate of hydrolysis was calculated from the slope of the initial zero-order phase of the progress curve, and expressed as turnover number per second. Tryptophan (Trp) fluorescence and quenching experiments Fluorescence measurements were carried out in 10 mm TRIS buffer (pH 8.0) at 25 °C on an AB-2 spectrofluorimeter (SLMAminco, Urbana, IL, USA) with constant stirring.Tryptophan fluorescence spectra were recorded with an excitation wavelength of 280 nm over an emission range 295–450 nm, with 4 nm slitwidths. S146A was added from a stock solution to a final concentration of 2.47 mm and titrated with vesicles and TB.The sensitivity (PMT voltage) was adjusted to 1% for the Raman peak from the buffer blank at the same excitation wavelength. The contribution from vesicles to the signal was negligible. Quenching of tryptophan fluorescence of S146A by iodide and acrylamide was recorded at 320 nm (excitation at 280 nm). Appropriate amounts of POPG vesicles were added to a solution of 1.1 mm enzyme and aliquots of quencher were added with continuous stirring.Acrylamide was added in increasing amounts from a 3.3 m stock solution in water, and potassium iodide was added from a 2 m stock solution containing 0.25 mm Na2S2O3 to avoid I32 formation. The final quencher concentration ranged from 0 to 350 mm. Quenching results were analyzed according to the Stern–Volmer equation for collisional quenching: F0/F = 1 + KSV[Q] where F0 and F are the fluorescence intensities in the absence and presence of quencher, respectively, [Q] is the molar concentration of quencher and KSV is the Stern–Volmer quenching constant.In this range of concentrations there is no deviation from linearity (r2 = 0.99). Binding isotherms The binding of S146A to POPG vesicles containing 2.5% of NBD-PE was determined as the increase in the resonance energy transfer (RET) signal from Trp residues in the enzyme to the labeled phospholipid in the interface at 535 nm (excitation at 285 nm).Vesicles in buffer were titrated with S146A HLL from a stock solution 23.2 mm in water, and the stoichiometry was determined directly from the plot of dF versus lipid/ enzyme (mol/mol). The relative change in fluorescence, dF, is defined as (F2F0)/F0, where F0 and F are the intensities without and with enzyme, respectively. Since the lipid concentration was very low (26.7 mm), the contribution from light scattering was less than 3%. Fluorescence anisotropy Steady-state fluorescence anisotropy measurements were carried out on an AB-2 spectrofluorimeter, with L-format fluorescence polarizers. The excitation wavelength was set at 280 nm and the emission at 340 nm with excitation and emission slitwidths at 4 nm.Titrations were carried out at 25 °C, adding aliquots of POPG vesicles and/or TB to a solution 2.47 mm S146A HLL in 10 mm TRIS buffer (pH 8.0). All solutions were stirred continuously during the measurements. The fluorescence anisotropy (r) was calculated automatically by the software provided with the instrument, according to r = (Ivv – I Vh)/(Ivv +2 IVh) where Ivv and IVh are the intensity of the emitted polarized light with the emission polarizer parallel or perpendicular to the excitation polarizer, respectively.Anisotropy values were automatically corrected for dependences in the detection system (G-factor correction). Changes in anisotropy were represented as (r 2 r0)/r0, where r0 and r are the anisotropy values before and after addition of vesicles, respectively.All measurements were carried out in triplicate. Penetration in the membrane by resonance energy transfer (RET) measurements A solution 2.5 mm S146A HLL in 10 mm TRIS buffer (pH 8.0) was titrated with aliquots of POPG SUVs labeled with 2.5% of the fluorescent probe TMA-DPH or DPH. The excitation wavelength was 280 nm and fluorescence emission was measured with excitation and emission slitwidths of 4 nm each. The relative decrease in fluorescence emission intensity of the enzyme (dF) at 340 nm was plotted.Results and discussion Activity measurements of Humicola lanuginosa lipase (HLL) in POPG vesicles as interfaces Substrates for HLL are di- and tri-glycerides, molecules that when suspended in water at concentrations above their 2230 Analyst, 1998, 123, 2229–2233solubility limit form droplets or aggregated forms of uncontrolled dispersity, or adsorb on the available surfaces.It is therefore almost impossible to generate well defined and controlled structures of known surface area. We have recently reported a new strategy to analyze steady-state kinetics for the hydrolysis by HLL8 of a soluble substrate (e.g., TB or PNPB) partitioned into the interface provided by POPG vesicles. As shown in Fig. 1, HLL hydrolyzes PNPB partitioned in the interface of POPG SUVs at a rate that is more than 100-fold higher than that for the monodisperse substrate (less than 20 s21 for PNPB in solution and 1200 s21 in the presence of 0.054 mm POPG).POPG is not a substrate for lipase, but POPG vesicles act as a neutral diluent interface to which the lipase binds and probably adopts its open or active conformation, in which its active site is accessible for the substrate.8 Therefore, POPG provides a well controlled interface to study the changes that take place in HLL lipase involving the opening of the lid when the enzyme adopts its fully open conformation.Changes in Trp fluorescence on binding to POPG vesicles The fluorescence of Trp is dependent on the polarity of its environment. This effect can be used to study the interaction of the HLL lipase in POPG vesicles. HLL has four Trp residues; one of them, Trp-89, is located in the lid that covers the active site in the closed or inactive form of the enzyme in solution. The major changes in the enzyme conformation on binding to the lipid–water interface are associated with the lipid contact zone region, which includes the lid.2 Therefore, changes in Trp fluorescence on binding to the POPG interface, where the enzyme adopts its open conformation, can give valuable information on the environment of the lid.As shown by the difference spectra (Fig. 2) as a function of POPG concentration, the binding of S146A HLL is accompanied by an increase in the emission intensity at 320 nm and a decrease at 360 nm. A well defined isosbestic point suggests that the fluorescence change is due to a one-step equilibrium between two forms of the enzyme, namely the free form in solution and the form bound to the interface. Addition of TB at the end of the titration with POPG does not result in additional changes in fluorescence; therefore, we can speculate that the lipase bound to POPG vesicles is already in the open form, even in the absence of substrate. The same experiment but without addition of TB was carried out with HLL wild type, and the changes in fluorescence were comparable (not shown).Fluorescence quenching of S146A HLL by iodide or acrylamide Iodide and acrylamide were used as aqueous-phase quenchers of the Trp fluorescence. The Stern–Volmer plots of the quenching of Trp fluorescence by iodide and acrylamide are shown in Fig. 3. In the presence of POPG vesicles, the Trp residues of S146A HLL become less accessible to the quenchers, as compared with the enzyme in buffer; this suggests penetration of the enzyme in the lipid bilayer and shielding of Trp from the aqueous quenchers.The Stern–Volmer quenching constants (KSV) for a bimolecular collisional quenching process were calculated from the apparent slopes of the plots of F0/F21 versus [Q] calculated by linear regression analysis, and are given in Table 1. No significant change in KSV was seen on addition of TB to the POPG–enzyme system, in agreement with the absence of additional changes in Trp fluorescence under the same conditions as described previously. Comparing the two quenchers, the decrease in KSV on binding to the interface was more pronounced when iodide was used.This can be explained because iodide is considered to access only the surface Trp Fig. 1 Reaction progress, monitored as change in absorbance at 400 nm, for the hydrolysis of 0.07 mm PNPB by 3.9 nmol HLL. (a) PNPB alone; and (b) PNPB in the presence of 0.054 mm POPG vesicles. Fig. 2 Change in the intensity of fluorescence in the emission spectra of the S146A mutant of HLL (2.47 mm) as a function of POPG concentration (from 0.013 to 0.266 mm) in 10 mm TRIS buffer (pH 8.0).Excitation was set at 280 nm. Fig. 3 Stern–Volmer plots showing the tryptophan fluorescence quenching of S146A HLL (1.1 mm) in 10 mm TRIS buffer (pH 8.0) by (A) iodide or (B) acrylamide. (8) S146A HLL in buffer; (2) S146A HLL with POPG vesicles; (×) S146A HLL with POPG vesicles and TB. POPG concentration 267 mm; TB concentration, 40 mm. The excitation wavelength was set at 280 nm and emission was monitored at 320 nm.Analyst, 1998, 123, 2229–2233 2231residues, whereas acrylamide would have good access to all but the most highly buried Trp residues.9 Therefore, it is possible that iodide will mainly quench the fluorescence of Trp-89 in the lid. When the enzyme binds to the vesicles, the lid possibly becomes more buried in the membrane and thus less accessible to iodide quenching.On the other hand, acrylamide may be able to access also the other more buried Trp residues of the enzyme, which most probably do not change their accessibility to the water on binding to POPG vesicles. Therefore, the small decrease in KSV (from 10.57 to 9.03 l mol21) probably reflects the shielding of Trp-89. This will be further analyzed with a new mutant enzyme which contains only Trp-89. The accessibility of the Trp fluorophore to acrylamide was also determined for the enzyme bound to 2 mm TB.At this concentration, well above its solubility limit, the substrate is in an aggregated form, and therefore the enzyme will bind in the active form. In this case, KSV was 8.9 l mol21 (Table 1), very similar to the value in POPG vesicles, and again pointing towards the idea that in both cases we are looking at the same active enzyme form. Fluorescence anisotropy of HLL Steady-state fluorescence anisotropy can be used for the characterization of the incorporation process of a protein into a vesicle bilayer.The rotational freedom of the protein will be restricted upon incorporation. Consequently, the anisotropy of the enzyme will increase upon addition of the vesicles.10,11 This phenomenon was used to monitor the lipase–vesicle interaction. For this experiment, as for all spectroscopic studies, S146A HLL mutant was used. The free enzyme exhibited an anisotropy value (r0) of 0.052 05 ± 0.003 26, and the anisotropy value of the bound protein (rmax) was 0.1137 ± 0.003 52.In Fig. 4, the relative change in anisotropy is shown as a function of lipid concentration. Incorporation of the enzyme to the POPG vesicles results in an increase in anisotropy until all the enzyme is bound. An estimate of the stoichiometry of the binding can be obtained directly from this plot, which gives a POPG to enzyme ratio of 80:1 (mol/mol). At the end of the titration with POPG, TB was added to the cuvette, and no additional changes in anisotropy were seen.Estimation of protein-to-vesicle ratio by RET Evidence for the interaction of S146A HLL with POPG vesicles comes from the resonance energy transfer signal resulting from the Trp donors of the mutant to the NBD–lipid acceptor at the interface, with a RET distance of about 15 Å.12 As shown in Fig. 5, the RET intensity at 535 nm increases with increase in the amount of protein added, and the intensity reaches a maximum when the surface is essentially covered with the protein.From the plot, a stoichiometry of 80 POPG molecules per enzyme molecule was calculated, in perfect agreement with the estimate from anisotropy measurements described above. If we consider the size of the SUV vesicles, with a diameter of approximately 50 nm, the number of lipid molecules that form each vesicle is approximately 8000; according to this calculation, the stoichiometry is 100–110 lipase molecules bound per vesicle. Measurement of penetration depth of HLL in the membrane Interaction of proteins with membranes can be monitored by RET.9 This method is based on the non-radiative transfer of the excited state energy from a donor to an acceptor molecule.The extent of energy transfer depends mainly on the extent of overlap between the emission spectrum of the donor and the absoption spectrum of the acceptor and on the orientation and distance between them. The donor molecules used in these experiments were the Trp residues from S146A HLL, and different molecules located in the lipid membrane were used as acceptors.The RET from enzyme molecules that remain free in solution can be considered negliglible, because the critical distance for RET to occur is in the range 15–30 Å, depending on the donor–acceptor pair. As acceptors for the enzyme excitation energy, 2.5% of the membrane probes DPH and TMA-DPH were incorporated in POPG vesicles. The location of these probes in the bilayer membrane is different.DPH is deeply buried in the hydrophobic core of the membrane, whereas TMA-DPH is located more shallowly in the membrane, in order to allow the TMA group to remain near the membrane surface.13 By the use of an internal label and a label close to the interface, information can be obtained on the penetration of the enzyme in the bilayer, as already demonstrated with Candida cylindracea lipase.14 In Fig. 6, a typical example of the change in the fluorescence spectra of S146A HLL upon titration with TMA-DPH vesicles Table 1 Stern–Volmer quenching constants for the quenching of S146A HLL fluorescence.Enzyme, 1.1 mm; POPG, 267 mm; TB 40, mm unless indicated otherwise KSV/l mol21 Conditions Iodide quenching Acrylamide quenching S146A 1.92 10.57 +POPG 0.97 9.03 +POPG+TB 0.85 9.18 +TB (2 mm) 2 8.90 Fig. 4 Fluorescence anisotropy change of S146A HLL as a function of POPG concentration. The excitation wavelength was set at 280 nm and the emission at 340 nm with excitation and emission slitwidths of 4 nm.Titrations were carried out at 25 °C, adding aliquots of POPG vesicles to a solution 2.47 mm of S146A in 10 mm TRIS buffer (pH 8.0). Fig. 5 Binding of S146A HLL to POPG vesicles (26.7 mm) containing 2.5% of NBD-PE. The RET intensity from Trp of enzyme to the label at the interface was measured at 535 nm (excitation at 280 nm). 2232 Analyst, 1998, 123, 2229–2233is shown. The decrease in the fluorescence of the enzyme at around 330 nm coincides with an increase in the signal emanating from the label, indicating that the lipase binds to the membrane and that it is in close proximity to the interfacial probe.Similar spectra were obtained with DPH as acceptor. In Fig. 7, the change in fluorescence of the Trp signal of S146A HLL bound to POPG vesicles with TMA-DPH or DPH are shown. The RET is more efficient with vesicles containing the shallow probe TMA-DPH, with a steep decrease in the enzyme fluorescence upon titration with the vesicles.This indicates that the enzyme is readily associated with the vesicles. The energy transfer to DPH labeled vesicles is less efficient, suggesting that the enzyme in the active form preferentially stays at the vesicle interface, with the a-helical lid possibly oriented parallel to the membrane surface and with the hydrophobic residues, such as Trp-89, slightly inserted in the hydrophobic part of the membrane but remaining close to the interface.This results are consistent with the fluorescence and quenching experiments described above. This is an important aspect in the activity of HLL, and it will be a topic for further investigation with mutants containing only one Trp residue. Acknowledgements Excellent technical assistance from and helpful discussions with Helena Carvajal Laiglesia in all the fluorescence experiments included in this paper are greatly appreciated. Y. Cajal is supported by a contract for the incorporation of Doctors in Spanish research groups (MEC, Spain).We thank Dr. J. Vind and Dr. S. A. Patkar for preparation of the enzymes. This work was supported by EEC Biotechnology Program BIO- 4-97-2365. References 1 L. Sarda and P. Desnuelle, Biochim. Biophys. Acta, 1958, 30, 513. 2 A. Svendsen, I. G. Clausen, S. A. Patkar, K. Borch and M. Thellersen, Methods Enzymol., 1997, 284, 317. 3 U. Derewenda, L. Swenson, R. Green, Y. Wei, G. G. Dodson, S. Yamaguchi, M. J. Haas and Z. S. Derewenda, Nature Struct. Biol., 1994, 1, 36. 4 A. M. Brzozowski, U. Derewenda, Z. S. Derewenda, G. Dodson, D. M. Lawson, J. P. Turkenburg, F. Björkling, B. Huge-Jensen, S. A. Patkare and L. Thim, Nature (London), 1991, 351, 491. 5 U. Derewenda, A. M. Brzozowski, D. M. Lawson and Z. S. Derewenda, Biochemistry, 1992, 31, 1532. 6 D. M. Lawson, A. M. Brzozowski, S. Rety, C. Verma and G. G. Dodson, Protein Eng., 1994, 7, 543. 7 M. Martinelle, M. Holmquist and K. Hult, Biochim. Biophys. Acta, 1995, 1258, 272. 8 O. G. Berg, Y. Cajal, G. L. Butterfoss, R. L. Grey, M. A. Alsina, B-Z. Yu and M. K. Jain, Biochemistry, 1998, 37, 6615. 9 G. P. Kurzban, G. Gitlin, E. A. Bayer, M. Wilcheck and P. M. Horowitz, Biochemistry, 1989, 30, 8537. 10 G. A. Roberts and M. P. Tombs, Biochim. Biophys. Acta, 1987, 902, 327. 11 J. M. Elsen, L. M. A. Unen, L. Bloois, M. A. Busquets, W. Jiskoot, P. Hoogerhout, J. Wilting, J. N. Herron and D. J. A. Crommelin, Biochemistry, 1997, 36, 12583. 12 M. K. Jain and W. L. C. Vaz, Biochim. Biophys. Acta, 1987, 906, 1. 13 R. D. Kaiser and E. London, Biochemistry, 1998, 37, 8180. 14 E. W. J. Mosmuller, E. H. W. Pap, A. J. W. G. Visser and J. F. J. Engbersen, Biochim. Biophys. Acta , 1994, 1189, 45. Paper 8/04575E Fig. 6 Changes in the fluorescence spectrum of S146A HLL (2.5 mm) upon addition of POPG vesicles containing 2.5% TMA-DPH. Lipid concentration ranges from 0 to 666 mM. Excitation wavelength, 280 nm. Fig. 7 Fluorescence emission at 340 nm monitored during the titration of S146A HLL (2.5 mm) with POPG vesicles containing (:) the internal label DPH or (-) the interfacial label TMA-DPH. Excitation wavelength, 280 nm. Analyst, 1998, 123, 2229–2233 2233
ISSN:0003-2654
DOI:10.1039/a804575e
出版商:RSC
年代:1998
数据来源: RSC
|
5. |
Resolution of a multicomponent polycyclic aromatic hydrocarbon system in micellar media by linear variable angle fluorescence applying distinct chemometric techniques† |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 2235-2242
J. Amador-Hernández,
Preview
|
PDF (104KB)
|
|
摘要:
Resolution of a multicomponent polycyclic aromatic hydrocarbon system in micellar media by linear variable angle fluorescence applying distinct chemometric techniques† J. Amador-Hernández,a A. Cladera,b J. M. Estela,b P. L. L�opez-de-Albaa and V. Cerd`a*b a Instituto de Investigaciones Científicas, Universidad de Guanajuato, 36000 Guanajuato, Gto., Mexico b Department of Chemistry, Universitat de les Illes Balears, E-07071 Palma de Mallorca, Spain Received 3rd June 1998, Accepted 5th August 1998 The simultaneous spectrofluorimetric determination of aromatic polycyclic hydrocarbons of environmental interest, namely benzo[a]pyrene, benzo[e]pyrene, benzo[ghi]perylene, coronene, dibenzo[a,h]anthracene and indeno[1,2,3-cd]pyrene, in micellar media by using the non-ionic surfactant polyoxyethylene 10 lauryl ether (POLE) was investigated.In order to employ the highest sensitivity signals for the determination of each of the compounds in the mixture, the corresponding linear variable angle fluorescence spectra were recorded.Owing to the high spectral overlap observed, resolution of the multicomponent system was carried out by applying known algorithms such as multiple linear regression, partial least squares regression type 1 and artificial neural networks. The results obtained for both synthetic mixtures and water samples of two different origins spiked with known amounts of hydrocarbons of interest were satisfactory. Introduction Since the discovery of their carcinogenic properties, polycyclic aromatic hydrocarbons (PAHs) have become a topic of great interest and have been widely studied.1–3 Owing to their occurrence during the incomplete combustion or pyrolysis of organic matter common in several natural or anthropogenic processes, these compounds are present in air, soil and water in different concentrations and international organizations such as the World Health Organization (WHO) and the US Environmental Protection Agency (EPA) have regulated the control of these compounds in the environment. Among the most commonly used instrumental techniques for the determination of these compounds are gas chromatography (GC) and high performance liquid chromatography (HPLC,4–6 together with UV/VIS absorption spectrophotometry and spectrofluorimetry; 3,7 however, conventional spectroscopic techniques require prior separation processes owing to the severe spectral interferences that occur. Synchronous fluorescence8 has become a tool of great sensitivity and selectivity in multicomponent analysis and has been satisfactorily applied to the simultaneous determination of several PAHs.9 Nevertheless, the high spectral overlap involving some hydrocarbons cannot be eliminated using this technique and alternatives are required.Over the last few years, the introduction of computers in laboratories has allowed the automation of numerous instrumental techniques. Thus, through the development of adequate software in fluorescence, fast and simple registration of spectra of variable angle is feasible without requiring hardware modification, increasing the possibilities of application of the former technique.10 On the other hand, the development of software specialized for chemometric techniques allows the application of powerful mathematical algorithms to resolve complex chemical systems by using multiple analytical signals which, owing to the interferences present in the determination, deviations from linearity, noise, etc., cannot be directly resolved.11,12 Among these algorithms are multiple linear regression (MLR), principal component regression (PCR), partial least squares regression (PLS) and artificial neural networks (ANN).13–15 Variable angle fluorescence combined with multicomponent analysis chemometric techniques has allowed the development of sensitive analytical methods for the simultaneous determination of PAH mixtures with satisfactory results, without requiring prior separation processes.16 The luminescence characteristics of various compounds change considerably in the presence of either ionic or non-ionic surfactants compared with an aqueous medium, giving rise to variations in the excitation and emisson spectra, increasing the fluorescence intensity and quenching effects, among others.17 Through the adequate exploitation of the photophysical and/or photochemical phenomena that occur in micellar systems, the development of more sensitive and/or selective fluorimetric, phosphorimetric and chemoluminescent methods is feasible.18 In this paper, the development of a spectroscopic method for the simultaneous determination of six PAHs of environmental interest is presented.This method is based on the linear variable angle fluorescence spectra obtained in a micellar medium with a non-ionic surfactant [polyoxyethylene 10 lauryl ether (POLE)].The corresponding spectral interferences are resolved through the application of three different multivariate algorithms. Two of them (MLR and PLS) consider the linear relationship between the analytical property of interest (fluorescence intensity) and the concentration of the analytes. The third algorithm (ANN) finds the relationship between the variables by means of training processes. This comparison was aimed at observing the predictive capacity of the methods studied in the resolution of a system with severe spectral interferences.† Presented at the VIIIth International Symposium on Luminescence Spectrometry in Biomedical and Environmental Analysis, Las Palmas de G.C., Spain, May 26–29, 1998. Analyst, 1998, 123, 2235–2241 2235Experimental Apparatus A Perkin-Elmer (Norwalk, CT, USA) LS-5 luminescence spectrometer was used, provided with a xenon discharge lamp (9.9 W) pulsed at line frequency and a 1 31 cm quartz cell. This instrument was connected to a compatible PC via an RS232C serial interface for instrument control and data acquisition, representation and storage.Fluoropack software, described previously,10 was used. A Selecta (Abrera, Barcelona, Spain) ultrasonic bath was also employed. MULTI3 (SCIWARE, Palma de Mallorca, Spain), Unscrambler II (Camo, Trondheim, Norway) and Winnet (Sciware) software packages were used for the mathematical treatment of data in a Pentium microcomputer. Reagents The polycyclic aromatic hydrocarbons benzo[a]pyrene (BAP), benzo[e]pyrene (BEP), benzo[ghi]perylene (BGP), coronene (COR), dibenzo[a,h]anthracene (DBA) and indeno[1,2,3- cd]pyrene (IND) were obtained from Sugelabor (Madrid, Spain), polyoxyethylene 10 lauryl ether (POLE) from Sigma (St.Louis, MO, USA) and methanol from Prochem (Wessel, Germany). Water purified with a Milli-Q system (Millipore, Bedford, MA, USA) was used throughout. 2.5 3 1022 m aqueous standard solution of POLE was prepared.Stock standard solutions of BAP (20 mg l21), BEP (10 mg l21), BGP (10 mg l21), COR (10 mg l21), DBA (20 mg l21) and IND (20 mg l21) were prepared by dissolving the appropriate amounts of the pure analytical reagent grade products in methanol and stored at 4 °C. Working standard solutions were prepared daily by appropriate dilution. Procedure PAH working standard solutions were prepared in 25 ml calibrated flasks with 5% methanol and 2.5 ml of 2.5 3 1022 m POLE. The concentration ranges of the PAHs used were 0.3–16 mg l21 BAP, 2–110 mg l21 BEP, 1–45 mg l21 BGP, 0.5–27 mg l21 COR, 0.5–29 mg l21 DBA and 1.6–106 mg l21 IND.The linear variable angle spectra were recorded by two proposed routes (A and B) with a resolution of 1 nm, maintaining the width of the excitation and emission slits at 5 and 10 nm, respectively. Results and discussion In order to establish the optimum experimental conditions for the simultaneous determination of the six PAHs, several factors were studied.According to some studies, the fluorescence intensity of several chemical species increases considerably in micellar media in relation to aqueous media because the micelles protect the excited singlet state of fluorophores, the radiant processes, therefore, competing favourably with other deactivation processes. In the present work, the non-ionic surfactant POLE was seleed to establish the micellar medium, since the former increases considerably, on some occasions, the analytical signal of the six PAHs without producing interferences in their respective signals.In Table 1 the relationships corresponding to the fluorescence intensities in micellar and aqueous media for each of the PAHs are shown at the excitation and emission wavelengths where the maximum fluorescence intensity in micellar media is observed. Second, the effect of the surfactant concentration was studied. A final concentration of 2.5 mm POLE was selected.This concentration is higher than the critical micellar concentration. Higher concentrations do not yield changes in the fluorescence intensity of the six PAHs. Subsequently, the effect of the methanol concentration was examined, a final concentration of 5% being selected as optimum, which has a negligible effect on the analytical signal of the compounds. The total fluorescence spectra corresponding to the six PAHs obtained under the experimental conditions described in the Procedure section are depicted in Fig. 1 as contour plots, where the linear curves represent excitation and emission wavelengths with isoluminescent signals. As can be observed, there are severe interferences between the analytical signals of these compounds, hence their simultaneous determination by conventional fluorimetric techniques is not feasible, even when using variable angle fluorescence, and the application of algorithms such as MLR, PLS or ANN is required to resolve the multicomponent system satisfactorily. Table 1 Ratio of the fluorescence intensity in micellar and aqueous media at the excitation and emission wavelengths, where a maximum signal in the micellar medium is observed for each of the PAHs Compound lex/nm lem/nm RIF BAP 264 408 8 BEP 289 392 3 BGP 300 417 20 COR 303 449 6 DBA 298 398 30 IND 256 360 4 Fig. 1 Contour plots of the total fluorescence spectra for the PAHs studied: (a) BAP; (b) BEP; (c) BGP; (d) COR; (e) BDA; and (f) IND. 2236 Analyst, 1998, 123, 2235–2241Linear variable angle spectra In order to obtain two-dimensional spectra representative of the six PAHs, variable angle fluorescence was employed. The program used for the automatic control of the fluorimeter (Fluoropack) allows the recording of variable angle spectra in a direct, fast and simple way without requiring modification of the instrumentation. Although these spectra can be defined by either a succession of linear trajectories or a circular trajectory, in this case the first type of spectra was chosen owing to the localization of the maximum fluorescence intensities.Two routes were defined for recording the linear variable angle spectra: route A, with five linear trajectories where the maximum fluorescence intensities corresponding to the six compounds are located at intermediate points regarding these trajectories (374 pairs of excitation and emission wavelengths), and route B, with eight linear trajectories where the maximum fluorescence intensities are located at the intersections (196 pairs of excitation and emission wavelengths), which are depicted in Fig. 2. Subsequently, the excitation and emission wavelengths were replaced with a series of progressive whole numbers from 1 to 374 for route A and from 1 to 196 for route B in order to identify the pairs of excitation and emission wavelengths with unrepeatable and equally spaced numbers for the adequate representation and analysis of the two-dimensional spectra.The linear variable angle spectra obtained through routes A and B pertaining to BAP, BEP, BGP, COR, DBA and IND are represented in Figs. 3 and 4, respectively. Series of samples and validation To establish the concentration ranges in which the compounds would be analyzed, the linear relationship between the fluorescence intensity located at the excitation and emission wavelengths where the maximum signal is present (Table 1) and its corresponding concentration was used as a criterion.The linear ranges obtained were 0.3–16 mg l21 for BAP, 2–110 mg l21 for BEP, 1–45 mg l21 for BGP, 0.5–27 mg l21 for COR, 0.5–29 mg l21 for DBA and 1.6–106 mg l21 for IND. However, neither the linearity in the full spectral range nor the effect of the internal filter were studied in order to establish the capacity of the different algorithms used to resolve such irregularities if present. In spite of the fact that in order to establish the calibration model through MLR a relatively small number of samples is required, the complexity of this system, due to the number of components and spectral overlapping requires a large number of calibration samples in order to apply the PLS or ANN algorithms.The composition of the series of calibration samples used in the present work is given in Table 2. For the external validation of the calibration models proposed by the different mathematical techniques, 10 synthetic mixtures, Fig. 2 Selected routes for recording the linear variable angle spectra, represented by the contour plot corresponding to a mixture of the six PAHs: solid line, route A, with 374 points; broken line, route B, with 196 points. Fig. 3 Linear variable angle spectra recorded through route A for the PAHs of interest under the proposed experimental conditions: (a) BAP; (b) BEP; (c) BGP; (d) COR; (e) BDA; and (f) IND. Fig. 4 Linear variable angle spectra obtained following the trajectory described by route B: (a) BAP; (b) BEP; (c) BGP; (d) COR; (e) BDA; and (f) IND.Analyst, 1998, 123, 2235–2241 2237whose compositions were at random within the proposed concentration ranges, are described in Table 3, were prepared. Several statistical parameters were calculated to validate both internally and externally each of the proposed calibration models. For internal validation purposes, the following parameters were calculated in the series of calibration samples, R (R of Pearson), SEC (standard error of calibration), RMSD (root mean square deviation) and REP (relative error of prediction), whereas for the external validation the average percentage recovery ± the standard deviation (%R ± s), SEP (standard error of prediction), RMSD and REP were calculated in the series of synthetic mixtures.14 The equations for the statistical parameters RMSD, SEC (and SEP) and REP are RMSD SEC and SEP REP e e e = - ( ) ( ) = - ( ) - = - ( ) S S S x x n x x n x x x n 2 2 2 1 100 (%) where x is the real concentration and xe the estimated concentration of one of the analytes in a series of samples (calibration or validation), �x is the average real concentration of the analyte of the former series and n corresponds to the number of samples evaluated.Multiple linear regression (MLR) Once the experimental conditions have been selected and the corresponding samples have been prepared, the different multicomponent analysis algorithms were applied, starting with the MLR.This method corresponds to the simplest type of the multivariate calibration techniques. In the calibration model of a multicomponent system, a linear dependence is assumed between the measured property (fluorescence intensity in this case) and the concentration of analytes, namely all the species which contribute to the analytical signals considered. For the application of the algorithm the MULTI311 program was used, which allows one to perform three different types of calibration: single standard, multiple standard and multiple standard addition.In the present work eight calibration models constructed from four series with different samples, whose compositions are described in Table 2, were studied, taking into account the full linear variable angle fluorescence spectra recorded through routes A and B: MLR/1 models, considering the singlecomponent spectra of the six compounds at medium concentration levels (samples 13–18); MLR/2 models, with the singlecomponent spectra of the six compounds at three different concentration levels (samples 1–18); MLR/3 models, constructed from the single-component spectra of the six compounds at medium concentration levels and six spectra of mixtures (samples 13–18 and 26–31); and R/4 models, constructed from 14 spectra of mixtures (samples 19–32). Subsequently, each proposed calibration model was validated both internally and externally.According to the analysis of the statistical parameters, the most satisfactory results in the simultaneous determination of the six PAHs correspond to the calibration model MLR/3 constructed with the linear variable angle spectra corresponding to route B (Table 4). However, the least favourable results were obtained in the determination of BAP, mainly when present at low concentrations in relation to another major compound, possibly owing to the severe spectral interferences of the remaining compounds or a possible deviation from linearity.Partial least squares regression type 1 (PLS-1) Of the existing multivariate calibration techniques based on the principal component analysis, PLS-1 was employed, which Table 2 Composition of the series of calibration samples. All concentrations are expressed as mg l21 Sample BAP BEP BGP COR DBA IND 1 0.3 0 0 0 0 0 2 16 0 0 0 0 0 3 0 2 0 0 0 0 4 0 110 0 0 0 0 5 0 0 1 0 0 0 6 0 0 45 0 0 0 7 0 0 0 0.5 0 0 8 0 0 0 27 0 0 9 0 0 0 0 0.5 0 10 0 0 0 0 29 0 11 0 0 0 0 0 1.6 12 0 0 0 0 0 90 13 8 0 0 0 0 0 14 0 55 0 0 0 0 15 0 0 22 0 0 0 16 0 0 0 13 0 0 17 0 0 0 0 14 0 18 0 0 0 0 0 45 19 6 40 0 0 0 40 20 0 0 19 11 6 0 21 0 110 4 0 1 0 22 0.4 4 2 27 0 0 23 0 2 2 2 0 106 24 16 0 0 1 1 6 25 0.3 0 45 0 2 2 26 0 64 10 6 3 16 27 2 12 0 19 8 32 28 9 8 1 1 0 61 29 0.5 5 24 2 15 0 30 3 48 13 12 4 26 31 10 28 27 7 2 13 32 1 20 5 4 18 67 33 6 72 3 5 7 35 34 4 44 14 17 5 8 35 2 10 32 6 10 48 36 8 16 3 5 11 29 37 4 56 6 2 10 51 38 3 8 18 10 19 19 39 11 52 16 5 2 10 40 4 6 29 12 3 37 41 4 28 11 7 7 26 42 10 10 10 10 10 10 43 2 24 6 8 6 70 44 7 36 7 3 6 45 45 6 33 4 4 13 22 46 7 14 16 9 14 11 Table 3 Composition of the series of synthetic mixtures used to validate the proposed methods.All concentrations are expressed as mg l21 Sample BAP BEP BGP COR DBA IND 1 5 11 21 2 11 23 2 5 34 11 3 3 5 3 2 31 13 16 4 24 4 3 23 4 9 10 54 5 5 76 3 6 2 5 6 3 6 7 11 7 38 7 1 46 12 6 9 43 8 5 38 8 0 5 18 9 0 32 26 5 8 0 10 3 0 18 0 0 40 2238 Analyst, 1998, 123, 2235–2241allows the analysis of compounds composing the system in an independent way from the most relevant spectral information.Although the linear relationship between the analytical property of interest (here fluorescence intensity) and the concentration of each of the analytes of the multicomponent system is a priori considered, it is possible to evaluate variables with deviations from linearity due to phenomena such as the presence of interfering substances or the internal filter effect.On the other hand, PLS-1 allows one to establish the number of principal components or factors required for the construction of the calibration model of each of the components in an independent way. In this case, Unscrambler II software was used for the application of the algorithm in the simultaneous determination of the six PAHs of interest. Initially, four calibration models were constructed: two PLS- 1/1 models with 46 calibration samples indicated in Table 2, by using the complete linear variable angle spectra of routes A and B, and two PLS-1/2 calibration models, the first 12 samples being eliminated (34 samples in total).Experimental data were subjected to a mathematical treatment prior to calibration, which consisted in the processes of mean centering and autoscaling of all the variables in relation to the inverse of the standard deviation.Once carried out, the data analysis was performed by using 15 factors for the construction of the corresponding calibration models, each one being evaluated by a cross validation leaving out one sample at time. For the selection of the optimum number of factors required for the construction of the calibration models by PLS-1, plots of the PRESS (prediction residual error sum of squares) versus the number of factors for each of the compounds were constructed and the first local minimum value criterion was applied.In addition, the statistical parameters SEC, RMSD and R together with the shape of the loading weight plots for the selected factors were analysed to avoid sub- or over-fitting of the model.14,19–22 Six were the factors selected for the construction of the calibration models of each of the compounds, in the four groups of analysed data. Subsequently, external validation of the proposed models was carried out through the 10 mixtures referred to in Table 3, with different concentration ratios of the components.The best results in both internal and external validations were obtained with the PLS-1/2 model using the linear variable angle spectra through route B, which are given in Table 4. As can be observed, application of PLS-1 allows the simultaneous determination of the six studied PAHs with satisfactory results. On the other hand, the least favourable results are obtained in the determination of BAP, as in the analysis by MLR.Artificial neural networks (ANN) The ANN is an algorithm designed somehow to ‘imitate’ the functioning of the human brain when extracting information of a data set to identify the structure and principles which rule the former. For this purpose, the ANN is formed by a series of processing units or ‘neurons’ interconnected and set in layers which iteratively perform certain mathematical operations. Thus, the input data pass through these units and are distributed, transformed and eventually collected to produce one or more outputs.In this way, the capacity of ANN for processing information does not correspond to the function of each individual processing unit but to the set of artificial neurons of which it is formed and the way in which they are interconnected. Although in order to establish the calibration model of a particular system the conventional techniques such as MLR and PLS take into account a mathematical function defined previously, ANNs only consider a data set which are transformed into one or more processing units with sufficient variable parameters (weights) to establish the relationship between the input and output data.15,23 In this case the algorithm of ANNs was applied to the simultaneous determination of the PAHs in a micellar medium, in order to observe its predictive capacity without initially taking into account the existing linear relationship between the fluorescence intensity and the concentration of the compounds in a system with severe spectral interferences.The training technique selected for the correction of the weights was the back-propagation of errors, because this algorithm is especially appropriate in the identification of the calibration model of a system when the composition of the samples whose spectra are given as input data is known. For the execution of the algorithm the Winnet software available under Windows was used, which allows the selection and normalization of the input data, together with the construction, training, validation and propagation of the neural networks by using the strategy of the back-propagation of errors for the correction of the weights.Since the program allows the use of a maximum of 50 signals per spectrum, six spectral regions were considered for the construction of the ANN with different architectures. The first Table 4 Statistical parameters calculated in the external and internal validation of the proposed calibration models Internal validation External validation Method PAH R SEC RMSD REP %R ± s SEP RMSD REP MLR BAP 0.996 0.36 0.34 12.85 8 ± 16 0.96 0.91 28.64 BEP 0.999 0.57 0.55 2.40 99 ± 3 0.78 0.74 2.48 BGP 0.999 0.26 0.24 3.02 100 ± 3 0.25 0.24 1.98 COR 0.999 0.12 0.11 2.27 100 ± 3 0.16 0.16 2.69 DBA 0.999 0.10 0.10 2.46 102 ± 4 0.21 0.20 3.44 IND 0.999 1.00 0.96 6.00 102 ± 13 2.45 2.32 9.25 PLS-1 BAP 0.993 0.49 0.48 12.35 8 ± 18 0.90 0.86 26.98 BEP 0.999 0.66 0.65 2.56 98 ± 2 1.03 0.98 3.28 BGP 0.999 0.34 0.33 3.21 100 ± 3 0.27 0.25 2.09 COR 0.999 0.18 0.17 2.85 97 ± 3 0.28 0.27 4.62 DBA 0.999 0.29 0.28 4.87 104 ± 5 0.30 0.28 4.94 IND 0.997 2.04 2.01 8.19 109 ± 12 1.85 1.76 6.99 ANN BAP 0.996 0.40 0.40 10.09 80 ± 46 1.50 1.43 45 BEP 0.997 2.59 2.55 10.08 87 ± 20 7.47 7.09 24 BGP 0.995 1.21 1.21 11.54 96 ± 21 2.17 2.06 17 COR 0.996 0.67 0.67 10.86 94 ± 18 0.93 0.88 15 DBA 0.989 0.95 0.95 16.01 107 ± 13 0.80 0.76 13 IND 0.992 3.42 3.42 13.71 11 ± 30 4.51 4.51 17 Analyst, 1998, 123, 2235–2241 2239two spectral regions comprised 50 signals uniformly distributed in the full variable angle spectra corresponding to routes A and B.Subsequently, 50 signals distributed within the intervals of positions 10–120, 155–200, 235–320 and 340–350 for route A and within the intervals of positions 6–75 and 117–195 for route B were selected. Finally, the two remaining spectral regions comprised 16 analytical signals for the fluorescence spectra of route A (position numbers 13, 24, 38, 86, 107, 113, 164, 177, 188, 197, 255, 263, 267, 297, 342 and 374) and six signals of route B (position numbers 13, 20, 27, 68, 128 and 196), which correspond to the signals of maximum spectral differences among the compounds.Subsequently, the input data were normalized. For this purpose the fluorescence intensities were divided by the maximum fluorescence intensities recorded for each data set.The concentrations of each compound were divided by the following values: 16 for BAP, 110 for BEP, 45 for BGP, 27 for COR, 29 for DBA and 106 for IND, which correspond to the highest concentration of each compound considered in the determination. In this way, the values of the input data corresponding to both the fluorescence intensities and the concentrations ranged between 0 and 1. Finally, the neural networks were constructed with architectures of (input layer 3 internal layer 3 output-layer) 50 3 0 3 6, 50 3 5 3 6, 50 3 15 3 6 and 50 3 30 3 6 by using the four set of spectra integrated by 50 analytical signals, architectures of 16 3 0 3 6 and 16 3 20 3 6 using the spectra of 16 analytical signals and architectures 6 3 0 3 6 and 6 3 12 3 6 with the spectra of six analytical signals.As can be observed, the network structure for each spectral group varied in relation to the number of processing units or neurons considered in the construction of the intermediate network layer.During the training of each network, the learning rate and the momentum parameter were both equal to 0.5. In order to carry out a sufficient number of iterations for the training of each of the proposed neural networks without reaching overtraining, the results were periodically recorded and compared. Once the networks had been trained, they were applied to the analysis of the synthetic mixtures whose compositions are described in Table 3.The statistical parameters of external and internal validation were calculated and according to the results it was proved that the most appropriate model for the simultaneous determination of the PAHs was that constructed with the spectra of 50 analytical signals uniformly distributed in the full spectral range, through route B using the architecture 50 3 30 3 6. The corresponding results are presented in Table 4. As can be observed, the predictive capacity of the calibration model proposed by neural networks was smaller than that of the models proposed by MLR or PLS-1.The use of this algorithm, which does not consider the existing linear relationship between fluorescence intensity and concentration, in addition to the high spectral overlap present in the system, yields larger errors in the determination. Analysis of real samples Finally, and in order to check the validity of the proposed models, the simultaneous determination of the six PAHs of interest was carried out under the experimental conditions described above by using as a matrix both sea-water and tap water samples.The sea-water samples were subjected to filtration prior to the determination. A Whatman filter-paper (No. 5 qualitative grade) was employed to retain the solid particles of size < 2.5 mm in suspension. Subsequently, eight samples were spiked with known amounts of the six PAHs, whose compositions are described in Table 5.The results obtained with the three selected mathematical models are presented in Table 6. As can be observed, there is good agreement between the expected and the estimated concentrations through the different multivariate calibration models, with satisfactory recovery results. Table 6 Concentrations (mg l21) and percentage recoveries obtained in the simultaneous determination of the six PAHs in micellar medium, performed on sea-water and tap water samples Method Source Sample BAP BEP BGP COR DBA IND MLR Sea-water 1 2.7 (67.5%) 12.5 (82.2%) 7.0 (81.4%) 6.0 (71.4%) 4.7 (81.0%) 25.8 (103.2%) 2 0.5 (—) 23.6 (95.2%) 11.2 (87.5%) 4.0 (83.3%) 7.1 (85.5%) 3.6 (—) 3 4.0 (66.7%) 18.1 (90.5%) 6.4 (83.1%) 5.1 (85.0%) 3.5 (87.5%) 43.0 (101.4%) 4 4.1 (75.9%) 14.7 (83.5%) 7.0 (72.9%) 5.3 (81.5%) 5.7 (85.1%) 33.2 (110.3%) Tap water 1 4.4 (73.3%) 20.2 (—) 13.0 (81.3%) 7.9 (88.8%) 7.6 (91.6%) 29.7 (103.1%) 2 3.1 (73.8%) 26.1 (88.2%) 7.5 (80.6%) 4.0 (83.3%) 5.7 (98.3%) 21.3 (107.6%) 3 4.6 (83.6%) 23.0 (95.8%) 9.5 (82.6%) 6.2 (80.5%) 20.7 (—) 59.9 (137.7%) 4 20.1 (—) 22.5 (86.5%) 6.9 (76.7%) 8.4 (76.4%) 4.5 (84.9%) 15.8 (112.1%) PLS-1 Sea-water 1 3.3 (82.5%) 17.7 (116.4%) 9.1 (105.8%) 7.7 (91.7%) 5.2 (89.7%) 20.8 (83.2%) 2 0.5 (—) 26.4 (106.5%) 12.8 (100.0%) 5.7 (118.8%) 7.3 (88.0%) 20.8 (—) 3 4.3 (71.7%) 22.6 (113.0%) 8.3 (107.8%) 6.8 (113.3%) 3.6 (90.0%) 38.7 (91.3%) 4 4.3 (79.6%) 22.5 (127.8%) 10.2 (106.3%) 8.4 (129.2%) 5.3 (79.1%) 25.3 (84.1%) Tap water 1 4.7 (78.3%) 4.4 (—) 14.9 (93.1%) 9.3 (104.5%) 7.5 (90.4%) 26.9 (93.4%) 2 3.5 (83.3%) 29.0 (98.0%) 8.7 (93.5%) 5.3 (110.4%) 5.9 (101.7%) 18.4 (92.9%) 3 4.6 (83.6%) 26.3 (109.6%) 11.6 (100.9%) 7.9 (102.6%) 20.3 (—) 40.0 (92.0%) 4 0.3 (—) 26.8 (103.1%) 8.7 (96.7%) 10.1 (91.8%) 4.6 (86.8%) 10.3 (73.0%) ANN Sea-water 1 2.2 (55.0%) 10.1 (66.4%) 5.9 (68.6%) 5.6 (66.7%) 3 (51.7%) 22 (88.0%) 2 0.6 (—) 14.2 (57.3%) 10.0 (78.1%) 3.3 (68.8%) 5.8 (69.9%) 6.1 (—) 3 3.8 (63.3%) 15.2 (76.0%) 5.0 (64.9%) 4.6 (76.7%) 2.7 (67.5%) 46.2 (109.0%) 4 4.6 (85.2%) 16.2 (92.0%) 4.3 (44.8%) 4.8 (73.8%) 2.3 (34.3%) 51.9 (172.4%) Tap water 1 3.9 (65.0%) 6.0 (—) 12.5 (78.1%) 8.0 (89.9%) 5.8 (69.9%) 25.9 (89.9%) 2 2.7 (64.3%) 21.2 (71.6%) 6.2 (66.7%) 3.4 (70.8%) 4.7 (81.0%) 20.0 (101.0%) 3 3.9 (70.9%) 12.2 (50.8%) 3.2 (27.8%) 2.8 (36.4%) 1.5 (—) 30.3 (69.7%) 4 0.6 (—) 14.4 (55.4%) 5.5 (61.1%) 8.8 (80.0%) 3.5 (66.0%) 12.7 (90.1%) Table 5 Composition of sea-water and tap water samples spiked with the six PAHs studied.All concentrations are expressed as mg l21 Source Sample BAP BEP BGP COR DBA IND Sea-water 1 4.0 15.2 8.6 8.4 5.8 25.0 2 0 24.8 12.8 4.8 8.3 0 3 6.0 20.0 7.7 6.0 4.0 42.4 4 5.4 17.6 9.6 6.5 6.7 30.1 Tap water 1 6.0 0 16.0 8.9 8.3 28.8 2 4.2 29.6 9.3 4.8 5.8 19.8 3 5.5 24.0 11.5 7.7 0 43.5 4 0 26.0 9.0 11.0 5.3 14.1 2240 Analyst, 1998, 123, 2235–2241Conclusions The presence of the non-ionic surfactant POLE in the proposed methanolic–aqueous medium gives rise to large variations in the luminescent properties of the six PAHs analysed.In all cases an increase in the fluorescence in different proportions is observed, which allows the development of more sensitive analytical methods for their determination. Regarding the use of spectroscopic techniques such as linear variable angle fluorescence, it is possible to record the spectra of multicomponent mixtures preserving the most relevant spectral information for each of the analytes without requiring the total excitation–emission spectra, their recording and analysis being time consuming.On the other hand, for systems with severe spectral interferences among the constituent compounds such as the present system, the application of multicomponent analysis algorithms allows their resolution without requiring prior separation processes. Of the algorithms considered in this work, those which consider a linear relationship between the optical property of interest and the species concentration in order to establish the calibration model allowed us to evaluate these interferences conveniently and resolve the mixture of six PAHs satisfactorily.The models obtained by the MLR and PLS-1 algorithms yield equally favourable results, after having adequately selected the calibration samples for each of the former. However, the calibration model proposed by the ANN algorithm had a less predictive capacity.It seems that the fact that this algorithm does not a priori consider an existing linear relationship between the variables analyzed together with the occurrence of severe spectral interferences made the establishment of the corresponding calibration model difficult. Acknowledgement The authors gratefully acknowledge the financial support of the CICyT (PETRI project, No. 95-0139-0P). References 1 J. W. Cook, I. Hieger, E. L. Kennaway and W. V. Mayneord, Proc. R. Soc. London, Ser. B, 1932, 3, 455. 2 E. J. Baum, Polycyclic Hydrocarbons and Cancer, Academic Press, New York, 1978. 3 D. J. Futoma, S. R. Smith, J. Tanaka and T. E. Smith, Polycyclic Aromatic Hydrocarbons in Water Systems, CRC Press, Boca Raton, FL, 1981. 4 W. J. Simonsick and R. A. Hites, Anal. Chem., 1986, 58, 2114. 5 J. C. Fetzer, W. R. Biggs and K. Jinno, Chromatographia, 1986, 21, 439. 6 P. Diercxens, PhD Thesis, D�epartement de Genie Rural et G�eom`etre, Ecole Polytechnique F�ed�erale de Lausanne, 1987. 7 K. D. Bartle, M. L. Lee and S. A. Wise, Chem. Soc. Rev., 1981, 10, 113. 8 J. B. F. Lloyd and I. W. Evett, Anal. Chem., 1977, 49, 1710. 9 T. Vo-Dihn, Anal. Chem., 1978, 50, 396. 10 M. T. Oms, R. Forteza, V. Cerda, F. García-Sánchez and A. L. Ramos, Comput. Chem., 1991, 15(1), 87. 11 G. Sala, S. Maspoch, H. Iturriaga, M. Blanco and V. Cerdá, J. Pharm. Biochem. Anal., 1988, 6, 765. 12 A. Cladera, J. Alpízar, J. M. Estela, V. Cerdá, M. Catasús, E. Lastres and L. García, Anal. Chim. Acta, 1997, 350, 163. 13 D. L. Massart, B. G. M. Vandeginste, S. N. Deming, Y. Michotte and L. Kaufman, Chemometrics: a textbook, Elsevier, Amsterdam, 1988. 14 H. Martens and N. Tormod, Multivariate Calibration, Wiley, Chichester, 1991. 15 J. Zupan and J. Gasteiger, Neural Networks for Chemists. An Introduction, VCH, Weinheim, 1993. 16 M. T. Oms, R. Forteza, V. Cerda, F. García and L. Ramos, Int. J. Environ. Anal. Chem., 1990, 42, 1. 17 R. V. Wandruszka, Crit. Rev. Anal. Chem., 1992, 23(3), 187. 18 A. J. Bermejo, PhD Thesis, Chemistry Department, Universidad de las Palmas de Gran Canaria, 1993. 19 D. M. Haaland and E. V. Thomas, Anal. Chem., 1988, 60, 1193. 20 D. M. Haaland and E. V. Thomas, Anal. Chem., 1988, 60, 1202. 21 E. V. Thomas and D. M. Haaland, Anal. Chem., 1990, 62, 1091. 22 P. L. López-de-Alba, L. López-Martínez, K. Wróbel-Kaczmarzyc, K. Wróbel-Zasada and J. Amador-Hernández, Anal. Chim. Acta, 1996, 330, 19. 23 D. Mukesh, J. Chem. Educ., 1996, 73 (5), 431. Paper 8/04183K Analyst, 1998, 123, 2235–2
ISSN:0003-2654
DOI:10.1039/a804183k
出版商:RSC
年代:1998
数据来源: RSC
|
6. |
Comparison of HPLC, capillary electrophoretic and direct spectrofluorimetric methods for the determination of temoporfin–poly(ethylene glycol) conjugates in plasma† |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 2243-2245
Hong Cai,
Preview
|
PDF (53KB)
|
|
摘要:
Comparison of HPLC, capillary electrophoretic and direct spectrofluorimetric methods for the determination of temoporfin–poly(ethylene glycol) conjugates in plasma† Hong Cai and C. K. Lim* MRC Toxicology Unit, Hodgkin Building, University of Leicester, P.O. Box 138, Lancaster Road, Leicester, UK LE1 9HN Received 4th June 1998, Accepted 8th July 1998 High-performance liquid chromatographic (HPLC), capillary electrophoretic (CE) and direct spectrofluorimetric methods for the determination of temoporfin–poly(ethylene glycol) 2000 conjugate (m-THPC–PEG 2000) in plasma are described and compared.m-THPC–PEG 2000 in plasma was quantitatively extracted (recovery 101–107%) with CH3OH–DMSO (4 + 1 v/v). The supernatant after centrifugation was used for HPLC, CE or direct spectrofluorimetric determination. The major drawback of the HPLC method was that it gave a broad and split peak even under gradient elution conditions, resulting in difficulty in detection and quantification.This is because m-THPC–PEG 2000 consists of a group of compounds with an average molecular mass of approximately 8680 Da owing to the wide molecular mass distribution of PEG 2000 used in the synthesis of the drug. m-THPC–PEG 2000 gave a single and relatively sharp peak when separated by CE with sodium tetraborate buffer (pH 9.45) in the presence of sodium dodecyl sulfate as the running buffer. However, this method lacks the necessary sensitivity for detecting the drug in plasma extract because of the limited sample volume that can be injected.Direct spectrofluorimetry is the method of choice because of its simplicity, specificity and sensitivity. Using an excitation wavelength of 423 nm and the specific emission maximum of 657 nm, the fluorescence intensity could be sensitively measured. The calibration curve constructed by plotting fluorescence intensity against concentration was linear within the range 1.32–1056 ng ml21. The detection limit (S/N = 3) was 1.32 ng ml21 and the limit of quantification (S/N = 10) was 2.24 ng ml21. The precision and reproducibility were assessed by repeated analysis (n = 24) of spiked plasma samples at 350.8 and 699.3 ng ml21.The RSD was 4.5% and 1.6%, respectively. Introduction Temoporfin, 5,10,15,20-tetra(m-hydroxyphenyl)chlorin (m- THPC), one of the most promising photodynamic therapeutic (PDT) agents developed in recent years,1 has been conjugated with poly(ethylene)glycol (PEG) via chlorotriazine–PEG to give temoporfin–poly(ethylene glycol) conjugates (m-THPC– PEG) (Fig. 1) in an attempt to improve its water solubility and possibly also to enhance its therapeutic efficacy. Preliminary studies with a mouse model system have shown that m-THPC– PEG conjugates were less potent than m-THPC on a molar basis but were capable of producing equivalent tumour necrosis accompanied by relatively low levels of muscle damage.2 In addition, tumour necrosis caused by m-THPC–PEG conjugates was seen over a wider range of drug–light intervals.PDT reactions with these m-THPC–PEG conjugates may therefore be less influenced by the biological variation and treatment conditions with respect to the therapeutic outcome.3 As interest in and pre-clinical studies on m-THPC–PEG 2000 increase, methods for the determination of these compounds need to be developed. We describe here three different assay procedures for the quantitative measurement of m-THPC–PEG conjugates in plasma, using the PEG 2000 derivative as an example.The methods, based on high-performance liquid chromatography (HPLC), capillary electrophoresis (CE) and direct spectrofluorimetry were compared and the spectrofluorimetric procedure was optimised and validated. Experimental Materials and reagents Temoprofin (m-THPC) and m-THPC–PEG 2000 were gifts from Scotia QuantaNova (Guildford, Surrey, UK). Acetonitrile and methanol were of HPLC grade from Rathburn Chemicals (Walkerburn, Peebleshire, UK). Dimethyl sulfoxide (DMSO) and tetrahydrofuran (THF) were of AnalaR grade from Merck (Poole, Dorset, UK).Trifluoroacetic acid (TFA) was obtained † Presented at the VIIIth International Symposium on Luminescence Spectrometry in Biomedical and Environmental Analysis, Las Palmas de G.C., Spain, May 26–29, 1998. Fig. 1 Structure of temoporfin–poly(ethylene glycol) conjugate (m- THPC–PEG 2000). Analyst, 1998, 123, 2243–2245 2243from Pierce and Warriner (Chester, UK).A 0.1% w/v solution was made by dissolving 1 ml (one ampoule) in 1 l of distilled water. Extraction of m-THPC–PEG 2000 from plasma A 1 ml volume of plasma was vortex mixed with 2 ml of methanol–DMSO (4 + 1 v/v). The mixture was centrifuged at 4000g for 10 min. The supernatant was transferred into a cuvette for spectrofluorimetric determination or injected into the HPLC or CE system for analysis. HPLC separation of m-THPC–PEG 2000 in plasma extract The HPLC system consisted of a Varian (Walton-on-Thames, Surrey, UK) Model 9012 chromatograph and a Varian Model 9050 UV/VIS detector set at 423 nm for peak detection. The separation was carried out on a 250 3 4.6 mm id Hypersil-BDS column with 0.1% w/v TFA (A), acetonitrile (B) and THF (C) as the gradient mobile phase mixture.The elution programme was as follows: time 0 to 10 min, from 70% A, 15% B, 15% C to 60% A, 15% B, 25 % C; time 10 to 30 min, from 60% A, 15% B, 25% C to 50% A, 15% B, 35% C; and time 30 to 40 min, from 50% A, 15% B, 35% C to 40% A, 15% B, 45% C.The flow rate was 1 ml min21. Capillary electrophoresis of m-THPC–PEG 2000 in plasma extract A Model 270A analytical electrophoresis system (Applied Biosystems, Cheshire, UK) was used with an on-column variable wavelength absorbance detector set at 423 nm for peak detection. The capillary was 72 cm in length (50 cm to the detector), and had an inner diameter of 50 mm and an outer diameter of 370 mm.The running voltage and temperature were 30 kV and 45 °C, respectively. The running buffer was 20 mm sodium borate buffer containing 50 mm sodium dodecyl sulfate (SDS) (pH 9.45). Spectrofluorimetric determination of m-THPC–PEG 2000 in plasma A Perkin-Elmer (Norwalk, CT, USA) (LS 50B) luminescence spectrofluorimeter was used for the measurement of the fluorescence intensity of m-THPC–PEG 2000. The supernatant of the plasma extract was transferred into a 3 ml quartz cuvette (1 cm pathlength) and the fluorescence intensity was measured using excitation and emission wavelengths of 423 and 657 nm, respectively.Methanol–DMSO–H2O (8 + 2 + 5 v/v/v) was used as the blank. A stock standard solution of m-THPC–PEG 2000 (batch BD130) was made by dissolving the drug (10.56 mg) in distilled water (100 ml). Aliquots of the stock standard solution were diluted with methanol–DMSO–H2O (8 + 2 + 5 v/v/v) to give working standard solutions with a range of concentrations between 1.32 and 1056.00 ng ml21.These were used to construct a calibration curve for the determination of m-THPC– PEG 2000. Results and discussion HPLC separation of m-THPC–PEG 2000 in plasma The separation of a standard solution of m-THPC–PEG 2000 by gradient elution reversed-phase HPLC is shown in Fig. 2. A broad and split peak was obtained. Attempts to improve the peak shape by modification of the mobile phase for elution were unsuccessful. The broad peak was due to the wide molecular mass distribution of the PEG 2000 used in the synthesis of the conjugate, resulting in a product (m-THPC–PEG 2000) which consisted of a group of compounds with an average molecular mass of approximately 8680 Da.HPLC determination of m- THPC–PEG 2000 in plasma is therefore difficult because the broad and split peak leads to difficulty in detection and quantification. Capillary electrophoresis of m-THPC–PEG 2000 Micellar electrokinetic chromatography (MEKC) with sodium borate buffer in the presence of SDS as the running buffer was found to be the best CE mode for the separation of m-THPC and its PEG conjugates. The optimised system, with 20 mm sodium borate buffer (pH 9.45) containing 50 mm SDS as running buffer and a running voltage and temperature of 30 kV and 45 °C, respectively, effectively separated m-THPC and m- THPC–PEG 2000 in less than 10 min (Fig. 3). m-THPC–PEG Fig. 2 HPLC separation of a standard solution of m-THPC–PEG 2000.The HPLC conditions are described in the Experimental section. Fig. 3 CE separation of a mixture of m-THPC and m-THPC–PEG 2000. The CE conditions are described in the Experimental section. 2244 Analyst, 1998, 123, 2243–22452000 was eluted as a single, relatively sharp peak without the broadening and splitting effects observed with HPLC separation. However, the CE system was insufficiently sensitive for the detection of m-THPC–PEG 2000 in plasma extracts because of the limited sample volume that can be injected. CE is therefore not suitable for the analysis of biological extracts for m-THPC and m-THPC–PEG conjugates.The CE method, however, may be ideal for monitoring the purity of the m-THPC–PEG conjugates during synthesis. Spectrofluorimetric determination of m-THPC–PEG 2000 in plasma Using the relatively high and specific excitation and emission wavelengths of 423 and 657 nm, respectively, m-THPC–PEG 2000 could be sensitively and selectively detected without interference from other compounds in the plasma extracts.The emission spectrum of m-THPC–PEG 2000 in a plasma extract (699.3 ng ml21) is shown in Fig. 4. The lack of interference was demonstrated when no emission peak at 657 nm was observed in blank plasma extracts. The recovery of the extraction method was assessed using control human plasma spiked with m-THPC–PEG 2000. The recovery was determined by comparison of the fluorescence intensity obtained from the plasma extract with that of an identical concentration of m-THPC–PEG 2000 in methanol– DMSO–H2O (8 + 2 + 5 v/v/v).The recoveries of plasma spiked with 350.8 and 699.3 ng ml21 of m-THPC–PEG 2000 were 101.0 ± 3.9 and 107.0 ± 2.6%, respectively (mean ± s, n = 24). The calibration curve for the determination of m-THPC–PEG 2000 was constructed by plotting fluorescence intensity against concentration using standard calibration solutions. The curve was linear in the range 1.32–1056 ng ml21.Linear regression analysis gave the equation of the line as y = 0.233x 2 0.9918 with a correlation coefficient (r2) of 0.9998. The precision of the method was determined using spiked control human plasma at two concentrations, 350.8 and 699.3 ng ml21, with 24 repeated analyses. The fluorescence intensities were measured following sample extraction and used to calculate the corresponding concentrations from a calibration curve. The mean concentration and the relative standard deviation (RSD) were determined for each concentration. For the 350.8 ng ml21 sample, these were 374.8 ng ml21 and 4.5%, respectively, and those for the 699.3 ng ml21 sample were 706.6 ng ml21 and 1.6%, respectively. The limit of detection was determined as 1.32 ng ml21 at a signal-to-noise ratio of 3.The limit of quantification, at a signalto- noise ratio of 10, was 2.24 ng ml21, which is more than adequate for the determination of plasma levels in metabolism and pharmacokinetic studies.The limits of detection and quantification are equivalent to 0.10 and 0.18 ng ml21 of unconjugated m-THPC, respectively. If required, the plasma sample size could be reduced to 100 ml and a micro-cuvette used for fluorescence measurement. The smaller volume of extract may also be diluted for measurement with a conventional cuvette. Although no fluorescent compounds in the plasma extract interfered with the determination of m-THPC–PEG 2000, it is important to ensure that the plasma used is free from contamination with haemolysed red cells.The haem liberated during the extraction procedure will quench and reduce the fluorescence intensity of m-THPC–PEG 2000 (Fig. 5), despite the fact that haem itself is a non-fluorescent compound. If haem has been found to be present in plasma, its concentration should be determined, e.g., by HPLC,4 and a correction factor applied to the calculation of results based on the graph shown in Fig. 5. Conclusions Three methods for the determination of m-THPC–PEG 2000 in plasma, HPLC, CE and spectrofluorimetric, were evaluated and compared. It was concluded that the spectrofluorimetric assay is sensitive, specific, accurate and reproducible and is the method of choice. Acknowledgement We thank Scotia QuantaNova for a visiting research fellowship to H. Cai and generous gifts of materials. References 1 R. Bonnett, Chem. Soc. Rev., 1995, 24, 19. 2 M. F. Grahn, A. McGuinness, M. L. de Jode, A. Giger, A. S. Dhiman, C.-M. Cheung, S. Pavitt, R. Benzie and N. S. Williams, Proc. SPIE, 1997, 3191, 180. 3 M. F. Grahn, A. Giger, A. McGuinness, M. L. de Jode, J. C. M. Stewart, H. B. Ris, H. J. Altermatt and N. S. Williams, Laser Med. Sci., in the press. 4 C. K. Lim, F. Li and T. J. Peters, J. Chromatogr., 1988, 429, 123. Paper 8/04226H Fig. 4 Fluorescence emission spectrum of m-THPC–PEG 2000 in a plasma extract (699.3 ng ml21). Excitation was at 423 nm. Fig. 5 Effect of haem on the fluorescence intensity of m-THPC–PEG 2000. Analyst, 1998, 123, 2243–2245 2245
ISSN:0003-2654
DOI:10.1039/a804226h
出版商:RSC
年代:1998
数据来源: RSC
|
7. |
Polarity studies on ormosils using a solvatochromic fluorescent probe† |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 2247-2250
Aleksandra Lobnik,
Preview
|
PDF (74KB)
|
|
摘要:
Polarity studies on ormosils using a solvatochromic fluorescent probe† Aleksandra Lobnik*a and Otto S. Wolfbeisb a Faculty of Mechanical Engineering, University of Maribor, Smetanova 17, 2000 Maribor, Slovenia. E-mail: aleksandra.lobnik@uni-mb.si b University of Regensburg, Institute of Analytical Chemistry, Chemo- and Biosensors, D-93040 Regensburg, Germany Received 16th June 1998, Accepted 1st September 1998 Ormosils (organically modified siloxanes) are a relatively new family of materials, prepared by the sol–gel method, with properties that are intermediate between those of glasses and polymers.To probe the micropolarity of various solvents and ormosils, a ketocyanine dye (KC) with unique solvatochromic properties in fluorescence was used. KC is an excellent probe and is sensitive to the polarity of its microenvironment. Fluorescence studies revealed remarkable changes in the fluorescence band positions or intensities as a function of the polarity of the ormosil, which depend on the different ormosil precursors used.Storage stability was also investigated. 1. Introduction The sol–gel process is an exciting new technology that permits the production of gels, glasses and ceramic materials at room temperature.1 Most sol–gel techniques use water and low molecular mass alkoxysilanes such as tetramethoxysilane (TMOS) or tetraethoxysilane (TEOS), or an equivalent organometallic alkoxide, as sol–gel precursors and the sol–gel is formed through hydrolysis and condensation reactions.2 Ormosils (organically modified siloxanes) represent hybrid systems in which several precursor types such as organotrialkoxysilane or diorganodialkoxysilane precursors [RASi(OR)3 or RA2Si- (OR)2, respectively], in which RA represents a non-hydrolyzable organic substituent, are combined.3 They have attracted much attention in recent years because they provide an easy introduction of reactive functional groups which can subsequently be used as covalent anchors for indicators,4 or they are used to reduce the functionality (potential number of sites able to form Si–O–Si bonds) of the alkoxide precursor, impart organic character or derivatize the siloxane network.Conventional sol–gels are hydrophilic and well suited for sensing ions. A more hydrophobic material is obtained on preparing ormosils. Ormosils have been attracting great interest recently in the area of chemical sensors, mostly with regard to the formation of ormosil-based optical oxygen5 or ammonia sensors.6 A critical parameter in gas sensing is the permeability and permeation selectivity of the polymer used.Silicones,7 PTFE8 and plasticized poly(vinyl chloride)9 are considered to be a good compromise, owing to their high permeability for gases and their impermeability to ions. Other materials, such as cellulose,10 poly(vinyl chloride)11 and poly(vinyl acetate)12 have also been used for gas sensing, but they are less gas permeable.In contrast to the rapid development of many applications, the exact nature of the entrapment and the properties of the encapsulating cage are still not fully understood. The detailed microstructure of sol–gel glasses depends on a large number of important processing parameters such as water-to-sol–gel precursor ratio, precursor type, nature of the catalyst, aging time, aging temperature, drying time and drying temperature.13 Sol–gels also undergo temporal changes in their physical and chemical structures. These effects have been demonstrated in numerous, mainly spectroscopic, studies. Luminescent molecules have been widely used as probes of the sol–gel process. 8-Hydroxy-1,3,6-pyrenetrisulfonate is a sensitive fluorimetric probe for following the kinetics of water consumption during the early stages of the TMOS sol–gel polymerization process.14 Fluorimetric probes can also be used to obtain unique insights regarding local chemistry and structure during the sol–gel– xerogel transition.15 Static and dynamic fluorescence spectroscopy of Rhodamine 6G (R6G) in a sol–gel matrix has been used to study the effect of aging time and hydrolysis pH on the local microviscosity.16 More interesting results are expected for molecules which are themselves very flexible and sense the microenvironment in which they are located.Recently, Rottman et al.17 reported the entrapment of the absorption polaritysensitive dye ET(30), which showed high sensitivity to the environmental polarity in different ormosils which were prepared from various proportions of methyltrimethoxysilane and tetramethoxysilane. Similarly to our observations presented here, they also concluded that the polarity of the ormosil cages decreased with a higher fraction of the Me-TriMOS precursor.Nile Red has been studied in organic solvents, binary solvent mixtures and polymers such as poly(methyl methacrylate) (PMMA) and poly(vinyl alcohol) (PVA).It has been found that the micropolarity in PVA is greater than that in PMMA.18 The idea of our work was to incorporate a fluorescent polarity-sensitive probe such as ketocyanine dye (KC) to obtain more information about the polarity of the ormosil cages. Fluorescence maxima and intensities were used to study KC in different organic solvents and ormosil monoliths, prepared from various proportions of tetramethoxysilane, methyltrimethoxysilane, phenyltrimethoxysilane, dimethyldimethoxysilane and diphenyldimethoxysilane.Ormosil properties such as storage stability were monitored. † Presented at the VIIIth International Symposium on Luminescence Spectrometry in Biomedical and Environmental Analysis, Las Palmas de G.C., Spain, May 26–29, 1998. Analyst, 1998, 123, 2247–2250 22472. Experimental 2.1. Reagents Tetramethoxysilane (TMOS; product No. 87682), methyltrimethoxysilane (Me-TriMOS; product No. 69471), phenyltrimethoxysilane (Ph-TriMOS; product No. 79240) and diphenyldimethoxysilane (DiPh-DiMOS; product No. 42940) were from Fluka (Buchs, Switzerland) and dimethyldimethoxysilane (DiMe-DiMOS; product No. 10,490-6) from Aldrich (Milwaukee, WI, USA). Hydrochloric acid and ethanol were of analytical-reagent grade [Merck (Darmstadt, Germany) and Fluka]. Ketocyanine (KC) (Fig. 1) was synthesized in our laboratory.19 2.2. Instrumentation All fluorescence measurements were performed with a Shimadzu (Kyoto, Japan) RF-5001 PC spectrofluorimeter.Monoliths were tested in plastic cuvettes. Light from a 150 W xenon source was directed on to the sensing position at an angle of typically 35°, and fluorescence was measured after passing the emission monochromator at a typical slitwidth of 1.5 nm. Various solvents were tested in glass cuvettes. Data were processed using Excel 5.0. The experimental set-up is shown in Fig. 2. 2.3. Preparation of ormosils In a typical experiment with a molar ratio of TMOS to ormosil of 2 : 8, a mixture of 1 ml of dye (1 mm), 1.5 ml of ethanol and 0.4 ml of TMOS, 1.6 ml of Me-TriMOS (or 2.0 ml of Ph- TriMOS, 1.3 g of DiMe-DiMOS or 2.5 ml of DiPh-DiMOS) was sonicated for 5 min, then, 1 ml of acidic water (pH 3) was added, followed by sonication for an additional 5 min.The water of pH 3 was prepared by adding hydrochloric acid to distilled water. The molar ratio of acidic water to sol–gel precursor (R) was fixed at R = 4 for all samples.20 In all cases, the ormosil solutions were dried at 70 °C for 24 h. 3. Results and discussion 3.1. Excitation/emission maxima of KC in various solvents To establish the relationship between solvent polarity and spectral properties, we compared the calculated energy of the excitation/emission maxima of KC with the standard ETN scale of solvents,21 which is based on the well known polarity indicator ET(30). In this scale, tetramethylsilane is defined as the most apolar solvent (ETN = 0.000) and water as the most polar solvent (ETN = 1.000).All other solvents have ETN values located between these values. In the presence of KC the excitation/emission maxima are shifted to longer wavelengths by about 100 nm on going from toluene to water (Fig. 3). 3.2. Fluorescence study of KC in various ormosils KC was doped into different ormosils, prepared by varying the ratio of tetramethoxysilane (TMOS) and organically modified sol–gel precursor (Me-TriMOS, Ph-TriMOS, DiMe-DiMOS and DiPh-DiMOS).Spectroscopic studies revealed remarkable changes in the fluorescence band positions and intensities as a function of the micropolarity of the sol–gel cages, caused by the organically modified sol–gel precursor such as Me-TriMOS. From Fig. 4, it is concluded that the Me-TriMOS-based ormosils show the most polar microenvironment compared with the polarities observed for Ph-TriMOS-, DiMe-DiMOS- and DiPh-DiMOSbased ormosils. The polarities of Ph-TriMOS- and DiMe- DiMOS-based ormosils are similar. DiPh-DiMOS-based ormosils show at lower fractions (20 and 40% of ormosil precursor) polarities which are similar to those observed for Me-TriMOS, whereas at higher fractions ( < 50%) the polarities are similar to those observed for Ph-TriMOS- and DiMe-DiMOS-based ormosils.Me-TriMOS-based ormosil cages show the most polar microenvironment at all ratios. The spectra of various proportions of TMOS and Me- TriMOS and Ph-TriMOS ormosil precursors are shown in Fig. 5 and the corresponding fluorescence maxima and fluorescence intensities of all studied ormosil precursors are given in Table 1. Changes in fluorescence maxima were significant only when DiPh-DiMOS precursor was used, whereas in all other series, when various ratios of Me-TriMOS, Ph-TriMOS and DiMe- DiMOS were used, small or no changes in fluorescence maxima were observed. Fig. 5 shows the differences in the profiles of the fluorescence spectra.In ormosils prepared with lower fractions of ormosil precursor ( > 50%) the excitation profile in the range 400–700 nm is broad and diffuse. The steady-state room temperature emission spectrum is also broad. These results Fig. 1 Structure of ketocyanine dye (KC). Fig. 2 Experimental set-up for fluorescence measurements: L, light source; MC, monochromator; MI, mirror; SM, sensor membrane; FC, flowthrough cell; S, sample solution; D, detector; and P, peristaltic pump.Fig. 3 Correlation of the standard ETN values of solvent polarity versus the calculated energy (in kcal mol21) of the excitation (2) and emission (-) maxima of KC in various solvents. 2248 Analyst, 1998, 123, 2247–2250demonstrate that the ormosils with more TMOS are composed of several distinct regions and the average local micropolarity sensed by KC is not homogeneous in all regions. Discrete microdomains are postulated, while a more homogeneous microenvironment is observed when higher fractions of ormosil precursors ( < 50%) are used.In the case of all ormosil monoliths we can observe a slight to significant increase in the intensities when the ratio of the ormosil monomers increases (Fig. 6). A 95% (98%) increase in fluorescence intensity was observed on going from DiMe2 (Me2) to DiMe10 (Me10), while DiPh-DiMOS and Ph-TriMOS showed 63 and 36% increases in fluorescence intensity, respectively. It is known that KC is almost non-fluorescent in water and polar solvents.However, in a non-polar environment, its fluorescence is enhanced, and it undergoes large blue shifts in both absorption and emission.19 It seems that methyl groups cause a higher fluorescence intensity than phenyl groups at the same dye concentrations and slitwidths. We have no explanation for this observation. We tested the prepared ormosils also as pH and ammonia optical sensors.4, 6 The addition of organically modified sol–gel precursor was found to increase the lipophilicity of the gel, leading to a much smaller or no effect of pH and a higher permeability to ammonia, and the TMOS-based sol–gel is an adequate hydrophilic material for sensing pH. 3.3. Storage stability The effect on the fluorescence band positions and fluorescence intensity of storage in air for 3 months was studied. Me2, similarly to a Ph2-based ormosil doped with KC, underwent a significant change in the excitation band positions (+20 nm), and showed a 30% decrease in the fluorescence intensity.Obviously, hydrolysis in these ormosils is incomplete and proceeds further, during the drying process, resulting in changes in the excitation maxima, while changes in the fluorescence intensities are caused by condensation reactions. These results are consistent with the assumption of continuing hydrolysis in Fig. 4 Cage polarity as a function of the molar % of Me-TriMOS (5), Ph- TriMOS (-), DiMe-DiMOS (:) and DiPh-DiMOS (3).Table 1 Fluorescence bands and fluorescence intensities (IF) of KC at different ratios of TMOS and ormosil precursors of type RASi(OMe)3 and RA2Si(OMe)2 Ormosil monoliths doped with KC Me-TriMOS (Me) Ph-TriMOS (Ph) DiMe-DiMOS (DiMe) DiPh-DiMOS (DiPh) Ormosil :TMOS (%) lex/lem (nm) IF lex/lem (nm) IF lex/lem (nm) IF lex/lem (nm) IF 20 : 80; O 2a 530/594 48 477/535 3 473/531 21 494/600 1.5 40 : 60; O4b 530/594 50 475/533 9 471/529 26 481/596 2.6 60 : 40; O6 531/594 143 471/529 16 473/527 36 473/529 3.2 80 : 20; O8 531/592 245 470/529 18 471, 483/540 154 465/535 4.1 100 : 0; O10 534/598 463 470/529 38 471/532 480 432/532 60.0 a O2, ormosil precursor such as Me-TriMOS (Me), Ph-TriMOS (Ph), DiPh-DiMOS (DiPh) or DiMe-DiMOS (DiMe) in ratio 20 : 80% TMOS.b O4, ormosil precursor such as Me-TriMOS (Me), Ph-TriMOS (Ph), DiPh-DiMOS (DiPh) or DiMe-DiMOS (DiMe) in ratio 40 : 60% TMOS. Fig. 5 Fluorescence spectra of KC incorporated into ormosils prepared from various ratios of TMOS and Me-TriMOS or Ph-TriMOS.Analyst, 1998, 123, 2247–2250 2249the presence of atmospheric moisture.5 Ormosils containing a > 50% fraction of organically modified precursors show no changes in the maxima and only about a 40% decrease in excitation intensity (Me6–Me10 layers). This is attributed to almost complete hydrolysis after preparation of the ormosil solution (smaller number of methoxy groups which are hydrolyzed), while the condensation reaction requires gelation times of several months. 4. Conclusions We have found that ormosil cages prepared from lower fractions of ormosil precursors ( > 50%), irrespective to the type of ormosil precursor used, show an undefined microenvironment. In such ormosil cages some microdomains with different polarity properties exist. The structure of such ormosil cages changes over a period of 3 months. More ‘stable’ ormosil cages can be prepared with higher ratios of ormosil precursors ( < 50%).Their microenvironment is more defined and they show a more homogeneous average polarity. 5. References 1 O. Lev, M. Tsionsky, L. Rabinovich, V. Glezer, S. Sampath, I. Pankratov and J. Gun, Anal. Chem., 1995, 67, 22A. 2 O. S. Wolfbeis, R. Reisfeld and I. Oehme, Struct. Bonding (Berlin), 1996, 85, 52. 3 C. J. Brinker and G. W. Scherer, Sol–Gel Science, Academic Press, San Diego, 1989. 4 A. Lobnik, I. Oehme, I. Murkovic and O. S. Wolfbeis, Anal. Chim. Acta, 1998, 367, 159. 5 A. K. McEvoy, C. M. McDonagh and B. D. MacCraith, Analyst, 1996, 121, 785. 6 A. Lobnik and O. S. Wolfbeis, Sens. Actuators, in the press. 7 M. Trinkel, W. Trettnak, F. Reininger, R. Benes, P. O’Leary and O. S. Wolfbeis, Anal. Chim. Acta, 1996, 320, 235. 8 J. Reichert, W. Sellien and H. J. Ache, Sens. Actuators, 1991, 25, 481. 9 A. Mills, L. Wild and Q. Chang, Mikrochim. Acta, 1995, 121, 225. 10 N. Nakano, K. Sugata and K. Nagashima, Anal. Chim. Acta, 1995, 302, 201. 11 S. Ozawa, P. C. Hauser, K. Seiler, S. S. Tan, W. E. Morf and W. Simon , Anal. Chem., 1991, 63, 640. 12 C. Preininger, G. J. Mohr, I. Klimant and O. S. Wolfbeis, Anal. Chim. Acta, 1996, 334, 113. 13 C. McDonagh, F. Sheridan, T. Butler and B. D. MacCraith, J. Non- Cryst. Solids, 1996, 194, 72. 14 V. R. Kaufman, D. Avnir, D. P. Rojanski and D. Huppert, J. Non- Cryst. Solids, 1988, 99, 379. 15 B. Dunn and J. I. Zink, J. Mat. Chem., 1991, 1, 903. 16 U. Narang, R. Wang, P. N. Prasad and F. V. Bright, J. Phys. Chem., 1994, 98, 17. 17 C. Rottman, G. S. Grader, Y. De Hazan and D. Avni, Langmuir, 1996, 2, 15. 18 A. K. Dutta, K. Kamada and K. Ohta, J. Photochem. Photobiol., A, 1996, 93, 57. 19 M. A. Kessler and O. S. Wolfbeis, Spectrochim. Acta, Part A, 1991, 47, 187. 20 C. McDonagh, F. Sheridan, T. Butler and B. D. MacCraith, J. Sol– Gel Sci. Technol., 1996, 5, 17. 21 Ch. Reichardt, Solvents and Solvent Effects in Organic Chemistry, VCH, Weinheim, 2nd edn., 1988. Paper 8/04583F Fig. 6 Fluorescence intensity as a function of the molar % of Me-TriMOS (5), Ph-TriMOS (-), DiMe-DiMOS (:) and DiPh-DiMOS (3). 2250 Analyst, 1998, 123, 2247–2250
ISSN:0003-2654
DOI:10.1039/a804583f
出版商:RSC
年代:1998
数据来源: RSC
|
8. |
pH-induced destabilization of lipid bilayers by a peptide from the VP3 protein of the capsid of hepatitis A virus† |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 2251-2256
Abelardo Chávez,
Preview
|
PDF (102KB)
|
|
摘要:
pH-induced destabilization of lipid bilayers by a peptide from the VP3 protein of the capsid of hepatitis A virus† Abelardo Ch�avez, Mar�ýa A. Busquets, Monserrat Pujol, M. Asunci�on Alsina* and Yolanda Cajal* Department of Physical Chemistry, School of Pharmacy, University of Barcelona, Avn. Joan XXIII s/n, 08028 Barcelona, Spain Received 16th June 1998, Accepted 7th September 1998 The membrane destabilizing and fusogenic properties of the synthetic peptide VP3(110–121), corresponding to an immunogenic sequence of the hepatitis A virus (HAV) VP3 capsid protein, were studied.By tryptophan fluorescence and acryalmide quenching it was demonstrated that the peptide binds liposomes of POPC–SM–DPPE (47 + 39 + 14) and POPC–SM–DPPE–DOTAP (40 + 33 + 12 + 15) and penetrates the membrane, at both neutral and acidic pH (POPC = 1-palmitoyl-2-oleoylglycero-sn-3-phosphocholine; SM = sphingomyelin; DPPE = 1,2-dipalmitoylphosphatidylethanolamine; DOTAP = 1,2-dioleoyl-3-trimethylammoniumpropane).VP3(110–121) did not have membrane-destabilizing properties at neutral pH. Acid-induced destabilization of the vesicles was demonstrated by fluorescence techniques and dynamic light scattering. VP3(110–121) induced aggregation of POPC–SM–DPPE–DOTAP (40 + 33 + 12 + 15) vesicles, lipid mixing and leakage of vesicle contents, all consistent with fusion of vesicles. In POPC–SM–DPPE (47 + 39 + 14) vesicles, at acidic pH, VP3(110–121) induced membrane destabilization with leakage of contents but without aggregation of vesicles or lipid mixing.The peptide only showed fusogenic properties when bound to the vesicles at neutral pH before acidification to pH below 6.0, and no effect was seen if the peptide was added to vesicles already set at acidic pH. These results may have physiological significance in the mechanism of infection of host hepatic cells by HAV. Introduction Hepatitis A virus is a non-enveloped virus from the Picornaviridae family.The current understanding of the process by which non-enveloped viruses gain entry into the host cell is limited. Viral uncoating can proceed by two possible mechanisms, involving interaction with cell membranes. One possibility is receptor-mediated endocytosis and subsequent acidification of the resulting clathrin-coated vesicles by fusion with endosomes, leading to unfolding or release of hydrophobic regions of the capsid proteins of the virus, with formation of a pore in the vesicle through which the viral RNA is transferred to the cytoplasm. Another mechanism involves interaction of the virus with the plasma membrane of the host cell, causing a rearrangement in the capsid proteins, freeing the RNA and allowing its transit through the membrane.All these effects on the membrane are generally attributed to a critical viral protein with a short hydrophobic peptide called the fusion peptide, which can cause viral infection and also fusion of any two adjacent membranes.1,2 Fusion peptides have been identified in several enveloped viruses.In vesicular stomatitis virus (VSV), the envelope glycoprotein G mediates VSV fusion with cell membranes at acidic pH.3 In influenza virus, several synthetic peptides based on N-terminal sequences of the HA2 subunit of influenza hemagglutinin have demonstrated membrane-destabilizing and fusion properties at the pH of the endosome.4–6 A better understanding of membrane fusion will provide a valuable insight into the mechanism of viral infection and virusinduced cell lysis, and also a better knowledge about more general problems of membrane biosynthesis, structure and function.In this paper, we report the membrane-destabilizing and fusogenic properties of the synthetic peptide VP3(110–121) from the capsid protein VP3 of hepatitis A virus, 110Phe–Trp– Arg–Gly–Asp–Leu–Val–Phe–Asp–Phe–Gln–Val121. Lipid vesicles were prepared from 1-palmitoyl-2-oleoylglycero- sn-3-phosphocholine–sphingomyelin–1,2-dipalmitoylphosphatidylethanolamine (POPC–SM–DPPE) (47 + 39 + 14), a phospholipid composition similar to that of the membranes of the hepatic host cells. Nevertheless, and owing to the anionic net charge of the peptides at pH 7.4, interaction with cationic membranes is of interest.To this end, vesicles were prepared including cationic 1,2-dioleoyl-3-trimethylammoniumpropane (DOTAP), POPC–SM–DPPE–DOTAP (40 + 33 + 12 + 15).Membrane destabilization and fusion of vesicles were demonstrated by fluorescent lipid mixing, contents leakage and changes in the vesicle size. Changes in peptide conformation upon binding to the vesicles and at different pH were measured as changes in tryptophan fluorescence and accessibility to aqueous collisional quenching. Experimental Chemicals VP3(110–121) was synthesized by solid-phase methodology and purified by reversed-phase chromatography as described elsewhere.7 Its purity was higher than 95% and its identity was confirmed by amino acid analysis and electrospray MS.Stock standard peptide solutions of 1–2 mm were prepared in 20% DMSO–water; the peptide was first dissolved in pure DMSO and then diluted with water to the final concentration (4% DMSO). 1-Palmitoyl-2-oleoylglycero-sn-3-phosphocholine (POPC), 1,2-dioleoyl-3-trimethylammoniumpropane (DOTAP), N-(7-nitro-2-1,3-benzoxadiazol-4-yl) dioleoylphos- † Presented at the VIIIth International Symposium on Luminescence Spectrometry in Biomedical and Environmental Analysis, Las Palmas de G.C., Spain, May 26–29, 1998.Analyst, 1998, 123, 2251–2256 2251phatidylethanolamine (NBD-PE) and N-(lissamine rhodamine B sulfonyl)dioleoylphosphatidylethanolamine (Rh-PE) were obtained from Avanti Polar Lipids (Alabaster, AL, USA). Sphingomyelin (SM) and 1,2-dipalmitoylphosphatidylethanolamine (DPPE) were purchased from Sigma (St. Louis, MO, USA). 1-Aminonaphthalene-3,6,8-trisulfonic acid (ANTS) and N,NA-p-xylenebis(pyridinium bromide) (DPX) were purchased from Molecular Probes (Eugene, OR, USA).HMA buffer (10 mm HEPES, 10 mm MES, 10 mm sodium acetate, 100 mm NaCl) (pH 7.4) was used throughout the experiments. During the assays, 1 m HCl was added to decrease the pH to the desired value. HMA buffer has a linear pH response to the volume of HCl added over the pH range 3.0–7.5. All buffers were obtained from Sigma. Vesicle preparation Small unilamellar vesicles (SUVs) of POPC–SM–DPPE (47 + 39 + 14), or POPC–SM–DPPE–DOTAP (40 + 33 + 12 + 15), alone or with the fluorescent probes NBD-PE and/or Rh-PE, were prepared by evaporation of a mixture of the lipids in CHCl3–CH3OH (2 + 1 v/v).The dried film was hydrated for a lipid concentration of 20 mm and then sonicated in a G112SPIT bath type sonicator (Laboratory Supplies, Hicksville, NY, USA) above the gel–fluid transition temperature until a clear dispersion was obtained (typically 2–4 min).Large unilamellar vesicles (LUVs) of the same lipid compositions were prepared by hydration of the lipid film and sonication in an ultrasonic bath, followed by 10 freeze–thaw cycles to ensure a homogeneous mixture of multilamellar vesicles (MLV). This preparation was extruded 10 times through two 100 nm pore-size polycarbonate filters (Nuclepore, Pleasanton, CA, USA) in a high-pressure extruder (Lipex Biomembranes, Vancouver, Canada). Vesicles were annealed for 1 h above their transition temperature.For the ANTS–DPX fusion assay the vesicles contained HMA buffer (pH 7.4) and either (i) 25 mm ANTS, (ii) 90 mm DPX or (iii) 12.5 mm ANTS and 45 mm DPX. The vesicles were separated from unencapsulated material on Sephadex G-25 (Pharmacia, Uppsala, Sweden), equilibrated with HMA buffer (pH 7.4). To calculate the lipid concentration after passage of the vesicles through the column, vesicles of the same composition but doped with 0.1% Rh-PE as a marker and without encapsulated material were filtered, and the amount of lipid recovered was calculated according to the Rh-PE fluorescence intensity taking the sample di into account.By this procedure, we calculated a 77 ± 10% lipid recovery and a dilution factor of 4 ± 0.4 (mean of six different samples). These vesicles were used within 10 h. Typically, the Sephadex G-25 column could be reused 4–5 times after washing. Light scattering Vesicle size was determined by dynamic light scattering with a Malvern II-C Autosizer at 25 °C.The mean diameter was 50 nm for SUV and 100 nm for LUV vesicles (polydispersity 0.1 and 0.2, respectively). The same instrument was used to measure changes in the size distribution of vesicles induced by VP3 and/ or HCl under the conditions where leakage of contents with or without lipid mixing occur. An aliquot of vesicles from a 15 mm standard solution was added to the cuvette containing HMA buffer, and then VP3(110–121) was added from a standard solution, followed by HCl to the desired pH.The effect of HCl alone on the scattering of the vesicles was negligible. Tryptophan fluorescence and quenching experiments Fluorescence measurements were carried out at 25 °C in HMA buffer (pH 7.4 or 3.4) on an AB-2 spectrofluorimeter (SLMAminco, Urbana, IL, USA) with constant stirring. Tryptophan fluorescence spectra were recorded with an excitation wavelength of 285 nm over an emission range of 300–420 nm.Peptide was added from a standard solution to a final concentration of 5.0 mm, and titrated with SUV or LUV vesicles. Spectra were corrected by subtracting the emission spectra from vesicles at the same lipid concentration. Quenching of tryptophan fluorescence of VP3(110–121) by acrylamide was recorded at 357 nm (excitation at 285 nm) in HMA buffer (pH 7.4). Appropriate amounts of lipid vesicles (LUVs) were added to a solution of 5.0 mm peptide, then HCl was added to pH 5.0 if needed.The quencher was added in increasing amounts from a 3.3 m standard solution in water (final concentration from 0 to 400 mm). Quenching results were analyzed according to the Stern–Volmer equation for static and collisional quenching; however, for practical reasons, we decided to use the slope of a linear fit of the Stern–Volmer plot for concentrations up to 150 mm as a criterion for VP3(110– 121) accessibility to acrylamide: F0/F = 1 + KSV[Q], where F0 and F are the fluorescence intensities in the absence and presence of quencher, respectively, [Q] is the molar concentration of quencher and KSV is the Stern–Volmer quenching constant. At this range of concentrations there is no deviation from linearity (r2 = 0.99).Lipid mixing by resonance energy transfer (RET) The vesicle mixture used for RET contained 0.6% of NBD-PE or Rh-PE codispersed with the unlabeled lipids to obtain SUVs of the desired lipid composition: POPC–SM–DPPE (47 + 39 + 14), or POPC–SM–DPPE–DOTAP (40 + 33 + 12 + 15).Excitation was at 460 nm and fluorescence emission from NBD was monitored at 530 nm, with 4 nm slitwidths. The sensitivity (PMT voltage) was adjusted to 1% for the Raman peak from the buffer blank at the same excitation wavelength. The change in fluorescence was calculated as dF = [(F 2 F0)/(Fmax 2 F0)] 3 100, where F0 and F are the fluorescence intensities before and after the addition of peptide, respectively, and Fmax is the fluorescence after total mixing of lipids, measured with covesicles containing 0.3 mol% of each of the probes at the same total bulk lipid concentration. The effect of lowering the pH was determined by adding 1 m HCl to the cuvette containing the mixture of vesicles either before or after addition of peptide.Control experiments on vesicles without peptide were carried out to determine the influence of the change in pH on the vesicles. If necessary, this contribution (always less than 3%) was subtracted.ANTS/DPX fusion assay for leakage or mixing of aqueous compartments Dequenching of co-encapsulated ANTS and DPX fluorescence resulting from dilution was measured to assess the leakage of aqueous contents.8 In these experiments, LUV vesicles were used to increase the volume of encapsulated probes. Excitation was set at 360 nm and the emission at 530 nm was recorded as a function of time. The scale was calibrated with the fluorescence of the 1 + 1 mixture of ANTS and DPX loaded vesicles taken as 100% leakage; the fluorescence of the same concentration of vesicles containing co-encapsulated ANTS and DPX was taken as 0% leakage.The fluorescence of the lysed vesicles containing both ANTS and DPX (using 2.2 mm deoxycholate) is the same as the fluorescence of the mixture of vesicles with ANTS and vesicles with DPX (1:1 ratio). To determine if non-leaky fusion occurs, ANTS and DPX were encapsulated in separate populations of vesicles and mixed in a 1:1 ratio.Non-leaky fusion results in mixing of aqueous contents and a decrease in ANTS fluorescence due to quenching of ANTS by DPX. 2252 Analyst, 1998, 123, 2251–2256Results Tryptophan fluorescence The fluorescence properties of tryptophan (Trp) depend on the polarity of its environment. The emission spectrum of VP3(110–121) in HMA buffer at pH 7.4 has an emission maximum at 350 nm (excitation for Trp was at 285 nm).The same maximum is seen on binding to POPC–SM–DPPE– DOTAP LUVs (Fig. 1) at this pH, but the intensity of fluorescence decreases significantly. This decrease depends on the lipid concentration, and it is not due to inner filter effects from the vesicles since the solution remains clear (absorbance at 285 nm < 0.2). Similar results were obtained with SUVs of the same lipid composition (not shown). All spectra were corrected from scattering by subtracting a spectrum of the same vesicle concentration without peptide.Also shown in Fig. 1 is the emission spectrum of VP3(110–121) bound to POPC–SM– DPPE–DOTAP LUVs at acidic pH; again, the spectra have the same emission maximum, at 350 nm, but the intensity of fluorescence is even lower. These results indicate binding of the peptide to cationic vesicles and penetration of the peptide in the bilayer at neutral and at acidic pH. Fluorescence quenching of VP3(110–121) by acrylamide Acrylamide was used as an aqueous-phase quencher of tryptophan fluorescence; acrylamide is considered to access all but the most highly buried Trp residues.9 Some Stern–Volmer plots of the quenching of Trp fluorescence of VP3 at different pH values by acrylamide are shown in Fig. 2 and the Stern– Volmer quenching constants for a bimolecular collisional quenching process are given in Table 1. In HMA buffer the Trp residue showed greater accessibility to acrylamide at pH 7.4 than at pH 5.0 or 3.4. Upon addition of cationic LUVs, Trp became less accessible to quenching, suggesting shielding by the lipid bilayer.The degree of shielding was higher at acidic pH, and is indicative of a higher affinity or a deeper penetration of the peptide in the lipid bilayer at this pH. Binding to vesicles of POPC–SM–DPPE also resulted in a decrease in the accessibility of Trp to the quencher; in this case the Stern–Volmer quenching constant was lower than for POPC–SM–DPPE–DOTAP vesicles at the same pH, indicating a different form of the bound peptide.Using SUVs of the same lipid composition gave similar results (not shown). Changes in vesicle size distribution induced by VP3(110–121) Addition of VP3(110–121) to LUV zwitterionic vesicles (POPC–SM–DPPE) did not result in any change in the size distribution of the vesicles determined by dynamic light scattering, even after acidification with HCl from pH 7.4 to 3.4. This suggests that vesicles are not aggregated or solubilized by the peptide.A very different picture emerged when the experiment was carried out with cationic vesicles (POPC–SM– DPPE–DOTAP). The size distribution of these vesicles [Fig. 3(A)] shows only one population of vesicles with a diameter centered at 100 nm and low polydispersity. Subsequent addition of VP3(110–121) did not induce vesicle aggregation at neutral pH, but significant scattering changes followed HCl addition, consistent with the formation of larger particles, as shown in Fig. 3(B) at a final pH of 5.0. Under these conditions, two populations of species are seen, one corresponding 73% of the total and formed by intact vesicles of 100 nm, and the other corresponding to around 25%, formed by larger aggregates of a mean diameter size of 500 nm and more polydisperse. This result is in agreement with fusion of vesicles, as will be described in the following set of experiments. The observed changes are very fast after HCl addition, and are complete in less than mixing time.Control experiments showed that HCl alone did not have any effect on the vesicles. Lipid mixing induced by VP3(110-121) Peptide-induced fusion of vesicles was studied by monitoring lipid mixing. SUV vesicles of POPC–SM–DPPE, or POPC– SM–DPPE–DOTAP containing 0.6% of either NBD-PE or Rh- PE were mixed in a 1 : 1 mole ratio in 1.5 ml of HMA buffer (pH 7.4). Lipid mixing was determined as the increase in RET between the fluorescently labeled lipids, monitoring the decrease in the NBD-PE signal.VP3(110–121) induced fusion of Fig. 1 Tryptophan fluorescence emission spectra of VP3(110–121). Peptide in (a) HMA buffer (pH 7.4) and in the presence of LUVs of POPC– SM–DPPE–DOTAP (40 + 33 + 12 + 15) at (b) pH 7.4 and (c) pH 3.4. Excitation at 285 nm. Peptide concentration, 5 mm; lipid concentration, 1.25 mm. Spectra were corrected by subtracting spectra of buffer or vesicles at the corresponding pH. Fig. 2 Stern–Volmer plots showing the tryptophan fluorescence quenching of VP3(110–121) by acrylamide. VP3(110–121) in buffer at pH (5) 7.4 and (-) 3.4; in the presence of POPC–SM–DPPE–DOTAP (40 + 33 + 12 + 15) at pH (2) 7.4 and (8) 3.4.Other conditions as in Fig. 1. Table 1 Stern–Volmer quenching constants for the quenching of VP3(110– 121) fluorescence by acrylamide. Peptide concentration, 5.0 mm; lipid concentration, 0.8 mm KSV/l mol21 pH In HMA buffer POPC–SM–DPPE–DOTAP POPC–SM–DPPE 7.4 17.46 8.80 7.61 5.0 13.93 7.57 6.65 3.4 13.27 5.90 — Analyst, 1998, 123, 2251–2256 2253POPC–SM–DPPE–DOTAP vesicles only at acidic pH.In Fig. 4, lipid mixing time courses of POPC–SM–DPPE–DOTAP induced by 3 mol% of VP3(110–121) are shown at different pH. The peptide was added to premixed vesicles at pH 7.4, followed by HCl to obtain the desired pH. No lipid mixing was observed at pH 7.4, and the extent of mixing increased on lowering the pH below 6.0. At pH 4.6, lipid mixing was slow and time dependent, but further decreases in pH gave greater and faster lipid mixing, with an initial rapid increase in mixing that was complete in about 200 s, followed by a slow time-dependent mixing that continued for several minutes.The initial phase is indicative of a transient destabilization of the membrane. Control experiments without peptide were carried out and the effect of HCl on the fluorescence of the vesicles, always less than 1%, was subtracted if necessary.The extent of lipid mixing depends on the concentration of peptide, as shown in Fig. 5 for POPC–SM–DPPE–DOTAP vesicles at pH 4.0. Even low peptide concentrations, such as 1 mol%, show a significant level of lipid mixing of around 6% at 1000 s after lowering the pH. If we consider the size of the SUV vesicles, with a diameter of 50 nm, the number of lipid molecules that form each vesicle is approximately 8000, and therefore 4000 molecules will be in the outer monolayer; according to this calculation, 1 mol% of VP3(110–121) represents 80 peptide molecules per vesicle. An interesting observation is the importance of the order of addition: lipid mixing occurs only when acidification follows binding of the peptide to the vesicles, and no effect is seen when VP3(110– 121) is added to the vesicle mixture at acidic pH (Fig. 5, bottom). Interestingly, VP3(110–121) did not induce lipid mixing between vesicles of POPC–SM–PE at any pH in the range 7.4–3.5 (not shown).The possibility that the peptide does not bind to these vesicles is ruled out by the observed change in the fluorescence spectra of the peptide and by the decrease in the Stern–Volmer quenching constant in the presence of the vesicles (Table 1), indicative of shielding of Trp by the lipid interface due to at least partial penetration of the peptide in the vesicle. Leakage or mixing of aqueous contents of the vesicles To determine whether the pH-induced destabilization of membranes by VP3(110–121) corresponded to fusion with contents mixing or leakage, in addition to the observed lipid mixing, we used the ANTS–DPX fusion assay8 with some modifications.10 To determine the ability of our system to detect fusion, vesicles containing both ANTS and DPX were incubated with CaCl2, which showed a time-dependent dequenching of ANTS fluorescence, indicative of calcium-induced fusion.VP3(110–121) does not cause leakage in PC–SM–PE–DOTAP (40 + 33 + 12 + 15) vesicles at pH 7.4, even at concentrations up to 10 mol%.As shown in Fig. 6(A), at 3.2 mol% peptide, leakage was observed immediately after decreasing the pH, under the same conditions where there is also lipid mixing. Leakage is very fast, being complete in less than 10 s, and the extent of leakage depends on the pH; for example, at pH 5.0, 30% of the contents are released within 10 s, followed by a plateau with almost no leakage. This behavior corresponds to the transient lipid mixing observed under the same conditions, Fig. 3 Dynamic light scattering of POPC–SM–DPPE–DOTAP (40 + 33 + 12 + 15) LUVs: (A) vesicle size distribution of the vesicles in HMA buffer (pH 7.4) determined at 25 °C; the same was obtained in the presence of VP3(110–121) at this pH; (B) same vesicles after addition of 3 mol% VP3 and subsequent acidification to pH 5.0. Fig. 4 Effect of pH on lipid mixing induced by VP3(110–121) in POPC– SM–DPPE–DOTAP (40 + 33 + 12 + 15) SUVs.The peptide (3 mol%) was added to a 1 + 1 mixture of vesicles containing 0.6% NBD-PE or Rh-PE in HMA buffer (pH 7.4), and then appropriate amounts of 1 m HCl were added to achieve the indicated pH. Lipid concentration, 134 mm. Excitation at 460 nm, emission at 530 nm. Fig. 5 Effect of VP3(110–121) concentration on lipid mixing in POPC– SM–DPPE–DOTAP (40 + 33 + 12 + 15) SUVs. Top: the indicated concentrations of peptide were added to the vesicles before acidification to a final pH of 4.0 with HCl.Bottom: the HCl was first added to the vesicles to pH 4.0 prior to addition of peptide. Other conditions as in Fig. 4. 2254 Analyst, 1998, 123, 2251–2256suggesting a rapid loss of the membrane-destabilizing properties of the peptide or a weaker interaction. Also here, the order of addition of the different components is essential, and no leakage is seen if the peptide is added to the vesicles previously set at acidic pH. In Fig. 6(B), it is shown that at a fixed final pH of 5.0, the extent of leakage depends on the peptide mole fraction; the kinetics are very fast in all cases, reaching a plateau in a few seconds.Control experiments showed no effect of HCl in the vesicles alone. In the case of PC–SM–PE (47 + 39 + 14) vesicles, leakage was also induced only at acidic pH [Fig. 7(A) at 3 mol% peptide]. In Fig. 7(B), the dependence of the extent of leakage on the peptide concentration is shown at pH 5.0. In these vesicles, the kinetics of leakage are different to those on cationic vesicles (Fig. 6); in this case there is also a fast phase, complete in few seconds, where most of the leakage takes place, but it is followed by a slow time-dependent phase. Surprisingly, in these vesicles, leakage of contents is not accompanied by lipid mixing. The possibility that VP3(110–121) induces non-leaky fusion (contents mixing) was tested by mixing the two populations of vesicles, one with encapsulated ANTS and the other with DPX.Addition of VP3(110–121) at concentrations up to 10 mol% did not induce contents mixing, even at acidic pH, for the two types of vesicles under study. Discussion The capsid protein VP3 of hepatitis A virus is highly exposed in infective virions, and it is also highly immunogenic. Several peptidic sequences from this protein have been synthesized and their immunogenicities evaluated.11 One of the more immunogenic sequences is VP3(110–121), corresponding to a maximum of hydrophilia in the protein7 according to the criteria of Chou and Fasman.Very little is known at the molecular level about the mode of infection of viruses in general and of picornaviruses in particular. Most probably, the virion is internalized in the host cell by endocytosis, and subsequent acidification of the resulting vesicle leads to partial unfolding of some regions of the capsid proteins, including VP3, exposing sequences previously buried within the capsid.12 One of these sequences can destabilize the membrane of the vesicle, or form a pore in such a way that the viral RNA can be transferred to the cytoplasm.In this work, we have shown that the immunogenic sequence VP3(110–121) is also able to cause fusion of lipid membranes, and we propose that it can be involved in the process of viral infection of the hepatic host cells. In this context, it should be noted that fusion proteins are immunogenic in vivo.13 We used unilamelar liposomes of a composition similar to the hepatic cells as membrane models: POPC–SM– DPPE (47 + 39 + 14), and also liposomes incorporating 15% cationic DOTAP. Cationic vesicles are also good models for biological membranes, because animal cells often have positively charged polymers adhering to the plasma membrane; these polymers might promote internalization of anionic compounds via the endocytic pathway,14 a step previous to the acidification of the vesicle by fusion with the endosome.Cationic vesicle interaction is useful for evaluating the electrostatic component in the interaction, given that VP3(110– 121) is a negatively charged peptide. VP3(110–121) does not have any membrane destabilizing effect on either of the two vesicle compositions under study at neutral pH, although it binds to the membranes, as suggested by the changes in peptide tryptophan fluorescence. Also, acrylamide quenching indicates that the peptide penetrates the Fig. 6 Leakage of ANTS–DPX co-encapsulated in POPC–SM–DPPE– DOTAP (40 + 33 + 12 + 15) LUVs (180 mm).(A) Effect of pH: 3.2 mol% VP3(110–121) was added to the vesicles in HMA buffer (pH 7.4) prior to acidification to the indicated final pH with 1 m HCl. (B) Effect of peptide concentration: the indicated concentrations of VP3(110–121) were added to the vesicles prior to acidification to pH 5.0 Excitation at 360 nm, emission at 530 nm. Fig. 7 Leakage of ANTS–DPX co-encapsulated in POPC–SM–DPPE (47 + 39 + 14) LUVs (180 mm).(A) Effect of pH with a constant 3.2 mol% VP3(110–121). (B) Effect of peptide concentration [VP3(110–121) concentrations (mol%) are given on the curves] at final pH 5.0. Other conditions and order of addition of the different components as in Fig. 5. Analyst, 1998, 123, 2251–2256 2255membrane at pH 7.4, at least partially, and more deeply in the vesicles of POPC–SM–DPPE than in the cationic vesicles of POPC–SM–DPPE–DOTAP. The destabilization of membranes accompanying the observed changes in peptide structure and penetration in the bilayer was studied by monitoring the fusion of lipid vesicles as measured by lipid mixing, contents mixing and leakage.At acidic pH, VP3(110–121) has strong destabilizing properties in both cationic and zwitterionic liposomes, as can be concluded from the leakage of aqueous contents from the vesicles’ interior induced by the peptide immediately after lowering the pH below 6.0. Leakage takes place in both cationic and zwitterionic vesicles, and is a very fast process, complete in less than 10 s.The peptide does not induce mixing of contents under any of the conditions studied here. Apparently, neutralization of the charges is required for expressing the membrane destabilizing properties of VP3(110–121), as has been described for other fusogenic peptides.6,15 In order to determine whether this destabilization of the lipid membranes at acidic pH was due to membrane fusion or other processes such as pore formation, vesicle aggregation and lipid mixing were also monitored.In POPC–SM–DPPE–DOTAP vesicles, VP3(110– 121) induced lipid mixing at acidic pH, and we can therefore conclude that the peptide causes leaky fusion of these vesicles at pH below 6.0. This is consistent with the changes in vesicle size and distribution seen by dynamic light scattering, where 25% of the vesicles were in an aggregated form with a mean diameter of 500 nm, but still almost 75% of the vesicles remained with 100 nm size.This indicates that the peptide loses its fusogenic properties, maybe remaining irreversibly bound to the aggregated vesicles. These changes may have physiological significance, in that the pH of the endosomal interior falls in the range 5–6,16 so it is possible that this is the route of viral infection used by HAV. In all cases, lipid mixing, leakage and vesicle aggregation are partial, and after a fast phase the processes are arrested almost completely, suggesting the loss of destabilizing properties of the peptide.The extent of lipid mixing and contents leakage depends on pH and peptide concentration. A very surprising observation is that VP3(110– 121) only displays its fusogenic properties if it was already bound to the vesicles at neutral pH and the medium is acidified afterwards, and absolutely no destabilizing activity is detected if the peptide binds to the vesicles at acidic pH.This indicates that the peptide cannot reach the same form in both cases, and we are planning new experiments to explore this point. The behavior in POPC–SM–DPPE vesicles is significantly different; VP3(110–121) binds to this vesicles, and at pH below 7.4 it also causes aqueous contents leakage, even at low concentrations. Nevertheless, no lipid mixing between these vesicles was observed even at acidic pH and high peptide concentrations. Also, no changes in vesicle size and distribution were observed by dynamic light scattering.Taken together, these observations imply that at acidic pH the peptide destabilizes the membrane of these vesicles, probably by spanning the bilayer and forming channels through which the contents leak, but the vesicles are not fusing. Solubilization can also be ruled out on the basis of the scattering results. We suggest that the ability of VP3(110–121) to induce fusion of membranes under acidic conditions can have a physiological significance in the understanding of the process of cellular infection by HAV.HAV is a non-enveloped virus, therefore it cannot penetrate the cell by inducing fusion of a viral membrane with the cell membrane, as is the case with enveloped viruses; it has been hypothesized that the viral infection in this case occurs by the endosomal route. The pH in the endosome is acidic in the range where VP3(110–121) shows membranedestabilizing properties, therefore it is possible that once in the endosome, this peptide becomes exposed and is able to interact with the membranes. Acknowledgements A. Ch�avez was supported by DGAPA UNAM. Y. Cajal was supported by a contract for the incorporation of Doctors in Spanish research groups (MEC, Spain). This work was supported by grants BIO95-0061-CO3-02 and BIO95- 0061-CO3-03 from CICYT, Spain. References 1 J. L. Nieva, S. Nir, A. Muga, F. M. Go�ni and J. Wilscut, Biochemistry, 1994, 33, 3201. 2 J. M. White, Science, 1992, 258, 917. 3 M. J. Clague, J. R. Knutson, R. Blumenthal and A. Hermann, Biochemistry, 1991, 30, 5491. 4 K. N. J. Burger, S. A. Wharton, R. A. Demel and A. J. Verkleij, Biochim. Biophys. Acta, 1991, 1065, 121. 5 M. Murata, S. Takahashi, Y. Shirai, S. Kagiwada, R. Hishda and S. Ohnishi, Biophys. J., 1993, 64, 724. 6 A. L. Bailey, M. A. Monck a. R. Cullis, Biochim. Biophys. Acta, 1997, 1324, 232. 7 M. García, M. Pujol, F. Reig, M. A. Alsina and I. Haro, Analyst, 1996, 121, 1583. 8 H. Ellens, J. Bentz and F. C. Szoka, Biochemistry, 1985, 24, 3099. 9 G. P. Kurzban, G. Gitlin, E. A. Bayer, M. Wilcheck and P. M. Horowitz, Biochemistry, 1989, 30, 8537. 10 Y. Cajal, J. M. Boggs and M. K. Jain, Biochemistry, 1997, 36, 2566. 11 I. Haro, R. M. Pint�o, J. F. Gonzalez-Dankaart, J. A. P�erez, F. Reig and A. Bosch, Microbiol. Immunol., 1995, 39, 845. 12 J. A. Levy, H. Fraenkel-Conrat and R. A. Owens, in Virology, Prentice Hall, Englewood Cliffs, NJ, 3rd edn., 1994, ch. 11, pp. 271- 292. 13 R. T. Earl, I. M. Hunneyball, E. E. Bilelett and R. J. Mayer, J. Pharm. Pharmacol., 1988, 40, 166. 14 S. Wei-Chiang and H. J.-P. Ryser, Proc. Natl. Acad. Sci. USA, 1978, 75, 1872. 15 C. C. Pak, A. Puri and R. Blumenthal, Biochemistry, 1997, 36, 8890. 16 M. Roeder, R. Bowswe and R. F. Murphy, J. Cell. Physiol., 1987, 131, 200. Paper 8/04562C 2256 Analyst, 1998, 123, 2251–
ISSN:0003-2654
DOI:10.1039/a804562c
出版商:RSC
年代:1998
数据来源: RSC
|
9. |
Fluorimetric flow-injection method for anionic surfactants based on protein–surfactant interactions† |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 2257-2261
D. L. Recalde Ruiz,
Preview
|
PDF (73KB)
|
|
摘要:
Fluorimetric flow-injection method for anionic surfactants based on protein–surfactant interactions† D. L. Recalde Ruiz,a A. L. Carvhalo Torres,b E. Andrés Garcíaa and M. E. Díaz García*a a Department of Physical and Analytical Chemistry, Faculty of Chemistry, University of Oviedo, 33006, Oviedo, Spain. E-mail: medg@dwarf1.quimica.uniovi.es b University of Salvador de Bahía, Chemistry Institute Rua Barao de Geremoabo, Sln, Ondina, Salvador de Bah�ýa, Brazil Received 4th June 1998, Accepted 10th August 1998 Surfactant–protein interactions have been widely used to study the composition of biopolymers, but their application in the quantitative analysis of surfactants has not been exploited.Here the analytical figures of merit of a sensitive spectrofluorimetric flow-injection system for the determination of an anionic surfactant, sodium dodecyl sulfate (SDS), are reported. The proposed method is based on the interactions of SDS with bovine serum albumin, a surface active protein, in the presence of a fluorescent probe, 8-anilino-1-naphthalenesulfonic acid.The linear dynamic range extends from the limit of quantification up to 1 3 1024 m SDS under selected conditions. The relative standard deviation is 4% with a detection limit of 2 3 1026 m SDS. The method was applied to the determination of SDS in river and tap water samples. A mechanism for the sensing chemistry involved is proposed. Surface active agents (surfactants) are a group of chemicals widely used both industrially and domestically.Even though at present most surfactants used are biodegradable, their accumulation or that of their biodegradation products in natural waters results in water pollution, which, in extreme cases, leads to the destruction of surface flora and fauna.123 Analytical methods for the determination of anionic surfactants are numerous. Those based on spectrophotometry constitute an important part, in particular that based on their liquid– liquid extraction as a lipophilic ion-association compound formed between the methylene blue cation and the surfactant anion.4,5 Methylene blue methods require multiple-phase separations from reagents themselves or to minimise interferences produced by anions such as chloride, nitrate or thiocyanate, which also form extractable ion pairs with the methylene blue cation.On the other hand, the batchwise methylene blue method is very troublesome and time consuming. Even the performance of flow-injection liquid–liquid extraction systems is limited by inadequate approaches to the phase separation step at the end of the procedure.6,7 In this paper, a rapid, sensitive and simple method is proposed for the determination of a typical anionic surfactant, sodium dodecyl sulfate (SDS), in natural waters.The method is based on the interaction of the fluorescent probe 8-anilino- 1-naphthalenesulfonate (ANS) with bovine serum albumin (BSA). The fluorescence of ANS is drastically increased on adding it to a BSA solution.8 This enhancement is caused by the movement of ANS molecules from water into the protein structure, the lower polarity of the environment of these molecules resulting in a pronounced increase in their quantum yield.ANS binds non-covalently to proteins and its fluorescence is extremely sensitive to changes in the probe environment, 9 which may result from changes in solvent accessibility, probe mobility and changes in the steric interactions with the surrounding protein.Hence the fluorescence of the ANS–BSA complex, similarly to that of other fluorescent compounds, may be altered by various substances present in the medium (e.g., some metal ions) or by those affecting protein structure. Here, we describe how the interaction of anionic surfactants with the ANS–BSA fluorescent complex can be used to determine the corresponding surfactant. The resulting microenvironmental changes manifest themselves as differences in emission intensity, depending on surfactant concentration.A single-line fluorimetric flow-injection (FI) system is described for the determination of SDS in river and tap waters. Our group has developed a flow-through biosensor system for anionic surfactant determination based on a sensitive roomtemperature phosphorescence phase (Al-Ferron) coated with BSA.10 The results of that investigation and those of the present study allowed us to propose a possible mechanism for the sensing chemistry involved in the approach developed in this work. Experimental Reagents and solutions All chemicals were of analytical reagent grade.SDS (Sigma, St. Louis, MO, USA), cetyltrimethylammonium chloride (CTAC) (Fluka, Buchs, Switzerland) and Triton X-100 (Aldrich, Milwaukee, WI, USA) were used as received. The critical micellar concentration (cmc) of the surfactants used in this study are reported to be 8.3 mm for SDS,11 0.12 mm for CTAC12 and 0.24 mm for Triton X-100.13 BSA, fraction V, purity grade 96–99%, and ANS ammonium salt (ANSA) were obtained from Sigma. 2,3-Naphthalenedicarboxylic acid was obtained from Aldrich, phenol from Merck (Darmstadt, Germany), diphenylamine from Doesder (Barcelona, Spain) and tannic acid from Hopkin and Williams (Barcelona, Spain). The carrier solution consisted of Tris buffer (20 mm, pH 8), BSA (0.1% m/v) and ANS (3 3 1025 m). Buffer solutions: AcOH/NaOAc (Merck), 0.2 m from pH 4 to 5, Na2HPO4–NaH2PO4 (Prolabo, † Presented at the VIIIth International Symposium on Luminescence Spectrometry in Biomedical and Environmental Analysis, Las Palmas de G.C., Spain, May 26–29, 1998.Analyst, 1998, 123, 2257–2261 2257Paris, France), 0.2 m from pH 6 to 7 and TRIS (Aldrich), 20 mm from pH 7.5 to 8.9, were used to study the influence of pH. Freshly prepared distilled, de-ionized water [obtained from a Milli-Q system (Millipore, Bedford, MA, USA)] of, 18 M ½ cm was used throughout.Instruments and flow-injection-set up All fluorescent measurements were made on a Perkin-Elmer (Norwalk, CT, USA) LS-50B fluorescence spectrometer which employs a xenon-pulsed light source. The excitation and emission wavelengths were set at 384 and 470 nm, respectively. The excitation and emission slits were both set at 15 nm. A Gilson (Worthington, OH, USA) Minipuls-2 four channel peristaltic pump, a PTFE rotary injection valve and a 25 ml quartz flow-cell (Hellma, Mulheim, Germany) were used to implement a single line flow-injection system.A reaction coil of 64 cm was used between the injection valve and the flow cell. PTFE tubing of 0.5 mm id was used. Results and discussion The fluorescence probe ANS has remained the most universally used compound in biomembrane research.8 In fact, its fluorescence is extremely sensitive to changes in the probe environment. ANS has been used for the study of the properties of different proteins such as albumin, peroxidase, chymotrypsin and other proteins.14 The fluorescence enhancement after addition of ANS to certain protein molecules can be explained by the stoichiometric binding of ANS to specific binding sites on the protein molecule. In this sense, the majority of protein molecules possess one or two specific binding sites for ANS but albumin has three or five sites.7 The ANS bound to BSA displays several interesting features that can be analytically useful.The fluorescence of the ANS– BSA complex is relatively independent of pH and of the ionic composition of the medium,8 but is quenched by anionic surfactants. The degree of quenching follows the Stern–Volmer equation describing steady state collisional quenching: I0/I = 1 + KC(anionic surfactant) where Io/I is the factorial decrease in fluorescence and K is the Stern–Volmer quenching constant. In order to use this ANS–BSA complex quenching for the fluorimetric FI determination of anionic surfactants, the composition of the ANS–BSA in the carrier was first studied.The results demonstrated that the fractional decrease in the fluorescence of ANS in the presence of SDS (6 3 1025 m) did not change substantially when the fluorophore concentration was in the range 0.3 3 1025–6.4 3 1025 m. As a rest, a carrier containing 3 3 1025 m ANS was used in subsequent experiments (Fig. 1). Increasing the concentration of BSA from 0.001 to 0.2% m/v showed no significant improvement in the fluorescence intensities (see Fig. 1). ANS is known to be a sensitive probe of microenvironment polarity.8 The increase in the relative decrease of fluorescence intensity observed even at very low BSA concentrations could be ascribed to a change in the microenvironment polarity experienced by the ANS probe on passing from a sub-micellar medium (the surfactant concentration is well below its cmc) to an unfolded protein environment.At very high BSA concentrations over that of SDS (e.g., 0.2% m/v), alteration of the protein structure by the surfactant was no longer observed and, consequently, the ratio I0/I started to decrease as ANS could bind to BSA molecules. Hence a concentration of 0.1% m/v BSA was chosen for further studies. The 0.1% m/v BSA solution was stable for 10–12 h. Therefore, freshly prepared solutions should be used daily. The fluorescence quenching of ANS–BSA in the presence of 3.5 3 1025 m SDS was independent of pH in the range 4–8.5.This study was carried out using different kinds of buffer solution and also the same TRIS solution at different pH values. Through this study a pH of 8 ± 0.2 was kept constant using 20 mm Tris buffer. The use of BSA as co-reagent has the advantage of minimising non-specific adsorption of the surfactant to the tubing employed in the FI system, so eliminating carry-over and, hence, irreproducibility. The ionic strength of the carrier solution was studied using NaCl. At pH 8, no influence on the fluorescence intensity was observed in the range 0.05–0.35 m NaCl.This is an important factor as the analytical system is potentially useful for the determination of anionic surfactants in samples of high saline content (e.g., sea-water). The effect of flow rate of the carrier solution was studied at flow rates of 0.4–3 ml min21. Increasing the flow rate caused the peak to narrow, although its height remained substantially constant. A flow rate of 2 ml min21 was selected as a compromise between peak width and reagent consumption.The effect of the volume of the sample injected was examined by varying the sample volume from 70 to 200 ml with a constant concentration of surfactant (3.0 3 1025 m). The fluorescence quenching increased gradually with increase in the sample loop up to 90 ml. However, a further increase in sample volume up to 200 ml caused a slight decrease in fluorescence quenching. Also, the peak became broader with increasing sample volume.A volume of 90 ml was chosen for subsequent experiments. Analytical figures of merit The analytical performance achieved with the proposed system, including calibration equations and linear dynamic range, is given in Table 1. The calibration equation was obtained with similar operational parameters (optimum conditions). The theoretical detection limit calculated on the basis of three times the standard deviation of the background signal was 2 3 1026 m.Fig. 1 Study of the composition of ANS–BSA in the carrier. [SDS] = 6 3 1025 m; (A) [ANS] = 3 3 10 25 m; (B) [BSA] = 0.1% m/v. 2258 Analyst, 1998, 123, 2257–2261Interferences The effect of various cations present in surface waters was studied. A range of solutions was prepared containing 3.5 3 1025 m SDS and different amounts of possible interfering cations. The solutions containing the surfactant plus the potential interfering cation were analysed by the proposed method.The response was compared with that obtained from a “clean” surfactant solution. The results are summarised in Table 2. As expected, of the cations studied only Fe(iii) interfered owing to its dynamic quenching effect on the ANS fluorescence. However, the preparation of similar solutions in the presence of 1% m/v 1,10-orthophenanthroline is effective in removing the interference of Fe(iii). The interference effects of a cationic surfactant, CTAC, and a non-ionic surfactant, Triton X-100, were investigated by a procedure similar to that used for metal cations with solutions containing 9 3 1025 m SDS.As shown in Table 3, the interferences of CTAC and Triton X-100 are strongly dependent on the BSA concentration in the carrier. The results showed that a 0.1% m/v BSA concentration in the carrier not only enhanced the performance of the system, but also minimised the interference of foreign surfactants (an error of 10% was considered).The effects of various species such as phenol, organic carboxylates, amines and other organic materials, e.g., tannin, potentially present in highly polluted water samples, were studied. In each case, solutions were prepared containing 7.0 3 1025 m SDS and increasing amounts of the possible interfering compounds. The results are summarized in Table 4. Sensing chemistry mechanism The results obtained in this work and those from other studies in our research group10 suggest that a possible mechanism for the sensing chemistry could be explained as shown in Fig. 2. In the absence of an anionic surfactant, the ANS probe is incorporated into the amphiphilic regions of albumin. Excitation of ANS bound to BSA results in a stronger fluorescent emission than that observed when ANS is free in an aqueous environment. This could be explained by taking into account the protective environment provided by the protein, which seems to prevent possible non-radiative decays of the excited fluorescence probe (route a).If SDS is present in the system (route b), the protein unfolding process takes place and the organic dye is set free. Consequently, the fluorophore shows the same spectral characteristics as in an aqueous medium. It is worth mentioning that commonly used anionic surfactants, such as SDS and sodium dodecyl sulfonate, generally denature proteins whereas non-ionic surfactants do not. For anionic surfactants, initial binding occurs to the cationic sites on the protein surface, inducing protein unfolding, thus exposing hydrophobic binding sites previously buried in the core of the tertiary structure.15 Taking into account the above mechanism, it is possible to explain the interference effect of cationic and non-ionic surfactants in the system.As is observed in Table 3, if the protein concentration is increased the selectivity of the method favours the anionic surfactant.This enhanced selectivity could be explained through a competitive mechanism (shown schematically in Fig. 3) between the anionic surfactant and the interfering micellar system. At low BSA concentration interfering micelles are able to solubilize the ANS probe molecules, thus leaving the protein “clean”, free of the fluorescent probe. Addition of anionic surfactant in this case could not have any affect on the ANS fluorescence. In contrast, at high BSA concentration levels, ANS is firmly retained into the protein structure so that foreign micelles are unable to extract them from the protein complex. Addition of anionic surfactant then follows the expected behaviour, quenching the BSA–ANS fluorescence.The proposed mechanism takes into account a situation in which micelles are present. However, the results of our studies, Table 1 Analytical figures of merit Parameter Value Linear dynamic range From LQa up to 1 3 1024 m Calibration equation (I0/I) 2 1 = 0.6766[SDS] + 0.007 Correlation coefficient 0.9980 Detection limit 2 3 1026 m RSD 4% Sample throughput 22 h21 a Limit of quantification.Table 2 Effect of co-existing cations on the determination of SDS. [SDS] = 3.5 3 1025 m; [ANS] = 3 3 1025 m; [BSA] = 0.1% m/v Molar ratio, Recovery of Cation cation: SDS SDS (%) Ca2+ 7 : 1 98 ± 2 Mg2+ 12 : 1 97 ± 2 Na+ 12 : 1 98 ± 2 Fe3+ 5 : 1 87 ± 2 Fe3+/orthophenanthroline 5 : 1 96 ± 2 Table 3 Influence of CTAC and Triton X-100 on the determination of SDS. [ANS] = 3 3 1025 m; [BSA] = 0.1% m/v; [SDS] = 9 3 10 25 m Molar ratio, Recovery of CTAC : SDS SDS (%) [BSA] (% m/v) 1 : 1 45 ± 4 1023 9 : 1 nda 1 : 1 110 ± 4 1022 9 : 1 32 ± 2 1 : 1 98 ± 5 1021 9 : 1 85 ± 2 18 : 1 86 ± 2 Molar ratio, Triton X-100 : SDS 0.5 : 1 80 ± 2 1023 1 : 1 67 ± 4 2 : 1 58 ± 4 0.5 : 1 96 ± 2 1022 1 : 1 80 ± 2 2 : 1 69 ± 2 0.5 : 1 94 ± 2 1021 1 : 1 82 ± 2 2 : 1 70 ± 4 a Not determined.Table 4 Influence of organic compounds in highly polluted waters on the determination of SDS.[ANS] = 3 3 1025 m; [BSA] = 0.1% m/v; [SDS] = 7 3 1025 m Recovery of SDS (%) Organic substance 1 : 1a 1 : 2a 1 : 5a 2,3-Naphthalenedicarboxylic acid 91 84 83 Tannic acid 103 104 104 Phenol 95 90 94 Diphenylamine 119 162 222 a Molar ratio : SDS : organic substance. Analyst, 1998, 123, 2257–2261 2259carried out both below and above the cmc of foreign surfactants, indicate that premicellar aggregates should be formed at surfactant concentrations below their cmc.In fact, evidence exists for the premicellar formation of dimers, trimers and smaller aggregates than micelles.16 Hence the proposed mechanism could be applied also for foreign surfactant concentrations below their reported cmc values, if premicellar structure are considered. Application to real samples To examine the applicability of the method, certain amounts of standard SDS solutions were added to tap, river and sea-water samples and analysed according to the proposed method.Samples were collected in glass bottles using formaldehyde as preservative and they were filtered prior to analysis using 0.45 mm nylon membranes (Supelco, Bellefonte, PA, USA). The results obtained, including recovery and precision, are summarised in Table 5. The results demonstrate that the method can be employed satisfactorily for the determination of SDS in surface waters, even in samples with a high saline content such as sea-water.Conclusions Flow injection analysis with fluorescence detection based on protein–surfactant interactions makes possible a sensitive and fast method for the determination of SDS in surface water samples. We have demonstrated that it is possible to manipulate the medium composition (BSA concentration) to increase the selectivity of the method and avoid a separation step. Acknowledgments The Instituto de Cooperación Iberoamericana is gratefully acknowledged for an ICI grant (D.L.R.R) and a MUTIS grant Fig. 2 Schematic diagram of the proposed sensing mechanism. Fig. 3 Schematic mechanism of surfactants interference from foreign surfactants. Table 5 Analytical results for spiked samples Recovery RSD Sample Added/1025 m (%) (%)a Tap water (March) 3.5 98 1 River waters (April and May) 3.5 98 1 River waters (December) 1.7 96 2 Sea-water 7.0 109 4 Sea-water 3.5 93 2 a Each value is the mean of three independent determinations. 2260 Analyst, 1998, 123, 2257–2261(A.L.C.T.).Financial support from CICYT (Project SAF 96/1484) is acknowledged. References 1 M. A. Lewis, Water Res., 1991, 25, 101 2 J. A. Field, T. M. Field, T. Poiger, H. Siegriut and W. Griger, Water Res., 1996, 29, 1301. 3 M. Ahel, J. MacEvoy and W. Griger, Environ. Pollut., 1993, 79, 243. 4 T. Aboul-Kassim and R. T. B. Simoneit, Crit. Rev. Environ. Sci. Technol., 1993, 23(4), 325. 5 L. Sánchez, Aspectos Ecológicos de los Detergentes, Gestió i Promoció, Barcelona, Spain, 1995. 6 K. Backstron, L. G. Danielsson and L. Nord, Anal. Chim. Acta, 1986, 187, 255. 7 L. N. Moskirn, J. Simon, P. Loffler, N. V. Michailova and D. N. Nicolaeva, Talanta, 1996, 43, 819. 8 J. Slavik, Biochim. Biophys. Acta, 1982, 694, 1. 9 D. V. Naik, W. L. Paul, R. M. Threatte and S. G. Schulman, Anal. Chem., 1975, 47, 267. 10 R. Badía and M. E. Díaz-García, Anal. Chim. Acta, 1998, 371, 73. 11 A. Berthod, S. H. Brooks and J. G. Dorsey, J. Colloid Interface Sci., 1988, 122 (2), 514. 12 L. J. Clive Love, J. G. Habarta and J. G. Dorsey, Anal. Chem., 1984, 56, 1132A. 13 D. Heinl and J. Sauer, Tetrahedron Lett., 1994, 35 (43), 7931. 14 M. N. Jones and A. Brass, in Food Polymers, Gels and Colloids, ed E. Dickinson, Special Publication No. 82, Royal Society of Chemistry, Cambridge, 1991, p. 65. 15 K. Ibel, R. P. May, K. Kirschner, H. Szadkowsky, E. Mascher and P. Lundahl, Eur. J. Biochem., 1990, 190, 31. 16 J. H. Fendler and E. J. Fendler, Catalysis in Micellar and Macromolecular Systems, Academic Press, London, 1975. Paper 8/04224A Analyst, 1998, 123, 2257–2261 2261
ISSN:0003-2654
DOI:10.1039/a804224a
出版商:RSC
年代:1998
数据来源: RSC
|
10. |
Spectrofluorimetric study of the degradation of α-amino β-lactam antibiotics catalysed by metal ions in methanol† |
|
Analyst,
Volume 123,
Issue 11,
1998,
Page 2263-2266
P. Gutiérez Navarro,
Preview
|
PDF (71KB)
|
|
摘要:
Spectrofluorimetric study of the degradation of a-amino b-lactam antibiotics catalysed by metal ions in methanol† P. Gutiérez Navarro,* A. El Bekkouri and E. Rodríguez Reinoso Department of Physical Chemistry, Faculty of Pharmacy, University of Granada, E-18071 Granada, Spain. E-mail: mpgn@platon.ugr.es Received 16th June 1998, Accepted 9th September 1998 The a-aminopenicillins ampicillin and amoxicillin and a cephalosporin, cephalothin, give rise to a fluorescent product when their methanolic solutions are incubated for prolonged time periods. The process also occurs in the presence of the metal ions Cd2+, Co2+ and Zn2+.The effects of the different ions on the emission and excitation wavelengths and the appearance rate of the fluorophore were studied. The appearance of the fluorescent product was zero order for ampicillin and amoxicillin in metal ion-free solution and solutions with Cd2+ and Zn2+, whereas in the presence of Co2+ ion it was first order under the experimental conditions used; for cephalothin it was first order in all cases.Apparent fluorescent compound formation rates were calculated in the zero-order reactions and rate constants in the first-order reactions. The activation energy of the formation reaction of the fluorescent products of amoxicillin and ampicillin was calculated from a study of the reactions at four temperatures; all the values recorded were between 34 and 118 kJ mol21. As a possible mechanism for the formation of these products, cyclization of the penamaldic derivative of the antibiotic, which is formed in the first stage of the methanolytic reaction, is proposed.Introduction Several fluorimetric methods have been described for the determination of a-aminopenicillins and a-aminocephalosporins in aqueous solution. These methods measure the fluorescence of particular reaction products of the antibiotics. Thus, when ampicillin was heated in a formaldehyde solution a fluorescent product was formed,1 later identified as 2-hydroxy- 3-phenyl-6-methylpyrazine.2 This same product was also formed when the a-aminobenzylpenicilloic acid produced by the alkaline hydrolisis of ampicillin was treated with mercury( ii) chloride.3 More recently, Baker4 reported the formation of a fluorescent product when sodium ampicilloate was treated with ascorbic acid, EDTA and modified Lowry A reagent.Furthermore, the reaction of a-aminocephalosporins, cephalexin, cefaclor and cephradine in alkaline medium also led to the formation of a fluorescent compound.5 In earlier work,6 our group found that a-aminopenicillins dissolved in methanol react with the solvent in a reaction catalysed by divalent metal ions, Co2+, Cd2+, Zn2+, etc., producing the corresponding penamaldic derivative of the antibiotic.Moreover, this compound in the catalysed reaction is coordinated with the metal ion as a 2 : 1 complex (ligand : metal). Later, we observed that when the solutions were incubated for a long time period, the formation of a fluorescent compound could be detected.In the present study, we analysed several physico-chemical issues related to the appearance of the fluorescent product from ampicillin, amoxicillin and cephalothin; the results may serve as a basis for further research on the development of a method to quantify the antibiotics assayed here. Experimental Substances and reagents Sodium ampicillin (99%) and 6-aminopenicillanic acid (6-APA) were purchased from Aldrich (Milwaukee, WI, USA) and amoxicillin acid, sodium cephalothin, sodium penicillin G and penicillin potassium from Sigma (St.Louis, MO, USA). Amoxicillin in the sodium salt form was kindly donated by Antibioticos (Barcelona, Spain). Cadmium nitrate tetrahydrate was obtained from Panreac (Barcelona, Spain), anhydrous zinc chloride from Sigma and cobalt nitrate hexahydrate from Merck (Darmstadt, Germany). Methanol of over 99% purity was purchased from Merck and atropine was supplied by Boehringer Ingelheim (Ingelheim, Germany).Instrumentation A Shimadzu (Kyoto, Japan) RF 5000 spectrofluorimeter was used for fluorescence measurements. We also used a Perkin- Elmer Lambda 16 spectrophotometer equipped with a Selecta thermostat (±0.1 °C) and a computer equipped with UV Winlab software from Perkin-Elmer (Norwalk, CT, USA). Measurements were performed with 1 cm thick spectrophotometric cells.Kinetic measurements and calculations We recorded the uncorrected fluorescence spectra of methanolic solutions of each antibiotic and their mixtures with metal ions (molar ratio 2 : 1) after different incubation times at a constant temperature of 25 °C. The concentration of antibiotic was 1 3 1024 mol dm23 in all cases. The fluorescence spectra † Presented at the VIIIth International Symposium on Luminescence Spectrometry in Biomedical and Environmental Analysis, Las Palmas de G.C., Spain, May 26–29, 1998.Analyst, 1998, 123, 2263–2266 2263thus obtained were not corrected, either for the solvent or for the lamp intensity. When the fluorescence intensity was used in quantitative measurements, the latter correction was performed by measurement of the fluorescence intensity of a solution of quinine sulfate. This correction had virtually no effect on the results. The apparent zero-order rate was obtained from the slope of the plot of the relative intensity of fluorescence versus time.The first-order rate constant was determined by adjusting the relative fluoresence intensities by the usual methods. The above-described spectral study was also carried out using atropine (alkaline substance) or NaOH. The fluorescence spectrum of 6-aminopenicillanic acid (6-APA) was also obtained in both the absence and presence of different metal ions at different incubation times. In order to study the influence of temperature and to determine the activation energy of the formation of the fluorescent product, two different methods were used, depending on the system under consideration. Thus, one method was employed for the reaction of ampicillin and amoxicillin in both the absence and presence of Cd2+ and Zn2+ ions, and another for the remaining cases.In both procedures the fluorescent compound formation was considered at four temperatures between 20 and 40 °C. In the first procedure, each of the four identically composed solutions was maintained at a different temperature until fluorescence spectra were of sufficient intensity for their interference with the spectrum of the solvent to be negligible.We then recorded the emission spectra of the four solutions at 25 °C for a set incubation time. The measurement of the fluorescence intensity at this temperature, different from the temperature of the kinetic study, was possible because the rate of appearance of the product was low in all cases.The fluorescence intensities at maximum emission were represented according to the Arrhenius equation and the Naperian logarithm of the relative fluorescence intensity was plotted against the inverse of the temperature. For the reaction of ampicillin and amoxicillin in the presence of Co2+, the fluorescence intensities corresponding to the four temperatures were fitted to first-order kinetics and a rate constant was obtained for each of the temperatures. The rate constants in logarithmic form were represented according to the Arrhenius equation.Excitation and emission spectra In all experiments the fluorescence measurements were made at 25 °C. The uncorrected excitation and emission spectra corresponding to each of the systems considered were obtained from methanolic solutions of the antibiotics alone and from their mixtures with each of the metal ions considered. The antibiotic concentration was always 1 3 1024 mol dm23 and that of the metal ion 5 3 1025 mol dm23, so that the molar ratio was 2 : 1, given that this same proportion was found in the methanolytic product obtained in the first stage of the reaction.6 The excitation and emission spectra of 6-APA were obtained under the same conditions.UV spectra To gather data on the simultaneous changes that occur in the UV spectra with the formation of the fluorescent compound, UV spectra were recorded at different reaction times at 25 °C. Moreover, the UV absorption spectrum of a kinetic mixture, cephalothin–Co2+, was recorded until the end of the reaction at four different temperatures, 20, 25, 30 and 40 °C.Results and discussion Properties of the fluorescent products A fluorescent product was obtained from a-aminopenicillins (amoxicillin and ampicillin) and an a-aminocephalosporin (cephalothin) in methanolic solutions that had been incubated for long time periods. The appearance of a fluorophore also occurred in methanolic solutions of these antibiotics in solutions with a metal ion (Cd2+, Zn2+, Co2+).These results are shown in Fig. 1(a) for a methanolic mixture of sodium amoxicillin and Cd (NO3)2·4H2O at a molar ratio of 2 : 1. Fig. 1 also shows the UV spectrum changes that occur simultaneously with the formation of the fluorescent compound. An increase of the UV band located towards 280 nm which has been assigned to the ligand-to-metal tranfer band (sulfur to metal ion) of the complex of the penamaldic derivative of a-aminopenicillin6,7 can be observed. Likewise, a small increase in the absorption at about 360 nm is also observed. This band is related to the fluorescent compound and was used in the fluorimetric analysis (Fig. 2). On the other hand, sodium penicillin G and potassium penicillin V, Fig. 1 (a) Plots of the fluorescence intensity versus wavelength recorded at time intervals of aproximately 24 h using a solution of sodium amoxicillin (1 3 1024 mol dm23) and cadmium nitrate (5 3 1025 mol dm23) in methanol (lex = 359 nm) and (b) absorption spectra measured for the same time range with the same solution (25 °C ).Fig. 2 Uncorrected excitation and emission fluorescence spectra of the compound of sodium ampicillin and zinc chloride in methanol (25 °C). 2264 Analyst, 1998, 123, 2263–2266which lack the a-amino group of ampicillin and amoxicillin, form no fluorescent product. These data show that the aminobenzyl group is necessary for the formation of the fluorescent product.The uncorrected excitation and emission spectra of each of the reaction products were obtained. Fig. 2 shows those corresponding to sodium ampicillin–Zn2+, and the maximum uncorrected excitation and emission wavelengths of the different fluorescent products are listed in Table 1. This table gives the excitation wavelengths used in the fluorescence measurements, since some excitation spectra presented two maxima. The fluorescence spectra of methanolic solutions of 6-APA and of this acid with the different metal ions studied, reveal that 6-APA gives rise to a fluorescent product whose properties are very different from those for antibiotics, in terms of both excitation and emission wavelengths and fluorescence intensity; the latter is negligible compared with that of antibiotic products.Therefore, these results rule out the idea that the fluorescent products obtained from antibiotics may be formed from 6-APA. Furthermore, 6-APA is not a product of the methanolytic reaction,6 and we therefore suggest that the fluorescent product is not associated with 6-APA.Effect of time on the formation of the fluorescent product As mentioned above, the fluorescence intensity increases with time (Fig. 1). A plot of the relative intensity of fluorescence versus time for ampicillin alone and in solution with the different metal ions reveals that the formation of the fluorogen is directly proportional to time, except for ampicillin–Co2+, where the fluorescence intensity–time curve is characteristic of a first-order reaction (Fig. 3). Similar behaviour is shown by amoxicillin, where the formation of the fluorescent product is zero order in all cases except for the amoxicillin–Co2+ mixture. On the other hand, the formation of the fluorescent product from cephalothin follows first-order kinetics in all cases. Table 2 gives apparent zero-order rates and first-order constant values for the corresponding reactions.The UV spectra recorded at the four temperatures for a cephalothin–Co2+ mixture incubated until the end of the reaction were parallel and very close to each other. This suggests that the formation reaction of the fluorescent compound is practically reversible and hence at the end of the reaction there is no equilibrium. Effect of the temperature on the rate of formation of the fluorescent product In those cases where the appearance of the fluorescent product is directly proportional to time, the fluorescence intensities of the kinetic solutions were measured at a set time, as described under Experimental.The fluorescence intensities were plotted logarithmically against the inverse of the temperature, and a straight line was obtained in each case, the slope giving the activation energy. The curves for the ampicillin–Zn2+ system, shown in Fig. 4(a) and (b), illustrate this procedure. When the formation reaction of the fluorescent product is adjusted to a first-order kinetic process, which occurred with amoxicillin–Co2+, ampicillin–Co2+ and cephalothin in both the presence and absence of the metal ions, the formation constants of the product were determined at the different temperatures.The rate constants were plotted according to the Arrhenius equation, and a straight line was obtained in all cases, the slopes giving the activation energy. The formation reactions of the fluorescent product of amoxicillin and ampicillin have relatively high values of the activation energy (34–118 kJ mol21; Table 3), as expected given the slowness of these reactions. The zinc results were much lower than the others.Effect of the addition of bases on the intensity of fluorescence and the appearance rate of the fluorescent product Fig. 5 shows the relative intensity of fluorescence plotted against time for methanolic solutions of ampicillin and of ampicillin in the presence of atropine.The data show that the addition of atropine to the reactive mixture greatly increases the fluorescence intensity, and that the effect is even greater in the case of NaOH. However, the presence of bases does not modify the reaction rates, which can be deduced from the parallelism of the two straight lines in Fig. 5. The fluorescent product To date, only one fluorescent product derived from aaminopenicillin antibiotics has been isolated and identified,2 namely 2-hydroxy-3-phenyl-6-methylpyrazine formed as a result of the reaction of ampicillin or ampicilloate with formaldehyde1 or of ampicilloate with mercury(ii) ion.3 In Table 1 Exitation and emission fluorescence wavelengths (nm) obtained after 72 h of reaction from different (antibiotic–metal) complexes and in the absence of the metal ion.All spectra are uncorrected Amoxicillin Ampicillin Cephalothin Metal ion lex lem lex lem lex lem None 362 442 350 438 349 447 Cd2+ 367 450 366 443 330 447 Co2+ 368 449 366 443 350 433 Zn2+ 367 448 365 444 362 445 Fig. 3 Relative intensity of fluorescence versus reaction time for the fluorescent product of the (a) ampicillin–Cd2+ and (b) ampicillin–Co2+ systems (25 °C). Table 2 Values of the apparent rate Vap (arbitrary units of fluorescence h21) and the rate constant k (h21) for the formation of the different fluorescent compounds at 25 °C Amoxicillin Ampicillin Cephalothin Metal ion k 3 102 Vap 3 102 k 3 103 Vap 3 102 k 3 103 Vap None — 7.205 — 3.669 2.370 — Cd2+ — 7.164 — 3.766 2.390 — Zn2+ — — — 4.171 1.300 — Co2+ 1.611 — 7.0 — 2.840 — Analyst, 1998, 123, 2263–2266 2265addition, the formation of this fluorescent compound has been suggested4 in the reaction of ampicillin and ampicilloate heated in presence of ascorbic acid, EDTA and modified Lowry reagent (prepared from copper sulfate and potassium sodium tartrate).The fluorescent properties of the compounds formed in the present study are different from those reported for other fluorescent derivatives of ampicillin.However, this comparison is not wholly accurate, given that we obtained the spectra in methanolic media, whereas the spectra described in the literature were obtained in aqueous medium. Nevertheless, the spectral properties of the different compounds depend to some degree on the presence of the metal ion, as can be deduced from Table 1, and this may indicate that this ion intervenes in the fluorescent product.Furthermore, in a previous study6 we isolated and identified the product formed in the first stage of the methanolytic reaction of ampicillin catalysed by Cd2+ ions as a complex formed by the coordination of the metal ions to two molecules of the penamaldic derivative from the methanolysis of the antibiotic. These facts suggest that at least for reactions in the presence of metal ions, the fluorescent product is the result of a cyclization involving the amino group of the lateral chain of the molecule of the compound formed in the first stage of the reaction.However, the exact structure of this compound remains unknown, and will be the subject of further research by our group. References 1 W. J. Jusko, J. Pharm. Sci., 1971, 60, 728. 2 R. H. Barbhaiya, J. Pharm. Pharmacol., 1978, 30, 224. 3 K. Miyazaki, O. Ogino and T. Arita, Chem. Pharm. Bull., 1974, 22, 1910. 4 W. L. Baker, Analyst, 1997, 122, 447. 5 F. A. Aly, M. M. Hefnawy and F. Belal, Anal. Lett., 1996, 29, 117. 6 A. Márquez, P. Gutiérrez and P. J. Martínez, Talanta, 1998, 46, 101. 7 R. S. Alexander, L. L. Kiefer, C. A. Fierke and D. W. Christianson, Biochemistry, 1993, 32, 1510. Paper 8/04577A Fig. 4 (a) Fluorescence spectra, recorded at 25 °C, of a methanolic solution of sodium ampicillin (1024 mol dm23) and zinc chloride, incubated for 72 h at different temperatures (20, 25, 30 and 40 °C); (b) Arrhenius plot using fluorescence values taken after a set reaction time, for (8) amipicillin–Cd2+, (5) ampicillin, (2) ampicillin–Zn2+, and (:) ampicillin–Co2+. Table 3 Activation energies (kJ mol21) for the formation of the different fluorescent products of amoxicillin and ampicillin. In all cases, except for zinc–antibiotic systems, the activation energy is the average of three values which were obtained from fluorescent data taken at time intervals of 24 h Metal ion Amoxicillin Ampicillin None 114.6 81.8 Cd2+ 87.4 99.4 Zn2+ 34.6 34.6 Co2+ 97.5 117.7 Fig. 5 Relative fluorescence intensity change versus reaction time in the formation of the fluorescent product of ampicillin (1024 mol dm23) (a) in the absence and (b) the presence of atropine (25 °C). 2266 Analyst, 1998, 123, 2263–2266
ISSN:0003-2654
DOI:10.1039/a804577a
出版商:RSC
年代:1998
数据来源: RSC
|
|